]>
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 | * | |
7ae1e1d0 GC |
13 | * Kernel Memory Controller |
14 | * Copyright (C) 2012 Parallels Inc. and Google Inc. | |
15 | * Authors: Glauber Costa and Suleiman Souhlal | |
16 | * | |
8cdea7c0 BS |
17 | * This program is free software; you can redistribute it and/or modify |
18 | * it under the terms of the GNU General Public License as published by | |
19 | * the Free Software Foundation; either version 2 of the License, or | |
20 | * (at your option) any later version. | |
21 | * | |
22 | * This program is distributed in the hope that it will be useful, | |
23 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
24 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
25 | * GNU General Public License for more details. | |
26 | */ | |
27 | ||
28 | #include <linux/res_counter.h> | |
29 | #include <linux/memcontrol.h> | |
30 | #include <linux/cgroup.h> | |
78fb7466 | 31 | #include <linux/mm.h> |
4ffef5fe | 32 | #include <linux/hugetlb.h> |
d13d1443 | 33 | #include <linux/pagemap.h> |
d52aa412 | 34 | #include <linux/smp.h> |
8a9f3ccd | 35 | #include <linux/page-flags.h> |
66e1707b | 36 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
37 | #include <linux/bit_spinlock.h> |
38 | #include <linux/rcupdate.h> | |
e222432b | 39 | #include <linux/limits.h> |
b9e15baf | 40 | #include <linux/export.h> |
8c7c6e34 | 41 | #include <linux/mutex.h> |
f64c3f54 | 42 | #include <linux/rbtree.h> |
b6ac57d5 | 43 | #include <linux/slab.h> |
66e1707b | 44 | #include <linux/swap.h> |
02491447 | 45 | #include <linux/swapops.h> |
66e1707b | 46 | #include <linux/spinlock.h> |
2e72b634 KS |
47 | #include <linux/eventfd.h> |
48 | #include <linux/sort.h> | |
66e1707b | 49 | #include <linux/fs.h> |
d2ceb9b7 | 50 | #include <linux/seq_file.h> |
33327948 | 51 | #include <linux/vmalloc.h> |
70ddf637 | 52 | #include <linux/vmpressure.h> |
b69408e8 | 53 | #include <linux/mm_inline.h> |
52d4b9ac | 54 | #include <linux/page_cgroup.h> |
cdec2e42 | 55 | #include <linux/cpu.h> |
158e0a2d | 56 | #include <linux/oom.h> |
08e552c6 | 57 | #include "internal.h" |
d1a4c0b3 | 58 | #include <net/sock.h> |
4bd2c1ee | 59 | #include <net/ip.h> |
d1a4c0b3 | 60 | #include <net/tcp_memcontrol.h> |
8cdea7c0 | 61 | |
8697d331 BS |
62 | #include <asm/uaccess.h> |
63 | ||
cc8e970c KM |
64 | #include <trace/events/vmscan.h> |
65 | ||
a181b0e8 | 66 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
68ae564b DR |
67 | EXPORT_SYMBOL(mem_cgroup_subsys); |
68 | ||
a181b0e8 | 69 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
6bbda35c | 70 | static struct mem_cgroup *root_mem_cgroup __read_mostly; |
8cdea7c0 | 71 | |
c255a458 | 72 | #ifdef CONFIG_MEMCG_SWAP |
338c8431 | 73 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ |
c077719b | 74 | int do_swap_account __read_mostly; |
a42c390c MH |
75 | |
76 | /* for remember boot option*/ | |
c255a458 | 77 | #ifdef CONFIG_MEMCG_SWAP_ENABLED |
a42c390c MH |
78 | static int really_do_swap_account __initdata = 1; |
79 | #else | |
80 | static int really_do_swap_account __initdata = 0; | |
81 | #endif | |
82 | ||
c077719b | 83 | #else |
a0db00fc | 84 | #define do_swap_account 0 |
c077719b KH |
85 | #endif |
86 | ||
87 | ||
d52aa412 KH |
88 | /* |
89 | * Statistics for memory cgroup. | |
90 | */ | |
91 | enum mem_cgroup_stat_index { | |
92 | /* | |
93 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
94 | */ | |
b070e65c DR |
95 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ |
96 | MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */ | |
97 | MEM_CGROUP_STAT_RSS_HUGE, /* # of pages charged as anon huge */ | |
98 | MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */ | |
99 | MEM_CGROUP_STAT_SWAP, /* # of pages, swapped out */ | |
d52aa412 KH |
100 | MEM_CGROUP_STAT_NSTATS, |
101 | }; | |
102 | ||
af7c4b0e JW |
103 | static const char * const mem_cgroup_stat_names[] = { |
104 | "cache", | |
105 | "rss", | |
b070e65c | 106 | "rss_huge", |
af7c4b0e JW |
107 | "mapped_file", |
108 | "swap", | |
109 | }; | |
110 | ||
e9f8974f JW |
111 | enum mem_cgroup_events_index { |
112 | MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */ | |
113 | MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */ | |
456f998e YH |
114 | MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */ |
115 | MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */ | |
e9f8974f JW |
116 | MEM_CGROUP_EVENTS_NSTATS, |
117 | }; | |
af7c4b0e JW |
118 | |
119 | static const char * const mem_cgroup_events_names[] = { | |
120 | "pgpgin", | |
121 | "pgpgout", | |
122 | "pgfault", | |
123 | "pgmajfault", | |
124 | }; | |
125 | ||
58cf188e SZ |
126 | static const char * const mem_cgroup_lru_names[] = { |
127 | "inactive_anon", | |
128 | "active_anon", | |
129 | "inactive_file", | |
130 | "active_file", | |
131 | "unevictable", | |
132 | }; | |
133 | ||
7a159cc9 JW |
134 | /* |
135 | * Per memcg event counter is incremented at every pagein/pageout. With THP, | |
136 | * it will be incremated by the number of pages. This counter is used for | |
137 | * for trigger some periodic events. This is straightforward and better | |
138 | * than using jiffies etc. to handle periodic memcg event. | |
139 | */ | |
140 | enum mem_cgroup_events_target { | |
141 | MEM_CGROUP_TARGET_THRESH, | |
142 | MEM_CGROUP_TARGET_SOFTLIMIT, | |
453a9bf3 | 143 | MEM_CGROUP_TARGET_NUMAINFO, |
7a159cc9 JW |
144 | MEM_CGROUP_NTARGETS, |
145 | }; | |
a0db00fc KS |
146 | #define THRESHOLDS_EVENTS_TARGET 128 |
147 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
148 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 149 | |
d52aa412 | 150 | struct mem_cgroup_stat_cpu { |
7a159cc9 | 151 | long count[MEM_CGROUP_STAT_NSTATS]; |
e9f8974f | 152 | unsigned long events[MEM_CGROUP_EVENTS_NSTATS]; |
13114716 | 153 | unsigned long nr_page_events; |
7a159cc9 | 154 | unsigned long targets[MEM_CGROUP_NTARGETS]; |
d52aa412 KH |
155 | }; |
156 | ||
527a5ec9 | 157 | struct mem_cgroup_reclaim_iter { |
5f578161 MH |
158 | /* |
159 | * last scanned hierarchy member. Valid only if last_dead_count | |
160 | * matches memcg->dead_count of the hierarchy root group. | |
161 | */ | |
542f85f9 | 162 | struct mem_cgroup *last_visited; |
5f578161 MH |
163 | unsigned long last_dead_count; |
164 | ||
527a5ec9 JW |
165 | /* scan generation, increased every round-trip */ |
166 | unsigned int generation; | |
167 | }; | |
168 | ||
6d12e2d8 KH |
169 | /* |
170 | * per-zone information in memory controller. | |
171 | */ | |
6d12e2d8 | 172 | struct mem_cgroup_per_zone { |
6290df54 | 173 | struct lruvec lruvec; |
1eb49272 | 174 | unsigned long lru_size[NR_LRU_LISTS]; |
3e2f41f1 | 175 | |
527a5ec9 JW |
176 | struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1]; |
177 | ||
f64c3f54 BS |
178 | struct rb_node tree_node; /* RB tree node */ |
179 | unsigned long long usage_in_excess;/* Set to the value by which */ | |
180 | /* the soft limit is exceeded*/ | |
181 | bool on_tree; | |
d79154bb | 182 | struct mem_cgroup *memcg; /* Back pointer, we cannot */ |
4e416953 | 183 | /* use container_of */ |
6d12e2d8 | 184 | }; |
6d12e2d8 KH |
185 | |
186 | struct mem_cgroup_per_node { | |
187 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
188 | }; | |
189 | ||
f64c3f54 BS |
190 | /* |
191 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
192 | * their hierarchy representation | |
193 | */ | |
194 | ||
195 | struct mem_cgroup_tree_per_zone { | |
196 | struct rb_root rb_root; | |
197 | spinlock_t lock; | |
198 | }; | |
199 | ||
200 | struct mem_cgroup_tree_per_node { | |
201 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
202 | }; | |
203 | ||
204 | struct mem_cgroup_tree { | |
205 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
206 | }; | |
207 | ||
208 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
209 | ||
2e72b634 KS |
210 | struct mem_cgroup_threshold { |
211 | struct eventfd_ctx *eventfd; | |
212 | u64 threshold; | |
213 | }; | |
214 | ||
9490ff27 | 215 | /* For threshold */ |
2e72b634 | 216 | struct mem_cgroup_threshold_ary { |
748dad36 | 217 | /* An array index points to threshold just below or equal to usage. */ |
5407a562 | 218 | int current_threshold; |
2e72b634 KS |
219 | /* Size of entries[] */ |
220 | unsigned int size; | |
221 | /* Array of thresholds */ | |
222 | struct mem_cgroup_threshold entries[0]; | |
223 | }; | |
2c488db2 KS |
224 | |
225 | struct mem_cgroup_thresholds { | |
226 | /* Primary thresholds array */ | |
227 | struct mem_cgroup_threshold_ary *primary; | |
228 | /* | |
229 | * Spare threshold array. | |
230 | * This is needed to make mem_cgroup_unregister_event() "never fail". | |
231 | * It must be able to store at least primary->size - 1 entries. | |
232 | */ | |
233 | struct mem_cgroup_threshold_ary *spare; | |
234 | }; | |
235 | ||
9490ff27 KH |
236 | /* for OOM */ |
237 | struct mem_cgroup_eventfd_list { | |
238 | struct list_head list; | |
239 | struct eventfd_ctx *eventfd; | |
240 | }; | |
2e72b634 | 241 | |
c0ff4b85 R |
242 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
243 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 244 | |
8cdea7c0 BS |
245 | /* |
246 | * The memory controller data structure. The memory controller controls both | |
247 | * page cache and RSS per cgroup. We would eventually like to provide | |
248 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
249 | * to help the administrator determine what knobs to tune. | |
250 | * | |
251 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
252 | * we hit the water mark. May be even add a low water mark, such that |
253 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
254 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
255 | */ |
256 | struct mem_cgroup { | |
257 | struct cgroup_subsys_state css; | |
258 | /* | |
259 | * the counter to account for memory usage | |
260 | */ | |
261 | struct res_counter res; | |
59927fb9 | 262 | |
70ddf637 AV |
263 | /* vmpressure notifications */ |
264 | struct vmpressure vmpressure; | |
265 | ||
465939a1 LZ |
266 | /* |
267 | * the counter to account for mem+swap usage. | |
268 | */ | |
269 | struct res_counter memsw; | |
59927fb9 | 270 | |
510fc4e1 GC |
271 | /* |
272 | * the counter to account for kernel memory usage. | |
273 | */ | |
274 | struct res_counter kmem; | |
18f59ea7 BS |
275 | /* |
276 | * Should the accounting and control be hierarchical, per subtree? | |
277 | */ | |
278 | bool use_hierarchy; | |
510fc4e1 | 279 | unsigned long kmem_account_flags; /* See KMEM_ACCOUNTED_*, below */ |
79dfdacc MH |
280 | |
281 | bool oom_lock; | |
282 | atomic_t under_oom; | |
283 | ||
1f4c025b | 284 | int swappiness; |
3c11ecf4 KH |
285 | /* OOM-Killer disable */ |
286 | int oom_kill_disable; | |
a7885eb8 | 287 | |
22a668d7 KH |
288 | /* set when res.limit == memsw.limit */ |
289 | bool memsw_is_minimum; | |
290 | ||
2e72b634 KS |
291 | /* protect arrays of thresholds */ |
292 | struct mutex thresholds_lock; | |
293 | ||
294 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 295 | struct mem_cgroup_thresholds thresholds; |
907860ed | 296 | |
2e72b634 | 297 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 298 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 299 | |
9490ff27 KH |
300 | /* For oom notifier event fd */ |
301 | struct list_head oom_notify; | |
185efc0f | 302 | |
7dc74be0 DN |
303 | /* |
304 | * Should we move charges of a task when a task is moved into this | |
305 | * mem_cgroup ? And what type of charges should we move ? | |
306 | */ | |
307 | unsigned long move_charge_at_immigrate; | |
619d094b KH |
308 | /* |
309 | * set > 0 if pages under this cgroup are moving to other cgroup. | |
310 | */ | |
311 | atomic_t moving_account; | |
312734c0 KH |
312 | /* taken only while moving_account > 0 */ |
313 | spinlock_t move_lock; | |
d52aa412 | 314 | /* |
c62b1a3b | 315 | * percpu counter. |
d52aa412 | 316 | */ |
3a7951b4 | 317 | struct mem_cgroup_stat_cpu __percpu *stat; |
711d3d2c KH |
318 | /* |
319 | * used when a cpu is offlined or other synchronizations | |
320 | * See mem_cgroup_read_stat(). | |
321 | */ | |
322 | struct mem_cgroup_stat_cpu nocpu_base; | |
323 | spinlock_t pcp_counter_lock; | |
d1a4c0b3 | 324 | |
5f578161 | 325 | atomic_t dead_count; |
4bd2c1ee | 326 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET) |
d1a4c0b3 GC |
327 | struct tcp_memcontrol tcp_mem; |
328 | #endif | |
2633d7a0 GC |
329 | #if defined(CONFIG_MEMCG_KMEM) |
330 | /* analogous to slab_common's slab_caches list. per-memcg */ | |
331 | struct list_head memcg_slab_caches; | |
332 | /* Not a spinlock, we can take a lot of time walking the list */ | |
333 | struct mutex slab_caches_mutex; | |
334 | /* Index in the kmem_cache->memcg_params->memcg_caches array */ | |
335 | int kmemcg_id; | |
336 | #endif | |
45cf7ebd GC |
337 | |
338 | int last_scanned_node; | |
339 | #if MAX_NUMNODES > 1 | |
340 | nodemask_t scan_nodes; | |
341 | atomic_t numainfo_events; | |
342 | atomic_t numainfo_updating; | |
343 | #endif | |
70ddf637 | 344 | |
54f72fe0 JW |
345 | struct mem_cgroup_per_node *nodeinfo[0]; |
346 | /* WARNING: nodeinfo must be the last member here */ | |
8cdea7c0 BS |
347 | }; |
348 | ||
45cf7ebd GC |
349 | static size_t memcg_size(void) |
350 | { | |
351 | return sizeof(struct mem_cgroup) + | |
352 | nr_node_ids * sizeof(struct mem_cgroup_per_node); | |
353 | } | |
354 | ||
510fc4e1 GC |
355 | /* internal only representation about the status of kmem accounting. */ |
356 | enum { | |
357 | KMEM_ACCOUNTED_ACTIVE = 0, /* accounted by this cgroup itself */ | |
a8964b9b | 358 | KMEM_ACCOUNTED_ACTIVATED, /* static key enabled. */ |
7de37682 | 359 | KMEM_ACCOUNTED_DEAD, /* dead memcg with pending kmem charges */ |
510fc4e1 GC |
360 | }; |
361 | ||
a8964b9b GC |
362 | /* We account when limit is on, but only after call sites are patched */ |
363 | #define KMEM_ACCOUNTED_MASK \ | |
364 | ((1 << KMEM_ACCOUNTED_ACTIVE) | (1 << KMEM_ACCOUNTED_ACTIVATED)) | |
510fc4e1 GC |
365 | |
366 | #ifdef CONFIG_MEMCG_KMEM | |
367 | static inline void memcg_kmem_set_active(struct mem_cgroup *memcg) | |
368 | { | |
369 | set_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
370 | } | |
7de37682 GC |
371 | |
372 | static bool memcg_kmem_is_active(struct mem_cgroup *memcg) | |
373 | { | |
374 | return test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
375 | } | |
376 | ||
a8964b9b GC |
377 | static void memcg_kmem_set_activated(struct mem_cgroup *memcg) |
378 | { | |
379 | set_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags); | |
380 | } | |
381 | ||
55007d84 GC |
382 | static void memcg_kmem_clear_activated(struct mem_cgroup *memcg) |
383 | { | |
384 | clear_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags); | |
385 | } | |
386 | ||
7de37682 GC |
387 | static void memcg_kmem_mark_dead(struct mem_cgroup *memcg) |
388 | { | |
10d5ebf4 LZ |
389 | /* |
390 | * Our caller must use css_get() first, because memcg_uncharge_kmem() | |
391 | * will call css_put() if it sees the memcg is dead. | |
392 | */ | |
393 | smp_wmb(); | |
7de37682 GC |
394 | if (test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags)) |
395 | set_bit(KMEM_ACCOUNTED_DEAD, &memcg->kmem_account_flags); | |
396 | } | |
397 | ||
398 | static bool memcg_kmem_test_and_clear_dead(struct mem_cgroup *memcg) | |
399 | { | |
400 | return test_and_clear_bit(KMEM_ACCOUNTED_DEAD, | |
401 | &memcg->kmem_account_flags); | |
402 | } | |
510fc4e1 GC |
403 | #endif |
404 | ||
7dc74be0 DN |
405 | /* Stuffs for move charges at task migration. */ |
406 | /* | |
ee5e8472 GC |
407 | * Types of charges to be moved. "move_charge_at_immitgrate" and |
408 | * "immigrate_flags" are treated as a left-shifted bitmap of these types. | |
7dc74be0 DN |
409 | */ |
410 | enum move_type { | |
4ffef5fe | 411 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 412 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
413 | NR_MOVE_TYPE, |
414 | }; | |
415 | ||
4ffef5fe DN |
416 | /* "mc" and its members are protected by cgroup_mutex */ |
417 | static struct move_charge_struct { | |
b1dd693e | 418 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
419 | struct mem_cgroup *from; |
420 | struct mem_cgroup *to; | |
ee5e8472 | 421 | unsigned long immigrate_flags; |
4ffef5fe | 422 | unsigned long precharge; |
854ffa8d | 423 | unsigned long moved_charge; |
483c30b5 | 424 | unsigned long moved_swap; |
8033b97c DN |
425 | struct task_struct *moving_task; /* a task moving charges */ |
426 | wait_queue_head_t waitq; /* a waitq for other context */ | |
427 | } mc = { | |
2bd9bb20 | 428 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
429 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
430 | }; | |
4ffef5fe | 431 | |
90254a65 DN |
432 | static bool move_anon(void) |
433 | { | |
ee5e8472 | 434 | return test_bit(MOVE_CHARGE_TYPE_ANON, &mc.immigrate_flags); |
90254a65 DN |
435 | } |
436 | ||
87946a72 DN |
437 | static bool move_file(void) |
438 | { | |
ee5e8472 | 439 | return test_bit(MOVE_CHARGE_TYPE_FILE, &mc.immigrate_flags); |
87946a72 DN |
440 | } |
441 | ||
4e416953 BS |
442 | /* |
443 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
444 | * limit reclaim to prevent infinite loops, if they ever occur. | |
445 | */ | |
a0db00fc KS |
446 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
447 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 | |
4e416953 | 448 | |
217bc319 KH |
449 | enum charge_type { |
450 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 451 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 452 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 453 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
454 | NR_CHARGE_TYPE, |
455 | }; | |
456 | ||
8c7c6e34 | 457 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
458 | enum res_type { |
459 | _MEM, | |
460 | _MEMSWAP, | |
461 | _OOM_TYPE, | |
510fc4e1 | 462 | _KMEM, |
86ae53e1 GC |
463 | }; |
464 | ||
a0db00fc KS |
465 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
466 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 467 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
468 | /* Used for OOM nofiier */ |
469 | #define OOM_CONTROL (0) | |
8c7c6e34 | 470 | |
75822b44 BS |
471 | /* |
472 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
473 | */ | |
474 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
475 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
476 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
477 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
478 | ||
0999821b GC |
479 | /* |
480 | * The memcg_create_mutex will be held whenever a new cgroup is created. | |
481 | * As a consequence, any change that needs to protect against new child cgroups | |
482 | * appearing has to hold it as well. | |
483 | */ | |
484 | static DEFINE_MUTEX(memcg_create_mutex); | |
485 | ||
b2145145 WL |
486 | static inline |
487 | struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s) | |
488 | { | |
489 | return container_of(s, struct mem_cgroup, css); | |
490 | } | |
491 | ||
70ddf637 AV |
492 | /* Some nice accessors for the vmpressure. */ |
493 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
494 | { | |
495 | if (!memcg) | |
496 | memcg = root_mem_cgroup; | |
497 | return &memcg->vmpressure; | |
498 | } | |
499 | ||
500 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
501 | { | |
502 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
503 | } | |
504 | ||
505 | struct vmpressure *css_to_vmpressure(struct cgroup_subsys_state *css) | |
506 | { | |
507 | return &mem_cgroup_from_css(css)->vmpressure; | |
508 | } | |
509 | ||
7ffc0edc MH |
510 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
511 | { | |
512 | return (memcg == root_mem_cgroup); | |
513 | } | |
514 | ||
e1aab161 | 515 | /* Writing them here to avoid exposing memcg's inner layout */ |
4bd2c1ee | 516 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
e1aab161 | 517 | |
e1aab161 GC |
518 | void sock_update_memcg(struct sock *sk) |
519 | { | |
376be5ff | 520 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 521 | struct mem_cgroup *memcg; |
3f134619 | 522 | struct cg_proto *cg_proto; |
e1aab161 GC |
523 | |
524 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
525 | ||
f3f511e1 GC |
526 | /* Socket cloning can throw us here with sk_cgrp already |
527 | * filled. It won't however, necessarily happen from | |
528 | * process context. So the test for root memcg given | |
529 | * the current task's memcg won't help us in this case. | |
530 | * | |
531 | * Respecting the original socket's memcg is a better | |
532 | * decision in this case. | |
533 | */ | |
534 | if (sk->sk_cgrp) { | |
535 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
5347e5ae | 536 | css_get(&sk->sk_cgrp->memcg->css); |
f3f511e1 GC |
537 | return; |
538 | } | |
539 | ||
e1aab161 GC |
540 | rcu_read_lock(); |
541 | memcg = mem_cgroup_from_task(current); | |
3f134619 | 542 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
5347e5ae LZ |
543 | if (!mem_cgroup_is_root(memcg) && |
544 | memcg_proto_active(cg_proto) && css_tryget(&memcg->css)) { | |
3f134619 | 545 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
546 | } |
547 | rcu_read_unlock(); | |
548 | } | |
549 | } | |
550 | EXPORT_SYMBOL(sock_update_memcg); | |
551 | ||
552 | void sock_release_memcg(struct sock *sk) | |
553 | { | |
376be5ff | 554 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
555 | struct mem_cgroup *memcg; |
556 | WARN_ON(!sk->sk_cgrp->memcg); | |
557 | memcg = sk->sk_cgrp->memcg; | |
5347e5ae | 558 | css_put(&sk->sk_cgrp->memcg->css); |
e1aab161 GC |
559 | } |
560 | } | |
d1a4c0b3 GC |
561 | |
562 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
563 | { | |
564 | if (!memcg || mem_cgroup_is_root(memcg)) | |
565 | return NULL; | |
566 | ||
567 | return &memcg->tcp_mem.cg_proto; | |
568 | } | |
569 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 570 | |
3f134619 GC |
571 | static void disarm_sock_keys(struct mem_cgroup *memcg) |
572 | { | |
573 | if (!memcg_proto_activated(&memcg->tcp_mem.cg_proto)) | |
574 | return; | |
575 | static_key_slow_dec(&memcg_socket_limit_enabled); | |
576 | } | |
577 | #else | |
578 | static void disarm_sock_keys(struct mem_cgroup *memcg) | |
579 | { | |
580 | } | |
581 | #endif | |
582 | ||
a8964b9b | 583 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 GC |
584 | /* |
585 | * This will be the memcg's index in each cache's ->memcg_params->memcg_caches. | |
586 | * There are two main reasons for not using the css_id for this: | |
587 | * 1) this works better in sparse environments, where we have a lot of memcgs, | |
588 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
589 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
590 | * 200 entry array for that. | |
591 | * | |
592 | * 2) In order not to violate the cgroup API, we would like to do all memory | |
593 | * allocation in ->create(). At that point, we haven't yet allocated the | |
594 | * css_id. Having a separate index prevents us from messing with the cgroup | |
595 | * core for this | |
596 | * | |
597 | * The current size of the caches array is stored in | |
598 | * memcg_limited_groups_array_size. It will double each time we have to | |
599 | * increase it. | |
600 | */ | |
601 | static DEFINE_IDA(kmem_limited_groups); | |
749c5415 GC |
602 | int memcg_limited_groups_array_size; |
603 | ||
55007d84 GC |
604 | /* |
605 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
606 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
607 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
608 | * tunable, but that is strictly not necessary. | |
609 | * | |
610 | * MAX_SIZE should be as large as the number of css_ids. Ideally, we could get | |
611 | * this constant directly from cgroup, but it is understandable that this is | |
612 | * better kept as an internal representation in cgroup.c. In any case, the | |
613 | * css_id space is not getting any smaller, and we don't have to necessarily | |
614 | * increase ours as well if it increases. | |
615 | */ | |
616 | #define MEMCG_CACHES_MIN_SIZE 4 | |
617 | #define MEMCG_CACHES_MAX_SIZE 65535 | |
618 | ||
d7f25f8a GC |
619 | /* |
620 | * A lot of the calls to the cache allocation functions are expected to be | |
621 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
622 | * conditional to this static branch, we'll have to allow modules that does | |
623 | * kmem_cache_alloc and the such to see this symbol as well | |
624 | */ | |
a8964b9b | 625 | struct static_key memcg_kmem_enabled_key; |
d7f25f8a | 626 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b GC |
627 | |
628 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
629 | { | |
55007d84 | 630 | if (memcg_kmem_is_active(memcg)) { |
a8964b9b | 631 | static_key_slow_dec(&memcg_kmem_enabled_key); |
55007d84 GC |
632 | ida_simple_remove(&kmem_limited_groups, memcg->kmemcg_id); |
633 | } | |
bea207c8 GC |
634 | /* |
635 | * This check can't live in kmem destruction function, | |
636 | * since the charges will outlive the cgroup | |
637 | */ | |
638 | WARN_ON(res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0); | |
a8964b9b GC |
639 | } |
640 | #else | |
641 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
642 | { | |
643 | } | |
644 | #endif /* CONFIG_MEMCG_KMEM */ | |
645 | ||
646 | static void disarm_static_keys(struct mem_cgroup *memcg) | |
647 | { | |
648 | disarm_sock_keys(memcg); | |
649 | disarm_kmem_keys(memcg); | |
650 | } | |
651 | ||
c0ff4b85 | 652 | static void drain_all_stock_async(struct mem_cgroup *memcg); |
8c7c6e34 | 653 | |
f64c3f54 | 654 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 655 | mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid) |
f64c3f54 | 656 | { |
45cf7ebd | 657 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
54f72fe0 | 658 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
659 | } |
660 | ||
c0ff4b85 | 661 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 662 | { |
c0ff4b85 | 663 | return &memcg->css; |
d324236b WF |
664 | } |
665 | ||
f64c3f54 | 666 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 667 | page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 668 | { |
97a6c37b JW |
669 | int nid = page_to_nid(page); |
670 | int zid = page_zonenum(page); | |
f64c3f54 | 671 | |
c0ff4b85 | 672 | return mem_cgroup_zoneinfo(memcg, nid, zid); |
f64c3f54 BS |
673 | } |
674 | ||
675 | static struct mem_cgroup_tree_per_zone * | |
676 | soft_limit_tree_node_zone(int nid, int zid) | |
677 | { | |
678 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
679 | } | |
680 | ||
681 | static struct mem_cgroup_tree_per_zone * | |
682 | soft_limit_tree_from_page(struct page *page) | |
683 | { | |
684 | int nid = page_to_nid(page); | |
685 | int zid = page_zonenum(page); | |
686 | ||
687 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
688 | } | |
689 | ||
690 | static void | |
c0ff4b85 | 691 | __mem_cgroup_insert_exceeded(struct mem_cgroup *memcg, |
f64c3f54 | 692 | struct mem_cgroup_per_zone *mz, |
ef8745c1 KH |
693 | struct mem_cgroup_tree_per_zone *mctz, |
694 | unsigned long long new_usage_in_excess) | |
f64c3f54 BS |
695 | { |
696 | struct rb_node **p = &mctz->rb_root.rb_node; | |
697 | struct rb_node *parent = NULL; | |
698 | struct mem_cgroup_per_zone *mz_node; | |
699 | ||
700 | if (mz->on_tree) | |
701 | return; | |
702 | ||
ef8745c1 KH |
703 | mz->usage_in_excess = new_usage_in_excess; |
704 | if (!mz->usage_in_excess) | |
705 | return; | |
f64c3f54 BS |
706 | while (*p) { |
707 | parent = *p; | |
708 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
709 | tree_node); | |
710 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
711 | p = &(*p)->rb_left; | |
712 | /* | |
713 | * We can't avoid mem cgroups that are over their soft | |
714 | * limit by the same amount | |
715 | */ | |
716 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
717 | p = &(*p)->rb_right; | |
718 | } | |
719 | rb_link_node(&mz->tree_node, parent, p); | |
720 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
721 | mz->on_tree = true; | |
4e416953 BS |
722 | } |
723 | ||
724 | static void | |
c0ff4b85 | 725 | __mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
4e416953 BS |
726 | struct mem_cgroup_per_zone *mz, |
727 | struct mem_cgroup_tree_per_zone *mctz) | |
728 | { | |
729 | if (!mz->on_tree) | |
730 | return; | |
731 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
732 | mz->on_tree = false; | |
733 | } | |
734 | ||
f64c3f54 | 735 | static void |
c0ff4b85 | 736 | mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
f64c3f54 BS |
737 | struct mem_cgroup_per_zone *mz, |
738 | struct mem_cgroup_tree_per_zone *mctz) | |
739 | { | |
740 | spin_lock(&mctz->lock); | |
c0ff4b85 | 741 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
742 | spin_unlock(&mctz->lock); |
743 | } | |
744 | ||
f64c3f54 | 745 | |
c0ff4b85 | 746 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 747 | { |
ef8745c1 | 748 | unsigned long long excess; |
f64c3f54 BS |
749 | struct mem_cgroup_per_zone *mz; |
750 | struct mem_cgroup_tree_per_zone *mctz; | |
4e649152 KH |
751 | int nid = page_to_nid(page); |
752 | int zid = page_zonenum(page); | |
f64c3f54 BS |
753 | mctz = soft_limit_tree_from_page(page); |
754 | ||
755 | /* | |
4e649152 KH |
756 | * Necessary to update all ancestors when hierarchy is used. |
757 | * because their event counter is not touched. | |
f64c3f54 | 758 | */ |
c0ff4b85 R |
759 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
760 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
761 | excess = res_counter_soft_limit_excess(&memcg->res); | |
4e649152 KH |
762 | /* |
763 | * We have to update the tree if mz is on RB-tree or | |
764 | * mem is over its softlimit. | |
765 | */ | |
ef8745c1 | 766 | if (excess || mz->on_tree) { |
4e649152 KH |
767 | spin_lock(&mctz->lock); |
768 | /* if on-tree, remove it */ | |
769 | if (mz->on_tree) | |
c0ff4b85 | 770 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
4e649152 | 771 | /* |
ef8745c1 KH |
772 | * Insert again. mz->usage_in_excess will be updated. |
773 | * If excess is 0, no tree ops. | |
4e649152 | 774 | */ |
c0ff4b85 | 775 | __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess); |
4e649152 KH |
776 | spin_unlock(&mctz->lock); |
777 | } | |
f64c3f54 BS |
778 | } |
779 | } | |
780 | ||
c0ff4b85 | 781 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) |
f64c3f54 BS |
782 | { |
783 | int node, zone; | |
784 | struct mem_cgroup_per_zone *mz; | |
785 | struct mem_cgroup_tree_per_zone *mctz; | |
786 | ||
3ed28fa1 | 787 | for_each_node(node) { |
f64c3f54 | 788 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
c0ff4b85 | 789 | mz = mem_cgroup_zoneinfo(memcg, node, zone); |
f64c3f54 | 790 | mctz = soft_limit_tree_node_zone(node, zone); |
c0ff4b85 | 791 | mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
792 | } |
793 | } | |
794 | } | |
795 | ||
4e416953 BS |
796 | static struct mem_cgroup_per_zone * |
797 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
798 | { | |
799 | struct rb_node *rightmost = NULL; | |
26251eaf | 800 | struct mem_cgroup_per_zone *mz; |
4e416953 BS |
801 | |
802 | retry: | |
26251eaf | 803 | mz = NULL; |
4e416953 BS |
804 | rightmost = rb_last(&mctz->rb_root); |
805 | if (!rightmost) | |
806 | goto done; /* Nothing to reclaim from */ | |
807 | ||
808 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
809 | /* | |
810 | * Remove the node now but someone else can add it back, | |
811 | * we will to add it back at the end of reclaim to its correct | |
812 | * position in the tree. | |
813 | */ | |
d79154bb HD |
814 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); |
815 | if (!res_counter_soft_limit_excess(&mz->memcg->res) || | |
816 | !css_tryget(&mz->memcg->css)) | |
4e416953 BS |
817 | goto retry; |
818 | done: | |
819 | return mz; | |
820 | } | |
821 | ||
822 | static struct mem_cgroup_per_zone * | |
823 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
824 | { | |
825 | struct mem_cgroup_per_zone *mz; | |
826 | ||
827 | spin_lock(&mctz->lock); | |
828 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
829 | spin_unlock(&mctz->lock); | |
830 | return mz; | |
831 | } | |
832 | ||
711d3d2c KH |
833 | /* |
834 | * Implementation Note: reading percpu statistics for memcg. | |
835 | * | |
836 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
837 | * synchronization to implement "quick" read. There are trade-off between | |
838 | * reading cost and precision of value. Then, we may have a chance to implement | |
839 | * a periodic synchronizion of counter in memcg's counter. | |
840 | * | |
841 | * But this _read() function is used for user interface now. The user accounts | |
842 | * memory usage by memory cgroup and he _always_ requires exact value because | |
843 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
844 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
845 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
846 | * | |
847 | * If there are kernel internal actions which can make use of some not-exact | |
848 | * value, and reading all cpu value can be performance bottleneck in some | |
849 | * common workload, threashold and synchonization as vmstat[] should be | |
850 | * implemented. | |
851 | */ | |
c0ff4b85 | 852 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 853 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 854 | { |
7a159cc9 | 855 | long val = 0; |
c62b1a3b | 856 | int cpu; |
c62b1a3b | 857 | |
711d3d2c KH |
858 | get_online_cpus(); |
859 | for_each_online_cpu(cpu) | |
c0ff4b85 | 860 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 861 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
862 | spin_lock(&memcg->pcp_counter_lock); |
863 | val += memcg->nocpu_base.count[idx]; | |
864 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
865 | #endif |
866 | put_online_cpus(); | |
c62b1a3b KH |
867 | return val; |
868 | } | |
869 | ||
c0ff4b85 | 870 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
0c3e73e8 BS |
871 | bool charge) |
872 | { | |
873 | int val = (charge) ? 1 : -1; | |
bff6bb83 | 874 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); |
0c3e73e8 BS |
875 | } |
876 | ||
c0ff4b85 | 877 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
878 | enum mem_cgroup_events_index idx) |
879 | { | |
880 | unsigned long val = 0; | |
881 | int cpu; | |
882 | ||
883 | for_each_online_cpu(cpu) | |
c0ff4b85 | 884 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 885 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
886 | spin_lock(&memcg->pcp_counter_lock); |
887 | val += memcg->nocpu_base.events[idx]; | |
888 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f JW |
889 | #endif |
890 | return val; | |
891 | } | |
892 | ||
c0ff4b85 | 893 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 894 | struct page *page, |
b2402857 | 895 | bool anon, int nr_pages) |
d52aa412 | 896 | { |
c62b1a3b KH |
897 | preempt_disable(); |
898 | ||
b2402857 KH |
899 | /* |
900 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
901 | * counted as CACHE even if it's on ANON LRU. | |
902 | */ | |
903 | if (anon) | |
904 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], | |
c0ff4b85 | 905 | nr_pages); |
d52aa412 | 906 | else |
b2402857 | 907 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 908 | nr_pages); |
55e462b0 | 909 | |
b070e65c DR |
910 | if (PageTransHuge(page)) |
911 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], | |
912 | nr_pages); | |
913 | ||
e401f176 KH |
914 | /* pagein of a big page is an event. So, ignore page size */ |
915 | if (nr_pages > 0) | |
c0ff4b85 | 916 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 917 | else { |
c0ff4b85 | 918 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
919 | nr_pages = -nr_pages; /* for event */ |
920 | } | |
e401f176 | 921 | |
13114716 | 922 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
2e72b634 | 923 | |
c62b1a3b | 924 | preempt_enable(); |
6d12e2d8 KH |
925 | } |
926 | ||
bb2a0de9 | 927 | unsigned long |
4d7dcca2 | 928 | mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
929 | { |
930 | struct mem_cgroup_per_zone *mz; | |
931 | ||
932 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
933 | return mz->lru_size[lru]; | |
934 | } | |
935 | ||
936 | static unsigned long | |
c0ff4b85 | 937 | mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid, |
bb2a0de9 | 938 | unsigned int lru_mask) |
889976db YH |
939 | { |
940 | struct mem_cgroup_per_zone *mz; | |
f156ab93 | 941 | enum lru_list lru; |
bb2a0de9 KH |
942 | unsigned long ret = 0; |
943 | ||
c0ff4b85 | 944 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
bb2a0de9 | 945 | |
f156ab93 HD |
946 | for_each_lru(lru) { |
947 | if (BIT(lru) & lru_mask) | |
948 | ret += mz->lru_size[lru]; | |
bb2a0de9 KH |
949 | } |
950 | return ret; | |
951 | } | |
952 | ||
953 | static unsigned long | |
c0ff4b85 | 954 | mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 KH |
955 | int nid, unsigned int lru_mask) |
956 | { | |
889976db YH |
957 | u64 total = 0; |
958 | int zid; | |
959 | ||
bb2a0de9 | 960 | for (zid = 0; zid < MAX_NR_ZONES; zid++) |
c0ff4b85 R |
961 | total += mem_cgroup_zone_nr_lru_pages(memcg, |
962 | nid, zid, lru_mask); | |
bb2a0de9 | 963 | |
889976db YH |
964 | return total; |
965 | } | |
bb2a0de9 | 966 | |
c0ff4b85 | 967 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 968 | unsigned int lru_mask) |
6d12e2d8 | 969 | { |
889976db | 970 | int nid; |
6d12e2d8 KH |
971 | u64 total = 0; |
972 | ||
31aaea4a | 973 | for_each_node_state(nid, N_MEMORY) |
c0ff4b85 | 974 | total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
6d12e2d8 | 975 | return total; |
d52aa412 KH |
976 | } |
977 | ||
f53d7ce3 JW |
978 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
979 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
980 | { |
981 | unsigned long val, next; | |
982 | ||
13114716 | 983 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 984 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 985 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
986 | if ((long)next - (long)val < 0) { |
987 | switch (target) { | |
988 | case MEM_CGROUP_TARGET_THRESH: | |
989 | next = val + THRESHOLDS_EVENTS_TARGET; | |
990 | break; | |
991 | case MEM_CGROUP_TARGET_SOFTLIMIT: | |
992 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
993 | break; | |
994 | case MEM_CGROUP_TARGET_NUMAINFO: | |
995 | next = val + NUMAINFO_EVENTS_TARGET; | |
996 | break; | |
997 | default: | |
998 | break; | |
999 | } | |
1000 | __this_cpu_write(memcg->stat->targets[target], next); | |
1001 | return true; | |
7a159cc9 | 1002 | } |
f53d7ce3 | 1003 | return false; |
d2265e6f KH |
1004 | } |
1005 | ||
1006 | /* | |
1007 | * Check events in order. | |
1008 | * | |
1009 | */ | |
c0ff4b85 | 1010 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f | 1011 | { |
4799401f | 1012 | preempt_disable(); |
d2265e6f | 1013 | /* threshold event is triggered in finer grain than soft limit */ |
f53d7ce3 JW |
1014 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
1015 | MEM_CGROUP_TARGET_THRESH))) { | |
82b3f2a7 AM |
1016 | bool do_softlimit; |
1017 | bool do_numainfo __maybe_unused; | |
f53d7ce3 JW |
1018 | |
1019 | do_softlimit = mem_cgroup_event_ratelimit(memcg, | |
1020 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
1021 | #if MAX_NUMNODES > 1 | |
1022 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
1023 | MEM_CGROUP_TARGET_NUMAINFO); | |
1024 | #endif | |
1025 | preempt_enable(); | |
1026 | ||
c0ff4b85 | 1027 | mem_cgroup_threshold(memcg); |
f53d7ce3 | 1028 | if (unlikely(do_softlimit)) |
c0ff4b85 | 1029 | mem_cgroup_update_tree(memcg, page); |
453a9bf3 | 1030 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 1031 | if (unlikely(do_numainfo)) |
c0ff4b85 | 1032 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 1033 | #endif |
f53d7ce3 JW |
1034 | } else |
1035 | preempt_enable(); | |
d2265e6f KH |
1036 | } |
1037 | ||
d1a4c0b3 | 1038 | struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 | 1039 | { |
b2145145 WL |
1040 | return mem_cgroup_from_css( |
1041 | cgroup_subsys_state(cont, mem_cgroup_subsys_id)); | |
8cdea7c0 BS |
1042 | } |
1043 | ||
cf475ad2 | 1044 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 1045 | { |
31a78f23 BS |
1046 | /* |
1047 | * mm_update_next_owner() may clear mm->owner to NULL | |
1048 | * if it races with swapoff, page migration, etc. | |
1049 | * So this can be called with p == NULL. | |
1050 | */ | |
1051 | if (unlikely(!p)) | |
1052 | return NULL; | |
1053 | ||
b2145145 | 1054 | return mem_cgroup_from_css(task_subsys_state(p, mem_cgroup_subsys_id)); |
78fb7466 PE |
1055 | } |
1056 | ||
a433658c | 1057 | struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 1058 | { |
c0ff4b85 | 1059 | struct mem_cgroup *memcg = NULL; |
0b7f569e KH |
1060 | |
1061 | if (!mm) | |
1062 | return NULL; | |
54595fe2 KH |
1063 | /* |
1064 | * Because we have no locks, mm->owner's may be being moved to other | |
1065 | * cgroup. We use css_tryget() here even if this looks | |
1066 | * pessimistic (rather than adding locks here). | |
1067 | */ | |
1068 | rcu_read_lock(); | |
1069 | do { | |
c0ff4b85 R |
1070 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1071 | if (unlikely(!memcg)) | |
54595fe2 | 1072 | break; |
c0ff4b85 | 1073 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 1074 | rcu_read_unlock(); |
c0ff4b85 | 1075 | return memcg; |
54595fe2 KH |
1076 | } |
1077 | ||
16248d8f MH |
1078 | /* |
1079 | * Returns a next (in a pre-order walk) alive memcg (with elevated css | |
1080 | * ref. count) or NULL if the whole root's subtree has been visited. | |
1081 | * | |
1082 | * helper function to be used by mem_cgroup_iter | |
1083 | */ | |
1084 | static struct mem_cgroup *__mem_cgroup_iter_next(struct mem_cgroup *root, | |
1085 | struct mem_cgroup *last_visited) | |
1086 | { | |
1087 | struct cgroup *prev_cgroup, *next_cgroup; | |
1088 | ||
1089 | /* | |
1090 | * Root is not visited by cgroup iterators so it needs an | |
1091 | * explicit visit. | |
1092 | */ | |
1093 | if (!last_visited) | |
1094 | return root; | |
1095 | ||
1096 | prev_cgroup = (last_visited == root) ? NULL | |
1097 | : last_visited->css.cgroup; | |
1098 | skip_node: | |
1099 | next_cgroup = cgroup_next_descendant_pre( | |
1100 | prev_cgroup, root->css.cgroup); | |
1101 | ||
1102 | /* | |
1103 | * Even if we found a group we have to make sure it is | |
1104 | * alive. css && !memcg means that the groups should be | |
1105 | * skipped and we should continue the tree walk. | |
1106 | * last_visited css is safe to use because it is | |
1107 | * protected by css_get and the tree walk is rcu safe. | |
1108 | */ | |
1109 | if (next_cgroup) { | |
1110 | struct mem_cgroup *mem = mem_cgroup_from_cont( | |
1111 | next_cgroup); | |
1112 | if (css_tryget(&mem->css)) | |
1113 | return mem; | |
1114 | else { | |
1115 | prev_cgroup = next_cgroup; | |
1116 | goto skip_node; | |
1117 | } | |
1118 | } | |
1119 | ||
1120 | return NULL; | |
1121 | } | |
1122 | ||
519ebea3 JW |
1123 | static void mem_cgroup_iter_invalidate(struct mem_cgroup *root) |
1124 | { | |
1125 | /* | |
1126 | * When a group in the hierarchy below root is destroyed, the | |
1127 | * hierarchy iterator can no longer be trusted since it might | |
1128 | * have pointed to the destroyed group. Invalidate it. | |
1129 | */ | |
1130 | atomic_inc(&root->dead_count); | |
1131 | } | |
1132 | ||
1133 | static struct mem_cgroup * | |
1134 | mem_cgroup_iter_load(struct mem_cgroup_reclaim_iter *iter, | |
1135 | struct mem_cgroup *root, | |
1136 | int *sequence) | |
1137 | { | |
1138 | struct mem_cgroup *position = NULL; | |
1139 | /* | |
1140 | * A cgroup destruction happens in two stages: offlining and | |
1141 | * release. They are separated by a RCU grace period. | |
1142 | * | |
1143 | * If the iterator is valid, we may still race with an | |
1144 | * offlining. The RCU lock ensures the object won't be | |
1145 | * released, tryget will fail if we lost the race. | |
1146 | */ | |
1147 | *sequence = atomic_read(&root->dead_count); | |
1148 | if (iter->last_dead_count == *sequence) { | |
1149 | smp_rmb(); | |
1150 | position = iter->last_visited; | |
1151 | if (position && !css_tryget(&position->css)) | |
1152 | position = NULL; | |
1153 | } | |
1154 | return position; | |
1155 | } | |
1156 | ||
1157 | static void mem_cgroup_iter_update(struct mem_cgroup_reclaim_iter *iter, | |
1158 | struct mem_cgroup *last_visited, | |
1159 | struct mem_cgroup *new_position, | |
1160 | int sequence) | |
1161 | { | |
1162 | if (last_visited) | |
1163 | css_put(&last_visited->css); | |
1164 | /* | |
1165 | * We store the sequence count from the time @last_visited was | |
1166 | * loaded successfully instead of rereading it here so that we | |
1167 | * don't lose destruction events in between. We could have | |
1168 | * raced with the destruction of @new_position after all. | |
1169 | */ | |
1170 | iter->last_visited = new_position; | |
1171 | smp_wmb(); | |
1172 | iter->last_dead_count = sequence; | |
1173 | } | |
1174 | ||
5660048c JW |
1175 | /** |
1176 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1177 | * @root: hierarchy root | |
1178 | * @prev: previously returned memcg, NULL on first invocation | |
1179 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1180 | * | |
1181 | * Returns references to children of the hierarchy below @root, or | |
1182 | * @root itself, or %NULL after a full round-trip. | |
1183 | * | |
1184 | * Caller must pass the return value in @prev on subsequent | |
1185 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1186 | * to cancel a hierarchy walk before the round-trip is complete. | |
1187 | * | |
1188 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
1189 | * divide up the memcgs in the hierarchy among all concurrent | |
1190 | * reclaimers operating on the same zone and priority. | |
1191 | */ | |
1192 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, | |
1193 | struct mem_cgroup *prev, | |
1194 | struct mem_cgroup_reclaim_cookie *reclaim) | |
14067bb3 | 1195 | { |
9f3a0d09 | 1196 | struct mem_cgroup *memcg = NULL; |
542f85f9 | 1197 | struct mem_cgroup *last_visited = NULL; |
711d3d2c | 1198 | |
5660048c JW |
1199 | if (mem_cgroup_disabled()) |
1200 | return NULL; | |
1201 | ||
9f3a0d09 JW |
1202 | if (!root) |
1203 | root = root_mem_cgroup; | |
7d74b06f | 1204 | |
9f3a0d09 | 1205 | if (prev && !reclaim) |
542f85f9 | 1206 | last_visited = prev; |
14067bb3 | 1207 | |
9f3a0d09 JW |
1208 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1209 | if (prev) | |
c40046f3 | 1210 | goto out_css_put; |
9f3a0d09 JW |
1211 | return root; |
1212 | } | |
14067bb3 | 1213 | |
542f85f9 | 1214 | rcu_read_lock(); |
9f3a0d09 | 1215 | while (!memcg) { |
527a5ec9 | 1216 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
519ebea3 | 1217 | int uninitialized_var(seq); |
711d3d2c | 1218 | |
527a5ec9 JW |
1219 | if (reclaim) { |
1220 | int nid = zone_to_nid(reclaim->zone); | |
1221 | int zid = zone_idx(reclaim->zone); | |
1222 | struct mem_cgroup_per_zone *mz; | |
1223 | ||
1224 | mz = mem_cgroup_zoneinfo(root, nid, zid); | |
1225 | iter = &mz->reclaim_iter[reclaim->priority]; | |
542f85f9 | 1226 | if (prev && reclaim->generation != iter->generation) { |
5f578161 | 1227 | iter->last_visited = NULL; |
542f85f9 MH |
1228 | goto out_unlock; |
1229 | } | |
5f578161 | 1230 | |
519ebea3 | 1231 | last_visited = mem_cgroup_iter_load(iter, root, &seq); |
527a5ec9 | 1232 | } |
7d74b06f | 1233 | |
16248d8f | 1234 | memcg = __mem_cgroup_iter_next(root, last_visited); |
14067bb3 | 1235 | |
527a5ec9 | 1236 | if (reclaim) { |
519ebea3 | 1237 | mem_cgroup_iter_update(iter, last_visited, memcg, seq); |
542f85f9 | 1238 | |
19f39402 | 1239 | if (!memcg) |
527a5ec9 JW |
1240 | iter->generation++; |
1241 | else if (!prev && memcg) | |
1242 | reclaim->generation = iter->generation; | |
1243 | } | |
9f3a0d09 | 1244 | |
19f39402 | 1245 | if (prev && !memcg) |
542f85f9 | 1246 | goto out_unlock; |
9f3a0d09 | 1247 | } |
542f85f9 MH |
1248 | out_unlock: |
1249 | rcu_read_unlock(); | |
c40046f3 MH |
1250 | out_css_put: |
1251 | if (prev && prev != root) | |
1252 | css_put(&prev->css); | |
1253 | ||
9f3a0d09 | 1254 | return memcg; |
14067bb3 | 1255 | } |
7d74b06f | 1256 | |
5660048c JW |
1257 | /** |
1258 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1259 | * @root: hierarchy root | |
1260 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1261 | */ | |
1262 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1263 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1264 | { |
1265 | if (!root) | |
1266 | root = root_mem_cgroup; | |
1267 | if (prev && prev != root) | |
1268 | css_put(&prev->css); | |
1269 | } | |
7d74b06f | 1270 | |
9f3a0d09 JW |
1271 | /* |
1272 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1273 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1274 | * be used for reference counting. | |
1275 | */ | |
1276 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1277 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1278 | iter != NULL; \ |
527a5ec9 | 1279 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1280 | |
9f3a0d09 | 1281 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1282 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1283 | iter != NULL; \ |
527a5ec9 | 1284 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1285 | |
68ae564b | 1286 | void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
456f998e | 1287 | { |
c0ff4b85 | 1288 | struct mem_cgroup *memcg; |
456f998e | 1289 | |
456f998e | 1290 | rcu_read_lock(); |
c0ff4b85 R |
1291 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1292 | if (unlikely(!memcg)) | |
456f998e YH |
1293 | goto out; |
1294 | ||
1295 | switch (idx) { | |
456f998e | 1296 | case PGFAULT: |
0e574a93 JW |
1297 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1298 | break; | |
1299 | case PGMAJFAULT: | |
1300 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1301 | break; |
1302 | default: | |
1303 | BUG(); | |
1304 | } | |
1305 | out: | |
1306 | rcu_read_unlock(); | |
1307 | } | |
68ae564b | 1308 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
456f998e | 1309 | |
925b7673 JW |
1310 | /** |
1311 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1312 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1313 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1314 | * |
1315 | * Returns the lru list vector holding pages for the given @zone and | |
1316 | * @mem. This can be the global zone lruvec, if the memory controller | |
1317 | * is disabled. | |
1318 | */ | |
1319 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1320 | struct mem_cgroup *memcg) | |
1321 | { | |
1322 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1323 | struct lruvec *lruvec; |
925b7673 | 1324 | |
bea8c150 HD |
1325 | if (mem_cgroup_disabled()) { |
1326 | lruvec = &zone->lruvec; | |
1327 | goto out; | |
1328 | } | |
925b7673 JW |
1329 | |
1330 | mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone)); | |
bea8c150 HD |
1331 | lruvec = &mz->lruvec; |
1332 | out: | |
1333 | /* | |
1334 | * Since a node can be onlined after the mem_cgroup was created, | |
1335 | * we have to be prepared to initialize lruvec->zone here; | |
1336 | * and if offlined then reonlined, we need to reinitialize it. | |
1337 | */ | |
1338 | if (unlikely(lruvec->zone != zone)) | |
1339 | lruvec->zone = zone; | |
1340 | return lruvec; | |
925b7673 JW |
1341 | } |
1342 | ||
08e552c6 KH |
1343 | /* |
1344 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
1345 | * Operations are called by routine of global LRU independently from memcg. | |
1346 | * What we have to take care of here is validness of pc->mem_cgroup. | |
1347 | * | |
1348 | * Changes to pc->mem_cgroup happens when | |
1349 | * 1. charge | |
1350 | * 2. moving account | |
1351 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
1352 | * It is added to LRU before charge. | |
1353 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
1354 | * When moving account, the page is not on LRU. It's isolated. | |
1355 | */ | |
4f98a2fe | 1356 | |
925b7673 | 1357 | /** |
fa9add64 | 1358 | * mem_cgroup_page_lruvec - return lruvec for adding an lru page |
925b7673 | 1359 | * @page: the page |
fa9add64 | 1360 | * @zone: zone of the page |
925b7673 | 1361 | */ |
fa9add64 | 1362 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1363 | { |
08e552c6 | 1364 | struct mem_cgroup_per_zone *mz; |
925b7673 JW |
1365 | struct mem_cgroup *memcg; |
1366 | struct page_cgroup *pc; | |
bea8c150 | 1367 | struct lruvec *lruvec; |
6d12e2d8 | 1368 | |
bea8c150 HD |
1369 | if (mem_cgroup_disabled()) { |
1370 | lruvec = &zone->lruvec; | |
1371 | goto out; | |
1372 | } | |
925b7673 | 1373 | |
08e552c6 | 1374 | pc = lookup_page_cgroup(page); |
38c5d72f | 1375 | memcg = pc->mem_cgroup; |
7512102c HD |
1376 | |
1377 | /* | |
fa9add64 | 1378 | * Surreptitiously switch any uncharged offlist page to root: |
7512102c HD |
1379 | * an uncharged page off lru does nothing to secure |
1380 | * its former mem_cgroup from sudden removal. | |
1381 | * | |
1382 | * Our caller holds lru_lock, and PageCgroupUsed is updated | |
1383 | * under page_cgroup lock: between them, they make all uses | |
1384 | * of pc->mem_cgroup safe. | |
1385 | */ | |
fa9add64 | 1386 | if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup) |
7512102c HD |
1387 | pc->mem_cgroup = memcg = root_mem_cgroup; |
1388 | ||
925b7673 | 1389 | mz = page_cgroup_zoneinfo(memcg, page); |
bea8c150 HD |
1390 | lruvec = &mz->lruvec; |
1391 | out: | |
1392 | /* | |
1393 | * Since a node can be onlined after the mem_cgroup was created, | |
1394 | * we have to be prepared to initialize lruvec->zone here; | |
1395 | * and if offlined then reonlined, we need to reinitialize it. | |
1396 | */ | |
1397 | if (unlikely(lruvec->zone != zone)) | |
1398 | lruvec->zone = zone; | |
1399 | return lruvec; | |
08e552c6 | 1400 | } |
b69408e8 | 1401 | |
925b7673 | 1402 | /** |
fa9add64 HD |
1403 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1404 | * @lruvec: mem_cgroup per zone lru vector | |
1405 | * @lru: index of lru list the page is sitting on | |
1406 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1407 | * |
fa9add64 HD |
1408 | * This function must be called when a page is added to or removed from an |
1409 | * lru list. | |
3f58a829 | 1410 | */ |
fa9add64 HD |
1411 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1412 | int nr_pages) | |
3f58a829 MK |
1413 | { |
1414 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1415 | unsigned long *lru_size; |
3f58a829 MK |
1416 | |
1417 | if (mem_cgroup_disabled()) | |
1418 | return; | |
1419 | ||
fa9add64 HD |
1420 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1421 | lru_size = mz->lru_size + lru; | |
1422 | *lru_size += nr_pages; | |
1423 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1424 | } |
544122e5 | 1425 | |
3e92041d | 1426 | /* |
c0ff4b85 | 1427 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1428 | * hierarchy subtree |
1429 | */ | |
c3ac9a8a JW |
1430 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1431 | struct mem_cgroup *memcg) | |
3e92041d | 1432 | { |
91c63734 JW |
1433 | if (root_memcg == memcg) |
1434 | return true; | |
3a981f48 | 1435 | if (!root_memcg->use_hierarchy || !memcg) |
91c63734 | 1436 | return false; |
c3ac9a8a JW |
1437 | return css_is_ancestor(&memcg->css, &root_memcg->css); |
1438 | } | |
1439 | ||
1440 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, | |
1441 | struct mem_cgroup *memcg) | |
1442 | { | |
1443 | bool ret; | |
1444 | ||
91c63734 | 1445 | rcu_read_lock(); |
c3ac9a8a | 1446 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
91c63734 JW |
1447 | rcu_read_unlock(); |
1448 | return ret; | |
3e92041d MH |
1449 | } |
1450 | ||
ffbdccf5 DR |
1451 | bool task_in_mem_cgroup(struct task_struct *task, |
1452 | const struct mem_cgroup *memcg) | |
4c4a2214 | 1453 | { |
0b7f569e | 1454 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1455 | struct task_struct *p; |
ffbdccf5 | 1456 | bool ret; |
4c4a2214 | 1457 | |
158e0a2d | 1458 | p = find_lock_task_mm(task); |
de077d22 DR |
1459 | if (p) { |
1460 | curr = try_get_mem_cgroup_from_mm(p->mm); | |
1461 | task_unlock(p); | |
1462 | } else { | |
1463 | /* | |
1464 | * All threads may have already detached their mm's, but the oom | |
1465 | * killer still needs to detect if they have already been oom | |
1466 | * killed to prevent needlessly killing additional tasks. | |
1467 | */ | |
ffbdccf5 | 1468 | rcu_read_lock(); |
de077d22 DR |
1469 | curr = mem_cgroup_from_task(task); |
1470 | if (curr) | |
1471 | css_get(&curr->css); | |
ffbdccf5 | 1472 | rcu_read_unlock(); |
de077d22 | 1473 | } |
0b7f569e | 1474 | if (!curr) |
ffbdccf5 | 1475 | return false; |
d31f56db | 1476 | /* |
c0ff4b85 | 1477 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1478 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1479 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1480 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1481 | */ |
c0ff4b85 | 1482 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1483 | css_put(&curr->css); |
4c4a2214 DR |
1484 | return ret; |
1485 | } | |
1486 | ||
c56d5c7d | 1487 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1488 | { |
9b272977 | 1489 | unsigned long inactive_ratio; |
14797e23 | 1490 | unsigned long inactive; |
9b272977 | 1491 | unsigned long active; |
c772be93 | 1492 | unsigned long gb; |
14797e23 | 1493 | |
4d7dcca2 HD |
1494 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1495 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1496 | |
c772be93 KM |
1497 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1498 | if (gb) | |
1499 | inactive_ratio = int_sqrt(10 * gb); | |
1500 | else | |
1501 | inactive_ratio = 1; | |
1502 | ||
9b272977 | 1503 | return inactive * inactive_ratio < active; |
14797e23 KM |
1504 | } |
1505 | ||
6d61ef40 BS |
1506 | #define mem_cgroup_from_res_counter(counter, member) \ |
1507 | container_of(counter, struct mem_cgroup, member) | |
1508 | ||
19942822 | 1509 | /** |
9d11ea9f | 1510 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1511 | * @memcg: the memory cgroup |
19942822 | 1512 | * |
9d11ea9f | 1513 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1514 | * pages. |
19942822 | 1515 | */ |
c0ff4b85 | 1516 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1517 | { |
9d11ea9f JW |
1518 | unsigned long long margin; |
1519 | ||
c0ff4b85 | 1520 | margin = res_counter_margin(&memcg->res); |
9d11ea9f | 1521 | if (do_swap_account) |
c0ff4b85 | 1522 | margin = min(margin, res_counter_margin(&memcg->memsw)); |
7ec99d62 | 1523 | return margin >> PAGE_SHIFT; |
19942822 JW |
1524 | } |
1525 | ||
1f4c025b | 1526 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 KM |
1527 | { |
1528 | struct cgroup *cgrp = memcg->css.cgroup; | |
a7885eb8 KM |
1529 | |
1530 | /* root ? */ | |
1531 | if (cgrp->parent == NULL) | |
1532 | return vm_swappiness; | |
1533 | ||
bf1ff263 | 1534 | return memcg->swappiness; |
a7885eb8 KM |
1535 | } |
1536 | ||
619d094b KH |
1537 | /* |
1538 | * memcg->moving_account is used for checking possibility that some thread is | |
1539 | * calling move_account(). When a thread on CPU-A starts moving pages under | |
1540 | * a memcg, other threads should check memcg->moving_account under | |
1541 | * rcu_read_lock(), like this: | |
1542 | * | |
1543 | * CPU-A CPU-B | |
1544 | * rcu_read_lock() | |
1545 | * memcg->moving_account+1 if (memcg->mocing_account) | |
1546 | * take heavy locks. | |
1547 | * synchronize_rcu() update something. | |
1548 | * rcu_read_unlock() | |
1549 | * start move here. | |
1550 | */ | |
4331f7d3 KH |
1551 | |
1552 | /* for quick checking without looking up memcg */ | |
1553 | atomic_t memcg_moving __read_mostly; | |
1554 | ||
c0ff4b85 | 1555 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a | 1556 | { |
4331f7d3 | 1557 | atomic_inc(&memcg_moving); |
619d094b | 1558 | atomic_inc(&memcg->moving_account); |
32047e2a KH |
1559 | synchronize_rcu(); |
1560 | } | |
1561 | ||
c0ff4b85 | 1562 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a | 1563 | { |
619d094b KH |
1564 | /* |
1565 | * Now, mem_cgroup_clear_mc() may call this function with NULL. | |
1566 | * We check NULL in callee rather than caller. | |
1567 | */ | |
4331f7d3 KH |
1568 | if (memcg) { |
1569 | atomic_dec(&memcg_moving); | |
619d094b | 1570 | atomic_dec(&memcg->moving_account); |
4331f7d3 | 1571 | } |
32047e2a | 1572 | } |
619d094b | 1573 | |
32047e2a KH |
1574 | /* |
1575 | * 2 routines for checking "mem" is under move_account() or not. | |
1576 | * | |
13fd1dd9 AM |
1577 | * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This |
1578 | * is used for avoiding races in accounting. If true, | |
32047e2a KH |
1579 | * pc->mem_cgroup may be overwritten. |
1580 | * | |
1581 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or | |
1582 | * under hierarchy of moving cgroups. This is for | |
1583 | * waiting at hith-memory prressure caused by "move". | |
1584 | */ | |
1585 | ||
13fd1dd9 | 1586 | static bool mem_cgroup_stolen(struct mem_cgroup *memcg) |
32047e2a KH |
1587 | { |
1588 | VM_BUG_ON(!rcu_read_lock_held()); | |
619d094b | 1589 | return atomic_read(&memcg->moving_account) > 0; |
32047e2a | 1590 | } |
4b534334 | 1591 | |
c0ff4b85 | 1592 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1593 | { |
2bd9bb20 KH |
1594 | struct mem_cgroup *from; |
1595 | struct mem_cgroup *to; | |
4b534334 | 1596 | bool ret = false; |
2bd9bb20 KH |
1597 | /* |
1598 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1599 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1600 | */ | |
1601 | spin_lock(&mc.lock); | |
1602 | from = mc.from; | |
1603 | to = mc.to; | |
1604 | if (!from) | |
1605 | goto unlock; | |
3e92041d | 1606 | |
c0ff4b85 R |
1607 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1608 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1609 | unlock: |
1610 | spin_unlock(&mc.lock); | |
4b534334 KH |
1611 | return ret; |
1612 | } | |
1613 | ||
c0ff4b85 | 1614 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1615 | { |
1616 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1617 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1618 | DEFINE_WAIT(wait); |
1619 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1620 | /* moving charge context might have finished. */ | |
1621 | if (mc.moving_task) | |
1622 | schedule(); | |
1623 | finish_wait(&mc.waitq, &wait); | |
1624 | return true; | |
1625 | } | |
1626 | } | |
1627 | return false; | |
1628 | } | |
1629 | ||
312734c0 KH |
1630 | /* |
1631 | * Take this lock when | |
1632 | * - a code tries to modify page's memcg while it's USED. | |
1633 | * - a code tries to modify page state accounting in a memcg. | |
13fd1dd9 | 1634 | * see mem_cgroup_stolen(), too. |
312734c0 KH |
1635 | */ |
1636 | static void move_lock_mem_cgroup(struct mem_cgroup *memcg, | |
1637 | unsigned long *flags) | |
1638 | { | |
1639 | spin_lock_irqsave(&memcg->move_lock, *flags); | |
1640 | } | |
1641 | ||
1642 | static void move_unlock_mem_cgroup(struct mem_cgroup *memcg, | |
1643 | unsigned long *flags) | |
1644 | { | |
1645 | spin_unlock_irqrestore(&memcg->move_lock, *flags); | |
1646 | } | |
1647 | ||
58cf188e | 1648 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1649 | /** |
58cf188e | 1650 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1651 | * @memcg: The memory cgroup that went over limit |
1652 | * @p: Task that is going to be killed | |
1653 | * | |
1654 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1655 | * enabled | |
1656 | */ | |
1657 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1658 | { | |
1659 | struct cgroup *task_cgrp; | |
1660 | struct cgroup *mem_cgrp; | |
1661 | /* | |
1662 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1663 | * on the assumption that OOM is serialized for memory controller. | |
1664 | * If this assumption is broken, revisit this code. | |
1665 | */ | |
1666 | static char memcg_name[PATH_MAX]; | |
1667 | int ret; | |
58cf188e SZ |
1668 | struct mem_cgroup *iter; |
1669 | unsigned int i; | |
e222432b | 1670 | |
58cf188e | 1671 | if (!p) |
e222432b BS |
1672 | return; |
1673 | ||
e222432b BS |
1674 | rcu_read_lock(); |
1675 | ||
1676 | mem_cgrp = memcg->css.cgroup; | |
1677 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1678 | ||
1679 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1680 | if (ret < 0) { | |
1681 | /* | |
1682 | * Unfortunately, we are unable to convert to a useful name | |
1683 | * But we'll still print out the usage information | |
1684 | */ | |
1685 | rcu_read_unlock(); | |
1686 | goto done; | |
1687 | } | |
1688 | rcu_read_unlock(); | |
1689 | ||
d045197f | 1690 | pr_info("Task in %s killed", memcg_name); |
e222432b BS |
1691 | |
1692 | rcu_read_lock(); | |
1693 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1694 | if (ret < 0) { | |
1695 | rcu_read_unlock(); | |
1696 | goto done; | |
1697 | } | |
1698 | rcu_read_unlock(); | |
1699 | ||
1700 | /* | |
1701 | * Continues from above, so we don't need an KERN_ level | |
1702 | */ | |
d045197f | 1703 | pr_cont(" as a result of limit of %s\n", memcg_name); |
e222432b BS |
1704 | done: |
1705 | ||
d045197f | 1706 | pr_info("memory: usage %llukB, limit %llukB, failcnt %llu\n", |
e222432b BS |
1707 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, |
1708 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1709 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
d045197f | 1710 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %llu\n", |
e222432b BS |
1711 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, |
1712 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1713 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
d045197f | 1714 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %llu\n", |
510fc4e1 GC |
1715 | res_counter_read_u64(&memcg->kmem, RES_USAGE) >> 10, |
1716 | res_counter_read_u64(&memcg->kmem, RES_LIMIT) >> 10, | |
1717 | res_counter_read_u64(&memcg->kmem, RES_FAILCNT)); | |
58cf188e SZ |
1718 | |
1719 | for_each_mem_cgroup_tree(iter, memcg) { | |
1720 | pr_info("Memory cgroup stats"); | |
1721 | ||
1722 | rcu_read_lock(); | |
1723 | ret = cgroup_path(iter->css.cgroup, memcg_name, PATH_MAX); | |
1724 | if (!ret) | |
1725 | pr_cont(" for %s", memcg_name); | |
1726 | rcu_read_unlock(); | |
1727 | pr_cont(":"); | |
1728 | ||
1729 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1730 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1731 | continue; | |
1732 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], | |
1733 | K(mem_cgroup_read_stat(iter, i))); | |
1734 | } | |
1735 | ||
1736 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1737 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1738 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1739 | ||
1740 | pr_cont("\n"); | |
1741 | } | |
e222432b BS |
1742 | } |
1743 | ||
81d39c20 KH |
1744 | /* |
1745 | * This function returns the number of memcg under hierarchy tree. Returns | |
1746 | * 1(self count) if no children. | |
1747 | */ | |
c0ff4b85 | 1748 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1749 | { |
1750 | int num = 0; | |
7d74b06f KH |
1751 | struct mem_cgroup *iter; |
1752 | ||
c0ff4b85 | 1753 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1754 | num++; |
81d39c20 KH |
1755 | return num; |
1756 | } | |
1757 | ||
a63d83f4 DR |
1758 | /* |
1759 | * Return the memory (and swap, if configured) limit for a memcg. | |
1760 | */ | |
9cbb78bb | 1761 | static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 DR |
1762 | { |
1763 | u64 limit; | |
a63d83f4 | 1764 | |
f3e8eb70 | 1765 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
f3e8eb70 | 1766 | |
a63d83f4 | 1767 | /* |
9a5a8f19 | 1768 | * Do not consider swap space if we cannot swap due to swappiness |
a63d83f4 | 1769 | */ |
9a5a8f19 MH |
1770 | if (mem_cgroup_swappiness(memcg)) { |
1771 | u64 memsw; | |
1772 | ||
1773 | limit += total_swap_pages << PAGE_SHIFT; | |
1774 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1775 | ||
1776 | /* | |
1777 | * If memsw is finite and limits the amount of swap space | |
1778 | * available to this memcg, return that limit. | |
1779 | */ | |
1780 | limit = min(limit, memsw); | |
1781 | } | |
1782 | ||
1783 | return limit; | |
a63d83f4 DR |
1784 | } |
1785 | ||
19965460 DR |
1786 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1787 | int order) | |
9cbb78bb DR |
1788 | { |
1789 | struct mem_cgroup *iter; | |
1790 | unsigned long chosen_points = 0; | |
1791 | unsigned long totalpages; | |
1792 | unsigned int points = 0; | |
1793 | struct task_struct *chosen = NULL; | |
1794 | ||
876aafbf | 1795 | /* |
465adcf1 DR |
1796 | * If current has a pending SIGKILL or is exiting, then automatically |
1797 | * select it. The goal is to allow it to allocate so that it may | |
1798 | * quickly exit and free its memory. | |
876aafbf | 1799 | */ |
465adcf1 | 1800 | if (fatal_signal_pending(current) || current->flags & PF_EXITING) { |
876aafbf DR |
1801 | set_thread_flag(TIF_MEMDIE); |
1802 | return; | |
1803 | } | |
1804 | ||
1805 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
9cbb78bb DR |
1806 | totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1; |
1807 | for_each_mem_cgroup_tree(iter, memcg) { | |
1808 | struct cgroup *cgroup = iter->css.cgroup; | |
1809 | struct cgroup_iter it; | |
1810 | struct task_struct *task; | |
1811 | ||
1812 | cgroup_iter_start(cgroup, &it); | |
1813 | while ((task = cgroup_iter_next(cgroup, &it))) { | |
1814 | switch (oom_scan_process_thread(task, totalpages, NULL, | |
1815 | false)) { | |
1816 | case OOM_SCAN_SELECT: | |
1817 | if (chosen) | |
1818 | put_task_struct(chosen); | |
1819 | chosen = task; | |
1820 | chosen_points = ULONG_MAX; | |
1821 | get_task_struct(chosen); | |
1822 | /* fall through */ | |
1823 | case OOM_SCAN_CONTINUE: | |
1824 | continue; | |
1825 | case OOM_SCAN_ABORT: | |
1826 | cgroup_iter_end(cgroup, &it); | |
1827 | mem_cgroup_iter_break(memcg, iter); | |
1828 | if (chosen) | |
1829 | put_task_struct(chosen); | |
1830 | return; | |
1831 | case OOM_SCAN_OK: | |
1832 | break; | |
1833 | }; | |
1834 | points = oom_badness(task, memcg, NULL, totalpages); | |
1835 | if (points > chosen_points) { | |
1836 | if (chosen) | |
1837 | put_task_struct(chosen); | |
1838 | chosen = task; | |
1839 | chosen_points = points; | |
1840 | get_task_struct(chosen); | |
1841 | } | |
1842 | } | |
1843 | cgroup_iter_end(cgroup, &it); | |
1844 | } | |
1845 | ||
1846 | if (!chosen) | |
1847 | return; | |
1848 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1849 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1850 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1851 | } |
1852 | ||
5660048c JW |
1853 | static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg, |
1854 | gfp_t gfp_mask, | |
1855 | unsigned long flags) | |
1856 | { | |
1857 | unsigned long total = 0; | |
1858 | bool noswap = false; | |
1859 | int loop; | |
1860 | ||
1861 | if (flags & MEM_CGROUP_RECLAIM_NOSWAP) | |
1862 | noswap = true; | |
1863 | if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum) | |
1864 | noswap = true; | |
1865 | ||
1866 | for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) { | |
1867 | if (loop) | |
1868 | drain_all_stock_async(memcg); | |
1869 | total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap); | |
1870 | /* | |
1871 | * Allow limit shrinkers, which are triggered directly | |
1872 | * by userspace, to catch signals and stop reclaim | |
1873 | * after minimal progress, regardless of the margin. | |
1874 | */ | |
1875 | if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK)) | |
1876 | break; | |
1877 | if (mem_cgroup_margin(memcg)) | |
1878 | break; | |
1879 | /* | |
1880 | * If nothing was reclaimed after two attempts, there | |
1881 | * may be no reclaimable pages in this hierarchy. | |
1882 | */ | |
1883 | if (loop && !total) | |
1884 | break; | |
1885 | } | |
1886 | return total; | |
1887 | } | |
1888 | ||
4d0c066d KH |
1889 | /** |
1890 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1891 | * @memcg: the target memcg |
4d0c066d KH |
1892 | * @nid: the node ID to be checked. |
1893 | * @noswap : specify true here if the user wants flle only information. | |
1894 | * | |
1895 | * This function returns whether the specified memcg contains any | |
1896 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1897 | * pages in the node. | |
1898 | */ | |
c0ff4b85 | 1899 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1900 | int nid, bool noswap) |
1901 | { | |
c0ff4b85 | 1902 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1903 | return true; |
1904 | if (noswap || !total_swap_pages) | |
1905 | return false; | |
c0ff4b85 | 1906 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1907 | return true; |
1908 | return false; | |
1909 | ||
1910 | } | |
889976db YH |
1911 | #if MAX_NUMNODES > 1 |
1912 | ||
1913 | /* | |
1914 | * Always updating the nodemask is not very good - even if we have an empty | |
1915 | * list or the wrong list here, we can start from some node and traverse all | |
1916 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1917 | * | |
1918 | */ | |
c0ff4b85 | 1919 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1920 | { |
1921 | int nid; | |
453a9bf3 KH |
1922 | /* |
1923 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1924 | * pagein/pageout changes since the last update. | |
1925 | */ | |
c0ff4b85 | 1926 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1927 | return; |
c0ff4b85 | 1928 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1929 | return; |
1930 | ||
889976db | 1931 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1932 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1933 | |
31aaea4a | 1934 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1935 | |
c0ff4b85 R |
1936 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1937 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1938 | } |
453a9bf3 | 1939 | |
c0ff4b85 R |
1940 | atomic_set(&memcg->numainfo_events, 0); |
1941 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1942 | } |
1943 | ||
1944 | /* | |
1945 | * Selecting a node where we start reclaim from. Because what we need is just | |
1946 | * reducing usage counter, start from anywhere is O,K. Considering | |
1947 | * memory reclaim from current node, there are pros. and cons. | |
1948 | * | |
1949 | * Freeing memory from current node means freeing memory from a node which | |
1950 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1951 | * hit limits, it will see a contention on a node. But freeing from remote | |
1952 | * node means more costs for memory reclaim because of memory latency. | |
1953 | * | |
1954 | * Now, we use round-robin. Better algorithm is welcomed. | |
1955 | */ | |
c0ff4b85 | 1956 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1957 | { |
1958 | int node; | |
1959 | ||
c0ff4b85 R |
1960 | mem_cgroup_may_update_nodemask(memcg); |
1961 | node = memcg->last_scanned_node; | |
889976db | 1962 | |
c0ff4b85 | 1963 | node = next_node(node, memcg->scan_nodes); |
889976db | 1964 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1965 | node = first_node(memcg->scan_nodes); |
889976db YH |
1966 | /* |
1967 | * We call this when we hit limit, not when pages are added to LRU. | |
1968 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1969 | * memcg is too small and all pages are not on LRU. In that case, | |
1970 | * we use curret node. | |
1971 | */ | |
1972 | if (unlikely(node == MAX_NUMNODES)) | |
1973 | node = numa_node_id(); | |
1974 | ||
c0ff4b85 | 1975 | memcg->last_scanned_node = node; |
889976db YH |
1976 | return node; |
1977 | } | |
1978 | ||
4d0c066d KH |
1979 | /* |
1980 | * Check all nodes whether it contains reclaimable pages or not. | |
1981 | * For quick scan, we make use of scan_nodes. This will allow us to skip | |
1982 | * unused nodes. But scan_nodes is lazily updated and may not cotain | |
1983 | * enough new information. We need to do double check. | |
1984 | */ | |
6bbda35c | 1985 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d KH |
1986 | { |
1987 | int nid; | |
1988 | ||
1989 | /* | |
1990 | * quick check...making use of scan_node. | |
1991 | * We can skip unused nodes. | |
1992 | */ | |
c0ff4b85 R |
1993 | if (!nodes_empty(memcg->scan_nodes)) { |
1994 | for (nid = first_node(memcg->scan_nodes); | |
4d0c066d | 1995 | nid < MAX_NUMNODES; |
c0ff4b85 | 1996 | nid = next_node(nid, memcg->scan_nodes)) { |
4d0c066d | 1997 | |
c0ff4b85 | 1998 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
1999 | return true; |
2000 | } | |
2001 | } | |
2002 | /* | |
2003 | * Check rest of nodes. | |
2004 | */ | |
31aaea4a | 2005 | for_each_node_state(nid, N_MEMORY) { |
c0ff4b85 | 2006 | if (node_isset(nid, memcg->scan_nodes)) |
4d0c066d | 2007 | continue; |
c0ff4b85 | 2008 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
2009 | return true; |
2010 | } | |
2011 | return false; | |
2012 | } | |
2013 | ||
889976db | 2014 | #else |
c0ff4b85 | 2015 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
2016 | { |
2017 | return 0; | |
2018 | } | |
4d0c066d | 2019 | |
6bbda35c | 2020 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d | 2021 | { |
c0ff4b85 | 2022 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); |
4d0c066d | 2023 | } |
889976db YH |
2024 | #endif |
2025 | ||
5660048c JW |
2026 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
2027 | struct zone *zone, | |
2028 | gfp_t gfp_mask, | |
2029 | unsigned long *total_scanned) | |
6d61ef40 | 2030 | { |
9f3a0d09 | 2031 | struct mem_cgroup *victim = NULL; |
5660048c | 2032 | int total = 0; |
04046e1a | 2033 | int loop = 0; |
9d11ea9f | 2034 | unsigned long excess; |
185efc0f | 2035 | unsigned long nr_scanned; |
527a5ec9 JW |
2036 | struct mem_cgroup_reclaim_cookie reclaim = { |
2037 | .zone = zone, | |
2038 | .priority = 0, | |
2039 | }; | |
9d11ea9f | 2040 | |
c0ff4b85 | 2041 | excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT; |
04046e1a | 2042 | |
4e416953 | 2043 | while (1) { |
527a5ec9 | 2044 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); |
9f3a0d09 | 2045 | if (!victim) { |
04046e1a | 2046 | loop++; |
4e416953 BS |
2047 | if (loop >= 2) { |
2048 | /* | |
2049 | * If we have not been able to reclaim | |
2050 | * anything, it might because there are | |
2051 | * no reclaimable pages under this hierarchy | |
2052 | */ | |
5660048c | 2053 | if (!total) |
4e416953 | 2054 | break; |
4e416953 | 2055 | /* |
25985edc | 2056 | * We want to do more targeted reclaim. |
4e416953 BS |
2057 | * excess >> 2 is not to excessive so as to |
2058 | * reclaim too much, nor too less that we keep | |
2059 | * coming back to reclaim from this cgroup | |
2060 | */ | |
2061 | if (total >= (excess >> 2) || | |
9f3a0d09 | 2062 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) |
4e416953 | 2063 | break; |
4e416953 | 2064 | } |
9f3a0d09 | 2065 | continue; |
4e416953 | 2066 | } |
5660048c | 2067 | if (!mem_cgroup_reclaimable(victim, false)) |
6d61ef40 | 2068 | continue; |
5660048c JW |
2069 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
2070 | zone, &nr_scanned); | |
2071 | *total_scanned += nr_scanned; | |
2072 | if (!res_counter_soft_limit_excess(&root_memcg->res)) | |
9f3a0d09 | 2073 | break; |
6d61ef40 | 2074 | } |
9f3a0d09 | 2075 | mem_cgroup_iter_break(root_memcg, victim); |
04046e1a | 2076 | return total; |
6d61ef40 BS |
2077 | } |
2078 | ||
867578cb KH |
2079 | /* |
2080 | * Check OOM-Killer is already running under our hierarchy. | |
2081 | * If someone is running, return false. | |
1af8efe9 | 2082 | * Has to be called with memcg_oom_lock |
867578cb | 2083 | */ |
c0ff4b85 | 2084 | static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg) |
867578cb | 2085 | { |
79dfdacc | 2086 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 2087 | |
9f3a0d09 | 2088 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 2089 | if (iter->oom_lock) { |
79dfdacc MH |
2090 | /* |
2091 | * this subtree of our hierarchy is already locked | |
2092 | * so we cannot give a lock. | |
2093 | */ | |
79dfdacc | 2094 | failed = iter; |
9f3a0d09 JW |
2095 | mem_cgroup_iter_break(memcg, iter); |
2096 | break; | |
23751be0 JW |
2097 | } else |
2098 | iter->oom_lock = true; | |
7d74b06f | 2099 | } |
867578cb | 2100 | |
79dfdacc | 2101 | if (!failed) |
23751be0 | 2102 | return true; |
79dfdacc MH |
2103 | |
2104 | /* | |
2105 | * OK, we failed to lock the whole subtree so we have to clean up | |
2106 | * what we set up to the failing subtree | |
2107 | */ | |
9f3a0d09 | 2108 | for_each_mem_cgroup_tree(iter, memcg) { |
79dfdacc | 2109 | if (iter == failed) { |
9f3a0d09 JW |
2110 | mem_cgroup_iter_break(memcg, iter); |
2111 | break; | |
79dfdacc MH |
2112 | } |
2113 | iter->oom_lock = false; | |
2114 | } | |
23751be0 | 2115 | return false; |
a636b327 | 2116 | } |
0b7f569e | 2117 | |
79dfdacc | 2118 | /* |
1af8efe9 | 2119 | * Has to be called with memcg_oom_lock |
79dfdacc | 2120 | */ |
c0ff4b85 | 2121 | static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 2122 | { |
7d74b06f KH |
2123 | struct mem_cgroup *iter; |
2124 | ||
c0ff4b85 | 2125 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
2126 | iter->oom_lock = false; |
2127 | return 0; | |
2128 | } | |
2129 | ||
c0ff4b85 | 2130 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2131 | { |
2132 | struct mem_cgroup *iter; | |
2133 | ||
c0ff4b85 | 2134 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
2135 | atomic_inc(&iter->under_oom); |
2136 | } | |
2137 | ||
c0ff4b85 | 2138 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2139 | { |
2140 | struct mem_cgroup *iter; | |
2141 | ||
867578cb KH |
2142 | /* |
2143 | * When a new child is created while the hierarchy is under oom, | |
2144 | * mem_cgroup_oom_lock() may not be called. We have to use | |
2145 | * atomic_add_unless() here. | |
2146 | */ | |
c0ff4b85 | 2147 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 2148 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
2149 | } |
2150 | ||
1af8efe9 | 2151 | static DEFINE_SPINLOCK(memcg_oom_lock); |
867578cb KH |
2152 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
2153 | ||
dc98df5a | 2154 | struct oom_wait_info { |
d79154bb | 2155 | struct mem_cgroup *memcg; |
dc98df5a KH |
2156 | wait_queue_t wait; |
2157 | }; | |
2158 | ||
2159 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
2160 | unsigned mode, int sync, void *arg) | |
2161 | { | |
d79154bb HD |
2162 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
2163 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
2164 | struct oom_wait_info *oom_wait_info; |
2165 | ||
2166 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 2167 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 2168 | |
dc98df5a | 2169 | /* |
d79154bb | 2170 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
dc98df5a KH |
2171 | * Then we can use css_is_ancestor without taking care of RCU. |
2172 | */ | |
c0ff4b85 R |
2173 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
2174 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 2175 | return 0; |
dc98df5a KH |
2176 | return autoremove_wake_function(wait, mode, sync, arg); |
2177 | } | |
2178 | ||
c0ff4b85 | 2179 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 2180 | { |
c0ff4b85 R |
2181 | /* for filtering, pass "memcg" as argument. */ |
2182 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
2183 | } |
2184 | ||
c0ff4b85 | 2185 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 2186 | { |
c0ff4b85 R |
2187 | if (memcg && atomic_read(&memcg->under_oom)) |
2188 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
2189 | } |
2190 | ||
867578cb KH |
2191 | /* |
2192 | * try to call OOM killer. returns false if we should exit memory-reclaim loop. | |
2193 | */ | |
6bbda35c KS |
2194 | static bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask, |
2195 | int order) | |
0b7f569e | 2196 | { |
dc98df5a | 2197 | struct oom_wait_info owait; |
3c11ecf4 | 2198 | bool locked, need_to_kill; |
867578cb | 2199 | |
d79154bb | 2200 | owait.memcg = memcg; |
dc98df5a KH |
2201 | owait.wait.flags = 0; |
2202 | owait.wait.func = memcg_oom_wake_function; | |
2203 | owait.wait.private = current; | |
2204 | INIT_LIST_HEAD(&owait.wait.task_list); | |
3c11ecf4 | 2205 | need_to_kill = true; |
c0ff4b85 | 2206 | mem_cgroup_mark_under_oom(memcg); |
79dfdacc | 2207 | |
c0ff4b85 | 2208 | /* At first, try to OOM lock hierarchy under memcg.*/ |
1af8efe9 | 2209 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 2210 | locked = mem_cgroup_oom_lock(memcg); |
867578cb KH |
2211 | /* |
2212 | * Even if signal_pending(), we can't quit charge() loop without | |
2213 | * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL | |
2214 | * under OOM is always welcomed, use TASK_KILLABLE here. | |
2215 | */ | |
3c11ecf4 | 2216 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
c0ff4b85 | 2217 | if (!locked || memcg->oom_kill_disable) |
3c11ecf4 KH |
2218 | need_to_kill = false; |
2219 | if (locked) | |
c0ff4b85 | 2220 | mem_cgroup_oom_notify(memcg); |
1af8efe9 | 2221 | spin_unlock(&memcg_oom_lock); |
867578cb | 2222 | |
3c11ecf4 KH |
2223 | if (need_to_kill) { |
2224 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
e845e199 | 2225 | mem_cgroup_out_of_memory(memcg, mask, order); |
3c11ecf4 | 2226 | } else { |
867578cb | 2227 | schedule(); |
dc98df5a | 2228 | finish_wait(&memcg_oom_waitq, &owait.wait); |
867578cb | 2229 | } |
1af8efe9 | 2230 | spin_lock(&memcg_oom_lock); |
79dfdacc | 2231 | if (locked) |
c0ff4b85 R |
2232 | mem_cgroup_oom_unlock(memcg); |
2233 | memcg_wakeup_oom(memcg); | |
1af8efe9 | 2234 | spin_unlock(&memcg_oom_lock); |
867578cb | 2235 | |
c0ff4b85 | 2236 | mem_cgroup_unmark_under_oom(memcg); |
79dfdacc | 2237 | |
867578cb KH |
2238 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) |
2239 | return false; | |
2240 | /* Give chance to dying process */ | |
715a5ee8 | 2241 | schedule_timeout_uninterruptible(1); |
867578cb | 2242 | return true; |
0b7f569e KH |
2243 | } |
2244 | ||
d69b042f BS |
2245 | /* |
2246 | * Currently used to update mapped file statistics, but the routine can be | |
2247 | * generalized to update other statistics as well. | |
32047e2a KH |
2248 | * |
2249 | * Notes: Race condition | |
2250 | * | |
2251 | * We usually use page_cgroup_lock() for accessing page_cgroup member but | |
2252 | * it tends to be costly. But considering some conditions, we doesn't need | |
2253 | * to do so _always_. | |
2254 | * | |
2255 | * Considering "charge", lock_page_cgroup() is not required because all | |
2256 | * file-stat operations happen after a page is attached to radix-tree. There | |
2257 | * are no race with "charge". | |
2258 | * | |
2259 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
2260 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
2261 | * if there are race with "uncharge". Statistics itself is properly handled | |
2262 | * by flags. | |
2263 | * | |
2264 | * Considering "move", this is an only case we see a race. To make the race | |
619d094b KH |
2265 | * small, we check mm->moving_account and detect there are possibility of race |
2266 | * If there is, we take a lock. | |
d69b042f | 2267 | */ |
26174efd | 2268 | |
89c06bd5 KH |
2269 | void __mem_cgroup_begin_update_page_stat(struct page *page, |
2270 | bool *locked, unsigned long *flags) | |
2271 | { | |
2272 | struct mem_cgroup *memcg; | |
2273 | struct page_cgroup *pc; | |
2274 | ||
2275 | pc = lookup_page_cgroup(page); | |
2276 | again: | |
2277 | memcg = pc->mem_cgroup; | |
2278 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
2279 | return; | |
2280 | /* | |
2281 | * If this memory cgroup is not under account moving, we don't | |
da92c47d | 2282 | * need to take move_lock_mem_cgroup(). Because we already hold |
89c06bd5 | 2283 | * rcu_read_lock(), any calls to move_account will be delayed until |
13fd1dd9 | 2284 | * rcu_read_unlock() if mem_cgroup_stolen() == true. |
89c06bd5 | 2285 | */ |
13fd1dd9 | 2286 | if (!mem_cgroup_stolen(memcg)) |
89c06bd5 KH |
2287 | return; |
2288 | ||
2289 | move_lock_mem_cgroup(memcg, flags); | |
2290 | if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) { | |
2291 | move_unlock_mem_cgroup(memcg, flags); | |
2292 | goto again; | |
2293 | } | |
2294 | *locked = true; | |
2295 | } | |
2296 | ||
2297 | void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags) | |
2298 | { | |
2299 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
2300 | ||
2301 | /* | |
2302 | * It's guaranteed that pc->mem_cgroup never changes while | |
2303 | * lock is held because a routine modifies pc->mem_cgroup | |
da92c47d | 2304 | * should take move_lock_mem_cgroup(). |
89c06bd5 KH |
2305 | */ |
2306 | move_unlock_mem_cgroup(pc->mem_cgroup, flags); | |
2307 | } | |
2308 | ||
2a7106f2 GT |
2309 | void mem_cgroup_update_page_stat(struct page *page, |
2310 | enum mem_cgroup_page_stat_item idx, int val) | |
d69b042f | 2311 | { |
c0ff4b85 | 2312 | struct mem_cgroup *memcg; |
32047e2a | 2313 | struct page_cgroup *pc = lookup_page_cgroup(page); |
dbd4ea78 | 2314 | unsigned long uninitialized_var(flags); |
d69b042f | 2315 | |
cfa44946 | 2316 | if (mem_cgroup_disabled()) |
d69b042f | 2317 | return; |
89c06bd5 | 2318 | |
c0ff4b85 R |
2319 | memcg = pc->mem_cgroup; |
2320 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
89c06bd5 | 2321 | return; |
26174efd | 2322 | |
26174efd | 2323 | switch (idx) { |
2a7106f2 | 2324 | case MEMCG_NR_FILE_MAPPED: |
2a7106f2 | 2325 | idx = MEM_CGROUP_STAT_FILE_MAPPED; |
26174efd KH |
2326 | break; |
2327 | default: | |
2328 | BUG(); | |
8725d541 | 2329 | } |
d69b042f | 2330 | |
c0ff4b85 | 2331 | this_cpu_add(memcg->stat->count[idx], val); |
d69b042f | 2332 | } |
26174efd | 2333 | |
cdec2e42 KH |
2334 | /* |
2335 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2336 | * TODO: maybe necessary to use big numbers in big irons. | |
2337 | */ | |
7ec99d62 | 2338 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2339 | struct memcg_stock_pcp { |
2340 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2341 | unsigned int nr_pages; |
cdec2e42 | 2342 | struct work_struct work; |
26fe6168 | 2343 | unsigned long flags; |
a0db00fc | 2344 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2345 | }; |
2346 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2347 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2348 | |
a0956d54 SS |
2349 | /** |
2350 | * consume_stock: Try to consume stocked charge on this cpu. | |
2351 | * @memcg: memcg to consume from. | |
2352 | * @nr_pages: how many pages to charge. | |
2353 | * | |
2354 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2355 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2356 | * service an allocation will refill the stock. | |
2357 | * | |
2358 | * returns true if successful, false otherwise. | |
cdec2e42 | 2359 | */ |
a0956d54 | 2360 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2361 | { |
2362 | struct memcg_stock_pcp *stock; | |
2363 | bool ret = true; | |
2364 | ||
a0956d54 SS |
2365 | if (nr_pages > CHARGE_BATCH) |
2366 | return false; | |
2367 | ||
cdec2e42 | 2368 | stock = &get_cpu_var(memcg_stock); |
a0956d54 SS |
2369 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) |
2370 | stock->nr_pages -= nr_pages; | |
cdec2e42 KH |
2371 | else /* need to call res_counter_charge */ |
2372 | ret = false; | |
2373 | put_cpu_var(memcg_stock); | |
2374 | return ret; | |
2375 | } | |
2376 | ||
2377 | /* | |
2378 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
2379 | */ | |
2380 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2381 | { | |
2382 | struct mem_cgroup *old = stock->cached; | |
2383 | ||
11c9ea4e JW |
2384 | if (stock->nr_pages) { |
2385 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
2386 | ||
2387 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 2388 | if (do_swap_account) |
11c9ea4e JW |
2389 | res_counter_uncharge(&old->memsw, bytes); |
2390 | stock->nr_pages = 0; | |
cdec2e42 KH |
2391 | } |
2392 | stock->cached = NULL; | |
cdec2e42 KH |
2393 | } |
2394 | ||
2395 | /* | |
2396 | * This must be called under preempt disabled or must be called by | |
2397 | * a thread which is pinned to local cpu. | |
2398 | */ | |
2399 | static void drain_local_stock(struct work_struct *dummy) | |
2400 | { | |
2401 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
2402 | drain_stock(stock); | |
26fe6168 | 2403 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2404 | } |
2405 | ||
e4777496 MH |
2406 | static void __init memcg_stock_init(void) |
2407 | { | |
2408 | int cpu; | |
2409 | ||
2410 | for_each_possible_cpu(cpu) { | |
2411 | struct memcg_stock_pcp *stock = | |
2412 | &per_cpu(memcg_stock, cpu); | |
2413 | INIT_WORK(&stock->work, drain_local_stock); | |
2414 | } | |
2415 | } | |
2416 | ||
cdec2e42 KH |
2417 | /* |
2418 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 2419 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2420 | */ |
c0ff4b85 | 2421 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2422 | { |
2423 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2424 | ||
c0ff4b85 | 2425 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2426 | drain_stock(stock); |
c0ff4b85 | 2427 | stock->cached = memcg; |
cdec2e42 | 2428 | } |
11c9ea4e | 2429 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2430 | put_cpu_var(memcg_stock); |
2431 | } | |
2432 | ||
2433 | /* | |
c0ff4b85 | 2434 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
d38144b7 MH |
2435 | * of the hierarchy under it. sync flag says whether we should block |
2436 | * until the work is done. | |
cdec2e42 | 2437 | */ |
c0ff4b85 | 2438 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
cdec2e42 | 2439 | { |
26fe6168 | 2440 | int cpu, curcpu; |
d38144b7 | 2441 | |
cdec2e42 | 2442 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2443 | get_online_cpus(); |
5af12d0e | 2444 | curcpu = get_cpu(); |
cdec2e42 KH |
2445 | for_each_online_cpu(cpu) { |
2446 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2447 | struct mem_cgroup *memcg; |
26fe6168 | 2448 | |
c0ff4b85 R |
2449 | memcg = stock->cached; |
2450 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2451 | continue; |
c0ff4b85 | 2452 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2453 | continue; |
d1a05b69 MH |
2454 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2455 | if (cpu == curcpu) | |
2456 | drain_local_stock(&stock->work); | |
2457 | else | |
2458 | schedule_work_on(cpu, &stock->work); | |
2459 | } | |
cdec2e42 | 2460 | } |
5af12d0e | 2461 | put_cpu(); |
d38144b7 MH |
2462 | |
2463 | if (!sync) | |
2464 | goto out; | |
2465 | ||
2466 | for_each_online_cpu(cpu) { | |
2467 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2468 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2469 | flush_work(&stock->work); |
2470 | } | |
2471 | out: | |
cdec2e42 | 2472 | put_online_cpus(); |
d38144b7 MH |
2473 | } |
2474 | ||
2475 | /* | |
2476 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2477 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
2478 | * expects some charges will be back to res_counter later but cannot wait for | |
2479 | * it. | |
2480 | */ | |
c0ff4b85 | 2481 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
d38144b7 | 2482 | { |
9f50fad6 MH |
2483 | /* |
2484 | * If someone calls draining, avoid adding more kworker runs. | |
2485 | */ | |
2486 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2487 | return; | |
c0ff4b85 | 2488 | drain_all_stock(root_memcg, false); |
9f50fad6 | 2489 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2490 | } |
2491 | ||
2492 | /* This is a synchronous drain interface. */ | |
c0ff4b85 | 2493 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
cdec2e42 KH |
2494 | { |
2495 | /* called when force_empty is called */ | |
9f50fad6 | 2496 | mutex_lock(&percpu_charge_mutex); |
c0ff4b85 | 2497 | drain_all_stock(root_memcg, true); |
9f50fad6 | 2498 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2499 | } |
2500 | ||
711d3d2c KH |
2501 | /* |
2502 | * This function drains percpu counter value from DEAD cpu and | |
2503 | * move it to local cpu. Note that this function can be preempted. | |
2504 | */ | |
c0ff4b85 | 2505 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2506 | { |
2507 | int i; | |
2508 | ||
c0ff4b85 | 2509 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2510 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2511 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2512 | |
c0ff4b85 R |
2513 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2514 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2515 | } |
e9f8974f | 2516 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2517 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2518 | |
c0ff4b85 R |
2519 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2520 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2521 | } |
c0ff4b85 | 2522 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2523 | } |
2524 | ||
2525 | static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb, | |
cdec2e42 KH |
2526 | unsigned long action, |
2527 | void *hcpu) | |
2528 | { | |
2529 | int cpu = (unsigned long)hcpu; | |
2530 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2531 | struct mem_cgroup *iter; |
cdec2e42 | 2532 | |
619d094b | 2533 | if (action == CPU_ONLINE) |
1489ebad | 2534 | return NOTIFY_OK; |
1489ebad | 2535 | |
d833049b | 2536 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2537 | return NOTIFY_OK; |
711d3d2c | 2538 | |
9f3a0d09 | 2539 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2540 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2541 | ||
cdec2e42 KH |
2542 | stock = &per_cpu(memcg_stock, cpu); |
2543 | drain_stock(stock); | |
2544 | return NOTIFY_OK; | |
2545 | } | |
2546 | ||
4b534334 KH |
2547 | |
2548 | /* See __mem_cgroup_try_charge() for details */ | |
2549 | enum { | |
2550 | CHARGE_OK, /* success */ | |
2551 | CHARGE_RETRY, /* need to retry but retry is not bad */ | |
2552 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ | |
2553 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ | |
2554 | CHARGE_OOM_DIE, /* the current is killed because of OOM */ | |
2555 | }; | |
2556 | ||
c0ff4b85 | 2557 | static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
4c9c5359 SS |
2558 | unsigned int nr_pages, unsigned int min_pages, |
2559 | bool oom_check) | |
4b534334 | 2560 | { |
7ec99d62 | 2561 | unsigned long csize = nr_pages * PAGE_SIZE; |
4b534334 KH |
2562 | struct mem_cgroup *mem_over_limit; |
2563 | struct res_counter *fail_res; | |
2564 | unsigned long flags = 0; | |
2565 | int ret; | |
2566 | ||
c0ff4b85 | 2567 | ret = res_counter_charge(&memcg->res, csize, &fail_res); |
4b534334 KH |
2568 | |
2569 | if (likely(!ret)) { | |
2570 | if (!do_swap_account) | |
2571 | return CHARGE_OK; | |
c0ff4b85 | 2572 | ret = res_counter_charge(&memcg->memsw, csize, &fail_res); |
4b534334 KH |
2573 | if (likely(!ret)) |
2574 | return CHARGE_OK; | |
2575 | ||
c0ff4b85 | 2576 | res_counter_uncharge(&memcg->res, csize); |
4b534334 KH |
2577 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
2578 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
2579 | } else | |
2580 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
9221edb7 | 2581 | /* |
9221edb7 JW |
2582 | * Never reclaim on behalf of optional batching, retry with a |
2583 | * single page instead. | |
2584 | */ | |
4c9c5359 | 2585 | if (nr_pages > min_pages) |
4b534334 KH |
2586 | return CHARGE_RETRY; |
2587 | ||
2588 | if (!(gfp_mask & __GFP_WAIT)) | |
2589 | return CHARGE_WOULDBLOCK; | |
2590 | ||
4c9c5359 SS |
2591 | if (gfp_mask & __GFP_NORETRY) |
2592 | return CHARGE_NOMEM; | |
2593 | ||
5660048c | 2594 | ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags); |
7ec99d62 | 2595 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
19942822 | 2596 | return CHARGE_RETRY; |
4b534334 | 2597 | /* |
19942822 JW |
2598 | * Even though the limit is exceeded at this point, reclaim |
2599 | * may have been able to free some pages. Retry the charge | |
2600 | * before killing the task. | |
2601 | * | |
2602 | * Only for regular pages, though: huge pages are rather | |
2603 | * unlikely to succeed so close to the limit, and we fall back | |
2604 | * to regular pages anyway in case of failure. | |
4b534334 | 2605 | */ |
4c9c5359 | 2606 | if (nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER) && ret) |
4b534334 KH |
2607 | return CHARGE_RETRY; |
2608 | ||
2609 | /* | |
2610 | * At task move, charge accounts can be doubly counted. So, it's | |
2611 | * better to wait until the end of task_move if something is going on. | |
2612 | */ | |
2613 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2614 | return CHARGE_RETRY; | |
2615 | ||
2616 | /* If we don't need to call oom-killer at el, return immediately */ | |
2617 | if (!oom_check) | |
2618 | return CHARGE_NOMEM; | |
2619 | /* check OOM */ | |
e845e199 | 2620 | if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask, get_order(csize))) |
4b534334 KH |
2621 | return CHARGE_OOM_DIE; |
2622 | ||
2623 | return CHARGE_RETRY; | |
2624 | } | |
2625 | ||
f817ed48 | 2626 | /* |
38c5d72f KH |
2627 | * __mem_cgroup_try_charge() does |
2628 | * 1. detect memcg to be charged against from passed *mm and *ptr, | |
2629 | * 2. update res_counter | |
2630 | * 3. call memory reclaim if necessary. | |
2631 | * | |
2632 | * In some special case, if the task is fatal, fatal_signal_pending() or | |
2633 | * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup | |
2634 | * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon | |
2635 | * as possible without any hazards. 2: all pages should have a valid | |
2636 | * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg | |
2637 | * pointer, that is treated as a charge to root_mem_cgroup. | |
2638 | * | |
2639 | * So __mem_cgroup_try_charge() will return | |
2640 | * 0 ... on success, filling *ptr with a valid memcg pointer. | |
2641 | * -ENOMEM ... charge failure because of resource limits. | |
2642 | * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup. | |
2643 | * | |
2644 | * Unlike the exported interface, an "oom" parameter is added. if oom==true, | |
2645 | * the oom-killer can be invoked. | |
8a9f3ccd | 2646 | */ |
f817ed48 | 2647 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
ec168510 | 2648 | gfp_t gfp_mask, |
7ec99d62 | 2649 | unsigned int nr_pages, |
c0ff4b85 | 2650 | struct mem_cgroup **ptr, |
7ec99d62 | 2651 | bool oom) |
8a9f3ccd | 2652 | { |
7ec99d62 | 2653 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
4b534334 | 2654 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
c0ff4b85 | 2655 | struct mem_cgroup *memcg = NULL; |
4b534334 | 2656 | int ret; |
a636b327 | 2657 | |
867578cb KH |
2658 | /* |
2659 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
2660 | * in system level. So, allow to go ahead dying process in addition to | |
2661 | * MEMDIE process. | |
2662 | */ | |
2663 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
2664 | || fatal_signal_pending(current))) | |
2665 | goto bypass; | |
a636b327 | 2666 | |
8a9f3ccd | 2667 | /* |
3be91277 HD |
2668 | * We always charge the cgroup the mm_struct belongs to. |
2669 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd | 2670 | * thread group leader migrates. It's possible that mm is not |
24467cac | 2671 | * set, if so charge the root memcg (happens for pagecache usage). |
8a9f3ccd | 2672 | */ |
c0ff4b85 | 2673 | if (!*ptr && !mm) |
38c5d72f | 2674 | *ptr = root_mem_cgroup; |
f75ca962 | 2675 | again: |
c0ff4b85 R |
2676 | if (*ptr) { /* css should be a valid one */ |
2677 | memcg = *ptr; | |
c0ff4b85 | 2678 | if (mem_cgroup_is_root(memcg)) |
f75ca962 | 2679 | goto done; |
a0956d54 | 2680 | if (consume_stock(memcg, nr_pages)) |
f75ca962 | 2681 | goto done; |
c0ff4b85 | 2682 | css_get(&memcg->css); |
4b534334 | 2683 | } else { |
f75ca962 | 2684 | struct task_struct *p; |
54595fe2 | 2685 | |
f75ca962 KH |
2686 | rcu_read_lock(); |
2687 | p = rcu_dereference(mm->owner); | |
f75ca962 | 2688 | /* |
ebb76ce1 | 2689 | * Because we don't have task_lock(), "p" can exit. |
c0ff4b85 | 2690 | * In that case, "memcg" can point to root or p can be NULL with |
ebb76ce1 KH |
2691 | * race with swapoff. Then, we have small risk of mis-accouning. |
2692 | * But such kind of mis-account by race always happens because | |
2693 | * we don't have cgroup_mutex(). It's overkill and we allo that | |
2694 | * small race, here. | |
2695 | * (*) swapoff at el will charge against mm-struct not against | |
2696 | * task-struct. So, mm->owner can be NULL. | |
f75ca962 | 2697 | */ |
c0ff4b85 | 2698 | memcg = mem_cgroup_from_task(p); |
38c5d72f KH |
2699 | if (!memcg) |
2700 | memcg = root_mem_cgroup; | |
2701 | if (mem_cgroup_is_root(memcg)) { | |
f75ca962 KH |
2702 | rcu_read_unlock(); |
2703 | goto done; | |
2704 | } | |
a0956d54 | 2705 | if (consume_stock(memcg, nr_pages)) { |
f75ca962 KH |
2706 | /* |
2707 | * It seems dagerous to access memcg without css_get(). | |
2708 | * But considering how consume_stok works, it's not | |
2709 | * necessary. If consume_stock success, some charges | |
2710 | * from this memcg are cached on this cpu. So, we | |
2711 | * don't need to call css_get()/css_tryget() before | |
2712 | * calling consume_stock(). | |
2713 | */ | |
2714 | rcu_read_unlock(); | |
2715 | goto done; | |
2716 | } | |
2717 | /* after here, we may be blocked. we need to get refcnt */ | |
c0ff4b85 | 2718 | if (!css_tryget(&memcg->css)) { |
f75ca962 KH |
2719 | rcu_read_unlock(); |
2720 | goto again; | |
2721 | } | |
2722 | rcu_read_unlock(); | |
2723 | } | |
8a9f3ccd | 2724 | |
4b534334 KH |
2725 | do { |
2726 | bool oom_check; | |
7a81b88c | 2727 | |
4b534334 | 2728 | /* If killed, bypass charge */ |
f75ca962 | 2729 | if (fatal_signal_pending(current)) { |
c0ff4b85 | 2730 | css_put(&memcg->css); |
4b534334 | 2731 | goto bypass; |
f75ca962 | 2732 | } |
6d61ef40 | 2733 | |
4b534334 KH |
2734 | oom_check = false; |
2735 | if (oom && !nr_oom_retries) { | |
2736 | oom_check = true; | |
2737 | nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
cdec2e42 | 2738 | } |
66e1707b | 2739 | |
4c9c5359 SS |
2740 | ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, nr_pages, |
2741 | oom_check); | |
4b534334 KH |
2742 | switch (ret) { |
2743 | case CHARGE_OK: | |
2744 | break; | |
2745 | case CHARGE_RETRY: /* not in OOM situation but retry */ | |
7ec99d62 | 2746 | batch = nr_pages; |
c0ff4b85 R |
2747 | css_put(&memcg->css); |
2748 | memcg = NULL; | |
f75ca962 | 2749 | goto again; |
4b534334 | 2750 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
c0ff4b85 | 2751 | css_put(&memcg->css); |
4b534334 KH |
2752 | goto nomem; |
2753 | case CHARGE_NOMEM: /* OOM routine works */ | |
f75ca962 | 2754 | if (!oom) { |
c0ff4b85 | 2755 | css_put(&memcg->css); |
867578cb | 2756 | goto nomem; |
f75ca962 | 2757 | } |
4b534334 KH |
2758 | /* If oom, we never return -ENOMEM */ |
2759 | nr_oom_retries--; | |
2760 | break; | |
2761 | case CHARGE_OOM_DIE: /* Killed by OOM Killer */ | |
c0ff4b85 | 2762 | css_put(&memcg->css); |
867578cb | 2763 | goto bypass; |
66e1707b | 2764 | } |
4b534334 KH |
2765 | } while (ret != CHARGE_OK); |
2766 | ||
7ec99d62 | 2767 | if (batch > nr_pages) |
c0ff4b85 R |
2768 | refill_stock(memcg, batch - nr_pages); |
2769 | css_put(&memcg->css); | |
0c3e73e8 | 2770 | done: |
c0ff4b85 | 2771 | *ptr = memcg; |
7a81b88c KH |
2772 | return 0; |
2773 | nomem: | |
c0ff4b85 | 2774 | *ptr = NULL; |
7a81b88c | 2775 | return -ENOMEM; |
867578cb | 2776 | bypass: |
38c5d72f KH |
2777 | *ptr = root_mem_cgroup; |
2778 | return -EINTR; | |
7a81b88c | 2779 | } |
8a9f3ccd | 2780 | |
a3032a2c DN |
2781 | /* |
2782 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2783 | * This function is for that and do uncharge, put css's refcnt. | |
2784 | * gotten by try_charge(). | |
2785 | */ | |
c0ff4b85 | 2786 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg, |
e7018b8d | 2787 | unsigned int nr_pages) |
a3032a2c | 2788 | { |
c0ff4b85 | 2789 | if (!mem_cgroup_is_root(memcg)) { |
e7018b8d JW |
2790 | unsigned long bytes = nr_pages * PAGE_SIZE; |
2791 | ||
c0ff4b85 | 2792 | res_counter_uncharge(&memcg->res, bytes); |
a3032a2c | 2793 | if (do_swap_account) |
c0ff4b85 | 2794 | res_counter_uncharge(&memcg->memsw, bytes); |
a3032a2c | 2795 | } |
854ffa8d DN |
2796 | } |
2797 | ||
d01dd17f KH |
2798 | /* |
2799 | * Cancel chrages in this cgroup....doesn't propagate to parent cgroup. | |
2800 | * This is useful when moving usage to parent cgroup. | |
2801 | */ | |
2802 | static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg, | |
2803 | unsigned int nr_pages) | |
2804 | { | |
2805 | unsigned long bytes = nr_pages * PAGE_SIZE; | |
2806 | ||
2807 | if (mem_cgroup_is_root(memcg)) | |
2808 | return; | |
2809 | ||
2810 | res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes); | |
2811 | if (do_swap_account) | |
2812 | res_counter_uncharge_until(&memcg->memsw, | |
2813 | memcg->memsw.parent, bytes); | |
2814 | } | |
2815 | ||
a3b2d692 KH |
2816 | /* |
2817 | * A helper function to get mem_cgroup from ID. must be called under | |
e9316080 TH |
2818 | * rcu_read_lock(). The caller is responsible for calling css_tryget if |
2819 | * the mem_cgroup is used for charging. (dropping refcnt from swap can be | |
2820 | * called against removed memcg.) | |
a3b2d692 KH |
2821 | */ |
2822 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2823 | { | |
2824 | struct cgroup_subsys_state *css; | |
2825 | ||
2826 | /* ID 0 is unused ID */ | |
2827 | if (!id) | |
2828 | return NULL; | |
2829 | css = css_lookup(&mem_cgroup_subsys, id); | |
2830 | if (!css) | |
2831 | return NULL; | |
b2145145 | 2832 | return mem_cgroup_from_css(css); |
a3b2d692 KH |
2833 | } |
2834 | ||
e42d9d5d | 2835 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2836 | { |
c0ff4b85 | 2837 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2838 | struct page_cgroup *pc; |
a3b2d692 | 2839 | unsigned short id; |
b5a84319 KH |
2840 | swp_entry_t ent; |
2841 | ||
3c776e64 DN |
2842 | VM_BUG_ON(!PageLocked(page)); |
2843 | ||
3c776e64 | 2844 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2845 | lock_page_cgroup(pc); |
a3b2d692 | 2846 | if (PageCgroupUsed(pc)) { |
c0ff4b85 R |
2847 | memcg = pc->mem_cgroup; |
2848 | if (memcg && !css_tryget(&memcg->css)) | |
2849 | memcg = NULL; | |
e42d9d5d | 2850 | } else if (PageSwapCache(page)) { |
3c776e64 | 2851 | ent.val = page_private(page); |
9fb4b7cc | 2852 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2853 | rcu_read_lock(); |
c0ff4b85 R |
2854 | memcg = mem_cgroup_lookup(id); |
2855 | if (memcg && !css_tryget(&memcg->css)) | |
2856 | memcg = NULL; | |
a3b2d692 | 2857 | rcu_read_unlock(); |
3c776e64 | 2858 | } |
c0bd3f63 | 2859 | unlock_page_cgroup(pc); |
c0ff4b85 | 2860 | return memcg; |
b5a84319 KH |
2861 | } |
2862 | ||
c0ff4b85 | 2863 | static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg, |
5564e88b | 2864 | struct page *page, |
7ec99d62 | 2865 | unsigned int nr_pages, |
9ce70c02 HD |
2866 | enum charge_type ctype, |
2867 | bool lrucare) | |
7a81b88c | 2868 | { |
ce587e65 | 2869 | struct page_cgroup *pc = lookup_page_cgroup(page); |
9ce70c02 | 2870 | struct zone *uninitialized_var(zone); |
fa9add64 | 2871 | struct lruvec *lruvec; |
9ce70c02 | 2872 | bool was_on_lru = false; |
b2402857 | 2873 | bool anon; |
9ce70c02 | 2874 | |
ca3e0214 | 2875 | lock_page_cgroup(pc); |
90deb788 | 2876 | VM_BUG_ON(PageCgroupUsed(pc)); |
ca3e0214 KH |
2877 | /* |
2878 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2879 | * accessed by any other context at this point. | |
2880 | */ | |
9ce70c02 HD |
2881 | |
2882 | /* | |
2883 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2884 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2885 | */ | |
2886 | if (lrucare) { | |
2887 | zone = page_zone(page); | |
2888 | spin_lock_irq(&zone->lru_lock); | |
2889 | if (PageLRU(page)) { | |
fa9add64 | 2890 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 | 2891 | ClearPageLRU(page); |
fa9add64 | 2892 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2893 | was_on_lru = true; |
2894 | } | |
2895 | } | |
2896 | ||
c0ff4b85 | 2897 | pc->mem_cgroup = memcg; |
261fb61a KH |
2898 | /* |
2899 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
2900 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
2901 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
2902 | * before USED bit, we need memory barrier here. | |
2903 | * See mem_cgroup_add_lru_list(), etc. | |
2904 | */ | |
08e552c6 | 2905 | smp_wmb(); |
b2402857 | 2906 | SetPageCgroupUsed(pc); |
3be91277 | 2907 | |
9ce70c02 HD |
2908 | if (lrucare) { |
2909 | if (was_on_lru) { | |
fa9add64 | 2910 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 HD |
2911 | VM_BUG_ON(PageLRU(page)); |
2912 | SetPageLRU(page); | |
fa9add64 | 2913 | add_page_to_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2914 | } |
2915 | spin_unlock_irq(&zone->lru_lock); | |
2916 | } | |
2917 | ||
41326c17 | 2918 | if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON) |
b2402857 KH |
2919 | anon = true; |
2920 | else | |
2921 | anon = false; | |
2922 | ||
b070e65c | 2923 | mem_cgroup_charge_statistics(memcg, page, anon, nr_pages); |
52d4b9ac | 2924 | unlock_page_cgroup(pc); |
9ce70c02 | 2925 | |
430e4863 KH |
2926 | /* |
2927 | * "charge_statistics" updated event counter. Then, check it. | |
2928 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
2929 | * if they exceeds softlimit. | |
2930 | */ | |
c0ff4b85 | 2931 | memcg_check_events(memcg, page); |
7a81b88c | 2932 | } |
66e1707b | 2933 | |
7cf27982 GC |
2934 | static DEFINE_MUTEX(set_limit_mutex); |
2935 | ||
7ae1e1d0 GC |
2936 | #ifdef CONFIG_MEMCG_KMEM |
2937 | static inline bool memcg_can_account_kmem(struct mem_cgroup *memcg) | |
2938 | { | |
2939 | return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg) && | |
2940 | (memcg->kmem_account_flags & KMEM_ACCOUNTED_MASK); | |
2941 | } | |
2942 | ||
1f458cbf GC |
2943 | /* |
2944 | * This is a bit cumbersome, but it is rarely used and avoids a backpointer | |
2945 | * in the memcg_cache_params struct. | |
2946 | */ | |
2947 | static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p) | |
2948 | { | |
2949 | struct kmem_cache *cachep; | |
2950 | ||
2951 | VM_BUG_ON(p->is_root_cache); | |
2952 | cachep = p->root_cache; | |
2953 | return cachep->memcg_params->memcg_caches[memcg_cache_id(p->memcg)]; | |
2954 | } | |
2955 | ||
749c5415 GC |
2956 | #ifdef CONFIG_SLABINFO |
2957 | static int mem_cgroup_slabinfo_read(struct cgroup *cont, struct cftype *cft, | |
2958 | struct seq_file *m) | |
2959 | { | |
2960 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); | |
2961 | struct memcg_cache_params *params; | |
2962 | ||
2963 | if (!memcg_can_account_kmem(memcg)) | |
2964 | return -EIO; | |
2965 | ||
2966 | print_slabinfo_header(m); | |
2967 | ||
2968 | mutex_lock(&memcg->slab_caches_mutex); | |
2969 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) | |
2970 | cache_show(memcg_params_to_cache(params), m); | |
2971 | mutex_unlock(&memcg->slab_caches_mutex); | |
2972 | ||
2973 | return 0; | |
2974 | } | |
2975 | #endif | |
2976 | ||
7ae1e1d0 GC |
2977 | static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, u64 size) |
2978 | { | |
2979 | struct res_counter *fail_res; | |
2980 | struct mem_cgroup *_memcg; | |
2981 | int ret = 0; | |
2982 | bool may_oom; | |
2983 | ||
2984 | ret = res_counter_charge(&memcg->kmem, size, &fail_res); | |
2985 | if (ret) | |
2986 | return ret; | |
2987 | ||
2988 | /* | |
2989 | * Conditions under which we can wait for the oom_killer. Those are | |
2990 | * the same conditions tested by the core page allocator | |
2991 | */ | |
2992 | may_oom = (gfp & __GFP_FS) && !(gfp & __GFP_NORETRY); | |
2993 | ||
2994 | _memcg = memcg; | |
2995 | ret = __mem_cgroup_try_charge(NULL, gfp, size >> PAGE_SHIFT, | |
2996 | &_memcg, may_oom); | |
2997 | ||
2998 | if (ret == -EINTR) { | |
2999 | /* | |
3000 | * __mem_cgroup_try_charge() chosed to bypass to root due to | |
3001 | * OOM kill or fatal signal. Since our only options are to | |
3002 | * either fail the allocation or charge it to this cgroup, do | |
3003 | * it as a temporary condition. But we can't fail. From a | |
3004 | * kmem/slab perspective, the cache has already been selected, | |
3005 | * by mem_cgroup_kmem_get_cache(), so it is too late to change | |
3006 | * our minds. | |
3007 | * | |
3008 | * This condition will only trigger if the task entered | |
3009 | * memcg_charge_kmem in a sane state, but was OOM-killed during | |
3010 | * __mem_cgroup_try_charge() above. Tasks that were already | |
3011 | * dying when the allocation triggers should have been already | |
3012 | * directed to the root cgroup in memcontrol.h | |
3013 | */ | |
3014 | res_counter_charge_nofail(&memcg->res, size, &fail_res); | |
3015 | if (do_swap_account) | |
3016 | res_counter_charge_nofail(&memcg->memsw, size, | |
3017 | &fail_res); | |
3018 | ret = 0; | |
3019 | } else if (ret) | |
3020 | res_counter_uncharge(&memcg->kmem, size); | |
3021 | ||
3022 | return ret; | |
3023 | } | |
3024 | ||
3025 | static void memcg_uncharge_kmem(struct mem_cgroup *memcg, u64 size) | |
3026 | { | |
7ae1e1d0 GC |
3027 | res_counter_uncharge(&memcg->res, size); |
3028 | if (do_swap_account) | |
3029 | res_counter_uncharge(&memcg->memsw, size); | |
7de37682 GC |
3030 | |
3031 | /* Not down to 0 */ | |
3032 | if (res_counter_uncharge(&memcg->kmem, size)) | |
3033 | return; | |
3034 | ||
10d5ebf4 LZ |
3035 | /* |
3036 | * Releases a reference taken in kmem_cgroup_css_offline in case | |
3037 | * this last uncharge is racing with the offlining code or it is | |
3038 | * outliving the memcg existence. | |
3039 | * | |
3040 | * The memory barrier imposed by test&clear is paired with the | |
3041 | * explicit one in memcg_kmem_mark_dead(). | |
3042 | */ | |
7de37682 | 3043 | if (memcg_kmem_test_and_clear_dead(memcg)) |
10d5ebf4 | 3044 | css_put(&memcg->css); |
7ae1e1d0 GC |
3045 | } |
3046 | ||
2633d7a0 GC |
3047 | void memcg_cache_list_add(struct mem_cgroup *memcg, struct kmem_cache *cachep) |
3048 | { | |
3049 | if (!memcg) | |
3050 | return; | |
3051 | ||
3052 | mutex_lock(&memcg->slab_caches_mutex); | |
3053 | list_add(&cachep->memcg_params->list, &memcg->memcg_slab_caches); | |
3054 | mutex_unlock(&memcg->slab_caches_mutex); | |
3055 | } | |
3056 | ||
3057 | /* | |
3058 | * helper for acessing a memcg's index. It will be used as an index in the | |
3059 | * child cache array in kmem_cache, and also to derive its name. This function | |
3060 | * will return -1 when this is not a kmem-limited memcg. | |
3061 | */ | |
3062 | int memcg_cache_id(struct mem_cgroup *memcg) | |
3063 | { | |
3064 | return memcg ? memcg->kmemcg_id : -1; | |
3065 | } | |
3066 | ||
55007d84 GC |
3067 | /* |
3068 | * This ends up being protected by the set_limit mutex, during normal | |
3069 | * operation, because that is its main call site. | |
3070 | * | |
3071 | * But when we create a new cache, we can call this as well if its parent | |
3072 | * is kmem-limited. That will have to hold set_limit_mutex as well. | |
3073 | */ | |
3074 | int memcg_update_cache_sizes(struct mem_cgroup *memcg) | |
3075 | { | |
3076 | int num, ret; | |
3077 | ||
3078 | num = ida_simple_get(&kmem_limited_groups, | |
3079 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); | |
3080 | if (num < 0) | |
3081 | return num; | |
3082 | /* | |
3083 | * After this point, kmem_accounted (that we test atomically in | |
3084 | * the beginning of this conditional), is no longer 0. This | |
3085 | * guarantees only one process will set the following boolean | |
3086 | * to true. We don't need test_and_set because we're protected | |
3087 | * by the set_limit_mutex anyway. | |
3088 | */ | |
3089 | memcg_kmem_set_activated(memcg); | |
3090 | ||
3091 | ret = memcg_update_all_caches(num+1); | |
3092 | if (ret) { | |
3093 | ida_simple_remove(&kmem_limited_groups, num); | |
3094 | memcg_kmem_clear_activated(memcg); | |
3095 | return ret; | |
3096 | } | |
3097 | ||
3098 | memcg->kmemcg_id = num; | |
3099 | INIT_LIST_HEAD(&memcg->memcg_slab_caches); | |
3100 | mutex_init(&memcg->slab_caches_mutex); | |
3101 | return 0; | |
3102 | } | |
3103 | ||
3104 | static size_t memcg_caches_array_size(int num_groups) | |
3105 | { | |
3106 | ssize_t size; | |
3107 | if (num_groups <= 0) | |
3108 | return 0; | |
3109 | ||
3110 | size = 2 * num_groups; | |
3111 | if (size < MEMCG_CACHES_MIN_SIZE) | |
3112 | size = MEMCG_CACHES_MIN_SIZE; | |
3113 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
3114 | size = MEMCG_CACHES_MAX_SIZE; | |
3115 | ||
3116 | return size; | |
3117 | } | |
3118 | ||
3119 | /* | |
3120 | * We should update the current array size iff all caches updates succeed. This | |
3121 | * can only be done from the slab side. The slab mutex needs to be held when | |
3122 | * calling this. | |
3123 | */ | |
3124 | void memcg_update_array_size(int num) | |
3125 | { | |
3126 | if (num > memcg_limited_groups_array_size) | |
3127 | memcg_limited_groups_array_size = memcg_caches_array_size(num); | |
3128 | } | |
3129 | ||
15cf17d2 KK |
3130 | static void kmem_cache_destroy_work_func(struct work_struct *w); |
3131 | ||
55007d84 GC |
3132 | int memcg_update_cache_size(struct kmem_cache *s, int num_groups) |
3133 | { | |
3134 | struct memcg_cache_params *cur_params = s->memcg_params; | |
3135 | ||
3136 | VM_BUG_ON(s->memcg_params && !s->memcg_params->is_root_cache); | |
3137 | ||
3138 | if (num_groups > memcg_limited_groups_array_size) { | |
3139 | int i; | |
3140 | ssize_t size = memcg_caches_array_size(num_groups); | |
3141 | ||
3142 | size *= sizeof(void *); | |
3143 | size += sizeof(struct memcg_cache_params); | |
3144 | ||
3145 | s->memcg_params = kzalloc(size, GFP_KERNEL); | |
3146 | if (!s->memcg_params) { | |
3147 | s->memcg_params = cur_params; | |
3148 | return -ENOMEM; | |
3149 | } | |
3150 | ||
3151 | s->memcg_params->is_root_cache = true; | |
3152 | ||
3153 | /* | |
3154 | * There is the chance it will be bigger than | |
3155 | * memcg_limited_groups_array_size, if we failed an allocation | |
3156 | * in a cache, in which case all caches updated before it, will | |
3157 | * have a bigger array. | |
3158 | * | |
3159 | * But if that is the case, the data after | |
3160 | * memcg_limited_groups_array_size is certainly unused | |
3161 | */ | |
3162 | for (i = 0; i < memcg_limited_groups_array_size; i++) { | |
3163 | if (!cur_params->memcg_caches[i]) | |
3164 | continue; | |
3165 | s->memcg_params->memcg_caches[i] = | |
3166 | cur_params->memcg_caches[i]; | |
3167 | } | |
3168 | ||
3169 | /* | |
3170 | * Ideally, we would wait until all caches succeed, and only | |
3171 | * then free the old one. But this is not worth the extra | |
3172 | * pointer per-cache we'd have to have for this. | |
3173 | * | |
3174 | * It is not a big deal if some caches are left with a size | |
3175 | * bigger than the others. And all updates will reset this | |
3176 | * anyway. | |
3177 | */ | |
3178 | kfree(cur_params); | |
3179 | } | |
3180 | return 0; | |
3181 | } | |
3182 | ||
943a451a GC |
3183 | int memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s, |
3184 | struct kmem_cache *root_cache) | |
2633d7a0 GC |
3185 | { |
3186 | size_t size = sizeof(struct memcg_cache_params); | |
3187 | ||
3188 | if (!memcg_kmem_enabled()) | |
3189 | return 0; | |
3190 | ||
55007d84 GC |
3191 | if (!memcg) |
3192 | size += memcg_limited_groups_array_size * sizeof(void *); | |
3193 | ||
2633d7a0 GC |
3194 | s->memcg_params = kzalloc(size, GFP_KERNEL); |
3195 | if (!s->memcg_params) | |
3196 | return -ENOMEM; | |
3197 | ||
15cf17d2 KK |
3198 | INIT_WORK(&s->memcg_params->destroy, |
3199 | kmem_cache_destroy_work_func); | |
943a451a | 3200 | if (memcg) { |
2633d7a0 | 3201 | s->memcg_params->memcg = memcg; |
943a451a | 3202 | s->memcg_params->root_cache = root_cache; |
4ba902b5 GC |
3203 | } else |
3204 | s->memcg_params->is_root_cache = true; | |
3205 | ||
2633d7a0 GC |
3206 | return 0; |
3207 | } | |
3208 | ||
3209 | void memcg_release_cache(struct kmem_cache *s) | |
3210 | { | |
d7f25f8a GC |
3211 | struct kmem_cache *root; |
3212 | struct mem_cgroup *memcg; | |
3213 | int id; | |
3214 | ||
3215 | /* | |
3216 | * This happens, for instance, when a root cache goes away before we | |
3217 | * add any memcg. | |
3218 | */ | |
3219 | if (!s->memcg_params) | |
3220 | return; | |
3221 | ||
3222 | if (s->memcg_params->is_root_cache) | |
3223 | goto out; | |
3224 | ||
3225 | memcg = s->memcg_params->memcg; | |
3226 | id = memcg_cache_id(memcg); | |
3227 | ||
3228 | root = s->memcg_params->root_cache; | |
3229 | root->memcg_params->memcg_caches[id] = NULL; | |
d7f25f8a GC |
3230 | |
3231 | mutex_lock(&memcg->slab_caches_mutex); | |
3232 | list_del(&s->memcg_params->list); | |
3233 | mutex_unlock(&memcg->slab_caches_mutex); | |
3234 | ||
20f05310 | 3235 | css_put(&memcg->css); |
d7f25f8a | 3236 | out: |
2633d7a0 GC |
3237 | kfree(s->memcg_params); |
3238 | } | |
3239 | ||
0e9d92f2 GC |
3240 | /* |
3241 | * During the creation a new cache, we need to disable our accounting mechanism | |
3242 | * altogether. This is true even if we are not creating, but rather just | |
3243 | * enqueing new caches to be created. | |
3244 | * | |
3245 | * This is because that process will trigger allocations; some visible, like | |
3246 | * explicit kmallocs to auxiliary data structures, name strings and internal | |
3247 | * cache structures; some well concealed, like INIT_WORK() that can allocate | |
3248 | * objects during debug. | |
3249 | * | |
3250 | * If any allocation happens during memcg_kmem_get_cache, we will recurse back | |
3251 | * to it. This may not be a bounded recursion: since the first cache creation | |
3252 | * failed to complete (waiting on the allocation), we'll just try to create the | |
3253 | * cache again, failing at the same point. | |
3254 | * | |
3255 | * memcg_kmem_get_cache is prepared to abort after seeing a positive count of | |
3256 | * memcg_kmem_skip_account. So we enclose anything that might allocate memory | |
3257 | * inside the following two functions. | |
3258 | */ | |
3259 | static inline void memcg_stop_kmem_account(void) | |
3260 | { | |
3261 | VM_BUG_ON(!current->mm); | |
3262 | current->memcg_kmem_skip_account++; | |
3263 | } | |
3264 | ||
3265 | static inline void memcg_resume_kmem_account(void) | |
3266 | { | |
3267 | VM_BUG_ON(!current->mm); | |
3268 | current->memcg_kmem_skip_account--; | |
3269 | } | |
3270 | ||
1f458cbf GC |
3271 | static void kmem_cache_destroy_work_func(struct work_struct *w) |
3272 | { | |
3273 | struct kmem_cache *cachep; | |
3274 | struct memcg_cache_params *p; | |
3275 | ||
3276 | p = container_of(w, struct memcg_cache_params, destroy); | |
3277 | ||
3278 | cachep = memcg_params_to_cache(p); | |
3279 | ||
22933152 GC |
3280 | /* |
3281 | * If we get down to 0 after shrink, we could delete right away. | |
3282 | * However, memcg_release_pages() already puts us back in the workqueue | |
3283 | * in that case. If we proceed deleting, we'll get a dangling | |
3284 | * reference, and removing the object from the workqueue in that case | |
3285 | * is unnecessary complication. We are not a fast path. | |
3286 | * | |
3287 | * Note that this case is fundamentally different from racing with | |
3288 | * shrink_slab(): if memcg_cgroup_destroy_cache() is called in | |
3289 | * kmem_cache_shrink, not only we would be reinserting a dead cache | |
3290 | * into the queue, but doing so from inside the worker racing to | |
3291 | * destroy it. | |
3292 | * | |
3293 | * So if we aren't down to zero, we'll just schedule a worker and try | |
3294 | * again | |
3295 | */ | |
3296 | if (atomic_read(&cachep->memcg_params->nr_pages) != 0) { | |
3297 | kmem_cache_shrink(cachep); | |
3298 | if (atomic_read(&cachep->memcg_params->nr_pages) == 0) | |
3299 | return; | |
3300 | } else | |
1f458cbf GC |
3301 | kmem_cache_destroy(cachep); |
3302 | } | |
3303 | ||
3304 | void mem_cgroup_destroy_cache(struct kmem_cache *cachep) | |
3305 | { | |
3306 | if (!cachep->memcg_params->dead) | |
3307 | return; | |
3308 | ||
22933152 GC |
3309 | /* |
3310 | * There are many ways in which we can get here. | |
3311 | * | |
3312 | * We can get to a memory-pressure situation while the delayed work is | |
3313 | * still pending to run. The vmscan shrinkers can then release all | |
3314 | * cache memory and get us to destruction. If this is the case, we'll | |
3315 | * be executed twice, which is a bug (the second time will execute over | |
3316 | * bogus data). In this case, cancelling the work should be fine. | |
3317 | * | |
3318 | * But we can also get here from the worker itself, if | |
3319 | * kmem_cache_shrink is enough to shake all the remaining objects and | |
3320 | * get the page count to 0. In this case, we'll deadlock if we try to | |
3321 | * cancel the work (the worker runs with an internal lock held, which | |
3322 | * is the same lock we would hold for cancel_work_sync().) | |
3323 | * | |
3324 | * Since we can't possibly know who got us here, just refrain from | |
3325 | * running if there is already work pending | |
3326 | */ | |
3327 | if (work_pending(&cachep->memcg_params->destroy)) | |
3328 | return; | |
1f458cbf GC |
3329 | /* |
3330 | * We have to defer the actual destroying to a workqueue, because | |
3331 | * we might currently be in a context that cannot sleep. | |
3332 | */ | |
3333 | schedule_work(&cachep->memcg_params->destroy); | |
3334 | } | |
3335 | ||
d9c10ddd MH |
3336 | /* |
3337 | * This lock protects updaters, not readers. We want readers to be as fast as | |
3338 | * they can, and they will either see NULL or a valid cache value. Our model | |
3339 | * allow them to see NULL, in which case the root memcg will be selected. | |
3340 | * | |
3341 | * We need this lock because multiple allocations to the same cache from a non | |
3342 | * will span more than one worker. Only one of them can create the cache. | |
3343 | */ | |
3344 | static DEFINE_MUTEX(memcg_cache_mutex); | |
d7f25f8a | 3345 | |
d9c10ddd MH |
3346 | /* |
3347 | * Called with memcg_cache_mutex held | |
3348 | */ | |
d7f25f8a GC |
3349 | static struct kmem_cache *kmem_cache_dup(struct mem_cgroup *memcg, |
3350 | struct kmem_cache *s) | |
3351 | { | |
d7f25f8a | 3352 | struct kmem_cache *new; |
d9c10ddd | 3353 | static char *tmp_name = NULL; |
d7f25f8a | 3354 | |
d9c10ddd MH |
3355 | lockdep_assert_held(&memcg_cache_mutex); |
3356 | ||
3357 | /* | |
3358 | * kmem_cache_create_memcg duplicates the given name and | |
3359 | * cgroup_name for this name requires RCU context. | |
3360 | * This static temporary buffer is used to prevent from | |
3361 | * pointless shortliving allocation. | |
3362 | */ | |
3363 | if (!tmp_name) { | |
3364 | tmp_name = kmalloc(PATH_MAX, GFP_KERNEL); | |
3365 | if (!tmp_name) | |
3366 | return NULL; | |
3367 | } | |
3368 | ||
3369 | rcu_read_lock(); | |
3370 | snprintf(tmp_name, PATH_MAX, "%s(%d:%s)", s->name, | |
3371 | memcg_cache_id(memcg), cgroup_name(memcg->css.cgroup)); | |
3372 | rcu_read_unlock(); | |
d7f25f8a | 3373 | |
d9c10ddd | 3374 | new = kmem_cache_create_memcg(memcg, tmp_name, s->object_size, s->align, |
943a451a | 3375 | (s->flags & ~SLAB_PANIC), s->ctor, s); |
d7f25f8a | 3376 | |
d79923fa GC |
3377 | if (new) |
3378 | new->allocflags |= __GFP_KMEMCG; | |
3379 | ||
d7f25f8a GC |
3380 | return new; |
3381 | } | |
3382 | ||
d7f25f8a GC |
3383 | static struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg, |
3384 | struct kmem_cache *cachep) | |
3385 | { | |
3386 | struct kmem_cache *new_cachep; | |
3387 | int idx; | |
3388 | ||
3389 | BUG_ON(!memcg_can_account_kmem(memcg)); | |
3390 | ||
3391 | idx = memcg_cache_id(memcg); | |
3392 | ||
3393 | mutex_lock(&memcg_cache_mutex); | |
3394 | new_cachep = cachep->memcg_params->memcg_caches[idx]; | |
20f05310 LZ |
3395 | if (new_cachep) { |
3396 | css_put(&memcg->css); | |
d7f25f8a | 3397 | goto out; |
20f05310 | 3398 | } |
d7f25f8a GC |
3399 | |
3400 | new_cachep = kmem_cache_dup(memcg, cachep); | |
d7f25f8a GC |
3401 | if (new_cachep == NULL) { |
3402 | new_cachep = cachep; | |
20f05310 | 3403 | css_put(&memcg->css); |
d7f25f8a GC |
3404 | goto out; |
3405 | } | |
3406 | ||
1f458cbf | 3407 | atomic_set(&new_cachep->memcg_params->nr_pages , 0); |
d7f25f8a GC |
3408 | |
3409 | cachep->memcg_params->memcg_caches[idx] = new_cachep; | |
3410 | /* | |
3411 | * the readers won't lock, make sure everybody sees the updated value, | |
3412 | * so they won't put stuff in the queue again for no reason | |
3413 | */ | |
3414 | wmb(); | |
3415 | out: | |
3416 | mutex_unlock(&memcg_cache_mutex); | |
3417 | return new_cachep; | |
3418 | } | |
3419 | ||
7cf27982 GC |
3420 | void kmem_cache_destroy_memcg_children(struct kmem_cache *s) |
3421 | { | |
3422 | struct kmem_cache *c; | |
3423 | int i; | |
3424 | ||
3425 | if (!s->memcg_params) | |
3426 | return; | |
3427 | if (!s->memcg_params->is_root_cache) | |
3428 | return; | |
3429 | ||
3430 | /* | |
3431 | * If the cache is being destroyed, we trust that there is no one else | |
3432 | * requesting objects from it. Even if there are, the sanity checks in | |
3433 | * kmem_cache_destroy should caught this ill-case. | |
3434 | * | |
3435 | * Still, we don't want anyone else freeing memcg_caches under our | |
3436 | * noses, which can happen if a new memcg comes to life. As usual, | |
3437 | * we'll take the set_limit_mutex to protect ourselves against this. | |
3438 | */ | |
3439 | mutex_lock(&set_limit_mutex); | |
3440 | for (i = 0; i < memcg_limited_groups_array_size; i++) { | |
3441 | c = s->memcg_params->memcg_caches[i]; | |
3442 | if (!c) | |
3443 | continue; | |
3444 | ||
3445 | /* | |
3446 | * We will now manually delete the caches, so to avoid races | |
3447 | * we need to cancel all pending destruction workers and | |
3448 | * proceed with destruction ourselves. | |
3449 | * | |
3450 | * kmem_cache_destroy() will call kmem_cache_shrink internally, | |
3451 | * and that could spawn the workers again: it is likely that | |
3452 | * the cache still have active pages until this very moment. | |
3453 | * This would lead us back to mem_cgroup_destroy_cache. | |
3454 | * | |
3455 | * But that will not execute at all if the "dead" flag is not | |
3456 | * set, so flip it down to guarantee we are in control. | |
3457 | */ | |
3458 | c->memcg_params->dead = false; | |
22933152 | 3459 | cancel_work_sync(&c->memcg_params->destroy); |
7cf27982 GC |
3460 | kmem_cache_destroy(c); |
3461 | } | |
3462 | mutex_unlock(&set_limit_mutex); | |
3463 | } | |
3464 | ||
d7f25f8a GC |
3465 | struct create_work { |
3466 | struct mem_cgroup *memcg; | |
3467 | struct kmem_cache *cachep; | |
3468 | struct work_struct work; | |
3469 | }; | |
3470 | ||
1f458cbf GC |
3471 | static void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg) |
3472 | { | |
3473 | struct kmem_cache *cachep; | |
3474 | struct memcg_cache_params *params; | |
3475 | ||
3476 | if (!memcg_kmem_is_active(memcg)) | |
3477 | return; | |
3478 | ||
3479 | mutex_lock(&memcg->slab_caches_mutex); | |
3480 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) { | |
3481 | cachep = memcg_params_to_cache(params); | |
3482 | cachep->memcg_params->dead = true; | |
1f458cbf GC |
3483 | schedule_work(&cachep->memcg_params->destroy); |
3484 | } | |
3485 | mutex_unlock(&memcg->slab_caches_mutex); | |
3486 | } | |
3487 | ||
d7f25f8a GC |
3488 | static void memcg_create_cache_work_func(struct work_struct *w) |
3489 | { | |
3490 | struct create_work *cw; | |
3491 | ||
3492 | cw = container_of(w, struct create_work, work); | |
3493 | memcg_create_kmem_cache(cw->memcg, cw->cachep); | |
d7f25f8a GC |
3494 | kfree(cw); |
3495 | } | |
3496 | ||
3497 | /* | |
3498 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 3499 | */ |
0e9d92f2 GC |
3500 | static void __memcg_create_cache_enqueue(struct mem_cgroup *memcg, |
3501 | struct kmem_cache *cachep) | |
d7f25f8a GC |
3502 | { |
3503 | struct create_work *cw; | |
3504 | ||
3505 | cw = kmalloc(sizeof(struct create_work), GFP_NOWAIT); | |
ca0dde97 LZ |
3506 | if (cw == NULL) { |
3507 | css_put(&memcg->css); | |
d7f25f8a GC |
3508 | return; |
3509 | } | |
3510 | ||
3511 | cw->memcg = memcg; | |
3512 | cw->cachep = cachep; | |
3513 | ||
3514 | INIT_WORK(&cw->work, memcg_create_cache_work_func); | |
3515 | schedule_work(&cw->work); | |
3516 | } | |
3517 | ||
0e9d92f2 GC |
3518 | static void memcg_create_cache_enqueue(struct mem_cgroup *memcg, |
3519 | struct kmem_cache *cachep) | |
3520 | { | |
3521 | /* | |
3522 | * We need to stop accounting when we kmalloc, because if the | |
3523 | * corresponding kmalloc cache is not yet created, the first allocation | |
3524 | * in __memcg_create_cache_enqueue will recurse. | |
3525 | * | |
3526 | * However, it is better to enclose the whole function. Depending on | |
3527 | * the debugging options enabled, INIT_WORK(), for instance, can | |
3528 | * trigger an allocation. This too, will make us recurse. Because at | |
3529 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
3530 | * the safest choice is to do it like this, wrapping the whole function. | |
3531 | */ | |
3532 | memcg_stop_kmem_account(); | |
3533 | __memcg_create_cache_enqueue(memcg, cachep); | |
3534 | memcg_resume_kmem_account(); | |
3535 | } | |
d7f25f8a GC |
3536 | /* |
3537 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
3538 | * We try to use the current memcg's version of the cache. | |
3539 | * | |
3540 | * If the cache does not exist yet, if we are the first user of it, | |
3541 | * we either create it immediately, if possible, or create it asynchronously | |
3542 | * in a workqueue. | |
3543 | * In the latter case, we will let the current allocation go through with | |
3544 | * the original cache. | |
3545 | * | |
3546 | * Can't be called in interrupt context or from kernel threads. | |
3547 | * This function needs to be called with rcu_read_lock() held. | |
3548 | */ | |
3549 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, | |
3550 | gfp_t gfp) | |
3551 | { | |
3552 | struct mem_cgroup *memcg; | |
3553 | int idx; | |
3554 | ||
3555 | VM_BUG_ON(!cachep->memcg_params); | |
3556 | VM_BUG_ON(!cachep->memcg_params->is_root_cache); | |
3557 | ||
0e9d92f2 GC |
3558 | if (!current->mm || current->memcg_kmem_skip_account) |
3559 | return cachep; | |
3560 | ||
d7f25f8a GC |
3561 | rcu_read_lock(); |
3562 | memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner)); | |
d7f25f8a GC |
3563 | |
3564 | if (!memcg_can_account_kmem(memcg)) | |
ca0dde97 | 3565 | goto out; |
d7f25f8a GC |
3566 | |
3567 | idx = memcg_cache_id(memcg); | |
3568 | ||
3569 | /* | |
3570 | * barrier to mare sure we're always seeing the up to date value. The | |
3571 | * code updating memcg_caches will issue a write barrier to match this. | |
3572 | */ | |
3573 | read_barrier_depends(); | |
ca0dde97 LZ |
3574 | if (likely(cachep->memcg_params->memcg_caches[idx])) { |
3575 | cachep = cachep->memcg_params->memcg_caches[idx]; | |
3576 | goto out; | |
d7f25f8a GC |
3577 | } |
3578 | ||
ca0dde97 LZ |
3579 | /* The corresponding put will be done in the workqueue. */ |
3580 | if (!css_tryget(&memcg->css)) | |
3581 | goto out; | |
3582 | rcu_read_unlock(); | |
3583 | ||
3584 | /* | |
3585 | * If we are in a safe context (can wait, and not in interrupt | |
3586 | * context), we could be be predictable and return right away. | |
3587 | * This would guarantee that the allocation being performed | |
3588 | * already belongs in the new cache. | |
3589 | * | |
3590 | * However, there are some clashes that can arrive from locking. | |
3591 | * For instance, because we acquire the slab_mutex while doing | |
3592 | * kmem_cache_dup, this means no further allocation could happen | |
3593 | * with the slab_mutex held. | |
3594 | * | |
3595 | * Also, because cache creation issue get_online_cpus(), this | |
3596 | * creates a lock chain: memcg_slab_mutex -> cpu_hotplug_mutex, | |
3597 | * that ends up reversed during cpu hotplug. (cpuset allocates | |
3598 | * a bunch of GFP_KERNEL memory during cpuup). Due to all that, | |
3599 | * better to defer everything. | |
3600 | */ | |
3601 | memcg_create_cache_enqueue(memcg, cachep); | |
3602 | return cachep; | |
3603 | out: | |
3604 | rcu_read_unlock(); | |
3605 | return cachep; | |
d7f25f8a GC |
3606 | } |
3607 | EXPORT_SYMBOL(__memcg_kmem_get_cache); | |
3608 | ||
7ae1e1d0 GC |
3609 | /* |
3610 | * We need to verify if the allocation against current->mm->owner's memcg is | |
3611 | * possible for the given order. But the page is not allocated yet, so we'll | |
3612 | * need a further commit step to do the final arrangements. | |
3613 | * | |
3614 | * It is possible for the task to switch cgroups in this mean time, so at | |
3615 | * commit time, we can't rely on task conversion any longer. We'll then use | |
3616 | * the handle argument to return to the caller which cgroup we should commit | |
3617 | * against. We could also return the memcg directly and avoid the pointer | |
3618 | * passing, but a boolean return value gives better semantics considering | |
3619 | * the compiled-out case as well. | |
3620 | * | |
3621 | * Returning true means the allocation is possible. | |
3622 | */ | |
3623 | bool | |
3624 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) | |
3625 | { | |
3626 | struct mem_cgroup *memcg; | |
3627 | int ret; | |
3628 | ||
3629 | *_memcg = NULL; | |
6d42c232 GC |
3630 | |
3631 | /* | |
3632 | * Disabling accounting is only relevant for some specific memcg | |
3633 | * internal allocations. Therefore we would initially not have such | |
3634 | * check here, since direct calls to the page allocator that are marked | |
3635 | * with GFP_KMEMCG only happen outside memcg core. We are mostly | |
3636 | * concerned with cache allocations, and by having this test at | |
3637 | * memcg_kmem_get_cache, we are already able to relay the allocation to | |
3638 | * the root cache and bypass the memcg cache altogether. | |
3639 | * | |
3640 | * There is one exception, though: the SLUB allocator does not create | |
3641 | * large order caches, but rather service large kmallocs directly from | |
3642 | * the page allocator. Therefore, the following sequence when backed by | |
3643 | * the SLUB allocator: | |
3644 | * | |
3645 | * memcg_stop_kmem_account(); | |
3646 | * kmalloc(<large_number>) | |
3647 | * memcg_resume_kmem_account(); | |
3648 | * | |
3649 | * would effectively ignore the fact that we should skip accounting, | |
3650 | * since it will drive us directly to this function without passing | |
3651 | * through the cache selector memcg_kmem_get_cache. Such large | |
3652 | * allocations are extremely rare but can happen, for instance, for the | |
3653 | * cache arrays. We bring this test here. | |
3654 | */ | |
3655 | if (!current->mm || current->memcg_kmem_skip_account) | |
3656 | return true; | |
3657 | ||
7ae1e1d0 GC |
3658 | memcg = try_get_mem_cgroup_from_mm(current->mm); |
3659 | ||
3660 | /* | |
3661 | * very rare case described in mem_cgroup_from_task. Unfortunately there | |
3662 | * isn't much we can do without complicating this too much, and it would | |
3663 | * be gfp-dependent anyway. Just let it go | |
3664 | */ | |
3665 | if (unlikely(!memcg)) | |
3666 | return true; | |
3667 | ||
3668 | if (!memcg_can_account_kmem(memcg)) { | |
3669 | css_put(&memcg->css); | |
3670 | return true; | |
3671 | } | |
3672 | ||
7ae1e1d0 GC |
3673 | ret = memcg_charge_kmem(memcg, gfp, PAGE_SIZE << order); |
3674 | if (!ret) | |
3675 | *_memcg = memcg; | |
7ae1e1d0 GC |
3676 | |
3677 | css_put(&memcg->css); | |
3678 | return (ret == 0); | |
3679 | } | |
3680 | ||
3681 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
3682 | int order) | |
3683 | { | |
3684 | struct page_cgroup *pc; | |
3685 | ||
3686 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
3687 | ||
3688 | /* The page allocation failed. Revert */ | |
3689 | if (!page) { | |
3690 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); | |
7ae1e1d0 GC |
3691 | return; |
3692 | } | |
3693 | ||
3694 | pc = lookup_page_cgroup(page); | |
3695 | lock_page_cgroup(pc); | |
3696 | pc->mem_cgroup = memcg; | |
3697 | SetPageCgroupUsed(pc); | |
3698 | unlock_page_cgroup(pc); | |
3699 | } | |
3700 | ||
3701 | void __memcg_kmem_uncharge_pages(struct page *page, int order) | |
3702 | { | |
3703 | struct mem_cgroup *memcg = NULL; | |
3704 | struct page_cgroup *pc; | |
3705 | ||
3706 | ||
3707 | pc = lookup_page_cgroup(page); | |
3708 | /* | |
3709 | * Fast unlocked return. Theoretically might have changed, have to | |
3710 | * check again after locking. | |
3711 | */ | |
3712 | if (!PageCgroupUsed(pc)) | |
3713 | return; | |
3714 | ||
3715 | lock_page_cgroup(pc); | |
3716 | if (PageCgroupUsed(pc)) { | |
3717 | memcg = pc->mem_cgroup; | |
3718 | ClearPageCgroupUsed(pc); | |
3719 | } | |
3720 | unlock_page_cgroup(pc); | |
3721 | ||
3722 | /* | |
3723 | * We trust that only if there is a memcg associated with the page, it | |
3724 | * is a valid allocation | |
3725 | */ | |
3726 | if (!memcg) | |
3727 | return; | |
3728 | ||
3729 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
3730 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); | |
7ae1e1d0 | 3731 | } |
1f458cbf GC |
3732 | #else |
3733 | static inline void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg) | |
3734 | { | |
3735 | } | |
7ae1e1d0 GC |
3736 | #endif /* CONFIG_MEMCG_KMEM */ |
3737 | ||
ca3e0214 KH |
3738 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3739 | ||
a0db00fc | 3740 | #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION) |
ca3e0214 KH |
3741 | /* |
3742 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
3743 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
3744 | * charge/uncharge will be never happen and move_account() is done under | |
3745 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 3746 | */ |
e94c8a9c | 3747 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 KH |
3748 | { |
3749 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
e94c8a9c | 3750 | struct page_cgroup *pc; |
b070e65c | 3751 | struct mem_cgroup *memcg; |
e94c8a9c | 3752 | int i; |
ca3e0214 | 3753 | |
3d37c4a9 KH |
3754 | if (mem_cgroup_disabled()) |
3755 | return; | |
b070e65c DR |
3756 | |
3757 | memcg = head_pc->mem_cgroup; | |
e94c8a9c KH |
3758 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
3759 | pc = head_pc + i; | |
b070e65c | 3760 | pc->mem_cgroup = memcg; |
e94c8a9c | 3761 | smp_wmb();/* see __commit_charge() */ |
e94c8a9c KH |
3762 | pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
3763 | } | |
b070e65c DR |
3764 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
3765 | HPAGE_PMD_NR); | |
ca3e0214 | 3766 | } |
12d27107 | 3767 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3768 | |
f817ed48 | 3769 | /** |
de3638d9 | 3770 | * mem_cgroup_move_account - move account of the page |
5564e88b | 3771 | * @page: the page |
7ec99d62 | 3772 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
3773 | * @pc: page_cgroup of the page. |
3774 | * @from: mem_cgroup which the page is moved from. | |
3775 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
3776 | * | |
3777 | * The caller must confirm following. | |
08e552c6 | 3778 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 3779 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 3780 | * |
2f3479b1 KH |
3781 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
3782 | * from old cgroup. | |
f817ed48 | 3783 | */ |
7ec99d62 JW |
3784 | static int mem_cgroup_move_account(struct page *page, |
3785 | unsigned int nr_pages, | |
3786 | struct page_cgroup *pc, | |
3787 | struct mem_cgroup *from, | |
2f3479b1 | 3788 | struct mem_cgroup *to) |
f817ed48 | 3789 | { |
de3638d9 JW |
3790 | unsigned long flags; |
3791 | int ret; | |
b2402857 | 3792 | bool anon = PageAnon(page); |
987eba66 | 3793 | |
f817ed48 | 3794 | VM_BUG_ON(from == to); |
5564e88b | 3795 | VM_BUG_ON(PageLRU(page)); |
de3638d9 JW |
3796 | /* |
3797 | * The page is isolated from LRU. So, collapse function | |
3798 | * will not handle this page. But page splitting can happen. | |
3799 | * Do this check under compound_page_lock(). The caller should | |
3800 | * hold it. | |
3801 | */ | |
3802 | ret = -EBUSY; | |
7ec99d62 | 3803 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
3804 | goto out; |
3805 | ||
3806 | lock_page_cgroup(pc); | |
3807 | ||
3808 | ret = -EINVAL; | |
3809 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
3810 | goto unlock; | |
3811 | ||
312734c0 | 3812 | move_lock_mem_cgroup(from, &flags); |
f817ed48 | 3813 | |
2ff76f11 | 3814 | if (!anon && page_mapped(page)) { |
c62b1a3b KH |
3815 | /* Update mapped_file data for mem_cgroup */ |
3816 | preempt_disable(); | |
3817 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
3818 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
3819 | preempt_enable(); | |
d69b042f | 3820 | } |
b070e65c | 3821 | mem_cgroup_charge_statistics(from, page, anon, -nr_pages); |
d69b042f | 3822 | |
854ffa8d | 3823 | /* caller should have done css_get */ |
08e552c6 | 3824 | pc->mem_cgroup = to; |
b070e65c | 3825 | mem_cgroup_charge_statistics(to, page, anon, nr_pages); |
312734c0 | 3826 | move_unlock_mem_cgroup(from, &flags); |
de3638d9 JW |
3827 | ret = 0; |
3828 | unlock: | |
57f9fd7d | 3829 | unlock_page_cgroup(pc); |
d2265e6f KH |
3830 | /* |
3831 | * check events | |
3832 | */ | |
5564e88b JW |
3833 | memcg_check_events(to, page); |
3834 | memcg_check_events(from, page); | |
de3638d9 | 3835 | out: |
f817ed48 KH |
3836 | return ret; |
3837 | } | |
3838 | ||
2ef37d3f MH |
3839 | /** |
3840 | * mem_cgroup_move_parent - moves page to the parent group | |
3841 | * @page: the page to move | |
3842 | * @pc: page_cgroup of the page | |
3843 | * @child: page's cgroup | |
3844 | * | |
3845 | * move charges to its parent or the root cgroup if the group has no | |
3846 | * parent (aka use_hierarchy==0). | |
3847 | * Although this might fail (get_page_unless_zero, isolate_lru_page or | |
3848 | * mem_cgroup_move_account fails) the failure is always temporary and | |
3849 | * it signals a race with a page removal/uncharge or migration. In the | |
3850 | * first case the page is on the way out and it will vanish from the LRU | |
3851 | * on the next attempt and the call should be retried later. | |
3852 | * Isolation from the LRU fails only if page has been isolated from | |
3853 | * the LRU since we looked at it and that usually means either global | |
3854 | * reclaim or migration going on. The page will either get back to the | |
3855 | * LRU or vanish. | |
3856 | * Finaly mem_cgroup_move_account fails only if the page got uncharged | |
3857 | * (!PageCgroupUsed) or moved to a different group. The page will | |
3858 | * disappear in the next attempt. | |
f817ed48 | 3859 | */ |
5564e88b JW |
3860 | static int mem_cgroup_move_parent(struct page *page, |
3861 | struct page_cgroup *pc, | |
6068bf01 | 3862 | struct mem_cgroup *child) |
f817ed48 | 3863 | { |
f817ed48 | 3864 | struct mem_cgroup *parent; |
7ec99d62 | 3865 | unsigned int nr_pages; |
4be4489f | 3866 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
3867 | int ret; |
3868 | ||
d8423011 | 3869 | VM_BUG_ON(mem_cgroup_is_root(child)); |
f817ed48 | 3870 | |
57f9fd7d DN |
3871 | ret = -EBUSY; |
3872 | if (!get_page_unless_zero(page)) | |
3873 | goto out; | |
3874 | if (isolate_lru_page(page)) | |
3875 | goto put; | |
52dbb905 | 3876 | |
7ec99d62 | 3877 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 3878 | |
cc926f78 KH |
3879 | parent = parent_mem_cgroup(child); |
3880 | /* | |
3881 | * If no parent, move charges to root cgroup. | |
3882 | */ | |
3883 | if (!parent) | |
3884 | parent = root_mem_cgroup; | |
f817ed48 | 3885 | |
2ef37d3f MH |
3886 | if (nr_pages > 1) { |
3887 | VM_BUG_ON(!PageTransHuge(page)); | |
987eba66 | 3888 | flags = compound_lock_irqsave(page); |
2ef37d3f | 3889 | } |
987eba66 | 3890 | |
cc926f78 | 3891 | ret = mem_cgroup_move_account(page, nr_pages, |
2f3479b1 | 3892 | pc, child, parent); |
cc926f78 KH |
3893 | if (!ret) |
3894 | __mem_cgroup_cancel_local_charge(child, nr_pages); | |
8dba474f | 3895 | |
7ec99d62 | 3896 | if (nr_pages > 1) |
987eba66 | 3897 | compound_unlock_irqrestore(page, flags); |
08e552c6 | 3898 | putback_lru_page(page); |
57f9fd7d | 3899 | put: |
40d58138 | 3900 | put_page(page); |
57f9fd7d | 3901 | out: |
f817ed48 KH |
3902 | return ret; |
3903 | } | |
3904 | ||
7a81b88c KH |
3905 | /* |
3906 | * Charge the memory controller for page usage. | |
3907 | * Return | |
3908 | * 0 if the charge was successful | |
3909 | * < 0 if the cgroup is over its limit | |
3910 | */ | |
3911 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
73045c47 | 3912 | gfp_t gfp_mask, enum charge_type ctype) |
7a81b88c | 3913 | { |
c0ff4b85 | 3914 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 3915 | unsigned int nr_pages = 1; |
8493ae43 | 3916 | bool oom = true; |
7a81b88c | 3917 | int ret; |
ec168510 | 3918 | |
37c2ac78 | 3919 | if (PageTransHuge(page)) { |
7ec99d62 | 3920 | nr_pages <<= compound_order(page); |
37c2ac78 | 3921 | VM_BUG_ON(!PageTransHuge(page)); |
8493ae43 JW |
3922 | /* |
3923 | * Never OOM-kill a process for a huge page. The | |
3924 | * fault handler will fall back to regular pages. | |
3925 | */ | |
3926 | oom = false; | |
37c2ac78 | 3927 | } |
7a81b88c | 3928 | |
c0ff4b85 | 3929 | ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom); |
38c5d72f | 3930 | if (ret == -ENOMEM) |
7a81b88c | 3931 | return ret; |
ce587e65 | 3932 | __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false); |
8a9f3ccd | 3933 | return 0; |
8a9f3ccd BS |
3934 | } |
3935 | ||
7a81b88c KH |
3936 | int mem_cgroup_newpage_charge(struct page *page, |
3937 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 3938 | { |
f8d66542 | 3939 | if (mem_cgroup_disabled()) |
cede86ac | 3940 | return 0; |
7a0524cf JW |
3941 | VM_BUG_ON(page_mapped(page)); |
3942 | VM_BUG_ON(page->mapping && !PageAnon(page)); | |
3943 | VM_BUG_ON(!mm); | |
217bc319 | 3944 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
41326c17 | 3945 | MEM_CGROUP_CHARGE_TYPE_ANON); |
217bc319 KH |
3946 | } |
3947 | ||
54595fe2 KH |
3948 | /* |
3949 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
3950 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 3951 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
3952 | * "commit()" or removed by "cancel()" |
3953 | */ | |
0435a2fd JW |
3954 | static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
3955 | struct page *page, | |
3956 | gfp_t mask, | |
3957 | struct mem_cgroup **memcgp) | |
8c7c6e34 | 3958 | { |
c0ff4b85 | 3959 | struct mem_cgroup *memcg; |
90deb788 | 3960 | struct page_cgroup *pc; |
54595fe2 | 3961 | int ret; |
8c7c6e34 | 3962 | |
90deb788 JW |
3963 | pc = lookup_page_cgroup(page); |
3964 | /* | |
3965 | * Every swap fault against a single page tries to charge the | |
3966 | * page, bail as early as possible. shmem_unuse() encounters | |
3967 | * already charged pages, too. The USED bit is protected by | |
3968 | * the page lock, which serializes swap cache removal, which | |
3969 | * in turn serializes uncharging. | |
3970 | */ | |
3971 | if (PageCgroupUsed(pc)) | |
3972 | return 0; | |
8c7c6e34 KH |
3973 | if (!do_swap_account) |
3974 | goto charge_cur_mm; | |
c0ff4b85 R |
3975 | memcg = try_get_mem_cgroup_from_page(page); |
3976 | if (!memcg) | |
54595fe2 | 3977 | goto charge_cur_mm; |
72835c86 JW |
3978 | *memcgp = memcg; |
3979 | ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true); | |
c0ff4b85 | 3980 | css_put(&memcg->css); |
38c5d72f KH |
3981 | if (ret == -EINTR) |
3982 | ret = 0; | |
54595fe2 | 3983 | return ret; |
8c7c6e34 | 3984 | charge_cur_mm: |
38c5d72f KH |
3985 | ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true); |
3986 | if (ret == -EINTR) | |
3987 | ret = 0; | |
3988 | return ret; | |
8c7c6e34 KH |
3989 | } |
3990 | ||
0435a2fd JW |
3991 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page, |
3992 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
3993 | { | |
3994 | *memcgp = NULL; | |
3995 | if (mem_cgroup_disabled()) | |
3996 | return 0; | |
bdf4f4d2 JW |
3997 | /* |
3998 | * A racing thread's fault, or swapoff, may have already | |
3999 | * updated the pte, and even removed page from swap cache: in | |
4000 | * those cases unuse_pte()'s pte_same() test will fail; but | |
4001 | * there's also a KSM case which does need to charge the page. | |
4002 | */ | |
4003 | if (!PageSwapCache(page)) { | |
4004 | int ret; | |
4005 | ||
4006 | ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, memcgp, true); | |
4007 | if (ret == -EINTR) | |
4008 | ret = 0; | |
4009 | return ret; | |
4010 | } | |
0435a2fd JW |
4011 | return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp); |
4012 | } | |
4013 | ||
827a03d2 JW |
4014 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg) |
4015 | { | |
4016 | if (mem_cgroup_disabled()) | |
4017 | return; | |
4018 | if (!memcg) | |
4019 | return; | |
4020 | __mem_cgroup_cancel_charge(memcg, 1); | |
4021 | } | |
4022 | ||
83aae4c7 | 4023 | static void |
72835c86 | 4024 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg, |
83aae4c7 | 4025 | enum charge_type ctype) |
7a81b88c | 4026 | { |
f8d66542 | 4027 | if (mem_cgroup_disabled()) |
7a81b88c | 4028 | return; |
72835c86 | 4029 | if (!memcg) |
7a81b88c | 4030 | return; |
5a6475a4 | 4031 | |
ce587e65 | 4032 | __mem_cgroup_commit_charge(memcg, page, 1, ctype, true); |
8c7c6e34 KH |
4033 | /* |
4034 | * Now swap is on-memory. This means this page may be | |
4035 | * counted both as mem and swap....double count. | |
03f3c433 KH |
4036 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
4037 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
4038 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 4039 | */ |
03f3c433 | 4040 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 4041 | swp_entry_t ent = {.val = page_private(page)}; |
86493009 | 4042 | mem_cgroup_uncharge_swap(ent); |
8c7c6e34 | 4043 | } |
7a81b88c KH |
4044 | } |
4045 | ||
72835c86 JW |
4046 | void mem_cgroup_commit_charge_swapin(struct page *page, |
4047 | struct mem_cgroup *memcg) | |
83aae4c7 | 4048 | { |
72835c86 | 4049 | __mem_cgroup_commit_charge_swapin(page, memcg, |
41326c17 | 4050 | MEM_CGROUP_CHARGE_TYPE_ANON); |
83aae4c7 DN |
4051 | } |
4052 | ||
827a03d2 JW |
4053 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
4054 | gfp_t gfp_mask) | |
7a81b88c | 4055 | { |
827a03d2 JW |
4056 | struct mem_cgroup *memcg = NULL; |
4057 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
4058 | int ret; | |
4059 | ||
f8d66542 | 4060 | if (mem_cgroup_disabled()) |
827a03d2 JW |
4061 | return 0; |
4062 | if (PageCompound(page)) | |
4063 | return 0; | |
4064 | ||
827a03d2 JW |
4065 | if (!PageSwapCache(page)) |
4066 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, type); | |
4067 | else { /* page is swapcache/shmem */ | |
0435a2fd JW |
4068 | ret = __mem_cgroup_try_charge_swapin(mm, page, |
4069 | gfp_mask, &memcg); | |
827a03d2 JW |
4070 | if (!ret) |
4071 | __mem_cgroup_commit_charge_swapin(page, memcg, type); | |
4072 | } | |
4073 | return ret; | |
7a81b88c KH |
4074 | } |
4075 | ||
c0ff4b85 | 4076 | static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg, |
7ec99d62 JW |
4077 | unsigned int nr_pages, |
4078 | const enum charge_type ctype) | |
569b846d KH |
4079 | { |
4080 | struct memcg_batch_info *batch = NULL; | |
4081 | bool uncharge_memsw = true; | |
7ec99d62 | 4082 | |
569b846d KH |
4083 | /* If swapout, usage of swap doesn't decrease */ |
4084 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
4085 | uncharge_memsw = false; | |
569b846d KH |
4086 | |
4087 | batch = ¤t->memcg_batch; | |
4088 | /* | |
4089 | * In usual, we do css_get() when we remember memcg pointer. | |
4090 | * But in this case, we keep res->usage until end of a series of | |
4091 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
4092 | */ | |
4093 | if (!batch->memcg) | |
c0ff4b85 | 4094 | batch->memcg = memcg; |
3c11ecf4 KH |
4095 | /* |
4096 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
25985edc | 4097 | * In those cases, all pages freed continuously can be expected to be in |
3c11ecf4 KH |
4098 | * the same cgroup and we have chance to coalesce uncharges. |
4099 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
4100 | * because we want to do uncharge as soon as possible. | |
4101 | */ | |
4102 | ||
4103 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
4104 | goto direct_uncharge; | |
4105 | ||
7ec99d62 | 4106 | if (nr_pages > 1) |
ec168510 AA |
4107 | goto direct_uncharge; |
4108 | ||
569b846d KH |
4109 | /* |
4110 | * In typical case, batch->memcg == mem. This means we can | |
4111 | * merge a series of uncharges to an uncharge of res_counter. | |
4112 | * If not, we uncharge res_counter ony by one. | |
4113 | */ | |
c0ff4b85 | 4114 | if (batch->memcg != memcg) |
569b846d KH |
4115 | goto direct_uncharge; |
4116 | /* remember freed charge and uncharge it later */ | |
7ffd4ca7 | 4117 | batch->nr_pages++; |
569b846d | 4118 | if (uncharge_memsw) |
7ffd4ca7 | 4119 | batch->memsw_nr_pages++; |
569b846d KH |
4120 | return; |
4121 | direct_uncharge: | |
c0ff4b85 | 4122 | res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE); |
569b846d | 4123 | if (uncharge_memsw) |
c0ff4b85 R |
4124 | res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE); |
4125 | if (unlikely(batch->memcg != memcg)) | |
4126 | memcg_oom_recover(memcg); | |
569b846d | 4127 | } |
7a81b88c | 4128 | |
8a9f3ccd | 4129 | /* |
69029cd5 | 4130 | * uncharge if !page_mapped(page) |
8a9f3ccd | 4131 | */ |
8c7c6e34 | 4132 | static struct mem_cgroup * |
0030f535 JW |
4133 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype, |
4134 | bool end_migration) | |
8a9f3ccd | 4135 | { |
c0ff4b85 | 4136 | struct mem_cgroup *memcg = NULL; |
7ec99d62 JW |
4137 | unsigned int nr_pages = 1; |
4138 | struct page_cgroup *pc; | |
b2402857 | 4139 | bool anon; |
8a9f3ccd | 4140 | |
f8d66542 | 4141 | if (mem_cgroup_disabled()) |
8c7c6e34 | 4142 | return NULL; |
4077960e | 4143 | |
37c2ac78 | 4144 | if (PageTransHuge(page)) { |
7ec99d62 | 4145 | nr_pages <<= compound_order(page); |
37c2ac78 AA |
4146 | VM_BUG_ON(!PageTransHuge(page)); |
4147 | } | |
8697d331 | 4148 | /* |
3c541e14 | 4149 | * Check if our page_cgroup is valid |
8697d331 | 4150 | */ |
52d4b9ac | 4151 | pc = lookup_page_cgroup(page); |
cfa44946 | 4152 | if (unlikely(!PageCgroupUsed(pc))) |
8c7c6e34 | 4153 | return NULL; |
b9c565d5 | 4154 | |
52d4b9ac | 4155 | lock_page_cgroup(pc); |
d13d1443 | 4156 | |
c0ff4b85 | 4157 | memcg = pc->mem_cgroup; |
8c7c6e34 | 4158 | |
d13d1443 KH |
4159 | if (!PageCgroupUsed(pc)) |
4160 | goto unlock_out; | |
4161 | ||
b2402857 KH |
4162 | anon = PageAnon(page); |
4163 | ||
d13d1443 | 4164 | switch (ctype) { |
41326c17 | 4165 | case MEM_CGROUP_CHARGE_TYPE_ANON: |
2ff76f11 KH |
4166 | /* |
4167 | * Generally PageAnon tells if it's the anon statistics to be | |
4168 | * updated; but sometimes e.g. mem_cgroup_uncharge_page() is | |
4169 | * used before page reached the stage of being marked PageAnon. | |
4170 | */ | |
b2402857 KH |
4171 | anon = true; |
4172 | /* fallthrough */ | |
8a9478ca | 4173 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c | 4174 | /* See mem_cgroup_prepare_migration() */ |
0030f535 JW |
4175 | if (page_mapped(page)) |
4176 | goto unlock_out; | |
4177 | /* | |
4178 | * Pages under migration may not be uncharged. But | |
4179 | * end_migration() /must/ be the one uncharging the | |
4180 | * unused post-migration page and so it has to call | |
4181 | * here with the migration bit still set. See the | |
4182 | * res_counter handling below. | |
4183 | */ | |
4184 | if (!end_migration && PageCgroupMigration(pc)) | |
d13d1443 KH |
4185 | goto unlock_out; |
4186 | break; | |
4187 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
4188 | if (!PageAnon(page)) { /* Shared memory */ | |
4189 | if (page->mapping && !page_is_file_cache(page)) | |
4190 | goto unlock_out; | |
4191 | } else if (page_mapped(page)) /* Anon */ | |
4192 | goto unlock_out; | |
4193 | break; | |
4194 | default: | |
4195 | break; | |
52d4b9ac | 4196 | } |
d13d1443 | 4197 | |
b070e65c | 4198 | mem_cgroup_charge_statistics(memcg, page, anon, -nr_pages); |
04046e1a | 4199 | |
52d4b9ac | 4200 | ClearPageCgroupUsed(pc); |
544122e5 KH |
4201 | /* |
4202 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
4203 | * freed from LRU. This is safe because uncharged page is expected not | |
4204 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
4205 | * special functions. | |
4206 | */ | |
b9c565d5 | 4207 | |
52d4b9ac | 4208 | unlock_page_cgroup(pc); |
f75ca962 | 4209 | /* |
c0ff4b85 | 4210 | * even after unlock, we have memcg->res.usage here and this memcg |
4050377b | 4211 | * will never be freed, so it's safe to call css_get(). |
f75ca962 | 4212 | */ |
c0ff4b85 | 4213 | memcg_check_events(memcg, page); |
f75ca962 | 4214 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
c0ff4b85 | 4215 | mem_cgroup_swap_statistics(memcg, true); |
4050377b | 4216 | css_get(&memcg->css); |
f75ca962 | 4217 | } |
0030f535 JW |
4218 | /* |
4219 | * Migration does not charge the res_counter for the | |
4220 | * replacement page, so leave it alone when phasing out the | |
4221 | * page that is unused after the migration. | |
4222 | */ | |
4223 | if (!end_migration && !mem_cgroup_is_root(memcg)) | |
c0ff4b85 | 4224 | mem_cgroup_do_uncharge(memcg, nr_pages, ctype); |
6d12e2d8 | 4225 | |
c0ff4b85 | 4226 | return memcg; |
d13d1443 KH |
4227 | |
4228 | unlock_out: | |
4229 | unlock_page_cgroup(pc); | |
8c7c6e34 | 4230 | return NULL; |
3c541e14 BS |
4231 | } |
4232 | ||
69029cd5 KH |
4233 | void mem_cgroup_uncharge_page(struct page *page) |
4234 | { | |
52d4b9ac KH |
4235 | /* early check. */ |
4236 | if (page_mapped(page)) | |
4237 | return; | |
40f23a21 | 4238 | VM_BUG_ON(page->mapping && !PageAnon(page)); |
28ccddf7 JW |
4239 | /* |
4240 | * If the page is in swap cache, uncharge should be deferred | |
4241 | * to the swap path, which also properly accounts swap usage | |
4242 | * and handles memcg lifetime. | |
4243 | * | |
4244 | * Note that this check is not stable and reclaim may add the | |
4245 | * page to swap cache at any time after this. However, if the | |
4246 | * page is not in swap cache by the time page->mapcount hits | |
4247 | * 0, there won't be any page table references to the swap | |
4248 | * slot, and reclaim will free it and not actually write the | |
4249 | * page to disk. | |
4250 | */ | |
0c59b89c JW |
4251 | if (PageSwapCache(page)) |
4252 | return; | |
0030f535 | 4253 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false); |
69029cd5 KH |
4254 | } |
4255 | ||
4256 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
4257 | { | |
4258 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 4259 | VM_BUG_ON(page->mapping); |
0030f535 | 4260 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false); |
69029cd5 KH |
4261 | } |
4262 | ||
569b846d KH |
4263 | /* |
4264 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
4265 | * In that cases, pages are freed continuously and we can expect pages | |
4266 | * are in the same memcg. All these calls itself limits the number of | |
4267 | * pages freed at once, then uncharge_start/end() is called properly. | |
4268 | * This may be called prural(2) times in a context, | |
4269 | */ | |
4270 | ||
4271 | void mem_cgroup_uncharge_start(void) | |
4272 | { | |
4273 | current->memcg_batch.do_batch++; | |
4274 | /* We can do nest. */ | |
4275 | if (current->memcg_batch.do_batch == 1) { | |
4276 | current->memcg_batch.memcg = NULL; | |
7ffd4ca7 JW |
4277 | current->memcg_batch.nr_pages = 0; |
4278 | current->memcg_batch.memsw_nr_pages = 0; | |
569b846d KH |
4279 | } |
4280 | } | |
4281 | ||
4282 | void mem_cgroup_uncharge_end(void) | |
4283 | { | |
4284 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
4285 | ||
4286 | if (!batch->do_batch) | |
4287 | return; | |
4288 | ||
4289 | batch->do_batch--; | |
4290 | if (batch->do_batch) /* If stacked, do nothing. */ | |
4291 | return; | |
4292 | ||
4293 | if (!batch->memcg) | |
4294 | return; | |
4295 | /* | |
4296 | * This "batch->memcg" is valid without any css_get/put etc... | |
4297 | * bacause we hide charges behind us. | |
4298 | */ | |
7ffd4ca7 JW |
4299 | if (batch->nr_pages) |
4300 | res_counter_uncharge(&batch->memcg->res, | |
4301 | batch->nr_pages * PAGE_SIZE); | |
4302 | if (batch->memsw_nr_pages) | |
4303 | res_counter_uncharge(&batch->memcg->memsw, | |
4304 | batch->memsw_nr_pages * PAGE_SIZE); | |
3c11ecf4 | 4305 | memcg_oom_recover(batch->memcg); |
569b846d KH |
4306 | /* forget this pointer (for sanity check) */ |
4307 | batch->memcg = NULL; | |
4308 | } | |
4309 | ||
e767e056 | 4310 | #ifdef CONFIG_SWAP |
8c7c6e34 | 4311 | /* |
e767e056 | 4312 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
4313 | * memcg information is recorded to swap_cgroup of "ent" |
4314 | */ | |
8a9478ca KH |
4315 | void |
4316 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
4317 | { |
4318 | struct mem_cgroup *memcg; | |
8a9478ca KH |
4319 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
4320 | ||
4321 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
4322 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
4323 | ||
0030f535 | 4324 | memcg = __mem_cgroup_uncharge_common(page, ctype, false); |
8c7c6e34 | 4325 | |
f75ca962 KH |
4326 | /* |
4327 | * record memcg information, if swapout && memcg != NULL, | |
4050377b | 4328 | * css_get() was called in uncharge(). |
f75ca962 KH |
4329 | */ |
4330 | if (do_swap_account && swapout && memcg) | |
a3b2d692 | 4331 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 | 4332 | } |
e767e056 | 4333 | #endif |
8c7c6e34 | 4334 | |
c255a458 | 4335 | #ifdef CONFIG_MEMCG_SWAP |
8c7c6e34 KH |
4336 | /* |
4337 | * called from swap_entry_free(). remove record in swap_cgroup and | |
4338 | * uncharge "memsw" account. | |
4339 | */ | |
4340 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 4341 | { |
8c7c6e34 | 4342 | struct mem_cgroup *memcg; |
a3b2d692 | 4343 | unsigned short id; |
8c7c6e34 KH |
4344 | |
4345 | if (!do_swap_account) | |
4346 | return; | |
4347 | ||
a3b2d692 KH |
4348 | id = swap_cgroup_record(ent, 0); |
4349 | rcu_read_lock(); | |
4350 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 4351 | if (memcg) { |
a3b2d692 KH |
4352 | /* |
4353 | * We uncharge this because swap is freed. | |
4354 | * This memcg can be obsolete one. We avoid calling css_tryget | |
4355 | */ | |
0c3e73e8 | 4356 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 4357 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 4358 | mem_cgroup_swap_statistics(memcg, false); |
4050377b | 4359 | css_put(&memcg->css); |
8c7c6e34 | 4360 | } |
a3b2d692 | 4361 | rcu_read_unlock(); |
d13d1443 | 4362 | } |
02491447 DN |
4363 | |
4364 | /** | |
4365 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
4366 | * @entry: swap entry to be moved | |
4367 | * @from: mem_cgroup which the entry is moved from | |
4368 | * @to: mem_cgroup which the entry is moved to | |
4369 | * | |
4370 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
4371 | * as the mem_cgroup's id of @from. | |
4372 | * | |
4373 | * Returns 0 on success, -EINVAL on failure. | |
4374 | * | |
4375 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
4376 | * both res and memsw, and called css_get(). | |
4377 | */ | |
4378 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 4379 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
4380 | { |
4381 | unsigned short old_id, new_id; | |
4382 | ||
4383 | old_id = css_id(&from->css); | |
4384 | new_id = css_id(&to->css); | |
4385 | ||
4386 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 4387 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 4388 | mem_cgroup_swap_statistics(to, true); |
02491447 | 4389 | /* |
483c30b5 DN |
4390 | * This function is only called from task migration context now. |
4391 | * It postpones res_counter and refcount handling till the end | |
4392 | * of task migration(mem_cgroup_clear_mc()) for performance | |
4050377b LZ |
4393 | * improvement. But we cannot postpone css_get(to) because if |
4394 | * the process that has been moved to @to does swap-in, the | |
4395 | * refcount of @to might be decreased to 0. | |
4396 | * | |
4397 | * We are in attach() phase, so the cgroup is guaranteed to be | |
4398 | * alive, so we can just call css_get(). | |
02491447 | 4399 | */ |
4050377b | 4400 | css_get(&to->css); |
02491447 DN |
4401 | return 0; |
4402 | } | |
4403 | return -EINVAL; | |
4404 | } | |
4405 | #else | |
4406 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 4407 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
4408 | { |
4409 | return -EINVAL; | |
4410 | } | |
8c7c6e34 | 4411 | #endif |
d13d1443 | 4412 | |
ae41be37 | 4413 | /* |
01b1ae63 KH |
4414 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
4415 | * page belongs to. | |
ae41be37 | 4416 | */ |
0030f535 JW |
4417 | void mem_cgroup_prepare_migration(struct page *page, struct page *newpage, |
4418 | struct mem_cgroup **memcgp) | |
ae41be37 | 4419 | { |
c0ff4b85 | 4420 | struct mem_cgroup *memcg = NULL; |
b32967ff | 4421 | unsigned int nr_pages = 1; |
7ec99d62 | 4422 | struct page_cgroup *pc; |
ac39cf8c | 4423 | enum charge_type ctype; |
8869b8f6 | 4424 | |
72835c86 | 4425 | *memcgp = NULL; |
56039efa | 4426 | |
f8d66542 | 4427 | if (mem_cgroup_disabled()) |
0030f535 | 4428 | return; |
4077960e | 4429 | |
b32967ff MG |
4430 | if (PageTransHuge(page)) |
4431 | nr_pages <<= compound_order(page); | |
4432 | ||
52d4b9ac KH |
4433 | pc = lookup_page_cgroup(page); |
4434 | lock_page_cgroup(pc); | |
4435 | if (PageCgroupUsed(pc)) { | |
c0ff4b85 R |
4436 | memcg = pc->mem_cgroup; |
4437 | css_get(&memcg->css); | |
ac39cf8c | 4438 | /* |
4439 | * At migrating an anonymous page, its mapcount goes down | |
4440 | * to 0 and uncharge() will be called. But, even if it's fully | |
4441 | * unmapped, migration may fail and this page has to be | |
4442 | * charged again. We set MIGRATION flag here and delay uncharge | |
4443 | * until end_migration() is called | |
4444 | * | |
4445 | * Corner Case Thinking | |
4446 | * A) | |
4447 | * When the old page was mapped as Anon and it's unmap-and-freed | |
4448 | * while migration was ongoing. | |
4449 | * If unmap finds the old page, uncharge() of it will be delayed | |
4450 | * until end_migration(). If unmap finds a new page, it's | |
4451 | * uncharged when it make mapcount to be 1->0. If unmap code | |
4452 | * finds swap_migration_entry, the new page will not be mapped | |
4453 | * and end_migration() will find it(mapcount==0). | |
4454 | * | |
4455 | * B) | |
4456 | * When the old page was mapped but migraion fails, the kernel | |
4457 | * remaps it. A charge for it is kept by MIGRATION flag even | |
4458 | * if mapcount goes down to 0. We can do remap successfully | |
4459 | * without charging it again. | |
4460 | * | |
4461 | * C) | |
4462 | * The "old" page is under lock_page() until the end of | |
4463 | * migration, so, the old page itself will not be swapped-out. | |
4464 | * If the new page is swapped out before end_migraton, our | |
4465 | * hook to usual swap-out path will catch the event. | |
4466 | */ | |
4467 | if (PageAnon(page)) | |
4468 | SetPageCgroupMigration(pc); | |
e8589cc1 | 4469 | } |
52d4b9ac | 4470 | unlock_page_cgroup(pc); |
ac39cf8c | 4471 | /* |
4472 | * If the page is not charged at this point, | |
4473 | * we return here. | |
4474 | */ | |
c0ff4b85 | 4475 | if (!memcg) |
0030f535 | 4476 | return; |
01b1ae63 | 4477 | |
72835c86 | 4478 | *memcgp = memcg; |
ac39cf8c | 4479 | /* |
4480 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
4481 | * is called before end_migration, we can catch all events on this new | |
4482 | * page. In the case new page is migrated but not remapped, new page's | |
4483 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
4484 | */ | |
ac39cf8c | 4485 | if (PageAnon(page)) |
41326c17 | 4486 | ctype = MEM_CGROUP_CHARGE_TYPE_ANON; |
ac39cf8c | 4487 | else |
62ba7442 | 4488 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; |
0030f535 JW |
4489 | /* |
4490 | * The page is committed to the memcg, but it's not actually | |
4491 | * charged to the res_counter since we plan on replacing the | |
4492 | * old one and only one page is going to be left afterwards. | |
4493 | */ | |
b32967ff | 4494 | __mem_cgroup_commit_charge(memcg, newpage, nr_pages, ctype, false); |
ae41be37 | 4495 | } |
8869b8f6 | 4496 | |
69029cd5 | 4497 | /* remove redundant charge if migration failed*/ |
c0ff4b85 | 4498 | void mem_cgroup_end_migration(struct mem_cgroup *memcg, |
50de1dd9 | 4499 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 4500 | { |
ac39cf8c | 4501 | struct page *used, *unused; |
01b1ae63 | 4502 | struct page_cgroup *pc; |
b2402857 | 4503 | bool anon; |
01b1ae63 | 4504 | |
c0ff4b85 | 4505 | if (!memcg) |
01b1ae63 | 4506 | return; |
b25ed609 | 4507 | |
50de1dd9 | 4508 | if (!migration_ok) { |
ac39cf8c | 4509 | used = oldpage; |
4510 | unused = newpage; | |
01b1ae63 | 4511 | } else { |
ac39cf8c | 4512 | used = newpage; |
01b1ae63 KH |
4513 | unused = oldpage; |
4514 | } | |
0030f535 | 4515 | anon = PageAnon(used); |
7d188958 JW |
4516 | __mem_cgroup_uncharge_common(unused, |
4517 | anon ? MEM_CGROUP_CHARGE_TYPE_ANON | |
4518 | : MEM_CGROUP_CHARGE_TYPE_CACHE, | |
4519 | true); | |
0030f535 | 4520 | css_put(&memcg->css); |
69029cd5 | 4521 | /* |
ac39cf8c | 4522 | * We disallowed uncharge of pages under migration because mapcount |
4523 | * of the page goes down to zero, temporarly. | |
4524 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 4525 | */ |
ac39cf8c | 4526 | pc = lookup_page_cgroup(oldpage); |
4527 | lock_page_cgroup(pc); | |
4528 | ClearPageCgroupMigration(pc); | |
4529 | unlock_page_cgroup(pc); | |
ac39cf8c | 4530 | |
01b1ae63 | 4531 | /* |
ac39cf8c | 4532 | * If a page is a file cache, radix-tree replacement is very atomic |
4533 | * and we can skip this check. When it was an Anon page, its mapcount | |
4534 | * goes down to 0. But because we added MIGRATION flage, it's not | |
4535 | * uncharged yet. There are several case but page->mapcount check | |
4536 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
4537 | * check. (see prepare_charge() also) | |
69029cd5 | 4538 | */ |
b2402857 | 4539 | if (anon) |
ac39cf8c | 4540 | mem_cgroup_uncharge_page(used); |
ae41be37 | 4541 | } |
78fb7466 | 4542 | |
ab936cbc KH |
4543 | /* |
4544 | * At replace page cache, newpage is not under any memcg but it's on | |
4545 | * LRU. So, this function doesn't touch res_counter but handles LRU | |
4546 | * in correct way. Both pages are locked so we cannot race with uncharge. | |
4547 | */ | |
4548 | void mem_cgroup_replace_page_cache(struct page *oldpage, | |
4549 | struct page *newpage) | |
4550 | { | |
bde05d1c | 4551 | struct mem_cgroup *memcg = NULL; |
ab936cbc | 4552 | struct page_cgroup *pc; |
ab936cbc | 4553 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
ab936cbc KH |
4554 | |
4555 | if (mem_cgroup_disabled()) | |
4556 | return; | |
4557 | ||
4558 | pc = lookup_page_cgroup(oldpage); | |
4559 | /* fix accounting on old pages */ | |
4560 | lock_page_cgroup(pc); | |
bde05d1c HD |
4561 | if (PageCgroupUsed(pc)) { |
4562 | memcg = pc->mem_cgroup; | |
b070e65c | 4563 | mem_cgroup_charge_statistics(memcg, oldpage, false, -1); |
bde05d1c HD |
4564 | ClearPageCgroupUsed(pc); |
4565 | } | |
ab936cbc KH |
4566 | unlock_page_cgroup(pc); |
4567 | ||
bde05d1c HD |
4568 | /* |
4569 | * When called from shmem_replace_page(), in some cases the | |
4570 | * oldpage has already been charged, and in some cases not. | |
4571 | */ | |
4572 | if (!memcg) | |
4573 | return; | |
ab936cbc KH |
4574 | /* |
4575 | * Even if newpage->mapping was NULL before starting replacement, | |
4576 | * the newpage may be on LRU(or pagevec for LRU) already. We lock | |
4577 | * LRU while we overwrite pc->mem_cgroup. | |
4578 | */ | |
ce587e65 | 4579 | __mem_cgroup_commit_charge(memcg, newpage, 1, type, true); |
ab936cbc KH |
4580 | } |
4581 | ||
f212ad7c DN |
4582 | #ifdef CONFIG_DEBUG_VM |
4583 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
4584 | { | |
4585 | struct page_cgroup *pc; | |
4586 | ||
4587 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
4588 | /* |
4589 | * Can be NULL while feeding pages into the page allocator for | |
4590 | * the first time, i.e. during boot or memory hotplug; | |
4591 | * or when mem_cgroup_disabled(). | |
4592 | */ | |
f212ad7c DN |
4593 | if (likely(pc) && PageCgroupUsed(pc)) |
4594 | return pc; | |
4595 | return NULL; | |
4596 | } | |
4597 | ||
4598 | bool mem_cgroup_bad_page_check(struct page *page) | |
4599 | { | |
4600 | if (mem_cgroup_disabled()) | |
4601 | return false; | |
4602 | ||
4603 | return lookup_page_cgroup_used(page) != NULL; | |
4604 | } | |
4605 | ||
4606 | void mem_cgroup_print_bad_page(struct page *page) | |
4607 | { | |
4608 | struct page_cgroup *pc; | |
4609 | ||
4610 | pc = lookup_page_cgroup_used(page); | |
4611 | if (pc) { | |
d045197f AM |
4612 | pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n", |
4613 | pc, pc->flags, pc->mem_cgroup); | |
f212ad7c DN |
4614 | } |
4615 | } | |
4616 | #endif | |
4617 | ||
d38d2a75 | 4618 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 4619 | unsigned long long val) |
628f4235 | 4620 | { |
81d39c20 | 4621 | int retry_count; |
3c11ecf4 | 4622 | u64 memswlimit, memlimit; |
628f4235 | 4623 | int ret = 0; |
81d39c20 KH |
4624 | int children = mem_cgroup_count_children(memcg); |
4625 | u64 curusage, oldusage; | |
3c11ecf4 | 4626 | int enlarge; |
81d39c20 KH |
4627 | |
4628 | /* | |
4629 | * For keeping hierarchical_reclaim simple, how long we should retry | |
4630 | * is depends on callers. We set our retry-count to be function | |
4631 | * of # of children which we should visit in this loop. | |
4632 | */ | |
4633 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
4634 | ||
4635 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 4636 | |
3c11ecf4 | 4637 | enlarge = 0; |
8c7c6e34 | 4638 | while (retry_count) { |
628f4235 KH |
4639 | if (signal_pending(current)) { |
4640 | ret = -EINTR; | |
4641 | break; | |
4642 | } | |
8c7c6e34 KH |
4643 | /* |
4644 | * Rather than hide all in some function, I do this in | |
4645 | * open coded manner. You see what this really does. | |
aaad153e | 4646 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
4647 | */ |
4648 | mutex_lock(&set_limit_mutex); | |
4649 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
4650 | if (memswlimit < val) { | |
4651 | ret = -EINVAL; | |
4652 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
4653 | break; |
4654 | } | |
3c11ecf4 KH |
4655 | |
4656 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4657 | if (memlimit < val) | |
4658 | enlarge = 1; | |
4659 | ||
8c7c6e34 | 4660 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
4661 | if (!ret) { |
4662 | if (memswlimit == val) | |
4663 | memcg->memsw_is_minimum = true; | |
4664 | else | |
4665 | memcg->memsw_is_minimum = false; | |
4666 | } | |
8c7c6e34 KH |
4667 | mutex_unlock(&set_limit_mutex); |
4668 | ||
4669 | if (!ret) | |
4670 | break; | |
4671 | ||
5660048c JW |
4672 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
4673 | MEM_CGROUP_RECLAIM_SHRINK); | |
81d39c20 KH |
4674 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
4675 | /* Usage is reduced ? */ | |
4676 | if (curusage >= oldusage) | |
4677 | retry_count--; | |
4678 | else | |
4679 | oldusage = curusage; | |
8c7c6e34 | 4680 | } |
3c11ecf4 KH |
4681 | if (!ret && enlarge) |
4682 | memcg_oom_recover(memcg); | |
14797e23 | 4683 | |
8c7c6e34 KH |
4684 | return ret; |
4685 | } | |
4686 | ||
338c8431 LZ |
4687 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
4688 | unsigned long long val) | |
8c7c6e34 | 4689 | { |
81d39c20 | 4690 | int retry_count; |
3c11ecf4 | 4691 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
4692 | int children = mem_cgroup_count_children(memcg); |
4693 | int ret = -EBUSY; | |
3c11ecf4 | 4694 | int enlarge = 0; |
8c7c6e34 | 4695 | |
81d39c20 KH |
4696 | /* see mem_cgroup_resize_res_limit */ |
4697 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
4698 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
4699 | while (retry_count) { |
4700 | if (signal_pending(current)) { | |
4701 | ret = -EINTR; | |
4702 | break; | |
4703 | } | |
4704 | /* | |
4705 | * Rather than hide all in some function, I do this in | |
4706 | * open coded manner. You see what this really does. | |
aaad153e | 4707 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
4708 | */ |
4709 | mutex_lock(&set_limit_mutex); | |
4710 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4711 | if (memlimit > val) { | |
4712 | ret = -EINVAL; | |
4713 | mutex_unlock(&set_limit_mutex); | |
4714 | break; | |
4715 | } | |
3c11ecf4 KH |
4716 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
4717 | if (memswlimit < val) | |
4718 | enlarge = 1; | |
8c7c6e34 | 4719 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
4720 | if (!ret) { |
4721 | if (memlimit == val) | |
4722 | memcg->memsw_is_minimum = true; | |
4723 | else | |
4724 | memcg->memsw_is_minimum = false; | |
4725 | } | |
8c7c6e34 KH |
4726 | mutex_unlock(&set_limit_mutex); |
4727 | ||
4728 | if (!ret) | |
4729 | break; | |
4730 | ||
5660048c JW |
4731 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
4732 | MEM_CGROUP_RECLAIM_NOSWAP | | |
4733 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 4734 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 4735 | /* Usage is reduced ? */ |
8c7c6e34 | 4736 | if (curusage >= oldusage) |
628f4235 | 4737 | retry_count--; |
81d39c20 KH |
4738 | else |
4739 | oldusage = curusage; | |
628f4235 | 4740 | } |
3c11ecf4 KH |
4741 | if (!ret && enlarge) |
4742 | memcg_oom_recover(memcg); | |
628f4235 KH |
4743 | return ret; |
4744 | } | |
4745 | ||
4e416953 | 4746 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
0ae5e89c YH |
4747 | gfp_t gfp_mask, |
4748 | unsigned long *total_scanned) | |
4e416953 BS |
4749 | { |
4750 | unsigned long nr_reclaimed = 0; | |
4751 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
4752 | unsigned long reclaimed; | |
4753 | int loop = 0; | |
4754 | struct mem_cgroup_tree_per_zone *mctz; | |
ef8745c1 | 4755 | unsigned long long excess; |
0ae5e89c | 4756 | unsigned long nr_scanned; |
4e416953 BS |
4757 | |
4758 | if (order > 0) | |
4759 | return 0; | |
4760 | ||
00918b6a | 4761 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); |
4e416953 BS |
4762 | /* |
4763 | * This loop can run a while, specially if mem_cgroup's continuously | |
4764 | * keep exceeding their soft limit and putting the system under | |
4765 | * pressure | |
4766 | */ | |
4767 | do { | |
4768 | if (next_mz) | |
4769 | mz = next_mz; | |
4770 | else | |
4771 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
4772 | if (!mz) | |
4773 | break; | |
4774 | ||
0ae5e89c | 4775 | nr_scanned = 0; |
d79154bb | 4776 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, |
5660048c | 4777 | gfp_mask, &nr_scanned); |
4e416953 | 4778 | nr_reclaimed += reclaimed; |
0ae5e89c | 4779 | *total_scanned += nr_scanned; |
4e416953 BS |
4780 | spin_lock(&mctz->lock); |
4781 | ||
4782 | /* | |
4783 | * If we failed to reclaim anything from this memory cgroup | |
4784 | * it is time to move on to the next cgroup | |
4785 | */ | |
4786 | next_mz = NULL; | |
4787 | if (!reclaimed) { | |
4788 | do { | |
4789 | /* | |
4790 | * Loop until we find yet another one. | |
4791 | * | |
4792 | * By the time we get the soft_limit lock | |
4793 | * again, someone might have aded the | |
4794 | * group back on the RB tree. Iterate to | |
4795 | * make sure we get a different mem. | |
4796 | * mem_cgroup_largest_soft_limit_node returns | |
4797 | * NULL if no other cgroup is present on | |
4798 | * the tree | |
4799 | */ | |
4800 | next_mz = | |
4801 | __mem_cgroup_largest_soft_limit_node(mctz); | |
39cc98f1 | 4802 | if (next_mz == mz) |
d79154bb | 4803 | css_put(&next_mz->memcg->css); |
39cc98f1 | 4804 | else /* next_mz == NULL or other memcg */ |
4e416953 BS |
4805 | break; |
4806 | } while (1); | |
4807 | } | |
d79154bb HD |
4808 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); |
4809 | excess = res_counter_soft_limit_excess(&mz->memcg->res); | |
4e416953 BS |
4810 | /* |
4811 | * One school of thought says that we should not add | |
4812 | * back the node to the tree if reclaim returns 0. | |
4813 | * But our reclaim could return 0, simply because due | |
4814 | * to priority we are exposing a smaller subset of | |
4815 | * memory to reclaim from. Consider this as a longer | |
4816 | * term TODO. | |
4817 | */ | |
ef8745c1 | 4818 | /* If excess == 0, no tree ops */ |
d79154bb | 4819 | __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess); |
4e416953 | 4820 | spin_unlock(&mctz->lock); |
d79154bb | 4821 | css_put(&mz->memcg->css); |
4e416953 BS |
4822 | loop++; |
4823 | /* | |
4824 | * Could not reclaim anything and there are no more | |
4825 | * mem cgroups to try or we seem to be looping without | |
4826 | * reclaiming anything. | |
4827 | */ | |
4828 | if (!nr_reclaimed && | |
4829 | (next_mz == NULL || | |
4830 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
4831 | break; | |
4832 | } while (!nr_reclaimed); | |
4833 | if (next_mz) | |
d79154bb | 4834 | css_put(&next_mz->memcg->css); |
4e416953 BS |
4835 | return nr_reclaimed; |
4836 | } | |
4837 | ||
2ef37d3f MH |
4838 | /** |
4839 | * mem_cgroup_force_empty_list - clears LRU of a group | |
4840 | * @memcg: group to clear | |
4841 | * @node: NUMA node | |
4842 | * @zid: zone id | |
4843 | * @lru: lru to to clear | |
4844 | * | |
3c935d18 | 4845 | * Traverse a specified page_cgroup list and try to drop them all. This doesn't |
2ef37d3f MH |
4846 | * reclaim the pages page themselves - pages are moved to the parent (or root) |
4847 | * group. | |
cc847582 | 4848 | */ |
2ef37d3f | 4849 | static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
08e552c6 | 4850 | int node, int zid, enum lru_list lru) |
cc847582 | 4851 | { |
bea8c150 | 4852 | struct lruvec *lruvec; |
2ef37d3f | 4853 | unsigned long flags; |
072c56c1 | 4854 | struct list_head *list; |
925b7673 JW |
4855 | struct page *busy; |
4856 | struct zone *zone; | |
072c56c1 | 4857 | |
08e552c6 | 4858 | zone = &NODE_DATA(node)->node_zones[zid]; |
bea8c150 HD |
4859 | lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
4860 | list = &lruvec->lists[lru]; | |
cc847582 | 4861 | |
f817ed48 | 4862 | busy = NULL; |
2ef37d3f | 4863 | do { |
925b7673 | 4864 | struct page_cgroup *pc; |
5564e88b JW |
4865 | struct page *page; |
4866 | ||
08e552c6 | 4867 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 4868 | if (list_empty(list)) { |
08e552c6 | 4869 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 4870 | break; |
f817ed48 | 4871 | } |
925b7673 JW |
4872 | page = list_entry(list->prev, struct page, lru); |
4873 | if (busy == page) { | |
4874 | list_move(&page->lru, list); | |
648bcc77 | 4875 | busy = NULL; |
08e552c6 | 4876 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
4877 | continue; |
4878 | } | |
08e552c6 | 4879 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 4880 | |
925b7673 | 4881 | pc = lookup_page_cgroup(page); |
5564e88b | 4882 | |
3c935d18 | 4883 | if (mem_cgroup_move_parent(page, pc, memcg)) { |
f817ed48 | 4884 | /* found lock contention or "pc" is obsolete. */ |
925b7673 | 4885 | busy = page; |
f817ed48 KH |
4886 | cond_resched(); |
4887 | } else | |
4888 | busy = NULL; | |
2ef37d3f | 4889 | } while (!list_empty(list)); |
cc847582 KH |
4890 | } |
4891 | ||
4892 | /* | |
c26251f9 MH |
4893 | * make mem_cgroup's charge to be 0 if there is no task by moving |
4894 | * all the charges and pages to the parent. | |
cc847582 | 4895 | * This enables deleting this mem_cgroup. |
c26251f9 MH |
4896 | * |
4897 | * Caller is responsible for holding css reference on the memcg. | |
cc847582 | 4898 | */ |
ab5196c2 | 4899 | static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg) |
cc847582 | 4900 | { |
c26251f9 | 4901 | int node, zid; |
bea207c8 | 4902 | u64 usage; |
f817ed48 | 4903 | |
fce66477 | 4904 | do { |
52d4b9ac KH |
4905 | /* This is for making all *used* pages to be on LRU. */ |
4906 | lru_add_drain_all(); | |
c0ff4b85 | 4907 | drain_all_stock_sync(memcg); |
c0ff4b85 | 4908 | mem_cgroup_start_move(memcg); |
31aaea4a | 4909 | for_each_node_state(node, N_MEMORY) { |
2ef37d3f | 4910 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
f156ab93 HD |
4911 | enum lru_list lru; |
4912 | for_each_lru(lru) { | |
2ef37d3f | 4913 | mem_cgroup_force_empty_list(memcg, |
f156ab93 | 4914 | node, zid, lru); |
f817ed48 | 4915 | } |
1ecaab2b | 4916 | } |
f817ed48 | 4917 | } |
c0ff4b85 R |
4918 | mem_cgroup_end_move(memcg); |
4919 | memcg_oom_recover(memcg); | |
52d4b9ac | 4920 | cond_resched(); |
f817ed48 | 4921 | |
2ef37d3f | 4922 | /* |
bea207c8 GC |
4923 | * Kernel memory may not necessarily be trackable to a specific |
4924 | * process. So they are not migrated, and therefore we can't | |
4925 | * expect their value to drop to 0 here. | |
4926 | * Having res filled up with kmem only is enough. | |
4927 | * | |
2ef37d3f MH |
4928 | * This is a safety check because mem_cgroup_force_empty_list |
4929 | * could have raced with mem_cgroup_replace_page_cache callers | |
4930 | * so the lru seemed empty but the page could have been added | |
4931 | * right after the check. RES_USAGE should be safe as we always | |
4932 | * charge before adding to the LRU. | |
4933 | */ | |
bea207c8 GC |
4934 | usage = res_counter_read_u64(&memcg->res, RES_USAGE) - |
4935 | res_counter_read_u64(&memcg->kmem, RES_USAGE); | |
4936 | } while (usage > 0); | |
c26251f9 MH |
4937 | } |
4938 | ||
b5f99b53 GC |
4939 | /* |
4940 | * This mainly exists for tests during the setting of set of use_hierarchy. | |
4941 | * Since this is the very setting we are changing, the current hierarchy value | |
4942 | * is meaningless | |
4943 | */ | |
4944 | static inline bool __memcg_has_children(struct mem_cgroup *memcg) | |
4945 | { | |
4946 | struct cgroup *pos; | |
4947 | ||
4948 | /* bounce at first found */ | |
4949 | cgroup_for_each_child(pos, memcg->css.cgroup) | |
4950 | return true; | |
4951 | return false; | |
4952 | } | |
4953 | ||
4954 | /* | |
0999821b GC |
4955 | * Must be called with memcg_create_mutex held, unless the cgroup is guaranteed |
4956 | * to be already dead (as in mem_cgroup_force_empty, for instance). This is | |
b5f99b53 GC |
4957 | * from mem_cgroup_count_children(), in the sense that we don't really care how |
4958 | * many children we have; we only need to know if we have any. It also counts | |
4959 | * any memcg without hierarchy as infertile. | |
4960 | */ | |
4961 | static inline bool memcg_has_children(struct mem_cgroup *memcg) | |
4962 | { | |
4963 | return memcg->use_hierarchy && __memcg_has_children(memcg); | |
4964 | } | |
4965 | ||
c26251f9 MH |
4966 | /* |
4967 | * Reclaims as many pages from the given memcg as possible and moves | |
4968 | * the rest to the parent. | |
4969 | * | |
4970 | * Caller is responsible for holding css reference for memcg. | |
4971 | */ | |
4972 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
4973 | { | |
4974 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
4975 | struct cgroup *cgrp = memcg->css.cgroup; | |
f817ed48 | 4976 | |
c1e862c1 | 4977 | /* returns EBUSY if there is a task or if we come here twice. */ |
c26251f9 MH |
4978 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
4979 | return -EBUSY; | |
4980 | ||
c1e862c1 KH |
4981 | /* we call try-to-free pages for make this cgroup empty */ |
4982 | lru_add_drain_all(); | |
f817ed48 | 4983 | /* try to free all pages in this cgroup */ |
569530fb | 4984 | while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) { |
f817ed48 | 4985 | int progress; |
c1e862c1 | 4986 | |
c26251f9 MH |
4987 | if (signal_pending(current)) |
4988 | return -EINTR; | |
4989 | ||
c0ff4b85 | 4990 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, |
185efc0f | 4991 | false); |
c1e862c1 | 4992 | if (!progress) { |
f817ed48 | 4993 | nr_retries--; |
c1e862c1 | 4994 | /* maybe some writeback is necessary */ |
8aa7e847 | 4995 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 4996 | } |
f817ed48 KH |
4997 | |
4998 | } | |
08e552c6 | 4999 | lru_add_drain(); |
ab5196c2 MH |
5000 | mem_cgroup_reparent_charges(memcg); |
5001 | ||
5002 | return 0; | |
cc847582 KH |
5003 | } |
5004 | ||
6bbda35c | 5005 | static int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
c1e862c1 | 5006 | { |
c26251f9 MH |
5007 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
5008 | int ret; | |
5009 | ||
d8423011 MH |
5010 | if (mem_cgroup_is_root(memcg)) |
5011 | return -EINVAL; | |
c26251f9 MH |
5012 | css_get(&memcg->css); |
5013 | ret = mem_cgroup_force_empty(memcg); | |
5014 | css_put(&memcg->css); | |
5015 | ||
5016 | return ret; | |
c1e862c1 KH |
5017 | } |
5018 | ||
5019 | ||
18f59ea7 BS |
5020 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
5021 | { | |
5022 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
5023 | } | |
5024 | ||
5025 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
5026 | u64 val) | |
5027 | { | |
5028 | int retval = 0; | |
c0ff4b85 | 5029 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
18f59ea7 | 5030 | struct cgroup *parent = cont->parent; |
c0ff4b85 | 5031 | struct mem_cgroup *parent_memcg = NULL; |
18f59ea7 BS |
5032 | |
5033 | if (parent) | |
c0ff4b85 | 5034 | parent_memcg = mem_cgroup_from_cont(parent); |
18f59ea7 | 5035 | |
0999821b | 5036 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
5037 | |
5038 | if (memcg->use_hierarchy == val) | |
5039 | goto out; | |
5040 | ||
18f59ea7 | 5041 | /* |
af901ca1 | 5042 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
5043 | * in the child subtrees. If it is unset, then the change can |
5044 | * occur, provided the current cgroup has no children. | |
5045 | * | |
5046 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
5047 | * set if there are no children. | |
5048 | */ | |
c0ff4b85 | 5049 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 5050 | (val == 1 || val == 0)) { |
b5f99b53 | 5051 | if (!__memcg_has_children(memcg)) |
c0ff4b85 | 5052 | memcg->use_hierarchy = val; |
18f59ea7 BS |
5053 | else |
5054 | retval = -EBUSY; | |
5055 | } else | |
5056 | retval = -EINVAL; | |
567fb435 GC |
5057 | |
5058 | out: | |
0999821b | 5059 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
5060 | |
5061 | return retval; | |
5062 | } | |
5063 | ||
0c3e73e8 | 5064 | |
c0ff4b85 | 5065 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg, |
7a159cc9 | 5066 | enum mem_cgroup_stat_index idx) |
0c3e73e8 | 5067 | { |
7d74b06f | 5068 | struct mem_cgroup *iter; |
7a159cc9 | 5069 | long val = 0; |
0c3e73e8 | 5070 | |
7a159cc9 | 5071 | /* Per-cpu values can be negative, use a signed accumulator */ |
c0ff4b85 | 5072 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f KH |
5073 | val += mem_cgroup_read_stat(iter, idx); |
5074 | ||
5075 | if (val < 0) /* race ? */ | |
5076 | val = 0; | |
5077 | return val; | |
0c3e73e8 BS |
5078 | } |
5079 | ||
c0ff4b85 | 5080 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
104f3928 | 5081 | { |
7d74b06f | 5082 | u64 val; |
104f3928 | 5083 | |
c0ff4b85 | 5084 | if (!mem_cgroup_is_root(memcg)) { |
104f3928 | 5085 | if (!swap) |
65c64ce8 | 5086 | return res_counter_read_u64(&memcg->res, RES_USAGE); |
104f3928 | 5087 | else |
65c64ce8 | 5088 | return res_counter_read_u64(&memcg->memsw, RES_USAGE); |
104f3928 KS |
5089 | } |
5090 | ||
b070e65c DR |
5091 | /* |
5092 | * Transparent hugepages are still accounted for in MEM_CGROUP_STAT_RSS | |
5093 | * as well as in MEM_CGROUP_STAT_RSS_HUGE. | |
5094 | */ | |
c0ff4b85 R |
5095 | val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE); |
5096 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS); | |
104f3928 | 5097 | |
7d74b06f | 5098 | if (swap) |
bff6bb83 | 5099 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP); |
104f3928 KS |
5100 | |
5101 | return val << PAGE_SHIFT; | |
5102 | } | |
5103 | ||
af36f906 TH |
5104 | static ssize_t mem_cgroup_read(struct cgroup *cont, struct cftype *cft, |
5105 | struct file *file, char __user *buf, | |
5106 | size_t nbytes, loff_t *ppos) | |
8cdea7c0 | 5107 | { |
c0ff4b85 | 5108 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
af36f906 | 5109 | char str[64]; |
104f3928 | 5110 | u64 val; |
86ae53e1 GC |
5111 | int name, len; |
5112 | enum res_type type; | |
8c7c6e34 KH |
5113 | |
5114 | type = MEMFILE_TYPE(cft->private); | |
5115 | name = MEMFILE_ATTR(cft->private); | |
af36f906 | 5116 | |
8c7c6e34 KH |
5117 | switch (type) { |
5118 | case _MEM: | |
104f3928 | 5119 | if (name == RES_USAGE) |
c0ff4b85 | 5120 | val = mem_cgroup_usage(memcg, false); |
104f3928 | 5121 | else |
c0ff4b85 | 5122 | val = res_counter_read_u64(&memcg->res, name); |
8c7c6e34 KH |
5123 | break; |
5124 | case _MEMSWAP: | |
104f3928 | 5125 | if (name == RES_USAGE) |
c0ff4b85 | 5126 | val = mem_cgroup_usage(memcg, true); |
104f3928 | 5127 | else |
c0ff4b85 | 5128 | val = res_counter_read_u64(&memcg->memsw, name); |
8c7c6e34 | 5129 | break; |
510fc4e1 GC |
5130 | case _KMEM: |
5131 | val = res_counter_read_u64(&memcg->kmem, name); | |
5132 | break; | |
8c7c6e34 KH |
5133 | default: |
5134 | BUG(); | |
8c7c6e34 | 5135 | } |
af36f906 TH |
5136 | |
5137 | len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val); | |
5138 | return simple_read_from_buffer(buf, nbytes, ppos, str, len); | |
8cdea7c0 | 5139 | } |
510fc4e1 GC |
5140 | |
5141 | static int memcg_update_kmem_limit(struct cgroup *cont, u64 val) | |
5142 | { | |
5143 | int ret = -EINVAL; | |
5144 | #ifdef CONFIG_MEMCG_KMEM | |
5145 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); | |
5146 | /* | |
5147 | * For simplicity, we won't allow this to be disabled. It also can't | |
5148 | * be changed if the cgroup has children already, or if tasks had | |
5149 | * already joined. | |
5150 | * | |
5151 | * If tasks join before we set the limit, a person looking at | |
5152 | * kmem.usage_in_bytes will have no way to determine when it took | |
5153 | * place, which makes the value quite meaningless. | |
5154 | * | |
5155 | * After it first became limited, changes in the value of the limit are | |
5156 | * of course permitted. | |
510fc4e1 | 5157 | */ |
0999821b | 5158 | mutex_lock(&memcg_create_mutex); |
510fc4e1 GC |
5159 | mutex_lock(&set_limit_mutex); |
5160 | if (!memcg->kmem_account_flags && val != RESOURCE_MAX) { | |
b5f99b53 | 5161 | if (cgroup_task_count(cont) || memcg_has_children(memcg)) { |
510fc4e1 GC |
5162 | ret = -EBUSY; |
5163 | goto out; | |
5164 | } | |
5165 | ret = res_counter_set_limit(&memcg->kmem, val); | |
5166 | VM_BUG_ON(ret); | |
5167 | ||
55007d84 GC |
5168 | ret = memcg_update_cache_sizes(memcg); |
5169 | if (ret) { | |
5170 | res_counter_set_limit(&memcg->kmem, RESOURCE_MAX); | |
5171 | goto out; | |
5172 | } | |
692e89ab GC |
5173 | static_key_slow_inc(&memcg_kmem_enabled_key); |
5174 | /* | |
5175 | * setting the active bit after the inc will guarantee no one | |
5176 | * starts accounting before all call sites are patched | |
5177 | */ | |
5178 | memcg_kmem_set_active(memcg); | |
510fc4e1 GC |
5179 | } else |
5180 | ret = res_counter_set_limit(&memcg->kmem, val); | |
5181 | out: | |
5182 | mutex_unlock(&set_limit_mutex); | |
0999821b | 5183 | mutex_unlock(&memcg_create_mutex); |
510fc4e1 GC |
5184 | #endif |
5185 | return ret; | |
5186 | } | |
5187 | ||
6d043990 | 5188 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 | 5189 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 5190 | { |
55007d84 | 5191 | int ret = 0; |
510fc4e1 GC |
5192 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
5193 | if (!parent) | |
55007d84 GC |
5194 | goto out; |
5195 | ||
510fc4e1 | 5196 | memcg->kmem_account_flags = parent->kmem_account_flags; |
a8964b9b GC |
5197 | /* |
5198 | * When that happen, we need to disable the static branch only on those | |
5199 | * memcgs that enabled it. To achieve this, we would be forced to | |
5200 | * complicate the code by keeping track of which memcgs were the ones | |
5201 | * that actually enabled limits, and which ones got it from its | |
5202 | * parents. | |
5203 | * | |
5204 | * It is a lot simpler just to do static_key_slow_inc() on every child | |
5205 | * that is accounted. | |
5206 | */ | |
55007d84 GC |
5207 | if (!memcg_kmem_is_active(memcg)) |
5208 | goto out; | |
5209 | ||
5210 | /* | |
10d5ebf4 LZ |
5211 | * __mem_cgroup_free() will issue static_key_slow_dec() because this |
5212 | * memcg is active already. If the later initialization fails then the | |
5213 | * cgroup core triggers the cleanup so we do not have to do it here. | |
55007d84 | 5214 | */ |
55007d84 GC |
5215 | static_key_slow_inc(&memcg_kmem_enabled_key); |
5216 | ||
5217 | mutex_lock(&set_limit_mutex); | |
425c598d | 5218 | memcg_stop_kmem_account(); |
55007d84 | 5219 | ret = memcg_update_cache_sizes(memcg); |
425c598d | 5220 | memcg_resume_kmem_account(); |
55007d84 | 5221 | mutex_unlock(&set_limit_mutex); |
55007d84 GC |
5222 | out: |
5223 | return ret; | |
510fc4e1 | 5224 | } |
6d043990 | 5225 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 5226 | |
628f4235 KH |
5227 | /* |
5228 | * The user of this function is... | |
5229 | * RES_LIMIT. | |
5230 | */ | |
856c13aa PM |
5231 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
5232 | const char *buffer) | |
8cdea7c0 | 5233 | { |
628f4235 | 5234 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
86ae53e1 GC |
5235 | enum res_type type; |
5236 | int name; | |
628f4235 KH |
5237 | unsigned long long val; |
5238 | int ret; | |
5239 | ||
8c7c6e34 KH |
5240 | type = MEMFILE_TYPE(cft->private); |
5241 | name = MEMFILE_ATTR(cft->private); | |
af36f906 | 5242 | |
8c7c6e34 | 5243 | switch (name) { |
628f4235 | 5244 | case RES_LIMIT: |
4b3bde4c BS |
5245 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
5246 | ret = -EINVAL; | |
5247 | break; | |
5248 | } | |
628f4235 KH |
5249 | /* This function does all necessary parse...reuse it */ |
5250 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
5251 | if (ret) |
5252 | break; | |
5253 | if (type == _MEM) | |
628f4235 | 5254 | ret = mem_cgroup_resize_limit(memcg, val); |
510fc4e1 | 5255 | else if (type == _MEMSWAP) |
8c7c6e34 | 5256 | ret = mem_cgroup_resize_memsw_limit(memcg, val); |
510fc4e1 GC |
5257 | else if (type == _KMEM) |
5258 | ret = memcg_update_kmem_limit(cont, val); | |
5259 | else | |
5260 | return -EINVAL; | |
628f4235 | 5261 | break; |
296c81d8 BS |
5262 | case RES_SOFT_LIMIT: |
5263 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
5264 | if (ret) | |
5265 | break; | |
5266 | /* | |
5267 | * For memsw, soft limits are hard to implement in terms | |
5268 | * of semantics, for now, we support soft limits for | |
5269 | * control without swap | |
5270 | */ | |
5271 | if (type == _MEM) | |
5272 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
5273 | else | |
5274 | ret = -EINVAL; | |
5275 | break; | |
628f4235 KH |
5276 | default: |
5277 | ret = -EINVAL; /* should be BUG() ? */ | |
5278 | break; | |
5279 | } | |
5280 | return ret; | |
8cdea7c0 BS |
5281 | } |
5282 | ||
fee7b548 KH |
5283 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
5284 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
5285 | { | |
5286 | struct cgroup *cgroup; | |
5287 | unsigned long long min_limit, min_memsw_limit, tmp; | |
5288 | ||
5289 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
5290 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
5291 | cgroup = memcg->css.cgroup; | |
5292 | if (!memcg->use_hierarchy) | |
5293 | goto out; | |
5294 | ||
5295 | while (cgroup->parent) { | |
5296 | cgroup = cgroup->parent; | |
5297 | memcg = mem_cgroup_from_cont(cgroup); | |
5298 | if (!memcg->use_hierarchy) | |
5299 | break; | |
5300 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
5301 | min_limit = min(min_limit, tmp); | |
5302 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
5303 | min_memsw_limit = min(min_memsw_limit, tmp); | |
5304 | } | |
5305 | out: | |
5306 | *mem_limit = min_limit; | |
5307 | *memsw_limit = min_memsw_limit; | |
fee7b548 KH |
5308 | } |
5309 | ||
29f2a4da | 5310 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 | 5311 | { |
af36f906 | 5312 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
86ae53e1 GC |
5313 | int name; |
5314 | enum res_type type; | |
c84872e1 | 5315 | |
8c7c6e34 KH |
5316 | type = MEMFILE_TYPE(event); |
5317 | name = MEMFILE_ATTR(event); | |
af36f906 | 5318 | |
8c7c6e34 | 5319 | switch (name) { |
29f2a4da | 5320 | case RES_MAX_USAGE: |
8c7c6e34 | 5321 | if (type == _MEM) |
c0ff4b85 | 5322 | res_counter_reset_max(&memcg->res); |
510fc4e1 | 5323 | else if (type == _MEMSWAP) |
c0ff4b85 | 5324 | res_counter_reset_max(&memcg->memsw); |
510fc4e1 GC |
5325 | else if (type == _KMEM) |
5326 | res_counter_reset_max(&memcg->kmem); | |
5327 | else | |
5328 | return -EINVAL; | |
29f2a4da PE |
5329 | break; |
5330 | case RES_FAILCNT: | |
8c7c6e34 | 5331 | if (type == _MEM) |
c0ff4b85 | 5332 | res_counter_reset_failcnt(&memcg->res); |
510fc4e1 | 5333 | else if (type == _MEMSWAP) |
c0ff4b85 | 5334 | res_counter_reset_failcnt(&memcg->memsw); |
510fc4e1 GC |
5335 | else if (type == _KMEM) |
5336 | res_counter_reset_failcnt(&memcg->kmem); | |
5337 | else | |
5338 | return -EINVAL; | |
29f2a4da PE |
5339 | break; |
5340 | } | |
f64c3f54 | 5341 | |
85cc59db | 5342 | return 0; |
c84872e1 PE |
5343 | } |
5344 | ||
7dc74be0 DN |
5345 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
5346 | struct cftype *cft) | |
5347 | { | |
5348 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; | |
5349 | } | |
5350 | ||
02491447 | 5351 | #ifdef CONFIG_MMU |
7dc74be0 DN |
5352 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
5353 | struct cftype *cft, u64 val) | |
5354 | { | |
c0ff4b85 | 5355 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
7dc74be0 DN |
5356 | |
5357 | if (val >= (1 << NR_MOVE_TYPE)) | |
5358 | return -EINVAL; | |
ee5e8472 | 5359 | |
7dc74be0 | 5360 | /* |
ee5e8472 GC |
5361 | * No kind of locking is needed in here, because ->can_attach() will |
5362 | * check this value once in the beginning of the process, and then carry | |
5363 | * on with stale data. This means that changes to this value will only | |
5364 | * affect task migrations starting after the change. | |
7dc74be0 | 5365 | */ |
c0ff4b85 | 5366 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
5367 | return 0; |
5368 | } | |
02491447 DN |
5369 | #else |
5370 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, | |
5371 | struct cftype *cft, u64 val) | |
5372 | { | |
5373 | return -ENOSYS; | |
5374 | } | |
5375 | #endif | |
7dc74be0 | 5376 | |
406eb0c9 | 5377 | #ifdef CONFIG_NUMA |
ab215884 | 5378 | static int memcg_numa_stat_show(struct cgroup *cont, struct cftype *cft, |
fada52ca | 5379 | struct seq_file *m) |
406eb0c9 YH |
5380 | { |
5381 | int nid; | |
5382 | unsigned long total_nr, file_nr, anon_nr, unevictable_nr; | |
5383 | unsigned long node_nr; | |
d79154bb | 5384 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
406eb0c9 | 5385 | |
d79154bb | 5386 | total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL); |
406eb0c9 | 5387 | seq_printf(m, "total=%lu", total_nr); |
31aaea4a | 5388 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5389 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL); |
406eb0c9 YH |
5390 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5391 | } | |
5392 | seq_putc(m, '\n'); | |
5393 | ||
d79154bb | 5394 | file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE); |
406eb0c9 | 5395 | seq_printf(m, "file=%lu", file_nr); |
31aaea4a | 5396 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5397 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 5398 | LRU_ALL_FILE); |
406eb0c9 YH |
5399 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5400 | } | |
5401 | seq_putc(m, '\n'); | |
5402 | ||
d79154bb | 5403 | anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON); |
406eb0c9 | 5404 | seq_printf(m, "anon=%lu", anon_nr); |
31aaea4a | 5405 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5406 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 5407 | LRU_ALL_ANON); |
406eb0c9 YH |
5408 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5409 | } | |
5410 | seq_putc(m, '\n'); | |
5411 | ||
d79154bb | 5412 | unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE)); |
406eb0c9 | 5413 | seq_printf(m, "unevictable=%lu", unevictable_nr); |
31aaea4a | 5414 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5415 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 5416 | BIT(LRU_UNEVICTABLE)); |
406eb0c9 YH |
5417 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5418 | } | |
5419 | seq_putc(m, '\n'); | |
5420 | return 0; | |
5421 | } | |
5422 | #endif /* CONFIG_NUMA */ | |
5423 | ||
af7c4b0e JW |
5424 | static inline void mem_cgroup_lru_names_not_uptodate(void) |
5425 | { | |
5426 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); | |
5427 | } | |
5428 | ||
ab215884 | 5429 | static int memcg_stat_show(struct cgroup *cont, struct cftype *cft, |
78ccf5b5 | 5430 | struct seq_file *m) |
d2ceb9b7 | 5431 | { |
d79154bb | 5432 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
af7c4b0e JW |
5433 | struct mem_cgroup *mi; |
5434 | unsigned int i; | |
406eb0c9 | 5435 | |
af7c4b0e | 5436 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 5437 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 5438 | continue; |
af7c4b0e JW |
5439 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
5440 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 5441 | } |
7b854121 | 5442 | |
af7c4b0e JW |
5443 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
5444 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
5445 | mem_cgroup_read_events(memcg, i)); | |
5446 | ||
5447 | for (i = 0; i < NR_LRU_LISTS; i++) | |
5448 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
5449 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
5450 | ||
14067bb3 | 5451 | /* Hierarchical information */ |
fee7b548 KH |
5452 | { |
5453 | unsigned long long limit, memsw_limit; | |
d79154bb | 5454 | memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit); |
78ccf5b5 | 5455 | seq_printf(m, "hierarchical_memory_limit %llu\n", limit); |
fee7b548 | 5456 | if (do_swap_account) |
78ccf5b5 JW |
5457 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
5458 | memsw_limit); | |
fee7b548 | 5459 | } |
7f016ee8 | 5460 | |
af7c4b0e JW |
5461 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
5462 | long long val = 0; | |
5463 | ||
bff6bb83 | 5464 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 5465 | continue; |
af7c4b0e JW |
5466 | for_each_mem_cgroup_tree(mi, memcg) |
5467 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
5468 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
5469 | } | |
5470 | ||
5471 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
5472 | unsigned long long val = 0; | |
5473 | ||
5474 | for_each_mem_cgroup_tree(mi, memcg) | |
5475 | val += mem_cgroup_read_events(mi, i); | |
5476 | seq_printf(m, "total_%s %llu\n", | |
5477 | mem_cgroup_events_names[i], val); | |
5478 | } | |
5479 | ||
5480 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
5481 | unsigned long long val = 0; | |
5482 | ||
5483 | for_each_mem_cgroup_tree(mi, memcg) | |
5484 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
5485 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 5486 | } |
14067bb3 | 5487 | |
7f016ee8 | 5488 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
5489 | { |
5490 | int nid, zid; | |
5491 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 5492 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
5493 | unsigned long recent_rotated[2] = {0, 0}; |
5494 | unsigned long recent_scanned[2] = {0, 0}; | |
5495 | ||
5496 | for_each_online_node(nid) | |
5497 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
d79154bb | 5498 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
89abfab1 | 5499 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 5500 | |
89abfab1 HD |
5501 | recent_rotated[0] += rstat->recent_rotated[0]; |
5502 | recent_rotated[1] += rstat->recent_rotated[1]; | |
5503 | recent_scanned[0] += rstat->recent_scanned[0]; | |
5504 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 5505 | } |
78ccf5b5 JW |
5506 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
5507 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
5508 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
5509 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
5510 | } |
5511 | #endif | |
5512 | ||
d2ceb9b7 KH |
5513 | return 0; |
5514 | } | |
5515 | ||
a7885eb8 KM |
5516 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
5517 | { | |
5518 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
5519 | ||
1f4c025b | 5520 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
5521 | } |
5522 | ||
5523 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
5524 | u64 val) | |
5525 | { | |
5526 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
5527 | struct mem_cgroup *parent; | |
068b38c1 | 5528 | |
a7885eb8 KM |
5529 | if (val > 100) |
5530 | return -EINVAL; | |
5531 | ||
5532 | if (cgrp->parent == NULL) | |
5533 | return -EINVAL; | |
5534 | ||
5535 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 | 5536 | |
0999821b | 5537 | mutex_lock(&memcg_create_mutex); |
068b38c1 | 5538 | |
a7885eb8 | 5539 | /* If under hierarchy, only empty-root can set this value */ |
b5f99b53 | 5540 | if ((parent->use_hierarchy) || memcg_has_children(memcg)) { |
0999821b | 5541 | mutex_unlock(&memcg_create_mutex); |
a7885eb8 | 5542 | return -EINVAL; |
068b38c1 | 5543 | } |
a7885eb8 | 5544 | |
a7885eb8 | 5545 | memcg->swappiness = val; |
a7885eb8 | 5546 | |
0999821b | 5547 | mutex_unlock(&memcg_create_mutex); |
068b38c1 | 5548 | |
a7885eb8 KM |
5549 | return 0; |
5550 | } | |
5551 | ||
2e72b634 KS |
5552 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
5553 | { | |
5554 | struct mem_cgroup_threshold_ary *t; | |
5555 | u64 usage; | |
5556 | int i; | |
5557 | ||
5558 | rcu_read_lock(); | |
5559 | if (!swap) | |
2c488db2 | 5560 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 5561 | else |
2c488db2 | 5562 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
5563 | |
5564 | if (!t) | |
5565 | goto unlock; | |
5566 | ||
5567 | usage = mem_cgroup_usage(memcg, swap); | |
5568 | ||
5569 | /* | |
748dad36 | 5570 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
5571 | * If it's not true, a threshold was crossed after last |
5572 | * call of __mem_cgroup_threshold(). | |
5573 | */ | |
5407a562 | 5574 | i = t->current_threshold; |
2e72b634 KS |
5575 | |
5576 | /* | |
5577 | * Iterate backward over array of thresholds starting from | |
5578 | * current_threshold and check if a threshold is crossed. | |
5579 | * If none of thresholds below usage is crossed, we read | |
5580 | * only one element of the array here. | |
5581 | */ | |
5582 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
5583 | eventfd_signal(t->entries[i].eventfd, 1); | |
5584 | ||
5585 | /* i = current_threshold + 1 */ | |
5586 | i++; | |
5587 | ||
5588 | /* | |
5589 | * Iterate forward over array of thresholds starting from | |
5590 | * current_threshold+1 and check if a threshold is crossed. | |
5591 | * If none of thresholds above usage is crossed, we read | |
5592 | * only one element of the array here. | |
5593 | */ | |
5594 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
5595 | eventfd_signal(t->entries[i].eventfd, 1); | |
5596 | ||
5597 | /* Update current_threshold */ | |
5407a562 | 5598 | t->current_threshold = i - 1; |
2e72b634 KS |
5599 | unlock: |
5600 | rcu_read_unlock(); | |
5601 | } | |
5602 | ||
5603 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
5604 | { | |
ad4ca5f4 KS |
5605 | while (memcg) { |
5606 | __mem_cgroup_threshold(memcg, false); | |
5607 | if (do_swap_account) | |
5608 | __mem_cgroup_threshold(memcg, true); | |
5609 | ||
5610 | memcg = parent_mem_cgroup(memcg); | |
5611 | } | |
2e72b634 KS |
5612 | } |
5613 | ||
5614 | static int compare_thresholds(const void *a, const void *b) | |
5615 | { | |
5616 | const struct mem_cgroup_threshold *_a = a; | |
5617 | const struct mem_cgroup_threshold *_b = b; | |
5618 | ||
5619 | return _a->threshold - _b->threshold; | |
5620 | } | |
5621 | ||
c0ff4b85 | 5622 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
5623 | { |
5624 | struct mem_cgroup_eventfd_list *ev; | |
5625 | ||
c0ff4b85 | 5626 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 KH |
5627 | eventfd_signal(ev->eventfd, 1); |
5628 | return 0; | |
5629 | } | |
5630 | ||
c0ff4b85 | 5631 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 5632 | { |
7d74b06f KH |
5633 | struct mem_cgroup *iter; |
5634 | ||
c0ff4b85 | 5635 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 5636 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
5637 | } |
5638 | ||
5639 | static int mem_cgroup_usage_register_event(struct cgroup *cgrp, | |
5640 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
2e72b634 KS |
5641 | { |
5642 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
5643 | struct mem_cgroup_thresholds *thresholds; |
5644 | struct mem_cgroup_threshold_ary *new; | |
86ae53e1 | 5645 | enum res_type type = MEMFILE_TYPE(cft->private); |
2e72b634 | 5646 | u64 threshold, usage; |
2c488db2 | 5647 | int i, size, ret; |
2e72b634 KS |
5648 | |
5649 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
5650 | if (ret) | |
5651 | return ret; | |
5652 | ||
5653 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 5654 | |
2e72b634 | 5655 | if (type == _MEM) |
2c488db2 | 5656 | thresholds = &memcg->thresholds; |
2e72b634 | 5657 | else if (type == _MEMSWAP) |
2c488db2 | 5658 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
5659 | else |
5660 | BUG(); | |
5661 | ||
5662 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
5663 | ||
5664 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 5665 | if (thresholds->primary) |
2e72b634 KS |
5666 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
5667 | ||
2c488db2 | 5668 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
5669 | |
5670 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 5671 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 5672 | GFP_KERNEL); |
2c488db2 | 5673 | if (!new) { |
2e72b634 KS |
5674 | ret = -ENOMEM; |
5675 | goto unlock; | |
5676 | } | |
2c488db2 | 5677 | new->size = size; |
2e72b634 KS |
5678 | |
5679 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
5680 | if (thresholds->primary) { |
5681 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 5682 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
5683 | } |
5684 | ||
2e72b634 | 5685 | /* Add new threshold */ |
2c488db2 KS |
5686 | new->entries[size - 1].eventfd = eventfd; |
5687 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
5688 | |
5689 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 5690 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
5691 | compare_thresholds, NULL); |
5692 | ||
5693 | /* Find current threshold */ | |
2c488db2 | 5694 | new->current_threshold = -1; |
2e72b634 | 5695 | for (i = 0; i < size; i++) { |
748dad36 | 5696 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 5697 | /* |
2c488db2 KS |
5698 | * new->current_threshold will not be used until |
5699 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
5700 | * it here. |
5701 | */ | |
2c488db2 | 5702 | ++new->current_threshold; |
748dad36 SZ |
5703 | } else |
5704 | break; | |
2e72b634 KS |
5705 | } |
5706 | ||
2c488db2 KS |
5707 | /* Free old spare buffer and save old primary buffer as spare */ |
5708 | kfree(thresholds->spare); | |
5709 | thresholds->spare = thresholds->primary; | |
5710 | ||
5711 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 5712 | |
907860ed | 5713 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
5714 | synchronize_rcu(); |
5715 | ||
2e72b634 KS |
5716 | unlock: |
5717 | mutex_unlock(&memcg->thresholds_lock); | |
5718 | ||
5719 | return ret; | |
5720 | } | |
5721 | ||
907860ed | 5722 | static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp, |
9490ff27 | 5723 | struct cftype *cft, struct eventfd_ctx *eventfd) |
2e72b634 KS |
5724 | { |
5725 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
5726 | struct mem_cgroup_thresholds *thresholds; |
5727 | struct mem_cgroup_threshold_ary *new; | |
86ae53e1 | 5728 | enum res_type type = MEMFILE_TYPE(cft->private); |
2e72b634 | 5729 | u64 usage; |
2c488db2 | 5730 | int i, j, size; |
2e72b634 KS |
5731 | |
5732 | mutex_lock(&memcg->thresholds_lock); | |
5733 | if (type == _MEM) | |
2c488db2 | 5734 | thresholds = &memcg->thresholds; |
2e72b634 | 5735 | else if (type == _MEMSWAP) |
2c488db2 | 5736 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
5737 | else |
5738 | BUG(); | |
5739 | ||
371528ca AV |
5740 | if (!thresholds->primary) |
5741 | goto unlock; | |
5742 | ||
2e72b634 KS |
5743 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); |
5744 | ||
5745 | /* Check if a threshold crossed before removing */ | |
5746 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
5747 | ||
5748 | /* Calculate new number of threshold */ | |
2c488db2 KS |
5749 | size = 0; |
5750 | for (i = 0; i < thresholds->primary->size; i++) { | |
5751 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
5752 | size++; |
5753 | } | |
5754 | ||
2c488db2 | 5755 | new = thresholds->spare; |
907860ed | 5756 | |
2e72b634 KS |
5757 | /* Set thresholds array to NULL if we don't have thresholds */ |
5758 | if (!size) { | |
2c488db2 KS |
5759 | kfree(new); |
5760 | new = NULL; | |
907860ed | 5761 | goto swap_buffers; |
2e72b634 KS |
5762 | } |
5763 | ||
2c488db2 | 5764 | new->size = size; |
2e72b634 KS |
5765 | |
5766 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
5767 | new->current_threshold = -1; |
5768 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
5769 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
5770 | continue; |
5771 | ||
2c488db2 | 5772 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 5773 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 5774 | /* |
2c488db2 | 5775 | * new->current_threshold will not be used |
2e72b634 KS |
5776 | * until rcu_assign_pointer(), so it's safe to increment |
5777 | * it here. | |
5778 | */ | |
2c488db2 | 5779 | ++new->current_threshold; |
2e72b634 KS |
5780 | } |
5781 | j++; | |
5782 | } | |
5783 | ||
907860ed | 5784 | swap_buffers: |
2c488db2 KS |
5785 | /* Swap primary and spare array */ |
5786 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
5787 | /* If all events are unregistered, free the spare array */ |
5788 | if (!new) { | |
5789 | kfree(thresholds->spare); | |
5790 | thresholds->spare = NULL; | |
5791 | } | |
5792 | ||
2c488db2 | 5793 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 5794 | |
907860ed | 5795 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 5796 | synchronize_rcu(); |
371528ca | 5797 | unlock: |
2e72b634 | 5798 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 5799 | } |
c1e862c1 | 5800 | |
9490ff27 KH |
5801 | static int mem_cgroup_oom_register_event(struct cgroup *cgrp, |
5802 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
5803 | { | |
5804 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
5805 | struct mem_cgroup_eventfd_list *event; | |
86ae53e1 | 5806 | enum res_type type = MEMFILE_TYPE(cft->private); |
9490ff27 KH |
5807 | |
5808 | BUG_ON(type != _OOM_TYPE); | |
5809 | event = kmalloc(sizeof(*event), GFP_KERNEL); | |
5810 | if (!event) | |
5811 | return -ENOMEM; | |
5812 | ||
1af8efe9 | 5813 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
5814 | |
5815 | event->eventfd = eventfd; | |
5816 | list_add(&event->list, &memcg->oom_notify); | |
5817 | ||
5818 | /* already in OOM ? */ | |
79dfdacc | 5819 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 5820 | eventfd_signal(eventfd, 1); |
1af8efe9 | 5821 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
5822 | |
5823 | return 0; | |
5824 | } | |
5825 | ||
907860ed | 5826 | static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp, |
9490ff27 KH |
5827 | struct cftype *cft, struct eventfd_ctx *eventfd) |
5828 | { | |
c0ff4b85 | 5829 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
9490ff27 | 5830 | struct mem_cgroup_eventfd_list *ev, *tmp; |
86ae53e1 | 5831 | enum res_type type = MEMFILE_TYPE(cft->private); |
9490ff27 KH |
5832 | |
5833 | BUG_ON(type != _OOM_TYPE); | |
5834 | ||
1af8efe9 | 5835 | spin_lock(&memcg_oom_lock); |
9490ff27 | 5836 | |
c0ff4b85 | 5837 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
5838 | if (ev->eventfd == eventfd) { |
5839 | list_del(&ev->list); | |
5840 | kfree(ev); | |
5841 | } | |
5842 | } | |
5843 | ||
1af8efe9 | 5844 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
5845 | } |
5846 | ||
3c11ecf4 KH |
5847 | static int mem_cgroup_oom_control_read(struct cgroup *cgrp, |
5848 | struct cftype *cft, struct cgroup_map_cb *cb) | |
5849 | { | |
c0ff4b85 | 5850 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 | 5851 | |
c0ff4b85 | 5852 | cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable); |
3c11ecf4 | 5853 | |
c0ff4b85 | 5854 | if (atomic_read(&memcg->under_oom)) |
3c11ecf4 KH |
5855 | cb->fill(cb, "under_oom", 1); |
5856 | else | |
5857 | cb->fill(cb, "under_oom", 0); | |
5858 | return 0; | |
5859 | } | |
5860 | ||
3c11ecf4 KH |
5861 | static int mem_cgroup_oom_control_write(struct cgroup *cgrp, |
5862 | struct cftype *cft, u64 val) | |
5863 | { | |
c0ff4b85 | 5864 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 KH |
5865 | struct mem_cgroup *parent; |
5866 | ||
5867 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
5868 | if (!cgrp->parent || !((val == 0) || (val == 1))) | |
5869 | return -EINVAL; | |
5870 | ||
5871 | parent = mem_cgroup_from_cont(cgrp->parent); | |
5872 | ||
0999821b | 5873 | mutex_lock(&memcg_create_mutex); |
3c11ecf4 | 5874 | /* oom-kill-disable is a flag for subhierarchy. */ |
b5f99b53 | 5875 | if ((parent->use_hierarchy) || memcg_has_children(memcg)) { |
0999821b | 5876 | mutex_unlock(&memcg_create_mutex); |
3c11ecf4 KH |
5877 | return -EINVAL; |
5878 | } | |
c0ff4b85 | 5879 | memcg->oom_kill_disable = val; |
4d845ebf | 5880 | if (!val) |
c0ff4b85 | 5881 | memcg_oom_recover(memcg); |
0999821b | 5882 | mutex_unlock(&memcg_create_mutex); |
3c11ecf4 KH |
5883 | return 0; |
5884 | } | |
5885 | ||
c255a458 | 5886 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 5887 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 5888 | { |
55007d84 GC |
5889 | int ret; |
5890 | ||
2633d7a0 | 5891 | memcg->kmemcg_id = -1; |
55007d84 GC |
5892 | ret = memcg_propagate_kmem(memcg); |
5893 | if (ret) | |
5894 | return ret; | |
2633d7a0 | 5895 | |
1d62e436 | 5896 | return mem_cgroup_sockets_init(memcg, ss); |
573b400d | 5897 | } |
e5671dfa | 5898 | |
10d5ebf4 | 5899 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 5900 | { |
1d62e436 | 5901 | mem_cgroup_sockets_destroy(memcg); |
10d5ebf4 LZ |
5902 | } |
5903 | ||
5904 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) | |
5905 | { | |
5906 | if (!memcg_kmem_is_active(memcg)) | |
5907 | return; | |
5908 | ||
5909 | /* | |
5910 | * kmem charges can outlive the cgroup. In the case of slab | |
5911 | * pages, for instance, a page contain objects from various | |
5912 | * processes. As we prevent from taking a reference for every | |
5913 | * such allocation we have to be careful when doing uncharge | |
5914 | * (see memcg_uncharge_kmem) and here during offlining. | |
5915 | * | |
5916 | * The idea is that that only the _last_ uncharge which sees | |
5917 | * the dead memcg will drop the last reference. An additional | |
5918 | * reference is taken here before the group is marked dead | |
5919 | * which is then paired with css_put during uncharge resp. here. | |
5920 | * | |
5921 | * Although this might sound strange as this path is called from | |
5922 | * css_offline() when the referencemight have dropped down to 0 | |
5923 | * and shouldn't be incremented anymore (css_tryget would fail) | |
5924 | * we do not have other options because of the kmem allocations | |
5925 | * lifetime. | |
5926 | */ | |
5927 | css_get(&memcg->css); | |
7de37682 GC |
5928 | |
5929 | memcg_kmem_mark_dead(memcg); | |
5930 | ||
5931 | if (res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0) | |
5932 | return; | |
5933 | ||
7de37682 | 5934 | if (memcg_kmem_test_and_clear_dead(memcg)) |
10d5ebf4 | 5935 | css_put(&memcg->css); |
d1a4c0b3 | 5936 | } |
e5671dfa | 5937 | #else |
cbe128e3 | 5938 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
5939 | { |
5940 | return 0; | |
5941 | } | |
d1a4c0b3 | 5942 | |
10d5ebf4 LZ |
5943 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
5944 | { | |
5945 | } | |
5946 | ||
5947 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) | |
d1a4c0b3 GC |
5948 | { |
5949 | } | |
e5671dfa GC |
5950 | #endif |
5951 | ||
8cdea7c0 BS |
5952 | static struct cftype mem_cgroup_files[] = { |
5953 | { | |
0eea1030 | 5954 | .name = "usage_in_bytes", |
8c7c6e34 | 5955 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
af36f906 | 5956 | .read = mem_cgroup_read, |
9490ff27 KH |
5957 | .register_event = mem_cgroup_usage_register_event, |
5958 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8cdea7c0 | 5959 | }, |
c84872e1 PE |
5960 | { |
5961 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 5962 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 5963 | .trigger = mem_cgroup_reset, |
af36f906 | 5964 | .read = mem_cgroup_read, |
c84872e1 | 5965 | }, |
8cdea7c0 | 5966 | { |
0eea1030 | 5967 | .name = "limit_in_bytes", |
8c7c6e34 | 5968 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 5969 | .write_string = mem_cgroup_write, |
af36f906 | 5970 | .read = mem_cgroup_read, |
8cdea7c0 | 5971 | }, |
296c81d8 BS |
5972 | { |
5973 | .name = "soft_limit_in_bytes", | |
5974 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
5975 | .write_string = mem_cgroup_write, | |
af36f906 | 5976 | .read = mem_cgroup_read, |
296c81d8 | 5977 | }, |
8cdea7c0 BS |
5978 | { |
5979 | .name = "failcnt", | |
8c7c6e34 | 5980 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 5981 | .trigger = mem_cgroup_reset, |
af36f906 | 5982 | .read = mem_cgroup_read, |
8cdea7c0 | 5983 | }, |
d2ceb9b7 KH |
5984 | { |
5985 | .name = "stat", | |
ab215884 | 5986 | .read_seq_string = memcg_stat_show, |
d2ceb9b7 | 5987 | }, |
c1e862c1 KH |
5988 | { |
5989 | .name = "force_empty", | |
5990 | .trigger = mem_cgroup_force_empty_write, | |
5991 | }, | |
18f59ea7 BS |
5992 | { |
5993 | .name = "use_hierarchy", | |
f00baae7 | 5994 | .flags = CFTYPE_INSANE, |
18f59ea7 BS |
5995 | .write_u64 = mem_cgroup_hierarchy_write, |
5996 | .read_u64 = mem_cgroup_hierarchy_read, | |
5997 | }, | |
a7885eb8 KM |
5998 | { |
5999 | .name = "swappiness", | |
6000 | .read_u64 = mem_cgroup_swappiness_read, | |
6001 | .write_u64 = mem_cgroup_swappiness_write, | |
6002 | }, | |
7dc74be0 DN |
6003 | { |
6004 | .name = "move_charge_at_immigrate", | |
6005 | .read_u64 = mem_cgroup_move_charge_read, | |
6006 | .write_u64 = mem_cgroup_move_charge_write, | |
6007 | }, | |
9490ff27 KH |
6008 | { |
6009 | .name = "oom_control", | |
3c11ecf4 KH |
6010 | .read_map = mem_cgroup_oom_control_read, |
6011 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
6012 | .register_event = mem_cgroup_oom_register_event, |
6013 | .unregister_event = mem_cgroup_oom_unregister_event, | |
6014 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), | |
6015 | }, | |
70ddf637 AV |
6016 | { |
6017 | .name = "pressure_level", | |
6018 | .register_event = vmpressure_register_event, | |
6019 | .unregister_event = vmpressure_unregister_event, | |
6020 | }, | |
406eb0c9 YH |
6021 | #ifdef CONFIG_NUMA |
6022 | { | |
6023 | .name = "numa_stat", | |
ab215884 | 6024 | .read_seq_string = memcg_numa_stat_show, |
406eb0c9 YH |
6025 | }, |
6026 | #endif | |
510fc4e1 GC |
6027 | #ifdef CONFIG_MEMCG_KMEM |
6028 | { | |
6029 | .name = "kmem.limit_in_bytes", | |
6030 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
6031 | .write_string = mem_cgroup_write, | |
6032 | .read = mem_cgroup_read, | |
6033 | }, | |
6034 | { | |
6035 | .name = "kmem.usage_in_bytes", | |
6036 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
6037 | .read = mem_cgroup_read, | |
6038 | }, | |
6039 | { | |
6040 | .name = "kmem.failcnt", | |
6041 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6042 | .trigger = mem_cgroup_reset, | |
6043 | .read = mem_cgroup_read, | |
6044 | }, | |
6045 | { | |
6046 | .name = "kmem.max_usage_in_bytes", | |
6047 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6048 | .trigger = mem_cgroup_reset, | |
6049 | .read = mem_cgroup_read, | |
6050 | }, | |
749c5415 GC |
6051 | #ifdef CONFIG_SLABINFO |
6052 | { | |
6053 | .name = "kmem.slabinfo", | |
6054 | .read_seq_string = mem_cgroup_slabinfo_read, | |
6055 | }, | |
6056 | #endif | |
8c7c6e34 | 6057 | #endif |
6bc10349 | 6058 | { }, /* terminate */ |
af36f906 | 6059 | }; |
8c7c6e34 | 6060 | |
2d11085e MH |
6061 | #ifdef CONFIG_MEMCG_SWAP |
6062 | static struct cftype memsw_cgroup_files[] = { | |
6063 | { | |
6064 | .name = "memsw.usage_in_bytes", | |
6065 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
6066 | .read = mem_cgroup_read, | |
6067 | .register_event = mem_cgroup_usage_register_event, | |
6068 | .unregister_event = mem_cgroup_usage_unregister_event, | |
6069 | }, | |
6070 | { | |
6071 | .name = "memsw.max_usage_in_bytes", | |
6072 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6073 | .trigger = mem_cgroup_reset, | |
6074 | .read = mem_cgroup_read, | |
6075 | }, | |
6076 | { | |
6077 | .name = "memsw.limit_in_bytes", | |
6078 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
6079 | .write_string = mem_cgroup_write, | |
6080 | .read = mem_cgroup_read, | |
6081 | }, | |
6082 | { | |
6083 | .name = "memsw.failcnt", | |
6084 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6085 | .trigger = mem_cgroup_reset, | |
6086 | .read = mem_cgroup_read, | |
6087 | }, | |
6088 | { }, /* terminate */ | |
6089 | }; | |
6090 | #endif | |
c0ff4b85 | 6091 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
6092 | { |
6093 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 6094 | struct mem_cgroup_per_zone *mz; |
41e3355d | 6095 | int zone, tmp = node; |
1ecaab2b KH |
6096 | /* |
6097 | * This routine is called against possible nodes. | |
6098 | * But it's BUG to call kmalloc() against offline node. | |
6099 | * | |
6100 | * TODO: this routine can waste much memory for nodes which will | |
6101 | * never be onlined. It's better to use memory hotplug callback | |
6102 | * function. | |
6103 | */ | |
41e3355d KH |
6104 | if (!node_state(node, N_NORMAL_MEMORY)) |
6105 | tmp = -1; | |
17295c88 | 6106 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
6107 | if (!pn) |
6108 | return 1; | |
1ecaab2b | 6109 | |
1ecaab2b KH |
6110 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
6111 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 6112 | lruvec_init(&mz->lruvec); |
f64c3f54 | 6113 | mz->usage_in_excess = 0; |
4e416953 | 6114 | mz->on_tree = false; |
d79154bb | 6115 | mz->memcg = memcg; |
1ecaab2b | 6116 | } |
54f72fe0 | 6117 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
6118 | return 0; |
6119 | } | |
6120 | ||
c0ff4b85 | 6121 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 6122 | { |
54f72fe0 | 6123 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
6124 | } |
6125 | ||
33327948 KH |
6126 | static struct mem_cgroup *mem_cgroup_alloc(void) |
6127 | { | |
d79154bb | 6128 | struct mem_cgroup *memcg; |
45cf7ebd | 6129 | size_t size = memcg_size(); |
33327948 | 6130 | |
45cf7ebd | 6131 | /* Can be very big if nr_node_ids is very big */ |
c8dad2bb | 6132 | if (size < PAGE_SIZE) |
d79154bb | 6133 | memcg = kzalloc(size, GFP_KERNEL); |
33327948 | 6134 | else |
d79154bb | 6135 | memcg = vzalloc(size); |
33327948 | 6136 | |
d79154bb | 6137 | if (!memcg) |
e7bbcdf3 DC |
6138 | return NULL; |
6139 | ||
d79154bb HD |
6140 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
6141 | if (!memcg->stat) | |
d2e61b8d | 6142 | goto out_free; |
d79154bb HD |
6143 | spin_lock_init(&memcg->pcp_counter_lock); |
6144 | return memcg; | |
d2e61b8d DC |
6145 | |
6146 | out_free: | |
6147 | if (size < PAGE_SIZE) | |
d79154bb | 6148 | kfree(memcg); |
d2e61b8d | 6149 | else |
d79154bb | 6150 | vfree(memcg); |
d2e61b8d | 6151 | return NULL; |
33327948 KH |
6152 | } |
6153 | ||
59927fb9 | 6154 | /* |
c8b2a36f GC |
6155 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
6156 | * (scanning all at force_empty is too costly...) | |
6157 | * | |
6158 | * Instead of clearing all references at force_empty, we remember | |
6159 | * the number of reference from swap_cgroup and free mem_cgroup when | |
6160 | * it goes down to 0. | |
6161 | * | |
6162 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 6163 | */ |
c8b2a36f GC |
6164 | |
6165 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 6166 | { |
c8b2a36f | 6167 | int node; |
45cf7ebd | 6168 | size_t size = memcg_size(); |
59927fb9 | 6169 | |
c8b2a36f GC |
6170 | mem_cgroup_remove_from_trees(memcg); |
6171 | free_css_id(&mem_cgroup_subsys, &memcg->css); | |
6172 | ||
6173 | for_each_node(node) | |
6174 | free_mem_cgroup_per_zone_info(memcg, node); | |
6175 | ||
6176 | free_percpu(memcg->stat); | |
6177 | ||
3f134619 GC |
6178 | /* |
6179 | * We need to make sure that (at least for now), the jump label | |
6180 | * destruction code runs outside of the cgroup lock. This is because | |
6181 | * get_online_cpus(), which is called from the static_branch update, | |
6182 | * can't be called inside the cgroup_lock. cpusets are the ones | |
6183 | * enforcing this dependency, so if they ever change, we might as well. | |
6184 | * | |
6185 | * schedule_work() will guarantee this happens. Be careful if you need | |
6186 | * to move this code around, and make sure it is outside | |
6187 | * the cgroup_lock. | |
6188 | */ | |
a8964b9b | 6189 | disarm_static_keys(memcg); |
3afe36b1 GC |
6190 | if (size < PAGE_SIZE) |
6191 | kfree(memcg); | |
6192 | else | |
6193 | vfree(memcg); | |
59927fb9 | 6194 | } |
3afe36b1 | 6195 | |
7bcc1bb1 DN |
6196 | /* |
6197 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
6198 | */ | |
e1aab161 | 6199 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 6200 | { |
c0ff4b85 | 6201 | if (!memcg->res.parent) |
7bcc1bb1 | 6202 | return NULL; |
c0ff4b85 | 6203 | return mem_cgroup_from_res_counter(memcg->res.parent, res); |
7bcc1bb1 | 6204 | } |
e1aab161 | 6205 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 6206 | |
8787a1df | 6207 | static void __init mem_cgroup_soft_limit_tree_init(void) |
f64c3f54 BS |
6208 | { |
6209 | struct mem_cgroup_tree_per_node *rtpn; | |
6210 | struct mem_cgroup_tree_per_zone *rtpz; | |
6211 | int tmp, node, zone; | |
6212 | ||
3ed28fa1 | 6213 | for_each_node(node) { |
f64c3f54 BS |
6214 | tmp = node; |
6215 | if (!node_state(node, N_NORMAL_MEMORY)) | |
6216 | tmp = -1; | |
6217 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
8787a1df | 6218 | BUG_ON(!rtpn); |
f64c3f54 BS |
6219 | |
6220 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
6221 | ||
6222 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
6223 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
6224 | rtpz->rb_root = RB_ROOT; | |
6225 | spin_lock_init(&rtpz->lock); | |
6226 | } | |
6227 | } | |
f64c3f54 BS |
6228 | } |
6229 | ||
0eb253e2 | 6230 | static struct cgroup_subsys_state * __ref |
92fb9748 | 6231 | mem_cgroup_css_alloc(struct cgroup *cont) |
8cdea7c0 | 6232 | { |
d142e3e6 | 6233 | struct mem_cgroup *memcg; |
04046e1a | 6234 | long error = -ENOMEM; |
6d12e2d8 | 6235 | int node; |
8cdea7c0 | 6236 | |
c0ff4b85 R |
6237 | memcg = mem_cgroup_alloc(); |
6238 | if (!memcg) | |
04046e1a | 6239 | return ERR_PTR(error); |
78fb7466 | 6240 | |
3ed28fa1 | 6241 | for_each_node(node) |
c0ff4b85 | 6242 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 6243 | goto free_out; |
f64c3f54 | 6244 | |
c077719b | 6245 | /* root ? */ |
28dbc4b6 | 6246 | if (cont->parent == NULL) { |
a41c58a6 | 6247 | root_mem_cgroup = memcg; |
d142e3e6 GC |
6248 | res_counter_init(&memcg->res, NULL); |
6249 | res_counter_init(&memcg->memsw, NULL); | |
6250 | res_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 6251 | } |
28dbc4b6 | 6252 | |
d142e3e6 GC |
6253 | memcg->last_scanned_node = MAX_NUMNODES; |
6254 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
6255 | memcg->move_charge_at_immigrate = 0; |
6256 | mutex_init(&memcg->thresholds_lock); | |
6257 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 6258 | vmpressure_init(&memcg->vmpressure); |
d142e3e6 GC |
6259 | |
6260 | return &memcg->css; | |
6261 | ||
6262 | free_out: | |
6263 | __mem_cgroup_free(memcg); | |
6264 | return ERR_PTR(error); | |
6265 | } | |
6266 | ||
6267 | static int | |
6268 | mem_cgroup_css_online(struct cgroup *cont) | |
6269 | { | |
6270 | struct mem_cgroup *memcg, *parent; | |
6271 | int error = 0; | |
6272 | ||
6273 | if (!cont->parent) | |
6274 | return 0; | |
6275 | ||
0999821b | 6276 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
6277 | memcg = mem_cgroup_from_cont(cont); |
6278 | parent = mem_cgroup_from_cont(cont->parent); | |
6279 | ||
6280 | memcg->use_hierarchy = parent->use_hierarchy; | |
6281 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
6282 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
6283 | ||
6284 | if (parent->use_hierarchy) { | |
c0ff4b85 R |
6285 | res_counter_init(&memcg->res, &parent->res); |
6286 | res_counter_init(&memcg->memsw, &parent->memsw); | |
510fc4e1 | 6287 | res_counter_init(&memcg->kmem, &parent->kmem); |
55007d84 | 6288 | |
7bcc1bb1 | 6289 | /* |
8d76a979 LZ |
6290 | * No need to take a reference to the parent because cgroup |
6291 | * core guarantees its existence. | |
7bcc1bb1 | 6292 | */ |
18f59ea7 | 6293 | } else { |
c0ff4b85 R |
6294 | res_counter_init(&memcg->res, NULL); |
6295 | res_counter_init(&memcg->memsw, NULL); | |
510fc4e1 | 6296 | res_counter_init(&memcg->kmem, NULL); |
8c7f6edb TH |
6297 | /* |
6298 | * Deeper hierachy with use_hierarchy == false doesn't make | |
6299 | * much sense so let cgroup subsystem know about this | |
6300 | * unfortunate state in our controller. | |
6301 | */ | |
d142e3e6 | 6302 | if (parent != root_mem_cgroup) |
8c7f6edb | 6303 | mem_cgroup_subsys.broken_hierarchy = true; |
18f59ea7 | 6304 | } |
cbe128e3 GC |
6305 | |
6306 | error = memcg_init_kmem(memcg, &mem_cgroup_subsys); | |
0999821b | 6307 | mutex_unlock(&memcg_create_mutex); |
d142e3e6 | 6308 | return error; |
8cdea7c0 BS |
6309 | } |
6310 | ||
5f578161 MH |
6311 | /* |
6312 | * Announce all parents that a group from their hierarchy is gone. | |
6313 | */ | |
6314 | static void mem_cgroup_invalidate_reclaim_iterators(struct mem_cgroup *memcg) | |
6315 | { | |
6316 | struct mem_cgroup *parent = memcg; | |
6317 | ||
6318 | while ((parent = parent_mem_cgroup(parent))) | |
519ebea3 | 6319 | mem_cgroup_iter_invalidate(parent); |
5f578161 MH |
6320 | |
6321 | /* | |
6322 | * if the root memcg is not hierarchical we have to check it | |
6323 | * explicitely. | |
6324 | */ | |
6325 | if (!root_mem_cgroup->use_hierarchy) | |
519ebea3 | 6326 | mem_cgroup_iter_invalidate(root_mem_cgroup); |
5f578161 MH |
6327 | } |
6328 | ||
92fb9748 | 6329 | static void mem_cgroup_css_offline(struct cgroup *cont) |
df878fb0 | 6330 | { |
c0ff4b85 | 6331 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
ec64f515 | 6332 | |
10d5ebf4 LZ |
6333 | kmem_cgroup_css_offline(memcg); |
6334 | ||
5f578161 | 6335 | mem_cgroup_invalidate_reclaim_iterators(memcg); |
ab5196c2 | 6336 | mem_cgroup_reparent_charges(memcg); |
1f458cbf | 6337 | mem_cgroup_destroy_all_caches(memcg); |
df878fb0 KH |
6338 | } |
6339 | ||
92fb9748 | 6340 | static void mem_cgroup_css_free(struct cgroup *cont) |
8cdea7c0 | 6341 | { |
c0ff4b85 | 6342 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
c268e994 | 6343 | |
10d5ebf4 | 6344 | memcg_destroy_kmem(memcg); |
465939a1 | 6345 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
6346 | } |
6347 | ||
02491447 | 6348 | #ifdef CONFIG_MMU |
7dc74be0 | 6349 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
6350 | #define PRECHARGE_COUNT_AT_ONCE 256 |
6351 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 6352 | { |
854ffa8d DN |
6353 | int ret = 0; |
6354 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
c0ff4b85 | 6355 | struct mem_cgroup *memcg = mc.to; |
4ffef5fe | 6356 | |
c0ff4b85 | 6357 | if (mem_cgroup_is_root(memcg)) { |
854ffa8d DN |
6358 | mc.precharge += count; |
6359 | /* we don't need css_get for root */ | |
6360 | return ret; | |
6361 | } | |
6362 | /* try to charge at once */ | |
6363 | if (count > 1) { | |
6364 | struct res_counter *dummy; | |
6365 | /* | |
c0ff4b85 | 6366 | * "memcg" cannot be under rmdir() because we've already checked |
854ffa8d DN |
6367 | * by cgroup_lock_live_cgroup() that it is not removed and we |
6368 | * are still under the same cgroup_mutex. So we can postpone | |
6369 | * css_get(). | |
6370 | */ | |
c0ff4b85 | 6371 | if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy)) |
854ffa8d | 6372 | goto one_by_one; |
c0ff4b85 | 6373 | if (do_swap_account && res_counter_charge(&memcg->memsw, |
854ffa8d | 6374 | PAGE_SIZE * count, &dummy)) { |
c0ff4b85 | 6375 | res_counter_uncharge(&memcg->res, PAGE_SIZE * count); |
854ffa8d DN |
6376 | goto one_by_one; |
6377 | } | |
6378 | mc.precharge += count; | |
854ffa8d DN |
6379 | return ret; |
6380 | } | |
6381 | one_by_one: | |
6382 | /* fall back to one by one charge */ | |
6383 | while (count--) { | |
6384 | if (signal_pending(current)) { | |
6385 | ret = -EINTR; | |
6386 | break; | |
6387 | } | |
6388 | if (!batch_count--) { | |
6389 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
6390 | cond_resched(); | |
6391 | } | |
c0ff4b85 R |
6392 | ret = __mem_cgroup_try_charge(NULL, |
6393 | GFP_KERNEL, 1, &memcg, false); | |
38c5d72f | 6394 | if (ret) |
854ffa8d | 6395 | /* mem_cgroup_clear_mc() will do uncharge later */ |
38c5d72f | 6396 | return ret; |
854ffa8d DN |
6397 | mc.precharge++; |
6398 | } | |
4ffef5fe DN |
6399 | return ret; |
6400 | } | |
6401 | ||
6402 | /** | |
8d32ff84 | 6403 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
6404 | * @vma: the vma the pte to be checked belongs |
6405 | * @addr: the address corresponding to the pte to be checked | |
6406 | * @ptent: the pte to be checked | |
02491447 | 6407 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
6408 | * |
6409 | * Returns | |
6410 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
6411 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
6412 | * move charge. if @target is not NULL, the page is stored in target->page | |
6413 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
6414 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
6415 | * target for charge migration. if @target is not NULL, the entry is stored | |
6416 | * in target->ent. | |
4ffef5fe DN |
6417 | * |
6418 | * Called with pte lock held. | |
6419 | */ | |
4ffef5fe DN |
6420 | union mc_target { |
6421 | struct page *page; | |
02491447 | 6422 | swp_entry_t ent; |
4ffef5fe DN |
6423 | }; |
6424 | ||
4ffef5fe | 6425 | enum mc_target_type { |
8d32ff84 | 6426 | MC_TARGET_NONE = 0, |
4ffef5fe | 6427 | MC_TARGET_PAGE, |
02491447 | 6428 | MC_TARGET_SWAP, |
4ffef5fe DN |
6429 | }; |
6430 | ||
90254a65 DN |
6431 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
6432 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 6433 | { |
90254a65 | 6434 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 6435 | |
90254a65 DN |
6436 | if (!page || !page_mapped(page)) |
6437 | return NULL; | |
6438 | if (PageAnon(page)) { | |
6439 | /* we don't move shared anon */ | |
4b91355e | 6440 | if (!move_anon()) |
90254a65 | 6441 | return NULL; |
87946a72 DN |
6442 | } else if (!move_file()) |
6443 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
6444 | return NULL; |
6445 | if (!get_page_unless_zero(page)) | |
6446 | return NULL; | |
6447 | ||
6448 | return page; | |
6449 | } | |
6450 | ||
4b91355e | 6451 | #ifdef CONFIG_SWAP |
90254a65 DN |
6452 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
6453 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6454 | { | |
90254a65 DN |
6455 | struct page *page = NULL; |
6456 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
6457 | ||
6458 | if (!move_anon() || non_swap_entry(ent)) | |
6459 | return NULL; | |
4b91355e KH |
6460 | /* |
6461 | * Because lookup_swap_cache() updates some statistics counter, | |
6462 | * we call find_get_page() with swapper_space directly. | |
6463 | */ | |
33806f06 | 6464 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
6465 | if (do_swap_account) |
6466 | entry->val = ent.val; | |
6467 | ||
6468 | return page; | |
6469 | } | |
4b91355e KH |
6470 | #else |
6471 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
6472 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6473 | { | |
6474 | return NULL; | |
6475 | } | |
6476 | #endif | |
90254a65 | 6477 | |
87946a72 DN |
6478 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
6479 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6480 | { | |
6481 | struct page *page = NULL; | |
87946a72 DN |
6482 | struct address_space *mapping; |
6483 | pgoff_t pgoff; | |
6484 | ||
6485 | if (!vma->vm_file) /* anonymous vma */ | |
6486 | return NULL; | |
6487 | if (!move_file()) | |
6488 | return NULL; | |
6489 | ||
87946a72 DN |
6490 | mapping = vma->vm_file->f_mapping; |
6491 | if (pte_none(ptent)) | |
6492 | pgoff = linear_page_index(vma, addr); | |
6493 | else /* pte_file(ptent) is true */ | |
6494 | pgoff = pte_to_pgoff(ptent); | |
6495 | ||
6496 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
6497 | page = find_get_page(mapping, pgoff); |
6498 | ||
6499 | #ifdef CONFIG_SWAP | |
6500 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
6501 | if (radix_tree_exceptional_entry(page)) { | |
6502 | swp_entry_t swap = radix_to_swp_entry(page); | |
87946a72 | 6503 | if (do_swap_account) |
aa3b1895 | 6504 | *entry = swap; |
33806f06 | 6505 | page = find_get_page(swap_address_space(swap), swap.val); |
87946a72 | 6506 | } |
aa3b1895 | 6507 | #endif |
87946a72 DN |
6508 | return page; |
6509 | } | |
6510 | ||
8d32ff84 | 6511 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
6512 | unsigned long addr, pte_t ptent, union mc_target *target) |
6513 | { | |
6514 | struct page *page = NULL; | |
6515 | struct page_cgroup *pc; | |
8d32ff84 | 6516 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
6517 | swp_entry_t ent = { .val = 0 }; |
6518 | ||
6519 | if (pte_present(ptent)) | |
6520 | page = mc_handle_present_pte(vma, addr, ptent); | |
6521 | else if (is_swap_pte(ptent)) | |
6522 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
6523 | else if (pte_none(ptent) || pte_file(ptent)) |
6524 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
6525 | |
6526 | if (!page && !ent.val) | |
8d32ff84 | 6527 | return ret; |
02491447 DN |
6528 | if (page) { |
6529 | pc = lookup_page_cgroup(page); | |
6530 | /* | |
6531 | * Do only loose check w/o page_cgroup lock. | |
6532 | * mem_cgroup_move_account() checks the pc is valid or not under | |
6533 | * the lock. | |
6534 | */ | |
6535 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
6536 | ret = MC_TARGET_PAGE; | |
6537 | if (target) | |
6538 | target->page = page; | |
6539 | } | |
6540 | if (!ret || !target) | |
6541 | put_page(page); | |
6542 | } | |
90254a65 DN |
6543 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
6544 | if (ent.val && !ret && | |
9fb4b7cc | 6545 | css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
6546 | ret = MC_TARGET_SWAP; |
6547 | if (target) | |
6548 | target->ent = ent; | |
4ffef5fe | 6549 | } |
4ffef5fe DN |
6550 | return ret; |
6551 | } | |
6552 | ||
12724850 NH |
6553 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
6554 | /* | |
6555 | * We don't consider swapping or file mapped pages because THP does not | |
6556 | * support them for now. | |
6557 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
6558 | */ | |
6559 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6560 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6561 | { | |
6562 | struct page *page = NULL; | |
6563 | struct page_cgroup *pc; | |
6564 | enum mc_target_type ret = MC_TARGET_NONE; | |
6565 | ||
6566 | page = pmd_page(pmd); | |
6567 | VM_BUG_ON(!page || !PageHead(page)); | |
6568 | if (!move_anon()) | |
6569 | return ret; | |
6570 | pc = lookup_page_cgroup(page); | |
6571 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
6572 | ret = MC_TARGET_PAGE; | |
6573 | if (target) { | |
6574 | get_page(page); | |
6575 | target->page = page; | |
6576 | } | |
6577 | } | |
6578 | return ret; | |
6579 | } | |
6580 | #else | |
6581 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6582 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6583 | { | |
6584 | return MC_TARGET_NONE; | |
6585 | } | |
6586 | #endif | |
6587 | ||
4ffef5fe DN |
6588 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
6589 | unsigned long addr, unsigned long end, | |
6590 | struct mm_walk *walk) | |
6591 | { | |
6592 | struct vm_area_struct *vma = walk->private; | |
6593 | pte_t *pte; | |
6594 | spinlock_t *ptl; | |
6595 | ||
12724850 NH |
6596 | if (pmd_trans_huge_lock(pmd, vma) == 1) { |
6597 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) | |
6598 | mc.precharge += HPAGE_PMD_NR; | |
6599 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 6600 | return 0; |
12724850 | 6601 | } |
03319327 | 6602 | |
45f83cef AA |
6603 | if (pmd_trans_unstable(pmd)) |
6604 | return 0; | |
4ffef5fe DN |
6605 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
6606 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 6607 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
6608 | mc.precharge++; /* increment precharge temporarily */ |
6609 | pte_unmap_unlock(pte - 1, ptl); | |
6610 | cond_resched(); | |
6611 | ||
7dc74be0 DN |
6612 | return 0; |
6613 | } | |
6614 | ||
4ffef5fe DN |
6615 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
6616 | { | |
6617 | unsigned long precharge; | |
6618 | struct vm_area_struct *vma; | |
6619 | ||
dfe076b0 | 6620 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
6621 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
6622 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
6623 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
6624 | .mm = mm, | |
6625 | .private = vma, | |
6626 | }; | |
6627 | if (is_vm_hugetlb_page(vma)) | |
6628 | continue; | |
4ffef5fe DN |
6629 | walk_page_range(vma->vm_start, vma->vm_end, |
6630 | &mem_cgroup_count_precharge_walk); | |
6631 | } | |
dfe076b0 | 6632 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
6633 | |
6634 | precharge = mc.precharge; | |
6635 | mc.precharge = 0; | |
6636 | ||
6637 | return precharge; | |
6638 | } | |
6639 | ||
4ffef5fe DN |
6640 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
6641 | { | |
dfe076b0 DN |
6642 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
6643 | ||
6644 | VM_BUG_ON(mc.moving_task); | |
6645 | mc.moving_task = current; | |
6646 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
6647 | } |
6648 | ||
dfe076b0 DN |
6649 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
6650 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 6651 | { |
2bd9bb20 KH |
6652 | struct mem_cgroup *from = mc.from; |
6653 | struct mem_cgroup *to = mc.to; | |
4050377b | 6654 | int i; |
2bd9bb20 | 6655 | |
4ffef5fe | 6656 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
6657 | if (mc.precharge) { |
6658 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
6659 | mc.precharge = 0; | |
6660 | } | |
6661 | /* | |
6662 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
6663 | * we must uncharge here. | |
6664 | */ | |
6665 | if (mc.moved_charge) { | |
6666 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
6667 | mc.moved_charge = 0; | |
4ffef5fe | 6668 | } |
483c30b5 DN |
6669 | /* we must fixup refcnts and charges */ |
6670 | if (mc.moved_swap) { | |
483c30b5 DN |
6671 | /* uncharge swap account from the old cgroup */ |
6672 | if (!mem_cgroup_is_root(mc.from)) | |
6673 | res_counter_uncharge(&mc.from->memsw, | |
6674 | PAGE_SIZE * mc.moved_swap); | |
4050377b LZ |
6675 | |
6676 | for (i = 0; i < mc.moved_swap; i++) | |
6677 | css_put(&mc.from->css); | |
483c30b5 DN |
6678 | |
6679 | if (!mem_cgroup_is_root(mc.to)) { | |
6680 | /* | |
6681 | * we charged both to->res and to->memsw, so we should | |
6682 | * uncharge to->res. | |
6683 | */ | |
6684 | res_counter_uncharge(&mc.to->res, | |
6685 | PAGE_SIZE * mc.moved_swap); | |
483c30b5 | 6686 | } |
4050377b | 6687 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
6688 | mc.moved_swap = 0; |
6689 | } | |
dfe076b0 DN |
6690 | memcg_oom_recover(from); |
6691 | memcg_oom_recover(to); | |
6692 | wake_up_all(&mc.waitq); | |
6693 | } | |
6694 | ||
6695 | static void mem_cgroup_clear_mc(void) | |
6696 | { | |
6697 | struct mem_cgroup *from = mc.from; | |
6698 | ||
6699 | /* | |
6700 | * we must clear moving_task before waking up waiters at the end of | |
6701 | * task migration. | |
6702 | */ | |
6703 | mc.moving_task = NULL; | |
6704 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 6705 | spin_lock(&mc.lock); |
4ffef5fe DN |
6706 | mc.from = NULL; |
6707 | mc.to = NULL; | |
2bd9bb20 | 6708 | spin_unlock(&mc.lock); |
32047e2a | 6709 | mem_cgroup_end_move(from); |
4ffef5fe DN |
6710 | } |
6711 | ||
761b3ef5 LZ |
6712 | static int mem_cgroup_can_attach(struct cgroup *cgroup, |
6713 | struct cgroup_taskset *tset) | |
7dc74be0 | 6714 | { |
2f7ee569 | 6715 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 6716 | int ret = 0; |
c0ff4b85 | 6717 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup); |
ee5e8472 | 6718 | unsigned long move_charge_at_immigrate; |
7dc74be0 | 6719 | |
ee5e8472 GC |
6720 | /* |
6721 | * We are now commited to this value whatever it is. Changes in this | |
6722 | * tunable will only affect upcoming migrations, not the current one. | |
6723 | * So we need to save it, and keep it going. | |
6724 | */ | |
6725 | move_charge_at_immigrate = memcg->move_charge_at_immigrate; | |
6726 | if (move_charge_at_immigrate) { | |
7dc74be0 DN |
6727 | struct mm_struct *mm; |
6728 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
6729 | ||
c0ff4b85 | 6730 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
6731 | |
6732 | mm = get_task_mm(p); | |
6733 | if (!mm) | |
6734 | return 0; | |
7dc74be0 | 6735 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
6736 | if (mm->owner == p) { |
6737 | VM_BUG_ON(mc.from); | |
6738 | VM_BUG_ON(mc.to); | |
6739 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 6740 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 6741 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 6742 | mem_cgroup_start_move(from); |
2bd9bb20 | 6743 | spin_lock(&mc.lock); |
4ffef5fe | 6744 | mc.from = from; |
c0ff4b85 | 6745 | mc.to = memcg; |
ee5e8472 | 6746 | mc.immigrate_flags = move_charge_at_immigrate; |
2bd9bb20 | 6747 | spin_unlock(&mc.lock); |
dfe076b0 | 6748 | /* We set mc.moving_task later */ |
4ffef5fe DN |
6749 | |
6750 | ret = mem_cgroup_precharge_mc(mm); | |
6751 | if (ret) | |
6752 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
6753 | } |
6754 | mmput(mm); | |
7dc74be0 DN |
6755 | } |
6756 | return ret; | |
6757 | } | |
6758 | ||
761b3ef5 LZ |
6759 | static void mem_cgroup_cancel_attach(struct cgroup *cgroup, |
6760 | struct cgroup_taskset *tset) | |
7dc74be0 | 6761 | { |
4ffef5fe | 6762 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
6763 | } |
6764 | ||
4ffef5fe DN |
6765 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
6766 | unsigned long addr, unsigned long end, | |
6767 | struct mm_walk *walk) | |
7dc74be0 | 6768 | { |
4ffef5fe DN |
6769 | int ret = 0; |
6770 | struct vm_area_struct *vma = walk->private; | |
6771 | pte_t *pte; | |
6772 | spinlock_t *ptl; | |
12724850 NH |
6773 | enum mc_target_type target_type; |
6774 | union mc_target target; | |
6775 | struct page *page; | |
6776 | struct page_cgroup *pc; | |
4ffef5fe | 6777 | |
12724850 NH |
6778 | /* |
6779 | * We don't take compound_lock() here but no race with splitting thp | |
6780 | * happens because: | |
6781 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
6782 | * under splitting, which means there's no concurrent thp split, | |
6783 | * - if another thread runs into split_huge_page() just after we | |
6784 | * entered this if-block, the thread must wait for page table lock | |
6785 | * to be unlocked in __split_huge_page_splitting(), where the main | |
6786 | * part of thp split is not executed yet. | |
6787 | */ | |
6788 | if (pmd_trans_huge_lock(pmd, vma) == 1) { | |
62ade86a | 6789 | if (mc.precharge < HPAGE_PMD_NR) { |
12724850 NH |
6790 | spin_unlock(&vma->vm_mm->page_table_lock); |
6791 | return 0; | |
6792 | } | |
6793 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
6794 | if (target_type == MC_TARGET_PAGE) { | |
6795 | page = target.page; | |
6796 | if (!isolate_lru_page(page)) { | |
6797 | pc = lookup_page_cgroup(page); | |
6798 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, | |
2f3479b1 | 6799 | pc, mc.from, mc.to)) { |
12724850 NH |
6800 | mc.precharge -= HPAGE_PMD_NR; |
6801 | mc.moved_charge += HPAGE_PMD_NR; | |
6802 | } | |
6803 | putback_lru_page(page); | |
6804 | } | |
6805 | put_page(page); | |
6806 | } | |
6807 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 6808 | return 0; |
12724850 NH |
6809 | } |
6810 | ||
45f83cef AA |
6811 | if (pmd_trans_unstable(pmd)) |
6812 | return 0; | |
4ffef5fe DN |
6813 | retry: |
6814 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
6815 | for (; addr != end; addr += PAGE_SIZE) { | |
6816 | pte_t ptent = *(pte++); | |
02491447 | 6817 | swp_entry_t ent; |
4ffef5fe DN |
6818 | |
6819 | if (!mc.precharge) | |
6820 | break; | |
6821 | ||
8d32ff84 | 6822 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
6823 | case MC_TARGET_PAGE: |
6824 | page = target.page; | |
6825 | if (isolate_lru_page(page)) | |
6826 | goto put; | |
6827 | pc = lookup_page_cgroup(page); | |
7ec99d62 | 6828 | if (!mem_cgroup_move_account(page, 1, pc, |
2f3479b1 | 6829 | mc.from, mc.to)) { |
4ffef5fe | 6830 | mc.precharge--; |
854ffa8d DN |
6831 | /* we uncharge from mc.from later. */ |
6832 | mc.moved_charge++; | |
4ffef5fe DN |
6833 | } |
6834 | putback_lru_page(page); | |
8d32ff84 | 6835 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
6836 | put_page(page); |
6837 | break; | |
02491447 DN |
6838 | case MC_TARGET_SWAP: |
6839 | ent = target.ent; | |
e91cbb42 | 6840 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6841 | mc.precharge--; |
483c30b5 DN |
6842 | /* we fixup refcnts and charges later. */ |
6843 | mc.moved_swap++; | |
6844 | } | |
02491447 | 6845 | break; |
4ffef5fe DN |
6846 | default: |
6847 | break; | |
6848 | } | |
6849 | } | |
6850 | pte_unmap_unlock(pte - 1, ptl); | |
6851 | cond_resched(); | |
6852 | ||
6853 | if (addr != end) { | |
6854 | /* | |
6855 | * We have consumed all precharges we got in can_attach(). | |
6856 | * We try charge one by one, but don't do any additional | |
6857 | * charges to mc.to if we have failed in charge once in attach() | |
6858 | * phase. | |
6859 | */ | |
854ffa8d | 6860 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6861 | if (!ret) |
6862 | goto retry; | |
6863 | } | |
6864 | ||
6865 | return ret; | |
6866 | } | |
6867 | ||
6868 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
6869 | { | |
6870 | struct vm_area_struct *vma; | |
6871 | ||
6872 | lru_add_drain_all(); | |
dfe076b0 DN |
6873 | retry: |
6874 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
6875 | /* | |
6876 | * Someone who are holding the mmap_sem might be waiting in | |
6877 | * waitq. So we cancel all extra charges, wake up all waiters, | |
6878 | * and retry. Because we cancel precharges, we might not be able | |
6879 | * to move enough charges, but moving charge is a best-effort | |
6880 | * feature anyway, so it wouldn't be a big problem. | |
6881 | */ | |
6882 | __mem_cgroup_clear_mc(); | |
6883 | cond_resched(); | |
6884 | goto retry; | |
6885 | } | |
4ffef5fe DN |
6886 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
6887 | int ret; | |
6888 | struct mm_walk mem_cgroup_move_charge_walk = { | |
6889 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6890 | .mm = mm, | |
6891 | .private = vma, | |
6892 | }; | |
6893 | if (is_vm_hugetlb_page(vma)) | |
6894 | continue; | |
4ffef5fe DN |
6895 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
6896 | &mem_cgroup_move_charge_walk); | |
6897 | if (ret) | |
6898 | /* | |
6899 | * means we have consumed all precharges and failed in | |
6900 | * doing additional charge. Just abandon here. | |
6901 | */ | |
6902 | break; | |
6903 | } | |
dfe076b0 | 6904 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
6905 | } |
6906 | ||
761b3ef5 LZ |
6907 | static void mem_cgroup_move_task(struct cgroup *cont, |
6908 | struct cgroup_taskset *tset) | |
67e465a7 | 6909 | { |
2f7ee569 | 6910 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 6911 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 6912 | |
dfe076b0 | 6913 | if (mm) { |
a433658c KM |
6914 | if (mc.to) |
6915 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
6916 | mmput(mm); |
6917 | } | |
a433658c KM |
6918 | if (mc.to) |
6919 | mem_cgroup_clear_mc(); | |
67e465a7 | 6920 | } |
5cfb80a7 | 6921 | #else /* !CONFIG_MMU */ |
761b3ef5 LZ |
6922 | static int mem_cgroup_can_attach(struct cgroup *cgroup, |
6923 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
6924 | { |
6925 | return 0; | |
6926 | } | |
761b3ef5 LZ |
6927 | static void mem_cgroup_cancel_attach(struct cgroup *cgroup, |
6928 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
6929 | { |
6930 | } | |
761b3ef5 LZ |
6931 | static void mem_cgroup_move_task(struct cgroup *cont, |
6932 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
6933 | { |
6934 | } | |
6935 | #endif | |
67e465a7 | 6936 | |
f00baae7 TH |
6937 | /* |
6938 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
6939 | * to verify sane_behavior flag on each mount attempt. | |
6940 | */ | |
6941 | static void mem_cgroup_bind(struct cgroup *root) | |
6942 | { | |
6943 | /* | |
6944 | * use_hierarchy is forced with sane_behavior. cgroup core | |
6945 | * guarantees that @root doesn't have any children, so turning it | |
6946 | * on for the root memcg is enough. | |
6947 | */ | |
6948 | if (cgroup_sane_behavior(root)) | |
6949 | mem_cgroup_from_cont(root)->use_hierarchy = true; | |
6950 | } | |
6951 | ||
8cdea7c0 BS |
6952 | struct cgroup_subsys mem_cgroup_subsys = { |
6953 | .name = "memory", | |
6954 | .subsys_id = mem_cgroup_subsys_id, | |
92fb9748 | 6955 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6956 | .css_online = mem_cgroup_css_online, |
92fb9748 TH |
6957 | .css_offline = mem_cgroup_css_offline, |
6958 | .css_free = mem_cgroup_css_free, | |
7dc74be0 DN |
6959 | .can_attach = mem_cgroup_can_attach, |
6960 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 6961 | .attach = mem_cgroup_move_task, |
f00baae7 | 6962 | .bind = mem_cgroup_bind, |
6bc10349 | 6963 | .base_cftypes = mem_cgroup_files, |
6d12e2d8 | 6964 | .early_init = 0, |
04046e1a | 6965 | .use_id = 1, |
8cdea7c0 | 6966 | }; |
c077719b | 6967 | |
c255a458 | 6968 | #ifdef CONFIG_MEMCG_SWAP |
a42c390c MH |
6969 | static int __init enable_swap_account(char *s) |
6970 | { | |
6971 | /* consider enabled if no parameter or 1 is given */ | |
a2c8990a | 6972 | if (!strcmp(s, "1")) |
a42c390c | 6973 | really_do_swap_account = 1; |
a2c8990a | 6974 | else if (!strcmp(s, "0")) |
a42c390c MH |
6975 | really_do_swap_account = 0; |
6976 | return 1; | |
6977 | } | |
a2c8990a | 6978 | __setup("swapaccount=", enable_swap_account); |
c077719b | 6979 | |
2d11085e MH |
6980 | static void __init memsw_file_init(void) |
6981 | { | |
6acc8b02 MH |
6982 | WARN_ON(cgroup_add_cftypes(&mem_cgroup_subsys, memsw_cgroup_files)); |
6983 | } | |
6984 | ||
6985 | static void __init enable_swap_cgroup(void) | |
6986 | { | |
6987 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
6988 | do_swap_account = 1; | |
6989 | memsw_file_init(); | |
6990 | } | |
2d11085e | 6991 | } |
6acc8b02 | 6992 | |
2d11085e | 6993 | #else |
6acc8b02 | 6994 | static void __init enable_swap_cgroup(void) |
2d11085e MH |
6995 | { |
6996 | } | |
c077719b | 6997 | #endif |
2d11085e MH |
6998 | |
6999 | /* | |
1081312f MH |
7000 | * subsys_initcall() for memory controller. |
7001 | * | |
7002 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
7003 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
7004 | * everything that doesn't depend on a specific mem_cgroup structure should | |
7005 | * be initialized from here. | |
2d11085e MH |
7006 | */ |
7007 | static int __init mem_cgroup_init(void) | |
7008 | { | |
7009 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); | |
6acc8b02 | 7010 | enable_swap_cgroup(); |
8787a1df | 7011 | mem_cgroup_soft_limit_tree_init(); |
e4777496 | 7012 | memcg_stock_init(); |
2d11085e MH |
7013 | return 0; |
7014 | } | |
7015 | subsys_initcall(mem_cgroup_init); |