]>
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 | ||
6d12e2d8 KH |
126 | /* |
127 | * per-zone information in memory controller. | |
128 | */ | |
6d12e2d8 | 129 | struct mem_cgroup_per_zone { |
072c56c1 KH |
130 | /* |
131 | * spin_lock to protect the per cgroup LRU | |
132 | */ | |
b69408e8 CL |
133 | struct list_head lists[NR_LRU_LISTS]; |
134 | unsigned long count[NR_LRU_LISTS]; | |
3e2f41f1 KM |
135 | |
136 | struct zone_reclaim_stat reclaim_stat; | |
f64c3f54 BS |
137 | struct rb_node tree_node; /* RB tree node */ |
138 | unsigned long long usage_in_excess;/* Set to the value by which */ | |
139 | /* the soft limit is exceeded*/ | |
140 | bool on_tree; | |
4e416953 BS |
141 | struct mem_cgroup *mem; /* Back pointer, we cannot */ |
142 | /* use container_of */ | |
6d12e2d8 KH |
143 | }; |
144 | /* Macro for accessing counter */ | |
145 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) | |
146 | ||
147 | struct mem_cgroup_per_node { | |
148 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
149 | }; | |
150 | ||
151 | struct mem_cgroup_lru_info { | |
152 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
153 | }; | |
154 | ||
f64c3f54 BS |
155 | /* |
156 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
157 | * their hierarchy representation | |
158 | */ | |
159 | ||
160 | struct mem_cgroup_tree_per_zone { | |
161 | struct rb_root rb_root; | |
162 | spinlock_t lock; | |
163 | }; | |
164 | ||
165 | struct mem_cgroup_tree_per_node { | |
166 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
167 | }; | |
168 | ||
169 | struct mem_cgroup_tree { | |
170 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
171 | }; | |
172 | ||
173 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
174 | ||
2e72b634 KS |
175 | struct mem_cgroup_threshold { |
176 | struct eventfd_ctx *eventfd; | |
177 | u64 threshold; | |
178 | }; | |
179 | ||
9490ff27 | 180 | /* For threshold */ |
2e72b634 KS |
181 | struct mem_cgroup_threshold_ary { |
182 | /* An array index points to threshold just below usage. */ | |
5407a562 | 183 | int current_threshold; |
2e72b634 KS |
184 | /* Size of entries[] */ |
185 | unsigned int size; | |
186 | /* Array of thresholds */ | |
187 | struct mem_cgroup_threshold entries[0]; | |
188 | }; | |
2c488db2 KS |
189 | |
190 | struct mem_cgroup_thresholds { | |
191 | /* Primary thresholds array */ | |
192 | struct mem_cgroup_threshold_ary *primary; | |
193 | /* | |
194 | * Spare threshold array. | |
195 | * This is needed to make mem_cgroup_unregister_event() "never fail". | |
196 | * It must be able to store at least primary->size - 1 entries. | |
197 | */ | |
198 | struct mem_cgroup_threshold_ary *spare; | |
199 | }; | |
200 | ||
9490ff27 KH |
201 | /* for OOM */ |
202 | struct mem_cgroup_eventfd_list { | |
203 | struct list_head list; | |
204 | struct eventfd_ctx *eventfd; | |
205 | }; | |
2e72b634 | 206 | |
c0ff4b85 R |
207 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
208 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 209 | |
8cdea7c0 BS |
210 | /* |
211 | * The memory controller data structure. The memory controller controls both | |
212 | * page cache and RSS per cgroup. We would eventually like to provide | |
213 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
214 | * to help the administrator determine what knobs to tune. | |
215 | * | |
216 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
217 | * we hit the water mark. May be even add a low water mark, such that |
218 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
219 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
220 | */ |
221 | struct mem_cgroup { | |
222 | struct cgroup_subsys_state css; | |
223 | /* | |
224 | * the counter to account for memory usage | |
225 | */ | |
226 | struct res_counter res; | |
8c7c6e34 KH |
227 | /* |
228 | * the counter to account for mem+swap usage. | |
229 | */ | |
230 | struct res_counter memsw; | |
78fb7466 PE |
231 | /* |
232 | * Per cgroup active and inactive list, similar to the | |
233 | * per zone LRU lists. | |
78fb7466 | 234 | */ |
6d12e2d8 | 235 | struct mem_cgroup_lru_info info; |
6d61ef40 | 236 | /* |
af901ca1 | 237 | * While reclaiming in a hierarchy, we cache the last child we |
04046e1a | 238 | * reclaimed from. |
6d61ef40 | 239 | */ |
04046e1a | 240 | int last_scanned_child; |
889976db YH |
241 | int last_scanned_node; |
242 | #if MAX_NUMNODES > 1 | |
243 | nodemask_t scan_nodes; | |
453a9bf3 KH |
244 | atomic_t numainfo_events; |
245 | atomic_t numainfo_updating; | |
889976db | 246 | #endif |
18f59ea7 BS |
247 | /* |
248 | * Should the accounting and control be hierarchical, per subtree? | |
249 | */ | |
250 | bool use_hierarchy; | |
79dfdacc MH |
251 | |
252 | bool oom_lock; | |
253 | atomic_t under_oom; | |
254 | ||
8c7c6e34 | 255 | atomic_t refcnt; |
14797e23 | 256 | |
1f4c025b | 257 | int swappiness; |
3c11ecf4 KH |
258 | /* OOM-Killer disable */ |
259 | int oom_kill_disable; | |
a7885eb8 | 260 | |
22a668d7 KH |
261 | /* set when res.limit == memsw.limit */ |
262 | bool memsw_is_minimum; | |
263 | ||
2e72b634 KS |
264 | /* protect arrays of thresholds */ |
265 | struct mutex thresholds_lock; | |
266 | ||
267 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 268 | struct mem_cgroup_thresholds thresholds; |
907860ed | 269 | |
2e72b634 | 270 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 271 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 272 | |
9490ff27 KH |
273 | /* For oom notifier event fd */ |
274 | struct list_head oom_notify; | |
185efc0f | 275 | |
7dc74be0 DN |
276 | /* |
277 | * Should we move charges of a task when a task is moved into this | |
278 | * mem_cgroup ? And what type of charges should we move ? | |
279 | */ | |
280 | unsigned long move_charge_at_immigrate; | |
d52aa412 | 281 | /* |
c62b1a3b | 282 | * percpu counter. |
d52aa412 | 283 | */ |
c62b1a3b | 284 | struct mem_cgroup_stat_cpu *stat; |
711d3d2c KH |
285 | /* |
286 | * used when a cpu is offlined or other synchronizations | |
287 | * See mem_cgroup_read_stat(). | |
288 | */ | |
289 | struct mem_cgroup_stat_cpu nocpu_base; | |
290 | spinlock_t pcp_counter_lock; | |
d1a4c0b3 GC |
291 | |
292 | #ifdef CONFIG_INET | |
293 | struct tcp_memcontrol tcp_mem; | |
294 | #endif | |
8cdea7c0 BS |
295 | }; |
296 | ||
7dc74be0 DN |
297 | /* Stuffs for move charges at task migration. */ |
298 | /* | |
299 | * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a | |
300 | * left-shifted bitmap of these types. | |
301 | */ | |
302 | enum move_type { | |
4ffef5fe | 303 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 304 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
305 | NR_MOVE_TYPE, |
306 | }; | |
307 | ||
4ffef5fe DN |
308 | /* "mc" and its members are protected by cgroup_mutex */ |
309 | static struct move_charge_struct { | |
b1dd693e | 310 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
311 | struct mem_cgroup *from; |
312 | struct mem_cgroup *to; | |
313 | unsigned long precharge; | |
854ffa8d | 314 | unsigned long moved_charge; |
483c30b5 | 315 | unsigned long moved_swap; |
8033b97c DN |
316 | struct task_struct *moving_task; /* a task moving charges */ |
317 | wait_queue_head_t waitq; /* a waitq for other context */ | |
318 | } mc = { | |
2bd9bb20 | 319 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
320 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
321 | }; | |
4ffef5fe | 322 | |
90254a65 DN |
323 | static bool move_anon(void) |
324 | { | |
325 | return test_bit(MOVE_CHARGE_TYPE_ANON, | |
326 | &mc.to->move_charge_at_immigrate); | |
327 | } | |
328 | ||
87946a72 DN |
329 | static bool move_file(void) |
330 | { | |
331 | return test_bit(MOVE_CHARGE_TYPE_FILE, | |
332 | &mc.to->move_charge_at_immigrate); | |
333 | } | |
334 | ||
4e416953 BS |
335 | /* |
336 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
337 | * limit reclaim to prevent infinite loops, if they ever occur. | |
338 | */ | |
339 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100) | |
340 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2) | |
341 | ||
217bc319 KH |
342 | enum charge_type { |
343 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
344 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
4f98a2fe | 345 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
c05555b5 | 346 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
d13d1443 | 347 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 348 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
349 | NR_CHARGE_TYPE, |
350 | }; | |
351 | ||
8c7c6e34 | 352 | /* for encoding cft->private value on file */ |
65c64ce8 GC |
353 | #define _MEM (0) |
354 | #define _MEMSWAP (1) | |
355 | #define _OOM_TYPE (2) | |
8c7c6e34 KH |
356 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) |
357 | #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) | |
358 | #define MEMFILE_ATTR(val) ((val) & 0xffff) | |
9490ff27 KH |
359 | /* Used for OOM nofiier */ |
360 | #define OOM_CONTROL (0) | |
8c7c6e34 | 361 | |
75822b44 BS |
362 | /* |
363 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
364 | */ | |
365 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
366 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
367 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
368 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
4e416953 BS |
369 | #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2 |
370 | #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT) | |
75822b44 | 371 | |
c0ff4b85 R |
372 | static void mem_cgroup_get(struct mem_cgroup *memcg); |
373 | static void mem_cgroup_put(struct mem_cgroup *memcg); | |
e1aab161 GC |
374 | |
375 | /* Writing them here to avoid exposing memcg's inner layout */ | |
376 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM | |
377 | #ifdef CONFIG_INET | |
378 | #include <net/sock.h> | |
d1a4c0b3 | 379 | #include <net/ip.h> |
e1aab161 GC |
380 | |
381 | static bool mem_cgroup_is_root(struct mem_cgroup *memcg); | |
382 | void sock_update_memcg(struct sock *sk) | |
383 | { | |
e1aab161 GC |
384 | if (static_branch(&memcg_socket_limit_enabled)) { |
385 | struct mem_cgroup *memcg; | |
386 | ||
387 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
388 | ||
f3f511e1 GC |
389 | /* Socket cloning can throw us here with sk_cgrp already |
390 | * filled. It won't however, necessarily happen from | |
391 | * process context. So the test for root memcg given | |
392 | * the current task's memcg won't help us in this case. | |
393 | * | |
394 | * Respecting the original socket's memcg is a better | |
395 | * decision in this case. | |
396 | */ | |
397 | if (sk->sk_cgrp) { | |
398 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
399 | mem_cgroup_get(sk->sk_cgrp->memcg); | |
400 | return; | |
401 | } | |
402 | ||
e1aab161 GC |
403 | rcu_read_lock(); |
404 | memcg = mem_cgroup_from_task(current); | |
405 | if (!mem_cgroup_is_root(memcg)) { | |
406 | mem_cgroup_get(memcg); | |
407 | sk->sk_cgrp = sk->sk_prot->proto_cgroup(memcg); | |
408 | } | |
409 | rcu_read_unlock(); | |
410 | } | |
411 | } | |
412 | EXPORT_SYMBOL(sock_update_memcg); | |
413 | ||
414 | void sock_release_memcg(struct sock *sk) | |
415 | { | |
416 | if (static_branch(&memcg_socket_limit_enabled) && sk->sk_cgrp) { | |
417 | struct mem_cgroup *memcg; | |
418 | WARN_ON(!sk->sk_cgrp->memcg); | |
419 | memcg = sk->sk_cgrp->memcg; | |
420 | mem_cgroup_put(memcg); | |
421 | } | |
422 | } | |
d1a4c0b3 GC |
423 | |
424 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
425 | { | |
426 | if (!memcg || mem_cgroup_is_root(memcg)) | |
427 | return NULL; | |
428 | ||
429 | return &memcg->tcp_mem.cg_proto; | |
430 | } | |
431 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 GC |
432 | #endif /* CONFIG_INET */ |
433 | #endif /* CONFIG_CGROUP_MEM_RES_CTLR_KMEM */ | |
434 | ||
c0ff4b85 | 435 | static void drain_all_stock_async(struct mem_cgroup *memcg); |
8c7c6e34 | 436 | |
f64c3f54 | 437 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 438 | mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid) |
f64c3f54 | 439 | { |
c0ff4b85 | 440 | return &memcg->info.nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
441 | } |
442 | ||
c0ff4b85 | 443 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 444 | { |
c0ff4b85 | 445 | return &memcg->css; |
d324236b WF |
446 | } |
447 | ||
f64c3f54 | 448 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 449 | page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 450 | { |
97a6c37b JW |
451 | int nid = page_to_nid(page); |
452 | int zid = page_zonenum(page); | |
f64c3f54 | 453 | |
c0ff4b85 | 454 | return mem_cgroup_zoneinfo(memcg, nid, zid); |
f64c3f54 BS |
455 | } |
456 | ||
457 | static struct mem_cgroup_tree_per_zone * | |
458 | soft_limit_tree_node_zone(int nid, int zid) | |
459 | { | |
460 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
461 | } | |
462 | ||
463 | static struct mem_cgroup_tree_per_zone * | |
464 | soft_limit_tree_from_page(struct page *page) | |
465 | { | |
466 | int nid = page_to_nid(page); | |
467 | int zid = page_zonenum(page); | |
468 | ||
469 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
470 | } | |
471 | ||
472 | static void | |
c0ff4b85 | 473 | __mem_cgroup_insert_exceeded(struct mem_cgroup *memcg, |
f64c3f54 | 474 | struct mem_cgroup_per_zone *mz, |
ef8745c1 KH |
475 | struct mem_cgroup_tree_per_zone *mctz, |
476 | unsigned long long new_usage_in_excess) | |
f64c3f54 BS |
477 | { |
478 | struct rb_node **p = &mctz->rb_root.rb_node; | |
479 | struct rb_node *parent = NULL; | |
480 | struct mem_cgroup_per_zone *mz_node; | |
481 | ||
482 | if (mz->on_tree) | |
483 | return; | |
484 | ||
ef8745c1 KH |
485 | mz->usage_in_excess = new_usage_in_excess; |
486 | if (!mz->usage_in_excess) | |
487 | return; | |
f64c3f54 BS |
488 | while (*p) { |
489 | parent = *p; | |
490 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
491 | tree_node); | |
492 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
493 | p = &(*p)->rb_left; | |
494 | /* | |
495 | * We can't avoid mem cgroups that are over their soft | |
496 | * limit by the same amount | |
497 | */ | |
498 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
499 | p = &(*p)->rb_right; | |
500 | } | |
501 | rb_link_node(&mz->tree_node, parent, p); | |
502 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
503 | mz->on_tree = true; | |
4e416953 BS |
504 | } |
505 | ||
506 | static void | |
c0ff4b85 | 507 | __mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
4e416953 BS |
508 | struct mem_cgroup_per_zone *mz, |
509 | struct mem_cgroup_tree_per_zone *mctz) | |
510 | { | |
511 | if (!mz->on_tree) | |
512 | return; | |
513 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
514 | mz->on_tree = false; | |
515 | } | |
516 | ||
f64c3f54 | 517 | static void |
c0ff4b85 | 518 | mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
f64c3f54 BS |
519 | struct mem_cgroup_per_zone *mz, |
520 | struct mem_cgroup_tree_per_zone *mctz) | |
521 | { | |
522 | spin_lock(&mctz->lock); | |
c0ff4b85 | 523 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
524 | spin_unlock(&mctz->lock); |
525 | } | |
526 | ||
f64c3f54 | 527 | |
c0ff4b85 | 528 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 529 | { |
ef8745c1 | 530 | unsigned long long excess; |
f64c3f54 BS |
531 | struct mem_cgroup_per_zone *mz; |
532 | struct mem_cgroup_tree_per_zone *mctz; | |
4e649152 KH |
533 | int nid = page_to_nid(page); |
534 | int zid = page_zonenum(page); | |
f64c3f54 BS |
535 | mctz = soft_limit_tree_from_page(page); |
536 | ||
537 | /* | |
4e649152 KH |
538 | * Necessary to update all ancestors when hierarchy is used. |
539 | * because their event counter is not touched. | |
f64c3f54 | 540 | */ |
c0ff4b85 R |
541 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
542 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
543 | excess = res_counter_soft_limit_excess(&memcg->res); | |
4e649152 KH |
544 | /* |
545 | * We have to update the tree if mz is on RB-tree or | |
546 | * mem is over its softlimit. | |
547 | */ | |
ef8745c1 | 548 | if (excess || mz->on_tree) { |
4e649152 KH |
549 | spin_lock(&mctz->lock); |
550 | /* if on-tree, remove it */ | |
551 | if (mz->on_tree) | |
c0ff4b85 | 552 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
4e649152 | 553 | /* |
ef8745c1 KH |
554 | * Insert again. mz->usage_in_excess will be updated. |
555 | * If excess is 0, no tree ops. | |
4e649152 | 556 | */ |
c0ff4b85 | 557 | __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess); |
4e649152 KH |
558 | spin_unlock(&mctz->lock); |
559 | } | |
f64c3f54 BS |
560 | } |
561 | } | |
562 | ||
c0ff4b85 | 563 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) |
f64c3f54 BS |
564 | { |
565 | int node, zone; | |
566 | struct mem_cgroup_per_zone *mz; | |
567 | struct mem_cgroup_tree_per_zone *mctz; | |
568 | ||
569 | for_each_node_state(node, N_POSSIBLE) { | |
570 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
c0ff4b85 | 571 | mz = mem_cgroup_zoneinfo(memcg, node, zone); |
f64c3f54 | 572 | mctz = soft_limit_tree_node_zone(node, zone); |
c0ff4b85 | 573 | mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
574 | } |
575 | } | |
576 | } | |
577 | ||
4e416953 BS |
578 | static struct mem_cgroup_per_zone * |
579 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
580 | { | |
581 | struct rb_node *rightmost = NULL; | |
26251eaf | 582 | struct mem_cgroup_per_zone *mz; |
4e416953 BS |
583 | |
584 | retry: | |
26251eaf | 585 | mz = NULL; |
4e416953 BS |
586 | rightmost = rb_last(&mctz->rb_root); |
587 | if (!rightmost) | |
588 | goto done; /* Nothing to reclaim from */ | |
589 | ||
590 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
591 | /* | |
592 | * Remove the node now but someone else can add it back, | |
593 | * we will to add it back at the end of reclaim to its correct | |
594 | * position in the tree. | |
595 | */ | |
596 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); | |
597 | if (!res_counter_soft_limit_excess(&mz->mem->res) || | |
598 | !css_tryget(&mz->mem->css)) | |
599 | goto retry; | |
600 | done: | |
601 | return mz; | |
602 | } | |
603 | ||
604 | static struct mem_cgroup_per_zone * | |
605 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
606 | { | |
607 | struct mem_cgroup_per_zone *mz; | |
608 | ||
609 | spin_lock(&mctz->lock); | |
610 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
611 | spin_unlock(&mctz->lock); | |
612 | return mz; | |
613 | } | |
614 | ||
711d3d2c KH |
615 | /* |
616 | * Implementation Note: reading percpu statistics for memcg. | |
617 | * | |
618 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
619 | * synchronization to implement "quick" read. There are trade-off between | |
620 | * reading cost and precision of value. Then, we may have a chance to implement | |
621 | * a periodic synchronizion of counter in memcg's counter. | |
622 | * | |
623 | * But this _read() function is used for user interface now. The user accounts | |
624 | * memory usage by memory cgroup and he _always_ requires exact value because | |
625 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
626 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
627 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
628 | * | |
629 | * If there are kernel internal actions which can make use of some not-exact | |
630 | * value, and reading all cpu value can be performance bottleneck in some | |
631 | * common workload, threashold and synchonization as vmstat[] should be | |
632 | * implemented. | |
633 | */ | |
c0ff4b85 | 634 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 635 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 636 | { |
7a159cc9 | 637 | long val = 0; |
c62b1a3b | 638 | int cpu; |
c62b1a3b | 639 | |
711d3d2c KH |
640 | get_online_cpus(); |
641 | for_each_online_cpu(cpu) | |
c0ff4b85 | 642 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 643 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
644 | spin_lock(&memcg->pcp_counter_lock); |
645 | val += memcg->nocpu_base.count[idx]; | |
646 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
647 | #endif |
648 | put_online_cpus(); | |
c62b1a3b KH |
649 | return val; |
650 | } | |
651 | ||
c0ff4b85 | 652 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
0c3e73e8 BS |
653 | bool charge) |
654 | { | |
655 | int val = (charge) ? 1 : -1; | |
c0ff4b85 | 656 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAPOUT], val); |
0c3e73e8 BS |
657 | } |
658 | ||
c0ff4b85 | 659 | void mem_cgroup_pgfault(struct mem_cgroup *memcg, int val) |
456f998e | 660 | { |
c0ff4b85 | 661 | this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT], val); |
456f998e YH |
662 | } |
663 | ||
c0ff4b85 | 664 | void mem_cgroup_pgmajfault(struct mem_cgroup *memcg, int val) |
456f998e | 665 | { |
c0ff4b85 | 666 | this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT], val); |
456f998e YH |
667 | } |
668 | ||
c0ff4b85 | 669 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
670 | enum mem_cgroup_events_index idx) |
671 | { | |
672 | unsigned long val = 0; | |
673 | int cpu; | |
674 | ||
675 | for_each_online_cpu(cpu) | |
c0ff4b85 | 676 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 677 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
678 | spin_lock(&memcg->pcp_counter_lock); |
679 | val += memcg->nocpu_base.events[idx]; | |
680 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f JW |
681 | #endif |
682 | return val; | |
683 | } | |
684 | ||
c0ff4b85 | 685 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
e401f176 | 686 | bool file, int nr_pages) |
d52aa412 | 687 | { |
c62b1a3b KH |
688 | preempt_disable(); |
689 | ||
e401f176 | 690 | if (file) |
c0ff4b85 R |
691 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
692 | nr_pages); | |
d52aa412 | 693 | else |
c0ff4b85 R |
694 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
695 | nr_pages); | |
55e462b0 | 696 | |
e401f176 KH |
697 | /* pagein of a big page is an event. So, ignore page size */ |
698 | if (nr_pages > 0) | |
c0ff4b85 | 699 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 700 | else { |
c0ff4b85 | 701 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
702 | nr_pages = -nr_pages; /* for event */ |
703 | } | |
e401f176 | 704 | |
c0ff4b85 | 705 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_COUNT], nr_pages); |
2e72b634 | 706 | |
c62b1a3b | 707 | preempt_enable(); |
6d12e2d8 KH |
708 | } |
709 | ||
bb2a0de9 | 710 | unsigned long |
c0ff4b85 | 711 | mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid, |
bb2a0de9 | 712 | unsigned int lru_mask) |
889976db YH |
713 | { |
714 | struct mem_cgroup_per_zone *mz; | |
bb2a0de9 KH |
715 | enum lru_list l; |
716 | unsigned long ret = 0; | |
717 | ||
c0ff4b85 | 718 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
bb2a0de9 KH |
719 | |
720 | for_each_lru(l) { | |
721 | if (BIT(l) & lru_mask) | |
722 | ret += MEM_CGROUP_ZSTAT(mz, l); | |
723 | } | |
724 | return ret; | |
725 | } | |
726 | ||
727 | static unsigned long | |
c0ff4b85 | 728 | mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 KH |
729 | int nid, unsigned int lru_mask) |
730 | { | |
889976db YH |
731 | u64 total = 0; |
732 | int zid; | |
733 | ||
bb2a0de9 | 734 | for (zid = 0; zid < MAX_NR_ZONES; zid++) |
c0ff4b85 R |
735 | total += mem_cgroup_zone_nr_lru_pages(memcg, |
736 | nid, zid, lru_mask); | |
bb2a0de9 | 737 | |
889976db YH |
738 | return total; |
739 | } | |
bb2a0de9 | 740 | |
c0ff4b85 | 741 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 742 | unsigned int lru_mask) |
6d12e2d8 | 743 | { |
889976db | 744 | int nid; |
6d12e2d8 KH |
745 | u64 total = 0; |
746 | ||
bb2a0de9 | 747 | for_each_node_state(nid, N_HIGH_MEMORY) |
c0ff4b85 | 748 | total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
6d12e2d8 | 749 | return total; |
d52aa412 KH |
750 | } |
751 | ||
c0ff4b85 | 752 | static bool __memcg_event_check(struct mem_cgroup *memcg, int target) |
7a159cc9 JW |
753 | { |
754 | unsigned long val, next; | |
755 | ||
4799401f SR |
756 | val = __this_cpu_read(memcg->stat->events[MEM_CGROUP_EVENTS_COUNT]); |
757 | next = __this_cpu_read(memcg->stat->targets[target]); | |
7a159cc9 JW |
758 | /* from time_after() in jiffies.h */ |
759 | return ((long)next - (long)val < 0); | |
760 | } | |
761 | ||
c0ff4b85 | 762 | static void __mem_cgroup_target_update(struct mem_cgroup *memcg, int target) |
d2265e6f | 763 | { |
7a159cc9 | 764 | unsigned long val, next; |
d2265e6f | 765 | |
4799401f | 766 | val = __this_cpu_read(memcg->stat->events[MEM_CGROUP_EVENTS_COUNT]); |
d2265e6f | 767 | |
7a159cc9 JW |
768 | switch (target) { |
769 | case MEM_CGROUP_TARGET_THRESH: | |
770 | next = val + THRESHOLDS_EVENTS_TARGET; | |
771 | break; | |
772 | case MEM_CGROUP_TARGET_SOFTLIMIT: | |
773 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
774 | break; | |
453a9bf3 KH |
775 | case MEM_CGROUP_TARGET_NUMAINFO: |
776 | next = val + NUMAINFO_EVENTS_TARGET; | |
777 | break; | |
7a159cc9 JW |
778 | default: |
779 | return; | |
780 | } | |
781 | ||
4799401f | 782 | __this_cpu_write(memcg->stat->targets[target], next); |
d2265e6f KH |
783 | } |
784 | ||
785 | /* | |
786 | * Check events in order. | |
787 | * | |
788 | */ | |
c0ff4b85 | 789 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f | 790 | { |
4799401f | 791 | preempt_disable(); |
d2265e6f | 792 | /* threshold event is triggered in finer grain than soft limit */ |
c0ff4b85 R |
793 | if (unlikely(__memcg_event_check(memcg, MEM_CGROUP_TARGET_THRESH))) { |
794 | mem_cgroup_threshold(memcg); | |
795 | __mem_cgroup_target_update(memcg, MEM_CGROUP_TARGET_THRESH); | |
796 | if (unlikely(__memcg_event_check(memcg, | |
453a9bf3 | 797 | MEM_CGROUP_TARGET_SOFTLIMIT))) { |
c0ff4b85 R |
798 | mem_cgroup_update_tree(memcg, page); |
799 | __mem_cgroup_target_update(memcg, | |
453a9bf3 KH |
800 | MEM_CGROUP_TARGET_SOFTLIMIT); |
801 | } | |
802 | #if MAX_NUMNODES > 1 | |
c0ff4b85 | 803 | if (unlikely(__memcg_event_check(memcg, |
453a9bf3 | 804 | MEM_CGROUP_TARGET_NUMAINFO))) { |
c0ff4b85 R |
805 | atomic_inc(&memcg->numainfo_events); |
806 | __mem_cgroup_target_update(memcg, | |
453a9bf3 | 807 | MEM_CGROUP_TARGET_NUMAINFO); |
7a159cc9 | 808 | } |
453a9bf3 | 809 | #endif |
d2265e6f | 810 | } |
4799401f | 811 | preempt_enable(); |
d2265e6f KH |
812 | } |
813 | ||
d1a4c0b3 | 814 | struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
815 | { |
816 | return container_of(cgroup_subsys_state(cont, | |
817 | mem_cgroup_subsys_id), struct mem_cgroup, | |
818 | css); | |
819 | } | |
820 | ||
cf475ad2 | 821 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 822 | { |
31a78f23 BS |
823 | /* |
824 | * mm_update_next_owner() may clear mm->owner to NULL | |
825 | * if it races with swapoff, page migration, etc. | |
826 | * So this can be called with p == NULL. | |
827 | */ | |
828 | if (unlikely(!p)) | |
829 | return NULL; | |
830 | ||
78fb7466 PE |
831 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
832 | struct mem_cgroup, css); | |
833 | } | |
834 | ||
a433658c | 835 | struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 836 | { |
c0ff4b85 | 837 | struct mem_cgroup *memcg = NULL; |
0b7f569e KH |
838 | |
839 | if (!mm) | |
840 | return NULL; | |
54595fe2 KH |
841 | /* |
842 | * Because we have no locks, mm->owner's may be being moved to other | |
843 | * cgroup. We use css_tryget() here even if this looks | |
844 | * pessimistic (rather than adding locks here). | |
845 | */ | |
846 | rcu_read_lock(); | |
847 | do { | |
c0ff4b85 R |
848 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
849 | if (unlikely(!memcg)) | |
54595fe2 | 850 | break; |
c0ff4b85 | 851 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 852 | rcu_read_unlock(); |
c0ff4b85 | 853 | return memcg; |
54595fe2 KH |
854 | } |
855 | ||
7d74b06f | 856 | /* The caller has to guarantee "mem" exists before calling this */ |
c0ff4b85 | 857 | static struct mem_cgroup *mem_cgroup_start_loop(struct mem_cgroup *memcg) |
14067bb3 | 858 | { |
711d3d2c KH |
859 | struct cgroup_subsys_state *css; |
860 | int found; | |
861 | ||
c0ff4b85 | 862 | if (!memcg) /* ROOT cgroup has the smallest ID */ |
711d3d2c | 863 | return root_mem_cgroup; /*css_put/get against root is ignored*/ |
c0ff4b85 R |
864 | if (!memcg->use_hierarchy) { |
865 | if (css_tryget(&memcg->css)) | |
866 | return memcg; | |
711d3d2c KH |
867 | return NULL; |
868 | } | |
869 | rcu_read_lock(); | |
870 | /* | |
871 | * searching a memory cgroup which has the smallest ID under given | |
872 | * ROOT cgroup. (ID >= 1) | |
873 | */ | |
c0ff4b85 | 874 | css = css_get_next(&mem_cgroup_subsys, 1, &memcg->css, &found); |
711d3d2c | 875 | if (css && css_tryget(css)) |
c0ff4b85 | 876 | memcg = container_of(css, struct mem_cgroup, css); |
711d3d2c | 877 | else |
c0ff4b85 | 878 | memcg = NULL; |
711d3d2c | 879 | rcu_read_unlock(); |
c0ff4b85 | 880 | return memcg; |
7d74b06f KH |
881 | } |
882 | ||
883 | static struct mem_cgroup *mem_cgroup_get_next(struct mem_cgroup *iter, | |
884 | struct mem_cgroup *root, | |
885 | bool cond) | |
886 | { | |
887 | int nextid = css_id(&iter->css) + 1; | |
888 | int found; | |
889 | int hierarchy_used; | |
14067bb3 | 890 | struct cgroup_subsys_state *css; |
14067bb3 | 891 | |
7d74b06f | 892 | hierarchy_used = iter->use_hierarchy; |
14067bb3 | 893 | |
7d74b06f | 894 | css_put(&iter->css); |
711d3d2c KH |
895 | /* If no ROOT, walk all, ignore hierarchy */ |
896 | if (!cond || (root && !hierarchy_used)) | |
7d74b06f | 897 | return NULL; |
14067bb3 | 898 | |
711d3d2c KH |
899 | if (!root) |
900 | root = root_mem_cgroup; | |
901 | ||
7d74b06f KH |
902 | do { |
903 | iter = NULL; | |
14067bb3 | 904 | rcu_read_lock(); |
7d74b06f KH |
905 | |
906 | css = css_get_next(&mem_cgroup_subsys, nextid, | |
907 | &root->css, &found); | |
14067bb3 | 908 | if (css && css_tryget(css)) |
7d74b06f | 909 | iter = container_of(css, struct mem_cgroup, css); |
14067bb3 | 910 | rcu_read_unlock(); |
7d74b06f | 911 | /* If css is NULL, no more cgroups will be found */ |
14067bb3 | 912 | nextid = found + 1; |
7d74b06f | 913 | } while (css && !iter); |
14067bb3 | 914 | |
7d74b06f | 915 | return iter; |
14067bb3 | 916 | } |
7d74b06f KH |
917 | /* |
918 | * for_eacn_mem_cgroup_tree() for visiting all cgroup under tree. Please | |
919 | * be careful that "break" loop is not allowed. We have reference count. | |
920 | * Instead of that modify "cond" to be false and "continue" to exit the loop. | |
921 | */ | |
922 | #define for_each_mem_cgroup_tree_cond(iter, root, cond) \ | |
923 | for (iter = mem_cgroup_start_loop(root);\ | |
924 | iter != NULL;\ | |
925 | iter = mem_cgroup_get_next(iter, root, cond)) | |
926 | ||
927 | #define for_each_mem_cgroup_tree(iter, root) \ | |
928 | for_each_mem_cgroup_tree_cond(iter, root, true) | |
929 | ||
711d3d2c KH |
930 | #define for_each_mem_cgroup_all(iter) \ |
931 | for_each_mem_cgroup_tree_cond(iter, NULL, true) | |
932 | ||
14067bb3 | 933 | |
c0ff4b85 | 934 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
4b3bde4c | 935 | { |
c0ff4b85 | 936 | return (memcg == root_mem_cgroup); |
4b3bde4c BS |
937 | } |
938 | ||
456f998e YH |
939 | void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
940 | { | |
c0ff4b85 | 941 | struct mem_cgroup *memcg; |
456f998e YH |
942 | |
943 | if (!mm) | |
944 | return; | |
945 | ||
946 | rcu_read_lock(); | |
c0ff4b85 R |
947 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
948 | if (unlikely(!memcg)) | |
456f998e YH |
949 | goto out; |
950 | ||
951 | switch (idx) { | |
952 | case PGMAJFAULT: | |
c0ff4b85 | 953 | mem_cgroup_pgmajfault(memcg, 1); |
456f998e YH |
954 | break; |
955 | case PGFAULT: | |
c0ff4b85 | 956 | mem_cgroup_pgfault(memcg, 1); |
456f998e YH |
957 | break; |
958 | default: | |
959 | BUG(); | |
960 | } | |
961 | out: | |
962 | rcu_read_unlock(); | |
963 | } | |
964 | EXPORT_SYMBOL(mem_cgroup_count_vm_event); | |
965 | ||
08e552c6 KH |
966 | /* |
967 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
968 | * Operations are called by routine of global LRU independently from memcg. | |
969 | * What we have to take care of here is validness of pc->mem_cgroup. | |
970 | * | |
971 | * Changes to pc->mem_cgroup happens when | |
972 | * 1. charge | |
973 | * 2. moving account | |
974 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
975 | * It is added to LRU before charge. | |
976 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
977 | * When moving account, the page is not on LRU. It's isolated. | |
978 | */ | |
4f98a2fe | 979 | |
08e552c6 KH |
980 | void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru) |
981 | { | |
982 | struct page_cgroup *pc; | |
08e552c6 | 983 | struct mem_cgroup_per_zone *mz; |
6d12e2d8 | 984 | |
f8d66542 | 985 | if (mem_cgroup_disabled()) |
08e552c6 KH |
986 | return; |
987 | pc = lookup_page_cgroup(page); | |
988 | /* can happen while we handle swapcache. */ | |
4b3bde4c | 989 | if (!TestClearPageCgroupAcctLRU(pc)) |
08e552c6 | 990 | return; |
4b3bde4c | 991 | VM_BUG_ON(!pc->mem_cgroup); |
544122e5 KH |
992 | /* |
993 | * We don't check PCG_USED bit. It's cleared when the "page" is finally | |
994 | * removed from global LRU. | |
995 | */ | |
97a6c37b | 996 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
ece35ca8 KH |
997 | /* huge page split is done under lru_lock. so, we have no races. */ |
998 | MEM_CGROUP_ZSTAT(mz, lru) -= 1 << compound_order(page); | |
4b3bde4c BS |
999 | if (mem_cgroup_is_root(pc->mem_cgroup)) |
1000 | return; | |
1001 | VM_BUG_ON(list_empty(&pc->lru)); | |
08e552c6 | 1002 | list_del_init(&pc->lru); |
6d12e2d8 KH |
1003 | } |
1004 | ||
08e552c6 | 1005 | void mem_cgroup_del_lru(struct page *page) |
6d12e2d8 | 1006 | { |
08e552c6 KH |
1007 | mem_cgroup_del_lru_list(page, page_lru(page)); |
1008 | } | |
b69408e8 | 1009 | |
3f58a829 MK |
1010 | /* |
1011 | * Writeback is about to end against a page which has been marked for immediate | |
1012 | * reclaim. If it still appears to be reclaimable, move it to the tail of the | |
1013 | * inactive list. | |
1014 | */ | |
1015 | void mem_cgroup_rotate_reclaimable_page(struct page *page) | |
1016 | { | |
1017 | struct mem_cgroup_per_zone *mz; | |
1018 | struct page_cgroup *pc; | |
1019 | enum lru_list lru = page_lru(page); | |
1020 | ||
1021 | if (mem_cgroup_disabled()) | |
1022 | return; | |
1023 | ||
1024 | pc = lookup_page_cgroup(page); | |
1025 | /* unused or root page is not rotated. */ | |
1026 | if (!PageCgroupUsed(pc)) | |
1027 | return; | |
1028 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ | |
1029 | smp_rmb(); | |
1030 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
1031 | return; | |
97a6c37b | 1032 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
3f58a829 MK |
1033 | list_move_tail(&pc->lru, &mz->lists[lru]); |
1034 | } | |
1035 | ||
08e552c6 KH |
1036 | void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru) |
1037 | { | |
1038 | struct mem_cgroup_per_zone *mz; | |
1039 | struct page_cgroup *pc; | |
b69408e8 | 1040 | |
f8d66542 | 1041 | if (mem_cgroup_disabled()) |
08e552c6 | 1042 | return; |
6d12e2d8 | 1043 | |
08e552c6 | 1044 | pc = lookup_page_cgroup(page); |
4b3bde4c | 1045 | /* unused or root page is not rotated. */ |
713735b4 JW |
1046 | if (!PageCgroupUsed(pc)) |
1047 | return; | |
1048 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ | |
1049 | smp_rmb(); | |
1050 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
08e552c6 | 1051 | return; |
97a6c37b | 1052 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
08e552c6 | 1053 | list_move(&pc->lru, &mz->lists[lru]); |
6d12e2d8 KH |
1054 | } |
1055 | ||
08e552c6 | 1056 | void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru) |
66e1707b | 1057 | { |
08e552c6 KH |
1058 | struct page_cgroup *pc; |
1059 | struct mem_cgroup_per_zone *mz; | |
6d12e2d8 | 1060 | |
f8d66542 | 1061 | if (mem_cgroup_disabled()) |
08e552c6 KH |
1062 | return; |
1063 | pc = lookup_page_cgroup(page); | |
4b3bde4c | 1064 | VM_BUG_ON(PageCgroupAcctLRU(pc)); |
a61ed3ce JW |
1065 | /* |
1066 | * putback: charge: | |
1067 | * SetPageLRU SetPageCgroupUsed | |
1068 | * smp_mb smp_mb | |
1069 | * PageCgroupUsed && add to memcg LRU PageLRU && add to memcg LRU | |
1070 | * | |
1071 | * Ensure that one of the two sides adds the page to the memcg | |
1072 | * LRU during a race. | |
1073 | */ | |
1074 | smp_mb(); | |
08e552c6 | 1075 | if (!PageCgroupUsed(pc)) |
894bc310 | 1076 | return; |
713735b4 JW |
1077 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ |
1078 | smp_rmb(); | |
97a6c37b | 1079 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
ece35ca8 KH |
1080 | /* huge page split is done under lru_lock. so, we have no races. */ |
1081 | MEM_CGROUP_ZSTAT(mz, lru) += 1 << compound_order(page); | |
4b3bde4c BS |
1082 | SetPageCgroupAcctLRU(pc); |
1083 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
1084 | return; | |
08e552c6 KH |
1085 | list_add(&pc->lru, &mz->lists[lru]); |
1086 | } | |
544122e5 | 1087 | |
08e552c6 | 1088 | /* |
5a6475a4 KH |
1089 | * At handling SwapCache and other FUSE stuff, pc->mem_cgroup may be changed |
1090 | * while it's linked to lru because the page may be reused after it's fully | |
1091 | * uncharged. To handle that, unlink page_cgroup from LRU when charge it again. | |
1092 | * It's done under lock_page and expected that zone->lru_lock isnever held. | |
08e552c6 | 1093 | */ |
5a6475a4 | 1094 | static void mem_cgroup_lru_del_before_commit(struct page *page) |
08e552c6 | 1095 | { |
544122e5 KH |
1096 | unsigned long flags; |
1097 | struct zone *zone = page_zone(page); | |
1098 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
1099 | ||
5a6475a4 KH |
1100 | /* |
1101 | * Doing this check without taking ->lru_lock seems wrong but this | |
1102 | * is safe. Because if page_cgroup's USED bit is unset, the page | |
1103 | * will not be added to any memcg's LRU. If page_cgroup's USED bit is | |
1104 | * set, the commit after this will fail, anyway. | |
1105 | * This all charge/uncharge is done under some mutual execustion. | |
1106 | * So, we don't need to taking care of changes in USED bit. | |
1107 | */ | |
1108 | if (likely(!PageLRU(page))) | |
1109 | return; | |
1110 | ||
544122e5 KH |
1111 | spin_lock_irqsave(&zone->lru_lock, flags); |
1112 | /* | |
1113 | * Forget old LRU when this page_cgroup is *not* used. This Used bit | |
1114 | * is guarded by lock_page() because the page is SwapCache. | |
1115 | */ | |
1116 | if (!PageCgroupUsed(pc)) | |
1117 | mem_cgroup_del_lru_list(page, page_lru(page)); | |
1118 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
08e552c6 KH |
1119 | } |
1120 | ||
5a6475a4 | 1121 | static void mem_cgroup_lru_add_after_commit(struct page *page) |
544122e5 KH |
1122 | { |
1123 | unsigned long flags; | |
1124 | struct zone *zone = page_zone(page); | |
1125 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
a61ed3ce JW |
1126 | /* |
1127 | * putback: charge: | |
1128 | * SetPageLRU SetPageCgroupUsed | |
1129 | * smp_mb smp_mb | |
1130 | * PageCgroupUsed && add to memcg LRU PageLRU && add to memcg LRU | |
1131 | * | |
1132 | * Ensure that one of the two sides adds the page to the memcg | |
1133 | * LRU during a race. | |
1134 | */ | |
1135 | smp_mb(); | |
5a6475a4 KH |
1136 | /* taking care of that the page is added to LRU while we commit it */ |
1137 | if (likely(!PageLRU(page))) | |
1138 | return; | |
544122e5 KH |
1139 | spin_lock_irqsave(&zone->lru_lock, flags); |
1140 | /* link when the page is linked to LRU but page_cgroup isn't */ | |
4b3bde4c | 1141 | if (PageLRU(page) && !PageCgroupAcctLRU(pc)) |
544122e5 KH |
1142 | mem_cgroup_add_lru_list(page, page_lru(page)); |
1143 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
1144 | } | |
1145 | ||
1146 | ||
08e552c6 KH |
1147 | void mem_cgroup_move_lists(struct page *page, |
1148 | enum lru_list from, enum lru_list to) | |
1149 | { | |
f8d66542 | 1150 | if (mem_cgroup_disabled()) |
08e552c6 KH |
1151 | return; |
1152 | mem_cgroup_del_lru_list(page, from); | |
1153 | mem_cgroup_add_lru_list(page, to); | |
66e1707b BS |
1154 | } |
1155 | ||
3e92041d | 1156 | /* |
c0ff4b85 | 1157 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1158 | * hierarchy subtree |
1159 | */ | |
c0ff4b85 R |
1160 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1161 | struct mem_cgroup *memcg) | |
3e92041d | 1162 | { |
c0ff4b85 R |
1163 | if (root_memcg != memcg) { |
1164 | return (root_memcg->use_hierarchy && | |
1165 | css_is_ancestor(&memcg->css, &root_memcg->css)); | |
3e92041d MH |
1166 | } |
1167 | ||
1168 | return true; | |
1169 | } | |
1170 | ||
c0ff4b85 | 1171 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg) |
4c4a2214 DR |
1172 | { |
1173 | int ret; | |
0b7f569e | 1174 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1175 | struct task_struct *p; |
4c4a2214 | 1176 | |
158e0a2d KH |
1177 | p = find_lock_task_mm(task); |
1178 | if (!p) | |
1179 | return 0; | |
1180 | curr = try_get_mem_cgroup_from_mm(p->mm); | |
1181 | task_unlock(p); | |
0b7f569e KH |
1182 | if (!curr) |
1183 | return 0; | |
d31f56db | 1184 | /* |
c0ff4b85 | 1185 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1186 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1187 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1188 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1189 | */ |
c0ff4b85 | 1190 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1191 | css_put(&curr->css); |
4c4a2214 DR |
1192 | return ret; |
1193 | } | |
1194 | ||
9b272977 | 1195 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg, struct zone *zone) |
14797e23 | 1196 | { |
9b272977 JW |
1197 | unsigned long inactive_ratio; |
1198 | int nid = zone_to_nid(zone); | |
1199 | int zid = zone_idx(zone); | |
14797e23 | 1200 | unsigned long inactive; |
9b272977 | 1201 | unsigned long active; |
c772be93 | 1202 | unsigned long gb; |
14797e23 | 1203 | |
9b272977 JW |
1204 | inactive = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid, |
1205 | BIT(LRU_INACTIVE_ANON)); | |
1206 | active = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid, | |
1207 | BIT(LRU_ACTIVE_ANON)); | |
14797e23 | 1208 | |
c772be93 KM |
1209 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1210 | if (gb) | |
1211 | inactive_ratio = int_sqrt(10 * gb); | |
1212 | else | |
1213 | inactive_ratio = 1; | |
1214 | ||
9b272977 | 1215 | return inactive * inactive_ratio < active; |
14797e23 KM |
1216 | } |
1217 | ||
9b272977 | 1218 | int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg, struct zone *zone) |
56e49d21 RR |
1219 | { |
1220 | unsigned long active; | |
1221 | unsigned long inactive; | |
9b272977 JW |
1222 | int zid = zone_idx(zone); |
1223 | int nid = zone_to_nid(zone); | |
56e49d21 | 1224 | |
9b272977 JW |
1225 | inactive = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid, |
1226 | BIT(LRU_INACTIVE_FILE)); | |
1227 | active = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid, | |
1228 | BIT(LRU_ACTIVE_FILE)); | |
56e49d21 RR |
1229 | |
1230 | return (active > inactive); | |
1231 | } | |
1232 | ||
3e2f41f1 KM |
1233 | struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, |
1234 | struct zone *zone) | |
1235 | { | |
13d7e3a2 | 1236 | int nid = zone_to_nid(zone); |
3e2f41f1 KM |
1237 | int zid = zone_idx(zone); |
1238 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
1239 | ||
1240 | return &mz->reclaim_stat; | |
1241 | } | |
1242 | ||
1243 | struct zone_reclaim_stat * | |
1244 | mem_cgroup_get_reclaim_stat_from_page(struct page *page) | |
1245 | { | |
1246 | struct page_cgroup *pc; | |
1247 | struct mem_cgroup_per_zone *mz; | |
1248 | ||
1249 | if (mem_cgroup_disabled()) | |
1250 | return NULL; | |
1251 | ||
1252 | pc = lookup_page_cgroup(page); | |
bd112db8 DN |
1253 | if (!PageCgroupUsed(pc)) |
1254 | return NULL; | |
713735b4 JW |
1255 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ |
1256 | smp_rmb(); | |
97a6c37b | 1257 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
3e2f41f1 KM |
1258 | return &mz->reclaim_stat; |
1259 | } | |
1260 | ||
66e1707b BS |
1261 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
1262 | struct list_head *dst, | |
1263 | unsigned long *scanned, int order, | |
4356f21d MK |
1264 | isolate_mode_t mode, |
1265 | struct zone *z, | |
66e1707b | 1266 | struct mem_cgroup *mem_cont, |
4f98a2fe | 1267 | int active, int file) |
66e1707b BS |
1268 | { |
1269 | unsigned long nr_taken = 0; | |
1270 | struct page *page; | |
1271 | unsigned long scan; | |
1272 | LIST_HEAD(pc_list); | |
1273 | struct list_head *src; | |
ff7283fa | 1274 | struct page_cgroup *pc, *tmp; |
13d7e3a2 | 1275 | int nid = zone_to_nid(z); |
1ecaab2b KH |
1276 | int zid = zone_idx(z); |
1277 | struct mem_cgroup_per_zone *mz; | |
b7c46d15 | 1278 | int lru = LRU_FILE * file + active; |
2ffebca6 | 1279 | int ret; |
66e1707b | 1280 | |
cf475ad2 | 1281 | BUG_ON(!mem_cont); |
1ecaab2b | 1282 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
b69408e8 | 1283 | src = &mz->lists[lru]; |
66e1707b | 1284 | |
ff7283fa KH |
1285 | scan = 0; |
1286 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 1287 | if (scan >= nr_to_scan) |
ff7283fa | 1288 | break; |
08e552c6 | 1289 | |
52d4b9ac KH |
1290 | if (unlikely(!PageCgroupUsed(pc))) |
1291 | continue; | |
5564e88b | 1292 | |
6b3ae58e | 1293 | page = lookup_cgroup_page(pc); |
5564e88b | 1294 | |
436c6541 | 1295 | if (unlikely(!PageLRU(page))) |
ff7283fa | 1296 | continue; |
ff7283fa | 1297 | |
436c6541 | 1298 | scan++; |
2ffebca6 KH |
1299 | ret = __isolate_lru_page(page, mode, file); |
1300 | switch (ret) { | |
1301 | case 0: | |
66e1707b | 1302 | list_move(&page->lru, dst); |
2ffebca6 | 1303 | mem_cgroup_del_lru(page); |
2c888cfb | 1304 | nr_taken += hpage_nr_pages(page); |
2ffebca6 KH |
1305 | break; |
1306 | case -EBUSY: | |
1307 | /* we don't affect global LRU but rotate in our LRU */ | |
1308 | mem_cgroup_rotate_lru_list(page, page_lru(page)); | |
1309 | break; | |
1310 | default: | |
1311 | break; | |
66e1707b BS |
1312 | } |
1313 | } | |
1314 | ||
66e1707b | 1315 | *scanned = scan; |
cc8e970c KM |
1316 | |
1317 | trace_mm_vmscan_memcg_isolate(0, nr_to_scan, scan, nr_taken, | |
1318 | 0, 0, 0, mode); | |
1319 | ||
66e1707b BS |
1320 | return nr_taken; |
1321 | } | |
1322 | ||
6d61ef40 BS |
1323 | #define mem_cgroup_from_res_counter(counter, member) \ |
1324 | container_of(counter, struct mem_cgroup, member) | |
1325 | ||
19942822 | 1326 | /** |
9d11ea9f JW |
1327 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
1328 | * @mem: the memory cgroup | |
19942822 | 1329 | * |
9d11ea9f | 1330 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1331 | * pages. |
19942822 | 1332 | */ |
c0ff4b85 | 1333 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1334 | { |
9d11ea9f JW |
1335 | unsigned long long margin; |
1336 | ||
c0ff4b85 | 1337 | margin = res_counter_margin(&memcg->res); |
9d11ea9f | 1338 | if (do_swap_account) |
c0ff4b85 | 1339 | margin = min(margin, res_counter_margin(&memcg->memsw)); |
7ec99d62 | 1340 | return margin >> PAGE_SHIFT; |
19942822 JW |
1341 | } |
1342 | ||
1f4c025b | 1343 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 KM |
1344 | { |
1345 | struct cgroup *cgrp = memcg->css.cgroup; | |
a7885eb8 KM |
1346 | |
1347 | /* root ? */ | |
1348 | if (cgrp->parent == NULL) | |
1349 | return vm_swappiness; | |
1350 | ||
bf1ff263 | 1351 | return memcg->swappiness; |
a7885eb8 KM |
1352 | } |
1353 | ||
c0ff4b85 | 1354 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a KH |
1355 | { |
1356 | int cpu; | |
1489ebad KH |
1357 | |
1358 | get_online_cpus(); | |
c0ff4b85 | 1359 | spin_lock(&memcg->pcp_counter_lock); |
1489ebad | 1360 | for_each_online_cpu(cpu) |
c0ff4b85 R |
1361 | per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) += 1; |
1362 | memcg->nocpu_base.count[MEM_CGROUP_ON_MOVE] += 1; | |
1363 | spin_unlock(&memcg->pcp_counter_lock); | |
1489ebad | 1364 | put_online_cpus(); |
32047e2a KH |
1365 | |
1366 | synchronize_rcu(); | |
1367 | } | |
1368 | ||
c0ff4b85 | 1369 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a KH |
1370 | { |
1371 | int cpu; | |
1372 | ||
c0ff4b85 | 1373 | if (!memcg) |
32047e2a | 1374 | return; |
1489ebad | 1375 | get_online_cpus(); |
c0ff4b85 | 1376 | spin_lock(&memcg->pcp_counter_lock); |
1489ebad | 1377 | for_each_online_cpu(cpu) |
c0ff4b85 R |
1378 | per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) -= 1; |
1379 | memcg->nocpu_base.count[MEM_CGROUP_ON_MOVE] -= 1; | |
1380 | spin_unlock(&memcg->pcp_counter_lock); | |
1489ebad | 1381 | put_online_cpus(); |
32047e2a KH |
1382 | } |
1383 | /* | |
1384 | * 2 routines for checking "mem" is under move_account() or not. | |
1385 | * | |
1386 | * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used | |
1387 | * for avoiding race in accounting. If true, | |
1388 | * pc->mem_cgroup may be overwritten. | |
1389 | * | |
1390 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or | |
1391 | * under hierarchy of moving cgroups. This is for | |
1392 | * waiting at hith-memory prressure caused by "move". | |
1393 | */ | |
1394 | ||
c0ff4b85 | 1395 | static bool mem_cgroup_stealed(struct mem_cgroup *memcg) |
32047e2a KH |
1396 | { |
1397 | VM_BUG_ON(!rcu_read_lock_held()); | |
c0ff4b85 | 1398 | return this_cpu_read(memcg->stat->count[MEM_CGROUP_ON_MOVE]) > 0; |
32047e2a | 1399 | } |
4b534334 | 1400 | |
c0ff4b85 | 1401 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1402 | { |
2bd9bb20 KH |
1403 | struct mem_cgroup *from; |
1404 | struct mem_cgroup *to; | |
4b534334 | 1405 | bool ret = false; |
2bd9bb20 KH |
1406 | /* |
1407 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1408 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1409 | */ | |
1410 | spin_lock(&mc.lock); | |
1411 | from = mc.from; | |
1412 | to = mc.to; | |
1413 | if (!from) | |
1414 | goto unlock; | |
3e92041d | 1415 | |
c0ff4b85 R |
1416 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1417 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1418 | unlock: |
1419 | spin_unlock(&mc.lock); | |
4b534334 KH |
1420 | return ret; |
1421 | } | |
1422 | ||
c0ff4b85 | 1423 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1424 | { |
1425 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1426 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1427 | DEFINE_WAIT(wait); |
1428 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1429 | /* moving charge context might have finished. */ | |
1430 | if (mc.moving_task) | |
1431 | schedule(); | |
1432 | finish_wait(&mc.waitq, &wait); | |
1433 | return true; | |
1434 | } | |
1435 | } | |
1436 | return false; | |
1437 | } | |
1438 | ||
e222432b | 1439 | /** |
6a6135b6 | 1440 | * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode. |
e222432b BS |
1441 | * @memcg: The memory cgroup that went over limit |
1442 | * @p: Task that is going to be killed | |
1443 | * | |
1444 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1445 | * enabled | |
1446 | */ | |
1447 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1448 | { | |
1449 | struct cgroup *task_cgrp; | |
1450 | struct cgroup *mem_cgrp; | |
1451 | /* | |
1452 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1453 | * on the assumption that OOM is serialized for memory controller. | |
1454 | * If this assumption is broken, revisit this code. | |
1455 | */ | |
1456 | static char memcg_name[PATH_MAX]; | |
1457 | int ret; | |
1458 | ||
d31f56db | 1459 | if (!memcg || !p) |
e222432b BS |
1460 | return; |
1461 | ||
1462 | ||
1463 | rcu_read_lock(); | |
1464 | ||
1465 | mem_cgrp = memcg->css.cgroup; | |
1466 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1467 | ||
1468 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1469 | if (ret < 0) { | |
1470 | /* | |
1471 | * Unfortunately, we are unable to convert to a useful name | |
1472 | * But we'll still print out the usage information | |
1473 | */ | |
1474 | rcu_read_unlock(); | |
1475 | goto done; | |
1476 | } | |
1477 | rcu_read_unlock(); | |
1478 | ||
1479 | printk(KERN_INFO "Task in %s killed", memcg_name); | |
1480 | ||
1481 | rcu_read_lock(); | |
1482 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1483 | if (ret < 0) { | |
1484 | rcu_read_unlock(); | |
1485 | goto done; | |
1486 | } | |
1487 | rcu_read_unlock(); | |
1488 | ||
1489 | /* | |
1490 | * Continues from above, so we don't need an KERN_ level | |
1491 | */ | |
1492 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); | |
1493 | done: | |
1494 | ||
1495 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", | |
1496 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, | |
1497 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1498 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
1499 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " | |
1500 | "failcnt %llu\n", | |
1501 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, | |
1502 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1503 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
1504 | } | |
1505 | ||
81d39c20 KH |
1506 | /* |
1507 | * This function returns the number of memcg under hierarchy tree. Returns | |
1508 | * 1(self count) if no children. | |
1509 | */ | |
c0ff4b85 | 1510 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1511 | { |
1512 | int num = 0; | |
7d74b06f KH |
1513 | struct mem_cgroup *iter; |
1514 | ||
c0ff4b85 | 1515 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1516 | num++; |
81d39c20 KH |
1517 | return num; |
1518 | } | |
1519 | ||
a63d83f4 DR |
1520 | /* |
1521 | * Return the memory (and swap, if configured) limit for a memcg. | |
1522 | */ | |
1523 | u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) | |
1524 | { | |
1525 | u64 limit; | |
1526 | u64 memsw; | |
1527 | ||
f3e8eb70 JW |
1528 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
1529 | limit += total_swap_pages << PAGE_SHIFT; | |
1530 | ||
a63d83f4 DR |
1531 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
1532 | /* | |
1533 | * If memsw is finite and limits the amount of swap space available | |
1534 | * to this memcg, return that limit. | |
1535 | */ | |
1536 | return min(limit, memsw); | |
1537 | } | |
1538 | ||
6d61ef40 | 1539 | /* |
04046e1a KH |
1540 | * Visit the first child (need not be the first child as per the ordering |
1541 | * of the cgroup list, since we track last_scanned_child) of @mem and use | |
1542 | * that to reclaim free pages from. | |
1543 | */ | |
1544 | static struct mem_cgroup * | |
c0ff4b85 | 1545 | mem_cgroup_select_victim(struct mem_cgroup *root_memcg) |
04046e1a KH |
1546 | { |
1547 | struct mem_cgroup *ret = NULL; | |
1548 | struct cgroup_subsys_state *css; | |
1549 | int nextid, found; | |
1550 | ||
c0ff4b85 R |
1551 | if (!root_memcg->use_hierarchy) { |
1552 | css_get(&root_memcg->css); | |
1553 | ret = root_memcg; | |
04046e1a KH |
1554 | } |
1555 | ||
1556 | while (!ret) { | |
1557 | rcu_read_lock(); | |
c0ff4b85 R |
1558 | nextid = root_memcg->last_scanned_child + 1; |
1559 | css = css_get_next(&mem_cgroup_subsys, nextid, &root_memcg->css, | |
04046e1a KH |
1560 | &found); |
1561 | if (css && css_tryget(css)) | |
1562 | ret = container_of(css, struct mem_cgroup, css); | |
1563 | ||
1564 | rcu_read_unlock(); | |
1565 | /* Updates scanning parameter */ | |
04046e1a KH |
1566 | if (!css) { |
1567 | /* this means start scan from ID:1 */ | |
c0ff4b85 | 1568 | root_memcg->last_scanned_child = 0; |
04046e1a | 1569 | } else |
c0ff4b85 | 1570 | root_memcg->last_scanned_child = found; |
04046e1a KH |
1571 | } |
1572 | ||
1573 | return ret; | |
1574 | } | |
1575 | ||
4d0c066d KH |
1576 | /** |
1577 | * test_mem_cgroup_node_reclaimable | |
1578 | * @mem: the target memcg | |
1579 | * @nid: the node ID to be checked. | |
1580 | * @noswap : specify true here if the user wants flle only information. | |
1581 | * | |
1582 | * This function returns whether the specified memcg contains any | |
1583 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1584 | * pages in the node. | |
1585 | */ | |
c0ff4b85 | 1586 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1587 | int nid, bool noswap) |
1588 | { | |
c0ff4b85 | 1589 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1590 | return true; |
1591 | if (noswap || !total_swap_pages) | |
1592 | return false; | |
c0ff4b85 | 1593 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1594 | return true; |
1595 | return false; | |
1596 | ||
1597 | } | |
889976db YH |
1598 | #if MAX_NUMNODES > 1 |
1599 | ||
1600 | /* | |
1601 | * Always updating the nodemask is not very good - even if we have an empty | |
1602 | * list or the wrong list here, we can start from some node and traverse all | |
1603 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1604 | * | |
1605 | */ | |
c0ff4b85 | 1606 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1607 | { |
1608 | int nid; | |
453a9bf3 KH |
1609 | /* |
1610 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1611 | * pagein/pageout changes since the last update. | |
1612 | */ | |
c0ff4b85 | 1613 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1614 | return; |
c0ff4b85 | 1615 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1616 | return; |
1617 | ||
889976db | 1618 | /* make a nodemask where this memcg uses memory from */ |
c0ff4b85 | 1619 | memcg->scan_nodes = node_states[N_HIGH_MEMORY]; |
889976db YH |
1620 | |
1621 | for_each_node_mask(nid, node_states[N_HIGH_MEMORY]) { | |
1622 | ||
c0ff4b85 R |
1623 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1624 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1625 | } |
453a9bf3 | 1626 | |
c0ff4b85 R |
1627 | atomic_set(&memcg->numainfo_events, 0); |
1628 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1629 | } |
1630 | ||
1631 | /* | |
1632 | * Selecting a node where we start reclaim from. Because what we need is just | |
1633 | * reducing usage counter, start from anywhere is O,K. Considering | |
1634 | * memory reclaim from current node, there are pros. and cons. | |
1635 | * | |
1636 | * Freeing memory from current node means freeing memory from a node which | |
1637 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1638 | * hit limits, it will see a contention on a node. But freeing from remote | |
1639 | * node means more costs for memory reclaim because of memory latency. | |
1640 | * | |
1641 | * Now, we use round-robin. Better algorithm is welcomed. | |
1642 | */ | |
c0ff4b85 | 1643 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1644 | { |
1645 | int node; | |
1646 | ||
c0ff4b85 R |
1647 | mem_cgroup_may_update_nodemask(memcg); |
1648 | node = memcg->last_scanned_node; | |
889976db | 1649 | |
c0ff4b85 | 1650 | node = next_node(node, memcg->scan_nodes); |
889976db | 1651 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1652 | node = first_node(memcg->scan_nodes); |
889976db YH |
1653 | /* |
1654 | * We call this when we hit limit, not when pages are added to LRU. | |
1655 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1656 | * memcg is too small and all pages are not on LRU. In that case, | |
1657 | * we use curret node. | |
1658 | */ | |
1659 | if (unlikely(node == MAX_NUMNODES)) | |
1660 | node = numa_node_id(); | |
1661 | ||
c0ff4b85 | 1662 | memcg->last_scanned_node = node; |
889976db YH |
1663 | return node; |
1664 | } | |
1665 | ||
4d0c066d KH |
1666 | /* |
1667 | * Check all nodes whether it contains reclaimable pages or not. | |
1668 | * For quick scan, we make use of scan_nodes. This will allow us to skip | |
1669 | * unused nodes. But scan_nodes is lazily updated and may not cotain | |
1670 | * enough new information. We need to do double check. | |
1671 | */ | |
c0ff4b85 | 1672 | bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d KH |
1673 | { |
1674 | int nid; | |
1675 | ||
1676 | /* | |
1677 | * quick check...making use of scan_node. | |
1678 | * We can skip unused nodes. | |
1679 | */ | |
c0ff4b85 R |
1680 | if (!nodes_empty(memcg->scan_nodes)) { |
1681 | for (nid = first_node(memcg->scan_nodes); | |
4d0c066d | 1682 | nid < MAX_NUMNODES; |
c0ff4b85 | 1683 | nid = next_node(nid, memcg->scan_nodes)) { |
4d0c066d | 1684 | |
c0ff4b85 | 1685 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
1686 | return true; |
1687 | } | |
1688 | } | |
1689 | /* | |
1690 | * Check rest of nodes. | |
1691 | */ | |
1692 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
c0ff4b85 | 1693 | if (node_isset(nid, memcg->scan_nodes)) |
4d0c066d | 1694 | continue; |
c0ff4b85 | 1695 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
1696 | return true; |
1697 | } | |
1698 | return false; | |
1699 | } | |
1700 | ||
889976db | 1701 | #else |
c0ff4b85 | 1702 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1703 | { |
1704 | return 0; | |
1705 | } | |
4d0c066d | 1706 | |
c0ff4b85 | 1707 | bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d | 1708 | { |
c0ff4b85 | 1709 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); |
4d0c066d | 1710 | } |
889976db YH |
1711 | #endif |
1712 | ||
04046e1a KH |
1713 | /* |
1714 | * Scan the hierarchy if needed to reclaim memory. We remember the last child | |
1715 | * we reclaimed from, so that we don't end up penalizing one child extensively | |
1716 | * based on its position in the children list. | |
6d61ef40 | 1717 | * |
c0ff4b85 | 1718 | * root_memcg is the original ancestor that we've been reclaim from. |
04046e1a | 1719 | * |
c0ff4b85 | 1720 | * We give up and return to the caller when we visit root_memcg twice. |
04046e1a | 1721 | * (other groups can be removed while we're walking....) |
81d39c20 KH |
1722 | * |
1723 | * If shrink==true, for avoiding to free too much, this returns immedieately. | |
6d61ef40 | 1724 | */ |
c0ff4b85 | 1725 | static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_memcg, |
4e416953 | 1726 | struct zone *zone, |
75822b44 | 1727 | gfp_t gfp_mask, |
0ae5e89c YH |
1728 | unsigned long reclaim_options, |
1729 | unsigned long *total_scanned) | |
6d61ef40 | 1730 | { |
04046e1a KH |
1731 | struct mem_cgroup *victim; |
1732 | int ret, total = 0; | |
1733 | int loop = 0; | |
75822b44 BS |
1734 | bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP; |
1735 | bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK; | |
4e416953 | 1736 | bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT; |
9d11ea9f | 1737 | unsigned long excess; |
185efc0f | 1738 | unsigned long nr_scanned; |
9d11ea9f | 1739 | |
c0ff4b85 | 1740 | excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT; |
04046e1a | 1741 | |
22a668d7 | 1742 | /* If memsw_is_minimum==1, swap-out is of-no-use. */ |
c0ff4b85 | 1743 | if (!check_soft && !shrink && root_memcg->memsw_is_minimum) |
22a668d7 KH |
1744 | noswap = true; |
1745 | ||
4e416953 | 1746 | while (1) { |
c0ff4b85 R |
1747 | victim = mem_cgroup_select_victim(root_memcg); |
1748 | if (victim == root_memcg) { | |
04046e1a | 1749 | loop++; |
fbc29a25 KH |
1750 | /* |
1751 | * We are not draining per cpu cached charges during | |
1752 | * soft limit reclaim because global reclaim doesn't | |
1753 | * care about charges. It tries to free some memory and | |
1754 | * charges will not give any. | |
1755 | */ | |
1756 | if (!check_soft && loop >= 1) | |
c0ff4b85 | 1757 | drain_all_stock_async(root_memcg); |
4e416953 BS |
1758 | if (loop >= 2) { |
1759 | /* | |
1760 | * If we have not been able to reclaim | |
1761 | * anything, it might because there are | |
1762 | * no reclaimable pages under this hierarchy | |
1763 | */ | |
1764 | if (!check_soft || !total) { | |
1765 | css_put(&victim->css); | |
1766 | break; | |
1767 | } | |
1768 | /* | |
25985edc | 1769 | * We want to do more targeted reclaim. |
4e416953 BS |
1770 | * excess >> 2 is not to excessive so as to |
1771 | * reclaim too much, nor too less that we keep | |
1772 | * coming back to reclaim from this cgroup | |
1773 | */ | |
1774 | if (total >= (excess >> 2) || | |
1775 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) { | |
1776 | css_put(&victim->css); | |
1777 | break; | |
1778 | } | |
1779 | } | |
1780 | } | |
4d0c066d | 1781 | if (!mem_cgroup_reclaimable(victim, noswap)) { |
04046e1a KH |
1782 | /* this cgroup's local usage == 0 */ |
1783 | css_put(&victim->css); | |
6d61ef40 BS |
1784 | continue; |
1785 | } | |
04046e1a | 1786 | /* we use swappiness of local cgroup */ |
0ae5e89c | 1787 | if (check_soft) { |
4e416953 | 1788 | ret = mem_cgroup_shrink_node_zone(victim, gfp_mask, |
185efc0f JW |
1789 | noswap, zone, &nr_scanned); |
1790 | *total_scanned += nr_scanned; | |
0ae5e89c | 1791 | } else |
4e416953 | 1792 | ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, |
185efc0f | 1793 | noswap); |
04046e1a | 1794 | css_put(&victim->css); |
81d39c20 KH |
1795 | /* |
1796 | * At shrinking usage, we can't check we should stop here or | |
1797 | * reclaim more. It's depends on callers. last_scanned_child | |
1798 | * will work enough for keeping fairness under tree. | |
1799 | */ | |
1800 | if (shrink) | |
1801 | return ret; | |
04046e1a | 1802 | total += ret; |
4e416953 | 1803 | if (check_soft) { |
c0ff4b85 | 1804 | if (!res_counter_soft_limit_excess(&root_memcg->res)) |
4e416953 | 1805 | return total; |
c0ff4b85 | 1806 | } else if (mem_cgroup_margin(root_memcg)) |
4fd14ebf | 1807 | return total; |
6d61ef40 | 1808 | } |
04046e1a | 1809 | return total; |
6d61ef40 BS |
1810 | } |
1811 | ||
867578cb KH |
1812 | /* |
1813 | * Check OOM-Killer is already running under our hierarchy. | |
1814 | * If someone is running, return false. | |
1af8efe9 | 1815 | * Has to be called with memcg_oom_lock |
867578cb | 1816 | */ |
c0ff4b85 | 1817 | static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg) |
867578cb | 1818 | { |
79dfdacc MH |
1819 | struct mem_cgroup *iter, *failed = NULL; |
1820 | bool cond = true; | |
a636b327 | 1821 | |
c0ff4b85 | 1822 | for_each_mem_cgroup_tree_cond(iter, memcg, cond) { |
23751be0 | 1823 | if (iter->oom_lock) { |
79dfdacc MH |
1824 | /* |
1825 | * this subtree of our hierarchy is already locked | |
1826 | * so we cannot give a lock. | |
1827 | */ | |
79dfdacc MH |
1828 | failed = iter; |
1829 | cond = false; | |
23751be0 JW |
1830 | } else |
1831 | iter->oom_lock = true; | |
7d74b06f | 1832 | } |
867578cb | 1833 | |
79dfdacc | 1834 | if (!failed) |
23751be0 | 1835 | return true; |
79dfdacc MH |
1836 | |
1837 | /* | |
1838 | * OK, we failed to lock the whole subtree so we have to clean up | |
1839 | * what we set up to the failing subtree | |
1840 | */ | |
1841 | cond = true; | |
c0ff4b85 | 1842 | for_each_mem_cgroup_tree_cond(iter, memcg, cond) { |
79dfdacc MH |
1843 | if (iter == failed) { |
1844 | cond = false; | |
1845 | continue; | |
1846 | } | |
1847 | iter->oom_lock = false; | |
1848 | } | |
23751be0 | 1849 | return false; |
a636b327 | 1850 | } |
0b7f569e | 1851 | |
79dfdacc | 1852 | /* |
1af8efe9 | 1853 | * Has to be called with memcg_oom_lock |
79dfdacc | 1854 | */ |
c0ff4b85 | 1855 | static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1856 | { |
7d74b06f KH |
1857 | struct mem_cgroup *iter; |
1858 | ||
c0ff4b85 | 1859 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1860 | iter->oom_lock = false; |
1861 | return 0; | |
1862 | } | |
1863 | ||
c0ff4b85 | 1864 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1865 | { |
1866 | struct mem_cgroup *iter; | |
1867 | ||
c0ff4b85 | 1868 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1869 | atomic_inc(&iter->under_oom); |
1870 | } | |
1871 | ||
c0ff4b85 | 1872 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1873 | { |
1874 | struct mem_cgroup *iter; | |
1875 | ||
867578cb KH |
1876 | /* |
1877 | * When a new child is created while the hierarchy is under oom, | |
1878 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1879 | * atomic_add_unless() here. | |
1880 | */ | |
c0ff4b85 | 1881 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1882 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
1883 | } |
1884 | ||
1af8efe9 | 1885 | static DEFINE_SPINLOCK(memcg_oom_lock); |
867578cb KH |
1886 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1887 | ||
dc98df5a KH |
1888 | struct oom_wait_info { |
1889 | struct mem_cgroup *mem; | |
1890 | wait_queue_t wait; | |
1891 | }; | |
1892 | ||
1893 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1894 | unsigned mode, int sync, void *arg) | |
1895 | { | |
c0ff4b85 R |
1896 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg, |
1897 | *oom_wait_memcg; | |
dc98df5a KH |
1898 | struct oom_wait_info *oom_wait_info; |
1899 | ||
1900 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
c0ff4b85 | 1901 | oom_wait_memcg = oom_wait_info->mem; |
dc98df5a | 1902 | |
dc98df5a KH |
1903 | /* |
1904 | * Both of oom_wait_info->mem and wake_mem are stable under us. | |
1905 | * Then we can use css_is_ancestor without taking care of RCU. | |
1906 | */ | |
c0ff4b85 R |
1907 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
1908 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 1909 | return 0; |
dc98df5a KH |
1910 | return autoremove_wake_function(wait, mode, sync, arg); |
1911 | } | |
1912 | ||
c0ff4b85 | 1913 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 1914 | { |
c0ff4b85 R |
1915 | /* for filtering, pass "memcg" as argument. */ |
1916 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
1917 | } |
1918 | ||
c0ff4b85 | 1919 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1920 | { |
c0ff4b85 R |
1921 | if (memcg && atomic_read(&memcg->under_oom)) |
1922 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
1923 | } |
1924 | ||
867578cb KH |
1925 | /* |
1926 | * try to call OOM killer. returns false if we should exit memory-reclaim loop. | |
1927 | */ | |
c0ff4b85 | 1928 | bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask) |
0b7f569e | 1929 | { |
dc98df5a | 1930 | struct oom_wait_info owait; |
3c11ecf4 | 1931 | bool locked, need_to_kill; |
867578cb | 1932 | |
c0ff4b85 | 1933 | owait.mem = memcg; |
dc98df5a KH |
1934 | owait.wait.flags = 0; |
1935 | owait.wait.func = memcg_oom_wake_function; | |
1936 | owait.wait.private = current; | |
1937 | INIT_LIST_HEAD(&owait.wait.task_list); | |
3c11ecf4 | 1938 | need_to_kill = true; |
c0ff4b85 | 1939 | mem_cgroup_mark_under_oom(memcg); |
79dfdacc | 1940 | |
c0ff4b85 | 1941 | /* At first, try to OOM lock hierarchy under memcg.*/ |
1af8efe9 | 1942 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1943 | locked = mem_cgroup_oom_lock(memcg); |
867578cb KH |
1944 | /* |
1945 | * Even if signal_pending(), we can't quit charge() loop without | |
1946 | * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL | |
1947 | * under OOM is always welcomed, use TASK_KILLABLE here. | |
1948 | */ | |
3c11ecf4 | 1949 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
c0ff4b85 | 1950 | if (!locked || memcg->oom_kill_disable) |
3c11ecf4 KH |
1951 | need_to_kill = false; |
1952 | if (locked) | |
c0ff4b85 | 1953 | mem_cgroup_oom_notify(memcg); |
1af8efe9 | 1954 | spin_unlock(&memcg_oom_lock); |
867578cb | 1955 | |
3c11ecf4 KH |
1956 | if (need_to_kill) { |
1957 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
c0ff4b85 | 1958 | mem_cgroup_out_of_memory(memcg, mask); |
3c11ecf4 | 1959 | } else { |
867578cb | 1960 | schedule(); |
dc98df5a | 1961 | finish_wait(&memcg_oom_waitq, &owait.wait); |
867578cb | 1962 | } |
1af8efe9 | 1963 | spin_lock(&memcg_oom_lock); |
79dfdacc | 1964 | if (locked) |
c0ff4b85 R |
1965 | mem_cgroup_oom_unlock(memcg); |
1966 | memcg_wakeup_oom(memcg); | |
1af8efe9 | 1967 | spin_unlock(&memcg_oom_lock); |
867578cb | 1968 | |
c0ff4b85 | 1969 | mem_cgroup_unmark_under_oom(memcg); |
79dfdacc | 1970 | |
867578cb KH |
1971 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) |
1972 | return false; | |
1973 | /* Give chance to dying process */ | |
715a5ee8 | 1974 | schedule_timeout_uninterruptible(1); |
867578cb | 1975 | return true; |
0b7f569e KH |
1976 | } |
1977 | ||
d69b042f BS |
1978 | /* |
1979 | * Currently used to update mapped file statistics, but the routine can be | |
1980 | * generalized to update other statistics as well. | |
32047e2a KH |
1981 | * |
1982 | * Notes: Race condition | |
1983 | * | |
1984 | * We usually use page_cgroup_lock() for accessing page_cgroup member but | |
1985 | * it tends to be costly. But considering some conditions, we doesn't need | |
1986 | * to do so _always_. | |
1987 | * | |
1988 | * Considering "charge", lock_page_cgroup() is not required because all | |
1989 | * file-stat operations happen after a page is attached to radix-tree. There | |
1990 | * are no race with "charge". | |
1991 | * | |
1992 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
1993 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
1994 | * if there are race with "uncharge". Statistics itself is properly handled | |
1995 | * by flags. | |
1996 | * | |
1997 | * Considering "move", this is an only case we see a race. To make the race | |
1998 | * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are | |
1999 | * possibility of race condition. If there is, we take a lock. | |
d69b042f | 2000 | */ |
26174efd | 2001 | |
2a7106f2 GT |
2002 | void mem_cgroup_update_page_stat(struct page *page, |
2003 | enum mem_cgroup_page_stat_item idx, int val) | |
d69b042f | 2004 | { |
c0ff4b85 | 2005 | struct mem_cgroup *memcg; |
32047e2a KH |
2006 | struct page_cgroup *pc = lookup_page_cgroup(page); |
2007 | bool need_unlock = false; | |
dbd4ea78 | 2008 | unsigned long uninitialized_var(flags); |
d69b042f | 2009 | |
d69b042f BS |
2010 | if (unlikely(!pc)) |
2011 | return; | |
2012 | ||
32047e2a | 2013 | rcu_read_lock(); |
c0ff4b85 R |
2014 | memcg = pc->mem_cgroup; |
2015 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
32047e2a KH |
2016 | goto out; |
2017 | /* pc->mem_cgroup is unstable ? */ | |
c0ff4b85 | 2018 | if (unlikely(mem_cgroup_stealed(memcg)) || PageTransHuge(page)) { |
32047e2a | 2019 | /* take a lock against to access pc->mem_cgroup */ |
dbd4ea78 | 2020 | move_lock_page_cgroup(pc, &flags); |
32047e2a | 2021 | need_unlock = true; |
c0ff4b85 R |
2022 | memcg = pc->mem_cgroup; |
2023 | if (!memcg || !PageCgroupUsed(pc)) | |
32047e2a KH |
2024 | goto out; |
2025 | } | |
26174efd | 2026 | |
26174efd | 2027 | switch (idx) { |
2a7106f2 | 2028 | case MEMCG_NR_FILE_MAPPED: |
26174efd KH |
2029 | if (val > 0) |
2030 | SetPageCgroupFileMapped(pc); | |
2031 | else if (!page_mapped(page)) | |
0c270f8f | 2032 | ClearPageCgroupFileMapped(pc); |
2a7106f2 | 2033 | idx = MEM_CGROUP_STAT_FILE_MAPPED; |
26174efd KH |
2034 | break; |
2035 | default: | |
2036 | BUG(); | |
8725d541 | 2037 | } |
d69b042f | 2038 | |
c0ff4b85 | 2039 | this_cpu_add(memcg->stat->count[idx], val); |
2a7106f2 | 2040 | |
32047e2a KH |
2041 | out: |
2042 | if (unlikely(need_unlock)) | |
dbd4ea78 | 2043 | move_unlock_page_cgroup(pc, &flags); |
32047e2a KH |
2044 | rcu_read_unlock(); |
2045 | return; | |
d69b042f | 2046 | } |
2a7106f2 | 2047 | EXPORT_SYMBOL(mem_cgroup_update_page_stat); |
26174efd | 2048 | |
cdec2e42 KH |
2049 | /* |
2050 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2051 | * TODO: maybe necessary to use big numbers in big irons. | |
2052 | */ | |
7ec99d62 | 2053 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2054 | struct memcg_stock_pcp { |
2055 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2056 | unsigned int nr_pages; |
cdec2e42 | 2057 | struct work_struct work; |
26fe6168 KH |
2058 | unsigned long flags; |
2059 | #define FLUSHING_CACHED_CHARGE (0) | |
cdec2e42 KH |
2060 | }; |
2061 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2062 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 KH |
2063 | |
2064 | /* | |
11c9ea4e | 2065 | * Try to consume stocked charge on this cpu. If success, one page is consumed |
cdec2e42 KH |
2066 | * from local stock and true is returned. If the stock is 0 or charges from a |
2067 | * cgroup which is not current target, returns false. This stock will be | |
2068 | * refilled. | |
2069 | */ | |
c0ff4b85 | 2070 | static bool consume_stock(struct mem_cgroup *memcg) |
cdec2e42 KH |
2071 | { |
2072 | struct memcg_stock_pcp *stock; | |
2073 | bool ret = true; | |
2074 | ||
2075 | stock = &get_cpu_var(memcg_stock); | |
c0ff4b85 | 2076 | if (memcg == stock->cached && stock->nr_pages) |
11c9ea4e | 2077 | stock->nr_pages--; |
cdec2e42 KH |
2078 | else /* need to call res_counter_charge */ |
2079 | ret = false; | |
2080 | put_cpu_var(memcg_stock); | |
2081 | return ret; | |
2082 | } | |
2083 | ||
2084 | /* | |
2085 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
2086 | */ | |
2087 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2088 | { | |
2089 | struct mem_cgroup *old = stock->cached; | |
2090 | ||
11c9ea4e JW |
2091 | if (stock->nr_pages) { |
2092 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
2093 | ||
2094 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 2095 | if (do_swap_account) |
11c9ea4e JW |
2096 | res_counter_uncharge(&old->memsw, bytes); |
2097 | stock->nr_pages = 0; | |
cdec2e42 KH |
2098 | } |
2099 | stock->cached = NULL; | |
cdec2e42 KH |
2100 | } |
2101 | ||
2102 | /* | |
2103 | * This must be called under preempt disabled or must be called by | |
2104 | * a thread which is pinned to local cpu. | |
2105 | */ | |
2106 | static void drain_local_stock(struct work_struct *dummy) | |
2107 | { | |
2108 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
2109 | drain_stock(stock); | |
26fe6168 | 2110 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2111 | } |
2112 | ||
2113 | /* | |
2114 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 2115 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2116 | */ |
c0ff4b85 | 2117 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2118 | { |
2119 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2120 | ||
c0ff4b85 | 2121 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2122 | drain_stock(stock); |
c0ff4b85 | 2123 | stock->cached = memcg; |
cdec2e42 | 2124 | } |
11c9ea4e | 2125 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2126 | put_cpu_var(memcg_stock); |
2127 | } | |
2128 | ||
2129 | /* | |
c0ff4b85 | 2130 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
d38144b7 MH |
2131 | * of the hierarchy under it. sync flag says whether we should block |
2132 | * until the work is done. | |
cdec2e42 | 2133 | */ |
c0ff4b85 | 2134 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
cdec2e42 | 2135 | { |
26fe6168 | 2136 | int cpu, curcpu; |
d38144b7 | 2137 | |
cdec2e42 | 2138 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2139 | get_online_cpus(); |
5af12d0e | 2140 | curcpu = get_cpu(); |
cdec2e42 KH |
2141 | for_each_online_cpu(cpu) { |
2142 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2143 | struct mem_cgroup *memcg; |
26fe6168 | 2144 | |
c0ff4b85 R |
2145 | memcg = stock->cached; |
2146 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2147 | continue; |
c0ff4b85 | 2148 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2149 | continue; |
d1a05b69 MH |
2150 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2151 | if (cpu == curcpu) | |
2152 | drain_local_stock(&stock->work); | |
2153 | else | |
2154 | schedule_work_on(cpu, &stock->work); | |
2155 | } | |
cdec2e42 | 2156 | } |
5af12d0e | 2157 | put_cpu(); |
d38144b7 MH |
2158 | |
2159 | if (!sync) | |
2160 | goto out; | |
2161 | ||
2162 | for_each_online_cpu(cpu) { | |
2163 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2164 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2165 | flush_work(&stock->work); |
2166 | } | |
2167 | out: | |
cdec2e42 | 2168 | put_online_cpus(); |
d38144b7 MH |
2169 | } |
2170 | ||
2171 | /* | |
2172 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2173 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
2174 | * expects some charges will be back to res_counter later but cannot wait for | |
2175 | * it. | |
2176 | */ | |
c0ff4b85 | 2177 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
d38144b7 | 2178 | { |
9f50fad6 MH |
2179 | /* |
2180 | * If someone calls draining, avoid adding more kworker runs. | |
2181 | */ | |
2182 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2183 | return; | |
c0ff4b85 | 2184 | drain_all_stock(root_memcg, false); |
9f50fad6 | 2185 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2186 | } |
2187 | ||
2188 | /* This is a synchronous drain interface. */ | |
c0ff4b85 | 2189 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
cdec2e42 KH |
2190 | { |
2191 | /* called when force_empty is called */ | |
9f50fad6 | 2192 | mutex_lock(&percpu_charge_mutex); |
c0ff4b85 | 2193 | drain_all_stock(root_memcg, true); |
9f50fad6 | 2194 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2195 | } |
2196 | ||
711d3d2c KH |
2197 | /* |
2198 | * This function drains percpu counter value from DEAD cpu and | |
2199 | * move it to local cpu. Note that this function can be preempted. | |
2200 | */ | |
c0ff4b85 | 2201 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2202 | { |
2203 | int i; | |
2204 | ||
c0ff4b85 | 2205 | spin_lock(&memcg->pcp_counter_lock); |
711d3d2c | 2206 | for (i = 0; i < MEM_CGROUP_STAT_DATA; i++) { |
c0ff4b85 | 2207 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2208 | |
c0ff4b85 R |
2209 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2210 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2211 | } |
e9f8974f | 2212 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2213 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2214 | |
c0ff4b85 R |
2215 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2216 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2217 | } |
1489ebad | 2218 | /* need to clear ON_MOVE value, works as a kind of lock. */ |
c0ff4b85 R |
2219 | per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) = 0; |
2220 | spin_unlock(&memcg->pcp_counter_lock); | |
1489ebad KH |
2221 | } |
2222 | ||
c0ff4b85 | 2223 | static void synchronize_mem_cgroup_on_move(struct mem_cgroup *memcg, int cpu) |
1489ebad KH |
2224 | { |
2225 | int idx = MEM_CGROUP_ON_MOVE; | |
2226 | ||
c0ff4b85 R |
2227 | spin_lock(&memcg->pcp_counter_lock); |
2228 | per_cpu(memcg->stat->count[idx], cpu) = memcg->nocpu_base.count[idx]; | |
2229 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
2230 | } |
2231 | ||
2232 | static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb, | |
cdec2e42 KH |
2233 | unsigned long action, |
2234 | void *hcpu) | |
2235 | { | |
2236 | int cpu = (unsigned long)hcpu; | |
2237 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2238 | struct mem_cgroup *iter; |
cdec2e42 | 2239 | |
1489ebad KH |
2240 | if ((action == CPU_ONLINE)) { |
2241 | for_each_mem_cgroup_all(iter) | |
2242 | synchronize_mem_cgroup_on_move(iter, cpu); | |
2243 | return NOTIFY_OK; | |
2244 | } | |
2245 | ||
711d3d2c | 2246 | if ((action != CPU_DEAD) || action != CPU_DEAD_FROZEN) |
cdec2e42 | 2247 | return NOTIFY_OK; |
711d3d2c KH |
2248 | |
2249 | for_each_mem_cgroup_all(iter) | |
2250 | mem_cgroup_drain_pcp_counter(iter, cpu); | |
2251 | ||
cdec2e42 KH |
2252 | stock = &per_cpu(memcg_stock, cpu); |
2253 | drain_stock(stock); | |
2254 | return NOTIFY_OK; | |
2255 | } | |
2256 | ||
4b534334 KH |
2257 | |
2258 | /* See __mem_cgroup_try_charge() for details */ | |
2259 | enum { | |
2260 | CHARGE_OK, /* success */ | |
2261 | CHARGE_RETRY, /* need to retry but retry is not bad */ | |
2262 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ | |
2263 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ | |
2264 | CHARGE_OOM_DIE, /* the current is killed because of OOM */ | |
2265 | }; | |
2266 | ||
c0ff4b85 | 2267 | static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
7ec99d62 | 2268 | unsigned int nr_pages, bool oom_check) |
4b534334 | 2269 | { |
7ec99d62 | 2270 | unsigned long csize = nr_pages * PAGE_SIZE; |
4b534334 KH |
2271 | struct mem_cgroup *mem_over_limit; |
2272 | struct res_counter *fail_res; | |
2273 | unsigned long flags = 0; | |
2274 | int ret; | |
2275 | ||
c0ff4b85 | 2276 | ret = res_counter_charge(&memcg->res, csize, &fail_res); |
4b534334 KH |
2277 | |
2278 | if (likely(!ret)) { | |
2279 | if (!do_swap_account) | |
2280 | return CHARGE_OK; | |
c0ff4b85 | 2281 | ret = res_counter_charge(&memcg->memsw, csize, &fail_res); |
4b534334 KH |
2282 | if (likely(!ret)) |
2283 | return CHARGE_OK; | |
2284 | ||
c0ff4b85 | 2285 | res_counter_uncharge(&memcg->res, csize); |
4b534334 KH |
2286 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
2287 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
2288 | } else | |
2289 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
9221edb7 | 2290 | /* |
7ec99d62 JW |
2291 | * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch |
2292 | * of regular pages (CHARGE_BATCH), or a single regular page (1). | |
9221edb7 JW |
2293 | * |
2294 | * Never reclaim on behalf of optional batching, retry with a | |
2295 | * single page instead. | |
2296 | */ | |
7ec99d62 | 2297 | if (nr_pages == CHARGE_BATCH) |
4b534334 KH |
2298 | return CHARGE_RETRY; |
2299 | ||
2300 | if (!(gfp_mask & __GFP_WAIT)) | |
2301 | return CHARGE_WOULDBLOCK; | |
2302 | ||
2303 | ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL, | |
0ae5e89c | 2304 | gfp_mask, flags, NULL); |
7ec99d62 | 2305 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
19942822 | 2306 | return CHARGE_RETRY; |
4b534334 | 2307 | /* |
19942822 JW |
2308 | * Even though the limit is exceeded at this point, reclaim |
2309 | * may have been able to free some pages. Retry the charge | |
2310 | * before killing the task. | |
2311 | * | |
2312 | * Only for regular pages, though: huge pages are rather | |
2313 | * unlikely to succeed so close to the limit, and we fall back | |
2314 | * to regular pages anyway in case of failure. | |
4b534334 | 2315 | */ |
7ec99d62 | 2316 | if (nr_pages == 1 && ret) |
4b534334 KH |
2317 | return CHARGE_RETRY; |
2318 | ||
2319 | /* | |
2320 | * At task move, charge accounts can be doubly counted. So, it's | |
2321 | * better to wait until the end of task_move if something is going on. | |
2322 | */ | |
2323 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2324 | return CHARGE_RETRY; | |
2325 | ||
2326 | /* If we don't need to call oom-killer at el, return immediately */ | |
2327 | if (!oom_check) | |
2328 | return CHARGE_NOMEM; | |
2329 | /* check OOM */ | |
2330 | if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask)) | |
2331 | return CHARGE_OOM_DIE; | |
2332 | ||
2333 | return CHARGE_RETRY; | |
2334 | } | |
2335 | ||
f817ed48 KH |
2336 | /* |
2337 | * Unlike exported interface, "oom" parameter is added. if oom==true, | |
2338 | * oom-killer can be invoked. | |
8a9f3ccd | 2339 | */ |
f817ed48 | 2340 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
ec168510 | 2341 | gfp_t gfp_mask, |
7ec99d62 | 2342 | unsigned int nr_pages, |
c0ff4b85 | 2343 | struct mem_cgroup **ptr, |
7ec99d62 | 2344 | bool oom) |
8a9f3ccd | 2345 | { |
7ec99d62 | 2346 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
4b534334 | 2347 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
c0ff4b85 | 2348 | struct mem_cgroup *memcg = NULL; |
4b534334 | 2349 | int ret; |
a636b327 | 2350 | |
867578cb KH |
2351 | /* |
2352 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
2353 | * in system level. So, allow to go ahead dying process in addition to | |
2354 | * MEMDIE process. | |
2355 | */ | |
2356 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
2357 | || fatal_signal_pending(current))) | |
2358 | goto bypass; | |
a636b327 | 2359 | |
8a9f3ccd | 2360 | /* |
3be91277 HD |
2361 | * We always charge the cgroup the mm_struct belongs to. |
2362 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
2363 | * thread group leader migrates. It's possible that mm is not |
2364 | * set, if so charge the init_mm (happens for pagecache usage). | |
2365 | */ | |
c0ff4b85 | 2366 | if (!*ptr && !mm) |
f75ca962 KH |
2367 | goto bypass; |
2368 | again: | |
c0ff4b85 R |
2369 | if (*ptr) { /* css should be a valid one */ |
2370 | memcg = *ptr; | |
2371 | VM_BUG_ON(css_is_removed(&memcg->css)); | |
2372 | if (mem_cgroup_is_root(memcg)) | |
f75ca962 | 2373 | goto done; |
c0ff4b85 | 2374 | if (nr_pages == 1 && consume_stock(memcg)) |
f75ca962 | 2375 | goto done; |
c0ff4b85 | 2376 | css_get(&memcg->css); |
4b534334 | 2377 | } else { |
f75ca962 | 2378 | struct task_struct *p; |
54595fe2 | 2379 | |
f75ca962 KH |
2380 | rcu_read_lock(); |
2381 | p = rcu_dereference(mm->owner); | |
f75ca962 | 2382 | /* |
ebb76ce1 | 2383 | * Because we don't have task_lock(), "p" can exit. |
c0ff4b85 | 2384 | * In that case, "memcg" can point to root or p can be NULL with |
ebb76ce1 KH |
2385 | * race with swapoff. Then, we have small risk of mis-accouning. |
2386 | * But such kind of mis-account by race always happens because | |
2387 | * we don't have cgroup_mutex(). It's overkill and we allo that | |
2388 | * small race, here. | |
2389 | * (*) swapoff at el will charge against mm-struct not against | |
2390 | * task-struct. So, mm->owner can be NULL. | |
f75ca962 | 2391 | */ |
c0ff4b85 R |
2392 | memcg = mem_cgroup_from_task(p); |
2393 | if (!memcg || mem_cgroup_is_root(memcg)) { | |
f75ca962 KH |
2394 | rcu_read_unlock(); |
2395 | goto done; | |
2396 | } | |
c0ff4b85 | 2397 | if (nr_pages == 1 && consume_stock(memcg)) { |
f75ca962 KH |
2398 | /* |
2399 | * It seems dagerous to access memcg without css_get(). | |
2400 | * But considering how consume_stok works, it's not | |
2401 | * necessary. If consume_stock success, some charges | |
2402 | * from this memcg are cached on this cpu. So, we | |
2403 | * don't need to call css_get()/css_tryget() before | |
2404 | * calling consume_stock(). | |
2405 | */ | |
2406 | rcu_read_unlock(); | |
2407 | goto done; | |
2408 | } | |
2409 | /* after here, we may be blocked. we need to get refcnt */ | |
c0ff4b85 | 2410 | if (!css_tryget(&memcg->css)) { |
f75ca962 KH |
2411 | rcu_read_unlock(); |
2412 | goto again; | |
2413 | } | |
2414 | rcu_read_unlock(); | |
2415 | } | |
8a9f3ccd | 2416 | |
4b534334 KH |
2417 | do { |
2418 | bool oom_check; | |
7a81b88c | 2419 | |
4b534334 | 2420 | /* If killed, bypass charge */ |
f75ca962 | 2421 | if (fatal_signal_pending(current)) { |
c0ff4b85 | 2422 | css_put(&memcg->css); |
4b534334 | 2423 | goto bypass; |
f75ca962 | 2424 | } |
6d61ef40 | 2425 | |
4b534334 KH |
2426 | oom_check = false; |
2427 | if (oom && !nr_oom_retries) { | |
2428 | oom_check = true; | |
2429 | nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
cdec2e42 | 2430 | } |
66e1707b | 2431 | |
c0ff4b85 | 2432 | ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, oom_check); |
4b534334 KH |
2433 | switch (ret) { |
2434 | case CHARGE_OK: | |
2435 | break; | |
2436 | case CHARGE_RETRY: /* not in OOM situation but retry */ | |
7ec99d62 | 2437 | batch = nr_pages; |
c0ff4b85 R |
2438 | css_put(&memcg->css); |
2439 | memcg = NULL; | |
f75ca962 | 2440 | goto again; |
4b534334 | 2441 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
c0ff4b85 | 2442 | css_put(&memcg->css); |
4b534334 KH |
2443 | goto nomem; |
2444 | case CHARGE_NOMEM: /* OOM routine works */ | |
f75ca962 | 2445 | if (!oom) { |
c0ff4b85 | 2446 | css_put(&memcg->css); |
867578cb | 2447 | goto nomem; |
f75ca962 | 2448 | } |
4b534334 KH |
2449 | /* If oom, we never return -ENOMEM */ |
2450 | nr_oom_retries--; | |
2451 | break; | |
2452 | case CHARGE_OOM_DIE: /* Killed by OOM Killer */ | |
c0ff4b85 | 2453 | css_put(&memcg->css); |
867578cb | 2454 | goto bypass; |
66e1707b | 2455 | } |
4b534334 KH |
2456 | } while (ret != CHARGE_OK); |
2457 | ||
7ec99d62 | 2458 | if (batch > nr_pages) |
c0ff4b85 R |
2459 | refill_stock(memcg, batch - nr_pages); |
2460 | css_put(&memcg->css); | |
0c3e73e8 | 2461 | done: |
c0ff4b85 | 2462 | *ptr = memcg; |
7a81b88c KH |
2463 | return 0; |
2464 | nomem: | |
c0ff4b85 | 2465 | *ptr = NULL; |
7a81b88c | 2466 | return -ENOMEM; |
867578cb | 2467 | bypass: |
c0ff4b85 | 2468 | *ptr = NULL; |
867578cb | 2469 | return 0; |
7a81b88c | 2470 | } |
8a9f3ccd | 2471 | |
a3032a2c DN |
2472 | /* |
2473 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2474 | * This function is for that and do uncharge, put css's refcnt. | |
2475 | * gotten by try_charge(). | |
2476 | */ | |
c0ff4b85 | 2477 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg, |
e7018b8d | 2478 | unsigned int nr_pages) |
a3032a2c | 2479 | { |
c0ff4b85 | 2480 | if (!mem_cgroup_is_root(memcg)) { |
e7018b8d JW |
2481 | unsigned long bytes = nr_pages * PAGE_SIZE; |
2482 | ||
c0ff4b85 | 2483 | res_counter_uncharge(&memcg->res, bytes); |
a3032a2c | 2484 | if (do_swap_account) |
c0ff4b85 | 2485 | res_counter_uncharge(&memcg->memsw, bytes); |
a3032a2c | 2486 | } |
854ffa8d DN |
2487 | } |
2488 | ||
a3b2d692 KH |
2489 | /* |
2490 | * A helper function to get mem_cgroup from ID. must be called under | |
2491 | * rcu_read_lock(). The caller must check css_is_removed() or some if | |
2492 | * it's concern. (dropping refcnt from swap can be called against removed | |
2493 | * memcg.) | |
2494 | */ | |
2495 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2496 | { | |
2497 | struct cgroup_subsys_state *css; | |
2498 | ||
2499 | /* ID 0 is unused ID */ | |
2500 | if (!id) | |
2501 | return NULL; | |
2502 | css = css_lookup(&mem_cgroup_subsys, id); | |
2503 | if (!css) | |
2504 | return NULL; | |
2505 | return container_of(css, struct mem_cgroup, css); | |
2506 | } | |
2507 | ||
e42d9d5d | 2508 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2509 | { |
c0ff4b85 | 2510 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2511 | struct page_cgroup *pc; |
a3b2d692 | 2512 | unsigned short id; |
b5a84319 KH |
2513 | swp_entry_t ent; |
2514 | ||
3c776e64 DN |
2515 | VM_BUG_ON(!PageLocked(page)); |
2516 | ||
3c776e64 | 2517 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2518 | lock_page_cgroup(pc); |
a3b2d692 | 2519 | if (PageCgroupUsed(pc)) { |
c0ff4b85 R |
2520 | memcg = pc->mem_cgroup; |
2521 | if (memcg && !css_tryget(&memcg->css)) | |
2522 | memcg = NULL; | |
e42d9d5d | 2523 | } else if (PageSwapCache(page)) { |
3c776e64 | 2524 | ent.val = page_private(page); |
a3b2d692 KH |
2525 | id = lookup_swap_cgroup(ent); |
2526 | rcu_read_lock(); | |
c0ff4b85 R |
2527 | memcg = mem_cgroup_lookup(id); |
2528 | if (memcg && !css_tryget(&memcg->css)) | |
2529 | memcg = NULL; | |
a3b2d692 | 2530 | rcu_read_unlock(); |
3c776e64 | 2531 | } |
c0bd3f63 | 2532 | unlock_page_cgroup(pc); |
c0ff4b85 | 2533 | return memcg; |
b5a84319 KH |
2534 | } |
2535 | ||
c0ff4b85 | 2536 | static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg, |
5564e88b | 2537 | struct page *page, |
7ec99d62 | 2538 | unsigned int nr_pages, |
ca3e0214 | 2539 | struct page_cgroup *pc, |
7ec99d62 | 2540 | enum charge_type ctype) |
7a81b88c | 2541 | { |
ca3e0214 KH |
2542 | lock_page_cgroup(pc); |
2543 | if (unlikely(PageCgroupUsed(pc))) { | |
2544 | unlock_page_cgroup(pc); | |
c0ff4b85 | 2545 | __mem_cgroup_cancel_charge(memcg, nr_pages); |
ca3e0214 KH |
2546 | return; |
2547 | } | |
2548 | /* | |
2549 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2550 | * accessed by any other context at this point. | |
2551 | */ | |
c0ff4b85 | 2552 | pc->mem_cgroup = memcg; |
261fb61a KH |
2553 | /* |
2554 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
2555 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
2556 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
2557 | * before USED bit, we need memory barrier here. | |
2558 | * See mem_cgroup_add_lru_list(), etc. | |
2559 | */ | |
08e552c6 | 2560 | smp_wmb(); |
4b3bde4c BS |
2561 | switch (ctype) { |
2562 | case MEM_CGROUP_CHARGE_TYPE_CACHE: | |
2563 | case MEM_CGROUP_CHARGE_TYPE_SHMEM: | |
2564 | SetPageCgroupCache(pc); | |
2565 | SetPageCgroupUsed(pc); | |
2566 | break; | |
2567 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
2568 | ClearPageCgroupCache(pc); | |
2569 | SetPageCgroupUsed(pc); | |
2570 | break; | |
2571 | default: | |
2572 | break; | |
2573 | } | |
3be91277 | 2574 | |
c0ff4b85 | 2575 | mem_cgroup_charge_statistics(memcg, PageCgroupCache(pc), nr_pages); |
52d4b9ac | 2576 | unlock_page_cgroup(pc); |
430e4863 KH |
2577 | /* |
2578 | * "charge_statistics" updated event counter. Then, check it. | |
2579 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
2580 | * if they exceeds softlimit. | |
2581 | */ | |
c0ff4b85 | 2582 | memcg_check_events(memcg, page); |
7a81b88c | 2583 | } |
66e1707b | 2584 | |
ca3e0214 KH |
2585 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2586 | ||
2587 | #define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\ | |
2588 | (1 << PCG_ACCT_LRU) | (1 << PCG_MIGRATION)) | |
2589 | /* | |
2590 | * Because tail pages are not marked as "used", set it. We're under | |
2591 | * zone->lru_lock, 'splitting on pmd' and compund_lock. | |
2592 | */ | |
2593 | void mem_cgroup_split_huge_fixup(struct page *head, struct page *tail) | |
2594 | { | |
2595 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
2596 | struct page_cgroup *tail_pc = lookup_page_cgroup(tail); | |
2597 | unsigned long flags; | |
2598 | ||
3d37c4a9 KH |
2599 | if (mem_cgroup_disabled()) |
2600 | return; | |
ca3e0214 | 2601 | /* |
ece35ca8 | 2602 | * We have no races with charge/uncharge but will have races with |
ca3e0214 KH |
2603 | * page state accounting. |
2604 | */ | |
2605 | move_lock_page_cgroup(head_pc, &flags); | |
2606 | ||
2607 | tail_pc->mem_cgroup = head_pc->mem_cgroup; | |
2608 | smp_wmb(); /* see __commit_charge() */ | |
ece35ca8 KH |
2609 | if (PageCgroupAcctLRU(head_pc)) { |
2610 | enum lru_list lru; | |
2611 | struct mem_cgroup_per_zone *mz; | |
2612 | ||
2613 | /* | |
2614 | * LRU flags cannot be copied because we need to add tail | |
2615 | *.page to LRU by generic call and our hook will be called. | |
2616 | * We hold lru_lock, then, reduce counter directly. | |
2617 | */ | |
2618 | lru = page_lru(head); | |
97a6c37b | 2619 | mz = page_cgroup_zoneinfo(head_pc->mem_cgroup, head); |
ece35ca8 KH |
2620 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
2621 | } | |
ca3e0214 KH |
2622 | tail_pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
2623 | move_unlock_page_cgroup(head_pc, &flags); | |
2624 | } | |
2625 | #endif | |
2626 | ||
f817ed48 | 2627 | /** |
de3638d9 | 2628 | * mem_cgroup_move_account - move account of the page |
5564e88b | 2629 | * @page: the page |
7ec99d62 | 2630 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
2631 | * @pc: page_cgroup of the page. |
2632 | * @from: mem_cgroup which the page is moved from. | |
2633 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
854ffa8d | 2634 | * @uncharge: whether we should call uncharge and css_put against @from. |
f817ed48 KH |
2635 | * |
2636 | * The caller must confirm following. | |
08e552c6 | 2637 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 2638 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 2639 | * |
854ffa8d | 2640 | * This function doesn't do "charge" nor css_get to new cgroup. It should be |
25985edc | 2641 | * done by a caller(__mem_cgroup_try_charge would be useful). If @uncharge is |
854ffa8d DN |
2642 | * true, this function does "uncharge" from old cgroup, but it doesn't if |
2643 | * @uncharge is false, so a caller should do "uncharge". | |
f817ed48 | 2644 | */ |
7ec99d62 JW |
2645 | static int mem_cgroup_move_account(struct page *page, |
2646 | unsigned int nr_pages, | |
2647 | struct page_cgroup *pc, | |
2648 | struct mem_cgroup *from, | |
2649 | struct mem_cgroup *to, | |
2650 | bool uncharge) | |
f817ed48 | 2651 | { |
de3638d9 JW |
2652 | unsigned long flags; |
2653 | int ret; | |
987eba66 | 2654 | |
f817ed48 | 2655 | VM_BUG_ON(from == to); |
5564e88b | 2656 | VM_BUG_ON(PageLRU(page)); |
de3638d9 JW |
2657 | /* |
2658 | * The page is isolated from LRU. So, collapse function | |
2659 | * will not handle this page. But page splitting can happen. | |
2660 | * Do this check under compound_page_lock(). The caller should | |
2661 | * hold it. | |
2662 | */ | |
2663 | ret = -EBUSY; | |
7ec99d62 | 2664 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
2665 | goto out; |
2666 | ||
2667 | lock_page_cgroup(pc); | |
2668 | ||
2669 | ret = -EINVAL; | |
2670 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
2671 | goto unlock; | |
2672 | ||
2673 | move_lock_page_cgroup(pc, &flags); | |
f817ed48 | 2674 | |
8725d541 | 2675 | if (PageCgroupFileMapped(pc)) { |
c62b1a3b KH |
2676 | /* Update mapped_file data for mem_cgroup */ |
2677 | preempt_disable(); | |
2678 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2679 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2680 | preempt_enable(); | |
d69b042f | 2681 | } |
987eba66 | 2682 | mem_cgroup_charge_statistics(from, PageCgroupCache(pc), -nr_pages); |
854ffa8d DN |
2683 | if (uncharge) |
2684 | /* This is not "cancel", but cancel_charge does all we need. */ | |
e7018b8d | 2685 | __mem_cgroup_cancel_charge(from, nr_pages); |
d69b042f | 2686 | |
854ffa8d | 2687 | /* caller should have done css_get */ |
08e552c6 | 2688 | pc->mem_cgroup = to; |
987eba66 | 2689 | mem_cgroup_charge_statistics(to, PageCgroupCache(pc), nr_pages); |
88703267 KH |
2690 | /* |
2691 | * We charges against "to" which may not have any tasks. Then, "to" | |
2692 | * can be under rmdir(). But in current implementation, caller of | |
4ffef5fe | 2693 | * this function is just force_empty() and move charge, so it's |
25985edc | 2694 | * guaranteed that "to" is never removed. So, we don't check rmdir |
4ffef5fe | 2695 | * status here. |
88703267 | 2696 | */ |
de3638d9 JW |
2697 | move_unlock_page_cgroup(pc, &flags); |
2698 | ret = 0; | |
2699 | unlock: | |
57f9fd7d | 2700 | unlock_page_cgroup(pc); |
d2265e6f KH |
2701 | /* |
2702 | * check events | |
2703 | */ | |
5564e88b JW |
2704 | memcg_check_events(to, page); |
2705 | memcg_check_events(from, page); | |
de3638d9 | 2706 | out: |
f817ed48 KH |
2707 | return ret; |
2708 | } | |
2709 | ||
2710 | /* | |
2711 | * move charges to its parent. | |
2712 | */ | |
2713 | ||
5564e88b JW |
2714 | static int mem_cgroup_move_parent(struct page *page, |
2715 | struct page_cgroup *pc, | |
f817ed48 KH |
2716 | struct mem_cgroup *child, |
2717 | gfp_t gfp_mask) | |
2718 | { | |
2719 | struct cgroup *cg = child->css.cgroup; | |
2720 | struct cgroup *pcg = cg->parent; | |
2721 | struct mem_cgroup *parent; | |
7ec99d62 | 2722 | unsigned int nr_pages; |
4be4489f | 2723 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
2724 | int ret; |
2725 | ||
2726 | /* Is ROOT ? */ | |
2727 | if (!pcg) | |
2728 | return -EINVAL; | |
2729 | ||
57f9fd7d DN |
2730 | ret = -EBUSY; |
2731 | if (!get_page_unless_zero(page)) | |
2732 | goto out; | |
2733 | if (isolate_lru_page(page)) | |
2734 | goto put; | |
52dbb905 | 2735 | |
7ec99d62 | 2736 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 2737 | |
f817ed48 | 2738 | parent = mem_cgroup_from_cont(pcg); |
7ec99d62 | 2739 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, nr_pages, &parent, false); |
a636b327 | 2740 | if (ret || !parent) |
57f9fd7d | 2741 | goto put_back; |
f817ed48 | 2742 | |
7ec99d62 | 2743 | if (nr_pages > 1) |
987eba66 KH |
2744 | flags = compound_lock_irqsave(page); |
2745 | ||
7ec99d62 | 2746 | ret = mem_cgroup_move_account(page, nr_pages, pc, child, parent, true); |
854ffa8d | 2747 | if (ret) |
7ec99d62 | 2748 | __mem_cgroup_cancel_charge(parent, nr_pages); |
8dba474f | 2749 | |
7ec99d62 | 2750 | if (nr_pages > 1) |
987eba66 | 2751 | compound_unlock_irqrestore(page, flags); |
8dba474f | 2752 | put_back: |
08e552c6 | 2753 | putback_lru_page(page); |
57f9fd7d | 2754 | put: |
40d58138 | 2755 | put_page(page); |
57f9fd7d | 2756 | out: |
f817ed48 KH |
2757 | return ret; |
2758 | } | |
2759 | ||
7a81b88c KH |
2760 | /* |
2761 | * Charge the memory controller for page usage. | |
2762 | * Return | |
2763 | * 0 if the charge was successful | |
2764 | * < 0 if the cgroup is over its limit | |
2765 | */ | |
2766 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
73045c47 | 2767 | gfp_t gfp_mask, enum charge_type ctype) |
7a81b88c | 2768 | { |
c0ff4b85 | 2769 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 2770 | unsigned int nr_pages = 1; |
7a81b88c | 2771 | struct page_cgroup *pc; |
8493ae43 | 2772 | bool oom = true; |
7a81b88c | 2773 | int ret; |
ec168510 | 2774 | |
37c2ac78 | 2775 | if (PageTransHuge(page)) { |
7ec99d62 | 2776 | nr_pages <<= compound_order(page); |
37c2ac78 | 2777 | VM_BUG_ON(!PageTransHuge(page)); |
8493ae43 JW |
2778 | /* |
2779 | * Never OOM-kill a process for a huge page. The | |
2780 | * fault handler will fall back to regular pages. | |
2781 | */ | |
2782 | oom = false; | |
37c2ac78 | 2783 | } |
7a81b88c KH |
2784 | |
2785 | pc = lookup_page_cgroup(page); | |
af4a6621 | 2786 | BUG_ON(!pc); /* XXX: remove this and move pc lookup into commit */ |
7a81b88c | 2787 | |
c0ff4b85 R |
2788 | ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom); |
2789 | if (ret || !memcg) | |
7a81b88c KH |
2790 | return ret; |
2791 | ||
c0ff4b85 | 2792 | __mem_cgroup_commit_charge(memcg, page, nr_pages, pc, ctype); |
8a9f3ccd | 2793 | return 0; |
8a9f3ccd BS |
2794 | } |
2795 | ||
7a81b88c KH |
2796 | int mem_cgroup_newpage_charge(struct page *page, |
2797 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 2798 | { |
f8d66542 | 2799 | if (mem_cgroup_disabled()) |
cede86ac | 2800 | return 0; |
69029cd5 KH |
2801 | /* |
2802 | * If already mapped, we don't have to account. | |
2803 | * If page cache, page->mapping has address_space. | |
2804 | * But page->mapping may have out-of-use anon_vma pointer, | |
2805 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
2806 | * is NULL. | |
2807 | */ | |
2808 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
2809 | return 0; | |
2810 | if (unlikely(!mm)) | |
2811 | mm = &init_mm; | |
217bc319 | 2812 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
73045c47 | 2813 | MEM_CGROUP_CHARGE_TYPE_MAPPED); |
217bc319 KH |
2814 | } |
2815 | ||
83aae4c7 DN |
2816 | static void |
2817 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
2818 | enum charge_type ctype); | |
2819 | ||
5a6475a4 | 2820 | static void |
c0ff4b85 | 2821 | __mem_cgroup_commit_charge_lrucare(struct page *page, struct mem_cgroup *memcg, |
5a6475a4 KH |
2822 | enum charge_type ctype) |
2823 | { | |
2824 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
2825 | /* | |
2826 | * In some case, SwapCache, FUSE(splice_buf->radixtree), the page | |
2827 | * is already on LRU. It means the page may on some other page_cgroup's | |
2828 | * LRU. Take care of it. | |
2829 | */ | |
2830 | mem_cgroup_lru_del_before_commit(page); | |
c0ff4b85 | 2831 | __mem_cgroup_commit_charge(memcg, page, 1, pc, ctype); |
5a6475a4 KH |
2832 | mem_cgroup_lru_add_after_commit(page); |
2833 | return; | |
2834 | } | |
2835 | ||
e1a1cd59 BS |
2836 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
2837 | gfp_t gfp_mask) | |
8697d331 | 2838 | { |
c0ff4b85 | 2839 | struct mem_cgroup *memcg = NULL; |
b5a84319 KH |
2840 | int ret; |
2841 | ||
f8d66542 | 2842 | if (mem_cgroup_disabled()) |
cede86ac | 2843 | return 0; |
52d4b9ac KH |
2844 | if (PageCompound(page)) |
2845 | return 0; | |
accf163e | 2846 | |
73045c47 | 2847 | if (unlikely(!mm)) |
8697d331 | 2848 | mm = &init_mm; |
accf163e | 2849 | |
5a6475a4 | 2850 | if (page_is_file_cache(page)) { |
c0ff4b85 R |
2851 | ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, &memcg, true); |
2852 | if (ret || !memcg) | |
5a6475a4 | 2853 | return ret; |
b5a84319 | 2854 | |
5a6475a4 KH |
2855 | /* |
2856 | * FUSE reuses pages without going through the final | |
2857 | * put that would remove them from the LRU list, make | |
2858 | * sure that they get relinked properly. | |
2859 | */ | |
c0ff4b85 | 2860 | __mem_cgroup_commit_charge_lrucare(page, memcg, |
5a6475a4 KH |
2861 | MEM_CGROUP_CHARGE_TYPE_CACHE); |
2862 | return ret; | |
2863 | } | |
83aae4c7 DN |
2864 | /* shmem */ |
2865 | if (PageSwapCache(page)) { | |
c0ff4b85 | 2866 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &memcg); |
83aae4c7 | 2867 | if (!ret) |
c0ff4b85 | 2868 | __mem_cgroup_commit_charge_swapin(page, memcg, |
83aae4c7 DN |
2869 | MEM_CGROUP_CHARGE_TYPE_SHMEM); |
2870 | } else | |
2871 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, | |
73045c47 | 2872 | MEM_CGROUP_CHARGE_TYPE_SHMEM); |
b5a84319 | 2873 | |
b5a84319 | 2874 | return ret; |
e8589cc1 KH |
2875 | } |
2876 | ||
54595fe2 KH |
2877 | /* |
2878 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
2879 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 2880 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
2881 | * "commit()" or removed by "cancel()" |
2882 | */ | |
8c7c6e34 KH |
2883 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
2884 | struct page *page, | |
2885 | gfp_t mask, struct mem_cgroup **ptr) | |
2886 | { | |
c0ff4b85 | 2887 | struct mem_cgroup *memcg; |
54595fe2 | 2888 | int ret; |
8c7c6e34 | 2889 | |
56039efa KH |
2890 | *ptr = NULL; |
2891 | ||
f8d66542 | 2892 | if (mem_cgroup_disabled()) |
8c7c6e34 KH |
2893 | return 0; |
2894 | ||
2895 | if (!do_swap_account) | |
2896 | goto charge_cur_mm; | |
8c7c6e34 KH |
2897 | /* |
2898 | * A racing thread's fault, or swapoff, may have already updated | |
407f9c8b HD |
2899 | * the pte, and even removed page from swap cache: in those cases |
2900 | * do_swap_page()'s pte_same() test will fail; but there's also a | |
2901 | * KSM case which does need to charge the page. | |
8c7c6e34 KH |
2902 | */ |
2903 | if (!PageSwapCache(page)) | |
407f9c8b | 2904 | goto charge_cur_mm; |
c0ff4b85 R |
2905 | memcg = try_get_mem_cgroup_from_page(page); |
2906 | if (!memcg) | |
54595fe2 | 2907 | goto charge_cur_mm; |
c0ff4b85 | 2908 | *ptr = memcg; |
7ec99d62 | 2909 | ret = __mem_cgroup_try_charge(NULL, mask, 1, ptr, true); |
c0ff4b85 | 2910 | css_put(&memcg->css); |
54595fe2 | 2911 | return ret; |
8c7c6e34 KH |
2912 | charge_cur_mm: |
2913 | if (unlikely(!mm)) | |
2914 | mm = &init_mm; | |
7ec99d62 | 2915 | return __mem_cgroup_try_charge(mm, mask, 1, ptr, true); |
8c7c6e34 KH |
2916 | } |
2917 | ||
83aae4c7 DN |
2918 | static void |
2919 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
2920 | enum charge_type ctype) | |
7a81b88c | 2921 | { |
f8d66542 | 2922 | if (mem_cgroup_disabled()) |
7a81b88c KH |
2923 | return; |
2924 | if (!ptr) | |
2925 | return; | |
88703267 | 2926 | cgroup_exclude_rmdir(&ptr->css); |
5a6475a4 KH |
2927 | |
2928 | __mem_cgroup_commit_charge_lrucare(page, ptr, ctype); | |
8c7c6e34 KH |
2929 | /* |
2930 | * Now swap is on-memory. This means this page may be | |
2931 | * counted both as mem and swap....double count. | |
03f3c433 KH |
2932 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
2933 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
2934 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 2935 | */ |
03f3c433 | 2936 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 2937 | swp_entry_t ent = {.val = page_private(page)}; |
a3b2d692 | 2938 | unsigned short id; |
8c7c6e34 | 2939 | struct mem_cgroup *memcg; |
a3b2d692 KH |
2940 | |
2941 | id = swap_cgroup_record(ent, 0); | |
2942 | rcu_read_lock(); | |
2943 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 2944 | if (memcg) { |
a3b2d692 KH |
2945 | /* |
2946 | * This recorded memcg can be obsolete one. So, avoid | |
2947 | * calling css_tryget | |
2948 | */ | |
0c3e73e8 | 2949 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 2950 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 2951 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
2952 | mem_cgroup_put(memcg); |
2953 | } | |
a3b2d692 | 2954 | rcu_read_unlock(); |
8c7c6e34 | 2955 | } |
88703267 KH |
2956 | /* |
2957 | * At swapin, we may charge account against cgroup which has no tasks. | |
2958 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
2959 | * In that case, we need to call pre_destroy() again. check it here. | |
2960 | */ | |
2961 | cgroup_release_and_wakeup_rmdir(&ptr->css); | |
7a81b88c KH |
2962 | } |
2963 | ||
83aae4c7 DN |
2964 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) |
2965 | { | |
2966 | __mem_cgroup_commit_charge_swapin(page, ptr, | |
2967 | MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
2968 | } | |
2969 | ||
c0ff4b85 | 2970 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg) |
7a81b88c | 2971 | { |
f8d66542 | 2972 | if (mem_cgroup_disabled()) |
7a81b88c | 2973 | return; |
c0ff4b85 | 2974 | if (!memcg) |
7a81b88c | 2975 | return; |
c0ff4b85 | 2976 | __mem_cgroup_cancel_charge(memcg, 1); |
7a81b88c KH |
2977 | } |
2978 | ||
c0ff4b85 | 2979 | static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg, |
7ec99d62 JW |
2980 | unsigned int nr_pages, |
2981 | const enum charge_type ctype) | |
569b846d KH |
2982 | { |
2983 | struct memcg_batch_info *batch = NULL; | |
2984 | bool uncharge_memsw = true; | |
7ec99d62 | 2985 | |
569b846d KH |
2986 | /* If swapout, usage of swap doesn't decrease */ |
2987 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2988 | uncharge_memsw = false; | |
569b846d KH |
2989 | |
2990 | batch = ¤t->memcg_batch; | |
2991 | /* | |
2992 | * In usual, we do css_get() when we remember memcg pointer. | |
2993 | * But in this case, we keep res->usage until end of a series of | |
2994 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
2995 | */ | |
2996 | if (!batch->memcg) | |
c0ff4b85 | 2997 | batch->memcg = memcg; |
3c11ecf4 KH |
2998 | /* |
2999 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
25985edc | 3000 | * In those cases, all pages freed continuously can be expected to be in |
3c11ecf4 KH |
3001 | * the same cgroup and we have chance to coalesce uncharges. |
3002 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
3003 | * because we want to do uncharge as soon as possible. | |
3004 | */ | |
3005 | ||
3006 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
3007 | goto direct_uncharge; | |
3008 | ||
7ec99d62 | 3009 | if (nr_pages > 1) |
ec168510 AA |
3010 | goto direct_uncharge; |
3011 | ||
569b846d KH |
3012 | /* |
3013 | * In typical case, batch->memcg == mem. This means we can | |
3014 | * merge a series of uncharges to an uncharge of res_counter. | |
3015 | * If not, we uncharge res_counter ony by one. | |
3016 | */ | |
c0ff4b85 | 3017 | if (batch->memcg != memcg) |
569b846d KH |
3018 | goto direct_uncharge; |
3019 | /* remember freed charge and uncharge it later */ | |
7ffd4ca7 | 3020 | batch->nr_pages++; |
569b846d | 3021 | if (uncharge_memsw) |
7ffd4ca7 | 3022 | batch->memsw_nr_pages++; |
569b846d KH |
3023 | return; |
3024 | direct_uncharge: | |
c0ff4b85 | 3025 | res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE); |
569b846d | 3026 | if (uncharge_memsw) |
c0ff4b85 R |
3027 | res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE); |
3028 | if (unlikely(batch->memcg != memcg)) | |
3029 | memcg_oom_recover(memcg); | |
569b846d KH |
3030 | return; |
3031 | } | |
7a81b88c | 3032 | |
8a9f3ccd | 3033 | /* |
69029cd5 | 3034 | * uncharge if !page_mapped(page) |
8a9f3ccd | 3035 | */ |
8c7c6e34 | 3036 | static struct mem_cgroup * |
69029cd5 | 3037 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
8a9f3ccd | 3038 | { |
c0ff4b85 | 3039 | struct mem_cgroup *memcg = NULL; |
7ec99d62 JW |
3040 | unsigned int nr_pages = 1; |
3041 | struct page_cgroup *pc; | |
8a9f3ccd | 3042 | |
f8d66542 | 3043 | if (mem_cgroup_disabled()) |
8c7c6e34 | 3044 | return NULL; |
4077960e | 3045 | |
d13d1443 | 3046 | if (PageSwapCache(page)) |
8c7c6e34 | 3047 | return NULL; |
d13d1443 | 3048 | |
37c2ac78 | 3049 | if (PageTransHuge(page)) { |
7ec99d62 | 3050 | nr_pages <<= compound_order(page); |
37c2ac78 AA |
3051 | VM_BUG_ON(!PageTransHuge(page)); |
3052 | } | |
8697d331 | 3053 | /* |
3c541e14 | 3054 | * Check if our page_cgroup is valid |
8697d331 | 3055 | */ |
52d4b9ac KH |
3056 | pc = lookup_page_cgroup(page); |
3057 | if (unlikely(!pc || !PageCgroupUsed(pc))) | |
8c7c6e34 | 3058 | return NULL; |
b9c565d5 | 3059 | |
52d4b9ac | 3060 | lock_page_cgroup(pc); |
d13d1443 | 3061 | |
c0ff4b85 | 3062 | memcg = pc->mem_cgroup; |
8c7c6e34 | 3063 | |
d13d1443 KH |
3064 | if (!PageCgroupUsed(pc)) |
3065 | goto unlock_out; | |
3066 | ||
3067 | switch (ctype) { | |
3068 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
8a9478ca | 3069 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c AM |
3070 | /* See mem_cgroup_prepare_migration() */ |
3071 | if (page_mapped(page) || PageCgroupMigration(pc)) | |
d13d1443 KH |
3072 | goto unlock_out; |
3073 | break; | |
3074 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
3075 | if (!PageAnon(page)) { /* Shared memory */ | |
3076 | if (page->mapping && !page_is_file_cache(page)) | |
3077 | goto unlock_out; | |
3078 | } else if (page_mapped(page)) /* Anon */ | |
3079 | goto unlock_out; | |
3080 | break; | |
3081 | default: | |
3082 | break; | |
52d4b9ac | 3083 | } |
d13d1443 | 3084 | |
c0ff4b85 | 3085 | mem_cgroup_charge_statistics(memcg, PageCgroupCache(pc), -nr_pages); |
04046e1a | 3086 | |
52d4b9ac | 3087 | ClearPageCgroupUsed(pc); |
544122e5 KH |
3088 | /* |
3089 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
3090 | * freed from LRU. This is safe because uncharged page is expected not | |
3091 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
3092 | * special functions. | |
3093 | */ | |
b9c565d5 | 3094 | |
52d4b9ac | 3095 | unlock_page_cgroup(pc); |
f75ca962 | 3096 | /* |
c0ff4b85 | 3097 | * even after unlock, we have memcg->res.usage here and this memcg |
f75ca962 KH |
3098 | * will never be freed. |
3099 | */ | |
c0ff4b85 | 3100 | memcg_check_events(memcg, page); |
f75ca962 | 3101 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
c0ff4b85 R |
3102 | mem_cgroup_swap_statistics(memcg, true); |
3103 | mem_cgroup_get(memcg); | |
f75ca962 | 3104 | } |
c0ff4b85 R |
3105 | if (!mem_cgroup_is_root(memcg)) |
3106 | mem_cgroup_do_uncharge(memcg, nr_pages, ctype); | |
6d12e2d8 | 3107 | |
c0ff4b85 | 3108 | return memcg; |
d13d1443 KH |
3109 | |
3110 | unlock_out: | |
3111 | unlock_page_cgroup(pc); | |
8c7c6e34 | 3112 | return NULL; |
3c541e14 BS |
3113 | } |
3114 | ||
69029cd5 KH |
3115 | void mem_cgroup_uncharge_page(struct page *page) |
3116 | { | |
52d4b9ac KH |
3117 | /* early check. */ |
3118 | if (page_mapped(page)) | |
3119 | return; | |
3120 | if (page->mapping && !PageAnon(page)) | |
3121 | return; | |
69029cd5 KH |
3122 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
3123 | } | |
3124 | ||
3125 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
3126 | { | |
3127 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 3128 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
3129 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
3130 | } | |
3131 | ||
569b846d KH |
3132 | /* |
3133 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
3134 | * In that cases, pages are freed continuously and we can expect pages | |
3135 | * are in the same memcg. All these calls itself limits the number of | |
3136 | * pages freed at once, then uncharge_start/end() is called properly. | |
3137 | * This may be called prural(2) times in a context, | |
3138 | */ | |
3139 | ||
3140 | void mem_cgroup_uncharge_start(void) | |
3141 | { | |
3142 | current->memcg_batch.do_batch++; | |
3143 | /* We can do nest. */ | |
3144 | if (current->memcg_batch.do_batch == 1) { | |
3145 | current->memcg_batch.memcg = NULL; | |
7ffd4ca7 JW |
3146 | current->memcg_batch.nr_pages = 0; |
3147 | current->memcg_batch.memsw_nr_pages = 0; | |
569b846d KH |
3148 | } |
3149 | } | |
3150 | ||
3151 | void mem_cgroup_uncharge_end(void) | |
3152 | { | |
3153 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
3154 | ||
3155 | if (!batch->do_batch) | |
3156 | return; | |
3157 | ||
3158 | batch->do_batch--; | |
3159 | if (batch->do_batch) /* If stacked, do nothing. */ | |
3160 | return; | |
3161 | ||
3162 | if (!batch->memcg) | |
3163 | return; | |
3164 | /* | |
3165 | * This "batch->memcg" is valid without any css_get/put etc... | |
3166 | * bacause we hide charges behind us. | |
3167 | */ | |
7ffd4ca7 JW |
3168 | if (batch->nr_pages) |
3169 | res_counter_uncharge(&batch->memcg->res, | |
3170 | batch->nr_pages * PAGE_SIZE); | |
3171 | if (batch->memsw_nr_pages) | |
3172 | res_counter_uncharge(&batch->memcg->memsw, | |
3173 | batch->memsw_nr_pages * PAGE_SIZE); | |
3c11ecf4 | 3174 | memcg_oom_recover(batch->memcg); |
569b846d KH |
3175 | /* forget this pointer (for sanity check) */ |
3176 | batch->memcg = NULL; | |
3177 | } | |
3178 | ||
e767e056 | 3179 | #ifdef CONFIG_SWAP |
8c7c6e34 | 3180 | /* |
e767e056 | 3181 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
3182 | * memcg information is recorded to swap_cgroup of "ent" |
3183 | */ | |
8a9478ca KH |
3184 | void |
3185 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
3186 | { |
3187 | struct mem_cgroup *memcg; | |
8a9478ca KH |
3188 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
3189 | ||
3190 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
3191 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
3192 | ||
3193 | memcg = __mem_cgroup_uncharge_common(page, ctype); | |
8c7c6e34 | 3194 | |
f75ca962 KH |
3195 | /* |
3196 | * record memcg information, if swapout && memcg != NULL, | |
3197 | * mem_cgroup_get() was called in uncharge(). | |
3198 | */ | |
3199 | if (do_swap_account && swapout && memcg) | |
a3b2d692 | 3200 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 | 3201 | } |
e767e056 | 3202 | #endif |
8c7c6e34 KH |
3203 | |
3204 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
3205 | /* | |
3206 | * called from swap_entry_free(). remove record in swap_cgroup and | |
3207 | * uncharge "memsw" account. | |
3208 | */ | |
3209 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 3210 | { |
8c7c6e34 | 3211 | struct mem_cgroup *memcg; |
a3b2d692 | 3212 | unsigned short id; |
8c7c6e34 KH |
3213 | |
3214 | if (!do_swap_account) | |
3215 | return; | |
3216 | ||
a3b2d692 KH |
3217 | id = swap_cgroup_record(ent, 0); |
3218 | rcu_read_lock(); | |
3219 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 3220 | if (memcg) { |
a3b2d692 KH |
3221 | /* |
3222 | * We uncharge this because swap is freed. | |
3223 | * This memcg can be obsolete one. We avoid calling css_tryget | |
3224 | */ | |
0c3e73e8 | 3225 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 3226 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 3227 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
3228 | mem_cgroup_put(memcg); |
3229 | } | |
a3b2d692 | 3230 | rcu_read_unlock(); |
d13d1443 | 3231 | } |
02491447 DN |
3232 | |
3233 | /** | |
3234 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3235 | * @entry: swap entry to be moved | |
3236 | * @from: mem_cgroup which the entry is moved from | |
3237 | * @to: mem_cgroup which the entry is moved to | |
483c30b5 | 3238 | * @need_fixup: whether we should fixup res_counters and refcounts. |
02491447 DN |
3239 | * |
3240 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3241 | * as the mem_cgroup's id of @from. | |
3242 | * | |
3243 | * Returns 0 on success, -EINVAL on failure. | |
3244 | * | |
3245 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
3246 | * both res and memsw, and called css_get(). | |
3247 | */ | |
3248 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
483c30b5 | 3249 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
02491447 DN |
3250 | { |
3251 | unsigned short old_id, new_id; | |
3252 | ||
3253 | old_id = css_id(&from->css); | |
3254 | new_id = css_id(&to->css); | |
3255 | ||
3256 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 3257 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 3258 | mem_cgroup_swap_statistics(to, true); |
02491447 | 3259 | /* |
483c30b5 DN |
3260 | * This function is only called from task migration context now. |
3261 | * It postpones res_counter and refcount handling till the end | |
3262 | * of task migration(mem_cgroup_clear_mc()) for performance | |
3263 | * improvement. But we cannot postpone mem_cgroup_get(to) | |
3264 | * because if the process that has been moved to @to does | |
3265 | * swap-in, the refcount of @to might be decreased to 0. | |
02491447 | 3266 | */ |
02491447 | 3267 | mem_cgroup_get(to); |
483c30b5 DN |
3268 | if (need_fixup) { |
3269 | if (!mem_cgroup_is_root(from)) | |
3270 | res_counter_uncharge(&from->memsw, PAGE_SIZE); | |
3271 | mem_cgroup_put(from); | |
3272 | /* | |
3273 | * we charged both to->res and to->memsw, so we should | |
3274 | * uncharge to->res. | |
3275 | */ | |
3276 | if (!mem_cgroup_is_root(to)) | |
3277 | res_counter_uncharge(&to->res, PAGE_SIZE); | |
483c30b5 | 3278 | } |
02491447 DN |
3279 | return 0; |
3280 | } | |
3281 | return -EINVAL; | |
3282 | } | |
3283 | #else | |
3284 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
483c30b5 | 3285 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
02491447 DN |
3286 | { |
3287 | return -EINVAL; | |
3288 | } | |
8c7c6e34 | 3289 | #endif |
d13d1443 | 3290 | |
ae41be37 | 3291 | /* |
01b1ae63 KH |
3292 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
3293 | * page belongs to. | |
ae41be37 | 3294 | */ |
ac39cf8c | 3295 | int mem_cgroup_prepare_migration(struct page *page, |
ef6a3c63 | 3296 | struct page *newpage, struct mem_cgroup **ptr, gfp_t gfp_mask) |
ae41be37 | 3297 | { |
c0ff4b85 | 3298 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 3299 | struct page_cgroup *pc; |
ac39cf8c | 3300 | enum charge_type ctype; |
e8589cc1 | 3301 | int ret = 0; |
8869b8f6 | 3302 | |
56039efa KH |
3303 | *ptr = NULL; |
3304 | ||
ec168510 | 3305 | VM_BUG_ON(PageTransHuge(page)); |
f8d66542 | 3306 | if (mem_cgroup_disabled()) |
4077960e BS |
3307 | return 0; |
3308 | ||
52d4b9ac KH |
3309 | pc = lookup_page_cgroup(page); |
3310 | lock_page_cgroup(pc); | |
3311 | if (PageCgroupUsed(pc)) { | |
c0ff4b85 R |
3312 | memcg = pc->mem_cgroup; |
3313 | css_get(&memcg->css); | |
ac39cf8c AM |
3314 | /* |
3315 | * At migrating an anonymous page, its mapcount goes down | |
3316 | * to 0 and uncharge() will be called. But, even if it's fully | |
3317 | * unmapped, migration may fail and this page has to be | |
3318 | * charged again. We set MIGRATION flag here and delay uncharge | |
3319 | * until end_migration() is called | |
3320 | * | |
3321 | * Corner Case Thinking | |
3322 | * A) | |
3323 | * When the old page was mapped as Anon and it's unmap-and-freed | |
3324 | * while migration was ongoing. | |
3325 | * If unmap finds the old page, uncharge() of it will be delayed | |
3326 | * until end_migration(). If unmap finds a new page, it's | |
3327 | * uncharged when it make mapcount to be 1->0. If unmap code | |
3328 | * finds swap_migration_entry, the new page will not be mapped | |
3329 | * and end_migration() will find it(mapcount==0). | |
3330 | * | |
3331 | * B) | |
3332 | * When the old page was mapped but migraion fails, the kernel | |
3333 | * remaps it. A charge for it is kept by MIGRATION flag even | |
3334 | * if mapcount goes down to 0. We can do remap successfully | |
3335 | * without charging it again. | |
3336 | * | |
3337 | * C) | |
3338 | * The "old" page is under lock_page() until the end of | |
3339 | * migration, so, the old page itself will not be swapped-out. | |
3340 | * If the new page is swapped out before end_migraton, our | |
3341 | * hook to usual swap-out path will catch the event. | |
3342 | */ | |
3343 | if (PageAnon(page)) | |
3344 | SetPageCgroupMigration(pc); | |
e8589cc1 | 3345 | } |
52d4b9ac | 3346 | unlock_page_cgroup(pc); |
ac39cf8c AM |
3347 | /* |
3348 | * If the page is not charged at this point, | |
3349 | * we return here. | |
3350 | */ | |
c0ff4b85 | 3351 | if (!memcg) |
ac39cf8c | 3352 | return 0; |
01b1ae63 | 3353 | |
c0ff4b85 | 3354 | *ptr = memcg; |
7ec99d62 | 3355 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, 1, ptr, false); |
c0ff4b85 | 3356 | css_put(&memcg->css);/* drop extra refcnt */ |
ac39cf8c AM |
3357 | if (ret || *ptr == NULL) { |
3358 | if (PageAnon(page)) { | |
3359 | lock_page_cgroup(pc); | |
3360 | ClearPageCgroupMigration(pc); | |
3361 | unlock_page_cgroup(pc); | |
3362 | /* | |
3363 | * The old page may be fully unmapped while we kept it. | |
3364 | */ | |
3365 | mem_cgroup_uncharge_page(page); | |
3366 | } | |
3367 | return -ENOMEM; | |
e8589cc1 | 3368 | } |
ac39cf8c AM |
3369 | /* |
3370 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
3371 | * is called before end_migration, we can catch all events on this new | |
3372 | * page. In the case new page is migrated but not remapped, new page's | |
3373 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
3374 | */ | |
3375 | pc = lookup_page_cgroup(newpage); | |
3376 | if (PageAnon(page)) | |
3377 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
3378 | else if (page_is_file_cache(page)) | |
3379 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
3380 | else | |
3381 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
c0ff4b85 | 3382 | __mem_cgroup_commit_charge(memcg, page, 1, pc, ctype); |
e8589cc1 | 3383 | return ret; |
ae41be37 | 3384 | } |
8869b8f6 | 3385 | |
69029cd5 | 3386 | /* remove redundant charge if migration failed*/ |
c0ff4b85 | 3387 | void mem_cgroup_end_migration(struct mem_cgroup *memcg, |
50de1dd9 | 3388 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 3389 | { |
ac39cf8c | 3390 | struct page *used, *unused; |
01b1ae63 | 3391 | struct page_cgroup *pc; |
01b1ae63 | 3392 | |
c0ff4b85 | 3393 | if (!memcg) |
01b1ae63 | 3394 | return; |
ac39cf8c | 3395 | /* blocks rmdir() */ |
c0ff4b85 | 3396 | cgroup_exclude_rmdir(&memcg->css); |
50de1dd9 | 3397 | if (!migration_ok) { |
ac39cf8c AM |
3398 | used = oldpage; |
3399 | unused = newpage; | |
01b1ae63 | 3400 | } else { |
ac39cf8c | 3401 | used = newpage; |
01b1ae63 KH |
3402 | unused = oldpage; |
3403 | } | |
69029cd5 | 3404 | /* |
ac39cf8c AM |
3405 | * We disallowed uncharge of pages under migration because mapcount |
3406 | * of the page goes down to zero, temporarly. | |
3407 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 3408 | */ |
ac39cf8c AM |
3409 | pc = lookup_page_cgroup(oldpage); |
3410 | lock_page_cgroup(pc); | |
3411 | ClearPageCgroupMigration(pc); | |
3412 | unlock_page_cgroup(pc); | |
01b1ae63 | 3413 | |
ac39cf8c AM |
3414 | __mem_cgroup_uncharge_common(unused, MEM_CGROUP_CHARGE_TYPE_FORCE); |
3415 | ||
01b1ae63 | 3416 | /* |
ac39cf8c AM |
3417 | * If a page is a file cache, radix-tree replacement is very atomic |
3418 | * and we can skip this check. When it was an Anon page, its mapcount | |
3419 | * goes down to 0. But because we added MIGRATION flage, it's not | |
3420 | * uncharged yet. There are several case but page->mapcount check | |
3421 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
3422 | * check. (see prepare_charge() also) | |
69029cd5 | 3423 | */ |
ac39cf8c AM |
3424 | if (PageAnon(used)) |
3425 | mem_cgroup_uncharge_page(used); | |
88703267 | 3426 | /* |
ac39cf8c AM |
3427 | * At migration, we may charge account against cgroup which has no |
3428 | * tasks. | |
88703267 KH |
3429 | * So, rmdir()->pre_destroy() can be called while we do this charge. |
3430 | * In that case, we need to call pre_destroy() again. check it here. | |
3431 | */ | |
c0ff4b85 | 3432 | cgroup_release_and_wakeup_rmdir(&memcg->css); |
ae41be37 | 3433 | } |
78fb7466 | 3434 | |
f212ad7c DN |
3435 | #ifdef CONFIG_DEBUG_VM |
3436 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
3437 | { | |
3438 | struct page_cgroup *pc; | |
3439 | ||
3440 | pc = lookup_page_cgroup(page); | |
3441 | if (likely(pc) && PageCgroupUsed(pc)) | |
3442 | return pc; | |
3443 | return NULL; | |
3444 | } | |
3445 | ||
3446 | bool mem_cgroup_bad_page_check(struct page *page) | |
3447 | { | |
3448 | if (mem_cgroup_disabled()) | |
3449 | return false; | |
3450 | ||
3451 | return lookup_page_cgroup_used(page) != NULL; | |
3452 | } | |
3453 | ||
3454 | void mem_cgroup_print_bad_page(struct page *page) | |
3455 | { | |
3456 | struct page_cgroup *pc; | |
3457 | ||
3458 | pc = lookup_page_cgroup_used(page); | |
3459 | if (pc) { | |
3460 | int ret = -1; | |
3461 | char *path; | |
3462 | ||
3463 | printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p", | |
3464 | pc, pc->flags, pc->mem_cgroup); | |
3465 | ||
3466 | path = kmalloc(PATH_MAX, GFP_KERNEL); | |
3467 | if (path) { | |
3468 | rcu_read_lock(); | |
3469 | ret = cgroup_path(pc->mem_cgroup->css.cgroup, | |
3470 | path, PATH_MAX); | |
3471 | rcu_read_unlock(); | |
3472 | } | |
3473 | ||
3474 | printk(KERN_CONT "(%s)\n", | |
3475 | (ret < 0) ? "cannot get the path" : path); | |
3476 | kfree(path); | |
3477 | } | |
3478 | } | |
3479 | #endif | |
3480 | ||
8c7c6e34 KH |
3481 | static DEFINE_MUTEX(set_limit_mutex); |
3482 | ||
d38d2a75 | 3483 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 3484 | unsigned long long val) |
628f4235 | 3485 | { |
81d39c20 | 3486 | int retry_count; |
3c11ecf4 | 3487 | u64 memswlimit, memlimit; |
628f4235 | 3488 | int ret = 0; |
81d39c20 KH |
3489 | int children = mem_cgroup_count_children(memcg); |
3490 | u64 curusage, oldusage; | |
3c11ecf4 | 3491 | int enlarge; |
81d39c20 KH |
3492 | |
3493 | /* | |
3494 | * For keeping hierarchical_reclaim simple, how long we should retry | |
3495 | * is depends on callers. We set our retry-count to be function | |
3496 | * of # of children which we should visit in this loop. | |
3497 | */ | |
3498 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
3499 | ||
3500 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 3501 | |
3c11ecf4 | 3502 | enlarge = 0; |
8c7c6e34 | 3503 | while (retry_count) { |
628f4235 KH |
3504 | if (signal_pending(current)) { |
3505 | ret = -EINTR; | |
3506 | break; | |
3507 | } | |
8c7c6e34 KH |
3508 | /* |
3509 | * Rather than hide all in some function, I do this in | |
3510 | * open coded manner. You see what this really does. | |
c0ff4b85 | 3511 | * We have to guarantee memcg->res.limit < memcg->memsw.limit. |
8c7c6e34 KH |
3512 | */ |
3513 | mutex_lock(&set_limit_mutex); | |
3514 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3515 | if (memswlimit < val) { | |
3516 | ret = -EINVAL; | |
3517 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
3518 | break; |
3519 | } | |
3c11ecf4 KH |
3520 | |
3521 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3522 | if (memlimit < val) | |
3523 | enlarge = 1; | |
3524 | ||
8c7c6e34 | 3525 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
3526 | if (!ret) { |
3527 | if (memswlimit == val) | |
3528 | memcg->memsw_is_minimum = true; | |
3529 | else | |
3530 | memcg->memsw_is_minimum = false; | |
3531 | } | |
8c7c6e34 KH |
3532 | mutex_unlock(&set_limit_mutex); |
3533 | ||
3534 | if (!ret) | |
3535 | break; | |
3536 | ||
aa20d489 | 3537 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
0ae5e89c YH |
3538 | MEM_CGROUP_RECLAIM_SHRINK, |
3539 | NULL); | |
81d39c20 KH |
3540 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
3541 | /* Usage is reduced ? */ | |
3542 | if (curusage >= oldusage) | |
3543 | retry_count--; | |
3544 | else | |
3545 | oldusage = curusage; | |
8c7c6e34 | 3546 | } |
3c11ecf4 KH |
3547 | if (!ret && enlarge) |
3548 | memcg_oom_recover(memcg); | |
14797e23 | 3549 | |
8c7c6e34 KH |
3550 | return ret; |
3551 | } | |
3552 | ||
338c8431 LZ |
3553 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3554 | unsigned long long val) | |
8c7c6e34 | 3555 | { |
81d39c20 | 3556 | int retry_count; |
3c11ecf4 | 3557 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
3558 | int children = mem_cgroup_count_children(memcg); |
3559 | int ret = -EBUSY; | |
3c11ecf4 | 3560 | int enlarge = 0; |
8c7c6e34 | 3561 | |
81d39c20 KH |
3562 | /* see mem_cgroup_resize_res_limit */ |
3563 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
3564 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
3565 | while (retry_count) { |
3566 | if (signal_pending(current)) { | |
3567 | ret = -EINTR; | |
3568 | break; | |
3569 | } | |
3570 | /* | |
3571 | * Rather than hide all in some function, I do this in | |
3572 | * open coded manner. You see what this really does. | |
c0ff4b85 | 3573 | * We have to guarantee memcg->res.limit < memcg->memsw.limit. |
8c7c6e34 KH |
3574 | */ |
3575 | mutex_lock(&set_limit_mutex); | |
3576 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3577 | if (memlimit > val) { | |
3578 | ret = -EINVAL; | |
3579 | mutex_unlock(&set_limit_mutex); | |
3580 | break; | |
3581 | } | |
3c11ecf4 KH |
3582 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
3583 | if (memswlimit < val) | |
3584 | enlarge = 1; | |
8c7c6e34 | 3585 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
3586 | if (!ret) { |
3587 | if (memlimit == val) | |
3588 | memcg->memsw_is_minimum = true; | |
3589 | else | |
3590 | memcg->memsw_is_minimum = false; | |
3591 | } | |
8c7c6e34 KH |
3592 | mutex_unlock(&set_limit_mutex); |
3593 | ||
3594 | if (!ret) | |
3595 | break; | |
3596 | ||
4e416953 | 3597 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
75822b44 | 3598 | MEM_CGROUP_RECLAIM_NOSWAP | |
0ae5e89c YH |
3599 | MEM_CGROUP_RECLAIM_SHRINK, |
3600 | NULL); | |
8c7c6e34 | 3601 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 3602 | /* Usage is reduced ? */ |
8c7c6e34 | 3603 | if (curusage >= oldusage) |
628f4235 | 3604 | retry_count--; |
81d39c20 KH |
3605 | else |
3606 | oldusage = curusage; | |
628f4235 | 3607 | } |
3c11ecf4 KH |
3608 | if (!ret && enlarge) |
3609 | memcg_oom_recover(memcg); | |
628f4235 KH |
3610 | return ret; |
3611 | } | |
3612 | ||
4e416953 | 3613 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
0ae5e89c YH |
3614 | gfp_t gfp_mask, |
3615 | unsigned long *total_scanned) | |
4e416953 BS |
3616 | { |
3617 | unsigned long nr_reclaimed = 0; | |
3618 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3619 | unsigned long reclaimed; | |
3620 | int loop = 0; | |
3621 | struct mem_cgroup_tree_per_zone *mctz; | |
ef8745c1 | 3622 | unsigned long long excess; |
0ae5e89c | 3623 | unsigned long nr_scanned; |
4e416953 BS |
3624 | |
3625 | if (order > 0) | |
3626 | return 0; | |
3627 | ||
00918b6a | 3628 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); |
4e416953 BS |
3629 | /* |
3630 | * This loop can run a while, specially if mem_cgroup's continuously | |
3631 | * keep exceeding their soft limit and putting the system under | |
3632 | * pressure | |
3633 | */ | |
3634 | do { | |
3635 | if (next_mz) | |
3636 | mz = next_mz; | |
3637 | else | |
3638 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3639 | if (!mz) | |
3640 | break; | |
3641 | ||
0ae5e89c | 3642 | nr_scanned = 0; |
4e416953 BS |
3643 | reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone, |
3644 | gfp_mask, | |
0ae5e89c YH |
3645 | MEM_CGROUP_RECLAIM_SOFT, |
3646 | &nr_scanned); | |
4e416953 | 3647 | nr_reclaimed += reclaimed; |
0ae5e89c | 3648 | *total_scanned += nr_scanned; |
4e416953 BS |
3649 | spin_lock(&mctz->lock); |
3650 | ||
3651 | /* | |
3652 | * If we failed to reclaim anything from this memory cgroup | |
3653 | * it is time to move on to the next cgroup | |
3654 | */ | |
3655 | next_mz = NULL; | |
3656 | if (!reclaimed) { | |
3657 | do { | |
3658 | /* | |
3659 | * Loop until we find yet another one. | |
3660 | * | |
3661 | * By the time we get the soft_limit lock | |
3662 | * again, someone might have aded the | |
3663 | * group back on the RB tree. Iterate to | |
3664 | * make sure we get a different mem. | |
3665 | * mem_cgroup_largest_soft_limit_node returns | |
3666 | * NULL if no other cgroup is present on | |
3667 | * the tree | |
3668 | */ | |
3669 | next_mz = | |
3670 | __mem_cgroup_largest_soft_limit_node(mctz); | |
39cc98f1 | 3671 | if (next_mz == mz) |
4e416953 | 3672 | css_put(&next_mz->mem->css); |
39cc98f1 | 3673 | else /* next_mz == NULL or other memcg */ |
4e416953 BS |
3674 | break; |
3675 | } while (1); | |
3676 | } | |
4e416953 | 3677 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); |
ef8745c1 | 3678 | excess = res_counter_soft_limit_excess(&mz->mem->res); |
4e416953 BS |
3679 | /* |
3680 | * One school of thought says that we should not add | |
3681 | * back the node to the tree if reclaim returns 0. | |
3682 | * But our reclaim could return 0, simply because due | |
3683 | * to priority we are exposing a smaller subset of | |
3684 | * memory to reclaim from. Consider this as a longer | |
3685 | * term TODO. | |
3686 | */ | |
ef8745c1 KH |
3687 | /* If excess == 0, no tree ops */ |
3688 | __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess); | |
4e416953 BS |
3689 | spin_unlock(&mctz->lock); |
3690 | css_put(&mz->mem->css); | |
3691 | loop++; | |
3692 | /* | |
3693 | * Could not reclaim anything and there are no more | |
3694 | * mem cgroups to try or we seem to be looping without | |
3695 | * reclaiming anything. | |
3696 | */ | |
3697 | if (!nr_reclaimed && | |
3698 | (next_mz == NULL || | |
3699 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3700 | break; | |
3701 | } while (!nr_reclaimed); | |
3702 | if (next_mz) | |
3703 | css_put(&next_mz->mem->css); | |
3704 | return nr_reclaimed; | |
3705 | } | |
3706 | ||
cc847582 KH |
3707 | /* |
3708 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
3709 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
3710 | */ | |
c0ff4b85 | 3711 | static int mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
08e552c6 | 3712 | int node, int zid, enum lru_list lru) |
cc847582 | 3713 | { |
08e552c6 KH |
3714 | struct zone *zone; |
3715 | struct mem_cgroup_per_zone *mz; | |
f817ed48 | 3716 | struct page_cgroup *pc, *busy; |
08e552c6 | 3717 | unsigned long flags, loop; |
072c56c1 | 3718 | struct list_head *list; |
f817ed48 | 3719 | int ret = 0; |
072c56c1 | 3720 | |
08e552c6 | 3721 | zone = &NODE_DATA(node)->node_zones[zid]; |
c0ff4b85 | 3722 | mz = mem_cgroup_zoneinfo(memcg, node, zid); |
b69408e8 | 3723 | list = &mz->lists[lru]; |
cc847582 | 3724 | |
f817ed48 KH |
3725 | loop = MEM_CGROUP_ZSTAT(mz, lru); |
3726 | /* give some margin against EBUSY etc...*/ | |
3727 | loop += 256; | |
3728 | busy = NULL; | |
3729 | while (loop--) { | |
5564e88b JW |
3730 | struct page *page; |
3731 | ||
f817ed48 | 3732 | ret = 0; |
08e552c6 | 3733 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 3734 | if (list_empty(list)) { |
08e552c6 | 3735 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 3736 | break; |
f817ed48 KH |
3737 | } |
3738 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
3739 | if (busy == pc) { | |
3740 | list_move(&pc->lru, list); | |
648bcc77 | 3741 | busy = NULL; |
08e552c6 | 3742 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
3743 | continue; |
3744 | } | |
08e552c6 | 3745 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 3746 | |
6b3ae58e | 3747 | page = lookup_cgroup_page(pc); |
5564e88b | 3748 | |
c0ff4b85 | 3749 | ret = mem_cgroup_move_parent(page, pc, memcg, GFP_KERNEL); |
f817ed48 | 3750 | if (ret == -ENOMEM) |
52d4b9ac | 3751 | break; |
f817ed48 KH |
3752 | |
3753 | if (ret == -EBUSY || ret == -EINVAL) { | |
3754 | /* found lock contention or "pc" is obsolete. */ | |
3755 | busy = pc; | |
3756 | cond_resched(); | |
3757 | } else | |
3758 | busy = NULL; | |
cc847582 | 3759 | } |
08e552c6 | 3760 | |
f817ed48 KH |
3761 | if (!ret && !list_empty(list)) |
3762 | return -EBUSY; | |
3763 | return ret; | |
cc847582 KH |
3764 | } |
3765 | ||
3766 | /* | |
3767 | * make mem_cgroup's charge to be 0 if there is no task. | |
3768 | * This enables deleting this mem_cgroup. | |
3769 | */ | |
c0ff4b85 | 3770 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg, bool free_all) |
cc847582 | 3771 | { |
f817ed48 KH |
3772 | int ret; |
3773 | int node, zid, shrink; | |
3774 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c0ff4b85 | 3775 | struct cgroup *cgrp = memcg->css.cgroup; |
8869b8f6 | 3776 | |
c0ff4b85 | 3777 | css_get(&memcg->css); |
f817ed48 KH |
3778 | |
3779 | shrink = 0; | |
c1e862c1 KH |
3780 | /* should free all ? */ |
3781 | if (free_all) | |
3782 | goto try_to_free; | |
f817ed48 | 3783 | move_account: |
fce66477 | 3784 | do { |
f817ed48 | 3785 | ret = -EBUSY; |
c1e862c1 KH |
3786 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
3787 | goto out; | |
3788 | ret = -EINTR; | |
3789 | if (signal_pending(current)) | |
cc847582 | 3790 | goto out; |
52d4b9ac KH |
3791 | /* This is for making all *used* pages to be on LRU. */ |
3792 | lru_add_drain_all(); | |
c0ff4b85 | 3793 | drain_all_stock_sync(memcg); |
f817ed48 | 3794 | ret = 0; |
c0ff4b85 | 3795 | mem_cgroup_start_move(memcg); |
299b4eaa | 3796 | for_each_node_state(node, N_HIGH_MEMORY) { |
f817ed48 | 3797 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { |
b69408e8 | 3798 | enum lru_list l; |
f817ed48 | 3799 | for_each_lru(l) { |
c0ff4b85 | 3800 | ret = mem_cgroup_force_empty_list(memcg, |
08e552c6 | 3801 | node, zid, l); |
f817ed48 KH |
3802 | if (ret) |
3803 | break; | |
3804 | } | |
1ecaab2b | 3805 | } |
f817ed48 KH |
3806 | if (ret) |
3807 | break; | |
3808 | } | |
c0ff4b85 R |
3809 | mem_cgroup_end_move(memcg); |
3810 | memcg_oom_recover(memcg); | |
f817ed48 KH |
3811 | /* it seems parent cgroup doesn't have enough mem */ |
3812 | if (ret == -ENOMEM) | |
3813 | goto try_to_free; | |
52d4b9ac | 3814 | cond_resched(); |
fce66477 | 3815 | /* "ret" should also be checked to ensure all lists are empty. */ |
c0ff4b85 | 3816 | } while (memcg->res.usage > 0 || ret); |
cc847582 | 3817 | out: |
c0ff4b85 | 3818 | css_put(&memcg->css); |
cc847582 | 3819 | return ret; |
f817ed48 KH |
3820 | |
3821 | try_to_free: | |
c1e862c1 KH |
3822 | /* returns EBUSY if there is a task or if we come here twice. */ |
3823 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
f817ed48 KH |
3824 | ret = -EBUSY; |
3825 | goto out; | |
3826 | } | |
c1e862c1 KH |
3827 | /* we call try-to-free pages for make this cgroup empty */ |
3828 | lru_add_drain_all(); | |
f817ed48 KH |
3829 | /* try to free all pages in this cgroup */ |
3830 | shrink = 1; | |
c0ff4b85 | 3831 | while (nr_retries && memcg->res.usage > 0) { |
f817ed48 | 3832 | int progress; |
c1e862c1 KH |
3833 | |
3834 | if (signal_pending(current)) { | |
3835 | ret = -EINTR; | |
3836 | goto out; | |
3837 | } | |
c0ff4b85 | 3838 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, |
185efc0f | 3839 | false); |
c1e862c1 | 3840 | if (!progress) { |
f817ed48 | 3841 | nr_retries--; |
c1e862c1 | 3842 | /* maybe some writeback is necessary */ |
8aa7e847 | 3843 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3844 | } |
f817ed48 KH |
3845 | |
3846 | } | |
08e552c6 | 3847 | lru_add_drain(); |
f817ed48 | 3848 | /* try move_account...there may be some *locked* pages. */ |
fce66477 | 3849 | goto move_account; |
cc847582 KH |
3850 | } |
3851 | ||
c1e862c1 KH |
3852 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
3853 | { | |
3854 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
3855 | } | |
3856 | ||
3857 | ||
18f59ea7 BS |
3858 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
3859 | { | |
3860 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
3861 | } | |
3862 | ||
3863 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
3864 | u64 val) | |
3865 | { | |
3866 | int retval = 0; | |
c0ff4b85 | 3867 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
18f59ea7 | 3868 | struct cgroup *parent = cont->parent; |
c0ff4b85 | 3869 | struct mem_cgroup *parent_memcg = NULL; |
18f59ea7 BS |
3870 | |
3871 | if (parent) | |
c0ff4b85 | 3872 | parent_memcg = mem_cgroup_from_cont(parent); |
18f59ea7 BS |
3873 | |
3874 | cgroup_lock(); | |
3875 | /* | |
af901ca1 | 3876 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3877 | * in the child subtrees. If it is unset, then the change can |
3878 | * occur, provided the current cgroup has no children. | |
3879 | * | |
3880 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3881 | * set if there are no children. | |
3882 | */ | |
c0ff4b85 | 3883 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 BS |
3884 | (val == 1 || val == 0)) { |
3885 | if (list_empty(&cont->children)) | |
c0ff4b85 | 3886 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3887 | else |
3888 | retval = -EBUSY; | |
3889 | } else | |
3890 | retval = -EINVAL; | |
3891 | cgroup_unlock(); | |
3892 | ||
3893 | return retval; | |
3894 | } | |
3895 | ||
0c3e73e8 | 3896 | |
c0ff4b85 | 3897 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg, |
7a159cc9 | 3898 | enum mem_cgroup_stat_index idx) |
0c3e73e8 | 3899 | { |
7d74b06f | 3900 | struct mem_cgroup *iter; |
7a159cc9 | 3901 | long val = 0; |
0c3e73e8 | 3902 | |
7a159cc9 | 3903 | /* Per-cpu values can be negative, use a signed accumulator */ |
c0ff4b85 | 3904 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f KH |
3905 | val += mem_cgroup_read_stat(iter, idx); |
3906 | ||
3907 | if (val < 0) /* race ? */ | |
3908 | val = 0; | |
3909 | return val; | |
0c3e73e8 BS |
3910 | } |
3911 | ||
c0ff4b85 | 3912 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
104f3928 | 3913 | { |
7d74b06f | 3914 | u64 val; |
104f3928 | 3915 | |
c0ff4b85 | 3916 | if (!mem_cgroup_is_root(memcg)) { |
104f3928 | 3917 | if (!swap) |
65c64ce8 | 3918 | return res_counter_read_u64(&memcg->res, RES_USAGE); |
104f3928 | 3919 | else |
65c64ce8 | 3920 | return res_counter_read_u64(&memcg->memsw, RES_USAGE); |
104f3928 KS |
3921 | } |
3922 | ||
c0ff4b85 R |
3923 | val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE); |
3924 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS); | |
104f3928 | 3925 | |
7d74b06f | 3926 | if (swap) |
c0ff4b85 | 3927 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAPOUT); |
104f3928 KS |
3928 | |
3929 | return val << PAGE_SHIFT; | |
3930 | } | |
3931 | ||
2c3daa72 | 3932 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 3933 | { |
c0ff4b85 | 3934 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
104f3928 | 3935 | u64 val; |
8c7c6e34 KH |
3936 | int type, name; |
3937 | ||
3938 | type = MEMFILE_TYPE(cft->private); | |
3939 | name = MEMFILE_ATTR(cft->private); | |
3940 | switch (type) { | |
3941 | case _MEM: | |
104f3928 | 3942 | if (name == RES_USAGE) |
c0ff4b85 | 3943 | val = mem_cgroup_usage(memcg, false); |
104f3928 | 3944 | else |
c0ff4b85 | 3945 | val = res_counter_read_u64(&memcg->res, name); |
8c7c6e34 KH |
3946 | break; |
3947 | case _MEMSWAP: | |
104f3928 | 3948 | if (name == RES_USAGE) |
c0ff4b85 | 3949 | val = mem_cgroup_usage(memcg, true); |
104f3928 | 3950 | else |
c0ff4b85 | 3951 | val = res_counter_read_u64(&memcg->memsw, name); |
8c7c6e34 KH |
3952 | break; |
3953 | default: | |
3954 | BUG(); | |
3955 | break; | |
3956 | } | |
3957 | return val; | |
8cdea7c0 | 3958 | } |
628f4235 KH |
3959 | /* |
3960 | * The user of this function is... | |
3961 | * RES_LIMIT. | |
3962 | */ | |
856c13aa PM |
3963 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
3964 | const char *buffer) | |
8cdea7c0 | 3965 | { |
628f4235 | 3966 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 3967 | int type, name; |
628f4235 KH |
3968 | unsigned long long val; |
3969 | int ret; | |
3970 | ||
8c7c6e34 KH |
3971 | type = MEMFILE_TYPE(cft->private); |
3972 | name = MEMFILE_ATTR(cft->private); | |
3973 | switch (name) { | |
628f4235 | 3974 | case RES_LIMIT: |
4b3bde4c BS |
3975 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3976 | ret = -EINVAL; | |
3977 | break; | |
3978 | } | |
628f4235 KH |
3979 | /* This function does all necessary parse...reuse it */ |
3980 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
3981 | if (ret) |
3982 | break; | |
3983 | if (type == _MEM) | |
628f4235 | 3984 | ret = mem_cgroup_resize_limit(memcg, val); |
8c7c6e34 KH |
3985 | else |
3986 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
628f4235 | 3987 | break; |
296c81d8 BS |
3988 | case RES_SOFT_LIMIT: |
3989 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
3990 | if (ret) | |
3991 | break; | |
3992 | /* | |
3993 | * For memsw, soft limits are hard to implement in terms | |
3994 | * of semantics, for now, we support soft limits for | |
3995 | * control without swap | |
3996 | */ | |
3997 | if (type == _MEM) | |
3998 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
3999 | else | |
4000 | ret = -EINVAL; | |
4001 | break; | |
628f4235 KH |
4002 | default: |
4003 | ret = -EINVAL; /* should be BUG() ? */ | |
4004 | break; | |
4005 | } | |
4006 | return ret; | |
8cdea7c0 BS |
4007 | } |
4008 | ||
fee7b548 KH |
4009 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
4010 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
4011 | { | |
4012 | struct cgroup *cgroup; | |
4013 | unsigned long long min_limit, min_memsw_limit, tmp; | |
4014 | ||
4015 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4016 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
4017 | cgroup = memcg->css.cgroup; | |
4018 | if (!memcg->use_hierarchy) | |
4019 | goto out; | |
4020 | ||
4021 | while (cgroup->parent) { | |
4022 | cgroup = cgroup->parent; | |
4023 | memcg = mem_cgroup_from_cont(cgroup); | |
4024 | if (!memcg->use_hierarchy) | |
4025 | break; | |
4026 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4027 | min_limit = min(min_limit, tmp); | |
4028 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
4029 | min_memsw_limit = min(min_memsw_limit, tmp); | |
4030 | } | |
4031 | out: | |
4032 | *mem_limit = min_limit; | |
4033 | *memsw_limit = min_memsw_limit; | |
4034 | return; | |
4035 | } | |
4036 | ||
29f2a4da | 4037 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 | 4038 | { |
c0ff4b85 | 4039 | struct mem_cgroup *memcg; |
8c7c6e34 | 4040 | int type, name; |
c84872e1 | 4041 | |
c0ff4b85 | 4042 | memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 KH |
4043 | type = MEMFILE_TYPE(event); |
4044 | name = MEMFILE_ATTR(event); | |
4045 | switch (name) { | |
29f2a4da | 4046 | case RES_MAX_USAGE: |
8c7c6e34 | 4047 | if (type == _MEM) |
c0ff4b85 | 4048 | res_counter_reset_max(&memcg->res); |
8c7c6e34 | 4049 | else |
c0ff4b85 | 4050 | res_counter_reset_max(&memcg->memsw); |
29f2a4da PE |
4051 | break; |
4052 | case RES_FAILCNT: | |
8c7c6e34 | 4053 | if (type == _MEM) |
c0ff4b85 | 4054 | res_counter_reset_failcnt(&memcg->res); |
8c7c6e34 | 4055 | else |
c0ff4b85 | 4056 | res_counter_reset_failcnt(&memcg->memsw); |
29f2a4da PE |
4057 | break; |
4058 | } | |
f64c3f54 | 4059 | |
85cc59db | 4060 | return 0; |
c84872e1 PE |
4061 | } |
4062 | ||
7dc74be0 DN |
4063 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
4064 | struct cftype *cft) | |
4065 | { | |
4066 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; | |
4067 | } | |
4068 | ||
02491447 | 4069 | #ifdef CONFIG_MMU |
7dc74be0 DN |
4070 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
4071 | struct cftype *cft, u64 val) | |
4072 | { | |
c0ff4b85 | 4073 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
7dc74be0 DN |
4074 | |
4075 | if (val >= (1 << NR_MOVE_TYPE)) | |
4076 | return -EINVAL; | |
4077 | /* | |
4078 | * We check this value several times in both in can_attach() and | |
4079 | * attach(), so we need cgroup lock to prevent this value from being | |
4080 | * inconsistent. | |
4081 | */ | |
4082 | cgroup_lock(); | |
c0ff4b85 | 4083 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
4084 | cgroup_unlock(); |
4085 | ||
4086 | return 0; | |
4087 | } | |
02491447 DN |
4088 | #else |
4089 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, | |
4090 | struct cftype *cft, u64 val) | |
4091 | { | |
4092 | return -ENOSYS; | |
4093 | } | |
4094 | #endif | |
7dc74be0 | 4095 | |
14067bb3 KH |
4096 | |
4097 | /* For read statistics */ | |
4098 | enum { | |
4099 | MCS_CACHE, | |
4100 | MCS_RSS, | |
d8046582 | 4101 | MCS_FILE_MAPPED, |
14067bb3 KH |
4102 | MCS_PGPGIN, |
4103 | MCS_PGPGOUT, | |
1dd3a273 | 4104 | MCS_SWAP, |
456f998e YH |
4105 | MCS_PGFAULT, |
4106 | MCS_PGMAJFAULT, | |
14067bb3 KH |
4107 | MCS_INACTIVE_ANON, |
4108 | MCS_ACTIVE_ANON, | |
4109 | MCS_INACTIVE_FILE, | |
4110 | MCS_ACTIVE_FILE, | |
4111 | MCS_UNEVICTABLE, | |
4112 | NR_MCS_STAT, | |
4113 | }; | |
4114 | ||
4115 | struct mcs_total_stat { | |
4116 | s64 stat[NR_MCS_STAT]; | |
d2ceb9b7 KH |
4117 | }; |
4118 | ||
14067bb3 KH |
4119 | struct { |
4120 | char *local_name; | |
4121 | char *total_name; | |
4122 | } memcg_stat_strings[NR_MCS_STAT] = { | |
4123 | {"cache", "total_cache"}, | |
4124 | {"rss", "total_rss"}, | |
d69b042f | 4125 | {"mapped_file", "total_mapped_file"}, |
14067bb3 KH |
4126 | {"pgpgin", "total_pgpgin"}, |
4127 | {"pgpgout", "total_pgpgout"}, | |
1dd3a273 | 4128 | {"swap", "total_swap"}, |
456f998e YH |
4129 | {"pgfault", "total_pgfault"}, |
4130 | {"pgmajfault", "total_pgmajfault"}, | |
14067bb3 KH |
4131 | {"inactive_anon", "total_inactive_anon"}, |
4132 | {"active_anon", "total_active_anon"}, | |
4133 | {"inactive_file", "total_inactive_file"}, | |
4134 | {"active_file", "total_active_file"}, | |
4135 | {"unevictable", "total_unevictable"} | |
4136 | }; | |
4137 | ||
4138 | ||
7d74b06f | 4139 | static void |
c0ff4b85 | 4140 | mem_cgroup_get_local_stat(struct mem_cgroup *memcg, struct mcs_total_stat *s) |
14067bb3 | 4141 | { |
14067bb3 KH |
4142 | s64 val; |
4143 | ||
4144 | /* per cpu stat */ | |
c0ff4b85 | 4145 | val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_CACHE); |
14067bb3 | 4146 | s->stat[MCS_CACHE] += val * PAGE_SIZE; |
c0ff4b85 | 4147 | val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_RSS); |
14067bb3 | 4148 | s->stat[MCS_RSS] += val * PAGE_SIZE; |
c0ff4b85 | 4149 | val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED); |
d8046582 | 4150 | s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE; |
c0ff4b85 | 4151 | val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGIN); |
14067bb3 | 4152 | s->stat[MCS_PGPGIN] += val; |
c0ff4b85 | 4153 | val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGOUT); |
14067bb3 | 4154 | s->stat[MCS_PGPGOUT] += val; |
1dd3a273 | 4155 | if (do_swap_account) { |
c0ff4b85 | 4156 | val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_SWAPOUT); |
1dd3a273 DN |
4157 | s->stat[MCS_SWAP] += val * PAGE_SIZE; |
4158 | } | |
c0ff4b85 | 4159 | val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGFAULT); |
456f998e | 4160 | s->stat[MCS_PGFAULT] += val; |
c0ff4b85 | 4161 | val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGMAJFAULT); |
456f998e | 4162 | s->stat[MCS_PGMAJFAULT] += val; |
14067bb3 KH |
4163 | |
4164 | /* per zone stat */ | |
c0ff4b85 | 4165 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_ANON)); |
14067bb3 | 4166 | s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE; |
c0ff4b85 | 4167 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_ANON)); |
14067bb3 | 4168 | s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE; |
c0ff4b85 | 4169 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_FILE)); |
14067bb3 | 4170 | s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE; |
c0ff4b85 | 4171 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_FILE)); |
14067bb3 | 4172 | s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE; |
c0ff4b85 | 4173 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE)); |
14067bb3 | 4174 | s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE; |
14067bb3 KH |
4175 | } |
4176 | ||
4177 | static void | |
c0ff4b85 | 4178 | mem_cgroup_get_total_stat(struct mem_cgroup *memcg, struct mcs_total_stat *s) |
14067bb3 | 4179 | { |
7d74b06f KH |
4180 | struct mem_cgroup *iter; |
4181 | ||
c0ff4b85 | 4182 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4183 | mem_cgroup_get_local_stat(iter, s); |
14067bb3 KH |
4184 | } |
4185 | ||
406eb0c9 YH |
4186 | #ifdef CONFIG_NUMA |
4187 | static int mem_control_numa_stat_show(struct seq_file *m, void *arg) | |
4188 | { | |
4189 | int nid; | |
4190 | unsigned long total_nr, file_nr, anon_nr, unevictable_nr; | |
4191 | unsigned long node_nr; | |
4192 | struct cgroup *cont = m->private; | |
4193 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); | |
4194 | ||
bb2a0de9 | 4195 | total_nr = mem_cgroup_nr_lru_pages(mem_cont, LRU_ALL); |
406eb0c9 YH |
4196 | seq_printf(m, "total=%lu", total_nr); |
4197 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
bb2a0de9 | 4198 | node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid, LRU_ALL); |
406eb0c9 YH |
4199 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4200 | } | |
4201 | seq_putc(m, '\n'); | |
4202 | ||
bb2a0de9 | 4203 | file_nr = mem_cgroup_nr_lru_pages(mem_cont, LRU_ALL_FILE); |
406eb0c9 YH |
4204 | seq_printf(m, "file=%lu", file_nr); |
4205 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
bb2a0de9 KH |
4206 | node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid, |
4207 | LRU_ALL_FILE); | |
406eb0c9 YH |
4208 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4209 | } | |
4210 | seq_putc(m, '\n'); | |
4211 | ||
bb2a0de9 | 4212 | anon_nr = mem_cgroup_nr_lru_pages(mem_cont, LRU_ALL_ANON); |
406eb0c9 YH |
4213 | seq_printf(m, "anon=%lu", anon_nr); |
4214 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
bb2a0de9 KH |
4215 | node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid, |
4216 | LRU_ALL_ANON); | |
406eb0c9 YH |
4217 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4218 | } | |
4219 | seq_putc(m, '\n'); | |
4220 | ||
bb2a0de9 | 4221 | unevictable_nr = mem_cgroup_nr_lru_pages(mem_cont, BIT(LRU_UNEVICTABLE)); |
406eb0c9 YH |
4222 | seq_printf(m, "unevictable=%lu", unevictable_nr); |
4223 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
bb2a0de9 KH |
4224 | node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid, |
4225 | BIT(LRU_UNEVICTABLE)); | |
406eb0c9 YH |
4226 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4227 | } | |
4228 | seq_putc(m, '\n'); | |
4229 | return 0; | |
4230 | } | |
4231 | #endif /* CONFIG_NUMA */ | |
4232 | ||
c64745cf PM |
4233 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
4234 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 4235 | { |
d2ceb9b7 | 4236 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
14067bb3 | 4237 | struct mcs_total_stat mystat; |
d2ceb9b7 KH |
4238 | int i; |
4239 | ||
14067bb3 KH |
4240 | memset(&mystat, 0, sizeof(mystat)); |
4241 | mem_cgroup_get_local_stat(mem_cont, &mystat); | |
d2ceb9b7 | 4242 | |
406eb0c9 | 4243 | |
1dd3a273 DN |
4244 | for (i = 0; i < NR_MCS_STAT; i++) { |
4245 | if (i == MCS_SWAP && !do_swap_account) | |
4246 | continue; | |
14067bb3 | 4247 | cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]); |
1dd3a273 | 4248 | } |
7b854121 | 4249 | |
14067bb3 | 4250 | /* Hierarchical information */ |
fee7b548 KH |
4251 | { |
4252 | unsigned long long limit, memsw_limit; | |
4253 | memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit); | |
4254 | cb->fill(cb, "hierarchical_memory_limit", limit); | |
4255 | if (do_swap_account) | |
4256 | cb->fill(cb, "hierarchical_memsw_limit", memsw_limit); | |
4257 | } | |
7f016ee8 | 4258 | |
14067bb3 KH |
4259 | memset(&mystat, 0, sizeof(mystat)); |
4260 | mem_cgroup_get_total_stat(mem_cont, &mystat); | |
1dd3a273 DN |
4261 | for (i = 0; i < NR_MCS_STAT; i++) { |
4262 | if (i == MCS_SWAP && !do_swap_account) | |
4263 | continue; | |
14067bb3 | 4264 | cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]); |
1dd3a273 | 4265 | } |
14067bb3 | 4266 | |
7f016ee8 | 4267 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
4268 | { |
4269 | int nid, zid; | |
4270 | struct mem_cgroup_per_zone *mz; | |
4271 | unsigned long recent_rotated[2] = {0, 0}; | |
4272 | unsigned long recent_scanned[2] = {0, 0}; | |
4273 | ||
4274 | for_each_online_node(nid) | |
4275 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
4276 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); | |
4277 | ||
4278 | recent_rotated[0] += | |
4279 | mz->reclaim_stat.recent_rotated[0]; | |
4280 | recent_rotated[1] += | |
4281 | mz->reclaim_stat.recent_rotated[1]; | |
4282 | recent_scanned[0] += | |
4283 | mz->reclaim_stat.recent_scanned[0]; | |
4284 | recent_scanned[1] += | |
4285 | mz->reclaim_stat.recent_scanned[1]; | |
4286 | } | |
4287 | cb->fill(cb, "recent_rotated_anon", recent_rotated[0]); | |
4288 | cb->fill(cb, "recent_rotated_file", recent_rotated[1]); | |
4289 | cb->fill(cb, "recent_scanned_anon", recent_scanned[0]); | |
4290 | cb->fill(cb, "recent_scanned_file", recent_scanned[1]); | |
4291 | } | |
4292 | #endif | |
4293 | ||
d2ceb9b7 KH |
4294 | return 0; |
4295 | } | |
4296 | ||
a7885eb8 KM |
4297 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
4298 | { | |
4299 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4300 | ||
1f4c025b | 4301 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4302 | } |
4303 | ||
4304 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
4305 | u64 val) | |
4306 | { | |
4307 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4308 | struct mem_cgroup *parent; | |
068b38c1 | 4309 | |
a7885eb8 KM |
4310 | if (val > 100) |
4311 | return -EINVAL; | |
4312 | ||
4313 | if (cgrp->parent == NULL) | |
4314 | return -EINVAL; | |
4315 | ||
4316 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 LZ |
4317 | |
4318 | cgroup_lock(); | |
4319 | ||
a7885eb8 KM |
4320 | /* If under hierarchy, only empty-root can set this value */ |
4321 | if ((parent->use_hierarchy) || | |
068b38c1 LZ |
4322 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
4323 | cgroup_unlock(); | |
a7885eb8 | 4324 | return -EINVAL; |
068b38c1 | 4325 | } |
a7885eb8 | 4326 | |
a7885eb8 | 4327 | memcg->swappiness = val; |
a7885eb8 | 4328 | |
068b38c1 LZ |
4329 | cgroup_unlock(); |
4330 | ||
a7885eb8 KM |
4331 | return 0; |
4332 | } | |
4333 | ||
2e72b634 KS |
4334 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4335 | { | |
4336 | struct mem_cgroup_threshold_ary *t; | |
4337 | u64 usage; | |
4338 | int i; | |
4339 | ||
4340 | rcu_read_lock(); | |
4341 | if (!swap) | |
2c488db2 | 4342 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4343 | else |
2c488db2 | 4344 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4345 | |
4346 | if (!t) | |
4347 | goto unlock; | |
4348 | ||
4349 | usage = mem_cgroup_usage(memcg, swap); | |
4350 | ||
4351 | /* | |
4352 | * current_threshold points to threshold just below usage. | |
4353 | * If it's not true, a threshold was crossed after last | |
4354 | * call of __mem_cgroup_threshold(). | |
4355 | */ | |
5407a562 | 4356 | i = t->current_threshold; |
2e72b634 KS |
4357 | |
4358 | /* | |
4359 | * Iterate backward over array of thresholds starting from | |
4360 | * current_threshold and check if a threshold is crossed. | |
4361 | * If none of thresholds below usage is crossed, we read | |
4362 | * only one element of the array here. | |
4363 | */ | |
4364 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4365 | eventfd_signal(t->entries[i].eventfd, 1); | |
4366 | ||
4367 | /* i = current_threshold + 1 */ | |
4368 | i++; | |
4369 | ||
4370 | /* | |
4371 | * Iterate forward over array of thresholds starting from | |
4372 | * current_threshold+1 and check if a threshold is crossed. | |
4373 | * If none of thresholds above usage is crossed, we read | |
4374 | * only one element of the array here. | |
4375 | */ | |
4376 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4377 | eventfd_signal(t->entries[i].eventfd, 1); | |
4378 | ||
4379 | /* Update current_threshold */ | |
5407a562 | 4380 | t->current_threshold = i - 1; |
2e72b634 KS |
4381 | unlock: |
4382 | rcu_read_unlock(); | |
4383 | } | |
4384 | ||
4385 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4386 | { | |
ad4ca5f4 KS |
4387 | while (memcg) { |
4388 | __mem_cgroup_threshold(memcg, false); | |
4389 | if (do_swap_account) | |
4390 | __mem_cgroup_threshold(memcg, true); | |
4391 | ||
4392 | memcg = parent_mem_cgroup(memcg); | |
4393 | } | |
2e72b634 KS |
4394 | } |
4395 | ||
4396 | static int compare_thresholds(const void *a, const void *b) | |
4397 | { | |
4398 | const struct mem_cgroup_threshold *_a = a; | |
4399 | const struct mem_cgroup_threshold *_b = b; | |
4400 | ||
4401 | return _a->threshold - _b->threshold; | |
4402 | } | |
4403 | ||
c0ff4b85 | 4404 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4405 | { |
4406 | struct mem_cgroup_eventfd_list *ev; | |
4407 | ||
c0ff4b85 | 4408 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 KH |
4409 | eventfd_signal(ev->eventfd, 1); |
4410 | return 0; | |
4411 | } | |
4412 | ||
c0ff4b85 | 4413 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4414 | { |
7d74b06f KH |
4415 | struct mem_cgroup *iter; |
4416 | ||
c0ff4b85 | 4417 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4418 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4419 | } |
4420 | ||
4421 | static int mem_cgroup_usage_register_event(struct cgroup *cgrp, | |
4422 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
2e72b634 KS |
4423 | { |
4424 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4425 | struct mem_cgroup_thresholds *thresholds; |
4426 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
4427 | int type = MEMFILE_TYPE(cft->private); |
4428 | u64 threshold, usage; | |
2c488db2 | 4429 | int i, size, ret; |
2e72b634 KS |
4430 | |
4431 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
4432 | if (ret) | |
4433 | return ret; | |
4434 | ||
4435 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4436 | |
2e72b634 | 4437 | if (type == _MEM) |
2c488db2 | 4438 | thresholds = &memcg->thresholds; |
2e72b634 | 4439 | else if (type == _MEMSWAP) |
2c488db2 | 4440 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4441 | else |
4442 | BUG(); | |
4443 | ||
4444 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
4445 | ||
4446 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 4447 | if (thresholds->primary) |
2e72b634 KS |
4448 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4449 | ||
2c488db2 | 4450 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4451 | |
4452 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 4453 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 4454 | GFP_KERNEL); |
2c488db2 | 4455 | if (!new) { |
2e72b634 KS |
4456 | ret = -ENOMEM; |
4457 | goto unlock; | |
4458 | } | |
2c488db2 | 4459 | new->size = size; |
2e72b634 KS |
4460 | |
4461 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4462 | if (thresholds->primary) { |
4463 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4464 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4465 | } |
4466 | ||
2e72b634 | 4467 | /* Add new threshold */ |
2c488db2 KS |
4468 | new->entries[size - 1].eventfd = eventfd; |
4469 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4470 | |
4471 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4472 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4473 | compare_thresholds, NULL); |
4474 | ||
4475 | /* Find current threshold */ | |
2c488db2 | 4476 | new->current_threshold = -1; |
2e72b634 | 4477 | for (i = 0; i < size; i++) { |
2c488db2 | 4478 | if (new->entries[i].threshold < usage) { |
2e72b634 | 4479 | /* |
2c488db2 KS |
4480 | * new->current_threshold will not be used until |
4481 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4482 | * it here. |
4483 | */ | |
2c488db2 | 4484 | ++new->current_threshold; |
2e72b634 KS |
4485 | } |
4486 | } | |
4487 | ||
2c488db2 KS |
4488 | /* Free old spare buffer and save old primary buffer as spare */ |
4489 | kfree(thresholds->spare); | |
4490 | thresholds->spare = thresholds->primary; | |
4491 | ||
4492 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4493 | |
907860ed | 4494 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4495 | synchronize_rcu(); |
4496 | ||
2e72b634 KS |
4497 | unlock: |
4498 | mutex_unlock(&memcg->thresholds_lock); | |
4499 | ||
4500 | return ret; | |
4501 | } | |
4502 | ||
907860ed | 4503 | static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp, |
9490ff27 | 4504 | struct cftype *cft, struct eventfd_ctx *eventfd) |
2e72b634 KS |
4505 | { |
4506 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4507 | struct mem_cgroup_thresholds *thresholds; |
4508 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
4509 | int type = MEMFILE_TYPE(cft->private); |
4510 | u64 usage; | |
2c488db2 | 4511 | int i, j, size; |
2e72b634 KS |
4512 | |
4513 | mutex_lock(&memcg->thresholds_lock); | |
4514 | if (type == _MEM) | |
2c488db2 | 4515 | thresholds = &memcg->thresholds; |
2e72b634 | 4516 | else if (type == _MEMSWAP) |
2c488db2 | 4517 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4518 | else |
4519 | BUG(); | |
4520 | ||
4521 | /* | |
4522 | * Something went wrong if we trying to unregister a threshold | |
4523 | * if we don't have thresholds | |
4524 | */ | |
4525 | BUG_ON(!thresholds); | |
4526 | ||
4527 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
4528 | ||
4529 | /* Check if a threshold crossed before removing */ | |
4530 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4531 | ||
4532 | /* Calculate new number of threshold */ | |
2c488db2 KS |
4533 | size = 0; |
4534 | for (i = 0; i < thresholds->primary->size; i++) { | |
4535 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
4536 | size++; |
4537 | } | |
4538 | ||
2c488db2 | 4539 | new = thresholds->spare; |
907860ed | 4540 | |
2e72b634 KS |
4541 | /* Set thresholds array to NULL if we don't have thresholds */ |
4542 | if (!size) { | |
2c488db2 KS |
4543 | kfree(new); |
4544 | new = NULL; | |
907860ed | 4545 | goto swap_buffers; |
2e72b634 KS |
4546 | } |
4547 | ||
2c488db2 | 4548 | new->size = size; |
2e72b634 KS |
4549 | |
4550 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4551 | new->current_threshold = -1; |
4552 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4553 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4554 | continue; |
4555 | ||
2c488db2 KS |
4556 | new->entries[j] = thresholds->primary->entries[i]; |
4557 | if (new->entries[j].threshold < usage) { | |
2e72b634 | 4558 | /* |
2c488db2 | 4559 | * new->current_threshold will not be used |
2e72b634 KS |
4560 | * until rcu_assign_pointer(), so it's safe to increment |
4561 | * it here. | |
4562 | */ | |
2c488db2 | 4563 | ++new->current_threshold; |
2e72b634 KS |
4564 | } |
4565 | j++; | |
4566 | } | |
4567 | ||
907860ed | 4568 | swap_buffers: |
2c488db2 KS |
4569 | /* Swap primary and spare array */ |
4570 | thresholds->spare = thresholds->primary; | |
4571 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4572 | |
907860ed | 4573 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4574 | synchronize_rcu(); |
4575 | ||
2e72b634 | 4576 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4577 | } |
c1e862c1 | 4578 | |
9490ff27 KH |
4579 | static int mem_cgroup_oom_register_event(struct cgroup *cgrp, |
4580 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
4581 | { | |
4582 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4583 | struct mem_cgroup_eventfd_list *event; | |
4584 | int type = MEMFILE_TYPE(cft->private); | |
4585 | ||
4586 | BUG_ON(type != _OOM_TYPE); | |
4587 | event = kmalloc(sizeof(*event), GFP_KERNEL); | |
4588 | if (!event) | |
4589 | return -ENOMEM; | |
4590 | ||
1af8efe9 | 4591 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4592 | |
4593 | event->eventfd = eventfd; | |
4594 | list_add(&event->list, &memcg->oom_notify); | |
4595 | ||
4596 | /* already in OOM ? */ | |
79dfdacc | 4597 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 4598 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4599 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4600 | |
4601 | return 0; | |
4602 | } | |
4603 | ||
907860ed | 4604 | static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp, |
9490ff27 KH |
4605 | struct cftype *cft, struct eventfd_ctx *eventfd) |
4606 | { | |
c0ff4b85 | 4607 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
9490ff27 KH |
4608 | struct mem_cgroup_eventfd_list *ev, *tmp; |
4609 | int type = MEMFILE_TYPE(cft->private); | |
4610 | ||
4611 | BUG_ON(type != _OOM_TYPE); | |
4612 | ||
1af8efe9 | 4613 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4614 | |
c0ff4b85 | 4615 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4616 | if (ev->eventfd == eventfd) { |
4617 | list_del(&ev->list); | |
4618 | kfree(ev); | |
4619 | } | |
4620 | } | |
4621 | ||
1af8efe9 | 4622 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4623 | } |
4624 | ||
3c11ecf4 KH |
4625 | static int mem_cgroup_oom_control_read(struct cgroup *cgrp, |
4626 | struct cftype *cft, struct cgroup_map_cb *cb) | |
4627 | { | |
c0ff4b85 | 4628 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 | 4629 | |
c0ff4b85 | 4630 | cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable); |
3c11ecf4 | 4631 | |
c0ff4b85 | 4632 | if (atomic_read(&memcg->under_oom)) |
3c11ecf4 KH |
4633 | cb->fill(cb, "under_oom", 1); |
4634 | else | |
4635 | cb->fill(cb, "under_oom", 0); | |
4636 | return 0; | |
4637 | } | |
4638 | ||
3c11ecf4 KH |
4639 | static int mem_cgroup_oom_control_write(struct cgroup *cgrp, |
4640 | struct cftype *cft, u64 val) | |
4641 | { | |
c0ff4b85 | 4642 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 KH |
4643 | struct mem_cgroup *parent; |
4644 | ||
4645 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
4646 | if (!cgrp->parent || !((val == 0) || (val == 1))) | |
4647 | return -EINVAL; | |
4648 | ||
4649 | parent = mem_cgroup_from_cont(cgrp->parent); | |
4650 | ||
4651 | cgroup_lock(); | |
4652 | /* oom-kill-disable is a flag for subhierarchy. */ | |
4653 | if ((parent->use_hierarchy) || | |
c0ff4b85 | 4654 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
3c11ecf4 KH |
4655 | cgroup_unlock(); |
4656 | return -EINVAL; | |
4657 | } | |
c0ff4b85 | 4658 | memcg->oom_kill_disable = val; |
4d845ebf | 4659 | if (!val) |
c0ff4b85 | 4660 | memcg_oom_recover(memcg); |
3c11ecf4 KH |
4661 | cgroup_unlock(); |
4662 | return 0; | |
4663 | } | |
4664 | ||
406eb0c9 YH |
4665 | #ifdef CONFIG_NUMA |
4666 | static const struct file_operations mem_control_numa_stat_file_operations = { | |
4667 | .read = seq_read, | |
4668 | .llseek = seq_lseek, | |
4669 | .release = single_release, | |
4670 | }; | |
4671 | ||
4672 | static int mem_control_numa_stat_open(struct inode *unused, struct file *file) | |
4673 | { | |
4674 | struct cgroup *cont = file->f_dentry->d_parent->d_fsdata; | |
4675 | ||
4676 | file->f_op = &mem_control_numa_stat_file_operations; | |
4677 | return single_open(file, mem_control_numa_stat_show, cont); | |
4678 | } | |
4679 | #endif /* CONFIG_NUMA */ | |
4680 | ||
e5671dfa | 4681 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM |
e5671dfa GC |
4682 | static int register_kmem_files(struct cgroup *cont, struct cgroup_subsys *ss) |
4683 | { | |
d1a4c0b3 GC |
4684 | /* |
4685 | * Part of this would be better living in a separate allocation | |
4686 | * function, leaving us with just the cgroup tree population work. | |
4687 | * We, however, depend on state such as network's proto_list that | |
4688 | * is only initialized after cgroup creation. I found the less | |
4689 | * cumbersome way to deal with it to defer it all to populate time | |
4690 | */ | |
65c64ce8 | 4691 | return mem_cgroup_sockets_init(cont, ss); |
e5671dfa GC |
4692 | }; |
4693 | ||
d1a4c0b3 GC |
4694 | static void kmem_cgroup_destroy(struct cgroup_subsys *ss, |
4695 | struct cgroup *cont) | |
4696 | { | |
4697 | mem_cgroup_sockets_destroy(cont, ss); | |
4698 | } | |
e5671dfa GC |
4699 | #else |
4700 | static int register_kmem_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
4701 | { | |
4702 | return 0; | |
4703 | } | |
d1a4c0b3 GC |
4704 | |
4705 | static void kmem_cgroup_destroy(struct cgroup_subsys *ss, | |
4706 | struct cgroup *cont) | |
4707 | { | |
4708 | } | |
e5671dfa GC |
4709 | #endif |
4710 | ||
8cdea7c0 BS |
4711 | static struct cftype mem_cgroup_files[] = { |
4712 | { | |
0eea1030 | 4713 | .name = "usage_in_bytes", |
8c7c6e34 | 4714 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
2c3daa72 | 4715 | .read_u64 = mem_cgroup_read, |
9490ff27 KH |
4716 | .register_event = mem_cgroup_usage_register_event, |
4717 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8cdea7c0 | 4718 | }, |
c84872e1 PE |
4719 | { |
4720 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4721 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 4722 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
4723 | .read_u64 = mem_cgroup_read, |
4724 | }, | |
8cdea7c0 | 4725 | { |
0eea1030 | 4726 | .name = "limit_in_bytes", |
8c7c6e34 | 4727 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 4728 | .write_string = mem_cgroup_write, |
2c3daa72 | 4729 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 4730 | }, |
296c81d8 BS |
4731 | { |
4732 | .name = "soft_limit_in_bytes", | |
4733 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
4734 | .write_string = mem_cgroup_write, | |
4735 | .read_u64 = mem_cgroup_read, | |
4736 | }, | |
8cdea7c0 BS |
4737 | { |
4738 | .name = "failcnt", | |
8c7c6e34 | 4739 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 4740 | .trigger = mem_cgroup_reset, |
2c3daa72 | 4741 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 4742 | }, |
d2ceb9b7 KH |
4743 | { |
4744 | .name = "stat", | |
c64745cf | 4745 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 4746 | }, |
c1e862c1 KH |
4747 | { |
4748 | .name = "force_empty", | |
4749 | .trigger = mem_cgroup_force_empty_write, | |
4750 | }, | |
18f59ea7 BS |
4751 | { |
4752 | .name = "use_hierarchy", | |
4753 | .write_u64 = mem_cgroup_hierarchy_write, | |
4754 | .read_u64 = mem_cgroup_hierarchy_read, | |
4755 | }, | |
a7885eb8 KM |
4756 | { |
4757 | .name = "swappiness", | |
4758 | .read_u64 = mem_cgroup_swappiness_read, | |
4759 | .write_u64 = mem_cgroup_swappiness_write, | |
4760 | }, | |
7dc74be0 DN |
4761 | { |
4762 | .name = "move_charge_at_immigrate", | |
4763 | .read_u64 = mem_cgroup_move_charge_read, | |
4764 | .write_u64 = mem_cgroup_move_charge_write, | |
4765 | }, | |
9490ff27 KH |
4766 | { |
4767 | .name = "oom_control", | |
3c11ecf4 KH |
4768 | .read_map = mem_cgroup_oom_control_read, |
4769 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
4770 | .register_event = mem_cgroup_oom_register_event, |
4771 | .unregister_event = mem_cgroup_oom_unregister_event, | |
4772 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), | |
4773 | }, | |
406eb0c9 YH |
4774 | #ifdef CONFIG_NUMA |
4775 | { | |
4776 | .name = "numa_stat", | |
4777 | .open = mem_control_numa_stat_open, | |
89577127 | 4778 | .mode = S_IRUGO, |
406eb0c9 YH |
4779 | }, |
4780 | #endif | |
8cdea7c0 BS |
4781 | }; |
4782 | ||
8c7c6e34 KH |
4783 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
4784 | static struct cftype memsw_cgroup_files[] = { | |
4785 | { | |
4786 | .name = "memsw.usage_in_bytes", | |
4787 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
4788 | .read_u64 = mem_cgroup_read, | |
9490ff27 KH |
4789 | .register_event = mem_cgroup_usage_register_event, |
4790 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8c7c6e34 KH |
4791 | }, |
4792 | { | |
4793 | .name = "memsw.max_usage_in_bytes", | |
4794 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
4795 | .trigger = mem_cgroup_reset, | |
4796 | .read_u64 = mem_cgroup_read, | |
4797 | }, | |
4798 | { | |
4799 | .name = "memsw.limit_in_bytes", | |
4800 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
4801 | .write_string = mem_cgroup_write, | |
4802 | .read_u64 = mem_cgroup_read, | |
4803 | }, | |
4804 | { | |
4805 | .name = "memsw.failcnt", | |
4806 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
4807 | .trigger = mem_cgroup_reset, | |
4808 | .read_u64 = mem_cgroup_read, | |
4809 | }, | |
4810 | }; | |
4811 | ||
4812 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
4813 | { | |
4814 | if (!do_swap_account) | |
4815 | return 0; | |
4816 | return cgroup_add_files(cont, ss, memsw_cgroup_files, | |
4817 | ARRAY_SIZE(memsw_cgroup_files)); | |
4818 | }; | |
4819 | #else | |
4820 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
4821 | { | |
4822 | return 0; | |
4823 | } | |
4824 | #endif | |
4825 | ||
c0ff4b85 | 4826 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4827 | { |
4828 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4829 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 4830 | enum lru_list l; |
41e3355d | 4831 | int zone, tmp = node; |
1ecaab2b KH |
4832 | /* |
4833 | * This routine is called against possible nodes. | |
4834 | * But it's BUG to call kmalloc() against offline node. | |
4835 | * | |
4836 | * TODO: this routine can waste much memory for nodes which will | |
4837 | * never be onlined. It's better to use memory hotplug callback | |
4838 | * function. | |
4839 | */ | |
41e3355d KH |
4840 | if (!node_state(node, N_NORMAL_MEMORY)) |
4841 | tmp = -1; | |
17295c88 | 4842 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4843 | if (!pn) |
4844 | return 1; | |
1ecaab2b | 4845 | |
1ecaab2b KH |
4846 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4847 | mz = &pn->zoneinfo[zone]; | |
b69408e8 CL |
4848 | for_each_lru(l) |
4849 | INIT_LIST_HEAD(&mz->lists[l]); | |
f64c3f54 | 4850 | mz->usage_in_excess = 0; |
4e416953 | 4851 | mz->on_tree = false; |
c0ff4b85 | 4852 | mz->mem = memcg; |
1ecaab2b | 4853 | } |
0a619e58 | 4854 | memcg->info.nodeinfo[node] = pn; |
6d12e2d8 KH |
4855 | return 0; |
4856 | } | |
4857 | ||
c0ff4b85 | 4858 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4859 | { |
c0ff4b85 | 4860 | kfree(memcg->info.nodeinfo[node]); |
1ecaab2b KH |
4861 | } |
4862 | ||
33327948 KH |
4863 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4864 | { | |
4865 | struct mem_cgroup *mem; | |
c62b1a3b | 4866 | int size = sizeof(struct mem_cgroup); |
33327948 | 4867 | |
c62b1a3b | 4868 | /* Can be very big if MAX_NUMNODES is very big */ |
c8dad2bb | 4869 | if (size < PAGE_SIZE) |
17295c88 | 4870 | mem = kzalloc(size, GFP_KERNEL); |
33327948 | 4871 | else |
17295c88 | 4872 | mem = vzalloc(size); |
33327948 | 4873 | |
e7bbcdf3 DC |
4874 | if (!mem) |
4875 | return NULL; | |
4876 | ||
c62b1a3b | 4877 | mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
d2e61b8d DC |
4878 | if (!mem->stat) |
4879 | goto out_free; | |
711d3d2c | 4880 | spin_lock_init(&mem->pcp_counter_lock); |
33327948 | 4881 | return mem; |
d2e61b8d DC |
4882 | |
4883 | out_free: | |
4884 | if (size < PAGE_SIZE) | |
4885 | kfree(mem); | |
4886 | else | |
4887 | vfree(mem); | |
4888 | return NULL; | |
33327948 KH |
4889 | } |
4890 | ||
8c7c6e34 KH |
4891 | /* |
4892 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
4893 | * (scanning all at force_empty is too costly...) | |
4894 | * | |
4895 | * Instead of clearing all references at force_empty, we remember | |
4896 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4897 | * it goes down to 0. | |
4898 | * | |
8c7c6e34 KH |
4899 | * Removal of cgroup itself succeeds regardless of refs from swap. |
4900 | */ | |
4901 | ||
c0ff4b85 | 4902 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
33327948 | 4903 | { |
08e552c6 KH |
4904 | int node; |
4905 | ||
c0ff4b85 R |
4906 | mem_cgroup_remove_from_trees(memcg); |
4907 | free_css_id(&mem_cgroup_subsys, &memcg->css); | |
04046e1a | 4908 | |
08e552c6 | 4909 | for_each_node_state(node, N_POSSIBLE) |
c0ff4b85 | 4910 | free_mem_cgroup_per_zone_info(memcg, node); |
08e552c6 | 4911 | |
c0ff4b85 | 4912 | free_percpu(memcg->stat); |
c62b1a3b | 4913 | if (sizeof(struct mem_cgroup) < PAGE_SIZE) |
c0ff4b85 | 4914 | kfree(memcg); |
33327948 | 4915 | else |
c0ff4b85 | 4916 | vfree(memcg); |
33327948 KH |
4917 | } |
4918 | ||
c0ff4b85 | 4919 | static void mem_cgroup_get(struct mem_cgroup *memcg) |
8c7c6e34 | 4920 | { |
c0ff4b85 | 4921 | atomic_inc(&memcg->refcnt); |
8c7c6e34 KH |
4922 | } |
4923 | ||
c0ff4b85 | 4924 | static void __mem_cgroup_put(struct mem_cgroup *memcg, int count) |
8c7c6e34 | 4925 | { |
c0ff4b85 R |
4926 | if (atomic_sub_and_test(count, &memcg->refcnt)) { |
4927 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); | |
4928 | __mem_cgroup_free(memcg); | |
7bcc1bb1 DN |
4929 | if (parent) |
4930 | mem_cgroup_put(parent); | |
4931 | } | |
8c7c6e34 KH |
4932 | } |
4933 | ||
c0ff4b85 | 4934 | static void mem_cgroup_put(struct mem_cgroup *memcg) |
483c30b5 | 4935 | { |
c0ff4b85 | 4936 | __mem_cgroup_put(memcg, 1); |
483c30b5 DN |
4937 | } |
4938 | ||
7bcc1bb1 DN |
4939 | /* |
4940 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4941 | */ | |
e1aab161 | 4942 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 4943 | { |
c0ff4b85 | 4944 | if (!memcg->res.parent) |
7bcc1bb1 | 4945 | return NULL; |
c0ff4b85 | 4946 | return mem_cgroup_from_res_counter(memcg->res.parent, res); |
7bcc1bb1 | 4947 | } |
e1aab161 | 4948 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 4949 | |
c077719b KH |
4950 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
4951 | static void __init enable_swap_cgroup(void) | |
4952 | { | |
f8d66542 | 4953 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
4954 | do_swap_account = 1; |
4955 | } | |
4956 | #else | |
4957 | static void __init enable_swap_cgroup(void) | |
4958 | { | |
4959 | } | |
4960 | #endif | |
4961 | ||
f64c3f54 BS |
4962 | static int mem_cgroup_soft_limit_tree_init(void) |
4963 | { | |
4964 | struct mem_cgroup_tree_per_node *rtpn; | |
4965 | struct mem_cgroup_tree_per_zone *rtpz; | |
4966 | int tmp, node, zone; | |
4967 | ||
4968 | for_each_node_state(node, N_POSSIBLE) { | |
4969 | tmp = node; | |
4970 | if (!node_state(node, N_NORMAL_MEMORY)) | |
4971 | tmp = -1; | |
4972 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
4973 | if (!rtpn) | |
4974 | return 1; | |
4975 | ||
4976 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
4977 | ||
4978 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
4979 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
4980 | rtpz->rb_root = RB_ROOT; | |
4981 | spin_lock_init(&rtpz->lock); | |
4982 | } | |
4983 | } | |
4984 | return 0; | |
4985 | } | |
4986 | ||
0eb253e2 | 4987 | static struct cgroup_subsys_state * __ref |
8cdea7c0 BS |
4988 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) |
4989 | { | |
c0ff4b85 | 4990 | struct mem_cgroup *memcg, *parent; |
04046e1a | 4991 | long error = -ENOMEM; |
6d12e2d8 | 4992 | int node; |
8cdea7c0 | 4993 | |
c0ff4b85 R |
4994 | memcg = mem_cgroup_alloc(); |
4995 | if (!memcg) | |
04046e1a | 4996 | return ERR_PTR(error); |
78fb7466 | 4997 | |
6d12e2d8 | 4998 | for_each_node_state(node, N_POSSIBLE) |
c0ff4b85 | 4999 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 5000 | goto free_out; |
f64c3f54 | 5001 | |
c077719b | 5002 | /* root ? */ |
28dbc4b6 | 5003 | if (cont->parent == NULL) { |
cdec2e42 | 5004 | int cpu; |
c077719b | 5005 | enable_swap_cgroup(); |
28dbc4b6 | 5006 | parent = NULL; |
f64c3f54 BS |
5007 | if (mem_cgroup_soft_limit_tree_init()) |
5008 | goto free_out; | |
a41c58a6 | 5009 | root_mem_cgroup = memcg; |
cdec2e42 KH |
5010 | for_each_possible_cpu(cpu) { |
5011 | struct memcg_stock_pcp *stock = | |
5012 | &per_cpu(memcg_stock, cpu); | |
5013 | INIT_WORK(&stock->work, drain_local_stock); | |
5014 | } | |
711d3d2c | 5015 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
18f59ea7 | 5016 | } else { |
28dbc4b6 | 5017 | parent = mem_cgroup_from_cont(cont->parent); |
c0ff4b85 R |
5018 | memcg->use_hierarchy = parent->use_hierarchy; |
5019 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
18f59ea7 | 5020 | } |
28dbc4b6 | 5021 | |
18f59ea7 | 5022 | if (parent && parent->use_hierarchy) { |
c0ff4b85 R |
5023 | res_counter_init(&memcg->res, &parent->res); |
5024 | res_counter_init(&memcg->memsw, &parent->memsw); | |
7bcc1bb1 DN |
5025 | /* |
5026 | * We increment refcnt of the parent to ensure that we can | |
5027 | * safely access it on res_counter_charge/uncharge. | |
5028 | * This refcnt will be decremented when freeing this | |
5029 | * mem_cgroup(see mem_cgroup_put). | |
5030 | */ | |
5031 | mem_cgroup_get(parent); | |
18f59ea7 | 5032 | } else { |
c0ff4b85 R |
5033 | res_counter_init(&memcg->res, NULL); |
5034 | res_counter_init(&memcg->memsw, NULL); | |
18f59ea7 | 5035 | } |
c0ff4b85 R |
5036 | memcg->last_scanned_child = 0; |
5037 | memcg->last_scanned_node = MAX_NUMNODES; | |
5038 | INIT_LIST_HEAD(&memcg->oom_notify); | |
6d61ef40 | 5039 | |
a7885eb8 | 5040 | if (parent) |
c0ff4b85 R |
5041 | memcg->swappiness = mem_cgroup_swappiness(parent); |
5042 | atomic_set(&memcg->refcnt, 1); | |
5043 | memcg->move_charge_at_immigrate = 0; | |
5044 | mutex_init(&memcg->thresholds_lock); | |
5045 | return &memcg->css; | |
6d12e2d8 | 5046 | free_out: |
c0ff4b85 | 5047 | __mem_cgroup_free(memcg); |
04046e1a | 5048 | return ERR_PTR(error); |
8cdea7c0 BS |
5049 | } |
5050 | ||
ec64f515 | 5051 | static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
df878fb0 KH |
5052 | struct cgroup *cont) |
5053 | { | |
c0ff4b85 | 5054 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
ec64f515 | 5055 | |
c0ff4b85 | 5056 | return mem_cgroup_force_empty(memcg, false); |
df878fb0 KH |
5057 | } |
5058 | ||
8cdea7c0 BS |
5059 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
5060 | struct cgroup *cont) | |
5061 | { | |
c0ff4b85 | 5062 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
c268e994 | 5063 | |
d1a4c0b3 GC |
5064 | kmem_cgroup_destroy(ss, cont); |
5065 | ||
c0ff4b85 | 5066 | mem_cgroup_put(memcg); |
8cdea7c0 BS |
5067 | } |
5068 | ||
5069 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
5070 | struct cgroup *cont) | |
5071 | { | |
8c7c6e34 KH |
5072 | int ret; |
5073 | ||
5074 | ret = cgroup_add_files(cont, ss, mem_cgroup_files, | |
5075 | ARRAY_SIZE(mem_cgroup_files)); | |
5076 | ||
5077 | if (!ret) | |
5078 | ret = register_memsw_files(cont, ss); | |
e5671dfa GC |
5079 | |
5080 | if (!ret) | |
5081 | ret = register_kmem_files(cont, ss); | |
5082 | ||
8c7c6e34 | 5083 | return ret; |
8cdea7c0 BS |
5084 | } |
5085 | ||
02491447 | 5086 | #ifdef CONFIG_MMU |
7dc74be0 | 5087 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
5088 | #define PRECHARGE_COUNT_AT_ONCE 256 |
5089 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 5090 | { |
854ffa8d DN |
5091 | int ret = 0; |
5092 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
c0ff4b85 | 5093 | struct mem_cgroup *memcg = mc.to; |
4ffef5fe | 5094 | |
c0ff4b85 | 5095 | if (mem_cgroup_is_root(memcg)) { |
854ffa8d DN |
5096 | mc.precharge += count; |
5097 | /* we don't need css_get for root */ | |
5098 | return ret; | |
5099 | } | |
5100 | /* try to charge at once */ | |
5101 | if (count > 1) { | |
5102 | struct res_counter *dummy; | |
5103 | /* | |
c0ff4b85 | 5104 | * "memcg" cannot be under rmdir() because we've already checked |
854ffa8d DN |
5105 | * by cgroup_lock_live_cgroup() that it is not removed and we |
5106 | * are still under the same cgroup_mutex. So we can postpone | |
5107 | * css_get(). | |
5108 | */ | |
c0ff4b85 | 5109 | if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy)) |
854ffa8d | 5110 | goto one_by_one; |
c0ff4b85 | 5111 | if (do_swap_account && res_counter_charge(&memcg->memsw, |
854ffa8d | 5112 | PAGE_SIZE * count, &dummy)) { |
c0ff4b85 | 5113 | res_counter_uncharge(&memcg->res, PAGE_SIZE * count); |
854ffa8d DN |
5114 | goto one_by_one; |
5115 | } | |
5116 | mc.precharge += count; | |
854ffa8d DN |
5117 | return ret; |
5118 | } | |
5119 | one_by_one: | |
5120 | /* fall back to one by one charge */ | |
5121 | while (count--) { | |
5122 | if (signal_pending(current)) { | |
5123 | ret = -EINTR; | |
5124 | break; | |
5125 | } | |
5126 | if (!batch_count--) { | |
5127 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
5128 | cond_resched(); | |
5129 | } | |
c0ff4b85 R |
5130 | ret = __mem_cgroup_try_charge(NULL, |
5131 | GFP_KERNEL, 1, &memcg, false); | |
5132 | if (ret || !memcg) | |
854ffa8d DN |
5133 | /* mem_cgroup_clear_mc() will do uncharge later */ |
5134 | return -ENOMEM; | |
5135 | mc.precharge++; | |
5136 | } | |
4ffef5fe DN |
5137 | return ret; |
5138 | } | |
5139 | ||
5140 | /** | |
5141 | * is_target_pte_for_mc - check a pte whether it is valid for move charge | |
5142 | * @vma: the vma the pte to be checked belongs | |
5143 | * @addr: the address corresponding to the pte to be checked | |
5144 | * @ptent: the pte to be checked | |
02491447 | 5145 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
5146 | * |
5147 | * Returns | |
5148 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5149 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5150 | * move charge. if @target is not NULL, the page is stored in target->page | |
5151 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
5152 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
5153 | * target for charge migration. if @target is not NULL, the entry is stored | |
5154 | * in target->ent. | |
4ffef5fe DN |
5155 | * |
5156 | * Called with pte lock held. | |
5157 | */ | |
4ffef5fe DN |
5158 | union mc_target { |
5159 | struct page *page; | |
02491447 | 5160 | swp_entry_t ent; |
4ffef5fe DN |
5161 | }; |
5162 | ||
4ffef5fe DN |
5163 | enum mc_target_type { |
5164 | MC_TARGET_NONE, /* not used */ | |
5165 | MC_TARGET_PAGE, | |
02491447 | 5166 | MC_TARGET_SWAP, |
4ffef5fe DN |
5167 | }; |
5168 | ||
90254a65 DN |
5169 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5170 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5171 | { |
90254a65 | 5172 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5173 | |
90254a65 DN |
5174 | if (!page || !page_mapped(page)) |
5175 | return NULL; | |
5176 | if (PageAnon(page)) { | |
5177 | /* we don't move shared anon */ | |
5178 | if (!move_anon() || page_mapcount(page) > 2) | |
5179 | return NULL; | |
87946a72 DN |
5180 | } else if (!move_file()) |
5181 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
5182 | return NULL; |
5183 | if (!get_page_unless_zero(page)) | |
5184 | return NULL; | |
5185 | ||
5186 | return page; | |
5187 | } | |
5188 | ||
5189 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
5190 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5191 | { | |
5192 | int usage_count; | |
5193 | struct page *page = NULL; | |
5194 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5195 | ||
5196 | if (!move_anon() || non_swap_entry(ent)) | |
5197 | return NULL; | |
5198 | usage_count = mem_cgroup_count_swap_user(ent, &page); | |
5199 | if (usage_count > 1) { /* we don't move shared anon */ | |
02491447 DN |
5200 | if (page) |
5201 | put_page(page); | |
90254a65 | 5202 | return NULL; |
02491447 | 5203 | } |
90254a65 DN |
5204 | if (do_swap_account) |
5205 | entry->val = ent.val; | |
5206 | ||
5207 | return page; | |
5208 | } | |
5209 | ||
87946a72 DN |
5210 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5211 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5212 | { | |
5213 | struct page *page = NULL; | |
5214 | struct inode *inode; | |
5215 | struct address_space *mapping; | |
5216 | pgoff_t pgoff; | |
5217 | ||
5218 | if (!vma->vm_file) /* anonymous vma */ | |
5219 | return NULL; | |
5220 | if (!move_file()) | |
5221 | return NULL; | |
5222 | ||
5223 | inode = vma->vm_file->f_path.dentry->d_inode; | |
5224 | mapping = vma->vm_file->f_mapping; | |
5225 | if (pte_none(ptent)) | |
5226 | pgoff = linear_page_index(vma, addr); | |
5227 | else /* pte_file(ptent) is true */ | |
5228 | pgoff = pte_to_pgoff(ptent); | |
5229 | ||
5230 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
5231 | page = find_get_page(mapping, pgoff); |
5232 | ||
5233 | #ifdef CONFIG_SWAP | |
5234 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
5235 | if (radix_tree_exceptional_entry(page)) { | |
5236 | swp_entry_t swap = radix_to_swp_entry(page); | |
87946a72 | 5237 | if (do_swap_account) |
aa3b1895 HD |
5238 | *entry = swap; |
5239 | page = find_get_page(&swapper_space, swap.val); | |
87946a72 | 5240 | } |
aa3b1895 | 5241 | #endif |
87946a72 DN |
5242 | return page; |
5243 | } | |
5244 | ||
90254a65 DN |
5245 | static int is_target_pte_for_mc(struct vm_area_struct *vma, |
5246 | unsigned long addr, pte_t ptent, union mc_target *target) | |
5247 | { | |
5248 | struct page *page = NULL; | |
5249 | struct page_cgroup *pc; | |
5250 | int ret = 0; | |
5251 | swp_entry_t ent = { .val = 0 }; | |
5252 | ||
5253 | if (pte_present(ptent)) | |
5254 | page = mc_handle_present_pte(vma, addr, ptent); | |
5255 | else if (is_swap_pte(ptent)) | |
5256 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
5257 | else if (pte_none(ptent) || pte_file(ptent)) |
5258 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
5259 | |
5260 | if (!page && !ent.val) | |
5261 | return 0; | |
02491447 DN |
5262 | if (page) { |
5263 | pc = lookup_page_cgroup(page); | |
5264 | /* | |
5265 | * Do only loose check w/o page_cgroup lock. | |
5266 | * mem_cgroup_move_account() checks the pc is valid or not under | |
5267 | * the lock. | |
5268 | */ | |
5269 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5270 | ret = MC_TARGET_PAGE; | |
5271 | if (target) | |
5272 | target->page = page; | |
5273 | } | |
5274 | if (!ret || !target) | |
5275 | put_page(page); | |
5276 | } | |
90254a65 DN |
5277 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
5278 | if (ent.val && !ret && | |
7f0f1546 KH |
5279 | css_id(&mc.from->css) == lookup_swap_cgroup(ent)) { |
5280 | ret = MC_TARGET_SWAP; | |
5281 | if (target) | |
5282 | target->ent = ent; | |
4ffef5fe | 5283 | } |
4ffef5fe DN |
5284 | return ret; |
5285 | } | |
5286 | ||
5287 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, | |
5288 | unsigned long addr, unsigned long end, | |
5289 | struct mm_walk *walk) | |
5290 | { | |
5291 | struct vm_area_struct *vma = walk->private; | |
5292 | pte_t *pte; | |
5293 | spinlock_t *ptl; | |
5294 | ||
03319327 DH |
5295 | split_huge_page_pmd(walk->mm, pmd); |
5296 | ||
4ffef5fe DN |
5297 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5298 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
5299 | if (is_target_pte_for_mc(vma, addr, *pte, NULL)) | |
5300 | mc.precharge++; /* increment precharge temporarily */ | |
5301 | pte_unmap_unlock(pte - 1, ptl); | |
5302 | cond_resched(); | |
5303 | ||
7dc74be0 DN |
5304 | return 0; |
5305 | } | |
5306 | ||
4ffef5fe DN |
5307 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5308 | { | |
5309 | unsigned long precharge; | |
5310 | struct vm_area_struct *vma; | |
5311 | ||
dfe076b0 | 5312 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
5313 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5314 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
5315 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5316 | .mm = mm, | |
5317 | .private = vma, | |
5318 | }; | |
5319 | if (is_vm_hugetlb_page(vma)) | |
5320 | continue; | |
4ffef5fe DN |
5321 | walk_page_range(vma->vm_start, vma->vm_end, |
5322 | &mem_cgroup_count_precharge_walk); | |
5323 | } | |
dfe076b0 | 5324 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5325 | |
5326 | precharge = mc.precharge; | |
5327 | mc.precharge = 0; | |
5328 | ||
5329 | return precharge; | |
5330 | } | |
5331 | ||
4ffef5fe DN |
5332 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5333 | { | |
dfe076b0 DN |
5334 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5335 | ||
5336 | VM_BUG_ON(mc.moving_task); | |
5337 | mc.moving_task = current; | |
5338 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5339 | } |
5340 | ||
dfe076b0 DN |
5341 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5342 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5343 | { |
2bd9bb20 KH |
5344 | struct mem_cgroup *from = mc.from; |
5345 | struct mem_cgroup *to = mc.to; | |
5346 | ||
4ffef5fe | 5347 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
5348 | if (mc.precharge) { |
5349 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
5350 | mc.precharge = 0; | |
5351 | } | |
5352 | /* | |
5353 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5354 | * we must uncharge here. | |
5355 | */ | |
5356 | if (mc.moved_charge) { | |
5357 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
5358 | mc.moved_charge = 0; | |
4ffef5fe | 5359 | } |
483c30b5 DN |
5360 | /* we must fixup refcnts and charges */ |
5361 | if (mc.moved_swap) { | |
483c30b5 DN |
5362 | /* uncharge swap account from the old cgroup */ |
5363 | if (!mem_cgroup_is_root(mc.from)) | |
5364 | res_counter_uncharge(&mc.from->memsw, | |
5365 | PAGE_SIZE * mc.moved_swap); | |
5366 | __mem_cgroup_put(mc.from, mc.moved_swap); | |
5367 | ||
5368 | if (!mem_cgroup_is_root(mc.to)) { | |
5369 | /* | |
5370 | * we charged both to->res and to->memsw, so we should | |
5371 | * uncharge to->res. | |
5372 | */ | |
5373 | res_counter_uncharge(&mc.to->res, | |
5374 | PAGE_SIZE * mc.moved_swap); | |
483c30b5 DN |
5375 | } |
5376 | /* we've already done mem_cgroup_get(mc.to) */ | |
483c30b5 DN |
5377 | mc.moved_swap = 0; |
5378 | } | |
dfe076b0 DN |
5379 | memcg_oom_recover(from); |
5380 | memcg_oom_recover(to); | |
5381 | wake_up_all(&mc.waitq); | |
5382 | } | |
5383 | ||
5384 | static void mem_cgroup_clear_mc(void) | |
5385 | { | |
5386 | struct mem_cgroup *from = mc.from; | |
5387 | ||
5388 | /* | |
5389 | * we must clear moving_task before waking up waiters at the end of | |
5390 | * task migration. | |
5391 | */ | |
5392 | mc.moving_task = NULL; | |
5393 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5394 | spin_lock(&mc.lock); |
4ffef5fe DN |
5395 | mc.from = NULL; |
5396 | mc.to = NULL; | |
2bd9bb20 | 5397 | spin_unlock(&mc.lock); |
32047e2a | 5398 | mem_cgroup_end_move(from); |
4ffef5fe DN |
5399 | } |
5400 | ||
7dc74be0 DN |
5401 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, |
5402 | struct cgroup *cgroup, | |
2f7ee569 | 5403 | struct cgroup_taskset *tset) |
7dc74be0 | 5404 | { |
2f7ee569 | 5405 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 5406 | int ret = 0; |
c0ff4b85 | 5407 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup); |
7dc74be0 | 5408 | |
c0ff4b85 | 5409 | if (memcg->move_charge_at_immigrate) { |
7dc74be0 DN |
5410 | struct mm_struct *mm; |
5411 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
5412 | ||
c0ff4b85 | 5413 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
5414 | |
5415 | mm = get_task_mm(p); | |
5416 | if (!mm) | |
5417 | return 0; | |
7dc74be0 | 5418 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
5419 | if (mm->owner == p) { |
5420 | VM_BUG_ON(mc.from); | |
5421 | VM_BUG_ON(mc.to); | |
5422 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 5423 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 5424 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 5425 | mem_cgroup_start_move(from); |
2bd9bb20 | 5426 | spin_lock(&mc.lock); |
4ffef5fe | 5427 | mc.from = from; |
c0ff4b85 | 5428 | mc.to = memcg; |
2bd9bb20 | 5429 | spin_unlock(&mc.lock); |
dfe076b0 | 5430 | /* We set mc.moving_task later */ |
4ffef5fe DN |
5431 | |
5432 | ret = mem_cgroup_precharge_mc(mm); | |
5433 | if (ret) | |
5434 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
5435 | } |
5436 | mmput(mm); | |
7dc74be0 DN |
5437 | } |
5438 | return ret; | |
5439 | } | |
5440 | ||
5441 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
5442 | struct cgroup *cgroup, | |
2f7ee569 | 5443 | struct cgroup_taskset *tset) |
7dc74be0 | 5444 | { |
4ffef5fe | 5445 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
5446 | } |
5447 | ||
4ffef5fe DN |
5448 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5449 | unsigned long addr, unsigned long end, | |
5450 | struct mm_walk *walk) | |
7dc74be0 | 5451 | { |
4ffef5fe DN |
5452 | int ret = 0; |
5453 | struct vm_area_struct *vma = walk->private; | |
5454 | pte_t *pte; | |
5455 | spinlock_t *ptl; | |
5456 | ||
03319327 | 5457 | split_huge_page_pmd(walk->mm, pmd); |
4ffef5fe DN |
5458 | retry: |
5459 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5460 | for (; addr != end; addr += PAGE_SIZE) { | |
5461 | pte_t ptent = *(pte++); | |
5462 | union mc_target target; | |
5463 | int type; | |
5464 | struct page *page; | |
5465 | struct page_cgroup *pc; | |
02491447 | 5466 | swp_entry_t ent; |
4ffef5fe DN |
5467 | |
5468 | if (!mc.precharge) | |
5469 | break; | |
5470 | ||
5471 | type = is_target_pte_for_mc(vma, addr, ptent, &target); | |
5472 | switch (type) { | |
5473 | case MC_TARGET_PAGE: | |
5474 | page = target.page; | |
5475 | if (isolate_lru_page(page)) | |
5476 | goto put; | |
5477 | pc = lookup_page_cgroup(page); | |
7ec99d62 JW |
5478 | if (!mem_cgroup_move_account(page, 1, pc, |
5479 | mc.from, mc.to, false)) { | |
4ffef5fe | 5480 | mc.precharge--; |
854ffa8d DN |
5481 | /* we uncharge from mc.from later. */ |
5482 | mc.moved_charge++; | |
4ffef5fe DN |
5483 | } |
5484 | putback_lru_page(page); | |
5485 | put: /* is_target_pte_for_mc() gets the page */ | |
5486 | put_page(page); | |
5487 | break; | |
02491447 DN |
5488 | case MC_TARGET_SWAP: |
5489 | ent = target.ent; | |
483c30b5 DN |
5490 | if (!mem_cgroup_move_swap_account(ent, |
5491 | mc.from, mc.to, false)) { | |
02491447 | 5492 | mc.precharge--; |
483c30b5 DN |
5493 | /* we fixup refcnts and charges later. */ |
5494 | mc.moved_swap++; | |
5495 | } | |
02491447 | 5496 | break; |
4ffef5fe DN |
5497 | default: |
5498 | break; | |
5499 | } | |
5500 | } | |
5501 | pte_unmap_unlock(pte - 1, ptl); | |
5502 | cond_resched(); | |
5503 | ||
5504 | if (addr != end) { | |
5505 | /* | |
5506 | * We have consumed all precharges we got in can_attach(). | |
5507 | * We try charge one by one, but don't do any additional | |
5508 | * charges to mc.to if we have failed in charge once in attach() | |
5509 | * phase. | |
5510 | */ | |
854ffa8d | 5511 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5512 | if (!ret) |
5513 | goto retry; | |
5514 | } | |
5515 | ||
5516 | return ret; | |
5517 | } | |
5518 | ||
5519 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
5520 | { | |
5521 | struct vm_area_struct *vma; | |
5522 | ||
5523 | lru_add_drain_all(); | |
dfe076b0 DN |
5524 | retry: |
5525 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5526 | /* | |
5527 | * Someone who are holding the mmap_sem might be waiting in | |
5528 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5529 | * and retry. Because we cancel precharges, we might not be able | |
5530 | * to move enough charges, but moving charge is a best-effort | |
5531 | * feature anyway, so it wouldn't be a big problem. | |
5532 | */ | |
5533 | __mem_cgroup_clear_mc(); | |
5534 | cond_resched(); | |
5535 | goto retry; | |
5536 | } | |
4ffef5fe DN |
5537 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5538 | int ret; | |
5539 | struct mm_walk mem_cgroup_move_charge_walk = { | |
5540 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
5541 | .mm = mm, | |
5542 | .private = vma, | |
5543 | }; | |
5544 | if (is_vm_hugetlb_page(vma)) | |
5545 | continue; | |
4ffef5fe DN |
5546 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
5547 | &mem_cgroup_move_charge_walk); | |
5548 | if (ret) | |
5549 | /* | |
5550 | * means we have consumed all precharges and failed in | |
5551 | * doing additional charge. Just abandon here. | |
5552 | */ | |
5553 | break; | |
5554 | } | |
dfe076b0 | 5555 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
5556 | } |
5557 | ||
67e465a7 BS |
5558 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
5559 | struct cgroup *cont, | |
2f7ee569 | 5560 | struct cgroup_taskset *tset) |
67e465a7 | 5561 | { |
2f7ee569 | 5562 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 5563 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 5564 | |
dfe076b0 | 5565 | if (mm) { |
a433658c KM |
5566 | if (mc.to) |
5567 | mem_cgroup_move_charge(mm); | |
5568 | put_swap_token(mm); | |
dfe076b0 DN |
5569 | mmput(mm); |
5570 | } | |
a433658c KM |
5571 | if (mc.to) |
5572 | mem_cgroup_clear_mc(); | |
67e465a7 | 5573 | } |
5cfb80a7 DN |
5574 | #else /* !CONFIG_MMU */ |
5575 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, | |
5576 | struct cgroup *cgroup, | |
2f7ee569 | 5577 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5578 | { |
5579 | return 0; | |
5580 | } | |
5581 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
5582 | struct cgroup *cgroup, | |
2f7ee569 | 5583 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5584 | { |
5585 | } | |
5586 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, | |
5587 | struct cgroup *cont, | |
2f7ee569 | 5588 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5589 | { |
5590 | } | |
5591 | #endif | |
67e465a7 | 5592 | |
8cdea7c0 BS |
5593 | struct cgroup_subsys mem_cgroup_subsys = { |
5594 | .name = "memory", | |
5595 | .subsys_id = mem_cgroup_subsys_id, | |
5596 | .create = mem_cgroup_create, | |
df878fb0 | 5597 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
5598 | .destroy = mem_cgroup_destroy, |
5599 | .populate = mem_cgroup_populate, | |
7dc74be0 DN |
5600 | .can_attach = mem_cgroup_can_attach, |
5601 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5602 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 5603 | .early_init = 0, |
04046e1a | 5604 | .use_id = 1, |
8cdea7c0 | 5605 | }; |
c077719b KH |
5606 | |
5607 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
a42c390c MH |
5608 | static int __init enable_swap_account(char *s) |
5609 | { | |
5610 | /* consider enabled if no parameter or 1 is given */ | |
a2c8990a | 5611 | if (!strcmp(s, "1")) |
a42c390c | 5612 | really_do_swap_account = 1; |
a2c8990a | 5613 | else if (!strcmp(s, "0")) |
a42c390c MH |
5614 | really_do_swap_account = 0; |
5615 | return 1; | |
5616 | } | |
a2c8990a | 5617 | __setup("swapaccount=", enable_swap_account); |
c077719b | 5618 | |
c077719b | 5619 | #endif |