]>
Commit | Line | Data |
---|---|---|
8cdea7c0 BS |
1 | /* memcontrol.c - Memory Controller |
2 | * | |
3 | * Copyright IBM Corporation, 2007 | |
4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
5 | * | |
78fb7466 PE |
6 | * Copyright 2007 OpenVZ SWsoft Inc |
7 | * Author: Pavel Emelianov <xemul@openvz.org> | |
8 | * | |
2e72b634 KS |
9 | * Memory thresholds |
10 | * Copyright (C) 2009 Nokia Corporation | |
11 | * Author: Kirill A. Shutemov | |
12 | * | |
8cdea7c0 BS |
13 | * This program is free software; you can redistribute it and/or modify |
14 | * it under the terms of the GNU General Public License as published by | |
15 | * the Free Software Foundation; either version 2 of the License, or | |
16 | * (at your option) any later version. | |
17 | * | |
18 | * This program is distributed in the hope that it will be useful, | |
19 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
20 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
21 | * GNU General Public License for more details. | |
22 | */ | |
23 | ||
24 | #include <linux/res_counter.h> | |
25 | #include <linux/memcontrol.h> | |
26 | #include <linux/cgroup.h> | |
78fb7466 | 27 | #include <linux/mm.h> |
4ffef5fe | 28 | #include <linux/hugetlb.h> |
d13d1443 | 29 | #include <linux/pagemap.h> |
d52aa412 | 30 | #include <linux/smp.h> |
8a9f3ccd | 31 | #include <linux/page-flags.h> |
66e1707b | 32 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
33 | #include <linux/bit_spinlock.h> |
34 | #include <linux/rcupdate.h> | |
e222432b | 35 | #include <linux/limits.h> |
b9e15baf | 36 | #include <linux/export.h> |
8c7c6e34 | 37 | #include <linux/mutex.h> |
f64c3f54 | 38 | #include <linux/rbtree.h> |
b6ac57d5 | 39 | #include <linux/slab.h> |
66e1707b | 40 | #include <linux/swap.h> |
02491447 | 41 | #include <linux/swapops.h> |
66e1707b | 42 | #include <linux/spinlock.h> |
2e72b634 KS |
43 | #include <linux/eventfd.h> |
44 | #include <linux/sort.h> | |
66e1707b | 45 | #include <linux/fs.h> |
d2ceb9b7 | 46 | #include <linux/seq_file.h> |
33327948 | 47 | #include <linux/vmalloc.h> |
b69408e8 | 48 | #include <linux/mm_inline.h> |
52d4b9ac | 49 | #include <linux/page_cgroup.h> |
cdec2e42 | 50 | #include <linux/cpu.h> |
158e0a2d | 51 | #include <linux/oom.h> |
08e552c6 | 52 | #include "internal.h" |
d1a4c0b3 GC |
53 | #include <net/sock.h> |
54 | #include <net/tcp_memcontrol.h> | |
8cdea7c0 | 55 | |
8697d331 BS |
56 | #include <asm/uaccess.h> |
57 | ||
cc8e970c KM |
58 | #include <trace/events/vmscan.h> |
59 | ||
a181b0e8 | 60 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
a181b0e8 | 61 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
6bbda35c | 62 | static struct mem_cgroup *root_mem_cgroup __read_mostly; |
8cdea7c0 | 63 | |
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 | 75 | #else |
a0db00fc | 76 | #define do_swap_account 0 |
c077719b KH |
77 | #endif |
78 | ||
79 | ||
d52aa412 KH |
80 | /* |
81 | * Statistics for memory cgroup. | |
82 | */ | |
83 | enum mem_cgroup_stat_index { | |
84 | /* | |
85 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
86 | */ | |
87 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
d69b042f | 88 | MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */ |
d8046582 | 89 | MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */ |
0c3e73e8 | 90 | MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */ |
711d3d2c | 91 | MEM_CGROUP_STAT_DATA, /* end of data requires synchronization */ |
d52aa412 KH |
92 | MEM_CGROUP_STAT_NSTATS, |
93 | }; | |
94 | ||
e9f8974f JW |
95 | enum mem_cgroup_events_index { |
96 | MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */ | |
97 | MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */ | |
98 | MEM_CGROUP_EVENTS_COUNT, /* # of pages paged in/out */ | |
456f998e YH |
99 | MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */ |
100 | MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */ | |
e9f8974f JW |
101 | MEM_CGROUP_EVENTS_NSTATS, |
102 | }; | |
7a159cc9 JW |
103 | /* |
104 | * Per memcg event counter is incremented at every pagein/pageout. With THP, | |
105 | * it will be incremated by the number of pages. This counter is used for | |
106 | * for trigger some periodic events. This is straightforward and better | |
107 | * than using jiffies etc. to handle periodic memcg event. | |
108 | */ | |
109 | enum mem_cgroup_events_target { | |
110 | MEM_CGROUP_TARGET_THRESH, | |
111 | MEM_CGROUP_TARGET_SOFTLIMIT, | |
453a9bf3 | 112 | MEM_CGROUP_TARGET_NUMAINFO, |
7a159cc9 JW |
113 | MEM_CGROUP_NTARGETS, |
114 | }; | |
a0db00fc KS |
115 | #define THRESHOLDS_EVENTS_TARGET 128 |
116 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
117 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 118 | |
d52aa412 | 119 | struct mem_cgroup_stat_cpu { |
7a159cc9 | 120 | long count[MEM_CGROUP_STAT_NSTATS]; |
e9f8974f | 121 | unsigned long events[MEM_CGROUP_EVENTS_NSTATS]; |
7a159cc9 | 122 | unsigned long targets[MEM_CGROUP_NTARGETS]; |
d52aa412 KH |
123 | }; |
124 | ||
527a5ec9 JW |
125 | struct mem_cgroup_reclaim_iter { |
126 | /* css_id of the last scanned hierarchy member */ | |
127 | int position; | |
128 | /* scan generation, increased every round-trip */ | |
129 | unsigned int generation; | |
130 | }; | |
131 | ||
6d12e2d8 KH |
132 | /* |
133 | * per-zone information in memory controller. | |
134 | */ | |
6d12e2d8 | 135 | struct mem_cgroup_per_zone { |
6290df54 | 136 | struct lruvec lruvec; |
1eb49272 | 137 | unsigned long lru_size[NR_LRU_LISTS]; |
3e2f41f1 | 138 | |
527a5ec9 JW |
139 | struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1]; |
140 | ||
f64c3f54 BS |
141 | struct rb_node tree_node; /* RB tree node */ |
142 | unsigned long long usage_in_excess;/* Set to the value by which */ | |
143 | /* the soft limit is exceeded*/ | |
144 | bool on_tree; | |
d79154bb | 145 | struct mem_cgroup *memcg; /* Back pointer, we cannot */ |
4e416953 | 146 | /* use container_of */ |
6d12e2d8 | 147 | }; |
6d12e2d8 KH |
148 | |
149 | struct mem_cgroup_per_node { | |
150 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
151 | }; | |
152 | ||
153 | struct mem_cgroup_lru_info { | |
154 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
155 | }; | |
156 | ||
f64c3f54 BS |
157 | /* |
158 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
159 | * their hierarchy representation | |
160 | */ | |
161 | ||
162 | struct mem_cgroup_tree_per_zone { | |
163 | struct rb_root rb_root; | |
164 | spinlock_t lock; | |
165 | }; | |
166 | ||
167 | struct mem_cgroup_tree_per_node { | |
168 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
169 | }; | |
170 | ||
171 | struct mem_cgroup_tree { | |
172 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
173 | }; | |
174 | ||
175 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
176 | ||
2e72b634 KS |
177 | struct mem_cgroup_threshold { |
178 | struct eventfd_ctx *eventfd; | |
179 | u64 threshold; | |
180 | }; | |
181 | ||
9490ff27 | 182 | /* For threshold */ |
2e72b634 | 183 | struct mem_cgroup_threshold_ary { |
748dad36 | 184 | /* An array index points to threshold just below or equal to usage. */ |
5407a562 | 185 | int current_threshold; |
2e72b634 KS |
186 | /* Size of entries[] */ |
187 | unsigned int size; | |
188 | /* Array of thresholds */ | |
189 | struct mem_cgroup_threshold entries[0]; | |
190 | }; | |
2c488db2 KS |
191 | |
192 | struct mem_cgroup_thresholds { | |
193 | /* Primary thresholds array */ | |
194 | struct mem_cgroup_threshold_ary *primary; | |
195 | /* | |
196 | * Spare threshold array. | |
197 | * This is needed to make mem_cgroup_unregister_event() "never fail". | |
198 | * It must be able to store at least primary->size - 1 entries. | |
199 | */ | |
200 | struct mem_cgroup_threshold_ary *spare; | |
201 | }; | |
202 | ||
9490ff27 KH |
203 | /* for OOM */ |
204 | struct mem_cgroup_eventfd_list { | |
205 | struct list_head list; | |
206 | struct eventfd_ctx *eventfd; | |
207 | }; | |
2e72b634 | 208 | |
c0ff4b85 R |
209 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
210 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 211 | |
8cdea7c0 BS |
212 | /* |
213 | * The memory controller data structure. The memory controller controls both | |
214 | * page cache and RSS per cgroup. We would eventually like to provide | |
215 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
216 | * to help the administrator determine what knobs to tune. | |
217 | * | |
218 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
219 | * we hit the water mark. May be even add a low water mark, such that |
220 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
221 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
222 | */ |
223 | struct mem_cgroup { | |
224 | struct cgroup_subsys_state css; | |
225 | /* | |
226 | * the counter to account for memory usage | |
227 | */ | |
228 | struct res_counter res; | |
59927fb9 HD |
229 | |
230 | union { | |
231 | /* | |
232 | * the counter to account for mem+swap usage. | |
233 | */ | |
234 | struct res_counter memsw; | |
235 | ||
236 | /* | |
237 | * rcu_freeing is used only when freeing struct mem_cgroup, | |
238 | * so put it into a union to avoid wasting more memory. | |
239 | * It must be disjoint from the css field. It could be | |
240 | * in a union with the res field, but res plays a much | |
241 | * larger part in mem_cgroup life than memsw, and might | |
242 | * be of interest, even at time of free, when debugging. | |
243 | * So share rcu_head with the less interesting memsw. | |
244 | */ | |
245 | struct rcu_head rcu_freeing; | |
246 | /* | |
247 | * But when using vfree(), that cannot be done at | |
248 | * interrupt time, so we must then queue the work. | |
249 | */ | |
250 | struct work_struct work_freeing; | |
251 | }; | |
252 | ||
78fb7466 PE |
253 | /* |
254 | * Per cgroup active and inactive list, similar to the | |
255 | * per zone LRU lists. | |
78fb7466 | 256 | */ |
6d12e2d8 | 257 | struct mem_cgroup_lru_info info; |
889976db YH |
258 | int last_scanned_node; |
259 | #if MAX_NUMNODES > 1 | |
260 | nodemask_t scan_nodes; | |
453a9bf3 KH |
261 | atomic_t numainfo_events; |
262 | atomic_t numainfo_updating; | |
889976db | 263 | #endif |
18f59ea7 BS |
264 | /* |
265 | * Should the accounting and control be hierarchical, per subtree? | |
266 | */ | |
267 | bool use_hierarchy; | |
79dfdacc MH |
268 | |
269 | bool oom_lock; | |
270 | atomic_t under_oom; | |
271 | ||
8c7c6e34 | 272 | atomic_t refcnt; |
14797e23 | 273 | |
1f4c025b | 274 | int swappiness; |
3c11ecf4 KH |
275 | /* OOM-Killer disable */ |
276 | int oom_kill_disable; | |
a7885eb8 | 277 | |
22a668d7 KH |
278 | /* set when res.limit == memsw.limit */ |
279 | bool memsw_is_minimum; | |
280 | ||
2e72b634 KS |
281 | /* protect arrays of thresholds */ |
282 | struct mutex thresholds_lock; | |
283 | ||
284 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 285 | struct mem_cgroup_thresholds thresholds; |
907860ed | 286 | |
2e72b634 | 287 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 288 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 289 | |
9490ff27 KH |
290 | /* For oom notifier event fd */ |
291 | struct list_head oom_notify; | |
185efc0f | 292 | |
7dc74be0 DN |
293 | /* |
294 | * Should we move charges of a task when a task is moved into this | |
295 | * mem_cgroup ? And what type of charges should we move ? | |
296 | */ | |
297 | unsigned long move_charge_at_immigrate; | |
619d094b KH |
298 | /* |
299 | * set > 0 if pages under this cgroup are moving to other cgroup. | |
300 | */ | |
301 | atomic_t moving_account; | |
312734c0 KH |
302 | /* taken only while moving_account > 0 */ |
303 | spinlock_t move_lock; | |
d52aa412 | 304 | /* |
c62b1a3b | 305 | * percpu counter. |
d52aa412 | 306 | */ |
3a7951b4 | 307 | struct mem_cgroup_stat_cpu __percpu *stat; |
711d3d2c KH |
308 | /* |
309 | * used when a cpu is offlined or other synchronizations | |
310 | * See mem_cgroup_read_stat(). | |
311 | */ | |
312 | struct mem_cgroup_stat_cpu nocpu_base; | |
313 | spinlock_t pcp_counter_lock; | |
d1a4c0b3 GC |
314 | |
315 | #ifdef CONFIG_INET | |
316 | struct tcp_memcontrol tcp_mem; | |
317 | #endif | |
8cdea7c0 BS |
318 | }; |
319 | ||
7dc74be0 DN |
320 | /* Stuffs for move charges at task migration. */ |
321 | /* | |
322 | * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a | |
323 | * left-shifted bitmap of these types. | |
324 | */ | |
325 | enum move_type { | |
4ffef5fe | 326 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 327 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
328 | NR_MOVE_TYPE, |
329 | }; | |
330 | ||
4ffef5fe DN |
331 | /* "mc" and its members are protected by cgroup_mutex */ |
332 | static struct move_charge_struct { | |
b1dd693e | 333 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
334 | struct mem_cgroup *from; |
335 | struct mem_cgroup *to; | |
336 | unsigned long precharge; | |
854ffa8d | 337 | unsigned long moved_charge; |
483c30b5 | 338 | unsigned long moved_swap; |
8033b97c DN |
339 | struct task_struct *moving_task; /* a task moving charges */ |
340 | wait_queue_head_t waitq; /* a waitq for other context */ | |
341 | } mc = { | |
2bd9bb20 | 342 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
343 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
344 | }; | |
4ffef5fe | 345 | |
90254a65 DN |
346 | static bool move_anon(void) |
347 | { | |
348 | return test_bit(MOVE_CHARGE_TYPE_ANON, | |
349 | &mc.to->move_charge_at_immigrate); | |
350 | } | |
351 | ||
87946a72 DN |
352 | static bool move_file(void) |
353 | { | |
354 | return test_bit(MOVE_CHARGE_TYPE_FILE, | |
355 | &mc.to->move_charge_at_immigrate); | |
356 | } | |
357 | ||
4e416953 BS |
358 | /* |
359 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
360 | * limit reclaim to prevent infinite loops, if they ever occur. | |
361 | */ | |
a0db00fc KS |
362 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
363 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 | |
4e416953 | 364 | |
217bc319 KH |
365 | enum charge_type { |
366 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
367 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
4f98a2fe | 368 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
c05555b5 | 369 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
d13d1443 | 370 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 371 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
372 | NR_CHARGE_TYPE, |
373 | }; | |
374 | ||
8c7c6e34 | 375 | /* for encoding cft->private value on file */ |
65c64ce8 GC |
376 | #define _MEM (0) |
377 | #define _MEMSWAP (1) | |
378 | #define _OOM_TYPE (2) | |
a0db00fc KS |
379 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
380 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 381 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
382 | /* Used for OOM nofiier */ |
383 | #define OOM_CONTROL (0) | |
8c7c6e34 | 384 | |
75822b44 BS |
385 | /* |
386 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
387 | */ | |
388 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
389 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
390 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
391 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
392 | ||
c0ff4b85 R |
393 | static void mem_cgroup_get(struct mem_cgroup *memcg); |
394 | static void mem_cgroup_put(struct mem_cgroup *memcg); | |
e1aab161 GC |
395 | |
396 | /* Writing them here to avoid exposing memcg's inner layout */ | |
397 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM | |
e1aab161 | 398 | #include <net/sock.h> |
d1a4c0b3 | 399 | #include <net/ip.h> |
e1aab161 GC |
400 | |
401 | static bool mem_cgroup_is_root(struct mem_cgroup *memcg); | |
402 | void sock_update_memcg(struct sock *sk) | |
403 | { | |
376be5ff | 404 | if (mem_cgroup_sockets_enabled) { |
e1aab161 GC |
405 | struct mem_cgroup *memcg; |
406 | ||
407 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
408 | ||
f3f511e1 GC |
409 | /* Socket cloning can throw us here with sk_cgrp already |
410 | * filled. It won't however, necessarily happen from | |
411 | * process context. So the test for root memcg given | |
412 | * the current task's memcg won't help us in this case. | |
413 | * | |
414 | * Respecting the original socket's memcg is a better | |
415 | * decision in this case. | |
416 | */ | |
417 | if (sk->sk_cgrp) { | |
418 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
419 | mem_cgroup_get(sk->sk_cgrp->memcg); | |
420 | return; | |
421 | } | |
422 | ||
e1aab161 GC |
423 | rcu_read_lock(); |
424 | memcg = mem_cgroup_from_task(current); | |
425 | if (!mem_cgroup_is_root(memcg)) { | |
426 | mem_cgroup_get(memcg); | |
427 | sk->sk_cgrp = sk->sk_prot->proto_cgroup(memcg); | |
428 | } | |
429 | rcu_read_unlock(); | |
430 | } | |
431 | } | |
432 | EXPORT_SYMBOL(sock_update_memcg); | |
433 | ||
434 | void sock_release_memcg(struct sock *sk) | |
435 | { | |
376be5ff | 436 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
437 | struct mem_cgroup *memcg; |
438 | WARN_ON(!sk->sk_cgrp->memcg); | |
439 | memcg = sk->sk_cgrp->memcg; | |
440 | mem_cgroup_put(memcg); | |
441 | } | |
442 | } | |
d1a4c0b3 | 443 | |
319d3b9c | 444 | #ifdef CONFIG_INET |
d1a4c0b3 GC |
445 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) |
446 | { | |
447 | if (!memcg || mem_cgroup_is_root(memcg)) | |
448 | return NULL; | |
449 | ||
450 | return &memcg->tcp_mem.cg_proto; | |
451 | } | |
452 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 GC |
453 | #endif /* CONFIG_INET */ |
454 | #endif /* CONFIG_CGROUP_MEM_RES_CTLR_KMEM */ | |
455 | ||
c0ff4b85 | 456 | static void drain_all_stock_async(struct mem_cgroup *memcg); |
8c7c6e34 | 457 | |
f64c3f54 | 458 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 459 | mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid) |
f64c3f54 | 460 | { |
c0ff4b85 | 461 | return &memcg->info.nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
462 | } |
463 | ||
c0ff4b85 | 464 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 465 | { |
c0ff4b85 | 466 | return &memcg->css; |
d324236b WF |
467 | } |
468 | ||
f64c3f54 | 469 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 470 | page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 471 | { |
97a6c37b JW |
472 | int nid = page_to_nid(page); |
473 | int zid = page_zonenum(page); | |
f64c3f54 | 474 | |
c0ff4b85 | 475 | return mem_cgroup_zoneinfo(memcg, nid, zid); |
f64c3f54 BS |
476 | } |
477 | ||
478 | static struct mem_cgroup_tree_per_zone * | |
479 | soft_limit_tree_node_zone(int nid, int zid) | |
480 | { | |
481 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
482 | } | |
483 | ||
484 | static struct mem_cgroup_tree_per_zone * | |
485 | soft_limit_tree_from_page(struct page *page) | |
486 | { | |
487 | int nid = page_to_nid(page); | |
488 | int zid = page_zonenum(page); | |
489 | ||
490 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
491 | } | |
492 | ||
493 | static void | |
c0ff4b85 | 494 | __mem_cgroup_insert_exceeded(struct mem_cgroup *memcg, |
f64c3f54 | 495 | struct mem_cgroup_per_zone *mz, |
ef8745c1 KH |
496 | struct mem_cgroup_tree_per_zone *mctz, |
497 | unsigned long long new_usage_in_excess) | |
f64c3f54 BS |
498 | { |
499 | struct rb_node **p = &mctz->rb_root.rb_node; | |
500 | struct rb_node *parent = NULL; | |
501 | struct mem_cgroup_per_zone *mz_node; | |
502 | ||
503 | if (mz->on_tree) | |
504 | return; | |
505 | ||
ef8745c1 KH |
506 | mz->usage_in_excess = new_usage_in_excess; |
507 | if (!mz->usage_in_excess) | |
508 | return; | |
f64c3f54 BS |
509 | while (*p) { |
510 | parent = *p; | |
511 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
512 | tree_node); | |
513 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
514 | p = &(*p)->rb_left; | |
515 | /* | |
516 | * We can't avoid mem cgroups that are over their soft | |
517 | * limit by the same amount | |
518 | */ | |
519 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
520 | p = &(*p)->rb_right; | |
521 | } | |
522 | rb_link_node(&mz->tree_node, parent, p); | |
523 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
524 | mz->on_tree = true; | |
4e416953 BS |
525 | } |
526 | ||
527 | static void | |
c0ff4b85 | 528 | __mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
4e416953 BS |
529 | struct mem_cgroup_per_zone *mz, |
530 | struct mem_cgroup_tree_per_zone *mctz) | |
531 | { | |
532 | if (!mz->on_tree) | |
533 | return; | |
534 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
535 | mz->on_tree = false; | |
536 | } | |
537 | ||
f64c3f54 | 538 | static void |
c0ff4b85 | 539 | mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
f64c3f54 BS |
540 | struct mem_cgroup_per_zone *mz, |
541 | struct mem_cgroup_tree_per_zone *mctz) | |
542 | { | |
543 | spin_lock(&mctz->lock); | |
c0ff4b85 | 544 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
545 | spin_unlock(&mctz->lock); |
546 | } | |
547 | ||
f64c3f54 | 548 | |
c0ff4b85 | 549 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 550 | { |
ef8745c1 | 551 | unsigned long long excess; |
f64c3f54 BS |
552 | struct mem_cgroup_per_zone *mz; |
553 | struct mem_cgroup_tree_per_zone *mctz; | |
4e649152 KH |
554 | int nid = page_to_nid(page); |
555 | int zid = page_zonenum(page); | |
f64c3f54 BS |
556 | mctz = soft_limit_tree_from_page(page); |
557 | ||
558 | /* | |
4e649152 KH |
559 | * Necessary to update all ancestors when hierarchy is used. |
560 | * because their event counter is not touched. | |
f64c3f54 | 561 | */ |
c0ff4b85 R |
562 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
563 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
564 | excess = res_counter_soft_limit_excess(&memcg->res); | |
4e649152 KH |
565 | /* |
566 | * We have to update the tree if mz is on RB-tree or | |
567 | * mem is over its softlimit. | |
568 | */ | |
ef8745c1 | 569 | if (excess || mz->on_tree) { |
4e649152 KH |
570 | spin_lock(&mctz->lock); |
571 | /* if on-tree, remove it */ | |
572 | if (mz->on_tree) | |
c0ff4b85 | 573 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
4e649152 | 574 | /* |
ef8745c1 KH |
575 | * Insert again. mz->usage_in_excess will be updated. |
576 | * If excess is 0, no tree ops. | |
4e649152 | 577 | */ |
c0ff4b85 | 578 | __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess); |
4e649152 KH |
579 | spin_unlock(&mctz->lock); |
580 | } | |
f64c3f54 BS |
581 | } |
582 | } | |
583 | ||
c0ff4b85 | 584 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) |
f64c3f54 BS |
585 | { |
586 | int node, zone; | |
587 | struct mem_cgroup_per_zone *mz; | |
588 | struct mem_cgroup_tree_per_zone *mctz; | |
589 | ||
3ed28fa1 | 590 | for_each_node(node) { |
f64c3f54 | 591 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
c0ff4b85 | 592 | mz = mem_cgroup_zoneinfo(memcg, node, zone); |
f64c3f54 | 593 | mctz = soft_limit_tree_node_zone(node, zone); |
c0ff4b85 | 594 | mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
595 | } |
596 | } | |
597 | } | |
598 | ||
4e416953 BS |
599 | static struct mem_cgroup_per_zone * |
600 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
601 | { | |
602 | struct rb_node *rightmost = NULL; | |
26251eaf | 603 | struct mem_cgroup_per_zone *mz; |
4e416953 BS |
604 | |
605 | retry: | |
26251eaf | 606 | mz = NULL; |
4e416953 BS |
607 | rightmost = rb_last(&mctz->rb_root); |
608 | if (!rightmost) | |
609 | goto done; /* Nothing to reclaim from */ | |
610 | ||
611 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
612 | /* | |
613 | * Remove the node now but someone else can add it back, | |
614 | * we will to add it back at the end of reclaim to its correct | |
615 | * position in the tree. | |
616 | */ | |
d79154bb HD |
617 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); |
618 | if (!res_counter_soft_limit_excess(&mz->memcg->res) || | |
619 | !css_tryget(&mz->memcg->css)) | |
4e416953 BS |
620 | goto retry; |
621 | done: | |
622 | return mz; | |
623 | } | |
624 | ||
625 | static struct mem_cgroup_per_zone * | |
626 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
627 | { | |
628 | struct mem_cgroup_per_zone *mz; | |
629 | ||
630 | spin_lock(&mctz->lock); | |
631 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
632 | spin_unlock(&mctz->lock); | |
633 | return mz; | |
634 | } | |
635 | ||
711d3d2c KH |
636 | /* |
637 | * Implementation Note: reading percpu statistics for memcg. | |
638 | * | |
639 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
640 | * synchronization to implement "quick" read. There are trade-off between | |
641 | * reading cost and precision of value. Then, we may have a chance to implement | |
642 | * a periodic synchronizion of counter in memcg's counter. | |
643 | * | |
644 | * But this _read() function is used for user interface now. The user accounts | |
645 | * memory usage by memory cgroup and he _always_ requires exact value because | |
646 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
647 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
648 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
649 | * | |
650 | * If there are kernel internal actions which can make use of some not-exact | |
651 | * value, and reading all cpu value can be performance bottleneck in some | |
652 | * common workload, threashold and synchonization as vmstat[] should be | |
653 | * implemented. | |
654 | */ | |
c0ff4b85 | 655 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 656 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 657 | { |
7a159cc9 | 658 | long val = 0; |
c62b1a3b | 659 | int cpu; |
c62b1a3b | 660 | |
711d3d2c KH |
661 | get_online_cpus(); |
662 | for_each_online_cpu(cpu) | |
c0ff4b85 | 663 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 664 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
665 | spin_lock(&memcg->pcp_counter_lock); |
666 | val += memcg->nocpu_base.count[idx]; | |
667 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
668 | #endif |
669 | put_online_cpus(); | |
c62b1a3b KH |
670 | return val; |
671 | } | |
672 | ||
c0ff4b85 | 673 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
0c3e73e8 BS |
674 | bool charge) |
675 | { | |
676 | int val = (charge) ? 1 : -1; | |
c0ff4b85 | 677 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAPOUT], val); |
0c3e73e8 BS |
678 | } |
679 | ||
c0ff4b85 | 680 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
681 | enum mem_cgroup_events_index idx) |
682 | { | |
683 | unsigned long val = 0; | |
684 | int cpu; | |
685 | ||
686 | for_each_online_cpu(cpu) | |
c0ff4b85 | 687 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 688 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
689 | spin_lock(&memcg->pcp_counter_lock); |
690 | val += memcg->nocpu_base.events[idx]; | |
691 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f JW |
692 | #endif |
693 | return val; | |
694 | } | |
695 | ||
c0ff4b85 | 696 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b2402857 | 697 | bool anon, int nr_pages) |
d52aa412 | 698 | { |
c62b1a3b KH |
699 | preempt_disable(); |
700 | ||
b2402857 KH |
701 | /* |
702 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
703 | * counted as CACHE even if it's on ANON LRU. | |
704 | */ | |
705 | if (anon) | |
706 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], | |
c0ff4b85 | 707 | nr_pages); |
d52aa412 | 708 | else |
b2402857 | 709 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 710 | nr_pages); |
55e462b0 | 711 | |
e401f176 KH |
712 | /* pagein of a big page is an event. So, ignore page size */ |
713 | if (nr_pages > 0) | |
c0ff4b85 | 714 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 715 | else { |
c0ff4b85 | 716 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
717 | nr_pages = -nr_pages; /* for event */ |
718 | } | |
e401f176 | 719 | |
c0ff4b85 | 720 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_COUNT], nr_pages); |
2e72b634 | 721 | |
c62b1a3b | 722 | preempt_enable(); |
6d12e2d8 KH |
723 | } |
724 | ||
bb2a0de9 | 725 | unsigned long |
c0ff4b85 | 726 | mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid, |
bb2a0de9 | 727 | unsigned int lru_mask) |
889976db YH |
728 | { |
729 | struct mem_cgroup_per_zone *mz; | |
f156ab93 | 730 | enum lru_list lru; |
bb2a0de9 KH |
731 | unsigned long ret = 0; |
732 | ||
c0ff4b85 | 733 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
bb2a0de9 | 734 | |
f156ab93 HD |
735 | for_each_lru(lru) { |
736 | if (BIT(lru) & lru_mask) | |
737 | ret += mz->lru_size[lru]; | |
bb2a0de9 KH |
738 | } |
739 | return ret; | |
740 | } | |
741 | ||
742 | static unsigned long | |
c0ff4b85 | 743 | mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 KH |
744 | int nid, unsigned int lru_mask) |
745 | { | |
889976db YH |
746 | u64 total = 0; |
747 | int zid; | |
748 | ||
bb2a0de9 | 749 | for (zid = 0; zid < MAX_NR_ZONES; zid++) |
c0ff4b85 R |
750 | total += mem_cgroup_zone_nr_lru_pages(memcg, |
751 | nid, zid, lru_mask); | |
bb2a0de9 | 752 | |
889976db YH |
753 | return total; |
754 | } | |
bb2a0de9 | 755 | |
c0ff4b85 | 756 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 757 | unsigned int lru_mask) |
6d12e2d8 | 758 | { |
889976db | 759 | int nid; |
6d12e2d8 KH |
760 | u64 total = 0; |
761 | ||
bb2a0de9 | 762 | for_each_node_state(nid, N_HIGH_MEMORY) |
c0ff4b85 | 763 | total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
6d12e2d8 | 764 | return total; |
d52aa412 KH |
765 | } |
766 | ||
f53d7ce3 JW |
767 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
768 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
769 | { |
770 | unsigned long val, next; | |
771 | ||
4799401f SR |
772 | val = __this_cpu_read(memcg->stat->events[MEM_CGROUP_EVENTS_COUNT]); |
773 | next = __this_cpu_read(memcg->stat->targets[target]); | |
7a159cc9 | 774 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
775 | if ((long)next - (long)val < 0) { |
776 | switch (target) { | |
777 | case MEM_CGROUP_TARGET_THRESH: | |
778 | next = val + THRESHOLDS_EVENTS_TARGET; | |
779 | break; | |
780 | case MEM_CGROUP_TARGET_SOFTLIMIT: | |
781 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
782 | break; | |
783 | case MEM_CGROUP_TARGET_NUMAINFO: | |
784 | next = val + NUMAINFO_EVENTS_TARGET; | |
785 | break; | |
786 | default: | |
787 | break; | |
788 | } | |
789 | __this_cpu_write(memcg->stat->targets[target], next); | |
790 | return true; | |
7a159cc9 | 791 | } |
f53d7ce3 | 792 | return false; |
d2265e6f KH |
793 | } |
794 | ||
795 | /* | |
796 | * Check events in order. | |
797 | * | |
798 | */ | |
c0ff4b85 | 799 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f | 800 | { |
4799401f | 801 | preempt_disable(); |
d2265e6f | 802 | /* threshold event is triggered in finer grain than soft limit */ |
f53d7ce3 JW |
803 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
804 | MEM_CGROUP_TARGET_THRESH))) { | |
82b3f2a7 AM |
805 | bool do_softlimit; |
806 | bool do_numainfo __maybe_unused; | |
f53d7ce3 JW |
807 | |
808 | do_softlimit = mem_cgroup_event_ratelimit(memcg, | |
809 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
810 | #if MAX_NUMNODES > 1 | |
811 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
812 | MEM_CGROUP_TARGET_NUMAINFO); | |
813 | #endif | |
814 | preempt_enable(); | |
815 | ||
c0ff4b85 | 816 | mem_cgroup_threshold(memcg); |
f53d7ce3 | 817 | if (unlikely(do_softlimit)) |
c0ff4b85 | 818 | mem_cgroup_update_tree(memcg, page); |
453a9bf3 | 819 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 820 | if (unlikely(do_numainfo)) |
c0ff4b85 | 821 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 822 | #endif |
f53d7ce3 JW |
823 | } else |
824 | preempt_enable(); | |
d2265e6f KH |
825 | } |
826 | ||
d1a4c0b3 | 827 | struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
828 | { |
829 | return container_of(cgroup_subsys_state(cont, | |
830 | mem_cgroup_subsys_id), struct mem_cgroup, | |
831 | css); | |
832 | } | |
833 | ||
cf475ad2 | 834 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 835 | { |
31a78f23 BS |
836 | /* |
837 | * mm_update_next_owner() may clear mm->owner to NULL | |
838 | * if it races with swapoff, page migration, etc. | |
839 | * So this can be called with p == NULL. | |
840 | */ | |
841 | if (unlikely(!p)) | |
842 | return NULL; | |
843 | ||
78fb7466 PE |
844 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
845 | struct mem_cgroup, css); | |
846 | } | |
847 | ||
a433658c | 848 | struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 849 | { |
c0ff4b85 | 850 | struct mem_cgroup *memcg = NULL; |
0b7f569e KH |
851 | |
852 | if (!mm) | |
853 | return NULL; | |
54595fe2 KH |
854 | /* |
855 | * Because we have no locks, mm->owner's may be being moved to other | |
856 | * cgroup. We use css_tryget() here even if this looks | |
857 | * pessimistic (rather than adding locks here). | |
858 | */ | |
859 | rcu_read_lock(); | |
860 | do { | |
c0ff4b85 R |
861 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
862 | if (unlikely(!memcg)) | |
54595fe2 | 863 | break; |
c0ff4b85 | 864 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 865 | rcu_read_unlock(); |
c0ff4b85 | 866 | return memcg; |
54595fe2 KH |
867 | } |
868 | ||
5660048c JW |
869 | /** |
870 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
871 | * @root: hierarchy root | |
872 | * @prev: previously returned memcg, NULL on first invocation | |
873 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
874 | * | |
875 | * Returns references to children of the hierarchy below @root, or | |
876 | * @root itself, or %NULL after a full round-trip. | |
877 | * | |
878 | * Caller must pass the return value in @prev on subsequent | |
879 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
880 | * to cancel a hierarchy walk before the round-trip is complete. | |
881 | * | |
882 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
883 | * divide up the memcgs in the hierarchy among all concurrent | |
884 | * reclaimers operating on the same zone and priority. | |
885 | */ | |
886 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, | |
887 | struct mem_cgroup *prev, | |
888 | struct mem_cgroup_reclaim_cookie *reclaim) | |
14067bb3 | 889 | { |
9f3a0d09 JW |
890 | struct mem_cgroup *memcg = NULL; |
891 | int id = 0; | |
711d3d2c | 892 | |
5660048c JW |
893 | if (mem_cgroup_disabled()) |
894 | return NULL; | |
895 | ||
9f3a0d09 JW |
896 | if (!root) |
897 | root = root_mem_cgroup; | |
7d74b06f | 898 | |
9f3a0d09 JW |
899 | if (prev && !reclaim) |
900 | id = css_id(&prev->css); | |
14067bb3 | 901 | |
9f3a0d09 JW |
902 | if (prev && prev != root) |
903 | css_put(&prev->css); | |
14067bb3 | 904 | |
9f3a0d09 JW |
905 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
906 | if (prev) | |
907 | return NULL; | |
908 | return root; | |
909 | } | |
14067bb3 | 910 | |
9f3a0d09 | 911 | while (!memcg) { |
527a5ec9 | 912 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
9f3a0d09 | 913 | struct cgroup_subsys_state *css; |
711d3d2c | 914 | |
527a5ec9 JW |
915 | if (reclaim) { |
916 | int nid = zone_to_nid(reclaim->zone); | |
917 | int zid = zone_idx(reclaim->zone); | |
918 | struct mem_cgroup_per_zone *mz; | |
919 | ||
920 | mz = mem_cgroup_zoneinfo(root, nid, zid); | |
921 | iter = &mz->reclaim_iter[reclaim->priority]; | |
922 | if (prev && reclaim->generation != iter->generation) | |
923 | return NULL; | |
924 | id = iter->position; | |
925 | } | |
7d74b06f | 926 | |
9f3a0d09 JW |
927 | rcu_read_lock(); |
928 | css = css_get_next(&mem_cgroup_subsys, id + 1, &root->css, &id); | |
929 | if (css) { | |
930 | if (css == &root->css || css_tryget(css)) | |
931 | memcg = container_of(css, | |
932 | struct mem_cgroup, css); | |
933 | } else | |
934 | id = 0; | |
14067bb3 | 935 | rcu_read_unlock(); |
14067bb3 | 936 | |
527a5ec9 JW |
937 | if (reclaim) { |
938 | iter->position = id; | |
939 | if (!css) | |
940 | iter->generation++; | |
941 | else if (!prev && memcg) | |
942 | reclaim->generation = iter->generation; | |
943 | } | |
9f3a0d09 JW |
944 | |
945 | if (prev && !css) | |
946 | return NULL; | |
947 | } | |
948 | return memcg; | |
14067bb3 | 949 | } |
7d74b06f | 950 | |
5660048c JW |
951 | /** |
952 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
953 | * @root: hierarchy root | |
954 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
955 | */ | |
956 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
957 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
958 | { |
959 | if (!root) | |
960 | root = root_mem_cgroup; | |
961 | if (prev && prev != root) | |
962 | css_put(&prev->css); | |
963 | } | |
7d74b06f | 964 | |
9f3a0d09 JW |
965 | /* |
966 | * Iteration constructs for visiting all cgroups (under a tree). If | |
967 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
968 | * be used for reference counting. | |
969 | */ | |
970 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 971 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 972 | iter != NULL; \ |
527a5ec9 | 973 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 974 | |
9f3a0d09 | 975 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 976 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 977 | iter != NULL; \ |
527a5ec9 | 978 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 979 | |
c0ff4b85 | 980 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
4b3bde4c | 981 | { |
c0ff4b85 | 982 | return (memcg == root_mem_cgroup); |
4b3bde4c BS |
983 | } |
984 | ||
456f998e YH |
985 | void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
986 | { | |
c0ff4b85 | 987 | struct mem_cgroup *memcg; |
456f998e YH |
988 | |
989 | if (!mm) | |
990 | return; | |
991 | ||
992 | rcu_read_lock(); | |
c0ff4b85 R |
993 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
994 | if (unlikely(!memcg)) | |
456f998e YH |
995 | goto out; |
996 | ||
997 | switch (idx) { | |
456f998e | 998 | case PGFAULT: |
0e574a93 JW |
999 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1000 | break; | |
1001 | case PGMAJFAULT: | |
1002 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1003 | break; |
1004 | default: | |
1005 | BUG(); | |
1006 | } | |
1007 | out: | |
1008 | rcu_read_unlock(); | |
1009 | } | |
1010 | EXPORT_SYMBOL(mem_cgroup_count_vm_event); | |
1011 | ||
925b7673 JW |
1012 | /** |
1013 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1014 | * @zone: zone of the wanted lruvec | |
1015 | * @mem: memcg of the wanted lruvec | |
1016 | * | |
1017 | * Returns the lru list vector holding pages for the given @zone and | |
1018 | * @mem. This can be the global zone lruvec, if the memory controller | |
1019 | * is disabled. | |
1020 | */ | |
1021 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1022 | struct mem_cgroup *memcg) | |
1023 | { | |
1024 | struct mem_cgroup_per_zone *mz; | |
1025 | ||
1026 | if (mem_cgroup_disabled()) | |
1027 | return &zone->lruvec; | |
1028 | ||
1029 | mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone)); | |
1030 | return &mz->lruvec; | |
1031 | } | |
1032 | ||
08e552c6 KH |
1033 | /* |
1034 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
1035 | * Operations are called by routine of global LRU independently from memcg. | |
1036 | * What we have to take care of here is validness of pc->mem_cgroup. | |
1037 | * | |
1038 | * Changes to pc->mem_cgroup happens when | |
1039 | * 1. charge | |
1040 | * 2. moving account | |
1041 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
1042 | * It is added to LRU before charge. | |
1043 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
1044 | * When moving account, the page is not on LRU. It's isolated. | |
1045 | */ | |
4f98a2fe | 1046 | |
925b7673 JW |
1047 | /** |
1048 | * mem_cgroup_lru_add_list - account for adding an lru page and return lruvec | |
1049 | * @zone: zone of the page | |
1050 | * @page: the page | |
1051 | * @lru: current lru | |
1052 | * | |
1053 | * This function accounts for @page being added to @lru, and returns | |
1054 | * the lruvec for the given @zone and the memcg @page is charged to. | |
1055 | * | |
1056 | * The callsite is then responsible for physically linking the page to | |
1057 | * the returned lruvec->lists[@lru]. | |
1058 | */ | |
1059 | struct lruvec *mem_cgroup_lru_add_list(struct zone *zone, struct page *page, | |
1060 | enum lru_list lru) | |
08e552c6 | 1061 | { |
08e552c6 | 1062 | struct mem_cgroup_per_zone *mz; |
925b7673 JW |
1063 | struct mem_cgroup *memcg; |
1064 | struct page_cgroup *pc; | |
6d12e2d8 | 1065 | |
f8d66542 | 1066 | if (mem_cgroup_disabled()) |
925b7673 JW |
1067 | return &zone->lruvec; |
1068 | ||
08e552c6 | 1069 | pc = lookup_page_cgroup(page); |
38c5d72f | 1070 | memcg = pc->mem_cgroup; |
7512102c HD |
1071 | |
1072 | /* | |
1073 | * Surreptitiously switch any uncharged page to root: | |
1074 | * an uncharged page off lru does nothing to secure | |
1075 | * its former mem_cgroup from sudden removal. | |
1076 | * | |
1077 | * Our caller holds lru_lock, and PageCgroupUsed is updated | |
1078 | * under page_cgroup lock: between them, they make all uses | |
1079 | * of pc->mem_cgroup safe. | |
1080 | */ | |
1081 | if (!PageCgroupUsed(pc) && memcg != root_mem_cgroup) | |
1082 | pc->mem_cgroup = memcg = root_mem_cgroup; | |
1083 | ||
925b7673 JW |
1084 | mz = page_cgroup_zoneinfo(memcg, page); |
1085 | /* compound_order() is stabilized through lru_lock */ | |
1eb49272 | 1086 | mz->lru_size[lru] += 1 << compound_order(page); |
925b7673 | 1087 | return &mz->lruvec; |
08e552c6 | 1088 | } |
b69408e8 | 1089 | |
925b7673 JW |
1090 | /** |
1091 | * mem_cgroup_lru_del_list - account for removing an lru page | |
1092 | * @page: the page | |
1093 | * @lru: target lru | |
1094 | * | |
1095 | * This function accounts for @page being removed from @lru. | |
1096 | * | |
1097 | * The callsite is then responsible for physically unlinking | |
1098 | * @page->lru. | |
3f58a829 | 1099 | */ |
925b7673 | 1100 | void mem_cgroup_lru_del_list(struct page *page, enum lru_list lru) |
3f58a829 MK |
1101 | { |
1102 | struct mem_cgroup_per_zone *mz; | |
925b7673 | 1103 | struct mem_cgroup *memcg; |
3f58a829 | 1104 | struct page_cgroup *pc; |
3f58a829 MK |
1105 | |
1106 | if (mem_cgroup_disabled()) | |
1107 | return; | |
1108 | ||
1109 | pc = lookup_page_cgroup(page); | |
38c5d72f KH |
1110 | memcg = pc->mem_cgroup; |
1111 | VM_BUG_ON(!memcg); | |
925b7673 JW |
1112 | mz = page_cgroup_zoneinfo(memcg, page); |
1113 | /* huge page split is done under lru_lock. so, we have no races. */ | |
1eb49272 HD |
1114 | VM_BUG_ON(mz->lru_size[lru] < (1 << compound_order(page))); |
1115 | mz->lru_size[lru] -= 1 << compound_order(page); | |
3f58a829 MK |
1116 | } |
1117 | ||
925b7673 JW |
1118 | /** |
1119 | * mem_cgroup_lru_move_lists - account for moving a page between lrus | |
1120 | * @zone: zone of the page | |
1121 | * @page: the page | |
1122 | * @from: current lru | |
1123 | * @to: target lru | |
1124 | * | |
1125 | * This function accounts for @page being moved between the lrus @from | |
1126 | * and @to, and returns the lruvec for the given @zone and the memcg | |
1127 | * @page is charged to. | |
1128 | * | |
1129 | * The callsite is then responsible for physically relinking | |
1130 | * @page->lru to the returned lruvec->lists[@to]. | |
1131 | */ | |
1132 | struct lruvec *mem_cgroup_lru_move_lists(struct zone *zone, | |
1133 | struct page *page, | |
1134 | enum lru_list from, | |
1135 | enum lru_list to) | |
66e1707b | 1136 | { |
925b7673 JW |
1137 | /* XXX: Optimize this, especially for @from == @to */ |
1138 | mem_cgroup_lru_del_list(page, from); | |
1139 | return mem_cgroup_lru_add_list(zone, page, to); | |
08e552c6 | 1140 | } |
544122e5 | 1141 | |
3e92041d | 1142 | /* |
c0ff4b85 | 1143 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1144 | * hierarchy subtree |
1145 | */ | |
c3ac9a8a JW |
1146 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1147 | struct mem_cgroup *memcg) | |
3e92041d | 1148 | { |
91c63734 JW |
1149 | if (root_memcg == memcg) |
1150 | return true; | |
1151 | if (!root_memcg->use_hierarchy) | |
1152 | return false; | |
c3ac9a8a JW |
1153 | return css_is_ancestor(&memcg->css, &root_memcg->css); |
1154 | } | |
1155 | ||
1156 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, | |
1157 | struct mem_cgroup *memcg) | |
1158 | { | |
1159 | bool ret; | |
1160 | ||
91c63734 | 1161 | rcu_read_lock(); |
c3ac9a8a | 1162 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
91c63734 JW |
1163 | rcu_read_unlock(); |
1164 | return ret; | |
3e92041d MH |
1165 | } |
1166 | ||
c0ff4b85 | 1167 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg) |
4c4a2214 DR |
1168 | { |
1169 | int ret; | |
0b7f569e | 1170 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1171 | struct task_struct *p; |
4c4a2214 | 1172 | |
158e0a2d | 1173 | p = find_lock_task_mm(task); |
de077d22 DR |
1174 | if (p) { |
1175 | curr = try_get_mem_cgroup_from_mm(p->mm); | |
1176 | task_unlock(p); | |
1177 | } else { | |
1178 | /* | |
1179 | * All threads may have already detached their mm's, but the oom | |
1180 | * killer still needs to detect if they have already been oom | |
1181 | * killed to prevent needlessly killing additional tasks. | |
1182 | */ | |
1183 | task_lock(task); | |
1184 | curr = mem_cgroup_from_task(task); | |
1185 | if (curr) | |
1186 | css_get(&curr->css); | |
1187 | task_unlock(task); | |
1188 | } | |
0b7f569e KH |
1189 | if (!curr) |
1190 | return 0; | |
d31f56db | 1191 | /* |
c0ff4b85 | 1192 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1193 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1194 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1195 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1196 | */ |
c0ff4b85 | 1197 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1198 | css_put(&curr->css); |
4c4a2214 DR |
1199 | return ret; |
1200 | } | |
1201 | ||
9b272977 | 1202 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg, struct zone *zone) |
14797e23 | 1203 | { |
9b272977 JW |
1204 | unsigned long inactive_ratio; |
1205 | int nid = zone_to_nid(zone); | |
1206 | int zid = zone_idx(zone); | |
14797e23 | 1207 | unsigned long inactive; |
9b272977 | 1208 | unsigned long active; |
c772be93 | 1209 | unsigned long gb; |
14797e23 | 1210 | |
9b272977 JW |
1211 | inactive = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid, |
1212 | BIT(LRU_INACTIVE_ANON)); | |
1213 | active = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid, | |
1214 | BIT(LRU_ACTIVE_ANON)); | |
14797e23 | 1215 | |
c772be93 KM |
1216 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1217 | if (gb) | |
1218 | inactive_ratio = int_sqrt(10 * gb); | |
1219 | else | |
1220 | inactive_ratio = 1; | |
1221 | ||
9b272977 | 1222 | return inactive * inactive_ratio < active; |
14797e23 KM |
1223 | } |
1224 | ||
9b272977 | 1225 | int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg, struct zone *zone) |
56e49d21 RR |
1226 | { |
1227 | unsigned long active; | |
1228 | unsigned long inactive; | |
9b272977 JW |
1229 | int zid = zone_idx(zone); |
1230 | int nid = zone_to_nid(zone); | |
56e49d21 | 1231 | |
9b272977 JW |
1232 | inactive = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid, |
1233 | BIT(LRU_INACTIVE_FILE)); | |
1234 | active = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid, | |
1235 | BIT(LRU_ACTIVE_FILE)); | |
56e49d21 RR |
1236 | |
1237 | return (active > inactive); | |
1238 | } | |
1239 | ||
3e2f41f1 KM |
1240 | struct zone_reclaim_stat * |
1241 | mem_cgroup_get_reclaim_stat_from_page(struct page *page) | |
1242 | { | |
1243 | struct page_cgroup *pc; | |
1244 | struct mem_cgroup_per_zone *mz; | |
1245 | ||
1246 | if (mem_cgroup_disabled()) | |
1247 | return NULL; | |
1248 | ||
1249 | pc = lookup_page_cgroup(page); | |
bd112db8 DN |
1250 | if (!PageCgroupUsed(pc)) |
1251 | return NULL; | |
713735b4 JW |
1252 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ |
1253 | smp_rmb(); | |
97a6c37b | 1254 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
89abfab1 | 1255 | return &mz->lruvec.reclaim_stat; |
3e2f41f1 KM |
1256 | } |
1257 | ||
6d61ef40 BS |
1258 | #define mem_cgroup_from_res_counter(counter, member) \ |
1259 | container_of(counter, struct mem_cgroup, member) | |
1260 | ||
19942822 | 1261 | /** |
9d11ea9f JW |
1262 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
1263 | * @mem: the memory cgroup | |
19942822 | 1264 | * |
9d11ea9f | 1265 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1266 | * pages. |
19942822 | 1267 | */ |
c0ff4b85 | 1268 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1269 | { |
9d11ea9f JW |
1270 | unsigned long long margin; |
1271 | ||
c0ff4b85 | 1272 | margin = res_counter_margin(&memcg->res); |
9d11ea9f | 1273 | if (do_swap_account) |
c0ff4b85 | 1274 | margin = min(margin, res_counter_margin(&memcg->memsw)); |
7ec99d62 | 1275 | return margin >> PAGE_SHIFT; |
19942822 JW |
1276 | } |
1277 | ||
1f4c025b | 1278 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 KM |
1279 | { |
1280 | struct cgroup *cgrp = memcg->css.cgroup; | |
a7885eb8 KM |
1281 | |
1282 | /* root ? */ | |
1283 | if (cgrp->parent == NULL) | |
1284 | return vm_swappiness; | |
1285 | ||
bf1ff263 | 1286 | return memcg->swappiness; |
a7885eb8 KM |
1287 | } |
1288 | ||
619d094b KH |
1289 | /* |
1290 | * memcg->moving_account is used for checking possibility that some thread is | |
1291 | * calling move_account(). When a thread on CPU-A starts moving pages under | |
1292 | * a memcg, other threads should check memcg->moving_account under | |
1293 | * rcu_read_lock(), like this: | |
1294 | * | |
1295 | * CPU-A CPU-B | |
1296 | * rcu_read_lock() | |
1297 | * memcg->moving_account+1 if (memcg->mocing_account) | |
1298 | * take heavy locks. | |
1299 | * synchronize_rcu() update something. | |
1300 | * rcu_read_unlock() | |
1301 | * start move here. | |
1302 | */ | |
4331f7d3 KH |
1303 | |
1304 | /* for quick checking without looking up memcg */ | |
1305 | atomic_t memcg_moving __read_mostly; | |
1306 | ||
c0ff4b85 | 1307 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a | 1308 | { |
4331f7d3 | 1309 | atomic_inc(&memcg_moving); |
619d094b | 1310 | atomic_inc(&memcg->moving_account); |
32047e2a KH |
1311 | synchronize_rcu(); |
1312 | } | |
1313 | ||
c0ff4b85 | 1314 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a | 1315 | { |
619d094b KH |
1316 | /* |
1317 | * Now, mem_cgroup_clear_mc() may call this function with NULL. | |
1318 | * We check NULL in callee rather than caller. | |
1319 | */ | |
4331f7d3 KH |
1320 | if (memcg) { |
1321 | atomic_dec(&memcg_moving); | |
619d094b | 1322 | atomic_dec(&memcg->moving_account); |
4331f7d3 | 1323 | } |
32047e2a | 1324 | } |
619d094b | 1325 | |
32047e2a KH |
1326 | /* |
1327 | * 2 routines for checking "mem" is under move_account() or not. | |
1328 | * | |
13fd1dd9 AM |
1329 | * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This |
1330 | * is used for avoiding races in accounting. If true, | |
32047e2a KH |
1331 | * pc->mem_cgroup may be overwritten. |
1332 | * | |
1333 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or | |
1334 | * under hierarchy of moving cgroups. This is for | |
1335 | * waiting at hith-memory prressure caused by "move". | |
1336 | */ | |
1337 | ||
13fd1dd9 | 1338 | static bool mem_cgroup_stolen(struct mem_cgroup *memcg) |
32047e2a KH |
1339 | { |
1340 | VM_BUG_ON(!rcu_read_lock_held()); | |
619d094b | 1341 | return atomic_read(&memcg->moving_account) > 0; |
32047e2a | 1342 | } |
4b534334 | 1343 | |
c0ff4b85 | 1344 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1345 | { |
2bd9bb20 KH |
1346 | struct mem_cgroup *from; |
1347 | struct mem_cgroup *to; | |
4b534334 | 1348 | bool ret = false; |
2bd9bb20 KH |
1349 | /* |
1350 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1351 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1352 | */ | |
1353 | spin_lock(&mc.lock); | |
1354 | from = mc.from; | |
1355 | to = mc.to; | |
1356 | if (!from) | |
1357 | goto unlock; | |
3e92041d | 1358 | |
c0ff4b85 R |
1359 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1360 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1361 | unlock: |
1362 | spin_unlock(&mc.lock); | |
4b534334 KH |
1363 | return ret; |
1364 | } | |
1365 | ||
c0ff4b85 | 1366 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1367 | { |
1368 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1369 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1370 | DEFINE_WAIT(wait); |
1371 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1372 | /* moving charge context might have finished. */ | |
1373 | if (mc.moving_task) | |
1374 | schedule(); | |
1375 | finish_wait(&mc.waitq, &wait); | |
1376 | return true; | |
1377 | } | |
1378 | } | |
1379 | return false; | |
1380 | } | |
1381 | ||
312734c0 KH |
1382 | /* |
1383 | * Take this lock when | |
1384 | * - a code tries to modify page's memcg while it's USED. | |
1385 | * - a code tries to modify page state accounting in a memcg. | |
13fd1dd9 | 1386 | * see mem_cgroup_stolen(), too. |
312734c0 KH |
1387 | */ |
1388 | static void move_lock_mem_cgroup(struct mem_cgroup *memcg, | |
1389 | unsigned long *flags) | |
1390 | { | |
1391 | spin_lock_irqsave(&memcg->move_lock, *flags); | |
1392 | } | |
1393 | ||
1394 | static void move_unlock_mem_cgroup(struct mem_cgroup *memcg, | |
1395 | unsigned long *flags) | |
1396 | { | |
1397 | spin_unlock_irqrestore(&memcg->move_lock, *flags); | |
1398 | } | |
1399 | ||
e222432b | 1400 | /** |
6a6135b6 | 1401 | * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode. |
e222432b BS |
1402 | * @memcg: The memory cgroup that went over limit |
1403 | * @p: Task that is going to be killed | |
1404 | * | |
1405 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1406 | * enabled | |
1407 | */ | |
1408 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1409 | { | |
1410 | struct cgroup *task_cgrp; | |
1411 | struct cgroup *mem_cgrp; | |
1412 | /* | |
1413 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1414 | * on the assumption that OOM is serialized for memory controller. | |
1415 | * If this assumption is broken, revisit this code. | |
1416 | */ | |
1417 | static char memcg_name[PATH_MAX]; | |
1418 | int ret; | |
1419 | ||
d31f56db | 1420 | if (!memcg || !p) |
e222432b BS |
1421 | return; |
1422 | ||
e222432b BS |
1423 | rcu_read_lock(); |
1424 | ||
1425 | mem_cgrp = memcg->css.cgroup; | |
1426 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1427 | ||
1428 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1429 | if (ret < 0) { | |
1430 | /* | |
1431 | * Unfortunately, we are unable to convert to a useful name | |
1432 | * But we'll still print out the usage information | |
1433 | */ | |
1434 | rcu_read_unlock(); | |
1435 | goto done; | |
1436 | } | |
1437 | rcu_read_unlock(); | |
1438 | ||
1439 | printk(KERN_INFO "Task in %s killed", memcg_name); | |
1440 | ||
1441 | rcu_read_lock(); | |
1442 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1443 | if (ret < 0) { | |
1444 | rcu_read_unlock(); | |
1445 | goto done; | |
1446 | } | |
1447 | rcu_read_unlock(); | |
1448 | ||
1449 | /* | |
1450 | * Continues from above, so we don't need an KERN_ level | |
1451 | */ | |
1452 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); | |
1453 | done: | |
1454 | ||
1455 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", | |
1456 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, | |
1457 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1458 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
1459 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " | |
1460 | "failcnt %llu\n", | |
1461 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, | |
1462 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1463 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
1464 | } | |
1465 | ||
81d39c20 KH |
1466 | /* |
1467 | * This function returns the number of memcg under hierarchy tree. Returns | |
1468 | * 1(self count) if no children. | |
1469 | */ | |
c0ff4b85 | 1470 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1471 | { |
1472 | int num = 0; | |
7d74b06f KH |
1473 | struct mem_cgroup *iter; |
1474 | ||
c0ff4b85 | 1475 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1476 | num++; |
81d39c20 KH |
1477 | return num; |
1478 | } | |
1479 | ||
a63d83f4 DR |
1480 | /* |
1481 | * Return the memory (and swap, if configured) limit for a memcg. | |
1482 | */ | |
1483 | u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) | |
1484 | { | |
1485 | u64 limit; | |
1486 | u64 memsw; | |
1487 | ||
f3e8eb70 JW |
1488 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
1489 | limit += total_swap_pages << PAGE_SHIFT; | |
1490 | ||
a63d83f4 DR |
1491 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
1492 | /* | |
1493 | * If memsw is finite and limits the amount of swap space available | |
1494 | * to this memcg, return that limit. | |
1495 | */ | |
1496 | return min(limit, memsw); | |
1497 | } | |
1498 | ||
5660048c JW |
1499 | static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg, |
1500 | gfp_t gfp_mask, | |
1501 | unsigned long flags) | |
1502 | { | |
1503 | unsigned long total = 0; | |
1504 | bool noswap = false; | |
1505 | int loop; | |
1506 | ||
1507 | if (flags & MEM_CGROUP_RECLAIM_NOSWAP) | |
1508 | noswap = true; | |
1509 | if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum) | |
1510 | noswap = true; | |
1511 | ||
1512 | for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) { | |
1513 | if (loop) | |
1514 | drain_all_stock_async(memcg); | |
1515 | total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap); | |
1516 | /* | |
1517 | * Allow limit shrinkers, which are triggered directly | |
1518 | * by userspace, to catch signals and stop reclaim | |
1519 | * after minimal progress, regardless of the margin. | |
1520 | */ | |
1521 | if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK)) | |
1522 | break; | |
1523 | if (mem_cgroup_margin(memcg)) | |
1524 | break; | |
1525 | /* | |
1526 | * If nothing was reclaimed after two attempts, there | |
1527 | * may be no reclaimable pages in this hierarchy. | |
1528 | */ | |
1529 | if (loop && !total) | |
1530 | break; | |
1531 | } | |
1532 | return total; | |
1533 | } | |
1534 | ||
4d0c066d KH |
1535 | /** |
1536 | * test_mem_cgroup_node_reclaimable | |
1537 | * @mem: the target memcg | |
1538 | * @nid: the node ID to be checked. | |
1539 | * @noswap : specify true here if the user wants flle only information. | |
1540 | * | |
1541 | * This function returns whether the specified memcg contains any | |
1542 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1543 | * pages in the node. | |
1544 | */ | |
c0ff4b85 | 1545 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1546 | int nid, bool noswap) |
1547 | { | |
c0ff4b85 | 1548 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1549 | return true; |
1550 | if (noswap || !total_swap_pages) | |
1551 | return false; | |
c0ff4b85 | 1552 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1553 | return true; |
1554 | return false; | |
1555 | ||
1556 | } | |
889976db YH |
1557 | #if MAX_NUMNODES > 1 |
1558 | ||
1559 | /* | |
1560 | * Always updating the nodemask is not very good - even if we have an empty | |
1561 | * list or the wrong list here, we can start from some node and traverse all | |
1562 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1563 | * | |
1564 | */ | |
c0ff4b85 | 1565 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1566 | { |
1567 | int nid; | |
453a9bf3 KH |
1568 | /* |
1569 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1570 | * pagein/pageout changes since the last update. | |
1571 | */ | |
c0ff4b85 | 1572 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1573 | return; |
c0ff4b85 | 1574 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1575 | return; |
1576 | ||
889976db | 1577 | /* make a nodemask where this memcg uses memory from */ |
c0ff4b85 | 1578 | memcg->scan_nodes = node_states[N_HIGH_MEMORY]; |
889976db YH |
1579 | |
1580 | for_each_node_mask(nid, node_states[N_HIGH_MEMORY]) { | |
1581 | ||
c0ff4b85 R |
1582 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1583 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1584 | } |
453a9bf3 | 1585 | |
c0ff4b85 R |
1586 | atomic_set(&memcg->numainfo_events, 0); |
1587 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1588 | } |
1589 | ||
1590 | /* | |
1591 | * Selecting a node where we start reclaim from. Because what we need is just | |
1592 | * reducing usage counter, start from anywhere is O,K. Considering | |
1593 | * memory reclaim from current node, there are pros. and cons. | |
1594 | * | |
1595 | * Freeing memory from current node means freeing memory from a node which | |
1596 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1597 | * hit limits, it will see a contention on a node. But freeing from remote | |
1598 | * node means more costs for memory reclaim because of memory latency. | |
1599 | * | |
1600 | * Now, we use round-robin. Better algorithm is welcomed. | |
1601 | */ | |
c0ff4b85 | 1602 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1603 | { |
1604 | int node; | |
1605 | ||
c0ff4b85 R |
1606 | mem_cgroup_may_update_nodemask(memcg); |
1607 | node = memcg->last_scanned_node; | |
889976db | 1608 | |
c0ff4b85 | 1609 | node = next_node(node, memcg->scan_nodes); |
889976db | 1610 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1611 | node = first_node(memcg->scan_nodes); |
889976db YH |
1612 | /* |
1613 | * We call this when we hit limit, not when pages are added to LRU. | |
1614 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1615 | * memcg is too small and all pages are not on LRU. In that case, | |
1616 | * we use curret node. | |
1617 | */ | |
1618 | if (unlikely(node == MAX_NUMNODES)) | |
1619 | node = numa_node_id(); | |
1620 | ||
c0ff4b85 | 1621 | memcg->last_scanned_node = node; |
889976db YH |
1622 | return node; |
1623 | } | |
1624 | ||
4d0c066d KH |
1625 | /* |
1626 | * Check all nodes whether it contains reclaimable pages or not. | |
1627 | * For quick scan, we make use of scan_nodes. This will allow us to skip | |
1628 | * unused nodes. But scan_nodes is lazily updated and may not cotain | |
1629 | * enough new information. We need to do double check. | |
1630 | */ | |
6bbda35c | 1631 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d KH |
1632 | { |
1633 | int nid; | |
1634 | ||
1635 | /* | |
1636 | * quick check...making use of scan_node. | |
1637 | * We can skip unused nodes. | |
1638 | */ | |
c0ff4b85 R |
1639 | if (!nodes_empty(memcg->scan_nodes)) { |
1640 | for (nid = first_node(memcg->scan_nodes); | |
4d0c066d | 1641 | nid < MAX_NUMNODES; |
c0ff4b85 | 1642 | nid = next_node(nid, memcg->scan_nodes)) { |
4d0c066d | 1643 | |
c0ff4b85 | 1644 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
1645 | return true; |
1646 | } | |
1647 | } | |
1648 | /* | |
1649 | * Check rest of nodes. | |
1650 | */ | |
1651 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
c0ff4b85 | 1652 | if (node_isset(nid, memcg->scan_nodes)) |
4d0c066d | 1653 | continue; |
c0ff4b85 | 1654 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
1655 | return true; |
1656 | } | |
1657 | return false; | |
1658 | } | |
1659 | ||
889976db | 1660 | #else |
c0ff4b85 | 1661 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1662 | { |
1663 | return 0; | |
1664 | } | |
4d0c066d | 1665 | |
6bbda35c | 1666 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d | 1667 | { |
c0ff4b85 | 1668 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); |
4d0c066d | 1669 | } |
889976db YH |
1670 | #endif |
1671 | ||
5660048c JW |
1672 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1673 | struct zone *zone, | |
1674 | gfp_t gfp_mask, | |
1675 | unsigned long *total_scanned) | |
6d61ef40 | 1676 | { |
9f3a0d09 | 1677 | struct mem_cgroup *victim = NULL; |
5660048c | 1678 | int total = 0; |
04046e1a | 1679 | int loop = 0; |
9d11ea9f | 1680 | unsigned long excess; |
185efc0f | 1681 | unsigned long nr_scanned; |
527a5ec9 JW |
1682 | struct mem_cgroup_reclaim_cookie reclaim = { |
1683 | .zone = zone, | |
1684 | .priority = 0, | |
1685 | }; | |
9d11ea9f | 1686 | |
c0ff4b85 | 1687 | excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT; |
04046e1a | 1688 | |
4e416953 | 1689 | while (1) { |
527a5ec9 | 1690 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); |
9f3a0d09 | 1691 | if (!victim) { |
04046e1a | 1692 | loop++; |
4e416953 BS |
1693 | if (loop >= 2) { |
1694 | /* | |
1695 | * If we have not been able to reclaim | |
1696 | * anything, it might because there are | |
1697 | * no reclaimable pages under this hierarchy | |
1698 | */ | |
5660048c | 1699 | if (!total) |
4e416953 | 1700 | break; |
4e416953 | 1701 | /* |
25985edc | 1702 | * We want to do more targeted reclaim. |
4e416953 BS |
1703 | * excess >> 2 is not to excessive so as to |
1704 | * reclaim too much, nor too less that we keep | |
1705 | * coming back to reclaim from this cgroup | |
1706 | */ | |
1707 | if (total >= (excess >> 2) || | |
9f3a0d09 | 1708 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) |
4e416953 | 1709 | break; |
4e416953 | 1710 | } |
9f3a0d09 | 1711 | continue; |
4e416953 | 1712 | } |
5660048c | 1713 | if (!mem_cgroup_reclaimable(victim, false)) |
6d61ef40 | 1714 | continue; |
5660048c JW |
1715 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1716 | zone, &nr_scanned); | |
1717 | *total_scanned += nr_scanned; | |
1718 | if (!res_counter_soft_limit_excess(&root_memcg->res)) | |
9f3a0d09 | 1719 | break; |
6d61ef40 | 1720 | } |
9f3a0d09 | 1721 | mem_cgroup_iter_break(root_memcg, victim); |
04046e1a | 1722 | return total; |
6d61ef40 BS |
1723 | } |
1724 | ||
867578cb KH |
1725 | /* |
1726 | * Check OOM-Killer is already running under our hierarchy. | |
1727 | * If someone is running, return false. | |
1af8efe9 | 1728 | * Has to be called with memcg_oom_lock |
867578cb | 1729 | */ |
c0ff4b85 | 1730 | static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg) |
867578cb | 1731 | { |
79dfdacc | 1732 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1733 | |
9f3a0d09 | 1734 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1735 | if (iter->oom_lock) { |
79dfdacc MH |
1736 | /* |
1737 | * this subtree of our hierarchy is already locked | |
1738 | * so we cannot give a lock. | |
1739 | */ | |
79dfdacc | 1740 | failed = iter; |
9f3a0d09 JW |
1741 | mem_cgroup_iter_break(memcg, iter); |
1742 | break; | |
23751be0 JW |
1743 | } else |
1744 | iter->oom_lock = true; | |
7d74b06f | 1745 | } |
867578cb | 1746 | |
79dfdacc | 1747 | if (!failed) |
23751be0 | 1748 | return true; |
79dfdacc MH |
1749 | |
1750 | /* | |
1751 | * OK, we failed to lock the whole subtree so we have to clean up | |
1752 | * what we set up to the failing subtree | |
1753 | */ | |
9f3a0d09 | 1754 | for_each_mem_cgroup_tree(iter, memcg) { |
79dfdacc | 1755 | if (iter == failed) { |
9f3a0d09 JW |
1756 | mem_cgroup_iter_break(memcg, iter); |
1757 | break; | |
79dfdacc MH |
1758 | } |
1759 | iter->oom_lock = false; | |
1760 | } | |
23751be0 | 1761 | return false; |
a636b327 | 1762 | } |
0b7f569e | 1763 | |
79dfdacc | 1764 | /* |
1af8efe9 | 1765 | * Has to be called with memcg_oom_lock |
79dfdacc | 1766 | */ |
c0ff4b85 | 1767 | static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1768 | { |
7d74b06f KH |
1769 | struct mem_cgroup *iter; |
1770 | ||
c0ff4b85 | 1771 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1772 | iter->oom_lock = false; |
1773 | return 0; | |
1774 | } | |
1775 | ||
c0ff4b85 | 1776 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1777 | { |
1778 | struct mem_cgroup *iter; | |
1779 | ||
c0ff4b85 | 1780 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1781 | atomic_inc(&iter->under_oom); |
1782 | } | |
1783 | ||
c0ff4b85 | 1784 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1785 | { |
1786 | struct mem_cgroup *iter; | |
1787 | ||
867578cb KH |
1788 | /* |
1789 | * When a new child is created while the hierarchy is under oom, | |
1790 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1791 | * atomic_add_unless() here. | |
1792 | */ | |
c0ff4b85 | 1793 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1794 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
1795 | } |
1796 | ||
1af8efe9 | 1797 | static DEFINE_SPINLOCK(memcg_oom_lock); |
867578cb KH |
1798 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1799 | ||
dc98df5a | 1800 | struct oom_wait_info { |
d79154bb | 1801 | struct mem_cgroup *memcg; |
dc98df5a KH |
1802 | wait_queue_t wait; |
1803 | }; | |
1804 | ||
1805 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1806 | unsigned mode, int sync, void *arg) | |
1807 | { | |
d79154bb HD |
1808 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1809 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1810 | struct oom_wait_info *oom_wait_info; |
1811 | ||
1812 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1813 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1814 | |
dc98df5a | 1815 | /* |
d79154bb | 1816 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
dc98df5a KH |
1817 | * Then we can use css_is_ancestor without taking care of RCU. |
1818 | */ | |
c0ff4b85 R |
1819 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
1820 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 1821 | return 0; |
dc98df5a KH |
1822 | return autoremove_wake_function(wait, mode, sync, arg); |
1823 | } | |
1824 | ||
c0ff4b85 | 1825 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 1826 | { |
c0ff4b85 R |
1827 | /* for filtering, pass "memcg" as argument. */ |
1828 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
1829 | } |
1830 | ||
c0ff4b85 | 1831 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1832 | { |
c0ff4b85 R |
1833 | if (memcg && atomic_read(&memcg->under_oom)) |
1834 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
1835 | } |
1836 | ||
867578cb KH |
1837 | /* |
1838 | * try to call OOM killer. returns false if we should exit memory-reclaim loop. | |
1839 | */ | |
6bbda35c KS |
1840 | static bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask, |
1841 | int order) | |
0b7f569e | 1842 | { |
dc98df5a | 1843 | struct oom_wait_info owait; |
3c11ecf4 | 1844 | bool locked, need_to_kill; |
867578cb | 1845 | |
d79154bb | 1846 | owait.memcg = memcg; |
dc98df5a KH |
1847 | owait.wait.flags = 0; |
1848 | owait.wait.func = memcg_oom_wake_function; | |
1849 | owait.wait.private = current; | |
1850 | INIT_LIST_HEAD(&owait.wait.task_list); | |
3c11ecf4 | 1851 | need_to_kill = true; |
c0ff4b85 | 1852 | mem_cgroup_mark_under_oom(memcg); |
79dfdacc | 1853 | |
c0ff4b85 | 1854 | /* At first, try to OOM lock hierarchy under memcg.*/ |
1af8efe9 | 1855 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1856 | locked = mem_cgroup_oom_lock(memcg); |
867578cb KH |
1857 | /* |
1858 | * Even if signal_pending(), we can't quit charge() loop without | |
1859 | * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL | |
1860 | * under OOM is always welcomed, use TASK_KILLABLE here. | |
1861 | */ | |
3c11ecf4 | 1862 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
c0ff4b85 | 1863 | if (!locked || memcg->oom_kill_disable) |
3c11ecf4 KH |
1864 | need_to_kill = false; |
1865 | if (locked) | |
c0ff4b85 | 1866 | mem_cgroup_oom_notify(memcg); |
1af8efe9 | 1867 | spin_unlock(&memcg_oom_lock); |
867578cb | 1868 | |
3c11ecf4 KH |
1869 | if (need_to_kill) { |
1870 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
e845e199 | 1871 | mem_cgroup_out_of_memory(memcg, mask, order); |
3c11ecf4 | 1872 | } else { |
867578cb | 1873 | schedule(); |
dc98df5a | 1874 | finish_wait(&memcg_oom_waitq, &owait.wait); |
867578cb | 1875 | } |
1af8efe9 | 1876 | spin_lock(&memcg_oom_lock); |
79dfdacc | 1877 | if (locked) |
c0ff4b85 R |
1878 | mem_cgroup_oom_unlock(memcg); |
1879 | memcg_wakeup_oom(memcg); | |
1af8efe9 | 1880 | spin_unlock(&memcg_oom_lock); |
867578cb | 1881 | |
c0ff4b85 | 1882 | mem_cgroup_unmark_under_oom(memcg); |
79dfdacc | 1883 | |
867578cb KH |
1884 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) |
1885 | return false; | |
1886 | /* Give chance to dying process */ | |
715a5ee8 | 1887 | schedule_timeout_uninterruptible(1); |
867578cb | 1888 | return true; |
0b7f569e KH |
1889 | } |
1890 | ||
d69b042f BS |
1891 | /* |
1892 | * Currently used to update mapped file statistics, but the routine can be | |
1893 | * generalized to update other statistics as well. | |
32047e2a KH |
1894 | * |
1895 | * Notes: Race condition | |
1896 | * | |
1897 | * We usually use page_cgroup_lock() for accessing page_cgroup member but | |
1898 | * it tends to be costly. But considering some conditions, we doesn't need | |
1899 | * to do so _always_. | |
1900 | * | |
1901 | * Considering "charge", lock_page_cgroup() is not required because all | |
1902 | * file-stat operations happen after a page is attached to radix-tree. There | |
1903 | * are no race with "charge". | |
1904 | * | |
1905 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
1906 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
1907 | * if there are race with "uncharge". Statistics itself is properly handled | |
1908 | * by flags. | |
1909 | * | |
1910 | * Considering "move", this is an only case we see a race. To make the race | |
619d094b KH |
1911 | * small, we check mm->moving_account and detect there are possibility of race |
1912 | * If there is, we take a lock. | |
d69b042f | 1913 | */ |
26174efd | 1914 | |
89c06bd5 KH |
1915 | void __mem_cgroup_begin_update_page_stat(struct page *page, |
1916 | bool *locked, unsigned long *flags) | |
1917 | { | |
1918 | struct mem_cgroup *memcg; | |
1919 | struct page_cgroup *pc; | |
1920 | ||
1921 | pc = lookup_page_cgroup(page); | |
1922 | again: | |
1923 | memcg = pc->mem_cgroup; | |
1924 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
1925 | return; | |
1926 | /* | |
1927 | * If this memory cgroup is not under account moving, we don't | |
1928 | * need to take move_lock_page_cgroup(). Because we already hold | |
1929 | * rcu_read_lock(), any calls to move_account will be delayed until | |
13fd1dd9 | 1930 | * rcu_read_unlock() if mem_cgroup_stolen() == true. |
89c06bd5 | 1931 | */ |
13fd1dd9 | 1932 | if (!mem_cgroup_stolen(memcg)) |
89c06bd5 KH |
1933 | return; |
1934 | ||
1935 | move_lock_mem_cgroup(memcg, flags); | |
1936 | if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) { | |
1937 | move_unlock_mem_cgroup(memcg, flags); | |
1938 | goto again; | |
1939 | } | |
1940 | *locked = true; | |
1941 | } | |
1942 | ||
1943 | void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags) | |
1944 | { | |
1945 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
1946 | ||
1947 | /* | |
1948 | * It's guaranteed that pc->mem_cgroup never changes while | |
1949 | * lock is held because a routine modifies pc->mem_cgroup | |
1950 | * should take move_lock_page_cgroup(). | |
1951 | */ | |
1952 | move_unlock_mem_cgroup(pc->mem_cgroup, flags); | |
1953 | } | |
1954 | ||
2a7106f2 GT |
1955 | void mem_cgroup_update_page_stat(struct page *page, |
1956 | enum mem_cgroup_page_stat_item idx, int val) | |
d69b042f | 1957 | { |
c0ff4b85 | 1958 | struct mem_cgroup *memcg; |
32047e2a | 1959 | struct page_cgroup *pc = lookup_page_cgroup(page); |
dbd4ea78 | 1960 | unsigned long uninitialized_var(flags); |
d69b042f | 1961 | |
cfa44946 | 1962 | if (mem_cgroup_disabled()) |
d69b042f | 1963 | return; |
89c06bd5 | 1964 | |
c0ff4b85 R |
1965 | memcg = pc->mem_cgroup; |
1966 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
89c06bd5 | 1967 | return; |
26174efd | 1968 | |
26174efd | 1969 | switch (idx) { |
2a7106f2 | 1970 | case MEMCG_NR_FILE_MAPPED: |
2a7106f2 | 1971 | idx = MEM_CGROUP_STAT_FILE_MAPPED; |
26174efd KH |
1972 | break; |
1973 | default: | |
1974 | BUG(); | |
8725d541 | 1975 | } |
d69b042f | 1976 | |
c0ff4b85 | 1977 | this_cpu_add(memcg->stat->count[idx], val); |
d69b042f | 1978 | } |
26174efd | 1979 | |
cdec2e42 KH |
1980 | /* |
1981 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1982 | * TODO: maybe necessary to use big numbers in big irons. | |
1983 | */ | |
7ec99d62 | 1984 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
1985 | struct memcg_stock_pcp { |
1986 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 1987 | unsigned int nr_pages; |
cdec2e42 | 1988 | struct work_struct work; |
26fe6168 | 1989 | unsigned long flags; |
a0db00fc | 1990 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
1991 | }; |
1992 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 1993 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 KH |
1994 | |
1995 | /* | |
11c9ea4e | 1996 | * Try to consume stocked charge on this cpu. If success, one page is consumed |
cdec2e42 KH |
1997 | * from local stock and true is returned. If the stock is 0 or charges from a |
1998 | * cgroup which is not current target, returns false. This stock will be | |
1999 | * refilled. | |
2000 | */ | |
c0ff4b85 | 2001 | static bool consume_stock(struct mem_cgroup *memcg) |
cdec2e42 KH |
2002 | { |
2003 | struct memcg_stock_pcp *stock; | |
2004 | bool ret = true; | |
2005 | ||
2006 | stock = &get_cpu_var(memcg_stock); | |
c0ff4b85 | 2007 | if (memcg == stock->cached && stock->nr_pages) |
11c9ea4e | 2008 | stock->nr_pages--; |
cdec2e42 KH |
2009 | else /* need to call res_counter_charge */ |
2010 | ret = false; | |
2011 | put_cpu_var(memcg_stock); | |
2012 | return ret; | |
2013 | } | |
2014 | ||
2015 | /* | |
2016 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
2017 | */ | |
2018 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2019 | { | |
2020 | struct mem_cgroup *old = stock->cached; | |
2021 | ||
11c9ea4e JW |
2022 | if (stock->nr_pages) { |
2023 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
2024 | ||
2025 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 2026 | if (do_swap_account) |
11c9ea4e JW |
2027 | res_counter_uncharge(&old->memsw, bytes); |
2028 | stock->nr_pages = 0; | |
cdec2e42 KH |
2029 | } |
2030 | stock->cached = NULL; | |
cdec2e42 KH |
2031 | } |
2032 | ||
2033 | /* | |
2034 | * This must be called under preempt disabled or must be called by | |
2035 | * a thread which is pinned to local cpu. | |
2036 | */ | |
2037 | static void drain_local_stock(struct work_struct *dummy) | |
2038 | { | |
2039 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
2040 | drain_stock(stock); | |
26fe6168 | 2041 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2042 | } |
2043 | ||
2044 | /* | |
2045 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 2046 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2047 | */ |
c0ff4b85 | 2048 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2049 | { |
2050 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2051 | ||
c0ff4b85 | 2052 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2053 | drain_stock(stock); |
c0ff4b85 | 2054 | stock->cached = memcg; |
cdec2e42 | 2055 | } |
11c9ea4e | 2056 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2057 | put_cpu_var(memcg_stock); |
2058 | } | |
2059 | ||
2060 | /* | |
c0ff4b85 | 2061 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
d38144b7 MH |
2062 | * of the hierarchy under it. sync flag says whether we should block |
2063 | * until the work is done. | |
cdec2e42 | 2064 | */ |
c0ff4b85 | 2065 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
cdec2e42 | 2066 | { |
26fe6168 | 2067 | int cpu, curcpu; |
d38144b7 | 2068 | |
cdec2e42 | 2069 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2070 | get_online_cpus(); |
5af12d0e | 2071 | curcpu = get_cpu(); |
cdec2e42 KH |
2072 | for_each_online_cpu(cpu) { |
2073 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2074 | struct mem_cgroup *memcg; |
26fe6168 | 2075 | |
c0ff4b85 R |
2076 | memcg = stock->cached; |
2077 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2078 | continue; |
c0ff4b85 | 2079 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2080 | continue; |
d1a05b69 MH |
2081 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2082 | if (cpu == curcpu) | |
2083 | drain_local_stock(&stock->work); | |
2084 | else | |
2085 | schedule_work_on(cpu, &stock->work); | |
2086 | } | |
cdec2e42 | 2087 | } |
5af12d0e | 2088 | put_cpu(); |
d38144b7 MH |
2089 | |
2090 | if (!sync) | |
2091 | goto out; | |
2092 | ||
2093 | for_each_online_cpu(cpu) { | |
2094 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2095 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2096 | flush_work(&stock->work); |
2097 | } | |
2098 | out: | |
cdec2e42 | 2099 | put_online_cpus(); |
d38144b7 MH |
2100 | } |
2101 | ||
2102 | /* | |
2103 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2104 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
2105 | * expects some charges will be back to res_counter later but cannot wait for | |
2106 | * it. | |
2107 | */ | |
c0ff4b85 | 2108 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
d38144b7 | 2109 | { |
9f50fad6 MH |
2110 | /* |
2111 | * If someone calls draining, avoid adding more kworker runs. | |
2112 | */ | |
2113 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2114 | return; | |
c0ff4b85 | 2115 | drain_all_stock(root_memcg, false); |
9f50fad6 | 2116 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2117 | } |
2118 | ||
2119 | /* This is a synchronous drain interface. */ | |
c0ff4b85 | 2120 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
cdec2e42 KH |
2121 | { |
2122 | /* called when force_empty is called */ | |
9f50fad6 | 2123 | mutex_lock(&percpu_charge_mutex); |
c0ff4b85 | 2124 | drain_all_stock(root_memcg, true); |
9f50fad6 | 2125 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2126 | } |
2127 | ||
711d3d2c KH |
2128 | /* |
2129 | * This function drains percpu counter value from DEAD cpu and | |
2130 | * move it to local cpu. Note that this function can be preempted. | |
2131 | */ | |
c0ff4b85 | 2132 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2133 | { |
2134 | int i; | |
2135 | ||
c0ff4b85 | 2136 | spin_lock(&memcg->pcp_counter_lock); |
711d3d2c | 2137 | for (i = 0; i < MEM_CGROUP_STAT_DATA; i++) { |
c0ff4b85 | 2138 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2139 | |
c0ff4b85 R |
2140 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2141 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2142 | } |
e9f8974f | 2143 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2144 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2145 | |
c0ff4b85 R |
2146 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2147 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2148 | } |
c0ff4b85 | 2149 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2150 | } |
2151 | ||
2152 | static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb, | |
cdec2e42 KH |
2153 | unsigned long action, |
2154 | void *hcpu) | |
2155 | { | |
2156 | int cpu = (unsigned long)hcpu; | |
2157 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2158 | struct mem_cgroup *iter; |
cdec2e42 | 2159 | |
619d094b | 2160 | if (action == CPU_ONLINE) |
1489ebad | 2161 | return NOTIFY_OK; |
1489ebad | 2162 | |
d833049b | 2163 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2164 | return NOTIFY_OK; |
711d3d2c | 2165 | |
9f3a0d09 | 2166 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2167 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2168 | ||
cdec2e42 KH |
2169 | stock = &per_cpu(memcg_stock, cpu); |
2170 | drain_stock(stock); | |
2171 | return NOTIFY_OK; | |
2172 | } | |
2173 | ||
4b534334 KH |
2174 | |
2175 | /* See __mem_cgroup_try_charge() for details */ | |
2176 | enum { | |
2177 | CHARGE_OK, /* success */ | |
2178 | CHARGE_RETRY, /* need to retry but retry is not bad */ | |
2179 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ | |
2180 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ | |
2181 | CHARGE_OOM_DIE, /* the current is killed because of OOM */ | |
2182 | }; | |
2183 | ||
c0ff4b85 | 2184 | static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
7ec99d62 | 2185 | unsigned int nr_pages, bool oom_check) |
4b534334 | 2186 | { |
7ec99d62 | 2187 | unsigned long csize = nr_pages * PAGE_SIZE; |
4b534334 KH |
2188 | struct mem_cgroup *mem_over_limit; |
2189 | struct res_counter *fail_res; | |
2190 | unsigned long flags = 0; | |
2191 | int ret; | |
2192 | ||
c0ff4b85 | 2193 | ret = res_counter_charge(&memcg->res, csize, &fail_res); |
4b534334 KH |
2194 | |
2195 | if (likely(!ret)) { | |
2196 | if (!do_swap_account) | |
2197 | return CHARGE_OK; | |
c0ff4b85 | 2198 | ret = res_counter_charge(&memcg->memsw, csize, &fail_res); |
4b534334 KH |
2199 | if (likely(!ret)) |
2200 | return CHARGE_OK; | |
2201 | ||
c0ff4b85 | 2202 | res_counter_uncharge(&memcg->res, csize); |
4b534334 KH |
2203 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
2204 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
2205 | } else | |
2206 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
9221edb7 | 2207 | /* |
7ec99d62 JW |
2208 | * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch |
2209 | * of regular pages (CHARGE_BATCH), or a single regular page (1). | |
9221edb7 JW |
2210 | * |
2211 | * Never reclaim on behalf of optional batching, retry with a | |
2212 | * single page instead. | |
2213 | */ | |
7ec99d62 | 2214 | if (nr_pages == CHARGE_BATCH) |
4b534334 KH |
2215 | return CHARGE_RETRY; |
2216 | ||
2217 | if (!(gfp_mask & __GFP_WAIT)) | |
2218 | return CHARGE_WOULDBLOCK; | |
2219 | ||
5660048c | 2220 | ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags); |
7ec99d62 | 2221 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
19942822 | 2222 | return CHARGE_RETRY; |
4b534334 | 2223 | /* |
19942822 JW |
2224 | * Even though the limit is exceeded at this point, reclaim |
2225 | * may have been able to free some pages. Retry the charge | |
2226 | * before killing the task. | |
2227 | * | |
2228 | * Only for regular pages, though: huge pages are rather | |
2229 | * unlikely to succeed so close to the limit, and we fall back | |
2230 | * to regular pages anyway in case of failure. | |
4b534334 | 2231 | */ |
7ec99d62 | 2232 | if (nr_pages == 1 && ret) |
4b534334 KH |
2233 | return CHARGE_RETRY; |
2234 | ||
2235 | /* | |
2236 | * At task move, charge accounts can be doubly counted. So, it's | |
2237 | * better to wait until the end of task_move if something is going on. | |
2238 | */ | |
2239 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2240 | return CHARGE_RETRY; | |
2241 | ||
2242 | /* If we don't need to call oom-killer at el, return immediately */ | |
2243 | if (!oom_check) | |
2244 | return CHARGE_NOMEM; | |
2245 | /* check OOM */ | |
e845e199 | 2246 | if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask, get_order(csize))) |
4b534334 KH |
2247 | return CHARGE_OOM_DIE; |
2248 | ||
2249 | return CHARGE_RETRY; | |
2250 | } | |
2251 | ||
f817ed48 | 2252 | /* |
38c5d72f KH |
2253 | * __mem_cgroup_try_charge() does |
2254 | * 1. detect memcg to be charged against from passed *mm and *ptr, | |
2255 | * 2. update res_counter | |
2256 | * 3. call memory reclaim if necessary. | |
2257 | * | |
2258 | * In some special case, if the task is fatal, fatal_signal_pending() or | |
2259 | * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup | |
2260 | * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon | |
2261 | * as possible without any hazards. 2: all pages should have a valid | |
2262 | * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg | |
2263 | * pointer, that is treated as a charge to root_mem_cgroup. | |
2264 | * | |
2265 | * So __mem_cgroup_try_charge() will return | |
2266 | * 0 ... on success, filling *ptr with a valid memcg pointer. | |
2267 | * -ENOMEM ... charge failure because of resource limits. | |
2268 | * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup. | |
2269 | * | |
2270 | * Unlike the exported interface, an "oom" parameter is added. if oom==true, | |
2271 | * the oom-killer can be invoked. | |
8a9f3ccd | 2272 | */ |
f817ed48 | 2273 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
ec168510 | 2274 | gfp_t gfp_mask, |
7ec99d62 | 2275 | unsigned int nr_pages, |
c0ff4b85 | 2276 | struct mem_cgroup **ptr, |
7ec99d62 | 2277 | bool oom) |
8a9f3ccd | 2278 | { |
7ec99d62 | 2279 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
4b534334 | 2280 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
c0ff4b85 | 2281 | struct mem_cgroup *memcg = NULL; |
4b534334 | 2282 | int ret; |
a636b327 | 2283 | |
867578cb KH |
2284 | /* |
2285 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
2286 | * in system level. So, allow to go ahead dying process in addition to | |
2287 | * MEMDIE process. | |
2288 | */ | |
2289 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
2290 | || fatal_signal_pending(current))) | |
2291 | goto bypass; | |
a636b327 | 2292 | |
8a9f3ccd | 2293 | /* |
3be91277 HD |
2294 | * We always charge the cgroup the mm_struct belongs to. |
2295 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
2296 | * thread group leader migrates. It's possible that mm is not |
2297 | * set, if so charge the init_mm (happens for pagecache usage). | |
2298 | */ | |
c0ff4b85 | 2299 | if (!*ptr && !mm) |
38c5d72f | 2300 | *ptr = root_mem_cgroup; |
f75ca962 | 2301 | again: |
c0ff4b85 R |
2302 | if (*ptr) { /* css should be a valid one */ |
2303 | memcg = *ptr; | |
2304 | VM_BUG_ON(css_is_removed(&memcg->css)); | |
2305 | if (mem_cgroup_is_root(memcg)) | |
f75ca962 | 2306 | goto done; |
c0ff4b85 | 2307 | if (nr_pages == 1 && consume_stock(memcg)) |
f75ca962 | 2308 | goto done; |
c0ff4b85 | 2309 | css_get(&memcg->css); |
4b534334 | 2310 | } else { |
f75ca962 | 2311 | struct task_struct *p; |
54595fe2 | 2312 | |
f75ca962 KH |
2313 | rcu_read_lock(); |
2314 | p = rcu_dereference(mm->owner); | |
f75ca962 | 2315 | /* |
ebb76ce1 | 2316 | * Because we don't have task_lock(), "p" can exit. |
c0ff4b85 | 2317 | * In that case, "memcg" can point to root or p can be NULL with |
ebb76ce1 KH |
2318 | * race with swapoff. Then, we have small risk of mis-accouning. |
2319 | * But such kind of mis-account by race always happens because | |
2320 | * we don't have cgroup_mutex(). It's overkill and we allo that | |
2321 | * small race, here. | |
2322 | * (*) swapoff at el will charge against mm-struct not against | |
2323 | * task-struct. So, mm->owner can be NULL. | |
f75ca962 | 2324 | */ |
c0ff4b85 | 2325 | memcg = mem_cgroup_from_task(p); |
38c5d72f KH |
2326 | if (!memcg) |
2327 | memcg = root_mem_cgroup; | |
2328 | if (mem_cgroup_is_root(memcg)) { | |
f75ca962 KH |
2329 | rcu_read_unlock(); |
2330 | goto done; | |
2331 | } | |
c0ff4b85 | 2332 | if (nr_pages == 1 && consume_stock(memcg)) { |
f75ca962 KH |
2333 | /* |
2334 | * It seems dagerous to access memcg without css_get(). | |
2335 | * But considering how consume_stok works, it's not | |
2336 | * necessary. If consume_stock success, some charges | |
2337 | * from this memcg are cached on this cpu. So, we | |
2338 | * don't need to call css_get()/css_tryget() before | |
2339 | * calling consume_stock(). | |
2340 | */ | |
2341 | rcu_read_unlock(); | |
2342 | goto done; | |
2343 | } | |
2344 | /* after here, we may be blocked. we need to get refcnt */ | |
c0ff4b85 | 2345 | if (!css_tryget(&memcg->css)) { |
f75ca962 KH |
2346 | rcu_read_unlock(); |
2347 | goto again; | |
2348 | } | |
2349 | rcu_read_unlock(); | |
2350 | } | |
8a9f3ccd | 2351 | |
4b534334 KH |
2352 | do { |
2353 | bool oom_check; | |
7a81b88c | 2354 | |
4b534334 | 2355 | /* If killed, bypass charge */ |
f75ca962 | 2356 | if (fatal_signal_pending(current)) { |
c0ff4b85 | 2357 | css_put(&memcg->css); |
4b534334 | 2358 | goto bypass; |
f75ca962 | 2359 | } |
6d61ef40 | 2360 | |
4b534334 KH |
2361 | oom_check = false; |
2362 | if (oom && !nr_oom_retries) { | |
2363 | oom_check = true; | |
2364 | nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
cdec2e42 | 2365 | } |
66e1707b | 2366 | |
c0ff4b85 | 2367 | ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, oom_check); |
4b534334 KH |
2368 | switch (ret) { |
2369 | case CHARGE_OK: | |
2370 | break; | |
2371 | case CHARGE_RETRY: /* not in OOM situation but retry */ | |
7ec99d62 | 2372 | batch = nr_pages; |
c0ff4b85 R |
2373 | css_put(&memcg->css); |
2374 | memcg = NULL; | |
f75ca962 | 2375 | goto again; |
4b534334 | 2376 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
c0ff4b85 | 2377 | css_put(&memcg->css); |
4b534334 KH |
2378 | goto nomem; |
2379 | case CHARGE_NOMEM: /* OOM routine works */ | |
f75ca962 | 2380 | if (!oom) { |
c0ff4b85 | 2381 | css_put(&memcg->css); |
867578cb | 2382 | goto nomem; |
f75ca962 | 2383 | } |
4b534334 KH |
2384 | /* If oom, we never return -ENOMEM */ |
2385 | nr_oom_retries--; | |
2386 | break; | |
2387 | case CHARGE_OOM_DIE: /* Killed by OOM Killer */ | |
c0ff4b85 | 2388 | css_put(&memcg->css); |
867578cb | 2389 | goto bypass; |
66e1707b | 2390 | } |
4b534334 KH |
2391 | } while (ret != CHARGE_OK); |
2392 | ||
7ec99d62 | 2393 | if (batch > nr_pages) |
c0ff4b85 R |
2394 | refill_stock(memcg, batch - nr_pages); |
2395 | css_put(&memcg->css); | |
0c3e73e8 | 2396 | done: |
c0ff4b85 | 2397 | *ptr = memcg; |
7a81b88c KH |
2398 | return 0; |
2399 | nomem: | |
c0ff4b85 | 2400 | *ptr = NULL; |
7a81b88c | 2401 | return -ENOMEM; |
867578cb | 2402 | bypass: |
38c5d72f KH |
2403 | *ptr = root_mem_cgroup; |
2404 | return -EINTR; | |
7a81b88c | 2405 | } |
8a9f3ccd | 2406 | |
a3032a2c DN |
2407 | /* |
2408 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2409 | * This function is for that and do uncharge, put css's refcnt. | |
2410 | * gotten by try_charge(). | |
2411 | */ | |
c0ff4b85 | 2412 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg, |
e7018b8d | 2413 | unsigned int nr_pages) |
a3032a2c | 2414 | { |
c0ff4b85 | 2415 | if (!mem_cgroup_is_root(memcg)) { |
e7018b8d JW |
2416 | unsigned long bytes = nr_pages * PAGE_SIZE; |
2417 | ||
c0ff4b85 | 2418 | res_counter_uncharge(&memcg->res, bytes); |
a3032a2c | 2419 | if (do_swap_account) |
c0ff4b85 | 2420 | res_counter_uncharge(&memcg->memsw, bytes); |
a3032a2c | 2421 | } |
854ffa8d DN |
2422 | } |
2423 | ||
a3b2d692 KH |
2424 | /* |
2425 | * A helper function to get mem_cgroup from ID. must be called under | |
2426 | * rcu_read_lock(). The caller must check css_is_removed() or some if | |
2427 | * it's concern. (dropping refcnt from swap can be called against removed | |
2428 | * memcg.) | |
2429 | */ | |
2430 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2431 | { | |
2432 | struct cgroup_subsys_state *css; | |
2433 | ||
2434 | /* ID 0 is unused ID */ | |
2435 | if (!id) | |
2436 | return NULL; | |
2437 | css = css_lookup(&mem_cgroup_subsys, id); | |
2438 | if (!css) | |
2439 | return NULL; | |
2440 | return container_of(css, struct mem_cgroup, css); | |
2441 | } | |
2442 | ||
e42d9d5d | 2443 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2444 | { |
c0ff4b85 | 2445 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2446 | struct page_cgroup *pc; |
a3b2d692 | 2447 | unsigned short id; |
b5a84319 KH |
2448 | swp_entry_t ent; |
2449 | ||
3c776e64 DN |
2450 | VM_BUG_ON(!PageLocked(page)); |
2451 | ||
3c776e64 | 2452 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2453 | lock_page_cgroup(pc); |
a3b2d692 | 2454 | if (PageCgroupUsed(pc)) { |
c0ff4b85 R |
2455 | memcg = pc->mem_cgroup; |
2456 | if (memcg && !css_tryget(&memcg->css)) | |
2457 | memcg = NULL; | |
e42d9d5d | 2458 | } else if (PageSwapCache(page)) { |
3c776e64 | 2459 | ent.val = page_private(page); |
9fb4b7cc | 2460 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2461 | rcu_read_lock(); |
c0ff4b85 R |
2462 | memcg = mem_cgroup_lookup(id); |
2463 | if (memcg && !css_tryget(&memcg->css)) | |
2464 | memcg = NULL; | |
a3b2d692 | 2465 | rcu_read_unlock(); |
3c776e64 | 2466 | } |
c0bd3f63 | 2467 | unlock_page_cgroup(pc); |
c0ff4b85 | 2468 | return memcg; |
b5a84319 KH |
2469 | } |
2470 | ||
c0ff4b85 | 2471 | static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg, |
5564e88b | 2472 | struct page *page, |
7ec99d62 | 2473 | unsigned int nr_pages, |
9ce70c02 HD |
2474 | enum charge_type ctype, |
2475 | bool lrucare) | |
7a81b88c | 2476 | { |
ce587e65 | 2477 | struct page_cgroup *pc = lookup_page_cgroup(page); |
9ce70c02 HD |
2478 | struct zone *uninitialized_var(zone); |
2479 | bool was_on_lru = false; | |
b2402857 | 2480 | bool anon; |
9ce70c02 | 2481 | |
ca3e0214 KH |
2482 | lock_page_cgroup(pc); |
2483 | if (unlikely(PageCgroupUsed(pc))) { | |
2484 | unlock_page_cgroup(pc); | |
c0ff4b85 | 2485 | __mem_cgroup_cancel_charge(memcg, nr_pages); |
ca3e0214 KH |
2486 | return; |
2487 | } | |
2488 | /* | |
2489 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2490 | * accessed by any other context at this point. | |
2491 | */ | |
9ce70c02 HD |
2492 | |
2493 | /* | |
2494 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2495 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2496 | */ | |
2497 | if (lrucare) { | |
2498 | zone = page_zone(page); | |
2499 | spin_lock_irq(&zone->lru_lock); | |
2500 | if (PageLRU(page)) { | |
2501 | ClearPageLRU(page); | |
2502 | del_page_from_lru_list(zone, page, page_lru(page)); | |
2503 | was_on_lru = true; | |
2504 | } | |
2505 | } | |
2506 | ||
c0ff4b85 | 2507 | pc->mem_cgroup = memcg; |
261fb61a KH |
2508 | /* |
2509 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
2510 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
2511 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
2512 | * before USED bit, we need memory barrier here. | |
2513 | * See mem_cgroup_add_lru_list(), etc. | |
2514 | */ | |
08e552c6 | 2515 | smp_wmb(); |
b2402857 | 2516 | SetPageCgroupUsed(pc); |
3be91277 | 2517 | |
9ce70c02 HD |
2518 | if (lrucare) { |
2519 | if (was_on_lru) { | |
2520 | VM_BUG_ON(PageLRU(page)); | |
2521 | SetPageLRU(page); | |
2522 | add_page_to_lru_list(zone, page, page_lru(page)); | |
2523 | } | |
2524 | spin_unlock_irq(&zone->lru_lock); | |
2525 | } | |
2526 | ||
b2402857 KH |
2527 | if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED) |
2528 | anon = true; | |
2529 | else | |
2530 | anon = false; | |
2531 | ||
2532 | mem_cgroup_charge_statistics(memcg, anon, nr_pages); | |
52d4b9ac | 2533 | unlock_page_cgroup(pc); |
9ce70c02 | 2534 | |
430e4863 KH |
2535 | /* |
2536 | * "charge_statistics" updated event counter. Then, check it. | |
2537 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
2538 | * if they exceeds softlimit. | |
2539 | */ | |
c0ff4b85 | 2540 | memcg_check_events(memcg, page); |
7a81b88c | 2541 | } |
66e1707b | 2542 | |
ca3e0214 KH |
2543 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2544 | ||
a0db00fc | 2545 | #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION) |
ca3e0214 KH |
2546 | /* |
2547 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
2548 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
2549 | * charge/uncharge will be never happen and move_account() is done under | |
2550 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 2551 | */ |
e94c8a9c | 2552 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 KH |
2553 | { |
2554 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
e94c8a9c KH |
2555 | struct page_cgroup *pc; |
2556 | int i; | |
ca3e0214 | 2557 | |
3d37c4a9 KH |
2558 | if (mem_cgroup_disabled()) |
2559 | return; | |
e94c8a9c KH |
2560 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
2561 | pc = head_pc + i; | |
2562 | pc->mem_cgroup = head_pc->mem_cgroup; | |
2563 | smp_wmb();/* see __commit_charge() */ | |
e94c8a9c KH |
2564 | pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
2565 | } | |
ca3e0214 | 2566 | } |
12d27107 | 2567 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2568 | |
f817ed48 | 2569 | /** |
de3638d9 | 2570 | * mem_cgroup_move_account - move account of the page |
5564e88b | 2571 | * @page: the page |
7ec99d62 | 2572 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
2573 | * @pc: page_cgroup of the page. |
2574 | * @from: mem_cgroup which the page is moved from. | |
2575 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
854ffa8d | 2576 | * @uncharge: whether we should call uncharge and css_put against @from. |
f817ed48 KH |
2577 | * |
2578 | * The caller must confirm following. | |
08e552c6 | 2579 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 2580 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 2581 | * |
854ffa8d | 2582 | * This function doesn't do "charge" nor css_get to new cgroup. It should be |
25985edc | 2583 | * done by a caller(__mem_cgroup_try_charge would be useful). If @uncharge is |
854ffa8d DN |
2584 | * true, this function does "uncharge" from old cgroup, but it doesn't if |
2585 | * @uncharge is false, so a caller should do "uncharge". | |
f817ed48 | 2586 | */ |
7ec99d62 JW |
2587 | static int mem_cgroup_move_account(struct page *page, |
2588 | unsigned int nr_pages, | |
2589 | struct page_cgroup *pc, | |
2590 | struct mem_cgroup *from, | |
2591 | struct mem_cgroup *to, | |
2592 | bool uncharge) | |
f817ed48 | 2593 | { |
de3638d9 JW |
2594 | unsigned long flags; |
2595 | int ret; | |
b2402857 | 2596 | bool anon = PageAnon(page); |
987eba66 | 2597 | |
f817ed48 | 2598 | VM_BUG_ON(from == to); |
5564e88b | 2599 | VM_BUG_ON(PageLRU(page)); |
de3638d9 JW |
2600 | /* |
2601 | * The page is isolated from LRU. So, collapse function | |
2602 | * will not handle this page. But page splitting can happen. | |
2603 | * Do this check under compound_page_lock(). The caller should | |
2604 | * hold it. | |
2605 | */ | |
2606 | ret = -EBUSY; | |
7ec99d62 | 2607 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
2608 | goto out; |
2609 | ||
2610 | lock_page_cgroup(pc); | |
2611 | ||
2612 | ret = -EINVAL; | |
2613 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
2614 | goto unlock; | |
2615 | ||
312734c0 | 2616 | move_lock_mem_cgroup(from, &flags); |
f817ed48 | 2617 | |
2ff76f11 | 2618 | if (!anon && page_mapped(page)) { |
c62b1a3b KH |
2619 | /* Update mapped_file data for mem_cgroup */ |
2620 | preempt_disable(); | |
2621 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2622 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2623 | preempt_enable(); | |
d69b042f | 2624 | } |
b2402857 | 2625 | mem_cgroup_charge_statistics(from, anon, -nr_pages); |
854ffa8d DN |
2626 | if (uncharge) |
2627 | /* This is not "cancel", but cancel_charge does all we need. */ | |
e7018b8d | 2628 | __mem_cgroup_cancel_charge(from, nr_pages); |
d69b042f | 2629 | |
854ffa8d | 2630 | /* caller should have done css_get */ |
08e552c6 | 2631 | pc->mem_cgroup = to; |
b2402857 | 2632 | mem_cgroup_charge_statistics(to, anon, nr_pages); |
88703267 KH |
2633 | /* |
2634 | * We charges against "to" which may not have any tasks. Then, "to" | |
2635 | * can be under rmdir(). But in current implementation, caller of | |
4ffef5fe | 2636 | * this function is just force_empty() and move charge, so it's |
25985edc | 2637 | * guaranteed that "to" is never removed. So, we don't check rmdir |
4ffef5fe | 2638 | * status here. |
88703267 | 2639 | */ |
312734c0 | 2640 | move_unlock_mem_cgroup(from, &flags); |
de3638d9 JW |
2641 | ret = 0; |
2642 | unlock: | |
57f9fd7d | 2643 | unlock_page_cgroup(pc); |
d2265e6f KH |
2644 | /* |
2645 | * check events | |
2646 | */ | |
5564e88b JW |
2647 | memcg_check_events(to, page); |
2648 | memcg_check_events(from, page); | |
de3638d9 | 2649 | out: |
f817ed48 KH |
2650 | return ret; |
2651 | } | |
2652 | ||
2653 | /* | |
2654 | * move charges to its parent. | |
2655 | */ | |
2656 | ||
5564e88b JW |
2657 | static int mem_cgroup_move_parent(struct page *page, |
2658 | struct page_cgroup *pc, | |
f817ed48 KH |
2659 | struct mem_cgroup *child, |
2660 | gfp_t gfp_mask) | |
2661 | { | |
2662 | struct cgroup *cg = child->css.cgroup; | |
2663 | struct cgroup *pcg = cg->parent; | |
2664 | struct mem_cgroup *parent; | |
7ec99d62 | 2665 | unsigned int nr_pages; |
4be4489f | 2666 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
2667 | int ret; |
2668 | ||
2669 | /* Is ROOT ? */ | |
2670 | if (!pcg) | |
2671 | return -EINVAL; | |
2672 | ||
57f9fd7d DN |
2673 | ret = -EBUSY; |
2674 | if (!get_page_unless_zero(page)) | |
2675 | goto out; | |
2676 | if (isolate_lru_page(page)) | |
2677 | goto put; | |
52dbb905 | 2678 | |
7ec99d62 | 2679 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 2680 | |
f817ed48 | 2681 | parent = mem_cgroup_from_cont(pcg); |
7ec99d62 | 2682 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, nr_pages, &parent, false); |
38c5d72f | 2683 | if (ret) |
57f9fd7d | 2684 | goto put_back; |
f817ed48 | 2685 | |
7ec99d62 | 2686 | if (nr_pages > 1) |
987eba66 KH |
2687 | flags = compound_lock_irqsave(page); |
2688 | ||
7ec99d62 | 2689 | ret = mem_cgroup_move_account(page, nr_pages, pc, child, parent, true); |
854ffa8d | 2690 | if (ret) |
7ec99d62 | 2691 | __mem_cgroup_cancel_charge(parent, nr_pages); |
8dba474f | 2692 | |
7ec99d62 | 2693 | if (nr_pages > 1) |
987eba66 | 2694 | compound_unlock_irqrestore(page, flags); |
8dba474f | 2695 | put_back: |
08e552c6 | 2696 | putback_lru_page(page); |
57f9fd7d | 2697 | put: |
40d58138 | 2698 | put_page(page); |
57f9fd7d | 2699 | out: |
f817ed48 KH |
2700 | return ret; |
2701 | } | |
2702 | ||
7a81b88c KH |
2703 | /* |
2704 | * Charge the memory controller for page usage. | |
2705 | * Return | |
2706 | * 0 if the charge was successful | |
2707 | * < 0 if the cgroup is over its limit | |
2708 | */ | |
2709 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
73045c47 | 2710 | gfp_t gfp_mask, enum charge_type ctype) |
7a81b88c | 2711 | { |
c0ff4b85 | 2712 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 2713 | unsigned int nr_pages = 1; |
8493ae43 | 2714 | bool oom = true; |
7a81b88c | 2715 | int ret; |
ec168510 | 2716 | |
37c2ac78 | 2717 | if (PageTransHuge(page)) { |
7ec99d62 | 2718 | nr_pages <<= compound_order(page); |
37c2ac78 | 2719 | VM_BUG_ON(!PageTransHuge(page)); |
8493ae43 JW |
2720 | /* |
2721 | * Never OOM-kill a process for a huge page. The | |
2722 | * fault handler will fall back to regular pages. | |
2723 | */ | |
2724 | oom = false; | |
37c2ac78 | 2725 | } |
7a81b88c | 2726 | |
c0ff4b85 | 2727 | ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom); |
38c5d72f | 2728 | if (ret == -ENOMEM) |
7a81b88c | 2729 | return ret; |
ce587e65 | 2730 | __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false); |
8a9f3ccd | 2731 | return 0; |
8a9f3ccd BS |
2732 | } |
2733 | ||
7a81b88c KH |
2734 | int mem_cgroup_newpage_charge(struct page *page, |
2735 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 2736 | { |
f8d66542 | 2737 | if (mem_cgroup_disabled()) |
cede86ac | 2738 | return 0; |
7a0524cf JW |
2739 | VM_BUG_ON(page_mapped(page)); |
2740 | VM_BUG_ON(page->mapping && !PageAnon(page)); | |
2741 | VM_BUG_ON(!mm); | |
217bc319 | 2742 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
7a0524cf | 2743 | MEM_CGROUP_CHARGE_TYPE_MAPPED); |
217bc319 KH |
2744 | } |
2745 | ||
83aae4c7 DN |
2746 | static void |
2747 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
2748 | enum charge_type ctype); | |
2749 | ||
e1a1cd59 BS |
2750 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
2751 | gfp_t gfp_mask) | |
8697d331 | 2752 | { |
c0ff4b85 | 2753 | struct mem_cgroup *memcg = NULL; |
dc67d504 | 2754 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
b5a84319 KH |
2755 | int ret; |
2756 | ||
f8d66542 | 2757 | if (mem_cgroup_disabled()) |
cede86ac | 2758 | return 0; |
52d4b9ac KH |
2759 | if (PageCompound(page)) |
2760 | return 0; | |
accf163e | 2761 | |
73045c47 | 2762 | if (unlikely(!mm)) |
8697d331 | 2763 | mm = &init_mm; |
dc67d504 KH |
2764 | if (!page_is_file_cache(page)) |
2765 | type = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
accf163e | 2766 | |
38c5d72f | 2767 | if (!PageSwapCache(page)) |
dc67d504 | 2768 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, type); |
38c5d72f | 2769 | else { /* page is swapcache/shmem */ |
c0ff4b85 | 2770 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &memcg); |
83aae4c7 | 2771 | if (!ret) |
dc67d504 KH |
2772 | __mem_cgroup_commit_charge_swapin(page, memcg, type); |
2773 | } | |
b5a84319 | 2774 | return ret; |
e8589cc1 KH |
2775 | } |
2776 | ||
54595fe2 KH |
2777 | /* |
2778 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
2779 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 2780 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
2781 | * "commit()" or removed by "cancel()" |
2782 | */ | |
8c7c6e34 KH |
2783 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
2784 | struct page *page, | |
72835c86 | 2785 | gfp_t mask, struct mem_cgroup **memcgp) |
8c7c6e34 | 2786 | { |
c0ff4b85 | 2787 | struct mem_cgroup *memcg; |
54595fe2 | 2788 | int ret; |
8c7c6e34 | 2789 | |
72835c86 | 2790 | *memcgp = NULL; |
56039efa | 2791 | |
f8d66542 | 2792 | if (mem_cgroup_disabled()) |
8c7c6e34 KH |
2793 | return 0; |
2794 | ||
2795 | if (!do_swap_account) | |
2796 | goto charge_cur_mm; | |
8c7c6e34 KH |
2797 | /* |
2798 | * A racing thread's fault, or swapoff, may have already updated | |
407f9c8b HD |
2799 | * the pte, and even removed page from swap cache: in those cases |
2800 | * do_swap_page()'s pte_same() test will fail; but there's also a | |
2801 | * KSM case which does need to charge the page. | |
8c7c6e34 KH |
2802 | */ |
2803 | if (!PageSwapCache(page)) | |
407f9c8b | 2804 | goto charge_cur_mm; |
c0ff4b85 R |
2805 | memcg = try_get_mem_cgroup_from_page(page); |
2806 | if (!memcg) | |
54595fe2 | 2807 | goto charge_cur_mm; |
72835c86 JW |
2808 | *memcgp = memcg; |
2809 | ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true); | |
c0ff4b85 | 2810 | css_put(&memcg->css); |
38c5d72f KH |
2811 | if (ret == -EINTR) |
2812 | ret = 0; | |
54595fe2 | 2813 | return ret; |
8c7c6e34 KH |
2814 | charge_cur_mm: |
2815 | if (unlikely(!mm)) | |
2816 | mm = &init_mm; | |
38c5d72f KH |
2817 | ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true); |
2818 | if (ret == -EINTR) | |
2819 | ret = 0; | |
2820 | return ret; | |
8c7c6e34 KH |
2821 | } |
2822 | ||
83aae4c7 | 2823 | static void |
72835c86 | 2824 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg, |
83aae4c7 | 2825 | enum charge_type ctype) |
7a81b88c | 2826 | { |
f8d66542 | 2827 | if (mem_cgroup_disabled()) |
7a81b88c | 2828 | return; |
72835c86 | 2829 | if (!memcg) |
7a81b88c | 2830 | return; |
72835c86 | 2831 | cgroup_exclude_rmdir(&memcg->css); |
5a6475a4 | 2832 | |
ce587e65 | 2833 | __mem_cgroup_commit_charge(memcg, page, 1, ctype, true); |
8c7c6e34 KH |
2834 | /* |
2835 | * Now swap is on-memory. This means this page may be | |
2836 | * counted both as mem and swap....double count. | |
03f3c433 KH |
2837 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
2838 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
2839 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 2840 | */ |
03f3c433 | 2841 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 2842 | swp_entry_t ent = {.val = page_private(page)}; |
86493009 | 2843 | mem_cgroup_uncharge_swap(ent); |
8c7c6e34 | 2844 | } |
88703267 KH |
2845 | /* |
2846 | * At swapin, we may charge account against cgroup which has no tasks. | |
2847 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
2848 | * In that case, we need to call pre_destroy() again. check it here. | |
2849 | */ | |
72835c86 | 2850 | cgroup_release_and_wakeup_rmdir(&memcg->css); |
7a81b88c KH |
2851 | } |
2852 | ||
72835c86 JW |
2853 | void mem_cgroup_commit_charge_swapin(struct page *page, |
2854 | struct mem_cgroup *memcg) | |
83aae4c7 | 2855 | { |
72835c86 JW |
2856 | __mem_cgroup_commit_charge_swapin(page, memcg, |
2857 | MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
83aae4c7 DN |
2858 | } |
2859 | ||
c0ff4b85 | 2860 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg) |
7a81b88c | 2861 | { |
f8d66542 | 2862 | if (mem_cgroup_disabled()) |
7a81b88c | 2863 | return; |
c0ff4b85 | 2864 | if (!memcg) |
7a81b88c | 2865 | return; |
c0ff4b85 | 2866 | __mem_cgroup_cancel_charge(memcg, 1); |
7a81b88c KH |
2867 | } |
2868 | ||
c0ff4b85 | 2869 | static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg, |
7ec99d62 JW |
2870 | unsigned int nr_pages, |
2871 | const enum charge_type ctype) | |
569b846d KH |
2872 | { |
2873 | struct memcg_batch_info *batch = NULL; | |
2874 | bool uncharge_memsw = true; | |
7ec99d62 | 2875 | |
569b846d KH |
2876 | /* If swapout, usage of swap doesn't decrease */ |
2877 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2878 | uncharge_memsw = false; | |
569b846d KH |
2879 | |
2880 | batch = ¤t->memcg_batch; | |
2881 | /* | |
2882 | * In usual, we do css_get() when we remember memcg pointer. | |
2883 | * But in this case, we keep res->usage until end of a series of | |
2884 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
2885 | */ | |
2886 | if (!batch->memcg) | |
c0ff4b85 | 2887 | batch->memcg = memcg; |
3c11ecf4 KH |
2888 | /* |
2889 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
25985edc | 2890 | * In those cases, all pages freed continuously can be expected to be in |
3c11ecf4 KH |
2891 | * the same cgroup and we have chance to coalesce uncharges. |
2892 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
2893 | * because we want to do uncharge as soon as possible. | |
2894 | */ | |
2895 | ||
2896 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
2897 | goto direct_uncharge; | |
2898 | ||
7ec99d62 | 2899 | if (nr_pages > 1) |
ec168510 AA |
2900 | goto direct_uncharge; |
2901 | ||
569b846d KH |
2902 | /* |
2903 | * In typical case, batch->memcg == mem. This means we can | |
2904 | * merge a series of uncharges to an uncharge of res_counter. | |
2905 | * If not, we uncharge res_counter ony by one. | |
2906 | */ | |
c0ff4b85 | 2907 | if (batch->memcg != memcg) |
569b846d KH |
2908 | goto direct_uncharge; |
2909 | /* remember freed charge and uncharge it later */ | |
7ffd4ca7 | 2910 | batch->nr_pages++; |
569b846d | 2911 | if (uncharge_memsw) |
7ffd4ca7 | 2912 | batch->memsw_nr_pages++; |
569b846d KH |
2913 | return; |
2914 | direct_uncharge: | |
c0ff4b85 | 2915 | res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE); |
569b846d | 2916 | if (uncharge_memsw) |
c0ff4b85 R |
2917 | res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE); |
2918 | if (unlikely(batch->memcg != memcg)) | |
2919 | memcg_oom_recover(memcg); | |
569b846d | 2920 | } |
7a81b88c | 2921 | |
8a9f3ccd | 2922 | /* |
69029cd5 | 2923 | * uncharge if !page_mapped(page) |
8a9f3ccd | 2924 | */ |
8c7c6e34 | 2925 | static struct mem_cgroup * |
69029cd5 | 2926 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
8a9f3ccd | 2927 | { |
c0ff4b85 | 2928 | struct mem_cgroup *memcg = NULL; |
7ec99d62 JW |
2929 | unsigned int nr_pages = 1; |
2930 | struct page_cgroup *pc; | |
b2402857 | 2931 | bool anon; |
8a9f3ccd | 2932 | |
f8d66542 | 2933 | if (mem_cgroup_disabled()) |
8c7c6e34 | 2934 | return NULL; |
4077960e | 2935 | |
d13d1443 | 2936 | if (PageSwapCache(page)) |
8c7c6e34 | 2937 | return NULL; |
d13d1443 | 2938 | |
37c2ac78 | 2939 | if (PageTransHuge(page)) { |
7ec99d62 | 2940 | nr_pages <<= compound_order(page); |
37c2ac78 AA |
2941 | VM_BUG_ON(!PageTransHuge(page)); |
2942 | } | |
8697d331 | 2943 | /* |
3c541e14 | 2944 | * Check if our page_cgroup is valid |
8697d331 | 2945 | */ |
52d4b9ac | 2946 | pc = lookup_page_cgroup(page); |
cfa44946 | 2947 | if (unlikely(!PageCgroupUsed(pc))) |
8c7c6e34 | 2948 | return NULL; |
b9c565d5 | 2949 | |
52d4b9ac | 2950 | lock_page_cgroup(pc); |
d13d1443 | 2951 | |
c0ff4b85 | 2952 | memcg = pc->mem_cgroup; |
8c7c6e34 | 2953 | |
d13d1443 KH |
2954 | if (!PageCgroupUsed(pc)) |
2955 | goto unlock_out; | |
2956 | ||
b2402857 KH |
2957 | anon = PageAnon(page); |
2958 | ||
d13d1443 KH |
2959 | switch (ctype) { |
2960 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
2ff76f11 KH |
2961 | /* |
2962 | * Generally PageAnon tells if it's the anon statistics to be | |
2963 | * updated; but sometimes e.g. mem_cgroup_uncharge_page() is | |
2964 | * used before page reached the stage of being marked PageAnon. | |
2965 | */ | |
b2402857 KH |
2966 | anon = true; |
2967 | /* fallthrough */ | |
8a9478ca | 2968 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c | 2969 | /* See mem_cgroup_prepare_migration() */ |
2970 | if (page_mapped(page) || PageCgroupMigration(pc)) | |
d13d1443 KH |
2971 | goto unlock_out; |
2972 | break; | |
2973 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
2974 | if (!PageAnon(page)) { /* Shared memory */ | |
2975 | if (page->mapping && !page_is_file_cache(page)) | |
2976 | goto unlock_out; | |
2977 | } else if (page_mapped(page)) /* Anon */ | |
2978 | goto unlock_out; | |
2979 | break; | |
2980 | default: | |
2981 | break; | |
52d4b9ac | 2982 | } |
d13d1443 | 2983 | |
b2402857 | 2984 | mem_cgroup_charge_statistics(memcg, anon, -nr_pages); |
04046e1a | 2985 | |
52d4b9ac | 2986 | ClearPageCgroupUsed(pc); |
544122e5 KH |
2987 | /* |
2988 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
2989 | * freed from LRU. This is safe because uncharged page is expected not | |
2990 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
2991 | * special functions. | |
2992 | */ | |
b9c565d5 | 2993 | |
52d4b9ac | 2994 | unlock_page_cgroup(pc); |
f75ca962 | 2995 | /* |
c0ff4b85 | 2996 | * even after unlock, we have memcg->res.usage here and this memcg |
f75ca962 KH |
2997 | * will never be freed. |
2998 | */ | |
c0ff4b85 | 2999 | memcg_check_events(memcg, page); |
f75ca962 | 3000 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
c0ff4b85 R |
3001 | mem_cgroup_swap_statistics(memcg, true); |
3002 | mem_cgroup_get(memcg); | |
f75ca962 | 3003 | } |
c0ff4b85 R |
3004 | if (!mem_cgroup_is_root(memcg)) |
3005 | mem_cgroup_do_uncharge(memcg, nr_pages, ctype); | |
6d12e2d8 | 3006 | |
c0ff4b85 | 3007 | return memcg; |
d13d1443 KH |
3008 | |
3009 | unlock_out: | |
3010 | unlock_page_cgroup(pc); | |
8c7c6e34 | 3011 | return NULL; |
3c541e14 BS |
3012 | } |
3013 | ||
69029cd5 KH |
3014 | void mem_cgroup_uncharge_page(struct page *page) |
3015 | { | |
52d4b9ac KH |
3016 | /* early check. */ |
3017 | if (page_mapped(page)) | |
3018 | return; | |
40f23a21 | 3019 | VM_BUG_ON(page->mapping && !PageAnon(page)); |
69029cd5 KH |
3020 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
3021 | } | |
3022 | ||
3023 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
3024 | { | |
3025 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 3026 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
3027 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
3028 | } | |
3029 | ||
569b846d KH |
3030 | /* |
3031 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
3032 | * In that cases, pages are freed continuously and we can expect pages | |
3033 | * are in the same memcg. All these calls itself limits the number of | |
3034 | * pages freed at once, then uncharge_start/end() is called properly. | |
3035 | * This may be called prural(2) times in a context, | |
3036 | */ | |
3037 | ||
3038 | void mem_cgroup_uncharge_start(void) | |
3039 | { | |
3040 | current->memcg_batch.do_batch++; | |
3041 | /* We can do nest. */ | |
3042 | if (current->memcg_batch.do_batch == 1) { | |
3043 | current->memcg_batch.memcg = NULL; | |
7ffd4ca7 JW |
3044 | current->memcg_batch.nr_pages = 0; |
3045 | current->memcg_batch.memsw_nr_pages = 0; | |
569b846d KH |
3046 | } |
3047 | } | |
3048 | ||
3049 | void mem_cgroup_uncharge_end(void) | |
3050 | { | |
3051 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
3052 | ||
3053 | if (!batch->do_batch) | |
3054 | return; | |
3055 | ||
3056 | batch->do_batch--; | |
3057 | if (batch->do_batch) /* If stacked, do nothing. */ | |
3058 | return; | |
3059 | ||
3060 | if (!batch->memcg) | |
3061 | return; | |
3062 | /* | |
3063 | * This "batch->memcg" is valid without any css_get/put etc... | |
3064 | * bacause we hide charges behind us. | |
3065 | */ | |
7ffd4ca7 JW |
3066 | if (batch->nr_pages) |
3067 | res_counter_uncharge(&batch->memcg->res, | |
3068 | batch->nr_pages * PAGE_SIZE); | |
3069 | if (batch->memsw_nr_pages) | |
3070 | res_counter_uncharge(&batch->memcg->memsw, | |
3071 | batch->memsw_nr_pages * PAGE_SIZE); | |
3c11ecf4 | 3072 | memcg_oom_recover(batch->memcg); |
569b846d KH |
3073 | /* forget this pointer (for sanity check) */ |
3074 | batch->memcg = NULL; | |
3075 | } | |
3076 | ||
e767e056 | 3077 | #ifdef CONFIG_SWAP |
8c7c6e34 | 3078 | /* |
e767e056 | 3079 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
3080 | * memcg information is recorded to swap_cgroup of "ent" |
3081 | */ | |
8a9478ca KH |
3082 | void |
3083 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
3084 | { |
3085 | struct mem_cgroup *memcg; | |
8a9478ca KH |
3086 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
3087 | ||
3088 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
3089 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
3090 | ||
3091 | memcg = __mem_cgroup_uncharge_common(page, ctype); | |
8c7c6e34 | 3092 | |
f75ca962 KH |
3093 | /* |
3094 | * record memcg information, if swapout && memcg != NULL, | |
3095 | * mem_cgroup_get() was called in uncharge(). | |
3096 | */ | |
3097 | if (do_swap_account && swapout && memcg) | |
a3b2d692 | 3098 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 | 3099 | } |
e767e056 | 3100 | #endif |
8c7c6e34 KH |
3101 | |
3102 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
3103 | /* | |
3104 | * called from swap_entry_free(). remove record in swap_cgroup and | |
3105 | * uncharge "memsw" account. | |
3106 | */ | |
3107 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 3108 | { |
8c7c6e34 | 3109 | struct mem_cgroup *memcg; |
a3b2d692 | 3110 | unsigned short id; |
8c7c6e34 KH |
3111 | |
3112 | if (!do_swap_account) | |
3113 | return; | |
3114 | ||
a3b2d692 KH |
3115 | id = swap_cgroup_record(ent, 0); |
3116 | rcu_read_lock(); | |
3117 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 3118 | if (memcg) { |
a3b2d692 KH |
3119 | /* |
3120 | * We uncharge this because swap is freed. | |
3121 | * This memcg can be obsolete one. We avoid calling css_tryget | |
3122 | */ | |
0c3e73e8 | 3123 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 3124 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 3125 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
3126 | mem_cgroup_put(memcg); |
3127 | } | |
a3b2d692 | 3128 | rcu_read_unlock(); |
d13d1443 | 3129 | } |
02491447 DN |
3130 | |
3131 | /** | |
3132 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3133 | * @entry: swap entry to be moved | |
3134 | * @from: mem_cgroup which the entry is moved from | |
3135 | * @to: mem_cgroup which the entry is moved to | |
3136 | * | |
3137 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3138 | * as the mem_cgroup's id of @from. | |
3139 | * | |
3140 | * Returns 0 on success, -EINVAL on failure. | |
3141 | * | |
3142 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
3143 | * both res and memsw, and called css_get(). | |
3144 | */ | |
3145 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3146 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3147 | { |
3148 | unsigned short old_id, new_id; | |
3149 | ||
3150 | old_id = css_id(&from->css); | |
3151 | new_id = css_id(&to->css); | |
3152 | ||
3153 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 3154 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 3155 | mem_cgroup_swap_statistics(to, true); |
02491447 | 3156 | /* |
483c30b5 DN |
3157 | * This function is only called from task migration context now. |
3158 | * It postpones res_counter and refcount handling till the end | |
3159 | * of task migration(mem_cgroup_clear_mc()) for performance | |
3160 | * improvement. But we cannot postpone mem_cgroup_get(to) | |
3161 | * because if the process that has been moved to @to does | |
3162 | * swap-in, the refcount of @to might be decreased to 0. | |
02491447 | 3163 | */ |
02491447 | 3164 | mem_cgroup_get(to); |
02491447 DN |
3165 | return 0; |
3166 | } | |
3167 | return -EINVAL; | |
3168 | } | |
3169 | #else | |
3170 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3171 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3172 | { |
3173 | return -EINVAL; | |
3174 | } | |
8c7c6e34 | 3175 | #endif |
d13d1443 | 3176 | |
ae41be37 | 3177 | /* |
01b1ae63 KH |
3178 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
3179 | * page belongs to. | |
ae41be37 | 3180 | */ |
ac39cf8c | 3181 | int mem_cgroup_prepare_migration(struct page *page, |
72835c86 | 3182 | struct page *newpage, struct mem_cgroup **memcgp, gfp_t gfp_mask) |
ae41be37 | 3183 | { |
c0ff4b85 | 3184 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 3185 | struct page_cgroup *pc; |
ac39cf8c | 3186 | enum charge_type ctype; |
e8589cc1 | 3187 | int ret = 0; |
8869b8f6 | 3188 | |
72835c86 | 3189 | *memcgp = NULL; |
56039efa | 3190 | |
ec168510 | 3191 | VM_BUG_ON(PageTransHuge(page)); |
f8d66542 | 3192 | if (mem_cgroup_disabled()) |
4077960e BS |
3193 | return 0; |
3194 | ||
52d4b9ac KH |
3195 | pc = lookup_page_cgroup(page); |
3196 | lock_page_cgroup(pc); | |
3197 | if (PageCgroupUsed(pc)) { | |
c0ff4b85 R |
3198 | memcg = pc->mem_cgroup; |
3199 | css_get(&memcg->css); | |
ac39cf8c | 3200 | /* |
3201 | * At migrating an anonymous page, its mapcount goes down | |
3202 | * to 0 and uncharge() will be called. But, even if it's fully | |
3203 | * unmapped, migration may fail and this page has to be | |
3204 | * charged again. We set MIGRATION flag here and delay uncharge | |
3205 | * until end_migration() is called | |
3206 | * | |
3207 | * Corner Case Thinking | |
3208 | * A) | |
3209 | * When the old page was mapped as Anon and it's unmap-and-freed | |
3210 | * while migration was ongoing. | |
3211 | * If unmap finds the old page, uncharge() of it will be delayed | |
3212 | * until end_migration(). If unmap finds a new page, it's | |
3213 | * uncharged when it make mapcount to be 1->0. If unmap code | |
3214 | * finds swap_migration_entry, the new page will not be mapped | |
3215 | * and end_migration() will find it(mapcount==0). | |
3216 | * | |
3217 | * B) | |
3218 | * When the old page was mapped but migraion fails, the kernel | |
3219 | * remaps it. A charge for it is kept by MIGRATION flag even | |
3220 | * if mapcount goes down to 0. We can do remap successfully | |
3221 | * without charging it again. | |
3222 | * | |
3223 | * C) | |
3224 | * The "old" page is under lock_page() until the end of | |
3225 | * migration, so, the old page itself will not be swapped-out. | |
3226 | * If the new page is swapped out before end_migraton, our | |
3227 | * hook to usual swap-out path will catch the event. | |
3228 | */ | |
3229 | if (PageAnon(page)) | |
3230 | SetPageCgroupMigration(pc); | |
e8589cc1 | 3231 | } |
52d4b9ac | 3232 | unlock_page_cgroup(pc); |
ac39cf8c | 3233 | /* |
3234 | * If the page is not charged at this point, | |
3235 | * we return here. | |
3236 | */ | |
c0ff4b85 | 3237 | if (!memcg) |
ac39cf8c | 3238 | return 0; |
01b1ae63 | 3239 | |
72835c86 JW |
3240 | *memcgp = memcg; |
3241 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, 1, memcgp, false); | |
c0ff4b85 | 3242 | css_put(&memcg->css);/* drop extra refcnt */ |
38c5d72f | 3243 | if (ret) { |
ac39cf8c | 3244 | if (PageAnon(page)) { |
3245 | lock_page_cgroup(pc); | |
3246 | ClearPageCgroupMigration(pc); | |
3247 | unlock_page_cgroup(pc); | |
3248 | /* | |
3249 | * The old page may be fully unmapped while we kept it. | |
3250 | */ | |
3251 | mem_cgroup_uncharge_page(page); | |
3252 | } | |
38c5d72f | 3253 | /* we'll need to revisit this error code (we have -EINTR) */ |
ac39cf8c | 3254 | return -ENOMEM; |
e8589cc1 | 3255 | } |
ac39cf8c | 3256 | /* |
3257 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
3258 | * is called before end_migration, we can catch all events on this new | |
3259 | * page. In the case new page is migrated but not remapped, new page's | |
3260 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
3261 | */ | |
ac39cf8c | 3262 | if (PageAnon(page)) |
3263 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
3264 | else if (page_is_file_cache(page)) | |
3265 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
3266 | else | |
3267 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
ce587e65 | 3268 | __mem_cgroup_commit_charge(memcg, newpage, 1, ctype, false); |
e8589cc1 | 3269 | return ret; |
ae41be37 | 3270 | } |
8869b8f6 | 3271 | |
69029cd5 | 3272 | /* remove redundant charge if migration failed*/ |
c0ff4b85 | 3273 | void mem_cgroup_end_migration(struct mem_cgroup *memcg, |
50de1dd9 | 3274 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 3275 | { |
ac39cf8c | 3276 | struct page *used, *unused; |
01b1ae63 | 3277 | struct page_cgroup *pc; |
b2402857 | 3278 | bool anon; |
01b1ae63 | 3279 | |
c0ff4b85 | 3280 | if (!memcg) |
01b1ae63 | 3281 | return; |
ac39cf8c | 3282 | /* blocks rmdir() */ |
c0ff4b85 | 3283 | cgroup_exclude_rmdir(&memcg->css); |
50de1dd9 | 3284 | if (!migration_ok) { |
ac39cf8c | 3285 | used = oldpage; |
3286 | unused = newpage; | |
01b1ae63 | 3287 | } else { |
ac39cf8c | 3288 | used = newpage; |
01b1ae63 KH |
3289 | unused = oldpage; |
3290 | } | |
69029cd5 | 3291 | /* |
ac39cf8c | 3292 | * We disallowed uncharge of pages under migration because mapcount |
3293 | * of the page goes down to zero, temporarly. | |
3294 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 3295 | */ |
ac39cf8c | 3296 | pc = lookup_page_cgroup(oldpage); |
3297 | lock_page_cgroup(pc); | |
3298 | ClearPageCgroupMigration(pc); | |
3299 | unlock_page_cgroup(pc); | |
b2402857 KH |
3300 | anon = PageAnon(used); |
3301 | __mem_cgroup_uncharge_common(unused, | |
3302 | anon ? MEM_CGROUP_CHARGE_TYPE_MAPPED | |
3303 | : MEM_CGROUP_CHARGE_TYPE_CACHE); | |
ac39cf8c | 3304 | |
01b1ae63 | 3305 | /* |
ac39cf8c | 3306 | * If a page is a file cache, radix-tree replacement is very atomic |
3307 | * and we can skip this check. When it was an Anon page, its mapcount | |
3308 | * goes down to 0. But because we added MIGRATION flage, it's not | |
3309 | * uncharged yet. There are several case but page->mapcount check | |
3310 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
3311 | * check. (see prepare_charge() also) | |
69029cd5 | 3312 | */ |
b2402857 | 3313 | if (anon) |
ac39cf8c | 3314 | mem_cgroup_uncharge_page(used); |
88703267 | 3315 | /* |
ac39cf8c | 3316 | * At migration, we may charge account against cgroup which has no |
3317 | * tasks. | |
88703267 KH |
3318 | * So, rmdir()->pre_destroy() can be called while we do this charge. |
3319 | * In that case, we need to call pre_destroy() again. check it here. | |
3320 | */ | |
c0ff4b85 | 3321 | cgroup_release_and_wakeup_rmdir(&memcg->css); |
ae41be37 | 3322 | } |
78fb7466 | 3323 | |
ab936cbc KH |
3324 | /* |
3325 | * At replace page cache, newpage is not under any memcg but it's on | |
3326 | * LRU. So, this function doesn't touch res_counter but handles LRU | |
3327 | * in correct way. Both pages are locked so we cannot race with uncharge. | |
3328 | */ | |
3329 | void mem_cgroup_replace_page_cache(struct page *oldpage, | |
3330 | struct page *newpage) | |
3331 | { | |
bde05d1c | 3332 | struct mem_cgroup *memcg = NULL; |
ab936cbc | 3333 | struct page_cgroup *pc; |
ab936cbc | 3334 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
ab936cbc KH |
3335 | |
3336 | if (mem_cgroup_disabled()) | |
3337 | return; | |
3338 | ||
3339 | pc = lookup_page_cgroup(oldpage); | |
3340 | /* fix accounting on old pages */ | |
3341 | lock_page_cgroup(pc); | |
bde05d1c HD |
3342 | if (PageCgroupUsed(pc)) { |
3343 | memcg = pc->mem_cgroup; | |
3344 | mem_cgroup_charge_statistics(memcg, false, -1); | |
3345 | ClearPageCgroupUsed(pc); | |
3346 | } | |
ab936cbc KH |
3347 | unlock_page_cgroup(pc); |
3348 | ||
bde05d1c HD |
3349 | /* |
3350 | * When called from shmem_replace_page(), in some cases the | |
3351 | * oldpage has already been charged, and in some cases not. | |
3352 | */ | |
3353 | if (!memcg) | |
3354 | return; | |
3355 | ||
ab936cbc KH |
3356 | if (PageSwapBacked(oldpage)) |
3357 | type = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
3358 | ||
ab936cbc KH |
3359 | /* |
3360 | * Even if newpage->mapping was NULL before starting replacement, | |
3361 | * the newpage may be on LRU(or pagevec for LRU) already. We lock | |
3362 | * LRU while we overwrite pc->mem_cgroup. | |
3363 | */ | |
ce587e65 | 3364 | __mem_cgroup_commit_charge(memcg, newpage, 1, type, true); |
ab936cbc KH |
3365 | } |
3366 | ||
f212ad7c DN |
3367 | #ifdef CONFIG_DEBUG_VM |
3368 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
3369 | { | |
3370 | struct page_cgroup *pc; | |
3371 | ||
3372 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
3373 | /* |
3374 | * Can be NULL while feeding pages into the page allocator for | |
3375 | * the first time, i.e. during boot or memory hotplug; | |
3376 | * or when mem_cgroup_disabled(). | |
3377 | */ | |
f212ad7c DN |
3378 | if (likely(pc) && PageCgroupUsed(pc)) |
3379 | return pc; | |
3380 | return NULL; | |
3381 | } | |
3382 | ||
3383 | bool mem_cgroup_bad_page_check(struct page *page) | |
3384 | { | |
3385 | if (mem_cgroup_disabled()) | |
3386 | return false; | |
3387 | ||
3388 | return lookup_page_cgroup_used(page) != NULL; | |
3389 | } | |
3390 | ||
3391 | void mem_cgroup_print_bad_page(struct page *page) | |
3392 | { | |
3393 | struct page_cgroup *pc; | |
3394 | ||
3395 | pc = lookup_page_cgroup_used(page); | |
3396 | if (pc) { | |
90b3feae | 3397 | printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p\n", |
f212ad7c | 3398 | pc, pc->flags, pc->mem_cgroup); |
f212ad7c DN |
3399 | } |
3400 | } | |
3401 | #endif | |
3402 | ||
8c7c6e34 KH |
3403 | static DEFINE_MUTEX(set_limit_mutex); |
3404 | ||
d38d2a75 | 3405 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 3406 | unsigned long long val) |
628f4235 | 3407 | { |
81d39c20 | 3408 | int retry_count; |
3c11ecf4 | 3409 | u64 memswlimit, memlimit; |
628f4235 | 3410 | int ret = 0; |
81d39c20 KH |
3411 | int children = mem_cgroup_count_children(memcg); |
3412 | u64 curusage, oldusage; | |
3c11ecf4 | 3413 | int enlarge; |
81d39c20 KH |
3414 | |
3415 | /* | |
3416 | * For keeping hierarchical_reclaim simple, how long we should retry | |
3417 | * is depends on callers. We set our retry-count to be function | |
3418 | * of # of children which we should visit in this loop. | |
3419 | */ | |
3420 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
3421 | ||
3422 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 3423 | |
3c11ecf4 | 3424 | enlarge = 0; |
8c7c6e34 | 3425 | while (retry_count) { |
628f4235 KH |
3426 | if (signal_pending(current)) { |
3427 | ret = -EINTR; | |
3428 | break; | |
3429 | } | |
8c7c6e34 KH |
3430 | /* |
3431 | * Rather than hide all in some function, I do this in | |
3432 | * open coded manner. You see what this really does. | |
c0ff4b85 | 3433 | * We have to guarantee memcg->res.limit < memcg->memsw.limit. |
8c7c6e34 KH |
3434 | */ |
3435 | mutex_lock(&set_limit_mutex); | |
3436 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3437 | if (memswlimit < val) { | |
3438 | ret = -EINVAL; | |
3439 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
3440 | break; |
3441 | } | |
3c11ecf4 KH |
3442 | |
3443 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3444 | if (memlimit < val) | |
3445 | enlarge = 1; | |
3446 | ||
8c7c6e34 | 3447 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
3448 | if (!ret) { |
3449 | if (memswlimit == val) | |
3450 | memcg->memsw_is_minimum = true; | |
3451 | else | |
3452 | memcg->memsw_is_minimum = false; | |
3453 | } | |
8c7c6e34 KH |
3454 | mutex_unlock(&set_limit_mutex); |
3455 | ||
3456 | if (!ret) | |
3457 | break; | |
3458 | ||
5660048c JW |
3459 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
3460 | MEM_CGROUP_RECLAIM_SHRINK); | |
81d39c20 KH |
3461 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
3462 | /* Usage is reduced ? */ | |
3463 | if (curusage >= oldusage) | |
3464 | retry_count--; | |
3465 | else | |
3466 | oldusage = curusage; | |
8c7c6e34 | 3467 | } |
3c11ecf4 KH |
3468 | if (!ret && enlarge) |
3469 | memcg_oom_recover(memcg); | |
14797e23 | 3470 | |
8c7c6e34 KH |
3471 | return ret; |
3472 | } | |
3473 | ||
338c8431 LZ |
3474 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3475 | unsigned long long val) | |
8c7c6e34 | 3476 | { |
81d39c20 | 3477 | int retry_count; |
3c11ecf4 | 3478 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
3479 | int children = mem_cgroup_count_children(memcg); |
3480 | int ret = -EBUSY; | |
3c11ecf4 | 3481 | int enlarge = 0; |
8c7c6e34 | 3482 | |
81d39c20 KH |
3483 | /* see mem_cgroup_resize_res_limit */ |
3484 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
3485 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
3486 | while (retry_count) { |
3487 | if (signal_pending(current)) { | |
3488 | ret = -EINTR; | |
3489 | break; | |
3490 | } | |
3491 | /* | |
3492 | * Rather than hide all in some function, I do this in | |
3493 | * open coded manner. You see what this really does. | |
c0ff4b85 | 3494 | * We have to guarantee memcg->res.limit < memcg->memsw.limit. |
8c7c6e34 KH |
3495 | */ |
3496 | mutex_lock(&set_limit_mutex); | |
3497 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3498 | if (memlimit > val) { | |
3499 | ret = -EINVAL; | |
3500 | mutex_unlock(&set_limit_mutex); | |
3501 | break; | |
3502 | } | |
3c11ecf4 KH |
3503 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
3504 | if (memswlimit < val) | |
3505 | enlarge = 1; | |
8c7c6e34 | 3506 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
3507 | if (!ret) { |
3508 | if (memlimit == val) | |
3509 | memcg->memsw_is_minimum = true; | |
3510 | else | |
3511 | memcg->memsw_is_minimum = false; | |
3512 | } | |
8c7c6e34 KH |
3513 | mutex_unlock(&set_limit_mutex); |
3514 | ||
3515 | if (!ret) | |
3516 | break; | |
3517 | ||
5660048c JW |
3518 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
3519 | MEM_CGROUP_RECLAIM_NOSWAP | | |
3520 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 3521 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 3522 | /* Usage is reduced ? */ |
8c7c6e34 | 3523 | if (curusage >= oldusage) |
628f4235 | 3524 | retry_count--; |
81d39c20 KH |
3525 | else |
3526 | oldusage = curusage; | |
628f4235 | 3527 | } |
3c11ecf4 KH |
3528 | if (!ret && enlarge) |
3529 | memcg_oom_recover(memcg); | |
628f4235 KH |
3530 | return ret; |
3531 | } | |
3532 | ||
4e416953 | 3533 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
0ae5e89c YH |
3534 | gfp_t gfp_mask, |
3535 | unsigned long *total_scanned) | |
4e416953 BS |
3536 | { |
3537 | unsigned long nr_reclaimed = 0; | |
3538 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3539 | unsigned long reclaimed; | |
3540 | int loop = 0; | |
3541 | struct mem_cgroup_tree_per_zone *mctz; | |
ef8745c1 | 3542 | unsigned long long excess; |
0ae5e89c | 3543 | unsigned long nr_scanned; |
4e416953 BS |
3544 | |
3545 | if (order > 0) | |
3546 | return 0; | |
3547 | ||
00918b6a | 3548 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); |
4e416953 BS |
3549 | /* |
3550 | * This loop can run a while, specially if mem_cgroup's continuously | |
3551 | * keep exceeding their soft limit and putting the system under | |
3552 | * pressure | |
3553 | */ | |
3554 | do { | |
3555 | if (next_mz) | |
3556 | mz = next_mz; | |
3557 | else | |
3558 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3559 | if (!mz) | |
3560 | break; | |
3561 | ||
0ae5e89c | 3562 | nr_scanned = 0; |
d79154bb | 3563 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, |
5660048c | 3564 | gfp_mask, &nr_scanned); |
4e416953 | 3565 | nr_reclaimed += reclaimed; |
0ae5e89c | 3566 | *total_scanned += nr_scanned; |
4e416953 BS |
3567 | spin_lock(&mctz->lock); |
3568 | ||
3569 | /* | |
3570 | * If we failed to reclaim anything from this memory cgroup | |
3571 | * it is time to move on to the next cgroup | |
3572 | */ | |
3573 | next_mz = NULL; | |
3574 | if (!reclaimed) { | |
3575 | do { | |
3576 | /* | |
3577 | * Loop until we find yet another one. | |
3578 | * | |
3579 | * By the time we get the soft_limit lock | |
3580 | * again, someone might have aded the | |
3581 | * group back on the RB tree. Iterate to | |
3582 | * make sure we get a different mem. | |
3583 | * mem_cgroup_largest_soft_limit_node returns | |
3584 | * NULL if no other cgroup is present on | |
3585 | * the tree | |
3586 | */ | |
3587 | next_mz = | |
3588 | __mem_cgroup_largest_soft_limit_node(mctz); | |
39cc98f1 | 3589 | if (next_mz == mz) |
d79154bb | 3590 | css_put(&next_mz->memcg->css); |
39cc98f1 | 3591 | else /* next_mz == NULL or other memcg */ |
4e416953 BS |
3592 | break; |
3593 | } while (1); | |
3594 | } | |
d79154bb HD |
3595 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); |
3596 | excess = res_counter_soft_limit_excess(&mz->memcg->res); | |
4e416953 BS |
3597 | /* |
3598 | * One school of thought says that we should not add | |
3599 | * back the node to the tree if reclaim returns 0. | |
3600 | * But our reclaim could return 0, simply because due | |
3601 | * to priority we are exposing a smaller subset of | |
3602 | * memory to reclaim from. Consider this as a longer | |
3603 | * term TODO. | |
3604 | */ | |
ef8745c1 | 3605 | /* If excess == 0, no tree ops */ |
d79154bb | 3606 | __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess); |
4e416953 | 3607 | spin_unlock(&mctz->lock); |
d79154bb | 3608 | css_put(&mz->memcg->css); |
4e416953 BS |
3609 | loop++; |
3610 | /* | |
3611 | * Could not reclaim anything and there are no more | |
3612 | * mem cgroups to try or we seem to be looping without | |
3613 | * reclaiming anything. | |
3614 | */ | |
3615 | if (!nr_reclaimed && | |
3616 | (next_mz == NULL || | |
3617 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3618 | break; | |
3619 | } while (!nr_reclaimed); | |
3620 | if (next_mz) | |
d79154bb | 3621 | css_put(&next_mz->memcg->css); |
4e416953 BS |
3622 | return nr_reclaimed; |
3623 | } | |
3624 | ||
cc847582 KH |
3625 | /* |
3626 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
3627 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
3628 | */ | |
c0ff4b85 | 3629 | static int mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
08e552c6 | 3630 | int node, int zid, enum lru_list lru) |
cc847582 | 3631 | { |
08e552c6 | 3632 | struct mem_cgroup_per_zone *mz; |
08e552c6 | 3633 | unsigned long flags, loop; |
072c56c1 | 3634 | struct list_head *list; |
925b7673 JW |
3635 | struct page *busy; |
3636 | struct zone *zone; | |
f817ed48 | 3637 | int ret = 0; |
072c56c1 | 3638 | |
08e552c6 | 3639 | zone = &NODE_DATA(node)->node_zones[zid]; |
c0ff4b85 | 3640 | mz = mem_cgroup_zoneinfo(memcg, node, zid); |
6290df54 | 3641 | list = &mz->lruvec.lists[lru]; |
cc847582 | 3642 | |
1eb49272 | 3643 | loop = mz->lru_size[lru]; |
f817ed48 KH |
3644 | /* give some margin against EBUSY etc...*/ |
3645 | loop += 256; | |
3646 | busy = NULL; | |
3647 | while (loop--) { | |
925b7673 | 3648 | struct page_cgroup *pc; |
5564e88b JW |
3649 | struct page *page; |
3650 | ||
f817ed48 | 3651 | ret = 0; |
08e552c6 | 3652 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 3653 | if (list_empty(list)) { |
08e552c6 | 3654 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 3655 | break; |
f817ed48 | 3656 | } |
925b7673 JW |
3657 | page = list_entry(list->prev, struct page, lru); |
3658 | if (busy == page) { | |
3659 | list_move(&page->lru, list); | |
648bcc77 | 3660 | busy = NULL; |
08e552c6 | 3661 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
3662 | continue; |
3663 | } | |
08e552c6 | 3664 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 3665 | |
925b7673 | 3666 | pc = lookup_page_cgroup(page); |
5564e88b | 3667 | |
c0ff4b85 | 3668 | ret = mem_cgroup_move_parent(page, pc, memcg, GFP_KERNEL); |
38c5d72f | 3669 | if (ret == -ENOMEM || ret == -EINTR) |
52d4b9ac | 3670 | break; |
f817ed48 KH |
3671 | |
3672 | if (ret == -EBUSY || ret == -EINVAL) { | |
3673 | /* found lock contention or "pc" is obsolete. */ | |
925b7673 | 3674 | busy = page; |
f817ed48 KH |
3675 | cond_resched(); |
3676 | } else | |
3677 | busy = NULL; | |
cc847582 | 3678 | } |
08e552c6 | 3679 | |
f817ed48 KH |
3680 | if (!ret && !list_empty(list)) |
3681 | return -EBUSY; | |
3682 | return ret; | |
cc847582 KH |
3683 | } |
3684 | ||
3685 | /* | |
3686 | * make mem_cgroup's charge to be 0 if there is no task. | |
3687 | * This enables deleting this mem_cgroup. | |
3688 | */ | |
c0ff4b85 | 3689 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg, bool free_all) |
cc847582 | 3690 | { |
f817ed48 KH |
3691 | int ret; |
3692 | int node, zid, shrink; | |
3693 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c0ff4b85 | 3694 | struct cgroup *cgrp = memcg->css.cgroup; |
8869b8f6 | 3695 | |
c0ff4b85 | 3696 | css_get(&memcg->css); |
f817ed48 KH |
3697 | |
3698 | shrink = 0; | |
c1e862c1 KH |
3699 | /* should free all ? */ |
3700 | if (free_all) | |
3701 | goto try_to_free; | |
f817ed48 | 3702 | move_account: |
fce66477 | 3703 | do { |
f817ed48 | 3704 | ret = -EBUSY; |
c1e862c1 KH |
3705 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
3706 | goto out; | |
3707 | ret = -EINTR; | |
3708 | if (signal_pending(current)) | |
cc847582 | 3709 | goto out; |
52d4b9ac KH |
3710 | /* This is for making all *used* pages to be on LRU. */ |
3711 | lru_add_drain_all(); | |
c0ff4b85 | 3712 | drain_all_stock_sync(memcg); |
f817ed48 | 3713 | ret = 0; |
c0ff4b85 | 3714 | mem_cgroup_start_move(memcg); |
299b4eaa | 3715 | for_each_node_state(node, N_HIGH_MEMORY) { |
f817ed48 | 3716 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { |
f156ab93 HD |
3717 | enum lru_list lru; |
3718 | for_each_lru(lru) { | |
c0ff4b85 | 3719 | ret = mem_cgroup_force_empty_list(memcg, |
f156ab93 | 3720 | node, zid, lru); |
f817ed48 KH |
3721 | if (ret) |
3722 | break; | |
3723 | } | |
1ecaab2b | 3724 | } |
f817ed48 KH |
3725 | if (ret) |
3726 | break; | |
3727 | } | |
c0ff4b85 R |
3728 | mem_cgroup_end_move(memcg); |
3729 | memcg_oom_recover(memcg); | |
f817ed48 KH |
3730 | /* it seems parent cgroup doesn't have enough mem */ |
3731 | if (ret == -ENOMEM) | |
3732 | goto try_to_free; | |
52d4b9ac | 3733 | cond_resched(); |
fce66477 | 3734 | /* "ret" should also be checked to ensure all lists are empty. */ |
569530fb | 3735 | } while (res_counter_read_u64(&memcg->res, RES_USAGE) > 0 || ret); |
cc847582 | 3736 | out: |
c0ff4b85 | 3737 | css_put(&memcg->css); |
cc847582 | 3738 | return ret; |
f817ed48 KH |
3739 | |
3740 | try_to_free: | |
c1e862c1 KH |
3741 | /* returns EBUSY if there is a task or if we come here twice. */ |
3742 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
f817ed48 KH |
3743 | ret = -EBUSY; |
3744 | goto out; | |
3745 | } | |
c1e862c1 KH |
3746 | /* we call try-to-free pages for make this cgroup empty */ |
3747 | lru_add_drain_all(); | |
f817ed48 KH |
3748 | /* try to free all pages in this cgroup */ |
3749 | shrink = 1; | |
569530fb | 3750 | while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) { |
f817ed48 | 3751 | int progress; |
c1e862c1 KH |
3752 | |
3753 | if (signal_pending(current)) { | |
3754 | ret = -EINTR; | |
3755 | goto out; | |
3756 | } | |
c0ff4b85 | 3757 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, |
185efc0f | 3758 | false); |
c1e862c1 | 3759 | if (!progress) { |
f817ed48 | 3760 | nr_retries--; |
c1e862c1 | 3761 | /* maybe some writeback is necessary */ |
8aa7e847 | 3762 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3763 | } |
f817ed48 KH |
3764 | |
3765 | } | |
08e552c6 | 3766 | lru_add_drain(); |
f817ed48 | 3767 | /* try move_account...there may be some *locked* pages. */ |
fce66477 | 3768 | goto move_account; |
cc847582 KH |
3769 | } |
3770 | ||
6bbda35c | 3771 | static int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
c1e862c1 KH |
3772 | { |
3773 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
3774 | } | |
3775 | ||
3776 | ||
18f59ea7 BS |
3777 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
3778 | { | |
3779 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
3780 | } | |
3781 | ||
3782 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
3783 | u64 val) | |
3784 | { | |
3785 | int retval = 0; | |
c0ff4b85 | 3786 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
18f59ea7 | 3787 | struct cgroup *parent = cont->parent; |
c0ff4b85 | 3788 | struct mem_cgroup *parent_memcg = NULL; |
18f59ea7 BS |
3789 | |
3790 | if (parent) | |
c0ff4b85 | 3791 | parent_memcg = mem_cgroup_from_cont(parent); |
18f59ea7 BS |
3792 | |
3793 | cgroup_lock(); | |
3794 | /* | |
af901ca1 | 3795 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3796 | * in the child subtrees. If it is unset, then the change can |
3797 | * occur, provided the current cgroup has no children. | |
3798 | * | |
3799 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3800 | * set if there are no children. | |
3801 | */ | |
c0ff4b85 | 3802 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 BS |
3803 | (val == 1 || val == 0)) { |
3804 | if (list_empty(&cont->children)) | |
c0ff4b85 | 3805 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3806 | else |
3807 | retval = -EBUSY; | |
3808 | } else | |
3809 | retval = -EINVAL; | |
3810 | cgroup_unlock(); | |
3811 | ||
3812 | return retval; | |
3813 | } | |
3814 | ||
0c3e73e8 | 3815 | |
c0ff4b85 | 3816 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg, |
7a159cc9 | 3817 | enum mem_cgroup_stat_index idx) |
0c3e73e8 | 3818 | { |
7d74b06f | 3819 | struct mem_cgroup *iter; |
7a159cc9 | 3820 | long val = 0; |
0c3e73e8 | 3821 | |
7a159cc9 | 3822 | /* Per-cpu values can be negative, use a signed accumulator */ |
c0ff4b85 | 3823 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f KH |
3824 | val += mem_cgroup_read_stat(iter, idx); |
3825 | ||
3826 | if (val < 0) /* race ? */ | |
3827 | val = 0; | |
3828 | return val; | |
0c3e73e8 BS |
3829 | } |
3830 | ||
c0ff4b85 | 3831 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
104f3928 | 3832 | { |
7d74b06f | 3833 | u64 val; |
104f3928 | 3834 | |
c0ff4b85 | 3835 | if (!mem_cgroup_is_root(memcg)) { |
104f3928 | 3836 | if (!swap) |
65c64ce8 | 3837 | return res_counter_read_u64(&memcg->res, RES_USAGE); |
104f3928 | 3838 | else |
65c64ce8 | 3839 | return res_counter_read_u64(&memcg->memsw, RES_USAGE); |
104f3928 KS |
3840 | } |
3841 | ||
c0ff4b85 R |
3842 | val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE); |
3843 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS); | |
104f3928 | 3844 | |
7d74b06f | 3845 | if (swap) |
c0ff4b85 | 3846 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAPOUT); |
104f3928 KS |
3847 | |
3848 | return val << PAGE_SHIFT; | |
3849 | } | |
3850 | ||
af36f906 TH |
3851 | static ssize_t mem_cgroup_read(struct cgroup *cont, struct cftype *cft, |
3852 | struct file *file, char __user *buf, | |
3853 | size_t nbytes, loff_t *ppos) | |
8cdea7c0 | 3854 | { |
c0ff4b85 | 3855 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
af36f906 | 3856 | char str[64]; |
104f3928 | 3857 | u64 val; |
af36f906 | 3858 | int type, name, len; |
8c7c6e34 KH |
3859 | |
3860 | type = MEMFILE_TYPE(cft->private); | |
3861 | name = MEMFILE_ATTR(cft->private); | |
af36f906 TH |
3862 | |
3863 | if (!do_swap_account && type == _MEMSWAP) | |
3864 | return -EOPNOTSUPP; | |
3865 | ||
8c7c6e34 KH |
3866 | switch (type) { |
3867 | case _MEM: | |
104f3928 | 3868 | if (name == RES_USAGE) |
c0ff4b85 | 3869 | val = mem_cgroup_usage(memcg, false); |
104f3928 | 3870 | else |
c0ff4b85 | 3871 | val = res_counter_read_u64(&memcg->res, name); |
8c7c6e34 KH |
3872 | break; |
3873 | case _MEMSWAP: | |
104f3928 | 3874 | if (name == RES_USAGE) |
c0ff4b85 | 3875 | val = mem_cgroup_usage(memcg, true); |
104f3928 | 3876 | else |
c0ff4b85 | 3877 | val = res_counter_read_u64(&memcg->memsw, name); |
8c7c6e34 KH |
3878 | break; |
3879 | default: | |
3880 | BUG(); | |
8c7c6e34 | 3881 | } |
af36f906 TH |
3882 | |
3883 | len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val); | |
3884 | return simple_read_from_buffer(buf, nbytes, ppos, str, len); | |
8cdea7c0 | 3885 | } |
628f4235 KH |
3886 | /* |
3887 | * The user of this function is... | |
3888 | * RES_LIMIT. | |
3889 | */ | |
856c13aa PM |
3890 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
3891 | const char *buffer) | |
8cdea7c0 | 3892 | { |
628f4235 | 3893 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 3894 | int type, name; |
628f4235 KH |
3895 | unsigned long long val; |
3896 | int ret; | |
3897 | ||
8c7c6e34 KH |
3898 | type = MEMFILE_TYPE(cft->private); |
3899 | name = MEMFILE_ATTR(cft->private); | |
af36f906 TH |
3900 | |
3901 | if (!do_swap_account && type == _MEMSWAP) | |
3902 | return -EOPNOTSUPP; | |
3903 | ||
8c7c6e34 | 3904 | switch (name) { |
628f4235 | 3905 | case RES_LIMIT: |
4b3bde4c BS |
3906 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3907 | ret = -EINVAL; | |
3908 | break; | |
3909 | } | |
628f4235 KH |
3910 | /* This function does all necessary parse...reuse it */ |
3911 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
3912 | if (ret) |
3913 | break; | |
3914 | if (type == _MEM) | |
628f4235 | 3915 | ret = mem_cgroup_resize_limit(memcg, val); |
8c7c6e34 KH |
3916 | else |
3917 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
628f4235 | 3918 | break; |
296c81d8 BS |
3919 | case RES_SOFT_LIMIT: |
3920 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
3921 | if (ret) | |
3922 | break; | |
3923 | /* | |
3924 | * For memsw, soft limits are hard to implement in terms | |
3925 | * of semantics, for now, we support soft limits for | |
3926 | * control without swap | |
3927 | */ | |
3928 | if (type == _MEM) | |
3929 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
3930 | else | |
3931 | ret = -EINVAL; | |
3932 | break; | |
628f4235 KH |
3933 | default: |
3934 | ret = -EINVAL; /* should be BUG() ? */ | |
3935 | break; | |
3936 | } | |
3937 | return ret; | |
8cdea7c0 BS |
3938 | } |
3939 | ||
fee7b548 KH |
3940 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
3941 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
3942 | { | |
3943 | struct cgroup *cgroup; | |
3944 | unsigned long long min_limit, min_memsw_limit, tmp; | |
3945 | ||
3946 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3947 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3948 | cgroup = memcg->css.cgroup; | |
3949 | if (!memcg->use_hierarchy) | |
3950 | goto out; | |
3951 | ||
3952 | while (cgroup->parent) { | |
3953 | cgroup = cgroup->parent; | |
3954 | memcg = mem_cgroup_from_cont(cgroup); | |
3955 | if (!memcg->use_hierarchy) | |
3956 | break; | |
3957 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3958 | min_limit = min(min_limit, tmp); | |
3959 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3960 | min_memsw_limit = min(min_memsw_limit, tmp); | |
3961 | } | |
3962 | out: | |
3963 | *mem_limit = min_limit; | |
3964 | *memsw_limit = min_memsw_limit; | |
fee7b548 KH |
3965 | } |
3966 | ||
29f2a4da | 3967 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 | 3968 | { |
af36f906 | 3969 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 3970 | int type, name; |
c84872e1 | 3971 | |
8c7c6e34 KH |
3972 | type = MEMFILE_TYPE(event); |
3973 | name = MEMFILE_ATTR(event); | |
af36f906 TH |
3974 | |
3975 | if (!do_swap_account && type == _MEMSWAP) | |
3976 | return -EOPNOTSUPP; | |
3977 | ||
8c7c6e34 | 3978 | switch (name) { |
29f2a4da | 3979 | case RES_MAX_USAGE: |
8c7c6e34 | 3980 | if (type == _MEM) |
c0ff4b85 | 3981 | res_counter_reset_max(&memcg->res); |
8c7c6e34 | 3982 | else |
c0ff4b85 | 3983 | res_counter_reset_max(&memcg->memsw); |
29f2a4da PE |
3984 | break; |
3985 | case RES_FAILCNT: | |
8c7c6e34 | 3986 | if (type == _MEM) |
c0ff4b85 | 3987 | res_counter_reset_failcnt(&memcg->res); |
8c7c6e34 | 3988 | else |
c0ff4b85 | 3989 | res_counter_reset_failcnt(&memcg->memsw); |
29f2a4da PE |
3990 | break; |
3991 | } | |
f64c3f54 | 3992 | |
85cc59db | 3993 | return 0; |
c84872e1 PE |
3994 | } |
3995 | ||
7dc74be0 DN |
3996 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
3997 | struct cftype *cft) | |
3998 | { | |
3999 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; | |
4000 | } | |
4001 | ||
02491447 | 4002 | #ifdef CONFIG_MMU |
7dc74be0 DN |
4003 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
4004 | struct cftype *cft, u64 val) | |
4005 | { | |
c0ff4b85 | 4006 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
7dc74be0 DN |
4007 | |
4008 | if (val >= (1 << NR_MOVE_TYPE)) | |
4009 | return -EINVAL; | |
4010 | /* | |
4011 | * We check this value several times in both in can_attach() and | |
4012 | * attach(), so we need cgroup lock to prevent this value from being | |
4013 | * inconsistent. | |
4014 | */ | |
4015 | cgroup_lock(); | |
c0ff4b85 | 4016 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
4017 | cgroup_unlock(); |
4018 | ||
4019 | return 0; | |
4020 | } | |
02491447 DN |
4021 | #else |
4022 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, | |
4023 | struct cftype *cft, u64 val) | |
4024 | { | |
4025 | return -ENOSYS; | |
4026 | } | |
4027 | #endif | |
7dc74be0 | 4028 | |
14067bb3 KH |
4029 | |
4030 | /* For read statistics */ | |
4031 | enum { | |
4032 | MCS_CACHE, | |
4033 | MCS_RSS, | |
d8046582 | 4034 | MCS_FILE_MAPPED, |
14067bb3 KH |
4035 | MCS_PGPGIN, |
4036 | MCS_PGPGOUT, | |
1dd3a273 | 4037 | MCS_SWAP, |
456f998e YH |
4038 | MCS_PGFAULT, |
4039 | MCS_PGMAJFAULT, | |
14067bb3 KH |
4040 | MCS_INACTIVE_ANON, |
4041 | MCS_ACTIVE_ANON, | |
4042 | MCS_INACTIVE_FILE, | |
4043 | MCS_ACTIVE_FILE, | |
4044 | MCS_UNEVICTABLE, | |
4045 | NR_MCS_STAT, | |
4046 | }; | |
4047 | ||
4048 | struct mcs_total_stat { | |
4049 | s64 stat[NR_MCS_STAT]; | |
d2ceb9b7 KH |
4050 | }; |
4051 | ||
6bbda35c | 4052 | static struct { |
14067bb3 KH |
4053 | char *local_name; |
4054 | char *total_name; | |
4055 | } memcg_stat_strings[NR_MCS_STAT] = { | |
4056 | {"cache", "total_cache"}, | |
4057 | {"rss", "total_rss"}, | |
d69b042f | 4058 | {"mapped_file", "total_mapped_file"}, |
14067bb3 KH |
4059 | {"pgpgin", "total_pgpgin"}, |
4060 | {"pgpgout", "total_pgpgout"}, | |
1dd3a273 | 4061 | {"swap", "total_swap"}, |
456f998e YH |
4062 | {"pgfault", "total_pgfault"}, |
4063 | {"pgmajfault", "total_pgmajfault"}, | |
14067bb3 KH |
4064 | {"inactive_anon", "total_inactive_anon"}, |
4065 | {"active_anon", "total_active_anon"}, | |
4066 | {"inactive_file", "total_inactive_file"}, | |
4067 | {"active_file", "total_active_file"}, | |
4068 | {"unevictable", "total_unevictable"} | |
4069 | }; | |
4070 | ||
4071 | ||
7d74b06f | 4072 | static void |
c0ff4b85 | 4073 | mem_cgroup_get_local_stat(struct mem_cgroup *memcg, struct mcs_total_stat *s) |
14067bb3 | 4074 | { |
14067bb3 KH |
4075 | s64 val; |
4076 | ||
4077 | /* per cpu stat */ | |
c0ff4b85 | 4078 | val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_CACHE); |
14067bb3 | 4079 | s->stat[MCS_CACHE] += val * PAGE_SIZE; |
c0ff4b85 | 4080 | val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_RSS); |
14067bb3 | 4081 | s->stat[MCS_RSS] += val * PAGE_SIZE; |
c0ff4b85 | 4082 | val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED); |
d8046582 | 4083 | s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE; |
c0ff4b85 | 4084 | val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGIN); |
14067bb3 | 4085 | s->stat[MCS_PGPGIN] += val; |
c0ff4b85 | 4086 | val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGOUT); |
14067bb3 | 4087 | s->stat[MCS_PGPGOUT] += val; |
1dd3a273 | 4088 | if (do_swap_account) { |
c0ff4b85 | 4089 | val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_SWAPOUT); |
1dd3a273 DN |
4090 | s->stat[MCS_SWAP] += val * PAGE_SIZE; |
4091 | } | |
c0ff4b85 | 4092 | val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGFAULT); |
456f998e | 4093 | s->stat[MCS_PGFAULT] += val; |
c0ff4b85 | 4094 | val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGMAJFAULT); |
456f998e | 4095 | s->stat[MCS_PGMAJFAULT] += val; |
14067bb3 KH |
4096 | |
4097 | /* per zone stat */ | |
c0ff4b85 | 4098 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_ANON)); |
14067bb3 | 4099 | s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE; |
c0ff4b85 | 4100 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_ANON)); |
14067bb3 | 4101 | s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE; |
c0ff4b85 | 4102 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_FILE)); |
14067bb3 | 4103 | s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE; |
c0ff4b85 | 4104 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_FILE)); |
14067bb3 | 4105 | s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE; |
c0ff4b85 | 4106 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE)); |
14067bb3 | 4107 | s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE; |
14067bb3 KH |
4108 | } |
4109 | ||
4110 | static void | |
c0ff4b85 | 4111 | mem_cgroup_get_total_stat(struct mem_cgroup *memcg, struct mcs_total_stat *s) |
14067bb3 | 4112 | { |
7d74b06f KH |
4113 | struct mem_cgroup *iter; |
4114 | ||
c0ff4b85 | 4115 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4116 | mem_cgroup_get_local_stat(iter, s); |
14067bb3 KH |
4117 | } |
4118 | ||
406eb0c9 YH |
4119 | #ifdef CONFIG_NUMA |
4120 | static int mem_control_numa_stat_show(struct seq_file *m, void *arg) | |
4121 | { | |
4122 | int nid; | |
4123 | unsigned long total_nr, file_nr, anon_nr, unevictable_nr; | |
4124 | unsigned long node_nr; | |
4125 | struct cgroup *cont = m->private; | |
d79154bb | 4126 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
406eb0c9 | 4127 | |
d79154bb | 4128 | total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL); |
406eb0c9 YH |
4129 | seq_printf(m, "total=%lu", total_nr); |
4130 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
d79154bb | 4131 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL); |
406eb0c9 YH |
4132 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4133 | } | |
4134 | seq_putc(m, '\n'); | |
4135 | ||
d79154bb | 4136 | file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE); |
406eb0c9 YH |
4137 | seq_printf(m, "file=%lu", file_nr); |
4138 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
d79154bb | 4139 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 4140 | LRU_ALL_FILE); |
406eb0c9 YH |
4141 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4142 | } | |
4143 | seq_putc(m, '\n'); | |
4144 | ||
d79154bb | 4145 | anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON); |
406eb0c9 YH |
4146 | seq_printf(m, "anon=%lu", anon_nr); |
4147 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
d79154bb | 4148 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 4149 | LRU_ALL_ANON); |
406eb0c9 YH |
4150 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4151 | } | |
4152 | seq_putc(m, '\n'); | |
4153 | ||
d79154bb | 4154 | unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE)); |
406eb0c9 YH |
4155 | seq_printf(m, "unevictable=%lu", unevictable_nr); |
4156 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
d79154bb | 4157 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 4158 | BIT(LRU_UNEVICTABLE)); |
406eb0c9 YH |
4159 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4160 | } | |
4161 | seq_putc(m, '\n'); | |
4162 | return 0; | |
4163 | } | |
4164 | #endif /* CONFIG_NUMA */ | |
4165 | ||
c64745cf PM |
4166 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
4167 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 4168 | { |
d79154bb | 4169 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
14067bb3 | 4170 | struct mcs_total_stat mystat; |
d2ceb9b7 KH |
4171 | int i; |
4172 | ||
14067bb3 | 4173 | memset(&mystat, 0, sizeof(mystat)); |
d79154bb | 4174 | mem_cgroup_get_local_stat(memcg, &mystat); |
d2ceb9b7 | 4175 | |
406eb0c9 | 4176 | |
1dd3a273 DN |
4177 | for (i = 0; i < NR_MCS_STAT; i++) { |
4178 | if (i == MCS_SWAP && !do_swap_account) | |
4179 | continue; | |
14067bb3 | 4180 | cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]); |
1dd3a273 | 4181 | } |
7b854121 | 4182 | |
14067bb3 | 4183 | /* Hierarchical information */ |
fee7b548 KH |
4184 | { |
4185 | unsigned long long limit, memsw_limit; | |
d79154bb | 4186 | memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit); |
fee7b548 KH |
4187 | cb->fill(cb, "hierarchical_memory_limit", limit); |
4188 | if (do_swap_account) | |
4189 | cb->fill(cb, "hierarchical_memsw_limit", memsw_limit); | |
4190 | } | |
7f016ee8 | 4191 | |
14067bb3 | 4192 | memset(&mystat, 0, sizeof(mystat)); |
d79154bb | 4193 | mem_cgroup_get_total_stat(memcg, &mystat); |
1dd3a273 DN |
4194 | for (i = 0; i < NR_MCS_STAT; i++) { |
4195 | if (i == MCS_SWAP && !do_swap_account) | |
4196 | continue; | |
14067bb3 | 4197 | cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]); |
1dd3a273 | 4198 | } |
14067bb3 | 4199 | |
7f016ee8 | 4200 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
4201 | { |
4202 | int nid, zid; | |
4203 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 4204 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
4205 | unsigned long recent_rotated[2] = {0, 0}; |
4206 | unsigned long recent_scanned[2] = {0, 0}; | |
4207 | ||
4208 | for_each_online_node(nid) | |
4209 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
d79154bb | 4210 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
89abfab1 | 4211 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 4212 | |
89abfab1 HD |
4213 | recent_rotated[0] += rstat->recent_rotated[0]; |
4214 | recent_rotated[1] += rstat->recent_rotated[1]; | |
4215 | recent_scanned[0] += rstat->recent_scanned[0]; | |
4216 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 KM |
4217 | } |
4218 | cb->fill(cb, "recent_rotated_anon", recent_rotated[0]); | |
4219 | cb->fill(cb, "recent_rotated_file", recent_rotated[1]); | |
4220 | cb->fill(cb, "recent_scanned_anon", recent_scanned[0]); | |
4221 | cb->fill(cb, "recent_scanned_file", recent_scanned[1]); | |
4222 | } | |
4223 | #endif | |
4224 | ||
d2ceb9b7 KH |
4225 | return 0; |
4226 | } | |
4227 | ||
a7885eb8 KM |
4228 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
4229 | { | |
4230 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4231 | ||
1f4c025b | 4232 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4233 | } |
4234 | ||
4235 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
4236 | u64 val) | |
4237 | { | |
4238 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4239 | struct mem_cgroup *parent; | |
068b38c1 | 4240 | |
a7885eb8 KM |
4241 | if (val > 100) |
4242 | return -EINVAL; | |
4243 | ||
4244 | if (cgrp->parent == NULL) | |
4245 | return -EINVAL; | |
4246 | ||
4247 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 LZ |
4248 | |
4249 | cgroup_lock(); | |
4250 | ||
a7885eb8 KM |
4251 | /* If under hierarchy, only empty-root can set this value */ |
4252 | if ((parent->use_hierarchy) || | |
068b38c1 LZ |
4253 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
4254 | cgroup_unlock(); | |
a7885eb8 | 4255 | return -EINVAL; |
068b38c1 | 4256 | } |
a7885eb8 | 4257 | |
a7885eb8 | 4258 | memcg->swappiness = val; |
a7885eb8 | 4259 | |
068b38c1 LZ |
4260 | cgroup_unlock(); |
4261 | ||
a7885eb8 KM |
4262 | return 0; |
4263 | } | |
4264 | ||
2e72b634 KS |
4265 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4266 | { | |
4267 | struct mem_cgroup_threshold_ary *t; | |
4268 | u64 usage; | |
4269 | int i; | |
4270 | ||
4271 | rcu_read_lock(); | |
4272 | if (!swap) | |
2c488db2 | 4273 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4274 | else |
2c488db2 | 4275 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4276 | |
4277 | if (!t) | |
4278 | goto unlock; | |
4279 | ||
4280 | usage = mem_cgroup_usage(memcg, swap); | |
4281 | ||
4282 | /* | |
748dad36 | 4283 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4284 | * If it's not true, a threshold was crossed after last |
4285 | * call of __mem_cgroup_threshold(). | |
4286 | */ | |
5407a562 | 4287 | i = t->current_threshold; |
2e72b634 KS |
4288 | |
4289 | /* | |
4290 | * Iterate backward over array of thresholds starting from | |
4291 | * current_threshold and check if a threshold is crossed. | |
4292 | * If none of thresholds below usage is crossed, we read | |
4293 | * only one element of the array here. | |
4294 | */ | |
4295 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4296 | eventfd_signal(t->entries[i].eventfd, 1); | |
4297 | ||
4298 | /* i = current_threshold + 1 */ | |
4299 | i++; | |
4300 | ||
4301 | /* | |
4302 | * Iterate forward over array of thresholds starting from | |
4303 | * current_threshold+1 and check if a threshold is crossed. | |
4304 | * If none of thresholds above usage is crossed, we read | |
4305 | * only one element of the array here. | |
4306 | */ | |
4307 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4308 | eventfd_signal(t->entries[i].eventfd, 1); | |
4309 | ||
4310 | /* Update current_threshold */ | |
5407a562 | 4311 | t->current_threshold = i - 1; |
2e72b634 KS |
4312 | unlock: |
4313 | rcu_read_unlock(); | |
4314 | } | |
4315 | ||
4316 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4317 | { | |
ad4ca5f4 KS |
4318 | while (memcg) { |
4319 | __mem_cgroup_threshold(memcg, false); | |
4320 | if (do_swap_account) | |
4321 | __mem_cgroup_threshold(memcg, true); | |
4322 | ||
4323 | memcg = parent_mem_cgroup(memcg); | |
4324 | } | |
2e72b634 KS |
4325 | } |
4326 | ||
4327 | static int compare_thresholds(const void *a, const void *b) | |
4328 | { | |
4329 | const struct mem_cgroup_threshold *_a = a; | |
4330 | const struct mem_cgroup_threshold *_b = b; | |
4331 | ||
4332 | return _a->threshold - _b->threshold; | |
4333 | } | |
4334 | ||
c0ff4b85 | 4335 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4336 | { |
4337 | struct mem_cgroup_eventfd_list *ev; | |
4338 | ||
c0ff4b85 | 4339 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 KH |
4340 | eventfd_signal(ev->eventfd, 1); |
4341 | return 0; | |
4342 | } | |
4343 | ||
c0ff4b85 | 4344 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4345 | { |
7d74b06f KH |
4346 | struct mem_cgroup *iter; |
4347 | ||
c0ff4b85 | 4348 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4349 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4350 | } |
4351 | ||
4352 | static int mem_cgroup_usage_register_event(struct cgroup *cgrp, | |
4353 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
2e72b634 KS |
4354 | { |
4355 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4356 | struct mem_cgroup_thresholds *thresholds; |
4357 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
4358 | int type = MEMFILE_TYPE(cft->private); |
4359 | u64 threshold, usage; | |
2c488db2 | 4360 | int i, size, ret; |
2e72b634 KS |
4361 | |
4362 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
4363 | if (ret) | |
4364 | return ret; | |
4365 | ||
4366 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4367 | |
2e72b634 | 4368 | if (type == _MEM) |
2c488db2 | 4369 | thresholds = &memcg->thresholds; |
2e72b634 | 4370 | else if (type == _MEMSWAP) |
2c488db2 | 4371 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4372 | else |
4373 | BUG(); | |
4374 | ||
4375 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
4376 | ||
4377 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 4378 | if (thresholds->primary) |
2e72b634 KS |
4379 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4380 | ||
2c488db2 | 4381 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4382 | |
4383 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 4384 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 4385 | GFP_KERNEL); |
2c488db2 | 4386 | if (!new) { |
2e72b634 KS |
4387 | ret = -ENOMEM; |
4388 | goto unlock; | |
4389 | } | |
2c488db2 | 4390 | new->size = size; |
2e72b634 KS |
4391 | |
4392 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4393 | if (thresholds->primary) { |
4394 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4395 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4396 | } |
4397 | ||
2e72b634 | 4398 | /* Add new threshold */ |
2c488db2 KS |
4399 | new->entries[size - 1].eventfd = eventfd; |
4400 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4401 | |
4402 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4403 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4404 | compare_thresholds, NULL); |
4405 | ||
4406 | /* Find current threshold */ | |
2c488db2 | 4407 | new->current_threshold = -1; |
2e72b634 | 4408 | for (i = 0; i < size; i++) { |
748dad36 | 4409 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4410 | /* |
2c488db2 KS |
4411 | * new->current_threshold will not be used until |
4412 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4413 | * it here. |
4414 | */ | |
2c488db2 | 4415 | ++new->current_threshold; |
748dad36 SZ |
4416 | } else |
4417 | break; | |
2e72b634 KS |
4418 | } |
4419 | ||
2c488db2 KS |
4420 | /* Free old spare buffer and save old primary buffer as spare */ |
4421 | kfree(thresholds->spare); | |
4422 | thresholds->spare = thresholds->primary; | |
4423 | ||
4424 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4425 | |
907860ed | 4426 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4427 | synchronize_rcu(); |
4428 | ||
2e72b634 KS |
4429 | unlock: |
4430 | mutex_unlock(&memcg->thresholds_lock); | |
4431 | ||
4432 | return ret; | |
4433 | } | |
4434 | ||
907860ed | 4435 | static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp, |
9490ff27 | 4436 | struct cftype *cft, struct eventfd_ctx *eventfd) |
2e72b634 KS |
4437 | { |
4438 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4439 | struct mem_cgroup_thresholds *thresholds; |
4440 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
4441 | int type = MEMFILE_TYPE(cft->private); |
4442 | u64 usage; | |
2c488db2 | 4443 | int i, j, size; |
2e72b634 KS |
4444 | |
4445 | mutex_lock(&memcg->thresholds_lock); | |
4446 | if (type == _MEM) | |
2c488db2 | 4447 | thresholds = &memcg->thresholds; |
2e72b634 | 4448 | else if (type == _MEMSWAP) |
2c488db2 | 4449 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4450 | else |
4451 | BUG(); | |
4452 | ||
371528ca AV |
4453 | if (!thresholds->primary) |
4454 | goto unlock; | |
4455 | ||
2e72b634 KS |
4456 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); |
4457 | ||
4458 | /* Check if a threshold crossed before removing */ | |
4459 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4460 | ||
4461 | /* Calculate new number of threshold */ | |
2c488db2 KS |
4462 | size = 0; |
4463 | for (i = 0; i < thresholds->primary->size; i++) { | |
4464 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
4465 | size++; |
4466 | } | |
4467 | ||
2c488db2 | 4468 | new = thresholds->spare; |
907860ed | 4469 | |
2e72b634 KS |
4470 | /* Set thresholds array to NULL if we don't have thresholds */ |
4471 | if (!size) { | |
2c488db2 KS |
4472 | kfree(new); |
4473 | new = NULL; | |
907860ed | 4474 | goto swap_buffers; |
2e72b634 KS |
4475 | } |
4476 | ||
2c488db2 | 4477 | new->size = size; |
2e72b634 KS |
4478 | |
4479 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4480 | new->current_threshold = -1; |
4481 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4482 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4483 | continue; |
4484 | ||
2c488db2 | 4485 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4486 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4487 | /* |
2c488db2 | 4488 | * new->current_threshold will not be used |
2e72b634 KS |
4489 | * until rcu_assign_pointer(), so it's safe to increment |
4490 | * it here. | |
4491 | */ | |
2c488db2 | 4492 | ++new->current_threshold; |
2e72b634 KS |
4493 | } |
4494 | j++; | |
4495 | } | |
4496 | ||
907860ed | 4497 | swap_buffers: |
2c488db2 KS |
4498 | /* Swap primary and spare array */ |
4499 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
4500 | /* If all events are unregistered, free the spare array */ |
4501 | if (!new) { | |
4502 | kfree(thresholds->spare); | |
4503 | thresholds->spare = NULL; | |
4504 | } | |
4505 | ||
2c488db2 | 4506 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4507 | |
907860ed | 4508 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4509 | synchronize_rcu(); |
371528ca | 4510 | unlock: |
2e72b634 | 4511 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4512 | } |
c1e862c1 | 4513 | |
9490ff27 KH |
4514 | static int mem_cgroup_oom_register_event(struct cgroup *cgrp, |
4515 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
4516 | { | |
4517 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4518 | struct mem_cgroup_eventfd_list *event; | |
4519 | int type = MEMFILE_TYPE(cft->private); | |
4520 | ||
4521 | BUG_ON(type != _OOM_TYPE); | |
4522 | event = kmalloc(sizeof(*event), GFP_KERNEL); | |
4523 | if (!event) | |
4524 | return -ENOMEM; | |
4525 | ||
1af8efe9 | 4526 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4527 | |
4528 | event->eventfd = eventfd; | |
4529 | list_add(&event->list, &memcg->oom_notify); | |
4530 | ||
4531 | /* already in OOM ? */ | |
79dfdacc | 4532 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 4533 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4534 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4535 | |
4536 | return 0; | |
4537 | } | |
4538 | ||
907860ed | 4539 | static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp, |
9490ff27 KH |
4540 | struct cftype *cft, struct eventfd_ctx *eventfd) |
4541 | { | |
c0ff4b85 | 4542 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
9490ff27 KH |
4543 | struct mem_cgroup_eventfd_list *ev, *tmp; |
4544 | int type = MEMFILE_TYPE(cft->private); | |
4545 | ||
4546 | BUG_ON(type != _OOM_TYPE); | |
4547 | ||
1af8efe9 | 4548 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4549 | |
c0ff4b85 | 4550 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4551 | if (ev->eventfd == eventfd) { |
4552 | list_del(&ev->list); | |
4553 | kfree(ev); | |
4554 | } | |
4555 | } | |
4556 | ||
1af8efe9 | 4557 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4558 | } |
4559 | ||
3c11ecf4 KH |
4560 | static int mem_cgroup_oom_control_read(struct cgroup *cgrp, |
4561 | struct cftype *cft, struct cgroup_map_cb *cb) | |
4562 | { | |
c0ff4b85 | 4563 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 | 4564 | |
c0ff4b85 | 4565 | cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable); |
3c11ecf4 | 4566 | |
c0ff4b85 | 4567 | if (atomic_read(&memcg->under_oom)) |
3c11ecf4 KH |
4568 | cb->fill(cb, "under_oom", 1); |
4569 | else | |
4570 | cb->fill(cb, "under_oom", 0); | |
4571 | return 0; | |
4572 | } | |
4573 | ||
3c11ecf4 KH |
4574 | static int mem_cgroup_oom_control_write(struct cgroup *cgrp, |
4575 | struct cftype *cft, u64 val) | |
4576 | { | |
c0ff4b85 | 4577 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 KH |
4578 | struct mem_cgroup *parent; |
4579 | ||
4580 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
4581 | if (!cgrp->parent || !((val == 0) || (val == 1))) | |
4582 | return -EINVAL; | |
4583 | ||
4584 | parent = mem_cgroup_from_cont(cgrp->parent); | |
4585 | ||
4586 | cgroup_lock(); | |
4587 | /* oom-kill-disable is a flag for subhierarchy. */ | |
4588 | if ((parent->use_hierarchy) || | |
c0ff4b85 | 4589 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
3c11ecf4 KH |
4590 | cgroup_unlock(); |
4591 | return -EINVAL; | |
4592 | } | |
c0ff4b85 | 4593 | memcg->oom_kill_disable = val; |
4d845ebf | 4594 | if (!val) |
c0ff4b85 | 4595 | memcg_oom_recover(memcg); |
3c11ecf4 KH |
4596 | cgroup_unlock(); |
4597 | return 0; | |
4598 | } | |
4599 | ||
406eb0c9 YH |
4600 | #ifdef CONFIG_NUMA |
4601 | static const struct file_operations mem_control_numa_stat_file_operations = { | |
4602 | .read = seq_read, | |
4603 | .llseek = seq_lseek, | |
4604 | .release = single_release, | |
4605 | }; | |
4606 | ||
4607 | static int mem_control_numa_stat_open(struct inode *unused, struct file *file) | |
4608 | { | |
4609 | struct cgroup *cont = file->f_dentry->d_parent->d_fsdata; | |
4610 | ||
4611 | file->f_op = &mem_control_numa_stat_file_operations; | |
4612 | return single_open(file, mem_control_numa_stat_show, cont); | |
4613 | } | |
4614 | #endif /* CONFIG_NUMA */ | |
4615 | ||
e5671dfa | 4616 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM |
cbe128e3 | 4617 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 4618 | { |
1d62e436 | 4619 | return mem_cgroup_sockets_init(memcg, ss); |
e5671dfa GC |
4620 | }; |
4621 | ||
1d62e436 | 4622 | static void kmem_cgroup_destroy(struct mem_cgroup *memcg) |
d1a4c0b3 | 4623 | { |
1d62e436 | 4624 | mem_cgroup_sockets_destroy(memcg); |
d1a4c0b3 | 4625 | } |
e5671dfa | 4626 | #else |
cbe128e3 | 4627 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
4628 | { |
4629 | return 0; | |
4630 | } | |
d1a4c0b3 | 4631 | |
1d62e436 | 4632 | static void kmem_cgroup_destroy(struct mem_cgroup *memcg) |
d1a4c0b3 GC |
4633 | { |
4634 | } | |
e5671dfa GC |
4635 | #endif |
4636 | ||
8cdea7c0 BS |
4637 | static struct cftype mem_cgroup_files[] = { |
4638 | { | |
0eea1030 | 4639 | .name = "usage_in_bytes", |
8c7c6e34 | 4640 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
af36f906 | 4641 | .read = mem_cgroup_read, |
9490ff27 KH |
4642 | .register_event = mem_cgroup_usage_register_event, |
4643 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8cdea7c0 | 4644 | }, |
c84872e1 PE |
4645 | { |
4646 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4647 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 4648 | .trigger = mem_cgroup_reset, |
af36f906 | 4649 | .read = mem_cgroup_read, |
c84872e1 | 4650 | }, |
8cdea7c0 | 4651 | { |
0eea1030 | 4652 | .name = "limit_in_bytes", |
8c7c6e34 | 4653 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 4654 | .write_string = mem_cgroup_write, |
af36f906 | 4655 | .read = mem_cgroup_read, |
8cdea7c0 | 4656 | }, |
296c81d8 BS |
4657 | { |
4658 | .name = "soft_limit_in_bytes", | |
4659 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
4660 | .write_string = mem_cgroup_write, | |
af36f906 | 4661 | .read = mem_cgroup_read, |
296c81d8 | 4662 | }, |
8cdea7c0 BS |
4663 | { |
4664 | .name = "failcnt", | |
8c7c6e34 | 4665 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 4666 | .trigger = mem_cgroup_reset, |
af36f906 | 4667 | .read = mem_cgroup_read, |
8cdea7c0 | 4668 | }, |
d2ceb9b7 KH |
4669 | { |
4670 | .name = "stat", | |
c64745cf | 4671 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 4672 | }, |
c1e862c1 KH |
4673 | { |
4674 | .name = "force_empty", | |
4675 | .trigger = mem_cgroup_force_empty_write, | |
4676 | }, | |
18f59ea7 BS |
4677 | { |
4678 | .name = "use_hierarchy", | |
4679 | .write_u64 = mem_cgroup_hierarchy_write, | |
4680 | .read_u64 = mem_cgroup_hierarchy_read, | |
4681 | }, | |
a7885eb8 KM |
4682 | { |
4683 | .name = "swappiness", | |
4684 | .read_u64 = mem_cgroup_swappiness_read, | |
4685 | .write_u64 = mem_cgroup_swappiness_write, | |
4686 | }, | |
7dc74be0 DN |
4687 | { |
4688 | .name = "move_charge_at_immigrate", | |
4689 | .read_u64 = mem_cgroup_move_charge_read, | |
4690 | .write_u64 = mem_cgroup_move_charge_write, | |
4691 | }, | |
9490ff27 KH |
4692 | { |
4693 | .name = "oom_control", | |
3c11ecf4 KH |
4694 | .read_map = mem_cgroup_oom_control_read, |
4695 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
4696 | .register_event = mem_cgroup_oom_register_event, |
4697 | .unregister_event = mem_cgroup_oom_unregister_event, | |
4698 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), | |
4699 | }, | |
406eb0c9 YH |
4700 | #ifdef CONFIG_NUMA |
4701 | { | |
4702 | .name = "numa_stat", | |
4703 | .open = mem_control_numa_stat_open, | |
89577127 | 4704 | .mode = S_IRUGO, |
406eb0c9 YH |
4705 | }, |
4706 | #endif | |
8c7c6e34 | 4707 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
8c7c6e34 KH |
4708 | { |
4709 | .name = "memsw.usage_in_bytes", | |
4710 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
af36f906 | 4711 | .read = mem_cgroup_read, |
9490ff27 KH |
4712 | .register_event = mem_cgroup_usage_register_event, |
4713 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8c7c6e34 KH |
4714 | }, |
4715 | { | |
4716 | .name = "memsw.max_usage_in_bytes", | |
4717 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
4718 | .trigger = mem_cgroup_reset, | |
af36f906 | 4719 | .read = mem_cgroup_read, |
8c7c6e34 KH |
4720 | }, |
4721 | { | |
4722 | .name = "memsw.limit_in_bytes", | |
4723 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
4724 | .write_string = mem_cgroup_write, | |
af36f906 | 4725 | .read = mem_cgroup_read, |
8c7c6e34 KH |
4726 | }, |
4727 | { | |
4728 | .name = "memsw.failcnt", | |
4729 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
4730 | .trigger = mem_cgroup_reset, | |
af36f906 | 4731 | .read = mem_cgroup_read, |
8c7c6e34 | 4732 | }, |
8c7c6e34 | 4733 | #endif |
6bc10349 | 4734 | { }, /* terminate */ |
af36f906 | 4735 | }; |
8c7c6e34 | 4736 | |
c0ff4b85 | 4737 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4738 | { |
4739 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4740 | struct mem_cgroup_per_zone *mz; |
f156ab93 | 4741 | enum lru_list lru; |
41e3355d | 4742 | int zone, tmp = node; |
1ecaab2b KH |
4743 | /* |
4744 | * This routine is called against possible nodes. | |
4745 | * But it's BUG to call kmalloc() against offline node. | |
4746 | * | |
4747 | * TODO: this routine can waste much memory for nodes which will | |
4748 | * never be onlined. It's better to use memory hotplug callback | |
4749 | * function. | |
4750 | */ | |
41e3355d KH |
4751 | if (!node_state(node, N_NORMAL_MEMORY)) |
4752 | tmp = -1; | |
17295c88 | 4753 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4754 | if (!pn) |
4755 | return 1; | |
1ecaab2b | 4756 | |
1ecaab2b KH |
4757 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4758 | mz = &pn->zoneinfo[zone]; | |
f156ab93 HD |
4759 | for_each_lru(lru) |
4760 | INIT_LIST_HEAD(&mz->lruvec.lists[lru]); | |
f64c3f54 | 4761 | mz->usage_in_excess = 0; |
4e416953 | 4762 | mz->on_tree = false; |
d79154bb | 4763 | mz->memcg = memcg; |
1ecaab2b | 4764 | } |
0a619e58 | 4765 | memcg->info.nodeinfo[node] = pn; |
6d12e2d8 KH |
4766 | return 0; |
4767 | } | |
4768 | ||
c0ff4b85 | 4769 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4770 | { |
c0ff4b85 | 4771 | kfree(memcg->info.nodeinfo[node]); |
1ecaab2b KH |
4772 | } |
4773 | ||
33327948 KH |
4774 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4775 | { | |
d79154bb | 4776 | struct mem_cgroup *memcg; |
c62b1a3b | 4777 | int size = sizeof(struct mem_cgroup); |
33327948 | 4778 | |
c62b1a3b | 4779 | /* Can be very big if MAX_NUMNODES is very big */ |
c8dad2bb | 4780 | if (size < PAGE_SIZE) |
d79154bb | 4781 | memcg = kzalloc(size, GFP_KERNEL); |
33327948 | 4782 | else |
d79154bb | 4783 | memcg = vzalloc(size); |
33327948 | 4784 | |
d79154bb | 4785 | if (!memcg) |
e7bbcdf3 DC |
4786 | return NULL; |
4787 | ||
d79154bb HD |
4788 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
4789 | if (!memcg->stat) | |
d2e61b8d | 4790 | goto out_free; |
d79154bb HD |
4791 | spin_lock_init(&memcg->pcp_counter_lock); |
4792 | return memcg; | |
d2e61b8d DC |
4793 | |
4794 | out_free: | |
4795 | if (size < PAGE_SIZE) | |
d79154bb | 4796 | kfree(memcg); |
d2e61b8d | 4797 | else |
d79154bb | 4798 | vfree(memcg); |
d2e61b8d | 4799 | return NULL; |
33327948 KH |
4800 | } |
4801 | ||
59927fb9 HD |
4802 | /* |
4803 | * Helpers for freeing a vzalloc()ed mem_cgroup by RCU, | |
4804 | * but in process context. The work_freeing structure is overlaid | |
4805 | * on the rcu_freeing structure, which itself is overlaid on memsw. | |
4806 | */ | |
4807 | static void vfree_work(struct work_struct *work) | |
4808 | { | |
4809 | struct mem_cgroup *memcg; | |
4810 | ||
4811 | memcg = container_of(work, struct mem_cgroup, work_freeing); | |
4812 | vfree(memcg); | |
4813 | } | |
4814 | static void vfree_rcu(struct rcu_head *rcu_head) | |
4815 | { | |
4816 | struct mem_cgroup *memcg; | |
4817 | ||
4818 | memcg = container_of(rcu_head, struct mem_cgroup, rcu_freeing); | |
4819 | INIT_WORK(&memcg->work_freeing, vfree_work); | |
4820 | schedule_work(&memcg->work_freeing); | |
4821 | } | |
4822 | ||
8c7c6e34 KH |
4823 | /* |
4824 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
4825 | * (scanning all at force_empty is too costly...) | |
4826 | * | |
4827 | * Instead of clearing all references at force_empty, we remember | |
4828 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4829 | * it goes down to 0. | |
4830 | * | |
8c7c6e34 KH |
4831 | * Removal of cgroup itself succeeds regardless of refs from swap. |
4832 | */ | |
4833 | ||
c0ff4b85 | 4834 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
33327948 | 4835 | { |
08e552c6 KH |
4836 | int node; |
4837 | ||
c0ff4b85 R |
4838 | mem_cgroup_remove_from_trees(memcg); |
4839 | free_css_id(&mem_cgroup_subsys, &memcg->css); | |
04046e1a | 4840 | |
3ed28fa1 | 4841 | for_each_node(node) |
c0ff4b85 | 4842 | free_mem_cgroup_per_zone_info(memcg, node); |
08e552c6 | 4843 | |
c0ff4b85 | 4844 | free_percpu(memcg->stat); |
c62b1a3b | 4845 | if (sizeof(struct mem_cgroup) < PAGE_SIZE) |
59927fb9 | 4846 | kfree_rcu(memcg, rcu_freeing); |
33327948 | 4847 | else |
59927fb9 | 4848 | call_rcu(&memcg->rcu_freeing, vfree_rcu); |
33327948 KH |
4849 | } |
4850 | ||
c0ff4b85 | 4851 | static void mem_cgroup_get(struct mem_cgroup *memcg) |
8c7c6e34 | 4852 | { |
c0ff4b85 | 4853 | atomic_inc(&memcg->refcnt); |
8c7c6e34 KH |
4854 | } |
4855 | ||
c0ff4b85 | 4856 | static void __mem_cgroup_put(struct mem_cgroup *memcg, int count) |
8c7c6e34 | 4857 | { |
c0ff4b85 R |
4858 | if (atomic_sub_and_test(count, &memcg->refcnt)) { |
4859 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); | |
4860 | __mem_cgroup_free(memcg); | |
7bcc1bb1 DN |
4861 | if (parent) |
4862 | mem_cgroup_put(parent); | |
4863 | } | |
8c7c6e34 KH |
4864 | } |
4865 | ||
c0ff4b85 | 4866 | static void mem_cgroup_put(struct mem_cgroup *memcg) |
483c30b5 | 4867 | { |
c0ff4b85 | 4868 | __mem_cgroup_put(memcg, 1); |
483c30b5 DN |
4869 | } |
4870 | ||
7bcc1bb1 DN |
4871 | /* |
4872 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4873 | */ | |
e1aab161 | 4874 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 4875 | { |
c0ff4b85 | 4876 | if (!memcg->res.parent) |
7bcc1bb1 | 4877 | return NULL; |
c0ff4b85 | 4878 | return mem_cgroup_from_res_counter(memcg->res.parent, res); |
7bcc1bb1 | 4879 | } |
e1aab161 | 4880 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 4881 | |
c077719b KH |
4882 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
4883 | static void __init enable_swap_cgroup(void) | |
4884 | { | |
f8d66542 | 4885 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
4886 | do_swap_account = 1; |
4887 | } | |
4888 | #else | |
4889 | static void __init enable_swap_cgroup(void) | |
4890 | { | |
4891 | } | |
4892 | #endif | |
4893 | ||
f64c3f54 BS |
4894 | static int mem_cgroup_soft_limit_tree_init(void) |
4895 | { | |
4896 | struct mem_cgroup_tree_per_node *rtpn; | |
4897 | struct mem_cgroup_tree_per_zone *rtpz; | |
4898 | int tmp, node, zone; | |
4899 | ||
3ed28fa1 | 4900 | for_each_node(node) { |
f64c3f54 BS |
4901 | tmp = node; |
4902 | if (!node_state(node, N_NORMAL_MEMORY)) | |
4903 | tmp = -1; | |
4904 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
4905 | if (!rtpn) | |
c3cecc68 | 4906 | goto err_cleanup; |
f64c3f54 BS |
4907 | |
4908 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
4909 | ||
4910 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
4911 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
4912 | rtpz->rb_root = RB_ROOT; | |
4913 | spin_lock_init(&rtpz->lock); | |
4914 | } | |
4915 | } | |
4916 | return 0; | |
c3cecc68 MH |
4917 | |
4918 | err_cleanup: | |
3ed28fa1 | 4919 | for_each_node(node) { |
c3cecc68 MH |
4920 | if (!soft_limit_tree.rb_tree_per_node[node]) |
4921 | break; | |
4922 | kfree(soft_limit_tree.rb_tree_per_node[node]); | |
4923 | soft_limit_tree.rb_tree_per_node[node] = NULL; | |
4924 | } | |
4925 | return 1; | |
4926 | ||
f64c3f54 BS |
4927 | } |
4928 | ||
0eb253e2 | 4929 | static struct cgroup_subsys_state * __ref |
761b3ef5 | 4930 | mem_cgroup_create(struct cgroup *cont) |
8cdea7c0 | 4931 | { |
c0ff4b85 | 4932 | struct mem_cgroup *memcg, *parent; |
04046e1a | 4933 | long error = -ENOMEM; |
6d12e2d8 | 4934 | int node; |
8cdea7c0 | 4935 | |
c0ff4b85 R |
4936 | memcg = mem_cgroup_alloc(); |
4937 | if (!memcg) | |
04046e1a | 4938 | return ERR_PTR(error); |
78fb7466 | 4939 | |
3ed28fa1 | 4940 | for_each_node(node) |
c0ff4b85 | 4941 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 4942 | goto free_out; |
f64c3f54 | 4943 | |
c077719b | 4944 | /* root ? */ |
28dbc4b6 | 4945 | if (cont->parent == NULL) { |
cdec2e42 | 4946 | int cpu; |
c077719b | 4947 | enable_swap_cgroup(); |
28dbc4b6 | 4948 | parent = NULL; |
f64c3f54 BS |
4949 | if (mem_cgroup_soft_limit_tree_init()) |
4950 | goto free_out; | |
a41c58a6 | 4951 | root_mem_cgroup = memcg; |
cdec2e42 KH |
4952 | for_each_possible_cpu(cpu) { |
4953 | struct memcg_stock_pcp *stock = | |
4954 | &per_cpu(memcg_stock, cpu); | |
4955 | INIT_WORK(&stock->work, drain_local_stock); | |
4956 | } | |
711d3d2c | 4957 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
18f59ea7 | 4958 | } else { |
28dbc4b6 | 4959 | parent = mem_cgroup_from_cont(cont->parent); |
c0ff4b85 R |
4960 | memcg->use_hierarchy = parent->use_hierarchy; |
4961 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
18f59ea7 | 4962 | } |
28dbc4b6 | 4963 | |
18f59ea7 | 4964 | if (parent && parent->use_hierarchy) { |
c0ff4b85 R |
4965 | res_counter_init(&memcg->res, &parent->res); |
4966 | res_counter_init(&memcg->memsw, &parent->memsw); | |
7bcc1bb1 DN |
4967 | /* |
4968 | * We increment refcnt of the parent to ensure that we can | |
4969 | * safely access it on res_counter_charge/uncharge. | |
4970 | * This refcnt will be decremented when freeing this | |
4971 | * mem_cgroup(see mem_cgroup_put). | |
4972 | */ | |
4973 | mem_cgroup_get(parent); | |
18f59ea7 | 4974 | } else { |
c0ff4b85 R |
4975 | res_counter_init(&memcg->res, NULL); |
4976 | res_counter_init(&memcg->memsw, NULL); | |
18f59ea7 | 4977 | } |
c0ff4b85 R |
4978 | memcg->last_scanned_node = MAX_NUMNODES; |
4979 | INIT_LIST_HEAD(&memcg->oom_notify); | |
6d61ef40 | 4980 | |
a7885eb8 | 4981 | if (parent) |
c0ff4b85 R |
4982 | memcg->swappiness = mem_cgroup_swappiness(parent); |
4983 | atomic_set(&memcg->refcnt, 1); | |
4984 | memcg->move_charge_at_immigrate = 0; | |
4985 | mutex_init(&memcg->thresholds_lock); | |
312734c0 | 4986 | spin_lock_init(&memcg->move_lock); |
cbe128e3 GC |
4987 | |
4988 | error = memcg_init_kmem(memcg, &mem_cgroup_subsys); | |
4989 | if (error) { | |
4990 | /* | |
4991 | * We call put now because our (and parent's) refcnts | |
4992 | * are already in place. mem_cgroup_put() will internally | |
4993 | * call __mem_cgroup_free, so return directly | |
4994 | */ | |
4995 | mem_cgroup_put(memcg); | |
4996 | return ERR_PTR(error); | |
4997 | } | |
c0ff4b85 | 4998 | return &memcg->css; |
6d12e2d8 | 4999 | free_out: |
c0ff4b85 | 5000 | __mem_cgroup_free(memcg); |
04046e1a | 5001 | return ERR_PTR(error); |
8cdea7c0 BS |
5002 | } |
5003 | ||
761b3ef5 | 5004 | static int mem_cgroup_pre_destroy(struct cgroup *cont) |
df878fb0 | 5005 | { |
c0ff4b85 | 5006 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
ec64f515 | 5007 | |
c0ff4b85 | 5008 | return mem_cgroup_force_empty(memcg, false); |
df878fb0 KH |
5009 | } |
5010 | ||
761b3ef5 | 5011 | static void mem_cgroup_destroy(struct cgroup *cont) |
8cdea7c0 | 5012 | { |
c0ff4b85 | 5013 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
c268e994 | 5014 | |
1d62e436 | 5015 | kmem_cgroup_destroy(memcg); |
d1a4c0b3 | 5016 | |
c0ff4b85 | 5017 | mem_cgroup_put(memcg); |
8cdea7c0 BS |
5018 | } |
5019 | ||
02491447 | 5020 | #ifdef CONFIG_MMU |
7dc74be0 | 5021 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
5022 | #define PRECHARGE_COUNT_AT_ONCE 256 |
5023 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 5024 | { |
854ffa8d DN |
5025 | int ret = 0; |
5026 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
c0ff4b85 | 5027 | struct mem_cgroup *memcg = mc.to; |
4ffef5fe | 5028 | |
c0ff4b85 | 5029 | if (mem_cgroup_is_root(memcg)) { |
854ffa8d DN |
5030 | mc.precharge += count; |
5031 | /* we don't need css_get for root */ | |
5032 | return ret; | |
5033 | } | |
5034 | /* try to charge at once */ | |
5035 | if (count > 1) { | |
5036 | struct res_counter *dummy; | |
5037 | /* | |
c0ff4b85 | 5038 | * "memcg" cannot be under rmdir() because we've already checked |
854ffa8d DN |
5039 | * by cgroup_lock_live_cgroup() that it is not removed and we |
5040 | * are still under the same cgroup_mutex. So we can postpone | |
5041 | * css_get(). | |
5042 | */ | |
c0ff4b85 | 5043 | if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy)) |
854ffa8d | 5044 | goto one_by_one; |
c0ff4b85 | 5045 | if (do_swap_account && res_counter_charge(&memcg->memsw, |
854ffa8d | 5046 | PAGE_SIZE * count, &dummy)) { |
c0ff4b85 | 5047 | res_counter_uncharge(&memcg->res, PAGE_SIZE * count); |
854ffa8d DN |
5048 | goto one_by_one; |
5049 | } | |
5050 | mc.precharge += count; | |
854ffa8d DN |
5051 | return ret; |
5052 | } | |
5053 | one_by_one: | |
5054 | /* fall back to one by one charge */ | |
5055 | while (count--) { | |
5056 | if (signal_pending(current)) { | |
5057 | ret = -EINTR; | |
5058 | break; | |
5059 | } | |
5060 | if (!batch_count--) { | |
5061 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
5062 | cond_resched(); | |
5063 | } | |
c0ff4b85 R |
5064 | ret = __mem_cgroup_try_charge(NULL, |
5065 | GFP_KERNEL, 1, &memcg, false); | |
38c5d72f | 5066 | if (ret) |
854ffa8d | 5067 | /* mem_cgroup_clear_mc() will do uncharge later */ |
38c5d72f | 5068 | return ret; |
854ffa8d DN |
5069 | mc.precharge++; |
5070 | } | |
4ffef5fe DN |
5071 | return ret; |
5072 | } | |
5073 | ||
5074 | /** | |
8d32ff84 | 5075 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
5076 | * @vma: the vma the pte to be checked belongs |
5077 | * @addr: the address corresponding to the pte to be checked | |
5078 | * @ptent: the pte to be checked | |
02491447 | 5079 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
5080 | * |
5081 | * Returns | |
5082 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5083 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5084 | * move charge. if @target is not NULL, the page is stored in target->page | |
5085 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
5086 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
5087 | * target for charge migration. if @target is not NULL, the entry is stored | |
5088 | * in target->ent. | |
4ffef5fe DN |
5089 | * |
5090 | * Called with pte lock held. | |
5091 | */ | |
4ffef5fe DN |
5092 | union mc_target { |
5093 | struct page *page; | |
02491447 | 5094 | swp_entry_t ent; |
4ffef5fe DN |
5095 | }; |
5096 | ||
4ffef5fe | 5097 | enum mc_target_type { |
8d32ff84 | 5098 | MC_TARGET_NONE = 0, |
4ffef5fe | 5099 | MC_TARGET_PAGE, |
02491447 | 5100 | MC_TARGET_SWAP, |
4ffef5fe DN |
5101 | }; |
5102 | ||
90254a65 DN |
5103 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5104 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5105 | { |
90254a65 | 5106 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5107 | |
90254a65 DN |
5108 | if (!page || !page_mapped(page)) |
5109 | return NULL; | |
5110 | if (PageAnon(page)) { | |
5111 | /* we don't move shared anon */ | |
4b91355e | 5112 | if (!move_anon()) |
90254a65 | 5113 | return NULL; |
87946a72 DN |
5114 | } else if (!move_file()) |
5115 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
5116 | return NULL; |
5117 | if (!get_page_unless_zero(page)) | |
5118 | return NULL; | |
5119 | ||
5120 | return page; | |
5121 | } | |
5122 | ||
4b91355e | 5123 | #ifdef CONFIG_SWAP |
90254a65 DN |
5124 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
5125 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5126 | { | |
90254a65 DN |
5127 | struct page *page = NULL; |
5128 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5129 | ||
5130 | if (!move_anon() || non_swap_entry(ent)) | |
5131 | return NULL; | |
4b91355e KH |
5132 | /* |
5133 | * Because lookup_swap_cache() updates some statistics counter, | |
5134 | * we call find_get_page() with swapper_space directly. | |
5135 | */ | |
5136 | page = find_get_page(&swapper_space, ent.val); | |
90254a65 DN |
5137 | if (do_swap_account) |
5138 | entry->val = ent.val; | |
5139 | ||
5140 | return page; | |
5141 | } | |
4b91355e KH |
5142 | #else |
5143 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
5144 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5145 | { | |
5146 | return NULL; | |
5147 | } | |
5148 | #endif | |
90254a65 | 5149 | |
87946a72 DN |
5150 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5151 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5152 | { | |
5153 | struct page *page = NULL; | |
87946a72 DN |
5154 | struct address_space *mapping; |
5155 | pgoff_t pgoff; | |
5156 | ||
5157 | if (!vma->vm_file) /* anonymous vma */ | |
5158 | return NULL; | |
5159 | if (!move_file()) | |
5160 | return NULL; | |
5161 | ||
87946a72 DN |
5162 | mapping = vma->vm_file->f_mapping; |
5163 | if (pte_none(ptent)) | |
5164 | pgoff = linear_page_index(vma, addr); | |
5165 | else /* pte_file(ptent) is true */ | |
5166 | pgoff = pte_to_pgoff(ptent); | |
5167 | ||
5168 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
5169 | page = find_get_page(mapping, pgoff); |
5170 | ||
5171 | #ifdef CONFIG_SWAP | |
5172 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
5173 | if (radix_tree_exceptional_entry(page)) { | |
5174 | swp_entry_t swap = radix_to_swp_entry(page); | |
87946a72 | 5175 | if (do_swap_account) |
aa3b1895 HD |
5176 | *entry = swap; |
5177 | page = find_get_page(&swapper_space, swap.val); | |
87946a72 | 5178 | } |
aa3b1895 | 5179 | #endif |
87946a72 DN |
5180 | return page; |
5181 | } | |
5182 | ||
8d32ff84 | 5183 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5184 | unsigned long addr, pte_t ptent, union mc_target *target) |
5185 | { | |
5186 | struct page *page = NULL; | |
5187 | struct page_cgroup *pc; | |
8d32ff84 | 5188 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5189 | swp_entry_t ent = { .val = 0 }; |
5190 | ||
5191 | if (pte_present(ptent)) | |
5192 | page = mc_handle_present_pte(vma, addr, ptent); | |
5193 | else if (is_swap_pte(ptent)) | |
5194 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
5195 | else if (pte_none(ptent) || pte_file(ptent)) |
5196 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
5197 | |
5198 | if (!page && !ent.val) | |
8d32ff84 | 5199 | return ret; |
02491447 DN |
5200 | if (page) { |
5201 | pc = lookup_page_cgroup(page); | |
5202 | /* | |
5203 | * Do only loose check w/o page_cgroup lock. | |
5204 | * mem_cgroup_move_account() checks the pc is valid or not under | |
5205 | * the lock. | |
5206 | */ | |
5207 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5208 | ret = MC_TARGET_PAGE; | |
5209 | if (target) | |
5210 | target->page = page; | |
5211 | } | |
5212 | if (!ret || !target) | |
5213 | put_page(page); | |
5214 | } | |
90254a65 DN |
5215 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
5216 | if (ent.val && !ret && | |
9fb4b7cc | 5217 | css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5218 | ret = MC_TARGET_SWAP; |
5219 | if (target) | |
5220 | target->ent = ent; | |
4ffef5fe | 5221 | } |
4ffef5fe DN |
5222 | return ret; |
5223 | } | |
5224 | ||
12724850 NH |
5225 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5226 | /* | |
5227 | * We don't consider swapping or file mapped pages because THP does not | |
5228 | * support them for now. | |
5229 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
5230 | */ | |
5231 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5232 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5233 | { | |
5234 | struct page *page = NULL; | |
5235 | struct page_cgroup *pc; | |
5236 | enum mc_target_type ret = MC_TARGET_NONE; | |
5237 | ||
5238 | page = pmd_page(pmd); | |
5239 | VM_BUG_ON(!page || !PageHead(page)); | |
5240 | if (!move_anon()) | |
5241 | return ret; | |
5242 | pc = lookup_page_cgroup(page); | |
5243 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5244 | ret = MC_TARGET_PAGE; | |
5245 | if (target) { | |
5246 | get_page(page); | |
5247 | target->page = page; | |
5248 | } | |
5249 | } | |
5250 | return ret; | |
5251 | } | |
5252 | #else | |
5253 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5254 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5255 | { | |
5256 | return MC_TARGET_NONE; | |
5257 | } | |
5258 | #endif | |
5259 | ||
4ffef5fe DN |
5260 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5261 | unsigned long addr, unsigned long end, | |
5262 | struct mm_walk *walk) | |
5263 | { | |
5264 | struct vm_area_struct *vma = walk->private; | |
5265 | pte_t *pte; | |
5266 | spinlock_t *ptl; | |
5267 | ||
12724850 NH |
5268 | if (pmd_trans_huge_lock(pmd, vma) == 1) { |
5269 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) | |
5270 | mc.precharge += HPAGE_PMD_NR; | |
5271 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 5272 | return 0; |
12724850 | 5273 | } |
03319327 | 5274 | |
45f83cef AA |
5275 | if (pmd_trans_unstable(pmd)) |
5276 | return 0; | |
4ffef5fe DN |
5277 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5278 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5279 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5280 | mc.precharge++; /* increment precharge temporarily */ |
5281 | pte_unmap_unlock(pte - 1, ptl); | |
5282 | cond_resched(); | |
5283 | ||
7dc74be0 DN |
5284 | return 0; |
5285 | } | |
5286 | ||
4ffef5fe DN |
5287 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5288 | { | |
5289 | unsigned long precharge; | |
5290 | struct vm_area_struct *vma; | |
5291 | ||
dfe076b0 | 5292 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
5293 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5294 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
5295 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5296 | .mm = mm, | |
5297 | .private = vma, | |
5298 | }; | |
5299 | if (is_vm_hugetlb_page(vma)) | |
5300 | continue; | |
4ffef5fe DN |
5301 | walk_page_range(vma->vm_start, vma->vm_end, |
5302 | &mem_cgroup_count_precharge_walk); | |
5303 | } | |
dfe076b0 | 5304 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5305 | |
5306 | precharge = mc.precharge; | |
5307 | mc.precharge = 0; | |
5308 | ||
5309 | return precharge; | |
5310 | } | |
5311 | ||
4ffef5fe DN |
5312 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5313 | { | |
dfe076b0 DN |
5314 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5315 | ||
5316 | VM_BUG_ON(mc.moving_task); | |
5317 | mc.moving_task = current; | |
5318 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5319 | } |
5320 | ||
dfe076b0 DN |
5321 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5322 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5323 | { |
2bd9bb20 KH |
5324 | struct mem_cgroup *from = mc.from; |
5325 | struct mem_cgroup *to = mc.to; | |
5326 | ||
4ffef5fe | 5327 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
5328 | if (mc.precharge) { |
5329 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
5330 | mc.precharge = 0; | |
5331 | } | |
5332 | /* | |
5333 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5334 | * we must uncharge here. | |
5335 | */ | |
5336 | if (mc.moved_charge) { | |
5337 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
5338 | mc.moved_charge = 0; | |
4ffef5fe | 5339 | } |
483c30b5 DN |
5340 | /* we must fixup refcnts and charges */ |
5341 | if (mc.moved_swap) { | |
483c30b5 DN |
5342 | /* uncharge swap account from the old cgroup */ |
5343 | if (!mem_cgroup_is_root(mc.from)) | |
5344 | res_counter_uncharge(&mc.from->memsw, | |
5345 | PAGE_SIZE * mc.moved_swap); | |
5346 | __mem_cgroup_put(mc.from, mc.moved_swap); | |
5347 | ||
5348 | if (!mem_cgroup_is_root(mc.to)) { | |
5349 | /* | |
5350 | * we charged both to->res and to->memsw, so we should | |
5351 | * uncharge to->res. | |
5352 | */ | |
5353 | res_counter_uncharge(&mc.to->res, | |
5354 | PAGE_SIZE * mc.moved_swap); | |
483c30b5 DN |
5355 | } |
5356 | /* we've already done mem_cgroup_get(mc.to) */ | |
483c30b5 DN |
5357 | mc.moved_swap = 0; |
5358 | } | |
dfe076b0 DN |
5359 | memcg_oom_recover(from); |
5360 | memcg_oom_recover(to); | |
5361 | wake_up_all(&mc.waitq); | |
5362 | } | |
5363 | ||
5364 | static void mem_cgroup_clear_mc(void) | |
5365 | { | |
5366 | struct mem_cgroup *from = mc.from; | |
5367 | ||
5368 | /* | |
5369 | * we must clear moving_task before waking up waiters at the end of | |
5370 | * task migration. | |
5371 | */ | |
5372 | mc.moving_task = NULL; | |
5373 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5374 | spin_lock(&mc.lock); |
4ffef5fe DN |
5375 | mc.from = NULL; |
5376 | mc.to = NULL; | |
2bd9bb20 | 5377 | spin_unlock(&mc.lock); |
32047e2a | 5378 | mem_cgroup_end_move(from); |
4ffef5fe DN |
5379 | } |
5380 | ||
761b3ef5 LZ |
5381 | static int mem_cgroup_can_attach(struct cgroup *cgroup, |
5382 | struct cgroup_taskset *tset) | |
7dc74be0 | 5383 | { |
2f7ee569 | 5384 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 5385 | int ret = 0; |
c0ff4b85 | 5386 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup); |
7dc74be0 | 5387 | |
c0ff4b85 | 5388 | if (memcg->move_charge_at_immigrate) { |
7dc74be0 DN |
5389 | struct mm_struct *mm; |
5390 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
5391 | ||
c0ff4b85 | 5392 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
5393 | |
5394 | mm = get_task_mm(p); | |
5395 | if (!mm) | |
5396 | return 0; | |
7dc74be0 | 5397 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
5398 | if (mm->owner == p) { |
5399 | VM_BUG_ON(mc.from); | |
5400 | VM_BUG_ON(mc.to); | |
5401 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 5402 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 5403 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 5404 | mem_cgroup_start_move(from); |
2bd9bb20 | 5405 | spin_lock(&mc.lock); |
4ffef5fe | 5406 | mc.from = from; |
c0ff4b85 | 5407 | mc.to = memcg; |
2bd9bb20 | 5408 | spin_unlock(&mc.lock); |
dfe076b0 | 5409 | /* We set mc.moving_task later */ |
4ffef5fe DN |
5410 | |
5411 | ret = mem_cgroup_precharge_mc(mm); | |
5412 | if (ret) | |
5413 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
5414 | } |
5415 | mmput(mm); | |
7dc74be0 DN |
5416 | } |
5417 | return ret; | |
5418 | } | |
5419 | ||
761b3ef5 LZ |
5420 | static void mem_cgroup_cancel_attach(struct cgroup *cgroup, |
5421 | struct cgroup_taskset *tset) | |
7dc74be0 | 5422 | { |
4ffef5fe | 5423 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
5424 | } |
5425 | ||
4ffef5fe DN |
5426 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5427 | unsigned long addr, unsigned long end, | |
5428 | struct mm_walk *walk) | |
7dc74be0 | 5429 | { |
4ffef5fe DN |
5430 | int ret = 0; |
5431 | struct vm_area_struct *vma = walk->private; | |
5432 | pte_t *pte; | |
5433 | spinlock_t *ptl; | |
12724850 NH |
5434 | enum mc_target_type target_type; |
5435 | union mc_target target; | |
5436 | struct page *page; | |
5437 | struct page_cgroup *pc; | |
4ffef5fe | 5438 | |
12724850 NH |
5439 | /* |
5440 | * We don't take compound_lock() here but no race with splitting thp | |
5441 | * happens because: | |
5442 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
5443 | * under splitting, which means there's no concurrent thp split, | |
5444 | * - if another thread runs into split_huge_page() just after we | |
5445 | * entered this if-block, the thread must wait for page table lock | |
5446 | * to be unlocked in __split_huge_page_splitting(), where the main | |
5447 | * part of thp split is not executed yet. | |
5448 | */ | |
5449 | if (pmd_trans_huge_lock(pmd, vma) == 1) { | |
62ade86a | 5450 | if (mc.precharge < HPAGE_PMD_NR) { |
12724850 NH |
5451 | spin_unlock(&vma->vm_mm->page_table_lock); |
5452 | return 0; | |
5453 | } | |
5454 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5455 | if (target_type == MC_TARGET_PAGE) { | |
5456 | page = target.page; | |
5457 | if (!isolate_lru_page(page)) { | |
5458 | pc = lookup_page_cgroup(page); | |
5459 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, | |
5460 | pc, mc.from, mc.to, | |
5461 | false)) { | |
5462 | mc.precharge -= HPAGE_PMD_NR; | |
5463 | mc.moved_charge += HPAGE_PMD_NR; | |
5464 | } | |
5465 | putback_lru_page(page); | |
5466 | } | |
5467 | put_page(page); | |
5468 | } | |
5469 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 5470 | return 0; |
12724850 NH |
5471 | } |
5472 | ||
45f83cef AA |
5473 | if (pmd_trans_unstable(pmd)) |
5474 | return 0; | |
4ffef5fe DN |
5475 | retry: |
5476 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5477 | for (; addr != end; addr += PAGE_SIZE) { | |
5478 | pte_t ptent = *(pte++); | |
02491447 | 5479 | swp_entry_t ent; |
4ffef5fe DN |
5480 | |
5481 | if (!mc.precharge) | |
5482 | break; | |
5483 | ||
8d32ff84 | 5484 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
5485 | case MC_TARGET_PAGE: |
5486 | page = target.page; | |
5487 | if (isolate_lru_page(page)) | |
5488 | goto put; | |
5489 | pc = lookup_page_cgroup(page); | |
7ec99d62 JW |
5490 | if (!mem_cgroup_move_account(page, 1, pc, |
5491 | mc.from, mc.to, false)) { | |
4ffef5fe | 5492 | mc.precharge--; |
854ffa8d DN |
5493 | /* we uncharge from mc.from later. */ |
5494 | mc.moved_charge++; | |
4ffef5fe DN |
5495 | } |
5496 | putback_lru_page(page); | |
8d32ff84 | 5497 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5498 | put_page(page); |
5499 | break; | |
02491447 DN |
5500 | case MC_TARGET_SWAP: |
5501 | ent = target.ent; | |
e91cbb42 | 5502 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5503 | mc.precharge--; |
483c30b5 DN |
5504 | /* we fixup refcnts and charges later. */ |
5505 | mc.moved_swap++; | |
5506 | } | |
02491447 | 5507 | break; |
4ffef5fe DN |
5508 | default: |
5509 | break; | |
5510 | } | |
5511 | } | |
5512 | pte_unmap_unlock(pte - 1, ptl); | |
5513 | cond_resched(); | |
5514 | ||
5515 | if (addr != end) { | |
5516 | /* | |
5517 | * We have consumed all precharges we got in can_attach(). | |
5518 | * We try charge one by one, but don't do any additional | |
5519 | * charges to mc.to if we have failed in charge once in attach() | |
5520 | * phase. | |
5521 | */ | |
854ffa8d | 5522 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5523 | if (!ret) |
5524 | goto retry; | |
5525 | } | |
5526 | ||
5527 | return ret; | |
5528 | } | |
5529 | ||
5530 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
5531 | { | |
5532 | struct vm_area_struct *vma; | |
5533 | ||
5534 | lru_add_drain_all(); | |
dfe076b0 DN |
5535 | retry: |
5536 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5537 | /* | |
5538 | * Someone who are holding the mmap_sem might be waiting in | |
5539 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5540 | * and retry. Because we cancel precharges, we might not be able | |
5541 | * to move enough charges, but moving charge is a best-effort | |
5542 | * feature anyway, so it wouldn't be a big problem. | |
5543 | */ | |
5544 | __mem_cgroup_clear_mc(); | |
5545 | cond_resched(); | |
5546 | goto retry; | |
5547 | } | |
4ffef5fe DN |
5548 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5549 | int ret; | |
5550 | struct mm_walk mem_cgroup_move_charge_walk = { | |
5551 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
5552 | .mm = mm, | |
5553 | .private = vma, | |
5554 | }; | |
5555 | if (is_vm_hugetlb_page(vma)) | |
5556 | continue; | |
4ffef5fe DN |
5557 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
5558 | &mem_cgroup_move_charge_walk); | |
5559 | if (ret) | |
5560 | /* | |
5561 | * means we have consumed all precharges and failed in | |
5562 | * doing additional charge. Just abandon here. | |
5563 | */ | |
5564 | break; | |
5565 | } | |
dfe076b0 | 5566 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
5567 | } |
5568 | ||
761b3ef5 LZ |
5569 | static void mem_cgroup_move_task(struct cgroup *cont, |
5570 | struct cgroup_taskset *tset) | |
67e465a7 | 5571 | { |
2f7ee569 | 5572 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 5573 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 5574 | |
dfe076b0 | 5575 | if (mm) { |
a433658c KM |
5576 | if (mc.to) |
5577 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
5578 | mmput(mm); |
5579 | } | |
a433658c KM |
5580 | if (mc.to) |
5581 | mem_cgroup_clear_mc(); | |
67e465a7 | 5582 | } |
5cfb80a7 | 5583 | #else /* !CONFIG_MMU */ |
761b3ef5 LZ |
5584 | static int mem_cgroup_can_attach(struct cgroup *cgroup, |
5585 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
5586 | { |
5587 | return 0; | |
5588 | } | |
761b3ef5 LZ |
5589 | static void mem_cgroup_cancel_attach(struct cgroup *cgroup, |
5590 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
5591 | { |
5592 | } | |
761b3ef5 LZ |
5593 | static void mem_cgroup_move_task(struct cgroup *cont, |
5594 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
5595 | { |
5596 | } | |
5597 | #endif | |
67e465a7 | 5598 | |
8cdea7c0 BS |
5599 | struct cgroup_subsys mem_cgroup_subsys = { |
5600 | .name = "memory", | |
5601 | .subsys_id = mem_cgroup_subsys_id, | |
5602 | .create = mem_cgroup_create, | |
df878fb0 | 5603 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 | 5604 | .destroy = mem_cgroup_destroy, |
7dc74be0 DN |
5605 | .can_attach = mem_cgroup_can_attach, |
5606 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5607 | .attach = mem_cgroup_move_task, |
6bc10349 | 5608 | .base_cftypes = mem_cgroup_files, |
6d12e2d8 | 5609 | .early_init = 0, |
04046e1a | 5610 | .use_id = 1, |
48ddbe19 | 5611 | .__DEPRECATED_clear_css_refs = true, |
8cdea7c0 | 5612 | }; |
c077719b KH |
5613 | |
5614 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
a42c390c MH |
5615 | static int __init enable_swap_account(char *s) |
5616 | { | |
5617 | /* consider enabled if no parameter or 1 is given */ | |
a2c8990a | 5618 | if (!strcmp(s, "1")) |
a42c390c | 5619 | really_do_swap_account = 1; |
a2c8990a | 5620 | else if (!strcmp(s, "0")) |
a42c390c MH |
5621 | really_do_swap_account = 0; |
5622 | return 1; | |
5623 | } | |
a2c8990a | 5624 | __setup("swapaccount=", enable_swap_account); |
c077719b | 5625 | |
c077719b | 5626 | #endif |