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> |
b69408e8 | 52 | #include <linux/mm_inline.h> |
52d4b9ac | 53 | #include <linux/page_cgroup.h> |
cdec2e42 | 54 | #include <linux/cpu.h> |
158e0a2d | 55 | #include <linux/oom.h> |
08e552c6 | 56 | #include "internal.h" |
d1a4c0b3 | 57 | #include <net/sock.h> |
4bd2c1ee | 58 | #include <net/ip.h> |
d1a4c0b3 | 59 | #include <net/tcp_memcontrol.h> |
8cdea7c0 | 60 | |
8697d331 BS |
61 | #include <asm/uaccess.h> |
62 | ||
cc8e970c KM |
63 | #include <trace/events/vmscan.h> |
64 | ||
a181b0e8 | 65 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
68ae564b DR |
66 | EXPORT_SYMBOL(mem_cgroup_subsys); |
67 | ||
a181b0e8 | 68 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
6bbda35c | 69 | static struct mem_cgroup *root_mem_cgroup __read_mostly; |
8cdea7c0 | 70 | |
c255a458 | 71 | #ifdef CONFIG_MEMCG_SWAP |
338c8431 | 72 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ |
c077719b | 73 | int do_swap_account __read_mostly; |
a42c390c MH |
74 | |
75 | /* for remember boot option*/ | |
c255a458 | 76 | #ifdef CONFIG_MEMCG_SWAP_ENABLED |
a42c390c MH |
77 | static int really_do_swap_account __initdata = 1; |
78 | #else | |
79 | static int really_do_swap_account __initdata = 0; | |
80 | #endif | |
81 | ||
c077719b | 82 | #else |
a0db00fc | 83 | #define do_swap_account 0 |
c077719b KH |
84 | #endif |
85 | ||
86 | ||
d52aa412 KH |
87 | /* |
88 | * Statistics for memory cgroup. | |
89 | */ | |
90 | enum mem_cgroup_stat_index { | |
91 | /* | |
92 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
93 | */ | |
94 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
d69b042f | 95 | MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */ |
d8046582 | 96 | MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */ |
bff6bb83 | 97 | MEM_CGROUP_STAT_SWAP, /* # of pages, swapped out */ |
d52aa412 KH |
98 | MEM_CGROUP_STAT_NSTATS, |
99 | }; | |
100 | ||
af7c4b0e JW |
101 | static const char * const mem_cgroup_stat_names[] = { |
102 | "cache", | |
103 | "rss", | |
104 | "mapped_file", | |
105 | "swap", | |
106 | }; | |
107 | ||
e9f8974f JW |
108 | enum mem_cgroup_events_index { |
109 | MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */ | |
110 | MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */ | |
456f998e YH |
111 | MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */ |
112 | MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */ | |
e9f8974f JW |
113 | MEM_CGROUP_EVENTS_NSTATS, |
114 | }; | |
af7c4b0e JW |
115 | |
116 | static const char * const mem_cgroup_events_names[] = { | |
117 | "pgpgin", | |
118 | "pgpgout", | |
119 | "pgfault", | |
120 | "pgmajfault", | |
121 | }; | |
122 | ||
7a159cc9 JW |
123 | /* |
124 | * Per memcg event counter is incremented at every pagein/pageout. With THP, | |
125 | * it will be incremated by the number of pages. This counter is used for | |
126 | * for trigger some periodic events. This is straightforward and better | |
127 | * than using jiffies etc. to handle periodic memcg event. | |
128 | */ | |
129 | enum mem_cgroup_events_target { | |
130 | MEM_CGROUP_TARGET_THRESH, | |
131 | MEM_CGROUP_TARGET_SOFTLIMIT, | |
453a9bf3 | 132 | MEM_CGROUP_TARGET_NUMAINFO, |
7a159cc9 JW |
133 | MEM_CGROUP_NTARGETS, |
134 | }; | |
a0db00fc KS |
135 | #define THRESHOLDS_EVENTS_TARGET 128 |
136 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
137 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 138 | |
d52aa412 | 139 | struct mem_cgroup_stat_cpu { |
7a159cc9 | 140 | long count[MEM_CGROUP_STAT_NSTATS]; |
e9f8974f | 141 | unsigned long events[MEM_CGROUP_EVENTS_NSTATS]; |
13114716 | 142 | unsigned long nr_page_events; |
7a159cc9 | 143 | unsigned long targets[MEM_CGROUP_NTARGETS]; |
d52aa412 KH |
144 | }; |
145 | ||
527a5ec9 JW |
146 | struct mem_cgroup_reclaim_iter { |
147 | /* css_id of the last scanned hierarchy member */ | |
148 | int position; | |
149 | /* scan generation, increased every round-trip */ | |
150 | unsigned int generation; | |
151 | }; | |
152 | ||
6d12e2d8 KH |
153 | /* |
154 | * per-zone information in memory controller. | |
155 | */ | |
6d12e2d8 | 156 | struct mem_cgroup_per_zone { |
6290df54 | 157 | struct lruvec lruvec; |
1eb49272 | 158 | unsigned long lru_size[NR_LRU_LISTS]; |
3e2f41f1 | 159 | |
527a5ec9 JW |
160 | struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1]; |
161 | ||
f64c3f54 BS |
162 | struct rb_node tree_node; /* RB tree node */ |
163 | unsigned long long usage_in_excess;/* Set to the value by which */ | |
164 | /* the soft limit is exceeded*/ | |
165 | bool on_tree; | |
d79154bb | 166 | struct mem_cgroup *memcg; /* Back pointer, we cannot */ |
4e416953 | 167 | /* use container_of */ |
6d12e2d8 | 168 | }; |
6d12e2d8 KH |
169 | |
170 | struct mem_cgroup_per_node { | |
171 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
172 | }; | |
173 | ||
174 | struct mem_cgroup_lru_info { | |
175 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
176 | }; | |
177 | ||
f64c3f54 BS |
178 | /* |
179 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
180 | * their hierarchy representation | |
181 | */ | |
182 | ||
183 | struct mem_cgroup_tree_per_zone { | |
184 | struct rb_root rb_root; | |
185 | spinlock_t lock; | |
186 | }; | |
187 | ||
188 | struct mem_cgroup_tree_per_node { | |
189 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
190 | }; | |
191 | ||
192 | struct mem_cgroup_tree { | |
193 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
194 | }; | |
195 | ||
196 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
197 | ||
2e72b634 KS |
198 | struct mem_cgroup_threshold { |
199 | struct eventfd_ctx *eventfd; | |
200 | u64 threshold; | |
201 | }; | |
202 | ||
9490ff27 | 203 | /* For threshold */ |
2e72b634 | 204 | struct mem_cgroup_threshold_ary { |
748dad36 | 205 | /* An array index points to threshold just below or equal to usage. */ |
5407a562 | 206 | int current_threshold; |
2e72b634 KS |
207 | /* Size of entries[] */ |
208 | unsigned int size; | |
209 | /* Array of thresholds */ | |
210 | struct mem_cgroup_threshold entries[0]; | |
211 | }; | |
2c488db2 KS |
212 | |
213 | struct mem_cgroup_thresholds { | |
214 | /* Primary thresholds array */ | |
215 | struct mem_cgroup_threshold_ary *primary; | |
216 | /* | |
217 | * Spare threshold array. | |
218 | * This is needed to make mem_cgroup_unregister_event() "never fail". | |
219 | * It must be able to store at least primary->size - 1 entries. | |
220 | */ | |
221 | struct mem_cgroup_threshold_ary *spare; | |
222 | }; | |
223 | ||
9490ff27 KH |
224 | /* for OOM */ |
225 | struct mem_cgroup_eventfd_list { | |
226 | struct list_head list; | |
227 | struct eventfd_ctx *eventfd; | |
228 | }; | |
2e72b634 | 229 | |
c0ff4b85 R |
230 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
231 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 232 | |
8cdea7c0 BS |
233 | /* |
234 | * The memory controller data structure. The memory controller controls both | |
235 | * page cache and RSS per cgroup. We would eventually like to provide | |
236 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
237 | * to help the administrator determine what knobs to tune. | |
238 | * | |
239 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
240 | * we hit the water mark. May be even add a low water mark, such that |
241 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
242 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
243 | */ |
244 | struct mem_cgroup { | |
245 | struct cgroup_subsys_state css; | |
246 | /* | |
247 | * the counter to account for memory usage | |
248 | */ | |
249 | struct res_counter res; | |
59927fb9 HD |
250 | |
251 | union { | |
252 | /* | |
253 | * the counter to account for mem+swap usage. | |
254 | */ | |
255 | struct res_counter memsw; | |
256 | ||
257 | /* | |
258 | * rcu_freeing is used only when freeing struct mem_cgroup, | |
259 | * so put it into a union to avoid wasting more memory. | |
260 | * It must be disjoint from the css field. It could be | |
261 | * in a union with the res field, but res plays a much | |
262 | * larger part in mem_cgroup life than memsw, and might | |
263 | * be of interest, even at time of free, when debugging. | |
264 | * So share rcu_head with the less interesting memsw. | |
265 | */ | |
266 | struct rcu_head rcu_freeing; | |
267 | /* | |
3afe36b1 GC |
268 | * We also need some space for a worker in deferred freeing. |
269 | * By the time we call it, rcu_freeing is no longer in use. | |
59927fb9 HD |
270 | */ |
271 | struct work_struct work_freeing; | |
272 | }; | |
273 | ||
510fc4e1 GC |
274 | /* |
275 | * the counter to account for kernel memory usage. | |
276 | */ | |
277 | struct res_counter kmem; | |
78fb7466 PE |
278 | /* |
279 | * Per cgroup active and inactive list, similar to the | |
280 | * per zone LRU lists. | |
78fb7466 | 281 | */ |
6d12e2d8 | 282 | struct mem_cgroup_lru_info info; |
889976db YH |
283 | int last_scanned_node; |
284 | #if MAX_NUMNODES > 1 | |
285 | nodemask_t scan_nodes; | |
453a9bf3 KH |
286 | atomic_t numainfo_events; |
287 | atomic_t numainfo_updating; | |
889976db | 288 | #endif |
18f59ea7 BS |
289 | /* |
290 | * Should the accounting and control be hierarchical, per subtree? | |
291 | */ | |
292 | bool use_hierarchy; | |
510fc4e1 | 293 | unsigned long kmem_account_flags; /* See KMEM_ACCOUNTED_*, below */ |
79dfdacc MH |
294 | |
295 | bool oom_lock; | |
296 | atomic_t under_oom; | |
297 | ||
8c7c6e34 | 298 | atomic_t refcnt; |
14797e23 | 299 | |
1f4c025b | 300 | int swappiness; |
3c11ecf4 KH |
301 | /* OOM-Killer disable */ |
302 | int oom_kill_disable; | |
a7885eb8 | 303 | |
22a668d7 KH |
304 | /* set when res.limit == memsw.limit */ |
305 | bool memsw_is_minimum; | |
306 | ||
2e72b634 KS |
307 | /* protect arrays of thresholds */ |
308 | struct mutex thresholds_lock; | |
309 | ||
310 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 311 | struct mem_cgroup_thresholds thresholds; |
907860ed | 312 | |
2e72b634 | 313 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 314 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 315 | |
9490ff27 KH |
316 | /* For oom notifier event fd */ |
317 | struct list_head oom_notify; | |
185efc0f | 318 | |
7dc74be0 DN |
319 | /* |
320 | * Should we move charges of a task when a task is moved into this | |
321 | * mem_cgroup ? And what type of charges should we move ? | |
322 | */ | |
323 | unsigned long move_charge_at_immigrate; | |
619d094b KH |
324 | /* |
325 | * set > 0 if pages under this cgroup are moving to other cgroup. | |
326 | */ | |
327 | atomic_t moving_account; | |
312734c0 KH |
328 | /* taken only while moving_account > 0 */ |
329 | spinlock_t move_lock; | |
d52aa412 | 330 | /* |
c62b1a3b | 331 | * percpu counter. |
d52aa412 | 332 | */ |
3a7951b4 | 333 | struct mem_cgroup_stat_cpu __percpu *stat; |
711d3d2c KH |
334 | /* |
335 | * used when a cpu is offlined or other synchronizations | |
336 | * See mem_cgroup_read_stat(). | |
337 | */ | |
338 | struct mem_cgroup_stat_cpu nocpu_base; | |
339 | spinlock_t pcp_counter_lock; | |
d1a4c0b3 | 340 | |
4bd2c1ee | 341 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET) |
d1a4c0b3 GC |
342 | struct tcp_memcontrol tcp_mem; |
343 | #endif | |
2633d7a0 GC |
344 | #if defined(CONFIG_MEMCG_KMEM) |
345 | /* analogous to slab_common's slab_caches list. per-memcg */ | |
346 | struct list_head memcg_slab_caches; | |
347 | /* Not a spinlock, we can take a lot of time walking the list */ | |
348 | struct mutex slab_caches_mutex; | |
349 | /* Index in the kmem_cache->memcg_params->memcg_caches array */ | |
350 | int kmemcg_id; | |
351 | #endif | |
8cdea7c0 BS |
352 | }; |
353 | ||
510fc4e1 GC |
354 | /* internal only representation about the status of kmem accounting. */ |
355 | enum { | |
356 | KMEM_ACCOUNTED_ACTIVE = 0, /* accounted by this cgroup itself */ | |
a8964b9b | 357 | KMEM_ACCOUNTED_ACTIVATED, /* static key enabled. */ |
7de37682 | 358 | KMEM_ACCOUNTED_DEAD, /* dead memcg with pending kmem charges */ |
510fc4e1 GC |
359 | }; |
360 | ||
a8964b9b GC |
361 | /* We account when limit is on, but only after call sites are patched */ |
362 | #define KMEM_ACCOUNTED_MASK \ | |
363 | ((1 << KMEM_ACCOUNTED_ACTIVE) | (1 << KMEM_ACCOUNTED_ACTIVATED)) | |
510fc4e1 GC |
364 | |
365 | #ifdef CONFIG_MEMCG_KMEM | |
366 | static inline void memcg_kmem_set_active(struct mem_cgroup *memcg) | |
367 | { | |
368 | set_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
369 | } | |
7de37682 GC |
370 | |
371 | static bool memcg_kmem_is_active(struct mem_cgroup *memcg) | |
372 | { | |
373 | return test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
374 | } | |
375 | ||
a8964b9b GC |
376 | static void memcg_kmem_set_activated(struct mem_cgroup *memcg) |
377 | { | |
378 | set_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags); | |
379 | } | |
380 | ||
7de37682 GC |
381 | static void memcg_kmem_mark_dead(struct mem_cgroup *memcg) |
382 | { | |
383 | if (test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags)) | |
384 | set_bit(KMEM_ACCOUNTED_DEAD, &memcg->kmem_account_flags); | |
385 | } | |
386 | ||
387 | static bool memcg_kmem_test_and_clear_dead(struct mem_cgroup *memcg) | |
388 | { | |
389 | return test_and_clear_bit(KMEM_ACCOUNTED_DEAD, | |
390 | &memcg->kmem_account_flags); | |
391 | } | |
510fc4e1 GC |
392 | #endif |
393 | ||
7dc74be0 DN |
394 | /* Stuffs for move charges at task migration. */ |
395 | /* | |
396 | * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a | |
397 | * left-shifted bitmap of these types. | |
398 | */ | |
399 | enum move_type { | |
4ffef5fe | 400 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 401 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
402 | NR_MOVE_TYPE, |
403 | }; | |
404 | ||
4ffef5fe DN |
405 | /* "mc" and its members are protected by cgroup_mutex */ |
406 | static struct move_charge_struct { | |
b1dd693e | 407 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
408 | struct mem_cgroup *from; |
409 | struct mem_cgroup *to; | |
410 | unsigned long precharge; | |
854ffa8d | 411 | unsigned long moved_charge; |
483c30b5 | 412 | unsigned long moved_swap; |
8033b97c DN |
413 | struct task_struct *moving_task; /* a task moving charges */ |
414 | wait_queue_head_t waitq; /* a waitq for other context */ | |
415 | } mc = { | |
2bd9bb20 | 416 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
417 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
418 | }; | |
4ffef5fe | 419 | |
90254a65 DN |
420 | static bool move_anon(void) |
421 | { | |
422 | return test_bit(MOVE_CHARGE_TYPE_ANON, | |
423 | &mc.to->move_charge_at_immigrate); | |
424 | } | |
425 | ||
87946a72 DN |
426 | static bool move_file(void) |
427 | { | |
428 | return test_bit(MOVE_CHARGE_TYPE_FILE, | |
429 | &mc.to->move_charge_at_immigrate); | |
430 | } | |
431 | ||
4e416953 BS |
432 | /* |
433 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
434 | * limit reclaim to prevent infinite loops, if they ever occur. | |
435 | */ | |
a0db00fc KS |
436 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
437 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 | |
4e416953 | 438 | |
217bc319 KH |
439 | enum charge_type { |
440 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 441 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 442 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 443 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
444 | NR_CHARGE_TYPE, |
445 | }; | |
446 | ||
8c7c6e34 | 447 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
448 | enum res_type { |
449 | _MEM, | |
450 | _MEMSWAP, | |
451 | _OOM_TYPE, | |
510fc4e1 | 452 | _KMEM, |
86ae53e1 GC |
453 | }; |
454 | ||
a0db00fc KS |
455 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
456 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 457 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
458 | /* Used for OOM nofiier */ |
459 | #define OOM_CONTROL (0) | |
8c7c6e34 | 460 | |
75822b44 BS |
461 | /* |
462 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
463 | */ | |
464 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
465 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
466 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
467 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
468 | ||
c0ff4b85 R |
469 | static void mem_cgroup_get(struct mem_cgroup *memcg); |
470 | static void mem_cgroup_put(struct mem_cgroup *memcg); | |
e1aab161 | 471 | |
b2145145 WL |
472 | static inline |
473 | struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s) | |
474 | { | |
475 | return container_of(s, struct mem_cgroup, css); | |
476 | } | |
477 | ||
7ffc0edc MH |
478 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
479 | { | |
480 | return (memcg == root_mem_cgroup); | |
481 | } | |
482 | ||
e1aab161 | 483 | /* Writing them here to avoid exposing memcg's inner layout */ |
4bd2c1ee | 484 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
e1aab161 | 485 | |
e1aab161 GC |
486 | void sock_update_memcg(struct sock *sk) |
487 | { | |
376be5ff | 488 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 489 | struct mem_cgroup *memcg; |
3f134619 | 490 | struct cg_proto *cg_proto; |
e1aab161 GC |
491 | |
492 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
493 | ||
f3f511e1 GC |
494 | /* Socket cloning can throw us here with sk_cgrp already |
495 | * filled. It won't however, necessarily happen from | |
496 | * process context. So the test for root memcg given | |
497 | * the current task's memcg won't help us in this case. | |
498 | * | |
499 | * Respecting the original socket's memcg is a better | |
500 | * decision in this case. | |
501 | */ | |
502 | if (sk->sk_cgrp) { | |
503 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
504 | mem_cgroup_get(sk->sk_cgrp->memcg); | |
505 | return; | |
506 | } | |
507 | ||
e1aab161 GC |
508 | rcu_read_lock(); |
509 | memcg = mem_cgroup_from_task(current); | |
3f134619 GC |
510 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
511 | if (!mem_cgroup_is_root(memcg) && memcg_proto_active(cg_proto)) { | |
e1aab161 | 512 | mem_cgroup_get(memcg); |
3f134619 | 513 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
514 | } |
515 | rcu_read_unlock(); | |
516 | } | |
517 | } | |
518 | EXPORT_SYMBOL(sock_update_memcg); | |
519 | ||
520 | void sock_release_memcg(struct sock *sk) | |
521 | { | |
376be5ff | 522 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
523 | struct mem_cgroup *memcg; |
524 | WARN_ON(!sk->sk_cgrp->memcg); | |
525 | memcg = sk->sk_cgrp->memcg; | |
526 | mem_cgroup_put(memcg); | |
527 | } | |
528 | } | |
d1a4c0b3 GC |
529 | |
530 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
531 | { | |
532 | if (!memcg || mem_cgroup_is_root(memcg)) | |
533 | return NULL; | |
534 | ||
535 | return &memcg->tcp_mem.cg_proto; | |
536 | } | |
537 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 538 | |
3f134619 GC |
539 | static void disarm_sock_keys(struct mem_cgroup *memcg) |
540 | { | |
541 | if (!memcg_proto_activated(&memcg->tcp_mem.cg_proto)) | |
542 | return; | |
543 | static_key_slow_dec(&memcg_socket_limit_enabled); | |
544 | } | |
545 | #else | |
546 | static void disarm_sock_keys(struct mem_cgroup *memcg) | |
547 | { | |
548 | } | |
549 | #endif | |
550 | ||
a8964b9b GC |
551 | #ifdef CONFIG_MEMCG_KMEM |
552 | struct static_key memcg_kmem_enabled_key; | |
553 | ||
554 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
555 | { | |
556 | if (memcg_kmem_is_active(memcg)) | |
557 | static_key_slow_dec(&memcg_kmem_enabled_key); | |
bea207c8 GC |
558 | /* |
559 | * This check can't live in kmem destruction function, | |
560 | * since the charges will outlive the cgroup | |
561 | */ | |
562 | WARN_ON(res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0); | |
a8964b9b GC |
563 | } |
564 | #else | |
565 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
566 | { | |
567 | } | |
568 | #endif /* CONFIG_MEMCG_KMEM */ | |
569 | ||
570 | static void disarm_static_keys(struct mem_cgroup *memcg) | |
571 | { | |
572 | disarm_sock_keys(memcg); | |
573 | disarm_kmem_keys(memcg); | |
574 | } | |
575 | ||
c0ff4b85 | 576 | static void drain_all_stock_async(struct mem_cgroup *memcg); |
8c7c6e34 | 577 | |
f64c3f54 | 578 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 579 | mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid) |
f64c3f54 | 580 | { |
c0ff4b85 | 581 | return &memcg->info.nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
582 | } |
583 | ||
c0ff4b85 | 584 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 585 | { |
c0ff4b85 | 586 | return &memcg->css; |
d324236b WF |
587 | } |
588 | ||
f64c3f54 | 589 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 590 | page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 591 | { |
97a6c37b JW |
592 | int nid = page_to_nid(page); |
593 | int zid = page_zonenum(page); | |
f64c3f54 | 594 | |
c0ff4b85 | 595 | return mem_cgroup_zoneinfo(memcg, nid, zid); |
f64c3f54 BS |
596 | } |
597 | ||
598 | static struct mem_cgroup_tree_per_zone * | |
599 | soft_limit_tree_node_zone(int nid, int zid) | |
600 | { | |
601 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
602 | } | |
603 | ||
604 | static struct mem_cgroup_tree_per_zone * | |
605 | soft_limit_tree_from_page(struct page *page) | |
606 | { | |
607 | int nid = page_to_nid(page); | |
608 | int zid = page_zonenum(page); | |
609 | ||
610 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
611 | } | |
612 | ||
613 | static void | |
c0ff4b85 | 614 | __mem_cgroup_insert_exceeded(struct mem_cgroup *memcg, |
f64c3f54 | 615 | struct mem_cgroup_per_zone *mz, |
ef8745c1 KH |
616 | struct mem_cgroup_tree_per_zone *mctz, |
617 | unsigned long long new_usage_in_excess) | |
f64c3f54 BS |
618 | { |
619 | struct rb_node **p = &mctz->rb_root.rb_node; | |
620 | struct rb_node *parent = NULL; | |
621 | struct mem_cgroup_per_zone *mz_node; | |
622 | ||
623 | if (mz->on_tree) | |
624 | return; | |
625 | ||
ef8745c1 KH |
626 | mz->usage_in_excess = new_usage_in_excess; |
627 | if (!mz->usage_in_excess) | |
628 | return; | |
f64c3f54 BS |
629 | while (*p) { |
630 | parent = *p; | |
631 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
632 | tree_node); | |
633 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
634 | p = &(*p)->rb_left; | |
635 | /* | |
636 | * We can't avoid mem cgroups that are over their soft | |
637 | * limit by the same amount | |
638 | */ | |
639 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
640 | p = &(*p)->rb_right; | |
641 | } | |
642 | rb_link_node(&mz->tree_node, parent, p); | |
643 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
644 | mz->on_tree = true; | |
4e416953 BS |
645 | } |
646 | ||
647 | static void | |
c0ff4b85 | 648 | __mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
4e416953 BS |
649 | struct mem_cgroup_per_zone *mz, |
650 | struct mem_cgroup_tree_per_zone *mctz) | |
651 | { | |
652 | if (!mz->on_tree) | |
653 | return; | |
654 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
655 | mz->on_tree = false; | |
656 | } | |
657 | ||
f64c3f54 | 658 | static void |
c0ff4b85 | 659 | mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
f64c3f54 BS |
660 | struct mem_cgroup_per_zone *mz, |
661 | struct mem_cgroup_tree_per_zone *mctz) | |
662 | { | |
663 | spin_lock(&mctz->lock); | |
c0ff4b85 | 664 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
665 | spin_unlock(&mctz->lock); |
666 | } | |
667 | ||
f64c3f54 | 668 | |
c0ff4b85 | 669 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 670 | { |
ef8745c1 | 671 | unsigned long long excess; |
f64c3f54 BS |
672 | struct mem_cgroup_per_zone *mz; |
673 | struct mem_cgroup_tree_per_zone *mctz; | |
4e649152 KH |
674 | int nid = page_to_nid(page); |
675 | int zid = page_zonenum(page); | |
f64c3f54 BS |
676 | mctz = soft_limit_tree_from_page(page); |
677 | ||
678 | /* | |
4e649152 KH |
679 | * Necessary to update all ancestors when hierarchy is used. |
680 | * because their event counter is not touched. | |
f64c3f54 | 681 | */ |
c0ff4b85 R |
682 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
683 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
684 | excess = res_counter_soft_limit_excess(&memcg->res); | |
4e649152 KH |
685 | /* |
686 | * We have to update the tree if mz is on RB-tree or | |
687 | * mem is over its softlimit. | |
688 | */ | |
ef8745c1 | 689 | if (excess || mz->on_tree) { |
4e649152 KH |
690 | spin_lock(&mctz->lock); |
691 | /* if on-tree, remove it */ | |
692 | if (mz->on_tree) | |
c0ff4b85 | 693 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
4e649152 | 694 | /* |
ef8745c1 KH |
695 | * Insert again. mz->usage_in_excess will be updated. |
696 | * If excess is 0, no tree ops. | |
4e649152 | 697 | */ |
c0ff4b85 | 698 | __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess); |
4e649152 KH |
699 | spin_unlock(&mctz->lock); |
700 | } | |
f64c3f54 BS |
701 | } |
702 | } | |
703 | ||
c0ff4b85 | 704 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) |
f64c3f54 BS |
705 | { |
706 | int node, zone; | |
707 | struct mem_cgroup_per_zone *mz; | |
708 | struct mem_cgroup_tree_per_zone *mctz; | |
709 | ||
3ed28fa1 | 710 | for_each_node(node) { |
f64c3f54 | 711 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
c0ff4b85 | 712 | mz = mem_cgroup_zoneinfo(memcg, node, zone); |
f64c3f54 | 713 | mctz = soft_limit_tree_node_zone(node, zone); |
c0ff4b85 | 714 | mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
715 | } |
716 | } | |
717 | } | |
718 | ||
4e416953 BS |
719 | static struct mem_cgroup_per_zone * |
720 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
721 | { | |
722 | struct rb_node *rightmost = NULL; | |
26251eaf | 723 | struct mem_cgroup_per_zone *mz; |
4e416953 BS |
724 | |
725 | retry: | |
26251eaf | 726 | mz = NULL; |
4e416953 BS |
727 | rightmost = rb_last(&mctz->rb_root); |
728 | if (!rightmost) | |
729 | goto done; /* Nothing to reclaim from */ | |
730 | ||
731 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
732 | /* | |
733 | * Remove the node now but someone else can add it back, | |
734 | * we will to add it back at the end of reclaim to its correct | |
735 | * position in the tree. | |
736 | */ | |
d79154bb HD |
737 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); |
738 | if (!res_counter_soft_limit_excess(&mz->memcg->res) || | |
739 | !css_tryget(&mz->memcg->css)) | |
4e416953 BS |
740 | goto retry; |
741 | done: | |
742 | return mz; | |
743 | } | |
744 | ||
745 | static struct mem_cgroup_per_zone * | |
746 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
747 | { | |
748 | struct mem_cgroup_per_zone *mz; | |
749 | ||
750 | spin_lock(&mctz->lock); | |
751 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
752 | spin_unlock(&mctz->lock); | |
753 | return mz; | |
754 | } | |
755 | ||
711d3d2c KH |
756 | /* |
757 | * Implementation Note: reading percpu statistics for memcg. | |
758 | * | |
759 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
760 | * synchronization to implement "quick" read. There are trade-off between | |
761 | * reading cost and precision of value. Then, we may have a chance to implement | |
762 | * a periodic synchronizion of counter in memcg's counter. | |
763 | * | |
764 | * But this _read() function is used for user interface now. The user accounts | |
765 | * memory usage by memory cgroup and he _always_ requires exact value because | |
766 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
767 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
768 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
769 | * | |
770 | * If there are kernel internal actions which can make use of some not-exact | |
771 | * value, and reading all cpu value can be performance bottleneck in some | |
772 | * common workload, threashold and synchonization as vmstat[] should be | |
773 | * implemented. | |
774 | */ | |
c0ff4b85 | 775 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 776 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 777 | { |
7a159cc9 | 778 | long val = 0; |
c62b1a3b | 779 | int cpu; |
c62b1a3b | 780 | |
711d3d2c KH |
781 | get_online_cpus(); |
782 | for_each_online_cpu(cpu) | |
c0ff4b85 | 783 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 784 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
785 | spin_lock(&memcg->pcp_counter_lock); |
786 | val += memcg->nocpu_base.count[idx]; | |
787 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
788 | #endif |
789 | put_online_cpus(); | |
c62b1a3b KH |
790 | return val; |
791 | } | |
792 | ||
c0ff4b85 | 793 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
0c3e73e8 BS |
794 | bool charge) |
795 | { | |
796 | int val = (charge) ? 1 : -1; | |
bff6bb83 | 797 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); |
0c3e73e8 BS |
798 | } |
799 | ||
c0ff4b85 | 800 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
801 | enum mem_cgroup_events_index idx) |
802 | { | |
803 | unsigned long val = 0; | |
804 | int cpu; | |
805 | ||
806 | for_each_online_cpu(cpu) | |
c0ff4b85 | 807 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 808 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
809 | spin_lock(&memcg->pcp_counter_lock); |
810 | val += memcg->nocpu_base.events[idx]; | |
811 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f JW |
812 | #endif |
813 | return val; | |
814 | } | |
815 | ||
c0ff4b85 | 816 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b2402857 | 817 | bool anon, int nr_pages) |
d52aa412 | 818 | { |
c62b1a3b KH |
819 | preempt_disable(); |
820 | ||
b2402857 KH |
821 | /* |
822 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
823 | * counted as CACHE even if it's on ANON LRU. | |
824 | */ | |
825 | if (anon) | |
826 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], | |
c0ff4b85 | 827 | nr_pages); |
d52aa412 | 828 | else |
b2402857 | 829 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 830 | nr_pages); |
55e462b0 | 831 | |
e401f176 KH |
832 | /* pagein of a big page is an event. So, ignore page size */ |
833 | if (nr_pages > 0) | |
c0ff4b85 | 834 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 835 | else { |
c0ff4b85 | 836 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
837 | nr_pages = -nr_pages; /* for event */ |
838 | } | |
e401f176 | 839 | |
13114716 | 840 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
2e72b634 | 841 | |
c62b1a3b | 842 | preempt_enable(); |
6d12e2d8 KH |
843 | } |
844 | ||
bb2a0de9 | 845 | unsigned long |
4d7dcca2 | 846 | mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
847 | { |
848 | struct mem_cgroup_per_zone *mz; | |
849 | ||
850 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
851 | return mz->lru_size[lru]; | |
852 | } | |
853 | ||
854 | static unsigned long | |
c0ff4b85 | 855 | mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid, |
bb2a0de9 | 856 | unsigned int lru_mask) |
889976db YH |
857 | { |
858 | struct mem_cgroup_per_zone *mz; | |
f156ab93 | 859 | enum lru_list lru; |
bb2a0de9 KH |
860 | unsigned long ret = 0; |
861 | ||
c0ff4b85 | 862 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
bb2a0de9 | 863 | |
f156ab93 HD |
864 | for_each_lru(lru) { |
865 | if (BIT(lru) & lru_mask) | |
866 | ret += mz->lru_size[lru]; | |
bb2a0de9 KH |
867 | } |
868 | return ret; | |
869 | } | |
870 | ||
871 | static unsigned long | |
c0ff4b85 | 872 | mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 KH |
873 | int nid, unsigned int lru_mask) |
874 | { | |
889976db YH |
875 | u64 total = 0; |
876 | int zid; | |
877 | ||
bb2a0de9 | 878 | for (zid = 0; zid < MAX_NR_ZONES; zid++) |
c0ff4b85 R |
879 | total += mem_cgroup_zone_nr_lru_pages(memcg, |
880 | nid, zid, lru_mask); | |
bb2a0de9 | 881 | |
889976db YH |
882 | return total; |
883 | } | |
bb2a0de9 | 884 | |
c0ff4b85 | 885 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 886 | unsigned int lru_mask) |
6d12e2d8 | 887 | { |
889976db | 888 | int nid; |
6d12e2d8 KH |
889 | u64 total = 0; |
890 | ||
31aaea4a | 891 | for_each_node_state(nid, N_MEMORY) |
c0ff4b85 | 892 | total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
6d12e2d8 | 893 | return total; |
d52aa412 KH |
894 | } |
895 | ||
f53d7ce3 JW |
896 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
897 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
898 | { |
899 | unsigned long val, next; | |
900 | ||
13114716 | 901 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 902 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 903 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
904 | if ((long)next - (long)val < 0) { |
905 | switch (target) { | |
906 | case MEM_CGROUP_TARGET_THRESH: | |
907 | next = val + THRESHOLDS_EVENTS_TARGET; | |
908 | break; | |
909 | case MEM_CGROUP_TARGET_SOFTLIMIT: | |
910 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
911 | break; | |
912 | case MEM_CGROUP_TARGET_NUMAINFO: | |
913 | next = val + NUMAINFO_EVENTS_TARGET; | |
914 | break; | |
915 | default: | |
916 | break; | |
917 | } | |
918 | __this_cpu_write(memcg->stat->targets[target], next); | |
919 | return true; | |
7a159cc9 | 920 | } |
f53d7ce3 | 921 | return false; |
d2265e6f KH |
922 | } |
923 | ||
924 | /* | |
925 | * Check events in order. | |
926 | * | |
927 | */ | |
c0ff4b85 | 928 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f | 929 | { |
4799401f | 930 | preempt_disable(); |
d2265e6f | 931 | /* threshold event is triggered in finer grain than soft limit */ |
f53d7ce3 JW |
932 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
933 | MEM_CGROUP_TARGET_THRESH))) { | |
82b3f2a7 AM |
934 | bool do_softlimit; |
935 | bool do_numainfo __maybe_unused; | |
f53d7ce3 JW |
936 | |
937 | do_softlimit = mem_cgroup_event_ratelimit(memcg, | |
938 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
939 | #if MAX_NUMNODES > 1 | |
940 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
941 | MEM_CGROUP_TARGET_NUMAINFO); | |
942 | #endif | |
943 | preempt_enable(); | |
944 | ||
c0ff4b85 | 945 | mem_cgroup_threshold(memcg); |
f53d7ce3 | 946 | if (unlikely(do_softlimit)) |
c0ff4b85 | 947 | mem_cgroup_update_tree(memcg, page); |
453a9bf3 | 948 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 949 | if (unlikely(do_numainfo)) |
c0ff4b85 | 950 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 951 | #endif |
f53d7ce3 JW |
952 | } else |
953 | preempt_enable(); | |
d2265e6f KH |
954 | } |
955 | ||
d1a4c0b3 | 956 | struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 | 957 | { |
b2145145 WL |
958 | return mem_cgroup_from_css( |
959 | cgroup_subsys_state(cont, mem_cgroup_subsys_id)); | |
8cdea7c0 BS |
960 | } |
961 | ||
cf475ad2 | 962 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 963 | { |
31a78f23 BS |
964 | /* |
965 | * mm_update_next_owner() may clear mm->owner to NULL | |
966 | * if it races with swapoff, page migration, etc. | |
967 | * So this can be called with p == NULL. | |
968 | */ | |
969 | if (unlikely(!p)) | |
970 | return NULL; | |
971 | ||
b2145145 | 972 | return mem_cgroup_from_css(task_subsys_state(p, mem_cgroup_subsys_id)); |
78fb7466 PE |
973 | } |
974 | ||
a433658c | 975 | struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 976 | { |
c0ff4b85 | 977 | struct mem_cgroup *memcg = NULL; |
0b7f569e KH |
978 | |
979 | if (!mm) | |
980 | return NULL; | |
54595fe2 KH |
981 | /* |
982 | * Because we have no locks, mm->owner's may be being moved to other | |
983 | * cgroup. We use css_tryget() here even if this looks | |
984 | * pessimistic (rather than adding locks here). | |
985 | */ | |
986 | rcu_read_lock(); | |
987 | do { | |
c0ff4b85 R |
988 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
989 | if (unlikely(!memcg)) | |
54595fe2 | 990 | break; |
c0ff4b85 | 991 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 992 | rcu_read_unlock(); |
c0ff4b85 | 993 | return memcg; |
54595fe2 KH |
994 | } |
995 | ||
5660048c JW |
996 | /** |
997 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
998 | * @root: hierarchy root | |
999 | * @prev: previously returned memcg, NULL on first invocation | |
1000 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1001 | * | |
1002 | * Returns references to children of the hierarchy below @root, or | |
1003 | * @root itself, or %NULL after a full round-trip. | |
1004 | * | |
1005 | * Caller must pass the return value in @prev on subsequent | |
1006 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1007 | * to cancel a hierarchy walk before the round-trip is complete. | |
1008 | * | |
1009 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
1010 | * divide up the memcgs in the hierarchy among all concurrent | |
1011 | * reclaimers operating on the same zone and priority. | |
1012 | */ | |
1013 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, | |
1014 | struct mem_cgroup *prev, | |
1015 | struct mem_cgroup_reclaim_cookie *reclaim) | |
14067bb3 | 1016 | { |
9f3a0d09 JW |
1017 | struct mem_cgroup *memcg = NULL; |
1018 | int id = 0; | |
711d3d2c | 1019 | |
5660048c JW |
1020 | if (mem_cgroup_disabled()) |
1021 | return NULL; | |
1022 | ||
9f3a0d09 JW |
1023 | if (!root) |
1024 | root = root_mem_cgroup; | |
7d74b06f | 1025 | |
9f3a0d09 JW |
1026 | if (prev && !reclaim) |
1027 | id = css_id(&prev->css); | |
14067bb3 | 1028 | |
9f3a0d09 JW |
1029 | if (prev && prev != root) |
1030 | css_put(&prev->css); | |
14067bb3 | 1031 | |
9f3a0d09 JW |
1032 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1033 | if (prev) | |
1034 | return NULL; | |
1035 | return root; | |
1036 | } | |
14067bb3 | 1037 | |
9f3a0d09 | 1038 | while (!memcg) { |
527a5ec9 | 1039 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
9f3a0d09 | 1040 | struct cgroup_subsys_state *css; |
711d3d2c | 1041 | |
527a5ec9 JW |
1042 | if (reclaim) { |
1043 | int nid = zone_to_nid(reclaim->zone); | |
1044 | int zid = zone_idx(reclaim->zone); | |
1045 | struct mem_cgroup_per_zone *mz; | |
1046 | ||
1047 | mz = mem_cgroup_zoneinfo(root, nid, zid); | |
1048 | iter = &mz->reclaim_iter[reclaim->priority]; | |
1049 | if (prev && reclaim->generation != iter->generation) | |
1050 | return NULL; | |
1051 | id = iter->position; | |
1052 | } | |
7d74b06f | 1053 | |
9f3a0d09 JW |
1054 | rcu_read_lock(); |
1055 | css = css_get_next(&mem_cgroup_subsys, id + 1, &root->css, &id); | |
1056 | if (css) { | |
1057 | if (css == &root->css || css_tryget(css)) | |
b2145145 | 1058 | memcg = mem_cgroup_from_css(css); |
9f3a0d09 JW |
1059 | } else |
1060 | id = 0; | |
14067bb3 | 1061 | rcu_read_unlock(); |
14067bb3 | 1062 | |
527a5ec9 JW |
1063 | if (reclaim) { |
1064 | iter->position = id; | |
1065 | if (!css) | |
1066 | iter->generation++; | |
1067 | else if (!prev && memcg) | |
1068 | reclaim->generation = iter->generation; | |
1069 | } | |
9f3a0d09 JW |
1070 | |
1071 | if (prev && !css) | |
1072 | return NULL; | |
1073 | } | |
1074 | return memcg; | |
14067bb3 | 1075 | } |
7d74b06f | 1076 | |
5660048c JW |
1077 | /** |
1078 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1079 | * @root: hierarchy root | |
1080 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1081 | */ | |
1082 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1083 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1084 | { |
1085 | if (!root) | |
1086 | root = root_mem_cgroup; | |
1087 | if (prev && prev != root) | |
1088 | css_put(&prev->css); | |
1089 | } | |
7d74b06f | 1090 | |
9f3a0d09 JW |
1091 | /* |
1092 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1093 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1094 | * be used for reference counting. | |
1095 | */ | |
1096 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1097 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1098 | iter != NULL; \ |
527a5ec9 | 1099 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1100 | |
9f3a0d09 | 1101 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1102 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1103 | iter != NULL; \ |
527a5ec9 | 1104 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1105 | |
68ae564b | 1106 | void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
456f998e | 1107 | { |
c0ff4b85 | 1108 | struct mem_cgroup *memcg; |
456f998e | 1109 | |
456f998e | 1110 | rcu_read_lock(); |
c0ff4b85 R |
1111 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1112 | if (unlikely(!memcg)) | |
456f998e YH |
1113 | goto out; |
1114 | ||
1115 | switch (idx) { | |
456f998e | 1116 | case PGFAULT: |
0e574a93 JW |
1117 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1118 | break; | |
1119 | case PGMAJFAULT: | |
1120 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1121 | break; |
1122 | default: | |
1123 | BUG(); | |
1124 | } | |
1125 | out: | |
1126 | rcu_read_unlock(); | |
1127 | } | |
68ae564b | 1128 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
456f998e | 1129 | |
925b7673 JW |
1130 | /** |
1131 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1132 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1133 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1134 | * |
1135 | * Returns the lru list vector holding pages for the given @zone and | |
1136 | * @mem. This can be the global zone lruvec, if the memory controller | |
1137 | * is disabled. | |
1138 | */ | |
1139 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1140 | struct mem_cgroup *memcg) | |
1141 | { | |
1142 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1143 | struct lruvec *lruvec; |
925b7673 | 1144 | |
bea8c150 HD |
1145 | if (mem_cgroup_disabled()) { |
1146 | lruvec = &zone->lruvec; | |
1147 | goto out; | |
1148 | } | |
925b7673 JW |
1149 | |
1150 | mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone)); | |
bea8c150 HD |
1151 | lruvec = &mz->lruvec; |
1152 | out: | |
1153 | /* | |
1154 | * Since a node can be onlined after the mem_cgroup was created, | |
1155 | * we have to be prepared to initialize lruvec->zone here; | |
1156 | * and if offlined then reonlined, we need to reinitialize it. | |
1157 | */ | |
1158 | if (unlikely(lruvec->zone != zone)) | |
1159 | lruvec->zone = zone; | |
1160 | return lruvec; | |
925b7673 JW |
1161 | } |
1162 | ||
08e552c6 KH |
1163 | /* |
1164 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
1165 | * Operations are called by routine of global LRU independently from memcg. | |
1166 | * What we have to take care of here is validness of pc->mem_cgroup. | |
1167 | * | |
1168 | * Changes to pc->mem_cgroup happens when | |
1169 | * 1. charge | |
1170 | * 2. moving account | |
1171 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
1172 | * It is added to LRU before charge. | |
1173 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
1174 | * When moving account, the page is not on LRU. It's isolated. | |
1175 | */ | |
4f98a2fe | 1176 | |
925b7673 | 1177 | /** |
fa9add64 | 1178 | * mem_cgroup_page_lruvec - return lruvec for adding an lru page |
925b7673 | 1179 | * @page: the page |
fa9add64 | 1180 | * @zone: zone of the page |
925b7673 | 1181 | */ |
fa9add64 | 1182 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1183 | { |
08e552c6 | 1184 | struct mem_cgroup_per_zone *mz; |
925b7673 JW |
1185 | struct mem_cgroup *memcg; |
1186 | struct page_cgroup *pc; | |
bea8c150 | 1187 | struct lruvec *lruvec; |
6d12e2d8 | 1188 | |
bea8c150 HD |
1189 | if (mem_cgroup_disabled()) { |
1190 | lruvec = &zone->lruvec; | |
1191 | goto out; | |
1192 | } | |
925b7673 | 1193 | |
08e552c6 | 1194 | pc = lookup_page_cgroup(page); |
38c5d72f | 1195 | memcg = pc->mem_cgroup; |
7512102c HD |
1196 | |
1197 | /* | |
fa9add64 | 1198 | * Surreptitiously switch any uncharged offlist page to root: |
7512102c HD |
1199 | * an uncharged page off lru does nothing to secure |
1200 | * its former mem_cgroup from sudden removal. | |
1201 | * | |
1202 | * Our caller holds lru_lock, and PageCgroupUsed is updated | |
1203 | * under page_cgroup lock: between them, they make all uses | |
1204 | * of pc->mem_cgroup safe. | |
1205 | */ | |
fa9add64 | 1206 | if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup) |
7512102c HD |
1207 | pc->mem_cgroup = memcg = root_mem_cgroup; |
1208 | ||
925b7673 | 1209 | mz = page_cgroup_zoneinfo(memcg, page); |
bea8c150 HD |
1210 | lruvec = &mz->lruvec; |
1211 | out: | |
1212 | /* | |
1213 | * Since a node can be onlined after the mem_cgroup was created, | |
1214 | * we have to be prepared to initialize lruvec->zone here; | |
1215 | * and if offlined then reonlined, we need to reinitialize it. | |
1216 | */ | |
1217 | if (unlikely(lruvec->zone != zone)) | |
1218 | lruvec->zone = zone; | |
1219 | return lruvec; | |
08e552c6 | 1220 | } |
b69408e8 | 1221 | |
925b7673 | 1222 | /** |
fa9add64 HD |
1223 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1224 | * @lruvec: mem_cgroup per zone lru vector | |
1225 | * @lru: index of lru list the page is sitting on | |
1226 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1227 | * |
fa9add64 HD |
1228 | * This function must be called when a page is added to or removed from an |
1229 | * lru list. | |
3f58a829 | 1230 | */ |
fa9add64 HD |
1231 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1232 | int nr_pages) | |
3f58a829 MK |
1233 | { |
1234 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1235 | unsigned long *lru_size; |
3f58a829 MK |
1236 | |
1237 | if (mem_cgroup_disabled()) | |
1238 | return; | |
1239 | ||
fa9add64 HD |
1240 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1241 | lru_size = mz->lru_size + lru; | |
1242 | *lru_size += nr_pages; | |
1243 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1244 | } |
544122e5 | 1245 | |
3e92041d | 1246 | /* |
c0ff4b85 | 1247 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1248 | * hierarchy subtree |
1249 | */ | |
c3ac9a8a JW |
1250 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1251 | struct mem_cgroup *memcg) | |
3e92041d | 1252 | { |
91c63734 JW |
1253 | if (root_memcg == memcg) |
1254 | return true; | |
3a981f48 | 1255 | if (!root_memcg->use_hierarchy || !memcg) |
91c63734 | 1256 | return false; |
c3ac9a8a JW |
1257 | return css_is_ancestor(&memcg->css, &root_memcg->css); |
1258 | } | |
1259 | ||
1260 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, | |
1261 | struct mem_cgroup *memcg) | |
1262 | { | |
1263 | bool ret; | |
1264 | ||
91c63734 | 1265 | rcu_read_lock(); |
c3ac9a8a | 1266 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
91c63734 JW |
1267 | rcu_read_unlock(); |
1268 | return ret; | |
3e92041d MH |
1269 | } |
1270 | ||
c0ff4b85 | 1271 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg) |
4c4a2214 DR |
1272 | { |
1273 | int ret; | |
0b7f569e | 1274 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1275 | struct task_struct *p; |
4c4a2214 | 1276 | |
158e0a2d | 1277 | p = find_lock_task_mm(task); |
de077d22 DR |
1278 | if (p) { |
1279 | curr = try_get_mem_cgroup_from_mm(p->mm); | |
1280 | task_unlock(p); | |
1281 | } else { | |
1282 | /* | |
1283 | * All threads may have already detached their mm's, but the oom | |
1284 | * killer still needs to detect if they have already been oom | |
1285 | * killed to prevent needlessly killing additional tasks. | |
1286 | */ | |
1287 | task_lock(task); | |
1288 | curr = mem_cgroup_from_task(task); | |
1289 | if (curr) | |
1290 | css_get(&curr->css); | |
1291 | task_unlock(task); | |
1292 | } | |
0b7f569e KH |
1293 | if (!curr) |
1294 | return 0; | |
d31f56db | 1295 | /* |
c0ff4b85 | 1296 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1297 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1298 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1299 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1300 | */ |
c0ff4b85 | 1301 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1302 | css_put(&curr->css); |
4c4a2214 DR |
1303 | return ret; |
1304 | } | |
1305 | ||
c56d5c7d | 1306 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1307 | { |
9b272977 | 1308 | unsigned long inactive_ratio; |
14797e23 | 1309 | unsigned long inactive; |
9b272977 | 1310 | unsigned long active; |
c772be93 | 1311 | unsigned long gb; |
14797e23 | 1312 | |
4d7dcca2 HD |
1313 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1314 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1315 | |
c772be93 KM |
1316 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1317 | if (gb) | |
1318 | inactive_ratio = int_sqrt(10 * gb); | |
1319 | else | |
1320 | inactive_ratio = 1; | |
1321 | ||
9b272977 | 1322 | return inactive * inactive_ratio < active; |
14797e23 KM |
1323 | } |
1324 | ||
c56d5c7d | 1325 | int mem_cgroup_inactive_file_is_low(struct lruvec *lruvec) |
56e49d21 RR |
1326 | { |
1327 | unsigned long active; | |
1328 | unsigned long inactive; | |
1329 | ||
4d7dcca2 HD |
1330 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_FILE); |
1331 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_FILE); | |
56e49d21 RR |
1332 | |
1333 | return (active > inactive); | |
1334 | } | |
1335 | ||
6d61ef40 BS |
1336 | #define mem_cgroup_from_res_counter(counter, member) \ |
1337 | container_of(counter, struct mem_cgroup, member) | |
1338 | ||
19942822 | 1339 | /** |
9d11ea9f | 1340 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1341 | * @memcg: the memory cgroup |
19942822 | 1342 | * |
9d11ea9f | 1343 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1344 | * pages. |
19942822 | 1345 | */ |
c0ff4b85 | 1346 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1347 | { |
9d11ea9f JW |
1348 | unsigned long long margin; |
1349 | ||
c0ff4b85 | 1350 | margin = res_counter_margin(&memcg->res); |
9d11ea9f | 1351 | if (do_swap_account) |
c0ff4b85 | 1352 | margin = min(margin, res_counter_margin(&memcg->memsw)); |
7ec99d62 | 1353 | return margin >> PAGE_SHIFT; |
19942822 JW |
1354 | } |
1355 | ||
1f4c025b | 1356 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 KM |
1357 | { |
1358 | struct cgroup *cgrp = memcg->css.cgroup; | |
a7885eb8 KM |
1359 | |
1360 | /* root ? */ | |
1361 | if (cgrp->parent == NULL) | |
1362 | return vm_swappiness; | |
1363 | ||
bf1ff263 | 1364 | return memcg->swappiness; |
a7885eb8 KM |
1365 | } |
1366 | ||
619d094b KH |
1367 | /* |
1368 | * memcg->moving_account is used for checking possibility that some thread is | |
1369 | * calling move_account(). When a thread on CPU-A starts moving pages under | |
1370 | * a memcg, other threads should check memcg->moving_account under | |
1371 | * rcu_read_lock(), like this: | |
1372 | * | |
1373 | * CPU-A CPU-B | |
1374 | * rcu_read_lock() | |
1375 | * memcg->moving_account+1 if (memcg->mocing_account) | |
1376 | * take heavy locks. | |
1377 | * synchronize_rcu() update something. | |
1378 | * rcu_read_unlock() | |
1379 | * start move here. | |
1380 | */ | |
4331f7d3 KH |
1381 | |
1382 | /* for quick checking without looking up memcg */ | |
1383 | atomic_t memcg_moving __read_mostly; | |
1384 | ||
c0ff4b85 | 1385 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a | 1386 | { |
4331f7d3 | 1387 | atomic_inc(&memcg_moving); |
619d094b | 1388 | atomic_inc(&memcg->moving_account); |
32047e2a KH |
1389 | synchronize_rcu(); |
1390 | } | |
1391 | ||
c0ff4b85 | 1392 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a | 1393 | { |
619d094b KH |
1394 | /* |
1395 | * Now, mem_cgroup_clear_mc() may call this function with NULL. | |
1396 | * We check NULL in callee rather than caller. | |
1397 | */ | |
4331f7d3 KH |
1398 | if (memcg) { |
1399 | atomic_dec(&memcg_moving); | |
619d094b | 1400 | atomic_dec(&memcg->moving_account); |
4331f7d3 | 1401 | } |
32047e2a | 1402 | } |
619d094b | 1403 | |
32047e2a KH |
1404 | /* |
1405 | * 2 routines for checking "mem" is under move_account() or not. | |
1406 | * | |
13fd1dd9 AM |
1407 | * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This |
1408 | * is used for avoiding races in accounting. If true, | |
32047e2a KH |
1409 | * pc->mem_cgroup may be overwritten. |
1410 | * | |
1411 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or | |
1412 | * under hierarchy of moving cgroups. This is for | |
1413 | * waiting at hith-memory prressure caused by "move". | |
1414 | */ | |
1415 | ||
13fd1dd9 | 1416 | static bool mem_cgroup_stolen(struct mem_cgroup *memcg) |
32047e2a KH |
1417 | { |
1418 | VM_BUG_ON(!rcu_read_lock_held()); | |
619d094b | 1419 | return atomic_read(&memcg->moving_account) > 0; |
32047e2a | 1420 | } |
4b534334 | 1421 | |
c0ff4b85 | 1422 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1423 | { |
2bd9bb20 KH |
1424 | struct mem_cgroup *from; |
1425 | struct mem_cgroup *to; | |
4b534334 | 1426 | bool ret = false; |
2bd9bb20 KH |
1427 | /* |
1428 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1429 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1430 | */ | |
1431 | spin_lock(&mc.lock); | |
1432 | from = mc.from; | |
1433 | to = mc.to; | |
1434 | if (!from) | |
1435 | goto unlock; | |
3e92041d | 1436 | |
c0ff4b85 R |
1437 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1438 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1439 | unlock: |
1440 | spin_unlock(&mc.lock); | |
4b534334 KH |
1441 | return ret; |
1442 | } | |
1443 | ||
c0ff4b85 | 1444 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1445 | { |
1446 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1447 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1448 | DEFINE_WAIT(wait); |
1449 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1450 | /* moving charge context might have finished. */ | |
1451 | if (mc.moving_task) | |
1452 | schedule(); | |
1453 | finish_wait(&mc.waitq, &wait); | |
1454 | return true; | |
1455 | } | |
1456 | } | |
1457 | return false; | |
1458 | } | |
1459 | ||
312734c0 KH |
1460 | /* |
1461 | * Take this lock when | |
1462 | * - a code tries to modify page's memcg while it's USED. | |
1463 | * - a code tries to modify page state accounting in a memcg. | |
13fd1dd9 | 1464 | * see mem_cgroup_stolen(), too. |
312734c0 KH |
1465 | */ |
1466 | static void move_lock_mem_cgroup(struct mem_cgroup *memcg, | |
1467 | unsigned long *flags) | |
1468 | { | |
1469 | spin_lock_irqsave(&memcg->move_lock, *flags); | |
1470 | } | |
1471 | ||
1472 | static void move_unlock_mem_cgroup(struct mem_cgroup *memcg, | |
1473 | unsigned long *flags) | |
1474 | { | |
1475 | spin_unlock_irqrestore(&memcg->move_lock, *flags); | |
1476 | } | |
1477 | ||
e222432b | 1478 | /** |
6a6135b6 | 1479 | * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode. |
e222432b BS |
1480 | * @memcg: The memory cgroup that went over limit |
1481 | * @p: Task that is going to be killed | |
1482 | * | |
1483 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1484 | * enabled | |
1485 | */ | |
1486 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1487 | { | |
1488 | struct cgroup *task_cgrp; | |
1489 | struct cgroup *mem_cgrp; | |
1490 | /* | |
1491 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1492 | * on the assumption that OOM is serialized for memory controller. | |
1493 | * If this assumption is broken, revisit this code. | |
1494 | */ | |
1495 | static char memcg_name[PATH_MAX]; | |
1496 | int ret; | |
1497 | ||
d31f56db | 1498 | if (!memcg || !p) |
e222432b BS |
1499 | return; |
1500 | ||
e222432b BS |
1501 | rcu_read_lock(); |
1502 | ||
1503 | mem_cgrp = memcg->css.cgroup; | |
1504 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1505 | ||
1506 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1507 | if (ret < 0) { | |
1508 | /* | |
1509 | * Unfortunately, we are unable to convert to a useful name | |
1510 | * But we'll still print out the usage information | |
1511 | */ | |
1512 | rcu_read_unlock(); | |
1513 | goto done; | |
1514 | } | |
1515 | rcu_read_unlock(); | |
1516 | ||
1517 | printk(KERN_INFO "Task in %s killed", memcg_name); | |
1518 | ||
1519 | rcu_read_lock(); | |
1520 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1521 | if (ret < 0) { | |
1522 | rcu_read_unlock(); | |
1523 | goto done; | |
1524 | } | |
1525 | rcu_read_unlock(); | |
1526 | ||
1527 | /* | |
1528 | * Continues from above, so we don't need an KERN_ level | |
1529 | */ | |
1530 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); | |
1531 | done: | |
1532 | ||
1533 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", | |
1534 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, | |
1535 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1536 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
1537 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " | |
1538 | "failcnt %llu\n", | |
1539 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, | |
1540 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1541 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
510fc4e1 GC |
1542 | printk(KERN_INFO "kmem: usage %llukB, limit %llukB, failcnt %llu\n", |
1543 | res_counter_read_u64(&memcg->kmem, RES_USAGE) >> 10, | |
1544 | res_counter_read_u64(&memcg->kmem, RES_LIMIT) >> 10, | |
1545 | res_counter_read_u64(&memcg->kmem, RES_FAILCNT)); | |
e222432b BS |
1546 | } |
1547 | ||
81d39c20 KH |
1548 | /* |
1549 | * This function returns the number of memcg under hierarchy tree. Returns | |
1550 | * 1(self count) if no children. | |
1551 | */ | |
c0ff4b85 | 1552 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1553 | { |
1554 | int num = 0; | |
7d74b06f KH |
1555 | struct mem_cgroup *iter; |
1556 | ||
c0ff4b85 | 1557 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1558 | num++; |
81d39c20 KH |
1559 | return num; |
1560 | } | |
1561 | ||
a63d83f4 DR |
1562 | /* |
1563 | * Return the memory (and swap, if configured) limit for a memcg. | |
1564 | */ | |
9cbb78bb | 1565 | static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 DR |
1566 | { |
1567 | u64 limit; | |
a63d83f4 | 1568 | |
f3e8eb70 | 1569 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
f3e8eb70 | 1570 | |
a63d83f4 | 1571 | /* |
9a5a8f19 | 1572 | * Do not consider swap space if we cannot swap due to swappiness |
a63d83f4 | 1573 | */ |
9a5a8f19 MH |
1574 | if (mem_cgroup_swappiness(memcg)) { |
1575 | u64 memsw; | |
1576 | ||
1577 | limit += total_swap_pages << PAGE_SHIFT; | |
1578 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1579 | ||
1580 | /* | |
1581 | * If memsw is finite and limits the amount of swap space | |
1582 | * available to this memcg, return that limit. | |
1583 | */ | |
1584 | limit = min(limit, memsw); | |
1585 | } | |
1586 | ||
1587 | return limit; | |
a63d83f4 DR |
1588 | } |
1589 | ||
19965460 DR |
1590 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1591 | int order) | |
9cbb78bb DR |
1592 | { |
1593 | struct mem_cgroup *iter; | |
1594 | unsigned long chosen_points = 0; | |
1595 | unsigned long totalpages; | |
1596 | unsigned int points = 0; | |
1597 | struct task_struct *chosen = NULL; | |
1598 | ||
876aafbf DR |
1599 | /* |
1600 | * If current has a pending SIGKILL, then automatically select it. The | |
1601 | * goal is to allow it to allocate so that it may quickly exit and free | |
1602 | * its memory. | |
1603 | */ | |
1604 | if (fatal_signal_pending(current)) { | |
1605 | set_thread_flag(TIF_MEMDIE); | |
1606 | return; | |
1607 | } | |
1608 | ||
1609 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
9cbb78bb DR |
1610 | totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1; |
1611 | for_each_mem_cgroup_tree(iter, memcg) { | |
1612 | struct cgroup *cgroup = iter->css.cgroup; | |
1613 | struct cgroup_iter it; | |
1614 | struct task_struct *task; | |
1615 | ||
1616 | cgroup_iter_start(cgroup, &it); | |
1617 | while ((task = cgroup_iter_next(cgroup, &it))) { | |
1618 | switch (oom_scan_process_thread(task, totalpages, NULL, | |
1619 | false)) { | |
1620 | case OOM_SCAN_SELECT: | |
1621 | if (chosen) | |
1622 | put_task_struct(chosen); | |
1623 | chosen = task; | |
1624 | chosen_points = ULONG_MAX; | |
1625 | get_task_struct(chosen); | |
1626 | /* fall through */ | |
1627 | case OOM_SCAN_CONTINUE: | |
1628 | continue; | |
1629 | case OOM_SCAN_ABORT: | |
1630 | cgroup_iter_end(cgroup, &it); | |
1631 | mem_cgroup_iter_break(memcg, iter); | |
1632 | if (chosen) | |
1633 | put_task_struct(chosen); | |
1634 | return; | |
1635 | case OOM_SCAN_OK: | |
1636 | break; | |
1637 | }; | |
1638 | points = oom_badness(task, memcg, NULL, totalpages); | |
1639 | if (points > chosen_points) { | |
1640 | if (chosen) | |
1641 | put_task_struct(chosen); | |
1642 | chosen = task; | |
1643 | chosen_points = points; | |
1644 | get_task_struct(chosen); | |
1645 | } | |
1646 | } | |
1647 | cgroup_iter_end(cgroup, &it); | |
1648 | } | |
1649 | ||
1650 | if (!chosen) | |
1651 | return; | |
1652 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1653 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1654 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1655 | } |
1656 | ||
5660048c JW |
1657 | static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg, |
1658 | gfp_t gfp_mask, | |
1659 | unsigned long flags) | |
1660 | { | |
1661 | unsigned long total = 0; | |
1662 | bool noswap = false; | |
1663 | int loop; | |
1664 | ||
1665 | if (flags & MEM_CGROUP_RECLAIM_NOSWAP) | |
1666 | noswap = true; | |
1667 | if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum) | |
1668 | noswap = true; | |
1669 | ||
1670 | for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) { | |
1671 | if (loop) | |
1672 | drain_all_stock_async(memcg); | |
1673 | total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap); | |
1674 | /* | |
1675 | * Allow limit shrinkers, which are triggered directly | |
1676 | * by userspace, to catch signals and stop reclaim | |
1677 | * after minimal progress, regardless of the margin. | |
1678 | */ | |
1679 | if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK)) | |
1680 | break; | |
1681 | if (mem_cgroup_margin(memcg)) | |
1682 | break; | |
1683 | /* | |
1684 | * If nothing was reclaimed after two attempts, there | |
1685 | * may be no reclaimable pages in this hierarchy. | |
1686 | */ | |
1687 | if (loop && !total) | |
1688 | break; | |
1689 | } | |
1690 | return total; | |
1691 | } | |
1692 | ||
4d0c066d KH |
1693 | /** |
1694 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1695 | * @memcg: the target memcg |
4d0c066d KH |
1696 | * @nid: the node ID to be checked. |
1697 | * @noswap : specify true here if the user wants flle only information. | |
1698 | * | |
1699 | * This function returns whether the specified memcg contains any | |
1700 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1701 | * pages in the node. | |
1702 | */ | |
c0ff4b85 | 1703 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1704 | int nid, bool noswap) |
1705 | { | |
c0ff4b85 | 1706 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1707 | return true; |
1708 | if (noswap || !total_swap_pages) | |
1709 | return false; | |
c0ff4b85 | 1710 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1711 | return true; |
1712 | return false; | |
1713 | ||
1714 | } | |
889976db YH |
1715 | #if MAX_NUMNODES > 1 |
1716 | ||
1717 | /* | |
1718 | * Always updating the nodemask is not very good - even if we have an empty | |
1719 | * list or the wrong list here, we can start from some node and traverse all | |
1720 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1721 | * | |
1722 | */ | |
c0ff4b85 | 1723 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1724 | { |
1725 | int nid; | |
453a9bf3 KH |
1726 | /* |
1727 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1728 | * pagein/pageout changes since the last update. | |
1729 | */ | |
c0ff4b85 | 1730 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1731 | return; |
c0ff4b85 | 1732 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1733 | return; |
1734 | ||
889976db | 1735 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1736 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1737 | |
31aaea4a | 1738 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1739 | |
c0ff4b85 R |
1740 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1741 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1742 | } |
453a9bf3 | 1743 | |
c0ff4b85 R |
1744 | atomic_set(&memcg->numainfo_events, 0); |
1745 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1746 | } |
1747 | ||
1748 | /* | |
1749 | * Selecting a node where we start reclaim from. Because what we need is just | |
1750 | * reducing usage counter, start from anywhere is O,K. Considering | |
1751 | * memory reclaim from current node, there are pros. and cons. | |
1752 | * | |
1753 | * Freeing memory from current node means freeing memory from a node which | |
1754 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1755 | * hit limits, it will see a contention on a node. But freeing from remote | |
1756 | * node means more costs for memory reclaim because of memory latency. | |
1757 | * | |
1758 | * Now, we use round-robin. Better algorithm is welcomed. | |
1759 | */ | |
c0ff4b85 | 1760 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1761 | { |
1762 | int node; | |
1763 | ||
c0ff4b85 R |
1764 | mem_cgroup_may_update_nodemask(memcg); |
1765 | node = memcg->last_scanned_node; | |
889976db | 1766 | |
c0ff4b85 | 1767 | node = next_node(node, memcg->scan_nodes); |
889976db | 1768 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1769 | node = first_node(memcg->scan_nodes); |
889976db YH |
1770 | /* |
1771 | * We call this when we hit limit, not when pages are added to LRU. | |
1772 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1773 | * memcg is too small and all pages are not on LRU. In that case, | |
1774 | * we use curret node. | |
1775 | */ | |
1776 | if (unlikely(node == MAX_NUMNODES)) | |
1777 | node = numa_node_id(); | |
1778 | ||
c0ff4b85 | 1779 | memcg->last_scanned_node = node; |
889976db YH |
1780 | return node; |
1781 | } | |
1782 | ||
4d0c066d KH |
1783 | /* |
1784 | * Check all nodes whether it contains reclaimable pages or not. | |
1785 | * For quick scan, we make use of scan_nodes. This will allow us to skip | |
1786 | * unused nodes. But scan_nodes is lazily updated and may not cotain | |
1787 | * enough new information. We need to do double check. | |
1788 | */ | |
6bbda35c | 1789 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d KH |
1790 | { |
1791 | int nid; | |
1792 | ||
1793 | /* | |
1794 | * quick check...making use of scan_node. | |
1795 | * We can skip unused nodes. | |
1796 | */ | |
c0ff4b85 R |
1797 | if (!nodes_empty(memcg->scan_nodes)) { |
1798 | for (nid = first_node(memcg->scan_nodes); | |
4d0c066d | 1799 | nid < MAX_NUMNODES; |
c0ff4b85 | 1800 | nid = next_node(nid, memcg->scan_nodes)) { |
4d0c066d | 1801 | |
c0ff4b85 | 1802 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
1803 | return true; |
1804 | } | |
1805 | } | |
1806 | /* | |
1807 | * Check rest of nodes. | |
1808 | */ | |
31aaea4a | 1809 | for_each_node_state(nid, N_MEMORY) { |
c0ff4b85 | 1810 | if (node_isset(nid, memcg->scan_nodes)) |
4d0c066d | 1811 | continue; |
c0ff4b85 | 1812 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
1813 | return true; |
1814 | } | |
1815 | return false; | |
1816 | } | |
1817 | ||
889976db | 1818 | #else |
c0ff4b85 | 1819 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1820 | { |
1821 | return 0; | |
1822 | } | |
4d0c066d | 1823 | |
6bbda35c | 1824 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d | 1825 | { |
c0ff4b85 | 1826 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); |
4d0c066d | 1827 | } |
889976db YH |
1828 | #endif |
1829 | ||
5660048c JW |
1830 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1831 | struct zone *zone, | |
1832 | gfp_t gfp_mask, | |
1833 | unsigned long *total_scanned) | |
6d61ef40 | 1834 | { |
9f3a0d09 | 1835 | struct mem_cgroup *victim = NULL; |
5660048c | 1836 | int total = 0; |
04046e1a | 1837 | int loop = 0; |
9d11ea9f | 1838 | unsigned long excess; |
185efc0f | 1839 | unsigned long nr_scanned; |
527a5ec9 JW |
1840 | struct mem_cgroup_reclaim_cookie reclaim = { |
1841 | .zone = zone, | |
1842 | .priority = 0, | |
1843 | }; | |
9d11ea9f | 1844 | |
c0ff4b85 | 1845 | excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT; |
04046e1a | 1846 | |
4e416953 | 1847 | while (1) { |
527a5ec9 | 1848 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); |
9f3a0d09 | 1849 | if (!victim) { |
04046e1a | 1850 | loop++; |
4e416953 BS |
1851 | if (loop >= 2) { |
1852 | /* | |
1853 | * If we have not been able to reclaim | |
1854 | * anything, it might because there are | |
1855 | * no reclaimable pages under this hierarchy | |
1856 | */ | |
5660048c | 1857 | if (!total) |
4e416953 | 1858 | break; |
4e416953 | 1859 | /* |
25985edc | 1860 | * We want to do more targeted reclaim. |
4e416953 BS |
1861 | * excess >> 2 is not to excessive so as to |
1862 | * reclaim too much, nor too less that we keep | |
1863 | * coming back to reclaim from this cgroup | |
1864 | */ | |
1865 | if (total >= (excess >> 2) || | |
9f3a0d09 | 1866 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) |
4e416953 | 1867 | break; |
4e416953 | 1868 | } |
9f3a0d09 | 1869 | continue; |
4e416953 | 1870 | } |
5660048c | 1871 | if (!mem_cgroup_reclaimable(victim, false)) |
6d61ef40 | 1872 | continue; |
5660048c JW |
1873 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1874 | zone, &nr_scanned); | |
1875 | *total_scanned += nr_scanned; | |
1876 | if (!res_counter_soft_limit_excess(&root_memcg->res)) | |
9f3a0d09 | 1877 | break; |
6d61ef40 | 1878 | } |
9f3a0d09 | 1879 | mem_cgroup_iter_break(root_memcg, victim); |
04046e1a | 1880 | return total; |
6d61ef40 BS |
1881 | } |
1882 | ||
867578cb KH |
1883 | /* |
1884 | * Check OOM-Killer is already running under our hierarchy. | |
1885 | * If someone is running, return false. | |
1af8efe9 | 1886 | * Has to be called with memcg_oom_lock |
867578cb | 1887 | */ |
c0ff4b85 | 1888 | static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg) |
867578cb | 1889 | { |
79dfdacc | 1890 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1891 | |
9f3a0d09 | 1892 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1893 | if (iter->oom_lock) { |
79dfdacc MH |
1894 | /* |
1895 | * this subtree of our hierarchy is already locked | |
1896 | * so we cannot give a lock. | |
1897 | */ | |
79dfdacc | 1898 | failed = iter; |
9f3a0d09 JW |
1899 | mem_cgroup_iter_break(memcg, iter); |
1900 | break; | |
23751be0 JW |
1901 | } else |
1902 | iter->oom_lock = true; | |
7d74b06f | 1903 | } |
867578cb | 1904 | |
79dfdacc | 1905 | if (!failed) |
23751be0 | 1906 | return true; |
79dfdacc MH |
1907 | |
1908 | /* | |
1909 | * OK, we failed to lock the whole subtree so we have to clean up | |
1910 | * what we set up to the failing subtree | |
1911 | */ | |
9f3a0d09 | 1912 | for_each_mem_cgroup_tree(iter, memcg) { |
79dfdacc | 1913 | if (iter == failed) { |
9f3a0d09 JW |
1914 | mem_cgroup_iter_break(memcg, iter); |
1915 | break; | |
79dfdacc MH |
1916 | } |
1917 | iter->oom_lock = false; | |
1918 | } | |
23751be0 | 1919 | return false; |
a636b327 | 1920 | } |
0b7f569e | 1921 | |
79dfdacc | 1922 | /* |
1af8efe9 | 1923 | * Has to be called with memcg_oom_lock |
79dfdacc | 1924 | */ |
c0ff4b85 | 1925 | static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1926 | { |
7d74b06f KH |
1927 | struct mem_cgroup *iter; |
1928 | ||
c0ff4b85 | 1929 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1930 | iter->oom_lock = false; |
1931 | return 0; | |
1932 | } | |
1933 | ||
c0ff4b85 | 1934 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1935 | { |
1936 | struct mem_cgroup *iter; | |
1937 | ||
c0ff4b85 | 1938 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1939 | atomic_inc(&iter->under_oom); |
1940 | } | |
1941 | ||
c0ff4b85 | 1942 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1943 | { |
1944 | struct mem_cgroup *iter; | |
1945 | ||
867578cb KH |
1946 | /* |
1947 | * When a new child is created while the hierarchy is under oom, | |
1948 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1949 | * atomic_add_unless() here. | |
1950 | */ | |
c0ff4b85 | 1951 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1952 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
1953 | } |
1954 | ||
1af8efe9 | 1955 | static DEFINE_SPINLOCK(memcg_oom_lock); |
867578cb KH |
1956 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1957 | ||
dc98df5a | 1958 | struct oom_wait_info { |
d79154bb | 1959 | struct mem_cgroup *memcg; |
dc98df5a KH |
1960 | wait_queue_t wait; |
1961 | }; | |
1962 | ||
1963 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1964 | unsigned mode, int sync, void *arg) | |
1965 | { | |
d79154bb HD |
1966 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1967 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1968 | struct oom_wait_info *oom_wait_info; |
1969 | ||
1970 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1971 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1972 | |
dc98df5a | 1973 | /* |
d79154bb | 1974 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
dc98df5a KH |
1975 | * Then we can use css_is_ancestor without taking care of RCU. |
1976 | */ | |
c0ff4b85 R |
1977 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
1978 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 1979 | return 0; |
dc98df5a KH |
1980 | return autoremove_wake_function(wait, mode, sync, arg); |
1981 | } | |
1982 | ||
c0ff4b85 | 1983 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 1984 | { |
c0ff4b85 R |
1985 | /* for filtering, pass "memcg" as argument. */ |
1986 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
1987 | } |
1988 | ||
c0ff4b85 | 1989 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1990 | { |
c0ff4b85 R |
1991 | if (memcg && atomic_read(&memcg->under_oom)) |
1992 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
1993 | } |
1994 | ||
867578cb KH |
1995 | /* |
1996 | * try to call OOM killer. returns false if we should exit memory-reclaim loop. | |
1997 | */ | |
6bbda35c KS |
1998 | static bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask, |
1999 | int order) | |
0b7f569e | 2000 | { |
dc98df5a | 2001 | struct oom_wait_info owait; |
3c11ecf4 | 2002 | bool locked, need_to_kill; |
867578cb | 2003 | |
d79154bb | 2004 | owait.memcg = memcg; |
dc98df5a KH |
2005 | owait.wait.flags = 0; |
2006 | owait.wait.func = memcg_oom_wake_function; | |
2007 | owait.wait.private = current; | |
2008 | INIT_LIST_HEAD(&owait.wait.task_list); | |
3c11ecf4 | 2009 | need_to_kill = true; |
c0ff4b85 | 2010 | mem_cgroup_mark_under_oom(memcg); |
79dfdacc | 2011 | |
c0ff4b85 | 2012 | /* At first, try to OOM lock hierarchy under memcg.*/ |
1af8efe9 | 2013 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 2014 | locked = mem_cgroup_oom_lock(memcg); |
867578cb KH |
2015 | /* |
2016 | * Even if signal_pending(), we can't quit charge() loop without | |
2017 | * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL | |
2018 | * under OOM is always welcomed, use TASK_KILLABLE here. | |
2019 | */ | |
3c11ecf4 | 2020 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
c0ff4b85 | 2021 | if (!locked || memcg->oom_kill_disable) |
3c11ecf4 KH |
2022 | need_to_kill = false; |
2023 | if (locked) | |
c0ff4b85 | 2024 | mem_cgroup_oom_notify(memcg); |
1af8efe9 | 2025 | spin_unlock(&memcg_oom_lock); |
867578cb | 2026 | |
3c11ecf4 KH |
2027 | if (need_to_kill) { |
2028 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
e845e199 | 2029 | mem_cgroup_out_of_memory(memcg, mask, order); |
3c11ecf4 | 2030 | } else { |
867578cb | 2031 | schedule(); |
dc98df5a | 2032 | finish_wait(&memcg_oom_waitq, &owait.wait); |
867578cb | 2033 | } |
1af8efe9 | 2034 | spin_lock(&memcg_oom_lock); |
79dfdacc | 2035 | if (locked) |
c0ff4b85 R |
2036 | mem_cgroup_oom_unlock(memcg); |
2037 | memcg_wakeup_oom(memcg); | |
1af8efe9 | 2038 | spin_unlock(&memcg_oom_lock); |
867578cb | 2039 | |
c0ff4b85 | 2040 | mem_cgroup_unmark_under_oom(memcg); |
79dfdacc | 2041 | |
867578cb KH |
2042 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) |
2043 | return false; | |
2044 | /* Give chance to dying process */ | |
715a5ee8 | 2045 | schedule_timeout_uninterruptible(1); |
867578cb | 2046 | return true; |
0b7f569e KH |
2047 | } |
2048 | ||
d69b042f BS |
2049 | /* |
2050 | * Currently used to update mapped file statistics, but the routine can be | |
2051 | * generalized to update other statistics as well. | |
32047e2a KH |
2052 | * |
2053 | * Notes: Race condition | |
2054 | * | |
2055 | * We usually use page_cgroup_lock() for accessing page_cgroup member but | |
2056 | * it tends to be costly. But considering some conditions, we doesn't need | |
2057 | * to do so _always_. | |
2058 | * | |
2059 | * Considering "charge", lock_page_cgroup() is not required because all | |
2060 | * file-stat operations happen after a page is attached to radix-tree. There | |
2061 | * are no race with "charge". | |
2062 | * | |
2063 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
2064 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
2065 | * if there are race with "uncharge". Statistics itself is properly handled | |
2066 | * by flags. | |
2067 | * | |
2068 | * Considering "move", this is an only case we see a race. To make the race | |
619d094b KH |
2069 | * small, we check mm->moving_account and detect there are possibility of race |
2070 | * If there is, we take a lock. | |
d69b042f | 2071 | */ |
26174efd | 2072 | |
89c06bd5 KH |
2073 | void __mem_cgroup_begin_update_page_stat(struct page *page, |
2074 | bool *locked, unsigned long *flags) | |
2075 | { | |
2076 | struct mem_cgroup *memcg; | |
2077 | struct page_cgroup *pc; | |
2078 | ||
2079 | pc = lookup_page_cgroup(page); | |
2080 | again: | |
2081 | memcg = pc->mem_cgroup; | |
2082 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
2083 | return; | |
2084 | /* | |
2085 | * If this memory cgroup is not under account moving, we don't | |
da92c47d | 2086 | * need to take move_lock_mem_cgroup(). Because we already hold |
89c06bd5 | 2087 | * rcu_read_lock(), any calls to move_account will be delayed until |
13fd1dd9 | 2088 | * rcu_read_unlock() if mem_cgroup_stolen() == true. |
89c06bd5 | 2089 | */ |
13fd1dd9 | 2090 | if (!mem_cgroup_stolen(memcg)) |
89c06bd5 KH |
2091 | return; |
2092 | ||
2093 | move_lock_mem_cgroup(memcg, flags); | |
2094 | if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) { | |
2095 | move_unlock_mem_cgroup(memcg, flags); | |
2096 | goto again; | |
2097 | } | |
2098 | *locked = true; | |
2099 | } | |
2100 | ||
2101 | void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags) | |
2102 | { | |
2103 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
2104 | ||
2105 | /* | |
2106 | * It's guaranteed that pc->mem_cgroup never changes while | |
2107 | * lock is held because a routine modifies pc->mem_cgroup | |
da92c47d | 2108 | * should take move_lock_mem_cgroup(). |
89c06bd5 KH |
2109 | */ |
2110 | move_unlock_mem_cgroup(pc->mem_cgroup, flags); | |
2111 | } | |
2112 | ||
2a7106f2 GT |
2113 | void mem_cgroup_update_page_stat(struct page *page, |
2114 | enum mem_cgroup_page_stat_item idx, int val) | |
d69b042f | 2115 | { |
c0ff4b85 | 2116 | struct mem_cgroup *memcg; |
32047e2a | 2117 | struct page_cgroup *pc = lookup_page_cgroup(page); |
dbd4ea78 | 2118 | unsigned long uninitialized_var(flags); |
d69b042f | 2119 | |
cfa44946 | 2120 | if (mem_cgroup_disabled()) |
d69b042f | 2121 | return; |
89c06bd5 | 2122 | |
c0ff4b85 R |
2123 | memcg = pc->mem_cgroup; |
2124 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
89c06bd5 | 2125 | return; |
26174efd | 2126 | |
26174efd | 2127 | switch (idx) { |
2a7106f2 | 2128 | case MEMCG_NR_FILE_MAPPED: |
2a7106f2 | 2129 | idx = MEM_CGROUP_STAT_FILE_MAPPED; |
26174efd KH |
2130 | break; |
2131 | default: | |
2132 | BUG(); | |
8725d541 | 2133 | } |
d69b042f | 2134 | |
c0ff4b85 | 2135 | this_cpu_add(memcg->stat->count[idx], val); |
d69b042f | 2136 | } |
26174efd | 2137 | |
cdec2e42 KH |
2138 | /* |
2139 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2140 | * TODO: maybe necessary to use big numbers in big irons. | |
2141 | */ | |
7ec99d62 | 2142 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2143 | struct memcg_stock_pcp { |
2144 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2145 | unsigned int nr_pages; |
cdec2e42 | 2146 | struct work_struct work; |
26fe6168 | 2147 | unsigned long flags; |
a0db00fc | 2148 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2149 | }; |
2150 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2151 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2152 | |
a0956d54 SS |
2153 | /** |
2154 | * consume_stock: Try to consume stocked charge on this cpu. | |
2155 | * @memcg: memcg to consume from. | |
2156 | * @nr_pages: how many pages to charge. | |
2157 | * | |
2158 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2159 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2160 | * service an allocation will refill the stock. | |
2161 | * | |
2162 | * returns true if successful, false otherwise. | |
cdec2e42 | 2163 | */ |
a0956d54 | 2164 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2165 | { |
2166 | struct memcg_stock_pcp *stock; | |
2167 | bool ret = true; | |
2168 | ||
a0956d54 SS |
2169 | if (nr_pages > CHARGE_BATCH) |
2170 | return false; | |
2171 | ||
cdec2e42 | 2172 | stock = &get_cpu_var(memcg_stock); |
a0956d54 SS |
2173 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) |
2174 | stock->nr_pages -= nr_pages; | |
cdec2e42 KH |
2175 | else /* need to call res_counter_charge */ |
2176 | ret = false; | |
2177 | put_cpu_var(memcg_stock); | |
2178 | return ret; | |
2179 | } | |
2180 | ||
2181 | /* | |
2182 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
2183 | */ | |
2184 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2185 | { | |
2186 | struct mem_cgroup *old = stock->cached; | |
2187 | ||
11c9ea4e JW |
2188 | if (stock->nr_pages) { |
2189 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
2190 | ||
2191 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 2192 | if (do_swap_account) |
11c9ea4e JW |
2193 | res_counter_uncharge(&old->memsw, bytes); |
2194 | stock->nr_pages = 0; | |
cdec2e42 KH |
2195 | } |
2196 | stock->cached = NULL; | |
cdec2e42 KH |
2197 | } |
2198 | ||
2199 | /* | |
2200 | * This must be called under preempt disabled or must be called by | |
2201 | * a thread which is pinned to local cpu. | |
2202 | */ | |
2203 | static void drain_local_stock(struct work_struct *dummy) | |
2204 | { | |
2205 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
2206 | drain_stock(stock); | |
26fe6168 | 2207 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2208 | } |
2209 | ||
2210 | /* | |
2211 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 2212 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2213 | */ |
c0ff4b85 | 2214 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2215 | { |
2216 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2217 | ||
c0ff4b85 | 2218 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2219 | drain_stock(stock); |
c0ff4b85 | 2220 | stock->cached = memcg; |
cdec2e42 | 2221 | } |
11c9ea4e | 2222 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2223 | put_cpu_var(memcg_stock); |
2224 | } | |
2225 | ||
2226 | /* | |
c0ff4b85 | 2227 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
d38144b7 MH |
2228 | * of the hierarchy under it. sync flag says whether we should block |
2229 | * until the work is done. | |
cdec2e42 | 2230 | */ |
c0ff4b85 | 2231 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
cdec2e42 | 2232 | { |
26fe6168 | 2233 | int cpu, curcpu; |
d38144b7 | 2234 | |
cdec2e42 | 2235 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2236 | get_online_cpus(); |
5af12d0e | 2237 | curcpu = get_cpu(); |
cdec2e42 KH |
2238 | for_each_online_cpu(cpu) { |
2239 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2240 | struct mem_cgroup *memcg; |
26fe6168 | 2241 | |
c0ff4b85 R |
2242 | memcg = stock->cached; |
2243 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2244 | continue; |
c0ff4b85 | 2245 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2246 | continue; |
d1a05b69 MH |
2247 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2248 | if (cpu == curcpu) | |
2249 | drain_local_stock(&stock->work); | |
2250 | else | |
2251 | schedule_work_on(cpu, &stock->work); | |
2252 | } | |
cdec2e42 | 2253 | } |
5af12d0e | 2254 | put_cpu(); |
d38144b7 MH |
2255 | |
2256 | if (!sync) | |
2257 | goto out; | |
2258 | ||
2259 | for_each_online_cpu(cpu) { | |
2260 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2261 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2262 | flush_work(&stock->work); |
2263 | } | |
2264 | out: | |
cdec2e42 | 2265 | put_online_cpus(); |
d38144b7 MH |
2266 | } |
2267 | ||
2268 | /* | |
2269 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2270 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
2271 | * expects some charges will be back to res_counter later but cannot wait for | |
2272 | * it. | |
2273 | */ | |
c0ff4b85 | 2274 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
d38144b7 | 2275 | { |
9f50fad6 MH |
2276 | /* |
2277 | * If someone calls draining, avoid adding more kworker runs. | |
2278 | */ | |
2279 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2280 | return; | |
c0ff4b85 | 2281 | drain_all_stock(root_memcg, false); |
9f50fad6 | 2282 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2283 | } |
2284 | ||
2285 | /* This is a synchronous drain interface. */ | |
c0ff4b85 | 2286 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
cdec2e42 KH |
2287 | { |
2288 | /* called when force_empty is called */ | |
9f50fad6 | 2289 | mutex_lock(&percpu_charge_mutex); |
c0ff4b85 | 2290 | drain_all_stock(root_memcg, true); |
9f50fad6 | 2291 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2292 | } |
2293 | ||
711d3d2c KH |
2294 | /* |
2295 | * This function drains percpu counter value from DEAD cpu and | |
2296 | * move it to local cpu. Note that this function can be preempted. | |
2297 | */ | |
c0ff4b85 | 2298 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2299 | { |
2300 | int i; | |
2301 | ||
c0ff4b85 | 2302 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2303 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2304 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2305 | |
c0ff4b85 R |
2306 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2307 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2308 | } |
e9f8974f | 2309 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2310 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2311 | |
c0ff4b85 R |
2312 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2313 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2314 | } |
c0ff4b85 | 2315 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2316 | } |
2317 | ||
2318 | static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb, | |
cdec2e42 KH |
2319 | unsigned long action, |
2320 | void *hcpu) | |
2321 | { | |
2322 | int cpu = (unsigned long)hcpu; | |
2323 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2324 | struct mem_cgroup *iter; |
cdec2e42 | 2325 | |
619d094b | 2326 | if (action == CPU_ONLINE) |
1489ebad | 2327 | return NOTIFY_OK; |
1489ebad | 2328 | |
d833049b | 2329 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2330 | return NOTIFY_OK; |
711d3d2c | 2331 | |
9f3a0d09 | 2332 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2333 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2334 | ||
cdec2e42 KH |
2335 | stock = &per_cpu(memcg_stock, cpu); |
2336 | drain_stock(stock); | |
2337 | return NOTIFY_OK; | |
2338 | } | |
2339 | ||
4b534334 KH |
2340 | |
2341 | /* See __mem_cgroup_try_charge() for details */ | |
2342 | enum { | |
2343 | CHARGE_OK, /* success */ | |
2344 | CHARGE_RETRY, /* need to retry but retry is not bad */ | |
2345 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ | |
2346 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ | |
2347 | CHARGE_OOM_DIE, /* the current is killed because of OOM */ | |
2348 | }; | |
2349 | ||
c0ff4b85 | 2350 | static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
4c9c5359 SS |
2351 | unsigned int nr_pages, unsigned int min_pages, |
2352 | bool oom_check) | |
4b534334 | 2353 | { |
7ec99d62 | 2354 | unsigned long csize = nr_pages * PAGE_SIZE; |
4b534334 KH |
2355 | struct mem_cgroup *mem_over_limit; |
2356 | struct res_counter *fail_res; | |
2357 | unsigned long flags = 0; | |
2358 | int ret; | |
2359 | ||
c0ff4b85 | 2360 | ret = res_counter_charge(&memcg->res, csize, &fail_res); |
4b534334 KH |
2361 | |
2362 | if (likely(!ret)) { | |
2363 | if (!do_swap_account) | |
2364 | return CHARGE_OK; | |
c0ff4b85 | 2365 | ret = res_counter_charge(&memcg->memsw, csize, &fail_res); |
4b534334 KH |
2366 | if (likely(!ret)) |
2367 | return CHARGE_OK; | |
2368 | ||
c0ff4b85 | 2369 | res_counter_uncharge(&memcg->res, csize); |
4b534334 KH |
2370 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
2371 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
2372 | } else | |
2373 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
9221edb7 | 2374 | /* |
9221edb7 JW |
2375 | * Never reclaim on behalf of optional batching, retry with a |
2376 | * single page instead. | |
2377 | */ | |
4c9c5359 | 2378 | if (nr_pages > min_pages) |
4b534334 KH |
2379 | return CHARGE_RETRY; |
2380 | ||
2381 | if (!(gfp_mask & __GFP_WAIT)) | |
2382 | return CHARGE_WOULDBLOCK; | |
2383 | ||
4c9c5359 SS |
2384 | if (gfp_mask & __GFP_NORETRY) |
2385 | return CHARGE_NOMEM; | |
2386 | ||
5660048c | 2387 | ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags); |
7ec99d62 | 2388 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
19942822 | 2389 | return CHARGE_RETRY; |
4b534334 | 2390 | /* |
19942822 JW |
2391 | * Even though the limit is exceeded at this point, reclaim |
2392 | * may have been able to free some pages. Retry the charge | |
2393 | * before killing the task. | |
2394 | * | |
2395 | * Only for regular pages, though: huge pages are rather | |
2396 | * unlikely to succeed so close to the limit, and we fall back | |
2397 | * to regular pages anyway in case of failure. | |
4b534334 | 2398 | */ |
4c9c5359 | 2399 | if (nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER) && ret) |
4b534334 KH |
2400 | return CHARGE_RETRY; |
2401 | ||
2402 | /* | |
2403 | * At task move, charge accounts can be doubly counted. So, it's | |
2404 | * better to wait until the end of task_move if something is going on. | |
2405 | */ | |
2406 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2407 | return CHARGE_RETRY; | |
2408 | ||
2409 | /* If we don't need to call oom-killer at el, return immediately */ | |
2410 | if (!oom_check) | |
2411 | return CHARGE_NOMEM; | |
2412 | /* check OOM */ | |
e845e199 | 2413 | if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask, get_order(csize))) |
4b534334 KH |
2414 | return CHARGE_OOM_DIE; |
2415 | ||
2416 | return CHARGE_RETRY; | |
2417 | } | |
2418 | ||
f817ed48 | 2419 | /* |
38c5d72f KH |
2420 | * __mem_cgroup_try_charge() does |
2421 | * 1. detect memcg to be charged against from passed *mm and *ptr, | |
2422 | * 2. update res_counter | |
2423 | * 3. call memory reclaim if necessary. | |
2424 | * | |
2425 | * In some special case, if the task is fatal, fatal_signal_pending() or | |
2426 | * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup | |
2427 | * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon | |
2428 | * as possible without any hazards. 2: all pages should have a valid | |
2429 | * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg | |
2430 | * pointer, that is treated as a charge to root_mem_cgroup. | |
2431 | * | |
2432 | * So __mem_cgroup_try_charge() will return | |
2433 | * 0 ... on success, filling *ptr with a valid memcg pointer. | |
2434 | * -ENOMEM ... charge failure because of resource limits. | |
2435 | * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup. | |
2436 | * | |
2437 | * Unlike the exported interface, an "oom" parameter is added. if oom==true, | |
2438 | * the oom-killer can be invoked. | |
8a9f3ccd | 2439 | */ |
f817ed48 | 2440 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
ec168510 | 2441 | gfp_t gfp_mask, |
7ec99d62 | 2442 | unsigned int nr_pages, |
c0ff4b85 | 2443 | struct mem_cgroup **ptr, |
7ec99d62 | 2444 | bool oom) |
8a9f3ccd | 2445 | { |
7ec99d62 | 2446 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
4b534334 | 2447 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
c0ff4b85 | 2448 | struct mem_cgroup *memcg = NULL; |
4b534334 | 2449 | int ret; |
a636b327 | 2450 | |
867578cb KH |
2451 | /* |
2452 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
2453 | * in system level. So, allow to go ahead dying process in addition to | |
2454 | * MEMDIE process. | |
2455 | */ | |
2456 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
2457 | || fatal_signal_pending(current))) | |
2458 | goto bypass; | |
a636b327 | 2459 | |
8a9f3ccd | 2460 | /* |
3be91277 HD |
2461 | * We always charge the cgroup the mm_struct belongs to. |
2462 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd | 2463 | * thread group leader migrates. It's possible that mm is not |
24467cac | 2464 | * set, if so charge the root memcg (happens for pagecache usage). |
8a9f3ccd | 2465 | */ |
c0ff4b85 | 2466 | if (!*ptr && !mm) |
38c5d72f | 2467 | *ptr = root_mem_cgroup; |
f75ca962 | 2468 | again: |
c0ff4b85 R |
2469 | if (*ptr) { /* css should be a valid one */ |
2470 | memcg = *ptr; | |
c0ff4b85 | 2471 | if (mem_cgroup_is_root(memcg)) |
f75ca962 | 2472 | goto done; |
a0956d54 | 2473 | if (consume_stock(memcg, nr_pages)) |
f75ca962 | 2474 | goto done; |
c0ff4b85 | 2475 | css_get(&memcg->css); |
4b534334 | 2476 | } else { |
f75ca962 | 2477 | struct task_struct *p; |
54595fe2 | 2478 | |
f75ca962 KH |
2479 | rcu_read_lock(); |
2480 | p = rcu_dereference(mm->owner); | |
f75ca962 | 2481 | /* |
ebb76ce1 | 2482 | * Because we don't have task_lock(), "p" can exit. |
c0ff4b85 | 2483 | * In that case, "memcg" can point to root or p can be NULL with |
ebb76ce1 KH |
2484 | * race with swapoff. Then, we have small risk of mis-accouning. |
2485 | * But such kind of mis-account by race always happens because | |
2486 | * we don't have cgroup_mutex(). It's overkill and we allo that | |
2487 | * small race, here. | |
2488 | * (*) swapoff at el will charge against mm-struct not against | |
2489 | * task-struct. So, mm->owner can be NULL. | |
f75ca962 | 2490 | */ |
c0ff4b85 | 2491 | memcg = mem_cgroup_from_task(p); |
38c5d72f KH |
2492 | if (!memcg) |
2493 | memcg = root_mem_cgroup; | |
2494 | if (mem_cgroup_is_root(memcg)) { | |
f75ca962 KH |
2495 | rcu_read_unlock(); |
2496 | goto done; | |
2497 | } | |
a0956d54 | 2498 | if (consume_stock(memcg, nr_pages)) { |
f75ca962 KH |
2499 | /* |
2500 | * It seems dagerous to access memcg without css_get(). | |
2501 | * But considering how consume_stok works, it's not | |
2502 | * necessary. If consume_stock success, some charges | |
2503 | * from this memcg are cached on this cpu. So, we | |
2504 | * don't need to call css_get()/css_tryget() before | |
2505 | * calling consume_stock(). | |
2506 | */ | |
2507 | rcu_read_unlock(); | |
2508 | goto done; | |
2509 | } | |
2510 | /* after here, we may be blocked. we need to get refcnt */ | |
c0ff4b85 | 2511 | if (!css_tryget(&memcg->css)) { |
f75ca962 KH |
2512 | rcu_read_unlock(); |
2513 | goto again; | |
2514 | } | |
2515 | rcu_read_unlock(); | |
2516 | } | |
8a9f3ccd | 2517 | |
4b534334 KH |
2518 | do { |
2519 | bool oom_check; | |
7a81b88c | 2520 | |
4b534334 | 2521 | /* If killed, bypass charge */ |
f75ca962 | 2522 | if (fatal_signal_pending(current)) { |
c0ff4b85 | 2523 | css_put(&memcg->css); |
4b534334 | 2524 | goto bypass; |
f75ca962 | 2525 | } |
6d61ef40 | 2526 | |
4b534334 KH |
2527 | oom_check = false; |
2528 | if (oom && !nr_oom_retries) { | |
2529 | oom_check = true; | |
2530 | nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
cdec2e42 | 2531 | } |
66e1707b | 2532 | |
4c9c5359 SS |
2533 | ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, nr_pages, |
2534 | oom_check); | |
4b534334 KH |
2535 | switch (ret) { |
2536 | case CHARGE_OK: | |
2537 | break; | |
2538 | case CHARGE_RETRY: /* not in OOM situation but retry */ | |
7ec99d62 | 2539 | batch = nr_pages; |
c0ff4b85 R |
2540 | css_put(&memcg->css); |
2541 | memcg = NULL; | |
f75ca962 | 2542 | goto again; |
4b534334 | 2543 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
c0ff4b85 | 2544 | css_put(&memcg->css); |
4b534334 KH |
2545 | goto nomem; |
2546 | case CHARGE_NOMEM: /* OOM routine works */ | |
f75ca962 | 2547 | if (!oom) { |
c0ff4b85 | 2548 | css_put(&memcg->css); |
867578cb | 2549 | goto nomem; |
f75ca962 | 2550 | } |
4b534334 KH |
2551 | /* If oom, we never return -ENOMEM */ |
2552 | nr_oom_retries--; | |
2553 | break; | |
2554 | case CHARGE_OOM_DIE: /* Killed by OOM Killer */ | |
c0ff4b85 | 2555 | css_put(&memcg->css); |
867578cb | 2556 | goto bypass; |
66e1707b | 2557 | } |
4b534334 KH |
2558 | } while (ret != CHARGE_OK); |
2559 | ||
7ec99d62 | 2560 | if (batch > nr_pages) |
c0ff4b85 R |
2561 | refill_stock(memcg, batch - nr_pages); |
2562 | css_put(&memcg->css); | |
0c3e73e8 | 2563 | done: |
c0ff4b85 | 2564 | *ptr = memcg; |
7a81b88c KH |
2565 | return 0; |
2566 | nomem: | |
c0ff4b85 | 2567 | *ptr = NULL; |
7a81b88c | 2568 | return -ENOMEM; |
867578cb | 2569 | bypass: |
38c5d72f KH |
2570 | *ptr = root_mem_cgroup; |
2571 | return -EINTR; | |
7a81b88c | 2572 | } |
8a9f3ccd | 2573 | |
a3032a2c DN |
2574 | /* |
2575 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2576 | * This function is for that and do uncharge, put css's refcnt. | |
2577 | * gotten by try_charge(). | |
2578 | */ | |
c0ff4b85 | 2579 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg, |
e7018b8d | 2580 | unsigned int nr_pages) |
a3032a2c | 2581 | { |
c0ff4b85 | 2582 | if (!mem_cgroup_is_root(memcg)) { |
e7018b8d JW |
2583 | unsigned long bytes = nr_pages * PAGE_SIZE; |
2584 | ||
c0ff4b85 | 2585 | res_counter_uncharge(&memcg->res, bytes); |
a3032a2c | 2586 | if (do_swap_account) |
c0ff4b85 | 2587 | res_counter_uncharge(&memcg->memsw, bytes); |
a3032a2c | 2588 | } |
854ffa8d DN |
2589 | } |
2590 | ||
d01dd17f KH |
2591 | /* |
2592 | * Cancel chrages in this cgroup....doesn't propagate to parent cgroup. | |
2593 | * This is useful when moving usage to parent cgroup. | |
2594 | */ | |
2595 | static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg, | |
2596 | unsigned int nr_pages) | |
2597 | { | |
2598 | unsigned long bytes = nr_pages * PAGE_SIZE; | |
2599 | ||
2600 | if (mem_cgroup_is_root(memcg)) | |
2601 | return; | |
2602 | ||
2603 | res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes); | |
2604 | if (do_swap_account) | |
2605 | res_counter_uncharge_until(&memcg->memsw, | |
2606 | memcg->memsw.parent, bytes); | |
2607 | } | |
2608 | ||
a3b2d692 KH |
2609 | /* |
2610 | * A helper function to get mem_cgroup from ID. must be called under | |
e9316080 TH |
2611 | * rcu_read_lock(). The caller is responsible for calling css_tryget if |
2612 | * the mem_cgroup is used for charging. (dropping refcnt from swap can be | |
2613 | * called against removed memcg.) | |
a3b2d692 KH |
2614 | */ |
2615 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2616 | { | |
2617 | struct cgroup_subsys_state *css; | |
2618 | ||
2619 | /* ID 0 is unused ID */ | |
2620 | if (!id) | |
2621 | return NULL; | |
2622 | css = css_lookup(&mem_cgroup_subsys, id); | |
2623 | if (!css) | |
2624 | return NULL; | |
b2145145 | 2625 | return mem_cgroup_from_css(css); |
a3b2d692 KH |
2626 | } |
2627 | ||
e42d9d5d | 2628 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2629 | { |
c0ff4b85 | 2630 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2631 | struct page_cgroup *pc; |
a3b2d692 | 2632 | unsigned short id; |
b5a84319 KH |
2633 | swp_entry_t ent; |
2634 | ||
3c776e64 DN |
2635 | VM_BUG_ON(!PageLocked(page)); |
2636 | ||
3c776e64 | 2637 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2638 | lock_page_cgroup(pc); |
a3b2d692 | 2639 | if (PageCgroupUsed(pc)) { |
c0ff4b85 R |
2640 | memcg = pc->mem_cgroup; |
2641 | if (memcg && !css_tryget(&memcg->css)) | |
2642 | memcg = NULL; | |
e42d9d5d | 2643 | } else if (PageSwapCache(page)) { |
3c776e64 | 2644 | ent.val = page_private(page); |
9fb4b7cc | 2645 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2646 | rcu_read_lock(); |
c0ff4b85 R |
2647 | memcg = mem_cgroup_lookup(id); |
2648 | if (memcg && !css_tryget(&memcg->css)) | |
2649 | memcg = NULL; | |
a3b2d692 | 2650 | rcu_read_unlock(); |
3c776e64 | 2651 | } |
c0bd3f63 | 2652 | unlock_page_cgroup(pc); |
c0ff4b85 | 2653 | return memcg; |
b5a84319 KH |
2654 | } |
2655 | ||
c0ff4b85 | 2656 | static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg, |
5564e88b | 2657 | struct page *page, |
7ec99d62 | 2658 | unsigned int nr_pages, |
9ce70c02 HD |
2659 | enum charge_type ctype, |
2660 | bool lrucare) | |
7a81b88c | 2661 | { |
ce587e65 | 2662 | struct page_cgroup *pc = lookup_page_cgroup(page); |
9ce70c02 | 2663 | struct zone *uninitialized_var(zone); |
fa9add64 | 2664 | struct lruvec *lruvec; |
9ce70c02 | 2665 | bool was_on_lru = false; |
b2402857 | 2666 | bool anon; |
9ce70c02 | 2667 | |
ca3e0214 | 2668 | lock_page_cgroup(pc); |
90deb788 | 2669 | VM_BUG_ON(PageCgroupUsed(pc)); |
ca3e0214 KH |
2670 | /* |
2671 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2672 | * accessed by any other context at this point. | |
2673 | */ | |
9ce70c02 HD |
2674 | |
2675 | /* | |
2676 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2677 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2678 | */ | |
2679 | if (lrucare) { | |
2680 | zone = page_zone(page); | |
2681 | spin_lock_irq(&zone->lru_lock); | |
2682 | if (PageLRU(page)) { | |
fa9add64 | 2683 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 | 2684 | ClearPageLRU(page); |
fa9add64 | 2685 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2686 | was_on_lru = true; |
2687 | } | |
2688 | } | |
2689 | ||
c0ff4b85 | 2690 | pc->mem_cgroup = memcg; |
261fb61a KH |
2691 | /* |
2692 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
2693 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
2694 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
2695 | * before USED bit, we need memory barrier here. | |
2696 | * See mem_cgroup_add_lru_list(), etc. | |
2697 | */ | |
08e552c6 | 2698 | smp_wmb(); |
b2402857 | 2699 | SetPageCgroupUsed(pc); |
3be91277 | 2700 | |
9ce70c02 HD |
2701 | if (lrucare) { |
2702 | if (was_on_lru) { | |
fa9add64 | 2703 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 HD |
2704 | VM_BUG_ON(PageLRU(page)); |
2705 | SetPageLRU(page); | |
fa9add64 | 2706 | add_page_to_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2707 | } |
2708 | spin_unlock_irq(&zone->lru_lock); | |
2709 | } | |
2710 | ||
41326c17 | 2711 | if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON) |
b2402857 KH |
2712 | anon = true; |
2713 | else | |
2714 | anon = false; | |
2715 | ||
2716 | mem_cgroup_charge_statistics(memcg, anon, nr_pages); | |
52d4b9ac | 2717 | unlock_page_cgroup(pc); |
9ce70c02 | 2718 | |
430e4863 KH |
2719 | /* |
2720 | * "charge_statistics" updated event counter. Then, check it. | |
2721 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
2722 | * if they exceeds softlimit. | |
2723 | */ | |
c0ff4b85 | 2724 | memcg_check_events(memcg, page); |
7a81b88c | 2725 | } |
66e1707b | 2726 | |
7ae1e1d0 GC |
2727 | #ifdef CONFIG_MEMCG_KMEM |
2728 | static inline bool memcg_can_account_kmem(struct mem_cgroup *memcg) | |
2729 | { | |
2730 | return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg) && | |
2731 | (memcg->kmem_account_flags & KMEM_ACCOUNTED_MASK); | |
2732 | } | |
2733 | ||
2734 | static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, u64 size) | |
2735 | { | |
2736 | struct res_counter *fail_res; | |
2737 | struct mem_cgroup *_memcg; | |
2738 | int ret = 0; | |
2739 | bool may_oom; | |
2740 | ||
2741 | ret = res_counter_charge(&memcg->kmem, size, &fail_res); | |
2742 | if (ret) | |
2743 | return ret; | |
2744 | ||
2745 | /* | |
2746 | * Conditions under which we can wait for the oom_killer. Those are | |
2747 | * the same conditions tested by the core page allocator | |
2748 | */ | |
2749 | may_oom = (gfp & __GFP_FS) && !(gfp & __GFP_NORETRY); | |
2750 | ||
2751 | _memcg = memcg; | |
2752 | ret = __mem_cgroup_try_charge(NULL, gfp, size >> PAGE_SHIFT, | |
2753 | &_memcg, may_oom); | |
2754 | ||
2755 | if (ret == -EINTR) { | |
2756 | /* | |
2757 | * __mem_cgroup_try_charge() chosed to bypass to root due to | |
2758 | * OOM kill or fatal signal. Since our only options are to | |
2759 | * either fail the allocation or charge it to this cgroup, do | |
2760 | * it as a temporary condition. But we can't fail. From a | |
2761 | * kmem/slab perspective, the cache has already been selected, | |
2762 | * by mem_cgroup_kmem_get_cache(), so it is too late to change | |
2763 | * our minds. | |
2764 | * | |
2765 | * This condition will only trigger if the task entered | |
2766 | * memcg_charge_kmem in a sane state, but was OOM-killed during | |
2767 | * __mem_cgroup_try_charge() above. Tasks that were already | |
2768 | * dying when the allocation triggers should have been already | |
2769 | * directed to the root cgroup in memcontrol.h | |
2770 | */ | |
2771 | res_counter_charge_nofail(&memcg->res, size, &fail_res); | |
2772 | if (do_swap_account) | |
2773 | res_counter_charge_nofail(&memcg->memsw, size, | |
2774 | &fail_res); | |
2775 | ret = 0; | |
2776 | } else if (ret) | |
2777 | res_counter_uncharge(&memcg->kmem, size); | |
2778 | ||
2779 | return ret; | |
2780 | } | |
2781 | ||
2782 | static void memcg_uncharge_kmem(struct mem_cgroup *memcg, u64 size) | |
2783 | { | |
7ae1e1d0 GC |
2784 | res_counter_uncharge(&memcg->res, size); |
2785 | if (do_swap_account) | |
2786 | res_counter_uncharge(&memcg->memsw, size); | |
7de37682 GC |
2787 | |
2788 | /* Not down to 0 */ | |
2789 | if (res_counter_uncharge(&memcg->kmem, size)) | |
2790 | return; | |
2791 | ||
2792 | if (memcg_kmem_test_and_clear_dead(memcg)) | |
2793 | mem_cgroup_put(memcg); | |
7ae1e1d0 GC |
2794 | } |
2795 | ||
2633d7a0 GC |
2796 | void memcg_cache_list_add(struct mem_cgroup *memcg, struct kmem_cache *cachep) |
2797 | { | |
2798 | if (!memcg) | |
2799 | return; | |
2800 | ||
2801 | mutex_lock(&memcg->slab_caches_mutex); | |
2802 | list_add(&cachep->memcg_params->list, &memcg->memcg_slab_caches); | |
2803 | mutex_unlock(&memcg->slab_caches_mutex); | |
2804 | } | |
2805 | ||
2806 | /* | |
2807 | * helper for acessing a memcg's index. It will be used as an index in the | |
2808 | * child cache array in kmem_cache, and also to derive its name. This function | |
2809 | * will return -1 when this is not a kmem-limited memcg. | |
2810 | */ | |
2811 | int memcg_cache_id(struct mem_cgroup *memcg) | |
2812 | { | |
2813 | return memcg ? memcg->kmemcg_id : -1; | |
2814 | } | |
2815 | ||
2816 | int memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s) | |
2817 | { | |
2818 | size_t size = sizeof(struct memcg_cache_params); | |
2819 | ||
2820 | if (!memcg_kmem_enabled()) | |
2821 | return 0; | |
2822 | ||
2823 | s->memcg_params = kzalloc(size, GFP_KERNEL); | |
2824 | if (!s->memcg_params) | |
2825 | return -ENOMEM; | |
2826 | ||
2827 | if (memcg) | |
2828 | s->memcg_params->memcg = memcg; | |
2829 | return 0; | |
2830 | } | |
2831 | ||
2832 | void memcg_release_cache(struct kmem_cache *s) | |
2833 | { | |
2834 | kfree(s->memcg_params); | |
2835 | } | |
2836 | ||
7ae1e1d0 GC |
2837 | /* |
2838 | * We need to verify if the allocation against current->mm->owner's memcg is | |
2839 | * possible for the given order. But the page is not allocated yet, so we'll | |
2840 | * need a further commit step to do the final arrangements. | |
2841 | * | |
2842 | * It is possible for the task to switch cgroups in this mean time, so at | |
2843 | * commit time, we can't rely on task conversion any longer. We'll then use | |
2844 | * the handle argument to return to the caller which cgroup we should commit | |
2845 | * against. We could also return the memcg directly and avoid the pointer | |
2846 | * passing, but a boolean return value gives better semantics considering | |
2847 | * the compiled-out case as well. | |
2848 | * | |
2849 | * Returning true means the allocation is possible. | |
2850 | */ | |
2851 | bool | |
2852 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) | |
2853 | { | |
2854 | struct mem_cgroup *memcg; | |
2855 | int ret; | |
2856 | ||
2857 | *_memcg = NULL; | |
2858 | memcg = try_get_mem_cgroup_from_mm(current->mm); | |
2859 | ||
2860 | /* | |
2861 | * very rare case described in mem_cgroup_from_task. Unfortunately there | |
2862 | * isn't much we can do without complicating this too much, and it would | |
2863 | * be gfp-dependent anyway. Just let it go | |
2864 | */ | |
2865 | if (unlikely(!memcg)) | |
2866 | return true; | |
2867 | ||
2868 | if (!memcg_can_account_kmem(memcg)) { | |
2869 | css_put(&memcg->css); | |
2870 | return true; | |
2871 | } | |
2872 | ||
7ae1e1d0 GC |
2873 | ret = memcg_charge_kmem(memcg, gfp, PAGE_SIZE << order); |
2874 | if (!ret) | |
2875 | *_memcg = memcg; | |
7ae1e1d0 GC |
2876 | |
2877 | css_put(&memcg->css); | |
2878 | return (ret == 0); | |
2879 | } | |
2880 | ||
2881 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
2882 | int order) | |
2883 | { | |
2884 | struct page_cgroup *pc; | |
2885 | ||
2886 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
2887 | ||
2888 | /* The page allocation failed. Revert */ | |
2889 | if (!page) { | |
2890 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); | |
7ae1e1d0 GC |
2891 | return; |
2892 | } | |
2893 | ||
2894 | pc = lookup_page_cgroup(page); | |
2895 | lock_page_cgroup(pc); | |
2896 | pc->mem_cgroup = memcg; | |
2897 | SetPageCgroupUsed(pc); | |
2898 | unlock_page_cgroup(pc); | |
2899 | } | |
2900 | ||
2901 | void __memcg_kmem_uncharge_pages(struct page *page, int order) | |
2902 | { | |
2903 | struct mem_cgroup *memcg = NULL; | |
2904 | struct page_cgroup *pc; | |
2905 | ||
2906 | ||
2907 | pc = lookup_page_cgroup(page); | |
2908 | /* | |
2909 | * Fast unlocked return. Theoretically might have changed, have to | |
2910 | * check again after locking. | |
2911 | */ | |
2912 | if (!PageCgroupUsed(pc)) | |
2913 | return; | |
2914 | ||
2915 | lock_page_cgroup(pc); | |
2916 | if (PageCgroupUsed(pc)) { | |
2917 | memcg = pc->mem_cgroup; | |
2918 | ClearPageCgroupUsed(pc); | |
2919 | } | |
2920 | unlock_page_cgroup(pc); | |
2921 | ||
2922 | /* | |
2923 | * We trust that only if there is a memcg associated with the page, it | |
2924 | * is a valid allocation | |
2925 | */ | |
2926 | if (!memcg) | |
2927 | return; | |
2928 | ||
2929 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
2930 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); | |
7ae1e1d0 GC |
2931 | } |
2932 | #endif /* CONFIG_MEMCG_KMEM */ | |
2933 | ||
ca3e0214 KH |
2934 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2935 | ||
a0db00fc | 2936 | #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION) |
ca3e0214 KH |
2937 | /* |
2938 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
2939 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
2940 | * charge/uncharge will be never happen and move_account() is done under | |
2941 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 2942 | */ |
e94c8a9c | 2943 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 KH |
2944 | { |
2945 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
e94c8a9c KH |
2946 | struct page_cgroup *pc; |
2947 | int i; | |
ca3e0214 | 2948 | |
3d37c4a9 KH |
2949 | if (mem_cgroup_disabled()) |
2950 | return; | |
e94c8a9c KH |
2951 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
2952 | pc = head_pc + i; | |
2953 | pc->mem_cgroup = head_pc->mem_cgroup; | |
2954 | smp_wmb();/* see __commit_charge() */ | |
e94c8a9c KH |
2955 | pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
2956 | } | |
ca3e0214 | 2957 | } |
12d27107 | 2958 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2959 | |
f817ed48 | 2960 | /** |
de3638d9 | 2961 | * mem_cgroup_move_account - move account of the page |
5564e88b | 2962 | * @page: the page |
7ec99d62 | 2963 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
2964 | * @pc: page_cgroup of the page. |
2965 | * @from: mem_cgroup which the page is moved from. | |
2966 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
2967 | * | |
2968 | * The caller must confirm following. | |
08e552c6 | 2969 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 2970 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 2971 | * |
2f3479b1 KH |
2972 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
2973 | * from old cgroup. | |
f817ed48 | 2974 | */ |
7ec99d62 JW |
2975 | static int mem_cgroup_move_account(struct page *page, |
2976 | unsigned int nr_pages, | |
2977 | struct page_cgroup *pc, | |
2978 | struct mem_cgroup *from, | |
2f3479b1 | 2979 | struct mem_cgroup *to) |
f817ed48 | 2980 | { |
de3638d9 JW |
2981 | unsigned long flags; |
2982 | int ret; | |
b2402857 | 2983 | bool anon = PageAnon(page); |
987eba66 | 2984 | |
f817ed48 | 2985 | VM_BUG_ON(from == to); |
5564e88b | 2986 | VM_BUG_ON(PageLRU(page)); |
de3638d9 JW |
2987 | /* |
2988 | * The page is isolated from LRU. So, collapse function | |
2989 | * will not handle this page. But page splitting can happen. | |
2990 | * Do this check under compound_page_lock(). The caller should | |
2991 | * hold it. | |
2992 | */ | |
2993 | ret = -EBUSY; | |
7ec99d62 | 2994 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
2995 | goto out; |
2996 | ||
2997 | lock_page_cgroup(pc); | |
2998 | ||
2999 | ret = -EINVAL; | |
3000 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
3001 | goto unlock; | |
3002 | ||
312734c0 | 3003 | move_lock_mem_cgroup(from, &flags); |
f817ed48 | 3004 | |
2ff76f11 | 3005 | if (!anon && page_mapped(page)) { |
c62b1a3b KH |
3006 | /* Update mapped_file data for mem_cgroup */ |
3007 | preempt_disable(); | |
3008 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
3009 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
3010 | preempt_enable(); | |
d69b042f | 3011 | } |
b2402857 | 3012 | mem_cgroup_charge_statistics(from, anon, -nr_pages); |
d69b042f | 3013 | |
854ffa8d | 3014 | /* caller should have done css_get */ |
08e552c6 | 3015 | pc->mem_cgroup = to; |
b2402857 | 3016 | mem_cgroup_charge_statistics(to, anon, nr_pages); |
312734c0 | 3017 | move_unlock_mem_cgroup(from, &flags); |
de3638d9 JW |
3018 | ret = 0; |
3019 | unlock: | |
57f9fd7d | 3020 | unlock_page_cgroup(pc); |
d2265e6f KH |
3021 | /* |
3022 | * check events | |
3023 | */ | |
5564e88b JW |
3024 | memcg_check_events(to, page); |
3025 | memcg_check_events(from, page); | |
de3638d9 | 3026 | out: |
f817ed48 KH |
3027 | return ret; |
3028 | } | |
3029 | ||
2ef37d3f MH |
3030 | /** |
3031 | * mem_cgroup_move_parent - moves page to the parent group | |
3032 | * @page: the page to move | |
3033 | * @pc: page_cgroup of the page | |
3034 | * @child: page's cgroup | |
3035 | * | |
3036 | * move charges to its parent or the root cgroup if the group has no | |
3037 | * parent (aka use_hierarchy==0). | |
3038 | * Although this might fail (get_page_unless_zero, isolate_lru_page or | |
3039 | * mem_cgroup_move_account fails) the failure is always temporary and | |
3040 | * it signals a race with a page removal/uncharge or migration. In the | |
3041 | * first case the page is on the way out and it will vanish from the LRU | |
3042 | * on the next attempt and the call should be retried later. | |
3043 | * Isolation from the LRU fails only if page has been isolated from | |
3044 | * the LRU since we looked at it and that usually means either global | |
3045 | * reclaim or migration going on. The page will either get back to the | |
3046 | * LRU or vanish. | |
3047 | * Finaly mem_cgroup_move_account fails only if the page got uncharged | |
3048 | * (!PageCgroupUsed) or moved to a different group. The page will | |
3049 | * disappear in the next attempt. | |
f817ed48 | 3050 | */ |
5564e88b JW |
3051 | static int mem_cgroup_move_parent(struct page *page, |
3052 | struct page_cgroup *pc, | |
6068bf01 | 3053 | struct mem_cgroup *child) |
f817ed48 | 3054 | { |
f817ed48 | 3055 | struct mem_cgroup *parent; |
7ec99d62 | 3056 | unsigned int nr_pages; |
4be4489f | 3057 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
3058 | int ret; |
3059 | ||
d8423011 | 3060 | VM_BUG_ON(mem_cgroup_is_root(child)); |
f817ed48 | 3061 | |
57f9fd7d DN |
3062 | ret = -EBUSY; |
3063 | if (!get_page_unless_zero(page)) | |
3064 | goto out; | |
3065 | if (isolate_lru_page(page)) | |
3066 | goto put; | |
52dbb905 | 3067 | |
7ec99d62 | 3068 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 3069 | |
cc926f78 KH |
3070 | parent = parent_mem_cgroup(child); |
3071 | /* | |
3072 | * If no parent, move charges to root cgroup. | |
3073 | */ | |
3074 | if (!parent) | |
3075 | parent = root_mem_cgroup; | |
f817ed48 | 3076 | |
2ef37d3f MH |
3077 | if (nr_pages > 1) { |
3078 | VM_BUG_ON(!PageTransHuge(page)); | |
987eba66 | 3079 | flags = compound_lock_irqsave(page); |
2ef37d3f | 3080 | } |
987eba66 | 3081 | |
cc926f78 | 3082 | ret = mem_cgroup_move_account(page, nr_pages, |
2f3479b1 | 3083 | pc, child, parent); |
cc926f78 KH |
3084 | if (!ret) |
3085 | __mem_cgroup_cancel_local_charge(child, nr_pages); | |
8dba474f | 3086 | |
7ec99d62 | 3087 | if (nr_pages > 1) |
987eba66 | 3088 | compound_unlock_irqrestore(page, flags); |
08e552c6 | 3089 | putback_lru_page(page); |
57f9fd7d | 3090 | put: |
40d58138 | 3091 | put_page(page); |
57f9fd7d | 3092 | out: |
f817ed48 KH |
3093 | return ret; |
3094 | } | |
3095 | ||
7a81b88c KH |
3096 | /* |
3097 | * Charge the memory controller for page usage. | |
3098 | * Return | |
3099 | * 0 if the charge was successful | |
3100 | * < 0 if the cgroup is over its limit | |
3101 | */ | |
3102 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
73045c47 | 3103 | gfp_t gfp_mask, enum charge_type ctype) |
7a81b88c | 3104 | { |
c0ff4b85 | 3105 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 3106 | unsigned int nr_pages = 1; |
8493ae43 | 3107 | bool oom = true; |
7a81b88c | 3108 | int ret; |
ec168510 | 3109 | |
37c2ac78 | 3110 | if (PageTransHuge(page)) { |
7ec99d62 | 3111 | nr_pages <<= compound_order(page); |
37c2ac78 | 3112 | VM_BUG_ON(!PageTransHuge(page)); |
8493ae43 JW |
3113 | /* |
3114 | * Never OOM-kill a process for a huge page. The | |
3115 | * fault handler will fall back to regular pages. | |
3116 | */ | |
3117 | oom = false; | |
37c2ac78 | 3118 | } |
7a81b88c | 3119 | |
c0ff4b85 | 3120 | ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom); |
38c5d72f | 3121 | if (ret == -ENOMEM) |
7a81b88c | 3122 | return ret; |
ce587e65 | 3123 | __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false); |
8a9f3ccd | 3124 | return 0; |
8a9f3ccd BS |
3125 | } |
3126 | ||
7a81b88c KH |
3127 | int mem_cgroup_newpage_charge(struct page *page, |
3128 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 3129 | { |
f8d66542 | 3130 | if (mem_cgroup_disabled()) |
cede86ac | 3131 | return 0; |
7a0524cf JW |
3132 | VM_BUG_ON(page_mapped(page)); |
3133 | VM_BUG_ON(page->mapping && !PageAnon(page)); | |
3134 | VM_BUG_ON(!mm); | |
217bc319 | 3135 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
41326c17 | 3136 | MEM_CGROUP_CHARGE_TYPE_ANON); |
217bc319 KH |
3137 | } |
3138 | ||
54595fe2 KH |
3139 | /* |
3140 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
3141 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 3142 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
3143 | * "commit()" or removed by "cancel()" |
3144 | */ | |
0435a2fd JW |
3145 | static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
3146 | struct page *page, | |
3147 | gfp_t mask, | |
3148 | struct mem_cgroup **memcgp) | |
8c7c6e34 | 3149 | { |
c0ff4b85 | 3150 | struct mem_cgroup *memcg; |
90deb788 | 3151 | struct page_cgroup *pc; |
54595fe2 | 3152 | int ret; |
8c7c6e34 | 3153 | |
90deb788 JW |
3154 | pc = lookup_page_cgroup(page); |
3155 | /* | |
3156 | * Every swap fault against a single page tries to charge the | |
3157 | * page, bail as early as possible. shmem_unuse() encounters | |
3158 | * already charged pages, too. The USED bit is protected by | |
3159 | * the page lock, which serializes swap cache removal, which | |
3160 | * in turn serializes uncharging. | |
3161 | */ | |
3162 | if (PageCgroupUsed(pc)) | |
3163 | return 0; | |
8c7c6e34 KH |
3164 | if (!do_swap_account) |
3165 | goto charge_cur_mm; | |
c0ff4b85 R |
3166 | memcg = try_get_mem_cgroup_from_page(page); |
3167 | if (!memcg) | |
54595fe2 | 3168 | goto charge_cur_mm; |
72835c86 JW |
3169 | *memcgp = memcg; |
3170 | ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true); | |
c0ff4b85 | 3171 | css_put(&memcg->css); |
38c5d72f KH |
3172 | if (ret == -EINTR) |
3173 | ret = 0; | |
54595fe2 | 3174 | return ret; |
8c7c6e34 | 3175 | charge_cur_mm: |
38c5d72f KH |
3176 | ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true); |
3177 | if (ret == -EINTR) | |
3178 | ret = 0; | |
3179 | return ret; | |
8c7c6e34 KH |
3180 | } |
3181 | ||
0435a2fd JW |
3182 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page, |
3183 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
3184 | { | |
3185 | *memcgp = NULL; | |
3186 | if (mem_cgroup_disabled()) | |
3187 | return 0; | |
bdf4f4d2 JW |
3188 | /* |
3189 | * A racing thread's fault, or swapoff, may have already | |
3190 | * updated the pte, and even removed page from swap cache: in | |
3191 | * those cases unuse_pte()'s pte_same() test will fail; but | |
3192 | * there's also a KSM case which does need to charge the page. | |
3193 | */ | |
3194 | if (!PageSwapCache(page)) { | |
3195 | int ret; | |
3196 | ||
3197 | ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, memcgp, true); | |
3198 | if (ret == -EINTR) | |
3199 | ret = 0; | |
3200 | return ret; | |
3201 | } | |
0435a2fd JW |
3202 | return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp); |
3203 | } | |
3204 | ||
827a03d2 JW |
3205 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg) |
3206 | { | |
3207 | if (mem_cgroup_disabled()) | |
3208 | return; | |
3209 | if (!memcg) | |
3210 | return; | |
3211 | __mem_cgroup_cancel_charge(memcg, 1); | |
3212 | } | |
3213 | ||
83aae4c7 | 3214 | static void |
72835c86 | 3215 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg, |
83aae4c7 | 3216 | enum charge_type ctype) |
7a81b88c | 3217 | { |
f8d66542 | 3218 | if (mem_cgroup_disabled()) |
7a81b88c | 3219 | return; |
72835c86 | 3220 | if (!memcg) |
7a81b88c | 3221 | return; |
5a6475a4 | 3222 | |
ce587e65 | 3223 | __mem_cgroup_commit_charge(memcg, page, 1, ctype, true); |
8c7c6e34 KH |
3224 | /* |
3225 | * Now swap is on-memory. This means this page may be | |
3226 | * counted both as mem and swap....double count. | |
03f3c433 KH |
3227 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
3228 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
3229 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 3230 | */ |
03f3c433 | 3231 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 3232 | swp_entry_t ent = {.val = page_private(page)}; |
86493009 | 3233 | mem_cgroup_uncharge_swap(ent); |
8c7c6e34 | 3234 | } |
7a81b88c KH |
3235 | } |
3236 | ||
72835c86 JW |
3237 | void mem_cgroup_commit_charge_swapin(struct page *page, |
3238 | struct mem_cgroup *memcg) | |
83aae4c7 | 3239 | { |
72835c86 | 3240 | __mem_cgroup_commit_charge_swapin(page, memcg, |
41326c17 | 3241 | MEM_CGROUP_CHARGE_TYPE_ANON); |
83aae4c7 DN |
3242 | } |
3243 | ||
827a03d2 JW |
3244 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
3245 | gfp_t gfp_mask) | |
7a81b88c | 3246 | { |
827a03d2 JW |
3247 | struct mem_cgroup *memcg = NULL; |
3248 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
3249 | int ret; | |
3250 | ||
f8d66542 | 3251 | if (mem_cgroup_disabled()) |
827a03d2 JW |
3252 | return 0; |
3253 | if (PageCompound(page)) | |
3254 | return 0; | |
3255 | ||
827a03d2 JW |
3256 | if (!PageSwapCache(page)) |
3257 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, type); | |
3258 | else { /* page is swapcache/shmem */ | |
0435a2fd JW |
3259 | ret = __mem_cgroup_try_charge_swapin(mm, page, |
3260 | gfp_mask, &memcg); | |
827a03d2 JW |
3261 | if (!ret) |
3262 | __mem_cgroup_commit_charge_swapin(page, memcg, type); | |
3263 | } | |
3264 | return ret; | |
7a81b88c KH |
3265 | } |
3266 | ||
c0ff4b85 | 3267 | static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg, |
7ec99d62 JW |
3268 | unsigned int nr_pages, |
3269 | const enum charge_type ctype) | |
569b846d KH |
3270 | { |
3271 | struct memcg_batch_info *batch = NULL; | |
3272 | bool uncharge_memsw = true; | |
7ec99d62 | 3273 | |
569b846d KH |
3274 | /* If swapout, usage of swap doesn't decrease */ |
3275 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
3276 | uncharge_memsw = false; | |
569b846d KH |
3277 | |
3278 | batch = ¤t->memcg_batch; | |
3279 | /* | |
3280 | * In usual, we do css_get() when we remember memcg pointer. | |
3281 | * But in this case, we keep res->usage until end of a series of | |
3282 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
3283 | */ | |
3284 | if (!batch->memcg) | |
c0ff4b85 | 3285 | batch->memcg = memcg; |
3c11ecf4 KH |
3286 | /* |
3287 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
25985edc | 3288 | * In those cases, all pages freed continuously can be expected to be in |
3c11ecf4 KH |
3289 | * the same cgroup and we have chance to coalesce uncharges. |
3290 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
3291 | * because we want to do uncharge as soon as possible. | |
3292 | */ | |
3293 | ||
3294 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
3295 | goto direct_uncharge; | |
3296 | ||
7ec99d62 | 3297 | if (nr_pages > 1) |
ec168510 AA |
3298 | goto direct_uncharge; |
3299 | ||
569b846d KH |
3300 | /* |
3301 | * In typical case, batch->memcg == mem. This means we can | |
3302 | * merge a series of uncharges to an uncharge of res_counter. | |
3303 | * If not, we uncharge res_counter ony by one. | |
3304 | */ | |
c0ff4b85 | 3305 | if (batch->memcg != memcg) |
569b846d KH |
3306 | goto direct_uncharge; |
3307 | /* remember freed charge and uncharge it later */ | |
7ffd4ca7 | 3308 | batch->nr_pages++; |
569b846d | 3309 | if (uncharge_memsw) |
7ffd4ca7 | 3310 | batch->memsw_nr_pages++; |
569b846d KH |
3311 | return; |
3312 | direct_uncharge: | |
c0ff4b85 | 3313 | res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE); |
569b846d | 3314 | if (uncharge_memsw) |
c0ff4b85 R |
3315 | res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE); |
3316 | if (unlikely(batch->memcg != memcg)) | |
3317 | memcg_oom_recover(memcg); | |
569b846d | 3318 | } |
7a81b88c | 3319 | |
8a9f3ccd | 3320 | /* |
69029cd5 | 3321 | * uncharge if !page_mapped(page) |
8a9f3ccd | 3322 | */ |
8c7c6e34 | 3323 | static struct mem_cgroup * |
0030f535 JW |
3324 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype, |
3325 | bool end_migration) | |
8a9f3ccd | 3326 | { |
c0ff4b85 | 3327 | struct mem_cgroup *memcg = NULL; |
7ec99d62 JW |
3328 | unsigned int nr_pages = 1; |
3329 | struct page_cgroup *pc; | |
b2402857 | 3330 | bool anon; |
8a9f3ccd | 3331 | |
f8d66542 | 3332 | if (mem_cgroup_disabled()) |
8c7c6e34 | 3333 | return NULL; |
4077960e | 3334 | |
0c59b89c | 3335 | VM_BUG_ON(PageSwapCache(page)); |
d13d1443 | 3336 | |
37c2ac78 | 3337 | if (PageTransHuge(page)) { |
7ec99d62 | 3338 | nr_pages <<= compound_order(page); |
37c2ac78 AA |
3339 | VM_BUG_ON(!PageTransHuge(page)); |
3340 | } | |
8697d331 | 3341 | /* |
3c541e14 | 3342 | * Check if our page_cgroup is valid |
8697d331 | 3343 | */ |
52d4b9ac | 3344 | pc = lookup_page_cgroup(page); |
cfa44946 | 3345 | if (unlikely(!PageCgroupUsed(pc))) |
8c7c6e34 | 3346 | return NULL; |
b9c565d5 | 3347 | |
52d4b9ac | 3348 | lock_page_cgroup(pc); |
d13d1443 | 3349 | |
c0ff4b85 | 3350 | memcg = pc->mem_cgroup; |
8c7c6e34 | 3351 | |
d13d1443 KH |
3352 | if (!PageCgroupUsed(pc)) |
3353 | goto unlock_out; | |
3354 | ||
b2402857 KH |
3355 | anon = PageAnon(page); |
3356 | ||
d13d1443 | 3357 | switch (ctype) { |
41326c17 | 3358 | case MEM_CGROUP_CHARGE_TYPE_ANON: |
2ff76f11 KH |
3359 | /* |
3360 | * Generally PageAnon tells if it's the anon statistics to be | |
3361 | * updated; but sometimes e.g. mem_cgroup_uncharge_page() is | |
3362 | * used before page reached the stage of being marked PageAnon. | |
3363 | */ | |
b2402857 KH |
3364 | anon = true; |
3365 | /* fallthrough */ | |
8a9478ca | 3366 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c | 3367 | /* See mem_cgroup_prepare_migration() */ |
0030f535 JW |
3368 | if (page_mapped(page)) |
3369 | goto unlock_out; | |
3370 | /* | |
3371 | * Pages under migration may not be uncharged. But | |
3372 | * end_migration() /must/ be the one uncharging the | |
3373 | * unused post-migration page and so it has to call | |
3374 | * here with the migration bit still set. See the | |
3375 | * res_counter handling below. | |
3376 | */ | |
3377 | if (!end_migration && PageCgroupMigration(pc)) | |
d13d1443 KH |
3378 | goto unlock_out; |
3379 | break; | |
3380 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
3381 | if (!PageAnon(page)) { /* Shared memory */ | |
3382 | if (page->mapping && !page_is_file_cache(page)) | |
3383 | goto unlock_out; | |
3384 | } else if (page_mapped(page)) /* Anon */ | |
3385 | goto unlock_out; | |
3386 | break; | |
3387 | default: | |
3388 | break; | |
52d4b9ac | 3389 | } |
d13d1443 | 3390 | |
b2402857 | 3391 | mem_cgroup_charge_statistics(memcg, anon, -nr_pages); |
04046e1a | 3392 | |
52d4b9ac | 3393 | ClearPageCgroupUsed(pc); |
544122e5 KH |
3394 | /* |
3395 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
3396 | * freed from LRU. This is safe because uncharged page is expected not | |
3397 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
3398 | * special functions. | |
3399 | */ | |
b9c565d5 | 3400 | |
52d4b9ac | 3401 | unlock_page_cgroup(pc); |
f75ca962 | 3402 | /* |
c0ff4b85 | 3403 | * even after unlock, we have memcg->res.usage here and this memcg |
f75ca962 KH |
3404 | * will never be freed. |
3405 | */ | |
c0ff4b85 | 3406 | memcg_check_events(memcg, page); |
f75ca962 | 3407 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
c0ff4b85 R |
3408 | mem_cgroup_swap_statistics(memcg, true); |
3409 | mem_cgroup_get(memcg); | |
f75ca962 | 3410 | } |
0030f535 JW |
3411 | /* |
3412 | * Migration does not charge the res_counter for the | |
3413 | * replacement page, so leave it alone when phasing out the | |
3414 | * page that is unused after the migration. | |
3415 | */ | |
3416 | if (!end_migration && !mem_cgroup_is_root(memcg)) | |
c0ff4b85 | 3417 | mem_cgroup_do_uncharge(memcg, nr_pages, ctype); |
6d12e2d8 | 3418 | |
c0ff4b85 | 3419 | return memcg; |
d13d1443 KH |
3420 | |
3421 | unlock_out: | |
3422 | unlock_page_cgroup(pc); | |
8c7c6e34 | 3423 | return NULL; |
3c541e14 BS |
3424 | } |
3425 | ||
69029cd5 KH |
3426 | void mem_cgroup_uncharge_page(struct page *page) |
3427 | { | |
52d4b9ac KH |
3428 | /* early check. */ |
3429 | if (page_mapped(page)) | |
3430 | return; | |
40f23a21 | 3431 | VM_BUG_ON(page->mapping && !PageAnon(page)); |
0c59b89c JW |
3432 | if (PageSwapCache(page)) |
3433 | return; | |
0030f535 | 3434 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false); |
69029cd5 KH |
3435 | } |
3436 | ||
3437 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
3438 | { | |
3439 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 3440 | VM_BUG_ON(page->mapping); |
0030f535 | 3441 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false); |
69029cd5 KH |
3442 | } |
3443 | ||
569b846d KH |
3444 | /* |
3445 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
3446 | * In that cases, pages are freed continuously and we can expect pages | |
3447 | * are in the same memcg. All these calls itself limits the number of | |
3448 | * pages freed at once, then uncharge_start/end() is called properly. | |
3449 | * This may be called prural(2) times in a context, | |
3450 | */ | |
3451 | ||
3452 | void mem_cgroup_uncharge_start(void) | |
3453 | { | |
3454 | current->memcg_batch.do_batch++; | |
3455 | /* We can do nest. */ | |
3456 | if (current->memcg_batch.do_batch == 1) { | |
3457 | current->memcg_batch.memcg = NULL; | |
7ffd4ca7 JW |
3458 | current->memcg_batch.nr_pages = 0; |
3459 | current->memcg_batch.memsw_nr_pages = 0; | |
569b846d KH |
3460 | } |
3461 | } | |
3462 | ||
3463 | void mem_cgroup_uncharge_end(void) | |
3464 | { | |
3465 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
3466 | ||
3467 | if (!batch->do_batch) | |
3468 | return; | |
3469 | ||
3470 | batch->do_batch--; | |
3471 | if (batch->do_batch) /* If stacked, do nothing. */ | |
3472 | return; | |
3473 | ||
3474 | if (!batch->memcg) | |
3475 | return; | |
3476 | /* | |
3477 | * This "batch->memcg" is valid without any css_get/put etc... | |
3478 | * bacause we hide charges behind us. | |
3479 | */ | |
7ffd4ca7 JW |
3480 | if (batch->nr_pages) |
3481 | res_counter_uncharge(&batch->memcg->res, | |
3482 | batch->nr_pages * PAGE_SIZE); | |
3483 | if (batch->memsw_nr_pages) | |
3484 | res_counter_uncharge(&batch->memcg->memsw, | |
3485 | batch->memsw_nr_pages * PAGE_SIZE); | |
3c11ecf4 | 3486 | memcg_oom_recover(batch->memcg); |
569b846d KH |
3487 | /* forget this pointer (for sanity check) */ |
3488 | batch->memcg = NULL; | |
3489 | } | |
3490 | ||
e767e056 | 3491 | #ifdef CONFIG_SWAP |
8c7c6e34 | 3492 | /* |
e767e056 | 3493 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
3494 | * memcg information is recorded to swap_cgroup of "ent" |
3495 | */ | |
8a9478ca KH |
3496 | void |
3497 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
3498 | { |
3499 | struct mem_cgroup *memcg; | |
8a9478ca KH |
3500 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
3501 | ||
3502 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
3503 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
3504 | ||
0030f535 | 3505 | memcg = __mem_cgroup_uncharge_common(page, ctype, false); |
8c7c6e34 | 3506 | |
f75ca962 KH |
3507 | /* |
3508 | * record memcg information, if swapout && memcg != NULL, | |
3509 | * mem_cgroup_get() was called in uncharge(). | |
3510 | */ | |
3511 | if (do_swap_account && swapout && memcg) | |
a3b2d692 | 3512 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 | 3513 | } |
e767e056 | 3514 | #endif |
8c7c6e34 | 3515 | |
c255a458 | 3516 | #ifdef CONFIG_MEMCG_SWAP |
8c7c6e34 KH |
3517 | /* |
3518 | * called from swap_entry_free(). remove record in swap_cgroup and | |
3519 | * uncharge "memsw" account. | |
3520 | */ | |
3521 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 3522 | { |
8c7c6e34 | 3523 | struct mem_cgroup *memcg; |
a3b2d692 | 3524 | unsigned short id; |
8c7c6e34 KH |
3525 | |
3526 | if (!do_swap_account) | |
3527 | return; | |
3528 | ||
a3b2d692 KH |
3529 | id = swap_cgroup_record(ent, 0); |
3530 | rcu_read_lock(); | |
3531 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 3532 | if (memcg) { |
a3b2d692 KH |
3533 | /* |
3534 | * We uncharge this because swap is freed. | |
3535 | * This memcg can be obsolete one. We avoid calling css_tryget | |
3536 | */ | |
0c3e73e8 | 3537 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 3538 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 3539 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
3540 | mem_cgroup_put(memcg); |
3541 | } | |
a3b2d692 | 3542 | rcu_read_unlock(); |
d13d1443 | 3543 | } |
02491447 DN |
3544 | |
3545 | /** | |
3546 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3547 | * @entry: swap entry to be moved | |
3548 | * @from: mem_cgroup which the entry is moved from | |
3549 | * @to: mem_cgroup which the entry is moved to | |
3550 | * | |
3551 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3552 | * as the mem_cgroup's id of @from. | |
3553 | * | |
3554 | * Returns 0 on success, -EINVAL on failure. | |
3555 | * | |
3556 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
3557 | * both res and memsw, and called css_get(). | |
3558 | */ | |
3559 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3560 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3561 | { |
3562 | unsigned short old_id, new_id; | |
3563 | ||
3564 | old_id = css_id(&from->css); | |
3565 | new_id = css_id(&to->css); | |
3566 | ||
3567 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 3568 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 3569 | mem_cgroup_swap_statistics(to, true); |
02491447 | 3570 | /* |
483c30b5 DN |
3571 | * This function is only called from task migration context now. |
3572 | * It postpones res_counter and refcount handling till the end | |
3573 | * of task migration(mem_cgroup_clear_mc()) for performance | |
3574 | * improvement. But we cannot postpone mem_cgroup_get(to) | |
3575 | * because if the process that has been moved to @to does | |
3576 | * swap-in, the refcount of @to might be decreased to 0. | |
02491447 | 3577 | */ |
02491447 | 3578 | mem_cgroup_get(to); |
02491447 DN |
3579 | return 0; |
3580 | } | |
3581 | return -EINVAL; | |
3582 | } | |
3583 | #else | |
3584 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3585 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3586 | { |
3587 | return -EINVAL; | |
3588 | } | |
8c7c6e34 | 3589 | #endif |
d13d1443 | 3590 | |
ae41be37 | 3591 | /* |
01b1ae63 KH |
3592 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
3593 | * page belongs to. | |
ae41be37 | 3594 | */ |
0030f535 JW |
3595 | void mem_cgroup_prepare_migration(struct page *page, struct page *newpage, |
3596 | struct mem_cgroup **memcgp) | |
ae41be37 | 3597 | { |
c0ff4b85 | 3598 | struct mem_cgroup *memcg = NULL; |
b32967ff | 3599 | unsigned int nr_pages = 1; |
7ec99d62 | 3600 | struct page_cgroup *pc; |
ac39cf8c | 3601 | enum charge_type ctype; |
8869b8f6 | 3602 | |
72835c86 | 3603 | *memcgp = NULL; |
56039efa | 3604 | |
f8d66542 | 3605 | if (mem_cgroup_disabled()) |
0030f535 | 3606 | return; |
4077960e | 3607 | |
b32967ff MG |
3608 | if (PageTransHuge(page)) |
3609 | nr_pages <<= compound_order(page); | |
3610 | ||
52d4b9ac KH |
3611 | pc = lookup_page_cgroup(page); |
3612 | lock_page_cgroup(pc); | |
3613 | if (PageCgroupUsed(pc)) { | |
c0ff4b85 R |
3614 | memcg = pc->mem_cgroup; |
3615 | css_get(&memcg->css); | |
ac39cf8c | 3616 | /* |
3617 | * At migrating an anonymous page, its mapcount goes down | |
3618 | * to 0 and uncharge() will be called. But, even if it's fully | |
3619 | * unmapped, migration may fail and this page has to be | |
3620 | * charged again. We set MIGRATION flag here and delay uncharge | |
3621 | * until end_migration() is called | |
3622 | * | |
3623 | * Corner Case Thinking | |
3624 | * A) | |
3625 | * When the old page was mapped as Anon and it's unmap-and-freed | |
3626 | * while migration was ongoing. | |
3627 | * If unmap finds the old page, uncharge() of it will be delayed | |
3628 | * until end_migration(). If unmap finds a new page, it's | |
3629 | * uncharged when it make mapcount to be 1->0. If unmap code | |
3630 | * finds swap_migration_entry, the new page will not be mapped | |
3631 | * and end_migration() will find it(mapcount==0). | |
3632 | * | |
3633 | * B) | |
3634 | * When the old page was mapped but migraion fails, the kernel | |
3635 | * remaps it. A charge for it is kept by MIGRATION flag even | |
3636 | * if mapcount goes down to 0. We can do remap successfully | |
3637 | * without charging it again. | |
3638 | * | |
3639 | * C) | |
3640 | * The "old" page is under lock_page() until the end of | |
3641 | * migration, so, the old page itself will not be swapped-out. | |
3642 | * If the new page is swapped out before end_migraton, our | |
3643 | * hook to usual swap-out path will catch the event. | |
3644 | */ | |
3645 | if (PageAnon(page)) | |
3646 | SetPageCgroupMigration(pc); | |
e8589cc1 | 3647 | } |
52d4b9ac | 3648 | unlock_page_cgroup(pc); |
ac39cf8c | 3649 | /* |
3650 | * If the page is not charged at this point, | |
3651 | * we return here. | |
3652 | */ | |
c0ff4b85 | 3653 | if (!memcg) |
0030f535 | 3654 | return; |
01b1ae63 | 3655 | |
72835c86 | 3656 | *memcgp = memcg; |
ac39cf8c | 3657 | /* |
3658 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
3659 | * is called before end_migration, we can catch all events on this new | |
3660 | * page. In the case new page is migrated but not remapped, new page's | |
3661 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
3662 | */ | |
ac39cf8c | 3663 | if (PageAnon(page)) |
41326c17 | 3664 | ctype = MEM_CGROUP_CHARGE_TYPE_ANON; |
ac39cf8c | 3665 | else |
62ba7442 | 3666 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; |
0030f535 JW |
3667 | /* |
3668 | * The page is committed to the memcg, but it's not actually | |
3669 | * charged to the res_counter since we plan on replacing the | |
3670 | * old one and only one page is going to be left afterwards. | |
3671 | */ | |
b32967ff | 3672 | __mem_cgroup_commit_charge(memcg, newpage, nr_pages, ctype, false); |
ae41be37 | 3673 | } |
8869b8f6 | 3674 | |
69029cd5 | 3675 | /* remove redundant charge if migration failed*/ |
c0ff4b85 | 3676 | void mem_cgroup_end_migration(struct mem_cgroup *memcg, |
50de1dd9 | 3677 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 3678 | { |
ac39cf8c | 3679 | struct page *used, *unused; |
01b1ae63 | 3680 | struct page_cgroup *pc; |
b2402857 | 3681 | bool anon; |
01b1ae63 | 3682 | |
c0ff4b85 | 3683 | if (!memcg) |
01b1ae63 | 3684 | return; |
b25ed609 | 3685 | |
50de1dd9 | 3686 | if (!migration_ok) { |
ac39cf8c | 3687 | used = oldpage; |
3688 | unused = newpage; | |
01b1ae63 | 3689 | } else { |
ac39cf8c | 3690 | used = newpage; |
01b1ae63 KH |
3691 | unused = oldpage; |
3692 | } | |
0030f535 | 3693 | anon = PageAnon(used); |
7d188958 JW |
3694 | __mem_cgroup_uncharge_common(unused, |
3695 | anon ? MEM_CGROUP_CHARGE_TYPE_ANON | |
3696 | : MEM_CGROUP_CHARGE_TYPE_CACHE, | |
3697 | true); | |
0030f535 | 3698 | css_put(&memcg->css); |
69029cd5 | 3699 | /* |
ac39cf8c | 3700 | * We disallowed uncharge of pages under migration because mapcount |
3701 | * of the page goes down to zero, temporarly. | |
3702 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 3703 | */ |
ac39cf8c | 3704 | pc = lookup_page_cgroup(oldpage); |
3705 | lock_page_cgroup(pc); | |
3706 | ClearPageCgroupMigration(pc); | |
3707 | unlock_page_cgroup(pc); | |
ac39cf8c | 3708 | |
01b1ae63 | 3709 | /* |
ac39cf8c | 3710 | * If a page is a file cache, radix-tree replacement is very atomic |
3711 | * and we can skip this check. When it was an Anon page, its mapcount | |
3712 | * goes down to 0. But because we added MIGRATION flage, it's not | |
3713 | * uncharged yet. There are several case but page->mapcount check | |
3714 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
3715 | * check. (see prepare_charge() also) | |
69029cd5 | 3716 | */ |
b2402857 | 3717 | if (anon) |
ac39cf8c | 3718 | mem_cgroup_uncharge_page(used); |
ae41be37 | 3719 | } |
78fb7466 | 3720 | |
ab936cbc KH |
3721 | /* |
3722 | * At replace page cache, newpage is not under any memcg but it's on | |
3723 | * LRU. So, this function doesn't touch res_counter but handles LRU | |
3724 | * in correct way. Both pages are locked so we cannot race with uncharge. | |
3725 | */ | |
3726 | void mem_cgroup_replace_page_cache(struct page *oldpage, | |
3727 | struct page *newpage) | |
3728 | { | |
bde05d1c | 3729 | struct mem_cgroup *memcg = NULL; |
ab936cbc | 3730 | struct page_cgroup *pc; |
ab936cbc | 3731 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
ab936cbc KH |
3732 | |
3733 | if (mem_cgroup_disabled()) | |
3734 | return; | |
3735 | ||
3736 | pc = lookup_page_cgroup(oldpage); | |
3737 | /* fix accounting on old pages */ | |
3738 | lock_page_cgroup(pc); | |
bde05d1c HD |
3739 | if (PageCgroupUsed(pc)) { |
3740 | memcg = pc->mem_cgroup; | |
3741 | mem_cgroup_charge_statistics(memcg, false, -1); | |
3742 | ClearPageCgroupUsed(pc); | |
3743 | } | |
ab936cbc KH |
3744 | unlock_page_cgroup(pc); |
3745 | ||
bde05d1c HD |
3746 | /* |
3747 | * When called from shmem_replace_page(), in some cases the | |
3748 | * oldpage has already been charged, and in some cases not. | |
3749 | */ | |
3750 | if (!memcg) | |
3751 | return; | |
ab936cbc KH |
3752 | /* |
3753 | * Even if newpage->mapping was NULL before starting replacement, | |
3754 | * the newpage may be on LRU(or pagevec for LRU) already. We lock | |
3755 | * LRU while we overwrite pc->mem_cgroup. | |
3756 | */ | |
ce587e65 | 3757 | __mem_cgroup_commit_charge(memcg, newpage, 1, type, true); |
ab936cbc KH |
3758 | } |
3759 | ||
f212ad7c DN |
3760 | #ifdef CONFIG_DEBUG_VM |
3761 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
3762 | { | |
3763 | struct page_cgroup *pc; | |
3764 | ||
3765 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
3766 | /* |
3767 | * Can be NULL while feeding pages into the page allocator for | |
3768 | * the first time, i.e. during boot or memory hotplug; | |
3769 | * or when mem_cgroup_disabled(). | |
3770 | */ | |
f212ad7c DN |
3771 | if (likely(pc) && PageCgroupUsed(pc)) |
3772 | return pc; | |
3773 | return NULL; | |
3774 | } | |
3775 | ||
3776 | bool mem_cgroup_bad_page_check(struct page *page) | |
3777 | { | |
3778 | if (mem_cgroup_disabled()) | |
3779 | return false; | |
3780 | ||
3781 | return lookup_page_cgroup_used(page) != NULL; | |
3782 | } | |
3783 | ||
3784 | void mem_cgroup_print_bad_page(struct page *page) | |
3785 | { | |
3786 | struct page_cgroup *pc; | |
3787 | ||
3788 | pc = lookup_page_cgroup_used(page); | |
3789 | if (pc) { | |
90b3feae | 3790 | printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p\n", |
f212ad7c | 3791 | pc, pc->flags, pc->mem_cgroup); |
f212ad7c DN |
3792 | } |
3793 | } | |
3794 | #endif | |
3795 | ||
8c7c6e34 KH |
3796 | static DEFINE_MUTEX(set_limit_mutex); |
3797 | ||
d38d2a75 | 3798 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 3799 | unsigned long long val) |
628f4235 | 3800 | { |
81d39c20 | 3801 | int retry_count; |
3c11ecf4 | 3802 | u64 memswlimit, memlimit; |
628f4235 | 3803 | int ret = 0; |
81d39c20 KH |
3804 | int children = mem_cgroup_count_children(memcg); |
3805 | u64 curusage, oldusage; | |
3c11ecf4 | 3806 | int enlarge; |
81d39c20 KH |
3807 | |
3808 | /* | |
3809 | * For keeping hierarchical_reclaim simple, how long we should retry | |
3810 | * is depends on callers. We set our retry-count to be function | |
3811 | * of # of children which we should visit in this loop. | |
3812 | */ | |
3813 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
3814 | ||
3815 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 3816 | |
3c11ecf4 | 3817 | enlarge = 0; |
8c7c6e34 | 3818 | while (retry_count) { |
628f4235 KH |
3819 | if (signal_pending(current)) { |
3820 | ret = -EINTR; | |
3821 | break; | |
3822 | } | |
8c7c6e34 KH |
3823 | /* |
3824 | * Rather than hide all in some function, I do this in | |
3825 | * open coded manner. You see what this really does. | |
aaad153e | 3826 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
3827 | */ |
3828 | mutex_lock(&set_limit_mutex); | |
3829 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3830 | if (memswlimit < val) { | |
3831 | ret = -EINVAL; | |
3832 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
3833 | break; |
3834 | } | |
3c11ecf4 KH |
3835 | |
3836 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3837 | if (memlimit < val) | |
3838 | enlarge = 1; | |
3839 | ||
8c7c6e34 | 3840 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
3841 | if (!ret) { |
3842 | if (memswlimit == val) | |
3843 | memcg->memsw_is_minimum = true; | |
3844 | else | |
3845 | memcg->memsw_is_minimum = false; | |
3846 | } | |
8c7c6e34 KH |
3847 | mutex_unlock(&set_limit_mutex); |
3848 | ||
3849 | if (!ret) | |
3850 | break; | |
3851 | ||
5660048c JW |
3852 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
3853 | MEM_CGROUP_RECLAIM_SHRINK); | |
81d39c20 KH |
3854 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
3855 | /* Usage is reduced ? */ | |
3856 | if (curusage >= oldusage) | |
3857 | retry_count--; | |
3858 | else | |
3859 | oldusage = curusage; | |
8c7c6e34 | 3860 | } |
3c11ecf4 KH |
3861 | if (!ret && enlarge) |
3862 | memcg_oom_recover(memcg); | |
14797e23 | 3863 | |
8c7c6e34 KH |
3864 | return ret; |
3865 | } | |
3866 | ||
338c8431 LZ |
3867 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3868 | unsigned long long val) | |
8c7c6e34 | 3869 | { |
81d39c20 | 3870 | int retry_count; |
3c11ecf4 | 3871 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
3872 | int children = mem_cgroup_count_children(memcg); |
3873 | int ret = -EBUSY; | |
3c11ecf4 | 3874 | int enlarge = 0; |
8c7c6e34 | 3875 | |
81d39c20 KH |
3876 | /* see mem_cgroup_resize_res_limit */ |
3877 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
3878 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
3879 | while (retry_count) { |
3880 | if (signal_pending(current)) { | |
3881 | ret = -EINTR; | |
3882 | break; | |
3883 | } | |
3884 | /* | |
3885 | * Rather than hide all in some function, I do this in | |
3886 | * open coded manner. You see what this really does. | |
aaad153e | 3887 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
3888 | */ |
3889 | mutex_lock(&set_limit_mutex); | |
3890 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3891 | if (memlimit > val) { | |
3892 | ret = -EINVAL; | |
3893 | mutex_unlock(&set_limit_mutex); | |
3894 | break; | |
3895 | } | |
3c11ecf4 KH |
3896 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
3897 | if (memswlimit < val) | |
3898 | enlarge = 1; | |
8c7c6e34 | 3899 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
3900 | if (!ret) { |
3901 | if (memlimit == val) | |
3902 | memcg->memsw_is_minimum = true; | |
3903 | else | |
3904 | memcg->memsw_is_minimum = false; | |
3905 | } | |
8c7c6e34 KH |
3906 | mutex_unlock(&set_limit_mutex); |
3907 | ||
3908 | if (!ret) | |
3909 | break; | |
3910 | ||
5660048c JW |
3911 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
3912 | MEM_CGROUP_RECLAIM_NOSWAP | | |
3913 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 3914 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 3915 | /* Usage is reduced ? */ |
8c7c6e34 | 3916 | if (curusage >= oldusage) |
628f4235 | 3917 | retry_count--; |
81d39c20 KH |
3918 | else |
3919 | oldusage = curusage; | |
628f4235 | 3920 | } |
3c11ecf4 KH |
3921 | if (!ret && enlarge) |
3922 | memcg_oom_recover(memcg); | |
628f4235 KH |
3923 | return ret; |
3924 | } | |
3925 | ||
4e416953 | 3926 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
0ae5e89c YH |
3927 | gfp_t gfp_mask, |
3928 | unsigned long *total_scanned) | |
4e416953 BS |
3929 | { |
3930 | unsigned long nr_reclaimed = 0; | |
3931 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3932 | unsigned long reclaimed; | |
3933 | int loop = 0; | |
3934 | struct mem_cgroup_tree_per_zone *mctz; | |
ef8745c1 | 3935 | unsigned long long excess; |
0ae5e89c | 3936 | unsigned long nr_scanned; |
4e416953 BS |
3937 | |
3938 | if (order > 0) | |
3939 | return 0; | |
3940 | ||
00918b6a | 3941 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); |
4e416953 BS |
3942 | /* |
3943 | * This loop can run a while, specially if mem_cgroup's continuously | |
3944 | * keep exceeding their soft limit and putting the system under | |
3945 | * pressure | |
3946 | */ | |
3947 | do { | |
3948 | if (next_mz) | |
3949 | mz = next_mz; | |
3950 | else | |
3951 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3952 | if (!mz) | |
3953 | break; | |
3954 | ||
0ae5e89c | 3955 | nr_scanned = 0; |
d79154bb | 3956 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, |
5660048c | 3957 | gfp_mask, &nr_scanned); |
4e416953 | 3958 | nr_reclaimed += reclaimed; |
0ae5e89c | 3959 | *total_scanned += nr_scanned; |
4e416953 BS |
3960 | spin_lock(&mctz->lock); |
3961 | ||
3962 | /* | |
3963 | * If we failed to reclaim anything from this memory cgroup | |
3964 | * it is time to move on to the next cgroup | |
3965 | */ | |
3966 | next_mz = NULL; | |
3967 | if (!reclaimed) { | |
3968 | do { | |
3969 | /* | |
3970 | * Loop until we find yet another one. | |
3971 | * | |
3972 | * By the time we get the soft_limit lock | |
3973 | * again, someone might have aded the | |
3974 | * group back on the RB tree. Iterate to | |
3975 | * make sure we get a different mem. | |
3976 | * mem_cgroup_largest_soft_limit_node returns | |
3977 | * NULL if no other cgroup is present on | |
3978 | * the tree | |
3979 | */ | |
3980 | next_mz = | |
3981 | __mem_cgroup_largest_soft_limit_node(mctz); | |
39cc98f1 | 3982 | if (next_mz == mz) |
d79154bb | 3983 | css_put(&next_mz->memcg->css); |
39cc98f1 | 3984 | else /* next_mz == NULL or other memcg */ |
4e416953 BS |
3985 | break; |
3986 | } while (1); | |
3987 | } | |
d79154bb HD |
3988 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); |
3989 | excess = res_counter_soft_limit_excess(&mz->memcg->res); | |
4e416953 BS |
3990 | /* |
3991 | * One school of thought says that we should not add | |
3992 | * back the node to the tree if reclaim returns 0. | |
3993 | * But our reclaim could return 0, simply because due | |
3994 | * to priority we are exposing a smaller subset of | |
3995 | * memory to reclaim from. Consider this as a longer | |
3996 | * term TODO. | |
3997 | */ | |
ef8745c1 | 3998 | /* If excess == 0, no tree ops */ |
d79154bb | 3999 | __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess); |
4e416953 | 4000 | spin_unlock(&mctz->lock); |
d79154bb | 4001 | css_put(&mz->memcg->css); |
4e416953 BS |
4002 | loop++; |
4003 | /* | |
4004 | * Could not reclaim anything and there are no more | |
4005 | * mem cgroups to try or we seem to be looping without | |
4006 | * reclaiming anything. | |
4007 | */ | |
4008 | if (!nr_reclaimed && | |
4009 | (next_mz == NULL || | |
4010 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
4011 | break; | |
4012 | } while (!nr_reclaimed); | |
4013 | if (next_mz) | |
d79154bb | 4014 | css_put(&next_mz->memcg->css); |
4e416953 BS |
4015 | return nr_reclaimed; |
4016 | } | |
4017 | ||
2ef37d3f MH |
4018 | /** |
4019 | * mem_cgroup_force_empty_list - clears LRU of a group | |
4020 | * @memcg: group to clear | |
4021 | * @node: NUMA node | |
4022 | * @zid: zone id | |
4023 | * @lru: lru to to clear | |
4024 | * | |
3c935d18 | 4025 | * Traverse a specified page_cgroup list and try to drop them all. This doesn't |
2ef37d3f MH |
4026 | * reclaim the pages page themselves - pages are moved to the parent (or root) |
4027 | * group. | |
cc847582 | 4028 | */ |
2ef37d3f | 4029 | static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
08e552c6 | 4030 | int node, int zid, enum lru_list lru) |
cc847582 | 4031 | { |
bea8c150 | 4032 | struct lruvec *lruvec; |
2ef37d3f | 4033 | unsigned long flags; |
072c56c1 | 4034 | struct list_head *list; |
925b7673 JW |
4035 | struct page *busy; |
4036 | struct zone *zone; | |
072c56c1 | 4037 | |
08e552c6 | 4038 | zone = &NODE_DATA(node)->node_zones[zid]; |
bea8c150 HD |
4039 | lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
4040 | list = &lruvec->lists[lru]; | |
cc847582 | 4041 | |
f817ed48 | 4042 | busy = NULL; |
2ef37d3f | 4043 | do { |
925b7673 | 4044 | struct page_cgroup *pc; |
5564e88b JW |
4045 | struct page *page; |
4046 | ||
08e552c6 | 4047 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 4048 | if (list_empty(list)) { |
08e552c6 | 4049 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 4050 | break; |
f817ed48 | 4051 | } |
925b7673 JW |
4052 | page = list_entry(list->prev, struct page, lru); |
4053 | if (busy == page) { | |
4054 | list_move(&page->lru, list); | |
648bcc77 | 4055 | busy = NULL; |
08e552c6 | 4056 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
4057 | continue; |
4058 | } | |
08e552c6 | 4059 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 4060 | |
925b7673 | 4061 | pc = lookup_page_cgroup(page); |
5564e88b | 4062 | |
3c935d18 | 4063 | if (mem_cgroup_move_parent(page, pc, memcg)) { |
f817ed48 | 4064 | /* found lock contention or "pc" is obsolete. */ |
925b7673 | 4065 | busy = page; |
f817ed48 KH |
4066 | cond_resched(); |
4067 | } else | |
4068 | busy = NULL; | |
2ef37d3f | 4069 | } while (!list_empty(list)); |
cc847582 KH |
4070 | } |
4071 | ||
4072 | /* | |
c26251f9 MH |
4073 | * make mem_cgroup's charge to be 0 if there is no task by moving |
4074 | * all the charges and pages to the parent. | |
cc847582 | 4075 | * This enables deleting this mem_cgroup. |
c26251f9 MH |
4076 | * |
4077 | * Caller is responsible for holding css reference on the memcg. | |
cc847582 | 4078 | */ |
ab5196c2 | 4079 | static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg) |
cc847582 | 4080 | { |
c26251f9 | 4081 | int node, zid; |
bea207c8 | 4082 | u64 usage; |
f817ed48 | 4083 | |
fce66477 | 4084 | do { |
52d4b9ac KH |
4085 | /* This is for making all *used* pages to be on LRU. */ |
4086 | lru_add_drain_all(); | |
c0ff4b85 | 4087 | drain_all_stock_sync(memcg); |
c0ff4b85 | 4088 | mem_cgroup_start_move(memcg); |
31aaea4a | 4089 | for_each_node_state(node, N_MEMORY) { |
2ef37d3f | 4090 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
f156ab93 HD |
4091 | enum lru_list lru; |
4092 | for_each_lru(lru) { | |
2ef37d3f | 4093 | mem_cgroup_force_empty_list(memcg, |
f156ab93 | 4094 | node, zid, lru); |
f817ed48 | 4095 | } |
1ecaab2b | 4096 | } |
f817ed48 | 4097 | } |
c0ff4b85 R |
4098 | mem_cgroup_end_move(memcg); |
4099 | memcg_oom_recover(memcg); | |
52d4b9ac | 4100 | cond_resched(); |
f817ed48 | 4101 | |
2ef37d3f | 4102 | /* |
bea207c8 GC |
4103 | * Kernel memory may not necessarily be trackable to a specific |
4104 | * process. So they are not migrated, and therefore we can't | |
4105 | * expect their value to drop to 0 here. | |
4106 | * Having res filled up with kmem only is enough. | |
4107 | * | |
2ef37d3f MH |
4108 | * This is a safety check because mem_cgroup_force_empty_list |
4109 | * could have raced with mem_cgroup_replace_page_cache callers | |
4110 | * so the lru seemed empty but the page could have been added | |
4111 | * right after the check. RES_USAGE should be safe as we always | |
4112 | * charge before adding to the LRU. | |
4113 | */ | |
bea207c8 GC |
4114 | usage = res_counter_read_u64(&memcg->res, RES_USAGE) - |
4115 | res_counter_read_u64(&memcg->kmem, RES_USAGE); | |
4116 | } while (usage > 0); | |
c26251f9 MH |
4117 | } |
4118 | ||
4119 | /* | |
4120 | * Reclaims as many pages from the given memcg as possible and moves | |
4121 | * the rest to the parent. | |
4122 | * | |
4123 | * Caller is responsible for holding css reference for memcg. | |
4124 | */ | |
4125 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
4126 | { | |
4127 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
4128 | struct cgroup *cgrp = memcg->css.cgroup; | |
f817ed48 | 4129 | |
c1e862c1 | 4130 | /* returns EBUSY if there is a task or if we come here twice. */ |
c26251f9 MH |
4131 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
4132 | return -EBUSY; | |
4133 | ||
c1e862c1 KH |
4134 | /* we call try-to-free pages for make this cgroup empty */ |
4135 | lru_add_drain_all(); | |
f817ed48 | 4136 | /* try to free all pages in this cgroup */ |
569530fb | 4137 | while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) { |
f817ed48 | 4138 | int progress; |
c1e862c1 | 4139 | |
c26251f9 MH |
4140 | if (signal_pending(current)) |
4141 | return -EINTR; | |
4142 | ||
c0ff4b85 | 4143 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, |
185efc0f | 4144 | false); |
c1e862c1 | 4145 | if (!progress) { |
f817ed48 | 4146 | nr_retries--; |
c1e862c1 | 4147 | /* maybe some writeback is necessary */ |
8aa7e847 | 4148 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 4149 | } |
f817ed48 KH |
4150 | |
4151 | } | |
08e552c6 | 4152 | lru_add_drain(); |
ab5196c2 MH |
4153 | mem_cgroup_reparent_charges(memcg); |
4154 | ||
4155 | return 0; | |
cc847582 KH |
4156 | } |
4157 | ||
6bbda35c | 4158 | static int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
c1e862c1 | 4159 | { |
c26251f9 MH |
4160 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
4161 | int ret; | |
4162 | ||
d8423011 MH |
4163 | if (mem_cgroup_is_root(memcg)) |
4164 | return -EINVAL; | |
c26251f9 MH |
4165 | css_get(&memcg->css); |
4166 | ret = mem_cgroup_force_empty(memcg); | |
4167 | css_put(&memcg->css); | |
4168 | ||
4169 | return ret; | |
c1e862c1 KH |
4170 | } |
4171 | ||
4172 | ||
18f59ea7 BS |
4173 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
4174 | { | |
4175 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
4176 | } | |
4177 | ||
4178 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
4179 | u64 val) | |
4180 | { | |
4181 | int retval = 0; | |
c0ff4b85 | 4182 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
18f59ea7 | 4183 | struct cgroup *parent = cont->parent; |
c0ff4b85 | 4184 | struct mem_cgroup *parent_memcg = NULL; |
18f59ea7 BS |
4185 | |
4186 | if (parent) | |
c0ff4b85 | 4187 | parent_memcg = mem_cgroup_from_cont(parent); |
18f59ea7 BS |
4188 | |
4189 | cgroup_lock(); | |
567fb435 GC |
4190 | |
4191 | if (memcg->use_hierarchy == val) | |
4192 | goto out; | |
4193 | ||
18f59ea7 | 4194 | /* |
af901ca1 | 4195 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
4196 | * in the child subtrees. If it is unset, then the change can |
4197 | * occur, provided the current cgroup has no children. | |
4198 | * | |
4199 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
4200 | * set if there are no children. | |
4201 | */ | |
c0ff4b85 | 4202 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 BS |
4203 | (val == 1 || val == 0)) { |
4204 | if (list_empty(&cont->children)) | |
c0ff4b85 | 4205 | memcg->use_hierarchy = val; |
18f59ea7 BS |
4206 | else |
4207 | retval = -EBUSY; | |
4208 | } else | |
4209 | retval = -EINVAL; | |
567fb435 GC |
4210 | |
4211 | out: | |
18f59ea7 BS |
4212 | cgroup_unlock(); |
4213 | ||
4214 | return retval; | |
4215 | } | |
4216 | ||
0c3e73e8 | 4217 | |
c0ff4b85 | 4218 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg, |
7a159cc9 | 4219 | enum mem_cgroup_stat_index idx) |
0c3e73e8 | 4220 | { |
7d74b06f | 4221 | struct mem_cgroup *iter; |
7a159cc9 | 4222 | long val = 0; |
0c3e73e8 | 4223 | |
7a159cc9 | 4224 | /* Per-cpu values can be negative, use a signed accumulator */ |
c0ff4b85 | 4225 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f KH |
4226 | val += mem_cgroup_read_stat(iter, idx); |
4227 | ||
4228 | if (val < 0) /* race ? */ | |
4229 | val = 0; | |
4230 | return val; | |
0c3e73e8 BS |
4231 | } |
4232 | ||
c0ff4b85 | 4233 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
104f3928 | 4234 | { |
7d74b06f | 4235 | u64 val; |
104f3928 | 4236 | |
c0ff4b85 | 4237 | if (!mem_cgroup_is_root(memcg)) { |
104f3928 | 4238 | if (!swap) |
65c64ce8 | 4239 | return res_counter_read_u64(&memcg->res, RES_USAGE); |
104f3928 | 4240 | else |
65c64ce8 | 4241 | return res_counter_read_u64(&memcg->memsw, RES_USAGE); |
104f3928 KS |
4242 | } |
4243 | ||
c0ff4b85 R |
4244 | val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE); |
4245 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS); | |
104f3928 | 4246 | |
7d74b06f | 4247 | if (swap) |
bff6bb83 | 4248 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP); |
104f3928 KS |
4249 | |
4250 | return val << PAGE_SHIFT; | |
4251 | } | |
4252 | ||
af36f906 TH |
4253 | static ssize_t mem_cgroup_read(struct cgroup *cont, struct cftype *cft, |
4254 | struct file *file, char __user *buf, | |
4255 | size_t nbytes, loff_t *ppos) | |
8cdea7c0 | 4256 | { |
c0ff4b85 | 4257 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
af36f906 | 4258 | char str[64]; |
104f3928 | 4259 | u64 val; |
86ae53e1 GC |
4260 | int name, len; |
4261 | enum res_type type; | |
8c7c6e34 KH |
4262 | |
4263 | type = MEMFILE_TYPE(cft->private); | |
4264 | name = MEMFILE_ATTR(cft->private); | |
af36f906 TH |
4265 | |
4266 | if (!do_swap_account && type == _MEMSWAP) | |
4267 | return -EOPNOTSUPP; | |
4268 | ||
8c7c6e34 KH |
4269 | switch (type) { |
4270 | case _MEM: | |
104f3928 | 4271 | if (name == RES_USAGE) |
c0ff4b85 | 4272 | val = mem_cgroup_usage(memcg, false); |
104f3928 | 4273 | else |
c0ff4b85 | 4274 | val = res_counter_read_u64(&memcg->res, name); |
8c7c6e34 KH |
4275 | break; |
4276 | case _MEMSWAP: | |
104f3928 | 4277 | if (name == RES_USAGE) |
c0ff4b85 | 4278 | val = mem_cgroup_usage(memcg, true); |
104f3928 | 4279 | else |
c0ff4b85 | 4280 | val = res_counter_read_u64(&memcg->memsw, name); |
8c7c6e34 | 4281 | break; |
510fc4e1 GC |
4282 | case _KMEM: |
4283 | val = res_counter_read_u64(&memcg->kmem, name); | |
4284 | break; | |
8c7c6e34 KH |
4285 | default: |
4286 | BUG(); | |
8c7c6e34 | 4287 | } |
af36f906 TH |
4288 | |
4289 | len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val); | |
4290 | return simple_read_from_buffer(buf, nbytes, ppos, str, len); | |
8cdea7c0 | 4291 | } |
510fc4e1 GC |
4292 | |
4293 | static int memcg_update_kmem_limit(struct cgroup *cont, u64 val) | |
4294 | { | |
4295 | int ret = -EINVAL; | |
4296 | #ifdef CONFIG_MEMCG_KMEM | |
a8964b9b GC |
4297 | bool must_inc_static_branch = false; |
4298 | ||
510fc4e1 GC |
4299 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
4300 | /* | |
4301 | * For simplicity, we won't allow this to be disabled. It also can't | |
4302 | * be changed if the cgroup has children already, or if tasks had | |
4303 | * already joined. | |
4304 | * | |
4305 | * If tasks join before we set the limit, a person looking at | |
4306 | * kmem.usage_in_bytes will have no way to determine when it took | |
4307 | * place, which makes the value quite meaningless. | |
4308 | * | |
4309 | * After it first became limited, changes in the value of the limit are | |
4310 | * of course permitted. | |
4311 | * | |
4312 | * Taking the cgroup_lock is really offensive, but it is so far the only | |
4313 | * way to guarantee that no children will appear. There are plenty of | |
4314 | * other offenders, and they should all go away. Fine grained locking | |
4315 | * is probably the way to go here. When we are fully hierarchical, we | |
4316 | * can also get rid of the use_hierarchy check. | |
4317 | */ | |
4318 | cgroup_lock(); | |
4319 | mutex_lock(&set_limit_mutex); | |
4320 | if (!memcg->kmem_account_flags && val != RESOURCE_MAX) { | |
4321 | if (cgroup_task_count(cont) || (memcg->use_hierarchy && | |
4322 | !list_empty(&cont->children))) { | |
4323 | ret = -EBUSY; | |
4324 | goto out; | |
4325 | } | |
4326 | ret = res_counter_set_limit(&memcg->kmem, val); | |
4327 | VM_BUG_ON(ret); | |
4328 | ||
a8964b9b GC |
4329 | /* |
4330 | * After this point, kmem_accounted (that we test atomically in | |
4331 | * the beginning of this conditional), is no longer 0. This | |
4332 | * guarantees only one process will set the following boolean | |
4333 | * to true. We don't need test_and_set because we're protected | |
4334 | * by the set_limit_mutex anyway. | |
4335 | */ | |
4336 | memcg_kmem_set_activated(memcg); | |
4337 | must_inc_static_branch = true; | |
7de37682 GC |
4338 | /* |
4339 | * kmem charges can outlive the cgroup. In the case of slab | |
4340 | * pages, for instance, a page contain objects from various | |
4341 | * processes, so it is unfeasible to migrate them away. We | |
4342 | * need to reference count the memcg because of that. | |
4343 | */ | |
4344 | mem_cgroup_get(memcg); | |
510fc4e1 GC |
4345 | } else |
4346 | ret = res_counter_set_limit(&memcg->kmem, val); | |
4347 | out: | |
4348 | mutex_unlock(&set_limit_mutex); | |
4349 | cgroup_unlock(); | |
a8964b9b GC |
4350 | |
4351 | /* | |
4352 | * We are by now familiar with the fact that we can't inc the static | |
4353 | * branch inside cgroup_lock. See disarm functions for details. A | |
4354 | * worker here is overkill, but also wrong: After the limit is set, we | |
4355 | * must start accounting right away. Since this operation can't fail, | |
4356 | * we can safely defer it to here - no rollback will be needed. | |
4357 | * | |
4358 | * The boolean used to control this is also safe, because | |
4359 | * KMEM_ACCOUNTED_ACTIVATED guarantees that only one process will be | |
4360 | * able to set it to true; | |
4361 | */ | |
4362 | if (must_inc_static_branch) { | |
4363 | static_key_slow_inc(&memcg_kmem_enabled_key); | |
4364 | /* | |
4365 | * setting the active bit after the inc will guarantee no one | |
4366 | * starts accounting before all call sites are patched | |
4367 | */ | |
4368 | memcg_kmem_set_active(memcg); | |
4369 | } | |
4370 | ||
510fc4e1 GC |
4371 | #endif |
4372 | return ret; | |
4373 | } | |
4374 | ||
4375 | static void memcg_propagate_kmem(struct mem_cgroup *memcg) | |
4376 | { | |
4377 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); | |
4378 | if (!parent) | |
4379 | return; | |
4380 | memcg->kmem_account_flags = parent->kmem_account_flags; | |
7de37682 | 4381 | #ifdef CONFIG_MEMCG_KMEM |
a8964b9b GC |
4382 | /* |
4383 | * When that happen, we need to disable the static branch only on those | |
4384 | * memcgs that enabled it. To achieve this, we would be forced to | |
4385 | * complicate the code by keeping track of which memcgs were the ones | |
4386 | * that actually enabled limits, and which ones got it from its | |
4387 | * parents. | |
4388 | * | |
4389 | * It is a lot simpler just to do static_key_slow_inc() on every child | |
4390 | * that is accounted. | |
4391 | */ | |
4392 | if (memcg_kmem_is_active(memcg)) { | |
7de37682 | 4393 | mem_cgroup_get(memcg); |
a8964b9b GC |
4394 | static_key_slow_inc(&memcg_kmem_enabled_key); |
4395 | } | |
7de37682 | 4396 | #endif |
510fc4e1 GC |
4397 | } |
4398 | ||
628f4235 KH |
4399 | /* |
4400 | * The user of this function is... | |
4401 | * RES_LIMIT. | |
4402 | */ | |
856c13aa PM |
4403 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
4404 | const char *buffer) | |
8cdea7c0 | 4405 | { |
628f4235 | 4406 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
86ae53e1 GC |
4407 | enum res_type type; |
4408 | int name; | |
628f4235 KH |
4409 | unsigned long long val; |
4410 | int ret; | |
4411 | ||
8c7c6e34 KH |
4412 | type = MEMFILE_TYPE(cft->private); |
4413 | name = MEMFILE_ATTR(cft->private); | |
af36f906 TH |
4414 | |
4415 | if (!do_swap_account && type == _MEMSWAP) | |
4416 | return -EOPNOTSUPP; | |
4417 | ||
8c7c6e34 | 4418 | switch (name) { |
628f4235 | 4419 | case RES_LIMIT: |
4b3bde4c BS |
4420 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
4421 | ret = -EINVAL; | |
4422 | break; | |
4423 | } | |
628f4235 KH |
4424 | /* This function does all necessary parse...reuse it */ |
4425 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
4426 | if (ret) |
4427 | break; | |
4428 | if (type == _MEM) | |
628f4235 | 4429 | ret = mem_cgroup_resize_limit(memcg, val); |
510fc4e1 | 4430 | else if (type == _MEMSWAP) |
8c7c6e34 | 4431 | ret = mem_cgroup_resize_memsw_limit(memcg, val); |
510fc4e1 GC |
4432 | else if (type == _KMEM) |
4433 | ret = memcg_update_kmem_limit(cont, val); | |
4434 | else | |
4435 | return -EINVAL; | |
628f4235 | 4436 | break; |
296c81d8 BS |
4437 | case RES_SOFT_LIMIT: |
4438 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
4439 | if (ret) | |
4440 | break; | |
4441 | /* | |
4442 | * For memsw, soft limits are hard to implement in terms | |
4443 | * of semantics, for now, we support soft limits for | |
4444 | * control without swap | |
4445 | */ | |
4446 | if (type == _MEM) | |
4447 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
4448 | else | |
4449 | ret = -EINVAL; | |
4450 | break; | |
628f4235 KH |
4451 | default: |
4452 | ret = -EINVAL; /* should be BUG() ? */ | |
4453 | break; | |
4454 | } | |
4455 | return ret; | |
8cdea7c0 BS |
4456 | } |
4457 | ||
fee7b548 KH |
4458 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
4459 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
4460 | { | |
4461 | struct cgroup *cgroup; | |
4462 | unsigned long long min_limit, min_memsw_limit, tmp; | |
4463 | ||
4464 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4465 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
4466 | cgroup = memcg->css.cgroup; | |
4467 | if (!memcg->use_hierarchy) | |
4468 | goto out; | |
4469 | ||
4470 | while (cgroup->parent) { | |
4471 | cgroup = cgroup->parent; | |
4472 | memcg = mem_cgroup_from_cont(cgroup); | |
4473 | if (!memcg->use_hierarchy) | |
4474 | break; | |
4475 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4476 | min_limit = min(min_limit, tmp); | |
4477 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
4478 | min_memsw_limit = min(min_memsw_limit, tmp); | |
4479 | } | |
4480 | out: | |
4481 | *mem_limit = min_limit; | |
4482 | *memsw_limit = min_memsw_limit; | |
fee7b548 KH |
4483 | } |
4484 | ||
29f2a4da | 4485 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 | 4486 | { |
af36f906 | 4487 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
86ae53e1 GC |
4488 | int name; |
4489 | enum res_type type; | |
c84872e1 | 4490 | |
8c7c6e34 KH |
4491 | type = MEMFILE_TYPE(event); |
4492 | name = MEMFILE_ATTR(event); | |
af36f906 TH |
4493 | |
4494 | if (!do_swap_account && type == _MEMSWAP) | |
4495 | return -EOPNOTSUPP; | |
4496 | ||
8c7c6e34 | 4497 | switch (name) { |
29f2a4da | 4498 | case RES_MAX_USAGE: |
8c7c6e34 | 4499 | if (type == _MEM) |
c0ff4b85 | 4500 | res_counter_reset_max(&memcg->res); |
510fc4e1 | 4501 | else if (type == _MEMSWAP) |
c0ff4b85 | 4502 | res_counter_reset_max(&memcg->memsw); |
510fc4e1 GC |
4503 | else if (type == _KMEM) |
4504 | res_counter_reset_max(&memcg->kmem); | |
4505 | else | |
4506 | return -EINVAL; | |
29f2a4da PE |
4507 | break; |
4508 | case RES_FAILCNT: | |
8c7c6e34 | 4509 | if (type == _MEM) |
c0ff4b85 | 4510 | res_counter_reset_failcnt(&memcg->res); |
510fc4e1 | 4511 | else if (type == _MEMSWAP) |
c0ff4b85 | 4512 | res_counter_reset_failcnt(&memcg->memsw); |
510fc4e1 GC |
4513 | else if (type == _KMEM) |
4514 | res_counter_reset_failcnt(&memcg->kmem); | |
4515 | else | |
4516 | return -EINVAL; | |
29f2a4da PE |
4517 | break; |
4518 | } | |
f64c3f54 | 4519 | |
85cc59db | 4520 | return 0; |
c84872e1 PE |
4521 | } |
4522 | ||
7dc74be0 DN |
4523 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
4524 | struct cftype *cft) | |
4525 | { | |
4526 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; | |
4527 | } | |
4528 | ||
02491447 | 4529 | #ifdef CONFIG_MMU |
7dc74be0 DN |
4530 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
4531 | struct cftype *cft, u64 val) | |
4532 | { | |
c0ff4b85 | 4533 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
7dc74be0 DN |
4534 | |
4535 | if (val >= (1 << NR_MOVE_TYPE)) | |
4536 | return -EINVAL; | |
4537 | /* | |
4538 | * We check this value several times in both in can_attach() and | |
4539 | * attach(), so we need cgroup lock to prevent this value from being | |
4540 | * inconsistent. | |
4541 | */ | |
4542 | cgroup_lock(); | |
c0ff4b85 | 4543 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
4544 | cgroup_unlock(); |
4545 | ||
4546 | return 0; | |
4547 | } | |
02491447 DN |
4548 | #else |
4549 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, | |
4550 | struct cftype *cft, u64 val) | |
4551 | { | |
4552 | return -ENOSYS; | |
4553 | } | |
4554 | #endif | |
7dc74be0 | 4555 | |
406eb0c9 | 4556 | #ifdef CONFIG_NUMA |
ab215884 | 4557 | static int memcg_numa_stat_show(struct cgroup *cont, struct cftype *cft, |
fada52ca | 4558 | struct seq_file *m) |
406eb0c9 YH |
4559 | { |
4560 | int nid; | |
4561 | unsigned long total_nr, file_nr, anon_nr, unevictable_nr; | |
4562 | unsigned long node_nr; | |
d79154bb | 4563 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
406eb0c9 | 4564 | |
d79154bb | 4565 | total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL); |
406eb0c9 | 4566 | seq_printf(m, "total=%lu", total_nr); |
31aaea4a | 4567 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 4568 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL); |
406eb0c9 YH |
4569 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4570 | } | |
4571 | seq_putc(m, '\n'); | |
4572 | ||
d79154bb | 4573 | file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE); |
406eb0c9 | 4574 | seq_printf(m, "file=%lu", file_nr); |
31aaea4a | 4575 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 4576 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 4577 | LRU_ALL_FILE); |
406eb0c9 YH |
4578 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4579 | } | |
4580 | seq_putc(m, '\n'); | |
4581 | ||
d79154bb | 4582 | anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON); |
406eb0c9 | 4583 | seq_printf(m, "anon=%lu", anon_nr); |
31aaea4a | 4584 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 4585 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 4586 | LRU_ALL_ANON); |
406eb0c9 YH |
4587 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4588 | } | |
4589 | seq_putc(m, '\n'); | |
4590 | ||
d79154bb | 4591 | unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE)); |
406eb0c9 | 4592 | seq_printf(m, "unevictable=%lu", unevictable_nr); |
31aaea4a | 4593 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 4594 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 4595 | BIT(LRU_UNEVICTABLE)); |
406eb0c9 YH |
4596 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4597 | } | |
4598 | seq_putc(m, '\n'); | |
4599 | return 0; | |
4600 | } | |
4601 | #endif /* CONFIG_NUMA */ | |
4602 | ||
af7c4b0e JW |
4603 | static const char * const mem_cgroup_lru_names[] = { |
4604 | "inactive_anon", | |
4605 | "active_anon", | |
4606 | "inactive_file", | |
4607 | "active_file", | |
4608 | "unevictable", | |
4609 | }; | |
4610 | ||
4611 | static inline void mem_cgroup_lru_names_not_uptodate(void) | |
4612 | { | |
4613 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); | |
4614 | } | |
4615 | ||
ab215884 | 4616 | static int memcg_stat_show(struct cgroup *cont, struct cftype *cft, |
78ccf5b5 | 4617 | struct seq_file *m) |
d2ceb9b7 | 4618 | { |
d79154bb | 4619 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
af7c4b0e JW |
4620 | struct mem_cgroup *mi; |
4621 | unsigned int i; | |
406eb0c9 | 4622 | |
af7c4b0e | 4623 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 4624 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 4625 | continue; |
af7c4b0e JW |
4626 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
4627 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 4628 | } |
7b854121 | 4629 | |
af7c4b0e JW |
4630 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
4631 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
4632 | mem_cgroup_read_events(memcg, i)); | |
4633 | ||
4634 | for (i = 0; i < NR_LRU_LISTS; i++) | |
4635 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
4636 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
4637 | ||
14067bb3 | 4638 | /* Hierarchical information */ |
fee7b548 KH |
4639 | { |
4640 | unsigned long long limit, memsw_limit; | |
d79154bb | 4641 | memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit); |
78ccf5b5 | 4642 | seq_printf(m, "hierarchical_memory_limit %llu\n", limit); |
fee7b548 | 4643 | if (do_swap_account) |
78ccf5b5 JW |
4644 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
4645 | memsw_limit); | |
fee7b548 | 4646 | } |
7f016ee8 | 4647 | |
af7c4b0e JW |
4648 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
4649 | long long val = 0; | |
4650 | ||
bff6bb83 | 4651 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 4652 | continue; |
af7c4b0e JW |
4653 | for_each_mem_cgroup_tree(mi, memcg) |
4654 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
4655 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
4656 | } | |
4657 | ||
4658 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
4659 | unsigned long long val = 0; | |
4660 | ||
4661 | for_each_mem_cgroup_tree(mi, memcg) | |
4662 | val += mem_cgroup_read_events(mi, i); | |
4663 | seq_printf(m, "total_%s %llu\n", | |
4664 | mem_cgroup_events_names[i], val); | |
4665 | } | |
4666 | ||
4667 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
4668 | unsigned long long val = 0; | |
4669 | ||
4670 | for_each_mem_cgroup_tree(mi, memcg) | |
4671 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
4672 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 4673 | } |
14067bb3 | 4674 | |
7f016ee8 | 4675 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
4676 | { |
4677 | int nid, zid; | |
4678 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 4679 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
4680 | unsigned long recent_rotated[2] = {0, 0}; |
4681 | unsigned long recent_scanned[2] = {0, 0}; | |
4682 | ||
4683 | for_each_online_node(nid) | |
4684 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
d79154bb | 4685 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
89abfab1 | 4686 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 4687 | |
89abfab1 HD |
4688 | recent_rotated[0] += rstat->recent_rotated[0]; |
4689 | recent_rotated[1] += rstat->recent_rotated[1]; | |
4690 | recent_scanned[0] += rstat->recent_scanned[0]; | |
4691 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 4692 | } |
78ccf5b5 JW |
4693 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
4694 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
4695 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
4696 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
4697 | } |
4698 | #endif | |
4699 | ||
d2ceb9b7 KH |
4700 | return 0; |
4701 | } | |
4702 | ||
a7885eb8 KM |
4703 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
4704 | { | |
4705 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4706 | ||
1f4c025b | 4707 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4708 | } |
4709 | ||
4710 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
4711 | u64 val) | |
4712 | { | |
4713 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4714 | struct mem_cgroup *parent; | |
068b38c1 | 4715 | |
a7885eb8 KM |
4716 | if (val > 100) |
4717 | return -EINVAL; | |
4718 | ||
4719 | if (cgrp->parent == NULL) | |
4720 | return -EINVAL; | |
4721 | ||
4722 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 LZ |
4723 | |
4724 | cgroup_lock(); | |
4725 | ||
a7885eb8 KM |
4726 | /* If under hierarchy, only empty-root can set this value */ |
4727 | if ((parent->use_hierarchy) || | |
068b38c1 LZ |
4728 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
4729 | cgroup_unlock(); | |
a7885eb8 | 4730 | return -EINVAL; |
068b38c1 | 4731 | } |
a7885eb8 | 4732 | |
a7885eb8 | 4733 | memcg->swappiness = val; |
a7885eb8 | 4734 | |
068b38c1 LZ |
4735 | cgroup_unlock(); |
4736 | ||
a7885eb8 KM |
4737 | return 0; |
4738 | } | |
4739 | ||
2e72b634 KS |
4740 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4741 | { | |
4742 | struct mem_cgroup_threshold_ary *t; | |
4743 | u64 usage; | |
4744 | int i; | |
4745 | ||
4746 | rcu_read_lock(); | |
4747 | if (!swap) | |
2c488db2 | 4748 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4749 | else |
2c488db2 | 4750 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4751 | |
4752 | if (!t) | |
4753 | goto unlock; | |
4754 | ||
4755 | usage = mem_cgroup_usage(memcg, swap); | |
4756 | ||
4757 | /* | |
748dad36 | 4758 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4759 | * If it's not true, a threshold was crossed after last |
4760 | * call of __mem_cgroup_threshold(). | |
4761 | */ | |
5407a562 | 4762 | i = t->current_threshold; |
2e72b634 KS |
4763 | |
4764 | /* | |
4765 | * Iterate backward over array of thresholds starting from | |
4766 | * current_threshold and check if a threshold is crossed. | |
4767 | * If none of thresholds below usage is crossed, we read | |
4768 | * only one element of the array here. | |
4769 | */ | |
4770 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4771 | eventfd_signal(t->entries[i].eventfd, 1); | |
4772 | ||
4773 | /* i = current_threshold + 1 */ | |
4774 | i++; | |
4775 | ||
4776 | /* | |
4777 | * Iterate forward over array of thresholds starting from | |
4778 | * current_threshold+1 and check if a threshold is crossed. | |
4779 | * If none of thresholds above usage is crossed, we read | |
4780 | * only one element of the array here. | |
4781 | */ | |
4782 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4783 | eventfd_signal(t->entries[i].eventfd, 1); | |
4784 | ||
4785 | /* Update current_threshold */ | |
5407a562 | 4786 | t->current_threshold = i - 1; |
2e72b634 KS |
4787 | unlock: |
4788 | rcu_read_unlock(); | |
4789 | } | |
4790 | ||
4791 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4792 | { | |
ad4ca5f4 KS |
4793 | while (memcg) { |
4794 | __mem_cgroup_threshold(memcg, false); | |
4795 | if (do_swap_account) | |
4796 | __mem_cgroup_threshold(memcg, true); | |
4797 | ||
4798 | memcg = parent_mem_cgroup(memcg); | |
4799 | } | |
2e72b634 KS |
4800 | } |
4801 | ||
4802 | static int compare_thresholds(const void *a, const void *b) | |
4803 | { | |
4804 | const struct mem_cgroup_threshold *_a = a; | |
4805 | const struct mem_cgroup_threshold *_b = b; | |
4806 | ||
4807 | return _a->threshold - _b->threshold; | |
4808 | } | |
4809 | ||
c0ff4b85 | 4810 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4811 | { |
4812 | struct mem_cgroup_eventfd_list *ev; | |
4813 | ||
c0ff4b85 | 4814 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 KH |
4815 | eventfd_signal(ev->eventfd, 1); |
4816 | return 0; | |
4817 | } | |
4818 | ||
c0ff4b85 | 4819 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4820 | { |
7d74b06f KH |
4821 | struct mem_cgroup *iter; |
4822 | ||
c0ff4b85 | 4823 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4824 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4825 | } |
4826 | ||
4827 | static int mem_cgroup_usage_register_event(struct cgroup *cgrp, | |
4828 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
2e72b634 KS |
4829 | { |
4830 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4831 | struct mem_cgroup_thresholds *thresholds; |
4832 | struct mem_cgroup_threshold_ary *new; | |
86ae53e1 | 4833 | enum res_type type = MEMFILE_TYPE(cft->private); |
2e72b634 | 4834 | u64 threshold, usage; |
2c488db2 | 4835 | int i, size, ret; |
2e72b634 KS |
4836 | |
4837 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
4838 | if (ret) | |
4839 | return ret; | |
4840 | ||
4841 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4842 | |
2e72b634 | 4843 | if (type == _MEM) |
2c488db2 | 4844 | thresholds = &memcg->thresholds; |
2e72b634 | 4845 | else if (type == _MEMSWAP) |
2c488db2 | 4846 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4847 | else |
4848 | BUG(); | |
4849 | ||
4850 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
4851 | ||
4852 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 4853 | if (thresholds->primary) |
2e72b634 KS |
4854 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4855 | ||
2c488db2 | 4856 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4857 | |
4858 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 4859 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 4860 | GFP_KERNEL); |
2c488db2 | 4861 | if (!new) { |
2e72b634 KS |
4862 | ret = -ENOMEM; |
4863 | goto unlock; | |
4864 | } | |
2c488db2 | 4865 | new->size = size; |
2e72b634 KS |
4866 | |
4867 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4868 | if (thresholds->primary) { |
4869 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4870 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4871 | } |
4872 | ||
2e72b634 | 4873 | /* Add new threshold */ |
2c488db2 KS |
4874 | new->entries[size - 1].eventfd = eventfd; |
4875 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4876 | |
4877 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4878 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4879 | compare_thresholds, NULL); |
4880 | ||
4881 | /* Find current threshold */ | |
2c488db2 | 4882 | new->current_threshold = -1; |
2e72b634 | 4883 | for (i = 0; i < size; i++) { |
748dad36 | 4884 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4885 | /* |
2c488db2 KS |
4886 | * new->current_threshold will not be used until |
4887 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4888 | * it here. |
4889 | */ | |
2c488db2 | 4890 | ++new->current_threshold; |
748dad36 SZ |
4891 | } else |
4892 | break; | |
2e72b634 KS |
4893 | } |
4894 | ||
2c488db2 KS |
4895 | /* Free old spare buffer and save old primary buffer as spare */ |
4896 | kfree(thresholds->spare); | |
4897 | thresholds->spare = thresholds->primary; | |
4898 | ||
4899 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4900 | |
907860ed | 4901 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4902 | synchronize_rcu(); |
4903 | ||
2e72b634 KS |
4904 | unlock: |
4905 | mutex_unlock(&memcg->thresholds_lock); | |
4906 | ||
4907 | return ret; | |
4908 | } | |
4909 | ||
907860ed | 4910 | static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp, |
9490ff27 | 4911 | struct cftype *cft, struct eventfd_ctx *eventfd) |
2e72b634 KS |
4912 | { |
4913 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4914 | struct mem_cgroup_thresholds *thresholds; |
4915 | struct mem_cgroup_threshold_ary *new; | |
86ae53e1 | 4916 | enum res_type type = MEMFILE_TYPE(cft->private); |
2e72b634 | 4917 | u64 usage; |
2c488db2 | 4918 | int i, j, size; |
2e72b634 KS |
4919 | |
4920 | mutex_lock(&memcg->thresholds_lock); | |
4921 | if (type == _MEM) | |
2c488db2 | 4922 | thresholds = &memcg->thresholds; |
2e72b634 | 4923 | else if (type == _MEMSWAP) |
2c488db2 | 4924 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4925 | else |
4926 | BUG(); | |
4927 | ||
371528ca AV |
4928 | if (!thresholds->primary) |
4929 | goto unlock; | |
4930 | ||
2e72b634 KS |
4931 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); |
4932 | ||
4933 | /* Check if a threshold crossed before removing */ | |
4934 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4935 | ||
4936 | /* Calculate new number of threshold */ | |
2c488db2 KS |
4937 | size = 0; |
4938 | for (i = 0; i < thresholds->primary->size; i++) { | |
4939 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
4940 | size++; |
4941 | } | |
4942 | ||
2c488db2 | 4943 | new = thresholds->spare; |
907860ed | 4944 | |
2e72b634 KS |
4945 | /* Set thresholds array to NULL if we don't have thresholds */ |
4946 | if (!size) { | |
2c488db2 KS |
4947 | kfree(new); |
4948 | new = NULL; | |
907860ed | 4949 | goto swap_buffers; |
2e72b634 KS |
4950 | } |
4951 | ||
2c488db2 | 4952 | new->size = size; |
2e72b634 KS |
4953 | |
4954 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4955 | new->current_threshold = -1; |
4956 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4957 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4958 | continue; |
4959 | ||
2c488db2 | 4960 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4961 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4962 | /* |
2c488db2 | 4963 | * new->current_threshold will not be used |
2e72b634 KS |
4964 | * until rcu_assign_pointer(), so it's safe to increment |
4965 | * it here. | |
4966 | */ | |
2c488db2 | 4967 | ++new->current_threshold; |
2e72b634 KS |
4968 | } |
4969 | j++; | |
4970 | } | |
4971 | ||
907860ed | 4972 | swap_buffers: |
2c488db2 KS |
4973 | /* Swap primary and spare array */ |
4974 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
4975 | /* If all events are unregistered, free the spare array */ |
4976 | if (!new) { | |
4977 | kfree(thresholds->spare); | |
4978 | thresholds->spare = NULL; | |
4979 | } | |
4980 | ||
2c488db2 | 4981 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4982 | |
907860ed | 4983 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4984 | synchronize_rcu(); |
371528ca | 4985 | unlock: |
2e72b634 | 4986 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4987 | } |
c1e862c1 | 4988 | |
9490ff27 KH |
4989 | static int mem_cgroup_oom_register_event(struct cgroup *cgrp, |
4990 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
4991 | { | |
4992 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4993 | struct mem_cgroup_eventfd_list *event; | |
86ae53e1 | 4994 | enum res_type type = MEMFILE_TYPE(cft->private); |
9490ff27 KH |
4995 | |
4996 | BUG_ON(type != _OOM_TYPE); | |
4997 | event = kmalloc(sizeof(*event), GFP_KERNEL); | |
4998 | if (!event) | |
4999 | return -ENOMEM; | |
5000 | ||
1af8efe9 | 5001 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
5002 | |
5003 | event->eventfd = eventfd; | |
5004 | list_add(&event->list, &memcg->oom_notify); | |
5005 | ||
5006 | /* already in OOM ? */ | |
79dfdacc | 5007 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 5008 | eventfd_signal(eventfd, 1); |
1af8efe9 | 5009 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
5010 | |
5011 | return 0; | |
5012 | } | |
5013 | ||
907860ed | 5014 | static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp, |
9490ff27 KH |
5015 | struct cftype *cft, struct eventfd_ctx *eventfd) |
5016 | { | |
c0ff4b85 | 5017 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
9490ff27 | 5018 | struct mem_cgroup_eventfd_list *ev, *tmp; |
86ae53e1 | 5019 | enum res_type type = MEMFILE_TYPE(cft->private); |
9490ff27 KH |
5020 | |
5021 | BUG_ON(type != _OOM_TYPE); | |
5022 | ||
1af8efe9 | 5023 | spin_lock(&memcg_oom_lock); |
9490ff27 | 5024 | |
c0ff4b85 | 5025 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
5026 | if (ev->eventfd == eventfd) { |
5027 | list_del(&ev->list); | |
5028 | kfree(ev); | |
5029 | } | |
5030 | } | |
5031 | ||
1af8efe9 | 5032 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
5033 | } |
5034 | ||
3c11ecf4 KH |
5035 | static int mem_cgroup_oom_control_read(struct cgroup *cgrp, |
5036 | struct cftype *cft, struct cgroup_map_cb *cb) | |
5037 | { | |
c0ff4b85 | 5038 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 | 5039 | |
c0ff4b85 | 5040 | cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable); |
3c11ecf4 | 5041 | |
c0ff4b85 | 5042 | if (atomic_read(&memcg->under_oom)) |
3c11ecf4 KH |
5043 | cb->fill(cb, "under_oom", 1); |
5044 | else | |
5045 | cb->fill(cb, "under_oom", 0); | |
5046 | return 0; | |
5047 | } | |
5048 | ||
3c11ecf4 KH |
5049 | static int mem_cgroup_oom_control_write(struct cgroup *cgrp, |
5050 | struct cftype *cft, u64 val) | |
5051 | { | |
c0ff4b85 | 5052 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 KH |
5053 | struct mem_cgroup *parent; |
5054 | ||
5055 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
5056 | if (!cgrp->parent || !((val == 0) || (val == 1))) | |
5057 | return -EINVAL; | |
5058 | ||
5059 | parent = mem_cgroup_from_cont(cgrp->parent); | |
5060 | ||
5061 | cgroup_lock(); | |
5062 | /* oom-kill-disable is a flag for subhierarchy. */ | |
5063 | if ((parent->use_hierarchy) || | |
c0ff4b85 | 5064 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
3c11ecf4 KH |
5065 | cgroup_unlock(); |
5066 | return -EINVAL; | |
5067 | } | |
c0ff4b85 | 5068 | memcg->oom_kill_disable = val; |
4d845ebf | 5069 | if (!val) |
c0ff4b85 | 5070 | memcg_oom_recover(memcg); |
3c11ecf4 KH |
5071 | cgroup_unlock(); |
5072 | return 0; | |
5073 | } | |
5074 | ||
c255a458 | 5075 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 5076 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 5077 | { |
2633d7a0 | 5078 | memcg->kmemcg_id = -1; |
510fc4e1 | 5079 | memcg_propagate_kmem(memcg); |
2633d7a0 | 5080 | |
1d62e436 | 5081 | return mem_cgroup_sockets_init(memcg, ss); |
e5671dfa GC |
5082 | }; |
5083 | ||
1d62e436 | 5084 | static void kmem_cgroup_destroy(struct mem_cgroup *memcg) |
d1a4c0b3 | 5085 | { |
1d62e436 | 5086 | mem_cgroup_sockets_destroy(memcg); |
7de37682 GC |
5087 | |
5088 | memcg_kmem_mark_dead(memcg); | |
5089 | ||
5090 | if (res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0) | |
5091 | return; | |
5092 | ||
5093 | /* | |
5094 | * Charges already down to 0, undo mem_cgroup_get() done in the charge | |
5095 | * path here, being careful not to race with memcg_uncharge_kmem: it is | |
5096 | * possible that the charges went down to 0 between mark_dead and the | |
5097 | * res_counter read, so in that case, we don't need the put | |
5098 | */ | |
5099 | if (memcg_kmem_test_and_clear_dead(memcg)) | |
5100 | mem_cgroup_put(memcg); | |
d1a4c0b3 | 5101 | } |
e5671dfa | 5102 | #else |
cbe128e3 | 5103 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
5104 | { |
5105 | return 0; | |
5106 | } | |
d1a4c0b3 | 5107 | |
1d62e436 | 5108 | static void kmem_cgroup_destroy(struct mem_cgroup *memcg) |
d1a4c0b3 GC |
5109 | { |
5110 | } | |
e5671dfa GC |
5111 | #endif |
5112 | ||
8cdea7c0 BS |
5113 | static struct cftype mem_cgroup_files[] = { |
5114 | { | |
0eea1030 | 5115 | .name = "usage_in_bytes", |
8c7c6e34 | 5116 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
af36f906 | 5117 | .read = mem_cgroup_read, |
9490ff27 KH |
5118 | .register_event = mem_cgroup_usage_register_event, |
5119 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8cdea7c0 | 5120 | }, |
c84872e1 PE |
5121 | { |
5122 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 5123 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 5124 | .trigger = mem_cgroup_reset, |
af36f906 | 5125 | .read = mem_cgroup_read, |
c84872e1 | 5126 | }, |
8cdea7c0 | 5127 | { |
0eea1030 | 5128 | .name = "limit_in_bytes", |
8c7c6e34 | 5129 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 5130 | .write_string = mem_cgroup_write, |
af36f906 | 5131 | .read = mem_cgroup_read, |
8cdea7c0 | 5132 | }, |
296c81d8 BS |
5133 | { |
5134 | .name = "soft_limit_in_bytes", | |
5135 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
5136 | .write_string = mem_cgroup_write, | |
af36f906 | 5137 | .read = mem_cgroup_read, |
296c81d8 | 5138 | }, |
8cdea7c0 BS |
5139 | { |
5140 | .name = "failcnt", | |
8c7c6e34 | 5141 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 5142 | .trigger = mem_cgroup_reset, |
af36f906 | 5143 | .read = mem_cgroup_read, |
8cdea7c0 | 5144 | }, |
d2ceb9b7 KH |
5145 | { |
5146 | .name = "stat", | |
ab215884 | 5147 | .read_seq_string = memcg_stat_show, |
d2ceb9b7 | 5148 | }, |
c1e862c1 KH |
5149 | { |
5150 | .name = "force_empty", | |
5151 | .trigger = mem_cgroup_force_empty_write, | |
5152 | }, | |
18f59ea7 BS |
5153 | { |
5154 | .name = "use_hierarchy", | |
5155 | .write_u64 = mem_cgroup_hierarchy_write, | |
5156 | .read_u64 = mem_cgroup_hierarchy_read, | |
5157 | }, | |
a7885eb8 KM |
5158 | { |
5159 | .name = "swappiness", | |
5160 | .read_u64 = mem_cgroup_swappiness_read, | |
5161 | .write_u64 = mem_cgroup_swappiness_write, | |
5162 | }, | |
7dc74be0 DN |
5163 | { |
5164 | .name = "move_charge_at_immigrate", | |
5165 | .read_u64 = mem_cgroup_move_charge_read, | |
5166 | .write_u64 = mem_cgroup_move_charge_write, | |
5167 | }, | |
9490ff27 KH |
5168 | { |
5169 | .name = "oom_control", | |
3c11ecf4 KH |
5170 | .read_map = mem_cgroup_oom_control_read, |
5171 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
5172 | .register_event = mem_cgroup_oom_register_event, |
5173 | .unregister_event = mem_cgroup_oom_unregister_event, | |
5174 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), | |
5175 | }, | |
406eb0c9 YH |
5176 | #ifdef CONFIG_NUMA |
5177 | { | |
5178 | .name = "numa_stat", | |
ab215884 | 5179 | .read_seq_string = memcg_numa_stat_show, |
406eb0c9 YH |
5180 | }, |
5181 | #endif | |
c255a458 | 5182 | #ifdef CONFIG_MEMCG_SWAP |
8c7c6e34 KH |
5183 | { |
5184 | .name = "memsw.usage_in_bytes", | |
5185 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
af36f906 | 5186 | .read = mem_cgroup_read, |
9490ff27 KH |
5187 | .register_event = mem_cgroup_usage_register_event, |
5188 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8c7c6e34 KH |
5189 | }, |
5190 | { | |
5191 | .name = "memsw.max_usage_in_bytes", | |
5192 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
5193 | .trigger = mem_cgroup_reset, | |
af36f906 | 5194 | .read = mem_cgroup_read, |
8c7c6e34 KH |
5195 | }, |
5196 | { | |
5197 | .name = "memsw.limit_in_bytes", | |
5198 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
5199 | .write_string = mem_cgroup_write, | |
af36f906 | 5200 | .read = mem_cgroup_read, |
8c7c6e34 KH |
5201 | }, |
5202 | { | |
5203 | .name = "memsw.failcnt", | |
5204 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
5205 | .trigger = mem_cgroup_reset, | |
af36f906 | 5206 | .read = mem_cgroup_read, |
8c7c6e34 | 5207 | }, |
510fc4e1 GC |
5208 | #endif |
5209 | #ifdef CONFIG_MEMCG_KMEM | |
5210 | { | |
5211 | .name = "kmem.limit_in_bytes", | |
5212 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
5213 | .write_string = mem_cgroup_write, | |
5214 | .read = mem_cgroup_read, | |
5215 | }, | |
5216 | { | |
5217 | .name = "kmem.usage_in_bytes", | |
5218 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
5219 | .read = mem_cgroup_read, | |
5220 | }, | |
5221 | { | |
5222 | .name = "kmem.failcnt", | |
5223 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
5224 | .trigger = mem_cgroup_reset, | |
5225 | .read = mem_cgroup_read, | |
5226 | }, | |
5227 | { | |
5228 | .name = "kmem.max_usage_in_bytes", | |
5229 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
5230 | .trigger = mem_cgroup_reset, | |
5231 | .read = mem_cgroup_read, | |
5232 | }, | |
8c7c6e34 | 5233 | #endif |
6bc10349 | 5234 | { }, /* terminate */ |
af36f906 | 5235 | }; |
8c7c6e34 | 5236 | |
c0ff4b85 | 5237 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
5238 | { |
5239 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 5240 | struct mem_cgroup_per_zone *mz; |
41e3355d | 5241 | int zone, tmp = node; |
1ecaab2b KH |
5242 | /* |
5243 | * This routine is called against possible nodes. | |
5244 | * But it's BUG to call kmalloc() against offline node. | |
5245 | * | |
5246 | * TODO: this routine can waste much memory for nodes which will | |
5247 | * never be onlined. It's better to use memory hotplug callback | |
5248 | * function. | |
5249 | */ | |
41e3355d KH |
5250 | if (!node_state(node, N_NORMAL_MEMORY)) |
5251 | tmp = -1; | |
17295c88 | 5252 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
5253 | if (!pn) |
5254 | return 1; | |
1ecaab2b | 5255 | |
1ecaab2b KH |
5256 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
5257 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 5258 | lruvec_init(&mz->lruvec); |
f64c3f54 | 5259 | mz->usage_in_excess = 0; |
4e416953 | 5260 | mz->on_tree = false; |
d79154bb | 5261 | mz->memcg = memcg; |
1ecaab2b | 5262 | } |
0a619e58 | 5263 | memcg->info.nodeinfo[node] = pn; |
6d12e2d8 KH |
5264 | return 0; |
5265 | } | |
5266 | ||
c0ff4b85 | 5267 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 5268 | { |
c0ff4b85 | 5269 | kfree(memcg->info.nodeinfo[node]); |
1ecaab2b KH |
5270 | } |
5271 | ||
33327948 KH |
5272 | static struct mem_cgroup *mem_cgroup_alloc(void) |
5273 | { | |
d79154bb | 5274 | struct mem_cgroup *memcg; |
c62b1a3b | 5275 | int size = sizeof(struct mem_cgroup); |
33327948 | 5276 | |
c62b1a3b | 5277 | /* Can be very big if MAX_NUMNODES is very big */ |
c8dad2bb | 5278 | if (size < PAGE_SIZE) |
d79154bb | 5279 | memcg = kzalloc(size, GFP_KERNEL); |
33327948 | 5280 | else |
d79154bb | 5281 | memcg = vzalloc(size); |
33327948 | 5282 | |
d79154bb | 5283 | if (!memcg) |
e7bbcdf3 DC |
5284 | return NULL; |
5285 | ||
d79154bb HD |
5286 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
5287 | if (!memcg->stat) | |
d2e61b8d | 5288 | goto out_free; |
d79154bb HD |
5289 | spin_lock_init(&memcg->pcp_counter_lock); |
5290 | return memcg; | |
d2e61b8d DC |
5291 | |
5292 | out_free: | |
5293 | if (size < PAGE_SIZE) | |
d79154bb | 5294 | kfree(memcg); |
d2e61b8d | 5295 | else |
d79154bb | 5296 | vfree(memcg); |
d2e61b8d | 5297 | return NULL; |
33327948 KH |
5298 | } |
5299 | ||
59927fb9 | 5300 | /* |
c8b2a36f GC |
5301 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
5302 | * (scanning all at force_empty is too costly...) | |
5303 | * | |
5304 | * Instead of clearing all references at force_empty, we remember | |
5305 | * the number of reference from swap_cgroup and free mem_cgroup when | |
5306 | * it goes down to 0. | |
5307 | * | |
5308 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 5309 | */ |
c8b2a36f GC |
5310 | |
5311 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 5312 | { |
c8b2a36f | 5313 | int node; |
3afe36b1 | 5314 | int size = sizeof(struct mem_cgroup); |
59927fb9 | 5315 | |
c8b2a36f GC |
5316 | mem_cgroup_remove_from_trees(memcg); |
5317 | free_css_id(&mem_cgroup_subsys, &memcg->css); | |
5318 | ||
5319 | for_each_node(node) | |
5320 | free_mem_cgroup_per_zone_info(memcg, node); | |
5321 | ||
5322 | free_percpu(memcg->stat); | |
5323 | ||
3f134619 GC |
5324 | /* |
5325 | * We need to make sure that (at least for now), the jump label | |
5326 | * destruction code runs outside of the cgroup lock. This is because | |
5327 | * get_online_cpus(), which is called from the static_branch update, | |
5328 | * can't be called inside the cgroup_lock. cpusets are the ones | |
5329 | * enforcing this dependency, so if they ever change, we might as well. | |
5330 | * | |
5331 | * schedule_work() will guarantee this happens. Be careful if you need | |
5332 | * to move this code around, and make sure it is outside | |
5333 | * the cgroup_lock. | |
5334 | */ | |
a8964b9b | 5335 | disarm_static_keys(memcg); |
3afe36b1 GC |
5336 | if (size < PAGE_SIZE) |
5337 | kfree(memcg); | |
5338 | else | |
5339 | vfree(memcg); | |
59927fb9 | 5340 | } |
3afe36b1 | 5341 | |
59927fb9 | 5342 | |
8c7c6e34 | 5343 | /* |
c8b2a36f GC |
5344 | * Helpers for freeing a kmalloc()ed/vzalloc()ed mem_cgroup by RCU, |
5345 | * but in process context. The work_freeing structure is overlaid | |
5346 | * on the rcu_freeing structure, which itself is overlaid on memsw. | |
8c7c6e34 | 5347 | */ |
c8b2a36f | 5348 | static void free_work(struct work_struct *work) |
33327948 | 5349 | { |
c8b2a36f | 5350 | struct mem_cgroup *memcg; |
08e552c6 | 5351 | |
c8b2a36f GC |
5352 | memcg = container_of(work, struct mem_cgroup, work_freeing); |
5353 | __mem_cgroup_free(memcg); | |
5354 | } | |
04046e1a | 5355 | |
c8b2a36f GC |
5356 | static void free_rcu(struct rcu_head *rcu_head) |
5357 | { | |
5358 | struct mem_cgroup *memcg; | |
08e552c6 | 5359 | |
c8b2a36f GC |
5360 | memcg = container_of(rcu_head, struct mem_cgroup, rcu_freeing); |
5361 | INIT_WORK(&memcg->work_freeing, free_work); | |
5362 | schedule_work(&memcg->work_freeing); | |
33327948 KH |
5363 | } |
5364 | ||
c0ff4b85 | 5365 | static void mem_cgroup_get(struct mem_cgroup *memcg) |
8c7c6e34 | 5366 | { |
c0ff4b85 | 5367 | atomic_inc(&memcg->refcnt); |
8c7c6e34 KH |
5368 | } |
5369 | ||
c0ff4b85 | 5370 | static void __mem_cgroup_put(struct mem_cgroup *memcg, int count) |
8c7c6e34 | 5371 | { |
c0ff4b85 R |
5372 | if (atomic_sub_and_test(count, &memcg->refcnt)) { |
5373 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); | |
c8b2a36f | 5374 | call_rcu(&memcg->rcu_freeing, free_rcu); |
7bcc1bb1 DN |
5375 | if (parent) |
5376 | mem_cgroup_put(parent); | |
5377 | } | |
8c7c6e34 KH |
5378 | } |
5379 | ||
c0ff4b85 | 5380 | static void mem_cgroup_put(struct mem_cgroup *memcg) |
483c30b5 | 5381 | { |
c0ff4b85 | 5382 | __mem_cgroup_put(memcg, 1); |
483c30b5 DN |
5383 | } |
5384 | ||
7bcc1bb1 DN |
5385 | /* |
5386 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
5387 | */ | |
e1aab161 | 5388 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 5389 | { |
c0ff4b85 | 5390 | if (!memcg->res.parent) |
7bcc1bb1 | 5391 | return NULL; |
c0ff4b85 | 5392 | return mem_cgroup_from_res_counter(memcg->res.parent, res); |
7bcc1bb1 | 5393 | } |
e1aab161 | 5394 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 5395 | |
c255a458 | 5396 | #ifdef CONFIG_MEMCG_SWAP |
c077719b KH |
5397 | static void __init enable_swap_cgroup(void) |
5398 | { | |
f8d66542 | 5399 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
5400 | do_swap_account = 1; |
5401 | } | |
5402 | #else | |
5403 | static void __init enable_swap_cgroup(void) | |
5404 | { | |
5405 | } | |
5406 | #endif | |
5407 | ||
f64c3f54 BS |
5408 | static int mem_cgroup_soft_limit_tree_init(void) |
5409 | { | |
5410 | struct mem_cgroup_tree_per_node *rtpn; | |
5411 | struct mem_cgroup_tree_per_zone *rtpz; | |
5412 | int tmp, node, zone; | |
5413 | ||
3ed28fa1 | 5414 | for_each_node(node) { |
f64c3f54 BS |
5415 | tmp = node; |
5416 | if (!node_state(node, N_NORMAL_MEMORY)) | |
5417 | tmp = -1; | |
5418 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
5419 | if (!rtpn) | |
c3cecc68 | 5420 | goto err_cleanup; |
f64c3f54 BS |
5421 | |
5422 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
5423 | ||
5424 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
5425 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
5426 | rtpz->rb_root = RB_ROOT; | |
5427 | spin_lock_init(&rtpz->lock); | |
5428 | } | |
5429 | } | |
5430 | return 0; | |
c3cecc68 MH |
5431 | |
5432 | err_cleanup: | |
3ed28fa1 | 5433 | for_each_node(node) { |
c3cecc68 MH |
5434 | if (!soft_limit_tree.rb_tree_per_node[node]) |
5435 | break; | |
5436 | kfree(soft_limit_tree.rb_tree_per_node[node]); | |
5437 | soft_limit_tree.rb_tree_per_node[node] = NULL; | |
5438 | } | |
5439 | return 1; | |
5440 | ||
f64c3f54 BS |
5441 | } |
5442 | ||
0eb253e2 | 5443 | static struct cgroup_subsys_state * __ref |
92fb9748 | 5444 | mem_cgroup_css_alloc(struct cgroup *cont) |
8cdea7c0 | 5445 | { |
c0ff4b85 | 5446 | struct mem_cgroup *memcg, *parent; |
04046e1a | 5447 | long error = -ENOMEM; |
6d12e2d8 | 5448 | int node; |
8cdea7c0 | 5449 | |
c0ff4b85 R |
5450 | memcg = mem_cgroup_alloc(); |
5451 | if (!memcg) | |
04046e1a | 5452 | return ERR_PTR(error); |
78fb7466 | 5453 | |
3ed28fa1 | 5454 | for_each_node(node) |
c0ff4b85 | 5455 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 5456 | goto free_out; |
f64c3f54 | 5457 | |
c077719b | 5458 | /* root ? */ |
28dbc4b6 | 5459 | if (cont->parent == NULL) { |
cdec2e42 | 5460 | int cpu; |
c077719b | 5461 | enable_swap_cgroup(); |
28dbc4b6 | 5462 | parent = NULL; |
f64c3f54 BS |
5463 | if (mem_cgroup_soft_limit_tree_init()) |
5464 | goto free_out; | |
a41c58a6 | 5465 | root_mem_cgroup = memcg; |
cdec2e42 KH |
5466 | for_each_possible_cpu(cpu) { |
5467 | struct memcg_stock_pcp *stock = | |
5468 | &per_cpu(memcg_stock, cpu); | |
5469 | INIT_WORK(&stock->work, drain_local_stock); | |
5470 | } | |
711d3d2c | 5471 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
18f59ea7 | 5472 | } else { |
28dbc4b6 | 5473 | parent = mem_cgroup_from_cont(cont->parent); |
c0ff4b85 R |
5474 | memcg->use_hierarchy = parent->use_hierarchy; |
5475 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
18f59ea7 | 5476 | } |
28dbc4b6 | 5477 | |
18f59ea7 | 5478 | if (parent && parent->use_hierarchy) { |
c0ff4b85 R |
5479 | res_counter_init(&memcg->res, &parent->res); |
5480 | res_counter_init(&memcg->memsw, &parent->memsw); | |
510fc4e1 | 5481 | res_counter_init(&memcg->kmem, &parent->kmem); |
7bcc1bb1 DN |
5482 | /* |
5483 | * We increment refcnt of the parent to ensure that we can | |
5484 | * safely access it on res_counter_charge/uncharge. | |
5485 | * This refcnt will be decremented when freeing this | |
5486 | * mem_cgroup(see mem_cgroup_put). | |
5487 | */ | |
5488 | mem_cgroup_get(parent); | |
18f59ea7 | 5489 | } else { |
c0ff4b85 R |
5490 | res_counter_init(&memcg->res, NULL); |
5491 | res_counter_init(&memcg->memsw, NULL); | |
510fc4e1 | 5492 | res_counter_init(&memcg->kmem, NULL); |
8c7f6edb TH |
5493 | /* |
5494 | * Deeper hierachy with use_hierarchy == false doesn't make | |
5495 | * much sense so let cgroup subsystem know about this | |
5496 | * unfortunate state in our controller. | |
5497 | */ | |
5498 | if (parent && parent != root_mem_cgroup) | |
5499 | mem_cgroup_subsys.broken_hierarchy = true; | |
18f59ea7 | 5500 | } |
c0ff4b85 R |
5501 | memcg->last_scanned_node = MAX_NUMNODES; |
5502 | INIT_LIST_HEAD(&memcg->oom_notify); | |
6d61ef40 | 5503 | |
a7885eb8 | 5504 | if (parent) |
c0ff4b85 R |
5505 | memcg->swappiness = mem_cgroup_swappiness(parent); |
5506 | atomic_set(&memcg->refcnt, 1); | |
5507 | memcg->move_charge_at_immigrate = 0; | |
5508 | mutex_init(&memcg->thresholds_lock); | |
312734c0 | 5509 | spin_lock_init(&memcg->move_lock); |
cbe128e3 GC |
5510 | |
5511 | error = memcg_init_kmem(memcg, &mem_cgroup_subsys); | |
5512 | if (error) { | |
5513 | /* | |
5514 | * We call put now because our (and parent's) refcnts | |
5515 | * are already in place. mem_cgroup_put() will internally | |
5516 | * call __mem_cgroup_free, so return directly | |
5517 | */ | |
5518 | mem_cgroup_put(memcg); | |
5519 | return ERR_PTR(error); | |
5520 | } | |
c0ff4b85 | 5521 | return &memcg->css; |
6d12e2d8 | 5522 | free_out: |
c0ff4b85 | 5523 | __mem_cgroup_free(memcg); |
04046e1a | 5524 | return ERR_PTR(error); |
8cdea7c0 BS |
5525 | } |
5526 | ||
92fb9748 | 5527 | static void mem_cgroup_css_offline(struct cgroup *cont) |
df878fb0 | 5528 | { |
c0ff4b85 | 5529 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
ec64f515 | 5530 | |
ab5196c2 | 5531 | mem_cgroup_reparent_charges(memcg); |
df878fb0 KH |
5532 | } |
5533 | ||
92fb9748 | 5534 | static void mem_cgroup_css_free(struct cgroup *cont) |
8cdea7c0 | 5535 | { |
c0ff4b85 | 5536 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
c268e994 | 5537 | |
1d62e436 | 5538 | kmem_cgroup_destroy(memcg); |
d1a4c0b3 | 5539 | |
c0ff4b85 | 5540 | mem_cgroup_put(memcg); |
8cdea7c0 BS |
5541 | } |
5542 | ||
02491447 | 5543 | #ifdef CONFIG_MMU |
7dc74be0 | 5544 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
5545 | #define PRECHARGE_COUNT_AT_ONCE 256 |
5546 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 5547 | { |
854ffa8d DN |
5548 | int ret = 0; |
5549 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
c0ff4b85 | 5550 | struct mem_cgroup *memcg = mc.to; |
4ffef5fe | 5551 | |
c0ff4b85 | 5552 | if (mem_cgroup_is_root(memcg)) { |
854ffa8d DN |
5553 | mc.precharge += count; |
5554 | /* we don't need css_get for root */ | |
5555 | return ret; | |
5556 | } | |
5557 | /* try to charge at once */ | |
5558 | if (count > 1) { | |
5559 | struct res_counter *dummy; | |
5560 | /* | |
c0ff4b85 | 5561 | * "memcg" cannot be under rmdir() because we've already checked |
854ffa8d DN |
5562 | * by cgroup_lock_live_cgroup() that it is not removed and we |
5563 | * are still under the same cgroup_mutex. So we can postpone | |
5564 | * css_get(). | |
5565 | */ | |
c0ff4b85 | 5566 | if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy)) |
854ffa8d | 5567 | goto one_by_one; |
c0ff4b85 | 5568 | if (do_swap_account && res_counter_charge(&memcg->memsw, |
854ffa8d | 5569 | PAGE_SIZE * count, &dummy)) { |
c0ff4b85 | 5570 | res_counter_uncharge(&memcg->res, PAGE_SIZE * count); |
854ffa8d DN |
5571 | goto one_by_one; |
5572 | } | |
5573 | mc.precharge += count; | |
854ffa8d DN |
5574 | return ret; |
5575 | } | |
5576 | one_by_one: | |
5577 | /* fall back to one by one charge */ | |
5578 | while (count--) { | |
5579 | if (signal_pending(current)) { | |
5580 | ret = -EINTR; | |
5581 | break; | |
5582 | } | |
5583 | if (!batch_count--) { | |
5584 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
5585 | cond_resched(); | |
5586 | } | |
c0ff4b85 R |
5587 | ret = __mem_cgroup_try_charge(NULL, |
5588 | GFP_KERNEL, 1, &memcg, false); | |
38c5d72f | 5589 | if (ret) |
854ffa8d | 5590 | /* mem_cgroup_clear_mc() will do uncharge later */ |
38c5d72f | 5591 | return ret; |
854ffa8d DN |
5592 | mc.precharge++; |
5593 | } | |
4ffef5fe DN |
5594 | return ret; |
5595 | } | |
5596 | ||
5597 | /** | |
8d32ff84 | 5598 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
5599 | * @vma: the vma the pte to be checked belongs |
5600 | * @addr: the address corresponding to the pte to be checked | |
5601 | * @ptent: the pte to be checked | |
02491447 | 5602 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
5603 | * |
5604 | * Returns | |
5605 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5606 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5607 | * move charge. if @target is not NULL, the page is stored in target->page | |
5608 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
5609 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
5610 | * target for charge migration. if @target is not NULL, the entry is stored | |
5611 | * in target->ent. | |
4ffef5fe DN |
5612 | * |
5613 | * Called with pte lock held. | |
5614 | */ | |
4ffef5fe DN |
5615 | union mc_target { |
5616 | struct page *page; | |
02491447 | 5617 | swp_entry_t ent; |
4ffef5fe DN |
5618 | }; |
5619 | ||
4ffef5fe | 5620 | enum mc_target_type { |
8d32ff84 | 5621 | MC_TARGET_NONE = 0, |
4ffef5fe | 5622 | MC_TARGET_PAGE, |
02491447 | 5623 | MC_TARGET_SWAP, |
4ffef5fe DN |
5624 | }; |
5625 | ||
90254a65 DN |
5626 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5627 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5628 | { |
90254a65 | 5629 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5630 | |
90254a65 DN |
5631 | if (!page || !page_mapped(page)) |
5632 | return NULL; | |
5633 | if (PageAnon(page)) { | |
5634 | /* we don't move shared anon */ | |
4b91355e | 5635 | if (!move_anon()) |
90254a65 | 5636 | return NULL; |
87946a72 DN |
5637 | } else if (!move_file()) |
5638 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
5639 | return NULL; |
5640 | if (!get_page_unless_zero(page)) | |
5641 | return NULL; | |
5642 | ||
5643 | return page; | |
5644 | } | |
5645 | ||
4b91355e | 5646 | #ifdef CONFIG_SWAP |
90254a65 DN |
5647 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
5648 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5649 | { | |
90254a65 DN |
5650 | struct page *page = NULL; |
5651 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5652 | ||
5653 | if (!move_anon() || non_swap_entry(ent)) | |
5654 | return NULL; | |
4b91355e KH |
5655 | /* |
5656 | * Because lookup_swap_cache() updates some statistics counter, | |
5657 | * we call find_get_page() with swapper_space directly. | |
5658 | */ | |
5659 | page = find_get_page(&swapper_space, ent.val); | |
90254a65 DN |
5660 | if (do_swap_account) |
5661 | entry->val = ent.val; | |
5662 | ||
5663 | return page; | |
5664 | } | |
4b91355e KH |
5665 | #else |
5666 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
5667 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5668 | { | |
5669 | return NULL; | |
5670 | } | |
5671 | #endif | |
90254a65 | 5672 | |
87946a72 DN |
5673 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5674 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5675 | { | |
5676 | struct page *page = NULL; | |
87946a72 DN |
5677 | struct address_space *mapping; |
5678 | pgoff_t pgoff; | |
5679 | ||
5680 | if (!vma->vm_file) /* anonymous vma */ | |
5681 | return NULL; | |
5682 | if (!move_file()) | |
5683 | return NULL; | |
5684 | ||
87946a72 DN |
5685 | mapping = vma->vm_file->f_mapping; |
5686 | if (pte_none(ptent)) | |
5687 | pgoff = linear_page_index(vma, addr); | |
5688 | else /* pte_file(ptent) is true */ | |
5689 | pgoff = pte_to_pgoff(ptent); | |
5690 | ||
5691 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
5692 | page = find_get_page(mapping, pgoff); |
5693 | ||
5694 | #ifdef CONFIG_SWAP | |
5695 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
5696 | if (radix_tree_exceptional_entry(page)) { | |
5697 | swp_entry_t swap = radix_to_swp_entry(page); | |
87946a72 | 5698 | if (do_swap_account) |
aa3b1895 HD |
5699 | *entry = swap; |
5700 | page = find_get_page(&swapper_space, swap.val); | |
87946a72 | 5701 | } |
aa3b1895 | 5702 | #endif |
87946a72 DN |
5703 | return page; |
5704 | } | |
5705 | ||
8d32ff84 | 5706 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5707 | unsigned long addr, pte_t ptent, union mc_target *target) |
5708 | { | |
5709 | struct page *page = NULL; | |
5710 | struct page_cgroup *pc; | |
8d32ff84 | 5711 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5712 | swp_entry_t ent = { .val = 0 }; |
5713 | ||
5714 | if (pte_present(ptent)) | |
5715 | page = mc_handle_present_pte(vma, addr, ptent); | |
5716 | else if (is_swap_pte(ptent)) | |
5717 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
5718 | else if (pte_none(ptent) || pte_file(ptent)) |
5719 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
5720 | |
5721 | if (!page && !ent.val) | |
8d32ff84 | 5722 | return ret; |
02491447 DN |
5723 | if (page) { |
5724 | pc = lookup_page_cgroup(page); | |
5725 | /* | |
5726 | * Do only loose check w/o page_cgroup lock. | |
5727 | * mem_cgroup_move_account() checks the pc is valid or not under | |
5728 | * the lock. | |
5729 | */ | |
5730 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5731 | ret = MC_TARGET_PAGE; | |
5732 | if (target) | |
5733 | target->page = page; | |
5734 | } | |
5735 | if (!ret || !target) | |
5736 | put_page(page); | |
5737 | } | |
90254a65 DN |
5738 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
5739 | if (ent.val && !ret && | |
9fb4b7cc | 5740 | css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5741 | ret = MC_TARGET_SWAP; |
5742 | if (target) | |
5743 | target->ent = ent; | |
4ffef5fe | 5744 | } |
4ffef5fe DN |
5745 | return ret; |
5746 | } | |
5747 | ||
12724850 NH |
5748 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5749 | /* | |
5750 | * We don't consider swapping or file mapped pages because THP does not | |
5751 | * support them for now. | |
5752 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
5753 | */ | |
5754 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5755 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5756 | { | |
5757 | struct page *page = NULL; | |
5758 | struct page_cgroup *pc; | |
5759 | enum mc_target_type ret = MC_TARGET_NONE; | |
5760 | ||
5761 | page = pmd_page(pmd); | |
5762 | VM_BUG_ON(!page || !PageHead(page)); | |
5763 | if (!move_anon()) | |
5764 | return ret; | |
5765 | pc = lookup_page_cgroup(page); | |
5766 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5767 | ret = MC_TARGET_PAGE; | |
5768 | if (target) { | |
5769 | get_page(page); | |
5770 | target->page = page; | |
5771 | } | |
5772 | } | |
5773 | return ret; | |
5774 | } | |
5775 | #else | |
5776 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5777 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5778 | { | |
5779 | return MC_TARGET_NONE; | |
5780 | } | |
5781 | #endif | |
5782 | ||
4ffef5fe DN |
5783 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5784 | unsigned long addr, unsigned long end, | |
5785 | struct mm_walk *walk) | |
5786 | { | |
5787 | struct vm_area_struct *vma = walk->private; | |
5788 | pte_t *pte; | |
5789 | spinlock_t *ptl; | |
5790 | ||
12724850 NH |
5791 | if (pmd_trans_huge_lock(pmd, vma) == 1) { |
5792 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) | |
5793 | mc.precharge += HPAGE_PMD_NR; | |
5794 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 5795 | return 0; |
12724850 | 5796 | } |
03319327 | 5797 | |
45f83cef AA |
5798 | if (pmd_trans_unstable(pmd)) |
5799 | return 0; | |
4ffef5fe DN |
5800 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5801 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5802 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5803 | mc.precharge++; /* increment precharge temporarily */ |
5804 | pte_unmap_unlock(pte - 1, ptl); | |
5805 | cond_resched(); | |
5806 | ||
7dc74be0 DN |
5807 | return 0; |
5808 | } | |
5809 | ||
4ffef5fe DN |
5810 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5811 | { | |
5812 | unsigned long precharge; | |
5813 | struct vm_area_struct *vma; | |
5814 | ||
dfe076b0 | 5815 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
5816 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5817 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
5818 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5819 | .mm = mm, | |
5820 | .private = vma, | |
5821 | }; | |
5822 | if (is_vm_hugetlb_page(vma)) | |
5823 | continue; | |
4ffef5fe DN |
5824 | walk_page_range(vma->vm_start, vma->vm_end, |
5825 | &mem_cgroup_count_precharge_walk); | |
5826 | } | |
dfe076b0 | 5827 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5828 | |
5829 | precharge = mc.precharge; | |
5830 | mc.precharge = 0; | |
5831 | ||
5832 | return precharge; | |
5833 | } | |
5834 | ||
4ffef5fe DN |
5835 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5836 | { | |
dfe076b0 DN |
5837 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5838 | ||
5839 | VM_BUG_ON(mc.moving_task); | |
5840 | mc.moving_task = current; | |
5841 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5842 | } |
5843 | ||
dfe076b0 DN |
5844 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5845 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5846 | { |
2bd9bb20 KH |
5847 | struct mem_cgroup *from = mc.from; |
5848 | struct mem_cgroup *to = mc.to; | |
5849 | ||
4ffef5fe | 5850 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
5851 | if (mc.precharge) { |
5852 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
5853 | mc.precharge = 0; | |
5854 | } | |
5855 | /* | |
5856 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5857 | * we must uncharge here. | |
5858 | */ | |
5859 | if (mc.moved_charge) { | |
5860 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
5861 | mc.moved_charge = 0; | |
4ffef5fe | 5862 | } |
483c30b5 DN |
5863 | /* we must fixup refcnts and charges */ |
5864 | if (mc.moved_swap) { | |
483c30b5 DN |
5865 | /* uncharge swap account from the old cgroup */ |
5866 | if (!mem_cgroup_is_root(mc.from)) | |
5867 | res_counter_uncharge(&mc.from->memsw, | |
5868 | PAGE_SIZE * mc.moved_swap); | |
5869 | __mem_cgroup_put(mc.from, mc.moved_swap); | |
5870 | ||
5871 | if (!mem_cgroup_is_root(mc.to)) { | |
5872 | /* | |
5873 | * we charged both to->res and to->memsw, so we should | |
5874 | * uncharge to->res. | |
5875 | */ | |
5876 | res_counter_uncharge(&mc.to->res, | |
5877 | PAGE_SIZE * mc.moved_swap); | |
483c30b5 DN |
5878 | } |
5879 | /* we've already done mem_cgroup_get(mc.to) */ | |
483c30b5 DN |
5880 | mc.moved_swap = 0; |
5881 | } | |
dfe076b0 DN |
5882 | memcg_oom_recover(from); |
5883 | memcg_oom_recover(to); | |
5884 | wake_up_all(&mc.waitq); | |
5885 | } | |
5886 | ||
5887 | static void mem_cgroup_clear_mc(void) | |
5888 | { | |
5889 | struct mem_cgroup *from = mc.from; | |
5890 | ||
5891 | /* | |
5892 | * we must clear moving_task before waking up waiters at the end of | |
5893 | * task migration. | |
5894 | */ | |
5895 | mc.moving_task = NULL; | |
5896 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5897 | spin_lock(&mc.lock); |
4ffef5fe DN |
5898 | mc.from = NULL; |
5899 | mc.to = NULL; | |
2bd9bb20 | 5900 | spin_unlock(&mc.lock); |
32047e2a | 5901 | mem_cgroup_end_move(from); |
4ffef5fe DN |
5902 | } |
5903 | ||
761b3ef5 LZ |
5904 | static int mem_cgroup_can_attach(struct cgroup *cgroup, |
5905 | struct cgroup_taskset *tset) | |
7dc74be0 | 5906 | { |
2f7ee569 | 5907 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 5908 | int ret = 0; |
c0ff4b85 | 5909 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup); |
7dc74be0 | 5910 | |
c0ff4b85 | 5911 | if (memcg->move_charge_at_immigrate) { |
7dc74be0 DN |
5912 | struct mm_struct *mm; |
5913 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
5914 | ||
c0ff4b85 | 5915 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
5916 | |
5917 | mm = get_task_mm(p); | |
5918 | if (!mm) | |
5919 | return 0; | |
7dc74be0 | 5920 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
5921 | if (mm->owner == p) { |
5922 | VM_BUG_ON(mc.from); | |
5923 | VM_BUG_ON(mc.to); | |
5924 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 5925 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 5926 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 5927 | mem_cgroup_start_move(from); |
2bd9bb20 | 5928 | spin_lock(&mc.lock); |
4ffef5fe | 5929 | mc.from = from; |
c0ff4b85 | 5930 | mc.to = memcg; |
2bd9bb20 | 5931 | spin_unlock(&mc.lock); |
dfe076b0 | 5932 | /* We set mc.moving_task later */ |
4ffef5fe DN |
5933 | |
5934 | ret = mem_cgroup_precharge_mc(mm); | |
5935 | if (ret) | |
5936 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
5937 | } |
5938 | mmput(mm); | |
7dc74be0 DN |
5939 | } |
5940 | return ret; | |
5941 | } | |
5942 | ||
761b3ef5 LZ |
5943 | static void mem_cgroup_cancel_attach(struct cgroup *cgroup, |
5944 | struct cgroup_taskset *tset) | |
7dc74be0 | 5945 | { |
4ffef5fe | 5946 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
5947 | } |
5948 | ||
4ffef5fe DN |
5949 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5950 | unsigned long addr, unsigned long end, | |
5951 | struct mm_walk *walk) | |
7dc74be0 | 5952 | { |
4ffef5fe DN |
5953 | int ret = 0; |
5954 | struct vm_area_struct *vma = walk->private; | |
5955 | pte_t *pte; | |
5956 | spinlock_t *ptl; | |
12724850 NH |
5957 | enum mc_target_type target_type; |
5958 | union mc_target target; | |
5959 | struct page *page; | |
5960 | struct page_cgroup *pc; | |
4ffef5fe | 5961 | |
12724850 NH |
5962 | /* |
5963 | * We don't take compound_lock() here but no race with splitting thp | |
5964 | * happens because: | |
5965 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
5966 | * under splitting, which means there's no concurrent thp split, | |
5967 | * - if another thread runs into split_huge_page() just after we | |
5968 | * entered this if-block, the thread must wait for page table lock | |
5969 | * to be unlocked in __split_huge_page_splitting(), where the main | |
5970 | * part of thp split is not executed yet. | |
5971 | */ | |
5972 | if (pmd_trans_huge_lock(pmd, vma) == 1) { | |
62ade86a | 5973 | if (mc.precharge < HPAGE_PMD_NR) { |
12724850 NH |
5974 | spin_unlock(&vma->vm_mm->page_table_lock); |
5975 | return 0; | |
5976 | } | |
5977 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5978 | if (target_type == MC_TARGET_PAGE) { | |
5979 | page = target.page; | |
5980 | if (!isolate_lru_page(page)) { | |
5981 | pc = lookup_page_cgroup(page); | |
5982 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, | |
2f3479b1 | 5983 | pc, mc.from, mc.to)) { |
12724850 NH |
5984 | mc.precharge -= HPAGE_PMD_NR; |
5985 | mc.moved_charge += HPAGE_PMD_NR; | |
5986 | } | |
5987 | putback_lru_page(page); | |
5988 | } | |
5989 | put_page(page); | |
5990 | } | |
5991 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 5992 | return 0; |
12724850 NH |
5993 | } |
5994 | ||
45f83cef AA |
5995 | if (pmd_trans_unstable(pmd)) |
5996 | return 0; | |
4ffef5fe DN |
5997 | retry: |
5998 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5999 | for (; addr != end; addr += PAGE_SIZE) { | |
6000 | pte_t ptent = *(pte++); | |
02491447 | 6001 | swp_entry_t ent; |
4ffef5fe DN |
6002 | |
6003 | if (!mc.precharge) | |
6004 | break; | |
6005 | ||
8d32ff84 | 6006 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
6007 | case MC_TARGET_PAGE: |
6008 | page = target.page; | |
6009 | if (isolate_lru_page(page)) | |
6010 | goto put; | |
6011 | pc = lookup_page_cgroup(page); | |
7ec99d62 | 6012 | if (!mem_cgroup_move_account(page, 1, pc, |
2f3479b1 | 6013 | mc.from, mc.to)) { |
4ffef5fe | 6014 | mc.precharge--; |
854ffa8d DN |
6015 | /* we uncharge from mc.from later. */ |
6016 | mc.moved_charge++; | |
4ffef5fe DN |
6017 | } |
6018 | putback_lru_page(page); | |
8d32ff84 | 6019 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
6020 | put_page(page); |
6021 | break; | |
02491447 DN |
6022 | case MC_TARGET_SWAP: |
6023 | ent = target.ent; | |
e91cbb42 | 6024 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6025 | mc.precharge--; |
483c30b5 DN |
6026 | /* we fixup refcnts and charges later. */ |
6027 | mc.moved_swap++; | |
6028 | } | |
02491447 | 6029 | break; |
4ffef5fe DN |
6030 | default: |
6031 | break; | |
6032 | } | |
6033 | } | |
6034 | pte_unmap_unlock(pte - 1, ptl); | |
6035 | cond_resched(); | |
6036 | ||
6037 | if (addr != end) { | |
6038 | /* | |
6039 | * We have consumed all precharges we got in can_attach(). | |
6040 | * We try charge one by one, but don't do any additional | |
6041 | * charges to mc.to if we have failed in charge once in attach() | |
6042 | * phase. | |
6043 | */ | |
854ffa8d | 6044 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6045 | if (!ret) |
6046 | goto retry; | |
6047 | } | |
6048 | ||
6049 | return ret; | |
6050 | } | |
6051 | ||
6052 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
6053 | { | |
6054 | struct vm_area_struct *vma; | |
6055 | ||
6056 | lru_add_drain_all(); | |
dfe076b0 DN |
6057 | retry: |
6058 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
6059 | /* | |
6060 | * Someone who are holding the mmap_sem might be waiting in | |
6061 | * waitq. So we cancel all extra charges, wake up all waiters, | |
6062 | * and retry. Because we cancel precharges, we might not be able | |
6063 | * to move enough charges, but moving charge is a best-effort | |
6064 | * feature anyway, so it wouldn't be a big problem. | |
6065 | */ | |
6066 | __mem_cgroup_clear_mc(); | |
6067 | cond_resched(); | |
6068 | goto retry; | |
6069 | } | |
4ffef5fe DN |
6070 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
6071 | int ret; | |
6072 | struct mm_walk mem_cgroup_move_charge_walk = { | |
6073 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6074 | .mm = mm, | |
6075 | .private = vma, | |
6076 | }; | |
6077 | if (is_vm_hugetlb_page(vma)) | |
6078 | continue; | |
4ffef5fe DN |
6079 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
6080 | &mem_cgroup_move_charge_walk); | |
6081 | if (ret) | |
6082 | /* | |
6083 | * means we have consumed all precharges and failed in | |
6084 | * doing additional charge. Just abandon here. | |
6085 | */ | |
6086 | break; | |
6087 | } | |
dfe076b0 | 6088 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
6089 | } |
6090 | ||
761b3ef5 LZ |
6091 | static void mem_cgroup_move_task(struct cgroup *cont, |
6092 | struct cgroup_taskset *tset) | |
67e465a7 | 6093 | { |
2f7ee569 | 6094 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 6095 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 6096 | |
dfe076b0 | 6097 | if (mm) { |
a433658c KM |
6098 | if (mc.to) |
6099 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
6100 | mmput(mm); |
6101 | } | |
a433658c KM |
6102 | if (mc.to) |
6103 | mem_cgroup_clear_mc(); | |
67e465a7 | 6104 | } |
5cfb80a7 | 6105 | #else /* !CONFIG_MMU */ |
761b3ef5 LZ |
6106 | static int mem_cgroup_can_attach(struct cgroup *cgroup, |
6107 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
6108 | { |
6109 | return 0; | |
6110 | } | |
761b3ef5 LZ |
6111 | static void mem_cgroup_cancel_attach(struct cgroup *cgroup, |
6112 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
6113 | { |
6114 | } | |
761b3ef5 LZ |
6115 | static void mem_cgroup_move_task(struct cgroup *cont, |
6116 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
6117 | { |
6118 | } | |
6119 | #endif | |
67e465a7 | 6120 | |
8cdea7c0 BS |
6121 | struct cgroup_subsys mem_cgroup_subsys = { |
6122 | .name = "memory", | |
6123 | .subsys_id = mem_cgroup_subsys_id, | |
92fb9748 TH |
6124 | .css_alloc = mem_cgroup_css_alloc, |
6125 | .css_offline = mem_cgroup_css_offline, | |
6126 | .css_free = mem_cgroup_css_free, | |
7dc74be0 DN |
6127 | .can_attach = mem_cgroup_can_attach, |
6128 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 6129 | .attach = mem_cgroup_move_task, |
6bc10349 | 6130 | .base_cftypes = mem_cgroup_files, |
6d12e2d8 | 6131 | .early_init = 0, |
04046e1a | 6132 | .use_id = 1, |
8cdea7c0 | 6133 | }; |
c077719b | 6134 | |
c255a458 | 6135 | #ifdef CONFIG_MEMCG_SWAP |
a42c390c MH |
6136 | static int __init enable_swap_account(char *s) |
6137 | { | |
6138 | /* consider enabled if no parameter or 1 is given */ | |
a2c8990a | 6139 | if (!strcmp(s, "1")) |
a42c390c | 6140 | really_do_swap_account = 1; |
a2c8990a | 6141 | else if (!strcmp(s, "0")) |
a42c390c MH |
6142 | really_do_swap_account = 0; |
6143 | return 1; | |
6144 | } | |
a2c8990a | 6145 | __setup("swapaccount=", enable_swap_account); |
c077719b | 6146 | |
c077719b | 6147 | #endif |