]>
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; | |
e5671dfa GC |
231 | /* |
232 | * the counter to account for kmem usage. | |
233 | */ | |
234 | struct res_counter kmem; | |
78fb7466 PE |
235 | /* |
236 | * Per cgroup active and inactive list, similar to the | |
237 | * per zone LRU lists. | |
78fb7466 | 238 | */ |
6d12e2d8 | 239 | struct mem_cgroup_lru_info info; |
6d61ef40 | 240 | /* |
af901ca1 | 241 | * While reclaiming in a hierarchy, we cache the last child we |
04046e1a | 242 | * reclaimed from. |
6d61ef40 | 243 | */ |
04046e1a | 244 | int last_scanned_child; |
889976db YH |
245 | int last_scanned_node; |
246 | #if MAX_NUMNODES > 1 | |
247 | nodemask_t scan_nodes; | |
453a9bf3 KH |
248 | atomic_t numainfo_events; |
249 | atomic_t numainfo_updating; | |
889976db | 250 | #endif |
18f59ea7 BS |
251 | /* |
252 | * Should the accounting and control be hierarchical, per subtree? | |
253 | */ | |
254 | bool use_hierarchy; | |
79dfdacc MH |
255 | |
256 | bool oom_lock; | |
257 | atomic_t under_oom; | |
258 | ||
8c7c6e34 | 259 | atomic_t refcnt; |
14797e23 | 260 | |
1f4c025b | 261 | int swappiness; |
3c11ecf4 KH |
262 | /* OOM-Killer disable */ |
263 | int oom_kill_disable; | |
a7885eb8 | 264 | |
22a668d7 KH |
265 | /* set when res.limit == memsw.limit */ |
266 | bool memsw_is_minimum; | |
267 | ||
2e72b634 KS |
268 | /* protect arrays of thresholds */ |
269 | struct mutex thresholds_lock; | |
270 | ||
271 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 272 | struct mem_cgroup_thresholds thresholds; |
907860ed | 273 | |
2e72b634 | 274 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 275 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 276 | |
9490ff27 KH |
277 | /* For oom notifier event fd */ |
278 | struct list_head oom_notify; | |
185efc0f | 279 | |
7dc74be0 DN |
280 | /* |
281 | * Should we move charges of a task when a task is moved into this | |
282 | * mem_cgroup ? And what type of charges should we move ? | |
283 | */ | |
284 | unsigned long move_charge_at_immigrate; | |
e5671dfa GC |
285 | /* |
286 | * Should kernel memory limits be stabilished independently | |
287 | * from user memory ? | |
288 | */ | |
289 | int kmem_independent_accounting; | |
d52aa412 | 290 | /* |
c62b1a3b | 291 | * percpu counter. |
d52aa412 | 292 | */ |
c62b1a3b | 293 | struct mem_cgroup_stat_cpu *stat; |
711d3d2c KH |
294 | /* |
295 | * used when a cpu is offlined or other synchronizations | |
296 | * See mem_cgroup_read_stat(). | |
297 | */ | |
298 | struct mem_cgroup_stat_cpu nocpu_base; | |
299 | spinlock_t pcp_counter_lock; | |
d1a4c0b3 GC |
300 | |
301 | #ifdef CONFIG_INET | |
302 | struct tcp_memcontrol tcp_mem; | |
303 | #endif | |
8cdea7c0 BS |
304 | }; |
305 | ||
7dc74be0 DN |
306 | /* Stuffs for move charges at task migration. */ |
307 | /* | |
308 | * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a | |
309 | * left-shifted bitmap of these types. | |
310 | */ | |
311 | enum move_type { | |
4ffef5fe | 312 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 313 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
314 | NR_MOVE_TYPE, |
315 | }; | |
316 | ||
4ffef5fe DN |
317 | /* "mc" and its members are protected by cgroup_mutex */ |
318 | static struct move_charge_struct { | |
b1dd693e | 319 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
320 | struct mem_cgroup *from; |
321 | struct mem_cgroup *to; | |
322 | unsigned long precharge; | |
854ffa8d | 323 | unsigned long moved_charge; |
483c30b5 | 324 | unsigned long moved_swap; |
8033b97c DN |
325 | struct task_struct *moving_task; /* a task moving charges */ |
326 | wait_queue_head_t waitq; /* a waitq for other context */ | |
327 | } mc = { | |
2bd9bb20 | 328 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
329 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
330 | }; | |
4ffef5fe | 331 | |
90254a65 DN |
332 | static bool move_anon(void) |
333 | { | |
334 | return test_bit(MOVE_CHARGE_TYPE_ANON, | |
335 | &mc.to->move_charge_at_immigrate); | |
336 | } | |
337 | ||
87946a72 DN |
338 | static bool move_file(void) |
339 | { | |
340 | return test_bit(MOVE_CHARGE_TYPE_FILE, | |
341 | &mc.to->move_charge_at_immigrate); | |
342 | } | |
343 | ||
4e416953 BS |
344 | /* |
345 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
346 | * limit reclaim to prevent infinite loops, if they ever occur. | |
347 | */ | |
348 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100) | |
349 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2) | |
350 | ||
217bc319 KH |
351 | enum charge_type { |
352 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
353 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
4f98a2fe | 354 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
c05555b5 | 355 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
d13d1443 | 356 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 357 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
358 | NR_CHARGE_TYPE, |
359 | }; | |
360 | ||
8c7c6e34 | 361 | /* for encoding cft->private value on file */ |
e5671dfa GC |
362 | |
363 | enum mem_type { | |
364 | _MEM = 0, | |
365 | _MEMSWAP, | |
366 | _OOM_TYPE, | |
367 | _KMEM, | |
368 | }; | |
369 | ||
8c7c6e34 KH |
370 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) |
371 | #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) | |
372 | #define MEMFILE_ATTR(val) ((val) & 0xffff) | |
9490ff27 KH |
373 | /* Used for OOM nofiier */ |
374 | #define OOM_CONTROL (0) | |
8c7c6e34 | 375 | |
75822b44 BS |
376 | /* |
377 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
378 | */ | |
379 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
380 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
381 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
382 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
4e416953 BS |
383 | #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2 |
384 | #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT) | |
75822b44 | 385 | |
c0ff4b85 R |
386 | static void mem_cgroup_get(struct mem_cgroup *memcg); |
387 | static void mem_cgroup_put(struct mem_cgroup *memcg); | |
e1aab161 GC |
388 | |
389 | /* Writing them here to avoid exposing memcg's inner layout */ | |
390 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM | |
391 | #ifdef CONFIG_INET | |
392 | #include <net/sock.h> | |
d1a4c0b3 | 393 | #include <net/ip.h> |
e1aab161 GC |
394 | |
395 | static bool mem_cgroup_is_root(struct mem_cgroup *memcg); | |
396 | void sock_update_memcg(struct sock *sk) | |
397 | { | |
398 | /* A socket spends its whole life in the same cgroup */ | |
399 | if (sk->sk_cgrp) { | |
400 | WARN_ON(1); | |
401 | return; | |
402 | } | |
403 | if (static_branch(&memcg_socket_limit_enabled)) { | |
404 | struct mem_cgroup *memcg; | |
405 | ||
406 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
407 | ||
408 | rcu_read_lock(); | |
409 | memcg = mem_cgroup_from_task(current); | |
410 | if (!mem_cgroup_is_root(memcg)) { | |
411 | mem_cgroup_get(memcg); | |
412 | sk->sk_cgrp = sk->sk_prot->proto_cgroup(memcg); | |
413 | } | |
414 | rcu_read_unlock(); | |
415 | } | |
416 | } | |
417 | EXPORT_SYMBOL(sock_update_memcg); | |
418 | ||
419 | void sock_release_memcg(struct sock *sk) | |
420 | { | |
421 | if (static_branch(&memcg_socket_limit_enabled) && sk->sk_cgrp) { | |
422 | struct mem_cgroup *memcg; | |
423 | WARN_ON(!sk->sk_cgrp->memcg); | |
424 | memcg = sk->sk_cgrp->memcg; | |
425 | mem_cgroup_put(memcg); | |
426 | } | |
427 | } | |
d1a4c0b3 GC |
428 | |
429 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
430 | { | |
431 | if (!memcg || mem_cgroup_is_root(memcg)) | |
432 | return NULL; | |
433 | ||
434 | return &memcg->tcp_mem.cg_proto; | |
435 | } | |
436 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 GC |
437 | #endif /* CONFIG_INET */ |
438 | #endif /* CONFIG_CGROUP_MEM_RES_CTLR_KMEM */ | |
439 | ||
c0ff4b85 | 440 | static void drain_all_stock_async(struct mem_cgroup *memcg); |
8c7c6e34 | 441 | |
f64c3f54 | 442 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 443 | mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid) |
f64c3f54 | 444 | { |
c0ff4b85 | 445 | return &memcg->info.nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
446 | } |
447 | ||
c0ff4b85 | 448 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 449 | { |
c0ff4b85 | 450 | return &memcg->css; |
d324236b WF |
451 | } |
452 | ||
f64c3f54 | 453 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 454 | page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 455 | { |
97a6c37b JW |
456 | int nid = page_to_nid(page); |
457 | int zid = page_zonenum(page); | |
f64c3f54 | 458 | |
c0ff4b85 | 459 | return mem_cgroup_zoneinfo(memcg, nid, zid); |
f64c3f54 BS |
460 | } |
461 | ||
462 | static struct mem_cgroup_tree_per_zone * | |
463 | soft_limit_tree_node_zone(int nid, int zid) | |
464 | { | |
465 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
466 | } | |
467 | ||
468 | static struct mem_cgroup_tree_per_zone * | |
469 | soft_limit_tree_from_page(struct page *page) | |
470 | { | |
471 | int nid = page_to_nid(page); | |
472 | int zid = page_zonenum(page); | |
473 | ||
474 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
475 | } | |
476 | ||
477 | static void | |
c0ff4b85 | 478 | __mem_cgroup_insert_exceeded(struct mem_cgroup *memcg, |
f64c3f54 | 479 | struct mem_cgroup_per_zone *mz, |
ef8745c1 KH |
480 | struct mem_cgroup_tree_per_zone *mctz, |
481 | unsigned long long new_usage_in_excess) | |
f64c3f54 BS |
482 | { |
483 | struct rb_node **p = &mctz->rb_root.rb_node; | |
484 | struct rb_node *parent = NULL; | |
485 | struct mem_cgroup_per_zone *mz_node; | |
486 | ||
487 | if (mz->on_tree) | |
488 | return; | |
489 | ||
ef8745c1 KH |
490 | mz->usage_in_excess = new_usage_in_excess; |
491 | if (!mz->usage_in_excess) | |
492 | return; | |
f64c3f54 BS |
493 | while (*p) { |
494 | parent = *p; | |
495 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
496 | tree_node); | |
497 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
498 | p = &(*p)->rb_left; | |
499 | /* | |
500 | * We can't avoid mem cgroups that are over their soft | |
501 | * limit by the same amount | |
502 | */ | |
503 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
504 | p = &(*p)->rb_right; | |
505 | } | |
506 | rb_link_node(&mz->tree_node, parent, p); | |
507 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
508 | mz->on_tree = true; | |
4e416953 BS |
509 | } |
510 | ||
511 | static void | |
c0ff4b85 | 512 | __mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
4e416953 BS |
513 | struct mem_cgroup_per_zone *mz, |
514 | struct mem_cgroup_tree_per_zone *mctz) | |
515 | { | |
516 | if (!mz->on_tree) | |
517 | return; | |
518 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
519 | mz->on_tree = false; | |
520 | } | |
521 | ||
f64c3f54 | 522 | static void |
c0ff4b85 | 523 | mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
f64c3f54 BS |
524 | struct mem_cgroup_per_zone *mz, |
525 | struct mem_cgroup_tree_per_zone *mctz) | |
526 | { | |
527 | spin_lock(&mctz->lock); | |
c0ff4b85 | 528 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
529 | spin_unlock(&mctz->lock); |
530 | } | |
531 | ||
f64c3f54 | 532 | |
c0ff4b85 | 533 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 534 | { |
ef8745c1 | 535 | unsigned long long excess; |
f64c3f54 BS |
536 | struct mem_cgroup_per_zone *mz; |
537 | struct mem_cgroup_tree_per_zone *mctz; | |
4e649152 KH |
538 | int nid = page_to_nid(page); |
539 | int zid = page_zonenum(page); | |
f64c3f54 BS |
540 | mctz = soft_limit_tree_from_page(page); |
541 | ||
542 | /* | |
4e649152 KH |
543 | * Necessary to update all ancestors when hierarchy is used. |
544 | * because their event counter is not touched. | |
f64c3f54 | 545 | */ |
c0ff4b85 R |
546 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
547 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
548 | excess = res_counter_soft_limit_excess(&memcg->res); | |
4e649152 KH |
549 | /* |
550 | * We have to update the tree if mz is on RB-tree or | |
551 | * mem is over its softlimit. | |
552 | */ | |
ef8745c1 | 553 | if (excess || mz->on_tree) { |
4e649152 KH |
554 | spin_lock(&mctz->lock); |
555 | /* if on-tree, remove it */ | |
556 | if (mz->on_tree) | |
c0ff4b85 | 557 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
4e649152 | 558 | /* |
ef8745c1 KH |
559 | * Insert again. mz->usage_in_excess will be updated. |
560 | * If excess is 0, no tree ops. | |
4e649152 | 561 | */ |
c0ff4b85 | 562 | __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess); |
4e649152 KH |
563 | spin_unlock(&mctz->lock); |
564 | } | |
f64c3f54 BS |
565 | } |
566 | } | |
567 | ||
c0ff4b85 | 568 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) |
f64c3f54 BS |
569 | { |
570 | int node, zone; | |
571 | struct mem_cgroup_per_zone *mz; | |
572 | struct mem_cgroup_tree_per_zone *mctz; | |
573 | ||
574 | for_each_node_state(node, N_POSSIBLE) { | |
575 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
c0ff4b85 | 576 | mz = mem_cgroup_zoneinfo(memcg, node, zone); |
f64c3f54 | 577 | mctz = soft_limit_tree_node_zone(node, zone); |
c0ff4b85 | 578 | mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
579 | } |
580 | } | |
581 | } | |
582 | ||
4e416953 BS |
583 | static struct mem_cgroup_per_zone * |
584 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
585 | { | |
586 | struct rb_node *rightmost = NULL; | |
26251eaf | 587 | struct mem_cgroup_per_zone *mz; |
4e416953 BS |
588 | |
589 | retry: | |
26251eaf | 590 | mz = NULL; |
4e416953 BS |
591 | rightmost = rb_last(&mctz->rb_root); |
592 | if (!rightmost) | |
593 | goto done; /* Nothing to reclaim from */ | |
594 | ||
595 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
596 | /* | |
597 | * Remove the node now but someone else can add it back, | |
598 | * we will to add it back at the end of reclaim to its correct | |
599 | * position in the tree. | |
600 | */ | |
601 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); | |
602 | if (!res_counter_soft_limit_excess(&mz->mem->res) || | |
603 | !css_tryget(&mz->mem->css)) | |
604 | goto retry; | |
605 | done: | |
606 | return mz; | |
607 | } | |
608 | ||
609 | static struct mem_cgroup_per_zone * | |
610 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
611 | { | |
612 | struct mem_cgroup_per_zone *mz; | |
613 | ||
614 | spin_lock(&mctz->lock); | |
615 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
616 | spin_unlock(&mctz->lock); | |
617 | return mz; | |
618 | } | |
619 | ||
711d3d2c KH |
620 | /* |
621 | * Implementation Note: reading percpu statistics for memcg. | |
622 | * | |
623 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
624 | * synchronization to implement "quick" read. There are trade-off between | |
625 | * reading cost and precision of value. Then, we may have a chance to implement | |
626 | * a periodic synchronizion of counter in memcg's counter. | |
627 | * | |
628 | * But this _read() function is used for user interface now. The user accounts | |
629 | * memory usage by memory cgroup and he _always_ requires exact value because | |
630 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
631 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
632 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
633 | * | |
634 | * If there are kernel internal actions which can make use of some not-exact | |
635 | * value, and reading all cpu value can be performance bottleneck in some | |
636 | * common workload, threashold and synchonization as vmstat[] should be | |
637 | * implemented. | |
638 | */ | |
c0ff4b85 | 639 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 640 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 641 | { |
7a159cc9 | 642 | long val = 0; |
c62b1a3b | 643 | int cpu; |
c62b1a3b | 644 | |
711d3d2c KH |
645 | get_online_cpus(); |
646 | for_each_online_cpu(cpu) | |
c0ff4b85 | 647 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 648 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
649 | spin_lock(&memcg->pcp_counter_lock); |
650 | val += memcg->nocpu_base.count[idx]; | |
651 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
652 | #endif |
653 | put_online_cpus(); | |
c62b1a3b KH |
654 | return val; |
655 | } | |
656 | ||
c0ff4b85 | 657 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
0c3e73e8 BS |
658 | bool charge) |
659 | { | |
660 | int val = (charge) ? 1 : -1; | |
c0ff4b85 | 661 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAPOUT], val); |
0c3e73e8 BS |
662 | } |
663 | ||
c0ff4b85 | 664 | void mem_cgroup_pgfault(struct mem_cgroup *memcg, int val) |
456f998e | 665 | { |
c0ff4b85 | 666 | this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT], val); |
456f998e YH |
667 | } |
668 | ||
c0ff4b85 | 669 | void mem_cgroup_pgmajfault(struct mem_cgroup *memcg, int val) |
456f998e | 670 | { |
c0ff4b85 | 671 | this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT], val); |
456f998e YH |
672 | } |
673 | ||
c0ff4b85 | 674 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
675 | enum mem_cgroup_events_index idx) |
676 | { | |
677 | unsigned long val = 0; | |
678 | int cpu; | |
679 | ||
680 | for_each_online_cpu(cpu) | |
c0ff4b85 | 681 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 682 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
683 | spin_lock(&memcg->pcp_counter_lock); |
684 | val += memcg->nocpu_base.events[idx]; | |
685 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f JW |
686 | #endif |
687 | return val; | |
688 | } | |
689 | ||
c0ff4b85 | 690 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
e401f176 | 691 | bool file, int nr_pages) |
d52aa412 | 692 | { |
c62b1a3b KH |
693 | preempt_disable(); |
694 | ||
e401f176 | 695 | if (file) |
c0ff4b85 R |
696 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
697 | nr_pages); | |
d52aa412 | 698 | else |
c0ff4b85 R |
699 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
700 | nr_pages); | |
55e462b0 | 701 | |
e401f176 KH |
702 | /* pagein of a big page is an event. So, ignore page size */ |
703 | if (nr_pages > 0) | |
c0ff4b85 | 704 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 705 | else { |
c0ff4b85 | 706 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
707 | nr_pages = -nr_pages; /* for event */ |
708 | } | |
e401f176 | 709 | |
c0ff4b85 | 710 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_COUNT], nr_pages); |
2e72b634 | 711 | |
c62b1a3b | 712 | preempt_enable(); |
6d12e2d8 KH |
713 | } |
714 | ||
bb2a0de9 | 715 | unsigned long |
c0ff4b85 | 716 | mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid, |
bb2a0de9 | 717 | unsigned int lru_mask) |
889976db YH |
718 | { |
719 | struct mem_cgroup_per_zone *mz; | |
bb2a0de9 KH |
720 | enum lru_list l; |
721 | unsigned long ret = 0; | |
722 | ||
c0ff4b85 | 723 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
bb2a0de9 KH |
724 | |
725 | for_each_lru(l) { | |
726 | if (BIT(l) & lru_mask) | |
727 | ret += MEM_CGROUP_ZSTAT(mz, l); | |
728 | } | |
729 | return ret; | |
730 | } | |
731 | ||
732 | static unsigned long | |
c0ff4b85 | 733 | mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 KH |
734 | int nid, unsigned int lru_mask) |
735 | { | |
889976db YH |
736 | u64 total = 0; |
737 | int zid; | |
738 | ||
bb2a0de9 | 739 | for (zid = 0; zid < MAX_NR_ZONES; zid++) |
c0ff4b85 R |
740 | total += mem_cgroup_zone_nr_lru_pages(memcg, |
741 | nid, zid, lru_mask); | |
bb2a0de9 | 742 | |
889976db YH |
743 | return total; |
744 | } | |
bb2a0de9 | 745 | |
c0ff4b85 | 746 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 747 | unsigned int lru_mask) |
6d12e2d8 | 748 | { |
889976db | 749 | int nid; |
6d12e2d8 KH |
750 | u64 total = 0; |
751 | ||
bb2a0de9 | 752 | for_each_node_state(nid, N_HIGH_MEMORY) |
c0ff4b85 | 753 | total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
6d12e2d8 | 754 | return total; |
d52aa412 KH |
755 | } |
756 | ||
c0ff4b85 | 757 | static bool __memcg_event_check(struct mem_cgroup *memcg, int target) |
7a159cc9 JW |
758 | { |
759 | unsigned long val, next; | |
760 | ||
4799401f SR |
761 | val = __this_cpu_read(memcg->stat->events[MEM_CGROUP_EVENTS_COUNT]); |
762 | next = __this_cpu_read(memcg->stat->targets[target]); | |
7a159cc9 JW |
763 | /* from time_after() in jiffies.h */ |
764 | return ((long)next - (long)val < 0); | |
765 | } | |
766 | ||
c0ff4b85 | 767 | static void __mem_cgroup_target_update(struct mem_cgroup *memcg, int target) |
d2265e6f | 768 | { |
7a159cc9 | 769 | unsigned long val, next; |
d2265e6f | 770 | |
4799401f | 771 | val = __this_cpu_read(memcg->stat->events[MEM_CGROUP_EVENTS_COUNT]); |
d2265e6f | 772 | |
7a159cc9 JW |
773 | switch (target) { |
774 | case MEM_CGROUP_TARGET_THRESH: | |
775 | next = val + THRESHOLDS_EVENTS_TARGET; | |
776 | break; | |
777 | case MEM_CGROUP_TARGET_SOFTLIMIT: | |
778 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
779 | break; | |
453a9bf3 KH |
780 | case MEM_CGROUP_TARGET_NUMAINFO: |
781 | next = val + NUMAINFO_EVENTS_TARGET; | |
782 | break; | |
7a159cc9 JW |
783 | default: |
784 | return; | |
785 | } | |
786 | ||
4799401f | 787 | __this_cpu_write(memcg->stat->targets[target], next); |
d2265e6f KH |
788 | } |
789 | ||
790 | /* | |
791 | * Check events in order. | |
792 | * | |
793 | */ | |
c0ff4b85 | 794 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f | 795 | { |
4799401f | 796 | preempt_disable(); |
d2265e6f | 797 | /* threshold event is triggered in finer grain than soft limit */ |
c0ff4b85 R |
798 | if (unlikely(__memcg_event_check(memcg, MEM_CGROUP_TARGET_THRESH))) { |
799 | mem_cgroup_threshold(memcg); | |
800 | __mem_cgroup_target_update(memcg, MEM_CGROUP_TARGET_THRESH); | |
801 | if (unlikely(__memcg_event_check(memcg, | |
453a9bf3 | 802 | MEM_CGROUP_TARGET_SOFTLIMIT))) { |
c0ff4b85 R |
803 | mem_cgroup_update_tree(memcg, page); |
804 | __mem_cgroup_target_update(memcg, | |
453a9bf3 KH |
805 | MEM_CGROUP_TARGET_SOFTLIMIT); |
806 | } | |
807 | #if MAX_NUMNODES > 1 | |
c0ff4b85 | 808 | if (unlikely(__memcg_event_check(memcg, |
453a9bf3 | 809 | MEM_CGROUP_TARGET_NUMAINFO))) { |
c0ff4b85 R |
810 | atomic_inc(&memcg->numainfo_events); |
811 | __mem_cgroup_target_update(memcg, | |
453a9bf3 | 812 | MEM_CGROUP_TARGET_NUMAINFO); |
7a159cc9 | 813 | } |
453a9bf3 | 814 | #endif |
d2265e6f | 815 | } |
4799401f | 816 | preempt_enable(); |
d2265e6f KH |
817 | } |
818 | ||
d1a4c0b3 | 819 | struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
820 | { |
821 | return container_of(cgroup_subsys_state(cont, | |
822 | mem_cgroup_subsys_id), struct mem_cgroup, | |
823 | css); | |
824 | } | |
825 | ||
cf475ad2 | 826 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 827 | { |
31a78f23 BS |
828 | /* |
829 | * mm_update_next_owner() may clear mm->owner to NULL | |
830 | * if it races with swapoff, page migration, etc. | |
831 | * So this can be called with p == NULL. | |
832 | */ | |
833 | if (unlikely(!p)) | |
834 | return NULL; | |
835 | ||
78fb7466 PE |
836 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
837 | struct mem_cgroup, css); | |
838 | } | |
839 | ||
a433658c | 840 | struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 841 | { |
c0ff4b85 | 842 | struct mem_cgroup *memcg = NULL; |
0b7f569e KH |
843 | |
844 | if (!mm) | |
845 | return NULL; | |
54595fe2 KH |
846 | /* |
847 | * Because we have no locks, mm->owner's may be being moved to other | |
848 | * cgroup. We use css_tryget() here even if this looks | |
849 | * pessimistic (rather than adding locks here). | |
850 | */ | |
851 | rcu_read_lock(); | |
852 | do { | |
c0ff4b85 R |
853 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
854 | if (unlikely(!memcg)) | |
54595fe2 | 855 | break; |
c0ff4b85 | 856 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 857 | rcu_read_unlock(); |
c0ff4b85 | 858 | return memcg; |
54595fe2 KH |
859 | } |
860 | ||
7d74b06f | 861 | /* The caller has to guarantee "mem" exists before calling this */ |
c0ff4b85 | 862 | static struct mem_cgroup *mem_cgroup_start_loop(struct mem_cgroup *memcg) |
14067bb3 | 863 | { |
711d3d2c KH |
864 | struct cgroup_subsys_state *css; |
865 | int found; | |
866 | ||
c0ff4b85 | 867 | if (!memcg) /* ROOT cgroup has the smallest ID */ |
711d3d2c | 868 | return root_mem_cgroup; /*css_put/get against root is ignored*/ |
c0ff4b85 R |
869 | if (!memcg->use_hierarchy) { |
870 | if (css_tryget(&memcg->css)) | |
871 | return memcg; | |
711d3d2c KH |
872 | return NULL; |
873 | } | |
874 | rcu_read_lock(); | |
875 | /* | |
876 | * searching a memory cgroup which has the smallest ID under given | |
877 | * ROOT cgroup. (ID >= 1) | |
878 | */ | |
c0ff4b85 | 879 | css = css_get_next(&mem_cgroup_subsys, 1, &memcg->css, &found); |
711d3d2c | 880 | if (css && css_tryget(css)) |
c0ff4b85 | 881 | memcg = container_of(css, struct mem_cgroup, css); |
711d3d2c | 882 | else |
c0ff4b85 | 883 | memcg = NULL; |
711d3d2c | 884 | rcu_read_unlock(); |
c0ff4b85 | 885 | return memcg; |
7d74b06f KH |
886 | } |
887 | ||
888 | static struct mem_cgroup *mem_cgroup_get_next(struct mem_cgroup *iter, | |
889 | struct mem_cgroup *root, | |
890 | bool cond) | |
891 | { | |
892 | int nextid = css_id(&iter->css) + 1; | |
893 | int found; | |
894 | int hierarchy_used; | |
14067bb3 | 895 | struct cgroup_subsys_state *css; |
14067bb3 | 896 | |
7d74b06f | 897 | hierarchy_used = iter->use_hierarchy; |
14067bb3 | 898 | |
7d74b06f | 899 | css_put(&iter->css); |
711d3d2c KH |
900 | /* If no ROOT, walk all, ignore hierarchy */ |
901 | if (!cond || (root && !hierarchy_used)) | |
7d74b06f | 902 | return NULL; |
14067bb3 | 903 | |
711d3d2c KH |
904 | if (!root) |
905 | root = root_mem_cgroup; | |
906 | ||
7d74b06f KH |
907 | do { |
908 | iter = NULL; | |
14067bb3 | 909 | rcu_read_lock(); |
7d74b06f KH |
910 | |
911 | css = css_get_next(&mem_cgroup_subsys, nextid, | |
912 | &root->css, &found); | |
14067bb3 | 913 | if (css && css_tryget(css)) |
7d74b06f | 914 | iter = container_of(css, struct mem_cgroup, css); |
14067bb3 | 915 | rcu_read_unlock(); |
7d74b06f | 916 | /* If css is NULL, no more cgroups will be found */ |
14067bb3 | 917 | nextid = found + 1; |
7d74b06f | 918 | } while (css && !iter); |
14067bb3 | 919 | |
7d74b06f | 920 | return iter; |
14067bb3 | 921 | } |
7d74b06f KH |
922 | /* |
923 | * for_eacn_mem_cgroup_tree() for visiting all cgroup under tree. Please | |
924 | * be careful that "break" loop is not allowed. We have reference count. | |
925 | * Instead of that modify "cond" to be false and "continue" to exit the loop. | |
926 | */ | |
927 | #define for_each_mem_cgroup_tree_cond(iter, root, cond) \ | |
928 | for (iter = mem_cgroup_start_loop(root);\ | |
929 | iter != NULL;\ | |
930 | iter = mem_cgroup_get_next(iter, root, cond)) | |
931 | ||
932 | #define for_each_mem_cgroup_tree(iter, root) \ | |
933 | for_each_mem_cgroup_tree_cond(iter, root, true) | |
934 | ||
711d3d2c KH |
935 | #define for_each_mem_cgroup_all(iter) \ |
936 | for_each_mem_cgroup_tree_cond(iter, NULL, true) | |
937 | ||
14067bb3 | 938 | |
c0ff4b85 | 939 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
4b3bde4c | 940 | { |
c0ff4b85 | 941 | return (memcg == root_mem_cgroup); |
4b3bde4c BS |
942 | } |
943 | ||
456f998e YH |
944 | void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
945 | { | |
c0ff4b85 | 946 | struct mem_cgroup *memcg; |
456f998e YH |
947 | |
948 | if (!mm) | |
949 | return; | |
950 | ||
951 | rcu_read_lock(); | |
c0ff4b85 R |
952 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
953 | if (unlikely(!memcg)) | |
456f998e YH |
954 | goto out; |
955 | ||
956 | switch (idx) { | |
957 | case PGMAJFAULT: | |
c0ff4b85 | 958 | mem_cgroup_pgmajfault(memcg, 1); |
456f998e YH |
959 | break; |
960 | case PGFAULT: | |
c0ff4b85 | 961 | mem_cgroup_pgfault(memcg, 1); |
456f998e YH |
962 | break; |
963 | default: | |
964 | BUG(); | |
965 | } | |
966 | out: | |
967 | rcu_read_unlock(); | |
968 | } | |
969 | EXPORT_SYMBOL(mem_cgroup_count_vm_event); | |
970 | ||
08e552c6 KH |
971 | /* |
972 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
973 | * Operations are called by routine of global LRU independently from memcg. | |
974 | * What we have to take care of here is validness of pc->mem_cgroup. | |
975 | * | |
976 | * Changes to pc->mem_cgroup happens when | |
977 | * 1. charge | |
978 | * 2. moving account | |
979 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
980 | * It is added to LRU before charge. | |
981 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
982 | * When moving account, the page is not on LRU. It's isolated. | |
983 | */ | |
4f98a2fe | 984 | |
08e552c6 KH |
985 | void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru) |
986 | { | |
987 | struct page_cgroup *pc; | |
08e552c6 | 988 | struct mem_cgroup_per_zone *mz; |
6d12e2d8 | 989 | |
f8d66542 | 990 | if (mem_cgroup_disabled()) |
08e552c6 KH |
991 | return; |
992 | pc = lookup_page_cgroup(page); | |
993 | /* can happen while we handle swapcache. */ | |
4b3bde4c | 994 | if (!TestClearPageCgroupAcctLRU(pc)) |
08e552c6 | 995 | return; |
4b3bde4c | 996 | VM_BUG_ON(!pc->mem_cgroup); |
544122e5 KH |
997 | /* |
998 | * We don't check PCG_USED bit. It's cleared when the "page" is finally | |
999 | * removed from global LRU. | |
1000 | */ | |
97a6c37b | 1001 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
ece35ca8 KH |
1002 | /* huge page split is done under lru_lock. so, we have no races. */ |
1003 | MEM_CGROUP_ZSTAT(mz, lru) -= 1 << compound_order(page); | |
4b3bde4c BS |
1004 | if (mem_cgroup_is_root(pc->mem_cgroup)) |
1005 | return; | |
1006 | VM_BUG_ON(list_empty(&pc->lru)); | |
08e552c6 | 1007 | list_del_init(&pc->lru); |
6d12e2d8 KH |
1008 | } |
1009 | ||
08e552c6 | 1010 | void mem_cgroup_del_lru(struct page *page) |
6d12e2d8 | 1011 | { |
08e552c6 KH |
1012 | mem_cgroup_del_lru_list(page, page_lru(page)); |
1013 | } | |
b69408e8 | 1014 | |
3f58a829 MK |
1015 | /* |
1016 | * Writeback is about to end against a page which has been marked for immediate | |
1017 | * reclaim. If it still appears to be reclaimable, move it to the tail of the | |
1018 | * inactive list. | |
1019 | */ | |
1020 | void mem_cgroup_rotate_reclaimable_page(struct page *page) | |
1021 | { | |
1022 | struct mem_cgroup_per_zone *mz; | |
1023 | struct page_cgroup *pc; | |
1024 | enum lru_list lru = page_lru(page); | |
1025 | ||
1026 | if (mem_cgroup_disabled()) | |
1027 | return; | |
1028 | ||
1029 | pc = lookup_page_cgroup(page); | |
1030 | /* unused or root page is not rotated. */ | |
1031 | if (!PageCgroupUsed(pc)) | |
1032 | return; | |
1033 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ | |
1034 | smp_rmb(); | |
1035 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
1036 | return; | |
97a6c37b | 1037 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
3f58a829 MK |
1038 | list_move_tail(&pc->lru, &mz->lists[lru]); |
1039 | } | |
1040 | ||
08e552c6 KH |
1041 | void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru) |
1042 | { | |
1043 | struct mem_cgroup_per_zone *mz; | |
1044 | struct page_cgroup *pc; | |
b69408e8 | 1045 | |
f8d66542 | 1046 | if (mem_cgroup_disabled()) |
08e552c6 | 1047 | return; |
6d12e2d8 | 1048 | |
08e552c6 | 1049 | pc = lookup_page_cgroup(page); |
4b3bde4c | 1050 | /* unused or root page is not rotated. */ |
713735b4 JW |
1051 | if (!PageCgroupUsed(pc)) |
1052 | return; | |
1053 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ | |
1054 | smp_rmb(); | |
1055 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
08e552c6 | 1056 | return; |
97a6c37b | 1057 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
08e552c6 | 1058 | list_move(&pc->lru, &mz->lists[lru]); |
6d12e2d8 KH |
1059 | } |
1060 | ||
08e552c6 | 1061 | void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru) |
66e1707b | 1062 | { |
08e552c6 KH |
1063 | struct page_cgroup *pc; |
1064 | struct mem_cgroup_per_zone *mz; | |
6d12e2d8 | 1065 | |
f8d66542 | 1066 | if (mem_cgroup_disabled()) |
08e552c6 KH |
1067 | return; |
1068 | pc = lookup_page_cgroup(page); | |
4b3bde4c | 1069 | VM_BUG_ON(PageCgroupAcctLRU(pc)); |
a61ed3ce JW |
1070 | /* |
1071 | * putback: charge: | |
1072 | * SetPageLRU SetPageCgroupUsed | |
1073 | * smp_mb smp_mb | |
1074 | * PageCgroupUsed && add to memcg LRU PageLRU && add to memcg LRU | |
1075 | * | |
1076 | * Ensure that one of the two sides adds the page to the memcg | |
1077 | * LRU during a race. | |
1078 | */ | |
1079 | smp_mb(); | |
08e552c6 | 1080 | if (!PageCgroupUsed(pc)) |
894bc310 | 1081 | return; |
713735b4 JW |
1082 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ |
1083 | smp_rmb(); | |
97a6c37b | 1084 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
ece35ca8 KH |
1085 | /* huge page split is done under lru_lock. so, we have no races. */ |
1086 | MEM_CGROUP_ZSTAT(mz, lru) += 1 << compound_order(page); | |
4b3bde4c BS |
1087 | SetPageCgroupAcctLRU(pc); |
1088 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
1089 | return; | |
08e552c6 KH |
1090 | list_add(&pc->lru, &mz->lists[lru]); |
1091 | } | |
544122e5 | 1092 | |
08e552c6 | 1093 | /* |
5a6475a4 KH |
1094 | * At handling SwapCache and other FUSE stuff, pc->mem_cgroup may be changed |
1095 | * while it's linked to lru because the page may be reused after it's fully | |
1096 | * uncharged. To handle that, unlink page_cgroup from LRU when charge it again. | |
1097 | * It's done under lock_page and expected that zone->lru_lock isnever held. | |
08e552c6 | 1098 | */ |
5a6475a4 | 1099 | static void mem_cgroup_lru_del_before_commit(struct page *page) |
08e552c6 | 1100 | { |
544122e5 KH |
1101 | unsigned long flags; |
1102 | struct zone *zone = page_zone(page); | |
1103 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
1104 | ||
5a6475a4 KH |
1105 | /* |
1106 | * Doing this check without taking ->lru_lock seems wrong but this | |
1107 | * is safe. Because if page_cgroup's USED bit is unset, the page | |
1108 | * will not be added to any memcg's LRU. If page_cgroup's USED bit is | |
1109 | * set, the commit after this will fail, anyway. | |
1110 | * This all charge/uncharge is done under some mutual execustion. | |
1111 | * So, we don't need to taking care of changes in USED bit. | |
1112 | */ | |
1113 | if (likely(!PageLRU(page))) | |
1114 | return; | |
1115 | ||
544122e5 KH |
1116 | spin_lock_irqsave(&zone->lru_lock, flags); |
1117 | /* | |
1118 | * Forget old LRU when this page_cgroup is *not* used. This Used bit | |
1119 | * is guarded by lock_page() because the page is SwapCache. | |
1120 | */ | |
1121 | if (!PageCgroupUsed(pc)) | |
1122 | mem_cgroup_del_lru_list(page, page_lru(page)); | |
1123 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
08e552c6 KH |
1124 | } |
1125 | ||
5a6475a4 | 1126 | static void mem_cgroup_lru_add_after_commit(struct page *page) |
544122e5 KH |
1127 | { |
1128 | unsigned long flags; | |
1129 | struct zone *zone = page_zone(page); | |
1130 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
a61ed3ce JW |
1131 | /* |
1132 | * putback: charge: | |
1133 | * SetPageLRU SetPageCgroupUsed | |
1134 | * smp_mb smp_mb | |
1135 | * PageCgroupUsed && add to memcg LRU PageLRU && add to memcg LRU | |
1136 | * | |
1137 | * Ensure that one of the two sides adds the page to the memcg | |
1138 | * LRU during a race. | |
1139 | */ | |
1140 | smp_mb(); | |
5a6475a4 KH |
1141 | /* taking care of that the page is added to LRU while we commit it */ |
1142 | if (likely(!PageLRU(page))) | |
1143 | return; | |
544122e5 KH |
1144 | spin_lock_irqsave(&zone->lru_lock, flags); |
1145 | /* link when the page is linked to LRU but page_cgroup isn't */ | |
4b3bde4c | 1146 | if (PageLRU(page) && !PageCgroupAcctLRU(pc)) |
544122e5 KH |
1147 | mem_cgroup_add_lru_list(page, page_lru(page)); |
1148 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
1149 | } | |
1150 | ||
1151 | ||
08e552c6 KH |
1152 | void mem_cgroup_move_lists(struct page *page, |
1153 | enum lru_list from, enum lru_list to) | |
1154 | { | |
f8d66542 | 1155 | if (mem_cgroup_disabled()) |
08e552c6 KH |
1156 | return; |
1157 | mem_cgroup_del_lru_list(page, from); | |
1158 | mem_cgroup_add_lru_list(page, to); | |
66e1707b BS |
1159 | } |
1160 | ||
3e92041d | 1161 | /* |
c0ff4b85 | 1162 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1163 | * hierarchy subtree |
1164 | */ | |
c0ff4b85 R |
1165 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1166 | struct mem_cgroup *memcg) | |
3e92041d | 1167 | { |
c0ff4b85 R |
1168 | if (root_memcg != memcg) { |
1169 | return (root_memcg->use_hierarchy && | |
1170 | css_is_ancestor(&memcg->css, &root_memcg->css)); | |
3e92041d MH |
1171 | } |
1172 | ||
1173 | return true; | |
1174 | } | |
1175 | ||
c0ff4b85 | 1176 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg) |
4c4a2214 DR |
1177 | { |
1178 | int ret; | |
0b7f569e | 1179 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1180 | struct task_struct *p; |
4c4a2214 | 1181 | |
158e0a2d KH |
1182 | p = find_lock_task_mm(task); |
1183 | if (!p) | |
1184 | return 0; | |
1185 | curr = try_get_mem_cgroup_from_mm(p->mm); | |
1186 | task_unlock(p); | |
0b7f569e KH |
1187 | if (!curr) |
1188 | return 0; | |
d31f56db | 1189 | /* |
c0ff4b85 | 1190 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1191 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1192 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1193 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1194 | */ |
c0ff4b85 | 1195 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1196 | css_put(&curr->css); |
4c4a2214 DR |
1197 | return ret; |
1198 | } | |
1199 | ||
9b272977 | 1200 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg, struct zone *zone) |
14797e23 | 1201 | { |
9b272977 JW |
1202 | unsigned long inactive_ratio; |
1203 | int nid = zone_to_nid(zone); | |
1204 | int zid = zone_idx(zone); | |
14797e23 | 1205 | unsigned long inactive; |
9b272977 | 1206 | unsigned long active; |
c772be93 | 1207 | unsigned long gb; |
14797e23 | 1208 | |
9b272977 JW |
1209 | inactive = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid, |
1210 | BIT(LRU_INACTIVE_ANON)); | |
1211 | active = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid, | |
1212 | BIT(LRU_ACTIVE_ANON)); | |
14797e23 | 1213 | |
c772be93 KM |
1214 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1215 | if (gb) | |
1216 | inactive_ratio = int_sqrt(10 * gb); | |
1217 | else | |
1218 | inactive_ratio = 1; | |
1219 | ||
9b272977 | 1220 | return inactive * inactive_ratio < active; |
14797e23 KM |
1221 | } |
1222 | ||
9b272977 | 1223 | int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg, struct zone *zone) |
56e49d21 RR |
1224 | { |
1225 | unsigned long active; | |
1226 | unsigned long inactive; | |
9b272977 JW |
1227 | int zid = zone_idx(zone); |
1228 | int nid = zone_to_nid(zone); | |
56e49d21 | 1229 | |
9b272977 JW |
1230 | inactive = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid, |
1231 | BIT(LRU_INACTIVE_FILE)); | |
1232 | active = mem_cgroup_zone_nr_lru_pages(memcg, nid, zid, | |
1233 | BIT(LRU_ACTIVE_FILE)); | |
56e49d21 RR |
1234 | |
1235 | return (active > inactive); | |
1236 | } | |
1237 | ||
3e2f41f1 KM |
1238 | struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, |
1239 | struct zone *zone) | |
1240 | { | |
13d7e3a2 | 1241 | int nid = zone_to_nid(zone); |
3e2f41f1 KM |
1242 | int zid = zone_idx(zone); |
1243 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
1244 | ||
1245 | return &mz->reclaim_stat; | |
1246 | } | |
1247 | ||
1248 | struct zone_reclaim_stat * | |
1249 | mem_cgroup_get_reclaim_stat_from_page(struct page *page) | |
1250 | { | |
1251 | struct page_cgroup *pc; | |
1252 | struct mem_cgroup_per_zone *mz; | |
1253 | ||
1254 | if (mem_cgroup_disabled()) | |
1255 | return NULL; | |
1256 | ||
1257 | pc = lookup_page_cgroup(page); | |
bd112db8 DN |
1258 | if (!PageCgroupUsed(pc)) |
1259 | return NULL; | |
713735b4 JW |
1260 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ |
1261 | smp_rmb(); | |
97a6c37b | 1262 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
3e2f41f1 KM |
1263 | return &mz->reclaim_stat; |
1264 | } | |
1265 | ||
66e1707b BS |
1266 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
1267 | struct list_head *dst, | |
1268 | unsigned long *scanned, int order, | |
4356f21d MK |
1269 | isolate_mode_t mode, |
1270 | struct zone *z, | |
66e1707b | 1271 | struct mem_cgroup *mem_cont, |
4f98a2fe | 1272 | int active, int file) |
66e1707b BS |
1273 | { |
1274 | unsigned long nr_taken = 0; | |
1275 | struct page *page; | |
1276 | unsigned long scan; | |
1277 | LIST_HEAD(pc_list); | |
1278 | struct list_head *src; | |
ff7283fa | 1279 | struct page_cgroup *pc, *tmp; |
13d7e3a2 | 1280 | int nid = zone_to_nid(z); |
1ecaab2b KH |
1281 | int zid = zone_idx(z); |
1282 | struct mem_cgroup_per_zone *mz; | |
b7c46d15 | 1283 | int lru = LRU_FILE * file + active; |
2ffebca6 | 1284 | int ret; |
66e1707b | 1285 | |
cf475ad2 | 1286 | BUG_ON(!mem_cont); |
1ecaab2b | 1287 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
b69408e8 | 1288 | src = &mz->lists[lru]; |
66e1707b | 1289 | |
ff7283fa KH |
1290 | scan = 0; |
1291 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 1292 | if (scan >= nr_to_scan) |
ff7283fa | 1293 | break; |
08e552c6 | 1294 | |
52d4b9ac KH |
1295 | if (unlikely(!PageCgroupUsed(pc))) |
1296 | continue; | |
5564e88b | 1297 | |
6b3ae58e | 1298 | page = lookup_cgroup_page(pc); |
5564e88b | 1299 | |
436c6541 | 1300 | if (unlikely(!PageLRU(page))) |
ff7283fa | 1301 | continue; |
ff7283fa | 1302 | |
436c6541 | 1303 | scan++; |
2ffebca6 KH |
1304 | ret = __isolate_lru_page(page, mode, file); |
1305 | switch (ret) { | |
1306 | case 0: | |
66e1707b | 1307 | list_move(&page->lru, dst); |
2ffebca6 | 1308 | mem_cgroup_del_lru(page); |
2c888cfb | 1309 | nr_taken += hpage_nr_pages(page); |
2ffebca6 KH |
1310 | break; |
1311 | case -EBUSY: | |
1312 | /* we don't affect global LRU but rotate in our LRU */ | |
1313 | mem_cgroup_rotate_lru_list(page, page_lru(page)); | |
1314 | break; | |
1315 | default: | |
1316 | break; | |
66e1707b BS |
1317 | } |
1318 | } | |
1319 | ||
66e1707b | 1320 | *scanned = scan; |
cc8e970c KM |
1321 | |
1322 | trace_mm_vmscan_memcg_isolate(0, nr_to_scan, scan, nr_taken, | |
1323 | 0, 0, 0, mode); | |
1324 | ||
66e1707b BS |
1325 | return nr_taken; |
1326 | } | |
1327 | ||
6d61ef40 BS |
1328 | #define mem_cgroup_from_res_counter(counter, member) \ |
1329 | container_of(counter, struct mem_cgroup, member) | |
1330 | ||
19942822 | 1331 | /** |
9d11ea9f JW |
1332 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
1333 | * @mem: the memory cgroup | |
19942822 | 1334 | * |
9d11ea9f | 1335 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1336 | * pages. |
19942822 | 1337 | */ |
c0ff4b85 | 1338 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1339 | { |
9d11ea9f JW |
1340 | unsigned long long margin; |
1341 | ||
c0ff4b85 | 1342 | margin = res_counter_margin(&memcg->res); |
9d11ea9f | 1343 | if (do_swap_account) |
c0ff4b85 | 1344 | margin = min(margin, res_counter_margin(&memcg->memsw)); |
7ec99d62 | 1345 | return margin >> PAGE_SHIFT; |
19942822 JW |
1346 | } |
1347 | ||
1f4c025b | 1348 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 KM |
1349 | { |
1350 | struct cgroup *cgrp = memcg->css.cgroup; | |
a7885eb8 KM |
1351 | |
1352 | /* root ? */ | |
1353 | if (cgrp->parent == NULL) | |
1354 | return vm_swappiness; | |
1355 | ||
bf1ff263 | 1356 | return memcg->swappiness; |
a7885eb8 KM |
1357 | } |
1358 | ||
c0ff4b85 | 1359 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a KH |
1360 | { |
1361 | int cpu; | |
1489ebad KH |
1362 | |
1363 | get_online_cpus(); | |
c0ff4b85 | 1364 | spin_lock(&memcg->pcp_counter_lock); |
1489ebad | 1365 | for_each_online_cpu(cpu) |
c0ff4b85 R |
1366 | per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) += 1; |
1367 | memcg->nocpu_base.count[MEM_CGROUP_ON_MOVE] += 1; | |
1368 | spin_unlock(&memcg->pcp_counter_lock); | |
1489ebad | 1369 | put_online_cpus(); |
32047e2a KH |
1370 | |
1371 | synchronize_rcu(); | |
1372 | } | |
1373 | ||
c0ff4b85 | 1374 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a KH |
1375 | { |
1376 | int cpu; | |
1377 | ||
c0ff4b85 | 1378 | if (!memcg) |
32047e2a | 1379 | return; |
1489ebad | 1380 | get_online_cpus(); |
c0ff4b85 | 1381 | spin_lock(&memcg->pcp_counter_lock); |
1489ebad | 1382 | for_each_online_cpu(cpu) |
c0ff4b85 R |
1383 | per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) -= 1; |
1384 | memcg->nocpu_base.count[MEM_CGROUP_ON_MOVE] -= 1; | |
1385 | spin_unlock(&memcg->pcp_counter_lock); | |
1489ebad | 1386 | put_online_cpus(); |
32047e2a KH |
1387 | } |
1388 | /* | |
1389 | * 2 routines for checking "mem" is under move_account() or not. | |
1390 | * | |
1391 | * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used | |
1392 | * for avoiding race in accounting. If true, | |
1393 | * pc->mem_cgroup may be overwritten. | |
1394 | * | |
1395 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or | |
1396 | * under hierarchy of moving cgroups. This is for | |
1397 | * waiting at hith-memory prressure caused by "move". | |
1398 | */ | |
1399 | ||
c0ff4b85 | 1400 | static bool mem_cgroup_stealed(struct mem_cgroup *memcg) |
32047e2a KH |
1401 | { |
1402 | VM_BUG_ON(!rcu_read_lock_held()); | |
c0ff4b85 | 1403 | return this_cpu_read(memcg->stat->count[MEM_CGROUP_ON_MOVE]) > 0; |
32047e2a | 1404 | } |
4b534334 | 1405 | |
c0ff4b85 | 1406 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1407 | { |
2bd9bb20 KH |
1408 | struct mem_cgroup *from; |
1409 | struct mem_cgroup *to; | |
4b534334 | 1410 | bool ret = false; |
2bd9bb20 KH |
1411 | /* |
1412 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1413 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1414 | */ | |
1415 | spin_lock(&mc.lock); | |
1416 | from = mc.from; | |
1417 | to = mc.to; | |
1418 | if (!from) | |
1419 | goto unlock; | |
3e92041d | 1420 | |
c0ff4b85 R |
1421 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1422 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1423 | unlock: |
1424 | spin_unlock(&mc.lock); | |
4b534334 KH |
1425 | return ret; |
1426 | } | |
1427 | ||
c0ff4b85 | 1428 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1429 | { |
1430 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1431 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1432 | DEFINE_WAIT(wait); |
1433 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1434 | /* moving charge context might have finished. */ | |
1435 | if (mc.moving_task) | |
1436 | schedule(); | |
1437 | finish_wait(&mc.waitq, &wait); | |
1438 | return true; | |
1439 | } | |
1440 | } | |
1441 | return false; | |
1442 | } | |
1443 | ||
e222432b | 1444 | /** |
6a6135b6 | 1445 | * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode. |
e222432b BS |
1446 | * @memcg: The memory cgroup that went over limit |
1447 | * @p: Task that is going to be killed | |
1448 | * | |
1449 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1450 | * enabled | |
1451 | */ | |
1452 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1453 | { | |
1454 | struct cgroup *task_cgrp; | |
1455 | struct cgroup *mem_cgrp; | |
1456 | /* | |
1457 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1458 | * on the assumption that OOM is serialized for memory controller. | |
1459 | * If this assumption is broken, revisit this code. | |
1460 | */ | |
1461 | static char memcg_name[PATH_MAX]; | |
1462 | int ret; | |
1463 | ||
d31f56db | 1464 | if (!memcg || !p) |
e222432b BS |
1465 | return; |
1466 | ||
1467 | ||
1468 | rcu_read_lock(); | |
1469 | ||
1470 | mem_cgrp = memcg->css.cgroup; | |
1471 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1472 | ||
1473 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1474 | if (ret < 0) { | |
1475 | /* | |
1476 | * Unfortunately, we are unable to convert to a useful name | |
1477 | * But we'll still print out the usage information | |
1478 | */ | |
1479 | rcu_read_unlock(); | |
1480 | goto done; | |
1481 | } | |
1482 | rcu_read_unlock(); | |
1483 | ||
1484 | printk(KERN_INFO "Task in %s killed", memcg_name); | |
1485 | ||
1486 | rcu_read_lock(); | |
1487 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1488 | if (ret < 0) { | |
1489 | rcu_read_unlock(); | |
1490 | goto done; | |
1491 | } | |
1492 | rcu_read_unlock(); | |
1493 | ||
1494 | /* | |
1495 | * Continues from above, so we don't need an KERN_ level | |
1496 | */ | |
1497 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); | |
1498 | done: | |
1499 | ||
1500 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", | |
1501 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, | |
1502 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1503 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
1504 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " | |
1505 | "failcnt %llu\n", | |
1506 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, | |
1507 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1508 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
1509 | } | |
1510 | ||
81d39c20 KH |
1511 | /* |
1512 | * This function returns the number of memcg under hierarchy tree. Returns | |
1513 | * 1(self count) if no children. | |
1514 | */ | |
c0ff4b85 | 1515 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1516 | { |
1517 | int num = 0; | |
7d74b06f KH |
1518 | struct mem_cgroup *iter; |
1519 | ||
c0ff4b85 | 1520 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1521 | num++; |
81d39c20 KH |
1522 | return num; |
1523 | } | |
1524 | ||
a63d83f4 DR |
1525 | /* |
1526 | * Return the memory (and swap, if configured) limit for a memcg. | |
1527 | */ | |
1528 | u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) | |
1529 | { | |
1530 | u64 limit; | |
1531 | u64 memsw; | |
1532 | ||
f3e8eb70 JW |
1533 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
1534 | limit += total_swap_pages << PAGE_SHIFT; | |
1535 | ||
a63d83f4 DR |
1536 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
1537 | /* | |
1538 | * If memsw is finite and limits the amount of swap space available | |
1539 | * to this memcg, return that limit. | |
1540 | */ | |
1541 | return min(limit, memsw); | |
1542 | } | |
1543 | ||
6d61ef40 | 1544 | /* |
04046e1a KH |
1545 | * Visit the first child (need not be the first child as per the ordering |
1546 | * of the cgroup list, since we track last_scanned_child) of @mem and use | |
1547 | * that to reclaim free pages from. | |
1548 | */ | |
1549 | static struct mem_cgroup * | |
c0ff4b85 | 1550 | mem_cgroup_select_victim(struct mem_cgroup *root_memcg) |
04046e1a KH |
1551 | { |
1552 | struct mem_cgroup *ret = NULL; | |
1553 | struct cgroup_subsys_state *css; | |
1554 | int nextid, found; | |
1555 | ||
c0ff4b85 R |
1556 | if (!root_memcg->use_hierarchy) { |
1557 | css_get(&root_memcg->css); | |
1558 | ret = root_memcg; | |
04046e1a KH |
1559 | } |
1560 | ||
1561 | while (!ret) { | |
1562 | rcu_read_lock(); | |
c0ff4b85 R |
1563 | nextid = root_memcg->last_scanned_child + 1; |
1564 | css = css_get_next(&mem_cgroup_subsys, nextid, &root_memcg->css, | |
04046e1a KH |
1565 | &found); |
1566 | if (css && css_tryget(css)) | |
1567 | ret = container_of(css, struct mem_cgroup, css); | |
1568 | ||
1569 | rcu_read_unlock(); | |
1570 | /* Updates scanning parameter */ | |
04046e1a KH |
1571 | if (!css) { |
1572 | /* this means start scan from ID:1 */ | |
c0ff4b85 | 1573 | root_memcg->last_scanned_child = 0; |
04046e1a | 1574 | } else |
c0ff4b85 | 1575 | root_memcg->last_scanned_child = found; |
04046e1a KH |
1576 | } |
1577 | ||
1578 | return ret; | |
1579 | } | |
1580 | ||
4d0c066d KH |
1581 | /** |
1582 | * test_mem_cgroup_node_reclaimable | |
1583 | * @mem: the target memcg | |
1584 | * @nid: the node ID to be checked. | |
1585 | * @noswap : specify true here if the user wants flle only information. | |
1586 | * | |
1587 | * This function returns whether the specified memcg contains any | |
1588 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1589 | * pages in the node. | |
1590 | */ | |
c0ff4b85 | 1591 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1592 | int nid, bool noswap) |
1593 | { | |
c0ff4b85 | 1594 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1595 | return true; |
1596 | if (noswap || !total_swap_pages) | |
1597 | return false; | |
c0ff4b85 | 1598 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1599 | return true; |
1600 | return false; | |
1601 | ||
1602 | } | |
889976db YH |
1603 | #if MAX_NUMNODES > 1 |
1604 | ||
1605 | /* | |
1606 | * Always updating the nodemask is not very good - even if we have an empty | |
1607 | * list or the wrong list here, we can start from some node and traverse all | |
1608 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1609 | * | |
1610 | */ | |
c0ff4b85 | 1611 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1612 | { |
1613 | int nid; | |
453a9bf3 KH |
1614 | /* |
1615 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1616 | * pagein/pageout changes since the last update. | |
1617 | */ | |
c0ff4b85 | 1618 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1619 | return; |
c0ff4b85 | 1620 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1621 | return; |
1622 | ||
889976db | 1623 | /* make a nodemask where this memcg uses memory from */ |
c0ff4b85 | 1624 | memcg->scan_nodes = node_states[N_HIGH_MEMORY]; |
889976db YH |
1625 | |
1626 | for_each_node_mask(nid, node_states[N_HIGH_MEMORY]) { | |
1627 | ||
c0ff4b85 R |
1628 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1629 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1630 | } |
453a9bf3 | 1631 | |
c0ff4b85 R |
1632 | atomic_set(&memcg->numainfo_events, 0); |
1633 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1634 | } |
1635 | ||
1636 | /* | |
1637 | * Selecting a node where we start reclaim from. Because what we need is just | |
1638 | * reducing usage counter, start from anywhere is O,K. Considering | |
1639 | * memory reclaim from current node, there are pros. and cons. | |
1640 | * | |
1641 | * Freeing memory from current node means freeing memory from a node which | |
1642 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1643 | * hit limits, it will see a contention on a node. But freeing from remote | |
1644 | * node means more costs for memory reclaim because of memory latency. | |
1645 | * | |
1646 | * Now, we use round-robin. Better algorithm is welcomed. | |
1647 | */ | |
c0ff4b85 | 1648 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1649 | { |
1650 | int node; | |
1651 | ||
c0ff4b85 R |
1652 | mem_cgroup_may_update_nodemask(memcg); |
1653 | node = memcg->last_scanned_node; | |
889976db | 1654 | |
c0ff4b85 | 1655 | node = next_node(node, memcg->scan_nodes); |
889976db | 1656 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1657 | node = first_node(memcg->scan_nodes); |
889976db YH |
1658 | /* |
1659 | * We call this when we hit limit, not when pages are added to LRU. | |
1660 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1661 | * memcg is too small and all pages are not on LRU. In that case, | |
1662 | * we use curret node. | |
1663 | */ | |
1664 | if (unlikely(node == MAX_NUMNODES)) | |
1665 | node = numa_node_id(); | |
1666 | ||
c0ff4b85 | 1667 | memcg->last_scanned_node = node; |
889976db YH |
1668 | return node; |
1669 | } | |
1670 | ||
4d0c066d KH |
1671 | /* |
1672 | * Check all nodes whether it contains reclaimable pages or not. | |
1673 | * For quick scan, we make use of scan_nodes. This will allow us to skip | |
1674 | * unused nodes. But scan_nodes is lazily updated and may not cotain | |
1675 | * enough new information. We need to do double check. | |
1676 | */ | |
c0ff4b85 | 1677 | bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d KH |
1678 | { |
1679 | int nid; | |
1680 | ||
1681 | /* | |
1682 | * quick check...making use of scan_node. | |
1683 | * We can skip unused nodes. | |
1684 | */ | |
c0ff4b85 R |
1685 | if (!nodes_empty(memcg->scan_nodes)) { |
1686 | for (nid = first_node(memcg->scan_nodes); | |
4d0c066d | 1687 | nid < MAX_NUMNODES; |
c0ff4b85 | 1688 | nid = next_node(nid, memcg->scan_nodes)) { |
4d0c066d | 1689 | |
c0ff4b85 | 1690 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
1691 | return true; |
1692 | } | |
1693 | } | |
1694 | /* | |
1695 | * Check rest of nodes. | |
1696 | */ | |
1697 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
c0ff4b85 | 1698 | if (node_isset(nid, memcg->scan_nodes)) |
4d0c066d | 1699 | continue; |
c0ff4b85 | 1700 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
1701 | return true; |
1702 | } | |
1703 | return false; | |
1704 | } | |
1705 | ||
889976db | 1706 | #else |
c0ff4b85 | 1707 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1708 | { |
1709 | return 0; | |
1710 | } | |
4d0c066d | 1711 | |
c0ff4b85 | 1712 | bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d | 1713 | { |
c0ff4b85 | 1714 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); |
4d0c066d | 1715 | } |
889976db YH |
1716 | #endif |
1717 | ||
04046e1a KH |
1718 | /* |
1719 | * Scan the hierarchy if needed to reclaim memory. We remember the last child | |
1720 | * we reclaimed from, so that we don't end up penalizing one child extensively | |
1721 | * based on its position in the children list. | |
6d61ef40 | 1722 | * |
c0ff4b85 | 1723 | * root_memcg is the original ancestor that we've been reclaim from. |
04046e1a | 1724 | * |
c0ff4b85 | 1725 | * We give up and return to the caller when we visit root_memcg twice. |
04046e1a | 1726 | * (other groups can be removed while we're walking....) |
81d39c20 KH |
1727 | * |
1728 | * If shrink==true, for avoiding to free too much, this returns immedieately. | |
6d61ef40 | 1729 | */ |
c0ff4b85 | 1730 | static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_memcg, |
4e416953 | 1731 | struct zone *zone, |
75822b44 | 1732 | gfp_t gfp_mask, |
0ae5e89c YH |
1733 | unsigned long reclaim_options, |
1734 | unsigned long *total_scanned) | |
6d61ef40 | 1735 | { |
04046e1a KH |
1736 | struct mem_cgroup *victim; |
1737 | int ret, total = 0; | |
1738 | int loop = 0; | |
75822b44 BS |
1739 | bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP; |
1740 | bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK; | |
4e416953 | 1741 | bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT; |
9d11ea9f | 1742 | unsigned long excess; |
185efc0f | 1743 | unsigned long nr_scanned; |
9d11ea9f | 1744 | |
c0ff4b85 | 1745 | excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT; |
04046e1a | 1746 | |
22a668d7 | 1747 | /* If memsw_is_minimum==1, swap-out is of-no-use. */ |
c0ff4b85 | 1748 | if (!check_soft && !shrink && root_memcg->memsw_is_minimum) |
22a668d7 KH |
1749 | noswap = true; |
1750 | ||
4e416953 | 1751 | while (1) { |
c0ff4b85 R |
1752 | victim = mem_cgroup_select_victim(root_memcg); |
1753 | if (victim == root_memcg) { | |
04046e1a | 1754 | loop++; |
fbc29a25 KH |
1755 | /* |
1756 | * We are not draining per cpu cached charges during | |
1757 | * soft limit reclaim because global reclaim doesn't | |
1758 | * care about charges. It tries to free some memory and | |
1759 | * charges will not give any. | |
1760 | */ | |
1761 | if (!check_soft && loop >= 1) | |
c0ff4b85 | 1762 | drain_all_stock_async(root_memcg); |
4e416953 BS |
1763 | if (loop >= 2) { |
1764 | /* | |
1765 | * If we have not been able to reclaim | |
1766 | * anything, it might because there are | |
1767 | * no reclaimable pages under this hierarchy | |
1768 | */ | |
1769 | if (!check_soft || !total) { | |
1770 | css_put(&victim->css); | |
1771 | break; | |
1772 | } | |
1773 | /* | |
25985edc | 1774 | * We want to do more targeted reclaim. |
4e416953 BS |
1775 | * excess >> 2 is not to excessive so as to |
1776 | * reclaim too much, nor too less that we keep | |
1777 | * coming back to reclaim from this cgroup | |
1778 | */ | |
1779 | if (total >= (excess >> 2) || | |
1780 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) { | |
1781 | css_put(&victim->css); | |
1782 | break; | |
1783 | } | |
1784 | } | |
1785 | } | |
4d0c066d | 1786 | if (!mem_cgroup_reclaimable(victim, noswap)) { |
04046e1a KH |
1787 | /* this cgroup's local usage == 0 */ |
1788 | css_put(&victim->css); | |
6d61ef40 BS |
1789 | continue; |
1790 | } | |
04046e1a | 1791 | /* we use swappiness of local cgroup */ |
0ae5e89c | 1792 | if (check_soft) { |
4e416953 | 1793 | ret = mem_cgroup_shrink_node_zone(victim, gfp_mask, |
185efc0f JW |
1794 | noswap, zone, &nr_scanned); |
1795 | *total_scanned += nr_scanned; | |
0ae5e89c | 1796 | } else |
4e416953 | 1797 | ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, |
185efc0f | 1798 | noswap); |
04046e1a | 1799 | css_put(&victim->css); |
81d39c20 KH |
1800 | /* |
1801 | * At shrinking usage, we can't check we should stop here or | |
1802 | * reclaim more. It's depends on callers. last_scanned_child | |
1803 | * will work enough for keeping fairness under tree. | |
1804 | */ | |
1805 | if (shrink) | |
1806 | return ret; | |
04046e1a | 1807 | total += ret; |
4e416953 | 1808 | if (check_soft) { |
c0ff4b85 | 1809 | if (!res_counter_soft_limit_excess(&root_memcg->res)) |
4e416953 | 1810 | return total; |
c0ff4b85 | 1811 | } else if (mem_cgroup_margin(root_memcg)) |
4fd14ebf | 1812 | return total; |
6d61ef40 | 1813 | } |
04046e1a | 1814 | return total; |
6d61ef40 BS |
1815 | } |
1816 | ||
867578cb KH |
1817 | /* |
1818 | * Check OOM-Killer is already running under our hierarchy. | |
1819 | * If someone is running, return false. | |
1af8efe9 | 1820 | * Has to be called with memcg_oom_lock |
867578cb | 1821 | */ |
c0ff4b85 | 1822 | static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg) |
867578cb | 1823 | { |
79dfdacc MH |
1824 | struct mem_cgroup *iter, *failed = NULL; |
1825 | bool cond = true; | |
a636b327 | 1826 | |
c0ff4b85 | 1827 | for_each_mem_cgroup_tree_cond(iter, memcg, cond) { |
23751be0 | 1828 | if (iter->oom_lock) { |
79dfdacc MH |
1829 | /* |
1830 | * this subtree of our hierarchy is already locked | |
1831 | * so we cannot give a lock. | |
1832 | */ | |
79dfdacc MH |
1833 | failed = iter; |
1834 | cond = false; | |
23751be0 JW |
1835 | } else |
1836 | iter->oom_lock = true; | |
7d74b06f | 1837 | } |
867578cb | 1838 | |
79dfdacc | 1839 | if (!failed) |
23751be0 | 1840 | return true; |
79dfdacc MH |
1841 | |
1842 | /* | |
1843 | * OK, we failed to lock the whole subtree so we have to clean up | |
1844 | * what we set up to the failing subtree | |
1845 | */ | |
1846 | cond = true; | |
c0ff4b85 | 1847 | for_each_mem_cgroup_tree_cond(iter, memcg, cond) { |
79dfdacc MH |
1848 | if (iter == failed) { |
1849 | cond = false; | |
1850 | continue; | |
1851 | } | |
1852 | iter->oom_lock = false; | |
1853 | } | |
23751be0 | 1854 | return false; |
a636b327 | 1855 | } |
0b7f569e | 1856 | |
79dfdacc | 1857 | /* |
1af8efe9 | 1858 | * Has to be called with memcg_oom_lock |
79dfdacc | 1859 | */ |
c0ff4b85 | 1860 | static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1861 | { |
7d74b06f KH |
1862 | struct mem_cgroup *iter; |
1863 | ||
c0ff4b85 | 1864 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1865 | iter->oom_lock = false; |
1866 | return 0; | |
1867 | } | |
1868 | ||
c0ff4b85 | 1869 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1870 | { |
1871 | struct mem_cgroup *iter; | |
1872 | ||
c0ff4b85 | 1873 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1874 | atomic_inc(&iter->under_oom); |
1875 | } | |
1876 | ||
c0ff4b85 | 1877 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1878 | { |
1879 | struct mem_cgroup *iter; | |
1880 | ||
867578cb KH |
1881 | /* |
1882 | * When a new child is created while the hierarchy is under oom, | |
1883 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1884 | * atomic_add_unless() here. | |
1885 | */ | |
c0ff4b85 | 1886 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1887 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
1888 | } |
1889 | ||
1af8efe9 | 1890 | static DEFINE_SPINLOCK(memcg_oom_lock); |
867578cb KH |
1891 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1892 | ||
dc98df5a KH |
1893 | struct oom_wait_info { |
1894 | struct mem_cgroup *mem; | |
1895 | wait_queue_t wait; | |
1896 | }; | |
1897 | ||
1898 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1899 | unsigned mode, int sync, void *arg) | |
1900 | { | |
c0ff4b85 R |
1901 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg, |
1902 | *oom_wait_memcg; | |
dc98df5a KH |
1903 | struct oom_wait_info *oom_wait_info; |
1904 | ||
1905 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
c0ff4b85 | 1906 | oom_wait_memcg = oom_wait_info->mem; |
dc98df5a | 1907 | |
dc98df5a KH |
1908 | /* |
1909 | * Both of oom_wait_info->mem and wake_mem are stable under us. | |
1910 | * Then we can use css_is_ancestor without taking care of RCU. | |
1911 | */ | |
c0ff4b85 R |
1912 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
1913 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 1914 | return 0; |
dc98df5a KH |
1915 | return autoremove_wake_function(wait, mode, sync, arg); |
1916 | } | |
1917 | ||
c0ff4b85 | 1918 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 1919 | { |
c0ff4b85 R |
1920 | /* for filtering, pass "memcg" as argument. */ |
1921 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
1922 | } |
1923 | ||
c0ff4b85 | 1924 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1925 | { |
c0ff4b85 R |
1926 | if (memcg && atomic_read(&memcg->under_oom)) |
1927 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
1928 | } |
1929 | ||
867578cb KH |
1930 | /* |
1931 | * try to call OOM killer. returns false if we should exit memory-reclaim loop. | |
1932 | */ | |
c0ff4b85 | 1933 | bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask) |
0b7f569e | 1934 | { |
dc98df5a | 1935 | struct oom_wait_info owait; |
3c11ecf4 | 1936 | bool locked, need_to_kill; |
867578cb | 1937 | |
c0ff4b85 | 1938 | owait.mem = memcg; |
dc98df5a KH |
1939 | owait.wait.flags = 0; |
1940 | owait.wait.func = memcg_oom_wake_function; | |
1941 | owait.wait.private = current; | |
1942 | INIT_LIST_HEAD(&owait.wait.task_list); | |
3c11ecf4 | 1943 | need_to_kill = true; |
c0ff4b85 | 1944 | mem_cgroup_mark_under_oom(memcg); |
79dfdacc | 1945 | |
c0ff4b85 | 1946 | /* At first, try to OOM lock hierarchy under memcg.*/ |
1af8efe9 | 1947 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1948 | locked = mem_cgroup_oom_lock(memcg); |
867578cb KH |
1949 | /* |
1950 | * Even if signal_pending(), we can't quit charge() loop without | |
1951 | * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL | |
1952 | * under OOM is always welcomed, use TASK_KILLABLE here. | |
1953 | */ | |
3c11ecf4 | 1954 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
c0ff4b85 | 1955 | if (!locked || memcg->oom_kill_disable) |
3c11ecf4 KH |
1956 | need_to_kill = false; |
1957 | if (locked) | |
c0ff4b85 | 1958 | mem_cgroup_oom_notify(memcg); |
1af8efe9 | 1959 | spin_unlock(&memcg_oom_lock); |
867578cb | 1960 | |
3c11ecf4 KH |
1961 | if (need_to_kill) { |
1962 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
c0ff4b85 | 1963 | mem_cgroup_out_of_memory(memcg, mask); |
3c11ecf4 | 1964 | } else { |
867578cb | 1965 | schedule(); |
dc98df5a | 1966 | finish_wait(&memcg_oom_waitq, &owait.wait); |
867578cb | 1967 | } |
1af8efe9 | 1968 | spin_lock(&memcg_oom_lock); |
79dfdacc | 1969 | if (locked) |
c0ff4b85 R |
1970 | mem_cgroup_oom_unlock(memcg); |
1971 | memcg_wakeup_oom(memcg); | |
1af8efe9 | 1972 | spin_unlock(&memcg_oom_lock); |
867578cb | 1973 | |
c0ff4b85 | 1974 | mem_cgroup_unmark_under_oom(memcg); |
79dfdacc | 1975 | |
867578cb KH |
1976 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) |
1977 | return false; | |
1978 | /* Give chance to dying process */ | |
715a5ee8 | 1979 | schedule_timeout_uninterruptible(1); |
867578cb | 1980 | return true; |
0b7f569e KH |
1981 | } |
1982 | ||
d69b042f BS |
1983 | /* |
1984 | * Currently used to update mapped file statistics, but the routine can be | |
1985 | * generalized to update other statistics as well. | |
32047e2a KH |
1986 | * |
1987 | * Notes: Race condition | |
1988 | * | |
1989 | * We usually use page_cgroup_lock() for accessing page_cgroup member but | |
1990 | * it tends to be costly. But considering some conditions, we doesn't need | |
1991 | * to do so _always_. | |
1992 | * | |
1993 | * Considering "charge", lock_page_cgroup() is not required because all | |
1994 | * file-stat operations happen after a page is attached to radix-tree. There | |
1995 | * are no race with "charge". | |
1996 | * | |
1997 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
1998 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
1999 | * if there are race with "uncharge". Statistics itself is properly handled | |
2000 | * by flags. | |
2001 | * | |
2002 | * Considering "move", this is an only case we see a race. To make the race | |
2003 | * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are | |
2004 | * possibility of race condition. If there is, we take a lock. | |
d69b042f | 2005 | */ |
26174efd | 2006 | |
2a7106f2 GT |
2007 | void mem_cgroup_update_page_stat(struct page *page, |
2008 | enum mem_cgroup_page_stat_item idx, int val) | |
d69b042f | 2009 | { |
c0ff4b85 | 2010 | struct mem_cgroup *memcg; |
32047e2a KH |
2011 | struct page_cgroup *pc = lookup_page_cgroup(page); |
2012 | bool need_unlock = false; | |
dbd4ea78 | 2013 | unsigned long uninitialized_var(flags); |
d69b042f | 2014 | |
d69b042f BS |
2015 | if (unlikely(!pc)) |
2016 | return; | |
2017 | ||
32047e2a | 2018 | rcu_read_lock(); |
c0ff4b85 R |
2019 | memcg = pc->mem_cgroup; |
2020 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
32047e2a KH |
2021 | goto out; |
2022 | /* pc->mem_cgroup is unstable ? */ | |
c0ff4b85 | 2023 | if (unlikely(mem_cgroup_stealed(memcg)) || PageTransHuge(page)) { |
32047e2a | 2024 | /* take a lock against to access pc->mem_cgroup */ |
dbd4ea78 | 2025 | move_lock_page_cgroup(pc, &flags); |
32047e2a | 2026 | need_unlock = true; |
c0ff4b85 R |
2027 | memcg = pc->mem_cgroup; |
2028 | if (!memcg || !PageCgroupUsed(pc)) | |
32047e2a KH |
2029 | goto out; |
2030 | } | |
26174efd | 2031 | |
26174efd | 2032 | switch (idx) { |
2a7106f2 | 2033 | case MEMCG_NR_FILE_MAPPED: |
26174efd KH |
2034 | if (val > 0) |
2035 | SetPageCgroupFileMapped(pc); | |
2036 | else if (!page_mapped(page)) | |
0c270f8f | 2037 | ClearPageCgroupFileMapped(pc); |
2a7106f2 | 2038 | idx = MEM_CGROUP_STAT_FILE_MAPPED; |
26174efd KH |
2039 | break; |
2040 | default: | |
2041 | BUG(); | |
8725d541 | 2042 | } |
d69b042f | 2043 | |
c0ff4b85 | 2044 | this_cpu_add(memcg->stat->count[idx], val); |
2a7106f2 | 2045 | |
32047e2a KH |
2046 | out: |
2047 | if (unlikely(need_unlock)) | |
dbd4ea78 | 2048 | move_unlock_page_cgroup(pc, &flags); |
32047e2a KH |
2049 | rcu_read_unlock(); |
2050 | return; | |
d69b042f | 2051 | } |
2a7106f2 | 2052 | EXPORT_SYMBOL(mem_cgroup_update_page_stat); |
26174efd | 2053 | |
cdec2e42 KH |
2054 | /* |
2055 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2056 | * TODO: maybe necessary to use big numbers in big irons. | |
2057 | */ | |
7ec99d62 | 2058 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2059 | struct memcg_stock_pcp { |
2060 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2061 | unsigned int nr_pages; |
cdec2e42 | 2062 | struct work_struct work; |
26fe6168 KH |
2063 | unsigned long flags; |
2064 | #define FLUSHING_CACHED_CHARGE (0) | |
cdec2e42 KH |
2065 | }; |
2066 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2067 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 KH |
2068 | |
2069 | /* | |
11c9ea4e | 2070 | * Try to consume stocked charge on this cpu. If success, one page is consumed |
cdec2e42 KH |
2071 | * from local stock and true is returned. If the stock is 0 or charges from a |
2072 | * cgroup which is not current target, returns false. This stock will be | |
2073 | * refilled. | |
2074 | */ | |
c0ff4b85 | 2075 | static bool consume_stock(struct mem_cgroup *memcg) |
cdec2e42 KH |
2076 | { |
2077 | struct memcg_stock_pcp *stock; | |
2078 | bool ret = true; | |
2079 | ||
2080 | stock = &get_cpu_var(memcg_stock); | |
c0ff4b85 | 2081 | if (memcg == stock->cached && stock->nr_pages) |
11c9ea4e | 2082 | stock->nr_pages--; |
cdec2e42 KH |
2083 | else /* need to call res_counter_charge */ |
2084 | ret = false; | |
2085 | put_cpu_var(memcg_stock); | |
2086 | return ret; | |
2087 | } | |
2088 | ||
2089 | /* | |
2090 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
2091 | */ | |
2092 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2093 | { | |
2094 | struct mem_cgroup *old = stock->cached; | |
2095 | ||
11c9ea4e JW |
2096 | if (stock->nr_pages) { |
2097 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
2098 | ||
2099 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 2100 | if (do_swap_account) |
11c9ea4e JW |
2101 | res_counter_uncharge(&old->memsw, bytes); |
2102 | stock->nr_pages = 0; | |
cdec2e42 KH |
2103 | } |
2104 | stock->cached = NULL; | |
cdec2e42 KH |
2105 | } |
2106 | ||
2107 | /* | |
2108 | * This must be called under preempt disabled or must be called by | |
2109 | * a thread which is pinned to local cpu. | |
2110 | */ | |
2111 | static void drain_local_stock(struct work_struct *dummy) | |
2112 | { | |
2113 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
2114 | drain_stock(stock); | |
26fe6168 | 2115 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2116 | } |
2117 | ||
2118 | /* | |
2119 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 2120 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2121 | */ |
c0ff4b85 | 2122 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2123 | { |
2124 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2125 | ||
c0ff4b85 | 2126 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2127 | drain_stock(stock); |
c0ff4b85 | 2128 | stock->cached = memcg; |
cdec2e42 | 2129 | } |
11c9ea4e | 2130 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2131 | put_cpu_var(memcg_stock); |
2132 | } | |
2133 | ||
2134 | /* | |
c0ff4b85 | 2135 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
d38144b7 MH |
2136 | * of the hierarchy under it. sync flag says whether we should block |
2137 | * until the work is done. | |
cdec2e42 | 2138 | */ |
c0ff4b85 | 2139 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
cdec2e42 | 2140 | { |
26fe6168 | 2141 | int cpu, curcpu; |
d38144b7 | 2142 | |
cdec2e42 | 2143 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2144 | get_online_cpus(); |
5af12d0e | 2145 | curcpu = get_cpu(); |
cdec2e42 KH |
2146 | for_each_online_cpu(cpu) { |
2147 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2148 | struct mem_cgroup *memcg; |
26fe6168 | 2149 | |
c0ff4b85 R |
2150 | memcg = stock->cached; |
2151 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2152 | continue; |
c0ff4b85 | 2153 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2154 | continue; |
d1a05b69 MH |
2155 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2156 | if (cpu == curcpu) | |
2157 | drain_local_stock(&stock->work); | |
2158 | else | |
2159 | schedule_work_on(cpu, &stock->work); | |
2160 | } | |
cdec2e42 | 2161 | } |
5af12d0e | 2162 | put_cpu(); |
d38144b7 MH |
2163 | |
2164 | if (!sync) | |
2165 | goto out; | |
2166 | ||
2167 | for_each_online_cpu(cpu) { | |
2168 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2169 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2170 | flush_work(&stock->work); |
2171 | } | |
2172 | out: | |
cdec2e42 | 2173 | put_online_cpus(); |
d38144b7 MH |
2174 | } |
2175 | ||
2176 | /* | |
2177 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2178 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
2179 | * expects some charges will be back to res_counter later but cannot wait for | |
2180 | * it. | |
2181 | */ | |
c0ff4b85 | 2182 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
d38144b7 | 2183 | { |
9f50fad6 MH |
2184 | /* |
2185 | * If someone calls draining, avoid adding more kworker runs. | |
2186 | */ | |
2187 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2188 | return; | |
c0ff4b85 | 2189 | drain_all_stock(root_memcg, false); |
9f50fad6 | 2190 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2191 | } |
2192 | ||
2193 | /* This is a synchronous drain interface. */ | |
c0ff4b85 | 2194 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
cdec2e42 KH |
2195 | { |
2196 | /* called when force_empty is called */ | |
9f50fad6 | 2197 | mutex_lock(&percpu_charge_mutex); |
c0ff4b85 | 2198 | drain_all_stock(root_memcg, true); |
9f50fad6 | 2199 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2200 | } |
2201 | ||
711d3d2c KH |
2202 | /* |
2203 | * This function drains percpu counter value from DEAD cpu and | |
2204 | * move it to local cpu. Note that this function can be preempted. | |
2205 | */ | |
c0ff4b85 | 2206 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2207 | { |
2208 | int i; | |
2209 | ||
c0ff4b85 | 2210 | spin_lock(&memcg->pcp_counter_lock); |
711d3d2c | 2211 | for (i = 0; i < MEM_CGROUP_STAT_DATA; i++) { |
c0ff4b85 | 2212 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2213 | |
c0ff4b85 R |
2214 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2215 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2216 | } |
e9f8974f | 2217 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2218 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2219 | |
c0ff4b85 R |
2220 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2221 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2222 | } |
1489ebad | 2223 | /* need to clear ON_MOVE value, works as a kind of lock. */ |
c0ff4b85 R |
2224 | per_cpu(memcg->stat->count[MEM_CGROUP_ON_MOVE], cpu) = 0; |
2225 | spin_unlock(&memcg->pcp_counter_lock); | |
1489ebad KH |
2226 | } |
2227 | ||
c0ff4b85 | 2228 | static void synchronize_mem_cgroup_on_move(struct mem_cgroup *memcg, int cpu) |
1489ebad KH |
2229 | { |
2230 | int idx = MEM_CGROUP_ON_MOVE; | |
2231 | ||
c0ff4b85 R |
2232 | spin_lock(&memcg->pcp_counter_lock); |
2233 | per_cpu(memcg->stat->count[idx], cpu) = memcg->nocpu_base.count[idx]; | |
2234 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
2235 | } |
2236 | ||
2237 | static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb, | |
cdec2e42 KH |
2238 | unsigned long action, |
2239 | void *hcpu) | |
2240 | { | |
2241 | int cpu = (unsigned long)hcpu; | |
2242 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2243 | struct mem_cgroup *iter; |
cdec2e42 | 2244 | |
1489ebad KH |
2245 | if ((action == CPU_ONLINE)) { |
2246 | for_each_mem_cgroup_all(iter) | |
2247 | synchronize_mem_cgroup_on_move(iter, cpu); | |
2248 | return NOTIFY_OK; | |
2249 | } | |
2250 | ||
711d3d2c | 2251 | if ((action != CPU_DEAD) || action != CPU_DEAD_FROZEN) |
cdec2e42 | 2252 | return NOTIFY_OK; |
711d3d2c KH |
2253 | |
2254 | for_each_mem_cgroup_all(iter) | |
2255 | mem_cgroup_drain_pcp_counter(iter, cpu); | |
2256 | ||
cdec2e42 KH |
2257 | stock = &per_cpu(memcg_stock, cpu); |
2258 | drain_stock(stock); | |
2259 | return NOTIFY_OK; | |
2260 | } | |
2261 | ||
4b534334 KH |
2262 | |
2263 | /* See __mem_cgroup_try_charge() for details */ | |
2264 | enum { | |
2265 | CHARGE_OK, /* success */ | |
2266 | CHARGE_RETRY, /* need to retry but retry is not bad */ | |
2267 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ | |
2268 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ | |
2269 | CHARGE_OOM_DIE, /* the current is killed because of OOM */ | |
2270 | }; | |
2271 | ||
c0ff4b85 | 2272 | static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
7ec99d62 | 2273 | unsigned int nr_pages, bool oom_check) |
4b534334 | 2274 | { |
7ec99d62 | 2275 | unsigned long csize = nr_pages * PAGE_SIZE; |
4b534334 KH |
2276 | struct mem_cgroup *mem_over_limit; |
2277 | struct res_counter *fail_res; | |
2278 | unsigned long flags = 0; | |
2279 | int ret; | |
2280 | ||
c0ff4b85 | 2281 | ret = res_counter_charge(&memcg->res, csize, &fail_res); |
4b534334 KH |
2282 | |
2283 | if (likely(!ret)) { | |
2284 | if (!do_swap_account) | |
2285 | return CHARGE_OK; | |
c0ff4b85 | 2286 | ret = res_counter_charge(&memcg->memsw, csize, &fail_res); |
4b534334 KH |
2287 | if (likely(!ret)) |
2288 | return CHARGE_OK; | |
2289 | ||
c0ff4b85 | 2290 | res_counter_uncharge(&memcg->res, csize); |
4b534334 KH |
2291 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
2292 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
2293 | } else | |
2294 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
9221edb7 | 2295 | /* |
7ec99d62 JW |
2296 | * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch |
2297 | * of regular pages (CHARGE_BATCH), or a single regular page (1). | |
9221edb7 JW |
2298 | * |
2299 | * Never reclaim on behalf of optional batching, retry with a | |
2300 | * single page instead. | |
2301 | */ | |
7ec99d62 | 2302 | if (nr_pages == CHARGE_BATCH) |
4b534334 KH |
2303 | return CHARGE_RETRY; |
2304 | ||
2305 | if (!(gfp_mask & __GFP_WAIT)) | |
2306 | return CHARGE_WOULDBLOCK; | |
2307 | ||
2308 | ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL, | |
0ae5e89c | 2309 | gfp_mask, flags, NULL); |
7ec99d62 | 2310 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
19942822 | 2311 | return CHARGE_RETRY; |
4b534334 | 2312 | /* |
19942822 JW |
2313 | * Even though the limit is exceeded at this point, reclaim |
2314 | * may have been able to free some pages. Retry the charge | |
2315 | * before killing the task. | |
2316 | * | |
2317 | * Only for regular pages, though: huge pages are rather | |
2318 | * unlikely to succeed so close to the limit, and we fall back | |
2319 | * to regular pages anyway in case of failure. | |
4b534334 | 2320 | */ |
7ec99d62 | 2321 | if (nr_pages == 1 && ret) |
4b534334 KH |
2322 | return CHARGE_RETRY; |
2323 | ||
2324 | /* | |
2325 | * At task move, charge accounts can be doubly counted. So, it's | |
2326 | * better to wait until the end of task_move if something is going on. | |
2327 | */ | |
2328 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2329 | return CHARGE_RETRY; | |
2330 | ||
2331 | /* If we don't need to call oom-killer at el, return immediately */ | |
2332 | if (!oom_check) | |
2333 | return CHARGE_NOMEM; | |
2334 | /* check OOM */ | |
2335 | if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask)) | |
2336 | return CHARGE_OOM_DIE; | |
2337 | ||
2338 | return CHARGE_RETRY; | |
2339 | } | |
2340 | ||
f817ed48 KH |
2341 | /* |
2342 | * Unlike exported interface, "oom" parameter is added. if oom==true, | |
2343 | * oom-killer can be invoked. | |
8a9f3ccd | 2344 | */ |
f817ed48 | 2345 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
ec168510 | 2346 | gfp_t gfp_mask, |
7ec99d62 | 2347 | unsigned int nr_pages, |
c0ff4b85 | 2348 | struct mem_cgroup **ptr, |
7ec99d62 | 2349 | bool oom) |
8a9f3ccd | 2350 | { |
7ec99d62 | 2351 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
4b534334 | 2352 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
c0ff4b85 | 2353 | struct mem_cgroup *memcg = NULL; |
4b534334 | 2354 | int ret; |
a636b327 | 2355 | |
867578cb KH |
2356 | /* |
2357 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
2358 | * in system level. So, allow to go ahead dying process in addition to | |
2359 | * MEMDIE process. | |
2360 | */ | |
2361 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
2362 | || fatal_signal_pending(current))) | |
2363 | goto bypass; | |
a636b327 | 2364 | |
8a9f3ccd | 2365 | /* |
3be91277 HD |
2366 | * We always charge the cgroup the mm_struct belongs to. |
2367 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
2368 | * thread group leader migrates. It's possible that mm is not |
2369 | * set, if so charge the init_mm (happens for pagecache usage). | |
2370 | */ | |
c0ff4b85 | 2371 | if (!*ptr && !mm) |
f75ca962 KH |
2372 | goto bypass; |
2373 | again: | |
c0ff4b85 R |
2374 | if (*ptr) { /* css should be a valid one */ |
2375 | memcg = *ptr; | |
2376 | VM_BUG_ON(css_is_removed(&memcg->css)); | |
2377 | if (mem_cgroup_is_root(memcg)) | |
f75ca962 | 2378 | goto done; |
c0ff4b85 | 2379 | if (nr_pages == 1 && consume_stock(memcg)) |
f75ca962 | 2380 | goto done; |
c0ff4b85 | 2381 | css_get(&memcg->css); |
4b534334 | 2382 | } else { |
f75ca962 | 2383 | struct task_struct *p; |
54595fe2 | 2384 | |
f75ca962 KH |
2385 | rcu_read_lock(); |
2386 | p = rcu_dereference(mm->owner); | |
f75ca962 | 2387 | /* |
ebb76ce1 | 2388 | * Because we don't have task_lock(), "p" can exit. |
c0ff4b85 | 2389 | * In that case, "memcg" can point to root or p can be NULL with |
ebb76ce1 KH |
2390 | * race with swapoff. Then, we have small risk of mis-accouning. |
2391 | * But such kind of mis-account by race always happens because | |
2392 | * we don't have cgroup_mutex(). It's overkill and we allo that | |
2393 | * small race, here. | |
2394 | * (*) swapoff at el will charge against mm-struct not against | |
2395 | * task-struct. So, mm->owner can be NULL. | |
f75ca962 | 2396 | */ |
c0ff4b85 R |
2397 | memcg = mem_cgroup_from_task(p); |
2398 | if (!memcg || mem_cgroup_is_root(memcg)) { | |
f75ca962 KH |
2399 | rcu_read_unlock(); |
2400 | goto done; | |
2401 | } | |
c0ff4b85 | 2402 | if (nr_pages == 1 && consume_stock(memcg)) { |
f75ca962 KH |
2403 | /* |
2404 | * It seems dagerous to access memcg without css_get(). | |
2405 | * But considering how consume_stok works, it's not | |
2406 | * necessary. If consume_stock success, some charges | |
2407 | * from this memcg are cached on this cpu. So, we | |
2408 | * don't need to call css_get()/css_tryget() before | |
2409 | * calling consume_stock(). | |
2410 | */ | |
2411 | rcu_read_unlock(); | |
2412 | goto done; | |
2413 | } | |
2414 | /* after here, we may be blocked. we need to get refcnt */ | |
c0ff4b85 | 2415 | if (!css_tryget(&memcg->css)) { |
f75ca962 KH |
2416 | rcu_read_unlock(); |
2417 | goto again; | |
2418 | } | |
2419 | rcu_read_unlock(); | |
2420 | } | |
8a9f3ccd | 2421 | |
4b534334 KH |
2422 | do { |
2423 | bool oom_check; | |
7a81b88c | 2424 | |
4b534334 | 2425 | /* If killed, bypass charge */ |
f75ca962 | 2426 | if (fatal_signal_pending(current)) { |
c0ff4b85 | 2427 | css_put(&memcg->css); |
4b534334 | 2428 | goto bypass; |
f75ca962 | 2429 | } |
6d61ef40 | 2430 | |
4b534334 KH |
2431 | oom_check = false; |
2432 | if (oom && !nr_oom_retries) { | |
2433 | oom_check = true; | |
2434 | nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
cdec2e42 | 2435 | } |
66e1707b | 2436 | |
c0ff4b85 | 2437 | ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, oom_check); |
4b534334 KH |
2438 | switch (ret) { |
2439 | case CHARGE_OK: | |
2440 | break; | |
2441 | case CHARGE_RETRY: /* not in OOM situation but retry */ | |
7ec99d62 | 2442 | batch = nr_pages; |
c0ff4b85 R |
2443 | css_put(&memcg->css); |
2444 | memcg = NULL; | |
f75ca962 | 2445 | goto again; |
4b534334 | 2446 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
c0ff4b85 | 2447 | css_put(&memcg->css); |
4b534334 KH |
2448 | goto nomem; |
2449 | case CHARGE_NOMEM: /* OOM routine works */ | |
f75ca962 | 2450 | if (!oom) { |
c0ff4b85 | 2451 | css_put(&memcg->css); |
867578cb | 2452 | goto nomem; |
f75ca962 | 2453 | } |
4b534334 KH |
2454 | /* If oom, we never return -ENOMEM */ |
2455 | nr_oom_retries--; | |
2456 | break; | |
2457 | case CHARGE_OOM_DIE: /* Killed by OOM Killer */ | |
c0ff4b85 | 2458 | css_put(&memcg->css); |
867578cb | 2459 | goto bypass; |
66e1707b | 2460 | } |
4b534334 KH |
2461 | } while (ret != CHARGE_OK); |
2462 | ||
7ec99d62 | 2463 | if (batch > nr_pages) |
c0ff4b85 R |
2464 | refill_stock(memcg, batch - nr_pages); |
2465 | css_put(&memcg->css); | |
0c3e73e8 | 2466 | done: |
c0ff4b85 | 2467 | *ptr = memcg; |
7a81b88c KH |
2468 | return 0; |
2469 | nomem: | |
c0ff4b85 | 2470 | *ptr = NULL; |
7a81b88c | 2471 | return -ENOMEM; |
867578cb | 2472 | bypass: |
c0ff4b85 | 2473 | *ptr = NULL; |
867578cb | 2474 | return 0; |
7a81b88c | 2475 | } |
8a9f3ccd | 2476 | |
a3032a2c DN |
2477 | /* |
2478 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2479 | * This function is for that and do uncharge, put css's refcnt. | |
2480 | * gotten by try_charge(). | |
2481 | */ | |
c0ff4b85 | 2482 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg, |
e7018b8d | 2483 | unsigned int nr_pages) |
a3032a2c | 2484 | { |
c0ff4b85 | 2485 | if (!mem_cgroup_is_root(memcg)) { |
e7018b8d JW |
2486 | unsigned long bytes = nr_pages * PAGE_SIZE; |
2487 | ||
c0ff4b85 | 2488 | res_counter_uncharge(&memcg->res, bytes); |
a3032a2c | 2489 | if (do_swap_account) |
c0ff4b85 | 2490 | res_counter_uncharge(&memcg->memsw, bytes); |
a3032a2c | 2491 | } |
854ffa8d DN |
2492 | } |
2493 | ||
a3b2d692 KH |
2494 | /* |
2495 | * A helper function to get mem_cgroup from ID. must be called under | |
2496 | * rcu_read_lock(). The caller must check css_is_removed() or some if | |
2497 | * it's concern. (dropping refcnt from swap can be called against removed | |
2498 | * memcg.) | |
2499 | */ | |
2500 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2501 | { | |
2502 | struct cgroup_subsys_state *css; | |
2503 | ||
2504 | /* ID 0 is unused ID */ | |
2505 | if (!id) | |
2506 | return NULL; | |
2507 | css = css_lookup(&mem_cgroup_subsys, id); | |
2508 | if (!css) | |
2509 | return NULL; | |
2510 | return container_of(css, struct mem_cgroup, css); | |
2511 | } | |
2512 | ||
e42d9d5d | 2513 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2514 | { |
c0ff4b85 | 2515 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2516 | struct page_cgroup *pc; |
a3b2d692 | 2517 | unsigned short id; |
b5a84319 KH |
2518 | swp_entry_t ent; |
2519 | ||
3c776e64 DN |
2520 | VM_BUG_ON(!PageLocked(page)); |
2521 | ||
3c776e64 | 2522 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2523 | lock_page_cgroup(pc); |
a3b2d692 | 2524 | if (PageCgroupUsed(pc)) { |
c0ff4b85 R |
2525 | memcg = pc->mem_cgroup; |
2526 | if (memcg && !css_tryget(&memcg->css)) | |
2527 | memcg = NULL; | |
e42d9d5d | 2528 | } else if (PageSwapCache(page)) { |
3c776e64 | 2529 | ent.val = page_private(page); |
a3b2d692 KH |
2530 | id = lookup_swap_cgroup(ent); |
2531 | rcu_read_lock(); | |
c0ff4b85 R |
2532 | memcg = mem_cgroup_lookup(id); |
2533 | if (memcg && !css_tryget(&memcg->css)) | |
2534 | memcg = NULL; | |
a3b2d692 | 2535 | rcu_read_unlock(); |
3c776e64 | 2536 | } |
c0bd3f63 | 2537 | unlock_page_cgroup(pc); |
c0ff4b85 | 2538 | return memcg; |
b5a84319 KH |
2539 | } |
2540 | ||
c0ff4b85 | 2541 | static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg, |
5564e88b | 2542 | struct page *page, |
7ec99d62 | 2543 | unsigned int nr_pages, |
ca3e0214 | 2544 | struct page_cgroup *pc, |
7ec99d62 | 2545 | enum charge_type ctype) |
7a81b88c | 2546 | { |
ca3e0214 KH |
2547 | lock_page_cgroup(pc); |
2548 | if (unlikely(PageCgroupUsed(pc))) { | |
2549 | unlock_page_cgroup(pc); | |
c0ff4b85 | 2550 | __mem_cgroup_cancel_charge(memcg, nr_pages); |
ca3e0214 KH |
2551 | return; |
2552 | } | |
2553 | /* | |
2554 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2555 | * accessed by any other context at this point. | |
2556 | */ | |
c0ff4b85 | 2557 | pc->mem_cgroup = memcg; |
261fb61a KH |
2558 | /* |
2559 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
2560 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
2561 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
2562 | * before USED bit, we need memory barrier here. | |
2563 | * See mem_cgroup_add_lru_list(), etc. | |
2564 | */ | |
08e552c6 | 2565 | smp_wmb(); |
4b3bde4c BS |
2566 | switch (ctype) { |
2567 | case MEM_CGROUP_CHARGE_TYPE_CACHE: | |
2568 | case MEM_CGROUP_CHARGE_TYPE_SHMEM: | |
2569 | SetPageCgroupCache(pc); | |
2570 | SetPageCgroupUsed(pc); | |
2571 | break; | |
2572 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
2573 | ClearPageCgroupCache(pc); | |
2574 | SetPageCgroupUsed(pc); | |
2575 | break; | |
2576 | default: | |
2577 | break; | |
2578 | } | |
3be91277 | 2579 | |
c0ff4b85 | 2580 | mem_cgroup_charge_statistics(memcg, PageCgroupCache(pc), nr_pages); |
52d4b9ac | 2581 | unlock_page_cgroup(pc); |
430e4863 KH |
2582 | /* |
2583 | * "charge_statistics" updated event counter. Then, check it. | |
2584 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
2585 | * if they exceeds softlimit. | |
2586 | */ | |
c0ff4b85 | 2587 | memcg_check_events(memcg, page); |
7a81b88c | 2588 | } |
66e1707b | 2589 | |
ca3e0214 KH |
2590 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2591 | ||
2592 | #define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\ | |
2593 | (1 << PCG_ACCT_LRU) | (1 << PCG_MIGRATION)) | |
2594 | /* | |
2595 | * Because tail pages are not marked as "used", set it. We're under | |
2596 | * zone->lru_lock, 'splitting on pmd' and compund_lock. | |
2597 | */ | |
2598 | void mem_cgroup_split_huge_fixup(struct page *head, struct page *tail) | |
2599 | { | |
2600 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
2601 | struct page_cgroup *tail_pc = lookup_page_cgroup(tail); | |
2602 | unsigned long flags; | |
2603 | ||
3d37c4a9 KH |
2604 | if (mem_cgroup_disabled()) |
2605 | return; | |
ca3e0214 | 2606 | /* |
ece35ca8 | 2607 | * We have no races with charge/uncharge but will have races with |
ca3e0214 KH |
2608 | * page state accounting. |
2609 | */ | |
2610 | move_lock_page_cgroup(head_pc, &flags); | |
2611 | ||
2612 | tail_pc->mem_cgroup = head_pc->mem_cgroup; | |
2613 | smp_wmb(); /* see __commit_charge() */ | |
ece35ca8 KH |
2614 | if (PageCgroupAcctLRU(head_pc)) { |
2615 | enum lru_list lru; | |
2616 | struct mem_cgroup_per_zone *mz; | |
2617 | ||
2618 | /* | |
2619 | * LRU flags cannot be copied because we need to add tail | |
2620 | *.page to LRU by generic call and our hook will be called. | |
2621 | * We hold lru_lock, then, reduce counter directly. | |
2622 | */ | |
2623 | lru = page_lru(head); | |
97a6c37b | 2624 | mz = page_cgroup_zoneinfo(head_pc->mem_cgroup, head); |
ece35ca8 KH |
2625 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
2626 | } | |
ca3e0214 KH |
2627 | tail_pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
2628 | move_unlock_page_cgroup(head_pc, &flags); | |
2629 | } | |
2630 | #endif | |
2631 | ||
f817ed48 | 2632 | /** |
de3638d9 | 2633 | * mem_cgroup_move_account - move account of the page |
5564e88b | 2634 | * @page: the page |
7ec99d62 | 2635 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
2636 | * @pc: page_cgroup of the page. |
2637 | * @from: mem_cgroup which the page is moved from. | |
2638 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
854ffa8d | 2639 | * @uncharge: whether we should call uncharge and css_put against @from. |
f817ed48 KH |
2640 | * |
2641 | * The caller must confirm following. | |
08e552c6 | 2642 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 2643 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 2644 | * |
854ffa8d | 2645 | * This function doesn't do "charge" nor css_get to new cgroup. It should be |
25985edc | 2646 | * done by a caller(__mem_cgroup_try_charge would be useful). If @uncharge is |
854ffa8d DN |
2647 | * true, this function does "uncharge" from old cgroup, but it doesn't if |
2648 | * @uncharge is false, so a caller should do "uncharge". | |
f817ed48 | 2649 | */ |
7ec99d62 JW |
2650 | static int mem_cgroup_move_account(struct page *page, |
2651 | unsigned int nr_pages, | |
2652 | struct page_cgroup *pc, | |
2653 | struct mem_cgroup *from, | |
2654 | struct mem_cgroup *to, | |
2655 | bool uncharge) | |
f817ed48 | 2656 | { |
de3638d9 JW |
2657 | unsigned long flags; |
2658 | int ret; | |
987eba66 | 2659 | |
f817ed48 | 2660 | VM_BUG_ON(from == to); |
5564e88b | 2661 | VM_BUG_ON(PageLRU(page)); |
de3638d9 JW |
2662 | /* |
2663 | * The page is isolated from LRU. So, collapse function | |
2664 | * will not handle this page. But page splitting can happen. | |
2665 | * Do this check under compound_page_lock(). The caller should | |
2666 | * hold it. | |
2667 | */ | |
2668 | ret = -EBUSY; | |
7ec99d62 | 2669 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
2670 | goto out; |
2671 | ||
2672 | lock_page_cgroup(pc); | |
2673 | ||
2674 | ret = -EINVAL; | |
2675 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
2676 | goto unlock; | |
2677 | ||
2678 | move_lock_page_cgroup(pc, &flags); | |
f817ed48 | 2679 | |
8725d541 | 2680 | if (PageCgroupFileMapped(pc)) { |
c62b1a3b KH |
2681 | /* Update mapped_file data for mem_cgroup */ |
2682 | preempt_disable(); | |
2683 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2684 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2685 | preempt_enable(); | |
d69b042f | 2686 | } |
987eba66 | 2687 | mem_cgroup_charge_statistics(from, PageCgroupCache(pc), -nr_pages); |
854ffa8d DN |
2688 | if (uncharge) |
2689 | /* This is not "cancel", but cancel_charge does all we need. */ | |
e7018b8d | 2690 | __mem_cgroup_cancel_charge(from, nr_pages); |
d69b042f | 2691 | |
854ffa8d | 2692 | /* caller should have done css_get */ |
08e552c6 | 2693 | pc->mem_cgroup = to; |
987eba66 | 2694 | mem_cgroup_charge_statistics(to, PageCgroupCache(pc), nr_pages); |
88703267 KH |
2695 | /* |
2696 | * We charges against "to" which may not have any tasks. Then, "to" | |
2697 | * can be under rmdir(). But in current implementation, caller of | |
4ffef5fe | 2698 | * this function is just force_empty() and move charge, so it's |
25985edc | 2699 | * guaranteed that "to" is never removed. So, we don't check rmdir |
4ffef5fe | 2700 | * status here. |
88703267 | 2701 | */ |
de3638d9 JW |
2702 | move_unlock_page_cgroup(pc, &flags); |
2703 | ret = 0; | |
2704 | unlock: | |
57f9fd7d | 2705 | unlock_page_cgroup(pc); |
d2265e6f KH |
2706 | /* |
2707 | * check events | |
2708 | */ | |
5564e88b JW |
2709 | memcg_check_events(to, page); |
2710 | memcg_check_events(from, page); | |
de3638d9 | 2711 | out: |
f817ed48 KH |
2712 | return ret; |
2713 | } | |
2714 | ||
2715 | /* | |
2716 | * move charges to its parent. | |
2717 | */ | |
2718 | ||
5564e88b JW |
2719 | static int mem_cgroup_move_parent(struct page *page, |
2720 | struct page_cgroup *pc, | |
f817ed48 KH |
2721 | struct mem_cgroup *child, |
2722 | gfp_t gfp_mask) | |
2723 | { | |
2724 | struct cgroup *cg = child->css.cgroup; | |
2725 | struct cgroup *pcg = cg->parent; | |
2726 | struct mem_cgroup *parent; | |
7ec99d62 | 2727 | unsigned int nr_pages; |
4be4489f | 2728 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
2729 | int ret; |
2730 | ||
2731 | /* Is ROOT ? */ | |
2732 | if (!pcg) | |
2733 | return -EINVAL; | |
2734 | ||
57f9fd7d DN |
2735 | ret = -EBUSY; |
2736 | if (!get_page_unless_zero(page)) | |
2737 | goto out; | |
2738 | if (isolate_lru_page(page)) | |
2739 | goto put; | |
52dbb905 | 2740 | |
7ec99d62 | 2741 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 2742 | |
f817ed48 | 2743 | parent = mem_cgroup_from_cont(pcg); |
7ec99d62 | 2744 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, nr_pages, &parent, false); |
a636b327 | 2745 | if (ret || !parent) |
57f9fd7d | 2746 | goto put_back; |
f817ed48 | 2747 | |
7ec99d62 | 2748 | if (nr_pages > 1) |
987eba66 KH |
2749 | flags = compound_lock_irqsave(page); |
2750 | ||
7ec99d62 | 2751 | ret = mem_cgroup_move_account(page, nr_pages, pc, child, parent, true); |
854ffa8d | 2752 | if (ret) |
7ec99d62 | 2753 | __mem_cgroup_cancel_charge(parent, nr_pages); |
8dba474f | 2754 | |
7ec99d62 | 2755 | if (nr_pages > 1) |
987eba66 | 2756 | compound_unlock_irqrestore(page, flags); |
8dba474f | 2757 | put_back: |
08e552c6 | 2758 | putback_lru_page(page); |
57f9fd7d | 2759 | put: |
40d58138 | 2760 | put_page(page); |
57f9fd7d | 2761 | out: |
f817ed48 KH |
2762 | return ret; |
2763 | } | |
2764 | ||
7a81b88c KH |
2765 | /* |
2766 | * Charge the memory controller for page usage. | |
2767 | * Return | |
2768 | * 0 if the charge was successful | |
2769 | * < 0 if the cgroup is over its limit | |
2770 | */ | |
2771 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
73045c47 | 2772 | gfp_t gfp_mask, enum charge_type ctype) |
7a81b88c | 2773 | { |
c0ff4b85 | 2774 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 2775 | unsigned int nr_pages = 1; |
7a81b88c | 2776 | struct page_cgroup *pc; |
8493ae43 | 2777 | bool oom = true; |
7a81b88c | 2778 | int ret; |
ec168510 | 2779 | |
37c2ac78 | 2780 | if (PageTransHuge(page)) { |
7ec99d62 | 2781 | nr_pages <<= compound_order(page); |
37c2ac78 | 2782 | VM_BUG_ON(!PageTransHuge(page)); |
8493ae43 JW |
2783 | /* |
2784 | * Never OOM-kill a process for a huge page. The | |
2785 | * fault handler will fall back to regular pages. | |
2786 | */ | |
2787 | oom = false; | |
37c2ac78 | 2788 | } |
7a81b88c KH |
2789 | |
2790 | pc = lookup_page_cgroup(page); | |
af4a6621 | 2791 | BUG_ON(!pc); /* XXX: remove this and move pc lookup into commit */ |
7a81b88c | 2792 | |
c0ff4b85 R |
2793 | ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom); |
2794 | if (ret || !memcg) | |
7a81b88c KH |
2795 | return ret; |
2796 | ||
c0ff4b85 | 2797 | __mem_cgroup_commit_charge(memcg, page, nr_pages, pc, ctype); |
8a9f3ccd | 2798 | return 0; |
8a9f3ccd BS |
2799 | } |
2800 | ||
7a81b88c KH |
2801 | int mem_cgroup_newpage_charge(struct page *page, |
2802 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 2803 | { |
f8d66542 | 2804 | if (mem_cgroup_disabled()) |
cede86ac | 2805 | return 0; |
69029cd5 KH |
2806 | /* |
2807 | * If already mapped, we don't have to account. | |
2808 | * If page cache, page->mapping has address_space. | |
2809 | * But page->mapping may have out-of-use anon_vma pointer, | |
2810 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
2811 | * is NULL. | |
2812 | */ | |
2813 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
2814 | return 0; | |
2815 | if (unlikely(!mm)) | |
2816 | mm = &init_mm; | |
217bc319 | 2817 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
73045c47 | 2818 | MEM_CGROUP_CHARGE_TYPE_MAPPED); |
217bc319 KH |
2819 | } |
2820 | ||
83aae4c7 DN |
2821 | static void |
2822 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
2823 | enum charge_type ctype); | |
2824 | ||
5a6475a4 | 2825 | static void |
c0ff4b85 | 2826 | __mem_cgroup_commit_charge_lrucare(struct page *page, struct mem_cgroup *memcg, |
5a6475a4 KH |
2827 | enum charge_type ctype) |
2828 | { | |
2829 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
2830 | /* | |
2831 | * In some case, SwapCache, FUSE(splice_buf->radixtree), the page | |
2832 | * is already on LRU. It means the page may on some other page_cgroup's | |
2833 | * LRU. Take care of it. | |
2834 | */ | |
2835 | mem_cgroup_lru_del_before_commit(page); | |
c0ff4b85 | 2836 | __mem_cgroup_commit_charge(memcg, page, 1, pc, ctype); |
5a6475a4 KH |
2837 | mem_cgroup_lru_add_after_commit(page); |
2838 | return; | |
2839 | } | |
2840 | ||
e1a1cd59 BS |
2841 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
2842 | gfp_t gfp_mask) | |
8697d331 | 2843 | { |
c0ff4b85 | 2844 | struct mem_cgroup *memcg = NULL; |
b5a84319 KH |
2845 | int ret; |
2846 | ||
f8d66542 | 2847 | if (mem_cgroup_disabled()) |
cede86ac | 2848 | return 0; |
52d4b9ac KH |
2849 | if (PageCompound(page)) |
2850 | return 0; | |
accf163e | 2851 | |
73045c47 | 2852 | if (unlikely(!mm)) |
8697d331 | 2853 | mm = &init_mm; |
accf163e | 2854 | |
5a6475a4 | 2855 | if (page_is_file_cache(page)) { |
c0ff4b85 R |
2856 | ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, &memcg, true); |
2857 | if (ret || !memcg) | |
5a6475a4 | 2858 | return ret; |
b5a84319 | 2859 | |
5a6475a4 KH |
2860 | /* |
2861 | * FUSE reuses pages without going through the final | |
2862 | * put that would remove them from the LRU list, make | |
2863 | * sure that they get relinked properly. | |
2864 | */ | |
c0ff4b85 | 2865 | __mem_cgroup_commit_charge_lrucare(page, memcg, |
5a6475a4 KH |
2866 | MEM_CGROUP_CHARGE_TYPE_CACHE); |
2867 | return ret; | |
2868 | } | |
83aae4c7 DN |
2869 | /* shmem */ |
2870 | if (PageSwapCache(page)) { | |
c0ff4b85 | 2871 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &memcg); |
83aae4c7 | 2872 | if (!ret) |
c0ff4b85 | 2873 | __mem_cgroup_commit_charge_swapin(page, memcg, |
83aae4c7 DN |
2874 | MEM_CGROUP_CHARGE_TYPE_SHMEM); |
2875 | } else | |
2876 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, | |
73045c47 | 2877 | MEM_CGROUP_CHARGE_TYPE_SHMEM); |
b5a84319 | 2878 | |
b5a84319 | 2879 | return ret; |
e8589cc1 KH |
2880 | } |
2881 | ||
54595fe2 KH |
2882 | /* |
2883 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
2884 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 2885 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
2886 | * "commit()" or removed by "cancel()" |
2887 | */ | |
8c7c6e34 KH |
2888 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
2889 | struct page *page, | |
2890 | gfp_t mask, struct mem_cgroup **ptr) | |
2891 | { | |
c0ff4b85 | 2892 | struct mem_cgroup *memcg; |
54595fe2 | 2893 | int ret; |
8c7c6e34 | 2894 | |
56039efa KH |
2895 | *ptr = NULL; |
2896 | ||
f8d66542 | 2897 | if (mem_cgroup_disabled()) |
8c7c6e34 KH |
2898 | return 0; |
2899 | ||
2900 | if (!do_swap_account) | |
2901 | goto charge_cur_mm; | |
8c7c6e34 KH |
2902 | /* |
2903 | * A racing thread's fault, or swapoff, may have already updated | |
407f9c8b HD |
2904 | * the pte, and even removed page from swap cache: in those cases |
2905 | * do_swap_page()'s pte_same() test will fail; but there's also a | |
2906 | * KSM case which does need to charge the page. | |
8c7c6e34 KH |
2907 | */ |
2908 | if (!PageSwapCache(page)) | |
407f9c8b | 2909 | goto charge_cur_mm; |
c0ff4b85 R |
2910 | memcg = try_get_mem_cgroup_from_page(page); |
2911 | if (!memcg) | |
54595fe2 | 2912 | goto charge_cur_mm; |
c0ff4b85 | 2913 | *ptr = memcg; |
7ec99d62 | 2914 | ret = __mem_cgroup_try_charge(NULL, mask, 1, ptr, true); |
c0ff4b85 | 2915 | css_put(&memcg->css); |
54595fe2 | 2916 | return ret; |
8c7c6e34 KH |
2917 | charge_cur_mm: |
2918 | if (unlikely(!mm)) | |
2919 | mm = &init_mm; | |
7ec99d62 | 2920 | return __mem_cgroup_try_charge(mm, mask, 1, ptr, true); |
8c7c6e34 KH |
2921 | } |
2922 | ||
83aae4c7 DN |
2923 | static void |
2924 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
2925 | enum charge_type ctype) | |
7a81b88c | 2926 | { |
f8d66542 | 2927 | if (mem_cgroup_disabled()) |
7a81b88c KH |
2928 | return; |
2929 | if (!ptr) | |
2930 | return; | |
88703267 | 2931 | cgroup_exclude_rmdir(&ptr->css); |
5a6475a4 KH |
2932 | |
2933 | __mem_cgroup_commit_charge_lrucare(page, ptr, ctype); | |
8c7c6e34 KH |
2934 | /* |
2935 | * Now swap is on-memory. This means this page may be | |
2936 | * counted both as mem and swap....double count. | |
03f3c433 KH |
2937 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
2938 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
2939 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 2940 | */ |
03f3c433 | 2941 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 2942 | swp_entry_t ent = {.val = page_private(page)}; |
a3b2d692 | 2943 | unsigned short id; |
8c7c6e34 | 2944 | struct mem_cgroup *memcg; |
a3b2d692 KH |
2945 | |
2946 | id = swap_cgroup_record(ent, 0); | |
2947 | rcu_read_lock(); | |
2948 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 2949 | if (memcg) { |
a3b2d692 KH |
2950 | /* |
2951 | * This recorded memcg can be obsolete one. So, avoid | |
2952 | * calling css_tryget | |
2953 | */ | |
0c3e73e8 | 2954 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 2955 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 2956 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
2957 | mem_cgroup_put(memcg); |
2958 | } | |
a3b2d692 | 2959 | rcu_read_unlock(); |
8c7c6e34 | 2960 | } |
88703267 KH |
2961 | /* |
2962 | * At swapin, we may charge account against cgroup which has no tasks. | |
2963 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
2964 | * In that case, we need to call pre_destroy() again. check it here. | |
2965 | */ | |
2966 | cgroup_release_and_wakeup_rmdir(&ptr->css); | |
7a81b88c KH |
2967 | } |
2968 | ||
83aae4c7 DN |
2969 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) |
2970 | { | |
2971 | __mem_cgroup_commit_charge_swapin(page, ptr, | |
2972 | MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
2973 | } | |
2974 | ||
c0ff4b85 | 2975 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg) |
7a81b88c | 2976 | { |
f8d66542 | 2977 | if (mem_cgroup_disabled()) |
7a81b88c | 2978 | return; |
c0ff4b85 | 2979 | if (!memcg) |
7a81b88c | 2980 | return; |
c0ff4b85 | 2981 | __mem_cgroup_cancel_charge(memcg, 1); |
7a81b88c KH |
2982 | } |
2983 | ||
c0ff4b85 | 2984 | static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg, |
7ec99d62 JW |
2985 | unsigned int nr_pages, |
2986 | const enum charge_type ctype) | |
569b846d KH |
2987 | { |
2988 | struct memcg_batch_info *batch = NULL; | |
2989 | bool uncharge_memsw = true; | |
7ec99d62 | 2990 | |
569b846d KH |
2991 | /* If swapout, usage of swap doesn't decrease */ |
2992 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2993 | uncharge_memsw = false; | |
569b846d KH |
2994 | |
2995 | batch = ¤t->memcg_batch; | |
2996 | /* | |
2997 | * In usual, we do css_get() when we remember memcg pointer. | |
2998 | * But in this case, we keep res->usage until end of a series of | |
2999 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
3000 | */ | |
3001 | if (!batch->memcg) | |
c0ff4b85 | 3002 | batch->memcg = memcg; |
3c11ecf4 KH |
3003 | /* |
3004 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
25985edc | 3005 | * In those cases, all pages freed continuously can be expected to be in |
3c11ecf4 KH |
3006 | * the same cgroup and we have chance to coalesce uncharges. |
3007 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
3008 | * because we want to do uncharge as soon as possible. | |
3009 | */ | |
3010 | ||
3011 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
3012 | goto direct_uncharge; | |
3013 | ||
7ec99d62 | 3014 | if (nr_pages > 1) |
ec168510 AA |
3015 | goto direct_uncharge; |
3016 | ||
569b846d KH |
3017 | /* |
3018 | * In typical case, batch->memcg == mem. This means we can | |
3019 | * merge a series of uncharges to an uncharge of res_counter. | |
3020 | * If not, we uncharge res_counter ony by one. | |
3021 | */ | |
c0ff4b85 | 3022 | if (batch->memcg != memcg) |
569b846d KH |
3023 | goto direct_uncharge; |
3024 | /* remember freed charge and uncharge it later */ | |
7ffd4ca7 | 3025 | batch->nr_pages++; |
569b846d | 3026 | if (uncharge_memsw) |
7ffd4ca7 | 3027 | batch->memsw_nr_pages++; |
569b846d KH |
3028 | return; |
3029 | direct_uncharge: | |
c0ff4b85 | 3030 | res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE); |
569b846d | 3031 | if (uncharge_memsw) |
c0ff4b85 R |
3032 | res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE); |
3033 | if (unlikely(batch->memcg != memcg)) | |
3034 | memcg_oom_recover(memcg); | |
569b846d KH |
3035 | return; |
3036 | } | |
7a81b88c | 3037 | |
8a9f3ccd | 3038 | /* |
69029cd5 | 3039 | * uncharge if !page_mapped(page) |
8a9f3ccd | 3040 | */ |
8c7c6e34 | 3041 | static struct mem_cgroup * |
69029cd5 | 3042 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
8a9f3ccd | 3043 | { |
c0ff4b85 | 3044 | struct mem_cgroup *memcg = NULL; |
7ec99d62 JW |
3045 | unsigned int nr_pages = 1; |
3046 | struct page_cgroup *pc; | |
8a9f3ccd | 3047 | |
f8d66542 | 3048 | if (mem_cgroup_disabled()) |
8c7c6e34 | 3049 | return NULL; |
4077960e | 3050 | |
d13d1443 | 3051 | if (PageSwapCache(page)) |
8c7c6e34 | 3052 | return NULL; |
d13d1443 | 3053 | |
37c2ac78 | 3054 | if (PageTransHuge(page)) { |
7ec99d62 | 3055 | nr_pages <<= compound_order(page); |
37c2ac78 AA |
3056 | VM_BUG_ON(!PageTransHuge(page)); |
3057 | } | |
8697d331 | 3058 | /* |
3c541e14 | 3059 | * Check if our page_cgroup is valid |
8697d331 | 3060 | */ |
52d4b9ac KH |
3061 | pc = lookup_page_cgroup(page); |
3062 | if (unlikely(!pc || !PageCgroupUsed(pc))) | |
8c7c6e34 | 3063 | return NULL; |
b9c565d5 | 3064 | |
52d4b9ac | 3065 | lock_page_cgroup(pc); |
d13d1443 | 3066 | |
c0ff4b85 | 3067 | memcg = pc->mem_cgroup; |
8c7c6e34 | 3068 | |
d13d1443 KH |
3069 | if (!PageCgroupUsed(pc)) |
3070 | goto unlock_out; | |
3071 | ||
3072 | switch (ctype) { | |
3073 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
8a9478ca | 3074 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c AM |
3075 | /* See mem_cgroup_prepare_migration() */ |
3076 | if (page_mapped(page) || PageCgroupMigration(pc)) | |
d13d1443 KH |
3077 | goto unlock_out; |
3078 | break; | |
3079 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
3080 | if (!PageAnon(page)) { /* Shared memory */ | |
3081 | if (page->mapping && !page_is_file_cache(page)) | |
3082 | goto unlock_out; | |
3083 | } else if (page_mapped(page)) /* Anon */ | |
3084 | goto unlock_out; | |
3085 | break; | |
3086 | default: | |
3087 | break; | |
52d4b9ac | 3088 | } |
d13d1443 | 3089 | |
c0ff4b85 | 3090 | mem_cgroup_charge_statistics(memcg, PageCgroupCache(pc), -nr_pages); |
04046e1a | 3091 | |
52d4b9ac | 3092 | ClearPageCgroupUsed(pc); |
544122e5 KH |
3093 | /* |
3094 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
3095 | * freed from LRU. This is safe because uncharged page is expected not | |
3096 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
3097 | * special functions. | |
3098 | */ | |
b9c565d5 | 3099 | |
52d4b9ac | 3100 | unlock_page_cgroup(pc); |
f75ca962 | 3101 | /* |
c0ff4b85 | 3102 | * even after unlock, we have memcg->res.usage here and this memcg |
f75ca962 KH |
3103 | * will never be freed. |
3104 | */ | |
c0ff4b85 | 3105 | memcg_check_events(memcg, page); |
f75ca962 | 3106 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
c0ff4b85 R |
3107 | mem_cgroup_swap_statistics(memcg, true); |
3108 | mem_cgroup_get(memcg); | |
f75ca962 | 3109 | } |
c0ff4b85 R |
3110 | if (!mem_cgroup_is_root(memcg)) |
3111 | mem_cgroup_do_uncharge(memcg, nr_pages, ctype); | |
6d12e2d8 | 3112 | |
c0ff4b85 | 3113 | return memcg; |
d13d1443 KH |
3114 | |
3115 | unlock_out: | |
3116 | unlock_page_cgroup(pc); | |
8c7c6e34 | 3117 | return NULL; |
3c541e14 BS |
3118 | } |
3119 | ||
69029cd5 KH |
3120 | void mem_cgroup_uncharge_page(struct page *page) |
3121 | { | |
52d4b9ac KH |
3122 | /* early check. */ |
3123 | if (page_mapped(page)) | |
3124 | return; | |
3125 | if (page->mapping && !PageAnon(page)) | |
3126 | return; | |
69029cd5 KH |
3127 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
3128 | } | |
3129 | ||
3130 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
3131 | { | |
3132 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 3133 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
3134 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
3135 | } | |
3136 | ||
569b846d KH |
3137 | /* |
3138 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
3139 | * In that cases, pages are freed continuously and we can expect pages | |
3140 | * are in the same memcg. All these calls itself limits the number of | |
3141 | * pages freed at once, then uncharge_start/end() is called properly. | |
3142 | * This may be called prural(2) times in a context, | |
3143 | */ | |
3144 | ||
3145 | void mem_cgroup_uncharge_start(void) | |
3146 | { | |
3147 | current->memcg_batch.do_batch++; | |
3148 | /* We can do nest. */ | |
3149 | if (current->memcg_batch.do_batch == 1) { | |
3150 | current->memcg_batch.memcg = NULL; | |
7ffd4ca7 JW |
3151 | current->memcg_batch.nr_pages = 0; |
3152 | current->memcg_batch.memsw_nr_pages = 0; | |
569b846d KH |
3153 | } |
3154 | } | |
3155 | ||
3156 | void mem_cgroup_uncharge_end(void) | |
3157 | { | |
3158 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
3159 | ||
3160 | if (!batch->do_batch) | |
3161 | return; | |
3162 | ||
3163 | batch->do_batch--; | |
3164 | if (batch->do_batch) /* If stacked, do nothing. */ | |
3165 | return; | |
3166 | ||
3167 | if (!batch->memcg) | |
3168 | return; | |
3169 | /* | |
3170 | * This "batch->memcg" is valid without any css_get/put etc... | |
3171 | * bacause we hide charges behind us. | |
3172 | */ | |
7ffd4ca7 JW |
3173 | if (batch->nr_pages) |
3174 | res_counter_uncharge(&batch->memcg->res, | |
3175 | batch->nr_pages * PAGE_SIZE); | |
3176 | if (batch->memsw_nr_pages) | |
3177 | res_counter_uncharge(&batch->memcg->memsw, | |
3178 | batch->memsw_nr_pages * PAGE_SIZE); | |
3c11ecf4 | 3179 | memcg_oom_recover(batch->memcg); |
569b846d KH |
3180 | /* forget this pointer (for sanity check) */ |
3181 | batch->memcg = NULL; | |
3182 | } | |
3183 | ||
e767e056 | 3184 | #ifdef CONFIG_SWAP |
8c7c6e34 | 3185 | /* |
e767e056 | 3186 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
3187 | * memcg information is recorded to swap_cgroup of "ent" |
3188 | */ | |
8a9478ca KH |
3189 | void |
3190 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
3191 | { |
3192 | struct mem_cgroup *memcg; | |
8a9478ca KH |
3193 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
3194 | ||
3195 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
3196 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
3197 | ||
3198 | memcg = __mem_cgroup_uncharge_common(page, ctype); | |
8c7c6e34 | 3199 | |
f75ca962 KH |
3200 | /* |
3201 | * record memcg information, if swapout && memcg != NULL, | |
3202 | * mem_cgroup_get() was called in uncharge(). | |
3203 | */ | |
3204 | if (do_swap_account && swapout && memcg) | |
a3b2d692 | 3205 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 | 3206 | } |
e767e056 | 3207 | #endif |
8c7c6e34 KH |
3208 | |
3209 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
3210 | /* | |
3211 | * called from swap_entry_free(). remove record in swap_cgroup and | |
3212 | * uncharge "memsw" account. | |
3213 | */ | |
3214 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 3215 | { |
8c7c6e34 | 3216 | struct mem_cgroup *memcg; |
a3b2d692 | 3217 | unsigned short id; |
8c7c6e34 KH |
3218 | |
3219 | if (!do_swap_account) | |
3220 | return; | |
3221 | ||
a3b2d692 KH |
3222 | id = swap_cgroup_record(ent, 0); |
3223 | rcu_read_lock(); | |
3224 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 3225 | if (memcg) { |
a3b2d692 KH |
3226 | /* |
3227 | * We uncharge this because swap is freed. | |
3228 | * This memcg can be obsolete one. We avoid calling css_tryget | |
3229 | */ | |
0c3e73e8 | 3230 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 3231 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 3232 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
3233 | mem_cgroup_put(memcg); |
3234 | } | |
a3b2d692 | 3235 | rcu_read_unlock(); |
d13d1443 | 3236 | } |
02491447 DN |
3237 | |
3238 | /** | |
3239 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3240 | * @entry: swap entry to be moved | |
3241 | * @from: mem_cgroup which the entry is moved from | |
3242 | * @to: mem_cgroup which the entry is moved to | |
483c30b5 | 3243 | * @need_fixup: whether we should fixup res_counters and refcounts. |
02491447 DN |
3244 | * |
3245 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3246 | * as the mem_cgroup's id of @from. | |
3247 | * | |
3248 | * Returns 0 on success, -EINVAL on failure. | |
3249 | * | |
3250 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
3251 | * both res and memsw, and called css_get(). | |
3252 | */ | |
3253 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
483c30b5 | 3254 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
02491447 DN |
3255 | { |
3256 | unsigned short old_id, new_id; | |
3257 | ||
3258 | old_id = css_id(&from->css); | |
3259 | new_id = css_id(&to->css); | |
3260 | ||
3261 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 3262 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 3263 | mem_cgroup_swap_statistics(to, true); |
02491447 | 3264 | /* |
483c30b5 DN |
3265 | * This function is only called from task migration context now. |
3266 | * It postpones res_counter and refcount handling till the end | |
3267 | * of task migration(mem_cgroup_clear_mc()) for performance | |
3268 | * improvement. But we cannot postpone mem_cgroup_get(to) | |
3269 | * because if the process that has been moved to @to does | |
3270 | * swap-in, the refcount of @to might be decreased to 0. | |
02491447 | 3271 | */ |
02491447 | 3272 | mem_cgroup_get(to); |
483c30b5 DN |
3273 | if (need_fixup) { |
3274 | if (!mem_cgroup_is_root(from)) | |
3275 | res_counter_uncharge(&from->memsw, PAGE_SIZE); | |
3276 | mem_cgroup_put(from); | |
3277 | /* | |
3278 | * we charged both to->res and to->memsw, so we should | |
3279 | * uncharge to->res. | |
3280 | */ | |
3281 | if (!mem_cgroup_is_root(to)) | |
3282 | res_counter_uncharge(&to->res, PAGE_SIZE); | |
483c30b5 | 3283 | } |
02491447 DN |
3284 | return 0; |
3285 | } | |
3286 | return -EINVAL; | |
3287 | } | |
3288 | #else | |
3289 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
483c30b5 | 3290 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
02491447 DN |
3291 | { |
3292 | return -EINVAL; | |
3293 | } | |
8c7c6e34 | 3294 | #endif |
d13d1443 | 3295 | |
ae41be37 | 3296 | /* |
01b1ae63 KH |
3297 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
3298 | * page belongs to. | |
ae41be37 | 3299 | */ |
ac39cf8c | 3300 | int mem_cgroup_prepare_migration(struct page *page, |
ef6a3c63 | 3301 | struct page *newpage, struct mem_cgroup **ptr, gfp_t gfp_mask) |
ae41be37 | 3302 | { |
c0ff4b85 | 3303 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 3304 | struct page_cgroup *pc; |
ac39cf8c | 3305 | enum charge_type ctype; |
e8589cc1 | 3306 | int ret = 0; |
8869b8f6 | 3307 | |
56039efa KH |
3308 | *ptr = NULL; |
3309 | ||
ec168510 | 3310 | VM_BUG_ON(PageTransHuge(page)); |
f8d66542 | 3311 | if (mem_cgroup_disabled()) |
4077960e BS |
3312 | return 0; |
3313 | ||
52d4b9ac KH |
3314 | pc = lookup_page_cgroup(page); |
3315 | lock_page_cgroup(pc); | |
3316 | if (PageCgroupUsed(pc)) { | |
c0ff4b85 R |
3317 | memcg = pc->mem_cgroup; |
3318 | css_get(&memcg->css); | |
ac39cf8c AM |
3319 | /* |
3320 | * At migrating an anonymous page, its mapcount goes down | |
3321 | * to 0 and uncharge() will be called. But, even if it's fully | |
3322 | * unmapped, migration may fail and this page has to be | |
3323 | * charged again. We set MIGRATION flag here and delay uncharge | |
3324 | * until end_migration() is called | |
3325 | * | |
3326 | * Corner Case Thinking | |
3327 | * A) | |
3328 | * When the old page was mapped as Anon and it's unmap-and-freed | |
3329 | * while migration was ongoing. | |
3330 | * If unmap finds the old page, uncharge() of it will be delayed | |
3331 | * until end_migration(). If unmap finds a new page, it's | |
3332 | * uncharged when it make mapcount to be 1->0. If unmap code | |
3333 | * finds swap_migration_entry, the new page will not be mapped | |
3334 | * and end_migration() will find it(mapcount==0). | |
3335 | * | |
3336 | * B) | |
3337 | * When the old page was mapped but migraion fails, the kernel | |
3338 | * remaps it. A charge for it is kept by MIGRATION flag even | |
3339 | * if mapcount goes down to 0. We can do remap successfully | |
3340 | * without charging it again. | |
3341 | * | |
3342 | * C) | |
3343 | * The "old" page is under lock_page() until the end of | |
3344 | * migration, so, the old page itself will not be swapped-out. | |
3345 | * If the new page is swapped out before end_migraton, our | |
3346 | * hook to usual swap-out path will catch the event. | |
3347 | */ | |
3348 | if (PageAnon(page)) | |
3349 | SetPageCgroupMigration(pc); | |
e8589cc1 | 3350 | } |
52d4b9ac | 3351 | unlock_page_cgroup(pc); |
ac39cf8c AM |
3352 | /* |
3353 | * If the page is not charged at this point, | |
3354 | * we return here. | |
3355 | */ | |
c0ff4b85 | 3356 | if (!memcg) |
ac39cf8c | 3357 | return 0; |
01b1ae63 | 3358 | |
c0ff4b85 | 3359 | *ptr = memcg; |
7ec99d62 | 3360 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, 1, ptr, false); |
c0ff4b85 | 3361 | css_put(&memcg->css);/* drop extra refcnt */ |
ac39cf8c AM |
3362 | if (ret || *ptr == NULL) { |
3363 | if (PageAnon(page)) { | |
3364 | lock_page_cgroup(pc); | |
3365 | ClearPageCgroupMigration(pc); | |
3366 | unlock_page_cgroup(pc); | |
3367 | /* | |
3368 | * The old page may be fully unmapped while we kept it. | |
3369 | */ | |
3370 | mem_cgroup_uncharge_page(page); | |
3371 | } | |
3372 | return -ENOMEM; | |
e8589cc1 | 3373 | } |
ac39cf8c AM |
3374 | /* |
3375 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
3376 | * is called before end_migration, we can catch all events on this new | |
3377 | * page. In the case new page is migrated but not remapped, new page's | |
3378 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
3379 | */ | |
3380 | pc = lookup_page_cgroup(newpage); | |
3381 | if (PageAnon(page)) | |
3382 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
3383 | else if (page_is_file_cache(page)) | |
3384 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
3385 | else | |
3386 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
c0ff4b85 | 3387 | __mem_cgroup_commit_charge(memcg, page, 1, pc, ctype); |
e8589cc1 | 3388 | return ret; |
ae41be37 | 3389 | } |
8869b8f6 | 3390 | |
69029cd5 | 3391 | /* remove redundant charge if migration failed*/ |
c0ff4b85 | 3392 | void mem_cgroup_end_migration(struct mem_cgroup *memcg, |
50de1dd9 | 3393 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 3394 | { |
ac39cf8c | 3395 | struct page *used, *unused; |
01b1ae63 | 3396 | struct page_cgroup *pc; |
01b1ae63 | 3397 | |
c0ff4b85 | 3398 | if (!memcg) |
01b1ae63 | 3399 | return; |
ac39cf8c | 3400 | /* blocks rmdir() */ |
c0ff4b85 | 3401 | cgroup_exclude_rmdir(&memcg->css); |
50de1dd9 | 3402 | if (!migration_ok) { |
ac39cf8c AM |
3403 | used = oldpage; |
3404 | unused = newpage; | |
01b1ae63 | 3405 | } else { |
ac39cf8c | 3406 | used = newpage; |
01b1ae63 KH |
3407 | unused = oldpage; |
3408 | } | |
69029cd5 | 3409 | /* |
ac39cf8c AM |
3410 | * We disallowed uncharge of pages under migration because mapcount |
3411 | * of the page goes down to zero, temporarly. | |
3412 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 3413 | */ |
ac39cf8c AM |
3414 | pc = lookup_page_cgroup(oldpage); |
3415 | lock_page_cgroup(pc); | |
3416 | ClearPageCgroupMigration(pc); | |
3417 | unlock_page_cgroup(pc); | |
01b1ae63 | 3418 | |
ac39cf8c AM |
3419 | __mem_cgroup_uncharge_common(unused, MEM_CGROUP_CHARGE_TYPE_FORCE); |
3420 | ||
01b1ae63 | 3421 | /* |
ac39cf8c AM |
3422 | * If a page is a file cache, radix-tree replacement is very atomic |
3423 | * and we can skip this check. When it was an Anon page, its mapcount | |
3424 | * goes down to 0. But because we added MIGRATION flage, it's not | |
3425 | * uncharged yet. There are several case but page->mapcount check | |
3426 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
3427 | * check. (see prepare_charge() also) | |
69029cd5 | 3428 | */ |
ac39cf8c AM |
3429 | if (PageAnon(used)) |
3430 | mem_cgroup_uncharge_page(used); | |
88703267 | 3431 | /* |
ac39cf8c AM |
3432 | * At migration, we may charge account against cgroup which has no |
3433 | * tasks. | |
88703267 KH |
3434 | * So, rmdir()->pre_destroy() can be called while we do this charge. |
3435 | * In that case, we need to call pre_destroy() again. check it here. | |
3436 | */ | |
c0ff4b85 | 3437 | cgroup_release_and_wakeup_rmdir(&memcg->css); |
ae41be37 | 3438 | } |
78fb7466 | 3439 | |
f212ad7c DN |
3440 | #ifdef CONFIG_DEBUG_VM |
3441 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
3442 | { | |
3443 | struct page_cgroup *pc; | |
3444 | ||
3445 | pc = lookup_page_cgroup(page); | |
3446 | if (likely(pc) && PageCgroupUsed(pc)) | |
3447 | return pc; | |
3448 | return NULL; | |
3449 | } | |
3450 | ||
3451 | bool mem_cgroup_bad_page_check(struct page *page) | |
3452 | { | |
3453 | if (mem_cgroup_disabled()) | |
3454 | return false; | |
3455 | ||
3456 | return lookup_page_cgroup_used(page) != NULL; | |
3457 | } | |
3458 | ||
3459 | void mem_cgroup_print_bad_page(struct page *page) | |
3460 | { | |
3461 | struct page_cgroup *pc; | |
3462 | ||
3463 | pc = lookup_page_cgroup_used(page); | |
3464 | if (pc) { | |
3465 | int ret = -1; | |
3466 | char *path; | |
3467 | ||
3468 | printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p", | |
3469 | pc, pc->flags, pc->mem_cgroup); | |
3470 | ||
3471 | path = kmalloc(PATH_MAX, GFP_KERNEL); | |
3472 | if (path) { | |
3473 | rcu_read_lock(); | |
3474 | ret = cgroup_path(pc->mem_cgroup->css.cgroup, | |
3475 | path, PATH_MAX); | |
3476 | rcu_read_unlock(); | |
3477 | } | |
3478 | ||
3479 | printk(KERN_CONT "(%s)\n", | |
3480 | (ret < 0) ? "cannot get the path" : path); | |
3481 | kfree(path); | |
3482 | } | |
3483 | } | |
3484 | #endif | |
3485 | ||
8c7c6e34 KH |
3486 | static DEFINE_MUTEX(set_limit_mutex); |
3487 | ||
d38d2a75 | 3488 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 3489 | unsigned long long val) |
628f4235 | 3490 | { |
81d39c20 | 3491 | int retry_count; |
3c11ecf4 | 3492 | u64 memswlimit, memlimit; |
628f4235 | 3493 | int ret = 0; |
81d39c20 KH |
3494 | int children = mem_cgroup_count_children(memcg); |
3495 | u64 curusage, oldusage; | |
3c11ecf4 | 3496 | int enlarge; |
81d39c20 KH |
3497 | |
3498 | /* | |
3499 | * For keeping hierarchical_reclaim simple, how long we should retry | |
3500 | * is depends on callers. We set our retry-count to be function | |
3501 | * of # of children which we should visit in this loop. | |
3502 | */ | |
3503 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
3504 | ||
3505 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 3506 | |
3c11ecf4 | 3507 | enlarge = 0; |
8c7c6e34 | 3508 | while (retry_count) { |
628f4235 KH |
3509 | if (signal_pending(current)) { |
3510 | ret = -EINTR; | |
3511 | break; | |
3512 | } | |
8c7c6e34 KH |
3513 | /* |
3514 | * Rather than hide all in some function, I do this in | |
3515 | * open coded manner. You see what this really does. | |
c0ff4b85 | 3516 | * We have to guarantee memcg->res.limit < memcg->memsw.limit. |
8c7c6e34 KH |
3517 | */ |
3518 | mutex_lock(&set_limit_mutex); | |
3519 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3520 | if (memswlimit < val) { | |
3521 | ret = -EINVAL; | |
3522 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
3523 | break; |
3524 | } | |
3c11ecf4 KH |
3525 | |
3526 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3527 | if (memlimit < val) | |
3528 | enlarge = 1; | |
3529 | ||
8c7c6e34 | 3530 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
3531 | if (!ret) { |
3532 | if (memswlimit == val) | |
3533 | memcg->memsw_is_minimum = true; | |
3534 | else | |
3535 | memcg->memsw_is_minimum = false; | |
3536 | } | |
8c7c6e34 KH |
3537 | mutex_unlock(&set_limit_mutex); |
3538 | ||
3539 | if (!ret) | |
3540 | break; | |
3541 | ||
aa20d489 | 3542 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
0ae5e89c YH |
3543 | MEM_CGROUP_RECLAIM_SHRINK, |
3544 | NULL); | |
81d39c20 KH |
3545 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
3546 | /* Usage is reduced ? */ | |
3547 | if (curusage >= oldusage) | |
3548 | retry_count--; | |
3549 | else | |
3550 | oldusage = curusage; | |
8c7c6e34 | 3551 | } |
3c11ecf4 KH |
3552 | if (!ret && enlarge) |
3553 | memcg_oom_recover(memcg); | |
14797e23 | 3554 | |
8c7c6e34 KH |
3555 | return ret; |
3556 | } | |
3557 | ||
338c8431 LZ |
3558 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3559 | unsigned long long val) | |
8c7c6e34 | 3560 | { |
81d39c20 | 3561 | int retry_count; |
3c11ecf4 | 3562 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
3563 | int children = mem_cgroup_count_children(memcg); |
3564 | int ret = -EBUSY; | |
3c11ecf4 | 3565 | int enlarge = 0; |
8c7c6e34 | 3566 | |
81d39c20 KH |
3567 | /* see mem_cgroup_resize_res_limit */ |
3568 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
3569 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
3570 | while (retry_count) { |
3571 | if (signal_pending(current)) { | |
3572 | ret = -EINTR; | |
3573 | break; | |
3574 | } | |
3575 | /* | |
3576 | * Rather than hide all in some function, I do this in | |
3577 | * open coded manner. You see what this really does. | |
c0ff4b85 | 3578 | * We have to guarantee memcg->res.limit < memcg->memsw.limit. |
8c7c6e34 KH |
3579 | */ |
3580 | mutex_lock(&set_limit_mutex); | |
3581 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3582 | if (memlimit > val) { | |
3583 | ret = -EINVAL; | |
3584 | mutex_unlock(&set_limit_mutex); | |
3585 | break; | |
3586 | } | |
3c11ecf4 KH |
3587 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
3588 | if (memswlimit < val) | |
3589 | enlarge = 1; | |
8c7c6e34 | 3590 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
3591 | if (!ret) { |
3592 | if (memlimit == val) | |
3593 | memcg->memsw_is_minimum = true; | |
3594 | else | |
3595 | memcg->memsw_is_minimum = false; | |
3596 | } | |
8c7c6e34 KH |
3597 | mutex_unlock(&set_limit_mutex); |
3598 | ||
3599 | if (!ret) | |
3600 | break; | |
3601 | ||
4e416953 | 3602 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
75822b44 | 3603 | MEM_CGROUP_RECLAIM_NOSWAP | |
0ae5e89c YH |
3604 | MEM_CGROUP_RECLAIM_SHRINK, |
3605 | NULL); | |
8c7c6e34 | 3606 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 3607 | /* Usage is reduced ? */ |
8c7c6e34 | 3608 | if (curusage >= oldusage) |
628f4235 | 3609 | retry_count--; |
81d39c20 KH |
3610 | else |
3611 | oldusage = curusage; | |
628f4235 | 3612 | } |
3c11ecf4 KH |
3613 | if (!ret && enlarge) |
3614 | memcg_oom_recover(memcg); | |
628f4235 KH |
3615 | return ret; |
3616 | } | |
3617 | ||
4e416953 | 3618 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
0ae5e89c YH |
3619 | gfp_t gfp_mask, |
3620 | unsigned long *total_scanned) | |
4e416953 BS |
3621 | { |
3622 | unsigned long nr_reclaimed = 0; | |
3623 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3624 | unsigned long reclaimed; | |
3625 | int loop = 0; | |
3626 | struct mem_cgroup_tree_per_zone *mctz; | |
ef8745c1 | 3627 | unsigned long long excess; |
0ae5e89c | 3628 | unsigned long nr_scanned; |
4e416953 BS |
3629 | |
3630 | if (order > 0) | |
3631 | return 0; | |
3632 | ||
00918b6a | 3633 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); |
4e416953 BS |
3634 | /* |
3635 | * This loop can run a while, specially if mem_cgroup's continuously | |
3636 | * keep exceeding their soft limit and putting the system under | |
3637 | * pressure | |
3638 | */ | |
3639 | do { | |
3640 | if (next_mz) | |
3641 | mz = next_mz; | |
3642 | else | |
3643 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3644 | if (!mz) | |
3645 | break; | |
3646 | ||
0ae5e89c | 3647 | nr_scanned = 0; |
4e416953 BS |
3648 | reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone, |
3649 | gfp_mask, | |
0ae5e89c YH |
3650 | MEM_CGROUP_RECLAIM_SOFT, |
3651 | &nr_scanned); | |
4e416953 | 3652 | nr_reclaimed += reclaimed; |
0ae5e89c | 3653 | *total_scanned += nr_scanned; |
4e416953 BS |
3654 | spin_lock(&mctz->lock); |
3655 | ||
3656 | /* | |
3657 | * If we failed to reclaim anything from this memory cgroup | |
3658 | * it is time to move on to the next cgroup | |
3659 | */ | |
3660 | next_mz = NULL; | |
3661 | if (!reclaimed) { | |
3662 | do { | |
3663 | /* | |
3664 | * Loop until we find yet another one. | |
3665 | * | |
3666 | * By the time we get the soft_limit lock | |
3667 | * again, someone might have aded the | |
3668 | * group back on the RB tree. Iterate to | |
3669 | * make sure we get a different mem. | |
3670 | * mem_cgroup_largest_soft_limit_node returns | |
3671 | * NULL if no other cgroup is present on | |
3672 | * the tree | |
3673 | */ | |
3674 | next_mz = | |
3675 | __mem_cgroup_largest_soft_limit_node(mctz); | |
39cc98f1 | 3676 | if (next_mz == mz) |
4e416953 | 3677 | css_put(&next_mz->mem->css); |
39cc98f1 | 3678 | else /* next_mz == NULL or other memcg */ |
4e416953 BS |
3679 | break; |
3680 | } while (1); | |
3681 | } | |
4e416953 | 3682 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); |
ef8745c1 | 3683 | excess = res_counter_soft_limit_excess(&mz->mem->res); |
4e416953 BS |
3684 | /* |
3685 | * One school of thought says that we should not add | |
3686 | * back the node to the tree if reclaim returns 0. | |
3687 | * But our reclaim could return 0, simply because due | |
3688 | * to priority we are exposing a smaller subset of | |
3689 | * memory to reclaim from. Consider this as a longer | |
3690 | * term TODO. | |
3691 | */ | |
ef8745c1 KH |
3692 | /* If excess == 0, no tree ops */ |
3693 | __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess); | |
4e416953 BS |
3694 | spin_unlock(&mctz->lock); |
3695 | css_put(&mz->mem->css); | |
3696 | loop++; | |
3697 | /* | |
3698 | * Could not reclaim anything and there are no more | |
3699 | * mem cgroups to try or we seem to be looping without | |
3700 | * reclaiming anything. | |
3701 | */ | |
3702 | if (!nr_reclaimed && | |
3703 | (next_mz == NULL || | |
3704 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3705 | break; | |
3706 | } while (!nr_reclaimed); | |
3707 | if (next_mz) | |
3708 | css_put(&next_mz->mem->css); | |
3709 | return nr_reclaimed; | |
3710 | } | |
3711 | ||
cc847582 KH |
3712 | /* |
3713 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
3714 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
3715 | */ | |
c0ff4b85 | 3716 | static int mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
08e552c6 | 3717 | int node, int zid, enum lru_list lru) |
cc847582 | 3718 | { |
08e552c6 KH |
3719 | struct zone *zone; |
3720 | struct mem_cgroup_per_zone *mz; | |
f817ed48 | 3721 | struct page_cgroup *pc, *busy; |
08e552c6 | 3722 | unsigned long flags, loop; |
072c56c1 | 3723 | struct list_head *list; |
f817ed48 | 3724 | int ret = 0; |
072c56c1 | 3725 | |
08e552c6 | 3726 | zone = &NODE_DATA(node)->node_zones[zid]; |
c0ff4b85 | 3727 | mz = mem_cgroup_zoneinfo(memcg, node, zid); |
b69408e8 | 3728 | list = &mz->lists[lru]; |
cc847582 | 3729 | |
f817ed48 KH |
3730 | loop = MEM_CGROUP_ZSTAT(mz, lru); |
3731 | /* give some margin against EBUSY etc...*/ | |
3732 | loop += 256; | |
3733 | busy = NULL; | |
3734 | while (loop--) { | |
5564e88b JW |
3735 | struct page *page; |
3736 | ||
f817ed48 | 3737 | ret = 0; |
08e552c6 | 3738 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 3739 | if (list_empty(list)) { |
08e552c6 | 3740 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 3741 | break; |
f817ed48 KH |
3742 | } |
3743 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
3744 | if (busy == pc) { | |
3745 | list_move(&pc->lru, list); | |
648bcc77 | 3746 | busy = NULL; |
08e552c6 | 3747 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
3748 | continue; |
3749 | } | |
08e552c6 | 3750 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 3751 | |
6b3ae58e | 3752 | page = lookup_cgroup_page(pc); |
5564e88b | 3753 | |
c0ff4b85 | 3754 | ret = mem_cgroup_move_parent(page, pc, memcg, GFP_KERNEL); |
f817ed48 | 3755 | if (ret == -ENOMEM) |
52d4b9ac | 3756 | break; |
f817ed48 KH |
3757 | |
3758 | if (ret == -EBUSY || ret == -EINVAL) { | |
3759 | /* found lock contention or "pc" is obsolete. */ | |
3760 | busy = pc; | |
3761 | cond_resched(); | |
3762 | } else | |
3763 | busy = NULL; | |
cc847582 | 3764 | } |
08e552c6 | 3765 | |
f817ed48 KH |
3766 | if (!ret && !list_empty(list)) |
3767 | return -EBUSY; | |
3768 | return ret; | |
cc847582 KH |
3769 | } |
3770 | ||
3771 | /* | |
3772 | * make mem_cgroup's charge to be 0 if there is no task. | |
3773 | * This enables deleting this mem_cgroup. | |
3774 | */ | |
c0ff4b85 | 3775 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg, bool free_all) |
cc847582 | 3776 | { |
f817ed48 KH |
3777 | int ret; |
3778 | int node, zid, shrink; | |
3779 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c0ff4b85 | 3780 | struct cgroup *cgrp = memcg->css.cgroup; |
8869b8f6 | 3781 | |
c0ff4b85 | 3782 | css_get(&memcg->css); |
f817ed48 KH |
3783 | |
3784 | shrink = 0; | |
c1e862c1 KH |
3785 | /* should free all ? */ |
3786 | if (free_all) | |
3787 | goto try_to_free; | |
f817ed48 | 3788 | move_account: |
fce66477 | 3789 | do { |
f817ed48 | 3790 | ret = -EBUSY; |
c1e862c1 KH |
3791 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
3792 | goto out; | |
3793 | ret = -EINTR; | |
3794 | if (signal_pending(current)) | |
cc847582 | 3795 | goto out; |
52d4b9ac KH |
3796 | /* This is for making all *used* pages to be on LRU. */ |
3797 | lru_add_drain_all(); | |
c0ff4b85 | 3798 | drain_all_stock_sync(memcg); |
f817ed48 | 3799 | ret = 0; |
c0ff4b85 | 3800 | mem_cgroup_start_move(memcg); |
299b4eaa | 3801 | for_each_node_state(node, N_HIGH_MEMORY) { |
f817ed48 | 3802 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { |
b69408e8 | 3803 | enum lru_list l; |
f817ed48 | 3804 | for_each_lru(l) { |
c0ff4b85 | 3805 | ret = mem_cgroup_force_empty_list(memcg, |
08e552c6 | 3806 | node, zid, l); |
f817ed48 KH |
3807 | if (ret) |
3808 | break; | |
3809 | } | |
1ecaab2b | 3810 | } |
f817ed48 KH |
3811 | if (ret) |
3812 | break; | |
3813 | } | |
c0ff4b85 R |
3814 | mem_cgroup_end_move(memcg); |
3815 | memcg_oom_recover(memcg); | |
f817ed48 KH |
3816 | /* it seems parent cgroup doesn't have enough mem */ |
3817 | if (ret == -ENOMEM) | |
3818 | goto try_to_free; | |
52d4b9ac | 3819 | cond_resched(); |
fce66477 | 3820 | /* "ret" should also be checked to ensure all lists are empty. */ |
c0ff4b85 | 3821 | } while (memcg->res.usage > 0 || ret); |
cc847582 | 3822 | out: |
c0ff4b85 | 3823 | css_put(&memcg->css); |
cc847582 | 3824 | return ret; |
f817ed48 KH |
3825 | |
3826 | try_to_free: | |
c1e862c1 KH |
3827 | /* returns EBUSY if there is a task or if we come here twice. */ |
3828 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
f817ed48 KH |
3829 | ret = -EBUSY; |
3830 | goto out; | |
3831 | } | |
c1e862c1 KH |
3832 | /* we call try-to-free pages for make this cgroup empty */ |
3833 | lru_add_drain_all(); | |
f817ed48 KH |
3834 | /* try to free all pages in this cgroup */ |
3835 | shrink = 1; | |
c0ff4b85 | 3836 | while (nr_retries && memcg->res.usage > 0) { |
f817ed48 | 3837 | int progress; |
c1e862c1 KH |
3838 | |
3839 | if (signal_pending(current)) { | |
3840 | ret = -EINTR; | |
3841 | goto out; | |
3842 | } | |
c0ff4b85 | 3843 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, |
185efc0f | 3844 | false); |
c1e862c1 | 3845 | if (!progress) { |
f817ed48 | 3846 | nr_retries--; |
c1e862c1 | 3847 | /* maybe some writeback is necessary */ |
8aa7e847 | 3848 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3849 | } |
f817ed48 KH |
3850 | |
3851 | } | |
08e552c6 | 3852 | lru_add_drain(); |
f817ed48 | 3853 | /* try move_account...there may be some *locked* pages. */ |
fce66477 | 3854 | goto move_account; |
cc847582 KH |
3855 | } |
3856 | ||
c1e862c1 KH |
3857 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
3858 | { | |
3859 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
3860 | } | |
3861 | ||
3862 | ||
18f59ea7 BS |
3863 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
3864 | { | |
3865 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
3866 | } | |
3867 | ||
3868 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
3869 | u64 val) | |
3870 | { | |
3871 | int retval = 0; | |
c0ff4b85 | 3872 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
18f59ea7 | 3873 | struct cgroup *parent = cont->parent; |
c0ff4b85 | 3874 | struct mem_cgroup *parent_memcg = NULL; |
18f59ea7 BS |
3875 | |
3876 | if (parent) | |
c0ff4b85 | 3877 | parent_memcg = mem_cgroup_from_cont(parent); |
18f59ea7 BS |
3878 | |
3879 | cgroup_lock(); | |
3880 | /* | |
af901ca1 | 3881 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3882 | * in the child subtrees. If it is unset, then the change can |
3883 | * occur, provided the current cgroup has no children. | |
3884 | * | |
3885 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3886 | * set if there are no children. | |
3887 | */ | |
c0ff4b85 | 3888 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 BS |
3889 | (val == 1 || val == 0)) { |
3890 | if (list_empty(&cont->children)) | |
c0ff4b85 | 3891 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3892 | else |
3893 | retval = -EBUSY; | |
3894 | } else | |
3895 | retval = -EINVAL; | |
3896 | cgroup_unlock(); | |
3897 | ||
3898 | return retval; | |
3899 | } | |
3900 | ||
0c3e73e8 | 3901 | |
c0ff4b85 | 3902 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg, |
7a159cc9 | 3903 | enum mem_cgroup_stat_index idx) |
0c3e73e8 | 3904 | { |
7d74b06f | 3905 | struct mem_cgroup *iter; |
7a159cc9 | 3906 | long val = 0; |
0c3e73e8 | 3907 | |
7a159cc9 | 3908 | /* Per-cpu values can be negative, use a signed accumulator */ |
c0ff4b85 | 3909 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f KH |
3910 | val += mem_cgroup_read_stat(iter, idx); |
3911 | ||
3912 | if (val < 0) /* race ? */ | |
3913 | val = 0; | |
3914 | return val; | |
0c3e73e8 BS |
3915 | } |
3916 | ||
c0ff4b85 | 3917 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
104f3928 | 3918 | { |
7d74b06f | 3919 | u64 val; |
104f3928 | 3920 | |
c0ff4b85 | 3921 | if (!mem_cgroup_is_root(memcg)) { |
e5671dfa GC |
3922 | val = 0; |
3923 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM | |
3924 | if (!memcg->kmem_independent_accounting) | |
3925 | val = res_counter_read_u64(&memcg->kmem, RES_USAGE); | |
3926 | #endif | |
104f3928 | 3927 | if (!swap) |
e5671dfa | 3928 | val += res_counter_read_u64(&memcg->res, RES_USAGE); |
104f3928 | 3929 | else |
e5671dfa GC |
3930 | val += res_counter_read_u64(&memcg->memsw, RES_USAGE); |
3931 | ||
3932 | return val; | |
104f3928 KS |
3933 | } |
3934 | ||
c0ff4b85 R |
3935 | val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE); |
3936 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS); | |
104f3928 | 3937 | |
7d74b06f | 3938 | if (swap) |
c0ff4b85 | 3939 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAPOUT); |
104f3928 KS |
3940 | |
3941 | return val << PAGE_SHIFT; | |
3942 | } | |
3943 | ||
2c3daa72 | 3944 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 3945 | { |
c0ff4b85 | 3946 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
104f3928 | 3947 | u64 val; |
8c7c6e34 KH |
3948 | int type, name; |
3949 | ||
3950 | type = MEMFILE_TYPE(cft->private); | |
3951 | name = MEMFILE_ATTR(cft->private); | |
3952 | switch (type) { | |
3953 | case _MEM: | |
104f3928 | 3954 | if (name == RES_USAGE) |
c0ff4b85 | 3955 | val = mem_cgroup_usage(memcg, false); |
104f3928 | 3956 | else |
c0ff4b85 | 3957 | val = res_counter_read_u64(&memcg->res, name); |
8c7c6e34 KH |
3958 | break; |
3959 | case _MEMSWAP: | |
104f3928 | 3960 | if (name == RES_USAGE) |
c0ff4b85 | 3961 | val = mem_cgroup_usage(memcg, true); |
104f3928 | 3962 | else |
c0ff4b85 | 3963 | val = res_counter_read_u64(&memcg->memsw, name); |
8c7c6e34 | 3964 | break; |
e5671dfa GC |
3965 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM |
3966 | case _KMEM: | |
3967 | val = res_counter_read_u64(&memcg->kmem, name); | |
3968 | break; | |
3969 | #endif | |
8c7c6e34 KH |
3970 | default: |
3971 | BUG(); | |
3972 | break; | |
3973 | } | |
3974 | return val; | |
8cdea7c0 | 3975 | } |
628f4235 KH |
3976 | /* |
3977 | * The user of this function is... | |
3978 | * RES_LIMIT. | |
3979 | */ | |
856c13aa PM |
3980 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
3981 | const char *buffer) | |
8cdea7c0 | 3982 | { |
628f4235 | 3983 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 3984 | int type, name; |
628f4235 KH |
3985 | unsigned long long val; |
3986 | int ret; | |
3987 | ||
8c7c6e34 KH |
3988 | type = MEMFILE_TYPE(cft->private); |
3989 | name = MEMFILE_ATTR(cft->private); | |
3990 | switch (name) { | |
628f4235 | 3991 | case RES_LIMIT: |
4b3bde4c BS |
3992 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3993 | ret = -EINVAL; | |
3994 | break; | |
3995 | } | |
628f4235 KH |
3996 | /* This function does all necessary parse...reuse it */ |
3997 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
3998 | if (ret) |
3999 | break; | |
4000 | if (type == _MEM) | |
628f4235 | 4001 | ret = mem_cgroup_resize_limit(memcg, val); |
8c7c6e34 KH |
4002 | else |
4003 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
628f4235 | 4004 | break; |
296c81d8 BS |
4005 | case RES_SOFT_LIMIT: |
4006 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
4007 | if (ret) | |
4008 | break; | |
4009 | /* | |
4010 | * For memsw, soft limits are hard to implement in terms | |
4011 | * of semantics, for now, we support soft limits for | |
4012 | * control without swap | |
4013 | */ | |
4014 | if (type == _MEM) | |
4015 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
4016 | else | |
4017 | ret = -EINVAL; | |
4018 | break; | |
628f4235 KH |
4019 | default: |
4020 | ret = -EINVAL; /* should be BUG() ? */ | |
4021 | break; | |
4022 | } | |
4023 | return ret; | |
8cdea7c0 BS |
4024 | } |
4025 | ||
fee7b548 KH |
4026 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
4027 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
4028 | { | |
4029 | struct cgroup *cgroup; | |
4030 | unsigned long long min_limit, min_memsw_limit, tmp; | |
4031 | ||
4032 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4033 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
4034 | cgroup = memcg->css.cgroup; | |
4035 | if (!memcg->use_hierarchy) | |
4036 | goto out; | |
4037 | ||
4038 | while (cgroup->parent) { | |
4039 | cgroup = cgroup->parent; | |
4040 | memcg = mem_cgroup_from_cont(cgroup); | |
4041 | if (!memcg->use_hierarchy) | |
4042 | break; | |
4043 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4044 | min_limit = min(min_limit, tmp); | |
4045 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
4046 | min_memsw_limit = min(min_memsw_limit, tmp); | |
4047 | } | |
4048 | out: | |
4049 | *mem_limit = min_limit; | |
4050 | *memsw_limit = min_memsw_limit; | |
4051 | return; | |
4052 | } | |
4053 | ||
29f2a4da | 4054 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 | 4055 | { |
c0ff4b85 | 4056 | struct mem_cgroup *memcg; |
8c7c6e34 | 4057 | int type, name; |
c84872e1 | 4058 | |
c0ff4b85 | 4059 | memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 KH |
4060 | type = MEMFILE_TYPE(event); |
4061 | name = MEMFILE_ATTR(event); | |
4062 | switch (name) { | |
29f2a4da | 4063 | case RES_MAX_USAGE: |
8c7c6e34 | 4064 | if (type == _MEM) |
c0ff4b85 | 4065 | res_counter_reset_max(&memcg->res); |
8c7c6e34 | 4066 | else |
c0ff4b85 | 4067 | res_counter_reset_max(&memcg->memsw); |
29f2a4da PE |
4068 | break; |
4069 | case RES_FAILCNT: | |
8c7c6e34 | 4070 | if (type == _MEM) |
c0ff4b85 | 4071 | res_counter_reset_failcnt(&memcg->res); |
8c7c6e34 | 4072 | else |
c0ff4b85 | 4073 | res_counter_reset_failcnt(&memcg->memsw); |
29f2a4da PE |
4074 | break; |
4075 | } | |
f64c3f54 | 4076 | |
85cc59db | 4077 | return 0; |
c84872e1 PE |
4078 | } |
4079 | ||
7dc74be0 DN |
4080 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
4081 | struct cftype *cft) | |
4082 | { | |
4083 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; | |
4084 | } | |
4085 | ||
02491447 | 4086 | #ifdef CONFIG_MMU |
7dc74be0 DN |
4087 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
4088 | struct cftype *cft, u64 val) | |
4089 | { | |
c0ff4b85 | 4090 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
7dc74be0 DN |
4091 | |
4092 | if (val >= (1 << NR_MOVE_TYPE)) | |
4093 | return -EINVAL; | |
4094 | /* | |
4095 | * We check this value several times in both in can_attach() and | |
4096 | * attach(), so we need cgroup lock to prevent this value from being | |
4097 | * inconsistent. | |
4098 | */ | |
4099 | cgroup_lock(); | |
c0ff4b85 | 4100 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
4101 | cgroup_unlock(); |
4102 | ||
4103 | return 0; | |
4104 | } | |
02491447 DN |
4105 | #else |
4106 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, | |
4107 | struct cftype *cft, u64 val) | |
4108 | { | |
4109 | return -ENOSYS; | |
4110 | } | |
4111 | #endif | |
7dc74be0 | 4112 | |
14067bb3 KH |
4113 | |
4114 | /* For read statistics */ | |
4115 | enum { | |
4116 | MCS_CACHE, | |
4117 | MCS_RSS, | |
d8046582 | 4118 | MCS_FILE_MAPPED, |
14067bb3 KH |
4119 | MCS_PGPGIN, |
4120 | MCS_PGPGOUT, | |
1dd3a273 | 4121 | MCS_SWAP, |
456f998e YH |
4122 | MCS_PGFAULT, |
4123 | MCS_PGMAJFAULT, | |
14067bb3 KH |
4124 | MCS_INACTIVE_ANON, |
4125 | MCS_ACTIVE_ANON, | |
4126 | MCS_INACTIVE_FILE, | |
4127 | MCS_ACTIVE_FILE, | |
4128 | MCS_UNEVICTABLE, | |
4129 | NR_MCS_STAT, | |
4130 | }; | |
4131 | ||
4132 | struct mcs_total_stat { | |
4133 | s64 stat[NR_MCS_STAT]; | |
d2ceb9b7 KH |
4134 | }; |
4135 | ||
14067bb3 KH |
4136 | struct { |
4137 | char *local_name; | |
4138 | char *total_name; | |
4139 | } memcg_stat_strings[NR_MCS_STAT] = { | |
4140 | {"cache", "total_cache"}, | |
4141 | {"rss", "total_rss"}, | |
d69b042f | 4142 | {"mapped_file", "total_mapped_file"}, |
14067bb3 KH |
4143 | {"pgpgin", "total_pgpgin"}, |
4144 | {"pgpgout", "total_pgpgout"}, | |
1dd3a273 | 4145 | {"swap", "total_swap"}, |
456f998e YH |
4146 | {"pgfault", "total_pgfault"}, |
4147 | {"pgmajfault", "total_pgmajfault"}, | |
14067bb3 KH |
4148 | {"inactive_anon", "total_inactive_anon"}, |
4149 | {"active_anon", "total_active_anon"}, | |
4150 | {"inactive_file", "total_inactive_file"}, | |
4151 | {"active_file", "total_active_file"}, | |
4152 | {"unevictable", "total_unevictable"} | |
4153 | }; | |
4154 | ||
4155 | ||
7d74b06f | 4156 | static void |
c0ff4b85 | 4157 | mem_cgroup_get_local_stat(struct mem_cgroup *memcg, struct mcs_total_stat *s) |
14067bb3 | 4158 | { |
14067bb3 KH |
4159 | s64 val; |
4160 | ||
4161 | /* per cpu stat */ | |
c0ff4b85 | 4162 | val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_CACHE); |
14067bb3 | 4163 | s->stat[MCS_CACHE] += val * PAGE_SIZE; |
c0ff4b85 | 4164 | val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_RSS); |
14067bb3 | 4165 | s->stat[MCS_RSS] += val * PAGE_SIZE; |
c0ff4b85 | 4166 | val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_FILE_MAPPED); |
d8046582 | 4167 | s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE; |
c0ff4b85 | 4168 | val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGIN); |
14067bb3 | 4169 | s->stat[MCS_PGPGIN] += val; |
c0ff4b85 | 4170 | val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGPGOUT); |
14067bb3 | 4171 | s->stat[MCS_PGPGOUT] += val; |
1dd3a273 | 4172 | if (do_swap_account) { |
c0ff4b85 | 4173 | val = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_SWAPOUT); |
1dd3a273 DN |
4174 | s->stat[MCS_SWAP] += val * PAGE_SIZE; |
4175 | } | |
c0ff4b85 | 4176 | val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGFAULT); |
456f998e | 4177 | s->stat[MCS_PGFAULT] += val; |
c0ff4b85 | 4178 | val = mem_cgroup_read_events(memcg, MEM_CGROUP_EVENTS_PGMAJFAULT); |
456f998e | 4179 | s->stat[MCS_PGMAJFAULT] += val; |
14067bb3 KH |
4180 | |
4181 | /* per zone stat */ | |
c0ff4b85 | 4182 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_ANON)); |
14067bb3 | 4183 | s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE; |
c0ff4b85 | 4184 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_ANON)); |
14067bb3 | 4185 | s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE; |
c0ff4b85 | 4186 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_FILE)); |
14067bb3 | 4187 | s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE; |
c0ff4b85 | 4188 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_FILE)); |
14067bb3 | 4189 | s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE; |
c0ff4b85 | 4190 | val = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE)); |
14067bb3 | 4191 | s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE; |
14067bb3 KH |
4192 | } |
4193 | ||
4194 | static void | |
c0ff4b85 | 4195 | mem_cgroup_get_total_stat(struct mem_cgroup *memcg, struct mcs_total_stat *s) |
14067bb3 | 4196 | { |
7d74b06f KH |
4197 | struct mem_cgroup *iter; |
4198 | ||
c0ff4b85 | 4199 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4200 | mem_cgroup_get_local_stat(iter, s); |
14067bb3 KH |
4201 | } |
4202 | ||
406eb0c9 YH |
4203 | #ifdef CONFIG_NUMA |
4204 | static int mem_control_numa_stat_show(struct seq_file *m, void *arg) | |
4205 | { | |
4206 | int nid; | |
4207 | unsigned long total_nr, file_nr, anon_nr, unevictable_nr; | |
4208 | unsigned long node_nr; | |
4209 | struct cgroup *cont = m->private; | |
4210 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); | |
4211 | ||
bb2a0de9 | 4212 | total_nr = mem_cgroup_nr_lru_pages(mem_cont, LRU_ALL); |
406eb0c9 YH |
4213 | seq_printf(m, "total=%lu", total_nr); |
4214 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
bb2a0de9 | 4215 | node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid, LRU_ALL); |
406eb0c9 YH |
4216 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4217 | } | |
4218 | seq_putc(m, '\n'); | |
4219 | ||
bb2a0de9 | 4220 | file_nr = mem_cgroup_nr_lru_pages(mem_cont, LRU_ALL_FILE); |
406eb0c9 YH |
4221 | seq_printf(m, "file=%lu", file_nr); |
4222 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
bb2a0de9 KH |
4223 | node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid, |
4224 | LRU_ALL_FILE); | |
406eb0c9 YH |
4225 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4226 | } | |
4227 | seq_putc(m, '\n'); | |
4228 | ||
bb2a0de9 | 4229 | anon_nr = mem_cgroup_nr_lru_pages(mem_cont, LRU_ALL_ANON); |
406eb0c9 YH |
4230 | seq_printf(m, "anon=%lu", anon_nr); |
4231 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
bb2a0de9 KH |
4232 | node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid, |
4233 | LRU_ALL_ANON); | |
406eb0c9 YH |
4234 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4235 | } | |
4236 | seq_putc(m, '\n'); | |
4237 | ||
bb2a0de9 | 4238 | unevictable_nr = mem_cgroup_nr_lru_pages(mem_cont, BIT(LRU_UNEVICTABLE)); |
406eb0c9 YH |
4239 | seq_printf(m, "unevictable=%lu", unevictable_nr); |
4240 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
bb2a0de9 KH |
4241 | node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid, |
4242 | BIT(LRU_UNEVICTABLE)); | |
406eb0c9 YH |
4243 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4244 | } | |
4245 | seq_putc(m, '\n'); | |
4246 | return 0; | |
4247 | } | |
4248 | #endif /* CONFIG_NUMA */ | |
4249 | ||
c64745cf PM |
4250 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
4251 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 4252 | { |
d2ceb9b7 | 4253 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
14067bb3 | 4254 | struct mcs_total_stat mystat; |
d2ceb9b7 KH |
4255 | int i; |
4256 | ||
14067bb3 KH |
4257 | memset(&mystat, 0, sizeof(mystat)); |
4258 | mem_cgroup_get_local_stat(mem_cont, &mystat); | |
d2ceb9b7 | 4259 | |
406eb0c9 | 4260 | |
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].local_name, mystat.stat[i]); |
1dd3a273 | 4265 | } |
7b854121 | 4266 | |
14067bb3 | 4267 | /* Hierarchical information */ |
fee7b548 KH |
4268 | { |
4269 | unsigned long long limit, memsw_limit; | |
4270 | memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit); | |
4271 | cb->fill(cb, "hierarchical_memory_limit", limit); | |
4272 | if (do_swap_account) | |
4273 | cb->fill(cb, "hierarchical_memsw_limit", memsw_limit); | |
4274 | } | |
7f016ee8 | 4275 | |
14067bb3 KH |
4276 | memset(&mystat, 0, sizeof(mystat)); |
4277 | mem_cgroup_get_total_stat(mem_cont, &mystat); | |
1dd3a273 DN |
4278 | for (i = 0; i < NR_MCS_STAT; i++) { |
4279 | if (i == MCS_SWAP && !do_swap_account) | |
4280 | continue; | |
14067bb3 | 4281 | cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]); |
1dd3a273 | 4282 | } |
14067bb3 | 4283 | |
7f016ee8 | 4284 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
4285 | { |
4286 | int nid, zid; | |
4287 | struct mem_cgroup_per_zone *mz; | |
4288 | unsigned long recent_rotated[2] = {0, 0}; | |
4289 | unsigned long recent_scanned[2] = {0, 0}; | |
4290 | ||
4291 | for_each_online_node(nid) | |
4292 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
4293 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); | |
4294 | ||
4295 | recent_rotated[0] += | |
4296 | mz->reclaim_stat.recent_rotated[0]; | |
4297 | recent_rotated[1] += | |
4298 | mz->reclaim_stat.recent_rotated[1]; | |
4299 | recent_scanned[0] += | |
4300 | mz->reclaim_stat.recent_scanned[0]; | |
4301 | recent_scanned[1] += | |
4302 | mz->reclaim_stat.recent_scanned[1]; | |
4303 | } | |
4304 | cb->fill(cb, "recent_rotated_anon", recent_rotated[0]); | |
4305 | cb->fill(cb, "recent_rotated_file", recent_rotated[1]); | |
4306 | cb->fill(cb, "recent_scanned_anon", recent_scanned[0]); | |
4307 | cb->fill(cb, "recent_scanned_file", recent_scanned[1]); | |
4308 | } | |
4309 | #endif | |
4310 | ||
d2ceb9b7 KH |
4311 | return 0; |
4312 | } | |
4313 | ||
a7885eb8 KM |
4314 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
4315 | { | |
4316 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4317 | ||
1f4c025b | 4318 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4319 | } |
4320 | ||
4321 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
4322 | u64 val) | |
4323 | { | |
4324 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4325 | struct mem_cgroup *parent; | |
068b38c1 | 4326 | |
a7885eb8 KM |
4327 | if (val > 100) |
4328 | return -EINVAL; | |
4329 | ||
4330 | if (cgrp->parent == NULL) | |
4331 | return -EINVAL; | |
4332 | ||
4333 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 LZ |
4334 | |
4335 | cgroup_lock(); | |
4336 | ||
a7885eb8 KM |
4337 | /* If under hierarchy, only empty-root can set this value */ |
4338 | if ((parent->use_hierarchy) || | |
068b38c1 LZ |
4339 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
4340 | cgroup_unlock(); | |
a7885eb8 | 4341 | return -EINVAL; |
068b38c1 | 4342 | } |
a7885eb8 | 4343 | |
a7885eb8 | 4344 | memcg->swappiness = val; |
a7885eb8 | 4345 | |
068b38c1 LZ |
4346 | cgroup_unlock(); |
4347 | ||
a7885eb8 KM |
4348 | return 0; |
4349 | } | |
4350 | ||
2e72b634 KS |
4351 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4352 | { | |
4353 | struct mem_cgroup_threshold_ary *t; | |
4354 | u64 usage; | |
4355 | int i; | |
4356 | ||
4357 | rcu_read_lock(); | |
4358 | if (!swap) | |
2c488db2 | 4359 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4360 | else |
2c488db2 | 4361 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4362 | |
4363 | if (!t) | |
4364 | goto unlock; | |
4365 | ||
4366 | usage = mem_cgroup_usage(memcg, swap); | |
4367 | ||
4368 | /* | |
4369 | * current_threshold points to threshold just below usage. | |
4370 | * If it's not true, a threshold was crossed after last | |
4371 | * call of __mem_cgroup_threshold(). | |
4372 | */ | |
5407a562 | 4373 | i = t->current_threshold; |
2e72b634 KS |
4374 | |
4375 | /* | |
4376 | * Iterate backward over array of thresholds starting from | |
4377 | * current_threshold and check if a threshold is crossed. | |
4378 | * If none of thresholds below usage is crossed, we read | |
4379 | * only one element of the array here. | |
4380 | */ | |
4381 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4382 | eventfd_signal(t->entries[i].eventfd, 1); | |
4383 | ||
4384 | /* i = current_threshold + 1 */ | |
4385 | i++; | |
4386 | ||
4387 | /* | |
4388 | * Iterate forward over array of thresholds starting from | |
4389 | * current_threshold+1 and check if a threshold is crossed. | |
4390 | * If none of thresholds above usage is crossed, we read | |
4391 | * only one element of the array here. | |
4392 | */ | |
4393 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4394 | eventfd_signal(t->entries[i].eventfd, 1); | |
4395 | ||
4396 | /* Update current_threshold */ | |
5407a562 | 4397 | t->current_threshold = i - 1; |
2e72b634 KS |
4398 | unlock: |
4399 | rcu_read_unlock(); | |
4400 | } | |
4401 | ||
4402 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4403 | { | |
ad4ca5f4 KS |
4404 | while (memcg) { |
4405 | __mem_cgroup_threshold(memcg, false); | |
4406 | if (do_swap_account) | |
4407 | __mem_cgroup_threshold(memcg, true); | |
4408 | ||
4409 | memcg = parent_mem_cgroup(memcg); | |
4410 | } | |
2e72b634 KS |
4411 | } |
4412 | ||
4413 | static int compare_thresholds(const void *a, const void *b) | |
4414 | { | |
4415 | const struct mem_cgroup_threshold *_a = a; | |
4416 | const struct mem_cgroup_threshold *_b = b; | |
4417 | ||
4418 | return _a->threshold - _b->threshold; | |
4419 | } | |
4420 | ||
c0ff4b85 | 4421 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4422 | { |
4423 | struct mem_cgroup_eventfd_list *ev; | |
4424 | ||
c0ff4b85 | 4425 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 KH |
4426 | eventfd_signal(ev->eventfd, 1); |
4427 | return 0; | |
4428 | } | |
4429 | ||
c0ff4b85 | 4430 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4431 | { |
7d74b06f KH |
4432 | struct mem_cgroup *iter; |
4433 | ||
c0ff4b85 | 4434 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4435 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4436 | } |
4437 | ||
4438 | static int mem_cgroup_usage_register_event(struct cgroup *cgrp, | |
4439 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
2e72b634 KS |
4440 | { |
4441 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4442 | struct mem_cgroup_thresholds *thresholds; |
4443 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
4444 | int type = MEMFILE_TYPE(cft->private); |
4445 | u64 threshold, usage; | |
2c488db2 | 4446 | int i, size, ret; |
2e72b634 KS |
4447 | |
4448 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
4449 | if (ret) | |
4450 | return ret; | |
4451 | ||
4452 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4453 | |
2e72b634 | 4454 | if (type == _MEM) |
2c488db2 | 4455 | thresholds = &memcg->thresholds; |
2e72b634 | 4456 | else if (type == _MEMSWAP) |
2c488db2 | 4457 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4458 | else |
4459 | BUG(); | |
4460 | ||
4461 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
4462 | ||
4463 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 4464 | if (thresholds->primary) |
2e72b634 KS |
4465 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4466 | ||
2c488db2 | 4467 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4468 | |
4469 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 4470 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 4471 | GFP_KERNEL); |
2c488db2 | 4472 | if (!new) { |
2e72b634 KS |
4473 | ret = -ENOMEM; |
4474 | goto unlock; | |
4475 | } | |
2c488db2 | 4476 | new->size = size; |
2e72b634 KS |
4477 | |
4478 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4479 | if (thresholds->primary) { |
4480 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4481 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4482 | } |
4483 | ||
2e72b634 | 4484 | /* Add new threshold */ |
2c488db2 KS |
4485 | new->entries[size - 1].eventfd = eventfd; |
4486 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4487 | |
4488 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4489 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4490 | compare_thresholds, NULL); |
4491 | ||
4492 | /* Find current threshold */ | |
2c488db2 | 4493 | new->current_threshold = -1; |
2e72b634 | 4494 | for (i = 0; i < size; i++) { |
2c488db2 | 4495 | if (new->entries[i].threshold < usage) { |
2e72b634 | 4496 | /* |
2c488db2 KS |
4497 | * new->current_threshold will not be used until |
4498 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4499 | * it here. |
4500 | */ | |
2c488db2 | 4501 | ++new->current_threshold; |
2e72b634 KS |
4502 | } |
4503 | } | |
4504 | ||
2c488db2 KS |
4505 | /* Free old spare buffer and save old primary buffer as spare */ |
4506 | kfree(thresholds->spare); | |
4507 | thresholds->spare = thresholds->primary; | |
4508 | ||
4509 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4510 | |
907860ed | 4511 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4512 | synchronize_rcu(); |
4513 | ||
2e72b634 KS |
4514 | unlock: |
4515 | mutex_unlock(&memcg->thresholds_lock); | |
4516 | ||
4517 | return ret; | |
4518 | } | |
4519 | ||
907860ed | 4520 | static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp, |
9490ff27 | 4521 | struct cftype *cft, struct eventfd_ctx *eventfd) |
2e72b634 KS |
4522 | { |
4523 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4524 | struct mem_cgroup_thresholds *thresholds; |
4525 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
4526 | int type = MEMFILE_TYPE(cft->private); |
4527 | u64 usage; | |
2c488db2 | 4528 | int i, j, size; |
2e72b634 KS |
4529 | |
4530 | mutex_lock(&memcg->thresholds_lock); | |
4531 | if (type == _MEM) | |
2c488db2 | 4532 | thresholds = &memcg->thresholds; |
2e72b634 | 4533 | else if (type == _MEMSWAP) |
2c488db2 | 4534 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4535 | else |
4536 | BUG(); | |
4537 | ||
4538 | /* | |
4539 | * Something went wrong if we trying to unregister a threshold | |
4540 | * if we don't have thresholds | |
4541 | */ | |
4542 | BUG_ON(!thresholds); | |
4543 | ||
4544 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
4545 | ||
4546 | /* Check if a threshold crossed before removing */ | |
4547 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4548 | ||
4549 | /* Calculate new number of threshold */ | |
2c488db2 KS |
4550 | size = 0; |
4551 | for (i = 0; i < thresholds->primary->size; i++) { | |
4552 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
4553 | size++; |
4554 | } | |
4555 | ||
2c488db2 | 4556 | new = thresholds->spare; |
907860ed | 4557 | |
2e72b634 KS |
4558 | /* Set thresholds array to NULL if we don't have thresholds */ |
4559 | if (!size) { | |
2c488db2 KS |
4560 | kfree(new); |
4561 | new = NULL; | |
907860ed | 4562 | goto swap_buffers; |
2e72b634 KS |
4563 | } |
4564 | ||
2c488db2 | 4565 | new->size = size; |
2e72b634 KS |
4566 | |
4567 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4568 | new->current_threshold = -1; |
4569 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4570 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4571 | continue; |
4572 | ||
2c488db2 KS |
4573 | new->entries[j] = thresholds->primary->entries[i]; |
4574 | if (new->entries[j].threshold < usage) { | |
2e72b634 | 4575 | /* |
2c488db2 | 4576 | * new->current_threshold will not be used |
2e72b634 KS |
4577 | * until rcu_assign_pointer(), so it's safe to increment |
4578 | * it here. | |
4579 | */ | |
2c488db2 | 4580 | ++new->current_threshold; |
2e72b634 KS |
4581 | } |
4582 | j++; | |
4583 | } | |
4584 | ||
907860ed | 4585 | swap_buffers: |
2c488db2 KS |
4586 | /* Swap primary and spare array */ |
4587 | thresholds->spare = thresholds->primary; | |
4588 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4589 | |
907860ed | 4590 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4591 | synchronize_rcu(); |
4592 | ||
2e72b634 | 4593 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4594 | } |
c1e862c1 | 4595 | |
9490ff27 KH |
4596 | static int mem_cgroup_oom_register_event(struct cgroup *cgrp, |
4597 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
4598 | { | |
4599 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4600 | struct mem_cgroup_eventfd_list *event; | |
4601 | int type = MEMFILE_TYPE(cft->private); | |
4602 | ||
4603 | BUG_ON(type != _OOM_TYPE); | |
4604 | event = kmalloc(sizeof(*event), GFP_KERNEL); | |
4605 | if (!event) | |
4606 | return -ENOMEM; | |
4607 | ||
1af8efe9 | 4608 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4609 | |
4610 | event->eventfd = eventfd; | |
4611 | list_add(&event->list, &memcg->oom_notify); | |
4612 | ||
4613 | /* already in OOM ? */ | |
79dfdacc | 4614 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 4615 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4616 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4617 | |
4618 | return 0; | |
4619 | } | |
4620 | ||
907860ed | 4621 | static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp, |
9490ff27 KH |
4622 | struct cftype *cft, struct eventfd_ctx *eventfd) |
4623 | { | |
c0ff4b85 | 4624 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
9490ff27 KH |
4625 | struct mem_cgroup_eventfd_list *ev, *tmp; |
4626 | int type = MEMFILE_TYPE(cft->private); | |
4627 | ||
4628 | BUG_ON(type != _OOM_TYPE); | |
4629 | ||
1af8efe9 | 4630 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4631 | |
c0ff4b85 | 4632 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4633 | if (ev->eventfd == eventfd) { |
4634 | list_del(&ev->list); | |
4635 | kfree(ev); | |
4636 | } | |
4637 | } | |
4638 | ||
1af8efe9 | 4639 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4640 | } |
4641 | ||
3c11ecf4 KH |
4642 | static int mem_cgroup_oom_control_read(struct cgroup *cgrp, |
4643 | struct cftype *cft, struct cgroup_map_cb *cb) | |
4644 | { | |
c0ff4b85 | 4645 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 | 4646 | |
c0ff4b85 | 4647 | cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable); |
3c11ecf4 | 4648 | |
c0ff4b85 | 4649 | if (atomic_read(&memcg->under_oom)) |
3c11ecf4 KH |
4650 | cb->fill(cb, "under_oom", 1); |
4651 | else | |
4652 | cb->fill(cb, "under_oom", 0); | |
4653 | return 0; | |
4654 | } | |
4655 | ||
3c11ecf4 KH |
4656 | static int mem_cgroup_oom_control_write(struct cgroup *cgrp, |
4657 | struct cftype *cft, u64 val) | |
4658 | { | |
c0ff4b85 | 4659 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 KH |
4660 | struct mem_cgroup *parent; |
4661 | ||
4662 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
4663 | if (!cgrp->parent || !((val == 0) || (val == 1))) | |
4664 | return -EINVAL; | |
4665 | ||
4666 | parent = mem_cgroup_from_cont(cgrp->parent); | |
4667 | ||
4668 | cgroup_lock(); | |
4669 | /* oom-kill-disable is a flag for subhierarchy. */ | |
4670 | if ((parent->use_hierarchy) || | |
c0ff4b85 | 4671 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
3c11ecf4 KH |
4672 | cgroup_unlock(); |
4673 | return -EINVAL; | |
4674 | } | |
c0ff4b85 | 4675 | memcg->oom_kill_disable = val; |
4d845ebf | 4676 | if (!val) |
c0ff4b85 | 4677 | memcg_oom_recover(memcg); |
3c11ecf4 KH |
4678 | cgroup_unlock(); |
4679 | return 0; | |
4680 | } | |
4681 | ||
406eb0c9 YH |
4682 | #ifdef CONFIG_NUMA |
4683 | static const struct file_operations mem_control_numa_stat_file_operations = { | |
4684 | .read = seq_read, | |
4685 | .llseek = seq_lseek, | |
4686 | .release = single_release, | |
4687 | }; | |
4688 | ||
4689 | static int mem_control_numa_stat_open(struct inode *unused, struct file *file) | |
4690 | { | |
4691 | struct cgroup *cont = file->f_dentry->d_parent->d_fsdata; | |
4692 | ||
4693 | file->f_op = &mem_control_numa_stat_file_operations; | |
4694 | return single_open(file, mem_control_numa_stat_show, cont); | |
4695 | } | |
4696 | #endif /* CONFIG_NUMA */ | |
4697 | ||
e5671dfa GC |
4698 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM |
4699 | static u64 kmem_limit_independent_read(struct cgroup *cgroup, struct cftype *cft) | |
4700 | { | |
4701 | return mem_cgroup_from_cont(cgroup)->kmem_independent_accounting; | |
4702 | } | |
4703 | ||
4704 | static int kmem_limit_independent_write(struct cgroup *cgroup, struct cftype *cft, | |
4705 | u64 val) | |
4706 | { | |
4707 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup); | |
4708 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); | |
4709 | ||
4710 | val = !!val; | |
4711 | ||
4712 | /* | |
4713 | * This follows the same hierarchy restrictions than | |
4714 | * mem_cgroup_hierarchy_write() | |
4715 | */ | |
4716 | if (!parent || !parent->use_hierarchy) { | |
4717 | if (list_empty(&cgroup->children)) | |
4718 | memcg->kmem_independent_accounting = val; | |
4719 | else | |
4720 | return -EBUSY; | |
4721 | } | |
4722 | else | |
4723 | return -EINVAL; | |
4724 | ||
4725 | return 0; | |
4726 | } | |
4727 | static struct cftype kmem_cgroup_files[] = { | |
4728 | { | |
4729 | .name = "independent_kmem_limit", | |
4730 | .read_u64 = kmem_limit_independent_read, | |
4731 | .write_u64 = kmem_limit_independent_write, | |
4732 | }, | |
4733 | { | |
4734 | .name = "kmem.usage_in_bytes", | |
4735 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
4736 | .read_u64 = mem_cgroup_read, | |
4737 | }, | |
4738 | { | |
4739 | .name = "kmem.limit_in_bytes", | |
4740 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
4741 | .read_u64 = mem_cgroup_read, | |
4742 | }, | |
4743 | }; | |
4744 | ||
4745 | static int register_kmem_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
4746 | { | |
4747 | int ret = 0; | |
4748 | ||
4749 | ret = cgroup_add_files(cont, ss, kmem_cgroup_files, | |
4750 | ARRAY_SIZE(kmem_cgroup_files)); | |
d1a4c0b3 GC |
4751 | |
4752 | /* | |
4753 | * Part of this would be better living in a separate allocation | |
4754 | * function, leaving us with just the cgroup tree population work. | |
4755 | * We, however, depend on state such as network's proto_list that | |
4756 | * is only initialized after cgroup creation. I found the less | |
4757 | * cumbersome way to deal with it to defer it all to populate time | |
4758 | */ | |
4759 | if (!ret) | |
4760 | ret = mem_cgroup_sockets_init(cont, ss); | |
e5671dfa GC |
4761 | return ret; |
4762 | }; | |
4763 | ||
d1a4c0b3 GC |
4764 | static void kmem_cgroup_destroy(struct cgroup_subsys *ss, |
4765 | struct cgroup *cont) | |
4766 | { | |
4767 | mem_cgroup_sockets_destroy(cont, ss); | |
4768 | } | |
e5671dfa GC |
4769 | #else |
4770 | static int register_kmem_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
4771 | { | |
4772 | return 0; | |
4773 | } | |
d1a4c0b3 GC |
4774 | |
4775 | static void kmem_cgroup_destroy(struct cgroup_subsys *ss, | |
4776 | struct cgroup *cont) | |
4777 | { | |
4778 | } | |
e5671dfa GC |
4779 | #endif |
4780 | ||
8cdea7c0 BS |
4781 | static struct cftype mem_cgroup_files[] = { |
4782 | { | |
0eea1030 | 4783 | .name = "usage_in_bytes", |
8c7c6e34 | 4784 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
2c3daa72 | 4785 | .read_u64 = mem_cgroup_read, |
9490ff27 KH |
4786 | .register_event = mem_cgroup_usage_register_event, |
4787 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8cdea7c0 | 4788 | }, |
c84872e1 PE |
4789 | { |
4790 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4791 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 4792 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
4793 | .read_u64 = mem_cgroup_read, |
4794 | }, | |
8cdea7c0 | 4795 | { |
0eea1030 | 4796 | .name = "limit_in_bytes", |
8c7c6e34 | 4797 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 4798 | .write_string = mem_cgroup_write, |
2c3daa72 | 4799 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 4800 | }, |
296c81d8 BS |
4801 | { |
4802 | .name = "soft_limit_in_bytes", | |
4803 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
4804 | .write_string = mem_cgroup_write, | |
4805 | .read_u64 = mem_cgroup_read, | |
4806 | }, | |
8cdea7c0 BS |
4807 | { |
4808 | .name = "failcnt", | |
8c7c6e34 | 4809 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 4810 | .trigger = mem_cgroup_reset, |
2c3daa72 | 4811 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 4812 | }, |
d2ceb9b7 KH |
4813 | { |
4814 | .name = "stat", | |
c64745cf | 4815 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 4816 | }, |
c1e862c1 KH |
4817 | { |
4818 | .name = "force_empty", | |
4819 | .trigger = mem_cgroup_force_empty_write, | |
4820 | }, | |
18f59ea7 BS |
4821 | { |
4822 | .name = "use_hierarchy", | |
4823 | .write_u64 = mem_cgroup_hierarchy_write, | |
4824 | .read_u64 = mem_cgroup_hierarchy_read, | |
4825 | }, | |
a7885eb8 KM |
4826 | { |
4827 | .name = "swappiness", | |
4828 | .read_u64 = mem_cgroup_swappiness_read, | |
4829 | .write_u64 = mem_cgroup_swappiness_write, | |
4830 | }, | |
7dc74be0 DN |
4831 | { |
4832 | .name = "move_charge_at_immigrate", | |
4833 | .read_u64 = mem_cgroup_move_charge_read, | |
4834 | .write_u64 = mem_cgroup_move_charge_write, | |
4835 | }, | |
9490ff27 KH |
4836 | { |
4837 | .name = "oom_control", | |
3c11ecf4 KH |
4838 | .read_map = mem_cgroup_oom_control_read, |
4839 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
4840 | .register_event = mem_cgroup_oom_register_event, |
4841 | .unregister_event = mem_cgroup_oom_unregister_event, | |
4842 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), | |
4843 | }, | |
406eb0c9 YH |
4844 | #ifdef CONFIG_NUMA |
4845 | { | |
4846 | .name = "numa_stat", | |
4847 | .open = mem_control_numa_stat_open, | |
89577127 | 4848 | .mode = S_IRUGO, |
406eb0c9 YH |
4849 | }, |
4850 | #endif | |
8cdea7c0 BS |
4851 | }; |
4852 | ||
8c7c6e34 KH |
4853 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
4854 | static struct cftype memsw_cgroup_files[] = { | |
4855 | { | |
4856 | .name = "memsw.usage_in_bytes", | |
4857 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
4858 | .read_u64 = mem_cgroup_read, | |
9490ff27 KH |
4859 | .register_event = mem_cgroup_usage_register_event, |
4860 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8c7c6e34 KH |
4861 | }, |
4862 | { | |
4863 | .name = "memsw.max_usage_in_bytes", | |
4864 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
4865 | .trigger = mem_cgroup_reset, | |
4866 | .read_u64 = mem_cgroup_read, | |
4867 | }, | |
4868 | { | |
4869 | .name = "memsw.limit_in_bytes", | |
4870 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
4871 | .write_string = mem_cgroup_write, | |
4872 | .read_u64 = mem_cgroup_read, | |
4873 | }, | |
4874 | { | |
4875 | .name = "memsw.failcnt", | |
4876 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
4877 | .trigger = mem_cgroup_reset, | |
4878 | .read_u64 = mem_cgroup_read, | |
4879 | }, | |
4880 | }; | |
4881 | ||
4882 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
4883 | { | |
4884 | if (!do_swap_account) | |
4885 | return 0; | |
4886 | return cgroup_add_files(cont, ss, memsw_cgroup_files, | |
4887 | ARRAY_SIZE(memsw_cgroup_files)); | |
4888 | }; | |
4889 | #else | |
4890 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
4891 | { | |
4892 | return 0; | |
4893 | } | |
4894 | #endif | |
4895 | ||
c0ff4b85 | 4896 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4897 | { |
4898 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4899 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 4900 | enum lru_list l; |
41e3355d | 4901 | int zone, tmp = node; |
1ecaab2b KH |
4902 | /* |
4903 | * This routine is called against possible nodes. | |
4904 | * But it's BUG to call kmalloc() against offline node. | |
4905 | * | |
4906 | * TODO: this routine can waste much memory for nodes which will | |
4907 | * never be onlined. It's better to use memory hotplug callback | |
4908 | * function. | |
4909 | */ | |
41e3355d KH |
4910 | if (!node_state(node, N_NORMAL_MEMORY)) |
4911 | tmp = -1; | |
17295c88 | 4912 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4913 | if (!pn) |
4914 | return 1; | |
1ecaab2b | 4915 | |
1ecaab2b KH |
4916 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4917 | mz = &pn->zoneinfo[zone]; | |
b69408e8 CL |
4918 | for_each_lru(l) |
4919 | INIT_LIST_HEAD(&mz->lists[l]); | |
f64c3f54 | 4920 | mz->usage_in_excess = 0; |
4e416953 | 4921 | mz->on_tree = false; |
c0ff4b85 | 4922 | mz->mem = memcg; |
1ecaab2b | 4923 | } |
0a619e58 | 4924 | memcg->info.nodeinfo[node] = pn; |
6d12e2d8 KH |
4925 | return 0; |
4926 | } | |
4927 | ||
c0ff4b85 | 4928 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4929 | { |
c0ff4b85 | 4930 | kfree(memcg->info.nodeinfo[node]); |
1ecaab2b KH |
4931 | } |
4932 | ||
33327948 KH |
4933 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4934 | { | |
4935 | struct mem_cgroup *mem; | |
c62b1a3b | 4936 | int size = sizeof(struct mem_cgroup); |
33327948 | 4937 | |
c62b1a3b | 4938 | /* Can be very big if MAX_NUMNODES is very big */ |
c8dad2bb | 4939 | if (size < PAGE_SIZE) |
17295c88 | 4940 | mem = kzalloc(size, GFP_KERNEL); |
33327948 | 4941 | else |
17295c88 | 4942 | mem = vzalloc(size); |
33327948 | 4943 | |
e7bbcdf3 DC |
4944 | if (!mem) |
4945 | return NULL; | |
4946 | ||
c62b1a3b | 4947 | mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
d2e61b8d DC |
4948 | if (!mem->stat) |
4949 | goto out_free; | |
711d3d2c | 4950 | spin_lock_init(&mem->pcp_counter_lock); |
33327948 | 4951 | return mem; |
d2e61b8d DC |
4952 | |
4953 | out_free: | |
4954 | if (size < PAGE_SIZE) | |
4955 | kfree(mem); | |
4956 | else | |
4957 | vfree(mem); | |
4958 | return NULL; | |
33327948 KH |
4959 | } |
4960 | ||
8c7c6e34 KH |
4961 | /* |
4962 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
4963 | * (scanning all at force_empty is too costly...) | |
4964 | * | |
4965 | * Instead of clearing all references at force_empty, we remember | |
4966 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4967 | * it goes down to 0. | |
4968 | * | |
8c7c6e34 KH |
4969 | * Removal of cgroup itself succeeds regardless of refs from swap. |
4970 | */ | |
4971 | ||
c0ff4b85 | 4972 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
33327948 | 4973 | { |
08e552c6 KH |
4974 | int node; |
4975 | ||
c0ff4b85 R |
4976 | mem_cgroup_remove_from_trees(memcg); |
4977 | free_css_id(&mem_cgroup_subsys, &memcg->css); | |
04046e1a | 4978 | |
08e552c6 | 4979 | for_each_node_state(node, N_POSSIBLE) |
c0ff4b85 | 4980 | free_mem_cgroup_per_zone_info(memcg, node); |
08e552c6 | 4981 | |
c0ff4b85 | 4982 | free_percpu(memcg->stat); |
c62b1a3b | 4983 | if (sizeof(struct mem_cgroup) < PAGE_SIZE) |
c0ff4b85 | 4984 | kfree(memcg); |
33327948 | 4985 | else |
c0ff4b85 | 4986 | vfree(memcg); |
33327948 KH |
4987 | } |
4988 | ||
c0ff4b85 | 4989 | static void mem_cgroup_get(struct mem_cgroup *memcg) |
8c7c6e34 | 4990 | { |
c0ff4b85 | 4991 | atomic_inc(&memcg->refcnt); |
8c7c6e34 KH |
4992 | } |
4993 | ||
c0ff4b85 | 4994 | static void __mem_cgroup_put(struct mem_cgroup *memcg, int count) |
8c7c6e34 | 4995 | { |
c0ff4b85 R |
4996 | if (atomic_sub_and_test(count, &memcg->refcnt)) { |
4997 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); | |
4998 | __mem_cgroup_free(memcg); | |
7bcc1bb1 DN |
4999 | if (parent) |
5000 | mem_cgroup_put(parent); | |
5001 | } | |
8c7c6e34 KH |
5002 | } |
5003 | ||
c0ff4b85 | 5004 | static void mem_cgroup_put(struct mem_cgroup *memcg) |
483c30b5 | 5005 | { |
c0ff4b85 | 5006 | __mem_cgroup_put(memcg, 1); |
483c30b5 DN |
5007 | } |
5008 | ||
7bcc1bb1 DN |
5009 | /* |
5010 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
5011 | */ | |
e1aab161 | 5012 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 5013 | { |
c0ff4b85 | 5014 | if (!memcg->res.parent) |
7bcc1bb1 | 5015 | return NULL; |
c0ff4b85 | 5016 | return mem_cgroup_from_res_counter(memcg->res.parent, res); |
7bcc1bb1 | 5017 | } |
e1aab161 | 5018 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 5019 | |
c077719b KH |
5020 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
5021 | static void __init enable_swap_cgroup(void) | |
5022 | { | |
f8d66542 | 5023 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
5024 | do_swap_account = 1; |
5025 | } | |
5026 | #else | |
5027 | static void __init enable_swap_cgroup(void) | |
5028 | { | |
5029 | } | |
5030 | #endif | |
5031 | ||
f64c3f54 BS |
5032 | static int mem_cgroup_soft_limit_tree_init(void) |
5033 | { | |
5034 | struct mem_cgroup_tree_per_node *rtpn; | |
5035 | struct mem_cgroup_tree_per_zone *rtpz; | |
5036 | int tmp, node, zone; | |
5037 | ||
5038 | for_each_node_state(node, N_POSSIBLE) { | |
5039 | tmp = node; | |
5040 | if (!node_state(node, N_NORMAL_MEMORY)) | |
5041 | tmp = -1; | |
5042 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
5043 | if (!rtpn) | |
5044 | return 1; | |
5045 | ||
5046 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
5047 | ||
5048 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
5049 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
5050 | rtpz->rb_root = RB_ROOT; | |
5051 | spin_lock_init(&rtpz->lock); | |
5052 | } | |
5053 | } | |
5054 | return 0; | |
5055 | } | |
5056 | ||
0eb253e2 | 5057 | static struct cgroup_subsys_state * __ref |
8cdea7c0 BS |
5058 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) |
5059 | { | |
c0ff4b85 | 5060 | struct mem_cgroup *memcg, *parent; |
04046e1a | 5061 | long error = -ENOMEM; |
6d12e2d8 | 5062 | int node; |
8cdea7c0 | 5063 | |
c0ff4b85 R |
5064 | memcg = mem_cgroup_alloc(); |
5065 | if (!memcg) | |
04046e1a | 5066 | return ERR_PTR(error); |
78fb7466 | 5067 | |
6d12e2d8 | 5068 | for_each_node_state(node, N_POSSIBLE) |
c0ff4b85 | 5069 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 5070 | goto free_out; |
f64c3f54 | 5071 | |
c077719b | 5072 | /* root ? */ |
28dbc4b6 | 5073 | if (cont->parent == NULL) { |
cdec2e42 | 5074 | int cpu; |
c077719b | 5075 | enable_swap_cgroup(); |
28dbc4b6 | 5076 | parent = NULL; |
c0ff4b85 | 5077 | root_mem_cgroup = memcg; |
f64c3f54 BS |
5078 | if (mem_cgroup_soft_limit_tree_init()) |
5079 | goto free_out; | |
cdec2e42 KH |
5080 | for_each_possible_cpu(cpu) { |
5081 | struct memcg_stock_pcp *stock = | |
5082 | &per_cpu(memcg_stock, cpu); | |
5083 | INIT_WORK(&stock->work, drain_local_stock); | |
5084 | } | |
711d3d2c | 5085 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
18f59ea7 | 5086 | } else { |
28dbc4b6 | 5087 | parent = mem_cgroup_from_cont(cont->parent); |
c0ff4b85 R |
5088 | memcg->use_hierarchy = parent->use_hierarchy; |
5089 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
18f59ea7 | 5090 | } |
28dbc4b6 | 5091 | |
18f59ea7 | 5092 | if (parent && parent->use_hierarchy) { |
c0ff4b85 R |
5093 | res_counter_init(&memcg->res, &parent->res); |
5094 | res_counter_init(&memcg->memsw, &parent->memsw); | |
e5671dfa | 5095 | res_counter_init(&memcg->kmem, &parent->kmem); |
7bcc1bb1 DN |
5096 | /* |
5097 | * We increment refcnt of the parent to ensure that we can | |
5098 | * safely access it on res_counter_charge/uncharge. | |
5099 | * This refcnt will be decremented when freeing this | |
5100 | * mem_cgroup(see mem_cgroup_put). | |
5101 | */ | |
5102 | mem_cgroup_get(parent); | |
18f59ea7 | 5103 | } else { |
c0ff4b85 R |
5104 | res_counter_init(&memcg->res, NULL); |
5105 | res_counter_init(&memcg->memsw, NULL); | |
e5671dfa | 5106 | res_counter_init(&memcg->kmem, NULL); |
18f59ea7 | 5107 | } |
c0ff4b85 R |
5108 | memcg->last_scanned_child = 0; |
5109 | memcg->last_scanned_node = MAX_NUMNODES; | |
5110 | INIT_LIST_HEAD(&memcg->oom_notify); | |
6d61ef40 | 5111 | |
a7885eb8 | 5112 | if (parent) |
c0ff4b85 R |
5113 | memcg->swappiness = mem_cgroup_swappiness(parent); |
5114 | atomic_set(&memcg->refcnt, 1); | |
5115 | memcg->move_charge_at_immigrate = 0; | |
5116 | mutex_init(&memcg->thresholds_lock); | |
5117 | return &memcg->css; | |
6d12e2d8 | 5118 | free_out: |
c0ff4b85 | 5119 | __mem_cgroup_free(memcg); |
4b3bde4c | 5120 | root_mem_cgroup = NULL; |
04046e1a | 5121 | return ERR_PTR(error); |
8cdea7c0 BS |
5122 | } |
5123 | ||
ec64f515 | 5124 | static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
df878fb0 KH |
5125 | struct cgroup *cont) |
5126 | { | |
c0ff4b85 | 5127 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
ec64f515 | 5128 | |
c0ff4b85 | 5129 | return mem_cgroup_force_empty(memcg, false); |
df878fb0 KH |
5130 | } |
5131 | ||
8cdea7c0 BS |
5132 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
5133 | struct cgroup *cont) | |
5134 | { | |
c0ff4b85 | 5135 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
c268e994 | 5136 | |
d1a4c0b3 GC |
5137 | kmem_cgroup_destroy(ss, cont); |
5138 | ||
c0ff4b85 | 5139 | mem_cgroup_put(memcg); |
8cdea7c0 BS |
5140 | } |
5141 | ||
5142 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
5143 | struct cgroup *cont) | |
5144 | { | |
8c7c6e34 KH |
5145 | int ret; |
5146 | ||
5147 | ret = cgroup_add_files(cont, ss, mem_cgroup_files, | |
5148 | ARRAY_SIZE(mem_cgroup_files)); | |
5149 | ||
5150 | if (!ret) | |
5151 | ret = register_memsw_files(cont, ss); | |
e5671dfa GC |
5152 | |
5153 | if (!ret) | |
5154 | ret = register_kmem_files(cont, ss); | |
5155 | ||
8c7c6e34 | 5156 | return ret; |
8cdea7c0 BS |
5157 | } |
5158 | ||
02491447 | 5159 | #ifdef CONFIG_MMU |
7dc74be0 | 5160 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
5161 | #define PRECHARGE_COUNT_AT_ONCE 256 |
5162 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 5163 | { |
854ffa8d DN |
5164 | int ret = 0; |
5165 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
c0ff4b85 | 5166 | struct mem_cgroup *memcg = mc.to; |
4ffef5fe | 5167 | |
c0ff4b85 | 5168 | if (mem_cgroup_is_root(memcg)) { |
854ffa8d DN |
5169 | mc.precharge += count; |
5170 | /* we don't need css_get for root */ | |
5171 | return ret; | |
5172 | } | |
5173 | /* try to charge at once */ | |
5174 | if (count > 1) { | |
5175 | struct res_counter *dummy; | |
5176 | /* | |
c0ff4b85 | 5177 | * "memcg" cannot be under rmdir() because we've already checked |
854ffa8d DN |
5178 | * by cgroup_lock_live_cgroup() that it is not removed and we |
5179 | * are still under the same cgroup_mutex. So we can postpone | |
5180 | * css_get(). | |
5181 | */ | |
c0ff4b85 | 5182 | if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy)) |
854ffa8d | 5183 | goto one_by_one; |
c0ff4b85 | 5184 | if (do_swap_account && res_counter_charge(&memcg->memsw, |
854ffa8d | 5185 | PAGE_SIZE * count, &dummy)) { |
c0ff4b85 | 5186 | res_counter_uncharge(&memcg->res, PAGE_SIZE * count); |
854ffa8d DN |
5187 | goto one_by_one; |
5188 | } | |
5189 | mc.precharge += count; | |
854ffa8d DN |
5190 | return ret; |
5191 | } | |
5192 | one_by_one: | |
5193 | /* fall back to one by one charge */ | |
5194 | while (count--) { | |
5195 | if (signal_pending(current)) { | |
5196 | ret = -EINTR; | |
5197 | break; | |
5198 | } | |
5199 | if (!batch_count--) { | |
5200 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
5201 | cond_resched(); | |
5202 | } | |
c0ff4b85 R |
5203 | ret = __mem_cgroup_try_charge(NULL, |
5204 | GFP_KERNEL, 1, &memcg, false); | |
5205 | if (ret || !memcg) | |
854ffa8d DN |
5206 | /* mem_cgroup_clear_mc() will do uncharge later */ |
5207 | return -ENOMEM; | |
5208 | mc.precharge++; | |
5209 | } | |
4ffef5fe DN |
5210 | return ret; |
5211 | } | |
5212 | ||
5213 | /** | |
5214 | * is_target_pte_for_mc - check a pte whether it is valid for move charge | |
5215 | * @vma: the vma the pte to be checked belongs | |
5216 | * @addr: the address corresponding to the pte to be checked | |
5217 | * @ptent: the pte to be checked | |
02491447 | 5218 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
5219 | * |
5220 | * Returns | |
5221 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5222 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5223 | * move charge. if @target is not NULL, the page is stored in target->page | |
5224 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
5225 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
5226 | * target for charge migration. if @target is not NULL, the entry is stored | |
5227 | * in target->ent. | |
4ffef5fe DN |
5228 | * |
5229 | * Called with pte lock held. | |
5230 | */ | |
4ffef5fe DN |
5231 | union mc_target { |
5232 | struct page *page; | |
02491447 | 5233 | swp_entry_t ent; |
4ffef5fe DN |
5234 | }; |
5235 | ||
4ffef5fe DN |
5236 | enum mc_target_type { |
5237 | MC_TARGET_NONE, /* not used */ | |
5238 | MC_TARGET_PAGE, | |
02491447 | 5239 | MC_TARGET_SWAP, |
4ffef5fe DN |
5240 | }; |
5241 | ||
90254a65 DN |
5242 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5243 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5244 | { |
90254a65 | 5245 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5246 | |
90254a65 DN |
5247 | if (!page || !page_mapped(page)) |
5248 | return NULL; | |
5249 | if (PageAnon(page)) { | |
5250 | /* we don't move shared anon */ | |
5251 | if (!move_anon() || page_mapcount(page) > 2) | |
5252 | return NULL; | |
87946a72 DN |
5253 | } else if (!move_file()) |
5254 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
5255 | return NULL; |
5256 | if (!get_page_unless_zero(page)) | |
5257 | return NULL; | |
5258 | ||
5259 | return page; | |
5260 | } | |
5261 | ||
5262 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
5263 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5264 | { | |
5265 | int usage_count; | |
5266 | struct page *page = NULL; | |
5267 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5268 | ||
5269 | if (!move_anon() || non_swap_entry(ent)) | |
5270 | return NULL; | |
5271 | usage_count = mem_cgroup_count_swap_user(ent, &page); | |
5272 | if (usage_count > 1) { /* we don't move shared anon */ | |
02491447 DN |
5273 | if (page) |
5274 | put_page(page); | |
90254a65 | 5275 | return NULL; |
02491447 | 5276 | } |
90254a65 DN |
5277 | if (do_swap_account) |
5278 | entry->val = ent.val; | |
5279 | ||
5280 | return page; | |
5281 | } | |
5282 | ||
87946a72 DN |
5283 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5284 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5285 | { | |
5286 | struct page *page = NULL; | |
5287 | struct inode *inode; | |
5288 | struct address_space *mapping; | |
5289 | pgoff_t pgoff; | |
5290 | ||
5291 | if (!vma->vm_file) /* anonymous vma */ | |
5292 | return NULL; | |
5293 | if (!move_file()) | |
5294 | return NULL; | |
5295 | ||
5296 | inode = vma->vm_file->f_path.dentry->d_inode; | |
5297 | mapping = vma->vm_file->f_mapping; | |
5298 | if (pte_none(ptent)) | |
5299 | pgoff = linear_page_index(vma, addr); | |
5300 | else /* pte_file(ptent) is true */ | |
5301 | pgoff = pte_to_pgoff(ptent); | |
5302 | ||
5303 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
5304 | page = find_get_page(mapping, pgoff); |
5305 | ||
5306 | #ifdef CONFIG_SWAP | |
5307 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
5308 | if (radix_tree_exceptional_entry(page)) { | |
5309 | swp_entry_t swap = radix_to_swp_entry(page); | |
87946a72 | 5310 | if (do_swap_account) |
aa3b1895 HD |
5311 | *entry = swap; |
5312 | page = find_get_page(&swapper_space, swap.val); | |
87946a72 | 5313 | } |
aa3b1895 | 5314 | #endif |
87946a72 DN |
5315 | return page; |
5316 | } | |
5317 | ||
90254a65 DN |
5318 | static int is_target_pte_for_mc(struct vm_area_struct *vma, |
5319 | unsigned long addr, pte_t ptent, union mc_target *target) | |
5320 | { | |
5321 | struct page *page = NULL; | |
5322 | struct page_cgroup *pc; | |
5323 | int ret = 0; | |
5324 | swp_entry_t ent = { .val = 0 }; | |
5325 | ||
5326 | if (pte_present(ptent)) | |
5327 | page = mc_handle_present_pte(vma, addr, ptent); | |
5328 | else if (is_swap_pte(ptent)) | |
5329 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
5330 | else if (pte_none(ptent) || pte_file(ptent)) |
5331 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
5332 | |
5333 | if (!page && !ent.val) | |
5334 | return 0; | |
02491447 DN |
5335 | if (page) { |
5336 | pc = lookup_page_cgroup(page); | |
5337 | /* | |
5338 | * Do only loose check w/o page_cgroup lock. | |
5339 | * mem_cgroup_move_account() checks the pc is valid or not under | |
5340 | * the lock. | |
5341 | */ | |
5342 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5343 | ret = MC_TARGET_PAGE; | |
5344 | if (target) | |
5345 | target->page = page; | |
5346 | } | |
5347 | if (!ret || !target) | |
5348 | put_page(page); | |
5349 | } | |
90254a65 DN |
5350 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
5351 | if (ent.val && !ret && | |
7f0f1546 KH |
5352 | css_id(&mc.from->css) == lookup_swap_cgroup(ent)) { |
5353 | ret = MC_TARGET_SWAP; | |
5354 | if (target) | |
5355 | target->ent = ent; | |
4ffef5fe | 5356 | } |
4ffef5fe DN |
5357 | return ret; |
5358 | } | |
5359 | ||
5360 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, | |
5361 | unsigned long addr, unsigned long end, | |
5362 | struct mm_walk *walk) | |
5363 | { | |
5364 | struct vm_area_struct *vma = walk->private; | |
5365 | pte_t *pte; | |
5366 | spinlock_t *ptl; | |
5367 | ||
03319327 DH |
5368 | split_huge_page_pmd(walk->mm, pmd); |
5369 | ||
4ffef5fe DN |
5370 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5371 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
5372 | if (is_target_pte_for_mc(vma, addr, *pte, NULL)) | |
5373 | mc.precharge++; /* increment precharge temporarily */ | |
5374 | pte_unmap_unlock(pte - 1, ptl); | |
5375 | cond_resched(); | |
5376 | ||
7dc74be0 DN |
5377 | return 0; |
5378 | } | |
5379 | ||
4ffef5fe DN |
5380 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5381 | { | |
5382 | unsigned long precharge; | |
5383 | struct vm_area_struct *vma; | |
5384 | ||
dfe076b0 | 5385 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
5386 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5387 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
5388 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5389 | .mm = mm, | |
5390 | .private = vma, | |
5391 | }; | |
5392 | if (is_vm_hugetlb_page(vma)) | |
5393 | continue; | |
4ffef5fe DN |
5394 | walk_page_range(vma->vm_start, vma->vm_end, |
5395 | &mem_cgroup_count_precharge_walk); | |
5396 | } | |
dfe076b0 | 5397 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5398 | |
5399 | precharge = mc.precharge; | |
5400 | mc.precharge = 0; | |
5401 | ||
5402 | return precharge; | |
5403 | } | |
5404 | ||
4ffef5fe DN |
5405 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5406 | { | |
dfe076b0 DN |
5407 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5408 | ||
5409 | VM_BUG_ON(mc.moving_task); | |
5410 | mc.moving_task = current; | |
5411 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5412 | } |
5413 | ||
dfe076b0 DN |
5414 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5415 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5416 | { |
2bd9bb20 KH |
5417 | struct mem_cgroup *from = mc.from; |
5418 | struct mem_cgroup *to = mc.to; | |
5419 | ||
4ffef5fe | 5420 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
5421 | if (mc.precharge) { |
5422 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
5423 | mc.precharge = 0; | |
5424 | } | |
5425 | /* | |
5426 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5427 | * we must uncharge here. | |
5428 | */ | |
5429 | if (mc.moved_charge) { | |
5430 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
5431 | mc.moved_charge = 0; | |
4ffef5fe | 5432 | } |
483c30b5 DN |
5433 | /* we must fixup refcnts and charges */ |
5434 | if (mc.moved_swap) { | |
483c30b5 DN |
5435 | /* uncharge swap account from the old cgroup */ |
5436 | if (!mem_cgroup_is_root(mc.from)) | |
5437 | res_counter_uncharge(&mc.from->memsw, | |
5438 | PAGE_SIZE * mc.moved_swap); | |
5439 | __mem_cgroup_put(mc.from, mc.moved_swap); | |
5440 | ||
5441 | if (!mem_cgroup_is_root(mc.to)) { | |
5442 | /* | |
5443 | * we charged both to->res and to->memsw, so we should | |
5444 | * uncharge to->res. | |
5445 | */ | |
5446 | res_counter_uncharge(&mc.to->res, | |
5447 | PAGE_SIZE * mc.moved_swap); | |
483c30b5 DN |
5448 | } |
5449 | /* we've already done mem_cgroup_get(mc.to) */ | |
483c30b5 DN |
5450 | mc.moved_swap = 0; |
5451 | } | |
dfe076b0 DN |
5452 | memcg_oom_recover(from); |
5453 | memcg_oom_recover(to); | |
5454 | wake_up_all(&mc.waitq); | |
5455 | } | |
5456 | ||
5457 | static void mem_cgroup_clear_mc(void) | |
5458 | { | |
5459 | struct mem_cgroup *from = mc.from; | |
5460 | ||
5461 | /* | |
5462 | * we must clear moving_task before waking up waiters at the end of | |
5463 | * task migration. | |
5464 | */ | |
5465 | mc.moving_task = NULL; | |
5466 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5467 | spin_lock(&mc.lock); |
4ffef5fe DN |
5468 | mc.from = NULL; |
5469 | mc.to = NULL; | |
2bd9bb20 | 5470 | spin_unlock(&mc.lock); |
32047e2a | 5471 | mem_cgroup_end_move(from); |
4ffef5fe DN |
5472 | } |
5473 | ||
7dc74be0 DN |
5474 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, |
5475 | struct cgroup *cgroup, | |
f780bdb7 | 5476 | struct task_struct *p) |
7dc74be0 DN |
5477 | { |
5478 | int ret = 0; | |
c0ff4b85 | 5479 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup); |
7dc74be0 | 5480 | |
c0ff4b85 | 5481 | if (memcg->move_charge_at_immigrate) { |
7dc74be0 DN |
5482 | struct mm_struct *mm; |
5483 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
5484 | ||
c0ff4b85 | 5485 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
5486 | |
5487 | mm = get_task_mm(p); | |
5488 | if (!mm) | |
5489 | return 0; | |
7dc74be0 | 5490 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
5491 | if (mm->owner == p) { |
5492 | VM_BUG_ON(mc.from); | |
5493 | VM_BUG_ON(mc.to); | |
5494 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 5495 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 5496 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 5497 | mem_cgroup_start_move(from); |
2bd9bb20 | 5498 | spin_lock(&mc.lock); |
4ffef5fe | 5499 | mc.from = from; |
c0ff4b85 | 5500 | mc.to = memcg; |
2bd9bb20 | 5501 | spin_unlock(&mc.lock); |
dfe076b0 | 5502 | /* We set mc.moving_task later */ |
4ffef5fe DN |
5503 | |
5504 | ret = mem_cgroup_precharge_mc(mm); | |
5505 | if (ret) | |
5506 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
5507 | } |
5508 | mmput(mm); | |
7dc74be0 DN |
5509 | } |
5510 | return ret; | |
5511 | } | |
5512 | ||
5513 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
5514 | struct cgroup *cgroup, | |
f780bdb7 | 5515 | struct task_struct *p) |
7dc74be0 | 5516 | { |
4ffef5fe | 5517 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
5518 | } |
5519 | ||
4ffef5fe DN |
5520 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5521 | unsigned long addr, unsigned long end, | |
5522 | struct mm_walk *walk) | |
7dc74be0 | 5523 | { |
4ffef5fe DN |
5524 | int ret = 0; |
5525 | struct vm_area_struct *vma = walk->private; | |
5526 | pte_t *pte; | |
5527 | spinlock_t *ptl; | |
5528 | ||
03319327 | 5529 | split_huge_page_pmd(walk->mm, pmd); |
4ffef5fe DN |
5530 | retry: |
5531 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5532 | for (; addr != end; addr += PAGE_SIZE) { | |
5533 | pte_t ptent = *(pte++); | |
5534 | union mc_target target; | |
5535 | int type; | |
5536 | struct page *page; | |
5537 | struct page_cgroup *pc; | |
02491447 | 5538 | swp_entry_t ent; |
4ffef5fe DN |
5539 | |
5540 | if (!mc.precharge) | |
5541 | break; | |
5542 | ||
5543 | type = is_target_pte_for_mc(vma, addr, ptent, &target); | |
5544 | switch (type) { | |
5545 | case MC_TARGET_PAGE: | |
5546 | page = target.page; | |
5547 | if (isolate_lru_page(page)) | |
5548 | goto put; | |
5549 | pc = lookup_page_cgroup(page); | |
7ec99d62 JW |
5550 | if (!mem_cgroup_move_account(page, 1, pc, |
5551 | mc.from, mc.to, false)) { | |
4ffef5fe | 5552 | mc.precharge--; |
854ffa8d DN |
5553 | /* we uncharge from mc.from later. */ |
5554 | mc.moved_charge++; | |
4ffef5fe DN |
5555 | } |
5556 | putback_lru_page(page); | |
5557 | put: /* is_target_pte_for_mc() gets the page */ | |
5558 | put_page(page); | |
5559 | break; | |
02491447 DN |
5560 | case MC_TARGET_SWAP: |
5561 | ent = target.ent; | |
483c30b5 DN |
5562 | if (!mem_cgroup_move_swap_account(ent, |
5563 | mc.from, mc.to, false)) { | |
02491447 | 5564 | mc.precharge--; |
483c30b5 DN |
5565 | /* we fixup refcnts and charges later. */ |
5566 | mc.moved_swap++; | |
5567 | } | |
02491447 | 5568 | break; |
4ffef5fe DN |
5569 | default: |
5570 | break; | |
5571 | } | |
5572 | } | |
5573 | pte_unmap_unlock(pte - 1, ptl); | |
5574 | cond_resched(); | |
5575 | ||
5576 | if (addr != end) { | |
5577 | /* | |
5578 | * We have consumed all precharges we got in can_attach(). | |
5579 | * We try charge one by one, but don't do any additional | |
5580 | * charges to mc.to if we have failed in charge once in attach() | |
5581 | * phase. | |
5582 | */ | |
854ffa8d | 5583 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5584 | if (!ret) |
5585 | goto retry; | |
5586 | } | |
5587 | ||
5588 | return ret; | |
5589 | } | |
5590 | ||
5591 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
5592 | { | |
5593 | struct vm_area_struct *vma; | |
5594 | ||
5595 | lru_add_drain_all(); | |
dfe076b0 DN |
5596 | retry: |
5597 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5598 | /* | |
5599 | * Someone who are holding the mmap_sem might be waiting in | |
5600 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5601 | * and retry. Because we cancel precharges, we might not be able | |
5602 | * to move enough charges, but moving charge is a best-effort | |
5603 | * feature anyway, so it wouldn't be a big problem. | |
5604 | */ | |
5605 | __mem_cgroup_clear_mc(); | |
5606 | cond_resched(); | |
5607 | goto retry; | |
5608 | } | |
4ffef5fe DN |
5609 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5610 | int ret; | |
5611 | struct mm_walk mem_cgroup_move_charge_walk = { | |
5612 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
5613 | .mm = mm, | |
5614 | .private = vma, | |
5615 | }; | |
5616 | if (is_vm_hugetlb_page(vma)) | |
5617 | continue; | |
4ffef5fe DN |
5618 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
5619 | &mem_cgroup_move_charge_walk); | |
5620 | if (ret) | |
5621 | /* | |
5622 | * means we have consumed all precharges and failed in | |
5623 | * doing additional charge. Just abandon here. | |
5624 | */ | |
5625 | break; | |
5626 | } | |
dfe076b0 | 5627 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
5628 | } |
5629 | ||
67e465a7 BS |
5630 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
5631 | struct cgroup *cont, | |
5632 | struct cgroup *old_cont, | |
f780bdb7 | 5633 | struct task_struct *p) |
67e465a7 | 5634 | { |
a433658c | 5635 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 5636 | |
dfe076b0 | 5637 | if (mm) { |
a433658c KM |
5638 | if (mc.to) |
5639 | mem_cgroup_move_charge(mm); | |
5640 | put_swap_token(mm); | |
dfe076b0 DN |
5641 | mmput(mm); |
5642 | } | |
a433658c KM |
5643 | if (mc.to) |
5644 | mem_cgroup_clear_mc(); | |
67e465a7 | 5645 | } |
5cfb80a7 DN |
5646 | #else /* !CONFIG_MMU */ |
5647 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, | |
5648 | struct cgroup *cgroup, | |
f780bdb7 | 5649 | struct task_struct *p) |
5cfb80a7 DN |
5650 | { |
5651 | return 0; | |
5652 | } | |
5653 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
5654 | struct cgroup *cgroup, | |
f780bdb7 | 5655 | struct task_struct *p) |
5cfb80a7 DN |
5656 | { |
5657 | } | |
5658 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, | |
5659 | struct cgroup *cont, | |
5660 | struct cgroup *old_cont, | |
f780bdb7 | 5661 | struct task_struct *p) |
5cfb80a7 DN |
5662 | { |
5663 | } | |
5664 | #endif | |
67e465a7 | 5665 | |
8cdea7c0 BS |
5666 | struct cgroup_subsys mem_cgroup_subsys = { |
5667 | .name = "memory", | |
5668 | .subsys_id = mem_cgroup_subsys_id, | |
5669 | .create = mem_cgroup_create, | |
df878fb0 | 5670 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
5671 | .destroy = mem_cgroup_destroy, |
5672 | .populate = mem_cgroup_populate, | |
7dc74be0 DN |
5673 | .can_attach = mem_cgroup_can_attach, |
5674 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5675 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 5676 | .early_init = 0, |
04046e1a | 5677 | .use_id = 1, |
8cdea7c0 | 5678 | }; |
c077719b KH |
5679 | |
5680 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
a42c390c MH |
5681 | static int __init enable_swap_account(char *s) |
5682 | { | |
5683 | /* consider enabled if no parameter or 1 is given */ | |
a2c8990a | 5684 | if (!strcmp(s, "1")) |
a42c390c | 5685 | really_do_swap_account = 1; |
a2c8990a | 5686 | else if (!strcmp(s, "0")) |
a42c390c MH |
5687 | really_do_swap_account = 0; |
5688 | return 1; | |
5689 | } | |
a2c8990a | 5690 | __setup("swapaccount=", enable_swap_account); |
c077719b | 5691 | |
c077719b | 5692 | #endif |