Commit | Line | Data |
---|---|---|
8cdea7c0 BS |
1 | /* memcontrol.c - Memory Controller |
2 | * | |
3 | * Copyright IBM Corporation, 2007 | |
4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
5 | * | |
78fb7466 PE |
6 | * Copyright 2007 OpenVZ SWsoft Inc |
7 | * Author: Pavel Emelianov <xemul@openvz.org> | |
8 | * | |
2e72b634 KS |
9 | * Memory thresholds |
10 | * Copyright (C) 2009 Nokia Corporation | |
11 | * Author: Kirill A. Shutemov | |
12 | * | |
8cdea7c0 BS |
13 | * This program is free software; you can redistribute it and/or modify |
14 | * it under the terms of the GNU General Public License as published by | |
15 | * the Free Software Foundation; either version 2 of the License, or | |
16 | * (at your option) any later version. | |
17 | * | |
18 | * This program is distributed in the hope that it will be useful, | |
19 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
20 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
21 | * GNU General Public License for more details. | |
22 | */ | |
23 | ||
24 | #include <linux/res_counter.h> | |
25 | #include <linux/memcontrol.h> | |
26 | #include <linux/cgroup.h> | |
78fb7466 | 27 | #include <linux/mm.h> |
4ffef5fe | 28 | #include <linux/hugetlb.h> |
d13d1443 | 29 | #include <linux/pagemap.h> |
d52aa412 | 30 | #include <linux/smp.h> |
8a9f3ccd | 31 | #include <linux/page-flags.h> |
66e1707b | 32 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
33 | #include <linux/bit_spinlock.h> |
34 | #include <linux/rcupdate.h> | |
e222432b | 35 | #include <linux/limits.h> |
b9e15baf | 36 | #include <linux/export.h> |
8c7c6e34 | 37 | #include <linux/mutex.h> |
f64c3f54 | 38 | #include <linux/rbtree.h> |
b6ac57d5 | 39 | #include <linux/slab.h> |
66e1707b | 40 | #include <linux/swap.h> |
02491447 | 41 | #include <linux/swapops.h> |
66e1707b | 42 | #include <linux/spinlock.h> |
2e72b634 KS |
43 | #include <linux/eventfd.h> |
44 | #include <linux/sort.h> | |
66e1707b | 45 | #include <linux/fs.h> |
d2ceb9b7 | 46 | #include <linux/seq_file.h> |
33327948 | 47 | #include <linux/vmalloc.h> |
b69408e8 | 48 | #include <linux/mm_inline.h> |
52d4b9ac | 49 | #include <linux/page_cgroup.h> |
cdec2e42 | 50 | #include <linux/cpu.h> |
158e0a2d | 51 | #include <linux/oom.h> |
08e552c6 | 52 | #include "internal.h" |
d1a4c0b3 GC |
53 | #include <net/sock.h> |
54 | #include <net/tcp_memcontrol.h> | |
8cdea7c0 | 55 | |
8697d331 BS |
56 | #include <asm/uaccess.h> |
57 | ||
cc8e970c KM |
58 | #include <trace/events/vmscan.h> |
59 | ||
a181b0e8 | 60 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
a181b0e8 | 61 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
4b3bde4c | 62 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
8cdea7c0 | 63 | |
c077719b | 64 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
338c8431 | 65 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ |
c077719b | 66 | int do_swap_account __read_mostly; |
a42c390c MH |
67 | |
68 | /* for remember boot option*/ | |
69 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED | |
70 | static int really_do_swap_account __initdata = 1; | |
71 | #else | |
72 | static int really_do_swap_account __initdata = 0; | |
73 | #endif | |
74 | ||
c077719b KH |
75 | #else |
76 | #define do_swap_account (0) | |
77 | #endif | |
78 | ||
79 | ||
d52aa412 KH |
80 | /* |
81 | * Statistics for memory cgroup. | |
82 | */ | |
83 | enum mem_cgroup_stat_index { | |
84 | /* | |
85 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
86 | */ | |
87 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
d69b042f | 88 | MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */ |
d8046582 | 89 | MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */ |
0c3e73e8 | 90 | MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */ |
711d3d2c | 91 | MEM_CGROUP_STAT_DATA, /* end of data requires synchronization */ |
32047e2a | 92 | MEM_CGROUP_ON_MOVE, /* someone is moving account between groups */ |
d52aa412 KH |
93 | MEM_CGROUP_STAT_NSTATS, |
94 | }; | |
95 | ||
e9f8974f JW |
96 | enum mem_cgroup_events_index { |
97 | MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */ | |
98 | MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */ | |
99 | MEM_CGROUP_EVENTS_COUNT, /* # of pages paged in/out */ | |
456f998e YH |
100 | MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */ |
101 | MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */ | |
e9f8974f JW |
102 | MEM_CGROUP_EVENTS_NSTATS, |
103 | }; | |
7a159cc9 JW |
104 | /* |
105 | * Per memcg event counter is incremented at every pagein/pageout. With THP, | |
106 | * it will be incremated by the number of pages. This counter is used for | |
107 | * for trigger some periodic events. This is straightforward and better | |
108 | * than using jiffies etc. to handle periodic memcg event. | |
109 | */ | |
110 | enum mem_cgroup_events_target { | |
111 | MEM_CGROUP_TARGET_THRESH, | |
112 | MEM_CGROUP_TARGET_SOFTLIMIT, | |
453a9bf3 | 113 | MEM_CGROUP_TARGET_NUMAINFO, |
7a159cc9 JW |
114 | MEM_CGROUP_NTARGETS, |
115 | }; | |
116 | #define THRESHOLDS_EVENTS_TARGET (128) | |
117 | #define SOFTLIMIT_EVENTS_TARGET (1024) | |
453a9bf3 | 118 | #define NUMAINFO_EVENTS_TARGET (1024) |
e9f8974f | 119 | |
d52aa412 | 120 | struct mem_cgroup_stat_cpu { |
7a159cc9 | 121 | long count[MEM_CGROUP_STAT_NSTATS]; |
e9f8974f | 122 | unsigned long events[MEM_CGROUP_EVENTS_NSTATS]; |
7a159cc9 | 123 | unsigned long targets[MEM_CGROUP_NTARGETS]; |
d52aa412 KH |
124 | }; |
125 | ||
527a5ec9 JW |
126 | struct mem_cgroup_reclaim_iter { |
127 | /* css_id of the last scanned hierarchy member */ | |
128 | int position; | |
129 | /* scan generation, increased every round-trip */ | |
130 | unsigned int generation; | |
131 | }; | |
132 | ||
6d12e2d8 KH |
133 | /* |
134 | * per-zone information in memory controller. | |
135 | */ | |
6d12e2d8 | 136 | struct mem_cgroup_per_zone { |
6290df54 | 137 | struct lruvec lruvec; |
1eb49272 | 138 | unsigned long lru_size[NR_LRU_LISTS]; |
3e2f41f1 | 139 | |
527a5ec9 JW |
140 | struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1]; |
141 | ||
3e2f41f1 | 142 | struct zone_reclaim_stat reclaim_stat; |
f64c3f54 BS |
143 | struct rb_node tree_node; /* RB tree node */ |
144 | unsigned long long usage_in_excess;/* Set to the value by which */ | |
145 | /* the soft limit is exceeded*/ | |
146 | bool on_tree; | |
d79154bb | 147 | struct mem_cgroup *memcg; /* Back pointer, we cannot */ |
4e416953 | 148 | /* use container_of */ |
6d12e2d8 | 149 | }; |
6d12e2d8 KH |
150 | |
151 | struct mem_cgroup_per_node { | |
152 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
153 | }; | |
154 | ||
155 | struct mem_cgroup_lru_info { | |
156 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
157 | }; | |
158 | ||
f64c3f54 BS |
159 | /* |
160 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
161 | * their hierarchy representation | |
162 | */ | |
163 | ||
164 | struct mem_cgroup_tree_per_zone { | |
165 | struct rb_root rb_root; | |
166 | spinlock_t lock; | |
167 | }; | |
168 | ||
169 | struct mem_cgroup_tree_per_node { | |
170 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
171 | }; | |
172 | ||
173 | struct mem_cgroup_tree { | |
174 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
175 | }; | |
176 | ||
177 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
178 | ||
2e72b634 KS |
179 | struct mem_cgroup_threshold { |
180 | struct eventfd_ctx *eventfd; | |
181 | u64 threshold; | |
182 | }; | |
183 | ||
9490ff27 | 184 | /* For threshold */ |
2e72b634 KS |
185 | struct mem_cgroup_threshold_ary { |
186 | /* An array index points to threshold just below usage. */ | |
5407a562 | 187 | int current_threshold; |
2e72b634 KS |
188 | /* Size of entries[] */ |
189 | unsigned int size; | |
190 | /* Array of thresholds */ | |
191 | struct mem_cgroup_threshold entries[0]; | |
192 | }; | |
2c488db2 KS |
193 | |
194 | struct mem_cgroup_thresholds { | |
195 | /* Primary thresholds array */ | |
196 | struct mem_cgroup_threshold_ary *primary; | |
197 | /* | |
198 | * Spare threshold array. | |
199 | * This is needed to make mem_cgroup_unregister_event() "never fail". | |
200 | * It must be able to store at least primary->size - 1 entries. | |
201 | */ | |
202 | struct mem_cgroup_threshold_ary *spare; | |
203 | }; | |
204 | ||
9490ff27 KH |
205 | /* for OOM */ |
206 | struct mem_cgroup_eventfd_list { | |
207 | struct list_head list; | |
208 | struct eventfd_ctx *eventfd; | |
209 | }; | |
2e72b634 | 210 | |
c0ff4b85 R |
211 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
212 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 213 | |
8cdea7c0 BS |
214 | /* |
215 | * The memory controller data structure. The memory controller controls both | |
216 | * page cache and RSS per cgroup. We would eventually like to provide | |
217 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
218 | * to help the administrator determine what knobs to tune. | |
219 | * | |
220 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
221 | * we hit the water mark. May be even add a low water mark, such that |
222 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
223 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
224 | */ |
225 | struct mem_cgroup { | |
226 | struct cgroup_subsys_state css; | |
227 | /* | |
228 | * the counter to account for memory usage | |
229 | */ | |
230 | struct res_counter res; | |
59927fb9 HD |
231 | |
232 | union { | |
233 | /* | |
234 | * the counter to account for mem+swap usage. | |
235 | */ | |
236 | struct res_counter memsw; | |
237 | ||
238 | /* | |
239 | * rcu_freeing is used only when freeing struct mem_cgroup, | |
240 | * so put it into a union to avoid wasting more memory. | |
241 | * It must be disjoint from the css field. It could be | |
242 | * in a union with the res field, but res plays a much | |
243 | * larger part in mem_cgroup life than memsw, and might | |
244 | * be of interest, even at time of free, when debugging. | |
245 | * So share rcu_head with the less interesting memsw. | |
246 | */ | |
247 | struct rcu_head rcu_freeing; | |
248 | /* | |
249 | * But when using vfree(), that cannot be done at | |
250 | * interrupt time, so we must then queue the work. | |
251 | */ | |
252 | struct work_struct work_freeing; | |
253 | }; | |
254 | ||
78fb7466 PE |
255 | /* |
256 | * Per cgroup active and inactive list, similar to the | |
257 | * per zone LRU lists. | |
78fb7466 | 258 | */ |
6d12e2d8 | 259 | struct mem_cgroup_lru_info info; |
889976db YH |
260 | int last_scanned_node; |
261 | #if MAX_NUMNODES > 1 | |
262 | nodemask_t scan_nodes; | |
453a9bf3 KH |
263 | atomic_t numainfo_events; |
264 | atomic_t numainfo_updating; | |
889976db | 265 | #endif |
18f59ea7 BS |
266 | /* |
267 | * Should the accounting and control be hierarchical, per subtree? | |
268 | */ | |
269 | bool use_hierarchy; | |
79dfdacc MH |
270 | |
271 | bool oom_lock; | |
272 | atomic_t under_oom; | |
273 | ||
8c7c6e34 | 274 | atomic_t refcnt; |
14797e23 | 275 | |
1f4c025b | 276 | int swappiness; |
3c11ecf4 KH |
277 | /* OOM-Killer disable */ |
278 | int oom_kill_disable; | |
a7885eb8 | 279 | |
22a668d7 KH |
280 | /* set when res.limit == memsw.limit */ |
281 | bool memsw_is_minimum; | |
282 | ||
2e72b634 KS |
283 | /* protect arrays of thresholds */ |
284 | struct mutex thresholds_lock; | |
285 | ||
286 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 287 | struct mem_cgroup_thresholds thresholds; |
907860ed | 288 | |
2e72b634 | 289 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 290 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 291 | |
9490ff27 KH |
292 | /* For oom notifier event fd */ |
293 | struct list_head oom_notify; | |
185efc0f | 294 | |
7dc74be0 DN |
295 | /* |
296 | * Should we move charges of a task when a task is moved into this | |
297 | * mem_cgroup ? And what type of charges should we move ? | |
298 | */ | |
299 | unsigned long move_charge_at_immigrate; | |
d52aa412 | 300 | /* |
c62b1a3b | 301 | * percpu counter. |
d52aa412 | 302 | */ |
c62b1a3b | 303 | struct mem_cgroup_stat_cpu *stat; |
711d3d2c KH |
304 | /* |
305 | * used when a cpu is offlined or other synchronizations | |
306 | * See mem_cgroup_read_stat(). | |
307 | */ | |
308 | struct mem_cgroup_stat_cpu nocpu_base; | |
309 | spinlock_t pcp_counter_lock; | |
d1a4c0b3 GC |
310 | |
311 | #ifdef CONFIG_INET | |
312 | struct tcp_memcontrol tcp_mem; | |
313 | #endif | |
8cdea7c0 BS |
314 | }; |
315 | ||
7dc74be0 DN |
316 | /* Stuffs for move charges at task migration. */ |
317 | /* | |
318 | * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a | |
319 | * left-shifted bitmap of these types. | |
320 | */ | |
321 | enum move_type { | |
4ffef5fe | 322 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 323 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
324 | NR_MOVE_TYPE, |
325 | }; | |
326 | ||
4ffef5fe DN |
327 | /* "mc" and its members are protected by cgroup_mutex */ |
328 | static struct move_charge_struct { | |
b1dd693e | 329 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
330 | struct mem_cgroup *from; |
331 | struct mem_cgroup *to; | |
332 | unsigned long precharge; | |
854ffa8d | 333 | unsigned long moved_charge; |
483c30b5 | 334 | unsigned long moved_swap; |
8033b97c DN |
335 | struct task_struct *moving_task; /* a task moving charges */ |
336 | wait_queue_head_t waitq; /* a waitq for other context */ | |
337 | } mc = { | |
2bd9bb20 | 338 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
339 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
340 | }; | |
4ffef5fe | 341 | |
90254a65 DN |
342 | static bool move_anon(void) |
343 | { | |
344 | return test_bit(MOVE_CHARGE_TYPE_ANON, | |
345 | &mc.to->move_charge_at_immigrate); | |
346 | } | |
347 | ||
87946a72 DN |
348 | static bool move_file(void) |
349 | { | |
350 | return test_bit(MOVE_CHARGE_TYPE_FILE, | |
351 | &mc.to->move_charge_at_immigrate); | |
352 | } | |
353 | ||
4e416953 BS |
354 | /* |
355 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
356 | * limit reclaim to prevent infinite loops, if they ever occur. | |
357 | */ | |
358 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100) | |
359 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2) | |
360 | ||
217bc319 KH |
361 | enum charge_type { |
362 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
363 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
4f98a2fe | 364 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
c05555b5 | 365 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
d13d1443 | 366 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 367 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
368 | NR_CHARGE_TYPE, |
369 | }; | |
370 | ||
8c7c6e34 | 371 | /* for encoding cft->private value on file */ |
65c64ce8 GC |
372 | #define _MEM (0) |
373 | #define _MEMSWAP (1) | |
374 | #define _OOM_TYPE (2) | |
8c7c6e34 KH |
375 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) |
376 | #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) | |
377 | #define MEMFILE_ATTR(val) ((val) & 0xffff) | |
9490ff27 KH |
378 | /* Used for OOM nofiier */ |
379 | #define OOM_CONTROL (0) | |
8c7c6e34 | 380 | |
75822b44 BS |
381 | /* |
382 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
383 | */ | |
384 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
385 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
386 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
387 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
388 | ||
c0ff4b85 R |
389 | static void mem_cgroup_get(struct mem_cgroup *memcg); |
390 | static void mem_cgroup_put(struct mem_cgroup *memcg); | |
e1aab161 GC |
391 | |
392 | /* Writing them here to avoid exposing memcg's inner layout */ | |
393 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM | |
e1aab161 | 394 | #include <net/sock.h> |
d1a4c0b3 | 395 | #include <net/ip.h> |
e1aab161 GC |
396 | |
397 | static bool mem_cgroup_is_root(struct mem_cgroup *memcg); | |
398 | void sock_update_memcg(struct sock *sk) | |
399 | { | |
376be5ff | 400 | if (mem_cgroup_sockets_enabled) { |
e1aab161 GC |
401 | struct mem_cgroup *memcg; |
402 | ||
403 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
404 | ||
f3f511e1 GC |
405 | /* Socket cloning can throw us here with sk_cgrp already |
406 | * filled. It won't however, necessarily happen from | |
407 | * process context. So the test for root memcg given | |
408 | * the current task's memcg won't help us in this case. | |
409 | * | |
410 | * Respecting the original socket's memcg is a better | |
411 | * decision in this case. | |
412 | */ | |
413 | if (sk->sk_cgrp) { | |
414 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
415 | mem_cgroup_get(sk->sk_cgrp->memcg); | |
416 | return; | |
417 | } | |
418 | ||
e1aab161 GC |
419 | rcu_read_lock(); |
420 | memcg = mem_cgroup_from_task(current); | |
421 | if (!mem_cgroup_is_root(memcg)) { | |
422 | mem_cgroup_get(memcg); | |
423 | sk->sk_cgrp = sk->sk_prot->proto_cgroup(memcg); | |
424 | } | |
425 | rcu_read_unlock(); | |
426 | } | |
427 | } | |
428 | EXPORT_SYMBOL(sock_update_memcg); | |
429 | ||
430 | void sock_release_memcg(struct sock *sk) | |
431 | { | |
376be5ff | 432 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
433 | struct mem_cgroup *memcg; |
434 | WARN_ON(!sk->sk_cgrp->memcg); | |
435 | memcg = sk->sk_cgrp->memcg; | |
436 | mem_cgroup_put(memcg); | |
437 | } | |
438 | } | |
d1a4c0b3 | 439 | |
319d3b9c | 440 | #ifdef CONFIG_INET |
d1a4c0b3 GC |
441 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) |
442 | { | |
443 | if (!memcg || mem_cgroup_is_root(memcg)) | |
444 | return NULL; | |
445 | ||
446 | return &memcg->tcp_mem.cg_proto; | |
447 | } | |
448 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 GC |
449 | #endif /* CONFIG_INET */ |
450 | #endif /* CONFIG_CGROUP_MEM_RES_CTLR_KMEM */ | |
451 | ||
c0ff4b85 | 452 | static void drain_all_stock_async(struct mem_cgroup *memcg); |
8c7c6e34 | 453 | |
f64c3f54 | 454 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 455 | mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid) |
f64c3f54 | 456 | { |
c0ff4b85 | 457 | return &memcg->info.nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
458 | } |
459 | ||
c0ff4b85 | 460 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 461 | { |
c0ff4b85 | 462 | return &memcg->css; |
d324236b WF |
463 | } |
464 | ||
f64c3f54 | 465 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 466 | page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 467 | { |
97a6c37b JW |
468 | int nid = page_to_nid(page); |
469 | int zid = page_zonenum(page); | |
f64c3f54 | 470 | |
c0ff4b85 | 471 | return mem_cgroup_zoneinfo(memcg, nid, zid); |
f64c3f54 BS |
472 | } |
473 | ||
474 | static struct mem_cgroup_tree_per_zone * | |
475 | soft_limit_tree_node_zone(int nid, int zid) | |
476 | { | |
477 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
478 | } | |
479 | ||
480 | static struct mem_cgroup_tree_per_zone * | |
481 | soft_limit_tree_from_page(struct page *page) | |
482 | { | |
483 | int nid = page_to_nid(page); | |
484 | int zid = page_zonenum(page); | |
485 | ||
486 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
487 | } | |
488 | ||
489 | static void | |
c0ff4b85 | 490 | __mem_cgroup_insert_exceeded(struct mem_cgroup *memcg, |
f64c3f54 | 491 | struct mem_cgroup_per_zone *mz, |
ef8745c1 KH |
492 | struct mem_cgroup_tree_per_zone *mctz, |
493 | unsigned long long new_usage_in_excess) | |
f64c3f54 BS |
494 | { |
495 | struct rb_node **p = &mctz->rb_root.rb_node; | |
496 | struct rb_node *parent = NULL; | |
497 | struct mem_cgroup_per_zone *mz_node; | |
498 | ||
499 | if (mz->on_tree) | |
500 | return; | |
501 | ||
ef8745c1 KH |
502 | mz->usage_in_excess = new_usage_in_excess; |
503 | if (!mz->usage_in_excess) | |
504 | return; | |
f64c3f54 BS |
505 | while (*p) { |
506 | parent = *p; | |
507 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
508 | tree_node); | |
509 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
510 | p = &(*p)->rb_left; | |
511 | /* | |
512 | * We can't avoid mem cgroups that are over their soft | |
513 | * limit by the same amount | |
514 | */ | |
515 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
516 | p = &(*p)->rb_right; | |
517 | } | |
518 | rb_link_node(&mz->tree_node, parent, p); | |
519 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
520 | mz->on_tree = true; | |
4e416953 BS |
521 | } |
522 | ||
523 | static void | |
c0ff4b85 | 524 | __mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
4e416953 BS |
525 | struct mem_cgroup_per_zone *mz, |
526 | struct mem_cgroup_tree_per_zone *mctz) | |
527 | { | |
528 | if (!mz->on_tree) | |
529 | return; | |
530 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
531 | mz->on_tree = false; | |
532 | } | |
533 | ||
f64c3f54 | 534 | static void |
c0ff4b85 | 535 | mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
f64c3f54 BS |
536 | struct mem_cgroup_per_zone *mz, |
537 | struct mem_cgroup_tree_per_zone *mctz) | |
538 | { | |
539 | spin_lock(&mctz->lock); | |
c0ff4b85 | 540 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
541 | spin_unlock(&mctz->lock); |
542 | } | |
543 | ||
f64c3f54 | 544 | |
c0ff4b85 | 545 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 546 | { |
ef8745c1 | 547 | unsigned long long excess; |
f64c3f54 BS |
548 | struct mem_cgroup_per_zone *mz; |
549 | struct mem_cgroup_tree_per_zone *mctz; | |
4e649152 KH |
550 | int nid = page_to_nid(page); |
551 | int zid = page_zonenum(page); | |
f64c3f54 BS |
552 | mctz = soft_limit_tree_from_page(page); |
553 | ||
554 | /* | |
4e649152 KH |
555 | * Necessary to update all ancestors when hierarchy is used. |
556 | * because their event counter is not touched. | |
f64c3f54 | 557 | */ |
c0ff4b85 R |
558 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
559 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
560 | excess = res_counter_soft_limit_excess(&memcg->res); | |
4e649152 KH |
561 | /* |
562 | * We have to update the tree if mz is on RB-tree or | |
563 | * mem is over its softlimit. | |
564 | */ | |
ef8745c1 | 565 | if (excess || mz->on_tree) { |
4e649152 KH |
566 | spin_lock(&mctz->lock); |
567 | /* if on-tree, remove it */ | |
568 | if (mz->on_tree) | |
c0ff4b85 | 569 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
4e649152 | 570 | /* |
ef8745c1 KH |
571 | * Insert again. mz->usage_in_excess will be updated. |
572 | * If excess is 0, no tree ops. | |
4e649152 | 573 | */ |
c0ff4b85 | 574 | __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess); |
4e649152 KH |
575 | spin_unlock(&mctz->lock); |
576 | } | |
f64c3f54 BS |
577 | } |
578 | } | |
579 | ||
c0ff4b85 | 580 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) |
f64c3f54 BS |
581 | { |
582 | int node, zone; | |
583 | struct mem_cgroup_per_zone *mz; | |
584 | struct mem_cgroup_tree_per_zone *mctz; | |
585 | ||
3ed28fa1 | 586 | for_each_node(node) { |
f64c3f54 | 587 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
c0ff4b85 | 588 | mz = mem_cgroup_zoneinfo(memcg, node, zone); |
f64c3f54 | 589 | mctz = soft_limit_tree_node_zone(node, zone); |
c0ff4b85 | 590 | mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
591 | } |
592 | } | |
593 | } | |
594 | ||
4e416953 BS |
595 | static struct mem_cgroup_per_zone * |
596 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
597 | { | |
598 | struct rb_node *rightmost = NULL; | |
26251eaf | 599 | struct mem_cgroup_per_zone *mz; |
4e416953 BS |
600 | |
601 | retry: | |
26251eaf | 602 | mz = NULL; |
4e416953 BS |
603 | rightmost = rb_last(&mctz->rb_root); |
604 | if (!rightmost) | |
605 | goto done; /* Nothing to reclaim from */ | |
606 | ||
607 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
608 | /* | |
609 | * Remove the node now but someone else can add it back, | |
610 | * we will to add it back at the end of reclaim to its correct | |
611 | * position in the tree. | |
612 | */ | |
d79154bb HD |
613 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); |
614 | if (!res_counter_soft_limit_excess(&mz->memcg->res) || | |
615 | !css_tryget(&mz->memcg->css)) | |
4e416953 BS |
616 | goto retry; |
617 | done: | |
618 | return mz; | |
619 | } | |
620 | ||
621 | static struct mem_cgroup_per_zone * | |
622 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
623 | { | |
624 | struct mem_cgroup_per_zone *mz; | |
625 | ||
626 | spin_lock(&mctz->lock); | |
627 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
628 | spin_unlock(&mctz->lock); | |
629 | return mz; | |
630 | } | |
631 | ||
711d3d2c KH |
632 | /* |
633 | * Implementation Note: reading percpu statistics for memcg. | |
634 | * | |
635 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
636 | * synchronization to implement "quick" read. There are trade-off between | |
637 | * reading cost and precision of value. Then, we may have a chance to implement | |
638 | * a periodic synchronizion of counter in memcg's counter. | |
639 | * | |
640 | * But this _read() function is used for user interface now. The user accounts | |
641 | * memory usage by memory cgroup and he _always_ requires exact value because | |
642 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
643 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
644 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
645 | * | |
646 | * If there are kernel internal actions which can make use of some not-exact | |
647 | * value, and reading all cpu value can be performance bottleneck in some | |
648 | * common workload, threashold and synchonization as vmstat[] should be | |
649 | * implemented. | |
650 | */ | |
c0ff4b85 | 651 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 652 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 653 | { |
7a159cc9 | 654 | long val = 0; |
c62b1a3b | 655 | int cpu; |
c62b1a3b | 656 | |
711d3d2c KH |
657 | get_online_cpus(); |
658 | for_each_online_cpu(cpu) | |
c0ff4b85 | 659 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 660 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
661 | spin_lock(&memcg->pcp_counter_lock); |
662 | val += memcg->nocpu_base.count[idx]; | |
663 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
664 | #endif |
665 | put_online_cpus(); | |
c62b1a3b KH |
666 | return val; |
667 | } | |
668 | ||
c0ff4b85 | 669 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
0c3e73e8 BS |
670 | bool charge) |
671 | { | |
672 | int val = (charge) ? 1 : -1; | |
c0ff4b85 | 673 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAPOUT], val); |
0c3e73e8 BS |
674 | } |
675 | ||
c0ff4b85 | 676 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
677 | enum mem_cgroup_events_index idx) |
678 | { | |
679 | unsigned long val = 0; | |
680 | int cpu; | |
681 | ||
682 | for_each_online_cpu(cpu) | |
c0ff4b85 | 683 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 684 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
685 | spin_lock(&memcg->pcp_counter_lock); |
686 | val += memcg->nocpu_base.events[idx]; | |
687 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f JW |
688 | #endif |
689 | return val; | |
690 | } | |
691 | ||
c0ff4b85 | 692 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b2402857 | 693 | bool anon, int nr_pages) |
d52aa412 | 694 | { |
c62b1a3b KH |
695 | preempt_disable(); |
696 | ||
b2402857 KH |
697 | /* |
698 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
699 | * counted as CACHE even if it's on ANON LRU. | |
700 | */ | |
701 | if (anon) | |
702 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], | |
c0ff4b85 | 703 | nr_pages); |
d52aa412 | 704 | else |
b2402857 | 705 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 706 | nr_pages); |
55e462b0 | 707 | |
e401f176 KH |
708 | /* pagein of a big page is an event. So, ignore page size */ |
709 | if (nr_pages > 0) | |
c0ff4b85 | 710 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 711 | else { |
c0ff4b85 | 712 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
713 | nr_pages = -nr_pages; /* for event */ |
714 | } | |
e401f176 | 715 | |
c0ff4b85 | 716 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_COUNT], nr_pages); |
2e72b634 | 717 | |
c62b1a3b | 718 | preempt_enable(); |
6d12e2d8 KH |
719 | } |
720 | ||
bb2a0de9 | 721 | unsigned long |
c0ff4b85 | 722 | mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid, |
bb2a0de9 | 723 | unsigned int lru_mask) |
889976db YH |
724 | { |
725 | struct mem_cgroup_per_zone *mz; | |
f156ab93 | 726 | enum lru_list lru; |
bb2a0de9 KH |
727 | unsigned long ret = 0; |
728 | ||
c0ff4b85 | 729 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
bb2a0de9 | 730 | |
f156ab93 HD |
731 | for_each_lru(lru) { |
732 | if (BIT(lru) & lru_mask) | |
733 | ret += mz->lru_size[lru]; | |
bb2a0de9 KH |
734 | } |
735 | return ret; | |
736 | } | |
737 | ||
738 | static unsigned long | |
c0ff4b85 | 739 | mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 KH |
740 | int nid, unsigned int lru_mask) |
741 | { | |
889976db YH |
742 | u64 total = 0; |
743 | int zid; | |
744 | ||
bb2a0de9 | 745 | for (zid = 0; zid < MAX_NR_ZONES; zid++) |
c0ff4b85 R |
746 | total += mem_cgroup_zone_nr_lru_pages(memcg, |
747 | nid, zid, lru_mask); | |
bb2a0de9 | 748 | |
889976db YH |
749 | return total; |
750 | } | |
bb2a0de9 | 751 | |
c0ff4b85 | 752 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 753 | unsigned int lru_mask) |
6d12e2d8 | 754 | { |
889976db | 755 | int nid; |
6d12e2d8 KH |
756 | u64 total = 0; |
757 | ||
bb2a0de9 | 758 | for_each_node_state(nid, N_HIGH_MEMORY) |
c0ff4b85 | 759 | total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
6d12e2d8 | 760 | return total; |
d52aa412 KH |
761 | } |
762 | ||
f53d7ce3 JW |
763 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
764 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
765 | { |
766 | unsigned long val, next; | |
767 | ||
4799401f SR |
768 | val = __this_cpu_read(memcg->stat->events[MEM_CGROUP_EVENTS_COUNT]); |
769 | next = __this_cpu_read(memcg->stat->targets[target]); | |
7a159cc9 | 770 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
771 | if ((long)next - (long)val < 0) { |
772 | switch (target) { | |
773 | case MEM_CGROUP_TARGET_THRESH: | |
774 | next = val + THRESHOLDS_EVENTS_TARGET; | |
775 | break; | |
776 | case MEM_CGROUP_TARGET_SOFTLIMIT: | |
777 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
778 | break; | |
779 | case MEM_CGROUP_TARGET_NUMAINFO: | |
780 | next = val + NUMAINFO_EVENTS_TARGET; | |
781 | break; | |
782 | default: | |
783 | break; | |
784 | } | |
785 | __this_cpu_write(memcg->stat->targets[target], next); | |
786 | return true; | |
7a159cc9 | 787 | } |
f53d7ce3 | 788 | return false; |
d2265e6f KH |
789 | } |
790 | ||
791 | /* | |
792 | * Check events in order. | |
793 | * | |
794 | */ | |
c0ff4b85 | 795 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f | 796 | { |
4799401f | 797 | preempt_disable(); |
d2265e6f | 798 | /* threshold event is triggered in finer grain than soft limit */ |
f53d7ce3 JW |
799 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
800 | MEM_CGROUP_TARGET_THRESH))) { | |
82b3f2a7 AM |
801 | bool do_softlimit; |
802 | bool do_numainfo __maybe_unused; | |
f53d7ce3 JW |
803 | |
804 | do_softlimit = mem_cgroup_event_ratelimit(memcg, | |
805 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
806 | #if MAX_NUMNODES > 1 | |
807 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
808 | MEM_CGROUP_TARGET_NUMAINFO); | |
809 | #endif | |
810 | preempt_enable(); | |
811 | ||
c0ff4b85 | 812 | mem_cgroup_threshold(memcg); |
f53d7ce3 | 813 | if (unlikely(do_softlimit)) |
c0ff4b85 | 814 | mem_cgroup_update_tree(memcg, page); |
453a9bf3 | 815 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 816 | if (unlikely(do_numainfo)) |
c0ff4b85 | 817 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 818 | #endif |
f53d7ce3 JW |
819 | } else |
820 | preempt_enable(); | |
d2265e6f KH |
821 | } |
822 | ||
d1a4c0b3 | 823 | struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
824 | { |
825 | return container_of(cgroup_subsys_state(cont, | |
826 | mem_cgroup_subsys_id), struct mem_cgroup, | |
827 | css); | |
828 | } | |
829 | ||
cf475ad2 | 830 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 831 | { |
31a78f23 BS |
832 | /* |
833 | * mm_update_next_owner() may clear mm->owner to NULL | |
834 | * if it races with swapoff, page migration, etc. | |
835 | * So this can be called with p == NULL. | |
836 | */ | |
837 | if (unlikely(!p)) | |
838 | return NULL; | |
839 | ||
78fb7466 PE |
840 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
841 | struct mem_cgroup, css); | |
842 | } | |
843 | ||
a433658c | 844 | struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 845 | { |
c0ff4b85 | 846 | struct mem_cgroup *memcg = NULL; |
0b7f569e KH |
847 | |
848 | if (!mm) | |
849 | return NULL; | |
54595fe2 KH |
850 | /* |
851 | * Because we have no locks, mm->owner's may be being moved to other | |
852 | * cgroup. We use css_tryget() here even if this looks | |
853 | * pessimistic (rather than adding locks here). | |
854 | */ | |
855 | rcu_read_lock(); | |
856 | do { | |
c0ff4b85 R |
857 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
858 | if (unlikely(!memcg)) | |
54595fe2 | 859 | break; |
c0ff4b85 | 860 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 861 | rcu_read_unlock(); |
c0ff4b85 | 862 | return memcg; |
54595fe2 KH |
863 | } |
864 | ||
5660048c JW |
865 | /** |
866 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
867 | * @root: hierarchy root | |
868 | * @prev: previously returned memcg, NULL on first invocation | |
869 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
870 | * | |
871 | * Returns references to children of the hierarchy below @root, or | |
872 | * @root itself, or %NULL after a full round-trip. | |
873 | * | |
874 | * Caller must pass the return value in @prev on subsequent | |
875 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
876 | * to cancel a hierarchy walk before the round-trip is complete. | |
877 | * | |
878 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
879 | * divide up the memcgs in the hierarchy among all concurrent | |
880 | * reclaimers operating on the same zone and priority. | |
881 | */ | |
882 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, | |
883 | struct mem_cgroup *prev, | |
884 | struct mem_cgroup_reclaim_cookie *reclaim) | |
14067bb3 | 885 | { |
9f3a0d09 JW |
886 | struct mem_cgroup *memcg = NULL; |
887 | int id = 0; | |
711d3d2c | 888 | |
5660048c JW |
889 | if (mem_cgroup_disabled()) |
890 | return NULL; | |
891 | ||
9f3a0d09 JW |
892 | if (!root) |
893 | root = root_mem_cgroup; | |
7d74b06f | 894 | |
9f3a0d09 JW |
895 | if (prev && !reclaim) |
896 | id = css_id(&prev->css); | |
14067bb3 | 897 | |
9f3a0d09 JW |
898 | if (prev && prev != root) |
899 | css_put(&prev->css); | |
14067bb3 | 900 | |
9f3a0d09 JW |
901 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
902 | if (prev) | |
903 | return NULL; | |
904 | return root; | |
905 | } | |
14067bb3 | 906 | |
9f3a0d09 | 907 | while (!memcg) { |
527a5ec9 | 908 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
9f3a0d09 | 909 | struct cgroup_subsys_state *css; |
711d3d2c | 910 | |
527a5ec9 JW |
911 | if (reclaim) { |
912 | int nid = zone_to_nid(reclaim->zone); | |
913 | int zid = zone_idx(reclaim->zone); | |
914 | struct mem_cgroup_per_zone *mz; | |
915 | ||
916 | mz = mem_cgroup_zoneinfo(root, nid, zid); | |
917 | iter = &mz->reclaim_iter[reclaim->priority]; | |
918 | if (prev && reclaim->generation != iter->generation) | |
919 | return NULL; | |
920 | id = iter->position; | |
921 | } | |
7d74b06f | 922 | |
9f3a0d09 JW |
923 | rcu_read_lock(); |
924 | css = css_get_next(&mem_cgroup_subsys, id + 1, &root->css, &id); | |
925 | if (css) { | |
926 | if (css == &root->css || css_tryget(css)) | |
927 | memcg = container_of(css, | |
928 | struct mem_cgroup, css); | |
929 | } else | |
930 | id = 0; | |
14067bb3 | 931 | rcu_read_unlock(); |
14067bb3 | 932 | |
527a5ec9 JW |
933 | if (reclaim) { |
934 | iter->position = id; | |
935 | if (!css) | |
936 | iter->generation++; | |
937 | else if (!prev && memcg) | |
938 | reclaim->generation = iter->generation; | |
939 | } | |
9f3a0d09 JW |
940 | |
941 | if (prev && !css) | |
942 | return NULL; | |
943 | } | |
944 | return memcg; | |
14067bb3 | 945 | } |
7d74b06f | 946 | |
5660048c JW |
947 | /** |
948 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
949 | * @root: hierarchy root | |
950 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
951 | */ | |
952 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
953 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
954 | { |
955 | if (!root) | |
956 | root = root_mem_cgroup; | |
957 | if (prev && prev != root) | |
958 | css_put(&prev->css); | |
959 | } | |
7d74b06f | 960 | |
9f3a0d09 JW |
961 | /* |
962 | * Iteration constructs for visiting all cgroups (under a tree). If | |
963 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
964 | * be used for reference counting. | |
965 | */ | |
966 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 967 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 968 | iter != NULL; \ |
527a5ec9 | 969 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 970 | |
9f3a0d09 | 971 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 972 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 973 | iter != NULL; \ |
527a5ec9 | 974 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 975 | |
c0ff4b85 | 976 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
4b3bde4c | 977 | { |
c0ff4b85 | 978 | return (memcg == root_mem_cgroup); |
4b3bde4c BS |
979 | } |
980 | ||
456f998e YH |
981 | void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
982 | { | |
c0ff4b85 | 983 | struct mem_cgroup *memcg; |
456f998e YH |
984 | |
985 | if (!mm) | |
986 | return; | |
987 | ||
988 | rcu_read_lock(); | |
c0ff4b85 R |
989 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
990 | if (unlikely(!memcg)) | |
456f998e YH |
991 | goto out; |
992 | ||
993 | switch (idx) { | |
456f998e | 994 | case PGFAULT: |
0e574a93 JW |
995 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
996 | break; | |
997 | case PGMAJFAULT: | |
998 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
999 | break; |
1000 | default: | |
1001 | BUG(); | |
1002 | } | |
1003 | out: | |
1004 | rcu_read_unlock(); | |
1005 | } | |
1006 | EXPORT_SYMBOL(mem_cgroup_count_vm_event); | |
1007 | ||
925b7673 JW |
1008 | /** |
1009 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1010 | * @zone: zone of the wanted lruvec | |
1011 | * @mem: memcg of the wanted lruvec | |
1012 | * | |
1013 | * Returns the lru list vector holding pages for the given @zone and | |
1014 | * @mem. This can be the global zone lruvec, if the memory controller | |
1015 | * is disabled. | |
1016 | */ | |
1017 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1018 | struct mem_cgroup *memcg) | |
1019 | { | |
1020 | struct mem_cgroup_per_zone *mz; | |
1021 | ||
1022 | if (mem_cgroup_disabled()) | |
1023 | return &zone->lruvec; | |
1024 | ||
1025 | mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone)); | |
1026 | return &mz->lruvec; | |
1027 | } | |
1028 | ||
08e552c6 KH |
1029 | /* |
1030 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
1031 | * Operations are called by routine of global LRU independently from memcg. | |
1032 | * What we have to take care of here is validness of pc->mem_cgroup. | |
1033 | * | |
1034 | * Changes to pc->mem_cgroup happens when | |
1035 | * 1. charge | |
1036 | * 2. moving account | |
1037 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
1038 | * It is added to LRU before charge. | |
1039 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
1040 | * When moving account, the page is not on LRU. It's isolated. | |
1041 | */ | |
4f98a2fe | 1042 | |
925b7673 JW |
1043 | /** |
1044 | * mem_cgroup_lru_add_list - account for adding an lru page and return lruvec | |
1045 | * @zone: zone of the page | |
1046 | * @page: the page | |
1047 | * @lru: current lru | |
1048 | * | |
1049 | * This function accounts for @page being added to @lru, and returns | |
1050 | * the lruvec for the given @zone and the memcg @page is charged to. | |
1051 | * | |
1052 | * The callsite is then responsible for physically linking the page to | |
1053 | * the returned lruvec->lists[@lru]. | |
1054 | */ | |
1055 | struct lruvec *mem_cgroup_lru_add_list(struct zone *zone, struct page *page, | |
1056 | enum lru_list lru) | |
08e552c6 | 1057 | { |
08e552c6 | 1058 | struct mem_cgroup_per_zone *mz; |
925b7673 JW |
1059 | struct mem_cgroup *memcg; |
1060 | struct page_cgroup *pc; | |
6d12e2d8 | 1061 | |
f8d66542 | 1062 | if (mem_cgroup_disabled()) |
925b7673 JW |
1063 | return &zone->lruvec; |
1064 | ||
08e552c6 | 1065 | pc = lookup_page_cgroup(page); |
38c5d72f | 1066 | memcg = pc->mem_cgroup; |
7512102c HD |
1067 | |
1068 | /* | |
1069 | * Surreptitiously switch any uncharged page to root: | |
1070 | * an uncharged page off lru does nothing to secure | |
1071 | * its former mem_cgroup from sudden removal. | |
1072 | * | |
1073 | * Our caller holds lru_lock, and PageCgroupUsed is updated | |
1074 | * under page_cgroup lock: between them, they make all uses | |
1075 | * of pc->mem_cgroup safe. | |
1076 | */ | |
1077 | if (!PageCgroupUsed(pc) && memcg != root_mem_cgroup) | |
1078 | pc->mem_cgroup = memcg = root_mem_cgroup; | |
1079 | ||
925b7673 JW |
1080 | mz = page_cgroup_zoneinfo(memcg, page); |
1081 | /* compound_order() is stabilized through lru_lock */ | |
1eb49272 | 1082 | mz->lru_size[lru] += 1 << compound_order(page); |
925b7673 | 1083 | return &mz->lruvec; |
08e552c6 | 1084 | } |
b69408e8 | 1085 | |
925b7673 JW |
1086 | /** |
1087 | * mem_cgroup_lru_del_list - account for removing an lru page | |
1088 | * @page: the page | |
1089 | * @lru: target lru | |
1090 | * | |
1091 | * This function accounts for @page being removed from @lru. | |
1092 | * | |
1093 | * The callsite is then responsible for physically unlinking | |
1094 | * @page->lru. | |
3f58a829 | 1095 | */ |
925b7673 | 1096 | void mem_cgroup_lru_del_list(struct page *page, enum lru_list lru) |
3f58a829 MK |
1097 | { |
1098 | struct mem_cgroup_per_zone *mz; | |
925b7673 | 1099 | struct mem_cgroup *memcg; |
3f58a829 | 1100 | struct page_cgroup *pc; |
3f58a829 MK |
1101 | |
1102 | if (mem_cgroup_disabled()) | |
1103 | return; | |
1104 | ||
1105 | pc = lookup_page_cgroup(page); | |
38c5d72f KH |
1106 | memcg = pc->mem_cgroup; |
1107 | VM_BUG_ON(!memcg); | |
925b7673 JW |
1108 | mz = page_cgroup_zoneinfo(memcg, page); |
1109 | /* huge page split is done under lru_lock. so, we have no races. */ | |
1eb49272 HD |
1110 | VM_BUG_ON(mz->lru_size[lru] < (1 << compound_order(page))); |
1111 | mz->lru_size[lru] -= 1 << compound_order(page); | |
3f58a829 MK |
1112 | } |
1113 | ||
925b7673 | 1114 | void mem_cgroup_lru_del(struct page *page) |
08e552c6 | 1115 | { |
925b7673 | 1116 | mem_cgroup_lru_del_list(page, page_lru(page)); |
6d12e2d8 KH |
1117 | } |
1118 | ||
925b7673 JW |
1119 | /** |
1120 | * mem_cgroup_lru_move_lists - account for moving a page between lrus | |
1121 | * @zone: zone of the page | |
1122 | * @page: the page | |
1123 | * @from: current lru | |
1124 | * @to: target lru | |
1125 | * | |
1126 | * This function accounts for @page being moved between the lrus @from | |
1127 | * and @to, and returns the lruvec for the given @zone and the memcg | |
1128 | * @page is charged to. | |
1129 | * | |
1130 | * The callsite is then responsible for physically relinking | |
1131 | * @page->lru to the returned lruvec->lists[@to]. | |
1132 | */ | |
1133 | struct lruvec *mem_cgroup_lru_move_lists(struct zone *zone, | |
1134 | struct page *page, | |
1135 | enum lru_list from, | |
1136 | enum lru_list to) | |
66e1707b | 1137 | { |
925b7673 JW |
1138 | /* XXX: Optimize this, especially for @from == @to */ |
1139 | mem_cgroup_lru_del_list(page, from); | |
1140 | return mem_cgroup_lru_add_list(zone, page, to); | |
08e552c6 | 1141 | } |
544122e5 | 1142 | |
3e92041d | 1143 | /* |
c0ff4b85 | 1144 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1145 | * hierarchy subtree |
1146 | */ | |
c0ff4b85 R |
1147 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1148 | struct mem_cgroup *memcg) | |
3e92041d | 1149 | { |
c0ff4b85 R |
1150 | if (root_memcg != memcg) { |
1151 | return (root_memcg->use_hierarchy && | |
1152 | css_is_ancestor(&memcg->css, &root_memcg->css)); | |
3e92041d MH |
1153 | } |
1154 | ||
1155 | return true; | |
1156 | } | |
1157 | ||
c0ff4b85 | 1158 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg) |
4c4a2214 DR |
1159 | { |
1160 | int ret; | |
0b7f569e | 1161 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1162 | struct task_struct *p; |
4c4a2214 | 1163 | |
158e0a2d | 1164 | p = find_lock_task_mm(task); |
de077d22 DR |
1165 | if (p) { |
1166 | curr = try_get_mem_cgroup_from_mm(p->mm); | |
1167 | task_unlock(p); | |
1168 | } else { | |
1169 | /* | |
1170 | * All threads may have already detached their mm's, but the oom | |
1171 | * killer still needs to detect if they have already been oom | |
1172 | * killed to prevent needlessly killing additional tasks. | |
1173 | */ | |
1174 | task_lock(task); | |
1175 | curr = mem_cgroup_from_task(task); | |
1176 | if (curr) | |
1177 | css_get(&curr->css); | |
1178 | task_unlock(task); | |
1179 | } | |
0b7f569e KH |
1180 | if (!curr) |
1181 | return 0; | |
d31f56db | 1182 | /* |
c0ff4b85 | 1183 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1184 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1185 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1186 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1187 | */ |
c0ff4b85 | 1188 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1189 | css_put(&curr->css); |
4c4a2214 DR |
1190 | return ret; |
1191 | } | |
1192 | ||
9b272977 | 1193 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg, struct zone *zone) |
14797e23 | 1194 | { |
9b272977 JW |
1195 | unsigned long inactive_ratio; |
1196 | int nid = zone_to_nid(zone); | |
1197 | int zid = zone_idx(zone); | |
14797e23 | 1198 | unsigned long inactive; |
9b272977 | 1199 | unsigned long active; |
c772be93 | 1200 | unsigned long gb; |
14797e23 | 1201 | |
9b272977 JW |
1202 | inactive = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid, |
1203 | BIT(LRU_INACTIVE_ANON)); | |
1204 | active = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid, | |
1205 | BIT(LRU_ACTIVE_ANON)); | |
14797e23 | 1206 | |
c772be93 KM |
1207 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1208 | if (gb) | |
1209 | inactive_ratio = int_sqrt(10 * gb); | |
1210 | else | |
1211 | inactive_ratio = 1; | |
1212 | ||
9b272977 | 1213 | return inactive * inactive_ratio < active; |
14797e23 KM |
1214 | } |
1215 | ||
9b272977 | 1216 | int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg, struct zone *zone) |
56e49d21 RR |
1217 | { |
1218 | unsigned long active; | |
1219 | unsigned long inactive; | |
9b272977 JW |
1220 | int zid = zone_idx(zone); |
1221 | int nid = zone_to_nid(zone); | |
56e49d21 | 1222 | |
9b272977 JW |
1223 | inactive = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid, |
1224 | BIT(LRU_INACTIVE_FILE)); | |
1225 | active = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid, | |
1226 | BIT(LRU_ACTIVE_FILE)); | |
56e49d21 RR |
1227 | |
1228 | return (active > inactive); | |
1229 | } | |
1230 | ||
3e2f41f1 KM |
1231 | struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, |
1232 | struct zone *zone) | |
1233 | { | |
13d7e3a2 | 1234 | int nid = zone_to_nid(zone); |
3e2f41f1 KM |
1235 | int zid = zone_idx(zone); |
1236 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
1237 | ||
1238 | return &mz->reclaim_stat; | |
1239 | } | |
1240 | ||
1241 | struct zone_reclaim_stat * | |
1242 | mem_cgroup_get_reclaim_stat_from_page(struct page *page) | |
1243 | { | |
1244 | struct page_cgroup *pc; | |
1245 | struct mem_cgroup_per_zone *mz; | |
1246 | ||
1247 | if (mem_cgroup_disabled()) | |
1248 | return NULL; | |
1249 | ||
1250 | pc = lookup_page_cgroup(page); | |
bd112db8 DN |
1251 | if (!PageCgroupUsed(pc)) |
1252 | return NULL; | |
713735b4 JW |
1253 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ |
1254 | smp_rmb(); | |
97a6c37b | 1255 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
3e2f41f1 KM |
1256 | return &mz->reclaim_stat; |
1257 | } | |
1258 | ||
6d61ef40 BS |
1259 | #define mem_cgroup_from_res_counter(counter, member) \ |
1260 | container_of(counter, struct mem_cgroup, member) | |
1261 | ||
19942822 | 1262 | /** |
9d11ea9f JW |
1263 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
1264 | * @mem: the memory cgroup | |
19942822 | 1265 | * |
9d11ea9f | 1266 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1267 | * pages. |
19942822 | 1268 | */ |
c0ff4b85 | 1269 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1270 | { |
9d11ea9f JW |
1271 | unsigned long long margin; |
1272 | ||
c0ff4b85 | 1273 | margin = res_counter_margin(&memcg->res); |
9d11ea9f | 1274 | if (do_swap_account) |
c0ff4b85 | 1275 | margin = min(margin, res_counter_margin(&memcg->memsw)); |
7ec99d62 | 1276 | return margin >> PAGE_SHIFT; |
19942822 JW |
1277 | } |
1278 | ||
1f4c025b | 1279 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 KM |
1280 | { |
1281 | struct cgroup *cgrp = memcg->css.cgroup; | |
a7885eb8 KM |
1282 | |
1283 | /* root ? */ | |
1284 | if (cgrp->parent == NULL) | |
1285 | return vm_swappiness; | |
1286 | ||
bf1ff263 | 1287 | return memcg->swappiness; |
a7885eb8 KM |
1288 | } |
1289 | ||
c0ff4b85 | 1290 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a KH |
1291 | { |
1292 | int cpu; | |
1489ebad KH |
1293 | |
1294 | get_online_cpus(); | |
c0ff4b85 | 1295 | spin_lock(&memcg->pcp_counter_lock); |
1489ebad | 1296 | for_each_online_cpu(cpu) |
c0ff4b85 R |
1297 | per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) += 1; |
1298 | memcg->nocpu_base.count[MEM_CGROUP_ON_MOVE] += 1; | |
1299 | spin_unlock(&memcg->pcp_counter_lock); | |
1489ebad | 1300 | put_online_cpus(); |
32047e2a KH |
1301 | |
1302 | synchronize_rcu(); | |
1303 | } | |
1304 | ||
c0ff4b85 | 1305 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a KH |
1306 | { |
1307 | int cpu; | |
1308 | ||
c0ff4b85 | 1309 | if (!memcg) |
32047e2a | 1310 | return; |
1489ebad | 1311 | get_online_cpus(); |
c0ff4b85 | 1312 | spin_lock(&memcg->pcp_counter_lock); |
1489ebad | 1313 | for_each_online_cpu(cpu) |
c0ff4b85 R |
1314 | per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) -= 1; |
1315 | memcg->nocpu_base.count[MEM_CGROUP_ON_MOVE] -= 1; | |
1316 | spin_unlock(&memcg->pcp_counter_lock); | |
1489ebad | 1317 | put_online_cpus(); |
32047e2a KH |
1318 | } |
1319 | /* | |
1320 | * 2 routines for checking "mem" is under move_account() or not. | |
1321 | * | |
1322 | * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used | |
1323 | * for avoiding race in accounting. If true, | |
1324 | * pc->mem_cgroup may be overwritten. | |
1325 | * | |
1326 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or | |
1327 | * under hierarchy of moving cgroups. This is for | |
1328 | * waiting at hith-memory prressure caused by "move". | |
1329 | */ | |
1330 | ||
c0ff4b85 | 1331 | static bool mem_cgroup_stealed(struct mem_cgroup *memcg) |
32047e2a KH |
1332 | { |
1333 | VM_BUG_ON(!rcu_read_lock_held()); | |
c0ff4b85 | 1334 | return this_cpu_read(memcg->stat->count[MEM_CGROUP_ON_MOVE]) > 0; |
32047e2a | 1335 | } |
4b534334 | 1336 | |
c0ff4b85 | 1337 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1338 | { |
2bd9bb20 KH |
1339 | struct mem_cgroup *from; |
1340 | struct mem_cgroup *to; | |
4b534334 | 1341 | bool ret = false; |
2bd9bb20 KH |
1342 | /* |
1343 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1344 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1345 | */ | |
1346 | spin_lock(&mc.lock); | |
1347 | from = mc.from; | |
1348 | to = mc.to; | |
1349 | if (!from) | |
1350 | goto unlock; | |
3e92041d | 1351 | |
c0ff4b85 R |
1352 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1353 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1354 | unlock: |
1355 | spin_unlock(&mc.lock); | |
4b534334 KH |
1356 | return ret; |
1357 | } | |
1358 | ||
c0ff4b85 | 1359 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1360 | { |
1361 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1362 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1363 | DEFINE_WAIT(wait); |
1364 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1365 | /* moving charge context might have finished. */ | |
1366 | if (mc.moving_task) | |
1367 | schedule(); | |
1368 | finish_wait(&mc.waitq, &wait); | |
1369 | return true; | |
1370 | } | |
1371 | } | |
1372 | return false; | |
1373 | } | |
1374 | ||
e222432b | 1375 | /** |
6a6135b6 | 1376 | * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode. |
e222432b BS |
1377 | * @memcg: The memory cgroup that went over limit |
1378 | * @p: Task that is going to be killed | |
1379 | * | |
1380 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1381 | * enabled | |
1382 | */ | |
1383 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1384 | { | |
1385 | struct cgroup *task_cgrp; | |
1386 | struct cgroup *mem_cgrp; | |
1387 | /* | |
1388 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1389 | * on the assumption that OOM is serialized for memory controller. | |
1390 | * If this assumption is broken, revisit this code. | |
1391 | */ | |
1392 | static char memcg_name[PATH_MAX]; | |
1393 | int ret; | |
1394 | ||
d31f56db | 1395 | if (!memcg || !p) |
e222432b BS |
1396 | return; |
1397 | ||
e222432b BS |
1398 | rcu_read_lock(); |
1399 | ||
1400 | mem_cgrp = memcg->css.cgroup; | |
1401 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1402 | ||
1403 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1404 | if (ret < 0) { | |
1405 | /* | |
1406 | * Unfortunately, we are unable to convert to a useful name | |
1407 | * But we'll still print out the usage information | |
1408 | */ | |
1409 | rcu_read_unlock(); | |
1410 | goto done; | |
1411 | } | |
1412 | rcu_read_unlock(); | |
1413 | ||
1414 | printk(KERN_INFO "Task in %s killed", memcg_name); | |
1415 | ||
1416 | rcu_read_lock(); | |
1417 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1418 | if (ret < 0) { | |
1419 | rcu_read_unlock(); | |
1420 | goto done; | |
1421 | } | |
1422 | rcu_read_unlock(); | |
1423 | ||
1424 | /* | |
1425 | * Continues from above, so we don't need an KERN_ level | |
1426 | */ | |
1427 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); | |
1428 | done: | |
1429 | ||
1430 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", | |
1431 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, | |
1432 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1433 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
1434 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " | |
1435 | "failcnt %llu\n", | |
1436 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, | |
1437 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1438 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
1439 | } | |
1440 | ||
81d39c20 KH |
1441 | /* |
1442 | * This function returns the number of memcg under hierarchy tree. Returns | |
1443 | * 1(self count) if no children. | |
1444 | */ | |
c0ff4b85 | 1445 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1446 | { |
1447 | int num = 0; | |
7d74b06f KH |
1448 | struct mem_cgroup *iter; |
1449 | ||
c0ff4b85 | 1450 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1451 | num++; |
81d39c20 KH |
1452 | return num; |
1453 | } | |
1454 | ||
a63d83f4 DR |
1455 | /* |
1456 | * Return the memory (and swap, if configured) limit for a memcg. | |
1457 | */ | |
1458 | u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) | |
1459 | { | |
1460 | u64 limit; | |
1461 | u64 memsw; | |
1462 | ||
f3e8eb70 JW |
1463 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
1464 | limit += total_swap_pages << PAGE_SHIFT; | |
1465 | ||
a63d83f4 DR |
1466 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
1467 | /* | |
1468 | * If memsw is finite and limits the amount of swap space available | |
1469 | * to this memcg, return that limit. | |
1470 | */ | |
1471 | return min(limit, memsw); | |
1472 | } | |
1473 | ||
5660048c JW |
1474 | static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg, |
1475 | gfp_t gfp_mask, | |
1476 | unsigned long flags) | |
1477 | { | |
1478 | unsigned long total = 0; | |
1479 | bool noswap = false; | |
1480 | int loop; | |
1481 | ||
1482 | if (flags & MEM_CGROUP_RECLAIM_NOSWAP) | |
1483 | noswap = true; | |
1484 | if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum) | |
1485 | noswap = true; | |
1486 | ||
1487 | for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) { | |
1488 | if (loop) | |
1489 | drain_all_stock_async(memcg); | |
1490 | total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap); | |
1491 | /* | |
1492 | * Allow limit shrinkers, which are triggered directly | |
1493 | * by userspace, to catch signals and stop reclaim | |
1494 | * after minimal progress, regardless of the margin. | |
1495 | */ | |
1496 | if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK)) | |
1497 | break; | |
1498 | if (mem_cgroup_margin(memcg)) | |
1499 | break; | |
1500 | /* | |
1501 | * If nothing was reclaimed after two attempts, there | |
1502 | * may be no reclaimable pages in this hierarchy. | |
1503 | */ | |
1504 | if (loop && !total) | |
1505 | break; | |
1506 | } | |
1507 | return total; | |
1508 | } | |
1509 | ||
4d0c066d KH |
1510 | /** |
1511 | * test_mem_cgroup_node_reclaimable | |
1512 | * @mem: the target memcg | |
1513 | * @nid: the node ID to be checked. | |
1514 | * @noswap : specify true here if the user wants flle only information. | |
1515 | * | |
1516 | * This function returns whether the specified memcg contains any | |
1517 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1518 | * pages in the node. | |
1519 | */ | |
c0ff4b85 | 1520 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1521 | int nid, bool noswap) |
1522 | { | |
c0ff4b85 | 1523 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1524 | return true; |
1525 | if (noswap || !total_swap_pages) | |
1526 | return false; | |
c0ff4b85 | 1527 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1528 | return true; |
1529 | return false; | |
1530 | ||
1531 | } | |
889976db YH |
1532 | #if MAX_NUMNODES > 1 |
1533 | ||
1534 | /* | |
1535 | * Always updating the nodemask is not very good - even if we have an empty | |
1536 | * list or the wrong list here, we can start from some node and traverse all | |
1537 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1538 | * | |
1539 | */ | |
c0ff4b85 | 1540 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1541 | { |
1542 | int nid; | |
453a9bf3 KH |
1543 | /* |
1544 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1545 | * pagein/pageout changes since the last update. | |
1546 | */ | |
c0ff4b85 | 1547 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1548 | return; |
c0ff4b85 | 1549 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1550 | return; |
1551 | ||
889976db | 1552 | /* make a nodemask where this memcg uses memory from */ |
c0ff4b85 | 1553 | memcg->scan_nodes = node_states[N_HIGH_MEMORY]; |
889976db YH |
1554 | |
1555 | for_each_node_mask(nid, node_states[N_HIGH_MEMORY]) { | |
1556 | ||
c0ff4b85 R |
1557 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1558 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1559 | } |
453a9bf3 | 1560 | |
c0ff4b85 R |
1561 | atomic_set(&memcg->numainfo_events, 0); |
1562 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1563 | } |
1564 | ||
1565 | /* | |
1566 | * Selecting a node where we start reclaim from. Because what we need is just | |
1567 | * reducing usage counter, start from anywhere is O,K. Considering | |
1568 | * memory reclaim from current node, there are pros. and cons. | |
1569 | * | |
1570 | * Freeing memory from current node means freeing memory from a node which | |
1571 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1572 | * hit limits, it will see a contention on a node. But freeing from remote | |
1573 | * node means more costs for memory reclaim because of memory latency. | |
1574 | * | |
1575 | * Now, we use round-robin. Better algorithm is welcomed. | |
1576 | */ | |
c0ff4b85 | 1577 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1578 | { |
1579 | int node; | |
1580 | ||
c0ff4b85 R |
1581 | mem_cgroup_may_update_nodemask(memcg); |
1582 | node = memcg->last_scanned_node; | |
889976db | 1583 | |
c0ff4b85 | 1584 | node = next_node(node, memcg->scan_nodes); |
889976db | 1585 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1586 | node = first_node(memcg->scan_nodes); |
889976db YH |
1587 | /* |
1588 | * We call this when we hit limit, not when pages are added to LRU. | |
1589 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1590 | * memcg is too small and all pages are not on LRU. In that case, | |
1591 | * we use curret node. | |
1592 | */ | |
1593 | if (unlikely(node == MAX_NUMNODES)) | |
1594 | node = numa_node_id(); | |
1595 | ||
c0ff4b85 | 1596 | memcg->last_scanned_node = node; |
889976db YH |
1597 | return node; |
1598 | } | |
1599 | ||
4d0c066d KH |
1600 | /* |
1601 | * Check all nodes whether it contains reclaimable pages or not. | |
1602 | * For quick scan, we make use of scan_nodes. This will allow us to skip | |
1603 | * unused nodes. But scan_nodes is lazily updated and may not cotain | |
1604 | * enough new information. We need to do double check. | |
1605 | */ | |
c0ff4b85 | 1606 | bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d KH |
1607 | { |
1608 | int nid; | |
1609 | ||
1610 | /* | |
1611 | * quick check...making use of scan_node. | |
1612 | * We can skip unused nodes. | |
1613 | */ | |
c0ff4b85 R |
1614 | if (!nodes_empty(memcg->scan_nodes)) { |
1615 | for (nid = first_node(memcg->scan_nodes); | |
4d0c066d | 1616 | nid < MAX_NUMNODES; |
c0ff4b85 | 1617 | nid = next_node(nid, memcg->scan_nodes)) { |
4d0c066d | 1618 | |
c0ff4b85 | 1619 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
1620 | return true; |
1621 | } | |
1622 | } | |
1623 | /* | |
1624 | * Check rest of nodes. | |
1625 | */ | |
1626 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
c0ff4b85 | 1627 | if (node_isset(nid, memcg->scan_nodes)) |
4d0c066d | 1628 | continue; |
c0ff4b85 | 1629 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
1630 | return true; |
1631 | } | |
1632 | return false; | |
1633 | } | |
1634 | ||
889976db | 1635 | #else |
c0ff4b85 | 1636 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1637 | { |
1638 | return 0; | |
1639 | } | |
4d0c066d | 1640 | |
c0ff4b85 | 1641 | bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d | 1642 | { |
c0ff4b85 | 1643 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); |
4d0c066d | 1644 | } |
889976db YH |
1645 | #endif |
1646 | ||
5660048c JW |
1647 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1648 | struct zone *zone, | |
1649 | gfp_t gfp_mask, | |
1650 | unsigned long *total_scanned) | |
6d61ef40 | 1651 | { |
9f3a0d09 | 1652 | struct mem_cgroup *victim = NULL; |
5660048c | 1653 | int total = 0; |
04046e1a | 1654 | int loop = 0; |
9d11ea9f | 1655 | unsigned long excess; |
185efc0f | 1656 | unsigned long nr_scanned; |
527a5ec9 JW |
1657 | struct mem_cgroup_reclaim_cookie reclaim = { |
1658 | .zone = zone, | |
1659 | .priority = 0, | |
1660 | }; | |
9d11ea9f | 1661 | |
c0ff4b85 | 1662 | excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT; |
04046e1a | 1663 | |
4e416953 | 1664 | while (1) { |
527a5ec9 | 1665 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); |
9f3a0d09 | 1666 | if (!victim) { |
04046e1a | 1667 | loop++; |
4e416953 BS |
1668 | if (loop >= 2) { |
1669 | /* | |
1670 | * If we have not been able to reclaim | |
1671 | * anything, it might because there are | |
1672 | * no reclaimable pages under this hierarchy | |
1673 | */ | |
5660048c | 1674 | if (!total) |
4e416953 | 1675 | break; |
4e416953 | 1676 | /* |
25985edc | 1677 | * We want to do more targeted reclaim. |
4e416953 BS |
1678 | * excess >> 2 is not to excessive so as to |
1679 | * reclaim too much, nor too less that we keep | |
1680 | * coming back to reclaim from this cgroup | |
1681 | */ | |
1682 | if (total >= (excess >> 2) || | |
9f3a0d09 | 1683 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) |
4e416953 | 1684 | break; |
4e416953 | 1685 | } |
9f3a0d09 | 1686 | continue; |
4e416953 | 1687 | } |
5660048c | 1688 | if (!mem_cgroup_reclaimable(victim, false)) |
6d61ef40 | 1689 | continue; |
5660048c JW |
1690 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1691 | zone, &nr_scanned); | |
1692 | *total_scanned += nr_scanned; | |
1693 | if (!res_counter_soft_limit_excess(&root_memcg->res)) | |
9f3a0d09 | 1694 | break; |
6d61ef40 | 1695 | } |
9f3a0d09 | 1696 | mem_cgroup_iter_break(root_memcg, victim); |
04046e1a | 1697 | return total; |
6d61ef40 BS |
1698 | } |
1699 | ||
867578cb KH |
1700 | /* |
1701 | * Check OOM-Killer is already running under our hierarchy. | |
1702 | * If someone is running, return false. | |
1af8efe9 | 1703 | * Has to be called with memcg_oom_lock |
867578cb | 1704 | */ |
c0ff4b85 | 1705 | static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg) |
867578cb | 1706 | { |
79dfdacc | 1707 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1708 | |
9f3a0d09 | 1709 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1710 | if (iter->oom_lock) { |
79dfdacc MH |
1711 | /* |
1712 | * this subtree of our hierarchy is already locked | |
1713 | * so we cannot give a lock. | |
1714 | */ | |
79dfdacc | 1715 | failed = iter; |
9f3a0d09 JW |
1716 | mem_cgroup_iter_break(memcg, iter); |
1717 | break; | |
23751be0 JW |
1718 | } else |
1719 | iter->oom_lock = true; | |
7d74b06f | 1720 | } |
867578cb | 1721 | |
79dfdacc | 1722 | if (!failed) |
23751be0 | 1723 | return true; |
79dfdacc MH |
1724 | |
1725 | /* | |
1726 | * OK, we failed to lock the whole subtree so we have to clean up | |
1727 | * what we set up to the failing subtree | |
1728 | */ | |
9f3a0d09 | 1729 | for_each_mem_cgroup_tree(iter, memcg) { |
79dfdacc | 1730 | if (iter == failed) { |
9f3a0d09 JW |
1731 | mem_cgroup_iter_break(memcg, iter); |
1732 | break; | |
79dfdacc MH |
1733 | } |
1734 | iter->oom_lock = false; | |
1735 | } | |
23751be0 | 1736 | return false; |
a636b327 | 1737 | } |
0b7f569e | 1738 | |
79dfdacc | 1739 | /* |
1af8efe9 | 1740 | * Has to be called with memcg_oom_lock |
79dfdacc | 1741 | */ |
c0ff4b85 | 1742 | static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1743 | { |
7d74b06f KH |
1744 | struct mem_cgroup *iter; |
1745 | ||
c0ff4b85 | 1746 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1747 | iter->oom_lock = false; |
1748 | return 0; | |
1749 | } | |
1750 | ||
c0ff4b85 | 1751 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1752 | { |
1753 | struct mem_cgroup *iter; | |
1754 | ||
c0ff4b85 | 1755 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1756 | atomic_inc(&iter->under_oom); |
1757 | } | |
1758 | ||
c0ff4b85 | 1759 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1760 | { |
1761 | struct mem_cgroup *iter; | |
1762 | ||
867578cb KH |
1763 | /* |
1764 | * When a new child is created while the hierarchy is under oom, | |
1765 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1766 | * atomic_add_unless() here. | |
1767 | */ | |
c0ff4b85 | 1768 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1769 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
1770 | } |
1771 | ||
1af8efe9 | 1772 | static DEFINE_SPINLOCK(memcg_oom_lock); |
867578cb KH |
1773 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1774 | ||
dc98df5a | 1775 | struct oom_wait_info { |
d79154bb | 1776 | struct mem_cgroup *memcg; |
dc98df5a KH |
1777 | wait_queue_t wait; |
1778 | }; | |
1779 | ||
1780 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1781 | unsigned mode, int sync, void *arg) | |
1782 | { | |
d79154bb HD |
1783 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1784 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1785 | struct oom_wait_info *oom_wait_info; |
1786 | ||
1787 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1788 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1789 | |
dc98df5a | 1790 | /* |
d79154bb | 1791 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
dc98df5a KH |
1792 | * Then we can use css_is_ancestor without taking care of RCU. |
1793 | */ | |
c0ff4b85 R |
1794 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
1795 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 1796 | return 0; |
dc98df5a KH |
1797 | return autoremove_wake_function(wait, mode, sync, arg); |
1798 | } | |
1799 | ||
c0ff4b85 | 1800 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 1801 | { |
c0ff4b85 R |
1802 | /* for filtering, pass "memcg" as argument. */ |
1803 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
1804 | } |
1805 | ||
c0ff4b85 | 1806 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1807 | { |
c0ff4b85 R |
1808 | if (memcg && atomic_read(&memcg->under_oom)) |
1809 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
1810 | } |
1811 | ||
867578cb KH |
1812 | /* |
1813 | * try to call OOM killer. returns false if we should exit memory-reclaim loop. | |
1814 | */ | |
e845e199 | 1815 | bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 1816 | { |
dc98df5a | 1817 | struct oom_wait_info owait; |
3c11ecf4 | 1818 | bool locked, need_to_kill; |
867578cb | 1819 | |
d79154bb | 1820 | owait.memcg = memcg; |
dc98df5a KH |
1821 | owait.wait.flags = 0; |
1822 | owait.wait.func = memcg_oom_wake_function; | |
1823 | owait.wait.private = current; | |
1824 | INIT_LIST_HEAD(&owait.wait.task_list); | |
3c11ecf4 | 1825 | need_to_kill = true; |
c0ff4b85 | 1826 | mem_cgroup_mark_under_oom(memcg); |
79dfdacc | 1827 | |
c0ff4b85 | 1828 | /* At first, try to OOM lock hierarchy under memcg.*/ |
1af8efe9 | 1829 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1830 | locked = mem_cgroup_oom_lock(memcg); |
867578cb KH |
1831 | /* |
1832 | * Even if signal_pending(), we can't quit charge() loop without | |
1833 | * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL | |
1834 | * under OOM is always welcomed, use TASK_KILLABLE here. | |
1835 | */ | |
3c11ecf4 | 1836 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
c0ff4b85 | 1837 | if (!locked || memcg->oom_kill_disable) |
3c11ecf4 KH |
1838 | need_to_kill = false; |
1839 | if (locked) | |
c0ff4b85 | 1840 | mem_cgroup_oom_notify(memcg); |
1af8efe9 | 1841 | spin_unlock(&memcg_oom_lock); |
867578cb | 1842 | |
3c11ecf4 KH |
1843 | if (need_to_kill) { |
1844 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
e845e199 | 1845 | mem_cgroup_out_of_memory(memcg, mask, order); |
3c11ecf4 | 1846 | } else { |
867578cb | 1847 | schedule(); |
dc98df5a | 1848 | finish_wait(&memcg_oom_waitq, &owait.wait); |
867578cb | 1849 | } |
1af8efe9 | 1850 | spin_lock(&memcg_oom_lock); |
79dfdacc | 1851 | if (locked) |
c0ff4b85 R |
1852 | mem_cgroup_oom_unlock(memcg); |
1853 | memcg_wakeup_oom(memcg); | |
1af8efe9 | 1854 | spin_unlock(&memcg_oom_lock); |
867578cb | 1855 | |
c0ff4b85 | 1856 | mem_cgroup_unmark_under_oom(memcg); |
79dfdacc | 1857 | |
867578cb KH |
1858 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) |
1859 | return false; | |
1860 | /* Give chance to dying process */ | |
715a5ee8 | 1861 | schedule_timeout_uninterruptible(1); |
867578cb | 1862 | return true; |
0b7f569e KH |
1863 | } |
1864 | ||
d69b042f BS |
1865 | /* |
1866 | * Currently used to update mapped file statistics, but the routine can be | |
1867 | * generalized to update other statistics as well. | |
32047e2a KH |
1868 | * |
1869 | * Notes: Race condition | |
1870 | * | |
1871 | * We usually use page_cgroup_lock() for accessing page_cgroup member but | |
1872 | * it tends to be costly. But considering some conditions, we doesn't need | |
1873 | * to do so _always_. | |
1874 | * | |
1875 | * Considering "charge", lock_page_cgroup() is not required because all | |
1876 | * file-stat operations happen after a page is attached to radix-tree. There | |
1877 | * are no race with "charge". | |
1878 | * | |
1879 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
1880 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
1881 | * if there are race with "uncharge". Statistics itself is properly handled | |
1882 | * by flags. | |
1883 | * | |
1884 | * Considering "move", this is an only case we see a race. To make the race | |
1885 | * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are | |
1886 | * possibility of race condition. If there is, we take a lock. | |
d69b042f | 1887 | */ |
26174efd | 1888 | |
2a7106f2 GT |
1889 | void mem_cgroup_update_page_stat(struct page *page, |
1890 | enum mem_cgroup_page_stat_item idx, int val) | |
d69b042f | 1891 | { |
c0ff4b85 | 1892 | struct mem_cgroup *memcg; |
32047e2a KH |
1893 | struct page_cgroup *pc = lookup_page_cgroup(page); |
1894 | bool need_unlock = false; | |
dbd4ea78 | 1895 | unsigned long uninitialized_var(flags); |
d69b042f | 1896 | |
cfa44946 | 1897 | if (mem_cgroup_disabled()) |
d69b042f BS |
1898 | return; |
1899 | ||
32047e2a | 1900 | rcu_read_lock(); |
c0ff4b85 R |
1901 | memcg = pc->mem_cgroup; |
1902 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
32047e2a KH |
1903 | goto out; |
1904 | /* pc->mem_cgroup is unstable ? */ | |
0e79dedd | 1905 | if (unlikely(mem_cgroup_stealed(memcg))) { |
32047e2a | 1906 | /* take a lock against to access pc->mem_cgroup */ |
dbd4ea78 | 1907 | move_lock_page_cgroup(pc, &flags); |
32047e2a | 1908 | need_unlock = true; |
c0ff4b85 R |
1909 | memcg = pc->mem_cgroup; |
1910 | if (!memcg || !PageCgroupUsed(pc)) | |
32047e2a KH |
1911 | goto out; |
1912 | } | |
26174efd | 1913 | |
26174efd | 1914 | switch (idx) { |
2a7106f2 | 1915 | case MEMCG_NR_FILE_MAPPED: |
26174efd KH |
1916 | if (val > 0) |
1917 | SetPageCgroupFileMapped(pc); | |
1918 | else if (!page_mapped(page)) | |
0c270f8f | 1919 | ClearPageCgroupFileMapped(pc); |
2a7106f2 | 1920 | idx = MEM_CGROUP_STAT_FILE_MAPPED; |
26174efd KH |
1921 | break; |
1922 | default: | |
1923 | BUG(); | |
8725d541 | 1924 | } |
d69b042f | 1925 | |
c0ff4b85 | 1926 | this_cpu_add(memcg->stat->count[idx], val); |
2a7106f2 | 1927 | |
32047e2a KH |
1928 | out: |
1929 | if (unlikely(need_unlock)) | |
dbd4ea78 | 1930 | move_unlock_page_cgroup(pc, &flags); |
32047e2a | 1931 | rcu_read_unlock(); |
d69b042f | 1932 | } |
26174efd | 1933 | |
cdec2e42 KH |
1934 | /* |
1935 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1936 | * TODO: maybe necessary to use big numbers in big irons. | |
1937 | */ | |
7ec99d62 | 1938 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
1939 | struct memcg_stock_pcp { |
1940 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 1941 | unsigned int nr_pages; |
cdec2e42 | 1942 | struct work_struct work; |
26fe6168 KH |
1943 | unsigned long flags; |
1944 | #define FLUSHING_CACHED_CHARGE (0) | |
cdec2e42 KH |
1945 | }; |
1946 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 1947 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 KH |
1948 | |
1949 | /* | |
11c9ea4e | 1950 | * Try to consume stocked charge on this cpu. If success, one page is consumed |
cdec2e42 KH |
1951 | * from local stock and true is returned. If the stock is 0 or charges from a |
1952 | * cgroup which is not current target, returns false. This stock will be | |
1953 | * refilled. | |
1954 | */ | |
c0ff4b85 | 1955 | static bool consume_stock(struct mem_cgroup *memcg) |
cdec2e42 KH |
1956 | { |
1957 | struct memcg_stock_pcp *stock; | |
1958 | bool ret = true; | |
1959 | ||
1960 | stock = &get_cpu_var(memcg_stock); | |
c0ff4b85 | 1961 | if (memcg == stock->cached && stock->nr_pages) |
11c9ea4e | 1962 | stock->nr_pages--; |
cdec2e42 KH |
1963 | else /* need to call res_counter_charge */ |
1964 | ret = false; | |
1965 | put_cpu_var(memcg_stock); | |
1966 | return ret; | |
1967 | } | |
1968 | ||
1969 | /* | |
1970 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
1971 | */ | |
1972 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1973 | { | |
1974 | struct mem_cgroup *old = stock->cached; | |
1975 | ||
11c9ea4e JW |
1976 | if (stock->nr_pages) { |
1977 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
1978 | ||
1979 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 1980 | if (do_swap_account) |
11c9ea4e JW |
1981 | res_counter_uncharge(&old->memsw, bytes); |
1982 | stock->nr_pages = 0; | |
cdec2e42 KH |
1983 | } |
1984 | stock->cached = NULL; | |
cdec2e42 KH |
1985 | } |
1986 | ||
1987 | /* | |
1988 | * This must be called under preempt disabled or must be called by | |
1989 | * a thread which is pinned to local cpu. | |
1990 | */ | |
1991 | static void drain_local_stock(struct work_struct *dummy) | |
1992 | { | |
1993 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
1994 | drain_stock(stock); | |
26fe6168 | 1995 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
1996 | } |
1997 | ||
1998 | /* | |
1999 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 2000 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2001 | */ |
c0ff4b85 | 2002 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2003 | { |
2004 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2005 | ||
c0ff4b85 | 2006 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2007 | drain_stock(stock); |
c0ff4b85 | 2008 | stock->cached = memcg; |
cdec2e42 | 2009 | } |
11c9ea4e | 2010 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2011 | put_cpu_var(memcg_stock); |
2012 | } | |
2013 | ||
2014 | /* | |
c0ff4b85 | 2015 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
d38144b7 MH |
2016 | * of the hierarchy under it. sync flag says whether we should block |
2017 | * until the work is done. | |
cdec2e42 | 2018 | */ |
c0ff4b85 | 2019 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
cdec2e42 | 2020 | { |
26fe6168 | 2021 | int cpu, curcpu; |
d38144b7 | 2022 | |
cdec2e42 | 2023 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2024 | get_online_cpus(); |
5af12d0e | 2025 | curcpu = get_cpu(); |
cdec2e42 KH |
2026 | for_each_online_cpu(cpu) { |
2027 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2028 | struct mem_cgroup *memcg; |
26fe6168 | 2029 | |
c0ff4b85 R |
2030 | memcg = stock->cached; |
2031 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2032 | continue; |
c0ff4b85 | 2033 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2034 | continue; |
d1a05b69 MH |
2035 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2036 | if (cpu == curcpu) | |
2037 | drain_local_stock(&stock->work); | |
2038 | else | |
2039 | schedule_work_on(cpu, &stock->work); | |
2040 | } | |
cdec2e42 | 2041 | } |
5af12d0e | 2042 | put_cpu(); |
d38144b7 MH |
2043 | |
2044 | if (!sync) | |
2045 | goto out; | |
2046 | ||
2047 | for_each_online_cpu(cpu) { | |
2048 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2049 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2050 | flush_work(&stock->work); |
2051 | } | |
2052 | out: | |
cdec2e42 | 2053 | put_online_cpus(); |
d38144b7 MH |
2054 | } |
2055 | ||
2056 | /* | |
2057 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2058 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
2059 | * expects some charges will be back to res_counter later but cannot wait for | |
2060 | * it. | |
2061 | */ | |
c0ff4b85 | 2062 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
d38144b7 | 2063 | { |
9f50fad6 MH |
2064 | /* |
2065 | * If someone calls draining, avoid adding more kworker runs. | |
2066 | */ | |
2067 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2068 | return; | |
c0ff4b85 | 2069 | drain_all_stock(root_memcg, false); |
9f50fad6 | 2070 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2071 | } |
2072 | ||
2073 | /* This is a synchronous drain interface. */ | |
c0ff4b85 | 2074 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
cdec2e42 KH |
2075 | { |
2076 | /* called when force_empty is called */ | |
9f50fad6 | 2077 | mutex_lock(&percpu_charge_mutex); |
c0ff4b85 | 2078 | drain_all_stock(root_memcg, true); |
9f50fad6 | 2079 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2080 | } |
2081 | ||
711d3d2c KH |
2082 | /* |
2083 | * This function drains percpu counter value from DEAD cpu and | |
2084 | * move it to local cpu. Note that this function can be preempted. | |
2085 | */ | |
c0ff4b85 | 2086 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2087 | { |
2088 | int i; | |
2089 | ||
c0ff4b85 | 2090 | spin_lock(&memcg->pcp_counter_lock); |
711d3d2c | 2091 | for (i = 0; i < MEM_CGROUP_STAT_DATA; i++) { |
c0ff4b85 | 2092 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2093 | |
c0ff4b85 R |
2094 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2095 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2096 | } |
e9f8974f | 2097 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2098 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2099 | |
c0ff4b85 R |
2100 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2101 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2102 | } |
1489ebad | 2103 | /* need to clear ON_MOVE value, works as a kind of lock. */ |
c0ff4b85 R |
2104 | per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) = 0; |
2105 | spin_unlock(&memcg->pcp_counter_lock); | |
1489ebad KH |
2106 | } |
2107 | ||
c0ff4b85 | 2108 | static void synchronize_mem_cgroup_on_move(struct mem_cgroup *memcg, int cpu) |
1489ebad KH |
2109 | { |
2110 | int idx = MEM_CGROUP_ON_MOVE; | |
2111 | ||
c0ff4b85 R |
2112 | spin_lock(&memcg->pcp_counter_lock); |
2113 | per_cpu(memcg->stat->count[idx], cpu) = memcg->nocpu_base.count[idx]; | |
2114 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
2115 | } |
2116 | ||
2117 | static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb, | |
cdec2e42 KH |
2118 | unsigned long action, |
2119 | void *hcpu) | |
2120 | { | |
2121 | int cpu = (unsigned long)hcpu; | |
2122 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2123 | struct mem_cgroup *iter; |
cdec2e42 | 2124 | |
1489ebad | 2125 | if ((action == CPU_ONLINE)) { |
9f3a0d09 | 2126 | for_each_mem_cgroup(iter) |
1489ebad KH |
2127 | synchronize_mem_cgroup_on_move(iter, cpu); |
2128 | return NOTIFY_OK; | |
2129 | } | |
2130 | ||
711d3d2c | 2131 | if ((action != CPU_DEAD) || action != CPU_DEAD_FROZEN) |
cdec2e42 | 2132 | return NOTIFY_OK; |
711d3d2c | 2133 | |
9f3a0d09 | 2134 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2135 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2136 | ||
cdec2e42 KH |
2137 | stock = &per_cpu(memcg_stock, cpu); |
2138 | drain_stock(stock); | |
2139 | return NOTIFY_OK; | |
2140 | } | |
2141 | ||
4b534334 KH |
2142 | |
2143 | /* See __mem_cgroup_try_charge() for details */ | |
2144 | enum { | |
2145 | CHARGE_OK, /* success */ | |
2146 | CHARGE_RETRY, /* need to retry but retry is not bad */ | |
2147 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ | |
2148 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ | |
2149 | CHARGE_OOM_DIE, /* the current is killed because of OOM */ | |
2150 | }; | |
2151 | ||
c0ff4b85 | 2152 | static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
7ec99d62 | 2153 | unsigned int nr_pages, bool oom_check) |
4b534334 | 2154 | { |
7ec99d62 | 2155 | unsigned long csize = nr_pages * PAGE_SIZE; |
4b534334 KH |
2156 | struct mem_cgroup *mem_over_limit; |
2157 | struct res_counter *fail_res; | |
2158 | unsigned long flags = 0; | |
2159 | int ret; | |
2160 | ||
c0ff4b85 | 2161 | ret = res_counter_charge(&memcg->res, csize, &fail_res); |
4b534334 KH |
2162 | |
2163 | if (likely(!ret)) { | |
2164 | if (!do_swap_account) | |
2165 | return CHARGE_OK; | |
c0ff4b85 | 2166 | ret = res_counter_charge(&memcg->memsw, csize, &fail_res); |
4b534334 KH |
2167 | if (likely(!ret)) |
2168 | return CHARGE_OK; | |
2169 | ||
c0ff4b85 | 2170 | res_counter_uncharge(&memcg->res, csize); |
4b534334 KH |
2171 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
2172 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
2173 | } else | |
2174 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
9221edb7 | 2175 | /* |
7ec99d62 JW |
2176 | * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch |
2177 | * of regular pages (CHARGE_BATCH), or a single regular page (1). | |
9221edb7 JW |
2178 | * |
2179 | * Never reclaim on behalf of optional batching, retry with a | |
2180 | * single page instead. | |
2181 | */ | |
7ec99d62 | 2182 | if (nr_pages == CHARGE_BATCH) |
4b534334 KH |
2183 | return CHARGE_RETRY; |
2184 | ||
2185 | if (!(gfp_mask & __GFP_WAIT)) | |
2186 | return CHARGE_WOULDBLOCK; | |
2187 | ||
5660048c | 2188 | ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags); |
7ec99d62 | 2189 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
19942822 | 2190 | return CHARGE_RETRY; |
4b534334 | 2191 | /* |
19942822 JW |
2192 | * Even though the limit is exceeded at this point, reclaim |
2193 | * may have been able to free some pages. Retry the charge | |
2194 | * before killing the task. | |
2195 | * | |
2196 | * Only for regular pages, though: huge pages are rather | |
2197 | * unlikely to succeed so close to the limit, and we fall back | |
2198 | * to regular pages anyway in case of failure. | |
4b534334 | 2199 | */ |
7ec99d62 | 2200 | if (nr_pages == 1 && ret) |
4b534334 KH |
2201 | return CHARGE_RETRY; |
2202 | ||
2203 | /* | |
2204 | * At task move, charge accounts can be doubly counted. So, it's | |
2205 | * better to wait until the end of task_move if something is going on. | |
2206 | */ | |
2207 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2208 | return CHARGE_RETRY; | |
2209 | ||
2210 | /* If we don't need to call oom-killer at el, return immediately */ | |
2211 | if (!oom_check) | |
2212 | return CHARGE_NOMEM; | |
2213 | /* check OOM */ | |
e845e199 | 2214 | if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask, get_order(csize))) |
4b534334 KH |
2215 | return CHARGE_OOM_DIE; |
2216 | ||
2217 | return CHARGE_RETRY; | |
2218 | } | |
2219 | ||
f817ed48 | 2220 | /* |
38c5d72f KH |
2221 | * __mem_cgroup_try_charge() does |
2222 | * 1. detect memcg to be charged against from passed *mm and *ptr, | |
2223 | * 2. update res_counter | |
2224 | * 3. call memory reclaim if necessary. | |
2225 | * | |
2226 | * In some special case, if the task is fatal, fatal_signal_pending() or | |
2227 | * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup | |
2228 | * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon | |
2229 | * as possible without any hazards. 2: all pages should have a valid | |
2230 | * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg | |
2231 | * pointer, that is treated as a charge to root_mem_cgroup. | |
2232 | * | |
2233 | * So __mem_cgroup_try_charge() will return | |
2234 | * 0 ... on success, filling *ptr with a valid memcg pointer. | |
2235 | * -ENOMEM ... charge failure because of resource limits. | |
2236 | * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup. | |
2237 | * | |
2238 | * Unlike the exported interface, an "oom" parameter is added. if oom==true, | |
2239 | * the oom-killer can be invoked. | |
8a9f3ccd | 2240 | */ |
f817ed48 | 2241 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
ec168510 | 2242 | gfp_t gfp_mask, |
7ec99d62 | 2243 | unsigned int nr_pages, |
c0ff4b85 | 2244 | struct mem_cgroup **ptr, |
7ec99d62 | 2245 | bool oom) |
8a9f3ccd | 2246 | { |
7ec99d62 | 2247 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
4b534334 | 2248 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
c0ff4b85 | 2249 | struct mem_cgroup *memcg = NULL; |
4b534334 | 2250 | int ret; |
a636b327 | 2251 | |
867578cb KH |
2252 | /* |
2253 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
2254 | * in system level. So, allow to go ahead dying process in addition to | |
2255 | * MEMDIE process. | |
2256 | */ | |
2257 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
2258 | || fatal_signal_pending(current))) | |
2259 | goto bypass; | |
a636b327 | 2260 | |
8a9f3ccd | 2261 | /* |
3be91277 HD |
2262 | * We always charge the cgroup the mm_struct belongs to. |
2263 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
2264 | * thread group leader migrates. It's possible that mm is not |
2265 | * set, if so charge the init_mm (happens for pagecache usage). | |
2266 | */ | |
c0ff4b85 | 2267 | if (!*ptr && !mm) |
38c5d72f | 2268 | *ptr = root_mem_cgroup; |
f75ca962 | 2269 | again: |
c0ff4b85 R |
2270 | if (*ptr) { /* css should be a valid one */ |
2271 | memcg = *ptr; | |
2272 | VM_BUG_ON(css_is_removed(&memcg->css)); | |
2273 | if (mem_cgroup_is_root(memcg)) | |
f75ca962 | 2274 | goto done; |
c0ff4b85 | 2275 | if (nr_pages == 1 && consume_stock(memcg)) |
f75ca962 | 2276 | goto done; |
c0ff4b85 | 2277 | css_get(&memcg->css); |
4b534334 | 2278 | } else { |
f75ca962 | 2279 | struct task_struct *p; |
54595fe2 | 2280 | |
f75ca962 KH |
2281 | rcu_read_lock(); |
2282 | p = rcu_dereference(mm->owner); | |
f75ca962 | 2283 | /* |
ebb76ce1 | 2284 | * Because we don't have task_lock(), "p" can exit. |
c0ff4b85 | 2285 | * In that case, "memcg" can point to root or p can be NULL with |
ebb76ce1 KH |
2286 | * race with swapoff. Then, we have small risk of mis-accouning. |
2287 | * But such kind of mis-account by race always happens because | |
2288 | * we don't have cgroup_mutex(). It's overkill and we allo that | |
2289 | * small race, here. | |
2290 | * (*) swapoff at el will charge against mm-struct not against | |
2291 | * task-struct. So, mm->owner can be NULL. | |
f75ca962 | 2292 | */ |
c0ff4b85 | 2293 | memcg = mem_cgroup_from_task(p); |
38c5d72f KH |
2294 | if (!memcg) |
2295 | memcg = root_mem_cgroup; | |
2296 | if (mem_cgroup_is_root(memcg)) { | |
f75ca962 KH |
2297 | rcu_read_unlock(); |
2298 | goto done; | |
2299 | } | |
c0ff4b85 | 2300 | if (nr_pages == 1 && consume_stock(memcg)) { |
f75ca962 KH |
2301 | /* |
2302 | * It seems dagerous to access memcg without css_get(). | |
2303 | * But considering how consume_stok works, it's not | |
2304 | * necessary. If consume_stock success, some charges | |
2305 | * from this memcg are cached on this cpu. So, we | |
2306 | * don't need to call css_get()/css_tryget() before | |
2307 | * calling consume_stock(). | |
2308 | */ | |
2309 | rcu_read_unlock(); | |
2310 | goto done; | |
2311 | } | |
2312 | /* after here, we may be blocked. we need to get refcnt */ | |
c0ff4b85 | 2313 | if (!css_tryget(&memcg->css)) { |
f75ca962 KH |
2314 | rcu_read_unlock(); |
2315 | goto again; | |
2316 | } | |
2317 | rcu_read_unlock(); | |
2318 | } | |
8a9f3ccd | 2319 | |
4b534334 KH |
2320 | do { |
2321 | bool oom_check; | |
7a81b88c | 2322 | |
4b534334 | 2323 | /* If killed, bypass charge */ |
f75ca962 | 2324 | if (fatal_signal_pending(current)) { |
c0ff4b85 | 2325 | css_put(&memcg->css); |
4b534334 | 2326 | goto bypass; |
f75ca962 | 2327 | } |
6d61ef40 | 2328 | |
4b534334 KH |
2329 | oom_check = false; |
2330 | if (oom && !nr_oom_retries) { | |
2331 | oom_check = true; | |
2332 | nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
cdec2e42 | 2333 | } |
66e1707b | 2334 | |
c0ff4b85 | 2335 | ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, oom_check); |
4b534334 KH |
2336 | switch (ret) { |
2337 | case CHARGE_OK: | |
2338 | break; | |
2339 | case CHARGE_RETRY: /* not in OOM situation but retry */ | |
7ec99d62 | 2340 | batch = nr_pages; |
c0ff4b85 R |
2341 | css_put(&memcg->css); |
2342 | memcg = NULL; | |
f75ca962 | 2343 | goto again; |
4b534334 | 2344 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
c0ff4b85 | 2345 | css_put(&memcg->css); |
4b534334 KH |
2346 | goto nomem; |
2347 | case CHARGE_NOMEM: /* OOM routine works */ | |
f75ca962 | 2348 | if (!oom) { |
c0ff4b85 | 2349 | css_put(&memcg->css); |
867578cb | 2350 | goto nomem; |
f75ca962 | 2351 | } |
4b534334 KH |
2352 | /* If oom, we never return -ENOMEM */ |
2353 | nr_oom_retries--; | |
2354 | break; | |
2355 | case CHARGE_OOM_DIE: /* Killed by OOM Killer */ | |
c0ff4b85 | 2356 | css_put(&memcg->css); |
867578cb | 2357 | goto bypass; |
66e1707b | 2358 | } |
4b534334 KH |
2359 | } while (ret != CHARGE_OK); |
2360 | ||
7ec99d62 | 2361 | if (batch > nr_pages) |
c0ff4b85 R |
2362 | refill_stock(memcg, batch - nr_pages); |
2363 | css_put(&memcg->css); | |
0c3e73e8 | 2364 | done: |
c0ff4b85 | 2365 | *ptr = memcg; |
7a81b88c KH |
2366 | return 0; |
2367 | nomem: | |
c0ff4b85 | 2368 | *ptr = NULL; |
7a81b88c | 2369 | return -ENOMEM; |
867578cb | 2370 | bypass: |
38c5d72f KH |
2371 | *ptr = root_mem_cgroup; |
2372 | return -EINTR; | |
7a81b88c | 2373 | } |
8a9f3ccd | 2374 | |
a3032a2c DN |
2375 | /* |
2376 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2377 | * This function is for that and do uncharge, put css's refcnt. | |
2378 | * gotten by try_charge(). | |
2379 | */ | |
c0ff4b85 | 2380 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg, |
e7018b8d | 2381 | unsigned int nr_pages) |
a3032a2c | 2382 | { |
c0ff4b85 | 2383 | if (!mem_cgroup_is_root(memcg)) { |
e7018b8d JW |
2384 | unsigned long bytes = nr_pages * PAGE_SIZE; |
2385 | ||
c0ff4b85 | 2386 | res_counter_uncharge(&memcg->res, bytes); |
a3032a2c | 2387 | if (do_swap_account) |
c0ff4b85 | 2388 | res_counter_uncharge(&memcg->memsw, bytes); |
a3032a2c | 2389 | } |
854ffa8d DN |
2390 | } |
2391 | ||
a3b2d692 KH |
2392 | /* |
2393 | * A helper function to get mem_cgroup from ID. must be called under | |
2394 | * rcu_read_lock(). The caller must check css_is_removed() or some if | |
2395 | * it's concern. (dropping refcnt from swap can be called against removed | |
2396 | * memcg.) | |
2397 | */ | |
2398 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2399 | { | |
2400 | struct cgroup_subsys_state *css; | |
2401 | ||
2402 | /* ID 0 is unused ID */ | |
2403 | if (!id) | |
2404 | return NULL; | |
2405 | css = css_lookup(&mem_cgroup_subsys, id); | |
2406 | if (!css) | |
2407 | return NULL; | |
2408 | return container_of(css, struct mem_cgroup, css); | |
2409 | } | |
2410 | ||
e42d9d5d | 2411 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2412 | { |
c0ff4b85 | 2413 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2414 | struct page_cgroup *pc; |
a3b2d692 | 2415 | unsigned short id; |
b5a84319 KH |
2416 | swp_entry_t ent; |
2417 | ||
3c776e64 DN |
2418 | VM_BUG_ON(!PageLocked(page)); |
2419 | ||
3c776e64 | 2420 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2421 | lock_page_cgroup(pc); |
a3b2d692 | 2422 | if (PageCgroupUsed(pc)) { |
c0ff4b85 R |
2423 | memcg = pc->mem_cgroup; |
2424 | if (memcg && !css_tryget(&memcg->css)) | |
2425 | memcg = NULL; | |
e42d9d5d | 2426 | } else if (PageSwapCache(page)) { |
3c776e64 | 2427 | ent.val = page_private(page); |
9fb4b7cc | 2428 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2429 | rcu_read_lock(); |
c0ff4b85 R |
2430 | memcg = mem_cgroup_lookup(id); |
2431 | if (memcg && !css_tryget(&memcg->css)) | |
2432 | memcg = NULL; | |
a3b2d692 | 2433 | rcu_read_unlock(); |
3c776e64 | 2434 | } |
c0bd3f63 | 2435 | unlock_page_cgroup(pc); |
c0ff4b85 | 2436 | return memcg; |
b5a84319 KH |
2437 | } |
2438 | ||
c0ff4b85 | 2439 | static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg, |
5564e88b | 2440 | struct page *page, |
7ec99d62 | 2441 | unsigned int nr_pages, |
ca3e0214 | 2442 | struct page_cgroup *pc, |
9ce70c02 HD |
2443 | enum charge_type ctype, |
2444 | bool lrucare) | |
7a81b88c | 2445 | { |
9ce70c02 HD |
2446 | struct zone *uninitialized_var(zone); |
2447 | bool was_on_lru = false; | |
b2402857 | 2448 | bool anon; |
9ce70c02 | 2449 | |
ca3e0214 KH |
2450 | lock_page_cgroup(pc); |
2451 | if (unlikely(PageCgroupUsed(pc))) { | |
2452 | unlock_page_cgroup(pc); | |
c0ff4b85 | 2453 | __mem_cgroup_cancel_charge(memcg, nr_pages); |
ca3e0214 KH |
2454 | return; |
2455 | } | |
2456 | /* | |
2457 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2458 | * accessed by any other context at this point. | |
2459 | */ | |
9ce70c02 HD |
2460 | |
2461 | /* | |
2462 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2463 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2464 | */ | |
2465 | if (lrucare) { | |
2466 | zone = page_zone(page); | |
2467 | spin_lock_irq(&zone->lru_lock); | |
2468 | if (PageLRU(page)) { | |
2469 | ClearPageLRU(page); | |
2470 | del_page_from_lru_list(zone, page, page_lru(page)); | |
2471 | was_on_lru = true; | |
2472 | } | |
2473 | } | |
2474 | ||
c0ff4b85 | 2475 | pc->mem_cgroup = memcg; |
261fb61a KH |
2476 | /* |
2477 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
2478 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
2479 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
2480 | * before USED bit, we need memory barrier here. | |
2481 | * See mem_cgroup_add_lru_list(), etc. | |
2482 | */ | |
08e552c6 | 2483 | smp_wmb(); |
b2402857 | 2484 | SetPageCgroupUsed(pc); |
3be91277 | 2485 | |
9ce70c02 HD |
2486 | if (lrucare) { |
2487 | if (was_on_lru) { | |
2488 | VM_BUG_ON(PageLRU(page)); | |
2489 | SetPageLRU(page); | |
2490 | add_page_to_lru_list(zone, page, page_lru(page)); | |
2491 | } | |
2492 | spin_unlock_irq(&zone->lru_lock); | |
2493 | } | |
2494 | ||
b2402857 KH |
2495 | if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED) |
2496 | anon = true; | |
2497 | else | |
2498 | anon = false; | |
2499 | ||
2500 | mem_cgroup_charge_statistics(memcg, anon, nr_pages); | |
52d4b9ac | 2501 | unlock_page_cgroup(pc); |
9ce70c02 | 2502 | |
430e4863 KH |
2503 | /* |
2504 | * "charge_statistics" updated event counter. Then, check it. | |
2505 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
2506 | * if they exceeds softlimit. | |
2507 | */ | |
c0ff4b85 | 2508 | memcg_check_events(memcg, page); |
7a81b88c | 2509 | } |
66e1707b | 2510 | |
ca3e0214 KH |
2511 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2512 | ||
2513 | #define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\ | |
38c5d72f | 2514 | (1 << PCG_MIGRATION)) |
ca3e0214 KH |
2515 | /* |
2516 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
2517 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
2518 | * charge/uncharge will be never happen and move_account() is done under | |
2519 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 2520 | */ |
e94c8a9c | 2521 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 KH |
2522 | { |
2523 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
e94c8a9c KH |
2524 | struct page_cgroup *pc; |
2525 | int i; | |
ca3e0214 | 2526 | |
3d37c4a9 KH |
2527 | if (mem_cgroup_disabled()) |
2528 | return; | |
e94c8a9c KH |
2529 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
2530 | pc = head_pc + i; | |
2531 | pc->mem_cgroup = head_pc->mem_cgroup; | |
2532 | smp_wmb();/* see __commit_charge() */ | |
e94c8a9c KH |
2533 | pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
2534 | } | |
ca3e0214 | 2535 | } |
12d27107 | 2536 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2537 | |
f817ed48 | 2538 | /** |
de3638d9 | 2539 | * mem_cgroup_move_account - move account of the page |
5564e88b | 2540 | * @page: the page |
7ec99d62 | 2541 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
2542 | * @pc: page_cgroup of the page. |
2543 | * @from: mem_cgroup which the page is moved from. | |
2544 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
854ffa8d | 2545 | * @uncharge: whether we should call uncharge and css_put against @from. |
f817ed48 KH |
2546 | * |
2547 | * The caller must confirm following. | |
08e552c6 | 2548 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 2549 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 2550 | * |
854ffa8d | 2551 | * This function doesn't do "charge" nor css_get to new cgroup. It should be |
25985edc | 2552 | * done by a caller(__mem_cgroup_try_charge would be useful). If @uncharge is |
854ffa8d DN |
2553 | * true, this function does "uncharge" from old cgroup, but it doesn't if |
2554 | * @uncharge is false, so a caller should do "uncharge". | |
f817ed48 | 2555 | */ |
7ec99d62 JW |
2556 | static int mem_cgroup_move_account(struct page *page, |
2557 | unsigned int nr_pages, | |
2558 | struct page_cgroup *pc, | |
2559 | struct mem_cgroup *from, | |
2560 | struct mem_cgroup *to, | |
2561 | bool uncharge) | |
f817ed48 | 2562 | { |
de3638d9 JW |
2563 | unsigned long flags; |
2564 | int ret; | |
b2402857 | 2565 | bool anon = PageAnon(page); |
987eba66 | 2566 | |
f817ed48 | 2567 | VM_BUG_ON(from == to); |
5564e88b | 2568 | VM_BUG_ON(PageLRU(page)); |
de3638d9 JW |
2569 | /* |
2570 | * The page is isolated from LRU. So, collapse function | |
2571 | * will not handle this page. But page splitting can happen. | |
2572 | * Do this check under compound_page_lock(). The caller should | |
2573 | * hold it. | |
2574 | */ | |
2575 | ret = -EBUSY; | |
7ec99d62 | 2576 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
2577 | goto out; |
2578 | ||
2579 | lock_page_cgroup(pc); | |
2580 | ||
2581 | ret = -EINVAL; | |
2582 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
2583 | goto unlock; | |
2584 | ||
2585 | move_lock_page_cgroup(pc, &flags); | |
f817ed48 | 2586 | |
8725d541 | 2587 | if (PageCgroupFileMapped(pc)) { |
c62b1a3b KH |
2588 | /* Update mapped_file data for mem_cgroup */ |
2589 | preempt_disable(); | |
2590 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2591 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2592 | preempt_enable(); | |
d69b042f | 2593 | } |
b2402857 | 2594 | mem_cgroup_charge_statistics(from, anon, -nr_pages); |
854ffa8d DN |
2595 | if (uncharge) |
2596 | /* This is not "cancel", but cancel_charge does all we need. */ | |
e7018b8d | 2597 | __mem_cgroup_cancel_charge(from, nr_pages); |
d69b042f | 2598 | |
854ffa8d | 2599 | /* caller should have done css_get */ |
08e552c6 | 2600 | pc->mem_cgroup = to; |
b2402857 | 2601 | mem_cgroup_charge_statistics(to, anon, nr_pages); |
88703267 KH |
2602 | /* |
2603 | * We charges against "to" which may not have any tasks. Then, "to" | |
2604 | * can be under rmdir(). But in current implementation, caller of | |
4ffef5fe | 2605 | * this function is just force_empty() and move charge, so it's |
25985edc | 2606 | * guaranteed that "to" is never removed. So, we don't check rmdir |
4ffef5fe | 2607 | * status here. |
88703267 | 2608 | */ |
de3638d9 JW |
2609 | move_unlock_page_cgroup(pc, &flags); |
2610 | ret = 0; | |
2611 | unlock: | |
57f9fd7d | 2612 | unlock_page_cgroup(pc); |
d2265e6f KH |
2613 | /* |
2614 | * check events | |
2615 | */ | |
5564e88b JW |
2616 | memcg_check_events(to, page); |
2617 | memcg_check_events(from, page); | |
de3638d9 | 2618 | out: |
f817ed48 KH |
2619 | return ret; |
2620 | } | |
2621 | ||
2622 | /* | |
2623 | * move charges to its parent. | |
2624 | */ | |
2625 | ||
5564e88b JW |
2626 | static int mem_cgroup_move_parent(struct page *page, |
2627 | struct page_cgroup *pc, | |
f817ed48 KH |
2628 | struct mem_cgroup *child, |
2629 | gfp_t gfp_mask) | |
2630 | { | |
2631 | struct cgroup *cg = child->css.cgroup; | |
2632 | struct cgroup *pcg = cg->parent; | |
2633 | struct mem_cgroup *parent; | |
7ec99d62 | 2634 | unsigned int nr_pages; |
4be4489f | 2635 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
2636 | int ret; |
2637 | ||
2638 | /* Is ROOT ? */ | |
2639 | if (!pcg) | |
2640 | return -EINVAL; | |
2641 | ||
57f9fd7d DN |
2642 | ret = -EBUSY; |
2643 | if (!get_page_unless_zero(page)) | |
2644 | goto out; | |
2645 | if (isolate_lru_page(page)) | |
2646 | goto put; | |
52dbb905 | 2647 | |
7ec99d62 | 2648 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 2649 | |
f817ed48 | 2650 | parent = mem_cgroup_from_cont(pcg); |
7ec99d62 | 2651 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, nr_pages, &parent, false); |
38c5d72f | 2652 | if (ret) |
57f9fd7d | 2653 | goto put_back; |
f817ed48 | 2654 | |
7ec99d62 | 2655 | if (nr_pages > 1) |
987eba66 KH |
2656 | flags = compound_lock_irqsave(page); |
2657 | ||
7ec99d62 | 2658 | ret = mem_cgroup_move_account(page, nr_pages, pc, child, parent, true); |
854ffa8d | 2659 | if (ret) |
7ec99d62 | 2660 | __mem_cgroup_cancel_charge(parent, nr_pages); |
8dba474f | 2661 | |
7ec99d62 | 2662 | if (nr_pages > 1) |
987eba66 | 2663 | compound_unlock_irqrestore(page, flags); |
8dba474f | 2664 | put_back: |
08e552c6 | 2665 | putback_lru_page(page); |
57f9fd7d | 2666 | put: |
40d58138 | 2667 | put_page(page); |
57f9fd7d | 2668 | out: |
f817ed48 KH |
2669 | return ret; |
2670 | } | |
2671 | ||
7a81b88c KH |
2672 | /* |
2673 | * Charge the memory controller for page usage. | |
2674 | * Return | |
2675 | * 0 if the charge was successful | |
2676 | * < 0 if the cgroup is over its limit | |
2677 | */ | |
2678 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
73045c47 | 2679 | gfp_t gfp_mask, enum charge_type ctype) |
7a81b88c | 2680 | { |
c0ff4b85 | 2681 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 2682 | unsigned int nr_pages = 1; |
7a81b88c | 2683 | struct page_cgroup *pc; |
8493ae43 | 2684 | bool oom = true; |
7a81b88c | 2685 | int ret; |
ec168510 | 2686 | |
37c2ac78 | 2687 | if (PageTransHuge(page)) { |
7ec99d62 | 2688 | nr_pages <<= compound_order(page); |
37c2ac78 | 2689 | VM_BUG_ON(!PageTransHuge(page)); |
8493ae43 JW |
2690 | /* |
2691 | * Never OOM-kill a process for a huge page. The | |
2692 | * fault handler will fall back to regular pages. | |
2693 | */ | |
2694 | oom = false; | |
37c2ac78 | 2695 | } |
7a81b88c KH |
2696 | |
2697 | pc = lookup_page_cgroup(page); | |
c0ff4b85 | 2698 | ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom); |
38c5d72f | 2699 | if (ret == -ENOMEM) |
7a81b88c | 2700 | return ret; |
9ce70c02 | 2701 | __mem_cgroup_commit_charge(memcg, page, nr_pages, pc, ctype, false); |
8a9f3ccd | 2702 | return 0; |
8a9f3ccd BS |
2703 | } |
2704 | ||
7a81b88c KH |
2705 | int mem_cgroup_newpage_charge(struct page *page, |
2706 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 2707 | { |
f8d66542 | 2708 | if (mem_cgroup_disabled()) |
cede86ac | 2709 | return 0; |
7a0524cf JW |
2710 | VM_BUG_ON(page_mapped(page)); |
2711 | VM_BUG_ON(page->mapping && !PageAnon(page)); | |
2712 | VM_BUG_ON(!mm); | |
217bc319 | 2713 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
7a0524cf | 2714 | MEM_CGROUP_CHARGE_TYPE_MAPPED); |
217bc319 KH |
2715 | } |
2716 | ||
83aae4c7 DN |
2717 | static void |
2718 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
2719 | enum charge_type ctype); | |
2720 | ||
e1a1cd59 BS |
2721 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
2722 | gfp_t gfp_mask) | |
8697d331 | 2723 | { |
c0ff4b85 | 2724 | struct mem_cgroup *memcg = NULL; |
dc67d504 | 2725 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
b5a84319 KH |
2726 | int ret; |
2727 | ||
f8d66542 | 2728 | if (mem_cgroup_disabled()) |
cede86ac | 2729 | return 0; |
52d4b9ac KH |
2730 | if (PageCompound(page)) |
2731 | return 0; | |
accf163e | 2732 | |
73045c47 | 2733 | if (unlikely(!mm)) |
8697d331 | 2734 | mm = &init_mm; |
dc67d504 KH |
2735 | if (!page_is_file_cache(page)) |
2736 | type = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
accf163e | 2737 | |
38c5d72f | 2738 | if (!PageSwapCache(page)) |
dc67d504 | 2739 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, type); |
38c5d72f | 2740 | else { /* page is swapcache/shmem */ |
c0ff4b85 | 2741 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &memcg); |
83aae4c7 | 2742 | if (!ret) |
dc67d504 KH |
2743 | __mem_cgroup_commit_charge_swapin(page, memcg, type); |
2744 | } | |
b5a84319 | 2745 | return ret; |
e8589cc1 KH |
2746 | } |
2747 | ||
54595fe2 KH |
2748 | /* |
2749 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
2750 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 2751 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
2752 | * "commit()" or removed by "cancel()" |
2753 | */ | |
8c7c6e34 KH |
2754 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
2755 | struct page *page, | |
72835c86 | 2756 | gfp_t mask, struct mem_cgroup **memcgp) |
8c7c6e34 | 2757 | { |
c0ff4b85 | 2758 | struct mem_cgroup *memcg; |
54595fe2 | 2759 | int ret; |
8c7c6e34 | 2760 | |
72835c86 | 2761 | *memcgp = NULL; |
56039efa | 2762 | |
f8d66542 | 2763 | if (mem_cgroup_disabled()) |
8c7c6e34 KH |
2764 | return 0; |
2765 | ||
2766 | if (!do_swap_account) | |
2767 | goto charge_cur_mm; | |
8c7c6e34 KH |
2768 | /* |
2769 | * A racing thread's fault, or swapoff, may have already updated | |
407f9c8b HD |
2770 | * the pte, and even removed page from swap cache: in those cases |
2771 | * do_swap_page()'s pte_same() test will fail; but there's also a | |
2772 | * KSM case which does need to charge the page. | |
8c7c6e34 KH |
2773 | */ |
2774 | if (!PageSwapCache(page)) | |
407f9c8b | 2775 | goto charge_cur_mm; |
c0ff4b85 R |
2776 | memcg = try_get_mem_cgroup_from_page(page); |
2777 | if (!memcg) | |
54595fe2 | 2778 | goto charge_cur_mm; |
72835c86 JW |
2779 | *memcgp = memcg; |
2780 | ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true); | |
c0ff4b85 | 2781 | css_put(&memcg->css); |
38c5d72f KH |
2782 | if (ret == -EINTR) |
2783 | ret = 0; | |
54595fe2 | 2784 | return ret; |
8c7c6e34 KH |
2785 | charge_cur_mm: |
2786 | if (unlikely(!mm)) | |
2787 | mm = &init_mm; | |
38c5d72f KH |
2788 | ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true); |
2789 | if (ret == -EINTR) | |
2790 | ret = 0; | |
2791 | return ret; | |
8c7c6e34 KH |
2792 | } |
2793 | ||
83aae4c7 | 2794 | static void |
72835c86 | 2795 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg, |
83aae4c7 | 2796 | enum charge_type ctype) |
7a81b88c | 2797 | { |
9ce70c02 HD |
2798 | struct page_cgroup *pc; |
2799 | ||
f8d66542 | 2800 | if (mem_cgroup_disabled()) |
7a81b88c | 2801 | return; |
72835c86 | 2802 | if (!memcg) |
7a81b88c | 2803 | return; |
72835c86 | 2804 | cgroup_exclude_rmdir(&memcg->css); |
5a6475a4 | 2805 | |
9ce70c02 HD |
2806 | pc = lookup_page_cgroup(page); |
2807 | __mem_cgroup_commit_charge(memcg, page, 1, pc, ctype, true); | |
8c7c6e34 KH |
2808 | /* |
2809 | * Now swap is on-memory. This means this page may be | |
2810 | * counted both as mem and swap....double count. | |
03f3c433 KH |
2811 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
2812 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
2813 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 2814 | */ |
03f3c433 | 2815 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 2816 | swp_entry_t ent = {.val = page_private(page)}; |
72835c86 | 2817 | struct mem_cgroup *swap_memcg; |
a3b2d692 | 2818 | unsigned short id; |
a3b2d692 KH |
2819 | |
2820 | id = swap_cgroup_record(ent, 0); | |
2821 | rcu_read_lock(); | |
72835c86 JW |
2822 | swap_memcg = mem_cgroup_lookup(id); |
2823 | if (swap_memcg) { | |
a3b2d692 KH |
2824 | /* |
2825 | * This recorded memcg can be obsolete one. So, avoid | |
2826 | * calling css_tryget | |
2827 | */ | |
72835c86 JW |
2828 | if (!mem_cgroup_is_root(swap_memcg)) |
2829 | res_counter_uncharge(&swap_memcg->memsw, | |
2830 | PAGE_SIZE); | |
2831 | mem_cgroup_swap_statistics(swap_memcg, false); | |
2832 | mem_cgroup_put(swap_memcg); | |
8c7c6e34 | 2833 | } |
a3b2d692 | 2834 | rcu_read_unlock(); |
8c7c6e34 | 2835 | } |
88703267 KH |
2836 | /* |
2837 | * At swapin, we may charge account against cgroup which has no tasks. | |
2838 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
2839 | * In that case, we need to call pre_destroy() again. check it here. | |
2840 | */ | |
72835c86 | 2841 | cgroup_release_and_wakeup_rmdir(&memcg->css); |
7a81b88c KH |
2842 | } |
2843 | ||
72835c86 JW |
2844 | void mem_cgroup_commit_charge_swapin(struct page *page, |
2845 | struct mem_cgroup *memcg) | |
83aae4c7 | 2846 | { |
72835c86 JW |
2847 | __mem_cgroup_commit_charge_swapin(page, memcg, |
2848 | MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
83aae4c7 DN |
2849 | } |
2850 | ||
c0ff4b85 | 2851 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg) |
7a81b88c | 2852 | { |
f8d66542 | 2853 | if (mem_cgroup_disabled()) |
7a81b88c | 2854 | return; |
c0ff4b85 | 2855 | if (!memcg) |
7a81b88c | 2856 | return; |
c0ff4b85 | 2857 | __mem_cgroup_cancel_charge(memcg, 1); |
7a81b88c KH |
2858 | } |
2859 | ||
c0ff4b85 | 2860 | static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg, |
7ec99d62 JW |
2861 | unsigned int nr_pages, |
2862 | const enum charge_type ctype) | |
569b846d KH |
2863 | { |
2864 | struct memcg_batch_info *batch = NULL; | |
2865 | bool uncharge_memsw = true; | |
7ec99d62 | 2866 | |
569b846d KH |
2867 | /* If swapout, usage of swap doesn't decrease */ |
2868 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2869 | uncharge_memsw = false; | |
569b846d KH |
2870 | |
2871 | batch = ¤t->memcg_batch; | |
2872 | /* | |
2873 | * In usual, we do css_get() when we remember memcg pointer. | |
2874 | * But in this case, we keep res->usage until end of a series of | |
2875 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
2876 | */ | |
2877 | if (!batch->memcg) | |
c0ff4b85 | 2878 | batch->memcg = memcg; |
3c11ecf4 KH |
2879 | /* |
2880 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
25985edc | 2881 | * In those cases, all pages freed continuously can be expected to be in |
3c11ecf4 KH |
2882 | * the same cgroup and we have chance to coalesce uncharges. |
2883 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
2884 | * because we want to do uncharge as soon as possible. | |
2885 | */ | |
2886 | ||
2887 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
2888 | goto direct_uncharge; | |
2889 | ||
7ec99d62 | 2890 | if (nr_pages > 1) |
ec168510 AA |
2891 | goto direct_uncharge; |
2892 | ||
569b846d KH |
2893 | /* |
2894 | * In typical case, batch->memcg == mem. This means we can | |
2895 | * merge a series of uncharges to an uncharge of res_counter. | |
2896 | * If not, we uncharge res_counter ony by one. | |
2897 | */ | |
c0ff4b85 | 2898 | if (batch->memcg != memcg) |
569b846d KH |
2899 | goto direct_uncharge; |
2900 | /* remember freed charge and uncharge it later */ | |
7ffd4ca7 | 2901 | batch->nr_pages++; |
569b846d | 2902 | if (uncharge_memsw) |
7ffd4ca7 | 2903 | batch->memsw_nr_pages++; |
569b846d KH |
2904 | return; |
2905 | direct_uncharge: | |
c0ff4b85 | 2906 | res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE); |
569b846d | 2907 | if (uncharge_memsw) |
c0ff4b85 R |
2908 | res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE); |
2909 | if (unlikely(batch->memcg != memcg)) | |
2910 | memcg_oom_recover(memcg); | |
569b846d | 2911 | } |
7a81b88c | 2912 | |
8a9f3ccd | 2913 | /* |
69029cd5 | 2914 | * uncharge if !page_mapped(page) |
8a9f3ccd | 2915 | */ |
8c7c6e34 | 2916 | static struct mem_cgroup * |
69029cd5 | 2917 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
8a9f3ccd | 2918 | { |
c0ff4b85 | 2919 | struct mem_cgroup *memcg = NULL; |
7ec99d62 JW |
2920 | unsigned int nr_pages = 1; |
2921 | struct page_cgroup *pc; | |
b2402857 | 2922 | bool anon; |
8a9f3ccd | 2923 | |
f8d66542 | 2924 | if (mem_cgroup_disabled()) |
8c7c6e34 | 2925 | return NULL; |
4077960e | 2926 | |
d13d1443 | 2927 | if (PageSwapCache(page)) |
8c7c6e34 | 2928 | return NULL; |
d13d1443 | 2929 | |
37c2ac78 | 2930 | if (PageTransHuge(page)) { |
7ec99d62 | 2931 | nr_pages <<= compound_order(page); |
37c2ac78 AA |
2932 | VM_BUG_ON(!PageTransHuge(page)); |
2933 | } | |
8697d331 | 2934 | /* |
3c541e14 | 2935 | * Check if our page_cgroup is valid |
8697d331 | 2936 | */ |
52d4b9ac | 2937 | pc = lookup_page_cgroup(page); |
cfa44946 | 2938 | if (unlikely(!PageCgroupUsed(pc))) |
8c7c6e34 | 2939 | return NULL; |
b9c565d5 | 2940 | |
52d4b9ac | 2941 | lock_page_cgroup(pc); |
d13d1443 | 2942 | |
c0ff4b85 | 2943 | memcg = pc->mem_cgroup; |
8c7c6e34 | 2944 | |
d13d1443 KH |
2945 | if (!PageCgroupUsed(pc)) |
2946 | goto unlock_out; | |
2947 | ||
b2402857 KH |
2948 | anon = PageAnon(page); |
2949 | ||
d13d1443 KH |
2950 | switch (ctype) { |
2951 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
b2402857 KH |
2952 | anon = true; |
2953 | /* fallthrough */ | |
8a9478ca | 2954 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c | 2955 | /* See mem_cgroup_prepare_migration() */ |
2956 | if (page_mapped(page) || PageCgroupMigration(pc)) | |
d13d1443 KH |
2957 | goto unlock_out; |
2958 | break; | |
2959 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
2960 | if (!PageAnon(page)) { /* Shared memory */ | |
2961 | if (page->mapping && !page_is_file_cache(page)) | |
2962 | goto unlock_out; | |
2963 | } else if (page_mapped(page)) /* Anon */ | |
2964 | goto unlock_out; | |
2965 | break; | |
2966 | default: | |
2967 | break; | |
52d4b9ac | 2968 | } |
d13d1443 | 2969 | |
b2402857 | 2970 | mem_cgroup_charge_statistics(memcg, anon, -nr_pages); |
04046e1a | 2971 | |
52d4b9ac | 2972 | ClearPageCgroupUsed(pc); |
544122e5 KH |
2973 | /* |
2974 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
2975 | * freed from LRU. This is safe because uncharged page is expected not | |
2976 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
2977 | * special functions. | |
2978 | */ | |
b9c565d5 | 2979 | |
52d4b9ac | 2980 | unlock_page_cgroup(pc); |
f75ca962 | 2981 | /* |
c0ff4b85 | 2982 | * even after unlock, we have memcg->res.usage here and this memcg |
f75ca962 KH |
2983 | * will never be freed. |
2984 | */ | |
c0ff4b85 | 2985 | memcg_check_events(memcg, page); |
f75ca962 | 2986 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
c0ff4b85 R |
2987 | mem_cgroup_swap_statistics(memcg, true); |
2988 | mem_cgroup_get(memcg); | |
f75ca962 | 2989 | } |
c0ff4b85 R |
2990 | if (!mem_cgroup_is_root(memcg)) |
2991 | mem_cgroup_do_uncharge(memcg, nr_pages, ctype); | |
6d12e2d8 | 2992 | |
c0ff4b85 | 2993 | return memcg; |
d13d1443 KH |
2994 | |
2995 | unlock_out: | |
2996 | unlock_page_cgroup(pc); | |
8c7c6e34 | 2997 | return NULL; |
3c541e14 BS |
2998 | } |
2999 | ||
69029cd5 KH |
3000 | void mem_cgroup_uncharge_page(struct page *page) |
3001 | { | |
52d4b9ac KH |
3002 | /* early check. */ |
3003 | if (page_mapped(page)) | |
3004 | return; | |
40f23a21 | 3005 | VM_BUG_ON(page->mapping && !PageAnon(page)); |
69029cd5 KH |
3006 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
3007 | } | |
3008 | ||
3009 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
3010 | { | |
3011 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 3012 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
3013 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
3014 | } | |
3015 | ||
569b846d KH |
3016 | /* |
3017 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
3018 | * In that cases, pages are freed continuously and we can expect pages | |
3019 | * are in the same memcg. All these calls itself limits the number of | |
3020 | * pages freed at once, then uncharge_start/end() is called properly. | |
3021 | * This may be called prural(2) times in a context, | |
3022 | */ | |
3023 | ||
3024 | void mem_cgroup_uncharge_start(void) | |
3025 | { | |
3026 | current->memcg_batch.do_batch++; | |
3027 | /* We can do nest. */ | |
3028 | if (current->memcg_batch.do_batch == 1) { | |
3029 | current->memcg_batch.memcg = NULL; | |
7ffd4ca7 JW |
3030 | current->memcg_batch.nr_pages = 0; |
3031 | current->memcg_batch.memsw_nr_pages = 0; | |
569b846d KH |
3032 | } |
3033 | } | |
3034 | ||
3035 | void mem_cgroup_uncharge_end(void) | |
3036 | { | |
3037 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
3038 | ||
3039 | if (!batch->do_batch) | |
3040 | return; | |
3041 | ||
3042 | batch->do_batch--; | |
3043 | if (batch->do_batch) /* If stacked, do nothing. */ | |
3044 | return; | |
3045 | ||
3046 | if (!batch->memcg) | |
3047 | return; | |
3048 | /* | |
3049 | * This "batch->memcg" is valid without any css_get/put etc... | |
3050 | * bacause we hide charges behind us. | |
3051 | */ | |
7ffd4ca7 JW |
3052 | if (batch->nr_pages) |
3053 | res_counter_uncharge(&batch->memcg->res, | |
3054 | batch->nr_pages * PAGE_SIZE); | |
3055 | if (batch->memsw_nr_pages) | |
3056 | res_counter_uncharge(&batch->memcg->memsw, | |
3057 | batch->memsw_nr_pages * PAGE_SIZE); | |
3c11ecf4 | 3058 | memcg_oom_recover(batch->memcg); |
569b846d KH |
3059 | /* forget this pointer (for sanity check) */ |
3060 | batch->memcg = NULL; | |
3061 | } | |
3062 | ||
e767e056 | 3063 | #ifdef CONFIG_SWAP |
8c7c6e34 | 3064 | /* |
e767e056 | 3065 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
3066 | * memcg information is recorded to swap_cgroup of "ent" |
3067 | */ | |
8a9478ca KH |
3068 | void |
3069 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
3070 | { |
3071 | struct mem_cgroup *memcg; | |
8a9478ca KH |
3072 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
3073 | ||
3074 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
3075 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
3076 | ||
3077 | memcg = __mem_cgroup_uncharge_common(page, ctype); | |
8c7c6e34 | 3078 | |
f75ca962 KH |
3079 | /* |
3080 | * record memcg information, if swapout && memcg != NULL, | |
3081 | * mem_cgroup_get() was called in uncharge(). | |
3082 | */ | |
3083 | if (do_swap_account && swapout && memcg) | |
a3b2d692 | 3084 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 | 3085 | } |
e767e056 | 3086 | #endif |
8c7c6e34 KH |
3087 | |
3088 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
3089 | /* | |
3090 | * called from swap_entry_free(). remove record in swap_cgroup and | |
3091 | * uncharge "memsw" account. | |
3092 | */ | |
3093 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 3094 | { |
8c7c6e34 | 3095 | struct mem_cgroup *memcg; |
a3b2d692 | 3096 | unsigned short id; |
8c7c6e34 KH |
3097 | |
3098 | if (!do_swap_account) | |
3099 | return; | |
3100 | ||
a3b2d692 KH |
3101 | id = swap_cgroup_record(ent, 0); |
3102 | rcu_read_lock(); | |
3103 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 3104 | if (memcg) { |
a3b2d692 KH |
3105 | /* |
3106 | * We uncharge this because swap is freed. | |
3107 | * This memcg can be obsolete one. We avoid calling css_tryget | |
3108 | */ | |
0c3e73e8 | 3109 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 3110 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 3111 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
3112 | mem_cgroup_put(memcg); |
3113 | } | |
a3b2d692 | 3114 | rcu_read_unlock(); |
d13d1443 | 3115 | } |
02491447 DN |
3116 | |
3117 | /** | |
3118 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3119 | * @entry: swap entry to be moved | |
3120 | * @from: mem_cgroup which the entry is moved from | |
3121 | * @to: mem_cgroup which the entry is moved to | |
483c30b5 | 3122 | * @need_fixup: whether we should fixup res_counters and refcounts. |
02491447 DN |
3123 | * |
3124 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3125 | * as the mem_cgroup's id of @from. | |
3126 | * | |
3127 | * Returns 0 on success, -EINVAL on failure. | |
3128 | * | |
3129 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
3130 | * both res and memsw, and called css_get(). | |
3131 | */ | |
3132 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
483c30b5 | 3133 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
02491447 DN |
3134 | { |
3135 | unsigned short old_id, new_id; | |
3136 | ||
3137 | old_id = css_id(&from->css); | |
3138 | new_id = css_id(&to->css); | |
3139 | ||
3140 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 3141 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 3142 | mem_cgroup_swap_statistics(to, true); |
02491447 | 3143 | /* |
483c30b5 DN |
3144 | * This function is only called from task migration context now. |
3145 | * It postpones res_counter and refcount handling till the end | |
3146 | * of task migration(mem_cgroup_clear_mc()) for performance | |
3147 | * improvement. But we cannot postpone mem_cgroup_get(to) | |
3148 | * because if the process that has been moved to @to does | |
3149 | * swap-in, the refcount of @to might be decreased to 0. | |
02491447 | 3150 | */ |
02491447 | 3151 | mem_cgroup_get(to); |
483c30b5 DN |
3152 | if (need_fixup) { |
3153 | if (!mem_cgroup_is_root(from)) | |
3154 | res_counter_uncharge(&from->memsw, PAGE_SIZE); | |
3155 | mem_cgroup_put(from); | |
3156 | /* | |
3157 | * we charged both to->res and to->memsw, so we should | |
3158 | * uncharge to->res. | |
3159 | */ | |
3160 | if (!mem_cgroup_is_root(to)) | |
3161 | res_counter_uncharge(&to->res, PAGE_SIZE); | |
483c30b5 | 3162 | } |
02491447 DN |
3163 | return 0; |
3164 | } | |
3165 | return -EINVAL; | |
3166 | } | |
3167 | #else | |
3168 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
483c30b5 | 3169 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
02491447 DN |
3170 | { |
3171 | return -EINVAL; | |
3172 | } | |
8c7c6e34 | 3173 | #endif |
d13d1443 | 3174 | |
ae41be37 | 3175 | /* |
01b1ae63 KH |
3176 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
3177 | * page belongs to. | |
ae41be37 | 3178 | */ |
ac39cf8c | 3179 | int mem_cgroup_prepare_migration(struct page *page, |
72835c86 | 3180 | struct page *newpage, struct mem_cgroup **memcgp, gfp_t gfp_mask) |
ae41be37 | 3181 | { |
c0ff4b85 | 3182 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 3183 | struct page_cgroup *pc; |
ac39cf8c | 3184 | enum charge_type ctype; |
e8589cc1 | 3185 | int ret = 0; |
8869b8f6 | 3186 | |
72835c86 | 3187 | *memcgp = NULL; |
56039efa | 3188 | |
ec168510 | 3189 | VM_BUG_ON(PageTransHuge(page)); |
f8d66542 | 3190 | if (mem_cgroup_disabled()) |
4077960e BS |
3191 | return 0; |
3192 | ||
52d4b9ac KH |
3193 | pc = lookup_page_cgroup(page); |
3194 | lock_page_cgroup(pc); | |
3195 | if (PageCgroupUsed(pc)) { | |
c0ff4b85 R |
3196 | memcg = pc->mem_cgroup; |
3197 | css_get(&memcg->css); | |
ac39cf8c | 3198 | /* |
3199 | * At migrating an anonymous page, its mapcount goes down | |
3200 | * to 0 and uncharge() will be called. But, even if it's fully | |
3201 | * unmapped, migration may fail and this page has to be | |
3202 | * charged again. We set MIGRATION flag here and delay uncharge | |
3203 | * until end_migration() is called | |
3204 | * | |
3205 | * Corner Case Thinking | |
3206 | * A) | |
3207 | * When the old page was mapped as Anon and it's unmap-and-freed | |
3208 | * while migration was ongoing. | |
3209 | * If unmap finds the old page, uncharge() of it will be delayed | |
3210 | * until end_migration(). If unmap finds a new page, it's | |
3211 | * uncharged when it make mapcount to be 1->0. If unmap code | |
3212 | * finds swap_migration_entry, the new page will not be mapped | |
3213 | * and end_migration() will find it(mapcount==0). | |
3214 | * | |
3215 | * B) | |
3216 | * When the old page was mapped but migraion fails, the kernel | |
3217 | * remaps it. A charge for it is kept by MIGRATION flag even | |
3218 | * if mapcount goes down to 0. We can do remap successfully | |
3219 | * without charging it again. | |
3220 | * | |
3221 | * C) | |
3222 | * The "old" page is under lock_page() until the end of | |
3223 | * migration, so, the old page itself will not be swapped-out. | |
3224 | * If the new page is swapped out before end_migraton, our | |
3225 | * hook to usual swap-out path will catch the event. | |
3226 | */ | |
3227 | if (PageAnon(page)) | |
3228 | SetPageCgroupMigration(pc); | |
e8589cc1 | 3229 | } |
52d4b9ac | 3230 | unlock_page_cgroup(pc); |
ac39cf8c | 3231 | /* |
3232 | * If the page is not charged at this point, | |
3233 | * we return here. | |
3234 | */ | |
c0ff4b85 | 3235 | if (!memcg) |
ac39cf8c | 3236 | return 0; |
01b1ae63 | 3237 | |
72835c86 JW |
3238 | *memcgp = memcg; |
3239 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, 1, memcgp, false); | |
c0ff4b85 | 3240 | css_put(&memcg->css);/* drop extra refcnt */ |
38c5d72f | 3241 | if (ret) { |
ac39cf8c | 3242 | if (PageAnon(page)) { |
3243 | lock_page_cgroup(pc); | |
3244 | ClearPageCgroupMigration(pc); | |
3245 | unlock_page_cgroup(pc); | |
3246 | /* | |
3247 | * The old page may be fully unmapped while we kept it. | |
3248 | */ | |
3249 | mem_cgroup_uncharge_page(page); | |
3250 | } | |
38c5d72f | 3251 | /* we'll need to revisit this error code (we have -EINTR) */ |
ac39cf8c | 3252 | return -ENOMEM; |
e8589cc1 | 3253 | } |
ac39cf8c | 3254 | /* |
3255 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
3256 | * is called before end_migration, we can catch all events on this new | |
3257 | * page. In the case new page is migrated but not remapped, new page's | |
3258 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
3259 | */ | |
3260 | pc = lookup_page_cgroup(newpage); | |
3261 | if (PageAnon(page)) | |
3262 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
3263 | else if (page_is_file_cache(page)) | |
3264 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
3265 | else | |
3266 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
9ce70c02 | 3267 | __mem_cgroup_commit_charge(memcg, newpage, 1, pc, ctype, false); |
e8589cc1 | 3268 | return ret; |
ae41be37 | 3269 | } |
8869b8f6 | 3270 | |
69029cd5 | 3271 | /* remove redundant charge if migration failed*/ |
c0ff4b85 | 3272 | void mem_cgroup_end_migration(struct mem_cgroup *memcg, |
50de1dd9 | 3273 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 3274 | { |
ac39cf8c | 3275 | struct page *used, *unused; |
01b1ae63 | 3276 | struct page_cgroup *pc; |
b2402857 | 3277 | bool anon; |
01b1ae63 | 3278 | |
c0ff4b85 | 3279 | if (!memcg) |
01b1ae63 | 3280 | return; |
ac39cf8c | 3281 | /* blocks rmdir() */ |
c0ff4b85 | 3282 | cgroup_exclude_rmdir(&memcg->css); |
50de1dd9 | 3283 | if (!migration_ok) { |
ac39cf8c | 3284 | used = oldpage; |
3285 | unused = newpage; | |
01b1ae63 | 3286 | } else { |
ac39cf8c | 3287 | used = newpage; |
01b1ae63 KH |
3288 | unused = oldpage; |
3289 | } | |
69029cd5 | 3290 | /* |
ac39cf8c | 3291 | * We disallowed uncharge of pages under migration because mapcount |
3292 | * of the page goes down to zero, temporarly. | |
3293 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 3294 | */ |
ac39cf8c | 3295 | pc = lookup_page_cgroup(oldpage); |
3296 | lock_page_cgroup(pc); | |
3297 | ClearPageCgroupMigration(pc); | |
3298 | unlock_page_cgroup(pc); | |
b2402857 KH |
3299 | anon = PageAnon(used); |
3300 | __mem_cgroup_uncharge_common(unused, | |
3301 | anon ? MEM_CGROUP_CHARGE_TYPE_MAPPED | |
3302 | : MEM_CGROUP_CHARGE_TYPE_CACHE); | |
ac39cf8c | 3303 | |
01b1ae63 | 3304 | /* |
ac39cf8c | 3305 | * If a page is a file cache, radix-tree replacement is very atomic |
3306 | * and we can skip this check. When it was an Anon page, its mapcount | |
3307 | * goes down to 0. But because we added MIGRATION flage, it's not | |
3308 | * uncharged yet. There are several case but page->mapcount check | |
3309 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
3310 | * check. (see prepare_charge() also) | |
69029cd5 | 3311 | */ |
b2402857 | 3312 | if (anon) |
ac39cf8c | 3313 | mem_cgroup_uncharge_page(used); |
88703267 | 3314 | /* |
ac39cf8c | 3315 | * At migration, we may charge account against cgroup which has no |
3316 | * tasks. | |
88703267 KH |
3317 | * So, rmdir()->pre_destroy() can be called while we do this charge. |
3318 | * In that case, we need to call pre_destroy() again. check it here. | |
3319 | */ | |
c0ff4b85 | 3320 | cgroup_release_and_wakeup_rmdir(&memcg->css); |
ae41be37 | 3321 | } |
78fb7466 | 3322 | |
ab936cbc KH |
3323 | /* |
3324 | * At replace page cache, newpage is not under any memcg but it's on | |
3325 | * LRU. So, this function doesn't touch res_counter but handles LRU | |
3326 | * in correct way. Both pages are locked so we cannot race with uncharge. | |
3327 | */ | |
3328 | void mem_cgroup_replace_page_cache(struct page *oldpage, | |
3329 | struct page *newpage) | |
3330 | { | |
3331 | struct mem_cgroup *memcg; | |
3332 | struct page_cgroup *pc; | |
ab936cbc | 3333 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
ab936cbc KH |
3334 | |
3335 | if (mem_cgroup_disabled()) | |
3336 | return; | |
3337 | ||
3338 | pc = lookup_page_cgroup(oldpage); | |
3339 | /* fix accounting on old pages */ | |
3340 | lock_page_cgroup(pc); | |
3341 | memcg = pc->mem_cgroup; | |
b2402857 | 3342 | mem_cgroup_charge_statistics(memcg, false, -1); |
ab936cbc KH |
3343 | ClearPageCgroupUsed(pc); |
3344 | unlock_page_cgroup(pc); | |
3345 | ||
3346 | if (PageSwapBacked(oldpage)) | |
3347 | type = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
3348 | ||
ab936cbc KH |
3349 | /* |
3350 | * Even if newpage->mapping was NULL before starting replacement, | |
3351 | * the newpage may be on LRU(or pagevec for LRU) already. We lock | |
3352 | * LRU while we overwrite pc->mem_cgroup. | |
3353 | */ | |
9ce70c02 | 3354 | __mem_cgroup_commit_charge(memcg, newpage, 1, pc, type, true); |
ab936cbc KH |
3355 | } |
3356 | ||
f212ad7c DN |
3357 | #ifdef CONFIG_DEBUG_VM |
3358 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
3359 | { | |
3360 | struct page_cgroup *pc; | |
3361 | ||
3362 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
3363 | /* |
3364 | * Can be NULL while feeding pages into the page allocator for | |
3365 | * the first time, i.e. during boot or memory hotplug; | |
3366 | * or when mem_cgroup_disabled(). | |
3367 | */ | |
f212ad7c DN |
3368 | if (likely(pc) && PageCgroupUsed(pc)) |
3369 | return pc; | |
3370 | return NULL; | |
3371 | } | |
3372 | ||
3373 | bool mem_cgroup_bad_page_check(struct page *page) | |
3374 | { | |
3375 | if (mem_cgroup_disabled()) | |
3376 | return false; | |
3377 | ||
3378 | return lookup_page_cgroup_used(page) != NULL; | |
3379 | } | |
3380 | ||
3381 | void mem_cgroup_print_bad_page(struct page *page) | |
3382 | { | |
3383 | struct page_cgroup *pc; | |
3384 | ||
3385 | pc = lookup_page_cgroup_used(page); | |
3386 | if (pc) { | |
90b3feae | 3387 | printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p\n", |
f212ad7c | 3388 | pc, pc->flags, pc->mem_cgroup); |
f212ad7c DN |
3389 | } |
3390 | } | |
3391 | #endif | |
3392 | ||
8c7c6e34 KH |
3393 | static DEFINE_MUTEX(set_limit_mutex); |
3394 | ||
d38d2a75 | 3395 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 3396 | unsigned long long val) |
628f4235 | 3397 | { |
81d39c20 | 3398 | int retry_count; |
3c11ecf4 | 3399 | u64 memswlimit, memlimit; |
628f4235 | 3400 | int ret = 0; |
81d39c20 KH |
3401 | int children = mem_cgroup_count_children(memcg); |
3402 | u64 curusage, oldusage; | |
3c11ecf4 | 3403 | int enlarge; |
81d39c20 KH |
3404 | |
3405 | /* | |
3406 | * For keeping hierarchical_reclaim simple, how long we should retry | |
3407 | * is depends on callers. We set our retry-count to be function | |
3408 | * of # of children which we should visit in this loop. | |
3409 | */ | |
3410 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
3411 | ||
3412 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 3413 | |
3c11ecf4 | 3414 | enlarge = 0; |
8c7c6e34 | 3415 | while (retry_count) { |
628f4235 KH |
3416 | if (signal_pending(current)) { |
3417 | ret = -EINTR; | |
3418 | break; | |
3419 | } | |
8c7c6e34 KH |
3420 | /* |
3421 | * Rather than hide all in some function, I do this in | |
3422 | * open coded manner. You see what this really does. | |
c0ff4b85 | 3423 | * We have to guarantee memcg->res.limit < memcg->memsw.limit. |
8c7c6e34 KH |
3424 | */ |
3425 | mutex_lock(&set_limit_mutex); | |
3426 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3427 | if (memswlimit < val) { | |
3428 | ret = -EINVAL; | |
3429 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
3430 | break; |
3431 | } | |
3c11ecf4 KH |
3432 | |
3433 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3434 | if (memlimit < val) | |
3435 | enlarge = 1; | |
3436 | ||
8c7c6e34 | 3437 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
3438 | if (!ret) { |
3439 | if (memswlimit == val) | |
3440 | memcg->memsw_is_minimum = true; | |
3441 | else | |
3442 | memcg->memsw_is_minimum = false; | |
3443 | } | |
8c7c6e34 KH |
3444 | mutex_unlock(&set_limit_mutex); |
3445 | ||
3446 | if (!ret) | |
3447 | break; | |
3448 | ||
5660048c JW |
3449 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
3450 | MEM_CGROUP_RECLAIM_SHRINK); | |
81d39c20 KH |
3451 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
3452 | /* Usage is reduced ? */ | |
3453 | if (curusage >= oldusage) | |
3454 | retry_count--; | |
3455 | else | |
3456 | oldusage = curusage; | |
8c7c6e34 | 3457 | } |
3c11ecf4 KH |
3458 | if (!ret && enlarge) |
3459 | memcg_oom_recover(memcg); | |
14797e23 | 3460 | |
8c7c6e34 KH |
3461 | return ret; |
3462 | } | |
3463 | ||
338c8431 LZ |
3464 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3465 | unsigned long long val) | |
8c7c6e34 | 3466 | { |
81d39c20 | 3467 | int retry_count; |
3c11ecf4 | 3468 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
3469 | int children = mem_cgroup_count_children(memcg); |
3470 | int ret = -EBUSY; | |
3c11ecf4 | 3471 | int enlarge = 0; |
8c7c6e34 | 3472 | |
81d39c20 KH |
3473 | /* see mem_cgroup_resize_res_limit */ |
3474 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
3475 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
3476 | while (retry_count) { |
3477 | if (signal_pending(current)) { | |
3478 | ret = -EINTR; | |
3479 | break; | |
3480 | } | |
3481 | /* | |
3482 | * Rather than hide all in some function, I do this in | |
3483 | * open coded manner. You see what this really does. | |
c0ff4b85 | 3484 | * We have to guarantee memcg->res.limit < memcg->memsw.limit. |
8c7c6e34 KH |
3485 | */ |
3486 | mutex_lock(&set_limit_mutex); | |
3487 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3488 | if (memlimit > val) { | |
3489 | ret = -EINVAL; | |
3490 | mutex_unlock(&set_limit_mutex); | |
3491 | break; | |
3492 | } | |
3c11ecf4 KH |
3493 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
3494 | if (memswlimit < val) | |
3495 | enlarge = 1; | |
8c7c6e34 | 3496 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
3497 | if (!ret) { |
3498 | if (memlimit == val) | |
3499 | memcg->memsw_is_minimum = true; | |
3500 | else | |
3501 | memcg->memsw_is_minimum = false; | |
3502 | } | |
8c7c6e34 KH |
3503 | mutex_unlock(&set_limit_mutex); |
3504 | ||
3505 | if (!ret) | |
3506 | break; | |
3507 | ||
5660048c JW |
3508 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
3509 | MEM_CGROUP_RECLAIM_NOSWAP | | |
3510 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 3511 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 3512 | /* Usage is reduced ? */ |
8c7c6e34 | 3513 | if (curusage >= oldusage) |
628f4235 | 3514 | retry_count--; |
81d39c20 KH |
3515 | else |
3516 | oldusage = curusage; | |
628f4235 | 3517 | } |
3c11ecf4 KH |
3518 | if (!ret && enlarge) |
3519 | memcg_oom_recover(memcg); | |
628f4235 KH |
3520 | return ret; |
3521 | } | |
3522 | ||
4e416953 | 3523 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
0ae5e89c YH |
3524 | gfp_t gfp_mask, |
3525 | unsigned long *total_scanned) | |
4e416953 BS |
3526 | { |
3527 | unsigned long nr_reclaimed = 0; | |
3528 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3529 | unsigned long reclaimed; | |
3530 | int loop = 0; | |
3531 | struct mem_cgroup_tree_per_zone *mctz; | |
ef8745c1 | 3532 | unsigned long long excess; |
0ae5e89c | 3533 | unsigned long nr_scanned; |
4e416953 BS |
3534 | |
3535 | if (order > 0) | |
3536 | return 0; | |
3537 | ||
00918b6a | 3538 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); |
4e416953 BS |
3539 | /* |
3540 | * This loop can run a while, specially if mem_cgroup's continuously | |
3541 | * keep exceeding their soft limit and putting the system under | |
3542 | * pressure | |
3543 | */ | |
3544 | do { | |
3545 | if (next_mz) | |
3546 | mz = next_mz; | |
3547 | else | |
3548 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3549 | if (!mz) | |
3550 | break; | |
3551 | ||
0ae5e89c | 3552 | nr_scanned = 0; |
d79154bb | 3553 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, |
5660048c | 3554 | gfp_mask, &nr_scanned); |
4e416953 | 3555 | nr_reclaimed += reclaimed; |
0ae5e89c | 3556 | *total_scanned += nr_scanned; |
4e416953 BS |
3557 | spin_lock(&mctz->lock); |
3558 | ||
3559 | /* | |
3560 | * If we failed to reclaim anything from this memory cgroup | |
3561 | * it is time to move on to the next cgroup | |
3562 | */ | |
3563 | next_mz = NULL; | |
3564 | if (!reclaimed) { | |
3565 | do { | |
3566 | /* | |
3567 | * Loop until we find yet another one. | |
3568 | * | |
3569 | * By the time we get the soft_limit lock | |
3570 | * again, someone might have aded the | |
3571 | * group back on the RB tree. Iterate to | |
3572 | * make sure we get a different mem. | |
3573 | * mem_cgroup_largest_soft_limit_node returns | |
3574 | * NULL if no other cgroup is present on | |
3575 | * the tree | |
3576 | */ | |
3577 | next_mz = | |
3578 | __mem_cgroup_largest_soft_limit_node(mctz); | |
39cc98f1 | 3579 | if (next_mz == mz) |
d79154bb | 3580 | css_put(&next_mz->memcg->css); |
39cc98f1 | 3581 | else /* next_mz == NULL or other memcg */ |
4e416953 BS |
3582 | break; |
3583 | } while (1); | |
3584 | } | |
d79154bb HD |
3585 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); |
3586 | excess = res_counter_soft_limit_excess(&mz->memcg->res); | |
4e416953 BS |
3587 | /* |
3588 | * One school of thought says that we should not add | |
3589 | * back the node to the tree if reclaim returns 0. | |
3590 | * But our reclaim could return 0, simply because due | |
3591 | * to priority we are exposing a smaller subset of | |
3592 | * memory to reclaim from. Consider this as a longer | |
3593 | * term TODO. | |
3594 | */ | |
ef8745c1 | 3595 | /* If excess == 0, no tree ops */ |
d79154bb | 3596 | __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess); |
4e416953 | 3597 | spin_unlock(&mctz->lock); |
d79154bb | 3598 | css_put(&mz->memcg->css); |
4e416953 BS |
3599 | loop++; |
3600 | /* | |
3601 | * Could not reclaim anything and there are no more | |
3602 | * mem cgroups to try or we seem to be looping without | |
3603 | * reclaiming anything. | |
3604 | */ | |
3605 | if (!nr_reclaimed && | |
3606 | (next_mz == NULL || | |
3607 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3608 | break; | |
3609 | } while (!nr_reclaimed); | |
3610 | if (next_mz) | |
d79154bb | 3611 | css_put(&next_mz->memcg->css); |
4e416953 BS |
3612 | return nr_reclaimed; |
3613 | } | |
3614 | ||
cc847582 KH |
3615 | /* |
3616 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
3617 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
3618 | */ | |
c0ff4b85 | 3619 | static int mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
08e552c6 | 3620 | int node, int zid, enum lru_list lru) |
cc847582 | 3621 | { |
08e552c6 | 3622 | struct mem_cgroup_per_zone *mz; |
08e552c6 | 3623 | unsigned long flags, loop; |
072c56c1 | 3624 | struct list_head *list; |
925b7673 JW |
3625 | struct page *busy; |
3626 | struct zone *zone; | |
f817ed48 | 3627 | int ret = 0; |
072c56c1 | 3628 | |
08e552c6 | 3629 | zone = &NODE_DATA(node)->node_zones[zid]; |
c0ff4b85 | 3630 | mz = mem_cgroup_zoneinfo(memcg, node, zid); |
6290df54 | 3631 | list = &mz->lruvec.lists[lru]; |
cc847582 | 3632 | |
1eb49272 | 3633 | loop = mz->lru_size[lru]; |
f817ed48 KH |
3634 | /* give some margin against EBUSY etc...*/ |
3635 | loop += 256; | |
3636 | busy = NULL; | |
3637 | while (loop--) { | |
925b7673 | 3638 | struct page_cgroup *pc; |
5564e88b JW |
3639 | struct page *page; |
3640 | ||
f817ed48 | 3641 | ret = 0; |
08e552c6 | 3642 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 3643 | if (list_empty(list)) { |
08e552c6 | 3644 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 3645 | break; |
f817ed48 | 3646 | } |
925b7673 JW |
3647 | page = list_entry(list->prev, struct page, lru); |
3648 | if (busy == page) { | |
3649 | list_move(&page->lru, list); | |
648bcc77 | 3650 | busy = NULL; |
08e552c6 | 3651 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
3652 | continue; |
3653 | } | |
08e552c6 | 3654 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 3655 | |
925b7673 | 3656 | pc = lookup_page_cgroup(page); |
5564e88b | 3657 | |
c0ff4b85 | 3658 | ret = mem_cgroup_move_parent(page, pc, memcg, GFP_KERNEL); |
38c5d72f | 3659 | if (ret == -ENOMEM || ret == -EINTR) |
52d4b9ac | 3660 | break; |
f817ed48 KH |
3661 | |
3662 | if (ret == -EBUSY || ret == -EINVAL) { | |
3663 | /* found lock contention or "pc" is obsolete. */ | |
925b7673 | 3664 | busy = page; |
f817ed48 KH |
3665 | cond_resched(); |
3666 | } else | |
3667 | busy = NULL; | |
cc847582 | 3668 | } |
08e552c6 | 3669 | |
f817ed48 KH |
3670 | if (!ret && !list_empty(list)) |
3671 | return -EBUSY; | |
3672 | return ret; | |
cc847582 KH |
3673 | } |
3674 | ||
3675 | /* | |
3676 | * make mem_cgroup's charge to be 0 if there is no task. | |
3677 | * This enables deleting this mem_cgroup. | |
3678 | */ | |
c0ff4b85 | 3679 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg, bool free_all) |
cc847582 | 3680 | { |
f817ed48 KH |
3681 | int ret; |
3682 | int node, zid, shrink; | |
3683 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c0ff4b85 | 3684 | struct cgroup *cgrp = memcg->css.cgroup; |
8869b8f6 | 3685 | |
c0ff4b85 | 3686 | css_get(&memcg->css); |
f817ed48 KH |
3687 | |
3688 | shrink = 0; | |
c1e862c1 KH |
3689 | /* should free all ? */ |
3690 | if (free_all) | |
3691 | goto try_to_free; | |
f817ed48 | 3692 | move_account: |
fce66477 | 3693 | do { |
f817ed48 | 3694 | ret = -EBUSY; |
c1e862c1 KH |
3695 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
3696 | goto out; | |
3697 | ret = -EINTR; | |
3698 | if (signal_pending(current)) | |
cc847582 | 3699 | goto out; |
52d4b9ac KH |
3700 | /* This is for making all *used* pages to be on LRU. */ |
3701 | lru_add_drain_all(); | |
c0ff4b85 | 3702 | drain_all_stock_sync(memcg); |
f817ed48 | 3703 | ret = 0; |
c0ff4b85 | 3704 | mem_cgroup_start_move(memcg); |
299b4eaa | 3705 | for_each_node_state(node, N_HIGH_MEMORY) { |
f817ed48 | 3706 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { |
f156ab93 HD |
3707 | enum lru_list lru; |
3708 | for_each_lru(lru) { | |
c0ff4b85 | 3709 | ret = mem_cgroup_force_empty_list(memcg, |
f156ab93 | 3710 | node, zid, lru); |
f817ed48 KH |
3711 | if (ret) |
3712 | break; | |
3713 | } | |
1ecaab2b | 3714 | } |
f817ed48 KH |
3715 | if (ret) |
3716 | break; | |
3717 | } | |
c0ff4b85 R |
3718 | mem_cgroup_end_move(memcg); |
3719 | memcg_oom_recover(memcg); | |
f817ed48 KH |
3720 | /* it seems parent cgroup doesn't have enough mem */ |
3721 | if (ret == -ENOMEM) | |
3722 | goto try_to_free; | |
52d4b9ac | 3723 | cond_resched(); |
fce66477 | 3724 | /* "ret" should also be checked to ensure all lists are empty. */ |
c0ff4b85 | 3725 | } while (memcg->res.usage > 0 || ret); |
cc847582 | 3726 | out: |
c0ff4b85 | 3727 | css_put(&memcg->css); |
cc847582 | 3728 | return ret; |
f817ed48 KH |
3729 | |
3730 | try_to_free: | |
c1e862c1 KH |
3731 | /* returns EBUSY if there is a task or if we come here twice. */ |
3732 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
f817ed48 KH |
3733 | ret = -EBUSY; |
3734 | goto out; | |
3735 | } | |
c1e862c1 KH |
3736 | /* we call try-to-free pages for make this cgroup empty */ |
3737 | lru_add_drain_all(); | |
f817ed48 KH |
3738 | /* try to free all pages in this cgroup */ |
3739 | shrink = 1; | |
c0ff4b85 | 3740 | while (nr_retries && memcg->res.usage > 0) { |
f817ed48 | 3741 | int progress; |
c1e862c1 KH |
3742 | |
3743 | if (signal_pending(current)) { | |
3744 | ret = -EINTR; | |
3745 | goto out; | |
3746 | } | |
c0ff4b85 | 3747 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, |
185efc0f | 3748 | false); |
c1e862c1 | 3749 | if (!progress) { |
f817ed48 | 3750 | nr_retries--; |
c1e862c1 | 3751 | /* maybe some writeback is necessary */ |
8aa7e847 | 3752 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3753 | } |
f817ed48 KH |
3754 | |
3755 | } | |
08e552c6 | 3756 | lru_add_drain(); |
f817ed48 | 3757 | /* try move_account...there may be some *locked* pages. */ |
fce66477 | 3758 | goto move_account; |
cc847582 KH |
3759 | } |
3760 | ||
c1e862c1 KH |
3761 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
3762 | { | |
3763 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
3764 | } | |
3765 | ||
3766 | ||
18f59ea7 BS |
3767 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
3768 | { | |
3769 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
3770 | } | |
3771 | ||
3772 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
3773 | u64 val) | |
3774 | { | |
3775 | int retval = 0; | |
c0ff4b85 | 3776 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
18f59ea7 | 3777 | struct cgroup *parent = cont->parent; |
c0ff4b85 | 3778 | struct mem_cgroup *parent_memcg = NULL; |
18f59ea7 BS |
3779 | |
3780 | if (parent) | |
c0ff4b85 | 3781 | parent_memcg = mem_cgroup_from_cont(parent); |
18f59ea7 BS |
3782 | |
3783 | cgroup_lock(); | |
3784 | /* | |
af901ca1 | 3785 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3786 | * in the child subtrees. If it is unset, then the change can |
3787 | * occur, provided the current cgroup has no children. | |
3788 | * | |
3789 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3790 | * set if there are no children. | |
3791 | */ | |
c0ff4b85 | 3792 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 BS |
3793 | (val == 1 || val == 0)) { |
3794 | if (list_empty(&cont->children)) | |
c0ff4b85 | 3795 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3796 | else |
3797 | retval = -EBUSY; | |
3798 | } else | |
3799 | retval = -EINVAL; | |
3800 | cgroup_unlock(); | |
3801 | ||
3802 | return retval; | |
3803 | } | |
3804 | ||
0c3e73e8 | 3805 | |
c0ff4b85 | 3806 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg, |
7a159cc9 | 3807 | enum mem_cgroup_stat_index idx) |
0c3e73e8 | 3808 | { |
7d74b06f | 3809 | struct mem_cgroup *iter; |
7a159cc9 | 3810 | long val = 0; |
0c3e73e8 | 3811 | |
7a159cc9 | 3812 | /* Per-cpu values can be negative, use a signed accumulator */ |
c0ff4b85 | 3813 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f KH |
3814 | val += mem_cgroup_read_stat(iter, idx); |
3815 | ||
3816 | if (val < 0) /* race ? */ | |
3817 | val = 0; | |
3818 | return val; | |
0c3e73e8 BS |
3819 | } |
3820 | ||
c0ff4b85 | 3821 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
104f3928 | 3822 | { |
7d74b06f | 3823 | u64 val; |
104f3928 | 3824 | |
c0ff4b85 | 3825 | if (!mem_cgroup_is_root(memcg)) { |
104f3928 | 3826 | if (!swap) |
65c64ce8 | 3827 | return res_counter_read_u64(&memcg->res, RES_USAGE); |
104f3928 | 3828 | else |
65c64ce8 | 3829 | return res_counter_read_u64(&memcg->memsw, RES_USAGE); |
104f3928 KS |
3830 | } |
3831 | ||
c0ff4b85 R |
3832 | val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE); |
3833 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS); | |
104f3928 | 3834 | |
7d74b06f | 3835 | if (swap) |
c0ff4b85 | 3836 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAPOUT); |
104f3928 KS |
3837 | |
3838 | return val << PAGE_SHIFT; | |
3839 | } | |
3840 | ||
2c3daa72 | 3841 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 3842 | { |
c0ff4b85 | 3843 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
104f3928 | 3844 | u64 val; |
8c7c6e34 KH |
3845 | int type, name; |
3846 | ||
3847 | type = MEMFILE_TYPE(cft->private); | |
3848 | name = MEMFILE_ATTR(cft->private); | |
3849 | switch (type) { | |
3850 | case _MEM: | |
104f3928 | 3851 | if (name == RES_USAGE) |
c0ff4b85 | 3852 | val = mem_cgroup_usage(memcg, false); |
104f3928 | 3853 | else |
c0ff4b85 | 3854 | val = res_counter_read_u64(&memcg->res, name); |
8c7c6e34 KH |
3855 | break; |
3856 | case _MEMSWAP: | |
104f3928 | 3857 | if (name == RES_USAGE) |
c0ff4b85 | 3858 | val = mem_cgroup_usage(memcg, true); |
104f3928 | 3859 | else |
c0ff4b85 | 3860 | val = res_counter_read_u64(&memcg->memsw, name); |
8c7c6e34 KH |
3861 | break; |
3862 | default: | |
3863 | BUG(); | |
3864 | break; | |
3865 | } | |
3866 | return val; | |
8cdea7c0 | 3867 | } |
628f4235 KH |
3868 | /* |
3869 | * The user of this function is... | |
3870 | * RES_LIMIT. | |
3871 | */ | |
856c13aa PM |
3872 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
3873 | const char *buffer) | |
8cdea7c0 | 3874 | { |
628f4235 | 3875 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 3876 | int type, name; |
628f4235 KH |
3877 | unsigned long long val; |
3878 | int ret; | |
3879 | ||
8c7c6e34 KH |
3880 | type = MEMFILE_TYPE(cft->private); |
3881 | name = MEMFILE_ATTR(cft->private); | |
3882 | switch (name) { | |
628f4235 | 3883 | case RES_LIMIT: |
4b3bde4c BS |
3884 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3885 | ret = -EINVAL; | |
3886 | break; | |
3887 | } | |
628f4235 KH |
3888 | /* This function does all necessary parse...reuse it */ |
3889 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
3890 | if (ret) |
3891 | break; | |
3892 | if (type == _MEM) | |
628f4235 | 3893 | ret = mem_cgroup_resize_limit(memcg, val); |
8c7c6e34 KH |
3894 | else |
3895 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
628f4235 | 3896 | break; |
296c81d8 BS |
3897 | case RES_SOFT_LIMIT: |
3898 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
3899 | if (ret) | |
3900 | break; | |
3901 | /* | |
3902 | * For memsw, soft limits are hard to implement in terms | |
3903 | * of semantics, for now, we support soft limits for | |
3904 | * control without swap | |
3905 | */ | |
3906 | if (type == _MEM) | |
3907 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
3908 | else | |
3909 | ret = -EINVAL; | |
3910 | break; | |
628f4235 KH |
3911 | default: |
3912 | ret = -EINVAL; /* should be BUG() ? */ | |
3913 | break; | |
3914 | } | |
3915 | return ret; | |
8cdea7c0 BS |
3916 | } |
3917 | ||
fee7b548 KH |
3918 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
3919 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
3920 | { | |
3921 | struct cgroup *cgroup; | |
3922 | unsigned long long min_limit, min_memsw_limit, tmp; | |
3923 | ||
3924 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3925 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3926 | cgroup = memcg->css.cgroup; | |
3927 | if (!memcg->use_hierarchy) | |
3928 | goto out; | |
3929 | ||
3930 | while (cgroup->parent) { | |
3931 | cgroup = cgroup->parent; | |
3932 | memcg = mem_cgroup_from_cont(cgroup); | |
3933 | if (!memcg->use_hierarchy) | |
3934 | break; | |
3935 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3936 | min_limit = min(min_limit, tmp); | |
3937 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3938 | min_memsw_limit = min(min_memsw_limit, tmp); | |
3939 | } | |
3940 | out: | |
3941 | *mem_limit = min_limit; | |
3942 | *memsw_limit = min_memsw_limit; | |
fee7b548 KH |
3943 | } |
3944 | ||
29f2a4da | 3945 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 | 3946 | { |
c0ff4b85 | 3947 | struct mem_cgroup *memcg; |
8c7c6e34 | 3948 | int type, name; |
c84872e1 | 3949 | |
c0ff4b85 | 3950 | memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 KH |
3951 | type = MEMFILE_TYPE(event); |
3952 | name = MEMFILE_ATTR(event); | |
3953 | switch (name) { | |
29f2a4da | 3954 | case RES_MAX_USAGE: |
8c7c6e34 | 3955 | if (type == _MEM) |
c0ff4b85 | 3956 | res_counter_reset_max(&memcg->res); |
8c7c6e34 | 3957 | else |
c0ff4b85 | 3958 | res_counter_reset_max(&memcg->memsw); |
29f2a4da PE |
3959 | break; |
3960 | case RES_FAILCNT: | |
8c7c6e34 | 3961 | if (type == _MEM) |
c0ff4b85 | 3962 | res_counter_reset_failcnt(&memcg->res); |
8c7c6e34 | 3963 | else |
c0ff4b85 | 3964 | res_counter_reset_failcnt(&memcg->memsw); |
29f2a4da PE |
3965 | break; |
3966 | } | |
f64c3f54 | 3967 | |
85cc59db | 3968 | return 0; |
c84872e1 PE |
3969 | } |
3970 | ||
7dc74be0 DN |
3971 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
3972 | struct cftype *cft) | |
3973 | { | |
3974 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; | |
3975 | } | |
3976 | ||
02491447 | 3977 | #ifdef CONFIG_MMU |
7dc74be0 DN |
3978 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
3979 | struct cftype *cft, u64 val) | |
3980 | { | |
c0ff4b85 | 3981 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
7dc74be0 DN |
3982 | |
3983 | if (val >= (1 << NR_MOVE_TYPE)) | |
3984 | return -EINVAL; | |
3985 | /* | |
3986 | * We check this value several times in both in can_attach() and | |
3987 | * attach(), so we need cgroup lock to prevent this value from being | |
3988 | * inconsistent. | |
3989 | */ | |
3990 | cgroup_lock(); | |
c0ff4b85 | 3991 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3992 | cgroup_unlock(); |
3993 | ||
3994 | return 0; | |
3995 | } | |
02491447 DN |
3996 | #else |
3997 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, | |
3998 | struct cftype *cft, u64 val) | |
3999 | { | |
4000 | return -ENOSYS; | |
4001 | } | |
4002 | #endif | |
7dc74be0 | 4003 | |
14067bb3 KH |
4004 | |
4005 | /* For read statistics */ | |
4006 | enum { | |
4007 | MCS_CACHE, | |
4008 | MCS_RSS, | |
d8046582 | 4009 | MCS_FILE_MAPPED, |
14067bb3 KH |
4010 | MCS_PGPGIN, |
4011 | MCS_PGPGOUT, | |
1dd3a273 | 4012 | MCS_SWAP, |
456f998e YH |
4013 | MCS_PGFAULT, |
4014 | MCS_PGMAJFAULT, | |
14067bb3 KH |
4015 | MCS_INACTIVE_ANON, |
4016 | MCS_ACTIVE_ANON, | |
4017 | MCS_INACTIVE_FILE, | |
4018 | MCS_ACTIVE_FILE, | |
4019 | MCS_UNEVICTABLE, | |
4020 | NR_MCS_STAT, | |
4021 | }; | |
4022 | ||
4023 | struct mcs_total_stat { | |
4024 | s64 stat[NR_MCS_STAT]; | |
d2ceb9b7 KH |
4025 | }; |
4026 | ||
14067bb3 KH |
4027 | struct { |
4028 | char *local_name; | |
4029 | char *total_name; | |
4030 | } memcg_stat_strings[NR_MCS_STAT] = { | |
4031 | {"cache", "total_cache"}, | |
4032 | {"rss", "total_rss"}, | |
d69b042f | 4033 | {"mapped_file", "total_mapped_file"}, |
14067bb3 KH |
4034 | {"pgpgin", "total_pgpgin"}, |
4035 | {"pgpgout", "total_pgpgout"}, | |
1dd3a273 | 4036 | {"swap", "total_swap"}, |
456f998e YH |
4037 | {"pgfault", "total_pgfault"}, |
4038 | {"pgmajfault", "total_pgmajfault"}, | |
14067bb3 KH |
4039 | {"inactive_anon", "total_inactive_anon"}, |
4040 | {"active_anon", "total_active_anon"}, | |
4041 | {"inactive_file", "total_inactive_file"}, | |
4042 | {"active_file", "total_active_file"}, | |
4043 | {"unevictable", "total_unevictable"} | |
4044 | }; | |
4045 | ||
4046 | ||
7d74b06f | 4047 | static void |
c0ff4b85 | 4048 | mem_cgroup_get_local_stat(struct mem_cgroup *memcg, struct mcs_total_stat *s) |
14067bb3 | 4049 | { |
14067bb3 KH |
4050 | s64 val; |
4051 | ||
4052 | /* per cpu stat */ | |
c0ff4b85 | 4053 | val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_CACHE); |
14067bb3 | 4054 | s->stat[MCS_CACHE] += val * PAGE_SIZE; |
c0ff4b85 | 4055 | val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_RSS); |
14067bb3 | 4056 | s->stat[MCS_RSS] += val * PAGE_SIZE; |
c0ff4b85 | 4057 | val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED); |
d8046582 | 4058 | s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE; |
c0ff4b85 | 4059 | val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGIN); |
14067bb3 | 4060 | s->stat[MCS_PGPGIN] += val; |
c0ff4b85 | 4061 | val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGOUT); |
14067bb3 | 4062 | s->stat[MCS_PGPGOUT] += val; |
1dd3a273 | 4063 | if (do_swap_account) { |
c0ff4b85 | 4064 | val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_SWAPOUT); |
1dd3a273 DN |
4065 | s->stat[MCS_SWAP] += val * PAGE_SIZE; |
4066 | } | |
c0ff4b85 | 4067 | val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGFAULT); |
456f998e | 4068 | s->stat[MCS_PGFAULT] += val; |
c0ff4b85 | 4069 | val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGMAJFAULT); |
456f998e | 4070 | s->stat[MCS_PGMAJFAULT] += val; |
14067bb3 KH |
4071 | |
4072 | /* per zone stat */ | |
c0ff4b85 | 4073 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_ANON)); |
14067bb3 | 4074 | s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE; |
c0ff4b85 | 4075 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_ANON)); |
14067bb3 | 4076 | s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE; |
c0ff4b85 | 4077 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_FILE)); |
14067bb3 | 4078 | s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE; |
c0ff4b85 | 4079 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_FILE)); |
14067bb3 | 4080 | s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE; |
c0ff4b85 | 4081 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE)); |
14067bb3 | 4082 | s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE; |
14067bb3 KH |
4083 | } |
4084 | ||
4085 | static void | |
c0ff4b85 | 4086 | mem_cgroup_get_total_stat(struct mem_cgroup *memcg, struct mcs_total_stat *s) |
14067bb3 | 4087 | { |
7d74b06f KH |
4088 | struct mem_cgroup *iter; |
4089 | ||
c0ff4b85 | 4090 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4091 | mem_cgroup_get_local_stat(iter, s); |
14067bb3 KH |
4092 | } |
4093 | ||
406eb0c9 YH |
4094 | #ifdef CONFIG_NUMA |
4095 | static int mem_control_numa_stat_show(struct seq_file *m, void *arg) | |
4096 | { | |
4097 | int nid; | |
4098 | unsigned long total_nr, file_nr, anon_nr, unevictable_nr; | |
4099 | unsigned long node_nr; | |
4100 | struct cgroup *cont = m->private; | |
d79154bb | 4101 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
406eb0c9 | 4102 | |
d79154bb | 4103 | total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL); |
406eb0c9 YH |
4104 | seq_printf(m, "total=%lu", total_nr); |
4105 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
d79154bb | 4106 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL); |
406eb0c9 YH |
4107 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4108 | } | |
4109 | seq_putc(m, '\n'); | |
4110 | ||
d79154bb | 4111 | file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE); |
406eb0c9 YH |
4112 | seq_printf(m, "file=%lu", file_nr); |
4113 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
d79154bb | 4114 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 4115 | LRU_ALL_FILE); |
406eb0c9 YH |
4116 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4117 | } | |
4118 | seq_putc(m, '\n'); | |
4119 | ||
d79154bb | 4120 | anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON); |
406eb0c9 YH |
4121 | seq_printf(m, "anon=%lu", anon_nr); |
4122 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
d79154bb | 4123 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 4124 | LRU_ALL_ANON); |
406eb0c9 YH |
4125 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4126 | } | |
4127 | seq_putc(m, '\n'); | |
4128 | ||
d79154bb | 4129 | unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE)); |
406eb0c9 YH |
4130 | seq_printf(m, "unevictable=%lu", unevictable_nr); |
4131 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
d79154bb | 4132 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 4133 | BIT(LRU_UNEVICTABLE)); |
406eb0c9 YH |
4134 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4135 | } | |
4136 | seq_putc(m, '\n'); | |
4137 | return 0; | |
4138 | } | |
4139 | #endif /* CONFIG_NUMA */ | |
4140 | ||
c64745cf PM |
4141 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
4142 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 4143 | { |
d79154bb | 4144 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
14067bb3 | 4145 | struct mcs_total_stat mystat; |
d2ceb9b7 KH |
4146 | int i; |
4147 | ||
14067bb3 | 4148 | memset(&mystat, 0, sizeof(mystat)); |
d79154bb | 4149 | mem_cgroup_get_local_stat(memcg, &mystat); |
d2ceb9b7 | 4150 | |
406eb0c9 | 4151 | |
1dd3a273 DN |
4152 | for (i = 0; i < NR_MCS_STAT; i++) { |
4153 | if (i == MCS_SWAP && !do_swap_account) | |
4154 | continue; | |
14067bb3 | 4155 | cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]); |
1dd3a273 | 4156 | } |
7b854121 | 4157 | |
14067bb3 | 4158 | /* Hierarchical information */ |
fee7b548 KH |
4159 | { |
4160 | unsigned long long limit, memsw_limit; | |
d79154bb | 4161 | memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit); |
fee7b548 KH |
4162 | cb->fill(cb, "hierarchical_memory_limit", limit); |
4163 | if (do_swap_account) | |
4164 | cb->fill(cb, "hierarchical_memsw_limit", memsw_limit); | |
4165 | } | |
7f016ee8 | 4166 | |
14067bb3 | 4167 | memset(&mystat, 0, sizeof(mystat)); |
d79154bb | 4168 | mem_cgroup_get_total_stat(memcg, &mystat); |
1dd3a273 DN |
4169 | for (i = 0; i < NR_MCS_STAT; i++) { |
4170 | if (i == MCS_SWAP && !do_swap_account) | |
4171 | continue; | |
14067bb3 | 4172 | cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]); |
1dd3a273 | 4173 | } |
14067bb3 | 4174 | |
7f016ee8 | 4175 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
4176 | { |
4177 | int nid, zid; | |
4178 | struct mem_cgroup_per_zone *mz; | |
4179 | unsigned long recent_rotated[2] = {0, 0}; | |
4180 | unsigned long recent_scanned[2] = {0, 0}; | |
4181 | ||
4182 | for_each_online_node(nid) | |
4183 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
d79154bb | 4184 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
7f016ee8 KM |
4185 | |
4186 | recent_rotated[0] += | |
4187 | mz->reclaim_stat.recent_rotated[0]; | |
4188 | recent_rotated[1] += | |
4189 | mz->reclaim_stat.recent_rotated[1]; | |
4190 | recent_scanned[0] += | |
4191 | mz->reclaim_stat.recent_scanned[0]; | |
4192 | recent_scanned[1] += | |
4193 | mz->reclaim_stat.recent_scanned[1]; | |
4194 | } | |
4195 | cb->fill(cb, "recent_rotated_anon", recent_rotated[0]); | |
4196 | cb->fill(cb, "recent_rotated_file", recent_rotated[1]); | |
4197 | cb->fill(cb, "recent_scanned_anon", recent_scanned[0]); | |
4198 | cb->fill(cb, "recent_scanned_file", recent_scanned[1]); | |
4199 | } | |
4200 | #endif | |
4201 | ||
d2ceb9b7 KH |
4202 | return 0; |
4203 | } | |
4204 | ||
a7885eb8 KM |
4205 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
4206 | { | |
4207 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4208 | ||
1f4c025b | 4209 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4210 | } |
4211 | ||
4212 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
4213 | u64 val) | |
4214 | { | |
4215 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4216 | struct mem_cgroup *parent; | |
068b38c1 | 4217 | |
a7885eb8 KM |
4218 | if (val > 100) |
4219 | return -EINVAL; | |
4220 | ||
4221 | if (cgrp->parent == NULL) | |
4222 | return -EINVAL; | |
4223 | ||
4224 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 LZ |
4225 | |
4226 | cgroup_lock(); | |
4227 | ||
a7885eb8 KM |
4228 | /* If under hierarchy, only empty-root can set this value */ |
4229 | if ((parent->use_hierarchy) || | |
068b38c1 LZ |
4230 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
4231 | cgroup_unlock(); | |
a7885eb8 | 4232 | return -EINVAL; |
068b38c1 | 4233 | } |
a7885eb8 | 4234 | |
a7885eb8 | 4235 | memcg->swappiness = val; |
a7885eb8 | 4236 | |
068b38c1 LZ |
4237 | cgroup_unlock(); |
4238 | ||
a7885eb8 KM |
4239 | return 0; |
4240 | } | |
4241 | ||
2e72b634 KS |
4242 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4243 | { | |
4244 | struct mem_cgroup_threshold_ary *t; | |
4245 | u64 usage; | |
4246 | int i; | |
4247 | ||
4248 | rcu_read_lock(); | |
4249 | if (!swap) | |
2c488db2 | 4250 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4251 | else |
2c488db2 | 4252 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4253 | |
4254 | if (!t) | |
4255 | goto unlock; | |
4256 | ||
4257 | usage = mem_cgroup_usage(memcg, swap); | |
4258 | ||
4259 | /* | |
4260 | * current_threshold points to threshold just below usage. | |
4261 | * If it's not true, a threshold was crossed after last | |
4262 | * call of __mem_cgroup_threshold(). | |
4263 | */ | |
5407a562 | 4264 | i = t->current_threshold; |
2e72b634 KS |
4265 | |
4266 | /* | |
4267 | * Iterate backward over array of thresholds starting from | |
4268 | * current_threshold and check if a threshold is crossed. | |
4269 | * If none of thresholds below usage is crossed, we read | |
4270 | * only one element of the array here. | |
4271 | */ | |
4272 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4273 | eventfd_signal(t->entries[i].eventfd, 1); | |
4274 | ||
4275 | /* i = current_threshold + 1 */ | |
4276 | i++; | |
4277 | ||
4278 | /* | |
4279 | * Iterate forward over array of thresholds starting from | |
4280 | * current_threshold+1 and check if a threshold is crossed. | |
4281 | * If none of thresholds above usage is crossed, we read | |
4282 | * only one element of the array here. | |
4283 | */ | |
4284 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4285 | eventfd_signal(t->entries[i].eventfd, 1); | |
4286 | ||
4287 | /* Update current_threshold */ | |
5407a562 | 4288 | t->current_threshold = i - 1; |
2e72b634 KS |
4289 | unlock: |
4290 | rcu_read_unlock(); | |
4291 | } | |
4292 | ||
4293 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4294 | { | |
ad4ca5f4 KS |
4295 | while (memcg) { |
4296 | __mem_cgroup_threshold(memcg, false); | |
4297 | if (do_swap_account) | |
4298 | __mem_cgroup_threshold(memcg, true); | |
4299 | ||
4300 | memcg = parent_mem_cgroup(memcg); | |
4301 | } | |
2e72b634 KS |
4302 | } |
4303 | ||
4304 | static int compare_thresholds(const void *a, const void *b) | |
4305 | { | |
4306 | const struct mem_cgroup_threshold *_a = a; | |
4307 | const struct mem_cgroup_threshold *_b = b; | |
4308 | ||
4309 | return _a->threshold - _b->threshold; | |
4310 | } | |
4311 | ||
c0ff4b85 | 4312 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4313 | { |
4314 | struct mem_cgroup_eventfd_list *ev; | |
4315 | ||
c0ff4b85 | 4316 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 KH |
4317 | eventfd_signal(ev->eventfd, 1); |
4318 | return 0; | |
4319 | } | |
4320 | ||
c0ff4b85 | 4321 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4322 | { |
7d74b06f KH |
4323 | struct mem_cgroup *iter; |
4324 | ||
c0ff4b85 | 4325 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4326 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4327 | } |
4328 | ||
4329 | static int mem_cgroup_usage_register_event(struct cgroup *cgrp, | |
4330 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
2e72b634 KS |
4331 | { |
4332 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4333 | struct mem_cgroup_thresholds *thresholds; |
4334 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
4335 | int type = MEMFILE_TYPE(cft->private); |
4336 | u64 threshold, usage; | |
2c488db2 | 4337 | int i, size, ret; |
2e72b634 KS |
4338 | |
4339 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
4340 | if (ret) | |
4341 | return ret; | |
4342 | ||
4343 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4344 | |
2e72b634 | 4345 | if (type == _MEM) |
2c488db2 | 4346 | thresholds = &memcg->thresholds; |
2e72b634 | 4347 | else if (type == _MEMSWAP) |
2c488db2 | 4348 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4349 | else |
4350 | BUG(); | |
4351 | ||
4352 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
4353 | ||
4354 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 4355 | if (thresholds->primary) |
2e72b634 KS |
4356 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4357 | ||
2c488db2 | 4358 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4359 | |
4360 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 4361 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 4362 | GFP_KERNEL); |
2c488db2 | 4363 | if (!new) { |
2e72b634 KS |
4364 | ret = -ENOMEM; |
4365 | goto unlock; | |
4366 | } | |
2c488db2 | 4367 | new->size = size; |
2e72b634 KS |
4368 | |
4369 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4370 | if (thresholds->primary) { |
4371 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4372 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4373 | } |
4374 | ||
2e72b634 | 4375 | /* Add new threshold */ |
2c488db2 KS |
4376 | new->entries[size - 1].eventfd = eventfd; |
4377 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4378 | |
4379 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4380 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4381 | compare_thresholds, NULL); |
4382 | ||
4383 | /* Find current threshold */ | |
2c488db2 | 4384 | new->current_threshold = -1; |
2e72b634 | 4385 | for (i = 0; i < size; i++) { |
2c488db2 | 4386 | if (new->entries[i].threshold < usage) { |
2e72b634 | 4387 | /* |
2c488db2 KS |
4388 | * new->current_threshold will not be used until |
4389 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4390 | * it here. |
4391 | */ | |
2c488db2 | 4392 | ++new->current_threshold; |
2e72b634 KS |
4393 | } |
4394 | } | |
4395 | ||
2c488db2 KS |
4396 | /* Free old spare buffer and save old primary buffer as spare */ |
4397 | kfree(thresholds->spare); | |
4398 | thresholds->spare = thresholds->primary; | |
4399 | ||
4400 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4401 | |
907860ed | 4402 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4403 | synchronize_rcu(); |
4404 | ||
2e72b634 KS |
4405 | unlock: |
4406 | mutex_unlock(&memcg->thresholds_lock); | |
4407 | ||
4408 | return ret; | |
4409 | } | |
4410 | ||
907860ed | 4411 | static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp, |
9490ff27 | 4412 | struct cftype *cft, struct eventfd_ctx *eventfd) |
2e72b634 KS |
4413 | { |
4414 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4415 | struct mem_cgroup_thresholds *thresholds; |
4416 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
4417 | int type = MEMFILE_TYPE(cft->private); |
4418 | u64 usage; | |
2c488db2 | 4419 | int i, j, size; |
2e72b634 KS |
4420 | |
4421 | mutex_lock(&memcg->thresholds_lock); | |
4422 | if (type == _MEM) | |
2c488db2 | 4423 | thresholds = &memcg->thresholds; |
2e72b634 | 4424 | else if (type == _MEMSWAP) |
2c488db2 | 4425 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4426 | else |
4427 | BUG(); | |
4428 | ||
4429 | /* | |
4430 | * Something went wrong if we trying to unregister a threshold | |
4431 | * if we don't have thresholds | |
4432 | */ | |
4433 | BUG_ON(!thresholds); | |
4434 | ||
371528ca AV |
4435 | if (!thresholds->primary) |
4436 | goto unlock; | |
4437 | ||
2e72b634 KS |
4438 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); |
4439 | ||
4440 | /* Check if a threshold crossed before removing */ | |
4441 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4442 | ||
4443 | /* Calculate new number of threshold */ | |
2c488db2 KS |
4444 | size = 0; |
4445 | for (i = 0; i < thresholds->primary->size; i++) { | |
4446 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
4447 | size++; |
4448 | } | |
4449 | ||
2c488db2 | 4450 | new = thresholds->spare; |
907860ed | 4451 | |
2e72b634 KS |
4452 | /* Set thresholds array to NULL if we don't have thresholds */ |
4453 | if (!size) { | |
2c488db2 KS |
4454 | kfree(new); |
4455 | new = NULL; | |
907860ed | 4456 | goto swap_buffers; |
2e72b634 KS |
4457 | } |
4458 | ||
2c488db2 | 4459 | new->size = size; |
2e72b634 KS |
4460 | |
4461 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4462 | new->current_threshold = -1; |
4463 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4464 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4465 | continue; |
4466 | ||
2c488db2 KS |
4467 | new->entries[j] = thresholds->primary->entries[i]; |
4468 | if (new->entries[j].threshold < usage) { | |
2e72b634 | 4469 | /* |
2c488db2 | 4470 | * new->current_threshold will not be used |
2e72b634 KS |
4471 | * until rcu_assign_pointer(), so it's safe to increment |
4472 | * it here. | |
4473 | */ | |
2c488db2 | 4474 | ++new->current_threshold; |
2e72b634 KS |
4475 | } |
4476 | j++; | |
4477 | } | |
4478 | ||
907860ed | 4479 | swap_buffers: |
2c488db2 KS |
4480 | /* Swap primary and spare array */ |
4481 | thresholds->spare = thresholds->primary; | |
4482 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4483 | |
907860ed | 4484 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4485 | synchronize_rcu(); |
371528ca | 4486 | unlock: |
2e72b634 | 4487 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4488 | } |
c1e862c1 | 4489 | |
9490ff27 KH |
4490 | static int mem_cgroup_oom_register_event(struct cgroup *cgrp, |
4491 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
4492 | { | |
4493 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4494 | struct mem_cgroup_eventfd_list *event; | |
4495 | int type = MEMFILE_TYPE(cft->private); | |
4496 | ||
4497 | BUG_ON(type != _OOM_TYPE); | |
4498 | event = kmalloc(sizeof(*event), GFP_KERNEL); | |
4499 | if (!event) | |
4500 | return -ENOMEM; | |
4501 | ||
1af8efe9 | 4502 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4503 | |
4504 | event->eventfd = eventfd; | |
4505 | list_add(&event->list, &memcg->oom_notify); | |
4506 | ||
4507 | /* already in OOM ? */ | |
79dfdacc | 4508 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 4509 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4510 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4511 | |
4512 | return 0; | |
4513 | } | |
4514 | ||
907860ed | 4515 | static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp, |
9490ff27 KH |
4516 | struct cftype *cft, struct eventfd_ctx *eventfd) |
4517 | { | |
c0ff4b85 | 4518 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
9490ff27 KH |
4519 | struct mem_cgroup_eventfd_list *ev, *tmp; |
4520 | int type = MEMFILE_TYPE(cft->private); | |
4521 | ||
4522 | BUG_ON(type != _OOM_TYPE); | |
4523 | ||
1af8efe9 | 4524 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4525 | |
c0ff4b85 | 4526 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4527 | if (ev->eventfd == eventfd) { |
4528 | list_del(&ev->list); | |
4529 | kfree(ev); | |
4530 | } | |
4531 | } | |
4532 | ||
1af8efe9 | 4533 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4534 | } |
4535 | ||
3c11ecf4 KH |
4536 | static int mem_cgroup_oom_control_read(struct cgroup *cgrp, |
4537 | struct cftype *cft, struct cgroup_map_cb *cb) | |
4538 | { | |
c0ff4b85 | 4539 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 | 4540 | |
c0ff4b85 | 4541 | cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable); |
3c11ecf4 | 4542 | |
c0ff4b85 | 4543 | if (atomic_read(&memcg->under_oom)) |
3c11ecf4 KH |
4544 | cb->fill(cb, "under_oom", 1); |
4545 | else | |
4546 | cb->fill(cb, "under_oom", 0); | |
4547 | return 0; | |
4548 | } | |
4549 | ||
3c11ecf4 KH |
4550 | static int mem_cgroup_oom_control_write(struct cgroup *cgrp, |
4551 | struct cftype *cft, u64 val) | |
4552 | { | |
c0ff4b85 | 4553 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 KH |
4554 | struct mem_cgroup *parent; |
4555 | ||
4556 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
4557 | if (!cgrp->parent || !((val == 0) || (val == 1))) | |
4558 | return -EINVAL; | |
4559 | ||
4560 | parent = mem_cgroup_from_cont(cgrp->parent); | |
4561 | ||
4562 | cgroup_lock(); | |
4563 | /* oom-kill-disable is a flag for subhierarchy. */ | |
4564 | if ((parent->use_hierarchy) || | |
c0ff4b85 | 4565 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
3c11ecf4 KH |
4566 | cgroup_unlock(); |
4567 | return -EINVAL; | |
4568 | } | |
c0ff4b85 | 4569 | memcg->oom_kill_disable = val; |
4d845ebf | 4570 | if (!val) |
c0ff4b85 | 4571 | memcg_oom_recover(memcg); |
3c11ecf4 KH |
4572 | cgroup_unlock(); |
4573 | return 0; | |
4574 | } | |
4575 | ||
406eb0c9 YH |
4576 | #ifdef CONFIG_NUMA |
4577 | static const struct file_operations mem_control_numa_stat_file_operations = { | |
4578 | .read = seq_read, | |
4579 | .llseek = seq_lseek, | |
4580 | .release = single_release, | |
4581 | }; | |
4582 | ||
4583 | static int mem_control_numa_stat_open(struct inode *unused, struct file *file) | |
4584 | { | |
4585 | struct cgroup *cont = file->f_dentry->d_parent->d_fsdata; | |
4586 | ||
4587 | file->f_op = &mem_control_numa_stat_file_operations; | |
4588 | return single_open(file, mem_control_numa_stat_show, cont); | |
4589 | } | |
4590 | #endif /* CONFIG_NUMA */ | |
4591 | ||
e5671dfa | 4592 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM |
e5671dfa GC |
4593 | static int register_kmem_files(struct cgroup *cont, struct cgroup_subsys *ss) |
4594 | { | |
d1a4c0b3 GC |
4595 | /* |
4596 | * Part of this would be better living in a separate allocation | |
4597 | * function, leaving us with just the cgroup tree population work. | |
4598 | * We, however, depend on state such as network's proto_list that | |
4599 | * is only initialized after cgroup creation. I found the less | |
4600 | * cumbersome way to deal with it to defer it all to populate time | |
4601 | */ | |
65c64ce8 | 4602 | return mem_cgroup_sockets_init(cont, ss); |
e5671dfa GC |
4603 | }; |
4604 | ||
761b3ef5 | 4605 | static void kmem_cgroup_destroy(struct cgroup *cont) |
d1a4c0b3 | 4606 | { |
761b3ef5 | 4607 | mem_cgroup_sockets_destroy(cont); |
d1a4c0b3 | 4608 | } |
e5671dfa GC |
4609 | #else |
4610 | static int register_kmem_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
4611 | { | |
4612 | return 0; | |
4613 | } | |
d1a4c0b3 | 4614 | |
761b3ef5 | 4615 | static void kmem_cgroup_destroy(struct cgroup *cont) |
d1a4c0b3 GC |
4616 | { |
4617 | } | |
e5671dfa GC |
4618 | #endif |
4619 | ||
8cdea7c0 BS |
4620 | static struct cftype mem_cgroup_files[] = { |
4621 | { | |
0eea1030 | 4622 | .name = "usage_in_bytes", |
8c7c6e34 | 4623 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
2c3daa72 | 4624 | .read_u64 = mem_cgroup_read, |
9490ff27 KH |
4625 | .register_event = mem_cgroup_usage_register_event, |
4626 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8cdea7c0 | 4627 | }, |
c84872e1 PE |
4628 | { |
4629 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4630 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 4631 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
4632 | .read_u64 = mem_cgroup_read, |
4633 | }, | |
8cdea7c0 | 4634 | { |
0eea1030 | 4635 | .name = "limit_in_bytes", |
8c7c6e34 | 4636 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 4637 | .write_string = mem_cgroup_write, |
2c3daa72 | 4638 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 4639 | }, |
296c81d8 BS |
4640 | { |
4641 | .name = "soft_limit_in_bytes", | |
4642 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
4643 | .write_string = mem_cgroup_write, | |
4644 | .read_u64 = mem_cgroup_read, | |
4645 | }, | |
8cdea7c0 BS |
4646 | { |
4647 | .name = "failcnt", | |
8c7c6e34 | 4648 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 4649 | .trigger = mem_cgroup_reset, |
2c3daa72 | 4650 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 4651 | }, |
d2ceb9b7 KH |
4652 | { |
4653 | .name = "stat", | |
c64745cf | 4654 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 4655 | }, |
c1e862c1 KH |
4656 | { |
4657 | .name = "force_empty", | |
4658 | .trigger = mem_cgroup_force_empty_write, | |
4659 | }, | |
18f59ea7 BS |
4660 | { |
4661 | .name = "use_hierarchy", | |
4662 | .write_u64 = mem_cgroup_hierarchy_write, | |
4663 | .read_u64 = mem_cgroup_hierarchy_read, | |
4664 | }, | |
a7885eb8 KM |
4665 | { |
4666 | .name = "swappiness", | |
4667 | .read_u64 = mem_cgroup_swappiness_read, | |
4668 | .write_u64 = mem_cgroup_swappiness_write, | |
4669 | }, | |
7dc74be0 DN |
4670 | { |
4671 | .name = "move_charge_at_immigrate", | |
4672 | .read_u64 = mem_cgroup_move_charge_read, | |
4673 | .write_u64 = mem_cgroup_move_charge_write, | |
4674 | }, | |
9490ff27 KH |
4675 | { |
4676 | .name = "oom_control", | |
3c11ecf4 KH |
4677 | .read_map = mem_cgroup_oom_control_read, |
4678 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
4679 | .register_event = mem_cgroup_oom_register_event, |
4680 | .unregister_event = mem_cgroup_oom_unregister_event, | |
4681 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), | |
4682 | }, | |
406eb0c9 YH |
4683 | #ifdef CONFIG_NUMA |
4684 | { | |
4685 | .name = "numa_stat", | |
4686 | .open = mem_control_numa_stat_open, | |
89577127 | 4687 | .mode = S_IRUGO, |
406eb0c9 YH |
4688 | }, |
4689 | #endif | |
8cdea7c0 BS |
4690 | }; |
4691 | ||
8c7c6e34 KH |
4692 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
4693 | static struct cftype memsw_cgroup_files[] = { | |
4694 | { | |
4695 | .name = "memsw.usage_in_bytes", | |
4696 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
4697 | .read_u64 = mem_cgroup_read, | |
9490ff27 KH |
4698 | .register_event = mem_cgroup_usage_register_event, |
4699 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8c7c6e34 KH |
4700 | }, |
4701 | { | |
4702 | .name = "memsw.max_usage_in_bytes", | |
4703 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
4704 | .trigger = mem_cgroup_reset, | |
4705 | .read_u64 = mem_cgroup_read, | |
4706 | }, | |
4707 | { | |
4708 | .name = "memsw.limit_in_bytes", | |
4709 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
4710 | .write_string = mem_cgroup_write, | |
4711 | .read_u64 = mem_cgroup_read, | |
4712 | }, | |
4713 | { | |
4714 | .name = "memsw.failcnt", | |
4715 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
4716 | .trigger = mem_cgroup_reset, | |
4717 | .read_u64 = mem_cgroup_read, | |
4718 | }, | |
4719 | }; | |
4720 | ||
4721 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
4722 | { | |
4723 | if (!do_swap_account) | |
4724 | return 0; | |
4725 | return cgroup_add_files(cont, ss, memsw_cgroup_files, | |
4726 | ARRAY_SIZE(memsw_cgroup_files)); | |
4727 | }; | |
4728 | #else | |
4729 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
4730 | { | |
4731 | return 0; | |
4732 | } | |
4733 | #endif | |
4734 | ||
c0ff4b85 | 4735 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4736 | { |
4737 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4738 | struct mem_cgroup_per_zone *mz; |
f156ab93 | 4739 | enum lru_list lru; |
41e3355d | 4740 | int zone, tmp = node; |
1ecaab2b KH |
4741 | /* |
4742 | * This routine is called against possible nodes. | |
4743 | * But it's BUG to call kmalloc() against offline node. | |
4744 | * | |
4745 | * TODO: this routine can waste much memory for nodes which will | |
4746 | * never be onlined. It's better to use memory hotplug callback | |
4747 | * function. | |
4748 | */ | |
41e3355d KH |
4749 | if (!node_state(node, N_NORMAL_MEMORY)) |
4750 | tmp = -1; | |
17295c88 | 4751 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4752 | if (!pn) |
4753 | return 1; | |
1ecaab2b | 4754 | |
1ecaab2b KH |
4755 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4756 | mz = &pn->zoneinfo[zone]; | |
f156ab93 HD |
4757 | for_each_lru(lru) |
4758 | INIT_LIST_HEAD(&mz->lruvec.lists[lru]); | |
f64c3f54 | 4759 | mz->usage_in_excess = 0; |
4e416953 | 4760 | mz->on_tree = false; |
d79154bb | 4761 | mz->memcg = memcg; |
1ecaab2b | 4762 | } |
0a619e58 | 4763 | memcg->info.nodeinfo[node] = pn; |
6d12e2d8 KH |
4764 | return 0; |
4765 | } | |
4766 | ||
c0ff4b85 | 4767 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4768 | { |
c0ff4b85 | 4769 | kfree(memcg->info.nodeinfo[node]); |
1ecaab2b KH |
4770 | } |
4771 | ||
33327948 KH |
4772 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4773 | { | |
d79154bb | 4774 | struct mem_cgroup *memcg; |
c62b1a3b | 4775 | int size = sizeof(struct mem_cgroup); |
33327948 | 4776 | |
c62b1a3b | 4777 | /* Can be very big if MAX_NUMNODES is very big */ |
c8dad2bb | 4778 | if (size < PAGE_SIZE) |
d79154bb | 4779 | memcg = kzalloc(size, GFP_KERNEL); |
33327948 | 4780 | else |
d79154bb | 4781 | memcg = vzalloc(size); |
33327948 | 4782 | |
d79154bb | 4783 | if (!memcg) |
e7bbcdf3 DC |
4784 | return NULL; |
4785 | ||
d79154bb HD |
4786 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
4787 | if (!memcg->stat) | |
d2e61b8d | 4788 | goto out_free; |
d79154bb HD |
4789 | spin_lock_init(&memcg->pcp_counter_lock); |
4790 | return memcg; | |
d2e61b8d DC |
4791 | |
4792 | out_free: | |
4793 | if (size < PAGE_SIZE) | |
d79154bb | 4794 | kfree(memcg); |
d2e61b8d | 4795 | else |
d79154bb | 4796 | vfree(memcg); |
d2e61b8d | 4797 | return NULL; |
33327948 KH |
4798 | } |
4799 | ||
59927fb9 HD |
4800 | /* |
4801 | * Helpers for freeing a vzalloc()ed mem_cgroup by RCU, | |
4802 | * but in process context. The work_freeing structure is overlaid | |
4803 | * on the rcu_freeing structure, which itself is overlaid on memsw. | |
4804 | */ | |
4805 | static void vfree_work(struct work_struct *work) | |
4806 | { | |
4807 | struct mem_cgroup *memcg; | |
4808 | ||
4809 | memcg = container_of(work, struct mem_cgroup, work_freeing); | |
4810 | vfree(memcg); | |
4811 | } | |
4812 | static void vfree_rcu(struct rcu_head *rcu_head) | |
4813 | { | |
4814 | struct mem_cgroup *memcg; | |
4815 | ||
4816 | memcg = container_of(rcu_head, struct mem_cgroup, rcu_freeing); | |
4817 | INIT_WORK(&memcg->work_freeing, vfree_work); | |
4818 | schedule_work(&memcg->work_freeing); | |
4819 | } | |
4820 | ||
8c7c6e34 KH |
4821 | /* |
4822 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
4823 | * (scanning all at force_empty is too costly...) | |
4824 | * | |
4825 | * Instead of clearing all references at force_empty, we remember | |
4826 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4827 | * it goes down to 0. | |
4828 | * | |
8c7c6e34 KH |
4829 | * Removal of cgroup itself succeeds regardless of refs from swap. |
4830 | */ | |
4831 | ||
c0ff4b85 | 4832 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
33327948 | 4833 | { |
08e552c6 KH |
4834 | int node; |
4835 | ||
c0ff4b85 R |
4836 | mem_cgroup_remove_from_trees(memcg); |
4837 | free_css_id(&mem_cgroup_subsys, &memcg->css); | |
04046e1a | 4838 | |
3ed28fa1 | 4839 | for_each_node(node) |
c0ff4b85 | 4840 | free_mem_cgroup_per_zone_info(memcg, node); |
08e552c6 | 4841 | |
c0ff4b85 | 4842 | free_percpu(memcg->stat); |
c62b1a3b | 4843 | if (sizeof(struct mem_cgroup) < PAGE_SIZE) |
59927fb9 | 4844 | kfree_rcu(memcg, rcu_freeing); |
33327948 | 4845 | else |
59927fb9 | 4846 | call_rcu(&memcg->rcu_freeing, vfree_rcu); |
33327948 KH |
4847 | } |
4848 | ||
c0ff4b85 | 4849 | static void mem_cgroup_get(struct mem_cgroup *memcg) |
8c7c6e34 | 4850 | { |
c0ff4b85 | 4851 | atomic_inc(&memcg->refcnt); |
8c7c6e34 KH |
4852 | } |
4853 | ||
c0ff4b85 | 4854 | static void __mem_cgroup_put(struct mem_cgroup *memcg, int count) |
8c7c6e34 | 4855 | { |
c0ff4b85 R |
4856 | if (atomic_sub_and_test(count, &memcg->refcnt)) { |
4857 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); | |
4858 | __mem_cgroup_free(memcg); | |
7bcc1bb1 DN |
4859 | if (parent) |
4860 | mem_cgroup_put(parent); | |
4861 | } | |
8c7c6e34 KH |
4862 | } |
4863 | ||
c0ff4b85 | 4864 | static void mem_cgroup_put(struct mem_cgroup *memcg) |
483c30b5 | 4865 | { |
c0ff4b85 | 4866 | __mem_cgroup_put(memcg, 1); |
483c30b5 DN |
4867 | } |
4868 | ||
7bcc1bb1 DN |
4869 | /* |
4870 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4871 | */ | |
e1aab161 | 4872 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 4873 | { |
c0ff4b85 | 4874 | if (!memcg->res.parent) |
7bcc1bb1 | 4875 | return NULL; |
c0ff4b85 | 4876 | return mem_cgroup_from_res_counter(memcg->res.parent, res); |
7bcc1bb1 | 4877 | } |
e1aab161 | 4878 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 4879 | |
c077719b KH |
4880 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
4881 | static void __init enable_swap_cgroup(void) | |
4882 | { | |
f8d66542 | 4883 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
4884 | do_swap_account = 1; |
4885 | } | |
4886 | #else | |
4887 | static void __init enable_swap_cgroup(void) | |
4888 | { | |
4889 | } | |
4890 | #endif | |
4891 | ||
f64c3f54 BS |
4892 | static int mem_cgroup_soft_limit_tree_init(void) |
4893 | { | |
4894 | struct mem_cgroup_tree_per_node *rtpn; | |
4895 | struct mem_cgroup_tree_per_zone *rtpz; | |
4896 | int tmp, node, zone; | |
4897 | ||
3ed28fa1 | 4898 | for_each_node(node) { |
f64c3f54 BS |
4899 | tmp = node; |
4900 | if (!node_state(node, N_NORMAL_MEMORY)) | |
4901 | tmp = -1; | |
4902 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
4903 | if (!rtpn) | |
c3cecc68 | 4904 | goto err_cleanup; |
f64c3f54 BS |
4905 | |
4906 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
4907 | ||
4908 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
4909 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
4910 | rtpz->rb_root = RB_ROOT; | |
4911 | spin_lock_init(&rtpz->lock); | |
4912 | } | |
4913 | } | |
4914 | return 0; | |
c3cecc68 MH |
4915 | |
4916 | err_cleanup: | |
3ed28fa1 | 4917 | for_each_node(node) { |
c3cecc68 MH |
4918 | if (!soft_limit_tree.rb_tree_per_node[node]) |
4919 | break; | |
4920 | kfree(soft_limit_tree.rb_tree_per_node[node]); | |
4921 | soft_limit_tree.rb_tree_per_node[node] = NULL; | |
4922 | } | |
4923 | return 1; | |
4924 | ||
f64c3f54 BS |
4925 | } |
4926 | ||
0eb253e2 | 4927 | static struct cgroup_subsys_state * __ref |
761b3ef5 | 4928 | mem_cgroup_create(struct cgroup *cont) |
8cdea7c0 | 4929 | { |
c0ff4b85 | 4930 | struct mem_cgroup *memcg, *parent; |
04046e1a | 4931 | long error = -ENOMEM; |
6d12e2d8 | 4932 | int node; |
8cdea7c0 | 4933 | |
c0ff4b85 R |
4934 | memcg = mem_cgroup_alloc(); |
4935 | if (!memcg) | |
04046e1a | 4936 | return ERR_PTR(error); |
78fb7466 | 4937 | |
3ed28fa1 | 4938 | for_each_node(node) |
c0ff4b85 | 4939 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 4940 | goto free_out; |
f64c3f54 | 4941 | |
c077719b | 4942 | /* root ? */ |
28dbc4b6 | 4943 | if (cont->parent == NULL) { |
cdec2e42 | 4944 | int cpu; |
c077719b | 4945 | enable_swap_cgroup(); |
28dbc4b6 | 4946 | parent = NULL; |
f64c3f54 BS |
4947 | if (mem_cgroup_soft_limit_tree_init()) |
4948 | goto free_out; | |
a41c58a6 | 4949 | root_mem_cgroup = memcg; |
cdec2e42 KH |
4950 | for_each_possible_cpu(cpu) { |
4951 | struct memcg_stock_pcp *stock = | |
4952 | &per_cpu(memcg_stock, cpu); | |
4953 | INIT_WORK(&stock->work, drain_local_stock); | |
4954 | } | |
711d3d2c | 4955 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
18f59ea7 | 4956 | } else { |
28dbc4b6 | 4957 | parent = mem_cgroup_from_cont(cont->parent); |
c0ff4b85 R |
4958 | memcg->use_hierarchy = parent->use_hierarchy; |
4959 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
18f59ea7 | 4960 | } |
28dbc4b6 | 4961 | |
18f59ea7 | 4962 | if (parent && parent->use_hierarchy) { |
c0ff4b85 R |
4963 | res_counter_init(&memcg->res, &parent->res); |
4964 | res_counter_init(&memcg->memsw, &parent->memsw); | |
7bcc1bb1 DN |
4965 | /* |
4966 | * We increment refcnt of the parent to ensure that we can | |
4967 | * safely access it on res_counter_charge/uncharge. | |
4968 | * This refcnt will be decremented when freeing this | |
4969 | * mem_cgroup(see mem_cgroup_put). | |
4970 | */ | |
4971 | mem_cgroup_get(parent); | |
18f59ea7 | 4972 | } else { |
c0ff4b85 R |
4973 | res_counter_init(&memcg->res, NULL); |
4974 | res_counter_init(&memcg->memsw, NULL); | |
18f59ea7 | 4975 | } |
c0ff4b85 R |
4976 | memcg->last_scanned_node = MAX_NUMNODES; |
4977 | INIT_LIST_HEAD(&memcg->oom_notify); | |
6d61ef40 | 4978 | |
a7885eb8 | 4979 | if (parent) |
c0ff4b85 R |
4980 | memcg->swappiness = mem_cgroup_swappiness(parent); |
4981 | atomic_set(&memcg->refcnt, 1); | |
4982 | memcg->move_charge_at_immigrate = 0; | |
4983 | mutex_init(&memcg->thresholds_lock); | |
4984 | return &memcg->css; | |
6d12e2d8 | 4985 | free_out: |
c0ff4b85 | 4986 | __mem_cgroup_free(memcg); |
04046e1a | 4987 | return ERR_PTR(error); |
8cdea7c0 BS |
4988 | } |
4989 | ||
761b3ef5 | 4990 | static int mem_cgroup_pre_destroy(struct cgroup *cont) |
df878fb0 | 4991 | { |
c0ff4b85 | 4992 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
ec64f515 | 4993 | |
c0ff4b85 | 4994 | return mem_cgroup_force_empty(memcg, false); |
df878fb0 KH |
4995 | } |
4996 | ||
761b3ef5 | 4997 | static void mem_cgroup_destroy(struct cgroup *cont) |
8cdea7c0 | 4998 | { |
c0ff4b85 | 4999 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
c268e994 | 5000 | |
761b3ef5 | 5001 | kmem_cgroup_destroy(cont); |
d1a4c0b3 | 5002 | |
c0ff4b85 | 5003 | mem_cgroup_put(memcg); |
8cdea7c0 BS |
5004 | } |
5005 | ||
5006 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
5007 | struct cgroup *cont) | |
5008 | { | |
8c7c6e34 KH |
5009 | int ret; |
5010 | ||
5011 | ret = cgroup_add_files(cont, ss, mem_cgroup_files, | |
5012 | ARRAY_SIZE(mem_cgroup_files)); | |
5013 | ||
5014 | if (!ret) | |
5015 | ret = register_memsw_files(cont, ss); | |
e5671dfa GC |
5016 | |
5017 | if (!ret) | |
5018 | ret = register_kmem_files(cont, ss); | |
5019 | ||
8c7c6e34 | 5020 | return ret; |
8cdea7c0 BS |
5021 | } |
5022 | ||
02491447 | 5023 | #ifdef CONFIG_MMU |
7dc74be0 | 5024 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
5025 | #define PRECHARGE_COUNT_AT_ONCE 256 |
5026 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 5027 | { |
854ffa8d DN |
5028 | int ret = 0; |
5029 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
c0ff4b85 | 5030 | struct mem_cgroup *memcg = mc.to; |
4ffef5fe | 5031 | |
c0ff4b85 | 5032 | if (mem_cgroup_is_root(memcg)) { |
854ffa8d DN |
5033 | mc.precharge += count; |
5034 | /* we don't need css_get for root */ | |
5035 | return ret; | |
5036 | } | |
5037 | /* try to charge at once */ | |
5038 | if (count > 1) { | |
5039 | struct res_counter *dummy; | |
5040 | /* | |
c0ff4b85 | 5041 | * "memcg" cannot be under rmdir() because we've already checked |
854ffa8d DN |
5042 | * by cgroup_lock_live_cgroup() that it is not removed and we |
5043 | * are still under the same cgroup_mutex. So we can postpone | |
5044 | * css_get(). | |
5045 | */ | |
c0ff4b85 | 5046 | if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy)) |
854ffa8d | 5047 | goto one_by_one; |
c0ff4b85 | 5048 | if (do_swap_account && res_counter_charge(&memcg->memsw, |
854ffa8d | 5049 | PAGE_SIZE * count, &dummy)) { |
c0ff4b85 | 5050 | res_counter_uncharge(&memcg->res, PAGE_SIZE * count); |
854ffa8d DN |
5051 | goto one_by_one; |
5052 | } | |
5053 | mc.precharge += count; | |
854ffa8d DN |
5054 | return ret; |
5055 | } | |
5056 | one_by_one: | |
5057 | /* fall back to one by one charge */ | |
5058 | while (count--) { | |
5059 | if (signal_pending(current)) { | |
5060 | ret = -EINTR; | |
5061 | break; | |
5062 | } | |
5063 | if (!batch_count--) { | |
5064 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
5065 | cond_resched(); | |
5066 | } | |
c0ff4b85 R |
5067 | ret = __mem_cgroup_try_charge(NULL, |
5068 | GFP_KERNEL, 1, &memcg, false); | |
38c5d72f | 5069 | if (ret) |
854ffa8d | 5070 | /* mem_cgroup_clear_mc() will do uncharge later */ |
38c5d72f | 5071 | return ret; |
854ffa8d DN |
5072 | mc.precharge++; |
5073 | } | |
4ffef5fe DN |
5074 | return ret; |
5075 | } | |
5076 | ||
5077 | /** | |
5078 | * is_target_pte_for_mc - check a pte whether it is valid for move charge | |
5079 | * @vma: the vma the pte to be checked belongs | |
5080 | * @addr: the address corresponding to the pte to be checked | |
5081 | * @ptent: the pte to be checked | |
02491447 | 5082 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
5083 | * |
5084 | * Returns | |
5085 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5086 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5087 | * move charge. if @target is not NULL, the page is stored in target->page | |
5088 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
5089 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
5090 | * target for charge migration. if @target is not NULL, the entry is stored | |
5091 | * in target->ent. | |
4ffef5fe DN |
5092 | * |
5093 | * Called with pte lock held. | |
5094 | */ | |
4ffef5fe DN |
5095 | union mc_target { |
5096 | struct page *page; | |
02491447 | 5097 | swp_entry_t ent; |
4ffef5fe DN |
5098 | }; |
5099 | ||
4ffef5fe DN |
5100 | enum mc_target_type { |
5101 | MC_TARGET_NONE, /* not used */ | |
5102 | MC_TARGET_PAGE, | |
02491447 | 5103 | MC_TARGET_SWAP, |
4ffef5fe DN |
5104 | }; |
5105 | ||
90254a65 DN |
5106 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5107 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5108 | { |
90254a65 | 5109 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5110 | |
90254a65 DN |
5111 | if (!page || !page_mapped(page)) |
5112 | return NULL; | |
5113 | if (PageAnon(page)) { | |
5114 | /* we don't move shared anon */ | |
be22aece | 5115 | if (!move_anon() || page_mapcount(page) > 2) |
90254a65 | 5116 | return NULL; |
87946a72 DN |
5117 | } else if (!move_file()) |
5118 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
5119 | return NULL; |
5120 | if (!get_page_unless_zero(page)) | |
5121 | return NULL; | |
5122 | ||
5123 | return page; | |
5124 | } | |
5125 | ||
5126 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
5127 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5128 | { | |
5129 | int usage_count; | |
5130 | struct page *page = NULL; | |
5131 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5132 | ||
5133 | if (!move_anon() || non_swap_entry(ent)) | |
5134 | return NULL; | |
5135 | usage_count = mem_cgroup_count_swap_user(ent, &page); | |
5136 | if (usage_count > 1) { /* we don't move shared anon */ | |
02491447 DN |
5137 | if (page) |
5138 | put_page(page); | |
90254a65 | 5139 | return NULL; |
02491447 | 5140 | } |
90254a65 DN |
5141 | if (do_swap_account) |
5142 | entry->val = ent.val; | |
5143 | ||
5144 | return page; | |
5145 | } | |
5146 | ||
87946a72 DN |
5147 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5148 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5149 | { | |
5150 | struct page *page = NULL; | |
5151 | struct inode *inode; | |
5152 | struct address_space *mapping; | |
5153 | pgoff_t pgoff; | |
5154 | ||
5155 | if (!vma->vm_file) /* anonymous vma */ | |
5156 | return NULL; | |
5157 | if (!move_file()) | |
5158 | return NULL; | |
5159 | ||
5160 | inode = vma->vm_file->f_path.dentry->d_inode; | |
5161 | mapping = vma->vm_file->f_mapping; | |
5162 | if (pte_none(ptent)) | |
5163 | pgoff = linear_page_index(vma, addr); | |
5164 | else /* pte_file(ptent) is true */ | |
5165 | pgoff = pte_to_pgoff(ptent); | |
5166 | ||
5167 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
5168 | page = find_get_page(mapping, pgoff); |
5169 | ||
5170 | #ifdef CONFIG_SWAP | |
5171 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
5172 | if (radix_tree_exceptional_entry(page)) { | |
5173 | swp_entry_t swap = radix_to_swp_entry(page); | |
87946a72 | 5174 | if (do_swap_account) |
aa3b1895 HD |
5175 | *entry = swap; |
5176 | page = find_get_page(&swapper_space, swap.val); | |
87946a72 | 5177 | } |
aa3b1895 | 5178 | #endif |
87946a72 DN |
5179 | return page; |
5180 | } | |
5181 | ||
90254a65 DN |
5182 | static int is_target_pte_for_mc(struct vm_area_struct *vma, |
5183 | unsigned long addr, pte_t ptent, union mc_target *target) | |
5184 | { | |
5185 | struct page *page = NULL; | |
5186 | struct page_cgroup *pc; | |
5187 | int ret = 0; | |
5188 | swp_entry_t ent = { .val = 0 }; | |
5189 | ||
5190 | if (pte_present(ptent)) | |
5191 | page = mc_handle_present_pte(vma, addr, ptent); | |
5192 | else if (is_swap_pte(ptent)) | |
5193 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
5194 | else if (pte_none(ptent) || pte_file(ptent)) |
5195 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
5196 | |
5197 | if (!page && !ent.val) | |
5198 | return 0; | |
02491447 DN |
5199 | if (page) { |
5200 | pc = lookup_page_cgroup(page); | |
5201 | /* | |
5202 | * Do only loose check w/o page_cgroup lock. | |
5203 | * mem_cgroup_move_account() checks the pc is valid or not under | |
5204 | * the lock. | |
5205 | */ | |
5206 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5207 | ret = MC_TARGET_PAGE; | |
5208 | if (target) | |
5209 | target->page = page; | |
5210 | } | |
5211 | if (!ret || !target) | |
5212 | put_page(page); | |
5213 | } | |
90254a65 DN |
5214 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
5215 | if (ent.val && !ret && | |
9fb4b7cc | 5216 | css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5217 | ret = MC_TARGET_SWAP; |
5218 | if (target) | |
5219 | target->ent = ent; | |
4ffef5fe | 5220 | } |
4ffef5fe DN |
5221 | return ret; |
5222 | } | |
5223 | ||
5224 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, | |
5225 | unsigned long addr, unsigned long end, | |
5226 | struct mm_walk *walk) | |
5227 | { | |
5228 | struct vm_area_struct *vma = walk->private; | |
5229 | pte_t *pte; | |
5230 | spinlock_t *ptl; | |
5231 | ||
03319327 | 5232 | split_huge_page_pmd(walk->mm, pmd); |
1a5a9906 AA |
5233 | if (pmd_trans_unstable(pmd)) |
5234 | return 0; | |
03319327 | 5235 | |
4ffef5fe DN |
5236 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5237 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
5238 | if (is_target_pte_for_mc(vma, addr, *pte, NULL)) | |
5239 | mc.precharge++; /* increment precharge temporarily */ | |
5240 | pte_unmap_unlock(pte - 1, ptl); | |
5241 | cond_resched(); | |
5242 | ||
7dc74be0 DN |
5243 | return 0; |
5244 | } | |
5245 | ||
4ffef5fe DN |
5246 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5247 | { | |
5248 | unsigned long precharge; | |
5249 | struct vm_area_struct *vma; | |
5250 | ||
dfe076b0 | 5251 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
5252 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5253 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
5254 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5255 | .mm = mm, | |
5256 | .private = vma, | |
5257 | }; | |
5258 | if (is_vm_hugetlb_page(vma)) | |
5259 | continue; | |
4ffef5fe DN |
5260 | walk_page_range(vma->vm_start, vma->vm_end, |
5261 | &mem_cgroup_count_precharge_walk); | |
5262 | } | |
dfe076b0 | 5263 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5264 | |
5265 | precharge = mc.precharge; | |
5266 | mc.precharge = 0; | |
5267 | ||
5268 | return precharge; | |
5269 | } | |
5270 | ||
4ffef5fe DN |
5271 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5272 | { | |
dfe076b0 DN |
5273 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5274 | ||
5275 | VM_BUG_ON(mc.moving_task); | |
5276 | mc.moving_task = current; | |
5277 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5278 | } |
5279 | ||
dfe076b0 DN |
5280 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5281 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5282 | { |
2bd9bb20 KH |
5283 | struct mem_cgroup *from = mc.from; |
5284 | struct mem_cgroup *to = mc.to; | |
5285 | ||
4ffef5fe | 5286 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
5287 | if (mc.precharge) { |
5288 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
5289 | mc.precharge = 0; | |
5290 | } | |
5291 | /* | |
5292 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5293 | * we must uncharge here. | |
5294 | */ | |
5295 | if (mc.moved_charge) { | |
5296 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
5297 | mc.moved_charge = 0; | |
4ffef5fe | 5298 | } |
483c30b5 DN |
5299 | /* we must fixup refcnts and charges */ |
5300 | if (mc.moved_swap) { | |
483c30b5 DN |
5301 | /* uncharge swap account from the old cgroup */ |
5302 | if (!mem_cgroup_is_root(mc.from)) | |
5303 | res_counter_uncharge(&mc.from->memsw, | |
5304 | PAGE_SIZE * mc.moved_swap); | |
5305 | __mem_cgroup_put(mc.from, mc.moved_swap); | |
5306 | ||
5307 | if (!mem_cgroup_is_root(mc.to)) { | |
5308 | /* | |
5309 | * we charged both to->res and to->memsw, so we should | |
5310 | * uncharge to->res. | |
5311 | */ | |
5312 | res_counter_uncharge(&mc.to->res, | |
5313 | PAGE_SIZE * mc.moved_swap); | |
483c30b5 DN |
5314 | } |
5315 | /* we've already done mem_cgroup_get(mc.to) */ | |
483c30b5 DN |
5316 | mc.moved_swap = 0; |
5317 | } | |
dfe076b0 DN |
5318 | memcg_oom_recover(from); |
5319 | memcg_oom_recover(to); | |
5320 | wake_up_all(&mc.waitq); | |
5321 | } | |
5322 | ||
5323 | static void mem_cgroup_clear_mc(void) | |
5324 | { | |
5325 | struct mem_cgroup *from = mc.from; | |
5326 | ||
5327 | /* | |
5328 | * we must clear moving_task before waking up waiters at the end of | |
5329 | * task migration. | |
5330 | */ | |
5331 | mc.moving_task = NULL; | |
5332 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5333 | spin_lock(&mc.lock); |
4ffef5fe DN |
5334 | mc.from = NULL; |
5335 | mc.to = NULL; | |
2bd9bb20 | 5336 | spin_unlock(&mc.lock); |
32047e2a | 5337 | mem_cgroup_end_move(from); |
4ffef5fe DN |
5338 | } |
5339 | ||
761b3ef5 LZ |
5340 | static int mem_cgroup_can_attach(struct cgroup *cgroup, |
5341 | struct cgroup_taskset *tset) | |
7dc74be0 | 5342 | { |
2f7ee569 | 5343 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 5344 | int ret = 0; |
c0ff4b85 | 5345 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup); |
7dc74be0 | 5346 | |
c0ff4b85 | 5347 | if (memcg->move_charge_at_immigrate) { |
7dc74be0 DN |
5348 | struct mm_struct *mm; |
5349 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
5350 | ||
c0ff4b85 | 5351 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
5352 | |
5353 | mm = get_task_mm(p); | |
5354 | if (!mm) | |
5355 | return 0; | |
7dc74be0 | 5356 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
5357 | if (mm->owner == p) { |
5358 | VM_BUG_ON(mc.from); | |
5359 | VM_BUG_ON(mc.to); | |
5360 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 5361 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 5362 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 5363 | mem_cgroup_start_move(from); |
2bd9bb20 | 5364 | spin_lock(&mc.lock); |
4ffef5fe | 5365 | mc.from = from; |
c0ff4b85 | 5366 | mc.to = memcg; |
2bd9bb20 | 5367 | spin_unlock(&mc.lock); |
dfe076b0 | 5368 | /* We set mc.moving_task later */ |
4ffef5fe DN |
5369 | |
5370 | ret = mem_cgroup_precharge_mc(mm); | |
5371 | if (ret) | |
5372 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
5373 | } |
5374 | mmput(mm); | |
7dc74be0 DN |
5375 | } |
5376 | return ret; | |
5377 | } | |
5378 | ||
761b3ef5 LZ |
5379 | static void mem_cgroup_cancel_attach(struct cgroup *cgroup, |
5380 | struct cgroup_taskset *tset) | |
7dc74be0 | 5381 | { |
4ffef5fe | 5382 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
5383 | } |
5384 | ||
4ffef5fe DN |
5385 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5386 | unsigned long addr, unsigned long end, | |
5387 | struct mm_walk *walk) | |
7dc74be0 | 5388 | { |
4ffef5fe DN |
5389 | int ret = 0; |
5390 | struct vm_area_struct *vma = walk->private; | |
5391 | pte_t *pte; | |
5392 | spinlock_t *ptl; | |
5393 | ||
03319327 | 5394 | split_huge_page_pmd(walk->mm, pmd); |
1a5a9906 AA |
5395 | if (pmd_trans_unstable(pmd)) |
5396 | return 0; | |
4ffef5fe DN |
5397 | retry: |
5398 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5399 | for (; addr != end; addr += PAGE_SIZE) { | |
5400 | pte_t ptent = *(pte++); | |
5401 | union mc_target target; | |
5402 | int type; | |
5403 | struct page *page; | |
5404 | struct page_cgroup *pc; | |
02491447 | 5405 | swp_entry_t ent; |
4ffef5fe DN |
5406 | |
5407 | if (!mc.precharge) | |
5408 | break; | |
5409 | ||
5410 | type = is_target_pte_for_mc(vma, addr, ptent, &target); | |
5411 | switch (type) { | |
5412 | case MC_TARGET_PAGE: | |
5413 | page = target.page; | |
5414 | if (isolate_lru_page(page)) | |
5415 | goto put; | |
5416 | pc = lookup_page_cgroup(page); | |
7ec99d62 JW |
5417 | if (!mem_cgroup_move_account(page, 1, pc, |
5418 | mc.from, mc.to, false)) { | |
4ffef5fe | 5419 | mc.precharge--; |
854ffa8d DN |
5420 | /* we uncharge from mc.from later. */ |
5421 | mc.moved_charge++; | |
4ffef5fe DN |
5422 | } |
5423 | putback_lru_page(page); | |
5424 | put: /* is_target_pte_for_mc() gets the page */ | |
5425 | put_page(page); | |
5426 | break; | |
02491447 DN |
5427 | case MC_TARGET_SWAP: |
5428 | ent = target.ent; | |
483c30b5 DN |
5429 | if (!mem_cgroup_move_swap_account(ent, |
5430 | mc.from, mc.to, false)) { | |
02491447 | 5431 | mc.precharge--; |
483c30b5 DN |
5432 | /* we fixup refcnts and charges later. */ |
5433 | mc.moved_swap++; | |
5434 | } | |
02491447 | 5435 | break; |
4ffef5fe DN |
5436 | default: |
5437 | break; | |
5438 | } | |
5439 | } | |
5440 | pte_unmap_unlock(pte - 1, ptl); | |
5441 | cond_resched(); | |
5442 | ||
5443 | if (addr != end) { | |
5444 | /* | |
5445 | * We have consumed all precharges we got in can_attach(). | |
5446 | * We try charge one by one, but don't do any additional | |
5447 | * charges to mc.to if we have failed in charge once in attach() | |
5448 | * phase. | |
5449 | */ | |
854ffa8d | 5450 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5451 | if (!ret) |
5452 | goto retry; | |
5453 | } | |
5454 | ||
5455 | return ret; | |
5456 | } | |
5457 | ||
5458 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
5459 | { | |
5460 | struct vm_area_struct *vma; | |
5461 | ||
5462 | lru_add_drain_all(); | |
dfe076b0 DN |
5463 | retry: |
5464 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5465 | /* | |
5466 | * Someone who are holding the mmap_sem might be waiting in | |
5467 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5468 | * and retry. Because we cancel precharges, we might not be able | |
5469 | * to move enough charges, but moving charge is a best-effort | |
5470 | * feature anyway, so it wouldn't be a big problem. | |
5471 | */ | |
5472 | __mem_cgroup_clear_mc(); | |
5473 | cond_resched(); | |
5474 | goto retry; | |
5475 | } | |
4ffef5fe DN |
5476 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5477 | int ret; | |
5478 | struct mm_walk mem_cgroup_move_charge_walk = { | |
5479 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
5480 | .mm = mm, | |
5481 | .private = vma, | |
5482 | }; | |
5483 | if (is_vm_hugetlb_page(vma)) | |
5484 | continue; | |
4ffef5fe DN |
5485 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
5486 | &mem_cgroup_move_charge_walk); | |
5487 | if (ret) | |
5488 | /* | |
5489 | * means we have consumed all precharges and failed in | |
5490 | * doing additional charge. Just abandon here. | |
5491 | */ | |
5492 | break; | |
5493 | } | |
dfe076b0 | 5494 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
5495 | } |
5496 | ||
761b3ef5 LZ |
5497 | static void mem_cgroup_move_task(struct cgroup *cont, |
5498 | struct cgroup_taskset *tset) | |
67e465a7 | 5499 | { |
2f7ee569 | 5500 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 5501 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 5502 | |
dfe076b0 | 5503 | if (mm) { |
a433658c KM |
5504 | if (mc.to) |
5505 | mem_cgroup_move_charge(mm); | |
5506 | put_swap_token(mm); | |
dfe076b0 DN |
5507 | mmput(mm); |
5508 | } | |
a433658c KM |
5509 | if (mc.to) |
5510 | mem_cgroup_clear_mc(); | |
67e465a7 | 5511 | } |
5cfb80a7 | 5512 | #else /* !CONFIG_MMU */ |
761b3ef5 LZ |
5513 | static int mem_cgroup_can_attach(struct cgroup *cgroup, |
5514 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
5515 | { |
5516 | return 0; | |
5517 | } | |
761b3ef5 LZ |
5518 | static void mem_cgroup_cancel_attach(struct cgroup *cgroup, |
5519 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
5520 | { |
5521 | } | |
761b3ef5 LZ |
5522 | static void mem_cgroup_move_task(struct cgroup *cont, |
5523 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
5524 | { |
5525 | } | |
5526 | #endif | |
67e465a7 | 5527 | |
8cdea7c0 BS |
5528 | struct cgroup_subsys mem_cgroup_subsys = { |
5529 | .name = "memory", | |
5530 | .subsys_id = mem_cgroup_subsys_id, | |
5531 | .create = mem_cgroup_create, | |
df878fb0 | 5532 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
5533 | .destroy = mem_cgroup_destroy, |
5534 | .populate = mem_cgroup_populate, | |
7dc74be0 DN |
5535 | .can_attach = mem_cgroup_can_attach, |
5536 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5537 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 5538 | .early_init = 0, |
04046e1a | 5539 | .use_id = 1, |
8cdea7c0 | 5540 | }; |
c077719b KH |
5541 | |
5542 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
a42c390c MH |
5543 | static int __init enable_swap_account(char *s) |
5544 | { | |
5545 | /* consider enabled if no parameter or 1 is given */ | |
a2c8990a | 5546 | if (!strcmp(s, "1")) |
a42c390c | 5547 | really_do_swap_account = 1; |
a2c8990a | 5548 | else if (!strcmp(s, "0")) |
a42c390c MH |
5549 | really_do_swap_account = 0; |
5550 | return 1; | |
5551 | } | |
a2c8990a | 5552 | __setup("swapaccount=", enable_swap_account); |
c077719b | 5553 | |
c077719b | 5554 | #endif |