Commit | Line | Data |
---|---|---|
8cdea7c0 BS |
1 | /* memcontrol.c - Memory Controller |
2 | * | |
3 | * Copyright IBM Corporation, 2007 | |
4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
5 | * | |
78fb7466 PE |
6 | * Copyright 2007 OpenVZ SWsoft Inc |
7 | * Author: Pavel Emelianov <xemul@openvz.org> | |
8 | * | |
2e72b634 KS |
9 | * Memory thresholds |
10 | * Copyright (C) 2009 Nokia Corporation | |
11 | * Author: Kirill A. Shutemov | |
12 | * | |
8cdea7c0 BS |
13 | * This program is free software; you can redistribute it and/or modify |
14 | * it under the terms of the GNU General Public License as published by | |
15 | * the Free Software Foundation; either version 2 of the License, or | |
16 | * (at your option) any later version. | |
17 | * | |
18 | * This program is distributed in the hope that it will be useful, | |
19 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
20 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
21 | * GNU General Public License for more details. | |
22 | */ | |
23 | ||
24 | #include <linux/res_counter.h> | |
25 | #include <linux/memcontrol.h> | |
26 | #include <linux/cgroup.h> | |
78fb7466 | 27 | #include <linux/mm.h> |
4ffef5fe | 28 | #include <linux/hugetlb.h> |
d13d1443 | 29 | #include <linux/pagemap.h> |
d52aa412 | 30 | #include <linux/smp.h> |
8a9f3ccd | 31 | #include <linux/page-flags.h> |
66e1707b | 32 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
33 | #include <linux/bit_spinlock.h> |
34 | #include <linux/rcupdate.h> | |
e222432b | 35 | #include <linux/limits.h> |
b9e15baf | 36 | #include <linux/export.h> |
8c7c6e34 | 37 | #include <linux/mutex.h> |
f64c3f54 | 38 | #include <linux/rbtree.h> |
b6ac57d5 | 39 | #include <linux/slab.h> |
66e1707b | 40 | #include <linux/swap.h> |
02491447 | 41 | #include <linux/swapops.h> |
66e1707b | 42 | #include <linux/spinlock.h> |
2e72b634 KS |
43 | #include <linux/eventfd.h> |
44 | #include <linux/sort.h> | |
66e1707b | 45 | #include <linux/fs.h> |
d2ceb9b7 | 46 | #include <linux/seq_file.h> |
33327948 | 47 | #include <linux/vmalloc.h> |
b69408e8 | 48 | #include <linux/mm_inline.h> |
52d4b9ac | 49 | #include <linux/page_cgroup.h> |
cdec2e42 | 50 | #include <linux/cpu.h> |
158e0a2d | 51 | #include <linux/oom.h> |
08e552c6 | 52 | #include "internal.h" |
d1a4c0b3 GC |
53 | #include <net/sock.h> |
54 | #include <net/tcp_memcontrol.h> | |
8cdea7c0 | 55 | |
8697d331 BS |
56 | #include <asm/uaccess.h> |
57 | ||
cc8e970c KM |
58 | #include <trace/events/vmscan.h> |
59 | ||
a181b0e8 | 60 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
a181b0e8 | 61 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
4b3bde4c | 62 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
8cdea7c0 | 63 | |
c077719b | 64 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
338c8431 | 65 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ |
c077719b | 66 | int do_swap_account __read_mostly; |
a42c390c MH |
67 | |
68 | /* for remember boot option*/ | |
69 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED | |
70 | static int really_do_swap_account __initdata = 1; | |
71 | #else | |
72 | static int really_do_swap_account __initdata = 0; | |
73 | #endif | |
74 | ||
c077719b KH |
75 | #else |
76 | #define do_swap_account (0) | |
77 | #endif | |
78 | ||
79 | ||
d52aa412 KH |
80 | /* |
81 | * Statistics for memory cgroup. | |
82 | */ | |
83 | enum mem_cgroup_stat_index { | |
84 | /* | |
85 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
86 | */ | |
87 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
d69b042f | 88 | MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */ |
d8046582 | 89 | MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */ |
0c3e73e8 | 90 | MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */ |
711d3d2c | 91 | MEM_CGROUP_STAT_DATA, /* end of data requires synchronization */ |
32047e2a | 92 | MEM_CGROUP_ON_MOVE, /* someone is moving account between groups */ |
d52aa412 KH |
93 | MEM_CGROUP_STAT_NSTATS, |
94 | }; | |
95 | ||
e9f8974f JW |
96 | enum mem_cgroup_events_index { |
97 | MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */ | |
98 | MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */ | |
99 | MEM_CGROUP_EVENTS_COUNT, /* # of pages paged in/out */ | |
456f998e YH |
100 | MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */ |
101 | MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */ | |
e9f8974f JW |
102 | MEM_CGROUP_EVENTS_NSTATS, |
103 | }; | |
7a159cc9 JW |
104 | /* |
105 | * Per memcg event counter is incremented at every pagein/pageout. With THP, | |
106 | * it will be incremated by the number of pages. This counter is used for | |
107 | * for trigger some periodic events. This is straightforward and better | |
108 | * than using jiffies etc. to handle periodic memcg event. | |
109 | */ | |
110 | enum mem_cgroup_events_target { | |
111 | MEM_CGROUP_TARGET_THRESH, | |
112 | MEM_CGROUP_TARGET_SOFTLIMIT, | |
453a9bf3 | 113 | MEM_CGROUP_TARGET_NUMAINFO, |
7a159cc9 JW |
114 | MEM_CGROUP_NTARGETS, |
115 | }; | |
116 | #define THRESHOLDS_EVENTS_TARGET (128) | |
117 | #define SOFTLIMIT_EVENTS_TARGET (1024) | |
453a9bf3 | 118 | #define NUMAINFO_EVENTS_TARGET (1024) |
e9f8974f | 119 | |
d52aa412 | 120 | struct mem_cgroup_stat_cpu { |
7a159cc9 | 121 | long count[MEM_CGROUP_STAT_NSTATS]; |
e9f8974f | 122 | unsigned long events[MEM_CGROUP_EVENTS_NSTATS]; |
7a159cc9 | 123 | unsigned long targets[MEM_CGROUP_NTARGETS]; |
d52aa412 KH |
124 | }; |
125 | ||
527a5ec9 JW |
126 | struct mem_cgroup_reclaim_iter { |
127 | /* css_id of the last scanned hierarchy member */ | |
128 | int position; | |
129 | /* scan generation, increased every round-trip */ | |
130 | unsigned int generation; | |
131 | }; | |
132 | ||
6d12e2d8 KH |
133 | /* |
134 | * per-zone information in memory controller. | |
135 | */ | |
6d12e2d8 | 136 | struct mem_cgroup_per_zone { |
6290df54 | 137 | struct lruvec lruvec; |
b69408e8 | 138 | unsigned long count[NR_LRU_LISTS]; |
3e2f41f1 | 139 | |
527a5ec9 JW |
140 | struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1]; |
141 | ||
3e2f41f1 | 142 | struct zone_reclaim_stat reclaim_stat; |
f64c3f54 BS |
143 | struct rb_node tree_node; /* RB tree node */ |
144 | unsigned long long usage_in_excess;/* Set to the value by which */ | |
145 | /* the soft limit is exceeded*/ | |
146 | bool on_tree; | |
4e416953 BS |
147 | struct mem_cgroup *mem; /* Back pointer, we cannot */ |
148 | /* use container_of */ | |
6d12e2d8 KH |
149 | }; |
150 | /* Macro for accessing counter */ | |
151 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) | |
152 | ||
153 | struct mem_cgroup_per_node { | |
154 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
155 | }; | |
156 | ||
157 | struct mem_cgroup_lru_info { | |
158 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
159 | }; | |
160 | ||
f64c3f54 BS |
161 | /* |
162 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
163 | * their hierarchy representation | |
164 | */ | |
165 | ||
166 | struct mem_cgroup_tree_per_zone { | |
167 | struct rb_root rb_root; | |
168 | spinlock_t lock; | |
169 | }; | |
170 | ||
171 | struct mem_cgroup_tree_per_node { | |
172 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
173 | }; | |
174 | ||
175 | struct mem_cgroup_tree { | |
176 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
177 | }; | |
178 | ||
179 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
180 | ||
2e72b634 KS |
181 | struct mem_cgroup_threshold { |
182 | struct eventfd_ctx *eventfd; | |
183 | u64 threshold; | |
184 | }; | |
185 | ||
9490ff27 | 186 | /* For threshold */ |
2e72b634 KS |
187 | struct mem_cgroup_threshold_ary { |
188 | /* An array index points to threshold just below usage. */ | |
5407a562 | 189 | int current_threshold; |
2e72b634 KS |
190 | /* Size of entries[] */ |
191 | unsigned int size; | |
192 | /* Array of thresholds */ | |
193 | struct mem_cgroup_threshold entries[0]; | |
194 | }; | |
2c488db2 KS |
195 | |
196 | struct mem_cgroup_thresholds { | |
197 | /* Primary thresholds array */ | |
198 | struct mem_cgroup_threshold_ary *primary; | |
199 | /* | |
200 | * Spare threshold array. | |
201 | * This is needed to make mem_cgroup_unregister_event() "never fail". | |
202 | * It must be able to store at least primary->size - 1 entries. | |
203 | */ | |
204 | struct mem_cgroup_threshold_ary *spare; | |
205 | }; | |
206 | ||
9490ff27 KH |
207 | /* for OOM */ |
208 | struct mem_cgroup_eventfd_list { | |
209 | struct list_head list; | |
210 | struct eventfd_ctx *eventfd; | |
211 | }; | |
2e72b634 | 212 | |
c0ff4b85 R |
213 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
214 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 215 | |
8cdea7c0 BS |
216 | /* |
217 | * The memory controller data structure. The memory controller controls both | |
218 | * page cache and RSS per cgroup. We would eventually like to provide | |
219 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
220 | * to help the administrator determine what knobs to tune. | |
221 | * | |
222 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
223 | * we hit the water mark. May be even add a low water mark, such that |
224 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
225 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
226 | */ |
227 | struct mem_cgroup { | |
228 | struct cgroup_subsys_state css; | |
229 | /* | |
230 | * the counter to account for memory usage | |
231 | */ | |
232 | struct res_counter res; | |
8c7c6e34 KH |
233 | /* |
234 | * the counter to account for mem+swap usage. | |
235 | */ | |
236 | struct res_counter memsw; | |
78fb7466 PE |
237 | /* |
238 | * Per cgroup active and inactive list, similar to the | |
239 | * per zone LRU lists. | |
78fb7466 | 240 | */ |
6d12e2d8 | 241 | struct mem_cgroup_lru_info info; |
889976db YH |
242 | int last_scanned_node; |
243 | #if MAX_NUMNODES > 1 | |
244 | nodemask_t scan_nodes; | |
453a9bf3 KH |
245 | atomic_t numainfo_events; |
246 | atomic_t numainfo_updating; | |
889976db | 247 | #endif |
18f59ea7 BS |
248 | /* |
249 | * Should the accounting and control be hierarchical, per subtree? | |
250 | */ | |
251 | bool use_hierarchy; | |
79dfdacc MH |
252 | |
253 | bool oom_lock; | |
254 | atomic_t under_oom; | |
255 | ||
8c7c6e34 | 256 | atomic_t refcnt; |
14797e23 | 257 | |
1f4c025b | 258 | int swappiness; |
3c11ecf4 KH |
259 | /* OOM-Killer disable */ |
260 | int oom_kill_disable; | |
a7885eb8 | 261 | |
22a668d7 KH |
262 | /* set when res.limit == memsw.limit */ |
263 | bool memsw_is_minimum; | |
264 | ||
2e72b634 KS |
265 | /* protect arrays of thresholds */ |
266 | struct mutex thresholds_lock; | |
267 | ||
268 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 269 | struct mem_cgroup_thresholds thresholds; |
907860ed | 270 | |
2e72b634 | 271 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 272 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 273 | |
9490ff27 KH |
274 | /* For oom notifier event fd */ |
275 | struct list_head oom_notify; | |
185efc0f | 276 | |
7dc74be0 DN |
277 | /* |
278 | * Should we move charges of a task when a task is moved into this | |
279 | * mem_cgroup ? And what type of charges should we move ? | |
280 | */ | |
281 | unsigned long move_charge_at_immigrate; | |
d52aa412 | 282 | /* |
c62b1a3b | 283 | * percpu counter. |
d52aa412 | 284 | */ |
c62b1a3b | 285 | struct mem_cgroup_stat_cpu *stat; |
711d3d2c KH |
286 | /* |
287 | * used when a cpu is offlined or other synchronizations | |
288 | * See mem_cgroup_read_stat(). | |
289 | */ | |
290 | struct mem_cgroup_stat_cpu nocpu_base; | |
291 | spinlock_t pcp_counter_lock; | |
d1a4c0b3 GC |
292 | |
293 | #ifdef CONFIG_INET | |
294 | struct tcp_memcontrol tcp_mem; | |
295 | #endif | |
8cdea7c0 BS |
296 | }; |
297 | ||
7dc74be0 DN |
298 | /* Stuffs for move charges at task migration. */ |
299 | /* | |
300 | * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a | |
301 | * left-shifted bitmap of these types. | |
302 | */ | |
303 | enum move_type { | |
4ffef5fe | 304 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 305 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
306 | NR_MOVE_TYPE, |
307 | }; | |
308 | ||
4ffef5fe DN |
309 | /* "mc" and its members are protected by cgroup_mutex */ |
310 | static struct move_charge_struct { | |
b1dd693e | 311 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
312 | struct mem_cgroup *from; |
313 | struct mem_cgroup *to; | |
314 | unsigned long precharge; | |
854ffa8d | 315 | unsigned long moved_charge; |
483c30b5 | 316 | unsigned long moved_swap; |
8033b97c DN |
317 | struct task_struct *moving_task; /* a task moving charges */ |
318 | wait_queue_head_t waitq; /* a waitq for other context */ | |
319 | } mc = { | |
2bd9bb20 | 320 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
321 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
322 | }; | |
4ffef5fe | 323 | |
90254a65 DN |
324 | static bool move_anon(void) |
325 | { | |
326 | return test_bit(MOVE_CHARGE_TYPE_ANON, | |
327 | &mc.to->move_charge_at_immigrate); | |
328 | } | |
329 | ||
87946a72 DN |
330 | static bool move_file(void) |
331 | { | |
332 | return test_bit(MOVE_CHARGE_TYPE_FILE, | |
333 | &mc.to->move_charge_at_immigrate); | |
334 | } | |
335 | ||
4e416953 BS |
336 | /* |
337 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
338 | * limit reclaim to prevent infinite loops, if they ever occur. | |
339 | */ | |
340 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100) | |
341 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2) | |
342 | ||
217bc319 KH |
343 | enum charge_type { |
344 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
345 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
4f98a2fe | 346 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
c05555b5 | 347 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
d13d1443 | 348 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 349 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
350 | NR_CHARGE_TYPE, |
351 | }; | |
352 | ||
8c7c6e34 | 353 | /* for encoding cft->private value on file */ |
65c64ce8 GC |
354 | #define _MEM (0) |
355 | #define _MEMSWAP (1) | |
356 | #define _OOM_TYPE (2) | |
8c7c6e34 KH |
357 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) |
358 | #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) | |
359 | #define MEMFILE_ATTR(val) ((val) & 0xffff) | |
9490ff27 KH |
360 | /* Used for OOM nofiier */ |
361 | #define OOM_CONTROL (0) | |
8c7c6e34 | 362 | |
75822b44 BS |
363 | /* |
364 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
365 | */ | |
366 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
367 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
368 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
369 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
370 | ||
c0ff4b85 R |
371 | static void mem_cgroup_get(struct mem_cgroup *memcg); |
372 | static void mem_cgroup_put(struct mem_cgroup *memcg); | |
e1aab161 GC |
373 | |
374 | /* Writing them here to avoid exposing memcg's inner layout */ | |
375 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM | |
376 | #ifdef CONFIG_INET | |
377 | #include <net/sock.h> | |
d1a4c0b3 | 378 | #include <net/ip.h> |
e1aab161 GC |
379 | |
380 | static bool mem_cgroup_is_root(struct mem_cgroup *memcg); | |
381 | void sock_update_memcg(struct sock *sk) | |
382 | { | |
e1aab161 GC |
383 | if (static_branch(&memcg_socket_limit_enabled)) { |
384 | struct mem_cgroup *memcg; | |
385 | ||
386 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
387 | ||
f3f511e1 GC |
388 | /* Socket cloning can throw us here with sk_cgrp already |
389 | * filled. It won't however, necessarily happen from | |
390 | * process context. So the test for root memcg given | |
391 | * the current task's memcg won't help us in this case. | |
392 | * | |
393 | * Respecting the original socket's memcg is a better | |
394 | * decision in this case. | |
395 | */ | |
396 | if (sk->sk_cgrp) { | |
397 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
398 | mem_cgroup_get(sk->sk_cgrp->memcg); | |
399 | return; | |
400 | } | |
401 | ||
e1aab161 GC |
402 | rcu_read_lock(); |
403 | memcg = mem_cgroup_from_task(current); | |
404 | if (!mem_cgroup_is_root(memcg)) { | |
405 | mem_cgroup_get(memcg); | |
406 | sk->sk_cgrp = sk->sk_prot->proto_cgroup(memcg); | |
407 | } | |
408 | rcu_read_unlock(); | |
409 | } | |
410 | } | |
411 | EXPORT_SYMBOL(sock_update_memcg); | |
412 | ||
413 | void sock_release_memcg(struct sock *sk) | |
414 | { | |
415 | if (static_branch(&memcg_socket_limit_enabled) && sk->sk_cgrp) { | |
416 | struct mem_cgroup *memcg; | |
417 | WARN_ON(!sk->sk_cgrp->memcg); | |
418 | memcg = sk->sk_cgrp->memcg; | |
419 | mem_cgroup_put(memcg); | |
420 | } | |
421 | } | |
d1a4c0b3 GC |
422 | |
423 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
424 | { | |
425 | if (!memcg || mem_cgroup_is_root(memcg)) | |
426 | return NULL; | |
427 | ||
428 | return &memcg->tcp_mem.cg_proto; | |
429 | } | |
430 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 GC |
431 | #endif /* CONFIG_INET */ |
432 | #endif /* CONFIG_CGROUP_MEM_RES_CTLR_KMEM */ | |
433 | ||
c0ff4b85 | 434 | static void drain_all_stock_async(struct mem_cgroup *memcg); |
8c7c6e34 | 435 | |
f64c3f54 | 436 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 437 | mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid) |
f64c3f54 | 438 | { |
c0ff4b85 | 439 | return &memcg->info.nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
440 | } |
441 | ||
c0ff4b85 | 442 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 443 | { |
c0ff4b85 | 444 | return &memcg->css; |
d324236b WF |
445 | } |
446 | ||
f64c3f54 | 447 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 448 | page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 449 | { |
97a6c37b JW |
450 | int nid = page_to_nid(page); |
451 | int zid = page_zonenum(page); | |
f64c3f54 | 452 | |
c0ff4b85 | 453 | return mem_cgroup_zoneinfo(memcg, nid, zid); |
f64c3f54 BS |
454 | } |
455 | ||
456 | static struct mem_cgroup_tree_per_zone * | |
457 | soft_limit_tree_node_zone(int nid, int zid) | |
458 | { | |
459 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
460 | } | |
461 | ||
462 | static struct mem_cgroup_tree_per_zone * | |
463 | soft_limit_tree_from_page(struct page *page) | |
464 | { | |
465 | int nid = page_to_nid(page); | |
466 | int zid = page_zonenum(page); | |
467 | ||
468 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
469 | } | |
470 | ||
471 | static void | |
c0ff4b85 | 472 | __mem_cgroup_insert_exceeded(struct mem_cgroup *memcg, |
f64c3f54 | 473 | struct mem_cgroup_per_zone *mz, |
ef8745c1 KH |
474 | struct mem_cgroup_tree_per_zone *mctz, |
475 | unsigned long long new_usage_in_excess) | |
f64c3f54 BS |
476 | { |
477 | struct rb_node **p = &mctz->rb_root.rb_node; | |
478 | struct rb_node *parent = NULL; | |
479 | struct mem_cgroup_per_zone *mz_node; | |
480 | ||
481 | if (mz->on_tree) | |
482 | return; | |
483 | ||
ef8745c1 KH |
484 | mz->usage_in_excess = new_usage_in_excess; |
485 | if (!mz->usage_in_excess) | |
486 | return; | |
f64c3f54 BS |
487 | while (*p) { |
488 | parent = *p; | |
489 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
490 | tree_node); | |
491 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
492 | p = &(*p)->rb_left; | |
493 | /* | |
494 | * We can't avoid mem cgroups that are over their soft | |
495 | * limit by the same amount | |
496 | */ | |
497 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
498 | p = &(*p)->rb_right; | |
499 | } | |
500 | rb_link_node(&mz->tree_node, parent, p); | |
501 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
502 | mz->on_tree = true; | |
4e416953 BS |
503 | } |
504 | ||
505 | static void | |
c0ff4b85 | 506 | __mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
4e416953 BS |
507 | struct mem_cgroup_per_zone *mz, |
508 | struct mem_cgroup_tree_per_zone *mctz) | |
509 | { | |
510 | if (!mz->on_tree) | |
511 | return; | |
512 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
513 | mz->on_tree = false; | |
514 | } | |
515 | ||
f64c3f54 | 516 | static void |
c0ff4b85 | 517 | mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
f64c3f54 BS |
518 | struct mem_cgroup_per_zone *mz, |
519 | struct mem_cgroup_tree_per_zone *mctz) | |
520 | { | |
521 | spin_lock(&mctz->lock); | |
c0ff4b85 | 522 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
523 | spin_unlock(&mctz->lock); |
524 | } | |
525 | ||
f64c3f54 | 526 | |
c0ff4b85 | 527 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 528 | { |
ef8745c1 | 529 | unsigned long long excess; |
f64c3f54 BS |
530 | struct mem_cgroup_per_zone *mz; |
531 | struct mem_cgroup_tree_per_zone *mctz; | |
4e649152 KH |
532 | int nid = page_to_nid(page); |
533 | int zid = page_zonenum(page); | |
f64c3f54 BS |
534 | mctz = soft_limit_tree_from_page(page); |
535 | ||
536 | /* | |
4e649152 KH |
537 | * Necessary to update all ancestors when hierarchy is used. |
538 | * because their event counter is not touched. | |
f64c3f54 | 539 | */ |
c0ff4b85 R |
540 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
541 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
542 | excess = res_counter_soft_limit_excess(&memcg->res); | |
4e649152 KH |
543 | /* |
544 | * We have to update the tree if mz is on RB-tree or | |
545 | * mem is over its softlimit. | |
546 | */ | |
ef8745c1 | 547 | if (excess || mz->on_tree) { |
4e649152 KH |
548 | spin_lock(&mctz->lock); |
549 | /* if on-tree, remove it */ | |
550 | if (mz->on_tree) | |
c0ff4b85 | 551 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
4e649152 | 552 | /* |
ef8745c1 KH |
553 | * Insert again. mz->usage_in_excess will be updated. |
554 | * If excess is 0, no tree ops. | |
4e649152 | 555 | */ |
c0ff4b85 | 556 | __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess); |
4e649152 KH |
557 | spin_unlock(&mctz->lock); |
558 | } | |
f64c3f54 BS |
559 | } |
560 | } | |
561 | ||
c0ff4b85 | 562 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) |
f64c3f54 BS |
563 | { |
564 | int node, zone; | |
565 | struct mem_cgroup_per_zone *mz; | |
566 | struct mem_cgroup_tree_per_zone *mctz; | |
567 | ||
568 | for_each_node_state(node, N_POSSIBLE) { | |
569 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
c0ff4b85 | 570 | mz = mem_cgroup_zoneinfo(memcg, node, zone); |
f64c3f54 | 571 | mctz = soft_limit_tree_node_zone(node, zone); |
c0ff4b85 | 572 | mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
573 | } |
574 | } | |
575 | } | |
576 | ||
4e416953 BS |
577 | static struct mem_cgroup_per_zone * |
578 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
579 | { | |
580 | struct rb_node *rightmost = NULL; | |
26251eaf | 581 | struct mem_cgroup_per_zone *mz; |
4e416953 BS |
582 | |
583 | retry: | |
26251eaf | 584 | mz = NULL; |
4e416953 BS |
585 | rightmost = rb_last(&mctz->rb_root); |
586 | if (!rightmost) | |
587 | goto done; /* Nothing to reclaim from */ | |
588 | ||
589 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
590 | /* | |
591 | * Remove the node now but someone else can add it back, | |
592 | * we will to add it back at the end of reclaim to its correct | |
593 | * position in the tree. | |
594 | */ | |
595 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); | |
596 | if (!res_counter_soft_limit_excess(&mz->mem->res) || | |
597 | !css_tryget(&mz->mem->css)) | |
598 | goto retry; | |
599 | done: | |
600 | return mz; | |
601 | } | |
602 | ||
603 | static struct mem_cgroup_per_zone * | |
604 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
605 | { | |
606 | struct mem_cgroup_per_zone *mz; | |
607 | ||
608 | spin_lock(&mctz->lock); | |
609 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
610 | spin_unlock(&mctz->lock); | |
611 | return mz; | |
612 | } | |
613 | ||
711d3d2c KH |
614 | /* |
615 | * Implementation Note: reading percpu statistics for memcg. | |
616 | * | |
617 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
618 | * synchronization to implement "quick" read. There are trade-off between | |
619 | * reading cost and precision of value. Then, we may have a chance to implement | |
620 | * a periodic synchronizion of counter in memcg's counter. | |
621 | * | |
622 | * But this _read() function is used for user interface now. The user accounts | |
623 | * memory usage by memory cgroup and he _always_ requires exact value because | |
624 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
625 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
626 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
627 | * | |
628 | * If there are kernel internal actions which can make use of some not-exact | |
629 | * value, and reading all cpu value can be performance bottleneck in some | |
630 | * common workload, threashold and synchonization as vmstat[] should be | |
631 | * implemented. | |
632 | */ | |
c0ff4b85 | 633 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 634 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 635 | { |
7a159cc9 | 636 | long val = 0; |
c62b1a3b | 637 | int cpu; |
c62b1a3b | 638 | |
711d3d2c KH |
639 | get_online_cpus(); |
640 | for_each_online_cpu(cpu) | |
c0ff4b85 | 641 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 642 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
643 | spin_lock(&memcg->pcp_counter_lock); |
644 | val += memcg->nocpu_base.count[idx]; | |
645 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
646 | #endif |
647 | put_online_cpus(); | |
c62b1a3b KH |
648 | return val; |
649 | } | |
650 | ||
c0ff4b85 | 651 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
0c3e73e8 BS |
652 | bool charge) |
653 | { | |
654 | int val = (charge) ? 1 : -1; | |
c0ff4b85 | 655 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAPOUT], val); |
0c3e73e8 BS |
656 | } |
657 | ||
c0ff4b85 | 658 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
659 | enum mem_cgroup_events_index idx) |
660 | { | |
661 | unsigned long val = 0; | |
662 | int cpu; | |
663 | ||
664 | for_each_online_cpu(cpu) | |
c0ff4b85 | 665 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 666 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
667 | spin_lock(&memcg->pcp_counter_lock); |
668 | val += memcg->nocpu_base.events[idx]; | |
669 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f JW |
670 | #endif |
671 | return val; | |
672 | } | |
673 | ||
c0ff4b85 | 674 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
e401f176 | 675 | bool file, int nr_pages) |
d52aa412 | 676 | { |
c62b1a3b KH |
677 | preempt_disable(); |
678 | ||
e401f176 | 679 | if (file) |
c0ff4b85 R |
680 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
681 | nr_pages); | |
d52aa412 | 682 | else |
c0ff4b85 R |
683 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
684 | nr_pages); | |
55e462b0 | 685 | |
e401f176 KH |
686 | /* pagein of a big page is an event. So, ignore page size */ |
687 | if (nr_pages > 0) | |
c0ff4b85 | 688 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 689 | else { |
c0ff4b85 | 690 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
691 | nr_pages = -nr_pages; /* for event */ |
692 | } | |
e401f176 | 693 | |
c0ff4b85 | 694 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_COUNT], nr_pages); |
2e72b634 | 695 | |
c62b1a3b | 696 | preempt_enable(); |
6d12e2d8 KH |
697 | } |
698 | ||
bb2a0de9 | 699 | unsigned long |
c0ff4b85 | 700 | mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid, |
bb2a0de9 | 701 | unsigned int lru_mask) |
889976db YH |
702 | { |
703 | struct mem_cgroup_per_zone *mz; | |
bb2a0de9 KH |
704 | enum lru_list l; |
705 | unsigned long ret = 0; | |
706 | ||
c0ff4b85 | 707 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
bb2a0de9 KH |
708 | |
709 | for_each_lru(l) { | |
710 | if (BIT(l) & lru_mask) | |
711 | ret += MEM_CGROUP_ZSTAT(mz, l); | |
712 | } | |
713 | return ret; | |
714 | } | |
715 | ||
716 | static unsigned long | |
c0ff4b85 | 717 | mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 KH |
718 | int nid, unsigned int lru_mask) |
719 | { | |
889976db YH |
720 | u64 total = 0; |
721 | int zid; | |
722 | ||
bb2a0de9 | 723 | for (zid = 0; zid < MAX_NR_ZONES; zid++) |
c0ff4b85 R |
724 | total += mem_cgroup_zone_nr_lru_pages(memcg, |
725 | nid, zid, lru_mask); | |
bb2a0de9 | 726 | |
889976db YH |
727 | return total; |
728 | } | |
bb2a0de9 | 729 | |
c0ff4b85 | 730 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 731 | unsigned int lru_mask) |
6d12e2d8 | 732 | { |
889976db | 733 | int nid; |
6d12e2d8 KH |
734 | u64 total = 0; |
735 | ||
bb2a0de9 | 736 | for_each_node_state(nid, N_HIGH_MEMORY) |
c0ff4b85 | 737 | total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
6d12e2d8 | 738 | return total; |
d52aa412 KH |
739 | } |
740 | ||
f53d7ce3 JW |
741 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
742 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
743 | { |
744 | unsigned long val, next; | |
745 | ||
4799401f SR |
746 | val = __this_cpu_read(memcg->stat->events[MEM_CGROUP_EVENTS_COUNT]); |
747 | next = __this_cpu_read(memcg->stat->targets[target]); | |
7a159cc9 | 748 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
749 | if ((long)next - (long)val < 0) { |
750 | switch (target) { | |
751 | case MEM_CGROUP_TARGET_THRESH: | |
752 | next = val + THRESHOLDS_EVENTS_TARGET; | |
753 | break; | |
754 | case MEM_CGROUP_TARGET_SOFTLIMIT: | |
755 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
756 | break; | |
757 | case MEM_CGROUP_TARGET_NUMAINFO: | |
758 | next = val + NUMAINFO_EVENTS_TARGET; | |
759 | break; | |
760 | default: | |
761 | break; | |
762 | } | |
763 | __this_cpu_write(memcg->stat->targets[target], next); | |
764 | return true; | |
7a159cc9 | 765 | } |
f53d7ce3 | 766 | return false; |
d2265e6f KH |
767 | } |
768 | ||
769 | /* | |
770 | * Check events in order. | |
771 | * | |
772 | */ | |
c0ff4b85 | 773 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f | 774 | { |
4799401f | 775 | preempt_disable(); |
d2265e6f | 776 | /* threshold event is triggered in finer grain than soft limit */ |
f53d7ce3 JW |
777 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
778 | MEM_CGROUP_TARGET_THRESH))) { | |
779 | bool do_softlimit, do_numainfo; | |
780 | ||
781 | do_softlimit = mem_cgroup_event_ratelimit(memcg, | |
782 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
783 | #if MAX_NUMNODES > 1 | |
784 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
785 | MEM_CGROUP_TARGET_NUMAINFO); | |
786 | #endif | |
787 | preempt_enable(); | |
788 | ||
c0ff4b85 | 789 | mem_cgroup_threshold(memcg); |
f53d7ce3 | 790 | if (unlikely(do_softlimit)) |
c0ff4b85 | 791 | mem_cgroup_update_tree(memcg, page); |
453a9bf3 | 792 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 793 | if (unlikely(do_numainfo)) |
c0ff4b85 | 794 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 795 | #endif |
f53d7ce3 JW |
796 | } else |
797 | preempt_enable(); | |
d2265e6f KH |
798 | } |
799 | ||
d1a4c0b3 | 800 | struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
801 | { |
802 | return container_of(cgroup_subsys_state(cont, | |
803 | mem_cgroup_subsys_id), struct mem_cgroup, | |
804 | css); | |
805 | } | |
806 | ||
cf475ad2 | 807 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 808 | { |
31a78f23 BS |
809 | /* |
810 | * mm_update_next_owner() may clear mm->owner to NULL | |
811 | * if it races with swapoff, page migration, etc. | |
812 | * So this can be called with p == NULL. | |
813 | */ | |
814 | if (unlikely(!p)) | |
815 | return NULL; | |
816 | ||
78fb7466 PE |
817 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
818 | struct mem_cgroup, css); | |
819 | } | |
820 | ||
a433658c | 821 | struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 822 | { |
c0ff4b85 | 823 | struct mem_cgroup *memcg = NULL; |
0b7f569e KH |
824 | |
825 | if (!mm) | |
826 | return NULL; | |
54595fe2 KH |
827 | /* |
828 | * Because we have no locks, mm->owner's may be being moved to other | |
829 | * cgroup. We use css_tryget() here even if this looks | |
830 | * pessimistic (rather than adding locks here). | |
831 | */ | |
832 | rcu_read_lock(); | |
833 | do { | |
c0ff4b85 R |
834 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
835 | if (unlikely(!memcg)) | |
54595fe2 | 836 | break; |
c0ff4b85 | 837 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 838 | rcu_read_unlock(); |
c0ff4b85 | 839 | return memcg; |
54595fe2 KH |
840 | } |
841 | ||
5660048c JW |
842 | /** |
843 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
844 | * @root: hierarchy root | |
845 | * @prev: previously returned memcg, NULL on first invocation | |
846 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
847 | * | |
848 | * Returns references to children of the hierarchy below @root, or | |
849 | * @root itself, or %NULL after a full round-trip. | |
850 | * | |
851 | * Caller must pass the return value in @prev on subsequent | |
852 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
853 | * to cancel a hierarchy walk before the round-trip is complete. | |
854 | * | |
855 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
856 | * divide up the memcgs in the hierarchy among all concurrent | |
857 | * reclaimers operating on the same zone and priority. | |
858 | */ | |
859 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, | |
860 | struct mem_cgroup *prev, | |
861 | struct mem_cgroup_reclaim_cookie *reclaim) | |
14067bb3 | 862 | { |
9f3a0d09 JW |
863 | struct mem_cgroup *memcg = NULL; |
864 | int id = 0; | |
711d3d2c | 865 | |
5660048c JW |
866 | if (mem_cgroup_disabled()) |
867 | return NULL; | |
868 | ||
9f3a0d09 JW |
869 | if (!root) |
870 | root = root_mem_cgroup; | |
7d74b06f | 871 | |
9f3a0d09 JW |
872 | if (prev && !reclaim) |
873 | id = css_id(&prev->css); | |
14067bb3 | 874 | |
9f3a0d09 JW |
875 | if (prev && prev != root) |
876 | css_put(&prev->css); | |
14067bb3 | 877 | |
9f3a0d09 JW |
878 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
879 | if (prev) | |
880 | return NULL; | |
881 | return root; | |
882 | } | |
14067bb3 | 883 | |
9f3a0d09 | 884 | while (!memcg) { |
527a5ec9 | 885 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
9f3a0d09 | 886 | struct cgroup_subsys_state *css; |
711d3d2c | 887 | |
527a5ec9 JW |
888 | if (reclaim) { |
889 | int nid = zone_to_nid(reclaim->zone); | |
890 | int zid = zone_idx(reclaim->zone); | |
891 | struct mem_cgroup_per_zone *mz; | |
892 | ||
893 | mz = mem_cgroup_zoneinfo(root, nid, zid); | |
894 | iter = &mz->reclaim_iter[reclaim->priority]; | |
895 | if (prev && reclaim->generation != iter->generation) | |
896 | return NULL; | |
897 | id = iter->position; | |
898 | } | |
7d74b06f | 899 | |
9f3a0d09 JW |
900 | rcu_read_lock(); |
901 | css = css_get_next(&mem_cgroup_subsys, id + 1, &root->css, &id); | |
902 | if (css) { | |
903 | if (css == &root->css || css_tryget(css)) | |
904 | memcg = container_of(css, | |
905 | struct mem_cgroup, css); | |
906 | } else | |
907 | id = 0; | |
14067bb3 | 908 | rcu_read_unlock(); |
14067bb3 | 909 | |
527a5ec9 JW |
910 | if (reclaim) { |
911 | iter->position = id; | |
912 | if (!css) | |
913 | iter->generation++; | |
914 | else if (!prev && memcg) | |
915 | reclaim->generation = iter->generation; | |
916 | } | |
9f3a0d09 JW |
917 | |
918 | if (prev && !css) | |
919 | return NULL; | |
920 | } | |
921 | return memcg; | |
14067bb3 | 922 | } |
7d74b06f | 923 | |
5660048c JW |
924 | /** |
925 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
926 | * @root: hierarchy root | |
927 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
928 | */ | |
929 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
930 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
931 | { |
932 | if (!root) | |
933 | root = root_mem_cgroup; | |
934 | if (prev && prev != root) | |
935 | css_put(&prev->css); | |
936 | } | |
7d74b06f | 937 | |
9f3a0d09 JW |
938 | /* |
939 | * Iteration constructs for visiting all cgroups (under a tree). If | |
940 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
941 | * be used for reference counting. | |
942 | */ | |
943 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 944 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 945 | iter != NULL; \ |
527a5ec9 | 946 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 947 | |
9f3a0d09 | 948 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 949 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 950 | iter != NULL; \ |
527a5ec9 | 951 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 952 | |
c0ff4b85 | 953 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
4b3bde4c | 954 | { |
c0ff4b85 | 955 | return (memcg == root_mem_cgroup); |
4b3bde4c BS |
956 | } |
957 | ||
456f998e YH |
958 | void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
959 | { | |
c0ff4b85 | 960 | struct mem_cgroup *memcg; |
456f998e YH |
961 | |
962 | if (!mm) | |
963 | return; | |
964 | ||
965 | rcu_read_lock(); | |
c0ff4b85 R |
966 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
967 | if (unlikely(!memcg)) | |
456f998e YH |
968 | goto out; |
969 | ||
970 | switch (idx) { | |
456f998e | 971 | case PGFAULT: |
0e574a93 JW |
972 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
973 | break; | |
974 | case PGMAJFAULT: | |
975 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
976 | break; |
977 | default: | |
978 | BUG(); | |
979 | } | |
980 | out: | |
981 | rcu_read_unlock(); | |
982 | } | |
983 | EXPORT_SYMBOL(mem_cgroup_count_vm_event); | |
984 | ||
925b7673 JW |
985 | /** |
986 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
987 | * @zone: zone of the wanted lruvec | |
988 | * @mem: memcg of the wanted lruvec | |
989 | * | |
990 | * Returns the lru list vector holding pages for the given @zone and | |
991 | * @mem. This can be the global zone lruvec, if the memory controller | |
992 | * is disabled. | |
993 | */ | |
994 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
995 | struct mem_cgroup *memcg) | |
996 | { | |
997 | struct mem_cgroup_per_zone *mz; | |
998 | ||
999 | if (mem_cgroup_disabled()) | |
1000 | return &zone->lruvec; | |
1001 | ||
1002 | mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone)); | |
1003 | return &mz->lruvec; | |
1004 | } | |
1005 | ||
08e552c6 KH |
1006 | /* |
1007 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
1008 | * Operations are called by routine of global LRU independently from memcg. | |
1009 | * What we have to take care of here is validness of pc->mem_cgroup. | |
1010 | * | |
1011 | * Changes to pc->mem_cgroup happens when | |
1012 | * 1. charge | |
1013 | * 2. moving account | |
1014 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
1015 | * It is added to LRU before charge. | |
1016 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
1017 | * When moving account, the page is not on LRU. It's isolated. | |
1018 | */ | |
4f98a2fe | 1019 | |
925b7673 JW |
1020 | /** |
1021 | * mem_cgroup_lru_add_list - account for adding an lru page and return lruvec | |
1022 | * @zone: zone of the page | |
1023 | * @page: the page | |
1024 | * @lru: current lru | |
1025 | * | |
1026 | * This function accounts for @page being added to @lru, and returns | |
1027 | * the lruvec for the given @zone and the memcg @page is charged to. | |
1028 | * | |
1029 | * The callsite is then responsible for physically linking the page to | |
1030 | * the returned lruvec->lists[@lru]. | |
1031 | */ | |
1032 | struct lruvec *mem_cgroup_lru_add_list(struct zone *zone, struct page *page, | |
1033 | enum lru_list lru) | |
08e552c6 | 1034 | { |
08e552c6 | 1035 | struct mem_cgroup_per_zone *mz; |
925b7673 JW |
1036 | struct mem_cgroup *memcg; |
1037 | struct page_cgroup *pc; | |
6d12e2d8 | 1038 | |
f8d66542 | 1039 | if (mem_cgroup_disabled()) |
925b7673 JW |
1040 | return &zone->lruvec; |
1041 | ||
08e552c6 | 1042 | pc = lookup_page_cgroup(page); |
38c5d72f | 1043 | memcg = pc->mem_cgroup; |
925b7673 JW |
1044 | mz = page_cgroup_zoneinfo(memcg, page); |
1045 | /* compound_order() is stabilized through lru_lock */ | |
1046 | MEM_CGROUP_ZSTAT(mz, lru) += 1 << compound_order(page); | |
1047 | return &mz->lruvec; | |
08e552c6 | 1048 | } |
b69408e8 | 1049 | |
925b7673 JW |
1050 | /** |
1051 | * mem_cgroup_lru_del_list - account for removing an lru page | |
1052 | * @page: the page | |
1053 | * @lru: target lru | |
1054 | * | |
1055 | * This function accounts for @page being removed from @lru. | |
1056 | * | |
1057 | * The callsite is then responsible for physically unlinking | |
1058 | * @page->lru. | |
3f58a829 | 1059 | */ |
925b7673 | 1060 | void mem_cgroup_lru_del_list(struct page *page, enum lru_list lru) |
3f58a829 MK |
1061 | { |
1062 | struct mem_cgroup_per_zone *mz; | |
925b7673 | 1063 | struct mem_cgroup *memcg; |
3f58a829 | 1064 | struct page_cgroup *pc; |
3f58a829 MK |
1065 | |
1066 | if (mem_cgroup_disabled()) | |
1067 | return; | |
1068 | ||
1069 | pc = lookup_page_cgroup(page); | |
38c5d72f KH |
1070 | memcg = pc->mem_cgroup; |
1071 | VM_BUG_ON(!memcg); | |
925b7673 JW |
1072 | mz = page_cgroup_zoneinfo(memcg, page); |
1073 | /* huge page split is done under lru_lock. so, we have no races. */ | |
1074 | MEM_CGROUP_ZSTAT(mz, lru) -= 1 << compound_order(page); | |
3f58a829 MK |
1075 | } |
1076 | ||
925b7673 | 1077 | void mem_cgroup_lru_del(struct page *page) |
08e552c6 | 1078 | { |
925b7673 | 1079 | mem_cgroup_lru_del_list(page, page_lru(page)); |
6d12e2d8 KH |
1080 | } |
1081 | ||
925b7673 JW |
1082 | /** |
1083 | * mem_cgroup_lru_move_lists - account for moving a page between lrus | |
1084 | * @zone: zone of the page | |
1085 | * @page: the page | |
1086 | * @from: current lru | |
1087 | * @to: target lru | |
1088 | * | |
1089 | * This function accounts for @page being moved between the lrus @from | |
1090 | * and @to, and returns the lruvec for the given @zone and the memcg | |
1091 | * @page is charged to. | |
1092 | * | |
1093 | * The callsite is then responsible for physically relinking | |
1094 | * @page->lru to the returned lruvec->lists[@to]. | |
1095 | */ | |
1096 | struct lruvec *mem_cgroup_lru_move_lists(struct zone *zone, | |
1097 | struct page *page, | |
1098 | enum lru_list from, | |
1099 | enum lru_list to) | |
66e1707b | 1100 | { |
925b7673 JW |
1101 | /* XXX: Optimize this, especially for @from == @to */ |
1102 | mem_cgroup_lru_del_list(page, from); | |
1103 | return mem_cgroup_lru_add_list(zone, page, to); | |
08e552c6 | 1104 | } |
544122e5 | 1105 | |
3e92041d | 1106 | /* |
c0ff4b85 | 1107 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1108 | * hierarchy subtree |
1109 | */ | |
c0ff4b85 R |
1110 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1111 | struct mem_cgroup *memcg) | |
3e92041d | 1112 | { |
c0ff4b85 R |
1113 | if (root_memcg != memcg) { |
1114 | return (root_memcg->use_hierarchy && | |
1115 | css_is_ancestor(&memcg->css, &root_memcg->css)); | |
3e92041d MH |
1116 | } |
1117 | ||
1118 | return true; | |
1119 | } | |
1120 | ||
c0ff4b85 | 1121 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg) |
4c4a2214 DR |
1122 | { |
1123 | int ret; | |
0b7f569e | 1124 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1125 | struct task_struct *p; |
4c4a2214 | 1126 | |
158e0a2d | 1127 | p = find_lock_task_mm(task); |
de077d22 DR |
1128 | if (p) { |
1129 | curr = try_get_mem_cgroup_from_mm(p->mm); | |
1130 | task_unlock(p); | |
1131 | } else { | |
1132 | /* | |
1133 | * All threads may have already detached their mm's, but the oom | |
1134 | * killer still needs to detect if they have already been oom | |
1135 | * killed to prevent needlessly killing additional tasks. | |
1136 | */ | |
1137 | task_lock(task); | |
1138 | curr = mem_cgroup_from_task(task); | |
1139 | if (curr) | |
1140 | css_get(&curr->css); | |
1141 | task_unlock(task); | |
1142 | } | |
0b7f569e KH |
1143 | if (!curr) |
1144 | return 0; | |
d31f56db | 1145 | /* |
c0ff4b85 | 1146 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1147 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1148 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1149 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1150 | */ |
c0ff4b85 | 1151 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1152 | css_put(&curr->css); |
4c4a2214 DR |
1153 | return ret; |
1154 | } | |
1155 | ||
9b272977 | 1156 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg, struct zone *zone) |
14797e23 | 1157 | { |
9b272977 JW |
1158 | unsigned long inactive_ratio; |
1159 | int nid = zone_to_nid(zone); | |
1160 | int zid = zone_idx(zone); | |
14797e23 | 1161 | unsigned long inactive; |
9b272977 | 1162 | unsigned long active; |
c772be93 | 1163 | unsigned long gb; |
14797e23 | 1164 | |
9b272977 JW |
1165 | inactive = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid, |
1166 | BIT(LRU_INACTIVE_ANON)); | |
1167 | active = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid, | |
1168 | BIT(LRU_ACTIVE_ANON)); | |
14797e23 | 1169 | |
c772be93 KM |
1170 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1171 | if (gb) | |
1172 | inactive_ratio = int_sqrt(10 * gb); | |
1173 | else | |
1174 | inactive_ratio = 1; | |
1175 | ||
9b272977 | 1176 | return inactive * inactive_ratio < active; |
14797e23 KM |
1177 | } |
1178 | ||
9b272977 | 1179 | int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg, struct zone *zone) |
56e49d21 RR |
1180 | { |
1181 | unsigned long active; | |
1182 | unsigned long inactive; | |
9b272977 JW |
1183 | int zid = zone_idx(zone); |
1184 | int nid = zone_to_nid(zone); | |
56e49d21 | 1185 | |
9b272977 JW |
1186 | inactive = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid, |
1187 | BIT(LRU_INACTIVE_FILE)); | |
1188 | active = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid, | |
1189 | BIT(LRU_ACTIVE_FILE)); | |
56e49d21 RR |
1190 | |
1191 | return (active > inactive); | |
1192 | } | |
1193 | ||
3e2f41f1 KM |
1194 | struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, |
1195 | struct zone *zone) | |
1196 | { | |
13d7e3a2 | 1197 | int nid = zone_to_nid(zone); |
3e2f41f1 KM |
1198 | int zid = zone_idx(zone); |
1199 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
1200 | ||
1201 | return &mz->reclaim_stat; | |
1202 | } | |
1203 | ||
1204 | struct zone_reclaim_stat * | |
1205 | mem_cgroup_get_reclaim_stat_from_page(struct page *page) | |
1206 | { | |
1207 | struct page_cgroup *pc; | |
1208 | struct mem_cgroup_per_zone *mz; | |
1209 | ||
1210 | if (mem_cgroup_disabled()) | |
1211 | return NULL; | |
1212 | ||
1213 | pc = lookup_page_cgroup(page); | |
bd112db8 DN |
1214 | if (!PageCgroupUsed(pc)) |
1215 | return NULL; | |
713735b4 JW |
1216 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ |
1217 | smp_rmb(); | |
97a6c37b | 1218 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
3e2f41f1 KM |
1219 | return &mz->reclaim_stat; |
1220 | } | |
1221 | ||
6d61ef40 BS |
1222 | #define mem_cgroup_from_res_counter(counter, member) \ |
1223 | container_of(counter, struct mem_cgroup, member) | |
1224 | ||
19942822 | 1225 | /** |
9d11ea9f JW |
1226 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
1227 | * @mem: the memory cgroup | |
19942822 | 1228 | * |
9d11ea9f | 1229 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1230 | * pages. |
19942822 | 1231 | */ |
c0ff4b85 | 1232 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1233 | { |
9d11ea9f JW |
1234 | unsigned long long margin; |
1235 | ||
c0ff4b85 | 1236 | margin = res_counter_margin(&memcg->res); |
9d11ea9f | 1237 | if (do_swap_account) |
c0ff4b85 | 1238 | margin = min(margin, res_counter_margin(&memcg->memsw)); |
7ec99d62 | 1239 | return margin >> PAGE_SHIFT; |
19942822 JW |
1240 | } |
1241 | ||
1f4c025b | 1242 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 KM |
1243 | { |
1244 | struct cgroup *cgrp = memcg->css.cgroup; | |
a7885eb8 KM |
1245 | |
1246 | /* root ? */ | |
1247 | if (cgrp->parent == NULL) | |
1248 | return vm_swappiness; | |
1249 | ||
bf1ff263 | 1250 | return memcg->swappiness; |
a7885eb8 KM |
1251 | } |
1252 | ||
c0ff4b85 | 1253 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a KH |
1254 | { |
1255 | int cpu; | |
1489ebad KH |
1256 | |
1257 | get_online_cpus(); | |
c0ff4b85 | 1258 | spin_lock(&memcg->pcp_counter_lock); |
1489ebad | 1259 | for_each_online_cpu(cpu) |
c0ff4b85 R |
1260 | per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) += 1; |
1261 | memcg->nocpu_base.count[MEM_CGROUP_ON_MOVE] += 1; | |
1262 | spin_unlock(&memcg->pcp_counter_lock); | |
1489ebad | 1263 | put_online_cpus(); |
32047e2a KH |
1264 | |
1265 | synchronize_rcu(); | |
1266 | } | |
1267 | ||
c0ff4b85 | 1268 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a KH |
1269 | { |
1270 | int cpu; | |
1271 | ||
c0ff4b85 | 1272 | if (!memcg) |
32047e2a | 1273 | return; |
1489ebad | 1274 | get_online_cpus(); |
c0ff4b85 | 1275 | spin_lock(&memcg->pcp_counter_lock); |
1489ebad | 1276 | for_each_online_cpu(cpu) |
c0ff4b85 R |
1277 | per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) -= 1; |
1278 | memcg->nocpu_base.count[MEM_CGROUP_ON_MOVE] -= 1; | |
1279 | spin_unlock(&memcg->pcp_counter_lock); | |
1489ebad | 1280 | put_online_cpus(); |
32047e2a KH |
1281 | } |
1282 | /* | |
1283 | * 2 routines for checking "mem" is under move_account() or not. | |
1284 | * | |
1285 | * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used | |
1286 | * for avoiding race in accounting. If true, | |
1287 | * pc->mem_cgroup may be overwritten. | |
1288 | * | |
1289 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or | |
1290 | * under hierarchy of moving cgroups. This is for | |
1291 | * waiting at hith-memory prressure caused by "move". | |
1292 | */ | |
1293 | ||
c0ff4b85 | 1294 | static bool mem_cgroup_stealed(struct mem_cgroup *memcg) |
32047e2a KH |
1295 | { |
1296 | VM_BUG_ON(!rcu_read_lock_held()); | |
c0ff4b85 | 1297 | return this_cpu_read(memcg->stat->count[MEM_CGROUP_ON_MOVE]) > 0; |
32047e2a | 1298 | } |
4b534334 | 1299 | |
c0ff4b85 | 1300 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1301 | { |
2bd9bb20 KH |
1302 | struct mem_cgroup *from; |
1303 | struct mem_cgroup *to; | |
4b534334 | 1304 | bool ret = false; |
2bd9bb20 KH |
1305 | /* |
1306 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1307 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1308 | */ | |
1309 | spin_lock(&mc.lock); | |
1310 | from = mc.from; | |
1311 | to = mc.to; | |
1312 | if (!from) | |
1313 | goto unlock; | |
3e92041d | 1314 | |
c0ff4b85 R |
1315 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1316 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1317 | unlock: |
1318 | spin_unlock(&mc.lock); | |
4b534334 KH |
1319 | return ret; |
1320 | } | |
1321 | ||
c0ff4b85 | 1322 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1323 | { |
1324 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1325 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1326 | DEFINE_WAIT(wait); |
1327 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1328 | /* moving charge context might have finished. */ | |
1329 | if (mc.moving_task) | |
1330 | schedule(); | |
1331 | finish_wait(&mc.waitq, &wait); | |
1332 | return true; | |
1333 | } | |
1334 | } | |
1335 | return false; | |
1336 | } | |
1337 | ||
e222432b | 1338 | /** |
6a6135b6 | 1339 | * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode. |
e222432b BS |
1340 | * @memcg: The memory cgroup that went over limit |
1341 | * @p: Task that is going to be killed | |
1342 | * | |
1343 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1344 | * enabled | |
1345 | */ | |
1346 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1347 | { | |
1348 | struct cgroup *task_cgrp; | |
1349 | struct cgroup *mem_cgrp; | |
1350 | /* | |
1351 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1352 | * on the assumption that OOM is serialized for memory controller. | |
1353 | * If this assumption is broken, revisit this code. | |
1354 | */ | |
1355 | static char memcg_name[PATH_MAX]; | |
1356 | int ret; | |
1357 | ||
d31f56db | 1358 | if (!memcg || !p) |
e222432b BS |
1359 | return; |
1360 | ||
1361 | ||
1362 | rcu_read_lock(); | |
1363 | ||
1364 | mem_cgrp = memcg->css.cgroup; | |
1365 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1366 | ||
1367 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1368 | if (ret < 0) { | |
1369 | /* | |
1370 | * Unfortunately, we are unable to convert to a useful name | |
1371 | * But we'll still print out the usage information | |
1372 | */ | |
1373 | rcu_read_unlock(); | |
1374 | goto done; | |
1375 | } | |
1376 | rcu_read_unlock(); | |
1377 | ||
1378 | printk(KERN_INFO "Task in %s killed", memcg_name); | |
1379 | ||
1380 | rcu_read_lock(); | |
1381 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1382 | if (ret < 0) { | |
1383 | rcu_read_unlock(); | |
1384 | goto done; | |
1385 | } | |
1386 | rcu_read_unlock(); | |
1387 | ||
1388 | /* | |
1389 | * Continues from above, so we don't need an KERN_ level | |
1390 | */ | |
1391 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); | |
1392 | done: | |
1393 | ||
1394 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", | |
1395 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, | |
1396 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1397 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
1398 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " | |
1399 | "failcnt %llu\n", | |
1400 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, | |
1401 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1402 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
1403 | } | |
1404 | ||
81d39c20 KH |
1405 | /* |
1406 | * This function returns the number of memcg under hierarchy tree. Returns | |
1407 | * 1(self count) if no children. | |
1408 | */ | |
c0ff4b85 | 1409 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1410 | { |
1411 | int num = 0; | |
7d74b06f KH |
1412 | struct mem_cgroup *iter; |
1413 | ||
c0ff4b85 | 1414 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1415 | num++; |
81d39c20 KH |
1416 | return num; |
1417 | } | |
1418 | ||
a63d83f4 DR |
1419 | /* |
1420 | * Return the memory (and swap, if configured) limit for a memcg. | |
1421 | */ | |
1422 | u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) | |
1423 | { | |
1424 | u64 limit; | |
1425 | u64 memsw; | |
1426 | ||
f3e8eb70 JW |
1427 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
1428 | limit += total_swap_pages << PAGE_SHIFT; | |
1429 | ||
a63d83f4 DR |
1430 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
1431 | /* | |
1432 | * If memsw is finite and limits the amount of swap space available | |
1433 | * to this memcg, return that limit. | |
1434 | */ | |
1435 | return min(limit, memsw); | |
1436 | } | |
1437 | ||
5660048c JW |
1438 | static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg, |
1439 | gfp_t gfp_mask, | |
1440 | unsigned long flags) | |
1441 | { | |
1442 | unsigned long total = 0; | |
1443 | bool noswap = false; | |
1444 | int loop; | |
1445 | ||
1446 | if (flags & MEM_CGROUP_RECLAIM_NOSWAP) | |
1447 | noswap = true; | |
1448 | if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum) | |
1449 | noswap = true; | |
1450 | ||
1451 | for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) { | |
1452 | if (loop) | |
1453 | drain_all_stock_async(memcg); | |
1454 | total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap); | |
1455 | /* | |
1456 | * Allow limit shrinkers, which are triggered directly | |
1457 | * by userspace, to catch signals and stop reclaim | |
1458 | * after minimal progress, regardless of the margin. | |
1459 | */ | |
1460 | if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK)) | |
1461 | break; | |
1462 | if (mem_cgroup_margin(memcg)) | |
1463 | break; | |
1464 | /* | |
1465 | * If nothing was reclaimed after two attempts, there | |
1466 | * may be no reclaimable pages in this hierarchy. | |
1467 | */ | |
1468 | if (loop && !total) | |
1469 | break; | |
1470 | } | |
1471 | return total; | |
1472 | } | |
1473 | ||
4d0c066d KH |
1474 | /** |
1475 | * test_mem_cgroup_node_reclaimable | |
1476 | * @mem: the target memcg | |
1477 | * @nid: the node ID to be checked. | |
1478 | * @noswap : specify true here if the user wants flle only information. | |
1479 | * | |
1480 | * This function returns whether the specified memcg contains any | |
1481 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1482 | * pages in the node. | |
1483 | */ | |
c0ff4b85 | 1484 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1485 | int nid, bool noswap) |
1486 | { | |
c0ff4b85 | 1487 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1488 | return true; |
1489 | if (noswap || !total_swap_pages) | |
1490 | return false; | |
c0ff4b85 | 1491 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1492 | return true; |
1493 | return false; | |
1494 | ||
1495 | } | |
889976db YH |
1496 | #if MAX_NUMNODES > 1 |
1497 | ||
1498 | /* | |
1499 | * Always updating the nodemask is not very good - even if we have an empty | |
1500 | * list or the wrong list here, we can start from some node and traverse all | |
1501 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1502 | * | |
1503 | */ | |
c0ff4b85 | 1504 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1505 | { |
1506 | int nid; | |
453a9bf3 KH |
1507 | /* |
1508 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1509 | * pagein/pageout changes since the last update. | |
1510 | */ | |
c0ff4b85 | 1511 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1512 | return; |
c0ff4b85 | 1513 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1514 | return; |
1515 | ||
889976db | 1516 | /* make a nodemask where this memcg uses memory from */ |
c0ff4b85 | 1517 | memcg->scan_nodes = node_states[N_HIGH_MEMORY]; |
889976db YH |
1518 | |
1519 | for_each_node_mask(nid, node_states[N_HIGH_MEMORY]) { | |
1520 | ||
c0ff4b85 R |
1521 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1522 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1523 | } |
453a9bf3 | 1524 | |
c0ff4b85 R |
1525 | atomic_set(&memcg->numainfo_events, 0); |
1526 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1527 | } |
1528 | ||
1529 | /* | |
1530 | * Selecting a node where we start reclaim from. Because what we need is just | |
1531 | * reducing usage counter, start from anywhere is O,K. Considering | |
1532 | * memory reclaim from current node, there are pros. and cons. | |
1533 | * | |
1534 | * Freeing memory from current node means freeing memory from a node which | |
1535 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1536 | * hit limits, it will see a contention on a node. But freeing from remote | |
1537 | * node means more costs for memory reclaim because of memory latency. | |
1538 | * | |
1539 | * Now, we use round-robin. Better algorithm is welcomed. | |
1540 | */ | |
c0ff4b85 | 1541 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1542 | { |
1543 | int node; | |
1544 | ||
c0ff4b85 R |
1545 | mem_cgroup_may_update_nodemask(memcg); |
1546 | node = memcg->last_scanned_node; | |
889976db | 1547 | |
c0ff4b85 | 1548 | node = next_node(node, memcg->scan_nodes); |
889976db | 1549 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1550 | node = first_node(memcg->scan_nodes); |
889976db YH |
1551 | /* |
1552 | * We call this when we hit limit, not when pages are added to LRU. | |
1553 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1554 | * memcg is too small and all pages are not on LRU. In that case, | |
1555 | * we use curret node. | |
1556 | */ | |
1557 | if (unlikely(node == MAX_NUMNODES)) | |
1558 | node = numa_node_id(); | |
1559 | ||
c0ff4b85 | 1560 | memcg->last_scanned_node = node; |
889976db YH |
1561 | return node; |
1562 | } | |
1563 | ||
4d0c066d KH |
1564 | /* |
1565 | * Check all nodes whether it contains reclaimable pages or not. | |
1566 | * For quick scan, we make use of scan_nodes. This will allow us to skip | |
1567 | * unused nodes. But scan_nodes is lazily updated and may not cotain | |
1568 | * enough new information. We need to do double check. | |
1569 | */ | |
c0ff4b85 | 1570 | bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d KH |
1571 | { |
1572 | int nid; | |
1573 | ||
1574 | /* | |
1575 | * quick check...making use of scan_node. | |
1576 | * We can skip unused nodes. | |
1577 | */ | |
c0ff4b85 R |
1578 | if (!nodes_empty(memcg->scan_nodes)) { |
1579 | for (nid = first_node(memcg->scan_nodes); | |
4d0c066d | 1580 | nid < MAX_NUMNODES; |
c0ff4b85 | 1581 | nid = next_node(nid, memcg->scan_nodes)) { |
4d0c066d | 1582 | |
c0ff4b85 | 1583 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
1584 | return true; |
1585 | } | |
1586 | } | |
1587 | /* | |
1588 | * Check rest of nodes. | |
1589 | */ | |
1590 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
c0ff4b85 | 1591 | if (node_isset(nid, memcg->scan_nodes)) |
4d0c066d | 1592 | continue; |
c0ff4b85 | 1593 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
1594 | return true; |
1595 | } | |
1596 | return false; | |
1597 | } | |
1598 | ||
889976db | 1599 | #else |
c0ff4b85 | 1600 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1601 | { |
1602 | return 0; | |
1603 | } | |
4d0c066d | 1604 | |
c0ff4b85 | 1605 | bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d | 1606 | { |
c0ff4b85 | 1607 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); |
4d0c066d | 1608 | } |
889976db YH |
1609 | #endif |
1610 | ||
5660048c JW |
1611 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1612 | struct zone *zone, | |
1613 | gfp_t gfp_mask, | |
1614 | unsigned long *total_scanned) | |
6d61ef40 | 1615 | { |
9f3a0d09 | 1616 | struct mem_cgroup *victim = NULL; |
5660048c | 1617 | int total = 0; |
04046e1a | 1618 | int loop = 0; |
9d11ea9f | 1619 | unsigned long excess; |
185efc0f | 1620 | unsigned long nr_scanned; |
527a5ec9 JW |
1621 | struct mem_cgroup_reclaim_cookie reclaim = { |
1622 | .zone = zone, | |
1623 | .priority = 0, | |
1624 | }; | |
9d11ea9f | 1625 | |
c0ff4b85 | 1626 | excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT; |
04046e1a | 1627 | |
4e416953 | 1628 | while (1) { |
527a5ec9 | 1629 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); |
9f3a0d09 | 1630 | if (!victim) { |
04046e1a | 1631 | loop++; |
4e416953 BS |
1632 | if (loop >= 2) { |
1633 | /* | |
1634 | * If we have not been able to reclaim | |
1635 | * anything, it might because there are | |
1636 | * no reclaimable pages under this hierarchy | |
1637 | */ | |
5660048c | 1638 | if (!total) |
4e416953 | 1639 | break; |
4e416953 | 1640 | /* |
25985edc | 1641 | * We want to do more targeted reclaim. |
4e416953 BS |
1642 | * excess >> 2 is not to excessive so as to |
1643 | * reclaim too much, nor too less that we keep | |
1644 | * coming back to reclaim from this cgroup | |
1645 | */ | |
1646 | if (total >= (excess >> 2) || | |
9f3a0d09 | 1647 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) |
4e416953 | 1648 | break; |
4e416953 | 1649 | } |
9f3a0d09 | 1650 | continue; |
4e416953 | 1651 | } |
5660048c | 1652 | if (!mem_cgroup_reclaimable(victim, false)) |
6d61ef40 | 1653 | continue; |
5660048c JW |
1654 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1655 | zone, &nr_scanned); | |
1656 | *total_scanned += nr_scanned; | |
1657 | if (!res_counter_soft_limit_excess(&root_memcg->res)) | |
9f3a0d09 | 1658 | break; |
6d61ef40 | 1659 | } |
9f3a0d09 | 1660 | mem_cgroup_iter_break(root_memcg, victim); |
04046e1a | 1661 | return total; |
6d61ef40 BS |
1662 | } |
1663 | ||
867578cb KH |
1664 | /* |
1665 | * Check OOM-Killer is already running under our hierarchy. | |
1666 | * If someone is running, return false. | |
1af8efe9 | 1667 | * Has to be called with memcg_oom_lock |
867578cb | 1668 | */ |
c0ff4b85 | 1669 | static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg) |
867578cb | 1670 | { |
79dfdacc | 1671 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1672 | |
9f3a0d09 | 1673 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1674 | if (iter->oom_lock) { |
79dfdacc MH |
1675 | /* |
1676 | * this subtree of our hierarchy is already locked | |
1677 | * so we cannot give a lock. | |
1678 | */ | |
79dfdacc | 1679 | failed = iter; |
9f3a0d09 JW |
1680 | mem_cgroup_iter_break(memcg, iter); |
1681 | break; | |
23751be0 JW |
1682 | } else |
1683 | iter->oom_lock = true; | |
7d74b06f | 1684 | } |
867578cb | 1685 | |
79dfdacc | 1686 | if (!failed) |
23751be0 | 1687 | return true; |
79dfdacc MH |
1688 | |
1689 | /* | |
1690 | * OK, we failed to lock the whole subtree so we have to clean up | |
1691 | * what we set up to the failing subtree | |
1692 | */ | |
9f3a0d09 | 1693 | for_each_mem_cgroup_tree(iter, memcg) { |
79dfdacc | 1694 | if (iter == failed) { |
9f3a0d09 JW |
1695 | mem_cgroup_iter_break(memcg, iter); |
1696 | break; | |
79dfdacc MH |
1697 | } |
1698 | iter->oom_lock = false; | |
1699 | } | |
23751be0 | 1700 | return false; |
a636b327 | 1701 | } |
0b7f569e | 1702 | |
79dfdacc | 1703 | /* |
1af8efe9 | 1704 | * Has to be called with memcg_oom_lock |
79dfdacc | 1705 | */ |
c0ff4b85 | 1706 | static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1707 | { |
7d74b06f KH |
1708 | struct mem_cgroup *iter; |
1709 | ||
c0ff4b85 | 1710 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1711 | iter->oom_lock = false; |
1712 | return 0; | |
1713 | } | |
1714 | ||
c0ff4b85 | 1715 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1716 | { |
1717 | struct mem_cgroup *iter; | |
1718 | ||
c0ff4b85 | 1719 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1720 | atomic_inc(&iter->under_oom); |
1721 | } | |
1722 | ||
c0ff4b85 | 1723 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1724 | { |
1725 | struct mem_cgroup *iter; | |
1726 | ||
867578cb KH |
1727 | /* |
1728 | * When a new child is created while the hierarchy is under oom, | |
1729 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1730 | * atomic_add_unless() here. | |
1731 | */ | |
c0ff4b85 | 1732 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1733 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
1734 | } |
1735 | ||
1af8efe9 | 1736 | static DEFINE_SPINLOCK(memcg_oom_lock); |
867578cb KH |
1737 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1738 | ||
dc98df5a KH |
1739 | struct oom_wait_info { |
1740 | struct mem_cgroup *mem; | |
1741 | wait_queue_t wait; | |
1742 | }; | |
1743 | ||
1744 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1745 | unsigned mode, int sync, void *arg) | |
1746 | { | |
c0ff4b85 R |
1747 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg, |
1748 | *oom_wait_memcg; | |
dc98df5a KH |
1749 | struct oom_wait_info *oom_wait_info; |
1750 | ||
1751 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
c0ff4b85 | 1752 | oom_wait_memcg = oom_wait_info->mem; |
dc98df5a | 1753 | |
dc98df5a KH |
1754 | /* |
1755 | * Both of oom_wait_info->mem and wake_mem are stable under us. | |
1756 | * Then we can use css_is_ancestor without taking care of RCU. | |
1757 | */ | |
c0ff4b85 R |
1758 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
1759 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 1760 | return 0; |
dc98df5a KH |
1761 | return autoremove_wake_function(wait, mode, sync, arg); |
1762 | } | |
1763 | ||
c0ff4b85 | 1764 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 1765 | { |
c0ff4b85 R |
1766 | /* for filtering, pass "memcg" as argument. */ |
1767 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
1768 | } |
1769 | ||
c0ff4b85 | 1770 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1771 | { |
c0ff4b85 R |
1772 | if (memcg && atomic_read(&memcg->under_oom)) |
1773 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
1774 | } |
1775 | ||
867578cb KH |
1776 | /* |
1777 | * try to call OOM killer. returns false if we should exit memory-reclaim loop. | |
1778 | */ | |
c0ff4b85 | 1779 | bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask) |
0b7f569e | 1780 | { |
dc98df5a | 1781 | struct oom_wait_info owait; |
3c11ecf4 | 1782 | bool locked, need_to_kill; |
867578cb | 1783 | |
c0ff4b85 | 1784 | owait.mem = memcg; |
dc98df5a KH |
1785 | owait.wait.flags = 0; |
1786 | owait.wait.func = memcg_oom_wake_function; | |
1787 | owait.wait.private = current; | |
1788 | INIT_LIST_HEAD(&owait.wait.task_list); | |
3c11ecf4 | 1789 | need_to_kill = true; |
c0ff4b85 | 1790 | mem_cgroup_mark_under_oom(memcg); |
79dfdacc | 1791 | |
c0ff4b85 | 1792 | /* At first, try to OOM lock hierarchy under memcg.*/ |
1af8efe9 | 1793 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1794 | locked = mem_cgroup_oom_lock(memcg); |
867578cb KH |
1795 | /* |
1796 | * Even if signal_pending(), we can't quit charge() loop without | |
1797 | * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL | |
1798 | * under OOM is always welcomed, use TASK_KILLABLE here. | |
1799 | */ | |
3c11ecf4 | 1800 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
c0ff4b85 | 1801 | if (!locked || memcg->oom_kill_disable) |
3c11ecf4 KH |
1802 | need_to_kill = false; |
1803 | if (locked) | |
c0ff4b85 | 1804 | mem_cgroup_oom_notify(memcg); |
1af8efe9 | 1805 | spin_unlock(&memcg_oom_lock); |
867578cb | 1806 | |
3c11ecf4 KH |
1807 | if (need_to_kill) { |
1808 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
c0ff4b85 | 1809 | mem_cgroup_out_of_memory(memcg, mask); |
3c11ecf4 | 1810 | } else { |
867578cb | 1811 | schedule(); |
dc98df5a | 1812 | finish_wait(&memcg_oom_waitq, &owait.wait); |
867578cb | 1813 | } |
1af8efe9 | 1814 | spin_lock(&memcg_oom_lock); |
79dfdacc | 1815 | if (locked) |
c0ff4b85 R |
1816 | mem_cgroup_oom_unlock(memcg); |
1817 | memcg_wakeup_oom(memcg); | |
1af8efe9 | 1818 | spin_unlock(&memcg_oom_lock); |
867578cb | 1819 | |
c0ff4b85 | 1820 | mem_cgroup_unmark_under_oom(memcg); |
79dfdacc | 1821 | |
867578cb KH |
1822 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) |
1823 | return false; | |
1824 | /* Give chance to dying process */ | |
715a5ee8 | 1825 | schedule_timeout_uninterruptible(1); |
867578cb | 1826 | return true; |
0b7f569e KH |
1827 | } |
1828 | ||
d69b042f BS |
1829 | /* |
1830 | * Currently used to update mapped file statistics, but the routine can be | |
1831 | * generalized to update other statistics as well. | |
32047e2a KH |
1832 | * |
1833 | * Notes: Race condition | |
1834 | * | |
1835 | * We usually use page_cgroup_lock() for accessing page_cgroup member but | |
1836 | * it tends to be costly. But considering some conditions, we doesn't need | |
1837 | * to do so _always_. | |
1838 | * | |
1839 | * Considering "charge", lock_page_cgroup() is not required because all | |
1840 | * file-stat operations happen after a page is attached to radix-tree. There | |
1841 | * are no race with "charge". | |
1842 | * | |
1843 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
1844 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
1845 | * if there are race with "uncharge". Statistics itself is properly handled | |
1846 | * by flags. | |
1847 | * | |
1848 | * Considering "move", this is an only case we see a race. To make the race | |
1849 | * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are | |
1850 | * possibility of race condition. If there is, we take a lock. | |
d69b042f | 1851 | */ |
26174efd | 1852 | |
2a7106f2 GT |
1853 | void mem_cgroup_update_page_stat(struct page *page, |
1854 | enum mem_cgroup_page_stat_item idx, int val) | |
d69b042f | 1855 | { |
c0ff4b85 | 1856 | struct mem_cgroup *memcg; |
32047e2a KH |
1857 | struct page_cgroup *pc = lookup_page_cgroup(page); |
1858 | bool need_unlock = false; | |
dbd4ea78 | 1859 | unsigned long uninitialized_var(flags); |
d69b042f | 1860 | |
cfa44946 | 1861 | if (mem_cgroup_disabled()) |
d69b042f BS |
1862 | return; |
1863 | ||
32047e2a | 1864 | rcu_read_lock(); |
c0ff4b85 R |
1865 | memcg = pc->mem_cgroup; |
1866 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
32047e2a KH |
1867 | goto out; |
1868 | /* pc->mem_cgroup is unstable ? */ | |
c0ff4b85 | 1869 | if (unlikely(mem_cgroup_stealed(memcg)) || PageTransHuge(page)) { |
32047e2a | 1870 | /* take a lock against to access pc->mem_cgroup */ |
dbd4ea78 | 1871 | move_lock_page_cgroup(pc, &flags); |
32047e2a | 1872 | need_unlock = true; |
c0ff4b85 R |
1873 | memcg = pc->mem_cgroup; |
1874 | if (!memcg || !PageCgroupUsed(pc)) | |
32047e2a KH |
1875 | goto out; |
1876 | } | |
26174efd | 1877 | |
26174efd | 1878 | switch (idx) { |
2a7106f2 | 1879 | case MEMCG_NR_FILE_MAPPED: |
26174efd KH |
1880 | if (val > 0) |
1881 | SetPageCgroupFileMapped(pc); | |
1882 | else if (!page_mapped(page)) | |
0c270f8f | 1883 | ClearPageCgroupFileMapped(pc); |
2a7106f2 | 1884 | idx = MEM_CGROUP_STAT_FILE_MAPPED; |
26174efd KH |
1885 | break; |
1886 | default: | |
1887 | BUG(); | |
8725d541 | 1888 | } |
d69b042f | 1889 | |
c0ff4b85 | 1890 | this_cpu_add(memcg->stat->count[idx], val); |
2a7106f2 | 1891 | |
32047e2a KH |
1892 | out: |
1893 | if (unlikely(need_unlock)) | |
dbd4ea78 | 1894 | move_unlock_page_cgroup(pc, &flags); |
32047e2a KH |
1895 | rcu_read_unlock(); |
1896 | return; | |
d69b042f | 1897 | } |
2a7106f2 | 1898 | EXPORT_SYMBOL(mem_cgroup_update_page_stat); |
26174efd | 1899 | |
cdec2e42 KH |
1900 | /* |
1901 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1902 | * TODO: maybe necessary to use big numbers in big irons. | |
1903 | */ | |
7ec99d62 | 1904 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
1905 | struct memcg_stock_pcp { |
1906 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 1907 | unsigned int nr_pages; |
cdec2e42 | 1908 | struct work_struct work; |
26fe6168 KH |
1909 | unsigned long flags; |
1910 | #define FLUSHING_CACHED_CHARGE (0) | |
cdec2e42 KH |
1911 | }; |
1912 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 1913 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 KH |
1914 | |
1915 | /* | |
11c9ea4e | 1916 | * Try to consume stocked charge on this cpu. If success, one page is consumed |
cdec2e42 KH |
1917 | * from local stock and true is returned. If the stock is 0 or charges from a |
1918 | * cgroup which is not current target, returns false. This stock will be | |
1919 | * refilled. | |
1920 | */ | |
c0ff4b85 | 1921 | static bool consume_stock(struct mem_cgroup *memcg) |
cdec2e42 KH |
1922 | { |
1923 | struct memcg_stock_pcp *stock; | |
1924 | bool ret = true; | |
1925 | ||
1926 | stock = &get_cpu_var(memcg_stock); | |
c0ff4b85 | 1927 | if (memcg == stock->cached && stock->nr_pages) |
11c9ea4e | 1928 | stock->nr_pages--; |
cdec2e42 KH |
1929 | else /* need to call res_counter_charge */ |
1930 | ret = false; | |
1931 | put_cpu_var(memcg_stock); | |
1932 | return ret; | |
1933 | } | |
1934 | ||
1935 | /* | |
1936 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
1937 | */ | |
1938 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1939 | { | |
1940 | struct mem_cgroup *old = stock->cached; | |
1941 | ||
11c9ea4e JW |
1942 | if (stock->nr_pages) { |
1943 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
1944 | ||
1945 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 1946 | if (do_swap_account) |
11c9ea4e JW |
1947 | res_counter_uncharge(&old->memsw, bytes); |
1948 | stock->nr_pages = 0; | |
cdec2e42 KH |
1949 | } |
1950 | stock->cached = NULL; | |
cdec2e42 KH |
1951 | } |
1952 | ||
1953 | /* | |
1954 | * This must be called under preempt disabled or must be called by | |
1955 | * a thread which is pinned to local cpu. | |
1956 | */ | |
1957 | static void drain_local_stock(struct work_struct *dummy) | |
1958 | { | |
1959 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
1960 | drain_stock(stock); | |
26fe6168 | 1961 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
1962 | } |
1963 | ||
1964 | /* | |
1965 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 1966 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 1967 | */ |
c0ff4b85 | 1968 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1969 | { |
1970 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
1971 | ||
c0ff4b85 | 1972 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 1973 | drain_stock(stock); |
c0ff4b85 | 1974 | stock->cached = memcg; |
cdec2e42 | 1975 | } |
11c9ea4e | 1976 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
1977 | put_cpu_var(memcg_stock); |
1978 | } | |
1979 | ||
1980 | /* | |
c0ff4b85 | 1981 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
d38144b7 MH |
1982 | * of the hierarchy under it. sync flag says whether we should block |
1983 | * until the work is done. | |
cdec2e42 | 1984 | */ |
c0ff4b85 | 1985 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
cdec2e42 | 1986 | { |
26fe6168 | 1987 | int cpu, curcpu; |
d38144b7 | 1988 | |
cdec2e42 | 1989 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 1990 | get_online_cpus(); |
5af12d0e | 1991 | curcpu = get_cpu(); |
cdec2e42 KH |
1992 | for_each_online_cpu(cpu) { |
1993 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 1994 | struct mem_cgroup *memcg; |
26fe6168 | 1995 | |
c0ff4b85 R |
1996 | memcg = stock->cached; |
1997 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 1998 | continue; |
c0ff4b85 | 1999 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2000 | continue; |
d1a05b69 MH |
2001 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2002 | if (cpu == curcpu) | |
2003 | drain_local_stock(&stock->work); | |
2004 | else | |
2005 | schedule_work_on(cpu, &stock->work); | |
2006 | } | |
cdec2e42 | 2007 | } |
5af12d0e | 2008 | put_cpu(); |
d38144b7 MH |
2009 | |
2010 | if (!sync) | |
2011 | goto out; | |
2012 | ||
2013 | for_each_online_cpu(cpu) { | |
2014 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2015 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2016 | flush_work(&stock->work); |
2017 | } | |
2018 | out: | |
cdec2e42 | 2019 | put_online_cpus(); |
d38144b7 MH |
2020 | } |
2021 | ||
2022 | /* | |
2023 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2024 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
2025 | * expects some charges will be back to res_counter later but cannot wait for | |
2026 | * it. | |
2027 | */ | |
c0ff4b85 | 2028 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
d38144b7 | 2029 | { |
9f50fad6 MH |
2030 | /* |
2031 | * If someone calls draining, avoid adding more kworker runs. | |
2032 | */ | |
2033 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2034 | return; | |
c0ff4b85 | 2035 | drain_all_stock(root_memcg, false); |
9f50fad6 | 2036 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2037 | } |
2038 | ||
2039 | /* This is a synchronous drain interface. */ | |
c0ff4b85 | 2040 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
cdec2e42 KH |
2041 | { |
2042 | /* called when force_empty is called */ | |
9f50fad6 | 2043 | mutex_lock(&percpu_charge_mutex); |
c0ff4b85 | 2044 | drain_all_stock(root_memcg, true); |
9f50fad6 | 2045 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2046 | } |
2047 | ||
711d3d2c KH |
2048 | /* |
2049 | * This function drains percpu counter value from DEAD cpu and | |
2050 | * move it to local cpu. Note that this function can be preempted. | |
2051 | */ | |
c0ff4b85 | 2052 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2053 | { |
2054 | int i; | |
2055 | ||
c0ff4b85 | 2056 | spin_lock(&memcg->pcp_counter_lock); |
711d3d2c | 2057 | for (i = 0; i < MEM_CGROUP_STAT_DATA; i++) { |
c0ff4b85 | 2058 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2059 | |
c0ff4b85 R |
2060 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2061 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2062 | } |
e9f8974f | 2063 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2064 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2065 | |
c0ff4b85 R |
2066 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2067 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2068 | } |
1489ebad | 2069 | /* need to clear ON_MOVE value, works as a kind of lock. */ |
c0ff4b85 R |
2070 | per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) = 0; |
2071 | spin_unlock(&memcg->pcp_counter_lock); | |
1489ebad KH |
2072 | } |
2073 | ||
c0ff4b85 | 2074 | static void synchronize_mem_cgroup_on_move(struct mem_cgroup *memcg, int cpu) |
1489ebad KH |
2075 | { |
2076 | int idx = MEM_CGROUP_ON_MOVE; | |
2077 | ||
c0ff4b85 R |
2078 | spin_lock(&memcg->pcp_counter_lock); |
2079 | per_cpu(memcg->stat->count[idx], cpu) = memcg->nocpu_base.count[idx]; | |
2080 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
2081 | } |
2082 | ||
2083 | static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb, | |
cdec2e42 KH |
2084 | unsigned long action, |
2085 | void *hcpu) | |
2086 | { | |
2087 | int cpu = (unsigned long)hcpu; | |
2088 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2089 | struct mem_cgroup *iter; |
cdec2e42 | 2090 | |
1489ebad | 2091 | if ((action == CPU_ONLINE)) { |
9f3a0d09 | 2092 | for_each_mem_cgroup(iter) |
1489ebad KH |
2093 | synchronize_mem_cgroup_on_move(iter, cpu); |
2094 | return NOTIFY_OK; | |
2095 | } | |
2096 | ||
711d3d2c | 2097 | if ((action != CPU_DEAD) || action != CPU_DEAD_FROZEN) |
cdec2e42 | 2098 | return NOTIFY_OK; |
711d3d2c | 2099 | |
9f3a0d09 | 2100 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2101 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2102 | ||
cdec2e42 KH |
2103 | stock = &per_cpu(memcg_stock, cpu); |
2104 | drain_stock(stock); | |
2105 | return NOTIFY_OK; | |
2106 | } | |
2107 | ||
4b534334 KH |
2108 | |
2109 | /* See __mem_cgroup_try_charge() for details */ | |
2110 | enum { | |
2111 | CHARGE_OK, /* success */ | |
2112 | CHARGE_RETRY, /* need to retry but retry is not bad */ | |
2113 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ | |
2114 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ | |
2115 | CHARGE_OOM_DIE, /* the current is killed because of OOM */ | |
2116 | }; | |
2117 | ||
c0ff4b85 | 2118 | static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
7ec99d62 | 2119 | unsigned int nr_pages, bool oom_check) |
4b534334 | 2120 | { |
7ec99d62 | 2121 | unsigned long csize = nr_pages * PAGE_SIZE; |
4b534334 KH |
2122 | struct mem_cgroup *mem_over_limit; |
2123 | struct res_counter *fail_res; | |
2124 | unsigned long flags = 0; | |
2125 | int ret; | |
2126 | ||
c0ff4b85 | 2127 | ret = res_counter_charge(&memcg->res, csize, &fail_res); |
4b534334 KH |
2128 | |
2129 | if (likely(!ret)) { | |
2130 | if (!do_swap_account) | |
2131 | return CHARGE_OK; | |
c0ff4b85 | 2132 | ret = res_counter_charge(&memcg->memsw, csize, &fail_res); |
4b534334 KH |
2133 | if (likely(!ret)) |
2134 | return CHARGE_OK; | |
2135 | ||
c0ff4b85 | 2136 | res_counter_uncharge(&memcg->res, csize); |
4b534334 KH |
2137 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
2138 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
2139 | } else | |
2140 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
9221edb7 | 2141 | /* |
7ec99d62 JW |
2142 | * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch |
2143 | * of regular pages (CHARGE_BATCH), or a single regular page (1). | |
9221edb7 JW |
2144 | * |
2145 | * Never reclaim on behalf of optional batching, retry with a | |
2146 | * single page instead. | |
2147 | */ | |
7ec99d62 | 2148 | if (nr_pages == CHARGE_BATCH) |
4b534334 KH |
2149 | return CHARGE_RETRY; |
2150 | ||
2151 | if (!(gfp_mask & __GFP_WAIT)) | |
2152 | return CHARGE_WOULDBLOCK; | |
2153 | ||
5660048c | 2154 | ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags); |
7ec99d62 | 2155 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
19942822 | 2156 | return CHARGE_RETRY; |
4b534334 | 2157 | /* |
19942822 JW |
2158 | * Even though the limit is exceeded at this point, reclaim |
2159 | * may have been able to free some pages. Retry the charge | |
2160 | * before killing the task. | |
2161 | * | |
2162 | * Only for regular pages, though: huge pages are rather | |
2163 | * unlikely to succeed so close to the limit, and we fall back | |
2164 | * to regular pages anyway in case of failure. | |
4b534334 | 2165 | */ |
7ec99d62 | 2166 | if (nr_pages == 1 && ret) |
4b534334 KH |
2167 | return CHARGE_RETRY; |
2168 | ||
2169 | /* | |
2170 | * At task move, charge accounts can be doubly counted. So, it's | |
2171 | * better to wait until the end of task_move if something is going on. | |
2172 | */ | |
2173 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2174 | return CHARGE_RETRY; | |
2175 | ||
2176 | /* If we don't need to call oom-killer at el, return immediately */ | |
2177 | if (!oom_check) | |
2178 | return CHARGE_NOMEM; | |
2179 | /* check OOM */ | |
2180 | if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask)) | |
2181 | return CHARGE_OOM_DIE; | |
2182 | ||
2183 | return CHARGE_RETRY; | |
2184 | } | |
2185 | ||
f817ed48 | 2186 | /* |
38c5d72f KH |
2187 | * __mem_cgroup_try_charge() does |
2188 | * 1. detect memcg to be charged against from passed *mm and *ptr, | |
2189 | * 2. update res_counter | |
2190 | * 3. call memory reclaim if necessary. | |
2191 | * | |
2192 | * In some special case, if the task is fatal, fatal_signal_pending() or | |
2193 | * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup | |
2194 | * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon | |
2195 | * as possible without any hazards. 2: all pages should have a valid | |
2196 | * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg | |
2197 | * pointer, that is treated as a charge to root_mem_cgroup. | |
2198 | * | |
2199 | * So __mem_cgroup_try_charge() will return | |
2200 | * 0 ... on success, filling *ptr with a valid memcg pointer. | |
2201 | * -ENOMEM ... charge failure because of resource limits. | |
2202 | * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup. | |
2203 | * | |
2204 | * Unlike the exported interface, an "oom" parameter is added. if oom==true, | |
2205 | * the oom-killer can be invoked. | |
8a9f3ccd | 2206 | */ |
f817ed48 | 2207 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
ec168510 | 2208 | gfp_t gfp_mask, |
7ec99d62 | 2209 | unsigned int nr_pages, |
c0ff4b85 | 2210 | struct mem_cgroup **ptr, |
7ec99d62 | 2211 | bool oom) |
8a9f3ccd | 2212 | { |
7ec99d62 | 2213 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
4b534334 | 2214 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
c0ff4b85 | 2215 | struct mem_cgroup *memcg = NULL; |
4b534334 | 2216 | int ret; |
a636b327 | 2217 | |
867578cb KH |
2218 | /* |
2219 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
2220 | * in system level. So, allow to go ahead dying process in addition to | |
2221 | * MEMDIE process. | |
2222 | */ | |
2223 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
2224 | || fatal_signal_pending(current))) | |
2225 | goto bypass; | |
a636b327 | 2226 | |
8a9f3ccd | 2227 | /* |
3be91277 HD |
2228 | * We always charge the cgroup the mm_struct belongs to. |
2229 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
2230 | * thread group leader migrates. It's possible that mm is not |
2231 | * set, if so charge the init_mm (happens for pagecache usage). | |
2232 | */ | |
c0ff4b85 | 2233 | if (!*ptr && !mm) |
38c5d72f | 2234 | *ptr = root_mem_cgroup; |
f75ca962 | 2235 | again: |
c0ff4b85 R |
2236 | if (*ptr) { /* css should be a valid one */ |
2237 | memcg = *ptr; | |
2238 | VM_BUG_ON(css_is_removed(&memcg->css)); | |
2239 | if (mem_cgroup_is_root(memcg)) | |
f75ca962 | 2240 | goto done; |
c0ff4b85 | 2241 | if (nr_pages == 1 && consume_stock(memcg)) |
f75ca962 | 2242 | goto done; |
c0ff4b85 | 2243 | css_get(&memcg->css); |
4b534334 | 2244 | } else { |
f75ca962 | 2245 | struct task_struct *p; |
54595fe2 | 2246 | |
f75ca962 KH |
2247 | rcu_read_lock(); |
2248 | p = rcu_dereference(mm->owner); | |
f75ca962 | 2249 | /* |
ebb76ce1 | 2250 | * Because we don't have task_lock(), "p" can exit. |
c0ff4b85 | 2251 | * In that case, "memcg" can point to root or p can be NULL with |
ebb76ce1 KH |
2252 | * race with swapoff. Then, we have small risk of mis-accouning. |
2253 | * But such kind of mis-account by race always happens because | |
2254 | * we don't have cgroup_mutex(). It's overkill and we allo that | |
2255 | * small race, here. | |
2256 | * (*) swapoff at el will charge against mm-struct not against | |
2257 | * task-struct. So, mm->owner can be NULL. | |
f75ca962 | 2258 | */ |
c0ff4b85 | 2259 | memcg = mem_cgroup_from_task(p); |
38c5d72f KH |
2260 | if (!memcg) |
2261 | memcg = root_mem_cgroup; | |
2262 | if (mem_cgroup_is_root(memcg)) { | |
f75ca962 KH |
2263 | rcu_read_unlock(); |
2264 | goto done; | |
2265 | } | |
c0ff4b85 | 2266 | if (nr_pages == 1 && consume_stock(memcg)) { |
f75ca962 KH |
2267 | /* |
2268 | * It seems dagerous to access memcg without css_get(). | |
2269 | * But considering how consume_stok works, it's not | |
2270 | * necessary. If consume_stock success, some charges | |
2271 | * from this memcg are cached on this cpu. So, we | |
2272 | * don't need to call css_get()/css_tryget() before | |
2273 | * calling consume_stock(). | |
2274 | */ | |
2275 | rcu_read_unlock(); | |
2276 | goto done; | |
2277 | } | |
2278 | /* after here, we may be blocked. we need to get refcnt */ | |
c0ff4b85 | 2279 | if (!css_tryget(&memcg->css)) { |
f75ca962 KH |
2280 | rcu_read_unlock(); |
2281 | goto again; | |
2282 | } | |
2283 | rcu_read_unlock(); | |
2284 | } | |
8a9f3ccd | 2285 | |
4b534334 KH |
2286 | do { |
2287 | bool oom_check; | |
7a81b88c | 2288 | |
4b534334 | 2289 | /* If killed, bypass charge */ |
f75ca962 | 2290 | if (fatal_signal_pending(current)) { |
c0ff4b85 | 2291 | css_put(&memcg->css); |
4b534334 | 2292 | goto bypass; |
f75ca962 | 2293 | } |
6d61ef40 | 2294 | |
4b534334 KH |
2295 | oom_check = false; |
2296 | if (oom && !nr_oom_retries) { | |
2297 | oom_check = true; | |
2298 | nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
cdec2e42 | 2299 | } |
66e1707b | 2300 | |
c0ff4b85 | 2301 | ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, oom_check); |
4b534334 KH |
2302 | switch (ret) { |
2303 | case CHARGE_OK: | |
2304 | break; | |
2305 | case CHARGE_RETRY: /* not in OOM situation but retry */ | |
7ec99d62 | 2306 | batch = nr_pages; |
c0ff4b85 R |
2307 | css_put(&memcg->css); |
2308 | memcg = NULL; | |
f75ca962 | 2309 | goto again; |
4b534334 | 2310 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
c0ff4b85 | 2311 | css_put(&memcg->css); |
4b534334 KH |
2312 | goto nomem; |
2313 | case CHARGE_NOMEM: /* OOM routine works */ | |
f75ca962 | 2314 | if (!oom) { |
c0ff4b85 | 2315 | css_put(&memcg->css); |
867578cb | 2316 | goto nomem; |
f75ca962 | 2317 | } |
4b534334 KH |
2318 | /* If oom, we never return -ENOMEM */ |
2319 | nr_oom_retries--; | |
2320 | break; | |
2321 | case CHARGE_OOM_DIE: /* Killed by OOM Killer */ | |
c0ff4b85 | 2322 | css_put(&memcg->css); |
867578cb | 2323 | goto bypass; |
66e1707b | 2324 | } |
4b534334 KH |
2325 | } while (ret != CHARGE_OK); |
2326 | ||
7ec99d62 | 2327 | if (batch > nr_pages) |
c0ff4b85 R |
2328 | refill_stock(memcg, batch - nr_pages); |
2329 | css_put(&memcg->css); | |
0c3e73e8 | 2330 | done: |
c0ff4b85 | 2331 | *ptr = memcg; |
7a81b88c KH |
2332 | return 0; |
2333 | nomem: | |
c0ff4b85 | 2334 | *ptr = NULL; |
7a81b88c | 2335 | return -ENOMEM; |
867578cb | 2336 | bypass: |
38c5d72f KH |
2337 | *ptr = root_mem_cgroup; |
2338 | return -EINTR; | |
7a81b88c | 2339 | } |
8a9f3ccd | 2340 | |
a3032a2c DN |
2341 | /* |
2342 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2343 | * This function is for that and do uncharge, put css's refcnt. | |
2344 | * gotten by try_charge(). | |
2345 | */ | |
c0ff4b85 | 2346 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg, |
e7018b8d | 2347 | unsigned int nr_pages) |
a3032a2c | 2348 | { |
c0ff4b85 | 2349 | if (!mem_cgroup_is_root(memcg)) { |
e7018b8d JW |
2350 | unsigned long bytes = nr_pages * PAGE_SIZE; |
2351 | ||
c0ff4b85 | 2352 | res_counter_uncharge(&memcg->res, bytes); |
a3032a2c | 2353 | if (do_swap_account) |
c0ff4b85 | 2354 | res_counter_uncharge(&memcg->memsw, bytes); |
a3032a2c | 2355 | } |
854ffa8d DN |
2356 | } |
2357 | ||
a3b2d692 KH |
2358 | /* |
2359 | * A helper function to get mem_cgroup from ID. must be called under | |
2360 | * rcu_read_lock(). The caller must check css_is_removed() or some if | |
2361 | * it's concern. (dropping refcnt from swap can be called against removed | |
2362 | * memcg.) | |
2363 | */ | |
2364 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2365 | { | |
2366 | struct cgroup_subsys_state *css; | |
2367 | ||
2368 | /* ID 0 is unused ID */ | |
2369 | if (!id) | |
2370 | return NULL; | |
2371 | css = css_lookup(&mem_cgroup_subsys, id); | |
2372 | if (!css) | |
2373 | return NULL; | |
2374 | return container_of(css, struct mem_cgroup, css); | |
2375 | } | |
2376 | ||
e42d9d5d | 2377 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2378 | { |
c0ff4b85 | 2379 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2380 | struct page_cgroup *pc; |
a3b2d692 | 2381 | unsigned short id; |
b5a84319 KH |
2382 | swp_entry_t ent; |
2383 | ||
3c776e64 DN |
2384 | VM_BUG_ON(!PageLocked(page)); |
2385 | ||
3c776e64 | 2386 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2387 | lock_page_cgroup(pc); |
a3b2d692 | 2388 | if (PageCgroupUsed(pc)) { |
c0ff4b85 R |
2389 | memcg = pc->mem_cgroup; |
2390 | if (memcg && !css_tryget(&memcg->css)) | |
2391 | memcg = NULL; | |
e42d9d5d | 2392 | } else if (PageSwapCache(page)) { |
3c776e64 | 2393 | ent.val = page_private(page); |
9fb4b7cc | 2394 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2395 | rcu_read_lock(); |
c0ff4b85 R |
2396 | memcg = mem_cgroup_lookup(id); |
2397 | if (memcg && !css_tryget(&memcg->css)) | |
2398 | memcg = NULL; | |
a3b2d692 | 2399 | rcu_read_unlock(); |
3c776e64 | 2400 | } |
c0bd3f63 | 2401 | unlock_page_cgroup(pc); |
c0ff4b85 | 2402 | return memcg; |
b5a84319 KH |
2403 | } |
2404 | ||
c0ff4b85 | 2405 | static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg, |
5564e88b | 2406 | struct page *page, |
7ec99d62 | 2407 | unsigned int nr_pages, |
ca3e0214 | 2408 | struct page_cgroup *pc, |
7ec99d62 | 2409 | enum charge_type ctype) |
7a81b88c | 2410 | { |
ca3e0214 KH |
2411 | lock_page_cgroup(pc); |
2412 | if (unlikely(PageCgroupUsed(pc))) { | |
2413 | unlock_page_cgroup(pc); | |
c0ff4b85 | 2414 | __mem_cgroup_cancel_charge(memcg, nr_pages); |
ca3e0214 KH |
2415 | return; |
2416 | } | |
2417 | /* | |
2418 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2419 | * accessed by any other context at this point. | |
2420 | */ | |
c0ff4b85 | 2421 | pc->mem_cgroup = memcg; |
261fb61a KH |
2422 | /* |
2423 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
2424 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
2425 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
2426 | * before USED bit, we need memory barrier here. | |
2427 | * See mem_cgroup_add_lru_list(), etc. | |
2428 | */ | |
08e552c6 | 2429 | smp_wmb(); |
4b3bde4c BS |
2430 | switch (ctype) { |
2431 | case MEM_CGROUP_CHARGE_TYPE_CACHE: | |
2432 | case MEM_CGROUP_CHARGE_TYPE_SHMEM: | |
2433 | SetPageCgroupCache(pc); | |
2434 | SetPageCgroupUsed(pc); | |
2435 | break; | |
2436 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
2437 | ClearPageCgroupCache(pc); | |
2438 | SetPageCgroupUsed(pc); | |
2439 | break; | |
2440 | default: | |
2441 | break; | |
2442 | } | |
3be91277 | 2443 | |
c0ff4b85 | 2444 | mem_cgroup_charge_statistics(memcg, PageCgroupCache(pc), nr_pages); |
52d4b9ac | 2445 | unlock_page_cgroup(pc); |
38c5d72f | 2446 | WARN_ON_ONCE(PageLRU(page)); |
430e4863 KH |
2447 | /* |
2448 | * "charge_statistics" updated event counter. Then, check it. | |
2449 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
2450 | * if they exceeds softlimit. | |
2451 | */ | |
c0ff4b85 | 2452 | memcg_check_events(memcg, page); |
7a81b88c | 2453 | } |
66e1707b | 2454 | |
ca3e0214 KH |
2455 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2456 | ||
2457 | #define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\ | |
38c5d72f | 2458 | (1 << PCG_MIGRATION)) |
ca3e0214 KH |
2459 | /* |
2460 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
2461 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
2462 | * charge/uncharge will be never happen and move_account() is done under | |
2463 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 2464 | */ |
e94c8a9c | 2465 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 KH |
2466 | { |
2467 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
38c5d72f | 2468 | struct mem_cgroup_per_zone *mz; |
e94c8a9c | 2469 | struct page_cgroup *pc; |
38c5d72f | 2470 | enum lru_list lru; |
e94c8a9c | 2471 | int i; |
ca3e0214 | 2472 | |
3d37c4a9 KH |
2473 | if (mem_cgroup_disabled()) |
2474 | return; | |
e94c8a9c KH |
2475 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
2476 | pc = head_pc + i; | |
2477 | pc->mem_cgroup = head_pc->mem_cgroup; | |
2478 | smp_wmb();/* see __commit_charge() */ | |
e94c8a9c KH |
2479 | pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
2480 | } | |
38c5d72f KH |
2481 | /* |
2482 | * Tail pages will be added to LRU. | |
2483 | * We hold lru_lock,then,reduce counter directly. | |
2484 | */ | |
2485 | lru = page_lru(head); | |
2486 | mz = page_cgroup_zoneinfo(head_pc->mem_cgroup, head); | |
2487 | MEM_CGROUP_ZSTAT(mz, lru) -= HPAGE_PMD_NR - 1; | |
ca3e0214 KH |
2488 | } |
2489 | #endif | |
2490 | ||
f817ed48 | 2491 | /** |
de3638d9 | 2492 | * mem_cgroup_move_account - move account of the page |
5564e88b | 2493 | * @page: the page |
7ec99d62 | 2494 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
2495 | * @pc: page_cgroup of the page. |
2496 | * @from: mem_cgroup which the page is moved from. | |
2497 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
854ffa8d | 2498 | * @uncharge: whether we should call uncharge and css_put against @from. |
f817ed48 KH |
2499 | * |
2500 | * The caller must confirm following. | |
08e552c6 | 2501 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 2502 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 2503 | * |
854ffa8d | 2504 | * This function doesn't do "charge" nor css_get to new cgroup. It should be |
25985edc | 2505 | * done by a caller(__mem_cgroup_try_charge would be useful). If @uncharge is |
854ffa8d DN |
2506 | * true, this function does "uncharge" from old cgroup, but it doesn't if |
2507 | * @uncharge is false, so a caller should do "uncharge". | |
f817ed48 | 2508 | */ |
7ec99d62 JW |
2509 | static int mem_cgroup_move_account(struct page *page, |
2510 | unsigned int nr_pages, | |
2511 | struct page_cgroup *pc, | |
2512 | struct mem_cgroup *from, | |
2513 | struct mem_cgroup *to, | |
2514 | bool uncharge) | |
f817ed48 | 2515 | { |
de3638d9 JW |
2516 | unsigned long flags; |
2517 | int ret; | |
987eba66 | 2518 | |
f817ed48 | 2519 | VM_BUG_ON(from == to); |
5564e88b | 2520 | VM_BUG_ON(PageLRU(page)); |
de3638d9 JW |
2521 | /* |
2522 | * The page is isolated from LRU. So, collapse function | |
2523 | * will not handle this page. But page splitting can happen. | |
2524 | * Do this check under compound_page_lock(). The caller should | |
2525 | * hold it. | |
2526 | */ | |
2527 | ret = -EBUSY; | |
7ec99d62 | 2528 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
2529 | goto out; |
2530 | ||
2531 | lock_page_cgroup(pc); | |
2532 | ||
2533 | ret = -EINVAL; | |
2534 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
2535 | goto unlock; | |
2536 | ||
2537 | move_lock_page_cgroup(pc, &flags); | |
f817ed48 | 2538 | |
8725d541 | 2539 | if (PageCgroupFileMapped(pc)) { |
c62b1a3b KH |
2540 | /* Update mapped_file data for mem_cgroup */ |
2541 | preempt_disable(); | |
2542 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2543 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2544 | preempt_enable(); | |
d69b042f | 2545 | } |
987eba66 | 2546 | mem_cgroup_charge_statistics(from, PageCgroupCache(pc), -nr_pages); |
854ffa8d DN |
2547 | if (uncharge) |
2548 | /* This is not "cancel", but cancel_charge does all we need. */ | |
e7018b8d | 2549 | __mem_cgroup_cancel_charge(from, nr_pages); |
d69b042f | 2550 | |
854ffa8d | 2551 | /* caller should have done css_get */ |
08e552c6 | 2552 | pc->mem_cgroup = to; |
987eba66 | 2553 | mem_cgroup_charge_statistics(to, PageCgroupCache(pc), nr_pages); |
88703267 KH |
2554 | /* |
2555 | * We charges against "to" which may not have any tasks. Then, "to" | |
2556 | * can be under rmdir(). But in current implementation, caller of | |
4ffef5fe | 2557 | * this function is just force_empty() and move charge, so it's |
25985edc | 2558 | * guaranteed that "to" is never removed. So, we don't check rmdir |
4ffef5fe | 2559 | * status here. |
88703267 | 2560 | */ |
de3638d9 JW |
2561 | move_unlock_page_cgroup(pc, &flags); |
2562 | ret = 0; | |
2563 | unlock: | |
57f9fd7d | 2564 | unlock_page_cgroup(pc); |
d2265e6f KH |
2565 | /* |
2566 | * check events | |
2567 | */ | |
5564e88b JW |
2568 | memcg_check_events(to, page); |
2569 | memcg_check_events(from, page); | |
de3638d9 | 2570 | out: |
f817ed48 KH |
2571 | return ret; |
2572 | } | |
2573 | ||
2574 | /* | |
2575 | * move charges to its parent. | |
2576 | */ | |
2577 | ||
5564e88b JW |
2578 | static int mem_cgroup_move_parent(struct page *page, |
2579 | struct page_cgroup *pc, | |
f817ed48 KH |
2580 | struct mem_cgroup *child, |
2581 | gfp_t gfp_mask) | |
2582 | { | |
2583 | struct cgroup *cg = child->css.cgroup; | |
2584 | struct cgroup *pcg = cg->parent; | |
2585 | struct mem_cgroup *parent; | |
7ec99d62 | 2586 | unsigned int nr_pages; |
4be4489f | 2587 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
2588 | int ret; |
2589 | ||
2590 | /* Is ROOT ? */ | |
2591 | if (!pcg) | |
2592 | return -EINVAL; | |
2593 | ||
57f9fd7d DN |
2594 | ret = -EBUSY; |
2595 | if (!get_page_unless_zero(page)) | |
2596 | goto out; | |
2597 | if (isolate_lru_page(page)) | |
2598 | goto put; | |
52dbb905 | 2599 | |
7ec99d62 | 2600 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 2601 | |
f817ed48 | 2602 | parent = mem_cgroup_from_cont(pcg); |
7ec99d62 | 2603 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, nr_pages, &parent, false); |
38c5d72f | 2604 | if (ret) |
57f9fd7d | 2605 | goto put_back; |
f817ed48 | 2606 | |
7ec99d62 | 2607 | if (nr_pages > 1) |
987eba66 KH |
2608 | flags = compound_lock_irqsave(page); |
2609 | ||
7ec99d62 | 2610 | ret = mem_cgroup_move_account(page, nr_pages, pc, child, parent, true); |
854ffa8d | 2611 | if (ret) |
7ec99d62 | 2612 | __mem_cgroup_cancel_charge(parent, nr_pages); |
8dba474f | 2613 | |
7ec99d62 | 2614 | if (nr_pages > 1) |
987eba66 | 2615 | compound_unlock_irqrestore(page, flags); |
8dba474f | 2616 | put_back: |
08e552c6 | 2617 | putback_lru_page(page); |
57f9fd7d | 2618 | put: |
40d58138 | 2619 | put_page(page); |
57f9fd7d | 2620 | out: |
f817ed48 KH |
2621 | return ret; |
2622 | } | |
2623 | ||
7a81b88c KH |
2624 | /* |
2625 | * Charge the memory controller for page usage. | |
2626 | * Return | |
2627 | * 0 if the charge was successful | |
2628 | * < 0 if the cgroup is over its limit | |
2629 | */ | |
2630 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
73045c47 | 2631 | gfp_t gfp_mask, enum charge_type ctype) |
7a81b88c | 2632 | { |
c0ff4b85 | 2633 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 2634 | unsigned int nr_pages = 1; |
7a81b88c | 2635 | struct page_cgroup *pc; |
8493ae43 | 2636 | bool oom = true; |
7a81b88c | 2637 | int ret; |
ec168510 | 2638 | |
37c2ac78 | 2639 | if (PageTransHuge(page)) { |
7ec99d62 | 2640 | nr_pages <<= compound_order(page); |
37c2ac78 | 2641 | VM_BUG_ON(!PageTransHuge(page)); |
8493ae43 JW |
2642 | /* |
2643 | * Never OOM-kill a process for a huge page. The | |
2644 | * fault handler will fall back to regular pages. | |
2645 | */ | |
2646 | oom = false; | |
37c2ac78 | 2647 | } |
7a81b88c KH |
2648 | |
2649 | pc = lookup_page_cgroup(page); | |
c0ff4b85 | 2650 | ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom); |
38c5d72f | 2651 | if (ret == -ENOMEM) |
7a81b88c | 2652 | return ret; |
c0ff4b85 | 2653 | __mem_cgroup_commit_charge(memcg, page, nr_pages, pc, ctype); |
8a9f3ccd | 2654 | return 0; |
8a9f3ccd BS |
2655 | } |
2656 | ||
7a81b88c KH |
2657 | int mem_cgroup_newpage_charge(struct page *page, |
2658 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 2659 | { |
f8d66542 | 2660 | if (mem_cgroup_disabled()) |
cede86ac | 2661 | return 0; |
7a0524cf JW |
2662 | VM_BUG_ON(page_mapped(page)); |
2663 | VM_BUG_ON(page->mapping && !PageAnon(page)); | |
2664 | VM_BUG_ON(!mm); | |
217bc319 | 2665 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
7a0524cf | 2666 | MEM_CGROUP_CHARGE_TYPE_MAPPED); |
217bc319 KH |
2667 | } |
2668 | ||
83aae4c7 DN |
2669 | static void |
2670 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
2671 | enum charge_type ctype); | |
2672 | ||
5a6475a4 | 2673 | static void |
c0ff4b85 | 2674 | __mem_cgroup_commit_charge_lrucare(struct page *page, struct mem_cgroup *memcg, |
5a6475a4 KH |
2675 | enum charge_type ctype) |
2676 | { | |
2677 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
36b62ad5 KH |
2678 | struct zone *zone = page_zone(page); |
2679 | unsigned long flags; | |
2680 | bool removed = false; | |
2681 | ||
5a6475a4 KH |
2682 | /* |
2683 | * In some case, SwapCache, FUSE(splice_buf->radixtree), the page | |
2684 | * is already on LRU. It means the page may on some other page_cgroup's | |
2685 | * LRU. Take care of it. | |
2686 | */ | |
36b62ad5 KH |
2687 | spin_lock_irqsave(&zone->lru_lock, flags); |
2688 | if (PageLRU(page)) { | |
2689 | del_page_from_lru_list(zone, page, page_lru(page)); | |
2690 | ClearPageLRU(page); | |
2691 | removed = true; | |
2692 | } | |
c0ff4b85 | 2693 | __mem_cgroup_commit_charge(memcg, page, 1, pc, ctype); |
36b62ad5 KH |
2694 | if (removed) { |
2695 | add_page_to_lru_list(zone, page, page_lru(page)); | |
2696 | SetPageLRU(page); | |
2697 | } | |
2698 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
5a6475a4 KH |
2699 | return; |
2700 | } | |
2701 | ||
e1a1cd59 BS |
2702 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
2703 | gfp_t gfp_mask) | |
8697d331 | 2704 | { |
c0ff4b85 | 2705 | struct mem_cgroup *memcg = NULL; |
dc67d504 | 2706 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
b5a84319 KH |
2707 | int ret; |
2708 | ||
f8d66542 | 2709 | if (mem_cgroup_disabled()) |
cede86ac | 2710 | return 0; |
52d4b9ac KH |
2711 | if (PageCompound(page)) |
2712 | return 0; | |
accf163e | 2713 | |
73045c47 | 2714 | if (unlikely(!mm)) |
8697d331 | 2715 | mm = &init_mm; |
dc67d504 KH |
2716 | if (!page_is_file_cache(page)) |
2717 | type = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
accf163e | 2718 | |
38c5d72f | 2719 | if (!PageSwapCache(page)) |
dc67d504 | 2720 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, type); |
38c5d72f | 2721 | else { /* page is swapcache/shmem */ |
c0ff4b85 | 2722 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &memcg); |
83aae4c7 | 2723 | if (!ret) |
dc67d504 KH |
2724 | __mem_cgroup_commit_charge_swapin(page, memcg, type); |
2725 | } | |
b5a84319 | 2726 | return ret; |
e8589cc1 KH |
2727 | } |
2728 | ||
54595fe2 KH |
2729 | /* |
2730 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
2731 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 2732 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
2733 | * "commit()" or removed by "cancel()" |
2734 | */ | |
8c7c6e34 KH |
2735 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
2736 | struct page *page, | |
72835c86 | 2737 | gfp_t mask, struct mem_cgroup **memcgp) |
8c7c6e34 | 2738 | { |
c0ff4b85 | 2739 | struct mem_cgroup *memcg; |
54595fe2 | 2740 | int ret; |
8c7c6e34 | 2741 | |
72835c86 | 2742 | *memcgp = NULL; |
56039efa | 2743 | |
f8d66542 | 2744 | if (mem_cgroup_disabled()) |
8c7c6e34 KH |
2745 | return 0; |
2746 | ||
2747 | if (!do_swap_account) | |
2748 | goto charge_cur_mm; | |
8c7c6e34 KH |
2749 | /* |
2750 | * A racing thread's fault, or swapoff, may have already updated | |
407f9c8b HD |
2751 | * the pte, and even removed page from swap cache: in those cases |
2752 | * do_swap_page()'s pte_same() test will fail; but there's also a | |
2753 | * KSM case which does need to charge the page. | |
8c7c6e34 KH |
2754 | */ |
2755 | if (!PageSwapCache(page)) | |
407f9c8b | 2756 | goto charge_cur_mm; |
c0ff4b85 R |
2757 | memcg = try_get_mem_cgroup_from_page(page); |
2758 | if (!memcg) | |
54595fe2 | 2759 | goto charge_cur_mm; |
72835c86 JW |
2760 | *memcgp = memcg; |
2761 | ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true); | |
c0ff4b85 | 2762 | css_put(&memcg->css); |
38c5d72f KH |
2763 | if (ret == -EINTR) |
2764 | ret = 0; | |
54595fe2 | 2765 | return ret; |
8c7c6e34 KH |
2766 | charge_cur_mm: |
2767 | if (unlikely(!mm)) | |
2768 | mm = &init_mm; | |
38c5d72f KH |
2769 | ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true); |
2770 | if (ret == -EINTR) | |
2771 | ret = 0; | |
2772 | return ret; | |
8c7c6e34 KH |
2773 | } |
2774 | ||
83aae4c7 | 2775 | static void |
72835c86 | 2776 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg, |
83aae4c7 | 2777 | enum charge_type ctype) |
7a81b88c | 2778 | { |
f8d66542 | 2779 | if (mem_cgroup_disabled()) |
7a81b88c | 2780 | return; |
72835c86 | 2781 | if (!memcg) |
7a81b88c | 2782 | return; |
72835c86 | 2783 | cgroup_exclude_rmdir(&memcg->css); |
5a6475a4 | 2784 | |
72835c86 | 2785 | __mem_cgroup_commit_charge_lrucare(page, memcg, ctype); |
8c7c6e34 KH |
2786 | /* |
2787 | * Now swap is on-memory. This means this page may be | |
2788 | * counted both as mem and swap....double count. | |
03f3c433 KH |
2789 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
2790 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
2791 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 2792 | */ |
03f3c433 | 2793 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 2794 | swp_entry_t ent = {.val = page_private(page)}; |
72835c86 | 2795 | struct mem_cgroup *swap_memcg; |
a3b2d692 | 2796 | unsigned short id; |
a3b2d692 KH |
2797 | |
2798 | id = swap_cgroup_record(ent, 0); | |
2799 | rcu_read_lock(); | |
72835c86 JW |
2800 | swap_memcg = mem_cgroup_lookup(id); |
2801 | if (swap_memcg) { | |
a3b2d692 KH |
2802 | /* |
2803 | * This recorded memcg can be obsolete one. So, avoid | |
2804 | * calling css_tryget | |
2805 | */ | |
72835c86 JW |
2806 | if (!mem_cgroup_is_root(swap_memcg)) |
2807 | res_counter_uncharge(&swap_memcg->memsw, | |
2808 | PAGE_SIZE); | |
2809 | mem_cgroup_swap_statistics(swap_memcg, false); | |
2810 | mem_cgroup_put(swap_memcg); | |
8c7c6e34 | 2811 | } |
a3b2d692 | 2812 | rcu_read_unlock(); |
8c7c6e34 | 2813 | } |
88703267 KH |
2814 | /* |
2815 | * At swapin, we may charge account against cgroup which has no tasks. | |
2816 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
2817 | * In that case, we need to call pre_destroy() again. check it here. | |
2818 | */ | |
72835c86 | 2819 | cgroup_release_and_wakeup_rmdir(&memcg->css); |
7a81b88c KH |
2820 | } |
2821 | ||
72835c86 JW |
2822 | void mem_cgroup_commit_charge_swapin(struct page *page, |
2823 | struct mem_cgroup *memcg) | |
83aae4c7 | 2824 | { |
72835c86 JW |
2825 | __mem_cgroup_commit_charge_swapin(page, memcg, |
2826 | MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
83aae4c7 DN |
2827 | } |
2828 | ||
c0ff4b85 | 2829 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg) |
7a81b88c | 2830 | { |
f8d66542 | 2831 | if (mem_cgroup_disabled()) |
7a81b88c | 2832 | return; |
c0ff4b85 | 2833 | if (!memcg) |
7a81b88c | 2834 | return; |
c0ff4b85 | 2835 | __mem_cgroup_cancel_charge(memcg, 1); |
7a81b88c KH |
2836 | } |
2837 | ||
c0ff4b85 | 2838 | static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg, |
7ec99d62 JW |
2839 | unsigned int nr_pages, |
2840 | const enum charge_type ctype) | |
569b846d KH |
2841 | { |
2842 | struct memcg_batch_info *batch = NULL; | |
2843 | bool uncharge_memsw = true; | |
7ec99d62 | 2844 | |
569b846d KH |
2845 | /* If swapout, usage of swap doesn't decrease */ |
2846 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2847 | uncharge_memsw = false; | |
569b846d KH |
2848 | |
2849 | batch = ¤t->memcg_batch; | |
2850 | /* | |
2851 | * In usual, we do css_get() when we remember memcg pointer. | |
2852 | * But in this case, we keep res->usage until end of a series of | |
2853 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
2854 | */ | |
2855 | if (!batch->memcg) | |
c0ff4b85 | 2856 | batch->memcg = memcg; |
3c11ecf4 KH |
2857 | /* |
2858 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
25985edc | 2859 | * In those cases, all pages freed continuously can be expected to be in |
3c11ecf4 KH |
2860 | * the same cgroup and we have chance to coalesce uncharges. |
2861 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
2862 | * because we want to do uncharge as soon as possible. | |
2863 | */ | |
2864 | ||
2865 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
2866 | goto direct_uncharge; | |
2867 | ||
7ec99d62 | 2868 | if (nr_pages > 1) |
ec168510 AA |
2869 | goto direct_uncharge; |
2870 | ||
569b846d KH |
2871 | /* |
2872 | * In typical case, batch->memcg == mem. This means we can | |
2873 | * merge a series of uncharges to an uncharge of res_counter. | |
2874 | * If not, we uncharge res_counter ony by one. | |
2875 | */ | |
c0ff4b85 | 2876 | if (batch->memcg != memcg) |
569b846d KH |
2877 | goto direct_uncharge; |
2878 | /* remember freed charge and uncharge it later */ | |
7ffd4ca7 | 2879 | batch->nr_pages++; |
569b846d | 2880 | if (uncharge_memsw) |
7ffd4ca7 | 2881 | batch->memsw_nr_pages++; |
569b846d KH |
2882 | return; |
2883 | direct_uncharge: | |
c0ff4b85 | 2884 | res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE); |
569b846d | 2885 | if (uncharge_memsw) |
c0ff4b85 R |
2886 | res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE); |
2887 | if (unlikely(batch->memcg != memcg)) | |
2888 | memcg_oom_recover(memcg); | |
569b846d KH |
2889 | return; |
2890 | } | |
7a81b88c | 2891 | |
8a9f3ccd | 2892 | /* |
69029cd5 | 2893 | * uncharge if !page_mapped(page) |
8a9f3ccd | 2894 | */ |
8c7c6e34 | 2895 | static struct mem_cgroup * |
69029cd5 | 2896 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
8a9f3ccd | 2897 | { |
c0ff4b85 | 2898 | struct mem_cgroup *memcg = NULL; |
7ec99d62 JW |
2899 | unsigned int nr_pages = 1; |
2900 | struct page_cgroup *pc; | |
8a9f3ccd | 2901 | |
f8d66542 | 2902 | if (mem_cgroup_disabled()) |
8c7c6e34 | 2903 | return NULL; |
4077960e | 2904 | |
d13d1443 | 2905 | if (PageSwapCache(page)) |
8c7c6e34 | 2906 | return NULL; |
d13d1443 | 2907 | |
37c2ac78 | 2908 | if (PageTransHuge(page)) { |
7ec99d62 | 2909 | nr_pages <<= compound_order(page); |
37c2ac78 AA |
2910 | VM_BUG_ON(!PageTransHuge(page)); |
2911 | } | |
8697d331 | 2912 | /* |
3c541e14 | 2913 | * Check if our page_cgroup is valid |
8697d331 | 2914 | */ |
52d4b9ac | 2915 | pc = lookup_page_cgroup(page); |
cfa44946 | 2916 | if (unlikely(!PageCgroupUsed(pc))) |
8c7c6e34 | 2917 | return NULL; |
b9c565d5 | 2918 | |
52d4b9ac | 2919 | lock_page_cgroup(pc); |
d13d1443 | 2920 | |
c0ff4b85 | 2921 | memcg = pc->mem_cgroup; |
8c7c6e34 | 2922 | |
d13d1443 KH |
2923 | if (!PageCgroupUsed(pc)) |
2924 | goto unlock_out; | |
2925 | ||
2926 | switch (ctype) { | |
2927 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
8a9478ca | 2928 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c | 2929 | /* See mem_cgroup_prepare_migration() */ |
2930 | if (page_mapped(page) || PageCgroupMigration(pc)) | |
d13d1443 KH |
2931 | goto unlock_out; |
2932 | break; | |
2933 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
2934 | if (!PageAnon(page)) { /* Shared memory */ | |
2935 | if (page->mapping && !page_is_file_cache(page)) | |
2936 | goto unlock_out; | |
2937 | } else if (page_mapped(page)) /* Anon */ | |
2938 | goto unlock_out; | |
2939 | break; | |
2940 | default: | |
2941 | break; | |
52d4b9ac | 2942 | } |
d13d1443 | 2943 | |
c0ff4b85 | 2944 | mem_cgroup_charge_statistics(memcg, PageCgroupCache(pc), -nr_pages); |
04046e1a | 2945 | |
52d4b9ac | 2946 | ClearPageCgroupUsed(pc); |
544122e5 KH |
2947 | /* |
2948 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
2949 | * freed from LRU. This is safe because uncharged page is expected not | |
2950 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
2951 | * special functions. | |
2952 | */ | |
b9c565d5 | 2953 | |
52d4b9ac | 2954 | unlock_page_cgroup(pc); |
f75ca962 | 2955 | /* |
c0ff4b85 | 2956 | * even after unlock, we have memcg->res.usage here and this memcg |
f75ca962 KH |
2957 | * will never be freed. |
2958 | */ | |
c0ff4b85 | 2959 | memcg_check_events(memcg, page); |
f75ca962 | 2960 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
c0ff4b85 R |
2961 | mem_cgroup_swap_statistics(memcg, true); |
2962 | mem_cgroup_get(memcg); | |
f75ca962 | 2963 | } |
c0ff4b85 R |
2964 | if (!mem_cgroup_is_root(memcg)) |
2965 | mem_cgroup_do_uncharge(memcg, nr_pages, ctype); | |
6d12e2d8 | 2966 | |
c0ff4b85 | 2967 | return memcg; |
d13d1443 KH |
2968 | |
2969 | unlock_out: | |
2970 | unlock_page_cgroup(pc); | |
8c7c6e34 | 2971 | return NULL; |
3c541e14 BS |
2972 | } |
2973 | ||
69029cd5 KH |
2974 | void mem_cgroup_uncharge_page(struct page *page) |
2975 | { | |
52d4b9ac KH |
2976 | /* early check. */ |
2977 | if (page_mapped(page)) | |
2978 | return; | |
40f23a21 | 2979 | VM_BUG_ON(page->mapping && !PageAnon(page)); |
69029cd5 KH |
2980 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
2981 | } | |
2982 | ||
2983 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
2984 | { | |
2985 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 2986 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
2987 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
2988 | } | |
2989 | ||
569b846d KH |
2990 | /* |
2991 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
2992 | * In that cases, pages are freed continuously and we can expect pages | |
2993 | * are in the same memcg. All these calls itself limits the number of | |
2994 | * pages freed at once, then uncharge_start/end() is called properly. | |
2995 | * This may be called prural(2) times in a context, | |
2996 | */ | |
2997 | ||
2998 | void mem_cgroup_uncharge_start(void) | |
2999 | { | |
3000 | current->memcg_batch.do_batch++; | |
3001 | /* We can do nest. */ | |
3002 | if (current->memcg_batch.do_batch == 1) { | |
3003 | current->memcg_batch.memcg = NULL; | |
7ffd4ca7 JW |
3004 | current->memcg_batch.nr_pages = 0; |
3005 | current->memcg_batch.memsw_nr_pages = 0; | |
569b846d KH |
3006 | } |
3007 | } | |
3008 | ||
3009 | void mem_cgroup_uncharge_end(void) | |
3010 | { | |
3011 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
3012 | ||
3013 | if (!batch->do_batch) | |
3014 | return; | |
3015 | ||
3016 | batch->do_batch--; | |
3017 | if (batch->do_batch) /* If stacked, do nothing. */ | |
3018 | return; | |
3019 | ||
3020 | if (!batch->memcg) | |
3021 | return; | |
3022 | /* | |
3023 | * This "batch->memcg" is valid without any css_get/put etc... | |
3024 | * bacause we hide charges behind us. | |
3025 | */ | |
7ffd4ca7 JW |
3026 | if (batch->nr_pages) |
3027 | res_counter_uncharge(&batch->memcg->res, | |
3028 | batch->nr_pages * PAGE_SIZE); | |
3029 | if (batch->memsw_nr_pages) | |
3030 | res_counter_uncharge(&batch->memcg->memsw, | |
3031 | batch->memsw_nr_pages * PAGE_SIZE); | |
3c11ecf4 | 3032 | memcg_oom_recover(batch->memcg); |
569b846d KH |
3033 | /* forget this pointer (for sanity check) */ |
3034 | batch->memcg = NULL; | |
3035 | } | |
3036 | ||
4e5f01c2 KH |
3037 | /* |
3038 | * A function for resetting pc->mem_cgroup for newly allocated pages. | |
3039 | * This function should be called if the newpage will be added to LRU | |
3040 | * before start accounting. | |
3041 | */ | |
3042 | void mem_cgroup_reset_owner(struct page *newpage) | |
3043 | { | |
3044 | struct page_cgroup *pc; | |
3045 | ||
3046 | if (mem_cgroup_disabled()) | |
3047 | return; | |
3048 | ||
3049 | pc = lookup_page_cgroup(newpage); | |
3050 | VM_BUG_ON(PageCgroupUsed(pc)); | |
3051 | pc->mem_cgroup = root_mem_cgroup; | |
3052 | } | |
3053 | ||
e767e056 | 3054 | #ifdef CONFIG_SWAP |
8c7c6e34 | 3055 | /* |
e767e056 | 3056 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
3057 | * memcg information is recorded to swap_cgroup of "ent" |
3058 | */ | |
8a9478ca KH |
3059 | void |
3060 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
3061 | { |
3062 | struct mem_cgroup *memcg; | |
8a9478ca KH |
3063 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
3064 | ||
3065 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
3066 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
3067 | ||
3068 | memcg = __mem_cgroup_uncharge_common(page, ctype); | |
8c7c6e34 | 3069 | |
f75ca962 KH |
3070 | /* |
3071 | * record memcg information, if swapout && memcg != NULL, | |
3072 | * mem_cgroup_get() was called in uncharge(). | |
3073 | */ | |
3074 | if (do_swap_account && swapout && memcg) | |
a3b2d692 | 3075 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 | 3076 | } |
e767e056 | 3077 | #endif |
8c7c6e34 KH |
3078 | |
3079 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
3080 | /* | |
3081 | * called from swap_entry_free(). remove record in swap_cgroup and | |
3082 | * uncharge "memsw" account. | |
3083 | */ | |
3084 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 3085 | { |
8c7c6e34 | 3086 | struct mem_cgroup *memcg; |
a3b2d692 | 3087 | unsigned short id; |
8c7c6e34 KH |
3088 | |
3089 | if (!do_swap_account) | |
3090 | return; | |
3091 | ||
a3b2d692 KH |
3092 | id = swap_cgroup_record(ent, 0); |
3093 | rcu_read_lock(); | |
3094 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 3095 | if (memcg) { |
a3b2d692 KH |
3096 | /* |
3097 | * We uncharge this because swap is freed. | |
3098 | * This memcg can be obsolete one. We avoid calling css_tryget | |
3099 | */ | |
0c3e73e8 | 3100 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 3101 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 3102 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
3103 | mem_cgroup_put(memcg); |
3104 | } | |
a3b2d692 | 3105 | rcu_read_unlock(); |
d13d1443 | 3106 | } |
02491447 DN |
3107 | |
3108 | /** | |
3109 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3110 | * @entry: swap entry to be moved | |
3111 | * @from: mem_cgroup which the entry is moved from | |
3112 | * @to: mem_cgroup which the entry is moved to | |
483c30b5 | 3113 | * @need_fixup: whether we should fixup res_counters and refcounts. |
02491447 DN |
3114 | * |
3115 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3116 | * as the mem_cgroup's id of @from. | |
3117 | * | |
3118 | * Returns 0 on success, -EINVAL on failure. | |
3119 | * | |
3120 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
3121 | * both res and memsw, and called css_get(). | |
3122 | */ | |
3123 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
483c30b5 | 3124 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
02491447 DN |
3125 | { |
3126 | unsigned short old_id, new_id; | |
3127 | ||
3128 | old_id = css_id(&from->css); | |
3129 | new_id = css_id(&to->css); | |
3130 | ||
3131 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 3132 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 3133 | mem_cgroup_swap_statistics(to, true); |
02491447 | 3134 | /* |
483c30b5 DN |
3135 | * This function is only called from task migration context now. |
3136 | * It postpones res_counter and refcount handling till the end | |
3137 | * of task migration(mem_cgroup_clear_mc()) for performance | |
3138 | * improvement. But we cannot postpone mem_cgroup_get(to) | |
3139 | * because if the process that has been moved to @to does | |
3140 | * swap-in, the refcount of @to might be decreased to 0. | |
02491447 | 3141 | */ |
02491447 | 3142 | mem_cgroup_get(to); |
483c30b5 DN |
3143 | if (need_fixup) { |
3144 | if (!mem_cgroup_is_root(from)) | |
3145 | res_counter_uncharge(&from->memsw, PAGE_SIZE); | |
3146 | mem_cgroup_put(from); | |
3147 | /* | |
3148 | * we charged both to->res and to->memsw, so we should | |
3149 | * uncharge to->res. | |
3150 | */ | |
3151 | if (!mem_cgroup_is_root(to)) | |
3152 | res_counter_uncharge(&to->res, PAGE_SIZE); | |
483c30b5 | 3153 | } |
02491447 DN |
3154 | return 0; |
3155 | } | |
3156 | return -EINVAL; | |
3157 | } | |
3158 | #else | |
3159 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
483c30b5 | 3160 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
02491447 DN |
3161 | { |
3162 | return -EINVAL; | |
3163 | } | |
8c7c6e34 | 3164 | #endif |
d13d1443 | 3165 | |
ae41be37 | 3166 | /* |
01b1ae63 KH |
3167 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
3168 | * page belongs to. | |
ae41be37 | 3169 | */ |
ac39cf8c | 3170 | int mem_cgroup_prepare_migration(struct page *page, |
72835c86 | 3171 | struct page *newpage, struct mem_cgroup **memcgp, gfp_t gfp_mask) |
ae41be37 | 3172 | { |
c0ff4b85 | 3173 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 3174 | struct page_cgroup *pc; |
ac39cf8c | 3175 | enum charge_type ctype; |
e8589cc1 | 3176 | int ret = 0; |
8869b8f6 | 3177 | |
72835c86 | 3178 | *memcgp = NULL; |
56039efa | 3179 | |
ec168510 | 3180 | VM_BUG_ON(PageTransHuge(page)); |
f8d66542 | 3181 | if (mem_cgroup_disabled()) |
4077960e BS |
3182 | return 0; |
3183 | ||
52d4b9ac KH |
3184 | pc = lookup_page_cgroup(page); |
3185 | lock_page_cgroup(pc); | |
3186 | if (PageCgroupUsed(pc)) { | |
c0ff4b85 R |
3187 | memcg = pc->mem_cgroup; |
3188 | css_get(&memcg->css); | |
ac39cf8c | 3189 | /* |
3190 | * At migrating an anonymous page, its mapcount goes down | |
3191 | * to 0 and uncharge() will be called. But, even if it's fully | |
3192 | * unmapped, migration may fail and this page has to be | |
3193 | * charged again. We set MIGRATION flag here and delay uncharge | |
3194 | * until end_migration() is called | |
3195 | * | |
3196 | * Corner Case Thinking | |
3197 | * A) | |
3198 | * When the old page was mapped as Anon and it's unmap-and-freed | |
3199 | * while migration was ongoing. | |
3200 | * If unmap finds the old page, uncharge() of it will be delayed | |
3201 | * until end_migration(). If unmap finds a new page, it's | |
3202 | * uncharged when it make mapcount to be 1->0. If unmap code | |
3203 | * finds swap_migration_entry, the new page will not be mapped | |
3204 | * and end_migration() will find it(mapcount==0). | |
3205 | * | |
3206 | * B) | |
3207 | * When the old page was mapped but migraion fails, the kernel | |
3208 | * remaps it. A charge for it is kept by MIGRATION flag even | |
3209 | * if mapcount goes down to 0. We can do remap successfully | |
3210 | * without charging it again. | |
3211 | * | |
3212 | * C) | |
3213 | * The "old" page is under lock_page() until the end of | |
3214 | * migration, so, the old page itself will not be swapped-out. | |
3215 | * If the new page is swapped out before end_migraton, our | |
3216 | * hook to usual swap-out path will catch the event. | |
3217 | */ | |
3218 | if (PageAnon(page)) | |
3219 | SetPageCgroupMigration(pc); | |
e8589cc1 | 3220 | } |
52d4b9ac | 3221 | unlock_page_cgroup(pc); |
ac39cf8c | 3222 | /* |
3223 | * If the page is not charged at this point, | |
3224 | * we return here. | |
3225 | */ | |
c0ff4b85 | 3226 | if (!memcg) |
ac39cf8c | 3227 | return 0; |
01b1ae63 | 3228 | |
72835c86 JW |
3229 | *memcgp = memcg; |
3230 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, 1, memcgp, false); | |
c0ff4b85 | 3231 | css_put(&memcg->css);/* drop extra refcnt */ |
38c5d72f | 3232 | if (ret) { |
ac39cf8c | 3233 | if (PageAnon(page)) { |
3234 | lock_page_cgroup(pc); | |
3235 | ClearPageCgroupMigration(pc); | |
3236 | unlock_page_cgroup(pc); | |
3237 | /* | |
3238 | * The old page may be fully unmapped while we kept it. | |
3239 | */ | |
3240 | mem_cgroup_uncharge_page(page); | |
3241 | } | |
38c5d72f | 3242 | /* we'll need to revisit this error code (we have -EINTR) */ |
ac39cf8c | 3243 | return -ENOMEM; |
e8589cc1 | 3244 | } |
ac39cf8c | 3245 | /* |
3246 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
3247 | * is called before end_migration, we can catch all events on this new | |
3248 | * page. In the case new page is migrated but not remapped, new page's | |
3249 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
3250 | */ | |
3251 | pc = lookup_page_cgroup(newpage); | |
3252 | if (PageAnon(page)) | |
3253 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
3254 | else if (page_is_file_cache(page)) | |
3255 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
3256 | else | |
3257 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
c0ff4b85 | 3258 | __mem_cgroup_commit_charge(memcg, page, 1, pc, ctype); |
e8589cc1 | 3259 | return ret; |
ae41be37 | 3260 | } |
8869b8f6 | 3261 | |
69029cd5 | 3262 | /* remove redundant charge if migration failed*/ |
c0ff4b85 | 3263 | void mem_cgroup_end_migration(struct mem_cgroup *memcg, |
50de1dd9 | 3264 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 3265 | { |
ac39cf8c | 3266 | struct page *used, *unused; |
01b1ae63 | 3267 | struct page_cgroup *pc; |
01b1ae63 | 3268 | |
c0ff4b85 | 3269 | if (!memcg) |
01b1ae63 | 3270 | return; |
ac39cf8c | 3271 | /* blocks rmdir() */ |
c0ff4b85 | 3272 | cgroup_exclude_rmdir(&memcg->css); |
50de1dd9 | 3273 | if (!migration_ok) { |
ac39cf8c | 3274 | used = oldpage; |
3275 | unused = newpage; | |
01b1ae63 | 3276 | } else { |
ac39cf8c | 3277 | used = newpage; |
01b1ae63 KH |
3278 | unused = oldpage; |
3279 | } | |
69029cd5 | 3280 | /* |
ac39cf8c | 3281 | * We disallowed uncharge of pages under migration because mapcount |
3282 | * of the page goes down to zero, temporarly. | |
3283 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 3284 | */ |
ac39cf8c | 3285 | pc = lookup_page_cgroup(oldpage); |
3286 | lock_page_cgroup(pc); | |
3287 | ClearPageCgroupMigration(pc); | |
3288 | unlock_page_cgroup(pc); | |
01b1ae63 | 3289 | |
ac39cf8c | 3290 | __mem_cgroup_uncharge_common(unused, MEM_CGROUP_CHARGE_TYPE_FORCE); |
3291 | ||
01b1ae63 | 3292 | /* |
ac39cf8c | 3293 | * If a page is a file cache, radix-tree replacement is very atomic |
3294 | * and we can skip this check. When it was an Anon page, its mapcount | |
3295 | * goes down to 0. But because we added MIGRATION flage, it's not | |
3296 | * uncharged yet. There are several case but page->mapcount check | |
3297 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
3298 | * check. (see prepare_charge() also) | |
69029cd5 | 3299 | */ |
ac39cf8c | 3300 | if (PageAnon(used)) |
3301 | mem_cgroup_uncharge_page(used); | |
88703267 | 3302 | /* |
ac39cf8c | 3303 | * At migration, we may charge account against cgroup which has no |
3304 | * tasks. | |
88703267 KH |
3305 | * So, rmdir()->pre_destroy() can be called while we do this charge. |
3306 | * In that case, we need to call pre_destroy() again. check it here. | |
3307 | */ | |
c0ff4b85 | 3308 | cgroup_release_and_wakeup_rmdir(&memcg->css); |
ae41be37 | 3309 | } |
78fb7466 | 3310 | |
ab936cbc KH |
3311 | /* |
3312 | * At replace page cache, newpage is not under any memcg but it's on | |
3313 | * LRU. So, this function doesn't touch res_counter but handles LRU | |
3314 | * in correct way. Both pages are locked so we cannot race with uncharge. | |
3315 | */ | |
3316 | void mem_cgroup_replace_page_cache(struct page *oldpage, | |
3317 | struct page *newpage) | |
3318 | { | |
3319 | struct mem_cgroup *memcg; | |
3320 | struct page_cgroup *pc; | |
ab936cbc | 3321 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
ab936cbc KH |
3322 | |
3323 | if (mem_cgroup_disabled()) | |
3324 | return; | |
3325 | ||
3326 | pc = lookup_page_cgroup(oldpage); | |
3327 | /* fix accounting on old pages */ | |
3328 | lock_page_cgroup(pc); | |
3329 | memcg = pc->mem_cgroup; | |
3330 | mem_cgroup_charge_statistics(memcg, PageCgroupCache(pc), -1); | |
3331 | ClearPageCgroupUsed(pc); | |
3332 | unlock_page_cgroup(pc); | |
3333 | ||
3334 | if (PageSwapBacked(oldpage)) | |
3335 | type = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
3336 | ||
ab936cbc KH |
3337 | /* |
3338 | * Even if newpage->mapping was NULL before starting replacement, | |
3339 | * the newpage may be on LRU(or pagevec for LRU) already. We lock | |
3340 | * LRU while we overwrite pc->mem_cgroup. | |
3341 | */ | |
36b62ad5 | 3342 | __mem_cgroup_commit_charge_lrucare(newpage, memcg, type); |
ab936cbc KH |
3343 | } |
3344 | ||
f212ad7c DN |
3345 | #ifdef CONFIG_DEBUG_VM |
3346 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
3347 | { | |
3348 | struct page_cgroup *pc; | |
3349 | ||
3350 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
3351 | /* |
3352 | * Can be NULL while feeding pages into the page allocator for | |
3353 | * the first time, i.e. during boot or memory hotplug; | |
3354 | * or when mem_cgroup_disabled(). | |
3355 | */ | |
f212ad7c DN |
3356 | if (likely(pc) && PageCgroupUsed(pc)) |
3357 | return pc; | |
3358 | return NULL; | |
3359 | } | |
3360 | ||
3361 | bool mem_cgroup_bad_page_check(struct page *page) | |
3362 | { | |
3363 | if (mem_cgroup_disabled()) | |
3364 | return false; | |
3365 | ||
3366 | return lookup_page_cgroup_used(page) != NULL; | |
3367 | } | |
3368 | ||
3369 | void mem_cgroup_print_bad_page(struct page *page) | |
3370 | { | |
3371 | struct page_cgroup *pc; | |
3372 | ||
3373 | pc = lookup_page_cgroup_used(page); | |
3374 | if (pc) { | |
3375 | int ret = -1; | |
3376 | char *path; | |
3377 | ||
3378 | printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p", | |
3379 | pc, pc->flags, pc->mem_cgroup); | |
3380 | ||
3381 | path = kmalloc(PATH_MAX, GFP_KERNEL); | |
3382 | if (path) { | |
3383 | rcu_read_lock(); | |
3384 | ret = cgroup_path(pc->mem_cgroup->css.cgroup, | |
3385 | path, PATH_MAX); | |
3386 | rcu_read_unlock(); | |
3387 | } | |
3388 | ||
3389 | printk(KERN_CONT "(%s)\n", | |
3390 | (ret < 0) ? "cannot get the path" : path); | |
3391 | kfree(path); | |
3392 | } | |
3393 | } | |
3394 | #endif | |
3395 | ||
8c7c6e34 KH |
3396 | static DEFINE_MUTEX(set_limit_mutex); |
3397 | ||
d38d2a75 | 3398 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 3399 | unsigned long long val) |
628f4235 | 3400 | { |
81d39c20 | 3401 | int retry_count; |
3c11ecf4 | 3402 | u64 memswlimit, memlimit; |
628f4235 | 3403 | int ret = 0; |
81d39c20 KH |
3404 | int children = mem_cgroup_count_children(memcg); |
3405 | u64 curusage, oldusage; | |
3c11ecf4 | 3406 | int enlarge; |
81d39c20 KH |
3407 | |
3408 | /* | |
3409 | * For keeping hierarchical_reclaim simple, how long we should retry | |
3410 | * is depends on callers. We set our retry-count to be function | |
3411 | * of # of children which we should visit in this loop. | |
3412 | */ | |
3413 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
3414 | ||
3415 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 3416 | |
3c11ecf4 | 3417 | enlarge = 0; |
8c7c6e34 | 3418 | while (retry_count) { |
628f4235 KH |
3419 | if (signal_pending(current)) { |
3420 | ret = -EINTR; | |
3421 | break; | |
3422 | } | |
8c7c6e34 KH |
3423 | /* |
3424 | * Rather than hide all in some function, I do this in | |
3425 | * open coded manner. You see what this really does. | |
c0ff4b85 | 3426 | * We have to guarantee memcg->res.limit < memcg->memsw.limit. |
8c7c6e34 KH |
3427 | */ |
3428 | mutex_lock(&set_limit_mutex); | |
3429 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3430 | if (memswlimit < val) { | |
3431 | ret = -EINVAL; | |
3432 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
3433 | break; |
3434 | } | |
3c11ecf4 KH |
3435 | |
3436 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3437 | if (memlimit < val) | |
3438 | enlarge = 1; | |
3439 | ||
8c7c6e34 | 3440 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
3441 | if (!ret) { |
3442 | if (memswlimit == val) | |
3443 | memcg->memsw_is_minimum = true; | |
3444 | else | |
3445 | memcg->memsw_is_minimum = false; | |
3446 | } | |
8c7c6e34 KH |
3447 | mutex_unlock(&set_limit_mutex); |
3448 | ||
3449 | if (!ret) | |
3450 | break; | |
3451 | ||
5660048c JW |
3452 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
3453 | MEM_CGROUP_RECLAIM_SHRINK); | |
81d39c20 KH |
3454 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
3455 | /* Usage is reduced ? */ | |
3456 | if (curusage >= oldusage) | |
3457 | retry_count--; | |
3458 | else | |
3459 | oldusage = curusage; | |
8c7c6e34 | 3460 | } |
3c11ecf4 KH |
3461 | if (!ret && enlarge) |
3462 | memcg_oom_recover(memcg); | |
14797e23 | 3463 | |
8c7c6e34 KH |
3464 | return ret; |
3465 | } | |
3466 | ||
338c8431 LZ |
3467 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3468 | unsigned long long val) | |
8c7c6e34 | 3469 | { |
81d39c20 | 3470 | int retry_count; |
3c11ecf4 | 3471 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
3472 | int children = mem_cgroup_count_children(memcg); |
3473 | int ret = -EBUSY; | |
3c11ecf4 | 3474 | int enlarge = 0; |
8c7c6e34 | 3475 | |
81d39c20 KH |
3476 | /* see mem_cgroup_resize_res_limit */ |
3477 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
3478 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
3479 | while (retry_count) { |
3480 | if (signal_pending(current)) { | |
3481 | ret = -EINTR; | |
3482 | break; | |
3483 | } | |
3484 | /* | |
3485 | * Rather than hide all in some function, I do this in | |
3486 | * open coded manner. You see what this really does. | |
c0ff4b85 | 3487 | * We have to guarantee memcg->res.limit < memcg->memsw.limit. |
8c7c6e34 KH |
3488 | */ |
3489 | mutex_lock(&set_limit_mutex); | |
3490 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3491 | if (memlimit > val) { | |
3492 | ret = -EINVAL; | |
3493 | mutex_unlock(&set_limit_mutex); | |
3494 | break; | |
3495 | } | |
3c11ecf4 KH |
3496 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
3497 | if (memswlimit < val) | |
3498 | enlarge = 1; | |
8c7c6e34 | 3499 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
3500 | if (!ret) { |
3501 | if (memlimit == val) | |
3502 | memcg->memsw_is_minimum = true; | |
3503 | else | |
3504 | memcg->memsw_is_minimum = false; | |
3505 | } | |
8c7c6e34 KH |
3506 | mutex_unlock(&set_limit_mutex); |
3507 | ||
3508 | if (!ret) | |
3509 | break; | |
3510 | ||
5660048c JW |
3511 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
3512 | MEM_CGROUP_RECLAIM_NOSWAP | | |
3513 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 3514 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 3515 | /* Usage is reduced ? */ |
8c7c6e34 | 3516 | if (curusage >= oldusage) |
628f4235 | 3517 | retry_count--; |
81d39c20 KH |
3518 | else |
3519 | oldusage = curusage; | |
628f4235 | 3520 | } |
3c11ecf4 KH |
3521 | if (!ret && enlarge) |
3522 | memcg_oom_recover(memcg); | |
628f4235 KH |
3523 | return ret; |
3524 | } | |
3525 | ||
4e416953 | 3526 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
0ae5e89c YH |
3527 | gfp_t gfp_mask, |
3528 | unsigned long *total_scanned) | |
4e416953 BS |
3529 | { |
3530 | unsigned long nr_reclaimed = 0; | |
3531 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3532 | unsigned long reclaimed; | |
3533 | int loop = 0; | |
3534 | struct mem_cgroup_tree_per_zone *mctz; | |
ef8745c1 | 3535 | unsigned long long excess; |
0ae5e89c | 3536 | unsigned long nr_scanned; |
4e416953 BS |
3537 | |
3538 | if (order > 0) | |
3539 | return 0; | |
3540 | ||
00918b6a | 3541 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); |
4e416953 BS |
3542 | /* |
3543 | * This loop can run a while, specially if mem_cgroup's continuously | |
3544 | * keep exceeding their soft limit and putting the system under | |
3545 | * pressure | |
3546 | */ | |
3547 | do { | |
3548 | if (next_mz) | |
3549 | mz = next_mz; | |
3550 | else | |
3551 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3552 | if (!mz) | |
3553 | break; | |
3554 | ||
0ae5e89c | 3555 | nr_scanned = 0; |
5660048c JW |
3556 | reclaimed = mem_cgroup_soft_reclaim(mz->mem, zone, |
3557 | gfp_mask, &nr_scanned); | |
4e416953 | 3558 | nr_reclaimed += reclaimed; |
0ae5e89c | 3559 | *total_scanned += nr_scanned; |
4e416953 BS |
3560 | spin_lock(&mctz->lock); |
3561 | ||
3562 | /* | |
3563 | * If we failed to reclaim anything from this memory cgroup | |
3564 | * it is time to move on to the next cgroup | |
3565 | */ | |
3566 | next_mz = NULL; | |
3567 | if (!reclaimed) { | |
3568 | do { | |
3569 | /* | |
3570 | * Loop until we find yet another one. | |
3571 | * | |
3572 | * By the time we get the soft_limit lock | |
3573 | * again, someone might have aded the | |
3574 | * group back on the RB tree. Iterate to | |
3575 | * make sure we get a different mem. | |
3576 | * mem_cgroup_largest_soft_limit_node returns | |
3577 | * NULL if no other cgroup is present on | |
3578 | * the tree | |
3579 | */ | |
3580 | next_mz = | |
3581 | __mem_cgroup_largest_soft_limit_node(mctz); | |
39cc98f1 | 3582 | if (next_mz == mz) |
4e416953 | 3583 | css_put(&next_mz->mem->css); |
39cc98f1 | 3584 | else /* next_mz == NULL or other memcg */ |
4e416953 BS |
3585 | break; |
3586 | } while (1); | |
3587 | } | |
4e416953 | 3588 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); |
ef8745c1 | 3589 | excess = res_counter_soft_limit_excess(&mz->mem->res); |
4e416953 BS |
3590 | /* |
3591 | * One school of thought says that we should not add | |
3592 | * back the node to the tree if reclaim returns 0. | |
3593 | * But our reclaim could return 0, simply because due | |
3594 | * to priority we are exposing a smaller subset of | |
3595 | * memory to reclaim from. Consider this as a longer | |
3596 | * term TODO. | |
3597 | */ | |
ef8745c1 KH |
3598 | /* If excess == 0, no tree ops */ |
3599 | __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess); | |
4e416953 BS |
3600 | spin_unlock(&mctz->lock); |
3601 | css_put(&mz->mem->css); | |
3602 | loop++; | |
3603 | /* | |
3604 | * Could not reclaim anything and there are no more | |
3605 | * mem cgroups to try or we seem to be looping without | |
3606 | * reclaiming anything. | |
3607 | */ | |
3608 | if (!nr_reclaimed && | |
3609 | (next_mz == NULL || | |
3610 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3611 | break; | |
3612 | } while (!nr_reclaimed); | |
3613 | if (next_mz) | |
3614 | css_put(&next_mz->mem->css); | |
3615 | return nr_reclaimed; | |
3616 | } | |
3617 | ||
cc847582 KH |
3618 | /* |
3619 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
3620 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
3621 | */ | |
c0ff4b85 | 3622 | static int mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
08e552c6 | 3623 | int node, int zid, enum lru_list lru) |
cc847582 | 3624 | { |
08e552c6 | 3625 | struct mem_cgroup_per_zone *mz; |
08e552c6 | 3626 | unsigned long flags, loop; |
072c56c1 | 3627 | struct list_head *list; |
925b7673 JW |
3628 | struct page *busy; |
3629 | struct zone *zone; | |
f817ed48 | 3630 | int ret = 0; |
072c56c1 | 3631 | |
08e552c6 | 3632 | zone = &NODE_DATA(node)->node_zones[zid]; |
c0ff4b85 | 3633 | mz = mem_cgroup_zoneinfo(memcg, node, zid); |
6290df54 | 3634 | list = &mz->lruvec.lists[lru]; |
cc847582 | 3635 | |
f817ed48 KH |
3636 | loop = MEM_CGROUP_ZSTAT(mz, lru); |
3637 | /* give some margin against EBUSY etc...*/ | |
3638 | loop += 256; | |
3639 | busy = NULL; | |
3640 | while (loop--) { | |
925b7673 | 3641 | struct page_cgroup *pc; |
5564e88b JW |
3642 | struct page *page; |
3643 | ||
f817ed48 | 3644 | ret = 0; |
08e552c6 | 3645 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 3646 | if (list_empty(list)) { |
08e552c6 | 3647 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 3648 | break; |
f817ed48 | 3649 | } |
925b7673 JW |
3650 | page = list_entry(list->prev, struct page, lru); |
3651 | if (busy == page) { | |
3652 | list_move(&page->lru, list); | |
648bcc77 | 3653 | busy = NULL; |
08e552c6 | 3654 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
3655 | continue; |
3656 | } | |
08e552c6 | 3657 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 3658 | |
925b7673 | 3659 | pc = lookup_page_cgroup(page); |
5564e88b | 3660 | |
c0ff4b85 | 3661 | ret = mem_cgroup_move_parent(page, pc, memcg, GFP_KERNEL); |
38c5d72f | 3662 | if (ret == -ENOMEM || ret == -EINTR) |
52d4b9ac | 3663 | break; |
f817ed48 KH |
3664 | |
3665 | if (ret == -EBUSY || ret == -EINVAL) { | |
3666 | /* found lock contention or "pc" is obsolete. */ | |
925b7673 | 3667 | busy = page; |
f817ed48 KH |
3668 | cond_resched(); |
3669 | } else | |
3670 | busy = NULL; | |
cc847582 | 3671 | } |
08e552c6 | 3672 | |
f817ed48 KH |
3673 | if (!ret && !list_empty(list)) |
3674 | return -EBUSY; | |
3675 | return ret; | |
cc847582 KH |
3676 | } |
3677 | ||
3678 | /* | |
3679 | * make mem_cgroup's charge to be 0 if there is no task. | |
3680 | * This enables deleting this mem_cgroup. | |
3681 | */ | |
c0ff4b85 | 3682 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg, bool free_all) |
cc847582 | 3683 | { |
f817ed48 KH |
3684 | int ret; |
3685 | int node, zid, shrink; | |
3686 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c0ff4b85 | 3687 | struct cgroup *cgrp = memcg->css.cgroup; |
8869b8f6 | 3688 | |
c0ff4b85 | 3689 | css_get(&memcg->css); |
f817ed48 KH |
3690 | |
3691 | shrink = 0; | |
c1e862c1 KH |
3692 | /* should free all ? */ |
3693 | if (free_all) | |
3694 | goto try_to_free; | |
f817ed48 | 3695 | move_account: |
fce66477 | 3696 | do { |
f817ed48 | 3697 | ret = -EBUSY; |
c1e862c1 KH |
3698 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
3699 | goto out; | |
3700 | ret = -EINTR; | |
3701 | if (signal_pending(current)) | |
cc847582 | 3702 | goto out; |
52d4b9ac KH |
3703 | /* This is for making all *used* pages to be on LRU. */ |
3704 | lru_add_drain_all(); | |
c0ff4b85 | 3705 | drain_all_stock_sync(memcg); |
f817ed48 | 3706 | ret = 0; |
c0ff4b85 | 3707 | mem_cgroup_start_move(memcg); |
299b4eaa | 3708 | for_each_node_state(node, N_HIGH_MEMORY) { |
f817ed48 | 3709 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { |
b69408e8 | 3710 | enum lru_list l; |
f817ed48 | 3711 | for_each_lru(l) { |
c0ff4b85 | 3712 | ret = mem_cgroup_force_empty_list(memcg, |
08e552c6 | 3713 | node, zid, l); |
f817ed48 KH |
3714 | if (ret) |
3715 | break; | |
3716 | } | |
1ecaab2b | 3717 | } |
f817ed48 KH |
3718 | if (ret) |
3719 | break; | |
3720 | } | |
c0ff4b85 R |
3721 | mem_cgroup_end_move(memcg); |
3722 | memcg_oom_recover(memcg); | |
f817ed48 KH |
3723 | /* it seems parent cgroup doesn't have enough mem */ |
3724 | if (ret == -ENOMEM) | |
3725 | goto try_to_free; | |
52d4b9ac | 3726 | cond_resched(); |
fce66477 | 3727 | /* "ret" should also be checked to ensure all lists are empty. */ |
c0ff4b85 | 3728 | } while (memcg->res.usage > 0 || ret); |
cc847582 | 3729 | out: |
c0ff4b85 | 3730 | css_put(&memcg->css); |
cc847582 | 3731 | return ret; |
f817ed48 KH |
3732 | |
3733 | try_to_free: | |
c1e862c1 KH |
3734 | /* returns EBUSY if there is a task or if we come here twice. */ |
3735 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
f817ed48 KH |
3736 | ret = -EBUSY; |
3737 | goto out; | |
3738 | } | |
c1e862c1 KH |
3739 | /* we call try-to-free pages for make this cgroup empty */ |
3740 | lru_add_drain_all(); | |
f817ed48 KH |
3741 | /* try to free all pages in this cgroup */ |
3742 | shrink = 1; | |
c0ff4b85 | 3743 | while (nr_retries && memcg->res.usage > 0) { |
f817ed48 | 3744 | int progress; |
c1e862c1 KH |
3745 | |
3746 | if (signal_pending(current)) { | |
3747 | ret = -EINTR; | |
3748 | goto out; | |
3749 | } | |
c0ff4b85 | 3750 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, |
185efc0f | 3751 | false); |
c1e862c1 | 3752 | if (!progress) { |
f817ed48 | 3753 | nr_retries--; |
c1e862c1 | 3754 | /* maybe some writeback is necessary */ |
8aa7e847 | 3755 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3756 | } |
f817ed48 KH |
3757 | |
3758 | } | |
08e552c6 | 3759 | lru_add_drain(); |
f817ed48 | 3760 | /* try move_account...there may be some *locked* pages. */ |
fce66477 | 3761 | goto move_account; |
cc847582 KH |
3762 | } |
3763 | ||
c1e862c1 KH |
3764 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
3765 | { | |
3766 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
3767 | } | |
3768 | ||
3769 | ||
18f59ea7 BS |
3770 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
3771 | { | |
3772 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
3773 | } | |
3774 | ||
3775 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
3776 | u64 val) | |
3777 | { | |
3778 | int retval = 0; | |
c0ff4b85 | 3779 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
18f59ea7 | 3780 | struct cgroup *parent = cont->parent; |
c0ff4b85 | 3781 | struct mem_cgroup *parent_memcg = NULL; |
18f59ea7 BS |
3782 | |
3783 | if (parent) | |
c0ff4b85 | 3784 | parent_memcg = mem_cgroup_from_cont(parent); |
18f59ea7 BS |
3785 | |
3786 | cgroup_lock(); | |
3787 | /* | |
af901ca1 | 3788 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3789 | * in the child subtrees. If it is unset, then the change can |
3790 | * occur, provided the current cgroup has no children. | |
3791 | * | |
3792 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3793 | * set if there are no children. | |
3794 | */ | |
c0ff4b85 | 3795 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 BS |
3796 | (val == 1 || val == 0)) { |
3797 | if (list_empty(&cont->children)) | |
c0ff4b85 | 3798 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3799 | else |
3800 | retval = -EBUSY; | |
3801 | } else | |
3802 | retval = -EINVAL; | |
3803 | cgroup_unlock(); | |
3804 | ||
3805 | return retval; | |
3806 | } | |
3807 | ||
0c3e73e8 | 3808 | |
c0ff4b85 | 3809 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg, |
7a159cc9 | 3810 | enum mem_cgroup_stat_index idx) |
0c3e73e8 | 3811 | { |
7d74b06f | 3812 | struct mem_cgroup *iter; |
7a159cc9 | 3813 | long val = 0; |
0c3e73e8 | 3814 | |
7a159cc9 | 3815 | /* Per-cpu values can be negative, use a signed accumulator */ |
c0ff4b85 | 3816 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f KH |
3817 | val += mem_cgroup_read_stat(iter, idx); |
3818 | ||
3819 | if (val < 0) /* race ? */ | |
3820 | val = 0; | |
3821 | return val; | |
0c3e73e8 BS |
3822 | } |
3823 | ||
c0ff4b85 | 3824 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
104f3928 | 3825 | { |
7d74b06f | 3826 | u64 val; |
104f3928 | 3827 | |
c0ff4b85 | 3828 | if (!mem_cgroup_is_root(memcg)) { |
104f3928 | 3829 | if (!swap) |
65c64ce8 | 3830 | return res_counter_read_u64(&memcg->res, RES_USAGE); |
104f3928 | 3831 | else |
65c64ce8 | 3832 | return res_counter_read_u64(&memcg->memsw, RES_USAGE); |
104f3928 KS |
3833 | } |
3834 | ||
c0ff4b85 R |
3835 | val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE); |
3836 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS); | |
104f3928 | 3837 | |
7d74b06f | 3838 | if (swap) |
c0ff4b85 | 3839 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAPOUT); |
104f3928 KS |
3840 | |
3841 | return val << PAGE_SHIFT; | |
3842 | } | |
3843 | ||
2c3daa72 | 3844 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 3845 | { |
c0ff4b85 | 3846 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
104f3928 | 3847 | u64 val; |
8c7c6e34 KH |
3848 | int type, name; |
3849 | ||
3850 | type = MEMFILE_TYPE(cft->private); | |
3851 | name = MEMFILE_ATTR(cft->private); | |
3852 | switch (type) { | |
3853 | case _MEM: | |
104f3928 | 3854 | if (name == RES_USAGE) |
c0ff4b85 | 3855 | val = mem_cgroup_usage(memcg, false); |
104f3928 | 3856 | else |
c0ff4b85 | 3857 | val = res_counter_read_u64(&memcg->res, name); |
8c7c6e34 KH |
3858 | break; |
3859 | case _MEMSWAP: | |
104f3928 | 3860 | if (name == RES_USAGE) |
c0ff4b85 | 3861 | val = mem_cgroup_usage(memcg, true); |
104f3928 | 3862 | else |
c0ff4b85 | 3863 | val = res_counter_read_u64(&memcg->memsw, name); |
8c7c6e34 KH |
3864 | break; |
3865 | default: | |
3866 | BUG(); | |
3867 | break; | |
3868 | } | |
3869 | return val; | |
8cdea7c0 | 3870 | } |
628f4235 KH |
3871 | /* |
3872 | * The user of this function is... | |
3873 | * RES_LIMIT. | |
3874 | */ | |
856c13aa PM |
3875 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
3876 | const char *buffer) | |
8cdea7c0 | 3877 | { |
628f4235 | 3878 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 3879 | int type, name; |
628f4235 KH |
3880 | unsigned long long val; |
3881 | int ret; | |
3882 | ||
8c7c6e34 KH |
3883 | type = MEMFILE_TYPE(cft->private); |
3884 | name = MEMFILE_ATTR(cft->private); | |
3885 | switch (name) { | |
628f4235 | 3886 | case RES_LIMIT: |
4b3bde4c BS |
3887 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3888 | ret = -EINVAL; | |
3889 | break; | |
3890 | } | |
628f4235 KH |
3891 | /* This function does all necessary parse...reuse it */ |
3892 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
3893 | if (ret) |
3894 | break; | |
3895 | if (type == _MEM) | |
628f4235 | 3896 | ret = mem_cgroup_resize_limit(memcg, val); |
8c7c6e34 KH |
3897 | else |
3898 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
628f4235 | 3899 | break; |
296c81d8 BS |
3900 | case RES_SOFT_LIMIT: |
3901 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
3902 | if (ret) | |
3903 | break; | |
3904 | /* | |
3905 | * For memsw, soft limits are hard to implement in terms | |
3906 | * of semantics, for now, we support soft limits for | |
3907 | * control without swap | |
3908 | */ | |
3909 | if (type == _MEM) | |
3910 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
3911 | else | |
3912 | ret = -EINVAL; | |
3913 | break; | |
628f4235 KH |
3914 | default: |
3915 | ret = -EINVAL; /* should be BUG() ? */ | |
3916 | break; | |
3917 | } | |
3918 | return ret; | |
8cdea7c0 BS |
3919 | } |
3920 | ||
fee7b548 KH |
3921 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
3922 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
3923 | { | |
3924 | struct cgroup *cgroup; | |
3925 | unsigned long long min_limit, min_memsw_limit, tmp; | |
3926 | ||
3927 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3928 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3929 | cgroup = memcg->css.cgroup; | |
3930 | if (!memcg->use_hierarchy) | |
3931 | goto out; | |
3932 | ||
3933 | while (cgroup->parent) { | |
3934 | cgroup = cgroup->parent; | |
3935 | memcg = mem_cgroup_from_cont(cgroup); | |
3936 | if (!memcg->use_hierarchy) | |
3937 | break; | |
3938 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3939 | min_limit = min(min_limit, tmp); | |
3940 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3941 | min_memsw_limit = min(min_memsw_limit, tmp); | |
3942 | } | |
3943 | out: | |
3944 | *mem_limit = min_limit; | |
3945 | *memsw_limit = min_memsw_limit; | |
3946 | return; | |
3947 | } | |
3948 | ||
29f2a4da | 3949 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 | 3950 | { |
c0ff4b85 | 3951 | struct mem_cgroup *memcg; |
8c7c6e34 | 3952 | int type, name; |
c84872e1 | 3953 | |
c0ff4b85 | 3954 | memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 KH |
3955 | type = MEMFILE_TYPE(event); |
3956 | name = MEMFILE_ATTR(event); | |
3957 | switch (name) { | |
29f2a4da | 3958 | case RES_MAX_USAGE: |
8c7c6e34 | 3959 | if (type == _MEM) |
c0ff4b85 | 3960 | res_counter_reset_max(&memcg->res); |
8c7c6e34 | 3961 | else |
c0ff4b85 | 3962 | res_counter_reset_max(&memcg->memsw); |
29f2a4da PE |
3963 | break; |
3964 | case RES_FAILCNT: | |
8c7c6e34 | 3965 | if (type == _MEM) |
c0ff4b85 | 3966 | res_counter_reset_failcnt(&memcg->res); |
8c7c6e34 | 3967 | else |
c0ff4b85 | 3968 | res_counter_reset_failcnt(&memcg->memsw); |
29f2a4da PE |
3969 | break; |
3970 | } | |
f64c3f54 | 3971 | |
85cc59db | 3972 | return 0; |
c84872e1 PE |
3973 | } |
3974 | ||
7dc74be0 DN |
3975 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
3976 | struct cftype *cft) | |
3977 | { | |
3978 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; | |
3979 | } | |
3980 | ||
02491447 | 3981 | #ifdef CONFIG_MMU |
7dc74be0 DN |
3982 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
3983 | struct cftype *cft, u64 val) | |
3984 | { | |
c0ff4b85 | 3985 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
7dc74be0 DN |
3986 | |
3987 | if (val >= (1 << NR_MOVE_TYPE)) | |
3988 | return -EINVAL; | |
3989 | /* | |
3990 | * We check this value several times in both in can_attach() and | |
3991 | * attach(), so we need cgroup lock to prevent this value from being | |
3992 | * inconsistent. | |
3993 | */ | |
3994 | cgroup_lock(); | |
c0ff4b85 | 3995 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3996 | cgroup_unlock(); |
3997 | ||
3998 | return 0; | |
3999 | } | |
02491447 DN |
4000 | #else |
4001 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, | |
4002 | struct cftype *cft, u64 val) | |
4003 | { | |
4004 | return -ENOSYS; | |
4005 | } | |
4006 | #endif | |
7dc74be0 | 4007 | |
14067bb3 KH |
4008 | |
4009 | /* For read statistics */ | |
4010 | enum { | |
4011 | MCS_CACHE, | |
4012 | MCS_RSS, | |
d8046582 | 4013 | MCS_FILE_MAPPED, |
14067bb3 KH |
4014 | MCS_PGPGIN, |
4015 | MCS_PGPGOUT, | |
1dd3a273 | 4016 | MCS_SWAP, |
456f998e YH |
4017 | MCS_PGFAULT, |
4018 | MCS_PGMAJFAULT, | |
14067bb3 KH |
4019 | MCS_INACTIVE_ANON, |
4020 | MCS_ACTIVE_ANON, | |
4021 | MCS_INACTIVE_FILE, | |
4022 | MCS_ACTIVE_FILE, | |
4023 | MCS_UNEVICTABLE, | |
4024 | NR_MCS_STAT, | |
4025 | }; | |
4026 | ||
4027 | struct mcs_total_stat { | |
4028 | s64 stat[NR_MCS_STAT]; | |
d2ceb9b7 KH |
4029 | }; |
4030 | ||
14067bb3 KH |
4031 | struct { |
4032 | char *local_name; | |
4033 | char *total_name; | |
4034 | } memcg_stat_strings[NR_MCS_STAT] = { | |
4035 | {"cache", "total_cache"}, | |
4036 | {"rss", "total_rss"}, | |
d69b042f | 4037 | {"mapped_file", "total_mapped_file"}, |
14067bb3 KH |
4038 | {"pgpgin", "total_pgpgin"}, |
4039 | {"pgpgout", "total_pgpgout"}, | |
1dd3a273 | 4040 | {"swap", "total_swap"}, |
456f998e YH |
4041 | {"pgfault", "total_pgfault"}, |
4042 | {"pgmajfault", "total_pgmajfault"}, | |
14067bb3 KH |
4043 | {"inactive_anon", "total_inactive_anon"}, |
4044 | {"active_anon", "total_active_anon"}, | |
4045 | {"inactive_file", "total_inactive_file"}, | |
4046 | {"active_file", "total_active_file"}, | |
4047 | {"unevictable", "total_unevictable"} | |
4048 | }; | |
4049 | ||
4050 | ||
7d74b06f | 4051 | static void |
c0ff4b85 | 4052 | mem_cgroup_get_local_stat(struct mem_cgroup *memcg, struct mcs_total_stat *s) |
14067bb3 | 4053 | { |
14067bb3 KH |
4054 | s64 val; |
4055 | ||
4056 | /* per cpu stat */ | |
c0ff4b85 | 4057 | val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_CACHE); |
14067bb3 | 4058 | s->stat[MCS_CACHE] += val * PAGE_SIZE; |
c0ff4b85 | 4059 | val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_RSS); |
14067bb3 | 4060 | s->stat[MCS_RSS] += val * PAGE_SIZE; |
c0ff4b85 | 4061 | val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED); |
d8046582 | 4062 | s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE; |
c0ff4b85 | 4063 | val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGIN); |
14067bb3 | 4064 | s->stat[MCS_PGPGIN] += val; |
c0ff4b85 | 4065 | val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGOUT); |
14067bb3 | 4066 | s->stat[MCS_PGPGOUT] += val; |
1dd3a273 | 4067 | if (do_swap_account) { |
c0ff4b85 | 4068 | val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_SWAPOUT); |
1dd3a273 DN |
4069 | s->stat[MCS_SWAP] += val * PAGE_SIZE; |
4070 | } | |
c0ff4b85 | 4071 | val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGFAULT); |
456f998e | 4072 | s->stat[MCS_PGFAULT] += val; |
c0ff4b85 | 4073 | val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGMAJFAULT); |
456f998e | 4074 | s->stat[MCS_PGMAJFAULT] += val; |
14067bb3 KH |
4075 | |
4076 | /* per zone stat */ | |
c0ff4b85 | 4077 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_ANON)); |
14067bb3 | 4078 | s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE; |
c0ff4b85 | 4079 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_ANON)); |
14067bb3 | 4080 | s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE; |
c0ff4b85 | 4081 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_FILE)); |
14067bb3 | 4082 | s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE; |
c0ff4b85 | 4083 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_FILE)); |
14067bb3 | 4084 | s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE; |
c0ff4b85 | 4085 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE)); |
14067bb3 | 4086 | s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE; |
14067bb3 KH |
4087 | } |
4088 | ||
4089 | static void | |
c0ff4b85 | 4090 | mem_cgroup_get_total_stat(struct mem_cgroup *memcg, struct mcs_total_stat *s) |
14067bb3 | 4091 | { |
7d74b06f KH |
4092 | struct mem_cgroup *iter; |
4093 | ||
c0ff4b85 | 4094 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4095 | mem_cgroup_get_local_stat(iter, s); |
14067bb3 KH |
4096 | } |
4097 | ||
406eb0c9 YH |
4098 | #ifdef CONFIG_NUMA |
4099 | static int mem_control_numa_stat_show(struct seq_file *m, void *arg) | |
4100 | { | |
4101 | int nid; | |
4102 | unsigned long total_nr, file_nr, anon_nr, unevictable_nr; | |
4103 | unsigned long node_nr; | |
4104 | struct cgroup *cont = m->private; | |
4105 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); | |
4106 | ||
bb2a0de9 | 4107 | total_nr = mem_cgroup_nr_lru_pages(mem_cont, LRU_ALL); |
406eb0c9 YH |
4108 | seq_printf(m, "total=%lu", total_nr); |
4109 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
bb2a0de9 | 4110 | node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid, LRU_ALL); |
406eb0c9 YH |
4111 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4112 | } | |
4113 | seq_putc(m, '\n'); | |
4114 | ||
bb2a0de9 | 4115 | file_nr = mem_cgroup_nr_lru_pages(mem_cont, LRU_ALL_FILE); |
406eb0c9 YH |
4116 | seq_printf(m, "file=%lu", file_nr); |
4117 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
bb2a0de9 KH |
4118 | node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid, |
4119 | LRU_ALL_FILE); | |
406eb0c9 YH |
4120 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4121 | } | |
4122 | seq_putc(m, '\n'); | |
4123 | ||
bb2a0de9 | 4124 | anon_nr = mem_cgroup_nr_lru_pages(mem_cont, LRU_ALL_ANON); |
406eb0c9 YH |
4125 | seq_printf(m, "anon=%lu", anon_nr); |
4126 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
bb2a0de9 KH |
4127 | node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid, |
4128 | LRU_ALL_ANON); | |
406eb0c9 YH |
4129 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4130 | } | |
4131 | seq_putc(m, '\n'); | |
4132 | ||
bb2a0de9 | 4133 | unevictable_nr = mem_cgroup_nr_lru_pages(mem_cont, BIT(LRU_UNEVICTABLE)); |
406eb0c9 YH |
4134 | seq_printf(m, "unevictable=%lu", unevictable_nr); |
4135 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
bb2a0de9 KH |
4136 | node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid, |
4137 | BIT(LRU_UNEVICTABLE)); | |
406eb0c9 YH |
4138 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4139 | } | |
4140 | seq_putc(m, '\n'); | |
4141 | return 0; | |
4142 | } | |
4143 | #endif /* CONFIG_NUMA */ | |
4144 | ||
c64745cf PM |
4145 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
4146 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 4147 | { |
d2ceb9b7 | 4148 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
14067bb3 | 4149 | struct mcs_total_stat mystat; |
d2ceb9b7 KH |
4150 | int i; |
4151 | ||
14067bb3 KH |
4152 | memset(&mystat, 0, sizeof(mystat)); |
4153 | mem_cgroup_get_local_stat(mem_cont, &mystat); | |
d2ceb9b7 | 4154 | |
406eb0c9 | 4155 | |
1dd3a273 DN |
4156 | for (i = 0; i < NR_MCS_STAT; i++) { |
4157 | if (i == MCS_SWAP && !do_swap_account) | |
4158 | continue; | |
14067bb3 | 4159 | cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]); |
1dd3a273 | 4160 | } |
7b854121 | 4161 | |
14067bb3 | 4162 | /* Hierarchical information */ |
fee7b548 KH |
4163 | { |
4164 | unsigned long long limit, memsw_limit; | |
4165 | memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit); | |
4166 | cb->fill(cb, "hierarchical_memory_limit", limit); | |
4167 | if (do_swap_account) | |
4168 | cb->fill(cb, "hierarchical_memsw_limit", memsw_limit); | |
4169 | } | |
7f016ee8 | 4170 | |
14067bb3 KH |
4171 | memset(&mystat, 0, sizeof(mystat)); |
4172 | mem_cgroup_get_total_stat(mem_cont, &mystat); | |
1dd3a273 DN |
4173 | for (i = 0; i < NR_MCS_STAT; i++) { |
4174 | if (i == MCS_SWAP && !do_swap_account) | |
4175 | continue; | |
14067bb3 | 4176 | cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]); |
1dd3a273 | 4177 | } |
14067bb3 | 4178 | |
7f016ee8 | 4179 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
4180 | { |
4181 | int nid, zid; | |
4182 | struct mem_cgroup_per_zone *mz; | |
4183 | unsigned long recent_rotated[2] = {0, 0}; | |
4184 | unsigned long recent_scanned[2] = {0, 0}; | |
4185 | ||
4186 | for_each_online_node(nid) | |
4187 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
4188 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); | |
4189 | ||
4190 | recent_rotated[0] += | |
4191 | mz->reclaim_stat.recent_rotated[0]; | |
4192 | recent_rotated[1] += | |
4193 | mz->reclaim_stat.recent_rotated[1]; | |
4194 | recent_scanned[0] += | |
4195 | mz->reclaim_stat.recent_scanned[0]; | |
4196 | recent_scanned[1] += | |
4197 | mz->reclaim_stat.recent_scanned[1]; | |
4198 | } | |
4199 | cb->fill(cb, "recent_rotated_anon", recent_rotated[0]); | |
4200 | cb->fill(cb, "recent_rotated_file", recent_rotated[1]); | |
4201 | cb->fill(cb, "recent_scanned_anon", recent_scanned[0]); | |
4202 | cb->fill(cb, "recent_scanned_file", recent_scanned[1]); | |
4203 | } | |
4204 | #endif | |
4205 | ||
d2ceb9b7 KH |
4206 | return 0; |
4207 | } | |
4208 | ||
a7885eb8 KM |
4209 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
4210 | { | |
4211 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4212 | ||
1f4c025b | 4213 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4214 | } |
4215 | ||
4216 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
4217 | u64 val) | |
4218 | { | |
4219 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4220 | struct mem_cgroup *parent; | |
068b38c1 | 4221 | |
a7885eb8 KM |
4222 | if (val > 100) |
4223 | return -EINVAL; | |
4224 | ||
4225 | if (cgrp->parent == NULL) | |
4226 | return -EINVAL; | |
4227 | ||
4228 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 LZ |
4229 | |
4230 | cgroup_lock(); | |
4231 | ||
a7885eb8 KM |
4232 | /* If under hierarchy, only empty-root can set this value */ |
4233 | if ((parent->use_hierarchy) || | |
068b38c1 LZ |
4234 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
4235 | cgroup_unlock(); | |
a7885eb8 | 4236 | return -EINVAL; |
068b38c1 | 4237 | } |
a7885eb8 | 4238 | |
a7885eb8 | 4239 | memcg->swappiness = val; |
a7885eb8 | 4240 | |
068b38c1 LZ |
4241 | cgroup_unlock(); |
4242 | ||
a7885eb8 KM |
4243 | return 0; |
4244 | } | |
4245 | ||
2e72b634 KS |
4246 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4247 | { | |
4248 | struct mem_cgroup_threshold_ary *t; | |
4249 | u64 usage; | |
4250 | int i; | |
4251 | ||
4252 | rcu_read_lock(); | |
4253 | if (!swap) | |
2c488db2 | 4254 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4255 | else |
2c488db2 | 4256 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4257 | |
4258 | if (!t) | |
4259 | goto unlock; | |
4260 | ||
4261 | usage = mem_cgroup_usage(memcg, swap); | |
4262 | ||
4263 | /* | |
4264 | * current_threshold points to threshold just below usage. | |
4265 | * If it's not true, a threshold was crossed after last | |
4266 | * call of __mem_cgroup_threshold(). | |
4267 | */ | |
5407a562 | 4268 | i = t->current_threshold; |
2e72b634 KS |
4269 | |
4270 | /* | |
4271 | * Iterate backward over array of thresholds starting from | |
4272 | * current_threshold and check if a threshold is crossed. | |
4273 | * If none of thresholds below usage is crossed, we read | |
4274 | * only one element of the array here. | |
4275 | */ | |
4276 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4277 | eventfd_signal(t->entries[i].eventfd, 1); | |
4278 | ||
4279 | /* i = current_threshold + 1 */ | |
4280 | i++; | |
4281 | ||
4282 | /* | |
4283 | * Iterate forward over array of thresholds starting from | |
4284 | * current_threshold+1 and check if a threshold is crossed. | |
4285 | * If none of thresholds above usage is crossed, we read | |
4286 | * only one element of the array here. | |
4287 | */ | |
4288 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4289 | eventfd_signal(t->entries[i].eventfd, 1); | |
4290 | ||
4291 | /* Update current_threshold */ | |
5407a562 | 4292 | t->current_threshold = i - 1; |
2e72b634 KS |
4293 | unlock: |
4294 | rcu_read_unlock(); | |
4295 | } | |
4296 | ||
4297 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4298 | { | |
ad4ca5f4 KS |
4299 | while (memcg) { |
4300 | __mem_cgroup_threshold(memcg, false); | |
4301 | if (do_swap_account) | |
4302 | __mem_cgroup_threshold(memcg, true); | |
4303 | ||
4304 | memcg = parent_mem_cgroup(memcg); | |
4305 | } | |
2e72b634 KS |
4306 | } |
4307 | ||
4308 | static int compare_thresholds(const void *a, const void *b) | |
4309 | { | |
4310 | const struct mem_cgroup_threshold *_a = a; | |
4311 | const struct mem_cgroup_threshold *_b = b; | |
4312 | ||
4313 | return _a->threshold - _b->threshold; | |
4314 | } | |
4315 | ||
c0ff4b85 | 4316 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4317 | { |
4318 | struct mem_cgroup_eventfd_list *ev; | |
4319 | ||
c0ff4b85 | 4320 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 KH |
4321 | eventfd_signal(ev->eventfd, 1); |
4322 | return 0; | |
4323 | } | |
4324 | ||
c0ff4b85 | 4325 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4326 | { |
7d74b06f KH |
4327 | struct mem_cgroup *iter; |
4328 | ||
c0ff4b85 | 4329 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4330 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4331 | } |
4332 | ||
4333 | static int mem_cgroup_usage_register_event(struct cgroup *cgrp, | |
4334 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
2e72b634 KS |
4335 | { |
4336 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4337 | struct mem_cgroup_thresholds *thresholds; |
4338 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
4339 | int type = MEMFILE_TYPE(cft->private); |
4340 | u64 threshold, usage; | |
2c488db2 | 4341 | int i, size, ret; |
2e72b634 KS |
4342 | |
4343 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
4344 | if (ret) | |
4345 | return ret; | |
4346 | ||
4347 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4348 | |
2e72b634 | 4349 | if (type == _MEM) |
2c488db2 | 4350 | thresholds = &memcg->thresholds; |
2e72b634 | 4351 | else if (type == _MEMSWAP) |
2c488db2 | 4352 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4353 | else |
4354 | BUG(); | |
4355 | ||
4356 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
4357 | ||
4358 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 4359 | if (thresholds->primary) |
2e72b634 KS |
4360 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4361 | ||
2c488db2 | 4362 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4363 | |
4364 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 4365 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 4366 | GFP_KERNEL); |
2c488db2 | 4367 | if (!new) { |
2e72b634 KS |
4368 | ret = -ENOMEM; |
4369 | goto unlock; | |
4370 | } | |
2c488db2 | 4371 | new->size = size; |
2e72b634 KS |
4372 | |
4373 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4374 | if (thresholds->primary) { |
4375 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4376 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4377 | } |
4378 | ||
2e72b634 | 4379 | /* Add new threshold */ |
2c488db2 KS |
4380 | new->entries[size - 1].eventfd = eventfd; |
4381 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4382 | |
4383 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4384 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4385 | compare_thresholds, NULL); |
4386 | ||
4387 | /* Find current threshold */ | |
2c488db2 | 4388 | new->current_threshold = -1; |
2e72b634 | 4389 | for (i = 0; i < size; i++) { |
2c488db2 | 4390 | if (new->entries[i].threshold < usage) { |
2e72b634 | 4391 | /* |
2c488db2 KS |
4392 | * new->current_threshold will not be used until |
4393 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4394 | * it here. |
4395 | */ | |
2c488db2 | 4396 | ++new->current_threshold; |
2e72b634 KS |
4397 | } |
4398 | } | |
4399 | ||
2c488db2 KS |
4400 | /* Free old spare buffer and save old primary buffer as spare */ |
4401 | kfree(thresholds->spare); | |
4402 | thresholds->spare = thresholds->primary; | |
4403 | ||
4404 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4405 | |
907860ed | 4406 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4407 | synchronize_rcu(); |
4408 | ||
2e72b634 KS |
4409 | unlock: |
4410 | mutex_unlock(&memcg->thresholds_lock); | |
4411 | ||
4412 | return ret; | |
4413 | } | |
4414 | ||
907860ed | 4415 | static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp, |
9490ff27 | 4416 | struct cftype *cft, struct eventfd_ctx *eventfd) |
2e72b634 KS |
4417 | { |
4418 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4419 | struct mem_cgroup_thresholds *thresholds; |
4420 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
4421 | int type = MEMFILE_TYPE(cft->private); |
4422 | u64 usage; | |
2c488db2 | 4423 | int i, j, size; |
2e72b634 KS |
4424 | |
4425 | mutex_lock(&memcg->thresholds_lock); | |
4426 | if (type == _MEM) | |
2c488db2 | 4427 | thresholds = &memcg->thresholds; |
2e72b634 | 4428 | else if (type == _MEMSWAP) |
2c488db2 | 4429 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4430 | else |
4431 | BUG(); | |
4432 | ||
4433 | /* | |
4434 | * Something went wrong if we trying to unregister a threshold | |
4435 | * if we don't have thresholds | |
4436 | */ | |
4437 | BUG_ON(!thresholds); | |
4438 | ||
4439 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
4440 | ||
4441 | /* Check if a threshold crossed before removing */ | |
4442 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4443 | ||
4444 | /* Calculate new number of threshold */ | |
2c488db2 KS |
4445 | size = 0; |
4446 | for (i = 0; i < thresholds->primary->size; i++) { | |
4447 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
4448 | size++; |
4449 | } | |
4450 | ||
2c488db2 | 4451 | new = thresholds->spare; |
907860ed | 4452 | |
2e72b634 KS |
4453 | /* Set thresholds array to NULL if we don't have thresholds */ |
4454 | if (!size) { | |
2c488db2 KS |
4455 | kfree(new); |
4456 | new = NULL; | |
907860ed | 4457 | goto swap_buffers; |
2e72b634 KS |
4458 | } |
4459 | ||
2c488db2 | 4460 | new->size = size; |
2e72b634 KS |
4461 | |
4462 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4463 | new->current_threshold = -1; |
4464 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4465 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4466 | continue; |
4467 | ||
2c488db2 KS |
4468 | new->entries[j] = thresholds->primary->entries[i]; |
4469 | if (new->entries[j].threshold < usage) { | |
2e72b634 | 4470 | /* |
2c488db2 | 4471 | * new->current_threshold will not be used |
2e72b634 KS |
4472 | * until rcu_assign_pointer(), so it's safe to increment |
4473 | * it here. | |
4474 | */ | |
2c488db2 | 4475 | ++new->current_threshold; |
2e72b634 KS |
4476 | } |
4477 | j++; | |
4478 | } | |
4479 | ||
907860ed | 4480 | swap_buffers: |
2c488db2 KS |
4481 | /* Swap primary and spare array */ |
4482 | thresholds->spare = thresholds->primary; | |
4483 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4484 | |
907860ed | 4485 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4486 | synchronize_rcu(); |
4487 | ||
2e72b634 | 4488 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4489 | } |
c1e862c1 | 4490 | |
9490ff27 KH |
4491 | static int mem_cgroup_oom_register_event(struct cgroup *cgrp, |
4492 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
4493 | { | |
4494 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4495 | struct mem_cgroup_eventfd_list *event; | |
4496 | int type = MEMFILE_TYPE(cft->private); | |
4497 | ||
4498 | BUG_ON(type != _OOM_TYPE); | |
4499 | event = kmalloc(sizeof(*event), GFP_KERNEL); | |
4500 | if (!event) | |
4501 | return -ENOMEM; | |
4502 | ||
1af8efe9 | 4503 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4504 | |
4505 | event->eventfd = eventfd; | |
4506 | list_add(&event->list, &memcg->oom_notify); | |
4507 | ||
4508 | /* already in OOM ? */ | |
79dfdacc | 4509 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 4510 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4511 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4512 | |
4513 | return 0; | |
4514 | } | |
4515 | ||
907860ed | 4516 | static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp, |
9490ff27 KH |
4517 | struct cftype *cft, struct eventfd_ctx *eventfd) |
4518 | { | |
c0ff4b85 | 4519 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
9490ff27 KH |
4520 | struct mem_cgroup_eventfd_list *ev, *tmp; |
4521 | int type = MEMFILE_TYPE(cft->private); | |
4522 | ||
4523 | BUG_ON(type != _OOM_TYPE); | |
4524 | ||
1af8efe9 | 4525 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4526 | |
c0ff4b85 | 4527 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4528 | if (ev->eventfd == eventfd) { |
4529 | list_del(&ev->list); | |
4530 | kfree(ev); | |
4531 | } | |
4532 | } | |
4533 | ||
1af8efe9 | 4534 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4535 | } |
4536 | ||
3c11ecf4 KH |
4537 | static int mem_cgroup_oom_control_read(struct cgroup *cgrp, |
4538 | struct cftype *cft, struct cgroup_map_cb *cb) | |
4539 | { | |
c0ff4b85 | 4540 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 | 4541 | |
c0ff4b85 | 4542 | cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable); |
3c11ecf4 | 4543 | |
c0ff4b85 | 4544 | if (atomic_read(&memcg->under_oom)) |
3c11ecf4 KH |
4545 | cb->fill(cb, "under_oom", 1); |
4546 | else | |
4547 | cb->fill(cb, "under_oom", 0); | |
4548 | return 0; | |
4549 | } | |
4550 | ||
3c11ecf4 KH |
4551 | static int mem_cgroup_oom_control_write(struct cgroup *cgrp, |
4552 | struct cftype *cft, u64 val) | |
4553 | { | |
c0ff4b85 | 4554 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 KH |
4555 | struct mem_cgroup *parent; |
4556 | ||
4557 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
4558 | if (!cgrp->parent || !((val == 0) || (val == 1))) | |
4559 | return -EINVAL; | |
4560 | ||
4561 | parent = mem_cgroup_from_cont(cgrp->parent); | |
4562 | ||
4563 | cgroup_lock(); | |
4564 | /* oom-kill-disable is a flag for subhierarchy. */ | |
4565 | if ((parent->use_hierarchy) || | |
c0ff4b85 | 4566 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
3c11ecf4 KH |
4567 | cgroup_unlock(); |
4568 | return -EINVAL; | |
4569 | } | |
c0ff4b85 | 4570 | memcg->oom_kill_disable = val; |
4d845ebf | 4571 | if (!val) |
c0ff4b85 | 4572 | memcg_oom_recover(memcg); |
3c11ecf4 KH |
4573 | cgroup_unlock(); |
4574 | return 0; | |
4575 | } | |
4576 | ||
406eb0c9 YH |
4577 | #ifdef CONFIG_NUMA |
4578 | static const struct file_operations mem_control_numa_stat_file_operations = { | |
4579 | .read = seq_read, | |
4580 | .llseek = seq_lseek, | |
4581 | .release = single_release, | |
4582 | }; | |
4583 | ||
4584 | static int mem_control_numa_stat_open(struct inode *unused, struct file *file) | |
4585 | { | |
4586 | struct cgroup *cont = file->f_dentry->d_parent->d_fsdata; | |
4587 | ||
4588 | file->f_op = &mem_control_numa_stat_file_operations; | |
4589 | return single_open(file, mem_control_numa_stat_show, cont); | |
4590 | } | |
4591 | #endif /* CONFIG_NUMA */ | |
4592 | ||
e5671dfa | 4593 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM |
e5671dfa GC |
4594 | static int register_kmem_files(struct cgroup *cont, struct cgroup_subsys *ss) |
4595 | { | |
d1a4c0b3 GC |
4596 | /* |
4597 | * Part of this would be better living in a separate allocation | |
4598 | * function, leaving us with just the cgroup tree population work. | |
4599 | * We, however, depend on state such as network's proto_list that | |
4600 | * is only initialized after cgroup creation. I found the less | |
4601 | * cumbersome way to deal with it to defer it all to populate time | |
4602 | */ | |
65c64ce8 | 4603 | return mem_cgroup_sockets_init(cont, ss); |
e5671dfa GC |
4604 | }; |
4605 | ||
d1a4c0b3 GC |
4606 | static void kmem_cgroup_destroy(struct cgroup_subsys *ss, |
4607 | struct cgroup *cont) | |
4608 | { | |
4609 | mem_cgroup_sockets_destroy(cont, ss); | |
4610 | } | |
e5671dfa GC |
4611 | #else |
4612 | static int register_kmem_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
4613 | { | |
4614 | return 0; | |
4615 | } | |
d1a4c0b3 GC |
4616 | |
4617 | static void kmem_cgroup_destroy(struct cgroup_subsys *ss, | |
4618 | struct cgroup *cont) | |
4619 | { | |
4620 | } | |
e5671dfa GC |
4621 | #endif |
4622 | ||
8cdea7c0 BS |
4623 | static struct cftype mem_cgroup_files[] = { |
4624 | { | |
0eea1030 | 4625 | .name = "usage_in_bytes", |
8c7c6e34 | 4626 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
2c3daa72 | 4627 | .read_u64 = mem_cgroup_read, |
9490ff27 KH |
4628 | .register_event = mem_cgroup_usage_register_event, |
4629 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8cdea7c0 | 4630 | }, |
c84872e1 PE |
4631 | { |
4632 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4633 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 4634 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
4635 | .read_u64 = mem_cgroup_read, |
4636 | }, | |
8cdea7c0 | 4637 | { |
0eea1030 | 4638 | .name = "limit_in_bytes", |
8c7c6e34 | 4639 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 4640 | .write_string = mem_cgroup_write, |
2c3daa72 | 4641 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 4642 | }, |
296c81d8 BS |
4643 | { |
4644 | .name = "soft_limit_in_bytes", | |
4645 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
4646 | .write_string = mem_cgroup_write, | |
4647 | .read_u64 = mem_cgroup_read, | |
4648 | }, | |
8cdea7c0 BS |
4649 | { |
4650 | .name = "failcnt", | |
8c7c6e34 | 4651 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 4652 | .trigger = mem_cgroup_reset, |
2c3daa72 | 4653 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 4654 | }, |
d2ceb9b7 KH |
4655 | { |
4656 | .name = "stat", | |
c64745cf | 4657 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 4658 | }, |
c1e862c1 KH |
4659 | { |
4660 | .name = "force_empty", | |
4661 | .trigger = mem_cgroup_force_empty_write, | |
4662 | }, | |
18f59ea7 BS |
4663 | { |
4664 | .name = "use_hierarchy", | |
4665 | .write_u64 = mem_cgroup_hierarchy_write, | |
4666 | .read_u64 = mem_cgroup_hierarchy_read, | |
4667 | }, | |
a7885eb8 KM |
4668 | { |
4669 | .name = "swappiness", | |
4670 | .read_u64 = mem_cgroup_swappiness_read, | |
4671 | .write_u64 = mem_cgroup_swappiness_write, | |
4672 | }, | |
7dc74be0 DN |
4673 | { |
4674 | .name = "move_charge_at_immigrate", | |
4675 | .read_u64 = mem_cgroup_move_charge_read, | |
4676 | .write_u64 = mem_cgroup_move_charge_write, | |
4677 | }, | |
9490ff27 KH |
4678 | { |
4679 | .name = "oom_control", | |
3c11ecf4 KH |
4680 | .read_map = mem_cgroup_oom_control_read, |
4681 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
4682 | .register_event = mem_cgroup_oom_register_event, |
4683 | .unregister_event = mem_cgroup_oom_unregister_event, | |
4684 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), | |
4685 | }, | |
406eb0c9 YH |
4686 | #ifdef CONFIG_NUMA |
4687 | { | |
4688 | .name = "numa_stat", | |
4689 | .open = mem_control_numa_stat_open, | |
89577127 | 4690 | .mode = S_IRUGO, |
406eb0c9 YH |
4691 | }, |
4692 | #endif | |
8cdea7c0 BS |
4693 | }; |
4694 | ||
8c7c6e34 KH |
4695 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
4696 | static struct cftype memsw_cgroup_files[] = { | |
4697 | { | |
4698 | .name = "memsw.usage_in_bytes", | |
4699 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
4700 | .read_u64 = mem_cgroup_read, | |
9490ff27 KH |
4701 | .register_event = mem_cgroup_usage_register_event, |
4702 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8c7c6e34 KH |
4703 | }, |
4704 | { | |
4705 | .name = "memsw.max_usage_in_bytes", | |
4706 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
4707 | .trigger = mem_cgroup_reset, | |
4708 | .read_u64 = mem_cgroup_read, | |
4709 | }, | |
4710 | { | |
4711 | .name = "memsw.limit_in_bytes", | |
4712 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
4713 | .write_string = mem_cgroup_write, | |
4714 | .read_u64 = mem_cgroup_read, | |
4715 | }, | |
4716 | { | |
4717 | .name = "memsw.failcnt", | |
4718 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
4719 | .trigger = mem_cgroup_reset, | |
4720 | .read_u64 = mem_cgroup_read, | |
4721 | }, | |
4722 | }; | |
4723 | ||
4724 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
4725 | { | |
4726 | if (!do_swap_account) | |
4727 | return 0; | |
4728 | return cgroup_add_files(cont, ss, memsw_cgroup_files, | |
4729 | ARRAY_SIZE(memsw_cgroup_files)); | |
4730 | }; | |
4731 | #else | |
4732 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
4733 | { | |
4734 | return 0; | |
4735 | } | |
4736 | #endif | |
4737 | ||
c0ff4b85 | 4738 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4739 | { |
4740 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4741 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 4742 | enum lru_list l; |
41e3355d | 4743 | int zone, tmp = node; |
1ecaab2b KH |
4744 | /* |
4745 | * This routine is called against possible nodes. | |
4746 | * But it's BUG to call kmalloc() against offline node. | |
4747 | * | |
4748 | * TODO: this routine can waste much memory for nodes which will | |
4749 | * never be onlined. It's better to use memory hotplug callback | |
4750 | * function. | |
4751 | */ | |
41e3355d KH |
4752 | if (!node_state(node, N_NORMAL_MEMORY)) |
4753 | tmp = -1; | |
17295c88 | 4754 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4755 | if (!pn) |
4756 | return 1; | |
1ecaab2b | 4757 | |
1ecaab2b KH |
4758 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4759 | mz = &pn->zoneinfo[zone]; | |
b69408e8 | 4760 | for_each_lru(l) |
6290df54 | 4761 | INIT_LIST_HEAD(&mz->lruvec.lists[l]); |
f64c3f54 | 4762 | mz->usage_in_excess = 0; |
4e416953 | 4763 | mz->on_tree = false; |
c0ff4b85 | 4764 | mz->mem = memcg; |
1ecaab2b | 4765 | } |
0a619e58 | 4766 | memcg->info.nodeinfo[node] = pn; |
6d12e2d8 KH |
4767 | return 0; |
4768 | } | |
4769 | ||
c0ff4b85 | 4770 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4771 | { |
c0ff4b85 | 4772 | kfree(memcg->info.nodeinfo[node]); |
1ecaab2b KH |
4773 | } |
4774 | ||
33327948 KH |
4775 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4776 | { | |
4777 | struct mem_cgroup *mem; | |
c62b1a3b | 4778 | int size = sizeof(struct mem_cgroup); |
33327948 | 4779 | |
c62b1a3b | 4780 | /* Can be very big if MAX_NUMNODES is very big */ |
c8dad2bb | 4781 | if (size < PAGE_SIZE) |
17295c88 | 4782 | mem = kzalloc(size, GFP_KERNEL); |
33327948 | 4783 | else |
17295c88 | 4784 | mem = vzalloc(size); |
33327948 | 4785 | |
e7bbcdf3 DC |
4786 | if (!mem) |
4787 | return NULL; | |
4788 | ||
c62b1a3b | 4789 | mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
d2e61b8d DC |
4790 | if (!mem->stat) |
4791 | goto out_free; | |
711d3d2c | 4792 | spin_lock_init(&mem->pcp_counter_lock); |
33327948 | 4793 | return mem; |
d2e61b8d DC |
4794 | |
4795 | out_free: | |
4796 | if (size < PAGE_SIZE) | |
4797 | kfree(mem); | |
4798 | else | |
4799 | vfree(mem); | |
4800 | return NULL; | |
33327948 KH |
4801 | } |
4802 | ||
8c7c6e34 KH |
4803 | /* |
4804 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
4805 | * (scanning all at force_empty is too costly...) | |
4806 | * | |
4807 | * Instead of clearing all references at force_empty, we remember | |
4808 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4809 | * it goes down to 0. | |
4810 | * | |
8c7c6e34 KH |
4811 | * Removal of cgroup itself succeeds regardless of refs from swap. |
4812 | */ | |
4813 | ||
c0ff4b85 | 4814 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
33327948 | 4815 | { |
08e552c6 KH |
4816 | int node; |
4817 | ||
c0ff4b85 R |
4818 | mem_cgroup_remove_from_trees(memcg); |
4819 | free_css_id(&mem_cgroup_subsys, &memcg->css); | |
04046e1a | 4820 | |
08e552c6 | 4821 | for_each_node_state(node, N_POSSIBLE) |
c0ff4b85 | 4822 | free_mem_cgroup_per_zone_info(memcg, node); |
08e552c6 | 4823 | |
c0ff4b85 | 4824 | free_percpu(memcg->stat); |
c62b1a3b | 4825 | if (sizeof(struct mem_cgroup) < PAGE_SIZE) |
c0ff4b85 | 4826 | kfree(memcg); |
33327948 | 4827 | else |
c0ff4b85 | 4828 | vfree(memcg); |
33327948 KH |
4829 | } |
4830 | ||
c0ff4b85 | 4831 | static void mem_cgroup_get(struct mem_cgroup *memcg) |
8c7c6e34 | 4832 | { |
c0ff4b85 | 4833 | atomic_inc(&memcg->refcnt); |
8c7c6e34 KH |
4834 | } |
4835 | ||
c0ff4b85 | 4836 | static void __mem_cgroup_put(struct mem_cgroup *memcg, int count) |
8c7c6e34 | 4837 | { |
c0ff4b85 R |
4838 | if (atomic_sub_and_test(count, &memcg->refcnt)) { |
4839 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); | |
4840 | __mem_cgroup_free(memcg); | |
7bcc1bb1 DN |
4841 | if (parent) |
4842 | mem_cgroup_put(parent); | |
4843 | } | |
8c7c6e34 KH |
4844 | } |
4845 | ||
c0ff4b85 | 4846 | static void mem_cgroup_put(struct mem_cgroup *memcg) |
483c30b5 | 4847 | { |
c0ff4b85 | 4848 | __mem_cgroup_put(memcg, 1); |
483c30b5 DN |
4849 | } |
4850 | ||
7bcc1bb1 DN |
4851 | /* |
4852 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4853 | */ | |
e1aab161 | 4854 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 4855 | { |
c0ff4b85 | 4856 | if (!memcg->res.parent) |
7bcc1bb1 | 4857 | return NULL; |
c0ff4b85 | 4858 | return mem_cgroup_from_res_counter(memcg->res.parent, res); |
7bcc1bb1 | 4859 | } |
e1aab161 | 4860 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 4861 | |
c077719b KH |
4862 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
4863 | static void __init enable_swap_cgroup(void) | |
4864 | { | |
f8d66542 | 4865 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
4866 | do_swap_account = 1; |
4867 | } | |
4868 | #else | |
4869 | static void __init enable_swap_cgroup(void) | |
4870 | { | |
4871 | } | |
4872 | #endif | |
4873 | ||
f64c3f54 BS |
4874 | static int mem_cgroup_soft_limit_tree_init(void) |
4875 | { | |
4876 | struct mem_cgroup_tree_per_node *rtpn; | |
4877 | struct mem_cgroup_tree_per_zone *rtpz; | |
4878 | int tmp, node, zone; | |
4879 | ||
4880 | for_each_node_state(node, N_POSSIBLE) { | |
4881 | tmp = node; | |
4882 | if (!node_state(node, N_NORMAL_MEMORY)) | |
4883 | tmp = -1; | |
4884 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
4885 | if (!rtpn) | |
c3cecc68 | 4886 | goto err_cleanup; |
f64c3f54 BS |
4887 | |
4888 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
4889 | ||
4890 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
4891 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
4892 | rtpz->rb_root = RB_ROOT; | |
4893 | spin_lock_init(&rtpz->lock); | |
4894 | } | |
4895 | } | |
4896 | return 0; | |
c3cecc68 MH |
4897 | |
4898 | err_cleanup: | |
4899 | for_each_node_state(node, N_POSSIBLE) { | |
4900 | if (!soft_limit_tree.rb_tree_per_node[node]) | |
4901 | break; | |
4902 | kfree(soft_limit_tree.rb_tree_per_node[node]); | |
4903 | soft_limit_tree.rb_tree_per_node[node] = NULL; | |
4904 | } | |
4905 | return 1; | |
4906 | ||
f64c3f54 BS |
4907 | } |
4908 | ||
0eb253e2 | 4909 | static struct cgroup_subsys_state * __ref |
8cdea7c0 BS |
4910 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) |
4911 | { | |
c0ff4b85 | 4912 | struct mem_cgroup *memcg, *parent; |
04046e1a | 4913 | long error = -ENOMEM; |
6d12e2d8 | 4914 | int node; |
8cdea7c0 | 4915 | |
c0ff4b85 R |
4916 | memcg = mem_cgroup_alloc(); |
4917 | if (!memcg) | |
04046e1a | 4918 | return ERR_PTR(error); |
78fb7466 | 4919 | |
6d12e2d8 | 4920 | for_each_node_state(node, N_POSSIBLE) |
c0ff4b85 | 4921 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 4922 | goto free_out; |
f64c3f54 | 4923 | |
c077719b | 4924 | /* root ? */ |
28dbc4b6 | 4925 | if (cont->parent == NULL) { |
cdec2e42 | 4926 | int cpu; |
c077719b | 4927 | enable_swap_cgroup(); |
28dbc4b6 | 4928 | parent = NULL; |
f64c3f54 BS |
4929 | if (mem_cgroup_soft_limit_tree_init()) |
4930 | goto free_out; | |
a41c58a6 | 4931 | root_mem_cgroup = memcg; |
cdec2e42 KH |
4932 | for_each_possible_cpu(cpu) { |
4933 | struct memcg_stock_pcp *stock = | |
4934 | &per_cpu(memcg_stock, cpu); | |
4935 | INIT_WORK(&stock->work, drain_local_stock); | |
4936 | } | |
711d3d2c | 4937 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
18f59ea7 | 4938 | } else { |
28dbc4b6 | 4939 | parent = mem_cgroup_from_cont(cont->parent); |
c0ff4b85 R |
4940 | memcg->use_hierarchy = parent->use_hierarchy; |
4941 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
18f59ea7 | 4942 | } |
28dbc4b6 | 4943 | |
18f59ea7 | 4944 | if (parent && parent->use_hierarchy) { |
c0ff4b85 R |
4945 | res_counter_init(&memcg->res, &parent->res); |
4946 | res_counter_init(&memcg->memsw, &parent->memsw); | |
7bcc1bb1 DN |
4947 | /* |
4948 | * We increment refcnt of the parent to ensure that we can | |
4949 | * safely access it on res_counter_charge/uncharge. | |
4950 | * This refcnt will be decremented when freeing this | |
4951 | * mem_cgroup(see mem_cgroup_put). | |
4952 | */ | |
4953 | mem_cgroup_get(parent); | |
18f59ea7 | 4954 | } else { |
c0ff4b85 R |
4955 | res_counter_init(&memcg->res, NULL); |
4956 | res_counter_init(&memcg->memsw, NULL); | |
18f59ea7 | 4957 | } |
c0ff4b85 R |
4958 | memcg->last_scanned_node = MAX_NUMNODES; |
4959 | INIT_LIST_HEAD(&memcg->oom_notify); | |
6d61ef40 | 4960 | |
a7885eb8 | 4961 | if (parent) |
c0ff4b85 R |
4962 | memcg->swappiness = mem_cgroup_swappiness(parent); |
4963 | atomic_set(&memcg->refcnt, 1); | |
4964 | memcg->move_charge_at_immigrate = 0; | |
4965 | mutex_init(&memcg->thresholds_lock); | |
4966 | return &memcg->css; | |
6d12e2d8 | 4967 | free_out: |
c0ff4b85 | 4968 | __mem_cgroup_free(memcg); |
04046e1a | 4969 | return ERR_PTR(error); |
8cdea7c0 BS |
4970 | } |
4971 | ||
ec64f515 | 4972 | static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
df878fb0 KH |
4973 | struct cgroup *cont) |
4974 | { | |
c0ff4b85 | 4975 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
ec64f515 | 4976 | |
c0ff4b85 | 4977 | return mem_cgroup_force_empty(memcg, false); |
df878fb0 KH |
4978 | } |
4979 | ||
8cdea7c0 BS |
4980 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
4981 | struct cgroup *cont) | |
4982 | { | |
c0ff4b85 | 4983 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
c268e994 | 4984 | |
d1a4c0b3 GC |
4985 | kmem_cgroup_destroy(ss, cont); |
4986 | ||
c0ff4b85 | 4987 | mem_cgroup_put(memcg); |
8cdea7c0 BS |
4988 | } |
4989 | ||
4990 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
4991 | struct cgroup *cont) | |
4992 | { | |
8c7c6e34 KH |
4993 | int ret; |
4994 | ||
4995 | ret = cgroup_add_files(cont, ss, mem_cgroup_files, | |
4996 | ARRAY_SIZE(mem_cgroup_files)); | |
4997 | ||
4998 | if (!ret) | |
4999 | ret = register_memsw_files(cont, ss); | |
e5671dfa GC |
5000 | |
5001 | if (!ret) | |
5002 | ret = register_kmem_files(cont, ss); | |
5003 | ||
8c7c6e34 | 5004 | return ret; |
8cdea7c0 BS |
5005 | } |
5006 | ||
02491447 | 5007 | #ifdef CONFIG_MMU |
7dc74be0 | 5008 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
5009 | #define PRECHARGE_COUNT_AT_ONCE 256 |
5010 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 5011 | { |
854ffa8d DN |
5012 | int ret = 0; |
5013 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
c0ff4b85 | 5014 | struct mem_cgroup *memcg = mc.to; |
4ffef5fe | 5015 | |
c0ff4b85 | 5016 | if (mem_cgroup_is_root(memcg)) { |
854ffa8d DN |
5017 | mc.precharge += count; |
5018 | /* we don't need css_get for root */ | |
5019 | return ret; | |
5020 | } | |
5021 | /* try to charge at once */ | |
5022 | if (count > 1) { | |
5023 | struct res_counter *dummy; | |
5024 | /* | |
c0ff4b85 | 5025 | * "memcg" cannot be under rmdir() because we've already checked |
854ffa8d DN |
5026 | * by cgroup_lock_live_cgroup() that it is not removed and we |
5027 | * are still under the same cgroup_mutex. So we can postpone | |
5028 | * css_get(). | |
5029 | */ | |
c0ff4b85 | 5030 | if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy)) |
854ffa8d | 5031 | goto one_by_one; |
c0ff4b85 | 5032 | if (do_swap_account && res_counter_charge(&memcg->memsw, |
854ffa8d | 5033 | PAGE_SIZE * count, &dummy)) { |
c0ff4b85 | 5034 | res_counter_uncharge(&memcg->res, PAGE_SIZE * count); |
854ffa8d DN |
5035 | goto one_by_one; |
5036 | } | |
5037 | mc.precharge += count; | |
854ffa8d DN |
5038 | return ret; |
5039 | } | |
5040 | one_by_one: | |
5041 | /* fall back to one by one charge */ | |
5042 | while (count--) { | |
5043 | if (signal_pending(current)) { | |
5044 | ret = -EINTR; | |
5045 | break; | |
5046 | } | |
5047 | if (!batch_count--) { | |
5048 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
5049 | cond_resched(); | |
5050 | } | |
c0ff4b85 R |
5051 | ret = __mem_cgroup_try_charge(NULL, |
5052 | GFP_KERNEL, 1, &memcg, false); | |
38c5d72f | 5053 | if (ret) |
854ffa8d | 5054 | /* mem_cgroup_clear_mc() will do uncharge later */ |
38c5d72f | 5055 | return ret; |
854ffa8d DN |
5056 | mc.precharge++; |
5057 | } | |
4ffef5fe DN |
5058 | return ret; |
5059 | } | |
5060 | ||
5061 | /** | |
5062 | * is_target_pte_for_mc - check a pte whether it is valid for move charge | |
5063 | * @vma: the vma the pte to be checked belongs | |
5064 | * @addr: the address corresponding to the pte to be checked | |
5065 | * @ptent: the pte to be checked | |
02491447 | 5066 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
5067 | * |
5068 | * Returns | |
5069 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5070 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5071 | * move charge. if @target is not NULL, the page is stored in target->page | |
5072 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
5073 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
5074 | * target for charge migration. if @target is not NULL, the entry is stored | |
5075 | * in target->ent. | |
4ffef5fe DN |
5076 | * |
5077 | * Called with pte lock held. | |
5078 | */ | |
4ffef5fe DN |
5079 | union mc_target { |
5080 | struct page *page; | |
02491447 | 5081 | swp_entry_t ent; |
4ffef5fe DN |
5082 | }; |
5083 | ||
4ffef5fe DN |
5084 | enum mc_target_type { |
5085 | MC_TARGET_NONE, /* not used */ | |
5086 | MC_TARGET_PAGE, | |
02491447 | 5087 | MC_TARGET_SWAP, |
4ffef5fe DN |
5088 | }; |
5089 | ||
90254a65 DN |
5090 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5091 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5092 | { |
90254a65 | 5093 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5094 | |
90254a65 DN |
5095 | if (!page || !page_mapped(page)) |
5096 | return NULL; | |
5097 | if (PageAnon(page)) { | |
5098 | /* we don't move shared anon */ | |
5099 | if (!move_anon() || page_mapcount(page) > 2) | |
5100 | return NULL; | |
87946a72 DN |
5101 | } else if (!move_file()) |
5102 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
5103 | return NULL; |
5104 | if (!get_page_unless_zero(page)) | |
5105 | return NULL; | |
5106 | ||
5107 | return page; | |
5108 | } | |
5109 | ||
5110 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
5111 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5112 | { | |
5113 | int usage_count; | |
5114 | struct page *page = NULL; | |
5115 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5116 | ||
5117 | if (!move_anon() || non_swap_entry(ent)) | |
5118 | return NULL; | |
5119 | usage_count = mem_cgroup_count_swap_user(ent, &page); | |
5120 | if (usage_count > 1) { /* we don't move shared anon */ | |
02491447 DN |
5121 | if (page) |
5122 | put_page(page); | |
90254a65 | 5123 | return NULL; |
02491447 | 5124 | } |
90254a65 DN |
5125 | if (do_swap_account) |
5126 | entry->val = ent.val; | |
5127 | ||
5128 | return page; | |
5129 | } | |
5130 | ||
87946a72 DN |
5131 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5132 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5133 | { | |
5134 | struct page *page = NULL; | |
5135 | struct inode *inode; | |
5136 | struct address_space *mapping; | |
5137 | pgoff_t pgoff; | |
5138 | ||
5139 | if (!vma->vm_file) /* anonymous vma */ | |
5140 | return NULL; | |
5141 | if (!move_file()) | |
5142 | return NULL; | |
5143 | ||
5144 | inode = vma->vm_file->f_path.dentry->d_inode; | |
5145 | mapping = vma->vm_file->f_mapping; | |
5146 | if (pte_none(ptent)) | |
5147 | pgoff = linear_page_index(vma, addr); | |
5148 | else /* pte_file(ptent) is true */ | |
5149 | pgoff = pte_to_pgoff(ptent); | |
5150 | ||
5151 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
5152 | page = find_get_page(mapping, pgoff); |
5153 | ||
5154 | #ifdef CONFIG_SWAP | |
5155 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
5156 | if (radix_tree_exceptional_entry(page)) { | |
5157 | swp_entry_t swap = radix_to_swp_entry(page); | |
87946a72 | 5158 | if (do_swap_account) |
aa3b1895 HD |
5159 | *entry = swap; |
5160 | page = find_get_page(&swapper_space, swap.val); | |
87946a72 | 5161 | } |
aa3b1895 | 5162 | #endif |
87946a72 DN |
5163 | return page; |
5164 | } | |
5165 | ||
90254a65 DN |
5166 | static int is_target_pte_for_mc(struct vm_area_struct *vma, |
5167 | unsigned long addr, pte_t ptent, union mc_target *target) | |
5168 | { | |
5169 | struct page *page = NULL; | |
5170 | struct page_cgroup *pc; | |
5171 | int ret = 0; | |
5172 | swp_entry_t ent = { .val = 0 }; | |
5173 | ||
5174 | if (pte_present(ptent)) | |
5175 | page = mc_handle_present_pte(vma, addr, ptent); | |
5176 | else if (is_swap_pte(ptent)) | |
5177 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
5178 | else if (pte_none(ptent) || pte_file(ptent)) |
5179 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
5180 | |
5181 | if (!page && !ent.val) | |
5182 | return 0; | |
02491447 DN |
5183 | if (page) { |
5184 | pc = lookup_page_cgroup(page); | |
5185 | /* | |
5186 | * Do only loose check w/o page_cgroup lock. | |
5187 | * mem_cgroup_move_account() checks the pc is valid or not under | |
5188 | * the lock. | |
5189 | */ | |
5190 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5191 | ret = MC_TARGET_PAGE; | |
5192 | if (target) | |
5193 | target->page = page; | |
5194 | } | |
5195 | if (!ret || !target) | |
5196 | put_page(page); | |
5197 | } | |
90254a65 DN |
5198 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
5199 | if (ent.val && !ret && | |
9fb4b7cc | 5200 | css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5201 | ret = MC_TARGET_SWAP; |
5202 | if (target) | |
5203 | target->ent = ent; | |
4ffef5fe | 5204 | } |
4ffef5fe DN |
5205 | return ret; |
5206 | } | |
5207 | ||
5208 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, | |
5209 | unsigned long addr, unsigned long end, | |
5210 | struct mm_walk *walk) | |
5211 | { | |
5212 | struct vm_area_struct *vma = walk->private; | |
5213 | pte_t *pte; | |
5214 | spinlock_t *ptl; | |
5215 | ||
03319327 DH |
5216 | split_huge_page_pmd(walk->mm, pmd); |
5217 | ||
4ffef5fe DN |
5218 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5219 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
5220 | if (is_target_pte_for_mc(vma, addr, *pte, NULL)) | |
5221 | mc.precharge++; /* increment precharge temporarily */ | |
5222 | pte_unmap_unlock(pte - 1, ptl); | |
5223 | cond_resched(); | |
5224 | ||
7dc74be0 DN |
5225 | return 0; |
5226 | } | |
5227 | ||
4ffef5fe DN |
5228 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5229 | { | |
5230 | unsigned long precharge; | |
5231 | struct vm_area_struct *vma; | |
5232 | ||
dfe076b0 | 5233 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
5234 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5235 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
5236 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5237 | .mm = mm, | |
5238 | .private = vma, | |
5239 | }; | |
5240 | if (is_vm_hugetlb_page(vma)) | |
5241 | continue; | |
4ffef5fe DN |
5242 | walk_page_range(vma->vm_start, vma->vm_end, |
5243 | &mem_cgroup_count_precharge_walk); | |
5244 | } | |
dfe076b0 | 5245 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5246 | |
5247 | precharge = mc.precharge; | |
5248 | mc.precharge = 0; | |
5249 | ||
5250 | return precharge; | |
5251 | } | |
5252 | ||
4ffef5fe DN |
5253 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5254 | { | |
dfe076b0 DN |
5255 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5256 | ||
5257 | VM_BUG_ON(mc.moving_task); | |
5258 | mc.moving_task = current; | |
5259 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5260 | } |
5261 | ||
dfe076b0 DN |
5262 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5263 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5264 | { |
2bd9bb20 KH |
5265 | struct mem_cgroup *from = mc.from; |
5266 | struct mem_cgroup *to = mc.to; | |
5267 | ||
4ffef5fe | 5268 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
5269 | if (mc.precharge) { |
5270 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
5271 | mc.precharge = 0; | |
5272 | } | |
5273 | /* | |
5274 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5275 | * we must uncharge here. | |
5276 | */ | |
5277 | if (mc.moved_charge) { | |
5278 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
5279 | mc.moved_charge = 0; | |
4ffef5fe | 5280 | } |
483c30b5 DN |
5281 | /* we must fixup refcnts and charges */ |
5282 | if (mc.moved_swap) { | |
483c30b5 DN |
5283 | /* uncharge swap account from the old cgroup */ |
5284 | if (!mem_cgroup_is_root(mc.from)) | |
5285 | res_counter_uncharge(&mc.from->memsw, | |
5286 | PAGE_SIZE * mc.moved_swap); | |
5287 | __mem_cgroup_put(mc.from, mc.moved_swap); | |
5288 | ||
5289 | if (!mem_cgroup_is_root(mc.to)) { | |
5290 | /* | |
5291 | * we charged both to->res and to->memsw, so we should | |
5292 | * uncharge to->res. | |
5293 | */ | |
5294 | res_counter_uncharge(&mc.to->res, | |
5295 | PAGE_SIZE * mc.moved_swap); | |
483c30b5 DN |
5296 | } |
5297 | /* we've already done mem_cgroup_get(mc.to) */ | |
483c30b5 DN |
5298 | mc.moved_swap = 0; |
5299 | } | |
dfe076b0 DN |
5300 | memcg_oom_recover(from); |
5301 | memcg_oom_recover(to); | |
5302 | wake_up_all(&mc.waitq); | |
5303 | } | |
5304 | ||
5305 | static void mem_cgroup_clear_mc(void) | |
5306 | { | |
5307 | struct mem_cgroup *from = mc.from; | |
5308 | ||
5309 | /* | |
5310 | * we must clear moving_task before waking up waiters at the end of | |
5311 | * task migration. | |
5312 | */ | |
5313 | mc.moving_task = NULL; | |
5314 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5315 | spin_lock(&mc.lock); |
4ffef5fe DN |
5316 | mc.from = NULL; |
5317 | mc.to = NULL; | |
2bd9bb20 | 5318 | spin_unlock(&mc.lock); |
32047e2a | 5319 | mem_cgroup_end_move(from); |
4ffef5fe DN |
5320 | } |
5321 | ||
7dc74be0 DN |
5322 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, |
5323 | struct cgroup *cgroup, | |
2f7ee569 | 5324 | struct cgroup_taskset *tset) |
7dc74be0 | 5325 | { |
2f7ee569 | 5326 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 5327 | int ret = 0; |
c0ff4b85 | 5328 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup); |
7dc74be0 | 5329 | |
c0ff4b85 | 5330 | if (memcg->move_charge_at_immigrate) { |
7dc74be0 DN |
5331 | struct mm_struct *mm; |
5332 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
5333 | ||
c0ff4b85 | 5334 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
5335 | |
5336 | mm = get_task_mm(p); | |
5337 | if (!mm) | |
5338 | return 0; | |
7dc74be0 | 5339 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
5340 | if (mm->owner == p) { |
5341 | VM_BUG_ON(mc.from); | |
5342 | VM_BUG_ON(mc.to); | |
5343 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 5344 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 5345 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 5346 | mem_cgroup_start_move(from); |
2bd9bb20 | 5347 | spin_lock(&mc.lock); |
4ffef5fe | 5348 | mc.from = from; |
c0ff4b85 | 5349 | mc.to = memcg; |
2bd9bb20 | 5350 | spin_unlock(&mc.lock); |
dfe076b0 | 5351 | /* We set mc.moving_task later */ |
4ffef5fe DN |
5352 | |
5353 | ret = mem_cgroup_precharge_mc(mm); | |
5354 | if (ret) | |
5355 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
5356 | } |
5357 | mmput(mm); | |
7dc74be0 DN |
5358 | } |
5359 | return ret; | |
5360 | } | |
5361 | ||
5362 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
5363 | struct cgroup *cgroup, | |
2f7ee569 | 5364 | struct cgroup_taskset *tset) |
7dc74be0 | 5365 | { |
4ffef5fe | 5366 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
5367 | } |
5368 | ||
4ffef5fe DN |
5369 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5370 | unsigned long addr, unsigned long end, | |
5371 | struct mm_walk *walk) | |
7dc74be0 | 5372 | { |
4ffef5fe DN |
5373 | int ret = 0; |
5374 | struct vm_area_struct *vma = walk->private; | |
5375 | pte_t *pte; | |
5376 | spinlock_t *ptl; | |
5377 | ||
03319327 | 5378 | split_huge_page_pmd(walk->mm, pmd); |
4ffef5fe DN |
5379 | retry: |
5380 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5381 | for (; addr != end; addr += PAGE_SIZE) { | |
5382 | pte_t ptent = *(pte++); | |
5383 | union mc_target target; | |
5384 | int type; | |
5385 | struct page *page; | |
5386 | struct page_cgroup *pc; | |
02491447 | 5387 | swp_entry_t ent; |
4ffef5fe DN |
5388 | |
5389 | if (!mc.precharge) | |
5390 | break; | |
5391 | ||
5392 | type = is_target_pte_for_mc(vma, addr, ptent, &target); | |
5393 | switch (type) { | |
5394 | case MC_TARGET_PAGE: | |
5395 | page = target.page; | |
5396 | if (isolate_lru_page(page)) | |
5397 | goto put; | |
5398 | pc = lookup_page_cgroup(page); | |
7ec99d62 JW |
5399 | if (!mem_cgroup_move_account(page, 1, pc, |
5400 | mc.from, mc.to, false)) { | |
4ffef5fe | 5401 | mc.precharge--; |
854ffa8d DN |
5402 | /* we uncharge from mc.from later. */ |
5403 | mc.moved_charge++; | |
4ffef5fe DN |
5404 | } |
5405 | putback_lru_page(page); | |
5406 | put: /* is_target_pte_for_mc() gets the page */ | |
5407 | put_page(page); | |
5408 | break; | |
02491447 DN |
5409 | case MC_TARGET_SWAP: |
5410 | ent = target.ent; | |
483c30b5 DN |
5411 | if (!mem_cgroup_move_swap_account(ent, |
5412 | mc.from, mc.to, false)) { | |
02491447 | 5413 | mc.precharge--; |
483c30b5 DN |
5414 | /* we fixup refcnts and charges later. */ |
5415 | mc.moved_swap++; | |
5416 | } | |
02491447 | 5417 | break; |
4ffef5fe DN |
5418 | default: |
5419 | break; | |
5420 | } | |
5421 | } | |
5422 | pte_unmap_unlock(pte - 1, ptl); | |
5423 | cond_resched(); | |
5424 | ||
5425 | if (addr != end) { | |
5426 | /* | |
5427 | * We have consumed all precharges we got in can_attach(). | |
5428 | * We try charge one by one, but don't do any additional | |
5429 | * charges to mc.to if we have failed in charge once in attach() | |
5430 | * phase. | |
5431 | */ | |
854ffa8d | 5432 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5433 | if (!ret) |
5434 | goto retry; | |
5435 | } | |
5436 | ||
5437 | return ret; | |
5438 | } | |
5439 | ||
5440 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
5441 | { | |
5442 | struct vm_area_struct *vma; | |
5443 | ||
5444 | lru_add_drain_all(); | |
dfe076b0 DN |
5445 | retry: |
5446 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5447 | /* | |
5448 | * Someone who are holding the mmap_sem might be waiting in | |
5449 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5450 | * and retry. Because we cancel precharges, we might not be able | |
5451 | * to move enough charges, but moving charge is a best-effort | |
5452 | * feature anyway, so it wouldn't be a big problem. | |
5453 | */ | |
5454 | __mem_cgroup_clear_mc(); | |
5455 | cond_resched(); | |
5456 | goto retry; | |
5457 | } | |
4ffef5fe DN |
5458 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5459 | int ret; | |
5460 | struct mm_walk mem_cgroup_move_charge_walk = { | |
5461 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
5462 | .mm = mm, | |
5463 | .private = vma, | |
5464 | }; | |
5465 | if (is_vm_hugetlb_page(vma)) | |
5466 | continue; | |
4ffef5fe DN |
5467 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
5468 | &mem_cgroup_move_charge_walk); | |
5469 | if (ret) | |
5470 | /* | |
5471 | * means we have consumed all precharges and failed in | |
5472 | * doing additional charge. Just abandon here. | |
5473 | */ | |
5474 | break; | |
5475 | } | |
dfe076b0 | 5476 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
5477 | } |
5478 | ||
67e465a7 BS |
5479 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
5480 | struct cgroup *cont, | |
2f7ee569 | 5481 | struct cgroup_taskset *tset) |
67e465a7 | 5482 | { |
2f7ee569 | 5483 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 5484 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 5485 | |
dfe076b0 | 5486 | if (mm) { |
a433658c KM |
5487 | if (mc.to) |
5488 | mem_cgroup_move_charge(mm); | |
5489 | put_swap_token(mm); | |
dfe076b0 DN |
5490 | mmput(mm); |
5491 | } | |
a433658c KM |
5492 | if (mc.to) |
5493 | mem_cgroup_clear_mc(); | |
67e465a7 | 5494 | } |
5cfb80a7 DN |
5495 | #else /* !CONFIG_MMU */ |
5496 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, | |
5497 | struct cgroup *cgroup, | |
2f7ee569 | 5498 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5499 | { |
5500 | return 0; | |
5501 | } | |
5502 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
5503 | struct cgroup *cgroup, | |
2f7ee569 | 5504 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5505 | { |
5506 | } | |
5507 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, | |
5508 | struct cgroup *cont, | |
2f7ee569 | 5509 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5510 | { |
5511 | } | |
5512 | #endif | |
67e465a7 | 5513 | |
8cdea7c0 BS |
5514 | struct cgroup_subsys mem_cgroup_subsys = { |
5515 | .name = "memory", | |
5516 | .subsys_id = mem_cgroup_subsys_id, | |
5517 | .create = mem_cgroup_create, | |
df878fb0 | 5518 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
5519 | .destroy = mem_cgroup_destroy, |
5520 | .populate = mem_cgroup_populate, | |
7dc74be0 DN |
5521 | .can_attach = mem_cgroup_can_attach, |
5522 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5523 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 5524 | .early_init = 0, |
04046e1a | 5525 | .use_id = 1, |
8cdea7c0 | 5526 | }; |
c077719b KH |
5527 | |
5528 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
a42c390c MH |
5529 | static int __init enable_swap_account(char *s) |
5530 | { | |
5531 | /* consider enabled if no parameter or 1 is given */ | |
a2c8990a | 5532 | if (!strcmp(s, "1")) |
a42c390c | 5533 | really_do_swap_account = 1; |
a2c8990a | 5534 | else if (!strcmp(s, "0")) |
a42c390c MH |
5535 | really_do_swap_account = 0; |
5536 | return 1; | |
5537 | } | |
a2c8990a | 5538 | __setup("swapaccount=", enable_swap_account); |
c077719b | 5539 | |
c077719b | 5540 | #endif |