]>
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> |
b6ac57d5 | 42 | #include <linux/slab.h> |
66e1707b | 43 | #include <linux/swap.h> |
02491447 | 44 | #include <linux/swapops.h> |
66e1707b | 45 | #include <linux/spinlock.h> |
2e72b634 KS |
46 | #include <linux/eventfd.h> |
47 | #include <linux/sort.h> | |
66e1707b | 48 | #include <linux/fs.h> |
d2ceb9b7 | 49 | #include <linux/seq_file.h> |
33327948 | 50 | #include <linux/vmalloc.h> |
70ddf637 | 51 | #include <linux/vmpressure.h> |
b69408e8 | 52 | #include <linux/mm_inline.h> |
52d4b9ac | 53 | #include <linux/page_cgroup.h> |
cdec2e42 | 54 | #include <linux/cpu.h> |
158e0a2d | 55 | #include <linux/oom.h> |
08e552c6 | 56 | #include "internal.h" |
d1a4c0b3 | 57 | #include <net/sock.h> |
4bd2c1ee | 58 | #include <net/ip.h> |
d1a4c0b3 | 59 | #include <net/tcp_memcontrol.h> |
8cdea7c0 | 60 | |
8697d331 BS |
61 | #include <asm/uaccess.h> |
62 | ||
cc8e970c KM |
63 | #include <trace/events/vmscan.h> |
64 | ||
a181b0e8 | 65 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
68ae564b DR |
66 | EXPORT_SYMBOL(mem_cgroup_subsys); |
67 | ||
a181b0e8 | 68 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
6bbda35c | 69 | static struct mem_cgroup *root_mem_cgroup __read_mostly; |
8cdea7c0 | 70 | |
c255a458 | 71 | #ifdef CONFIG_MEMCG_SWAP |
338c8431 | 72 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ |
c077719b | 73 | int do_swap_account __read_mostly; |
a42c390c MH |
74 | |
75 | /* for remember boot option*/ | |
c255a458 | 76 | #ifdef CONFIG_MEMCG_SWAP_ENABLED |
a42c390c MH |
77 | static int really_do_swap_account __initdata = 1; |
78 | #else | |
79 | static int really_do_swap_account __initdata = 0; | |
80 | #endif | |
81 | ||
c077719b | 82 | #else |
a0db00fc | 83 | #define do_swap_account 0 |
c077719b KH |
84 | #endif |
85 | ||
86 | ||
d52aa412 KH |
87 | /* |
88 | * Statistics for memory cgroup. | |
89 | */ | |
90 | enum mem_cgroup_stat_index { | |
91 | /* | |
92 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
93 | */ | |
b070e65c DR |
94 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ |
95 | MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */ | |
96 | MEM_CGROUP_STAT_RSS_HUGE, /* # of pages charged as anon huge */ | |
97 | MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */ | |
98 | MEM_CGROUP_STAT_SWAP, /* # of pages, swapped out */ | |
d52aa412 KH |
99 | MEM_CGROUP_STAT_NSTATS, |
100 | }; | |
101 | ||
af7c4b0e JW |
102 | static const char * const mem_cgroup_stat_names[] = { |
103 | "cache", | |
104 | "rss", | |
b070e65c | 105 | "rss_huge", |
af7c4b0e JW |
106 | "mapped_file", |
107 | "swap", | |
108 | }; | |
109 | ||
e9f8974f JW |
110 | enum mem_cgroup_events_index { |
111 | MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */ | |
112 | MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */ | |
456f998e YH |
113 | MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */ |
114 | MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */ | |
e9f8974f JW |
115 | MEM_CGROUP_EVENTS_NSTATS, |
116 | }; | |
af7c4b0e JW |
117 | |
118 | static const char * const mem_cgroup_events_names[] = { | |
119 | "pgpgin", | |
120 | "pgpgout", | |
121 | "pgfault", | |
122 | "pgmajfault", | |
123 | }; | |
124 | ||
58cf188e SZ |
125 | static const char * const mem_cgroup_lru_names[] = { |
126 | "inactive_anon", | |
127 | "active_anon", | |
128 | "inactive_file", | |
129 | "active_file", | |
130 | "unevictable", | |
131 | }; | |
132 | ||
7a159cc9 JW |
133 | /* |
134 | * Per memcg event counter is incremented at every pagein/pageout. With THP, | |
135 | * it will be incremated by the number of pages. This counter is used for | |
136 | * for trigger some periodic events. This is straightforward and better | |
137 | * than using jiffies etc. to handle periodic memcg event. | |
138 | */ | |
139 | enum mem_cgroup_events_target { | |
140 | MEM_CGROUP_TARGET_THRESH, | |
7d910c05 | 141 | MEM_CGROUP_TARGET_SOFTLIMIT, |
453a9bf3 | 142 | MEM_CGROUP_TARGET_NUMAINFO, |
7a159cc9 JW |
143 | MEM_CGROUP_NTARGETS, |
144 | }; | |
a0db00fc KS |
145 | #define THRESHOLDS_EVENTS_TARGET 128 |
146 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
147 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 148 | |
d52aa412 | 149 | struct mem_cgroup_stat_cpu { |
7a159cc9 | 150 | long count[MEM_CGROUP_STAT_NSTATS]; |
e9f8974f | 151 | unsigned long events[MEM_CGROUP_EVENTS_NSTATS]; |
13114716 | 152 | unsigned long nr_page_events; |
7a159cc9 | 153 | unsigned long targets[MEM_CGROUP_NTARGETS]; |
d52aa412 KH |
154 | }; |
155 | ||
527a5ec9 | 156 | struct mem_cgroup_reclaim_iter { |
5f578161 MH |
157 | /* |
158 | * last scanned hierarchy member. Valid only if last_dead_count | |
159 | * matches memcg->dead_count of the hierarchy root group. | |
160 | */ | |
542f85f9 | 161 | struct mem_cgroup *last_visited; |
5f578161 MH |
162 | unsigned long last_dead_count; |
163 | ||
527a5ec9 JW |
164 | /* scan generation, increased every round-trip */ |
165 | unsigned int generation; | |
166 | }; | |
167 | ||
6d12e2d8 KH |
168 | /* |
169 | * per-zone information in memory controller. | |
170 | */ | |
6d12e2d8 | 171 | struct mem_cgroup_per_zone { |
6290df54 | 172 | struct lruvec lruvec; |
1eb49272 | 173 | unsigned long lru_size[NR_LRU_LISTS]; |
3e2f41f1 | 174 | |
527a5ec9 JW |
175 | struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1]; |
176 | ||
d79154bb | 177 | struct mem_cgroup *memcg; /* Back pointer, we cannot */ |
4e416953 | 178 | /* use container_of */ |
6d12e2d8 | 179 | }; |
6d12e2d8 KH |
180 | |
181 | struct mem_cgroup_per_node { | |
182 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
183 | }; | |
184 | ||
2e72b634 KS |
185 | struct mem_cgroup_threshold { |
186 | struct eventfd_ctx *eventfd; | |
187 | u64 threshold; | |
188 | }; | |
189 | ||
9490ff27 | 190 | /* For threshold */ |
2e72b634 | 191 | struct mem_cgroup_threshold_ary { |
748dad36 | 192 | /* An array index points to threshold just below or equal to usage. */ |
5407a562 | 193 | int current_threshold; |
2e72b634 KS |
194 | /* Size of entries[] */ |
195 | unsigned int size; | |
196 | /* Array of thresholds */ | |
197 | struct mem_cgroup_threshold entries[0]; | |
198 | }; | |
2c488db2 KS |
199 | |
200 | struct mem_cgroup_thresholds { | |
201 | /* Primary thresholds array */ | |
202 | struct mem_cgroup_threshold_ary *primary; | |
203 | /* | |
204 | * Spare threshold array. | |
205 | * This is needed to make mem_cgroup_unregister_event() "never fail". | |
206 | * It must be able to store at least primary->size - 1 entries. | |
207 | */ | |
208 | struct mem_cgroup_threshold_ary *spare; | |
209 | }; | |
210 | ||
9490ff27 KH |
211 | /* for OOM */ |
212 | struct mem_cgroup_eventfd_list { | |
213 | struct list_head list; | |
214 | struct eventfd_ctx *eventfd; | |
215 | }; | |
2e72b634 | 216 | |
c0ff4b85 R |
217 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
218 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 219 | |
8cdea7c0 BS |
220 | /* |
221 | * The memory controller data structure. The memory controller controls both | |
222 | * page cache and RSS per cgroup. We would eventually like to provide | |
223 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
224 | * to help the administrator determine what knobs to tune. | |
225 | * | |
226 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
227 | * we hit the water mark. May be even add a low water mark, such that |
228 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
229 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
230 | */ |
231 | struct mem_cgroup { | |
232 | struct cgroup_subsys_state css; | |
233 | /* | |
234 | * the counter to account for memory usage | |
235 | */ | |
236 | struct res_counter res; | |
59927fb9 | 237 | |
70ddf637 AV |
238 | /* vmpressure notifications */ |
239 | struct vmpressure vmpressure; | |
240 | ||
465939a1 LZ |
241 | /* |
242 | * the counter to account for mem+swap usage. | |
243 | */ | |
244 | struct res_counter memsw; | |
59927fb9 | 245 | |
510fc4e1 GC |
246 | /* |
247 | * the counter to account for kernel memory usage. | |
248 | */ | |
249 | struct res_counter kmem; | |
18f59ea7 BS |
250 | /* |
251 | * Should the accounting and control be hierarchical, per subtree? | |
252 | */ | |
253 | bool use_hierarchy; | |
510fc4e1 | 254 | unsigned long kmem_account_flags; /* See KMEM_ACCOUNTED_*, below */ |
79dfdacc MH |
255 | |
256 | bool oom_lock; | |
257 | atomic_t under_oom; | |
258 | ||
1f4c025b | 259 | int swappiness; |
3c11ecf4 KH |
260 | /* OOM-Killer disable */ |
261 | int oom_kill_disable; | |
a7885eb8 | 262 | |
22a668d7 KH |
263 | /* set when res.limit == memsw.limit */ |
264 | bool memsw_is_minimum; | |
265 | ||
2e72b634 KS |
266 | /* protect arrays of thresholds */ |
267 | struct mutex thresholds_lock; | |
268 | ||
269 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 270 | struct mem_cgroup_thresholds thresholds; |
907860ed | 271 | |
2e72b634 | 272 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 273 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 274 | |
9490ff27 KH |
275 | /* For oom notifier event fd */ |
276 | struct list_head oom_notify; | |
185efc0f | 277 | |
7dc74be0 DN |
278 | /* |
279 | * Should we move charges of a task when a task is moved into this | |
280 | * mem_cgroup ? And what type of charges should we move ? | |
281 | */ | |
282 | unsigned long move_charge_at_immigrate; | |
619d094b KH |
283 | /* |
284 | * set > 0 if pages under this cgroup are moving to other cgroup. | |
285 | */ | |
286 | atomic_t moving_account; | |
312734c0 KH |
287 | /* taken only while moving_account > 0 */ |
288 | spinlock_t move_lock; | |
d52aa412 | 289 | /* |
c62b1a3b | 290 | * percpu counter. |
d52aa412 | 291 | */ |
3a7951b4 | 292 | struct mem_cgroup_stat_cpu __percpu *stat; |
711d3d2c KH |
293 | /* |
294 | * used when a cpu is offlined or other synchronizations | |
295 | * See mem_cgroup_read_stat(). | |
296 | */ | |
297 | struct mem_cgroup_stat_cpu nocpu_base; | |
298 | spinlock_t pcp_counter_lock; | |
d1a4c0b3 | 299 | |
5f578161 | 300 | atomic_t dead_count; |
4bd2c1ee | 301 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET) |
d1a4c0b3 GC |
302 | struct tcp_memcontrol tcp_mem; |
303 | #endif | |
2633d7a0 GC |
304 | #if defined(CONFIG_MEMCG_KMEM) |
305 | /* analogous to slab_common's slab_caches list. per-memcg */ | |
306 | struct list_head memcg_slab_caches; | |
307 | /* Not a spinlock, we can take a lot of time walking the list */ | |
308 | struct mutex slab_caches_mutex; | |
309 | /* Index in the kmem_cache->memcg_params->memcg_caches array */ | |
310 | int kmemcg_id; | |
311 | #endif | |
45cf7ebd GC |
312 | |
313 | int last_scanned_node; | |
314 | #if MAX_NUMNODES > 1 | |
315 | nodemask_t scan_nodes; | |
316 | atomic_t numainfo_events; | |
317 | atomic_t numainfo_updating; | |
318 | #endif | |
7d910c05 MH |
319 | /* |
320 | * Protects soft_contributed transitions. | |
321 | * See mem_cgroup_update_soft_limit | |
322 | */ | |
323 | spinlock_t soft_lock; | |
324 | ||
325 | /* | |
326 | * If true then this group has increased parents' children_in_excess | |
327 | * when it got over the soft limit. | |
328 | * When a group falls bellow the soft limit, parents' children_in_excess | |
329 | * is decreased and soft_contributed changed to false. | |
330 | */ | |
331 | bool soft_contributed; | |
332 | ||
333 | /* Number of children that are in soft limit excess */ | |
334 | atomic_t children_in_excess; | |
70ddf637 | 335 | |
54f72fe0 JW |
336 | struct mem_cgroup_per_node *nodeinfo[0]; |
337 | /* WARNING: nodeinfo must be the last member here */ | |
8cdea7c0 BS |
338 | }; |
339 | ||
45cf7ebd GC |
340 | static size_t memcg_size(void) |
341 | { | |
342 | return sizeof(struct mem_cgroup) + | |
343 | nr_node_ids * sizeof(struct mem_cgroup_per_node); | |
344 | } | |
345 | ||
510fc4e1 GC |
346 | /* internal only representation about the status of kmem accounting. */ |
347 | enum { | |
348 | KMEM_ACCOUNTED_ACTIVE = 0, /* accounted by this cgroup itself */ | |
a8964b9b | 349 | KMEM_ACCOUNTED_ACTIVATED, /* static key enabled. */ |
7de37682 | 350 | KMEM_ACCOUNTED_DEAD, /* dead memcg with pending kmem charges */ |
510fc4e1 GC |
351 | }; |
352 | ||
a8964b9b GC |
353 | /* We account when limit is on, but only after call sites are patched */ |
354 | #define KMEM_ACCOUNTED_MASK \ | |
355 | ((1 << KMEM_ACCOUNTED_ACTIVE) | (1 << KMEM_ACCOUNTED_ACTIVATED)) | |
510fc4e1 GC |
356 | |
357 | #ifdef CONFIG_MEMCG_KMEM | |
358 | static inline void memcg_kmem_set_active(struct mem_cgroup *memcg) | |
359 | { | |
360 | set_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
361 | } | |
7de37682 GC |
362 | |
363 | static bool memcg_kmem_is_active(struct mem_cgroup *memcg) | |
364 | { | |
365 | return test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
366 | } | |
367 | ||
a8964b9b GC |
368 | static void memcg_kmem_set_activated(struct mem_cgroup *memcg) |
369 | { | |
370 | set_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags); | |
371 | } | |
372 | ||
55007d84 GC |
373 | static void memcg_kmem_clear_activated(struct mem_cgroup *memcg) |
374 | { | |
375 | clear_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags); | |
376 | } | |
377 | ||
7de37682 GC |
378 | static void memcg_kmem_mark_dead(struct mem_cgroup *memcg) |
379 | { | |
10d5ebf4 LZ |
380 | /* |
381 | * Our caller must use css_get() first, because memcg_uncharge_kmem() | |
382 | * will call css_put() if it sees the memcg is dead. | |
383 | */ | |
384 | smp_wmb(); | |
7de37682 GC |
385 | if (test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags)) |
386 | set_bit(KMEM_ACCOUNTED_DEAD, &memcg->kmem_account_flags); | |
387 | } | |
388 | ||
389 | static bool memcg_kmem_test_and_clear_dead(struct mem_cgroup *memcg) | |
390 | { | |
391 | return test_and_clear_bit(KMEM_ACCOUNTED_DEAD, | |
392 | &memcg->kmem_account_flags); | |
393 | } | |
510fc4e1 GC |
394 | #endif |
395 | ||
7dc74be0 DN |
396 | /* Stuffs for move charges at task migration. */ |
397 | /* | |
ee5e8472 GC |
398 | * Types of charges to be moved. "move_charge_at_immitgrate" and |
399 | * "immigrate_flags" are treated as a left-shifted bitmap of these types. | |
7dc74be0 DN |
400 | */ |
401 | enum move_type { | |
4ffef5fe | 402 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 403 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
404 | NR_MOVE_TYPE, |
405 | }; | |
406 | ||
4ffef5fe DN |
407 | /* "mc" and its members are protected by cgroup_mutex */ |
408 | static struct move_charge_struct { | |
b1dd693e | 409 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
410 | struct mem_cgroup *from; |
411 | struct mem_cgroup *to; | |
ee5e8472 | 412 | unsigned long immigrate_flags; |
4ffef5fe | 413 | unsigned long precharge; |
854ffa8d | 414 | unsigned long moved_charge; |
483c30b5 | 415 | unsigned long moved_swap; |
8033b97c DN |
416 | struct task_struct *moving_task; /* a task moving charges */ |
417 | wait_queue_head_t waitq; /* a waitq for other context */ | |
418 | } mc = { | |
2bd9bb20 | 419 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
420 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
421 | }; | |
4ffef5fe | 422 | |
90254a65 DN |
423 | static bool move_anon(void) |
424 | { | |
ee5e8472 | 425 | return test_bit(MOVE_CHARGE_TYPE_ANON, &mc.immigrate_flags); |
90254a65 DN |
426 | } |
427 | ||
87946a72 DN |
428 | static bool move_file(void) |
429 | { | |
ee5e8472 | 430 | return test_bit(MOVE_CHARGE_TYPE_FILE, &mc.immigrate_flags); |
87946a72 DN |
431 | } |
432 | ||
4e416953 BS |
433 | /* |
434 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
435 | * limit reclaim to prevent infinite loops, if they ever occur. | |
436 | */ | |
a0db00fc | 437 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
4e416953 | 438 | |
217bc319 KH |
439 | enum charge_type { |
440 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 441 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 442 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 443 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
444 | NR_CHARGE_TYPE, |
445 | }; | |
446 | ||
8c7c6e34 | 447 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
448 | enum res_type { |
449 | _MEM, | |
450 | _MEMSWAP, | |
451 | _OOM_TYPE, | |
510fc4e1 | 452 | _KMEM, |
86ae53e1 GC |
453 | }; |
454 | ||
a0db00fc KS |
455 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
456 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 457 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
458 | /* Used for OOM nofiier */ |
459 | #define OOM_CONTROL (0) | |
8c7c6e34 | 460 | |
75822b44 BS |
461 | /* |
462 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
463 | */ | |
464 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
465 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
466 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
467 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
468 | ||
0999821b GC |
469 | /* |
470 | * The memcg_create_mutex will be held whenever a new cgroup is created. | |
471 | * As a consequence, any change that needs to protect against new child cgroups | |
472 | * appearing has to hold it as well. | |
473 | */ | |
474 | static DEFINE_MUTEX(memcg_create_mutex); | |
475 | ||
b2145145 WL |
476 | struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s) |
477 | { | |
a7c6d554 | 478 | return s ? container_of(s, struct mem_cgroup, css) : NULL; |
b2145145 WL |
479 | } |
480 | ||
70ddf637 AV |
481 | /* Some nice accessors for the vmpressure. */ |
482 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
483 | { | |
484 | if (!memcg) | |
485 | memcg = root_mem_cgroup; | |
486 | return &memcg->vmpressure; | |
487 | } | |
488 | ||
489 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
490 | { | |
491 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
492 | } | |
493 | ||
494 | struct vmpressure *css_to_vmpressure(struct cgroup_subsys_state *css) | |
495 | { | |
496 | return &mem_cgroup_from_css(css)->vmpressure; | |
497 | } | |
498 | ||
7ffc0edc MH |
499 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
500 | { | |
501 | return (memcg == root_mem_cgroup); | |
502 | } | |
503 | ||
e1aab161 | 504 | /* Writing them here to avoid exposing memcg's inner layout */ |
4bd2c1ee | 505 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
e1aab161 | 506 | |
e1aab161 GC |
507 | void sock_update_memcg(struct sock *sk) |
508 | { | |
376be5ff | 509 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 510 | struct mem_cgroup *memcg; |
3f134619 | 511 | struct cg_proto *cg_proto; |
e1aab161 GC |
512 | |
513 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
514 | ||
f3f511e1 GC |
515 | /* Socket cloning can throw us here with sk_cgrp already |
516 | * filled. It won't however, necessarily happen from | |
517 | * process context. So the test for root memcg given | |
518 | * the current task's memcg won't help us in this case. | |
519 | * | |
520 | * Respecting the original socket's memcg is a better | |
521 | * decision in this case. | |
522 | */ | |
523 | if (sk->sk_cgrp) { | |
524 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
5347e5ae | 525 | css_get(&sk->sk_cgrp->memcg->css); |
f3f511e1 GC |
526 | return; |
527 | } | |
528 | ||
e1aab161 GC |
529 | rcu_read_lock(); |
530 | memcg = mem_cgroup_from_task(current); | |
3f134619 | 531 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
5347e5ae LZ |
532 | if (!mem_cgroup_is_root(memcg) && |
533 | memcg_proto_active(cg_proto) && css_tryget(&memcg->css)) { | |
3f134619 | 534 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
535 | } |
536 | rcu_read_unlock(); | |
537 | } | |
538 | } | |
539 | EXPORT_SYMBOL(sock_update_memcg); | |
540 | ||
541 | void sock_release_memcg(struct sock *sk) | |
542 | { | |
376be5ff | 543 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
544 | struct mem_cgroup *memcg; |
545 | WARN_ON(!sk->sk_cgrp->memcg); | |
546 | memcg = sk->sk_cgrp->memcg; | |
5347e5ae | 547 | css_put(&sk->sk_cgrp->memcg->css); |
e1aab161 GC |
548 | } |
549 | } | |
d1a4c0b3 GC |
550 | |
551 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
552 | { | |
553 | if (!memcg || mem_cgroup_is_root(memcg)) | |
554 | return NULL; | |
555 | ||
556 | return &memcg->tcp_mem.cg_proto; | |
557 | } | |
558 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 559 | |
3f134619 GC |
560 | static void disarm_sock_keys(struct mem_cgroup *memcg) |
561 | { | |
562 | if (!memcg_proto_activated(&memcg->tcp_mem.cg_proto)) | |
563 | return; | |
564 | static_key_slow_dec(&memcg_socket_limit_enabled); | |
565 | } | |
566 | #else | |
567 | static void disarm_sock_keys(struct mem_cgroup *memcg) | |
568 | { | |
569 | } | |
570 | #endif | |
571 | ||
a8964b9b | 572 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 GC |
573 | /* |
574 | * This will be the memcg's index in each cache's ->memcg_params->memcg_caches. | |
575 | * There are two main reasons for not using the css_id for this: | |
576 | * 1) this works better in sparse environments, where we have a lot of memcgs, | |
577 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
578 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
579 | * 200 entry array for that. | |
580 | * | |
581 | * 2) In order not to violate the cgroup API, we would like to do all memory | |
582 | * allocation in ->create(). At that point, we haven't yet allocated the | |
583 | * css_id. Having a separate index prevents us from messing with the cgroup | |
584 | * core for this | |
585 | * | |
586 | * The current size of the caches array is stored in | |
587 | * memcg_limited_groups_array_size. It will double each time we have to | |
588 | * increase it. | |
589 | */ | |
590 | static DEFINE_IDA(kmem_limited_groups); | |
749c5415 GC |
591 | int memcg_limited_groups_array_size; |
592 | ||
55007d84 GC |
593 | /* |
594 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
595 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
596 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
597 | * tunable, but that is strictly not necessary. | |
598 | * | |
599 | * MAX_SIZE should be as large as the number of css_ids. Ideally, we could get | |
600 | * this constant directly from cgroup, but it is understandable that this is | |
601 | * better kept as an internal representation in cgroup.c. In any case, the | |
602 | * css_id space is not getting any smaller, and we don't have to necessarily | |
603 | * increase ours as well if it increases. | |
604 | */ | |
605 | #define MEMCG_CACHES_MIN_SIZE 4 | |
606 | #define MEMCG_CACHES_MAX_SIZE 65535 | |
607 | ||
d7f25f8a GC |
608 | /* |
609 | * A lot of the calls to the cache allocation functions are expected to be | |
610 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
611 | * conditional to this static branch, we'll have to allow modules that does | |
612 | * kmem_cache_alloc and the such to see this symbol as well | |
613 | */ | |
a8964b9b | 614 | struct static_key memcg_kmem_enabled_key; |
d7f25f8a | 615 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b GC |
616 | |
617 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
618 | { | |
55007d84 | 619 | if (memcg_kmem_is_active(memcg)) { |
a8964b9b | 620 | static_key_slow_dec(&memcg_kmem_enabled_key); |
55007d84 GC |
621 | ida_simple_remove(&kmem_limited_groups, memcg->kmemcg_id); |
622 | } | |
bea207c8 GC |
623 | /* |
624 | * This check can't live in kmem destruction function, | |
625 | * since the charges will outlive the cgroup | |
626 | */ | |
627 | WARN_ON(res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0); | |
a8964b9b GC |
628 | } |
629 | #else | |
630 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
631 | { | |
632 | } | |
633 | #endif /* CONFIG_MEMCG_KMEM */ | |
634 | ||
635 | static void disarm_static_keys(struct mem_cgroup *memcg) | |
636 | { | |
637 | disarm_sock_keys(memcg); | |
638 | disarm_kmem_keys(memcg); | |
639 | } | |
640 | ||
c0ff4b85 | 641 | static void drain_all_stock_async(struct mem_cgroup *memcg); |
8c7c6e34 | 642 | |
f64c3f54 | 643 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 644 | mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid) |
f64c3f54 | 645 | { |
45cf7ebd | 646 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
54f72fe0 | 647 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
648 | } |
649 | ||
c0ff4b85 | 650 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 651 | { |
c0ff4b85 | 652 | return &memcg->css; |
d324236b WF |
653 | } |
654 | ||
f64c3f54 | 655 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 656 | page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 657 | { |
97a6c37b JW |
658 | int nid = page_to_nid(page); |
659 | int zid = page_zonenum(page); | |
f64c3f54 | 660 | |
c0ff4b85 | 661 | return mem_cgroup_zoneinfo(memcg, nid, zid); |
f64c3f54 BS |
662 | } |
663 | ||
711d3d2c KH |
664 | /* |
665 | * Implementation Note: reading percpu statistics for memcg. | |
666 | * | |
667 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
668 | * synchronization to implement "quick" read. There are trade-off between | |
669 | * reading cost and precision of value. Then, we may have a chance to implement | |
670 | * a periodic synchronizion of counter in memcg's counter. | |
671 | * | |
672 | * But this _read() function is used for user interface now. The user accounts | |
673 | * memory usage by memory cgroup and he _always_ requires exact value because | |
674 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
675 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
676 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
677 | * | |
678 | * If there are kernel internal actions which can make use of some not-exact | |
679 | * value, and reading all cpu value can be performance bottleneck in some | |
680 | * common workload, threashold and synchonization as vmstat[] should be | |
681 | * implemented. | |
682 | */ | |
c0ff4b85 | 683 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 684 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 685 | { |
7a159cc9 | 686 | long val = 0; |
c62b1a3b | 687 | int cpu; |
c62b1a3b | 688 | |
711d3d2c KH |
689 | get_online_cpus(); |
690 | for_each_online_cpu(cpu) | |
c0ff4b85 | 691 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 692 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
693 | spin_lock(&memcg->pcp_counter_lock); |
694 | val += memcg->nocpu_base.count[idx]; | |
695 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
696 | #endif |
697 | put_online_cpus(); | |
c62b1a3b KH |
698 | return val; |
699 | } | |
700 | ||
c0ff4b85 | 701 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
0c3e73e8 BS |
702 | bool charge) |
703 | { | |
704 | int val = (charge) ? 1 : -1; | |
bff6bb83 | 705 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); |
0c3e73e8 BS |
706 | } |
707 | ||
c0ff4b85 | 708 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
709 | enum mem_cgroup_events_index idx) |
710 | { | |
711 | unsigned long val = 0; | |
712 | int cpu; | |
713 | ||
714 | for_each_online_cpu(cpu) | |
c0ff4b85 | 715 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 716 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
717 | spin_lock(&memcg->pcp_counter_lock); |
718 | val += memcg->nocpu_base.events[idx]; | |
719 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f JW |
720 | #endif |
721 | return val; | |
722 | } | |
723 | ||
c0ff4b85 | 724 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 725 | struct page *page, |
b2402857 | 726 | bool anon, int nr_pages) |
d52aa412 | 727 | { |
c62b1a3b KH |
728 | preempt_disable(); |
729 | ||
b2402857 KH |
730 | /* |
731 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
732 | * counted as CACHE even if it's on ANON LRU. | |
733 | */ | |
734 | if (anon) | |
735 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], | |
c0ff4b85 | 736 | nr_pages); |
d52aa412 | 737 | else |
b2402857 | 738 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 739 | nr_pages); |
55e462b0 | 740 | |
b070e65c DR |
741 | if (PageTransHuge(page)) |
742 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], | |
743 | nr_pages); | |
744 | ||
e401f176 KH |
745 | /* pagein of a big page is an event. So, ignore page size */ |
746 | if (nr_pages > 0) | |
c0ff4b85 | 747 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 748 | else { |
c0ff4b85 | 749 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
750 | nr_pages = -nr_pages; /* for event */ |
751 | } | |
e401f176 | 752 | |
13114716 | 753 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
2e72b634 | 754 | |
c62b1a3b | 755 | preempt_enable(); |
6d12e2d8 KH |
756 | } |
757 | ||
bb2a0de9 | 758 | unsigned long |
4d7dcca2 | 759 | mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
760 | { |
761 | struct mem_cgroup_per_zone *mz; | |
762 | ||
763 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
764 | return mz->lru_size[lru]; | |
765 | } | |
766 | ||
767 | static unsigned long | |
c0ff4b85 | 768 | mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid, |
bb2a0de9 | 769 | unsigned int lru_mask) |
889976db YH |
770 | { |
771 | struct mem_cgroup_per_zone *mz; | |
f156ab93 | 772 | enum lru_list lru; |
bb2a0de9 KH |
773 | unsigned long ret = 0; |
774 | ||
c0ff4b85 | 775 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
bb2a0de9 | 776 | |
f156ab93 HD |
777 | for_each_lru(lru) { |
778 | if (BIT(lru) & lru_mask) | |
779 | ret += mz->lru_size[lru]; | |
bb2a0de9 KH |
780 | } |
781 | return ret; | |
782 | } | |
783 | ||
784 | static unsigned long | |
c0ff4b85 | 785 | mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 KH |
786 | int nid, unsigned int lru_mask) |
787 | { | |
889976db YH |
788 | u64 total = 0; |
789 | int zid; | |
790 | ||
bb2a0de9 | 791 | for (zid = 0; zid < MAX_NR_ZONES; zid++) |
c0ff4b85 R |
792 | total += mem_cgroup_zone_nr_lru_pages(memcg, |
793 | nid, zid, lru_mask); | |
bb2a0de9 | 794 | |
889976db YH |
795 | return total; |
796 | } | |
bb2a0de9 | 797 | |
c0ff4b85 | 798 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 799 | unsigned int lru_mask) |
6d12e2d8 | 800 | { |
889976db | 801 | int nid; |
6d12e2d8 KH |
802 | u64 total = 0; |
803 | ||
31aaea4a | 804 | for_each_node_state(nid, N_MEMORY) |
c0ff4b85 | 805 | total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
6d12e2d8 | 806 | return total; |
d52aa412 KH |
807 | } |
808 | ||
f53d7ce3 JW |
809 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
810 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
811 | { |
812 | unsigned long val, next; | |
813 | ||
13114716 | 814 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 815 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 816 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
817 | if ((long)next - (long)val < 0) { |
818 | switch (target) { | |
819 | case MEM_CGROUP_TARGET_THRESH: | |
820 | next = val + THRESHOLDS_EVENTS_TARGET; | |
821 | break; | |
7d910c05 MH |
822 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
823 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
824 | break; | |
f53d7ce3 JW |
825 | case MEM_CGROUP_TARGET_NUMAINFO: |
826 | next = val + NUMAINFO_EVENTS_TARGET; | |
827 | break; | |
828 | default: | |
829 | break; | |
830 | } | |
831 | __this_cpu_write(memcg->stat->targets[target], next); | |
832 | return true; | |
7a159cc9 | 833 | } |
f53d7ce3 | 834 | return false; |
d2265e6f KH |
835 | } |
836 | ||
7d910c05 MH |
837 | /* |
838 | * Called from rate-limitted memcg_check_events when enough | |
839 | * MEM_CGROUP_TARGET_SOFTLIMIT events are accumulated and it makes sure | |
840 | * that all the parents up the hierarchy will be noticed that this group | |
841 | * is in excess or that it is not in excess anymore. mmecg->soft_contributed | |
842 | * makes the transition a single action whenever the state flips from one to | |
843 | * other. | |
844 | */ | |
845 | static void mem_cgroup_update_soft_limit(struct mem_cgroup *memcg) | |
846 | { | |
847 | unsigned long long excess = res_counter_soft_limit_excess(&memcg->res); | |
848 | struct mem_cgroup *parent = memcg; | |
849 | int delta = 0; | |
850 | ||
851 | spin_lock(&memcg->soft_lock); | |
852 | if (excess) { | |
853 | if (!memcg->soft_contributed) { | |
854 | delta = 1; | |
855 | memcg->soft_contributed = true; | |
856 | } | |
857 | } else { | |
858 | if (memcg->soft_contributed) { | |
859 | delta = -1; | |
860 | memcg->soft_contributed = false; | |
861 | } | |
862 | } | |
863 | ||
864 | /* | |
865 | * Necessary to update all ancestors when hierarchy is used | |
866 | * because their event counter is not touched. | |
867 | */ | |
868 | while (delta && (parent = parent_mem_cgroup(parent))) | |
869 | atomic_add(delta, &parent->children_in_excess); | |
870 | spin_unlock(&memcg->soft_lock); | |
871 | } | |
872 | ||
d2265e6f KH |
873 | /* |
874 | * Check events in order. | |
875 | * | |
876 | */ | |
c0ff4b85 | 877 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f | 878 | { |
4799401f | 879 | preempt_disable(); |
d2265e6f | 880 | /* threshold event is triggered in finer grain than soft limit */ |
f53d7ce3 JW |
881 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
882 | MEM_CGROUP_TARGET_THRESH))) { | |
7d910c05 | 883 | bool do_softlimit; |
82b3f2a7 | 884 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 885 | |
7d910c05 MH |
886 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
887 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
888 | #if MAX_NUMNODES > 1 |
889 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
890 | MEM_CGROUP_TARGET_NUMAINFO); | |
891 | #endif | |
892 | preempt_enable(); | |
893 | ||
c0ff4b85 | 894 | mem_cgroup_threshold(memcg); |
7d910c05 MH |
895 | if (unlikely(do_softlimit)) |
896 | mem_cgroup_update_soft_limit(memcg); | |
453a9bf3 | 897 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 898 | if (unlikely(do_numainfo)) |
c0ff4b85 | 899 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 900 | #endif |
f53d7ce3 JW |
901 | } else |
902 | preempt_enable(); | |
d2265e6f KH |
903 | } |
904 | ||
cf475ad2 | 905 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 906 | { |
31a78f23 BS |
907 | /* |
908 | * mm_update_next_owner() may clear mm->owner to NULL | |
909 | * if it races with swapoff, page migration, etc. | |
910 | * So this can be called with p == NULL. | |
911 | */ | |
912 | if (unlikely(!p)) | |
913 | return NULL; | |
914 | ||
8af01f56 | 915 | return mem_cgroup_from_css(task_css(p, mem_cgroup_subsys_id)); |
78fb7466 PE |
916 | } |
917 | ||
a433658c | 918 | struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 919 | { |
c0ff4b85 | 920 | struct mem_cgroup *memcg = NULL; |
0b7f569e KH |
921 | |
922 | if (!mm) | |
923 | return NULL; | |
54595fe2 KH |
924 | /* |
925 | * Because we have no locks, mm->owner's may be being moved to other | |
926 | * cgroup. We use css_tryget() here even if this looks | |
927 | * pessimistic (rather than adding locks here). | |
928 | */ | |
929 | rcu_read_lock(); | |
930 | do { | |
c0ff4b85 R |
931 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
932 | if (unlikely(!memcg)) | |
54595fe2 | 933 | break; |
c0ff4b85 | 934 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 935 | rcu_read_unlock(); |
c0ff4b85 | 936 | return memcg; |
54595fe2 KH |
937 | } |
938 | ||
de57780d MH |
939 | static enum mem_cgroup_filter_t |
940 | mem_cgroup_filter(struct mem_cgroup *memcg, struct mem_cgroup *root, | |
941 | mem_cgroup_iter_filter cond) | |
942 | { | |
943 | if (!cond) | |
944 | return VISIT; | |
945 | return cond(memcg, root); | |
946 | } | |
947 | ||
16248d8f MH |
948 | /* |
949 | * Returns a next (in a pre-order walk) alive memcg (with elevated css | |
950 | * ref. count) or NULL if the whole root's subtree has been visited. | |
951 | * | |
952 | * helper function to be used by mem_cgroup_iter | |
953 | */ | |
954 | static struct mem_cgroup *__mem_cgroup_iter_next(struct mem_cgroup *root, | |
de57780d | 955 | struct mem_cgroup *last_visited, mem_cgroup_iter_filter cond) |
16248d8f | 956 | { |
492eb21b | 957 | struct cgroup_subsys_state *prev_css, *next_css; |
16248d8f | 958 | |
bd8815a6 | 959 | prev_css = last_visited ? &last_visited->css : NULL; |
16248d8f | 960 | skip_node: |
492eb21b | 961 | next_css = css_next_descendant_pre(prev_css, &root->css); |
16248d8f MH |
962 | |
963 | /* | |
964 | * Even if we found a group we have to make sure it is | |
965 | * alive. css && !memcg means that the groups should be | |
966 | * skipped and we should continue the tree walk. | |
967 | * last_visited css is safe to use because it is | |
968 | * protected by css_get and the tree walk is rcu safe. | |
969 | */ | |
492eb21b TH |
970 | if (next_css) { |
971 | struct mem_cgroup *mem = mem_cgroup_from_css(next_css); | |
972 | ||
de57780d MH |
973 | switch (mem_cgroup_filter(mem, root, cond)) { |
974 | case SKIP: | |
492eb21b | 975 | prev_css = next_css; |
16248d8f | 976 | goto skip_node; |
de57780d MH |
977 | case SKIP_TREE: |
978 | if (mem == root) | |
979 | return NULL; | |
980 | /* | |
981 | * css_rightmost_descendant is not an optimal way to | |
982 | * skip through a subtree (especially for imbalanced | |
983 | * trees leaning to right) but that's what we have right | |
984 | * now. More effective solution would be traversing | |
985 | * right-up for first non-NULL without calling | |
986 | * css_next_descendant_pre afterwards. | |
987 | */ | |
988 | prev_css = css_rightmost_descendant(next_css); | |
989 | goto skip_node; | |
990 | case VISIT: | |
991 | if (css_tryget(&mem->css)) | |
992 | return mem; | |
993 | else { | |
994 | prev_css = next_css; | |
995 | goto skip_node; | |
996 | } | |
997 | break; | |
16248d8f MH |
998 | } |
999 | } | |
1000 | ||
1001 | return NULL; | |
1002 | } | |
1003 | ||
519ebea3 JW |
1004 | static void mem_cgroup_iter_invalidate(struct mem_cgroup *root) |
1005 | { | |
1006 | /* | |
1007 | * When a group in the hierarchy below root is destroyed, the | |
1008 | * hierarchy iterator can no longer be trusted since it might | |
1009 | * have pointed to the destroyed group. Invalidate it. | |
1010 | */ | |
1011 | atomic_inc(&root->dead_count); | |
1012 | } | |
1013 | ||
1014 | static struct mem_cgroup * | |
1015 | mem_cgroup_iter_load(struct mem_cgroup_reclaim_iter *iter, | |
1016 | struct mem_cgroup *root, | |
1017 | int *sequence) | |
1018 | { | |
1019 | struct mem_cgroup *position = NULL; | |
1020 | /* | |
1021 | * A cgroup destruction happens in two stages: offlining and | |
1022 | * release. They are separated by a RCU grace period. | |
1023 | * | |
1024 | * If the iterator is valid, we may still race with an | |
1025 | * offlining. The RCU lock ensures the object won't be | |
1026 | * released, tryget will fail if we lost the race. | |
1027 | */ | |
1028 | *sequence = atomic_read(&root->dead_count); | |
1029 | if (iter->last_dead_count == *sequence) { | |
1030 | smp_rmb(); | |
1031 | position = iter->last_visited; | |
1032 | if (position && !css_tryget(&position->css)) | |
1033 | position = NULL; | |
1034 | } | |
1035 | return position; | |
1036 | } | |
1037 | ||
1038 | static void mem_cgroup_iter_update(struct mem_cgroup_reclaim_iter *iter, | |
1039 | struct mem_cgroup *last_visited, | |
1040 | struct mem_cgroup *new_position, | |
1041 | int sequence) | |
1042 | { | |
1043 | if (last_visited) | |
1044 | css_put(&last_visited->css); | |
1045 | /* | |
1046 | * We store the sequence count from the time @last_visited was | |
1047 | * loaded successfully instead of rereading it here so that we | |
1048 | * don't lose destruction events in between. We could have | |
1049 | * raced with the destruction of @new_position after all. | |
1050 | */ | |
1051 | iter->last_visited = new_position; | |
1052 | smp_wmb(); | |
1053 | iter->last_dead_count = sequence; | |
1054 | } | |
1055 | ||
5660048c JW |
1056 | /** |
1057 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1058 | * @root: hierarchy root | |
1059 | * @prev: previously returned memcg, NULL on first invocation | |
1060 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
de57780d | 1061 | * @cond: filter for visited nodes, NULL for no filter |
5660048c JW |
1062 | * |
1063 | * Returns references to children of the hierarchy below @root, or | |
1064 | * @root itself, or %NULL after a full round-trip. | |
1065 | * | |
1066 | * Caller must pass the return value in @prev on subsequent | |
1067 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1068 | * to cancel a hierarchy walk before the round-trip is complete. | |
1069 | * | |
1070 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
1071 | * divide up the memcgs in the hierarchy among all concurrent | |
1072 | * reclaimers operating on the same zone and priority. | |
1073 | */ | |
de57780d | 1074 | struct mem_cgroup *mem_cgroup_iter_cond(struct mem_cgroup *root, |
5660048c | 1075 | struct mem_cgroup *prev, |
de57780d MH |
1076 | struct mem_cgroup_reclaim_cookie *reclaim, |
1077 | mem_cgroup_iter_filter cond) | |
14067bb3 | 1078 | { |
9f3a0d09 | 1079 | struct mem_cgroup *memcg = NULL; |
542f85f9 | 1080 | struct mem_cgroup *last_visited = NULL; |
711d3d2c | 1081 | |
de57780d MH |
1082 | if (mem_cgroup_disabled()) { |
1083 | /* first call must return non-NULL, second return NULL */ | |
1084 | return (struct mem_cgroup *)(unsigned long)!prev; | |
1085 | } | |
5660048c | 1086 | |
9f3a0d09 JW |
1087 | if (!root) |
1088 | root = root_mem_cgroup; | |
7d74b06f | 1089 | |
9f3a0d09 | 1090 | if (prev && !reclaim) |
542f85f9 | 1091 | last_visited = prev; |
14067bb3 | 1092 | |
9f3a0d09 JW |
1093 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1094 | if (prev) | |
c40046f3 | 1095 | goto out_css_put; |
de57780d MH |
1096 | if (mem_cgroup_filter(root, root, cond) == VISIT) |
1097 | return root; | |
1098 | return NULL; | |
9f3a0d09 | 1099 | } |
14067bb3 | 1100 | |
542f85f9 | 1101 | rcu_read_lock(); |
9f3a0d09 | 1102 | while (!memcg) { |
527a5ec9 | 1103 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
519ebea3 | 1104 | int uninitialized_var(seq); |
711d3d2c | 1105 | |
527a5ec9 JW |
1106 | if (reclaim) { |
1107 | int nid = zone_to_nid(reclaim->zone); | |
1108 | int zid = zone_idx(reclaim->zone); | |
1109 | struct mem_cgroup_per_zone *mz; | |
1110 | ||
1111 | mz = mem_cgroup_zoneinfo(root, nid, zid); | |
1112 | iter = &mz->reclaim_iter[reclaim->priority]; | |
542f85f9 | 1113 | if (prev && reclaim->generation != iter->generation) { |
5f578161 | 1114 | iter->last_visited = NULL; |
542f85f9 MH |
1115 | goto out_unlock; |
1116 | } | |
5f578161 | 1117 | |
519ebea3 | 1118 | last_visited = mem_cgroup_iter_load(iter, root, &seq); |
527a5ec9 | 1119 | } |
7d74b06f | 1120 | |
de57780d | 1121 | memcg = __mem_cgroup_iter_next(root, last_visited, cond); |
14067bb3 | 1122 | |
527a5ec9 | 1123 | if (reclaim) { |
519ebea3 | 1124 | mem_cgroup_iter_update(iter, last_visited, memcg, seq); |
542f85f9 | 1125 | |
19f39402 | 1126 | if (!memcg) |
527a5ec9 JW |
1127 | iter->generation++; |
1128 | else if (!prev && memcg) | |
1129 | reclaim->generation = iter->generation; | |
1130 | } | |
9f3a0d09 | 1131 | |
de57780d MH |
1132 | /* |
1133 | * We have finished the whole tree walk or no group has been | |
1134 | * visited because filter told us to skip the root node. | |
1135 | */ | |
1136 | if (!memcg && (prev || (cond && !last_visited))) | |
542f85f9 | 1137 | goto out_unlock; |
9f3a0d09 | 1138 | } |
542f85f9 MH |
1139 | out_unlock: |
1140 | rcu_read_unlock(); | |
c40046f3 MH |
1141 | out_css_put: |
1142 | if (prev && prev != root) | |
1143 | css_put(&prev->css); | |
1144 | ||
9f3a0d09 | 1145 | return memcg; |
14067bb3 | 1146 | } |
7d74b06f | 1147 | |
5660048c JW |
1148 | /** |
1149 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1150 | * @root: hierarchy root | |
1151 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1152 | */ | |
1153 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1154 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1155 | { |
1156 | if (!root) | |
1157 | root = root_mem_cgroup; | |
1158 | if (prev && prev != root) | |
1159 | css_put(&prev->css); | |
1160 | } | |
7d74b06f | 1161 | |
9f3a0d09 JW |
1162 | /* |
1163 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1164 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1165 | * be used for reference counting. | |
1166 | */ | |
1167 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1168 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1169 | iter != NULL; \ |
527a5ec9 | 1170 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1171 | |
9f3a0d09 | 1172 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1173 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1174 | iter != NULL; \ |
527a5ec9 | 1175 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1176 | |
68ae564b | 1177 | void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
456f998e | 1178 | { |
c0ff4b85 | 1179 | struct mem_cgroup *memcg; |
456f998e | 1180 | |
456f998e | 1181 | rcu_read_lock(); |
c0ff4b85 R |
1182 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1183 | if (unlikely(!memcg)) | |
456f998e YH |
1184 | goto out; |
1185 | ||
1186 | switch (idx) { | |
456f998e | 1187 | case PGFAULT: |
0e574a93 JW |
1188 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1189 | break; | |
1190 | case PGMAJFAULT: | |
1191 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1192 | break; |
1193 | default: | |
1194 | BUG(); | |
1195 | } | |
1196 | out: | |
1197 | rcu_read_unlock(); | |
1198 | } | |
68ae564b | 1199 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
456f998e | 1200 | |
925b7673 JW |
1201 | /** |
1202 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1203 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1204 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1205 | * |
1206 | * Returns the lru list vector holding pages for the given @zone and | |
1207 | * @mem. This can be the global zone lruvec, if the memory controller | |
1208 | * is disabled. | |
1209 | */ | |
1210 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1211 | struct mem_cgroup *memcg) | |
1212 | { | |
1213 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1214 | struct lruvec *lruvec; |
925b7673 | 1215 | |
bea8c150 HD |
1216 | if (mem_cgroup_disabled()) { |
1217 | lruvec = &zone->lruvec; | |
1218 | goto out; | |
1219 | } | |
925b7673 JW |
1220 | |
1221 | mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone)); | |
bea8c150 HD |
1222 | lruvec = &mz->lruvec; |
1223 | out: | |
1224 | /* | |
1225 | * Since a node can be onlined after the mem_cgroup was created, | |
1226 | * we have to be prepared to initialize lruvec->zone here; | |
1227 | * and if offlined then reonlined, we need to reinitialize it. | |
1228 | */ | |
1229 | if (unlikely(lruvec->zone != zone)) | |
1230 | lruvec->zone = zone; | |
1231 | return lruvec; | |
925b7673 JW |
1232 | } |
1233 | ||
08e552c6 KH |
1234 | /* |
1235 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
1236 | * Operations are called by routine of global LRU independently from memcg. | |
1237 | * What we have to take care of here is validness of pc->mem_cgroup. | |
1238 | * | |
1239 | * Changes to pc->mem_cgroup happens when | |
1240 | * 1. charge | |
1241 | * 2. moving account | |
1242 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
1243 | * It is added to LRU before charge. | |
1244 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
1245 | * When moving account, the page is not on LRU. It's isolated. | |
1246 | */ | |
4f98a2fe | 1247 | |
925b7673 | 1248 | /** |
fa9add64 | 1249 | * mem_cgroup_page_lruvec - return lruvec for adding an lru page |
925b7673 | 1250 | * @page: the page |
fa9add64 | 1251 | * @zone: zone of the page |
925b7673 | 1252 | */ |
fa9add64 | 1253 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1254 | { |
08e552c6 | 1255 | struct mem_cgroup_per_zone *mz; |
925b7673 JW |
1256 | struct mem_cgroup *memcg; |
1257 | struct page_cgroup *pc; | |
bea8c150 | 1258 | struct lruvec *lruvec; |
6d12e2d8 | 1259 | |
bea8c150 HD |
1260 | if (mem_cgroup_disabled()) { |
1261 | lruvec = &zone->lruvec; | |
1262 | goto out; | |
1263 | } | |
925b7673 | 1264 | |
08e552c6 | 1265 | pc = lookup_page_cgroup(page); |
38c5d72f | 1266 | memcg = pc->mem_cgroup; |
7512102c HD |
1267 | |
1268 | /* | |
fa9add64 | 1269 | * Surreptitiously switch any uncharged offlist page to root: |
7512102c HD |
1270 | * an uncharged page off lru does nothing to secure |
1271 | * its former mem_cgroup from sudden removal. | |
1272 | * | |
1273 | * Our caller holds lru_lock, and PageCgroupUsed is updated | |
1274 | * under page_cgroup lock: between them, they make all uses | |
1275 | * of pc->mem_cgroup safe. | |
1276 | */ | |
fa9add64 | 1277 | if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup) |
7512102c HD |
1278 | pc->mem_cgroup = memcg = root_mem_cgroup; |
1279 | ||
925b7673 | 1280 | mz = page_cgroup_zoneinfo(memcg, page); |
bea8c150 HD |
1281 | lruvec = &mz->lruvec; |
1282 | out: | |
1283 | /* | |
1284 | * Since a node can be onlined after the mem_cgroup was created, | |
1285 | * we have to be prepared to initialize lruvec->zone here; | |
1286 | * and if offlined then reonlined, we need to reinitialize it. | |
1287 | */ | |
1288 | if (unlikely(lruvec->zone != zone)) | |
1289 | lruvec->zone = zone; | |
1290 | return lruvec; | |
08e552c6 | 1291 | } |
b69408e8 | 1292 | |
925b7673 | 1293 | /** |
fa9add64 HD |
1294 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1295 | * @lruvec: mem_cgroup per zone lru vector | |
1296 | * @lru: index of lru list the page is sitting on | |
1297 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1298 | * |
fa9add64 HD |
1299 | * This function must be called when a page is added to or removed from an |
1300 | * lru list. | |
3f58a829 | 1301 | */ |
fa9add64 HD |
1302 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1303 | int nr_pages) | |
3f58a829 MK |
1304 | { |
1305 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1306 | unsigned long *lru_size; |
3f58a829 MK |
1307 | |
1308 | if (mem_cgroup_disabled()) | |
1309 | return; | |
1310 | ||
fa9add64 HD |
1311 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1312 | lru_size = mz->lru_size + lru; | |
1313 | *lru_size += nr_pages; | |
1314 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1315 | } |
544122e5 | 1316 | |
3e92041d | 1317 | /* |
c0ff4b85 | 1318 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1319 | * hierarchy subtree |
1320 | */ | |
c3ac9a8a JW |
1321 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1322 | struct mem_cgroup *memcg) | |
3e92041d | 1323 | { |
91c63734 JW |
1324 | if (root_memcg == memcg) |
1325 | return true; | |
3a981f48 | 1326 | if (!root_memcg->use_hierarchy || !memcg) |
91c63734 | 1327 | return false; |
c3ac9a8a JW |
1328 | return css_is_ancestor(&memcg->css, &root_memcg->css); |
1329 | } | |
1330 | ||
1331 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, | |
1332 | struct mem_cgroup *memcg) | |
1333 | { | |
1334 | bool ret; | |
1335 | ||
91c63734 | 1336 | rcu_read_lock(); |
c3ac9a8a | 1337 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
91c63734 JW |
1338 | rcu_read_unlock(); |
1339 | return ret; | |
3e92041d MH |
1340 | } |
1341 | ||
ffbdccf5 DR |
1342 | bool task_in_mem_cgroup(struct task_struct *task, |
1343 | const struct mem_cgroup *memcg) | |
4c4a2214 | 1344 | { |
0b7f569e | 1345 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1346 | struct task_struct *p; |
ffbdccf5 | 1347 | bool ret; |
4c4a2214 | 1348 | |
158e0a2d | 1349 | p = find_lock_task_mm(task); |
de077d22 DR |
1350 | if (p) { |
1351 | curr = try_get_mem_cgroup_from_mm(p->mm); | |
1352 | task_unlock(p); | |
1353 | } else { | |
1354 | /* | |
1355 | * All threads may have already detached their mm's, but the oom | |
1356 | * killer still needs to detect if they have already been oom | |
1357 | * killed to prevent needlessly killing additional tasks. | |
1358 | */ | |
ffbdccf5 | 1359 | rcu_read_lock(); |
de077d22 DR |
1360 | curr = mem_cgroup_from_task(task); |
1361 | if (curr) | |
1362 | css_get(&curr->css); | |
ffbdccf5 | 1363 | rcu_read_unlock(); |
de077d22 | 1364 | } |
0b7f569e | 1365 | if (!curr) |
ffbdccf5 | 1366 | return false; |
d31f56db | 1367 | /* |
c0ff4b85 | 1368 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1369 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1370 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1371 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1372 | */ |
c0ff4b85 | 1373 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1374 | css_put(&curr->css); |
4c4a2214 DR |
1375 | return ret; |
1376 | } | |
1377 | ||
c56d5c7d | 1378 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1379 | { |
9b272977 | 1380 | unsigned long inactive_ratio; |
14797e23 | 1381 | unsigned long inactive; |
9b272977 | 1382 | unsigned long active; |
c772be93 | 1383 | unsigned long gb; |
14797e23 | 1384 | |
4d7dcca2 HD |
1385 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1386 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1387 | |
c772be93 KM |
1388 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1389 | if (gb) | |
1390 | inactive_ratio = int_sqrt(10 * gb); | |
1391 | else | |
1392 | inactive_ratio = 1; | |
1393 | ||
9b272977 | 1394 | return inactive * inactive_ratio < active; |
14797e23 KM |
1395 | } |
1396 | ||
6d61ef40 BS |
1397 | #define mem_cgroup_from_res_counter(counter, member) \ |
1398 | container_of(counter, struct mem_cgroup, member) | |
1399 | ||
19942822 | 1400 | /** |
9d11ea9f | 1401 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1402 | * @memcg: the memory cgroup |
19942822 | 1403 | * |
9d11ea9f | 1404 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1405 | * pages. |
19942822 | 1406 | */ |
c0ff4b85 | 1407 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1408 | { |
9d11ea9f JW |
1409 | unsigned long long margin; |
1410 | ||
c0ff4b85 | 1411 | margin = res_counter_margin(&memcg->res); |
9d11ea9f | 1412 | if (do_swap_account) |
c0ff4b85 | 1413 | margin = min(margin, res_counter_margin(&memcg->memsw)); |
7ec99d62 | 1414 | return margin >> PAGE_SHIFT; |
19942822 JW |
1415 | } |
1416 | ||
1f4c025b | 1417 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 | 1418 | { |
a7885eb8 | 1419 | /* root ? */ |
63876986 | 1420 | if (!css_parent(&memcg->css)) |
a7885eb8 KM |
1421 | return vm_swappiness; |
1422 | ||
bf1ff263 | 1423 | return memcg->swappiness; |
a7885eb8 KM |
1424 | } |
1425 | ||
619d094b KH |
1426 | /* |
1427 | * memcg->moving_account is used for checking possibility that some thread is | |
1428 | * calling move_account(). When a thread on CPU-A starts moving pages under | |
1429 | * a memcg, other threads should check memcg->moving_account under | |
1430 | * rcu_read_lock(), like this: | |
1431 | * | |
1432 | * CPU-A CPU-B | |
1433 | * rcu_read_lock() | |
1434 | * memcg->moving_account+1 if (memcg->mocing_account) | |
1435 | * take heavy locks. | |
1436 | * synchronize_rcu() update something. | |
1437 | * rcu_read_unlock() | |
1438 | * start move here. | |
1439 | */ | |
4331f7d3 KH |
1440 | |
1441 | /* for quick checking without looking up memcg */ | |
1442 | atomic_t memcg_moving __read_mostly; | |
1443 | ||
c0ff4b85 | 1444 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a | 1445 | { |
4331f7d3 | 1446 | atomic_inc(&memcg_moving); |
619d094b | 1447 | atomic_inc(&memcg->moving_account); |
32047e2a KH |
1448 | synchronize_rcu(); |
1449 | } | |
1450 | ||
c0ff4b85 | 1451 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a | 1452 | { |
619d094b KH |
1453 | /* |
1454 | * Now, mem_cgroup_clear_mc() may call this function with NULL. | |
1455 | * We check NULL in callee rather than caller. | |
1456 | */ | |
4331f7d3 KH |
1457 | if (memcg) { |
1458 | atomic_dec(&memcg_moving); | |
619d094b | 1459 | atomic_dec(&memcg->moving_account); |
4331f7d3 | 1460 | } |
32047e2a | 1461 | } |
619d094b | 1462 | |
32047e2a KH |
1463 | /* |
1464 | * 2 routines for checking "mem" is under move_account() or not. | |
1465 | * | |
13fd1dd9 AM |
1466 | * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This |
1467 | * is used for avoiding races in accounting. If true, | |
32047e2a KH |
1468 | * pc->mem_cgroup may be overwritten. |
1469 | * | |
1470 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or | |
1471 | * under hierarchy of moving cgroups. This is for | |
1472 | * waiting at hith-memory prressure caused by "move". | |
1473 | */ | |
1474 | ||
13fd1dd9 | 1475 | static bool mem_cgroup_stolen(struct mem_cgroup *memcg) |
32047e2a KH |
1476 | { |
1477 | VM_BUG_ON(!rcu_read_lock_held()); | |
619d094b | 1478 | return atomic_read(&memcg->moving_account) > 0; |
32047e2a | 1479 | } |
4b534334 | 1480 | |
c0ff4b85 | 1481 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1482 | { |
2bd9bb20 KH |
1483 | struct mem_cgroup *from; |
1484 | struct mem_cgroup *to; | |
4b534334 | 1485 | bool ret = false; |
2bd9bb20 KH |
1486 | /* |
1487 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1488 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1489 | */ | |
1490 | spin_lock(&mc.lock); | |
1491 | from = mc.from; | |
1492 | to = mc.to; | |
1493 | if (!from) | |
1494 | goto unlock; | |
3e92041d | 1495 | |
c0ff4b85 R |
1496 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1497 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1498 | unlock: |
1499 | spin_unlock(&mc.lock); | |
4b534334 KH |
1500 | return ret; |
1501 | } | |
1502 | ||
c0ff4b85 | 1503 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1504 | { |
1505 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1506 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1507 | DEFINE_WAIT(wait); |
1508 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1509 | /* moving charge context might have finished. */ | |
1510 | if (mc.moving_task) | |
1511 | schedule(); | |
1512 | finish_wait(&mc.waitq, &wait); | |
1513 | return true; | |
1514 | } | |
1515 | } | |
1516 | return false; | |
1517 | } | |
1518 | ||
312734c0 KH |
1519 | /* |
1520 | * Take this lock when | |
1521 | * - a code tries to modify page's memcg while it's USED. | |
1522 | * - a code tries to modify page state accounting in a memcg. | |
13fd1dd9 | 1523 | * see mem_cgroup_stolen(), too. |
312734c0 KH |
1524 | */ |
1525 | static void move_lock_mem_cgroup(struct mem_cgroup *memcg, | |
1526 | unsigned long *flags) | |
1527 | { | |
1528 | spin_lock_irqsave(&memcg->move_lock, *flags); | |
1529 | } | |
1530 | ||
1531 | static void move_unlock_mem_cgroup(struct mem_cgroup *memcg, | |
1532 | unsigned long *flags) | |
1533 | { | |
1534 | spin_unlock_irqrestore(&memcg->move_lock, *flags); | |
1535 | } | |
1536 | ||
58cf188e | 1537 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1538 | /** |
58cf188e | 1539 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1540 | * @memcg: The memory cgroup that went over limit |
1541 | * @p: Task that is going to be killed | |
1542 | * | |
1543 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1544 | * enabled | |
1545 | */ | |
1546 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1547 | { | |
1548 | struct cgroup *task_cgrp; | |
1549 | struct cgroup *mem_cgrp; | |
1550 | /* | |
1551 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1552 | * on the assumption that OOM is serialized for memory controller. | |
1553 | * If this assumption is broken, revisit this code. | |
1554 | */ | |
1555 | static char memcg_name[PATH_MAX]; | |
1556 | int ret; | |
58cf188e SZ |
1557 | struct mem_cgroup *iter; |
1558 | unsigned int i; | |
e222432b | 1559 | |
58cf188e | 1560 | if (!p) |
e222432b BS |
1561 | return; |
1562 | ||
e222432b BS |
1563 | rcu_read_lock(); |
1564 | ||
1565 | mem_cgrp = memcg->css.cgroup; | |
1566 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1567 | ||
1568 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1569 | if (ret < 0) { | |
1570 | /* | |
1571 | * Unfortunately, we are unable to convert to a useful name | |
1572 | * But we'll still print out the usage information | |
1573 | */ | |
1574 | rcu_read_unlock(); | |
1575 | goto done; | |
1576 | } | |
1577 | rcu_read_unlock(); | |
1578 | ||
d045197f | 1579 | pr_info("Task in %s killed", memcg_name); |
e222432b BS |
1580 | |
1581 | rcu_read_lock(); | |
1582 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1583 | if (ret < 0) { | |
1584 | rcu_read_unlock(); | |
1585 | goto done; | |
1586 | } | |
1587 | rcu_read_unlock(); | |
1588 | ||
1589 | /* | |
1590 | * Continues from above, so we don't need an KERN_ level | |
1591 | */ | |
d045197f | 1592 | pr_cont(" as a result of limit of %s\n", memcg_name); |
e222432b BS |
1593 | done: |
1594 | ||
d045197f | 1595 | pr_info("memory: usage %llukB, limit %llukB, failcnt %llu\n", |
e222432b BS |
1596 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, |
1597 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1598 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
d045197f | 1599 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %llu\n", |
e222432b BS |
1600 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, |
1601 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1602 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
d045197f | 1603 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %llu\n", |
510fc4e1 GC |
1604 | res_counter_read_u64(&memcg->kmem, RES_USAGE) >> 10, |
1605 | res_counter_read_u64(&memcg->kmem, RES_LIMIT) >> 10, | |
1606 | res_counter_read_u64(&memcg->kmem, RES_FAILCNT)); | |
58cf188e SZ |
1607 | |
1608 | for_each_mem_cgroup_tree(iter, memcg) { | |
1609 | pr_info("Memory cgroup stats"); | |
1610 | ||
1611 | rcu_read_lock(); | |
1612 | ret = cgroup_path(iter->css.cgroup, memcg_name, PATH_MAX); | |
1613 | if (!ret) | |
1614 | pr_cont(" for %s", memcg_name); | |
1615 | rcu_read_unlock(); | |
1616 | pr_cont(":"); | |
1617 | ||
1618 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1619 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1620 | continue; | |
1621 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], | |
1622 | K(mem_cgroup_read_stat(iter, i))); | |
1623 | } | |
1624 | ||
1625 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1626 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1627 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1628 | ||
1629 | pr_cont("\n"); | |
1630 | } | |
e222432b BS |
1631 | } |
1632 | ||
81d39c20 KH |
1633 | /* |
1634 | * This function returns the number of memcg under hierarchy tree. Returns | |
1635 | * 1(self count) if no children. | |
1636 | */ | |
c0ff4b85 | 1637 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1638 | { |
1639 | int num = 0; | |
7d74b06f KH |
1640 | struct mem_cgroup *iter; |
1641 | ||
c0ff4b85 | 1642 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1643 | num++; |
81d39c20 KH |
1644 | return num; |
1645 | } | |
1646 | ||
a63d83f4 DR |
1647 | /* |
1648 | * Return the memory (and swap, if configured) limit for a memcg. | |
1649 | */ | |
9cbb78bb | 1650 | static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 DR |
1651 | { |
1652 | u64 limit; | |
a63d83f4 | 1653 | |
f3e8eb70 | 1654 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
f3e8eb70 | 1655 | |
a63d83f4 | 1656 | /* |
9a5a8f19 | 1657 | * Do not consider swap space if we cannot swap due to swappiness |
a63d83f4 | 1658 | */ |
9a5a8f19 MH |
1659 | if (mem_cgroup_swappiness(memcg)) { |
1660 | u64 memsw; | |
1661 | ||
1662 | limit += total_swap_pages << PAGE_SHIFT; | |
1663 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1664 | ||
1665 | /* | |
1666 | * If memsw is finite and limits the amount of swap space | |
1667 | * available to this memcg, return that limit. | |
1668 | */ | |
1669 | limit = min(limit, memsw); | |
1670 | } | |
1671 | ||
1672 | return limit; | |
a63d83f4 DR |
1673 | } |
1674 | ||
19965460 DR |
1675 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1676 | int order) | |
9cbb78bb DR |
1677 | { |
1678 | struct mem_cgroup *iter; | |
1679 | unsigned long chosen_points = 0; | |
1680 | unsigned long totalpages; | |
1681 | unsigned int points = 0; | |
1682 | struct task_struct *chosen = NULL; | |
1683 | ||
876aafbf | 1684 | /* |
465adcf1 DR |
1685 | * If current has a pending SIGKILL or is exiting, then automatically |
1686 | * select it. The goal is to allow it to allocate so that it may | |
1687 | * quickly exit and free its memory. | |
876aafbf | 1688 | */ |
465adcf1 | 1689 | if (fatal_signal_pending(current) || current->flags & PF_EXITING) { |
876aafbf DR |
1690 | set_thread_flag(TIF_MEMDIE); |
1691 | return; | |
1692 | } | |
1693 | ||
1694 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
9cbb78bb DR |
1695 | totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1; |
1696 | for_each_mem_cgroup_tree(iter, memcg) { | |
72ec7029 | 1697 | struct css_task_iter it; |
9cbb78bb DR |
1698 | struct task_struct *task; |
1699 | ||
72ec7029 TH |
1700 | css_task_iter_start(&iter->css, &it); |
1701 | while ((task = css_task_iter_next(&it))) { | |
9cbb78bb DR |
1702 | switch (oom_scan_process_thread(task, totalpages, NULL, |
1703 | false)) { | |
1704 | case OOM_SCAN_SELECT: | |
1705 | if (chosen) | |
1706 | put_task_struct(chosen); | |
1707 | chosen = task; | |
1708 | chosen_points = ULONG_MAX; | |
1709 | get_task_struct(chosen); | |
1710 | /* fall through */ | |
1711 | case OOM_SCAN_CONTINUE: | |
1712 | continue; | |
1713 | case OOM_SCAN_ABORT: | |
72ec7029 | 1714 | css_task_iter_end(&it); |
9cbb78bb DR |
1715 | mem_cgroup_iter_break(memcg, iter); |
1716 | if (chosen) | |
1717 | put_task_struct(chosen); | |
1718 | return; | |
1719 | case OOM_SCAN_OK: | |
1720 | break; | |
1721 | }; | |
1722 | points = oom_badness(task, memcg, NULL, totalpages); | |
1723 | if (points > chosen_points) { | |
1724 | if (chosen) | |
1725 | put_task_struct(chosen); | |
1726 | chosen = task; | |
1727 | chosen_points = points; | |
1728 | get_task_struct(chosen); | |
1729 | } | |
1730 | } | |
72ec7029 | 1731 | css_task_iter_end(&it); |
9cbb78bb DR |
1732 | } |
1733 | ||
1734 | if (!chosen) | |
1735 | return; | |
1736 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1737 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1738 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1739 | } |
1740 | ||
5660048c JW |
1741 | static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg, |
1742 | gfp_t gfp_mask, | |
1743 | unsigned long flags) | |
1744 | { | |
1745 | unsigned long total = 0; | |
1746 | bool noswap = false; | |
1747 | int loop; | |
1748 | ||
1749 | if (flags & MEM_CGROUP_RECLAIM_NOSWAP) | |
1750 | noswap = true; | |
1751 | if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum) | |
1752 | noswap = true; | |
1753 | ||
1754 | for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) { | |
1755 | if (loop) | |
1756 | drain_all_stock_async(memcg); | |
1757 | total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap); | |
1758 | /* | |
1759 | * Allow limit shrinkers, which are triggered directly | |
1760 | * by userspace, to catch signals and stop reclaim | |
1761 | * after minimal progress, regardless of the margin. | |
1762 | */ | |
1763 | if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK)) | |
1764 | break; | |
1765 | if (mem_cgroup_margin(memcg)) | |
1766 | break; | |
1767 | /* | |
1768 | * If nothing was reclaimed after two attempts, there | |
1769 | * may be no reclaimable pages in this hierarchy. | |
1770 | */ | |
1771 | if (loop && !total) | |
1772 | break; | |
1773 | } | |
1774 | return total; | |
1775 | } | |
1776 | ||
e883110a | 1777 | #if MAX_NUMNODES > 1 |
4d0c066d KH |
1778 | /** |
1779 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1780 | * @memcg: the target memcg |
4d0c066d KH |
1781 | * @nid: the node ID to be checked. |
1782 | * @noswap : specify true here if the user wants flle only information. | |
1783 | * | |
1784 | * This function returns whether the specified memcg contains any | |
1785 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1786 | * pages in the node. | |
1787 | */ | |
c0ff4b85 | 1788 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1789 | int nid, bool noswap) |
1790 | { | |
c0ff4b85 | 1791 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1792 | return true; |
1793 | if (noswap || !total_swap_pages) | |
1794 | return false; | |
c0ff4b85 | 1795 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1796 | return true; |
1797 | return false; | |
1798 | ||
1799 | } | |
889976db YH |
1800 | |
1801 | /* | |
1802 | * Always updating the nodemask is not very good - even if we have an empty | |
1803 | * list or the wrong list here, we can start from some node and traverse all | |
1804 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1805 | * | |
1806 | */ | |
c0ff4b85 | 1807 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1808 | { |
1809 | int nid; | |
453a9bf3 KH |
1810 | /* |
1811 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1812 | * pagein/pageout changes since the last update. | |
1813 | */ | |
c0ff4b85 | 1814 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1815 | return; |
c0ff4b85 | 1816 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1817 | return; |
1818 | ||
889976db | 1819 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1820 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1821 | |
31aaea4a | 1822 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1823 | |
c0ff4b85 R |
1824 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1825 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1826 | } |
453a9bf3 | 1827 | |
c0ff4b85 R |
1828 | atomic_set(&memcg->numainfo_events, 0); |
1829 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1830 | } |
1831 | ||
1832 | /* | |
1833 | * Selecting a node where we start reclaim from. Because what we need is just | |
1834 | * reducing usage counter, start from anywhere is O,K. Considering | |
1835 | * memory reclaim from current node, there are pros. and cons. | |
1836 | * | |
1837 | * Freeing memory from current node means freeing memory from a node which | |
1838 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1839 | * hit limits, it will see a contention on a node. But freeing from remote | |
1840 | * node means more costs for memory reclaim because of memory latency. | |
1841 | * | |
1842 | * Now, we use round-robin. Better algorithm is welcomed. | |
1843 | */ | |
c0ff4b85 | 1844 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1845 | { |
1846 | int node; | |
1847 | ||
c0ff4b85 R |
1848 | mem_cgroup_may_update_nodemask(memcg); |
1849 | node = memcg->last_scanned_node; | |
889976db | 1850 | |
c0ff4b85 | 1851 | node = next_node(node, memcg->scan_nodes); |
889976db | 1852 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1853 | node = first_node(memcg->scan_nodes); |
889976db YH |
1854 | /* |
1855 | * We call this when we hit limit, not when pages are added to LRU. | |
1856 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1857 | * memcg is too small and all pages are not on LRU. In that case, | |
1858 | * we use curret node. | |
1859 | */ | |
1860 | if (unlikely(node == MAX_NUMNODES)) | |
1861 | node = numa_node_id(); | |
1862 | ||
c0ff4b85 | 1863 | memcg->last_scanned_node = node; |
889976db YH |
1864 | return node; |
1865 | } | |
1866 | ||
1867 | #else | |
c0ff4b85 | 1868 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1869 | { |
1870 | return 0; | |
1871 | } | |
4d0c066d | 1872 | |
889976db YH |
1873 | #endif |
1874 | ||
3b38722e | 1875 | /* |
a5b7c87f MH |
1876 | * A group is eligible for the soft limit reclaim under the given root |
1877 | * hierarchy if | |
1878 | * a) it is over its soft limit | |
3b38722e | 1879 | * b) any parent up the hierarchy is over its soft limit |
7d910c05 MH |
1880 | * |
1881 | * If the given group doesn't have any children over the limit then it | |
1882 | * doesn't make any sense to iterate its subtree. | |
3b38722e | 1883 | */ |
de57780d MH |
1884 | enum mem_cgroup_filter_t |
1885 | mem_cgroup_soft_reclaim_eligible(struct mem_cgroup *memcg, | |
a5b7c87f | 1886 | struct mem_cgroup *root) |
3b38722e MH |
1887 | { |
1888 | struct mem_cgroup *parent = memcg; | |
1889 | ||
1890 | if (res_counter_soft_limit_excess(&memcg->res)) | |
de57780d | 1891 | return VISIT; |
3b38722e MH |
1892 | |
1893 | /* | |
a5b7c87f MH |
1894 | * If any parent up to the root in the hierarchy is over its soft limit |
1895 | * then we have to obey and reclaim from this group as well. | |
3b38722e MH |
1896 | */ |
1897 | while((parent = parent_mem_cgroup(parent))) { | |
1898 | if (res_counter_soft_limit_excess(&parent->res)) | |
de57780d | 1899 | return VISIT; |
a5b7c87f MH |
1900 | if (parent == root) |
1901 | break; | |
6d61ef40 | 1902 | } |
3b38722e | 1903 | |
7d910c05 MH |
1904 | if (!atomic_read(&memcg->children_in_excess)) |
1905 | return SKIP_TREE; | |
de57780d | 1906 | return SKIP; |
6d61ef40 BS |
1907 | } |
1908 | ||
867578cb KH |
1909 | /* |
1910 | * Check OOM-Killer is already running under our hierarchy. | |
1911 | * If someone is running, return false. | |
1af8efe9 | 1912 | * Has to be called with memcg_oom_lock |
867578cb | 1913 | */ |
c0ff4b85 | 1914 | static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg) |
867578cb | 1915 | { |
79dfdacc | 1916 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1917 | |
9f3a0d09 | 1918 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1919 | if (iter->oom_lock) { |
79dfdacc MH |
1920 | /* |
1921 | * this subtree of our hierarchy is already locked | |
1922 | * so we cannot give a lock. | |
1923 | */ | |
79dfdacc | 1924 | failed = iter; |
9f3a0d09 JW |
1925 | mem_cgroup_iter_break(memcg, iter); |
1926 | break; | |
23751be0 JW |
1927 | } else |
1928 | iter->oom_lock = true; | |
7d74b06f | 1929 | } |
867578cb | 1930 | |
79dfdacc | 1931 | if (!failed) |
23751be0 | 1932 | return true; |
79dfdacc MH |
1933 | |
1934 | /* | |
1935 | * OK, we failed to lock the whole subtree so we have to clean up | |
1936 | * what we set up to the failing subtree | |
1937 | */ | |
9f3a0d09 | 1938 | for_each_mem_cgroup_tree(iter, memcg) { |
79dfdacc | 1939 | if (iter == failed) { |
9f3a0d09 JW |
1940 | mem_cgroup_iter_break(memcg, iter); |
1941 | break; | |
79dfdacc MH |
1942 | } |
1943 | iter->oom_lock = false; | |
1944 | } | |
23751be0 | 1945 | return false; |
a636b327 | 1946 | } |
0b7f569e | 1947 | |
79dfdacc | 1948 | /* |
1af8efe9 | 1949 | * Has to be called with memcg_oom_lock |
79dfdacc | 1950 | */ |
c0ff4b85 | 1951 | static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1952 | { |
7d74b06f KH |
1953 | struct mem_cgroup *iter; |
1954 | ||
c0ff4b85 | 1955 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1956 | iter->oom_lock = false; |
1957 | return 0; | |
1958 | } | |
1959 | ||
c0ff4b85 | 1960 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1961 | { |
1962 | struct mem_cgroup *iter; | |
1963 | ||
c0ff4b85 | 1964 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1965 | atomic_inc(&iter->under_oom); |
1966 | } | |
1967 | ||
c0ff4b85 | 1968 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1969 | { |
1970 | struct mem_cgroup *iter; | |
1971 | ||
867578cb KH |
1972 | /* |
1973 | * When a new child is created while the hierarchy is under oom, | |
1974 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1975 | * atomic_add_unless() here. | |
1976 | */ | |
c0ff4b85 | 1977 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1978 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
1979 | } |
1980 | ||
1af8efe9 | 1981 | static DEFINE_SPINLOCK(memcg_oom_lock); |
867578cb KH |
1982 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1983 | ||
dc98df5a | 1984 | struct oom_wait_info { |
d79154bb | 1985 | struct mem_cgroup *memcg; |
dc98df5a KH |
1986 | wait_queue_t wait; |
1987 | }; | |
1988 | ||
1989 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1990 | unsigned mode, int sync, void *arg) | |
1991 | { | |
d79154bb HD |
1992 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1993 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1994 | struct oom_wait_info *oom_wait_info; |
1995 | ||
1996 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1997 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1998 | |
dc98df5a | 1999 | /* |
d79154bb | 2000 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
dc98df5a KH |
2001 | * Then we can use css_is_ancestor without taking care of RCU. |
2002 | */ | |
c0ff4b85 R |
2003 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
2004 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 2005 | return 0; |
dc98df5a KH |
2006 | return autoremove_wake_function(wait, mode, sync, arg); |
2007 | } | |
2008 | ||
c0ff4b85 | 2009 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 2010 | { |
c0ff4b85 R |
2011 | /* for filtering, pass "memcg" as argument. */ |
2012 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
2013 | } |
2014 | ||
c0ff4b85 | 2015 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 2016 | { |
c0ff4b85 R |
2017 | if (memcg && atomic_read(&memcg->under_oom)) |
2018 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
2019 | } |
2020 | ||
867578cb KH |
2021 | /* |
2022 | * try to call OOM killer. returns false if we should exit memory-reclaim loop. | |
2023 | */ | |
6bbda35c KS |
2024 | static bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask, |
2025 | int order) | |
0b7f569e | 2026 | { |
dc98df5a | 2027 | struct oom_wait_info owait; |
3c11ecf4 | 2028 | bool locked, need_to_kill; |
867578cb | 2029 | |
d79154bb | 2030 | owait.memcg = memcg; |
dc98df5a KH |
2031 | owait.wait.flags = 0; |
2032 | owait.wait.func = memcg_oom_wake_function; | |
2033 | owait.wait.private = current; | |
2034 | INIT_LIST_HEAD(&owait.wait.task_list); | |
3c11ecf4 | 2035 | need_to_kill = true; |
c0ff4b85 | 2036 | mem_cgroup_mark_under_oom(memcg); |
79dfdacc | 2037 | |
c0ff4b85 | 2038 | /* At first, try to OOM lock hierarchy under memcg.*/ |
1af8efe9 | 2039 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 2040 | locked = mem_cgroup_oom_lock(memcg); |
867578cb KH |
2041 | /* |
2042 | * Even if signal_pending(), we can't quit charge() loop without | |
2043 | * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL | |
2044 | * under OOM is always welcomed, use TASK_KILLABLE here. | |
2045 | */ | |
3c11ecf4 | 2046 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
c0ff4b85 | 2047 | if (!locked || memcg->oom_kill_disable) |
3c11ecf4 KH |
2048 | need_to_kill = false; |
2049 | if (locked) | |
c0ff4b85 | 2050 | mem_cgroup_oom_notify(memcg); |
1af8efe9 | 2051 | spin_unlock(&memcg_oom_lock); |
867578cb | 2052 | |
3c11ecf4 KH |
2053 | if (need_to_kill) { |
2054 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
e845e199 | 2055 | mem_cgroup_out_of_memory(memcg, mask, order); |
3c11ecf4 | 2056 | } else { |
867578cb | 2057 | schedule(); |
dc98df5a | 2058 | finish_wait(&memcg_oom_waitq, &owait.wait); |
867578cb | 2059 | } |
1af8efe9 | 2060 | spin_lock(&memcg_oom_lock); |
79dfdacc | 2061 | if (locked) |
c0ff4b85 R |
2062 | mem_cgroup_oom_unlock(memcg); |
2063 | memcg_wakeup_oom(memcg); | |
1af8efe9 | 2064 | spin_unlock(&memcg_oom_lock); |
867578cb | 2065 | |
c0ff4b85 | 2066 | mem_cgroup_unmark_under_oom(memcg); |
79dfdacc | 2067 | |
867578cb KH |
2068 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) |
2069 | return false; | |
2070 | /* Give chance to dying process */ | |
715a5ee8 | 2071 | schedule_timeout_uninterruptible(1); |
867578cb | 2072 | return true; |
0b7f569e KH |
2073 | } |
2074 | ||
d69b042f BS |
2075 | /* |
2076 | * Currently used to update mapped file statistics, but the routine can be | |
2077 | * generalized to update other statistics as well. | |
32047e2a KH |
2078 | * |
2079 | * Notes: Race condition | |
2080 | * | |
2081 | * We usually use page_cgroup_lock() for accessing page_cgroup member but | |
2082 | * it tends to be costly. But considering some conditions, we doesn't need | |
2083 | * to do so _always_. | |
2084 | * | |
2085 | * Considering "charge", lock_page_cgroup() is not required because all | |
2086 | * file-stat operations happen after a page is attached to radix-tree. There | |
2087 | * are no race with "charge". | |
2088 | * | |
2089 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
2090 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
2091 | * if there are race with "uncharge". Statistics itself is properly handled | |
2092 | * by flags. | |
2093 | * | |
2094 | * Considering "move", this is an only case we see a race. To make the race | |
619d094b KH |
2095 | * small, we check mm->moving_account and detect there are possibility of race |
2096 | * If there is, we take a lock. | |
d69b042f | 2097 | */ |
26174efd | 2098 | |
89c06bd5 KH |
2099 | void __mem_cgroup_begin_update_page_stat(struct page *page, |
2100 | bool *locked, unsigned long *flags) | |
2101 | { | |
2102 | struct mem_cgroup *memcg; | |
2103 | struct page_cgroup *pc; | |
2104 | ||
2105 | pc = lookup_page_cgroup(page); | |
2106 | again: | |
2107 | memcg = pc->mem_cgroup; | |
2108 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
2109 | return; | |
2110 | /* | |
2111 | * If this memory cgroup is not under account moving, we don't | |
da92c47d | 2112 | * need to take move_lock_mem_cgroup(). Because we already hold |
89c06bd5 | 2113 | * rcu_read_lock(), any calls to move_account will be delayed until |
13fd1dd9 | 2114 | * rcu_read_unlock() if mem_cgroup_stolen() == true. |
89c06bd5 | 2115 | */ |
13fd1dd9 | 2116 | if (!mem_cgroup_stolen(memcg)) |
89c06bd5 KH |
2117 | return; |
2118 | ||
2119 | move_lock_mem_cgroup(memcg, flags); | |
2120 | if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) { | |
2121 | move_unlock_mem_cgroup(memcg, flags); | |
2122 | goto again; | |
2123 | } | |
2124 | *locked = true; | |
2125 | } | |
2126 | ||
2127 | void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags) | |
2128 | { | |
2129 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
2130 | ||
2131 | /* | |
2132 | * It's guaranteed that pc->mem_cgroup never changes while | |
2133 | * lock is held because a routine modifies pc->mem_cgroup | |
da92c47d | 2134 | * should take move_lock_mem_cgroup(). |
89c06bd5 KH |
2135 | */ |
2136 | move_unlock_mem_cgroup(pc->mem_cgroup, flags); | |
2137 | } | |
2138 | ||
2a7106f2 GT |
2139 | void mem_cgroup_update_page_stat(struct page *page, |
2140 | enum mem_cgroup_page_stat_item idx, int val) | |
d69b042f | 2141 | { |
c0ff4b85 | 2142 | struct mem_cgroup *memcg; |
32047e2a | 2143 | struct page_cgroup *pc = lookup_page_cgroup(page); |
dbd4ea78 | 2144 | unsigned long uninitialized_var(flags); |
d69b042f | 2145 | |
cfa44946 | 2146 | if (mem_cgroup_disabled()) |
d69b042f | 2147 | return; |
89c06bd5 | 2148 | |
c0ff4b85 R |
2149 | memcg = pc->mem_cgroup; |
2150 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
89c06bd5 | 2151 | return; |
26174efd | 2152 | |
26174efd | 2153 | switch (idx) { |
2a7106f2 | 2154 | case MEMCG_NR_FILE_MAPPED: |
2a7106f2 | 2155 | idx = MEM_CGROUP_STAT_FILE_MAPPED; |
26174efd KH |
2156 | break; |
2157 | default: | |
2158 | BUG(); | |
8725d541 | 2159 | } |
d69b042f | 2160 | |
c0ff4b85 | 2161 | this_cpu_add(memcg->stat->count[idx], val); |
d69b042f | 2162 | } |
26174efd | 2163 | |
cdec2e42 KH |
2164 | /* |
2165 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2166 | * TODO: maybe necessary to use big numbers in big irons. | |
2167 | */ | |
7ec99d62 | 2168 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2169 | struct memcg_stock_pcp { |
2170 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2171 | unsigned int nr_pages; |
cdec2e42 | 2172 | struct work_struct work; |
26fe6168 | 2173 | unsigned long flags; |
a0db00fc | 2174 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2175 | }; |
2176 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2177 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2178 | |
a0956d54 SS |
2179 | /** |
2180 | * consume_stock: Try to consume stocked charge on this cpu. | |
2181 | * @memcg: memcg to consume from. | |
2182 | * @nr_pages: how many pages to charge. | |
2183 | * | |
2184 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2185 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2186 | * service an allocation will refill the stock. | |
2187 | * | |
2188 | * returns true if successful, false otherwise. | |
cdec2e42 | 2189 | */ |
a0956d54 | 2190 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2191 | { |
2192 | struct memcg_stock_pcp *stock; | |
2193 | bool ret = true; | |
2194 | ||
a0956d54 SS |
2195 | if (nr_pages > CHARGE_BATCH) |
2196 | return false; | |
2197 | ||
cdec2e42 | 2198 | stock = &get_cpu_var(memcg_stock); |
a0956d54 SS |
2199 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) |
2200 | stock->nr_pages -= nr_pages; | |
cdec2e42 KH |
2201 | else /* need to call res_counter_charge */ |
2202 | ret = false; | |
2203 | put_cpu_var(memcg_stock); | |
2204 | return ret; | |
2205 | } | |
2206 | ||
2207 | /* | |
2208 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
2209 | */ | |
2210 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2211 | { | |
2212 | struct mem_cgroup *old = stock->cached; | |
2213 | ||
11c9ea4e JW |
2214 | if (stock->nr_pages) { |
2215 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
2216 | ||
2217 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 2218 | if (do_swap_account) |
11c9ea4e JW |
2219 | res_counter_uncharge(&old->memsw, bytes); |
2220 | stock->nr_pages = 0; | |
cdec2e42 KH |
2221 | } |
2222 | stock->cached = NULL; | |
cdec2e42 KH |
2223 | } |
2224 | ||
2225 | /* | |
2226 | * This must be called under preempt disabled or must be called by | |
2227 | * a thread which is pinned to local cpu. | |
2228 | */ | |
2229 | static void drain_local_stock(struct work_struct *dummy) | |
2230 | { | |
2231 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
2232 | drain_stock(stock); | |
26fe6168 | 2233 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2234 | } |
2235 | ||
e4777496 MH |
2236 | static void __init memcg_stock_init(void) |
2237 | { | |
2238 | int cpu; | |
2239 | ||
2240 | for_each_possible_cpu(cpu) { | |
2241 | struct memcg_stock_pcp *stock = | |
2242 | &per_cpu(memcg_stock, cpu); | |
2243 | INIT_WORK(&stock->work, drain_local_stock); | |
2244 | } | |
2245 | } | |
2246 | ||
cdec2e42 KH |
2247 | /* |
2248 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 2249 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2250 | */ |
c0ff4b85 | 2251 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2252 | { |
2253 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2254 | ||
c0ff4b85 | 2255 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2256 | drain_stock(stock); |
c0ff4b85 | 2257 | stock->cached = memcg; |
cdec2e42 | 2258 | } |
11c9ea4e | 2259 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2260 | put_cpu_var(memcg_stock); |
2261 | } | |
2262 | ||
2263 | /* | |
c0ff4b85 | 2264 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
d38144b7 MH |
2265 | * of the hierarchy under it. sync flag says whether we should block |
2266 | * until the work is done. | |
cdec2e42 | 2267 | */ |
c0ff4b85 | 2268 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
cdec2e42 | 2269 | { |
26fe6168 | 2270 | int cpu, curcpu; |
d38144b7 | 2271 | |
cdec2e42 | 2272 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2273 | get_online_cpus(); |
5af12d0e | 2274 | curcpu = get_cpu(); |
cdec2e42 KH |
2275 | for_each_online_cpu(cpu) { |
2276 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2277 | struct mem_cgroup *memcg; |
26fe6168 | 2278 | |
c0ff4b85 R |
2279 | memcg = stock->cached; |
2280 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2281 | continue; |
c0ff4b85 | 2282 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2283 | continue; |
d1a05b69 MH |
2284 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2285 | if (cpu == curcpu) | |
2286 | drain_local_stock(&stock->work); | |
2287 | else | |
2288 | schedule_work_on(cpu, &stock->work); | |
2289 | } | |
cdec2e42 | 2290 | } |
5af12d0e | 2291 | put_cpu(); |
d38144b7 MH |
2292 | |
2293 | if (!sync) | |
2294 | goto out; | |
2295 | ||
2296 | for_each_online_cpu(cpu) { | |
2297 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2298 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2299 | flush_work(&stock->work); |
2300 | } | |
2301 | out: | |
cdec2e42 | 2302 | put_online_cpus(); |
d38144b7 MH |
2303 | } |
2304 | ||
2305 | /* | |
2306 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2307 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
2308 | * expects some charges will be back to res_counter later but cannot wait for | |
2309 | * it. | |
2310 | */ | |
c0ff4b85 | 2311 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
d38144b7 | 2312 | { |
9f50fad6 MH |
2313 | /* |
2314 | * If someone calls draining, avoid adding more kworker runs. | |
2315 | */ | |
2316 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2317 | return; | |
c0ff4b85 | 2318 | drain_all_stock(root_memcg, false); |
9f50fad6 | 2319 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2320 | } |
2321 | ||
2322 | /* This is a synchronous drain interface. */ | |
c0ff4b85 | 2323 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
cdec2e42 KH |
2324 | { |
2325 | /* called when force_empty is called */ | |
9f50fad6 | 2326 | mutex_lock(&percpu_charge_mutex); |
c0ff4b85 | 2327 | drain_all_stock(root_memcg, true); |
9f50fad6 | 2328 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2329 | } |
2330 | ||
711d3d2c KH |
2331 | /* |
2332 | * This function drains percpu counter value from DEAD cpu and | |
2333 | * move it to local cpu. Note that this function can be preempted. | |
2334 | */ | |
c0ff4b85 | 2335 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2336 | { |
2337 | int i; | |
2338 | ||
c0ff4b85 | 2339 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2340 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2341 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2342 | |
c0ff4b85 R |
2343 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2344 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2345 | } |
e9f8974f | 2346 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2347 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2348 | |
c0ff4b85 R |
2349 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2350 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2351 | } |
c0ff4b85 | 2352 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2353 | } |
2354 | ||
0db0628d | 2355 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
2356 | unsigned long action, |
2357 | void *hcpu) | |
2358 | { | |
2359 | int cpu = (unsigned long)hcpu; | |
2360 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2361 | struct mem_cgroup *iter; |
cdec2e42 | 2362 | |
619d094b | 2363 | if (action == CPU_ONLINE) |
1489ebad | 2364 | return NOTIFY_OK; |
1489ebad | 2365 | |
d833049b | 2366 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2367 | return NOTIFY_OK; |
711d3d2c | 2368 | |
9f3a0d09 | 2369 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2370 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2371 | ||
cdec2e42 KH |
2372 | stock = &per_cpu(memcg_stock, cpu); |
2373 | drain_stock(stock); | |
2374 | return NOTIFY_OK; | |
2375 | } | |
2376 | ||
4b534334 KH |
2377 | |
2378 | /* See __mem_cgroup_try_charge() for details */ | |
2379 | enum { | |
2380 | CHARGE_OK, /* success */ | |
2381 | CHARGE_RETRY, /* need to retry but retry is not bad */ | |
2382 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ | |
2383 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ | |
2384 | CHARGE_OOM_DIE, /* the current is killed because of OOM */ | |
2385 | }; | |
2386 | ||
c0ff4b85 | 2387 | static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
4c9c5359 SS |
2388 | unsigned int nr_pages, unsigned int min_pages, |
2389 | bool oom_check) | |
4b534334 | 2390 | { |
7ec99d62 | 2391 | unsigned long csize = nr_pages * PAGE_SIZE; |
4b534334 KH |
2392 | struct mem_cgroup *mem_over_limit; |
2393 | struct res_counter *fail_res; | |
2394 | unsigned long flags = 0; | |
2395 | int ret; | |
2396 | ||
c0ff4b85 | 2397 | ret = res_counter_charge(&memcg->res, csize, &fail_res); |
4b534334 KH |
2398 | |
2399 | if (likely(!ret)) { | |
2400 | if (!do_swap_account) | |
2401 | return CHARGE_OK; | |
c0ff4b85 | 2402 | ret = res_counter_charge(&memcg->memsw, csize, &fail_res); |
4b534334 KH |
2403 | if (likely(!ret)) |
2404 | return CHARGE_OK; | |
2405 | ||
c0ff4b85 | 2406 | res_counter_uncharge(&memcg->res, csize); |
4b534334 KH |
2407 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
2408 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
2409 | } else | |
2410 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
9221edb7 | 2411 | /* |
9221edb7 JW |
2412 | * Never reclaim on behalf of optional batching, retry with a |
2413 | * single page instead. | |
2414 | */ | |
4c9c5359 | 2415 | if (nr_pages > min_pages) |
4b534334 KH |
2416 | return CHARGE_RETRY; |
2417 | ||
2418 | if (!(gfp_mask & __GFP_WAIT)) | |
2419 | return CHARGE_WOULDBLOCK; | |
2420 | ||
4c9c5359 SS |
2421 | if (gfp_mask & __GFP_NORETRY) |
2422 | return CHARGE_NOMEM; | |
2423 | ||
5660048c | 2424 | ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags); |
7ec99d62 | 2425 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
19942822 | 2426 | return CHARGE_RETRY; |
4b534334 | 2427 | /* |
19942822 JW |
2428 | * Even though the limit is exceeded at this point, reclaim |
2429 | * may have been able to free some pages. Retry the charge | |
2430 | * before killing the task. | |
2431 | * | |
2432 | * Only for regular pages, though: huge pages are rather | |
2433 | * unlikely to succeed so close to the limit, and we fall back | |
2434 | * to regular pages anyway in case of failure. | |
4b534334 | 2435 | */ |
4c9c5359 | 2436 | if (nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER) && ret) |
4b534334 KH |
2437 | return CHARGE_RETRY; |
2438 | ||
2439 | /* | |
2440 | * At task move, charge accounts can be doubly counted. So, it's | |
2441 | * better to wait until the end of task_move if something is going on. | |
2442 | */ | |
2443 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2444 | return CHARGE_RETRY; | |
2445 | ||
2446 | /* If we don't need to call oom-killer at el, return immediately */ | |
2447 | if (!oom_check) | |
2448 | return CHARGE_NOMEM; | |
2449 | /* check OOM */ | |
e845e199 | 2450 | if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask, get_order(csize))) |
4b534334 KH |
2451 | return CHARGE_OOM_DIE; |
2452 | ||
2453 | return CHARGE_RETRY; | |
2454 | } | |
2455 | ||
f817ed48 | 2456 | /* |
38c5d72f KH |
2457 | * __mem_cgroup_try_charge() does |
2458 | * 1. detect memcg to be charged against from passed *mm and *ptr, | |
2459 | * 2. update res_counter | |
2460 | * 3. call memory reclaim if necessary. | |
2461 | * | |
2462 | * In some special case, if the task is fatal, fatal_signal_pending() or | |
2463 | * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup | |
2464 | * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon | |
2465 | * as possible without any hazards. 2: all pages should have a valid | |
2466 | * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg | |
2467 | * pointer, that is treated as a charge to root_mem_cgroup. | |
2468 | * | |
2469 | * So __mem_cgroup_try_charge() will return | |
2470 | * 0 ... on success, filling *ptr with a valid memcg pointer. | |
2471 | * -ENOMEM ... charge failure because of resource limits. | |
2472 | * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup. | |
2473 | * | |
2474 | * Unlike the exported interface, an "oom" parameter is added. if oom==true, | |
2475 | * the oom-killer can be invoked. | |
8a9f3ccd | 2476 | */ |
f817ed48 | 2477 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
ec168510 | 2478 | gfp_t gfp_mask, |
7ec99d62 | 2479 | unsigned int nr_pages, |
c0ff4b85 | 2480 | struct mem_cgroup **ptr, |
7ec99d62 | 2481 | bool oom) |
8a9f3ccd | 2482 | { |
7ec99d62 | 2483 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
4b534334 | 2484 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
c0ff4b85 | 2485 | struct mem_cgroup *memcg = NULL; |
4b534334 | 2486 | int ret; |
a636b327 | 2487 | |
867578cb KH |
2488 | /* |
2489 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
2490 | * in system level. So, allow to go ahead dying process in addition to | |
2491 | * MEMDIE process. | |
2492 | */ | |
2493 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
2494 | || fatal_signal_pending(current))) | |
2495 | goto bypass; | |
a636b327 | 2496 | |
8a9f3ccd | 2497 | /* |
3be91277 HD |
2498 | * We always charge the cgroup the mm_struct belongs to. |
2499 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd | 2500 | * thread group leader migrates. It's possible that mm is not |
24467cac | 2501 | * set, if so charge the root memcg (happens for pagecache usage). |
8a9f3ccd | 2502 | */ |
c0ff4b85 | 2503 | if (!*ptr && !mm) |
38c5d72f | 2504 | *ptr = root_mem_cgroup; |
f75ca962 | 2505 | again: |
c0ff4b85 R |
2506 | if (*ptr) { /* css should be a valid one */ |
2507 | memcg = *ptr; | |
c0ff4b85 | 2508 | if (mem_cgroup_is_root(memcg)) |
f75ca962 | 2509 | goto done; |
a0956d54 | 2510 | if (consume_stock(memcg, nr_pages)) |
f75ca962 | 2511 | goto done; |
c0ff4b85 | 2512 | css_get(&memcg->css); |
4b534334 | 2513 | } else { |
f75ca962 | 2514 | struct task_struct *p; |
54595fe2 | 2515 | |
f75ca962 KH |
2516 | rcu_read_lock(); |
2517 | p = rcu_dereference(mm->owner); | |
f75ca962 | 2518 | /* |
ebb76ce1 | 2519 | * Because we don't have task_lock(), "p" can exit. |
c0ff4b85 | 2520 | * In that case, "memcg" can point to root or p can be NULL with |
ebb76ce1 KH |
2521 | * race with swapoff. Then, we have small risk of mis-accouning. |
2522 | * But such kind of mis-account by race always happens because | |
2523 | * we don't have cgroup_mutex(). It's overkill and we allo that | |
2524 | * small race, here. | |
2525 | * (*) swapoff at el will charge against mm-struct not against | |
2526 | * task-struct. So, mm->owner can be NULL. | |
f75ca962 | 2527 | */ |
c0ff4b85 | 2528 | memcg = mem_cgroup_from_task(p); |
38c5d72f KH |
2529 | if (!memcg) |
2530 | memcg = root_mem_cgroup; | |
2531 | if (mem_cgroup_is_root(memcg)) { | |
f75ca962 KH |
2532 | rcu_read_unlock(); |
2533 | goto done; | |
2534 | } | |
a0956d54 | 2535 | if (consume_stock(memcg, nr_pages)) { |
f75ca962 KH |
2536 | /* |
2537 | * It seems dagerous to access memcg without css_get(). | |
2538 | * But considering how consume_stok works, it's not | |
2539 | * necessary. If consume_stock success, some charges | |
2540 | * from this memcg are cached on this cpu. So, we | |
2541 | * don't need to call css_get()/css_tryget() before | |
2542 | * calling consume_stock(). | |
2543 | */ | |
2544 | rcu_read_unlock(); | |
2545 | goto done; | |
2546 | } | |
2547 | /* after here, we may be blocked. we need to get refcnt */ | |
c0ff4b85 | 2548 | if (!css_tryget(&memcg->css)) { |
f75ca962 KH |
2549 | rcu_read_unlock(); |
2550 | goto again; | |
2551 | } | |
2552 | rcu_read_unlock(); | |
2553 | } | |
8a9f3ccd | 2554 | |
4b534334 KH |
2555 | do { |
2556 | bool oom_check; | |
7a81b88c | 2557 | |
4b534334 | 2558 | /* If killed, bypass charge */ |
f75ca962 | 2559 | if (fatal_signal_pending(current)) { |
c0ff4b85 | 2560 | css_put(&memcg->css); |
4b534334 | 2561 | goto bypass; |
f75ca962 | 2562 | } |
6d61ef40 | 2563 | |
4b534334 KH |
2564 | oom_check = false; |
2565 | if (oom && !nr_oom_retries) { | |
2566 | oom_check = true; | |
2567 | nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
cdec2e42 | 2568 | } |
66e1707b | 2569 | |
4c9c5359 SS |
2570 | ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, nr_pages, |
2571 | oom_check); | |
4b534334 KH |
2572 | switch (ret) { |
2573 | case CHARGE_OK: | |
2574 | break; | |
2575 | case CHARGE_RETRY: /* not in OOM situation but retry */ | |
7ec99d62 | 2576 | batch = nr_pages; |
c0ff4b85 R |
2577 | css_put(&memcg->css); |
2578 | memcg = NULL; | |
f75ca962 | 2579 | goto again; |
4b534334 | 2580 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
c0ff4b85 | 2581 | css_put(&memcg->css); |
4b534334 KH |
2582 | goto nomem; |
2583 | case CHARGE_NOMEM: /* OOM routine works */ | |
f75ca962 | 2584 | if (!oom) { |
c0ff4b85 | 2585 | css_put(&memcg->css); |
867578cb | 2586 | goto nomem; |
f75ca962 | 2587 | } |
4b534334 KH |
2588 | /* If oom, we never return -ENOMEM */ |
2589 | nr_oom_retries--; | |
2590 | break; | |
2591 | case CHARGE_OOM_DIE: /* Killed by OOM Killer */ | |
c0ff4b85 | 2592 | css_put(&memcg->css); |
867578cb | 2593 | goto bypass; |
66e1707b | 2594 | } |
4b534334 KH |
2595 | } while (ret != CHARGE_OK); |
2596 | ||
7ec99d62 | 2597 | if (batch > nr_pages) |
c0ff4b85 R |
2598 | refill_stock(memcg, batch - nr_pages); |
2599 | css_put(&memcg->css); | |
0c3e73e8 | 2600 | done: |
c0ff4b85 | 2601 | *ptr = memcg; |
7a81b88c KH |
2602 | return 0; |
2603 | nomem: | |
c0ff4b85 | 2604 | *ptr = NULL; |
7a81b88c | 2605 | return -ENOMEM; |
867578cb | 2606 | bypass: |
38c5d72f KH |
2607 | *ptr = root_mem_cgroup; |
2608 | return -EINTR; | |
7a81b88c | 2609 | } |
8a9f3ccd | 2610 | |
a3032a2c DN |
2611 | /* |
2612 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2613 | * This function is for that and do uncharge, put css's refcnt. | |
2614 | * gotten by try_charge(). | |
2615 | */ | |
c0ff4b85 | 2616 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg, |
e7018b8d | 2617 | unsigned int nr_pages) |
a3032a2c | 2618 | { |
c0ff4b85 | 2619 | if (!mem_cgroup_is_root(memcg)) { |
e7018b8d JW |
2620 | unsigned long bytes = nr_pages * PAGE_SIZE; |
2621 | ||
c0ff4b85 | 2622 | res_counter_uncharge(&memcg->res, bytes); |
a3032a2c | 2623 | if (do_swap_account) |
c0ff4b85 | 2624 | res_counter_uncharge(&memcg->memsw, bytes); |
a3032a2c | 2625 | } |
854ffa8d DN |
2626 | } |
2627 | ||
d01dd17f KH |
2628 | /* |
2629 | * Cancel chrages in this cgroup....doesn't propagate to parent cgroup. | |
2630 | * This is useful when moving usage to parent cgroup. | |
2631 | */ | |
2632 | static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg, | |
2633 | unsigned int nr_pages) | |
2634 | { | |
2635 | unsigned long bytes = nr_pages * PAGE_SIZE; | |
2636 | ||
2637 | if (mem_cgroup_is_root(memcg)) | |
2638 | return; | |
2639 | ||
2640 | res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes); | |
2641 | if (do_swap_account) | |
2642 | res_counter_uncharge_until(&memcg->memsw, | |
2643 | memcg->memsw.parent, bytes); | |
2644 | } | |
2645 | ||
a3b2d692 KH |
2646 | /* |
2647 | * A helper function to get mem_cgroup from ID. must be called under | |
e9316080 TH |
2648 | * rcu_read_lock(). The caller is responsible for calling css_tryget if |
2649 | * the mem_cgroup is used for charging. (dropping refcnt from swap can be | |
2650 | * called against removed memcg.) | |
a3b2d692 KH |
2651 | */ |
2652 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2653 | { | |
2654 | struct cgroup_subsys_state *css; | |
2655 | ||
2656 | /* ID 0 is unused ID */ | |
2657 | if (!id) | |
2658 | return NULL; | |
2659 | css = css_lookup(&mem_cgroup_subsys, id); | |
2660 | if (!css) | |
2661 | return NULL; | |
b2145145 | 2662 | return mem_cgroup_from_css(css); |
a3b2d692 KH |
2663 | } |
2664 | ||
e42d9d5d | 2665 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2666 | { |
c0ff4b85 | 2667 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2668 | struct page_cgroup *pc; |
a3b2d692 | 2669 | unsigned short id; |
b5a84319 KH |
2670 | swp_entry_t ent; |
2671 | ||
3c776e64 DN |
2672 | VM_BUG_ON(!PageLocked(page)); |
2673 | ||
3c776e64 | 2674 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2675 | lock_page_cgroup(pc); |
a3b2d692 | 2676 | if (PageCgroupUsed(pc)) { |
c0ff4b85 R |
2677 | memcg = pc->mem_cgroup; |
2678 | if (memcg && !css_tryget(&memcg->css)) | |
2679 | memcg = NULL; | |
e42d9d5d | 2680 | } else if (PageSwapCache(page)) { |
3c776e64 | 2681 | ent.val = page_private(page); |
9fb4b7cc | 2682 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2683 | rcu_read_lock(); |
c0ff4b85 R |
2684 | memcg = mem_cgroup_lookup(id); |
2685 | if (memcg && !css_tryget(&memcg->css)) | |
2686 | memcg = NULL; | |
a3b2d692 | 2687 | rcu_read_unlock(); |
3c776e64 | 2688 | } |
c0bd3f63 | 2689 | unlock_page_cgroup(pc); |
c0ff4b85 | 2690 | return memcg; |
b5a84319 KH |
2691 | } |
2692 | ||
c0ff4b85 | 2693 | static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg, |
5564e88b | 2694 | struct page *page, |
7ec99d62 | 2695 | unsigned int nr_pages, |
9ce70c02 HD |
2696 | enum charge_type ctype, |
2697 | bool lrucare) | |
7a81b88c | 2698 | { |
ce587e65 | 2699 | struct page_cgroup *pc = lookup_page_cgroup(page); |
9ce70c02 | 2700 | struct zone *uninitialized_var(zone); |
fa9add64 | 2701 | struct lruvec *lruvec; |
9ce70c02 | 2702 | bool was_on_lru = false; |
b2402857 | 2703 | bool anon; |
9ce70c02 | 2704 | |
ca3e0214 | 2705 | lock_page_cgroup(pc); |
90deb788 | 2706 | VM_BUG_ON(PageCgroupUsed(pc)); |
ca3e0214 KH |
2707 | /* |
2708 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2709 | * accessed by any other context at this point. | |
2710 | */ | |
9ce70c02 HD |
2711 | |
2712 | /* | |
2713 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2714 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2715 | */ | |
2716 | if (lrucare) { | |
2717 | zone = page_zone(page); | |
2718 | spin_lock_irq(&zone->lru_lock); | |
2719 | if (PageLRU(page)) { | |
fa9add64 | 2720 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 | 2721 | ClearPageLRU(page); |
fa9add64 | 2722 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2723 | was_on_lru = true; |
2724 | } | |
2725 | } | |
2726 | ||
c0ff4b85 | 2727 | pc->mem_cgroup = memcg; |
261fb61a KH |
2728 | /* |
2729 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
2730 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
2731 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
2732 | * before USED bit, we need memory barrier here. | |
2733 | * See mem_cgroup_add_lru_list(), etc. | |
2734 | */ | |
08e552c6 | 2735 | smp_wmb(); |
b2402857 | 2736 | SetPageCgroupUsed(pc); |
3be91277 | 2737 | |
9ce70c02 HD |
2738 | if (lrucare) { |
2739 | if (was_on_lru) { | |
fa9add64 | 2740 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 HD |
2741 | VM_BUG_ON(PageLRU(page)); |
2742 | SetPageLRU(page); | |
fa9add64 | 2743 | add_page_to_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2744 | } |
2745 | spin_unlock_irq(&zone->lru_lock); | |
2746 | } | |
2747 | ||
41326c17 | 2748 | if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON) |
b2402857 KH |
2749 | anon = true; |
2750 | else | |
2751 | anon = false; | |
2752 | ||
b070e65c | 2753 | mem_cgroup_charge_statistics(memcg, page, anon, nr_pages); |
52d4b9ac | 2754 | unlock_page_cgroup(pc); |
9ce70c02 | 2755 | |
430e4863 | 2756 | /* |
e883110a | 2757 | * "charge_statistics" updated event counter. |
430e4863 | 2758 | */ |
c0ff4b85 | 2759 | memcg_check_events(memcg, page); |
7a81b88c | 2760 | } |
66e1707b | 2761 | |
7cf27982 GC |
2762 | static DEFINE_MUTEX(set_limit_mutex); |
2763 | ||
7ae1e1d0 GC |
2764 | #ifdef CONFIG_MEMCG_KMEM |
2765 | static inline bool memcg_can_account_kmem(struct mem_cgroup *memcg) | |
2766 | { | |
2767 | return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg) && | |
2768 | (memcg->kmem_account_flags & KMEM_ACCOUNTED_MASK); | |
2769 | } | |
2770 | ||
1f458cbf GC |
2771 | /* |
2772 | * This is a bit cumbersome, but it is rarely used and avoids a backpointer | |
2773 | * in the memcg_cache_params struct. | |
2774 | */ | |
2775 | static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p) | |
2776 | { | |
2777 | struct kmem_cache *cachep; | |
2778 | ||
2779 | VM_BUG_ON(p->is_root_cache); | |
2780 | cachep = p->root_cache; | |
2781 | return cachep->memcg_params->memcg_caches[memcg_cache_id(p->memcg)]; | |
2782 | } | |
2783 | ||
749c5415 | 2784 | #ifdef CONFIG_SLABINFO |
182446d0 TH |
2785 | static int mem_cgroup_slabinfo_read(struct cgroup_subsys_state *css, |
2786 | struct cftype *cft, struct seq_file *m) | |
749c5415 | 2787 | { |
182446d0 | 2788 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
749c5415 GC |
2789 | struct memcg_cache_params *params; |
2790 | ||
2791 | if (!memcg_can_account_kmem(memcg)) | |
2792 | return -EIO; | |
2793 | ||
2794 | print_slabinfo_header(m); | |
2795 | ||
2796 | mutex_lock(&memcg->slab_caches_mutex); | |
2797 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) | |
2798 | cache_show(memcg_params_to_cache(params), m); | |
2799 | mutex_unlock(&memcg->slab_caches_mutex); | |
2800 | ||
2801 | return 0; | |
2802 | } | |
2803 | #endif | |
2804 | ||
7ae1e1d0 GC |
2805 | static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, u64 size) |
2806 | { | |
2807 | struct res_counter *fail_res; | |
2808 | struct mem_cgroup *_memcg; | |
2809 | int ret = 0; | |
2810 | bool may_oom; | |
2811 | ||
2812 | ret = res_counter_charge(&memcg->kmem, size, &fail_res); | |
2813 | if (ret) | |
2814 | return ret; | |
2815 | ||
2816 | /* | |
2817 | * Conditions under which we can wait for the oom_killer. Those are | |
2818 | * the same conditions tested by the core page allocator | |
2819 | */ | |
2820 | may_oom = (gfp & __GFP_FS) && !(gfp & __GFP_NORETRY); | |
2821 | ||
2822 | _memcg = memcg; | |
2823 | ret = __mem_cgroup_try_charge(NULL, gfp, size >> PAGE_SHIFT, | |
2824 | &_memcg, may_oom); | |
2825 | ||
2826 | if (ret == -EINTR) { | |
2827 | /* | |
2828 | * __mem_cgroup_try_charge() chosed to bypass to root due to | |
2829 | * OOM kill or fatal signal. Since our only options are to | |
2830 | * either fail the allocation or charge it to this cgroup, do | |
2831 | * it as a temporary condition. But we can't fail. From a | |
2832 | * kmem/slab perspective, the cache has already been selected, | |
2833 | * by mem_cgroup_kmem_get_cache(), so it is too late to change | |
2834 | * our minds. | |
2835 | * | |
2836 | * This condition will only trigger if the task entered | |
2837 | * memcg_charge_kmem in a sane state, but was OOM-killed during | |
2838 | * __mem_cgroup_try_charge() above. Tasks that were already | |
2839 | * dying when the allocation triggers should have been already | |
2840 | * directed to the root cgroup in memcontrol.h | |
2841 | */ | |
2842 | res_counter_charge_nofail(&memcg->res, size, &fail_res); | |
2843 | if (do_swap_account) | |
2844 | res_counter_charge_nofail(&memcg->memsw, size, | |
2845 | &fail_res); | |
2846 | ret = 0; | |
2847 | } else if (ret) | |
2848 | res_counter_uncharge(&memcg->kmem, size); | |
2849 | ||
2850 | return ret; | |
2851 | } | |
2852 | ||
2853 | static void memcg_uncharge_kmem(struct mem_cgroup *memcg, u64 size) | |
2854 | { | |
7ae1e1d0 GC |
2855 | res_counter_uncharge(&memcg->res, size); |
2856 | if (do_swap_account) | |
2857 | res_counter_uncharge(&memcg->memsw, size); | |
7de37682 GC |
2858 | |
2859 | /* Not down to 0 */ | |
2860 | if (res_counter_uncharge(&memcg->kmem, size)) | |
2861 | return; | |
2862 | ||
10d5ebf4 LZ |
2863 | /* |
2864 | * Releases a reference taken in kmem_cgroup_css_offline in case | |
2865 | * this last uncharge is racing with the offlining code or it is | |
2866 | * outliving the memcg existence. | |
2867 | * | |
2868 | * The memory barrier imposed by test&clear is paired with the | |
2869 | * explicit one in memcg_kmem_mark_dead(). | |
2870 | */ | |
7de37682 | 2871 | if (memcg_kmem_test_and_clear_dead(memcg)) |
10d5ebf4 | 2872 | css_put(&memcg->css); |
7ae1e1d0 GC |
2873 | } |
2874 | ||
2633d7a0 GC |
2875 | void memcg_cache_list_add(struct mem_cgroup *memcg, struct kmem_cache *cachep) |
2876 | { | |
2877 | if (!memcg) | |
2878 | return; | |
2879 | ||
2880 | mutex_lock(&memcg->slab_caches_mutex); | |
2881 | list_add(&cachep->memcg_params->list, &memcg->memcg_slab_caches); | |
2882 | mutex_unlock(&memcg->slab_caches_mutex); | |
2883 | } | |
2884 | ||
2885 | /* | |
2886 | * helper for acessing a memcg's index. It will be used as an index in the | |
2887 | * child cache array in kmem_cache, and also to derive its name. This function | |
2888 | * will return -1 when this is not a kmem-limited memcg. | |
2889 | */ | |
2890 | int memcg_cache_id(struct mem_cgroup *memcg) | |
2891 | { | |
2892 | return memcg ? memcg->kmemcg_id : -1; | |
2893 | } | |
2894 | ||
55007d84 GC |
2895 | /* |
2896 | * This ends up being protected by the set_limit mutex, during normal | |
2897 | * operation, because that is its main call site. | |
2898 | * | |
2899 | * But when we create a new cache, we can call this as well if its parent | |
2900 | * is kmem-limited. That will have to hold set_limit_mutex as well. | |
2901 | */ | |
2902 | int memcg_update_cache_sizes(struct mem_cgroup *memcg) | |
2903 | { | |
2904 | int num, ret; | |
2905 | ||
2906 | num = ida_simple_get(&kmem_limited_groups, | |
2907 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); | |
2908 | if (num < 0) | |
2909 | return num; | |
2910 | /* | |
2911 | * After this point, kmem_accounted (that we test atomically in | |
2912 | * the beginning of this conditional), is no longer 0. This | |
2913 | * guarantees only one process will set the following boolean | |
2914 | * to true. We don't need test_and_set because we're protected | |
2915 | * by the set_limit_mutex anyway. | |
2916 | */ | |
2917 | memcg_kmem_set_activated(memcg); | |
2918 | ||
2919 | ret = memcg_update_all_caches(num+1); | |
2920 | if (ret) { | |
2921 | ida_simple_remove(&kmem_limited_groups, num); | |
2922 | memcg_kmem_clear_activated(memcg); | |
2923 | return ret; | |
2924 | } | |
2925 | ||
2926 | memcg->kmemcg_id = num; | |
2927 | INIT_LIST_HEAD(&memcg->memcg_slab_caches); | |
2928 | mutex_init(&memcg->slab_caches_mutex); | |
2929 | return 0; | |
2930 | } | |
2931 | ||
2932 | static size_t memcg_caches_array_size(int num_groups) | |
2933 | { | |
2934 | ssize_t size; | |
2935 | if (num_groups <= 0) | |
2936 | return 0; | |
2937 | ||
2938 | size = 2 * num_groups; | |
2939 | if (size < MEMCG_CACHES_MIN_SIZE) | |
2940 | size = MEMCG_CACHES_MIN_SIZE; | |
2941 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2942 | size = MEMCG_CACHES_MAX_SIZE; | |
2943 | ||
2944 | return size; | |
2945 | } | |
2946 | ||
2947 | /* | |
2948 | * We should update the current array size iff all caches updates succeed. This | |
2949 | * can only be done from the slab side. The slab mutex needs to be held when | |
2950 | * calling this. | |
2951 | */ | |
2952 | void memcg_update_array_size(int num) | |
2953 | { | |
2954 | if (num > memcg_limited_groups_array_size) | |
2955 | memcg_limited_groups_array_size = memcg_caches_array_size(num); | |
2956 | } | |
2957 | ||
15cf17d2 KK |
2958 | static void kmem_cache_destroy_work_func(struct work_struct *w); |
2959 | ||
55007d84 GC |
2960 | int memcg_update_cache_size(struct kmem_cache *s, int num_groups) |
2961 | { | |
2962 | struct memcg_cache_params *cur_params = s->memcg_params; | |
2963 | ||
2964 | VM_BUG_ON(s->memcg_params && !s->memcg_params->is_root_cache); | |
2965 | ||
2966 | if (num_groups > memcg_limited_groups_array_size) { | |
2967 | int i; | |
2968 | ssize_t size = memcg_caches_array_size(num_groups); | |
2969 | ||
2970 | size *= sizeof(void *); | |
90c7a79c | 2971 | size += offsetof(struct memcg_cache_params, memcg_caches); |
55007d84 GC |
2972 | |
2973 | s->memcg_params = kzalloc(size, GFP_KERNEL); | |
2974 | if (!s->memcg_params) { | |
2975 | s->memcg_params = cur_params; | |
2976 | return -ENOMEM; | |
2977 | } | |
2978 | ||
2979 | s->memcg_params->is_root_cache = true; | |
2980 | ||
2981 | /* | |
2982 | * There is the chance it will be bigger than | |
2983 | * memcg_limited_groups_array_size, if we failed an allocation | |
2984 | * in a cache, in which case all caches updated before it, will | |
2985 | * have a bigger array. | |
2986 | * | |
2987 | * But if that is the case, the data after | |
2988 | * memcg_limited_groups_array_size is certainly unused | |
2989 | */ | |
2990 | for (i = 0; i < memcg_limited_groups_array_size; i++) { | |
2991 | if (!cur_params->memcg_caches[i]) | |
2992 | continue; | |
2993 | s->memcg_params->memcg_caches[i] = | |
2994 | cur_params->memcg_caches[i]; | |
2995 | } | |
2996 | ||
2997 | /* | |
2998 | * Ideally, we would wait until all caches succeed, and only | |
2999 | * then free the old one. But this is not worth the extra | |
3000 | * pointer per-cache we'd have to have for this. | |
3001 | * | |
3002 | * It is not a big deal if some caches are left with a size | |
3003 | * bigger than the others. And all updates will reset this | |
3004 | * anyway. | |
3005 | */ | |
3006 | kfree(cur_params); | |
3007 | } | |
3008 | return 0; | |
3009 | } | |
3010 | ||
943a451a GC |
3011 | int memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s, |
3012 | struct kmem_cache *root_cache) | |
2633d7a0 | 3013 | { |
90c7a79c | 3014 | size_t size; |
2633d7a0 GC |
3015 | |
3016 | if (!memcg_kmem_enabled()) | |
3017 | return 0; | |
3018 | ||
90c7a79c AV |
3019 | if (!memcg) { |
3020 | size = offsetof(struct memcg_cache_params, memcg_caches); | |
55007d84 | 3021 | size += memcg_limited_groups_array_size * sizeof(void *); |
90c7a79c AV |
3022 | } else |
3023 | size = sizeof(struct memcg_cache_params); | |
55007d84 | 3024 | |
2633d7a0 GC |
3025 | s->memcg_params = kzalloc(size, GFP_KERNEL); |
3026 | if (!s->memcg_params) | |
3027 | return -ENOMEM; | |
3028 | ||
943a451a | 3029 | if (memcg) { |
2633d7a0 | 3030 | s->memcg_params->memcg = memcg; |
943a451a | 3031 | s->memcg_params->root_cache = root_cache; |
3e6b11df AV |
3032 | INIT_WORK(&s->memcg_params->destroy, |
3033 | kmem_cache_destroy_work_func); | |
4ba902b5 GC |
3034 | } else |
3035 | s->memcg_params->is_root_cache = true; | |
3036 | ||
2633d7a0 GC |
3037 | return 0; |
3038 | } | |
3039 | ||
3040 | void memcg_release_cache(struct kmem_cache *s) | |
3041 | { | |
d7f25f8a GC |
3042 | struct kmem_cache *root; |
3043 | struct mem_cgroup *memcg; | |
3044 | int id; | |
3045 | ||
3046 | /* | |
3047 | * This happens, for instance, when a root cache goes away before we | |
3048 | * add any memcg. | |
3049 | */ | |
3050 | if (!s->memcg_params) | |
3051 | return; | |
3052 | ||
3053 | if (s->memcg_params->is_root_cache) | |
3054 | goto out; | |
3055 | ||
3056 | memcg = s->memcg_params->memcg; | |
3057 | id = memcg_cache_id(memcg); | |
3058 | ||
3059 | root = s->memcg_params->root_cache; | |
3060 | root->memcg_params->memcg_caches[id] = NULL; | |
d7f25f8a GC |
3061 | |
3062 | mutex_lock(&memcg->slab_caches_mutex); | |
3063 | list_del(&s->memcg_params->list); | |
3064 | mutex_unlock(&memcg->slab_caches_mutex); | |
3065 | ||
20f05310 | 3066 | css_put(&memcg->css); |
d7f25f8a | 3067 | out: |
2633d7a0 GC |
3068 | kfree(s->memcg_params); |
3069 | } | |
3070 | ||
0e9d92f2 GC |
3071 | /* |
3072 | * During the creation a new cache, we need to disable our accounting mechanism | |
3073 | * altogether. This is true even if we are not creating, but rather just | |
3074 | * enqueing new caches to be created. | |
3075 | * | |
3076 | * This is because that process will trigger allocations; some visible, like | |
3077 | * explicit kmallocs to auxiliary data structures, name strings and internal | |
3078 | * cache structures; some well concealed, like INIT_WORK() that can allocate | |
3079 | * objects during debug. | |
3080 | * | |
3081 | * If any allocation happens during memcg_kmem_get_cache, we will recurse back | |
3082 | * to it. This may not be a bounded recursion: since the first cache creation | |
3083 | * failed to complete (waiting on the allocation), we'll just try to create the | |
3084 | * cache again, failing at the same point. | |
3085 | * | |
3086 | * memcg_kmem_get_cache is prepared to abort after seeing a positive count of | |
3087 | * memcg_kmem_skip_account. So we enclose anything that might allocate memory | |
3088 | * inside the following two functions. | |
3089 | */ | |
3090 | static inline void memcg_stop_kmem_account(void) | |
3091 | { | |
3092 | VM_BUG_ON(!current->mm); | |
3093 | current->memcg_kmem_skip_account++; | |
3094 | } | |
3095 | ||
3096 | static inline void memcg_resume_kmem_account(void) | |
3097 | { | |
3098 | VM_BUG_ON(!current->mm); | |
3099 | current->memcg_kmem_skip_account--; | |
3100 | } | |
3101 | ||
1f458cbf GC |
3102 | static void kmem_cache_destroy_work_func(struct work_struct *w) |
3103 | { | |
3104 | struct kmem_cache *cachep; | |
3105 | struct memcg_cache_params *p; | |
3106 | ||
3107 | p = container_of(w, struct memcg_cache_params, destroy); | |
3108 | ||
3109 | cachep = memcg_params_to_cache(p); | |
3110 | ||
22933152 GC |
3111 | /* |
3112 | * If we get down to 0 after shrink, we could delete right away. | |
3113 | * However, memcg_release_pages() already puts us back in the workqueue | |
3114 | * in that case. If we proceed deleting, we'll get a dangling | |
3115 | * reference, and removing the object from the workqueue in that case | |
3116 | * is unnecessary complication. We are not a fast path. | |
3117 | * | |
3118 | * Note that this case is fundamentally different from racing with | |
3119 | * shrink_slab(): if memcg_cgroup_destroy_cache() is called in | |
3120 | * kmem_cache_shrink, not only we would be reinserting a dead cache | |
3121 | * into the queue, but doing so from inside the worker racing to | |
3122 | * destroy it. | |
3123 | * | |
3124 | * So if we aren't down to zero, we'll just schedule a worker and try | |
3125 | * again | |
3126 | */ | |
3127 | if (atomic_read(&cachep->memcg_params->nr_pages) != 0) { | |
3128 | kmem_cache_shrink(cachep); | |
3129 | if (atomic_read(&cachep->memcg_params->nr_pages) == 0) | |
3130 | return; | |
3131 | } else | |
1f458cbf GC |
3132 | kmem_cache_destroy(cachep); |
3133 | } | |
3134 | ||
3135 | void mem_cgroup_destroy_cache(struct kmem_cache *cachep) | |
3136 | { | |
3137 | if (!cachep->memcg_params->dead) | |
3138 | return; | |
3139 | ||
22933152 GC |
3140 | /* |
3141 | * There are many ways in which we can get here. | |
3142 | * | |
3143 | * We can get to a memory-pressure situation while the delayed work is | |
3144 | * still pending to run. The vmscan shrinkers can then release all | |
3145 | * cache memory and get us to destruction. If this is the case, we'll | |
3146 | * be executed twice, which is a bug (the second time will execute over | |
3147 | * bogus data). In this case, cancelling the work should be fine. | |
3148 | * | |
3149 | * But we can also get here from the worker itself, if | |
3150 | * kmem_cache_shrink is enough to shake all the remaining objects and | |
3151 | * get the page count to 0. In this case, we'll deadlock if we try to | |
3152 | * cancel the work (the worker runs with an internal lock held, which | |
3153 | * is the same lock we would hold for cancel_work_sync().) | |
3154 | * | |
3155 | * Since we can't possibly know who got us here, just refrain from | |
3156 | * running if there is already work pending | |
3157 | */ | |
3158 | if (work_pending(&cachep->memcg_params->destroy)) | |
3159 | return; | |
1f458cbf GC |
3160 | /* |
3161 | * We have to defer the actual destroying to a workqueue, because | |
3162 | * we might currently be in a context that cannot sleep. | |
3163 | */ | |
3164 | schedule_work(&cachep->memcg_params->destroy); | |
3165 | } | |
3166 | ||
d9c10ddd MH |
3167 | /* |
3168 | * This lock protects updaters, not readers. We want readers to be as fast as | |
3169 | * they can, and they will either see NULL or a valid cache value. Our model | |
3170 | * allow them to see NULL, in which case the root memcg will be selected. | |
3171 | * | |
3172 | * We need this lock because multiple allocations to the same cache from a non | |
3173 | * will span more than one worker. Only one of them can create the cache. | |
3174 | */ | |
3175 | static DEFINE_MUTEX(memcg_cache_mutex); | |
d7f25f8a | 3176 | |
d9c10ddd MH |
3177 | /* |
3178 | * Called with memcg_cache_mutex held | |
3179 | */ | |
d7f25f8a GC |
3180 | static struct kmem_cache *kmem_cache_dup(struct mem_cgroup *memcg, |
3181 | struct kmem_cache *s) | |
3182 | { | |
d7f25f8a | 3183 | struct kmem_cache *new; |
d9c10ddd | 3184 | static char *tmp_name = NULL; |
d7f25f8a | 3185 | |
d9c10ddd MH |
3186 | lockdep_assert_held(&memcg_cache_mutex); |
3187 | ||
3188 | /* | |
3189 | * kmem_cache_create_memcg duplicates the given name and | |
3190 | * cgroup_name for this name requires RCU context. | |
3191 | * This static temporary buffer is used to prevent from | |
3192 | * pointless shortliving allocation. | |
3193 | */ | |
3194 | if (!tmp_name) { | |
3195 | tmp_name = kmalloc(PATH_MAX, GFP_KERNEL); | |
3196 | if (!tmp_name) | |
3197 | return NULL; | |
3198 | } | |
3199 | ||
3200 | rcu_read_lock(); | |
3201 | snprintf(tmp_name, PATH_MAX, "%s(%d:%s)", s->name, | |
3202 | memcg_cache_id(memcg), cgroup_name(memcg->css.cgroup)); | |
3203 | rcu_read_unlock(); | |
d7f25f8a | 3204 | |
d9c10ddd | 3205 | new = kmem_cache_create_memcg(memcg, tmp_name, s->object_size, s->align, |
943a451a | 3206 | (s->flags & ~SLAB_PANIC), s->ctor, s); |
d7f25f8a | 3207 | |
d79923fa GC |
3208 | if (new) |
3209 | new->allocflags |= __GFP_KMEMCG; | |
3210 | ||
d7f25f8a GC |
3211 | return new; |
3212 | } | |
3213 | ||
d7f25f8a GC |
3214 | static struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg, |
3215 | struct kmem_cache *cachep) | |
3216 | { | |
3217 | struct kmem_cache *new_cachep; | |
3218 | int idx; | |
3219 | ||
3220 | BUG_ON(!memcg_can_account_kmem(memcg)); | |
3221 | ||
3222 | idx = memcg_cache_id(memcg); | |
3223 | ||
3224 | mutex_lock(&memcg_cache_mutex); | |
3225 | new_cachep = cachep->memcg_params->memcg_caches[idx]; | |
20f05310 LZ |
3226 | if (new_cachep) { |
3227 | css_put(&memcg->css); | |
d7f25f8a | 3228 | goto out; |
20f05310 | 3229 | } |
d7f25f8a GC |
3230 | |
3231 | new_cachep = kmem_cache_dup(memcg, cachep); | |
d7f25f8a GC |
3232 | if (new_cachep == NULL) { |
3233 | new_cachep = cachep; | |
20f05310 | 3234 | css_put(&memcg->css); |
d7f25f8a GC |
3235 | goto out; |
3236 | } | |
3237 | ||
1f458cbf | 3238 | atomic_set(&new_cachep->memcg_params->nr_pages , 0); |
d7f25f8a GC |
3239 | |
3240 | cachep->memcg_params->memcg_caches[idx] = new_cachep; | |
3241 | /* | |
3242 | * the readers won't lock, make sure everybody sees the updated value, | |
3243 | * so they won't put stuff in the queue again for no reason | |
3244 | */ | |
3245 | wmb(); | |
3246 | out: | |
3247 | mutex_unlock(&memcg_cache_mutex); | |
3248 | return new_cachep; | |
3249 | } | |
3250 | ||
7cf27982 GC |
3251 | void kmem_cache_destroy_memcg_children(struct kmem_cache *s) |
3252 | { | |
3253 | struct kmem_cache *c; | |
3254 | int i; | |
3255 | ||
3256 | if (!s->memcg_params) | |
3257 | return; | |
3258 | if (!s->memcg_params->is_root_cache) | |
3259 | return; | |
3260 | ||
3261 | /* | |
3262 | * If the cache is being destroyed, we trust that there is no one else | |
3263 | * requesting objects from it. Even if there are, the sanity checks in | |
3264 | * kmem_cache_destroy should caught this ill-case. | |
3265 | * | |
3266 | * Still, we don't want anyone else freeing memcg_caches under our | |
3267 | * noses, which can happen if a new memcg comes to life. As usual, | |
3268 | * we'll take the set_limit_mutex to protect ourselves against this. | |
3269 | */ | |
3270 | mutex_lock(&set_limit_mutex); | |
3271 | for (i = 0; i < memcg_limited_groups_array_size; i++) { | |
3272 | c = s->memcg_params->memcg_caches[i]; | |
3273 | if (!c) | |
3274 | continue; | |
3275 | ||
3276 | /* | |
3277 | * We will now manually delete the caches, so to avoid races | |
3278 | * we need to cancel all pending destruction workers and | |
3279 | * proceed with destruction ourselves. | |
3280 | * | |
3281 | * kmem_cache_destroy() will call kmem_cache_shrink internally, | |
3282 | * and that could spawn the workers again: it is likely that | |
3283 | * the cache still have active pages until this very moment. | |
3284 | * This would lead us back to mem_cgroup_destroy_cache. | |
3285 | * | |
3286 | * But that will not execute at all if the "dead" flag is not | |
3287 | * set, so flip it down to guarantee we are in control. | |
3288 | */ | |
3289 | c->memcg_params->dead = false; | |
22933152 | 3290 | cancel_work_sync(&c->memcg_params->destroy); |
7cf27982 GC |
3291 | kmem_cache_destroy(c); |
3292 | } | |
3293 | mutex_unlock(&set_limit_mutex); | |
3294 | } | |
3295 | ||
d7f25f8a GC |
3296 | struct create_work { |
3297 | struct mem_cgroup *memcg; | |
3298 | struct kmem_cache *cachep; | |
3299 | struct work_struct work; | |
3300 | }; | |
3301 | ||
1f458cbf GC |
3302 | static void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg) |
3303 | { | |
3304 | struct kmem_cache *cachep; | |
3305 | struct memcg_cache_params *params; | |
3306 | ||
3307 | if (!memcg_kmem_is_active(memcg)) | |
3308 | return; | |
3309 | ||
3310 | mutex_lock(&memcg->slab_caches_mutex); | |
3311 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) { | |
3312 | cachep = memcg_params_to_cache(params); | |
3313 | cachep->memcg_params->dead = true; | |
1f458cbf GC |
3314 | schedule_work(&cachep->memcg_params->destroy); |
3315 | } | |
3316 | mutex_unlock(&memcg->slab_caches_mutex); | |
3317 | } | |
3318 | ||
d7f25f8a GC |
3319 | static void memcg_create_cache_work_func(struct work_struct *w) |
3320 | { | |
3321 | struct create_work *cw; | |
3322 | ||
3323 | cw = container_of(w, struct create_work, work); | |
3324 | memcg_create_kmem_cache(cw->memcg, cw->cachep); | |
d7f25f8a GC |
3325 | kfree(cw); |
3326 | } | |
3327 | ||
3328 | /* | |
3329 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 3330 | */ |
0e9d92f2 GC |
3331 | static void __memcg_create_cache_enqueue(struct mem_cgroup *memcg, |
3332 | struct kmem_cache *cachep) | |
d7f25f8a GC |
3333 | { |
3334 | struct create_work *cw; | |
3335 | ||
3336 | cw = kmalloc(sizeof(struct create_work), GFP_NOWAIT); | |
ca0dde97 LZ |
3337 | if (cw == NULL) { |
3338 | css_put(&memcg->css); | |
d7f25f8a GC |
3339 | return; |
3340 | } | |
3341 | ||
3342 | cw->memcg = memcg; | |
3343 | cw->cachep = cachep; | |
3344 | ||
3345 | INIT_WORK(&cw->work, memcg_create_cache_work_func); | |
3346 | schedule_work(&cw->work); | |
3347 | } | |
3348 | ||
0e9d92f2 GC |
3349 | static void memcg_create_cache_enqueue(struct mem_cgroup *memcg, |
3350 | struct kmem_cache *cachep) | |
3351 | { | |
3352 | /* | |
3353 | * We need to stop accounting when we kmalloc, because if the | |
3354 | * corresponding kmalloc cache is not yet created, the first allocation | |
3355 | * in __memcg_create_cache_enqueue will recurse. | |
3356 | * | |
3357 | * However, it is better to enclose the whole function. Depending on | |
3358 | * the debugging options enabled, INIT_WORK(), for instance, can | |
3359 | * trigger an allocation. This too, will make us recurse. Because at | |
3360 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
3361 | * the safest choice is to do it like this, wrapping the whole function. | |
3362 | */ | |
3363 | memcg_stop_kmem_account(); | |
3364 | __memcg_create_cache_enqueue(memcg, cachep); | |
3365 | memcg_resume_kmem_account(); | |
3366 | } | |
d7f25f8a GC |
3367 | /* |
3368 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
3369 | * We try to use the current memcg's version of the cache. | |
3370 | * | |
3371 | * If the cache does not exist yet, if we are the first user of it, | |
3372 | * we either create it immediately, if possible, or create it asynchronously | |
3373 | * in a workqueue. | |
3374 | * In the latter case, we will let the current allocation go through with | |
3375 | * the original cache. | |
3376 | * | |
3377 | * Can't be called in interrupt context or from kernel threads. | |
3378 | * This function needs to be called with rcu_read_lock() held. | |
3379 | */ | |
3380 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, | |
3381 | gfp_t gfp) | |
3382 | { | |
3383 | struct mem_cgroup *memcg; | |
3384 | int idx; | |
3385 | ||
3386 | VM_BUG_ON(!cachep->memcg_params); | |
3387 | VM_BUG_ON(!cachep->memcg_params->is_root_cache); | |
3388 | ||
0e9d92f2 GC |
3389 | if (!current->mm || current->memcg_kmem_skip_account) |
3390 | return cachep; | |
3391 | ||
d7f25f8a GC |
3392 | rcu_read_lock(); |
3393 | memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner)); | |
d7f25f8a GC |
3394 | |
3395 | if (!memcg_can_account_kmem(memcg)) | |
ca0dde97 | 3396 | goto out; |
d7f25f8a GC |
3397 | |
3398 | idx = memcg_cache_id(memcg); | |
3399 | ||
3400 | /* | |
3401 | * barrier to mare sure we're always seeing the up to date value. The | |
3402 | * code updating memcg_caches will issue a write barrier to match this. | |
3403 | */ | |
3404 | read_barrier_depends(); | |
ca0dde97 LZ |
3405 | if (likely(cachep->memcg_params->memcg_caches[idx])) { |
3406 | cachep = cachep->memcg_params->memcg_caches[idx]; | |
3407 | goto out; | |
d7f25f8a GC |
3408 | } |
3409 | ||
ca0dde97 LZ |
3410 | /* The corresponding put will be done in the workqueue. */ |
3411 | if (!css_tryget(&memcg->css)) | |
3412 | goto out; | |
3413 | rcu_read_unlock(); | |
3414 | ||
3415 | /* | |
3416 | * If we are in a safe context (can wait, and not in interrupt | |
3417 | * context), we could be be predictable and return right away. | |
3418 | * This would guarantee that the allocation being performed | |
3419 | * already belongs in the new cache. | |
3420 | * | |
3421 | * However, there are some clashes that can arrive from locking. | |
3422 | * For instance, because we acquire the slab_mutex while doing | |
3423 | * kmem_cache_dup, this means no further allocation could happen | |
3424 | * with the slab_mutex held. | |
3425 | * | |
3426 | * Also, because cache creation issue get_online_cpus(), this | |
3427 | * creates a lock chain: memcg_slab_mutex -> cpu_hotplug_mutex, | |
3428 | * that ends up reversed during cpu hotplug. (cpuset allocates | |
3429 | * a bunch of GFP_KERNEL memory during cpuup). Due to all that, | |
3430 | * better to defer everything. | |
3431 | */ | |
3432 | memcg_create_cache_enqueue(memcg, cachep); | |
3433 | return cachep; | |
3434 | out: | |
3435 | rcu_read_unlock(); | |
3436 | return cachep; | |
d7f25f8a GC |
3437 | } |
3438 | EXPORT_SYMBOL(__memcg_kmem_get_cache); | |
3439 | ||
7ae1e1d0 GC |
3440 | /* |
3441 | * We need to verify if the allocation against current->mm->owner's memcg is | |
3442 | * possible for the given order. But the page is not allocated yet, so we'll | |
3443 | * need a further commit step to do the final arrangements. | |
3444 | * | |
3445 | * It is possible for the task to switch cgroups in this mean time, so at | |
3446 | * commit time, we can't rely on task conversion any longer. We'll then use | |
3447 | * the handle argument to return to the caller which cgroup we should commit | |
3448 | * against. We could also return the memcg directly and avoid the pointer | |
3449 | * passing, but a boolean return value gives better semantics considering | |
3450 | * the compiled-out case as well. | |
3451 | * | |
3452 | * Returning true means the allocation is possible. | |
3453 | */ | |
3454 | bool | |
3455 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) | |
3456 | { | |
3457 | struct mem_cgroup *memcg; | |
3458 | int ret; | |
3459 | ||
3460 | *_memcg = NULL; | |
6d42c232 GC |
3461 | |
3462 | /* | |
3463 | * Disabling accounting is only relevant for some specific memcg | |
3464 | * internal allocations. Therefore we would initially not have such | |
3465 | * check here, since direct calls to the page allocator that are marked | |
3466 | * with GFP_KMEMCG only happen outside memcg core. We are mostly | |
3467 | * concerned with cache allocations, and by having this test at | |
3468 | * memcg_kmem_get_cache, we are already able to relay the allocation to | |
3469 | * the root cache and bypass the memcg cache altogether. | |
3470 | * | |
3471 | * There is one exception, though: the SLUB allocator does not create | |
3472 | * large order caches, but rather service large kmallocs directly from | |
3473 | * the page allocator. Therefore, the following sequence when backed by | |
3474 | * the SLUB allocator: | |
3475 | * | |
3476 | * memcg_stop_kmem_account(); | |
3477 | * kmalloc(<large_number>) | |
3478 | * memcg_resume_kmem_account(); | |
3479 | * | |
3480 | * would effectively ignore the fact that we should skip accounting, | |
3481 | * since it will drive us directly to this function without passing | |
3482 | * through the cache selector memcg_kmem_get_cache. Such large | |
3483 | * allocations are extremely rare but can happen, for instance, for the | |
3484 | * cache arrays. We bring this test here. | |
3485 | */ | |
3486 | if (!current->mm || current->memcg_kmem_skip_account) | |
3487 | return true; | |
3488 | ||
7ae1e1d0 GC |
3489 | memcg = try_get_mem_cgroup_from_mm(current->mm); |
3490 | ||
3491 | /* | |
3492 | * very rare case described in mem_cgroup_from_task. Unfortunately there | |
3493 | * isn't much we can do without complicating this too much, and it would | |
3494 | * be gfp-dependent anyway. Just let it go | |
3495 | */ | |
3496 | if (unlikely(!memcg)) | |
3497 | return true; | |
3498 | ||
3499 | if (!memcg_can_account_kmem(memcg)) { | |
3500 | css_put(&memcg->css); | |
3501 | return true; | |
3502 | } | |
3503 | ||
7ae1e1d0 GC |
3504 | ret = memcg_charge_kmem(memcg, gfp, PAGE_SIZE << order); |
3505 | if (!ret) | |
3506 | *_memcg = memcg; | |
7ae1e1d0 GC |
3507 | |
3508 | css_put(&memcg->css); | |
3509 | return (ret == 0); | |
3510 | } | |
3511 | ||
3512 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
3513 | int order) | |
3514 | { | |
3515 | struct page_cgroup *pc; | |
3516 | ||
3517 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
3518 | ||
3519 | /* The page allocation failed. Revert */ | |
3520 | if (!page) { | |
3521 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); | |
7ae1e1d0 GC |
3522 | return; |
3523 | } | |
3524 | ||
3525 | pc = lookup_page_cgroup(page); | |
3526 | lock_page_cgroup(pc); | |
3527 | pc->mem_cgroup = memcg; | |
3528 | SetPageCgroupUsed(pc); | |
3529 | unlock_page_cgroup(pc); | |
3530 | } | |
3531 | ||
3532 | void __memcg_kmem_uncharge_pages(struct page *page, int order) | |
3533 | { | |
3534 | struct mem_cgroup *memcg = NULL; | |
3535 | struct page_cgroup *pc; | |
3536 | ||
3537 | ||
3538 | pc = lookup_page_cgroup(page); | |
3539 | /* | |
3540 | * Fast unlocked return. Theoretically might have changed, have to | |
3541 | * check again after locking. | |
3542 | */ | |
3543 | if (!PageCgroupUsed(pc)) | |
3544 | return; | |
3545 | ||
3546 | lock_page_cgroup(pc); | |
3547 | if (PageCgroupUsed(pc)) { | |
3548 | memcg = pc->mem_cgroup; | |
3549 | ClearPageCgroupUsed(pc); | |
3550 | } | |
3551 | unlock_page_cgroup(pc); | |
3552 | ||
3553 | /* | |
3554 | * We trust that only if there is a memcg associated with the page, it | |
3555 | * is a valid allocation | |
3556 | */ | |
3557 | if (!memcg) | |
3558 | return; | |
3559 | ||
3560 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
3561 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); | |
7ae1e1d0 | 3562 | } |
1f458cbf GC |
3563 | #else |
3564 | static inline void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg) | |
3565 | { | |
3566 | } | |
7ae1e1d0 GC |
3567 | #endif /* CONFIG_MEMCG_KMEM */ |
3568 | ||
ca3e0214 KH |
3569 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3570 | ||
a0db00fc | 3571 | #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION) |
ca3e0214 KH |
3572 | /* |
3573 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
3574 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
3575 | * charge/uncharge will be never happen and move_account() is done under | |
3576 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 3577 | */ |
e94c8a9c | 3578 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 KH |
3579 | { |
3580 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
e94c8a9c | 3581 | struct page_cgroup *pc; |
b070e65c | 3582 | struct mem_cgroup *memcg; |
e94c8a9c | 3583 | int i; |
ca3e0214 | 3584 | |
3d37c4a9 KH |
3585 | if (mem_cgroup_disabled()) |
3586 | return; | |
b070e65c DR |
3587 | |
3588 | memcg = head_pc->mem_cgroup; | |
e94c8a9c KH |
3589 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
3590 | pc = head_pc + i; | |
b070e65c | 3591 | pc->mem_cgroup = memcg; |
e94c8a9c | 3592 | smp_wmb();/* see __commit_charge() */ |
e94c8a9c KH |
3593 | pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
3594 | } | |
b070e65c DR |
3595 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
3596 | HPAGE_PMD_NR); | |
ca3e0214 | 3597 | } |
12d27107 | 3598 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3599 | |
f817ed48 | 3600 | /** |
de3638d9 | 3601 | * mem_cgroup_move_account - move account of the page |
5564e88b | 3602 | * @page: the page |
7ec99d62 | 3603 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
3604 | * @pc: page_cgroup of the page. |
3605 | * @from: mem_cgroup which the page is moved from. | |
3606 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
3607 | * | |
3608 | * The caller must confirm following. | |
08e552c6 | 3609 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 3610 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 3611 | * |
2f3479b1 KH |
3612 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
3613 | * from old cgroup. | |
f817ed48 | 3614 | */ |
7ec99d62 JW |
3615 | static int mem_cgroup_move_account(struct page *page, |
3616 | unsigned int nr_pages, | |
3617 | struct page_cgroup *pc, | |
3618 | struct mem_cgroup *from, | |
2f3479b1 | 3619 | struct mem_cgroup *to) |
f817ed48 | 3620 | { |
de3638d9 JW |
3621 | unsigned long flags; |
3622 | int ret; | |
b2402857 | 3623 | bool anon = PageAnon(page); |
987eba66 | 3624 | |
f817ed48 | 3625 | VM_BUG_ON(from == to); |
5564e88b | 3626 | VM_BUG_ON(PageLRU(page)); |
de3638d9 JW |
3627 | /* |
3628 | * The page is isolated from LRU. So, collapse function | |
3629 | * will not handle this page. But page splitting can happen. | |
3630 | * Do this check under compound_page_lock(). The caller should | |
3631 | * hold it. | |
3632 | */ | |
3633 | ret = -EBUSY; | |
7ec99d62 | 3634 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
3635 | goto out; |
3636 | ||
3637 | lock_page_cgroup(pc); | |
3638 | ||
3639 | ret = -EINVAL; | |
3640 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
3641 | goto unlock; | |
3642 | ||
312734c0 | 3643 | move_lock_mem_cgroup(from, &flags); |
f817ed48 | 3644 | |
2ff76f11 | 3645 | if (!anon && page_mapped(page)) { |
c62b1a3b KH |
3646 | /* Update mapped_file data for mem_cgroup */ |
3647 | preempt_disable(); | |
3648 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
3649 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
3650 | preempt_enable(); | |
d69b042f | 3651 | } |
b070e65c | 3652 | mem_cgroup_charge_statistics(from, page, anon, -nr_pages); |
d69b042f | 3653 | |
854ffa8d | 3654 | /* caller should have done css_get */ |
08e552c6 | 3655 | pc->mem_cgroup = to; |
b070e65c | 3656 | mem_cgroup_charge_statistics(to, page, anon, nr_pages); |
312734c0 | 3657 | move_unlock_mem_cgroup(from, &flags); |
de3638d9 JW |
3658 | ret = 0; |
3659 | unlock: | |
57f9fd7d | 3660 | unlock_page_cgroup(pc); |
d2265e6f KH |
3661 | /* |
3662 | * check events | |
3663 | */ | |
5564e88b JW |
3664 | memcg_check_events(to, page); |
3665 | memcg_check_events(from, page); | |
de3638d9 | 3666 | out: |
f817ed48 KH |
3667 | return ret; |
3668 | } | |
3669 | ||
2ef37d3f MH |
3670 | /** |
3671 | * mem_cgroup_move_parent - moves page to the parent group | |
3672 | * @page: the page to move | |
3673 | * @pc: page_cgroup of the page | |
3674 | * @child: page's cgroup | |
3675 | * | |
3676 | * move charges to its parent or the root cgroup if the group has no | |
3677 | * parent (aka use_hierarchy==0). | |
3678 | * Although this might fail (get_page_unless_zero, isolate_lru_page or | |
3679 | * mem_cgroup_move_account fails) the failure is always temporary and | |
3680 | * it signals a race with a page removal/uncharge or migration. In the | |
3681 | * first case the page is on the way out and it will vanish from the LRU | |
3682 | * on the next attempt and the call should be retried later. | |
3683 | * Isolation from the LRU fails only if page has been isolated from | |
3684 | * the LRU since we looked at it and that usually means either global | |
3685 | * reclaim or migration going on. The page will either get back to the | |
3686 | * LRU or vanish. | |
3687 | * Finaly mem_cgroup_move_account fails only if the page got uncharged | |
3688 | * (!PageCgroupUsed) or moved to a different group. The page will | |
3689 | * disappear in the next attempt. | |
f817ed48 | 3690 | */ |
5564e88b JW |
3691 | static int mem_cgroup_move_parent(struct page *page, |
3692 | struct page_cgroup *pc, | |
6068bf01 | 3693 | struct mem_cgroup *child) |
f817ed48 | 3694 | { |
f817ed48 | 3695 | struct mem_cgroup *parent; |
7ec99d62 | 3696 | unsigned int nr_pages; |
4be4489f | 3697 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
3698 | int ret; |
3699 | ||
d8423011 | 3700 | VM_BUG_ON(mem_cgroup_is_root(child)); |
f817ed48 | 3701 | |
57f9fd7d DN |
3702 | ret = -EBUSY; |
3703 | if (!get_page_unless_zero(page)) | |
3704 | goto out; | |
3705 | if (isolate_lru_page(page)) | |
3706 | goto put; | |
52dbb905 | 3707 | |
7ec99d62 | 3708 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 3709 | |
cc926f78 KH |
3710 | parent = parent_mem_cgroup(child); |
3711 | /* | |
3712 | * If no parent, move charges to root cgroup. | |
3713 | */ | |
3714 | if (!parent) | |
3715 | parent = root_mem_cgroup; | |
f817ed48 | 3716 | |
2ef37d3f MH |
3717 | if (nr_pages > 1) { |
3718 | VM_BUG_ON(!PageTransHuge(page)); | |
987eba66 | 3719 | flags = compound_lock_irqsave(page); |
2ef37d3f | 3720 | } |
987eba66 | 3721 | |
cc926f78 | 3722 | ret = mem_cgroup_move_account(page, nr_pages, |
2f3479b1 | 3723 | pc, child, parent); |
cc926f78 KH |
3724 | if (!ret) |
3725 | __mem_cgroup_cancel_local_charge(child, nr_pages); | |
8dba474f | 3726 | |
7ec99d62 | 3727 | if (nr_pages > 1) |
987eba66 | 3728 | compound_unlock_irqrestore(page, flags); |
08e552c6 | 3729 | putback_lru_page(page); |
57f9fd7d | 3730 | put: |
40d58138 | 3731 | put_page(page); |
57f9fd7d | 3732 | out: |
f817ed48 KH |
3733 | return ret; |
3734 | } | |
3735 | ||
7a81b88c KH |
3736 | /* |
3737 | * Charge the memory controller for page usage. | |
3738 | * Return | |
3739 | * 0 if the charge was successful | |
3740 | * < 0 if the cgroup is over its limit | |
3741 | */ | |
3742 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
73045c47 | 3743 | gfp_t gfp_mask, enum charge_type ctype) |
7a81b88c | 3744 | { |
c0ff4b85 | 3745 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 3746 | unsigned int nr_pages = 1; |
8493ae43 | 3747 | bool oom = true; |
7a81b88c | 3748 | int ret; |
ec168510 | 3749 | |
37c2ac78 | 3750 | if (PageTransHuge(page)) { |
7ec99d62 | 3751 | nr_pages <<= compound_order(page); |
37c2ac78 | 3752 | VM_BUG_ON(!PageTransHuge(page)); |
8493ae43 JW |
3753 | /* |
3754 | * Never OOM-kill a process for a huge page. The | |
3755 | * fault handler will fall back to regular pages. | |
3756 | */ | |
3757 | oom = false; | |
37c2ac78 | 3758 | } |
7a81b88c | 3759 | |
c0ff4b85 | 3760 | ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom); |
38c5d72f | 3761 | if (ret == -ENOMEM) |
7a81b88c | 3762 | return ret; |
ce587e65 | 3763 | __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false); |
8a9f3ccd | 3764 | return 0; |
8a9f3ccd BS |
3765 | } |
3766 | ||
7a81b88c KH |
3767 | int mem_cgroup_newpage_charge(struct page *page, |
3768 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 3769 | { |
f8d66542 | 3770 | if (mem_cgroup_disabled()) |
cede86ac | 3771 | return 0; |
7a0524cf JW |
3772 | VM_BUG_ON(page_mapped(page)); |
3773 | VM_BUG_ON(page->mapping && !PageAnon(page)); | |
3774 | VM_BUG_ON(!mm); | |
217bc319 | 3775 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
41326c17 | 3776 | MEM_CGROUP_CHARGE_TYPE_ANON); |
217bc319 KH |
3777 | } |
3778 | ||
54595fe2 KH |
3779 | /* |
3780 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
3781 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 3782 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
3783 | * "commit()" or removed by "cancel()" |
3784 | */ | |
0435a2fd JW |
3785 | static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
3786 | struct page *page, | |
3787 | gfp_t mask, | |
3788 | struct mem_cgroup **memcgp) | |
8c7c6e34 | 3789 | { |
c0ff4b85 | 3790 | struct mem_cgroup *memcg; |
90deb788 | 3791 | struct page_cgroup *pc; |
54595fe2 | 3792 | int ret; |
8c7c6e34 | 3793 | |
90deb788 JW |
3794 | pc = lookup_page_cgroup(page); |
3795 | /* | |
3796 | * Every swap fault against a single page tries to charge the | |
3797 | * page, bail as early as possible. shmem_unuse() encounters | |
3798 | * already charged pages, too. The USED bit is protected by | |
3799 | * the page lock, which serializes swap cache removal, which | |
3800 | * in turn serializes uncharging. | |
3801 | */ | |
3802 | if (PageCgroupUsed(pc)) | |
3803 | return 0; | |
8c7c6e34 KH |
3804 | if (!do_swap_account) |
3805 | goto charge_cur_mm; | |
c0ff4b85 R |
3806 | memcg = try_get_mem_cgroup_from_page(page); |
3807 | if (!memcg) | |
54595fe2 | 3808 | goto charge_cur_mm; |
72835c86 JW |
3809 | *memcgp = memcg; |
3810 | ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true); | |
c0ff4b85 | 3811 | css_put(&memcg->css); |
38c5d72f KH |
3812 | if (ret == -EINTR) |
3813 | ret = 0; | |
54595fe2 | 3814 | return ret; |
8c7c6e34 | 3815 | charge_cur_mm: |
38c5d72f KH |
3816 | ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true); |
3817 | if (ret == -EINTR) | |
3818 | ret = 0; | |
3819 | return ret; | |
8c7c6e34 KH |
3820 | } |
3821 | ||
0435a2fd JW |
3822 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page, |
3823 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
3824 | { | |
3825 | *memcgp = NULL; | |
3826 | if (mem_cgroup_disabled()) | |
3827 | return 0; | |
bdf4f4d2 JW |
3828 | /* |
3829 | * A racing thread's fault, or swapoff, may have already | |
3830 | * updated the pte, and even removed page from swap cache: in | |
3831 | * those cases unuse_pte()'s pte_same() test will fail; but | |
3832 | * there's also a KSM case which does need to charge the page. | |
3833 | */ | |
3834 | if (!PageSwapCache(page)) { | |
3835 | int ret; | |
3836 | ||
3837 | ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, memcgp, true); | |
3838 | if (ret == -EINTR) | |
3839 | ret = 0; | |
3840 | return ret; | |
3841 | } | |
0435a2fd JW |
3842 | return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp); |
3843 | } | |
3844 | ||
827a03d2 JW |
3845 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg) |
3846 | { | |
3847 | if (mem_cgroup_disabled()) | |
3848 | return; | |
3849 | if (!memcg) | |
3850 | return; | |
3851 | __mem_cgroup_cancel_charge(memcg, 1); | |
3852 | } | |
3853 | ||
83aae4c7 | 3854 | static void |
72835c86 | 3855 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg, |
83aae4c7 | 3856 | enum charge_type ctype) |
7a81b88c | 3857 | { |
f8d66542 | 3858 | if (mem_cgroup_disabled()) |
7a81b88c | 3859 | return; |
72835c86 | 3860 | if (!memcg) |
7a81b88c | 3861 | return; |
5a6475a4 | 3862 | |
ce587e65 | 3863 | __mem_cgroup_commit_charge(memcg, page, 1, ctype, true); |
8c7c6e34 KH |
3864 | /* |
3865 | * Now swap is on-memory. This means this page may be | |
3866 | * counted both as mem and swap....double count. | |
03f3c433 KH |
3867 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
3868 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
3869 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 3870 | */ |
03f3c433 | 3871 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 3872 | swp_entry_t ent = {.val = page_private(page)}; |
86493009 | 3873 | mem_cgroup_uncharge_swap(ent); |
8c7c6e34 | 3874 | } |
7a81b88c KH |
3875 | } |
3876 | ||
72835c86 JW |
3877 | void mem_cgroup_commit_charge_swapin(struct page *page, |
3878 | struct mem_cgroup *memcg) | |
83aae4c7 | 3879 | { |
72835c86 | 3880 | __mem_cgroup_commit_charge_swapin(page, memcg, |
41326c17 | 3881 | MEM_CGROUP_CHARGE_TYPE_ANON); |
83aae4c7 DN |
3882 | } |
3883 | ||
827a03d2 JW |
3884 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
3885 | gfp_t gfp_mask) | |
7a81b88c | 3886 | { |
827a03d2 JW |
3887 | struct mem_cgroup *memcg = NULL; |
3888 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
3889 | int ret; | |
3890 | ||
f8d66542 | 3891 | if (mem_cgroup_disabled()) |
827a03d2 JW |
3892 | return 0; |
3893 | if (PageCompound(page)) | |
3894 | return 0; | |
3895 | ||
827a03d2 JW |
3896 | if (!PageSwapCache(page)) |
3897 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, type); | |
3898 | else { /* page is swapcache/shmem */ | |
0435a2fd JW |
3899 | ret = __mem_cgroup_try_charge_swapin(mm, page, |
3900 | gfp_mask, &memcg); | |
827a03d2 JW |
3901 | if (!ret) |
3902 | __mem_cgroup_commit_charge_swapin(page, memcg, type); | |
3903 | } | |
3904 | return ret; | |
7a81b88c KH |
3905 | } |
3906 | ||
c0ff4b85 | 3907 | static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg, |
7ec99d62 JW |
3908 | unsigned int nr_pages, |
3909 | const enum charge_type ctype) | |
569b846d KH |
3910 | { |
3911 | struct memcg_batch_info *batch = NULL; | |
3912 | bool uncharge_memsw = true; | |
7ec99d62 | 3913 | |
569b846d KH |
3914 | /* If swapout, usage of swap doesn't decrease */ |
3915 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
3916 | uncharge_memsw = false; | |
569b846d KH |
3917 | |
3918 | batch = ¤t->memcg_batch; | |
3919 | /* | |
3920 | * In usual, we do css_get() when we remember memcg pointer. | |
3921 | * But in this case, we keep res->usage until end of a series of | |
3922 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
3923 | */ | |
3924 | if (!batch->memcg) | |
c0ff4b85 | 3925 | batch->memcg = memcg; |
3c11ecf4 KH |
3926 | /* |
3927 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
25985edc | 3928 | * In those cases, all pages freed continuously can be expected to be in |
3c11ecf4 KH |
3929 | * the same cgroup and we have chance to coalesce uncharges. |
3930 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
3931 | * because we want to do uncharge as soon as possible. | |
3932 | */ | |
3933 | ||
3934 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
3935 | goto direct_uncharge; | |
3936 | ||
7ec99d62 | 3937 | if (nr_pages > 1) |
ec168510 AA |
3938 | goto direct_uncharge; |
3939 | ||
569b846d KH |
3940 | /* |
3941 | * In typical case, batch->memcg == mem. This means we can | |
3942 | * merge a series of uncharges to an uncharge of res_counter. | |
3943 | * If not, we uncharge res_counter ony by one. | |
3944 | */ | |
c0ff4b85 | 3945 | if (batch->memcg != memcg) |
569b846d KH |
3946 | goto direct_uncharge; |
3947 | /* remember freed charge and uncharge it later */ | |
7ffd4ca7 | 3948 | batch->nr_pages++; |
569b846d | 3949 | if (uncharge_memsw) |
7ffd4ca7 | 3950 | batch->memsw_nr_pages++; |
569b846d KH |
3951 | return; |
3952 | direct_uncharge: | |
c0ff4b85 | 3953 | res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE); |
569b846d | 3954 | if (uncharge_memsw) |
c0ff4b85 R |
3955 | res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE); |
3956 | if (unlikely(batch->memcg != memcg)) | |
3957 | memcg_oom_recover(memcg); | |
569b846d | 3958 | } |
7a81b88c | 3959 | |
8a9f3ccd | 3960 | /* |
69029cd5 | 3961 | * uncharge if !page_mapped(page) |
8a9f3ccd | 3962 | */ |
8c7c6e34 | 3963 | static struct mem_cgroup * |
0030f535 JW |
3964 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype, |
3965 | bool end_migration) | |
8a9f3ccd | 3966 | { |
c0ff4b85 | 3967 | struct mem_cgroup *memcg = NULL; |
7ec99d62 JW |
3968 | unsigned int nr_pages = 1; |
3969 | struct page_cgroup *pc; | |
b2402857 | 3970 | bool anon; |
8a9f3ccd | 3971 | |
f8d66542 | 3972 | if (mem_cgroup_disabled()) |
8c7c6e34 | 3973 | return NULL; |
4077960e | 3974 | |
37c2ac78 | 3975 | if (PageTransHuge(page)) { |
7ec99d62 | 3976 | nr_pages <<= compound_order(page); |
37c2ac78 AA |
3977 | VM_BUG_ON(!PageTransHuge(page)); |
3978 | } | |
8697d331 | 3979 | /* |
3c541e14 | 3980 | * Check if our page_cgroup is valid |
8697d331 | 3981 | */ |
52d4b9ac | 3982 | pc = lookup_page_cgroup(page); |
cfa44946 | 3983 | if (unlikely(!PageCgroupUsed(pc))) |
8c7c6e34 | 3984 | return NULL; |
b9c565d5 | 3985 | |
52d4b9ac | 3986 | lock_page_cgroup(pc); |
d13d1443 | 3987 | |
c0ff4b85 | 3988 | memcg = pc->mem_cgroup; |
8c7c6e34 | 3989 | |
d13d1443 KH |
3990 | if (!PageCgroupUsed(pc)) |
3991 | goto unlock_out; | |
3992 | ||
b2402857 KH |
3993 | anon = PageAnon(page); |
3994 | ||
d13d1443 | 3995 | switch (ctype) { |
41326c17 | 3996 | case MEM_CGROUP_CHARGE_TYPE_ANON: |
2ff76f11 KH |
3997 | /* |
3998 | * Generally PageAnon tells if it's the anon statistics to be | |
3999 | * updated; but sometimes e.g. mem_cgroup_uncharge_page() is | |
4000 | * used before page reached the stage of being marked PageAnon. | |
4001 | */ | |
b2402857 KH |
4002 | anon = true; |
4003 | /* fallthrough */ | |
8a9478ca | 4004 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c | 4005 | /* See mem_cgroup_prepare_migration() */ |
0030f535 JW |
4006 | if (page_mapped(page)) |
4007 | goto unlock_out; | |
4008 | /* | |
4009 | * Pages under migration may not be uncharged. But | |
4010 | * end_migration() /must/ be the one uncharging the | |
4011 | * unused post-migration page and so it has to call | |
4012 | * here with the migration bit still set. See the | |
4013 | * res_counter handling below. | |
4014 | */ | |
4015 | if (!end_migration && PageCgroupMigration(pc)) | |
d13d1443 KH |
4016 | goto unlock_out; |
4017 | break; | |
4018 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
4019 | if (!PageAnon(page)) { /* Shared memory */ | |
4020 | if (page->mapping && !page_is_file_cache(page)) | |
4021 | goto unlock_out; | |
4022 | } else if (page_mapped(page)) /* Anon */ | |
4023 | goto unlock_out; | |
4024 | break; | |
4025 | default: | |
4026 | break; | |
52d4b9ac | 4027 | } |
d13d1443 | 4028 | |
b070e65c | 4029 | mem_cgroup_charge_statistics(memcg, page, anon, -nr_pages); |
04046e1a | 4030 | |
52d4b9ac | 4031 | ClearPageCgroupUsed(pc); |
544122e5 KH |
4032 | /* |
4033 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
4034 | * freed from LRU. This is safe because uncharged page is expected not | |
4035 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
4036 | * special functions. | |
4037 | */ | |
b9c565d5 | 4038 | |
52d4b9ac | 4039 | unlock_page_cgroup(pc); |
f75ca962 | 4040 | /* |
c0ff4b85 | 4041 | * even after unlock, we have memcg->res.usage here and this memcg |
4050377b | 4042 | * will never be freed, so it's safe to call css_get(). |
f75ca962 | 4043 | */ |
c0ff4b85 | 4044 | memcg_check_events(memcg, page); |
f75ca962 | 4045 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
c0ff4b85 | 4046 | mem_cgroup_swap_statistics(memcg, true); |
4050377b | 4047 | css_get(&memcg->css); |
f75ca962 | 4048 | } |
0030f535 JW |
4049 | /* |
4050 | * Migration does not charge the res_counter for the | |
4051 | * replacement page, so leave it alone when phasing out the | |
4052 | * page that is unused after the migration. | |
4053 | */ | |
4054 | if (!end_migration && !mem_cgroup_is_root(memcg)) | |
c0ff4b85 | 4055 | mem_cgroup_do_uncharge(memcg, nr_pages, ctype); |
6d12e2d8 | 4056 | |
c0ff4b85 | 4057 | return memcg; |
d13d1443 KH |
4058 | |
4059 | unlock_out: | |
4060 | unlock_page_cgroup(pc); | |
8c7c6e34 | 4061 | return NULL; |
3c541e14 BS |
4062 | } |
4063 | ||
69029cd5 KH |
4064 | void mem_cgroup_uncharge_page(struct page *page) |
4065 | { | |
52d4b9ac KH |
4066 | /* early check. */ |
4067 | if (page_mapped(page)) | |
4068 | return; | |
40f23a21 | 4069 | VM_BUG_ON(page->mapping && !PageAnon(page)); |
28ccddf7 JW |
4070 | /* |
4071 | * If the page is in swap cache, uncharge should be deferred | |
4072 | * to the swap path, which also properly accounts swap usage | |
4073 | * and handles memcg lifetime. | |
4074 | * | |
4075 | * Note that this check is not stable and reclaim may add the | |
4076 | * page to swap cache at any time after this. However, if the | |
4077 | * page is not in swap cache by the time page->mapcount hits | |
4078 | * 0, there won't be any page table references to the swap | |
4079 | * slot, and reclaim will free it and not actually write the | |
4080 | * page to disk. | |
4081 | */ | |
0c59b89c JW |
4082 | if (PageSwapCache(page)) |
4083 | return; | |
0030f535 | 4084 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false); |
69029cd5 KH |
4085 | } |
4086 | ||
4087 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
4088 | { | |
4089 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 4090 | VM_BUG_ON(page->mapping); |
0030f535 | 4091 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false); |
69029cd5 KH |
4092 | } |
4093 | ||
569b846d KH |
4094 | /* |
4095 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
4096 | * In that cases, pages are freed continuously and we can expect pages | |
4097 | * are in the same memcg. All these calls itself limits the number of | |
4098 | * pages freed at once, then uncharge_start/end() is called properly. | |
4099 | * This may be called prural(2) times in a context, | |
4100 | */ | |
4101 | ||
4102 | void mem_cgroup_uncharge_start(void) | |
4103 | { | |
4104 | current->memcg_batch.do_batch++; | |
4105 | /* We can do nest. */ | |
4106 | if (current->memcg_batch.do_batch == 1) { | |
4107 | current->memcg_batch.memcg = NULL; | |
7ffd4ca7 JW |
4108 | current->memcg_batch.nr_pages = 0; |
4109 | current->memcg_batch.memsw_nr_pages = 0; | |
569b846d KH |
4110 | } |
4111 | } | |
4112 | ||
4113 | void mem_cgroup_uncharge_end(void) | |
4114 | { | |
4115 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
4116 | ||
4117 | if (!batch->do_batch) | |
4118 | return; | |
4119 | ||
4120 | batch->do_batch--; | |
4121 | if (batch->do_batch) /* If stacked, do nothing. */ | |
4122 | return; | |
4123 | ||
4124 | if (!batch->memcg) | |
4125 | return; | |
4126 | /* | |
4127 | * This "batch->memcg" is valid without any css_get/put etc... | |
4128 | * bacause we hide charges behind us. | |
4129 | */ | |
7ffd4ca7 JW |
4130 | if (batch->nr_pages) |
4131 | res_counter_uncharge(&batch->memcg->res, | |
4132 | batch->nr_pages * PAGE_SIZE); | |
4133 | if (batch->memsw_nr_pages) | |
4134 | res_counter_uncharge(&batch->memcg->memsw, | |
4135 | batch->memsw_nr_pages * PAGE_SIZE); | |
3c11ecf4 | 4136 | memcg_oom_recover(batch->memcg); |
569b846d KH |
4137 | /* forget this pointer (for sanity check) */ |
4138 | batch->memcg = NULL; | |
4139 | } | |
4140 | ||
e767e056 | 4141 | #ifdef CONFIG_SWAP |
8c7c6e34 | 4142 | /* |
e767e056 | 4143 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
4144 | * memcg information is recorded to swap_cgroup of "ent" |
4145 | */ | |
8a9478ca KH |
4146 | void |
4147 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
4148 | { |
4149 | struct mem_cgroup *memcg; | |
8a9478ca KH |
4150 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
4151 | ||
4152 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
4153 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
4154 | ||
0030f535 | 4155 | memcg = __mem_cgroup_uncharge_common(page, ctype, false); |
8c7c6e34 | 4156 | |
f75ca962 KH |
4157 | /* |
4158 | * record memcg information, if swapout && memcg != NULL, | |
4050377b | 4159 | * css_get() was called in uncharge(). |
f75ca962 KH |
4160 | */ |
4161 | if (do_swap_account && swapout && memcg) | |
a3b2d692 | 4162 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 | 4163 | } |
e767e056 | 4164 | #endif |
8c7c6e34 | 4165 | |
c255a458 | 4166 | #ifdef CONFIG_MEMCG_SWAP |
8c7c6e34 KH |
4167 | /* |
4168 | * called from swap_entry_free(). remove record in swap_cgroup and | |
4169 | * uncharge "memsw" account. | |
4170 | */ | |
4171 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 4172 | { |
8c7c6e34 | 4173 | struct mem_cgroup *memcg; |
a3b2d692 | 4174 | unsigned short id; |
8c7c6e34 KH |
4175 | |
4176 | if (!do_swap_account) | |
4177 | return; | |
4178 | ||
a3b2d692 KH |
4179 | id = swap_cgroup_record(ent, 0); |
4180 | rcu_read_lock(); | |
4181 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 4182 | if (memcg) { |
a3b2d692 KH |
4183 | /* |
4184 | * We uncharge this because swap is freed. | |
4185 | * This memcg can be obsolete one. We avoid calling css_tryget | |
4186 | */ | |
0c3e73e8 | 4187 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 4188 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 4189 | mem_cgroup_swap_statistics(memcg, false); |
4050377b | 4190 | css_put(&memcg->css); |
8c7c6e34 | 4191 | } |
a3b2d692 | 4192 | rcu_read_unlock(); |
d13d1443 | 4193 | } |
02491447 DN |
4194 | |
4195 | /** | |
4196 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
4197 | * @entry: swap entry to be moved | |
4198 | * @from: mem_cgroup which the entry is moved from | |
4199 | * @to: mem_cgroup which the entry is moved to | |
4200 | * | |
4201 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
4202 | * as the mem_cgroup's id of @from. | |
4203 | * | |
4204 | * Returns 0 on success, -EINVAL on failure. | |
4205 | * | |
4206 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
4207 | * both res and memsw, and called css_get(). | |
4208 | */ | |
4209 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 4210 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
4211 | { |
4212 | unsigned short old_id, new_id; | |
4213 | ||
4214 | old_id = css_id(&from->css); | |
4215 | new_id = css_id(&to->css); | |
4216 | ||
4217 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 4218 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 4219 | mem_cgroup_swap_statistics(to, true); |
02491447 | 4220 | /* |
483c30b5 DN |
4221 | * This function is only called from task migration context now. |
4222 | * It postpones res_counter and refcount handling till the end | |
4223 | * of task migration(mem_cgroup_clear_mc()) for performance | |
4050377b LZ |
4224 | * improvement. But we cannot postpone css_get(to) because if |
4225 | * the process that has been moved to @to does swap-in, the | |
4226 | * refcount of @to might be decreased to 0. | |
4227 | * | |
4228 | * We are in attach() phase, so the cgroup is guaranteed to be | |
4229 | * alive, so we can just call css_get(). | |
02491447 | 4230 | */ |
4050377b | 4231 | css_get(&to->css); |
02491447 DN |
4232 | return 0; |
4233 | } | |
4234 | return -EINVAL; | |
4235 | } | |
4236 | #else | |
4237 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 4238 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
4239 | { |
4240 | return -EINVAL; | |
4241 | } | |
8c7c6e34 | 4242 | #endif |
d13d1443 | 4243 | |
ae41be37 | 4244 | /* |
01b1ae63 KH |
4245 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
4246 | * page belongs to. | |
ae41be37 | 4247 | */ |
0030f535 JW |
4248 | void mem_cgroup_prepare_migration(struct page *page, struct page *newpage, |
4249 | struct mem_cgroup **memcgp) | |
ae41be37 | 4250 | { |
c0ff4b85 | 4251 | struct mem_cgroup *memcg = NULL; |
b32967ff | 4252 | unsigned int nr_pages = 1; |
7ec99d62 | 4253 | struct page_cgroup *pc; |
ac39cf8c | 4254 | enum charge_type ctype; |
8869b8f6 | 4255 | |
72835c86 | 4256 | *memcgp = NULL; |
56039efa | 4257 | |
f8d66542 | 4258 | if (mem_cgroup_disabled()) |
0030f535 | 4259 | return; |
4077960e | 4260 | |
b32967ff MG |
4261 | if (PageTransHuge(page)) |
4262 | nr_pages <<= compound_order(page); | |
4263 | ||
52d4b9ac KH |
4264 | pc = lookup_page_cgroup(page); |
4265 | lock_page_cgroup(pc); | |
4266 | if (PageCgroupUsed(pc)) { | |
c0ff4b85 R |
4267 | memcg = pc->mem_cgroup; |
4268 | css_get(&memcg->css); | |
ac39cf8c AM |
4269 | /* |
4270 | * At migrating an anonymous page, its mapcount goes down | |
4271 | * to 0 and uncharge() will be called. But, even if it's fully | |
4272 | * unmapped, migration may fail and this page has to be | |
4273 | * charged again. We set MIGRATION flag here and delay uncharge | |
4274 | * until end_migration() is called | |
4275 | * | |
4276 | * Corner Case Thinking | |
4277 | * A) | |
4278 | * When the old page was mapped as Anon and it's unmap-and-freed | |
4279 | * while migration was ongoing. | |
4280 | * If unmap finds the old page, uncharge() of it will be delayed | |
4281 | * until end_migration(). If unmap finds a new page, it's | |
4282 | * uncharged when it make mapcount to be 1->0. If unmap code | |
4283 | * finds swap_migration_entry, the new page will not be mapped | |
4284 | * and end_migration() will find it(mapcount==0). | |
4285 | * | |
4286 | * B) | |
4287 | * When the old page was mapped but migraion fails, the kernel | |
4288 | * remaps it. A charge for it is kept by MIGRATION flag even | |
4289 | * if mapcount goes down to 0. We can do remap successfully | |
4290 | * without charging it again. | |
4291 | * | |
4292 | * C) | |
4293 | * The "old" page is under lock_page() until the end of | |
4294 | * migration, so, the old page itself will not be swapped-out. | |
4295 | * If the new page is swapped out before end_migraton, our | |
4296 | * hook to usual swap-out path will catch the event. | |
4297 | */ | |
4298 | if (PageAnon(page)) | |
4299 | SetPageCgroupMigration(pc); | |
e8589cc1 | 4300 | } |
52d4b9ac | 4301 | unlock_page_cgroup(pc); |
ac39cf8c AM |
4302 | /* |
4303 | * If the page is not charged at this point, | |
4304 | * we return here. | |
4305 | */ | |
c0ff4b85 | 4306 | if (!memcg) |
0030f535 | 4307 | return; |
01b1ae63 | 4308 | |
72835c86 | 4309 | *memcgp = memcg; |
ac39cf8c AM |
4310 | /* |
4311 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
4312 | * is called before end_migration, we can catch all events on this new | |
4313 | * page. In the case new page is migrated but not remapped, new page's | |
4314 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
4315 | */ | |
ac39cf8c | 4316 | if (PageAnon(page)) |
41326c17 | 4317 | ctype = MEM_CGROUP_CHARGE_TYPE_ANON; |
ac39cf8c | 4318 | else |
62ba7442 | 4319 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; |
0030f535 JW |
4320 | /* |
4321 | * The page is committed to the memcg, but it's not actually | |
4322 | * charged to the res_counter since we plan on replacing the | |
4323 | * old one and only one page is going to be left afterwards. | |
4324 | */ | |
b32967ff | 4325 | __mem_cgroup_commit_charge(memcg, newpage, nr_pages, ctype, false); |
ae41be37 | 4326 | } |
8869b8f6 | 4327 | |
69029cd5 | 4328 | /* remove redundant charge if migration failed*/ |
c0ff4b85 | 4329 | void mem_cgroup_end_migration(struct mem_cgroup *memcg, |
50de1dd9 | 4330 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 4331 | { |
ac39cf8c | 4332 | struct page *used, *unused; |
01b1ae63 | 4333 | struct page_cgroup *pc; |
b2402857 | 4334 | bool anon; |
01b1ae63 | 4335 | |
c0ff4b85 | 4336 | if (!memcg) |
01b1ae63 | 4337 | return; |
b25ed609 | 4338 | |
50de1dd9 | 4339 | if (!migration_ok) { |
ac39cf8c AM |
4340 | used = oldpage; |
4341 | unused = newpage; | |
01b1ae63 | 4342 | } else { |
ac39cf8c | 4343 | used = newpage; |
01b1ae63 KH |
4344 | unused = oldpage; |
4345 | } | |
0030f535 | 4346 | anon = PageAnon(used); |
7d188958 JW |
4347 | __mem_cgroup_uncharge_common(unused, |
4348 | anon ? MEM_CGROUP_CHARGE_TYPE_ANON | |
4349 | : MEM_CGROUP_CHARGE_TYPE_CACHE, | |
4350 | true); | |
0030f535 | 4351 | css_put(&memcg->css); |
69029cd5 | 4352 | /* |
ac39cf8c AM |
4353 | * We disallowed uncharge of pages under migration because mapcount |
4354 | * of the page goes down to zero, temporarly. | |
4355 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 4356 | */ |
ac39cf8c AM |
4357 | pc = lookup_page_cgroup(oldpage); |
4358 | lock_page_cgroup(pc); | |
4359 | ClearPageCgroupMigration(pc); | |
4360 | unlock_page_cgroup(pc); | |
ac39cf8c | 4361 | |
01b1ae63 | 4362 | /* |
ac39cf8c AM |
4363 | * If a page is a file cache, radix-tree replacement is very atomic |
4364 | * and we can skip this check. When it was an Anon page, its mapcount | |
4365 | * goes down to 0. But because we added MIGRATION flage, it's not | |
4366 | * uncharged yet. There are several case but page->mapcount check | |
4367 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
4368 | * check. (see prepare_charge() also) | |
69029cd5 | 4369 | */ |
b2402857 | 4370 | if (anon) |
ac39cf8c | 4371 | mem_cgroup_uncharge_page(used); |
ae41be37 | 4372 | } |
78fb7466 | 4373 | |
ab936cbc KH |
4374 | /* |
4375 | * At replace page cache, newpage is not under any memcg but it's on | |
4376 | * LRU. So, this function doesn't touch res_counter but handles LRU | |
4377 | * in correct way. Both pages are locked so we cannot race with uncharge. | |
4378 | */ | |
4379 | void mem_cgroup_replace_page_cache(struct page *oldpage, | |
4380 | struct page *newpage) | |
4381 | { | |
bde05d1c | 4382 | struct mem_cgroup *memcg = NULL; |
ab936cbc | 4383 | struct page_cgroup *pc; |
ab936cbc | 4384 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
ab936cbc KH |
4385 | |
4386 | if (mem_cgroup_disabled()) | |
4387 | return; | |
4388 | ||
4389 | pc = lookup_page_cgroup(oldpage); | |
4390 | /* fix accounting on old pages */ | |
4391 | lock_page_cgroup(pc); | |
bde05d1c HD |
4392 | if (PageCgroupUsed(pc)) { |
4393 | memcg = pc->mem_cgroup; | |
b070e65c | 4394 | mem_cgroup_charge_statistics(memcg, oldpage, false, -1); |
bde05d1c HD |
4395 | ClearPageCgroupUsed(pc); |
4396 | } | |
ab936cbc KH |
4397 | unlock_page_cgroup(pc); |
4398 | ||
bde05d1c HD |
4399 | /* |
4400 | * When called from shmem_replace_page(), in some cases the | |
4401 | * oldpage has already been charged, and in some cases not. | |
4402 | */ | |
4403 | if (!memcg) | |
4404 | return; | |
ab936cbc KH |
4405 | /* |
4406 | * Even if newpage->mapping was NULL before starting replacement, | |
4407 | * the newpage may be on LRU(or pagevec for LRU) already. We lock | |
4408 | * LRU while we overwrite pc->mem_cgroup. | |
4409 | */ | |
ce587e65 | 4410 | __mem_cgroup_commit_charge(memcg, newpage, 1, type, true); |
ab936cbc KH |
4411 | } |
4412 | ||
f212ad7c DN |
4413 | #ifdef CONFIG_DEBUG_VM |
4414 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
4415 | { | |
4416 | struct page_cgroup *pc; | |
4417 | ||
4418 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
4419 | /* |
4420 | * Can be NULL while feeding pages into the page allocator for | |
4421 | * the first time, i.e. during boot or memory hotplug; | |
4422 | * or when mem_cgroup_disabled(). | |
4423 | */ | |
f212ad7c DN |
4424 | if (likely(pc) && PageCgroupUsed(pc)) |
4425 | return pc; | |
4426 | return NULL; | |
4427 | } | |
4428 | ||
4429 | bool mem_cgroup_bad_page_check(struct page *page) | |
4430 | { | |
4431 | if (mem_cgroup_disabled()) | |
4432 | return false; | |
4433 | ||
4434 | return lookup_page_cgroup_used(page) != NULL; | |
4435 | } | |
4436 | ||
4437 | void mem_cgroup_print_bad_page(struct page *page) | |
4438 | { | |
4439 | struct page_cgroup *pc; | |
4440 | ||
4441 | pc = lookup_page_cgroup_used(page); | |
4442 | if (pc) { | |
d045197f AM |
4443 | pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n", |
4444 | pc, pc->flags, pc->mem_cgroup); | |
f212ad7c DN |
4445 | } |
4446 | } | |
4447 | #endif | |
4448 | ||
d38d2a75 | 4449 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 4450 | unsigned long long val) |
628f4235 | 4451 | { |
81d39c20 | 4452 | int retry_count; |
3c11ecf4 | 4453 | u64 memswlimit, memlimit; |
628f4235 | 4454 | int ret = 0; |
81d39c20 KH |
4455 | int children = mem_cgroup_count_children(memcg); |
4456 | u64 curusage, oldusage; | |
3c11ecf4 | 4457 | int enlarge; |
81d39c20 KH |
4458 | |
4459 | /* | |
4460 | * For keeping hierarchical_reclaim simple, how long we should retry | |
4461 | * is depends on callers. We set our retry-count to be function | |
4462 | * of # of children which we should visit in this loop. | |
4463 | */ | |
4464 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
4465 | ||
4466 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 4467 | |
3c11ecf4 | 4468 | enlarge = 0; |
8c7c6e34 | 4469 | while (retry_count) { |
628f4235 KH |
4470 | if (signal_pending(current)) { |
4471 | ret = -EINTR; | |
4472 | break; | |
4473 | } | |
8c7c6e34 KH |
4474 | /* |
4475 | * Rather than hide all in some function, I do this in | |
4476 | * open coded manner. You see what this really does. | |
aaad153e | 4477 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
4478 | */ |
4479 | mutex_lock(&set_limit_mutex); | |
4480 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
4481 | if (memswlimit < val) { | |
4482 | ret = -EINVAL; | |
4483 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
4484 | break; |
4485 | } | |
3c11ecf4 KH |
4486 | |
4487 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4488 | if (memlimit < val) | |
4489 | enlarge = 1; | |
4490 | ||
8c7c6e34 | 4491 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
4492 | if (!ret) { |
4493 | if (memswlimit == val) | |
4494 | memcg->memsw_is_minimum = true; | |
4495 | else | |
4496 | memcg->memsw_is_minimum = false; | |
4497 | } | |
8c7c6e34 KH |
4498 | mutex_unlock(&set_limit_mutex); |
4499 | ||
4500 | if (!ret) | |
4501 | break; | |
4502 | ||
5660048c JW |
4503 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
4504 | MEM_CGROUP_RECLAIM_SHRINK); | |
81d39c20 KH |
4505 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
4506 | /* Usage is reduced ? */ | |
4507 | if (curusage >= oldusage) | |
4508 | retry_count--; | |
4509 | else | |
4510 | oldusage = curusage; | |
8c7c6e34 | 4511 | } |
3c11ecf4 KH |
4512 | if (!ret && enlarge) |
4513 | memcg_oom_recover(memcg); | |
14797e23 | 4514 | |
8c7c6e34 KH |
4515 | return ret; |
4516 | } | |
4517 | ||
338c8431 LZ |
4518 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
4519 | unsigned long long val) | |
8c7c6e34 | 4520 | { |
81d39c20 | 4521 | int retry_count; |
3c11ecf4 | 4522 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
4523 | int children = mem_cgroup_count_children(memcg); |
4524 | int ret = -EBUSY; | |
3c11ecf4 | 4525 | int enlarge = 0; |
8c7c6e34 | 4526 | |
81d39c20 KH |
4527 | /* see mem_cgroup_resize_res_limit */ |
4528 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
4529 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
4530 | while (retry_count) { |
4531 | if (signal_pending(current)) { | |
4532 | ret = -EINTR; | |
4533 | break; | |
4534 | } | |
4535 | /* | |
4536 | * Rather than hide all in some function, I do this in | |
4537 | * open coded manner. You see what this really does. | |
aaad153e | 4538 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
4539 | */ |
4540 | mutex_lock(&set_limit_mutex); | |
4541 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4542 | if (memlimit > val) { | |
4543 | ret = -EINVAL; | |
4544 | mutex_unlock(&set_limit_mutex); | |
4545 | break; | |
4546 | } | |
3c11ecf4 KH |
4547 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
4548 | if (memswlimit < val) | |
4549 | enlarge = 1; | |
8c7c6e34 | 4550 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
4551 | if (!ret) { |
4552 | if (memlimit == val) | |
4553 | memcg->memsw_is_minimum = true; | |
4554 | else | |
4555 | memcg->memsw_is_minimum = false; | |
4556 | } | |
8c7c6e34 KH |
4557 | mutex_unlock(&set_limit_mutex); |
4558 | ||
4559 | if (!ret) | |
4560 | break; | |
4561 | ||
5660048c JW |
4562 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
4563 | MEM_CGROUP_RECLAIM_NOSWAP | | |
4564 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 4565 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 4566 | /* Usage is reduced ? */ |
8c7c6e34 | 4567 | if (curusage >= oldusage) |
628f4235 | 4568 | retry_count--; |
81d39c20 KH |
4569 | else |
4570 | oldusage = curusage; | |
628f4235 | 4571 | } |
3c11ecf4 KH |
4572 | if (!ret && enlarge) |
4573 | memcg_oom_recover(memcg); | |
628f4235 KH |
4574 | return ret; |
4575 | } | |
4576 | ||
2ef37d3f MH |
4577 | /** |
4578 | * mem_cgroup_force_empty_list - clears LRU of a group | |
4579 | * @memcg: group to clear | |
4580 | * @node: NUMA node | |
4581 | * @zid: zone id | |
4582 | * @lru: lru to to clear | |
4583 | * | |
3c935d18 | 4584 | * Traverse a specified page_cgroup list and try to drop them all. This doesn't |
2ef37d3f MH |
4585 | * reclaim the pages page themselves - pages are moved to the parent (or root) |
4586 | * group. | |
cc847582 | 4587 | */ |
2ef37d3f | 4588 | static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
08e552c6 | 4589 | int node, int zid, enum lru_list lru) |
cc847582 | 4590 | { |
bea8c150 | 4591 | struct lruvec *lruvec; |
2ef37d3f | 4592 | unsigned long flags; |
072c56c1 | 4593 | struct list_head *list; |
925b7673 JW |
4594 | struct page *busy; |
4595 | struct zone *zone; | |
072c56c1 | 4596 | |
08e552c6 | 4597 | zone = &NODE_DATA(node)->node_zones[zid]; |
bea8c150 HD |
4598 | lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
4599 | list = &lruvec->lists[lru]; | |
cc847582 | 4600 | |
f817ed48 | 4601 | busy = NULL; |
2ef37d3f | 4602 | do { |
925b7673 | 4603 | struct page_cgroup *pc; |
5564e88b JW |
4604 | struct page *page; |
4605 | ||
08e552c6 | 4606 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 4607 | if (list_empty(list)) { |
08e552c6 | 4608 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 4609 | break; |
f817ed48 | 4610 | } |
925b7673 JW |
4611 | page = list_entry(list->prev, struct page, lru); |
4612 | if (busy == page) { | |
4613 | list_move(&page->lru, list); | |
648bcc77 | 4614 | busy = NULL; |
08e552c6 | 4615 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
4616 | continue; |
4617 | } | |
08e552c6 | 4618 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 4619 | |
925b7673 | 4620 | pc = lookup_page_cgroup(page); |
5564e88b | 4621 | |
3c935d18 | 4622 | if (mem_cgroup_move_parent(page, pc, memcg)) { |
f817ed48 | 4623 | /* found lock contention or "pc" is obsolete. */ |
925b7673 | 4624 | busy = page; |
f817ed48 KH |
4625 | cond_resched(); |
4626 | } else | |
4627 | busy = NULL; | |
2ef37d3f | 4628 | } while (!list_empty(list)); |
cc847582 KH |
4629 | } |
4630 | ||
4631 | /* | |
c26251f9 MH |
4632 | * make mem_cgroup's charge to be 0 if there is no task by moving |
4633 | * all the charges and pages to the parent. | |
cc847582 | 4634 | * This enables deleting this mem_cgroup. |
c26251f9 MH |
4635 | * |
4636 | * Caller is responsible for holding css reference on the memcg. | |
cc847582 | 4637 | */ |
ab5196c2 | 4638 | static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg) |
cc847582 | 4639 | { |
c26251f9 | 4640 | int node, zid; |
bea207c8 | 4641 | u64 usage; |
f817ed48 | 4642 | |
fce66477 | 4643 | do { |
52d4b9ac KH |
4644 | /* This is for making all *used* pages to be on LRU. */ |
4645 | lru_add_drain_all(); | |
c0ff4b85 | 4646 | drain_all_stock_sync(memcg); |
c0ff4b85 | 4647 | mem_cgroup_start_move(memcg); |
31aaea4a | 4648 | for_each_node_state(node, N_MEMORY) { |
2ef37d3f | 4649 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
f156ab93 HD |
4650 | enum lru_list lru; |
4651 | for_each_lru(lru) { | |
2ef37d3f | 4652 | mem_cgroup_force_empty_list(memcg, |
f156ab93 | 4653 | node, zid, lru); |
f817ed48 | 4654 | } |
1ecaab2b | 4655 | } |
f817ed48 | 4656 | } |
c0ff4b85 R |
4657 | mem_cgroup_end_move(memcg); |
4658 | memcg_oom_recover(memcg); | |
52d4b9ac | 4659 | cond_resched(); |
f817ed48 | 4660 | |
2ef37d3f | 4661 | /* |
bea207c8 GC |
4662 | * Kernel memory may not necessarily be trackable to a specific |
4663 | * process. So they are not migrated, and therefore we can't | |
4664 | * expect their value to drop to 0 here. | |
4665 | * Having res filled up with kmem only is enough. | |
4666 | * | |
2ef37d3f MH |
4667 | * This is a safety check because mem_cgroup_force_empty_list |
4668 | * could have raced with mem_cgroup_replace_page_cache callers | |
4669 | * so the lru seemed empty but the page could have been added | |
4670 | * right after the check. RES_USAGE should be safe as we always | |
4671 | * charge before adding to the LRU. | |
4672 | */ | |
bea207c8 GC |
4673 | usage = res_counter_read_u64(&memcg->res, RES_USAGE) - |
4674 | res_counter_read_u64(&memcg->kmem, RES_USAGE); | |
4675 | } while (usage > 0); | |
c26251f9 MH |
4676 | } |
4677 | ||
b5f99b53 GC |
4678 | /* |
4679 | * This mainly exists for tests during the setting of set of use_hierarchy. | |
4680 | * Since this is the very setting we are changing, the current hierarchy value | |
4681 | * is meaningless | |
4682 | */ | |
4683 | static inline bool __memcg_has_children(struct mem_cgroup *memcg) | |
4684 | { | |
492eb21b | 4685 | struct cgroup_subsys_state *pos; |
b5f99b53 GC |
4686 | |
4687 | /* bounce at first found */ | |
492eb21b | 4688 | css_for_each_child(pos, &memcg->css) |
b5f99b53 GC |
4689 | return true; |
4690 | return false; | |
4691 | } | |
4692 | ||
4693 | /* | |
0999821b GC |
4694 | * Must be called with memcg_create_mutex held, unless the cgroup is guaranteed |
4695 | * to be already dead (as in mem_cgroup_force_empty, for instance). This is | |
b5f99b53 GC |
4696 | * from mem_cgroup_count_children(), in the sense that we don't really care how |
4697 | * many children we have; we only need to know if we have any. It also counts | |
4698 | * any memcg without hierarchy as infertile. | |
4699 | */ | |
4700 | static inline bool memcg_has_children(struct mem_cgroup *memcg) | |
4701 | { | |
4702 | return memcg->use_hierarchy && __memcg_has_children(memcg); | |
4703 | } | |
4704 | ||
c26251f9 MH |
4705 | /* |
4706 | * Reclaims as many pages from the given memcg as possible and moves | |
4707 | * the rest to the parent. | |
4708 | * | |
4709 | * Caller is responsible for holding css reference for memcg. | |
4710 | */ | |
4711 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
4712 | { | |
4713 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
4714 | struct cgroup *cgrp = memcg->css.cgroup; | |
f817ed48 | 4715 | |
c1e862c1 | 4716 | /* returns EBUSY if there is a task or if we come here twice. */ |
c26251f9 MH |
4717 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
4718 | return -EBUSY; | |
4719 | ||
c1e862c1 KH |
4720 | /* we call try-to-free pages for make this cgroup empty */ |
4721 | lru_add_drain_all(); | |
f817ed48 | 4722 | /* try to free all pages in this cgroup */ |
569530fb | 4723 | while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) { |
f817ed48 | 4724 | int progress; |
c1e862c1 | 4725 | |
c26251f9 MH |
4726 | if (signal_pending(current)) |
4727 | return -EINTR; | |
4728 | ||
c0ff4b85 | 4729 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, |
185efc0f | 4730 | false); |
c1e862c1 | 4731 | if (!progress) { |
f817ed48 | 4732 | nr_retries--; |
c1e862c1 | 4733 | /* maybe some writeback is necessary */ |
8aa7e847 | 4734 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 4735 | } |
f817ed48 KH |
4736 | |
4737 | } | |
08e552c6 | 4738 | lru_add_drain(); |
ab5196c2 MH |
4739 | mem_cgroup_reparent_charges(memcg); |
4740 | ||
4741 | return 0; | |
cc847582 KH |
4742 | } |
4743 | ||
182446d0 TH |
4744 | static int mem_cgroup_force_empty_write(struct cgroup_subsys_state *css, |
4745 | unsigned int event) | |
c1e862c1 | 4746 | { |
182446d0 | 4747 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c26251f9 | 4748 | |
d8423011 MH |
4749 | if (mem_cgroup_is_root(memcg)) |
4750 | return -EINVAL; | |
c33bd835 | 4751 | return mem_cgroup_force_empty(memcg); |
c1e862c1 KH |
4752 | } |
4753 | ||
182446d0 TH |
4754 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
4755 | struct cftype *cft) | |
18f59ea7 | 4756 | { |
182446d0 | 4757 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
4758 | } |
4759 | ||
182446d0 TH |
4760 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
4761 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
4762 | { |
4763 | int retval = 0; | |
182446d0 | 4764 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
63876986 | 4765 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(css_parent(&memcg->css)); |
18f59ea7 | 4766 | |
0999821b | 4767 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
4768 | |
4769 | if (memcg->use_hierarchy == val) | |
4770 | goto out; | |
4771 | ||
18f59ea7 | 4772 | /* |
af901ca1 | 4773 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
4774 | * in the child subtrees. If it is unset, then the change can |
4775 | * occur, provided the current cgroup has no children. | |
4776 | * | |
4777 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
4778 | * set if there are no children. | |
4779 | */ | |
c0ff4b85 | 4780 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 4781 | (val == 1 || val == 0)) { |
b5f99b53 | 4782 | if (!__memcg_has_children(memcg)) |
c0ff4b85 | 4783 | memcg->use_hierarchy = val; |
18f59ea7 BS |
4784 | else |
4785 | retval = -EBUSY; | |
4786 | } else | |
4787 | retval = -EINVAL; | |
567fb435 GC |
4788 | |
4789 | out: | |
0999821b | 4790 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
4791 | |
4792 | return retval; | |
4793 | } | |
4794 | ||
0c3e73e8 | 4795 | |
c0ff4b85 | 4796 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg, |
7a159cc9 | 4797 | enum mem_cgroup_stat_index idx) |
0c3e73e8 | 4798 | { |
7d74b06f | 4799 | struct mem_cgroup *iter; |
7a159cc9 | 4800 | long val = 0; |
0c3e73e8 | 4801 | |
7a159cc9 | 4802 | /* Per-cpu values can be negative, use a signed accumulator */ |
c0ff4b85 | 4803 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f KH |
4804 | val += mem_cgroup_read_stat(iter, idx); |
4805 | ||
4806 | if (val < 0) /* race ? */ | |
4807 | val = 0; | |
4808 | return val; | |
0c3e73e8 BS |
4809 | } |
4810 | ||
c0ff4b85 | 4811 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
104f3928 | 4812 | { |
7d74b06f | 4813 | u64 val; |
104f3928 | 4814 | |
c0ff4b85 | 4815 | if (!mem_cgroup_is_root(memcg)) { |
104f3928 | 4816 | if (!swap) |
65c64ce8 | 4817 | return res_counter_read_u64(&memcg->res, RES_USAGE); |
104f3928 | 4818 | else |
65c64ce8 | 4819 | return res_counter_read_u64(&memcg->memsw, RES_USAGE); |
104f3928 KS |
4820 | } |
4821 | ||
b070e65c DR |
4822 | /* |
4823 | * Transparent hugepages are still accounted for in MEM_CGROUP_STAT_RSS | |
4824 | * as well as in MEM_CGROUP_STAT_RSS_HUGE. | |
4825 | */ | |
c0ff4b85 R |
4826 | val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE); |
4827 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS); | |
104f3928 | 4828 | |
7d74b06f | 4829 | if (swap) |
bff6bb83 | 4830 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP); |
104f3928 KS |
4831 | |
4832 | return val << PAGE_SHIFT; | |
4833 | } | |
4834 | ||
182446d0 TH |
4835 | static ssize_t mem_cgroup_read(struct cgroup_subsys_state *css, |
4836 | struct cftype *cft, struct file *file, | |
4837 | char __user *buf, size_t nbytes, loff_t *ppos) | |
8cdea7c0 | 4838 | { |
182446d0 | 4839 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
af36f906 | 4840 | char str[64]; |
104f3928 | 4841 | u64 val; |
86ae53e1 GC |
4842 | int name, len; |
4843 | enum res_type type; | |
8c7c6e34 KH |
4844 | |
4845 | type = MEMFILE_TYPE(cft->private); | |
4846 | name = MEMFILE_ATTR(cft->private); | |
af36f906 | 4847 | |
8c7c6e34 KH |
4848 | switch (type) { |
4849 | case _MEM: | |
104f3928 | 4850 | if (name == RES_USAGE) |
c0ff4b85 | 4851 | val = mem_cgroup_usage(memcg, false); |
104f3928 | 4852 | else |
c0ff4b85 | 4853 | val = res_counter_read_u64(&memcg->res, name); |
8c7c6e34 KH |
4854 | break; |
4855 | case _MEMSWAP: | |
104f3928 | 4856 | if (name == RES_USAGE) |
c0ff4b85 | 4857 | val = mem_cgroup_usage(memcg, true); |
104f3928 | 4858 | else |
c0ff4b85 | 4859 | val = res_counter_read_u64(&memcg->memsw, name); |
8c7c6e34 | 4860 | break; |
510fc4e1 GC |
4861 | case _KMEM: |
4862 | val = res_counter_read_u64(&memcg->kmem, name); | |
4863 | break; | |
8c7c6e34 KH |
4864 | default: |
4865 | BUG(); | |
8c7c6e34 | 4866 | } |
af36f906 TH |
4867 | |
4868 | len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val); | |
4869 | return simple_read_from_buffer(buf, nbytes, ppos, str, len); | |
8cdea7c0 | 4870 | } |
510fc4e1 | 4871 | |
182446d0 | 4872 | static int memcg_update_kmem_limit(struct cgroup_subsys_state *css, u64 val) |
510fc4e1 GC |
4873 | { |
4874 | int ret = -EINVAL; | |
4875 | #ifdef CONFIG_MEMCG_KMEM | |
182446d0 | 4876 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
510fc4e1 GC |
4877 | /* |
4878 | * For simplicity, we won't allow this to be disabled. It also can't | |
4879 | * be changed if the cgroup has children already, or if tasks had | |
4880 | * already joined. | |
4881 | * | |
4882 | * If tasks join before we set the limit, a person looking at | |
4883 | * kmem.usage_in_bytes will have no way to determine when it took | |
4884 | * place, which makes the value quite meaningless. | |
4885 | * | |
4886 | * After it first became limited, changes in the value of the limit are | |
4887 | * of course permitted. | |
510fc4e1 | 4888 | */ |
0999821b | 4889 | mutex_lock(&memcg_create_mutex); |
510fc4e1 GC |
4890 | mutex_lock(&set_limit_mutex); |
4891 | if (!memcg->kmem_account_flags && val != RESOURCE_MAX) { | |
182446d0 | 4892 | if (cgroup_task_count(css->cgroup) || memcg_has_children(memcg)) { |
510fc4e1 GC |
4893 | ret = -EBUSY; |
4894 | goto out; | |
4895 | } | |
4896 | ret = res_counter_set_limit(&memcg->kmem, val); | |
4897 | VM_BUG_ON(ret); | |
4898 | ||
55007d84 GC |
4899 | ret = memcg_update_cache_sizes(memcg); |
4900 | if (ret) { | |
4901 | res_counter_set_limit(&memcg->kmem, RESOURCE_MAX); | |
4902 | goto out; | |
4903 | } | |
692e89ab GC |
4904 | static_key_slow_inc(&memcg_kmem_enabled_key); |
4905 | /* | |
4906 | * setting the active bit after the inc will guarantee no one | |
4907 | * starts accounting before all call sites are patched | |
4908 | */ | |
4909 | memcg_kmem_set_active(memcg); | |
510fc4e1 GC |
4910 | } else |
4911 | ret = res_counter_set_limit(&memcg->kmem, val); | |
4912 | out: | |
4913 | mutex_unlock(&set_limit_mutex); | |
0999821b | 4914 | mutex_unlock(&memcg_create_mutex); |
510fc4e1 GC |
4915 | #endif |
4916 | return ret; | |
4917 | } | |
4918 | ||
6d043990 | 4919 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 | 4920 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 4921 | { |
55007d84 | 4922 | int ret = 0; |
510fc4e1 GC |
4923 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
4924 | if (!parent) | |
55007d84 GC |
4925 | goto out; |
4926 | ||
510fc4e1 | 4927 | memcg->kmem_account_flags = parent->kmem_account_flags; |
a8964b9b GC |
4928 | /* |
4929 | * When that happen, we need to disable the static branch only on those | |
4930 | * memcgs that enabled it. To achieve this, we would be forced to | |
4931 | * complicate the code by keeping track of which memcgs were the ones | |
4932 | * that actually enabled limits, and which ones got it from its | |
4933 | * parents. | |
4934 | * | |
4935 | * It is a lot simpler just to do static_key_slow_inc() on every child | |
4936 | * that is accounted. | |
4937 | */ | |
55007d84 GC |
4938 | if (!memcg_kmem_is_active(memcg)) |
4939 | goto out; | |
4940 | ||
4941 | /* | |
10d5ebf4 LZ |
4942 | * __mem_cgroup_free() will issue static_key_slow_dec() because this |
4943 | * memcg is active already. If the later initialization fails then the | |
4944 | * cgroup core triggers the cleanup so we do not have to do it here. | |
55007d84 | 4945 | */ |
55007d84 GC |
4946 | static_key_slow_inc(&memcg_kmem_enabled_key); |
4947 | ||
4948 | mutex_lock(&set_limit_mutex); | |
425c598d | 4949 | memcg_stop_kmem_account(); |
55007d84 | 4950 | ret = memcg_update_cache_sizes(memcg); |
425c598d | 4951 | memcg_resume_kmem_account(); |
55007d84 | 4952 | mutex_unlock(&set_limit_mutex); |
55007d84 GC |
4953 | out: |
4954 | return ret; | |
510fc4e1 | 4955 | } |
6d043990 | 4956 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 4957 | |
628f4235 KH |
4958 | /* |
4959 | * The user of this function is... | |
4960 | * RES_LIMIT. | |
4961 | */ | |
182446d0 | 4962 | static int mem_cgroup_write(struct cgroup_subsys_state *css, struct cftype *cft, |
856c13aa | 4963 | const char *buffer) |
8cdea7c0 | 4964 | { |
182446d0 | 4965 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
86ae53e1 GC |
4966 | enum res_type type; |
4967 | int name; | |
628f4235 KH |
4968 | unsigned long long val; |
4969 | int ret; | |
4970 | ||
8c7c6e34 KH |
4971 | type = MEMFILE_TYPE(cft->private); |
4972 | name = MEMFILE_ATTR(cft->private); | |
af36f906 | 4973 | |
8c7c6e34 | 4974 | switch (name) { |
628f4235 | 4975 | case RES_LIMIT: |
4b3bde4c BS |
4976 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
4977 | ret = -EINVAL; | |
4978 | break; | |
4979 | } | |
628f4235 KH |
4980 | /* This function does all necessary parse...reuse it */ |
4981 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
4982 | if (ret) |
4983 | break; | |
4984 | if (type == _MEM) | |
628f4235 | 4985 | ret = mem_cgroup_resize_limit(memcg, val); |
510fc4e1 | 4986 | else if (type == _MEMSWAP) |
8c7c6e34 | 4987 | ret = mem_cgroup_resize_memsw_limit(memcg, val); |
510fc4e1 | 4988 | else if (type == _KMEM) |
182446d0 | 4989 | ret = memcg_update_kmem_limit(css, val); |
510fc4e1 GC |
4990 | else |
4991 | return -EINVAL; | |
628f4235 | 4992 | break; |
296c81d8 BS |
4993 | case RES_SOFT_LIMIT: |
4994 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
4995 | if (ret) | |
4996 | break; | |
4997 | /* | |
4998 | * For memsw, soft limits are hard to implement in terms | |
4999 | * of semantics, for now, we support soft limits for | |
5000 | * control without swap | |
5001 | */ | |
5002 | if (type == _MEM) | |
5003 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
5004 | else | |
5005 | ret = -EINVAL; | |
5006 | break; | |
628f4235 KH |
5007 | default: |
5008 | ret = -EINVAL; /* should be BUG() ? */ | |
5009 | break; | |
5010 | } | |
5011 | return ret; | |
8cdea7c0 BS |
5012 | } |
5013 | ||
fee7b548 KH |
5014 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
5015 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
5016 | { | |
fee7b548 KH |
5017 | unsigned long long min_limit, min_memsw_limit, tmp; |
5018 | ||
5019 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
5020 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
fee7b548 KH |
5021 | if (!memcg->use_hierarchy) |
5022 | goto out; | |
5023 | ||
63876986 TH |
5024 | while (css_parent(&memcg->css)) { |
5025 | memcg = mem_cgroup_from_css(css_parent(&memcg->css)); | |
fee7b548 KH |
5026 | if (!memcg->use_hierarchy) |
5027 | break; | |
5028 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
5029 | min_limit = min(min_limit, tmp); | |
5030 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
5031 | min_memsw_limit = min(min_memsw_limit, tmp); | |
5032 | } | |
5033 | out: | |
5034 | *mem_limit = min_limit; | |
5035 | *memsw_limit = min_memsw_limit; | |
fee7b548 KH |
5036 | } |
5037 | ||
182446d0 | 5038 | static int mem_cgroup_reset(struct cgroup_subsys_state *css, unsigned int event) |
c84872e1 | 5039 | { |
182446d0 | 5040 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
86ae53e1 GC |
5041 | int name; |
5042 | enum res_type type; | |
c84872e1 | 5043 | |
8c7c6e34 KH |
5044 | type = MEMFILE_TYPE(event); |
5045 | name = MEMFILE_ATTR(event); | |
af36f906 | 5046 | |
8c7c6e34 | 5047 | switch (name) { |
29f2a4da | 5048 | case RES_MAX_USAGE: |
8c7c6e34 | 5049 | if (type == _MEM) |
c0ff4b85 | 5050 | res_counter_reset_max(&memcg->res); |
510fc4e1 | 5051 | else if (type == _MEMSWAP) |
c0ff4b85 | 5052 | res_counter_reset_max(&memcg->memsw); |
510fc4e1 GC |
5053 | else if (type == _KMEM) |
5054 | res_counter_reset_max(&memcg->kmem); | |
5055 | else | |
5056 | return -EINVAL; | |
29f2a4da PE |
5057 | break; |
5058 | case RES_FAILCNT: | |
8c7c6e34 | 5059 | if (type == _MEM) |
c0ff4b85 | 5060 | res_counter_reset_failcnt(&memcg->res); |
510fc4e1 | 5061 | else if (type == _MEMSWAP) |
c0ff4b85 | 5062 | res_counter_reset_failcnt(&memcg->memsw); |
510fc4e1 GC |
5063 | else if (type == _KMEM) |
5064 | res_counter_reset_failcnt(&memcg->kmem); | |
5065 | else | |
5066 | return -EINVAL; | |
29f2a4da PE |
5067 | break; |
5068 | } | |
f64c3f54 | 5069 | |
85cc59db | 5070 | return 0; |
c84872e1 PE |
5071 | } |
5072 | ||
182446d0 | 5073 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
5074 | struct cftype *cft) |
5075 | { | |
182446d0 | 5076 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
5077 | } |
5078 | ||
02491447 | 5079 | #ifdef CONFIG_MMU |
182446d0 | 5080 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
5081 | struct cftype *cft, u64 val) |
5082 | { | |
182446d0 | 5083 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 DN |
5084 | |
5085 | if (val >= (1 << NR_MOVE_TYPE)) | |
5086 | return -EINVAL; | |
ee5e8472 | 5087 | |
7dc74be0 | 5088 | /* |
ee5e8472 GC |
5089 | * No kind of locking is needed in here, because ->can_attach() will |
5090 | * check this value once in the beginning of the process, and then carry | |
5091 | * on with stale data. This means that changes to this value will only | |
5092 | * affect task migrations starting after the change. | |
7dc74be0 | 5093 | */ |
c0ff4b85 | 5094 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
5095 | return 0; |
5096 | } | |
02491447 | 5097 | #else |
182446d0 | 5098 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
5099 | struct cftype *cft, u64 val) |
5100 | { | |
5101 | return -ENOSYS; | |
5102 | } | |
5103 | #endif | |
7dc74be0 | 5104 | |
406eb0c9 | 5105 | #ifdef CONFIG_NUMA |
182446d0 TH |
5106 | static int memcg_numa_stat_show(struct cgroup_subsys_state *css, |
5107 | struct cftype *cft, struct seq_file *m) | |
406eb0c9 YH |
5108 | { |
5109 | int nid; | |
5110 | unsigned long total_nr, file_nr, anon_nr, unevictable_nr; | |
5111 | unsigned long node_nr; | |
182446d0 | 5112 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
406eb0c9 | 5113 | |
d79154bb | 5114 | total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL); |
406eb0c9 | 5115 | seq_printf(m, "total=%lu", total_nr); |
31aaea4a | 5116 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5117 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL); |
406eb0c9 YH |
5118 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5119 | } | |
5120 | seq_putc(m, '\n'); | |
5121 | ||
d79154bb | 5122 | file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE); |
406eb0c9 | 5123 | seq_printf(m, "file=%lu", file_nr); |
31aaea4a | 5124 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5125 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 5126 | LRU_ALL_FILE); |
406eb0c9 YH |
5127 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5128 | } | |
5129 | seq_putc(m, '\n'); | |
5130 | ||
d79154bb | 5131 | anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON); |
406eb0c9 | 5132 | seq_printf(m, "anon=%lu", anon_nr); |
31aaea4a | 5133 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5134 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 5135 | LRU_ALL_ANON); |
406eb0c9 YH |
5136 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5137 | } | |
5138 | seq_putc(m, '\n'); | |
5139 | ||
d79154bb | 5140 | unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE)); |
406eb0c9 | 5141 | seq_printf(m, "unevictable=%lu", unevictable_nr); |
31aaea4a | 5142 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5143 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 5144 | BIT(LRU_UNEVICTABLE)); |
406eb0c9 YH |
5145 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5146 | } | |
5147 | seq_putc(m, '\n'); | |
5148 | return 0; | |
5149 | } | |
5150 | #endif /* CONFIG_NUMA */ | |
5151 | ||
af7c4b0e JW |
5152 | static inline void mem_cgroup_lru_names_not_uptodate(void) |
5153 | { | |
5154 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); | |
5155 | } | |
5156 | ||
182446d0 | 5157 | static int memcg_stat_show(struct cgroup_subsys_state *css, struct cftype *cft, |
78ccf5b5 | 5158 | struct seq_file *m) |
d2ceb9b7 | 5159 | { |
182446d0 | 5160 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
af7c4b0e JW |
5161 | struct mem_cgroup *mi; |
5162 | unsigned int i; | |
406eb0c9 | 5163 | |
af7c4b0e | 5164 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 5165 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 5166 | continue; |
af7c4b0e JW |
5167 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
5168 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 5169 | } |
7b854121 | 5170 | |
af7c4b0e JW |
5171 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
5172 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
5173 | mem_cgroup_read_events(memcg, i)); | |
5174 | ||
5175 | for (i = 0; i < NR_LRU_LISTS; i++) | |
5176 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
5177 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
5178 | ||
14067bb3 | 5179 | /* Hierarchical information */ |
fee7b548 KH |
5180 | { |
5181 | unsigned long long limit, memsw_limit; | |
d79154bb | 5182 | memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit); |
78ccf5b5 | 5183 | seq_printf(m, "hierarchical_memory_limit %llu\n", limit); |
fee7b548 | 5184 | if (do_swap_account) |
78ccf5b5 JW |
5185 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
5186 | memsw_limit); | |
fee7b548 | 5187 | } |
7f016ee8 | 5188 | |
af7c4b0e JW |
5189 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
5190 | long long val = 0; | |
5191 | ||
bff6bb83 | 5192 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 5193 | continue; |
af7c4b0e JW |
5194 | for_each_mem_cgroup_tree(mi, memcg) |
5195 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
5196 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
5197 | } | |
5198 | ||
5199 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
5200 | unsigned long long val = 0; | |
5201 | ||
5202 | for_each_mem_cgroup_tree(mi, memcg) | |
5203 | val += mem_cgroup_read_events(mi, i); | |
5204 | seq_printf(m, "total_%s %llu\n", | |
5205 | mem_cgroup_events_names[i], val); | |
5206 | } | |
5207 | ||
5208 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
5209 | unsigned long long val = 0; | |
5210 | ||
5211 | for_each_mem_cgroup_tree(mi, memcg) | |
5212 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
5213 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 5214 | } |
14067bb3 | 5215 | |
7f016ee8 | 5216 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
5217 | { |
5218 | int nid, zid; | |
5219 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 5220 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
5221 | unsigned long recent_rotated[2] = {0, 0}; |
5222 | unsigned long recent_scanned[2] = {0, 0}; | |
5223 | ||
5224 | for_each_online_node(nid) | |
5225 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
d79154bb | 5226 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
89abfab1 | 5227 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 5228 | |
89abfab1 HD |
5229 | recent_rotated[0] += rstat->recent_rotated[0]; |
5230 | recent_rotated[1] += rstat->recent_rotated[1]; | |
5231 | recent_scanned[0] += rstat->recent_scanned[0]; | |
5232 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 5233 | } |
78ccf5b5 JW |
5234 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
5235 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
5236 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
5237 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
5238 | } |
5239 | #endif | |
5240 | ||
d2ceb9b7 KH |
5241 | return 0; |
5242 | } | |
5243 | ||
182446d0 TH |
5244 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
5245 | struct cftype *cft) | |
a7885eb8 | 5246 | { |
182446d0 | 5247 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 5248 | |
1f4c025b | 5249 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
5250 | } |
5251 | ||
182446d0 TH |
5252 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
5253 | struct cftype *cft, u64 val) | |
a7885eb8 | 5254 | { |
182446d0 | 5255 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
63876986 | 5256 | struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css)); |
a7885eb8 | 5257 | |
63876986 | 5258 | if (val > 100 || !parent) |
a7885eb8 KM |
5259 | return -EINVAL; |
5260 | ||
0999821b | 5261 | mutex_lock(&memcg_create_mutex); |
068b38c1 | 5262 | |
a7885eb8 | 5263 | /* If under hierarchy, only empty-root can set this value */ |
b5f99b53 | 5264 | if ((parent->use_hierarchy) || memcg_has_children(memcg)) { |
0999821b | 5265 | mutex_unlock(&memcg_create_mutex); |
a7885eb8 | 5266 | return -EINVAL; |
068b38c1 | 5267 | } |
a7885eb8 | 5268 | |
a7885eb8 | 5269 | memcg->swappiness = val; |
a7885eb8 | 5270 | |
0999821b | 5271 | mutex_unlock(&memcg_create_mutex); |
068b38c1 | 5272 | |
a7885eb8 KM |
5273 | return 0; |
5274 | } | |
5275 | ||
2e72b634 KS |
5276 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
5277 | { | |
5278 | struct mem_cgroup_threshold_ary *t; | |
5279 | u64 usage; | |
5280 | int i; | |
5281 | ||
5282 | rcu_read_lock(); | |
5283 | if (!swap) | |
2c488db2 | 5284 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 5285 | else |
2c488db2 | 5286 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
5287 | |
5288 | if (!t) | |
5289 | goto unlock; | |
5290 | ||
5291 | usage = mem_cgroup_usage(memcg, swap); | |
5292 | ||
5293 | /* | |
748dad36 | 5294 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
5295 | * If it's not true, a threshold was crossed after last |
5296 | * call of __mem_cgroup_threshold(). | |
5297 | */ | |
5407a562 | 5298 | i = t->current_threshold; |
2e72b634 KS |
5299 | |
5300 | /* | |
5301 | * Iterate backward over array of thresholds starting from | |
5302 | * current_threshold and check if a threshold is crossed. | |
5303 | * If none of thresholds below usage is crossed, we read | |
5304 | * only one element of the array here. | |
5305 | */ | |
5306 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
5307 | eventfd_signal(t->entries[i].eventfd, 1); | |
5308 | ||
5309 | /* i = current_threshold + 1 */ | |
5310 | i++; | |
5311 | ||
5312 | /* | |
5313 | * Iterate forward over array of thresholds starting from | |
5314 | * current_threshold+1 and check if a threshold is crossed. | |
5315 | * If none of thresholds above usage is crossed, we read | |
5316 | * only one element of the array here. | |
5317 | */ | |
5318 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
5319 | eventfd_signal(t->entries[i].eventfd, 1); | |
5320 | ||
5321 | /* Update current_threshold */ | |
5407a562 | 5322 | t->current_threshold = i - 1; |
2e72b634 KS |
5323 | unlock: |
5324 | rcu_read_unlock(); | |
5325 | } | |
5326 | ||
5327 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
5328 | { | |
ad4ca5f4 KS |
5329 | while (memcg) { |
5330 | __mem_cgroup_threshold(memcg, false); | |
5331 | if (do_swap_account) | |
5332 | __mem_cgroup_threshold(memcg, true); | |
5333 | ||
5334 | memcg = parent_mem_cgroup(memcg); | |
5335 | } | |
2e72b634 KS |
5336 | } |
5337 | ||
5338 | static int compare_thresholds(const void *a, const void *b) | |
5339 | { | |
5340 | const struct mem_cgroup_threshold *_a = a; | |
5341 | const struct mem_cgroup_threshold *_b = b; | |
5342 | ||
2bff24a3 GT |
5343 | if (_a->threshold > _b->threshold) |
5344 | return 1; | |
5345 | ||
5346 | if (_a->threshold < _b->threshold) | |
5347 | return -1; | |
5348 | ||
5349 | return 0; | |
2e72b634 KS |
5350 | } |
5351 | ||
c0ff4b85 | 5352 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
5353 | { |
5354 | struct mem_cgroup_eventfd_list *ev; | |
5355 | ||
c0ff4b85 | 5356 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 KH |
5357 | eventfd_signal(ev->eventfd, 1); |
5358 | return 0; | |
5359 | } | |
5360 | ||
c0ff4b85 | 5361 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 5362 | { |
7d74b06f KH |
5363 | struct mem_cgroup *iter; |
5364 | ||
c0ff4b85 | 5365 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 5366 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
5367 | } |
5368 | ||
81eeaf04 | 5369 | static int mem_cgroup_usage_register_event(struct cgroup_subsys_state *css, |
9490ff27 | 5370 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) |
2e72b634 | 5371 | { |
81eeaf04 | 5372 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
2c488db2 KS |
5373 | struct mem_cgroup_thresholds *thresholds; |
5374 | struct mem_cgroup_threshold_ary *new; | |
86ae53e1 | 5375 | enum res_type type = MEMFILE_TYPE(cft->private); |
2e72b634 | 5376 | u64 threshold, usage; |
2c488db2 | 5377 | int i, size, ret; |
2e72b634 KS |
5378 | |
5379 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
5380 | if (ret) | |
5381 | return ret; | |
5382 | ||
5383 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 5384 | |
2e72b634 | 5385 | if (type == _MEM) |
2c488db2 | 5386 | thresholds = &memcg->thresholds; |
2e72b634 | 5387 | else if (type == _MEMSWAP) |
2c488db2 | 5388 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
5389 | else |
5390 | BUG(); | |
5391 | ||
5392 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
5393 | ||
5394 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 5395 | if (thresholds->primary) |
2e72b634 KS |
5396 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
5397 | ||
2c488db2 | 5398 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
5399 | |
5400 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 5401 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 5402 | GFP_KERNEL); |
2c488db2 | 5403 | if (!new) { |
2e72b634 KS |
5404 | ret = -ENOMEM; |
5405 | goto unlock; | |
5406 | } | |
2c488db2 | 5407 | new->size = size; |
2e72b634 KS |
5408 | |
5409 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
5410 | if (thresholds->primary) { |
5411 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 5412 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
5413 | } |
5414 | ||
2e72b634 | 5415 | /* Add new threshold */ |
2c488db2 KS |
5416 | new->entries[size - 1].eventfd = eventfd; |
5417 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
5418 | |
5419 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 5420 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
5421 | compare_thresholds, NULL); |
5422 | ||
5423 | /* Find current threshold */ | |
2c488db2 | 5424 | new->current_threshold = -1; |
2e72b634 | 5425 | for (i = 0; i < size; i++) { |
748dad36 | 5426 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 5427 | /* |
2c488db2 KS |
5428 | * new->current_threshold will not be used until |
5429 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
5430 | * it here. |
5431 | */ | |
2c488db2 | 5432 | ++new->current_threshold; |
748dad36 SZ |
5433 | } else |
5434 | break; | |
2e72b634 KS |
5435 | } |
5436 | ||
2c488db2 KS |
5437 | /* Free old spare buffer and save old primary buffer as spare */ |
5438 | kfree(thresholds->spare); | |
5439 | thresholds->spare = thresholds->primary; | |
5440 | ||
5441 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 5442 | |
907860ed | 5443 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
5444 | synchronize_rcu(); |
5445 | ||
2e72b634 KS |
5446 | unlock: |
5447 | mutex_unlock(&memcg->thresholds_lock); | |
5448 | ||
5449 | return ret; | |
5450 | } | |
5451 | ||
81eeaf04 | 5452 | static void mem_cgroup_usage_unregister_event(struct cgroup_subsys_state *css, |
9490ff27 | 5453 | struct cftype *cft, struct eventfd_ctx *eventfd) |
2e72b634 | 5454 | { |
81eeaf04 | 5455 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
2c488db2 KS |
5456 | struct mem_cgroup_thresholds *thresholds; |
5457 | struct mem_cgroup_threshold_ary *new; | |
86ae53e1 | 5458 | enum res_type type = MEMFILE_TYPE(cft->private); |
2e72b634 | 5459 | u64 usage; |
2c488db2 | 5460 | int i, j, size; |
2e72b634 KS |
5461 | |
5462 | mutex_lock(&memcg->thresholds_lock); | |
5463 | if (type == _MEM) | |
2c488db2 | 5464 | thresholds = &memcg->thresholds; |
2e72b634 | 5465 | else if (type == _MEMSWAP) |
2c488db2 | 5466 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
5467 | else |
5468 | BUG(); | |
5469 | ||
371528ca AV |
5470 | if (!thresholds->primary) |
5471 | goto unlock; | |
5472 | ||
2e72b634 KS |
5473 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); |
5474 | ||
5475 | /* Check if a threshold crossed before removing */ | |
5476 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
5477 | ||
5478 | /* Calculate new number of threshold */ | |
2c488db2 KS |
5479 | size = 0; |
5480 | for (i = 0; i < thresholds->primary->size; i++) { | |
5481 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
5482 | size++; |
5483 | } | |
5484 | ||
2c488db2 | 5485 | new = thresholds->spare; |
907860ed | 5486 | |
2e72b634 KS |
5487 | /* Set thresholds array to NULL if we don't have thresholds */ |
5488 | if (!size) { | |
2c488db2 KS |
5489 | kfree(new); |
5490 | new = NULL; | |
907860ed | 5491 | goto swap_buffers; |
2e72b634 KS |
5492 | } |
5493 | ||
2c488db2 | 5494 | new->size = size; |
2e72b634 KS |
5495 | |
5496 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
5497 | new->current_threshold = -1; |
5498 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
5499 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
5500 | continue; |
5501 | ||
2c488db2 | 5502 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 5503 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 5504 | /* |
2c488db2 | 5505 | * new->current_threshold will not be used |
2e72b634 KS |
5506 | * until rcu_assign_pointer(), so it's safe to increment |
5507 | * it here. | |
5508 | */ | |
2c488db2 | 5509 | ++new->current_threshold; |
2e72b634 KS |
5510 | } |
5511 | j++; | |
5512 | } | |
5513 | ||
907860ed | 5514 | swap_buffers: |
2c488db2 KS |
5515 | /* Swap primary and spare array */ |
5516 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
5517 | /* If all events are unregistered, free the spare array */ |
5518 | if (!new) { | |
5519 | kfree(thresholds->spare); | |
5520 | thresholds->spare = NULL; | |
5521 | } | |
5522 | ||
2c488db2 | 5523 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 5524 | |
907860ed | 5525 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 5526 | synchronize_rcu(); |
371528ca | 5527 | unlock: |
2e72b634 | 5528 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 5529 | } |
c1e862c1 | 5530 | |
81eeaf04 | 5531 | static int mem_cgroup_oom_register_event(struct cgroup_subsys_state *css, |
9490ff27 KH |
5532 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) |
5533 | { | |
81eeaf04 | 5534 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
9490ff27 | 5535 | struct mem_cgroup_eventfd_list *event; |
86ae53e1 | 5536 | enum res_type type = MEMFILE_TYPE(cft->private); |
9490ff27 KH |
5537 | |
5538 | BUG_ON(type != _OOM_TYPE); | |
5539 | event = kmalloc(sizeof(*event), GFP_KERNEL); | |
5540 | if (!event) | |
5541 | return -ENOMEM; | |
5542 | ||
1af8efe9 | 5543 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
5544 | |
5545 | event->eventfd = eventfd; | |
5546 | list_add(&event->list, &memcg->oom_notify); | |
5547 | ||
5548 | /* already in OOM ? */ | |
79dfdacc | 5549 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 5550 | eventfd_signal(eventfd, 1); |
1af8efe9 | 5551 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
5552 | |
5553 | return 0; | |
5554 | } | |
5555 | ||
81eeaf04 | 5556 | static void mem_cgroup_oom_unregister_event(struct cgroup_subsys_state *css, |
9490ff27 KH |
5557 | struct cftype *cft, struct eventfd_ctx *eventfd) |
5558 | { | |
81eeaf04 | 5559 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
9490ff27 | 5560 | struct mem_cgroup_eventfd_list *ev, *tmp; |
86ae53e1 | 5561 | enum res_type type = MEMFILE_TYPE(cft->private); |
9490ff27 KH |
5562 | |
5563 | BUG_ON(type != _OOM_TYPE); | |
5564 | ||
1af8efe9 | 5565 | spin_lock(&memcg_oom_lock); |
9490ff27 | 5566 | |
c0ff4b85 | 5567 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
5568 | if (ev->eventfd == eventfd) { |
5569 | list_del(&ev->list); | |
5570 | kfree(ev); | |
5571 | } | |
5572 | } | |
5573 | ||
1af8efe9 | 5574 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
5575 | } |
5576 | ||
182446d0 | 5577 | static int mem_cgroup_oom_control_read(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
5578 | struct cftype *cft, struct cgroup_map_cb *cb) |
5579 | { | |
182446d0 | 5580 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 | 5581 | |
c0ff4b85 | 5582 | cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable); |
3c11ecf4 | 5583 | |
c0ff4b85 | 5584 | if (atomic_read(&memcg->under_oom)) |
3c11ecf4 KH |
5585 | cb->fill(cb, "under_oom", 1); |
5586 | else | |
5587 | cb->fill(cb, "under_oom", 0); | |
5588 | return 0; | |
5589 | } | |
5590 | ||
182446d0 | 5591 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
5592 | struct cftype *cft, u64 val) |
5593 | { | |
182446d0 | 5594 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
63876986 | 5595 | struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css)); |
3c11ecf4 KH |
5596 | |
5597 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
63876986 | 5598 | if (!parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
5599 | return -EINVAL; |
5600 | ||
0999821b | 5601 | mutex_lock(&memcg_create_mutex); |
3c11ecf4 | 5602 | /* oom-kill-disable is a flag for subhierarchy. */ |
b5f99b53 | 5603 | if ((parent->use_hierarchy) || memcg_has_children(memcg)) { |
0999821b | 5604 | mutex_unlock(&memcg_create_mutex); |
3c11ecf4 KH |
5605 | return -EINVAL; |
5606 | } | |
c0ff4b85 | 5607 | memcg->oom_kill_disable = val; |
4d845ebf | 5608 | if (!val) |
c0ff4b85 | 5609 | memcg_oom_recover(memcg); |
0999821b | 5610 | mutex_unlock(&memcg_create_mutex); |
3c11ecf4 KH |
5611 | return 0; |
5612 | } | |
5613 | ||
c255a458 | 5614 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 5615 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 5616 | { |
55007d84 GC |
5617 | int ret; |
5618 | ||
2633d7a0 | 5619 | memcg->kmemcg_id = -1; |
55007d84 GC |
5620 | ret = memcg_propagate_kmem(memcg); |
5621 | if (ret) | |
5622 | return ret; | |
2633d7a0 | 5623 | |
1d62e436 | 5624 | return mem_cgroup_sockets_init(memcg, ss); |
573b400d | 5625 | } |
e5671dfa | 5626 | |
10d5ebf4 | 5627 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 5628 | { |
1d62e436 | 5629 | mem_cgroup_sockets_destroy(memcg); |
10d5ebf4 LZ |
5630 | } |
5631 | ||
5632 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) | |
5633 | { | |
5634 | if (!memcg_kmem_is_active(memcg)) | |
5635 | return; | |
5636 | ||
5637 | /* | |
5638 | * kmem charges can outlive the cgroup. In the case of slab | |
5639 | * pages, for instance, a page contain objects from various | |
5640 | * processes. As we prevent from taking a reference for every | |
5641 | * such allocation we have to be careful when doing uncharge | |
5642 | * (see memcg_uncharge_kmem) and here during offlining. | |
5643 | * | |
5644 | * The idea is that that only the _last_ uncharge which sees | |
5645 | * the dead memcg will drop the last reference. An additional | |
5646 | * reference is taken here before the group is marked dead | |
5647 | * which is then paired with css_put during uncharge resp. here. | |
5648 | * | |
5649 | * Although this might sound strange as this path is called from | |
5650 | * css_offline() when the referencemight have dropped down to 0 | |
5651 | * and shouldn't be incremented anymore (css_tryget would fail) | |
5652 | * we do not have other options because of the kmem allocations | |
5653 | * lifetime. | |
5654 | */ | |
5655 | css_get(&memcg->css); | |
7de37682 GC |
5656 | |
5657 | memcg_kmem_mark_dead(memcg); | |
5658 | ||
5659 | if (res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0) | |
5660 | return; | |
5661 | ||
7de37682 | 5662 | if (memcg_kmem_test_and_clear_dead(memcg)) |
10d5ebf4 | 5663 | css_put(&memcg->css); |
d1a4c0b3 | 5664 | } |
e5671dfa | 5665 | #else |
cbe128e3 | 5666 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
5667 | { |
5668 | return 0; | |
5669 | } | |
d1a4c0b3 | 5670 | |
10d5ebf4 LZ |
5671 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
5672 | { | |
5673 | } | |
5674 | ||
5675 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) | |
d1a4c0b3 GC |
5676 | { |
5677 | } | |
e5671dfa GC |
5678 | #endif |
5679 | ||
8cdea7c0 BS |
5680 | static struct cftype mem_cgroup_files[] = { |
5681 | { | |
0eea1030 | 5682 | .name = "usage_in_bytes", |
8c7c6e34 | 5683 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
af36f906 | 5684 | .read = mem_cgroup_read, |
9490ff27 KH |
5685 | .register_event = mem_cgroup_usage_register_event, |
5686 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8cdea7c0 | 5687 | }, |
c84872e1 PE |
5688 | { |
5689 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 5690 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 5691 | .trigger = mem_cgroup_reset, |
af36f906 | 5692 | .read = mem_cgroup_read, |
c84872e1 | 5693 | }, |
8cdea7c0 | 5694 | { |
0eea1030 | 5695 | .name = "limit_in_bytes", |
8c7c6e34 | 5696 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 5697 | .write_string = mem_cgroup_write, |
af36f906 | 5698 | .read = mem_cgroup_read, |
8cdea7c0 | 5699 | }, |
296c81d8 BS |
5700 | { |
5701 | .name = "soft_limit_in_bytes", | |
5702 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
5703 | .write_string = mem_cgroup_write, | |
af36f906 | 5704 | .read = mem_cgroup_read, |
296c81d8 | 5705 | }, |
8cdea7c0 BS |
5706 | { |
5707 | .name = "failcnt", | |
8c7c6e34 | 5708 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 5709 | .trigger = mem_cgroup_reset, |
af36f906 | 5710 | .read = mem_cgroup_read, |
8cdea7c0 | 5711 | }, |
d2ceb9b7 KH |
5712 | { |
5713 | .name = "stat", | |
ab215884 | 5714 | .read_seq_string = memcg_stat_show, |
d2ceb9b7 | 5715 | }, |
c1e862c1 KH |
5716 | { |
5717 | .name = "force_empty", | |
5718 | .trigger = mem_cgroup_force_empty_write, | |
5719 | }, | |
18f59ea7 BS |
5720 | { |
5721 | .name = "use_hierarchy", | |
f00baae7 | 5722 | .flags = CFTYPE_INSANE, |
18f59ea7 BS |
5723 | .write_u64 = mem_cgroup_hierarchy_write, |
5724 | .read_u64 = mem_cgroup_hierarchy_read, | |
5725 | }, | |
a7885eb8 KM |
5726 | { |
5727 | .name = "swappiness", | |
5728 | .read_u64 = mem_cgroup_swappiness_read, | |
5729 | .write_u64 = mem_cgroup_swappiness_write, | |
5730 | }, | |
7dc74be0 DN |
5731 | { |
5732 | .name = "move_charge_at_immigrate", | |
5733 | .read_u64 = mem_cgroup_move_charge_read, | |
5734 | .write_u64 = mem_cgroup_move_charge_write, | |
5735 | }, | |
9490ff27 KH |
5736 | { |
5737 | .name = "oom_control", | |
3c11ecf4 KH |
5738 | .read_map = mem_cgroup_oom_control_read, |
5739 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
5740 | .register_event = mem_cgroup_oom_register_event, |
5741 | .unregister_event = mem_cgroup_oom_unregister_event, | |
5742 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), | |
5743 | }, | |
70ddf637 AV |
5744 | { |
5745 | .name = "pressure_level", | |
5746 | .register_event = vmpressure_register_event, | |
5747 | .unregister_event = vmpressure_unregister_event, | |
5748 | }, | |
406eb0c9 YH |
5749 | #ifdef CONFIG_NUMA |
5750 | { | |
5751 | .name = "numa_stat", | |
ab215884 | 5752 | .read_seq_string = memcg_numa_stat_show, |
406eb0c9 YH |
5753 | }, |
5754 | #endif | |
510fc4e1 GC |
5755 | #ifdef CONFIG_MEMCG_KMEM |
5756 | { | |
5757 | .name = "kmem.limit_in_bytes", | |
5758 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
5759 | .write_string = mem_cgroup_write, | |
5760 | .read = mem_cgroup_read, | |
5761 | }, | |
5762 | { | |
5763 | .name = "kmem.usage_in_bytes", | |
5764 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
5765 | .read = mem_cgroup_read, | |
5766 | }, | |
5767 | { | |
5768 | .name = "kmem.failcnt", | |
5769 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
5770 | .trigger = mem_cgroup_reset, | |
5771 | .read = mem_cgroup_read, | |
5772 | }, | |
5773 | { | |
5774 | .name = "kmem.max_usage_in_bytes", | |
5775 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
5776 | .trigger = mem_cgroup_reset, | |
5777 | .read = mem_cgroup_read, | |
5778 | }, | |
749c5415 GC |
5779 | #ifdef CONFIG_SLABINFO |
5780 | { | |
5781 | .name = "kmem.slabinfo", | |
5782 | .read_seq_string = mem_cgroup_slabinfo_read, | |
5783 | }, | |
5784 | #endif | |
8c7c6e34 | 5785 | #endif |
6bc10349 | 5786 | { }, /* terminate */ |
af36f906 | 5787 | }; |
8c7c6e34 | 5788 | |
2d11085e MH |
5789 | #ifdef CONFIG_MEMCG_SWAP |
5790 | static struct cftype memsw_cgroup_files[] = { | |
5791 | { | |
5792 | .name = "memsw.usage_in_bytes", | |
5793 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
5794 | .read = mem_cgroup_read, | |
5795 | .register_event = mem_cgroup_usage_register_event, | |
5796 | .unregister_event = mem_cgroup_usage_unregister_event, | |
5797 | }, | |
5798 | { | |
5799 | .name = "memsw.max_usage_in_bytes", | |
5800 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
5801 | .trigger = mem_cgroup_reset, | |
5802 | .read = mem_cgroup_read, | |
5803 | }, | |
5804 | { | |
5805 | .name = "memsw.limit_in_bytes", | |
5806 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
5807 | .write_string = mem_cgroup_write, | |
5808 | .read = mem_cgroup_read, | |
5809 | }, | |
5810 | { | |
5811 | .name = "memsw.failcnt", | |
5812 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
5813 | .trigger = mem_cgroup_reset, | |
5814 | .read = mem_cgroup_read, | |
5815 | }, | |
5816 | { }, /* terminate */ | |
5817 | }; | |
5818 | #endif | |
c0ff4b85 | 5819 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
5820 | { |
5821 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 5822 | struct mem_cgroup_per_zone *mz; |
41e3355d | 5823 | int zone, tmp = node; |
1ecaab2b KH |
5824 | /* |
5825 | * This routine is called against possible nodes. | |
5826 | * But it's BUG to call kmalloc() against offline node. | |
5827 | * | |
5828 | * TODO: this routine can waste much memory for nodes which will | |
5829 | * never be onlined. It's better to use memory hotplug callback | |
5830 | * function. | |
5831 | */ | |
41e3355d KH |
5832 | if (!node_state(node, N_NORMAL_MEMORY)) |
5833 | tmp = -1; | |
17295c88 | 5834 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
5835 | if (!pn) |
5836 | return 1; | |
1ecaab2b | 5837 | |
1ecaab2b KH |
5838 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
5839 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 5840 | lruvec_init(&mz->lruvec); |
d79154bb | 5841 | mz->memcg = memcg; |
1ecaab2b | 5842 | } |
54f72fe0 | 5843 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
5844 | return 0; |
5845 | } | |
5846 | ||
c0ff4b85 | 5847 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 5848 | { |
54f72fe0 | 5849 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
5850 | } |
5851 | ||
33327948 KH |
5852 | static struct mem_cgroup *mem_cgroup_alloc(void) |
5853 | { | |
d79154bb | 5854 | struct mem_cgroup *memcg; |
45cf7ebd | 5855 | size_t size = memcg_size(); |
33327948 | 5856 | |
45cf7ebd | 5857 | /* Can be very big if nr_node_ids is very big */ |
c8dad2bb | 5858 | if (size < PAGE_SIZE) |
d79154bb | 5859 | memcg = kzalloc(size, GFP_KERNEL); |
33327948 | 5860 | else |
d79154bb | 5861 | memcg = vzalloc(size); |
33327948 | 5862 | |
d79154bb | 5863 | if (!memcg) |
e7bbcdf3 DC |
5864 | return NULL; |
5865 | ||
d79154bb HD |
5866 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
5867 | if (!memcg->stat) | |
d2e61b8d | 5868 | goto out_free; |
d79154bb HD |
5869 | spin_lock_init(&memcg->pcp_counter_lock); |
5870 | return memcg; | |
d2e61b8d DC |
5871 | |
5872 | out_free: | |
5873 | if (size < PAGE_SIZE) | |
d79154bb | 5874 | kfree(memcg); |
d2e61b8d | 5875 | else |
d79154bb | 5876 | vfree(memcg); |
d2e61b8d | 5877 | return NULL; |
33327948 KH |
5878 | } |
5879 | ||
59927fb9 | 5880 | /* |
c8b2a36f GC |
5881 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
5882 | * (scanning all at force_empty is too costly...) | |
5883 | * | |
5884 | * Instead of clearing all references at force_empty, we remember | |
5885 | * the number of reference from swap_cgroup and free mem_cgroup when | |
5886 | * it goes down to 0. | |
5887 | * | |
5888 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 5889 | */ |
c8b2a36f GC |
5890 | |
5891 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 5892 | { |
c8b2a36f | 5893 | int node; |
45cf7ebd | 5894 | size_t size = memcg_size(); |
59927fb9 | 5895 | |
c8b2a36f GC |
5896 | free_css_id(&mem_cgroup_subsys, &memcg->css); |
5897 | ||
5898 | for_each_node(node) | |
5899 | free_mem_cgroup_per_zone_info(memcg, node); | |
5900 | ||
5901 | free_percpu(memcg->stat); | |
5902 | ||
3f134619 GC |
5903 | /* |
5904 | * We need to make sure that (at least for now), the jump label | |
5905 | * destruction code runs outside of the cgroup lock. This is because | |
5906 | * get_online_cpus(), which is called from the static_branch update, | |
5907 | * can't be called inside the cgroup_lock. cpusets are the ones | |
5908 | * enforcing this dependency, so if they ever change, we might as well. | |
5909 | * | |
5910 | * schedule_work() will guarantee this happens. Be careful if you need | |
5911 | * to move this code around, and make sure it is outside | |
5912 | * the cgroup_lock. | |
5913 | */ | |
a8964b9b | 5914 | disarm_static_keys(memcg); |
3afe36b1 GC |
5915 | if (size < PAGE_SIZE) |
5916 | kfree(memcg); | |
5917 | else | |
5918 | vfree(memcg); | |
59927fb9 | 5919 | } |
3afe36b1 | 5920 | |
7bcc1bb1 DN |
5921 | /* |
5922 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
5923 | */ | |
e1aab161 | 5924 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 5925 | { |
c0ff4b85 | 5926 | if (!memcg->res.parent) |
7bcc1bb1 | 5927 | return NULL; |
c0ff4b85 | 5928 | return mem_cgroup_from_res_counter(memcg->res.parent, res); |
7bcc1bb1 | 5929 | } |
e1aab161 | 5930 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 5931 | |
0eb253e2 | 5932 | static struct cgroup_subsys_state * __ref |
eb95419b | 5933 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 5934 | { |
d142e3e6 | 5935 | struct mem_cgroup *memcg; |
04046e1a | 5936 | long error = -ENOMEM; |
6d12e2d8 | 5937 | int node; |
8cdea7c0 | 5938 | |
c0ff4b85 R |
5939 | memcg = mem_cgroup_alloc(); |
5940 | if (!memcg) | |
04046e1a | 5941 | return ERR_PTR(error); |
78fb7466 | 5942 | |
3ed28fa1 | 5943 | for_each_node(node) |
c0ff4b85 | 5944 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 5945 | goto free_out; |
f64c3f54 | 5946 | |
c077719b | 5947 | /* root ? */ |
eb95419b | 5948 | if (parent_css == NULL) { |
a41c58a6 | 5949 | root_mem_cgroup = memcg; |
d142e3e6 GC |
5950 | res_counter_init(&memcg->res, NULL); |
5951 | res_counter_init(&memcg->memsw, NULL); | |
5952 | res_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 5953 | } |
28dbc4b6 | 5954 | |
d142e3e6 GC |
5955 | memcg->last_scanned_node = MAX_NUMNODES; |
5956 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
5957 | memcg->move_charge_at_immigrate = 0; |
5958 | mutex_init(&memcg->thresholds_lock); | |
5959 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 5960 | vmpressure_init(&memcg->vmpressure); |
7d910c05 | 5961 | spin_lock_init(&memcg->soft_lock); |
d142e3e6 GC |
5962 | |
5963 | return &memcg->css; | |
5964 | ||
5965 | free_out: | |
5966 | __mem_cgroup_free(memcg); | |
5967 | return ERR_PTR(error); | |
5968 | } | |
5969 | ||
5970 | static int | |
eb95419b | 5971 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 5972 | { |
eb95419b TH |
5973 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5974 | struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(css)); | |
d142e3e6 GC |
5975 | int error = 0; |
5976 | ||
63876986 | 5977 | if (!parent) |
d142e3e6 GC |
5978 | return 0; |
5979 | ||
0999821b | 5980 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
5981 | |
5982 | memcg->use_hierarchy = parent->use_hierarchy; | |
5983 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
5984 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
5985 | ||
5986 | if (parent->use_hierarchy) { | |
c0ff4b85 R |
5987 | res_counter_init(&memcg->res, &parent->res); |
5988 | res_counter_init(&memcg->memsw, &parent->memsw); | |
510fc4e1 | 5989 | res_counter_init(&memcg->kmem, &parent->kmem); |
55007d84 | 5990 | |
7bcc1bb1 | 5991 | /* |
8d76a979 LZ |
5992 | * No need to take a reference to the parent because cgroup |
5993 | * core guarantees its existence. | |
7bcc1bb1 | 5994 | */ |
18f59ea7 | 5995 | } else { |
c0ff4b85 R |
5996 | res_counter_init(&memcg->res, NULL); |
5997 | res_counter_init(&memcg->memsw, NULL); | |
510fc4e1 | 5998 | res_counter_init(&memcg->kmem, NULL); |
8c7f6edb TH |
5999 | /* |
6000 | * Deeper hierachy with use_hierarchy == false doesn't make | |
6001 | * much sense so let cgroup subsystem know about this | |
6002 | * unfortunate state in our controller. | |
6003 | */ | |
d142e3e6 | 6004 | if (parent != root_mem_cgroup) |
8c7f6edb | 6005 | mem_cgroup_subsys.broken_hierarchy = true; |
18f59ea7 | 6006 | } |
cbe128e3 GC |
6007 | |
6008 | error = memcg_init_kmem(memcg, &mem_cgroup_subsys); | |
0999821b | 6009 | mutex_unlock(&memcg_create_mutex); |
d142e3e6 | 6010 | return error; |
8cdea7c0 BS |
6011 | } |
6012 | ||
5f578161 MH |
6013 | /* |
6014 | * Announce all parents that a group from their hierarchy is gone. | |
6015 | */ | |
6016 | static void mem_cgroup_invalidate_reclaim_iterators(struct mem_cgroup *memcg) | |
6017 | { | |
6018 | struct mem_cgroup *parent = memcg; | |
6019 | ||
6020 | while ((parent = parent_mem_cgroup(parent))) | |
519ebea3 | 6021 | mem_cgroup_iter_invalidate(parent); |
5f578161 MH |
6022 | |
6023 | /* | |
6024 | * if the root memcg is not hierarchical we have to check it | |
6025 | * explicitely. | |
6026 | */ | |
6027 | if (!root_mem_cgroup->use_hierarchy) | |
519ebea3 | 6028 | mem_cgroup_iter_invalidate(root_mem_cgroup); |
5f578161 MH |
6029 | } |
6030 | ||
eb95419b | 6031 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 6032 | { |
eb95419b | 6033 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
ec64f515 | 6034 | |
10d5ebf4 LZ |
6035 | kmem_cgroup_css_offline(memcg); |
6036 | ||
5f578161 | 6037 | mem_cgroup_invalidate_reclaim_iterators(memcg); |
ab5196c2 | 6038 | mem_cgroup_reparent_charges(memcg); |
7d910c05 MH |
6039 | if (memcg->soft_contributed) { |
6040 | while ((memcg = parent_mem_cgroup(memcg))) | |
6041 | atomic_dec(&memcg->children_in_excess); | |
6042 | } | |
1f458cbf | 6043 | mem_cgroup_destroy_all_caches(memcg); |
33cb876e | 6044 | vmpressure_cleanup(&memcg->vmpressure); |
df878fb0 KH |
6045 | } |
6046 | ||
eb95419b | 6047 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 6048 | { |
eb95419b | 6049 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 6050 | |
10d5ebf4 | 6051 | memcg_destroy_kmem(memcg); |
465939a1 | 6052 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
6053 | } |
6054 | ||
02491447 | 6055 | #ifdef CONFIG_MMU |
7dc74be0 | 6056 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
6057 | #define PRECHARGE_COUNT_AT_ONCE 256 |
6058 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 6059 | { |
854ffa8d DN |
6060 | int ret = 0; |
6061 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
c0ff4b85 | 6062 | struct mem_cgroup *memcg = mc.to; |
4ffef5fe | 6063 | |
c0ff4b85 | 6064 | if (mem_cgroup_is_root(memcg)) { |
854ffa8d DN |
6065 | mc.precharge += count; |
6066 | /* we don't need css_get for root */ | |
6067 | return ret; | |
6068 | } | |
6069 | /* try to charge at once */ | |
6070 | if (count > 1) { | |
6071 | struct res_counter *dummy; | |
6072 | /* | |
c0ff4b85 | 6073 | * "memcg" cannot be under rmdir() because we've already checked |
854ffa8d DN |
6074 | * by cgroup_lock_live_cgroup() that it is not removed and we |
6075 | * are still under the same cgroup_mutex. So we can postpone | |
6076 | * css_get(). | |
6077 | */ | |
c0ff4b85 | 6078 | if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy)) |
854ffa8d | 6079 | goto one_by_one; |
c0ff4b85 | 6080 | if (do_swap_account && res_counter_charge(&memcg->memsw, |
854ffa8d | 6081 | PAGE_SIZE * count, &dummy)) { |
c0ff4b85 | 6082 | res_counter_uncharge(&memcg->res, PAGE_SIZE * count); |
854ffa8d DN |
6083 | goto one_by_one; |
6084 | } | |
6085 | mc.precharge += count; | |
854ffa8d DN |
6086 | return ret; |
6087 | } | |
6088 | one_by_one: | |
6089 | /* fall back to one by one charge */ | |
6090 | while (count--) { | |
6091 | if (signal_pending(current)) { | |
6092 | ret = -EINTR; | |
6093 | break; | |
6094 | } | |
6095 | if (!batch_count--) { | |
6096 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
6097 | cond_resched(); | |
6098 | } | |
c0ff4b85 R |
6099 | ret = __mem_cgroup_try_charge(NULL, |
6100 | GFP_KERNEL, 1, &memcg, false); | |
38c5d72f | 6101 | if (ret) |
854ffa8d | 6102 | /* mem_cgroup_clear_mc() will do uncharge later */ |
38c5d72f | 6103 | return ret; |
854ffa8d DN |
6104 | mc.precharge++; |
6105 | } | |
4ffef5fe DN |
6106 | return ret; |
6107 | } | |
6108 | ||
6109 | /** | |
8d32ff84 | 6110 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
6111 | * @vma: the vma the pte to be checked belongs |
6112 | * @addr: the address corresponding to the pte to be checked | |
6113 | * @ptent: the pte to be checked | |
02491447 | 6114 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
6115 | * |
6116 | * Returns | |
6117 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
6118 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
6119 | * move charge. if @target is not NULL, the page is stored in target->page | |
6120 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
6121 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
6122 | * target for charge migration. if @target is not NULL, the entry is stored | |
6123 | * in target->ent. | |
4ffef5fe DN |
6124 | * |
6125 | * Called with pte lock held. | |
6126 | */ | |
4ffef5fe DN |
6127 | union mc_target { |
6128 | struct page *page; | |
02491447 | 6129 | swp_entry_t ent; |
4ffef5fe DN |
6130 | }; |
6131 | ||
4ffef5fe | 6132 | enum mc_target_type { |
8d32ff84 | 6133 | MC_TARGET_NONE = 0, |
4ffef5fe | 6134 | MC_TARGET_PAGE, |
02491447 | 6135 | MC_TARGET_SWAP, |
4ffef5fe DN |
6136 | }; |
6137 | ||
90254a65 DN |
6138 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
6139 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 6140 | { |
90254a65 | 6141 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 6142 | |
90254a65 DN |
6143 | if (!page || !page_mapped(page)) |
6144 | return NULL; | |
6145 | if (PageAnon(page)) { | |
6146 | /* we don't move shared anon */ | |
4b91355e | 6147 | if (!move_anon()) |
90254a65 | 6148 | return NULL; |
87946a72 DN |
6149 | } else if (!move_file()) |
6150 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
6151 | return NULL; |
6152 | if (!get_page_unless_zero(page)) | |
6153 | return NULL; | |
6154 | ||
6155 | return page; | |
6156 | } | |
6157 | ||
4b91355e | 6158 | #ifdef CONFIG_SWAP |
90254a65 DN |
6159 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
6160 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6161 | { | |
90254a65 DN |
6162 | struct page *page = NULL; |
6163 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
6164 | ||
6165 | if (!move_anon() || non_swap_entry(ent)) | |
6166 | return NULL; | |
4b91355e KH |
6167 | /* |
6168 | * Because lookup_swap_cache() updates some statistics counter, | |
6169 | * we call find_get_page() with swapper_space directly. | |
6170 | */ | |
33806f06 | 6171 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
6172 | if (do_swap_account) |
6173 | entry->val = ent.val; | |
6174 | ||
6175 | return page; | |
6176 | } | |
4b91355e KH |
6177 | #else |
6178 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
6179 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6180 | { | |
6181 | return NULL; | |
6182 | } | |
6183 | #endif | |
90254a65 | 6184 | |
87946a72 DN |
6185 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
6186 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6187 | { | |
6188 | struct page *page = NULL; | |
87946a72 DN |
6189 | struct address_space *mapping; |
6190 | pgoff_t pgoff; | |
6191 | ||
6192 | if (!vma->vm_file) /* anonymous vma */ | |
6193 | return NULL; | |
6194 | if (!move_file()) | |
6195 | return NULL; | |
6196 | ||
87946a72 DN |
6197 | mapping = vma->vm_file->f_mapping; |
6198 | if (pte_none(ptent)) | |
6199 | pgoff = linear_page_index(vma, addr); | |
6200 | else /* pte_file(ptent) is true */ | |
6201 | pgoff = pte_to_pgoff(ptent); | |
6202 | ||
6203 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
6204 | page = find_get_page(mapping, pgoff); |
6205 | ||
6206 | #ifdef CONFIG_SWAP | |
6207 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
6208 | if (radix_tree_exceptional_entry(page)) { | |
6209 | swp_entry_t swap = radix_to_swp_entry(page); | |
87946a72 | 6210 | if (do_swap_account) |
aa3b1895 | 6211 | *entry = swap; |
33806f06 | 6212 | page = find_get_page(swap_address_space(swap), swap.val); |
87946a72 | 6213 | } |
aa3b1895 | 6214 | #endif |
87946a72 DN |
6215 | return page; |
6216 | } | |
6217 | ||
8d32ff84 | 6218 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
6219 | unsigned long addr, pte_t ptent, union mc_target *target) |
6220 | { | |
6221 | struct page *page = NULL; | |
6222 | struct page_cgroup *pc; | |
8d32ff84 | 6223 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
6224 | swp_entry_t ent = { .val = 0 }; |
6225 | ||
6226 | if (pte_present(ptent)) | |
6227 | page = mc_handle_present_pte(vma, addr, ptent); | |
6228 | else if (is_swap_pte(ptent)) | |
6229 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
6230 | else if (pte_none(ptent) || pte_file(ptent)) |
6231 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
6232 | |
6233 | if (!page && !ent.val) | |
8d32ff84 | 6234 | return ret; |
02491447 DN |
6235 | if (page) { |
6236 | pc = lookup_page_cgroup(page); | |
6237 | /* | |
6238 | * Do only loose check w/o page_cgroup lock. | |
6239 | * mem_cgroup_move_account() checks the pc is valid or not under | |
6240 | * the lock. | |
6241 | */ | |
6242 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
6243 | ret = MC_TARGET_PAGE; | |
6244 | if (target) | |
6245 | target->page = page; | |
6246 | } | |
6247 | if (!ret || !target) | |
6248 | put_page(page); | |
6249 | } | |
90254a65 DN |
6250 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
6251 | if (ent.val && !ret && | |
9fb4b7cc | 6252 | css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
6253 | ret = MC_TARGET_SWAP; |
6254 | if (target) | |
6255 | target->ent = ent; | |
4ffef5fe | 6256 | } |
4ffef5fe DN |
6257 | return ret; |
6258 | } | |
6259 | ||
12724850 NH |
6260 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
6261 | /* | |
6262 | * We don't consider swapping or file mapped pages because THP does not | |
6263 | * support them for now. | |
6264 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
6265 | */ | |
6266 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6267 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6268 | { | |
6269 | struct page *page = NULL; | |
6270 | struct page_cgroup *pc; | |
6271 | enum mc_target_type ret = MC_TARGET_NONE; | |
6272 | ||
6273 | page = pmd_page(pmd); | |
6274 | VM_BUG_ON(!page || !PageHead(page)); | |
6275 | if (!move_anon()) | |
6276 | return ret; | |
6277 | pc = lookup_page_cgroup(page); | |
6278 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
6279 | ret = MC_TARGET_PAGE; | |
6280 | if (target) { | |
6281 | get_page(page); | |
6282 | target->page = page; | |
6283 | } | |
6284 | } | |
6285 | return ret; | |
6286 | } | |
6287 | #else | |
6288 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6289 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6290 | { | |
6291 | return MC_TARGET_NONE; | |
6292 | } | |
6293 | #endif | |
6294 | ||
4ffef5fe DN |
6295 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
6296 | unsigned long addr, unsigned long end, | |
6297 | struct mm_walk *walk) | |
6298 | { | |
6299 | struct vm_area_struct *vma = walk->private; | |
6300 | pte_t *pte; | |
6301 | spinlock_t *ptl; | |
6302 | ||
12724850 NH |
6303 | if (pmd_trans_huge_lock(pmd, vma) == 1) { |
6304 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) | |
6305 | mc.precharge += HPAGE_PMD_NR; | |
6306 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 6307 | return 0; |
12724850 | 6308 | } |
03319327 | 6309 | |
45f83cef AA |
6310 | if (pmd_trans_unstable(pmd)) |
6311 | return 0; | |
4ffef5fe DN |
6312 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
6313 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 6314 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
6315 | mc.precharge++; /* increment precharge temporarily */ |
6316 | pte_unmap_unlock(pte - 1, ptl); | |
6317 | cond_resched(); | |
6318 | ||
7dc74be0 DN |
6319 | return 0; |
6320 | } | |
6321 | ||
4ffef5fe DN |
6322 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
6323 | { | |
6324 | unsigned long precharge; | |
6325 | struct vm_area_struct *vma; | |
6326 | ||
dfe076b0 | 6327 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
6328 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
6329 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
6330 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
6331 | .mm = mm, | |
6332 | .private = vma, | |
6333 | }; | |
6334 | if (is_vm_hugetlb_page(vma)) | |
6335 | continue; | |
4ffef5fe DN |
6336 | walk_page_range(vma->vm_start, vma->vm_end, |
6337 | &mem_cgroup_count_precharge_walk); | |
6338 | } | |
dfe076b0 | 6339 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
6340 | |
6341 | precharge = mc.precharge; | |
6342 | mc.precharge = 0; | |
6343 | ||
6344 | return precharge; | |
6345 | } | |
6346 | ||
4ffef5fe DN |
6347 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
6348 | { | |
dfe076b0 DN |
6349 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
6350 | ||
6351 | VM_BUG_ON(mc.moving_task); | |
6352 | mc.moving_task = current; | |
6353 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
6354 | } |
6355 | ||
dfe076b0 DN |
6356 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
6357 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 6358 | { |
2bd9bb20 KH |
6359 | struct mem_cgroup *from = mc.from; |
6360 | struct mem_cgroup *to = mc.to; | |
4050377b | 6361 | int i; |
2bd9bb20 | 6362 | |
4ffef5fe | 6363 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
6364 | if (mc.precharge) { |
6365 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
6366 | mc.precharge = 0; | |
6367 | } | |
6368 | /* | |
6369 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
6370 | * we must uncharge here. | |
6371 | */ | |
6372 | if (mc.moved_charge) { | |
6373 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
6374 | mc.moved_charge = 0; | |
4ffef5fe | 6375 | } |
483c30b5 DN |
6376 | /* we must fixup refcnts and charges */ |
6377 | if (mc.moved_swap) { | |
483c30b5 DN |
6378 | /* uncharge swap account from the old cgroup */ |
6379 | if (!mem_cgroup_is_root(mc.from)) | |
6380 | res_counter_uncharge(&mc.from->memsw, | |
6381 | PAGE_SIZE * mc.moved_swap); | |
4050377b LZ |
6382 | |
6383 | for (i = 0; i < mc.moved_swap; i++) | |
6384 | css_put(&mc.from->css); | |
483c30b5 DN |
6385 | |
6386 | if (!mem_cgroup_is_root(mc.to)) { | |
6387 | /* | |
6388 | * we charged both to->res and to->memsw, so we should | |
6389 | * uncharge to->res. | |
6390 | */ | |
6391 | res_counter_uncharge(&mc.to->res, | |
6392 | PAGE_SIZE * mc.moved_swap); | |
483c30b5 | 6393 | } |
4050377b | 6394 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
6395 | mc.moved_swap = 0; |
6396 | } | |
dfe076b0 DN |
6397 | memcg_oom_recover(from); |
6398 | memcg_oom_recover(to); | |
6399 | wake_up_all(&mc.waitq); | |
6400 | } | |
6401 | ||
6402 | static void mem_cgroup_clear_mc(void) | |
6403 | { | |
6404 | struct mem_cgroup *from = mc.from; | |
6405 | ||
6406 | /* | |
6407 | * we must clear moving_task before waking up waiters at the end of | |
6408 | * task migration. | |
6409 | */ | |
6410 | mc.moving_task = NULL; | |
6411 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 6412 | spin_lock(&mc.lock); |
4ffef5fe DN |
6413 | mc.from = NULL; |
6414 | mc.to = NULL; | |
2bd9bb20 | 6415 | spin_unlock(&mc.lock); |
32047e2a | 6416 | mem_cgroup_end_move(from); |
4ffef5fe DN |
6417 | } |
6418 | ||
eb95419b | 6419 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6420 | struct cgroup_taskset *tset) |
7dc74be0 | 6421 | { |
2f7ee569 | 6422 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 6423 | int ret = 0; |
eb95419b | 6424 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
ee5e8472 | 6425 | unsigned long move_charge_at_immigrate; |
7dc74be0 | 6426 | |
ee5e8472 GC |
6427 | /* |
6428 | * We are now commited to this value whatever it is. Changes in this | |
6429 | * tunable will only affect upcoming migrations, not the current one. | |
6430 | * So we need to save it, and keep it going. | |
6431 | */ | |
6432 | move_charge_at_immigrate = memcg->move_charge_at_immigrate; | |
6433 | if (move_charge_at_immigrate) { | |
7dc74be0 DN |
6434 | struct mm_struct *mm; |
6435 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
6436 | ||
c0ff4b85 | 6437 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
6438 | |
6439 | mm = get_task_mm(p); | |
6440 | if (!mm) | |
6441 | return 0; | |
7dc74be0 | 6442 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
6443 | if (mm->owner == p) { |
6444 | VM_BUG_ON(mc.from); | |
6445 | VM_BUG_ON(mc.to); | |
6446 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 6447 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 6448 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 6449 | mem_cgroup_start_move(from); |
2bd9bb20 | 6450 | spin_lock(&mc.lock); |
4ffef5fe | 6451 | mc.from = from; |
c0ff4b85 | 6452 | mc.to = memcg; |
ee5e8472 | 6453 | mc.immigrate_flags = move_charge_at_immigrate; |
2bd9bb20 | 6454 | spin_unlock(&mc.lock); |
dfe076b0 | 6455 | /* We set mc.moving_task later */ |
4ffef5fe DN |
6456 | |
6457 | ret = mem_cgroup_precharge_mc(mm); | |
6458 | if (ret) | |
6459 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
6460 | } |
6461 | mmput(mm); | |
7dc74be0 DN |
6462 | } |
6463 | return ret; | |
6464 | } | |
6465 | ||
eb95419b | 6466 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6467 | struct cgroup_taskset *tset) |
7dc74be0 | 6468 | { |
4ffef5fe | 6469 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
6470 | } |
6471 | ||
4ffef5fe DN |
6472 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
6473 | unsigned long addr, unsigned long end, | |
6474 | struct mm_walk *walk) | |
7dc74be0 | 6475 | { |
4ffef5fe DN |
6476 | int ret = 0; |
6477 | struct vm_area_struct *vma = walk->private; | |
6478 | pte_t *pte; | |
6479 | spinlock_t *ptl; | |
12724850 NH |
6480 | enum mc_target_type target_type; |
6481 | union mc_target target; | |
6482 | struct page *page; | |
6483 | struct page_cgroup *pc; | |
4ffef5fe | 6484 | |
12724850 NH |
6485 | /* |
6486 | * We don't take compound_lock() here but no race with splitting thp | |
6487 | * happens because: | |
6488 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
6489 | * under splitting, which means there's no concurrent thp split, | |
6490 | * - if another thread runs into split_huge_page() just after we | |
6491 | * entered this if-block, the thread must wait for page table lock | |
6492 | * to be unlocked in __split_huge_page_splitting(), where the main | |
6493 | * part of thp split is not executed yet. | |
6494 | */ | |
6495 | if (pmd_trans_huge_lock(pmd, vma) == 1) { | |
62ade86a | 6496 | if (mc.precharge < HPAGE_PMD_NR) { |
12724850 NH |
6497 | spin_unlock(&vma->vm_mm->page_table_lock); |
6498 | return 0; | |
6499 | } | |
6500 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
6501 | if (target_type == MC_TARGET_PAGE) { | |
6502 | page = target.page; | |
6503 | if (!isolate_lru_page(page)) { | |
6504 | pc = lookup_page_cgroup(page); | |
6505 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, | |
2f3479b1 | 6506 | pc, mc.from, mc.to)) { |
12724850 NH |
6507 | mc.precharge -= HPAGE_PMD_NR; |
6508 | mc.moved_charge += HPAGE_PMD_NR; | |
6509 | } | |
6510 | putback_lru_page(page); | |
6511 | } | |
6512 | put_page(page); | |
6513 | } | |
6514 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 6515 | return 0; |
12724850 NH |
6516 | } |
6517 | ||
45f83cef AA |
6518 | if (pmd_trans_unstable(pmd)) |
6519 | return 0; | |
4ffef5fe DN |
6520 | retry: |
6521 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
6522 | for (; addr != end; addr += PAGE_SIZE) { | |
6523 | pte_t ptent = *(pte++); | |
02491447 | 6524 | swp_entry_t ent; |
4ffef5fe DN |
6525 | |
6526 | if (!mc.precharge) | |
6527 | break; | |
6528 | ||
8d32ff84 | 6529 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
6530 | case MC_TARGET_PAGE: |
6531 | page = target.page; | |
6532 | if (isolate_lru_page(page)) | |
6533 | goto put; | |
6534 | pc = lookup_page_cgroup(page); | |
7ec99d62 | 6535 | if (!mem_cgroup_move_account(page, 1, pc, |
2f3479b1 | 6536 | mc.from, mc.to)) { |
4ffef5fe | 6537 | mc.precharge--; |
854ffa8d DN |
6538 | /* we uncharge from mc.from later. */ |
6539 | mc.moved_charge++; | |
4ffef5fe DN |
6540 | } |
6541 | putback_lru_page(page); | |
8d32ff84 | 6542 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
6543 | put_page(page); |
6544 | break; | |
02491447 DN |
6545 | case MC_TARGET_SWAP: |
6546 | ent = target.ent; | |
e91cbb42 | 6547 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6548 | mc.precharge--; |
483c30b5 DN |
6549 | /* we fixup refcnts and charges later. */ |
6550 | mc.moved_swap++; | |
6551 | } | |
02491447 | 6552 | break; |
4ffef5fe DN |
6553 | default: |
6554 | break; | |
6555 | } | |
6556 | } | |
6557 | pte_unmap_unlock(pte - 1, ptl); | |
6558 | cond_resched(); | |
6559 | ||
6560 | if (addr != end) { | |
6561 | /* | |
6562 | * We have consumed all precharges we got in can_attach(). | |
6563 | * We try charge one by one, but don't do any additional | |
6564 | * charges to mc.to if we have failed in charge once in attach() | |
6565 | * phase. | |
6566 | */ | |
854ffa8d | 6567 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6568 | if (!ret) |
6569 | goto retry; | |
6570 | } | |
6571 | ||
6572 | return ret; | |
6573 | } | |
6574 | ||
6575 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
6576 | { | |
6577 | struct vm_area_struct *vma; | |
6578 | ||
6579 | lru_add_drain_all(); | |
dfe076b0 DN |
6580 | retry: |
6581 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
6582 | /* | |
6583 | * Someone who are holding the mmap_sem might be waiting in | |
6584 | * waitq. So we cancel all extra charges, wake up all waiters, | |
6585 | * and retry. Because we cancel precharges, we might not be able | |
6586 | * to move enough charges, but moving charge is a best-effort | |
6587 | * feature anyway, so it wouldn't be a big problem. | |
6588 | */ | |
6589 | __mem_cgroup_clear_mc(); | |
6590 | cond_resched(); | |
6591 | goto retry; | |
6592 | } | |
4ffef5fe DN |
6593 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
6594 | int ret; | |
6595 | struct mm_walk mem_cgroup_move_charge_walk = { | |
6596 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6597 | .mm = mm, | |
6598 | .private = vma, | |
6599 | }; | |
6600 | if (is_vm_hugetlb_page(vma)) | |
6601 | continue; | |
4ffef5fe DN |
6602 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
6603 | &mem_cgroup_move_charge_walk); | |
6604 | if (ret) | |
6605 | /* | |
6606 | * means we have consumed all precharges and failed in | |
6607 | * doing additional charge. Just abandon here. | |
6608 | */ | |
6609 | break; | |
6610 | } | |
dfe076b0 | 6611 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
6612 | } |
6613 | ||
eb95419b | 6614 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 6615 | struct cgroup_taskset *tset) |
67e465a7 | 6616 | { |
2f7ee569 | 6617 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 6618 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 6619 | |
dfe076b0 | 6620 | if (mm) { |
a433658c KM |
6621 | if (mc.to) |
6622 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
6623 | mmput(mm); |
6624 | } | |
a433658c KM |
6625 | if (mc.to) |
6626 | mem_cgroup_clear_mc(); | |
67e465a7 | 6627 | } |
5cfb80a7 | 6628 | #else /* !CONFIG_MMU */ |
eb95419b | 6629 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6630 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
6631 | { |
6632 | return 0; | |
6633 | } | |
eb95419b | 6634 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6635 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
6636 | { |
6637 | } | |
eb95419b | 6638 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 6639 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
6640 | { |
6641 | } | |
6642 | #endif | |
67e465a7 | 6643 | |
f00baae7 TH |
6644 | /* |
6645 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
6646 | * to verify sane_behavior flag on each mount attempt. | |
6647 | */ | |
eb95419b | 6648 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
6649 | { |
6650 | /* | |
6651 | * use_hierarchy is forced with sane_behavior. cgroup core | |
6652 | * guarantees that @root doesn't have any children, so turning it | |
6653 | * on for the root memcg is enough. | |
6654 | */ | |
eb95419b TH |
6655 | if (cgroup_sane_behavior(root_css->cgroup)) |
6656 | mem_cgroup_from_css(root_css)->use_hierarchy = true; | |
f00baae7 TH |
6657 | } |
6658 | ||
8cdea7c0 BS |
6659 | struct cgroup_subsys mem_cgroup_subsys = { |
6660 | .name = "memory", | |
6661 | .subsys_id = mem_cgroup_subsys_id, | |
92fb9748 | 6662 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6663 | .css_online = mem_cgroup_css_online, |
92fb9748 TH |
6664 | .css_offline = mem_cgroup_css_offline, |
6665 | .css_free = mem_cgroup_css_free, | |
7dc74be0 DN |
6666 | .can_attach = mem_cgroup_can_attach, |
6667 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 6668 | .attach = mem_cgroup_move_task, |
f00baae7 | 6669 | .bind = mem_cgroup_bind, |
6bc10349 | 6670 | .base_cftypes = mem_cgroup_files, |
6d12e2d8 | 6671 | .early_init = 0, |
04046e1a | 6672 | .use_id = 1, |
8cdea7c0 | 6673 | }; |
c077719b | 6674 | |
c255a458 | 6675 | #ifdef CONFIG_MEMCG_SWAP |
a42c390c MH |
6676 | static int __init enable_swap_account(char *s) |
6677 | { | |
a2c8990a | 6678 | if (!strcmp(s, "1")) |
a42c390c | 6679 | really_do_swap_account = 1; |
a2c8990a | 6680 | else if (!strcmp(s, "0")) |
a42c390c MH |
6681 | really_do_swap_account = 0; |
6682 | return 1; | |
6683 | } | |
a2c8990a | 6684 | __setup("swapaccount=", enable_swap_account); |
c077719b | 6685 | |
2d11085e MH |
6686 | static void __init memsw_file_init(void) |
6687 | { | |
6acc8b02 MH |
6688 | WARN_ON(cgroup_add_cftypes(&mem_cgroup_subsys, memsw_cgroup_files)); |
6689 | } | |
6690 | ||
6691 | static void __init enable_swap_cgroup(void) | |
6692 | { | |
6693 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
6694 | do_swap_account = 1; | |
6695 | memsw_file_init(); | |
6696 | } | |
2d11085e | 6697 | } |
6acc8b02 | 6698 | |
2d11085e | 6699 | #else |
6acc8b02 | 6700 | static void __init enable_swap_cgroup(void) |
2d11085e MH |
6701 | { |
6702 | } | |
c077719b | 6703 | #endif |
2d11085e MH |
6704 | |
6705 | /* | |
1081312f MH |
6706 | * subsys_initcall() for memory controller. |
6707 | * | |
6708 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
6709 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
6710 | * everything that doesn't depend on a specific mem_cgroup structure should | |
6711 | * be initialized from here. | |
2d11085e MH |
6712 | */ |
6713 | static int __init mem_cgroup_init(void) | |
6714 | { | |
6715 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); | |
6acc8b02 | 6716 | enable_swap_cgroup(); |
e4777496 | 6717 | memcg_stock_init(); |
2d11085e MH |
6718 | return 0; |
6719 | } | |
6720 | subsys_initcall(mem_cgroup_init); |