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; | |
3812c8c8 | 258 | atomic_t oom_wakeups; |
79dfdacc | 259 | |
1f4c025b | 260 | int swappiness; |
3c11ecf4 KH |
261 | /* OOM-Killer disable */ |
262 | int oom_kill_disable; | |
a7885eb8 | 263 | |
22a668d7 KH |
264 | /* set when res.limit == memsw.limit */ |
265 | bool memsw_is_minimum; | |
266 | ||
2e72b634 KS |
267 | /* protect arrays of thresholds */ |
268 | struct mutex thresholds_lock; | |
269 | ||
270 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 271 | struct mem_cgroup_thresholds thresholds; |
907860ed | 272 | |
2e72b634 | 273 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 274 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 275 | |
9490ff27 KH |
276 | /* For oom notifier event fd */ |
277 | struct list_head oom_notify; | |
185efc0f | 278 | |
7dc74be0 DN |
279 | /* |
280 | * Should we move charges of a task when a task is moved into this | |
281 | * mem_cgroup ? And what type of charges should we move ? | |
282 | */ | |
f894ffa8 | 283 | unsigned long move_charge_at_immigrate; |
619d094b KH |
284 | /* |
285 | * set > 0 if pages under this cgroup are moving to other cgroup. | |
286 | */ | |
287 | atomic_t moving_account; | |
312734c0 KH |
288 | /* taken only while moving_account > 0 */ |
289 | spinlock_t move_lock; | |
d52aa412 | 290 | /* |
c62b1a3b | 291 | * percpu counter. |
d52aa412 | 292 | */ |
3a7951b4 | 293 | struct mem_cgroup_stat_cpu __percpu *stat; |
711d3d2c KH |
294 | /* |
295 | * used when a cpu is offlined or other synchronizations | |
296 | * See mem_cgroup_read_stat(). | |
297 | */ | |
298 | struct mem_cgroup_stat_cpu nocpu_base; | |
299 | spinlock_t pcp_counter_lock; | |
d1a4c0b3 | 300 | |
5f578161 | 301 | atomic_t dead_count; |
4bd2c1ee | 302 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET) |
d1a4c0b3 GC |
303 | struct tcp_memcontrol tcp_mem; |
304 | #endif | |
2633d7a0 GC |
305 | #if defined(CONFIG_MEMCG_KMEM) |
306 | /* analogous to slab_common's slab_caches list. per-memcg */ | |
307 | struct list_head memcg_slab_caches; | |
308 | /* Not a spinlock, we can take a lot of time walking the list */ | |
309 | struct mutex slab_caches_mutex; | |
310 | /* Index in the kmem_cache->memcg_params->memcg_caches array */ | |
311 | int kmemcg_id; | |
312 | #endif | |
45cf7ebd GC |
313 | |
314 | int last_scanned_node; | |
315 | #if MAX_NUMNODES > 1 | |
316 | nodemask_t scan_nodes; | |
317 | atomic_t numainfo_events; | |
318 | atomic_t numainfo_updating; | |
319 | #endif | |
7d910c05 MH |
320 | /* |
321 | * Protects soft_contributed transitions. | |
322 | * See mem_cgroup_update_soft_limit | |
323 | */ | |
324 | spinlock_t soft_lock; | |
325 | ||
326 | /* | |
327 | * If true then this group has increased parents' children_in_excess | |
f894ffa8 | 328 | * when it got over the soft limit. |
7d910c05 MH |
329 | * When a group falls bellow the soft limit, parents' children_in_excess |
330 | * is decreased and soft_contributed changed to false. | |
331 | */ | |
332 | bool soft_contributed; | |
333 | ||
334 | /* Number of children that are in soft limit excess */ | |
335 | atomic_t children_in_excess; | |
70ddf637 | 336 | |
54f72fe0 JW |
337 | struct mem_cgroup_per_node *nodeinfo[0]; |
338 | /* WARNING: nodeinfo must be the last member here */ | |
8cdea7c0 BS |
339 | }; |
340 | ||
45cf7ebd GC |
341 | static size_t memcg_size(void) |
342 | { | |
343 | return sizeof(struct mem_cgroup) + | |
344 | nr_node_ids * sizeof(struct mem_cgroup_per_node); | |
345 | } | |
346 | ||
510fc4e1 GC |
347 | /* internal only representation about the status of kmem accounting. */ |
348 | enum { | |
349 | KMEM_ACCOUNTED_ACTIVE = 0, /* accounted by this cgroup itself */ | |
a8964b9b | 350 | KMEM_ACCOUNTED_ACTIVATED, /* static key enabled. */ |
7de37682 | 351 | KMEM_ACCOUNTED_DEAD, /* dead memcg with pending kmem charges */ |
510fc4e1 GC |
352 | }; |
353 | ||
a8964b9b GC |
354 | /* We account when limit is on, but only after call sites are patched */ |
355 | #define KMEM_ACCOUNTED_MASK \ | |
356 | ((1 << KMEM_ACCOUNTED_ACTIVE) | (1 << KMEM_ACCOUNTED_ACTIVATED)) | |
510fc4e1 GC |
357 | |
358 | #ifdef CONFIG_MEMCG_KMEM | |
359 | static inline void memcg_kmem_set_active(struct mem_cgroup *memcg) | |
360 | { | |
361 | set_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
362 | } | |
7de37682 GC |
363 | |
364 | static bool memcg_kmem_is_active(struct mem_cgroup *memcg) | |
365 | { | |
366 | return test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
367 | } | |
368 | ||
a8964b9b GC |
369 | static void memcg_kmem_set_activated(struct mem_cgroup *memcg) |
370 | { | |
371 | set_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags); | |
372 | } | |
373 | ||
55007d84 GC |
374 | static void memcg_kmem_clear_activated(struct mem_cgroup *memcg) |
375 | { | |
376 | clear_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags); | |
377 | } | |
378 | ||
7de37682 GC |
379 | static void memcg_kmem_mark_dead(struct mem_cgroup *memcg) |
380 | { | |
10d5ebf4 LZ |
381 | /* |
382 | * Our caller must use css_get() first, because memcg_uncharge_kmem() | |
383 | * will call css_put() if it sees the memcg is dead. | |
384 | */ | |
385 | smp_wmb(); | |
7de37682 GC |
386 | if (test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags)) |
387 | set_bit(KMEM_ACCOUNTED_DEAD, &memcg->kmem_account_flags); | |
388 | } | |
389 | ||
390 | static bool memcg_kmem_test_and_clear_dead(struct mem_cgroup *memcg) | |
391 | { | |
392 | return test_and_clear_bit(KMEM_ACCOUNTED_DEAD, | |
393 | &memcg->kmem_account_flags); | |
394 | } | |
510fc4e1 GC |
395 | #endif |
396 | ||
7dc74be0 DN |
397 | /* Stuffs for move charges at task migration. */ |
398 | /* | |
ee5e8472 GC |
399 | * Types of charges to be moved. "move_charge_at_immitgrate" and |
400 | * "immigrate_flags" are treated as a left-shifted bitmap of these types. | |
7dc74be0 DN |
401 | */ |
402 | enum move_type { | |
4ffef5fe | 403 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 404 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
405 | NR_MOVE_TYPE, |
406 | }; | |
407 | ||
4ffef5fe DN |
408 | /* "mc" and its members are protected by cgroup_mutex */ |
409 | static struct move_charge_struct { | |
b1dd693e | 410 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
411 | struct mem_cgroup *from; |
412 | struct mem_cgroup *to; | |
ee5e8472 | 413 | unsigned long immigrate_flags; |
4ffef5fe | 414 | unsigned long precharge; |
854ffa8d | 415 | unsigned long moved_charge; |
483c30b5 | 416 | unsigned long moved_swap; |
8033b97c DN |
417 | struct task_struct *moving_task; /* a task moving charges */ |
418 | wait_queue_head_t waitq; /* a waitq for other context */ | |
419 | } mc = { | |
2bd9bb20 | 420 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
421 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
422 | }; | |
4ffef5fe | 423 | |
90254a65 DN |
424 | static bool move_anon(void) |
425 | { | |
ee5e8472 | 426 | return test_bit(MOVE_CHARGE_TYPE_ANON, &mc.immigrate_flags); |
90254a65 DN |
427 | } |
428 | ||
87946a72 DN |
429 | static bool move_file(void) |
430 | { | |
ee5e8472 | 431 | return test_bit(MOVE_CHARGE_TYPE_FILE, &mc.immigrate_flags); |
87946a72 DN |
432 | } |
433 | ||
4e416953 BS |
434 | /* |
435 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
436 | * limit reclaim to prevent infinite loops, if they ever occur. | |
437 | */ | |
a0db00fc | 438 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
4e416953 | 439 | |
217bc319 KH |
440 | enum charge_type { |
441 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 442 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 443 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 444 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
445 | NR_CHARGE_TYPE, |
446 | }; | |
447 | ||
8c7c6e34 | 448 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
449 | enum res_type { |
450 | _MEM, | |
451 | _MEMSWAP, | |
452 | _OOM_TYPE, | |
510fc4e1 | 453 | _KMEM, |
86ae53e1 GC |
454 | }; |
455 | ||
a0db00fc KS |
456 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
457 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 458 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
459 | /* Used for OOM nofiier */ |
460 | #define OOM_CONTROL (0) | |
8c7c6e34 | 461 | |
75822b44 BS |
462 | /* |
463 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
464 | */ | |
465 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
466 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
467 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
468 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
469 | ||
0999821b GC |
470 | /* |
471 | * The memcg_create_mutex will be held whenever a new cgroup is created. | |
472 | * As a consequence, any change that needs to protect against new child cgroups | |
473 | * appearing has to hold it as well. | |
474 | */ | |
475 | static DEFINE_MUTEX(memcg_create_mutex); | |
476 | ||
b2145145 WL |
477 | struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s) |
478 | { | |
a7c6d554 | 479 | return s ? container_of(s, struct mem_cgroup, css) : NULL; |
b2145145 WL |
480 | } |
481 | ||
70ddf637 AV |
482 | /* Some nice accessors for the vmpressure. */ |
483 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
484 | { | |
485 | if (!memcg) | |
486 | memcg = root_mem_cgroup; | |
487 | return &memcg->vmpressure; | |
488 | } | |
489 | ||
490 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
491 | { | |
492 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
493 | } | |
494 | ||
495 | struct vmpressure *css_to_vmpressure(struct cgroup_subsys_state *css) | |
496 | { | |
497 | return &mem_cgroup_from_css(css)->vmpressure; | |
498 | } | |
499 | ||
7ffc0edc MH |
500 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
501 | { | |
502 | return (memcg == root_mem_cgroup); | |
503 | } | |
504 | ||
e1aab161 | 505 | /* Writing them here to avoid exposing memcg's inner layout */ |
4bd2c1ee | 506 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
e1aab161 | 507 | |
e1aab161 GC |
508 | void sock_update_memcg(struct sock *sk) |
509 | { | |
376be5ff | 510 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 511 | struct mem_cgroup *memcg; |
3f134619 | 512 | struct cg_proto *cg_proto; |
e1aab161 GC |
513 | |
514 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
515 | ||
f3f511e1 GC |
516 | /* Socket cloning can throw us here with sk_cgrp already |
517 | * filled. It won't however, necessarily happen from | |
518 | * process context. So the test for root memcg given | |
519 | * the current task's memcg won't help us in this case. | |
520 | * | |
521 | * Respecting the original socket's memcg is a better | |
522 | * decision in this case. | |
523 | */ | |
524 | if (sk->sk_cgrp) { | |
525 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
5347e5ae | 526 | css_get(&sk->sk_cgrp->memcg->css); |
f3f511e1 GC |
527 | return; |
528 | } | |
529 | ||
e1aab161 GC |
530 | rcu_read_lock(); |
531 | memcg = mem_cgroup_from_task(current); | |
3f134619 | 532 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
5347e5ae LZ |
533 | if (!mem_cgroup_is_root(memcg) && |
534 | memcg_proto_active(cg_proto) && css_tryget(&memcg->css)) { | |
3f134619 | 535 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
536 | } |
537 | rcu_read_unlock(); | |
538 | } | |
539 | } | |
540 | EXPORT_SYMBOL(sock_update_memcg); | |
541 | ||
542 | void sock_release_memcg(struct sock *sk) | |
543 | { | |
376be5ff | 544 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
545 | struct mem_cgroup *memcg; |
546 | WARN_ON(!sk->sk_cgrp->memcg); | |
547 | memcg = sk->sk_cgrp->memcg; | |
5347e5ae | 548 | css_put(&sk->sk_cgrp->memcg->css); |
e1aab161 GC |
549 | } |
550 | } | |
d1a4c0b3 GC |
551 | |
552 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
553 | { | |
554 | if (!memcg || mem_cgroup_is_root(memcg)) | |
555 | return NULL; | |
556 | ||
557 | return &memcg->tcp_mem.cg_proto; | |
558 | } | |
559 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 560 | |
3f134619 GC |
561 | static void disarm_sock_keys(struct mem_cgroup *memcg) |
562 | { | |
563 | if (!memcg_proto_activated(&memcg->tcp_mem.cg_proto)) | |
564 | return; | |
565 | static_key_slow_dec(&memcg_socket_limit_enabled); | |
566 | } | |
567 | #else | |
568 | static void disarm_sock_keys(struct mem_cgroup *memcg) | |
569 | { | |
570 | } | |
571 | #endif | |
572 | ||
a8964b9b | 573 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 GC |
574 | /* |
575 | * This will be the memcg's index in each cache's ->memcg_params->memcg_caches. | |
576 | * There are two main reasons for not using the css_id for this: | |
577 | * 1) this works better in sparse environments, where we have a lot of memcgs, | |
578 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
579 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
580 | * 200 entry array for that. | |
581 | * | |
582 | * 2) In order not to violate the cgroup API, we would like to do all memory | |
583 | * allocation in ->create(). At that point, we haven't yet allocated the | |
584 | * css_id. Having a separate index prevents us from messing with the cgroup | |
585 | * core for this | |
586 | * | |
587 | * The current size of the caches array is stored in | |
588 | * memcg_limited_groups_array_size. It will double each time we have to | |
589 | * increase it. | |
590 | */ | |
591 | static DEFINE_IDA(kmem_limited_groups); | |
749c5415 GC |
592 | int memcg_limited_groups_array_size; |
593 | ||
55007d84 GC |
594 | /* |
595 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
596 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
597 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
598 | * tunable, but that is strictly not necessary. | |
599 | * | |
600 | * MAX_SIZE should be as large as the number of css_ids. Ideally, we could get | |
601 | * this constant directly from cgroup, but it is understandable that this is | |
602 | * better kept as an internal representation in cgroup.c. In any case, the | |
603 | * css_id space is not getting any smaller, and we don't have to necessarily | |
604 | * increase ours as well if it increases. | |
605 | */ | |
606 | #define MEMCG_CACHES_MIN_SIZE 4 | |
607 | #define MEMCG_CACHES_MAX_SIZE 65535 | |
608 | ||
d7f25f8a GC |
609 | /* |
610 | * A lot of the calls to the cache allocation functions are expected to be | |
611 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
612 | * conditional to this static branch, we'll have to allow modules that does | |
613 | * kmem_cache_alloc and the such to see this symbol as well | |
614 | */ | |
a8964b9b | 615 | struct static_key memcg_kmem_enabled_key; |
d7f25f8a | 616 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b GC |
617 | |
618 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
619 | { | |
55007d84 | 620 | if (memcg_kmem_is_active(memcg)) { |
a8964b9b | 621 | static_key_slow_dec(&memcg_kmem_enabled_key); |
55007d84 GC |
622 | ida_simple_remove(&kmem_limited_groups, memcg->kmemcg_id); |
623 | } | |
bea207c8 GC |
624 | /* |
625 | * This check can't live in kmem destruction function, | |
626 | * since the charges will outlive the cgroup | |
627 | */ | |
628 | WARN_ON(res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0); | |
a8964b9b GC |
629 | } |
630 | #else | |
631 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
632 | { | |
633 | } | |
634 | #endif /* CONFIG_MEMCG_KMEM */ | |
635 | ||
636 | static void disarm_static_keys(struct mem_cgroup *memcg) | |
637 | { | |
638 | disarm_sock_keys(memcg); | |
639 | disarm_kmem_keys(memcg); | |
640 | } | |
641 | ||
c0ff4b85 | 642 | static void drain_all_stock_async(struct mem_cgroup *memcg); |
8c7c6e34 | 643 | |
f64c3f54 | 644 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 645 | mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid) |
f64c3f54 | 646 | { |
45cf7ebd | 647 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
54f72fe0 | 648 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
649 | } |
650 | ||
c0ff4b85 | 651 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 652 | { |
c0ff4b85 | 653 | return &memcg->css; |
d324236b WF |
654 | } |
655 | ||
f64c3f54 | 656 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 657 | page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 658 | { |
97a6c37b JW |
659 | int nid = page_to_nid(page); |
660 | int zid = page_zonenum(page); | |
f64c3f54 | 661 | |
c0ff4b85 | 662 | return mem_cgroup_zoneinfo(memcg, nid, zid); |
f64c3f54 BS |
663 | } |
664 | ||
711d3d2c KH |
665 | /* |
666 | * Implementation Note: reading percpu statistics for memcg. | |
667 | * | |
668 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
669 | * synchronization to implement "quick" read. There are trade-off between | |
670 | * reading cost and precision of value. Then, we may have a chance to implement | |
671 | * a periodic synchronizion of counter in memcg's counter. | |
672 | * | |
673 | * But this _read() function is used for user interface now. The user accounts | |
674 | * memory usage by memory cgroup and he _always_ requires exact value because | |
675 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
676 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
677 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
678 | * | |
679 | * If there are kernel internal actions which can make use of some not-exact | |
680 | * value, and reading all cpu value can be performance bottleneck in some | |
681 | * common workload, threashold and synchonization as vmstat[] should be | |
682 | * implemented. | |
683 | */ | |
c0ff4b85 | 684 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 685 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 686 | { |
7a159cc9 | 687 | long val = 0; |
c62b1a3b | 688 | int cpu; |
c62b1a3b | 689 | |
711d3d2c KH |
690 | get_online_cpus(); |
691 | for_each_online_cpu(cpu) | |
c0ff4b85 | 692 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 693 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
694 | spin_lock(&memcg->pcp_counter_lock); |
695 | val += memcg->nocpu_base.count[idx]; | |
696 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
697 | #endif |
698 | put_online_cpus(); | |
c62b1a3b KH |
699 | return val; |
700 | } | |
701 | ||
c0ff4b85 | 702 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
0c3e73e8 BS |
703 | bool charge) |
704 | { | |
705 | int val = (charge) ? 1 : -1; | |
bff6bb83 | 706 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); |
0c3e73e8 BS |
707 | } |
708 | ||
c0ff4b85 | 709 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
710 | enum mem_cgroup_events_index idx) |
711 | { | |
712 | unsigned long val = 0; | |
713 | int cpu; | |
714 | ||
715 | for_each_online_cpu(cpu) | |
c0ff4b85 | 716 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 717 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
718 | spin_lock(&memcg->pcp_counter_lock); |
719 | val += memcg->nocpu_base.events[idx]; | |
720 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f JW |
721 | #endif |
722 | return val; | |
723 | } | |
724 | ||
c0ff4b85 | 725 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 726 | struct page *page, |
b2402857 | 727 | bool anon, int nr_pages) |
d52aa412 | 728 | { |
c62b1a3b KH |
729 | preempt_disable(); |
730 | ||
b2402857 KH |
731 | /* |
732 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
733 | * counted as CACHE even if it's on ANON LRU. | |
734 | */ | |
735 | if (anon) | |
736 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], | |
c0ff4b85 | 737 | nr_pages); |
d52aa412 | 738 | else |
b2402857 | 739 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 740 | nr_pages); |
55e462b0 | 741 | |
b070e65c DR |
742 | if (PageTransHuge(page)) |
743 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], | |
744 | nr_pages); | |
745 | ||
e401f176 KH |
746 | /* pagein of a big page is an event. So, ignore page size */ |
747 | if (nr_pages > 0) | |
c0ff4b85 | 748 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 749 | else { |
c0ff4b85 | 750 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
751 | nr_pages = -nr_pages; /* for event */ |
752 | } | |
e401f176 | 753 | |
13114716 | 754 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
2e72b634 | 755 | |
c62b1a3b | 756 | preempt_enable(); |
6d12e2d8 KH |
757 | } |
758 | ||
bb2a0de9 | 759 | unsigned long |
4d7dcca2 | 760 | mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
761 | { |
762 | struct mem_cgroup_per_zone *mz; | |
763 | ||
764 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
765 | return mz->lru_size[lru]; | |
766 | } | |
767 | ||
768 | static unsigned long | |
c0ff4b85 | 769 | mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid, |
bb2a0de9 | 770 | unsigned int lru_mask) |
889976db YH |
771 | { |
772 | struct mem_cgroup_per_zone *mz; | |
f156ab93 | 773 | enum lru_list lru; |
bb2a0de9 KH |
774 | unsigned long ret = 0; |
775 | ||
c0ff4b85 | 776 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
bb2a0de9 | 777 | |
f156ab93 HD |
778 | for_each_lru(lru) { |
779 | if (BIT(lru) & lru_mask) | |
780 | ret += mz->lru_size[lru]; | |
bb2a0de9 KH |
781 | } |
782 | return ret; | |
783 | } | |
784 | ||
785 | static unsigned long | |
c0ff4b85 | 786 | mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 KH |
787 | int nid, unsigned int lru_mask) |
788 | { | |
889976db YH |
789 | u64 total = 0; |
790 | int zid; | |
791 | ||
bb2a0de9 | 792 | for (zid = 0; zid < MAX_NR_ZONES; zid++) |
c0ff4b85 R |
793 | total += mem_cgroup_zone_nr_lru_pages(memcg, |
794 | nid, zid, lru_mask); | |
bb2a0de9 | 795 | |
889976db YH |
796 | return total; |
797 | } | |
bb2a0de9 | 798 | |
c0ff4b85 | 799 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 800 | unsigned int lru_mask) |
6d12e2d8 | 801 | { |
889976db | 802 | int nid; |
6d12e2d8 KH |
803 | u64 total = 0; |
804 | ||
31aaea4a | 805 | for_each_node_state(nid, N_MEMORY) |
c0ff4b85 | 806 | total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
6d12e2d8 | 807 | return total; |
d52aa412 KH |
808 | } |
809 | ||
f53d7ce3 JW |
810 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
811 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
812 | { |
813 | unsigned long val, next; | |
814 | ||
13114716 | 815 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 816 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 817 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
818 | if ((long)next - (long)val < 0) { |
819 | switch (target) { | |
820 | case MEM_CGROUP_TARGET_THRESH: | |
821 | next = val + THRESHOLDS_EVENTS_TARGET; | |
822 | break; | |
7d910c05 MH |
823 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
824 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
825 | break; | |
f53d7ce3 JW |
826 | case MEM_CGROUP_TARGET_NUMAINFO: |
827 | next = val + NUMAINFO_EVENTS_TARGET; | |
828 | break; | |
829 | default: | |
830 | break; | |
831 | } | |
832 | __this_cpu_write(memcg->stat->targets[target], next); | |
833 | return true; | |
7a159cc9 | 834 | } |
f53d7ce3 | 835 | return false; |
d2265e6f KH |
836 | } |
837 | ||
7d910c05 | 838 | /* |
f894ffa8 | 839 | * Called from rate-limited memcg_check_events when enough |
7d910c05 | 840 | * MEM_CGROUP_TARGET_SOFTLIMIT events are accumulated and it makes sure |
f894ffa8 | 841 | * that all the parents up the hierarchy will be notified that this group |
7d910c05 MH |
842 | * is in excess or that it is not in excess anymore. mmecg->soft_contributed |
843 | * makes the transition a single action whenever the state flips from one to | |
f894ffa8 | 844 | * the other. |
7d910c05 MH |
845 | */ |
846 | static void mem_cgroup_update_soft_limit(struct mem_cgroup *memcg) | |
847 | { | |
848 | unsigned long long excess = res_counter_soft_limit_excess(&memcg->res); | |
849 | struct mem_cgroup *parent = memcg; | |
850 | int delta = 0; | |
851 | ||
852 | spin_lock(&memcg->soft_lock); | |
853 | if (excess) { | |
854 | if (!memcg->soft_contributed) { | |
855 | delta = 1; | |
856 | memcg->soft_contributed = true; | |
857 | } | |
858 | } else { | |
859 | if (memcg->soft_contributed) { | |
860 | delta = -1; | |
861 | memcg->soft_contributed = false; | |
862 | } | |
863 | } | |
864 | ||
865 | /* | |
866 | * Necessary to update all ancestors when hierarchy is used | |
867 | * because their event counter is not touched. | |
1be171d6 MH |
868 | * We track children even outside the hierarchy for the root |
869 | * cgroup because tree walk starting at root should visit | |
870 | * all cgroups and we want to prevent from pointless tree | |
871 | * walk if no children is below the limit. | |
7d910c05 MH |
872 | */ |
873 | while (delta && (parent = parent_mem_cgroup(parent))) | |
874 | atomic_add(delta, &parent->children_in_excess); | |
1be171d6 MH |
875 | if (memcg != root_mem_cgroup && !root_mem_cgroup->use_hierarchy) |
876 | atomic_add(delta, &root_mem_cgroup->children_in_excess); | |
7d910c05 MH |
877 | spin_unlock(&memcg->soft_lock); |
878 | } | |
879 | ||
d2265e6f KH |
880 | /* |
881 | * Check events in order. | |
882 | * | |
883 | */ | |
c0ff4b85 | 884 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f | 885 | { |
4799401f | 886 | preempt_disable(); |
d2265e6f | 887 | /* threshold event is triggered in finer grain than soft limit */ |
f53d7ce3 JW |
888 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
889 | MEM_CGROUP_TARGET_THRESH))) { | |
7d910c05 | 890 | bool do_softlimit; |
82b3f2a7 | 891 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 892 | |
7d910c05 MH |
893 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
894 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
895 | #if MAX_NUMNODES > 1 |
896 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
897 | MEM_CGROUP_TARGET_NUMAINFO); | |
898 | #endif | |
899 | preempt_enable(); | |
900 | ||
c0ff4b85 | 901 | mem_cgroup_threshold(memcg); |
7d910c05 MH |
902 | if (unlikely(do_softlimit)) |
903 | mem_cgroup_update_soft_limit(memcg); | |
453a9bf3 | 904 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 905 | if (unlikely(do_numainfo)) |
c0ff4b85 | 906 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 907 | #endif |
f53d7ce3 JW |
908 | } else |
909 | preempt_enable(); | |
d2265e6f KH |
910 | } |
911 | ||
cf475ad2 | 912 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 913 | { |
31a78f23 BS |
914 | /* |
915 | * mm_update_next_owner() may clear mm->owner to NULL | |
916 | * if it races with swapoff, page migration, etc. | |
917 | * So this can be called with p == NULL. | |
918 | */ | |
919 | if (unlikely(!p)) | |
920 | return NULL; | |
921 | ||
8af01f56 | 922 | return mem_cgroup_from_css(task_css(p, mem_cgroup_subsys_id)); |
78fb7466 PE |
923 | } |
924 | ||
a433658c | 925 | struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 926 | { |
c0ff4b85 | 927 | struct mem_cgroup *memcg = NULL; |
0b7f569e KH |
928 | |
929 | if (!mm) | |
930 | return NULL; | |
54595fe2 KH |
931 | /* |
932 | * Because we have no locks, mm->owner's may be being moved to other | |
933 | * cgroup. We use css_tryget() here even if this looks | |
934 | * pessimistic (rather than adding locks here). | |
935 | */ | |
936 | rcu_read_lock(); | |
937 | do { | |
c0ff4b85 R |
938 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
939 | if (unlikely(!memcg)) | |
54595fe2 | 940 | break; |
c0ff4b85 | 941 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 942 | rcu_read_unlock(); |
c0ff4b85 | 943 | return memcg; |
54595fe2 KH |
944 | } |
945 | ||
de57780d MH |
946 | static enum mem_cgroup_filter_t |
947 | mem_cgroup_filter(struct mem_cgroup *memcg, struct mem_cgroup *root, | |
948 | mem_cgroup_iter_filter cond) | |
949 | { | |
950 | if (!cond) | |
951 | return VISIT; | |
952 | return cond(memcg, root); | |
953 | } | |
954 | ||
16248d8f MH |
955 | /* |
956 | * Returns a next (in a pre-order walk) alive memcg (with elevated css | |
957 | * ref. count) or NULL if the whole root's subtree has been visited. | |
958 | * | |
959 | * helper function to be used by mem_cgroup_iter | |
960 | */ | |
961 | static struct mem_cgroup *__mem_cgroup_iter_next(struct mem_cgroup *root, | |
de57780d | 962 | struct mem_cgroup *last_visited, mem_cgroup_iter_filter cond) |
16248d8f | 963 | { |
492eb21b | 964 | struct cgroup_subsys_state *prev_css, *next_css; |
16248d8f | 965 | |
bd8815a6 | 966 | prev_css = last_visited ? &last_visited->css : NULL; |
16248d8f | 967 | skip_node: |
492eb21b | 968 | next_css = css_next_descendant_pre(prev_css, &root->css); |
16248d8f MH |
969 | |
970 | /* | |
971 | * Even if we found a group we have to make sure it is | |
972 | * alive. css && !memcg means that the groups should be | |
973 | * skipped and we should continue the tree walk. | |
974 | * last_visited css is safe to use because it is | |
975 | * protected by css_get and the tree walk is rcu safe. | |
976 | */ | |
492eb21b TH |
977 | if (next_css) { |
978 | struct mem_cgroup *mem = mem_cgroup_from_css(next_css); | |
979 | ||
de57780d MH |
980 | switch (mem_cgroup_filter(mem, root, cond)) { |
981 | case SKIP: | |
492eb21b | 982 | prev_css = next_css; |
16248d8f | 983 | goto skip_node; |
de57780d MH |
984 | case SKIP_TREE: |
985 | if (mem == root) | |
986 | return NULL; | |
987 | /* | |
988 | * css_rightmost_descendant is not an optimal way to | |
989 | * skip through a subtree (especially for imbalanced | |
990 | * trees leaning to right) but that's what we have right | |
991 | * now. More effective solution would be traversing | |
992 | * right-up for first non-NULL without calling | |
993 | * css_next_descendant_pre afterwards. | |
994 | */ | |
995 | prev_css = css_rightmost_descendant(next_css); | |
996 | goto skip_node; | |
997 | case VISIT: | |
998 | if (css_tryget(&mem->css)) | |
999 | return mem; | |
1000 | else { | |
1001 | prev_css = next_css; | |
1002 | goto skip_node; | |
1003 | } | |
1004 | break; | |
16248d8f MH |
1005 | } |
1006 | } | |
1007 | ||
1008 | return NULL; | |
1009 | } | |
1010 | ||
519ebea3 JW |
1011 | static void mem_cgroup_iter_invalidate(struct mem_cgroup *root) |
1012 | { | |
1013 | /* | |
1014 | * When a group in the hierarchy below root is destroyed, the | |
1015 | * hierarchy iterator can no longer be trusted since it might | |
1016 | * have pointed to the destroyed group. Invalidate it. | |
1017 | */ | |
1018 | atomic_inc(&root->dead_count); | |
1019 | } | |
1020 | ||
1021 | static struct mem_cgroup * | |
1022 | mem_cgroup_iter_load(struct mem_cgroup_reclaim_iter *iter, | |
1023 | struct mem_cgroup *root, | |
1024 | int *sequence) | |
1025 | { | |
1026 | struct mem_cgroup *position = NULL; | |
1027 | /* | |
1028 | * A cgroup destruction happens in two stages: offlining and | |
1029 | * release. They are separated by a RCU grace period. | |
1030 | * | |
1031 | * If the iterator is valid, we may still race with an | |
1032 | * offlining. The RCU lock ensures the object won't be | |
1033 | * released, tryget will fail if we lost the race. | |
1034 | */ | |
1035 | *sequence = atomic_read(&root->dead_count); | |
1036 | if (iter->last_dead_count == *sequence) { | |
1037 | smp_rmb(); | |
1038 | position = iter->last_visited; | |
1039 | if (position && !css_tryget(&position->css)) | |
1040 | position = NULL; | |
1041 | } | |
1042 | return position; | |
1043 | } | |
1044 | ||
1045 | static void mem_cgroup_iter_update(struct mem_cgroup_reclaim_iter *iter, | |
1046 | struct mem_cgroup *last_visited, | |
1047 | struct mem_cgroup *new_position, | |
1048 | int sequence) | |
1049 | { | |
1050 | if (last_visited) | |
1051 | css_put(&last_visited->css); | |
1052 | /* | |
1053 | * We store the sequence count from the time @last_visited was | |
1054 | * loaded successfully instead of rereading it here so that we | |
1055 | * don't lose destruction events in between. We could have | |
1056 | * raced with the destruction of @new_position after all. | |
1057 | */ | |
1058 | iter->last_visited = new_position; | |
1059 | smp_wmb(); | |
1060 | iter->last_dead_count = sequence; | |
1061 | } | |
1062 | ||
5660048c JW |
1063 | /** |
1064 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1065 | * @root: hierarchy root | |
1066 | * @prev: previously returned memcg, NULL on first invocation | |
1067 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
de57780d | 1068 | * @cond: filter for visited nodes, NULL for no filter |
5660048c JW |
1069 | * |
1070 | * Returns references to children of the hierarchy below @root, or | |
1071 | * @root itself, or %NULL after a full round-trip. | |
1072 | * | |
1073 | * Caller must pass the return value in @prev on subsequent | |
1074 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1075 | * to cancel a hierarchy walk before the round-trip is complete. | |
1076 | * | |
1077 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
1078 | * divide up the memcgs in the hierarchy among all concurrent | |
1079 | * reclaimers operating on the same zone and priority. | |
1080 | */ | |
de57780d | 1081 | struct mem_cgroup *mem_cgroup_iter_cond(struct mem_cgroup *root, |
5660048c | 1082 | struct mem_cgroup *prev, |
de57780d MH |
1083 | struct mem_cgroup_reclaim_cookie *reclaim, |
1084 | mem_cgroup_iter_filter cond) | |
14067bb3 | 1085 | { |
9f3a0d09 | 1086 | struct mem_cgroup *memcg = NULL; |
542f85f9 | 1087 | struct mem_cgroup *last_visited = NULL; |
711d3d2c | 1088 | |
de57780d MH |
1089 | if (mem_cgroup_disabled()) { |
1090 | /* first call must return non-NULL, second return NULL */ | |
1091 | return (struct mem_cgroup *)(unsigned long)!prev; | |
1092 | } | |
5660048c | 1093 | |
9f3a0d09 JW |
1094 | if (!root) |
1095 | root = root_mem_cgroup; | |
7d74b06f | 1096 | |
9f3a0d09 | 1097 | if (prev && !reclaim) |
542f85f9 | 1098 | last_visited = prev; |
14067bb3 | 1099 | |
9f3a0d09 JW |
1100 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1101 | if (prev) | |
c40046f3 | 1102 | goto out_css_put; |
de57780d MH |
1103 | if (mem_cgroup_filter(root, root, cond) == VISIT) |
1104 | return root; | |
1105 | return NULL; | |
9f3a0d09 | 1106 | } |
14067bb3 | 1107 | |
542f85f9 | 1108 | rcu_read_lock(); |
9f3a0d09 | 1109 | while (!memcg) { |
527a5ec9 | 1110 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
519ebea3 | 1111 | int uninitialized_var(seq); |
711d3d2c | 1112 | |
527a5ec9 JW |
1113 | if (reclaim) { |
1114 | int nid = zone_to_nid(reclaim->zone); | |
1115 | int zid = zone_idx(reclaim->zone); | |
1116 | struct mem_cgroup_per_zone *mz; | |
1117 | ||
1118 | mz = mem_cgroup_zoneinfo(root, nid, zid); | |
1119 | iter = &mz->reclaim_iter[reclaim->priority]; | |
542f85f9 | 1120 | if (prev && reclaim->generation != iter->generation) { |
5f578161 | 1121 | iter->last_visited = NULL; |
542f85f9 MH |
1122 | goto out_unlock; |
1123 | } | |
5f578161 | 1124 | |
519ebea3 | 1125 | last_visited = mem_cgroup_iter_load(iter, root, &seq); |
527a5ec9 | 1126 | } |
7d74b06f | 1127 | |
de57780d | 1128 | memcg = __mem_cgroup_iter_next(root, last_visited, cond); |
14067bb3 | 1129 | |
527a5ec9 | 1130 | if (reclaim) { |
519ebea3 | 1131 | mem_cgroup_iter_update(iter, last_visited, memcg, seq); |
542f85f9 | 1132 | |
19f39402 | 1133 | if (!memcg) |
527a5ec9 JW |
1134 | iter->generation++; |
1135 | else if (!prev && memcg) | |
1136 | reclaim->generation = iter->generation; | |
1137 | } | |
9f3a0d09 | 1138 | |
de57780d MH |
1139 | /* |
1140 | * We have finished the whole tree walk or no group has been | |
1141 | * visited because filter told us to skip the root node. | |
1142 | */ | |
1143 | if (!memcg && (prev || (cond && !last_visited))) | |
542f85f9 | 1144 | goto out_unlock; |
9f3a0d09 | 1145 | } |
542f85f9 MH |
1146 | out_unlock: |
1147 | rcu_read_unlock(); | |
c40046f3 MH |
1148 | out_css_put: |
1149 | if (prev && prev != root) | |
1150 | css_put(&prev->css); | |
1151 | ||
9f3a0d09 | 1152 | return memcg; |
14067bb3 | 1153 | } |
7d74b06f | 1154 | |
5660048c JW |
1155 | /** |
1156 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1157 | * @root: hierarchy root | |
1158 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1159 | */ | |
1160 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1161 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1162 | { |
1163 | if (!root) | |
1164 | root = root_mem_cgroup; | |
1165 | if (prev && prev != root) | |
1166 | css_put(&prev->css); | |
1167 | } | |
7d74b06f | 1168 | |
9f3a0d09 JW |
1169 | /* |
1170 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1171 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1172 | * be used for reference counting. | |
1173 | */ | |
1174 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1175 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1176 | iter != NULL; \ |
527a5ec9 | 1177 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1178 | |
9f3a0d09 | 1179 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1180 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1181 | iter != NULL; \ |
527a5ec9 | 1182 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1183 | |
68ae564b | 1184 | void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
456f998e | 1185 | { |
c0ff4b85 | 1186 | struct mem_cgroup *memcg; |
456f998e | 1187 | |
456f998e | 1188 | rcu_read_lock(); |
c0ff4b85 R |
1189 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1190 | if (unlikely(!memcg)) | |
456f998e YH |
1191 | goto out; |
1192 | ||
1193 | switch (idx) { | |
456f998e | 1194 | case PGFAULT: |
0e574a93 JW |
1195 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1196 | break; | |
1197 | case PGMAJFAULT: | |
1198 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1199 | break; |
1200 | default: | |
1201 | BUG(); | |
1202 | } | |
1203 | out: | |
1204 | rcu_read_unlock(); | |
1205 | } | |
68ae564b | 1206 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
456f998e | 1207 | |
925b7673 JW |
1208 | /** |
1209 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1210 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1211 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1212 | * |
1213 | * Returns the lru list vector holding pages for the given @zone and | |
1214 | * @mem. This can be the global zone lruvec, if the memory controller | |
1215 | * is disabled. | |
1216 | */ | |
1217 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1218 | struct mem_cgroup *memcg) | |
1219 | { | |
1220 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1221 | struct lruvec *lruvec; |
925b7673 | 1222 | |
bea8c150 HD |
1223 | if (mem_cgroup_disabled()) { |
1224 | lruvec = &zone->lruvec; | |
1225 | goto out; | |
1226 | } | |
925b7673 JW |
1227 | |
1228 | mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone)); | |
bea8c150 HD |
1229 | lruvec = &mz->lruvec; |
1230 | out: | |
1231 | /* | |
1232 | * Since a node can be onlined after the mem_cgroup was created, | |
1233 | * we have to be prepared to initialize lruvec->zone here; | |
1234 | * and if offlined then reonlined, we need to reinitialize it. | |
1235 | */ | |
1236 | if (unlikely(lruvec->zone != zone)) | |
1237 | lruvec->zone = zone; | |
1238 | return lruvec; | |
925b7673 JW |
1239 | } |
1240 | ||
08e552c6 KH |
1241 | /* |
1242 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
1243 | * Operations are called by routine of global LRU independently from memcg. | |
1244 | * What we have to take care of here is validness of pc->mem_cgroup. | |
1245 | * | |
1246 | * Changes to pc->mem_cgroup happens when | |
1247 | * 1. charge | |
1248 | * 2. moving account | |
1249 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
1250 | * It is added to LRU before charge. | |
1251 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
1252 | * When moving account, the page is not on LRU. It's isolated. | |
1253 | */ | |
4f98a2fe | 1254 | |
925b7673 | 1255 | /** |
fa9add64 | 1256 | * mem_cgroup_page_lruvec - return lruvec for adding an lru page |
925b7673 | 1257 | * @page: the page |
fa9add64 | 1258 | * @zone: zone of the page |
925b7673 | 1259 | */ |
fa9add64 | 1260 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1261 | { |
08e552c6 | 1262 | struct mem_cgroup_per_zone *mz; |
925b7673 JW |
1263 | struct mem_cgroup *memcg; |
1264 | struct page_cgroup *pc; | |
bea8c150 | 1265 | struct lruvec *lruvec; |
6d12e2d8 | 1266 | |
bea8c150 HD |
1267 | if (mem_cgroup_disabled()) { |
1268 | lruvec = &zone->lruvec; | |
1269 | goto out; | |
1270 | } | |
925b7673 | 1271 | |
08e552c6 | 1272 | pc = lookup_page_cgroup(page); |
38c5d72f | 1273 | memcg = pc->mem_cgroup; |
7512102c HD |
1274 | |
1275 | /* | |
fa9add64 | 1276 | * Surreptitiously switch any uncharged offlist page to root: |
7512102c HD |
1277 | * an uncharged page off lru does nothing to secure |
1278 | * its former mem_cgroup from sudden removal. | |
1279 | * | |
1280 | * Our caller holds lru_lock, and PageCgroupUsed is updated | |
1281 | * under page_cgroup lock: between them, they make all uses | |
1282 | * of pc->mem_cgroup safe. | |
1283 | */ | |
fa9add64 | 1284 | if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup) |
7512102c HD |
1285 | pc->mem_cgroup = memcg = root_mem_cgroup; |
1286 | ||
925b7673 | 1287 | mz = page_cgroup_zoneinfo(memcg, page); |
bea8c150 HD |
1288 | lruvec = &mz->lruvec; |
1289 | out: | |
1290 | /* | |
1291 | * Since a node can be onlined after the mem_cgroup was created, | |
1292 | * we have to be prepared to initialize lruvec->zone here; | |
1293 | * and if offlined then reonlined, we need to reinitialize it. | |
1294 | */ | |
1295 | if (unlikely(lruvec->zone != zone)) | |
1296 | lruvec->zone = zone; | |
1297 | return lruvec; | |
08e552c6 | 1298 | } |
b69408e8 | 1299 | |
925b7673 | 1300 | /** |
fa9add64 HD |
1301 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1302 | * @lruvec: mem_cgroup per zone lru vector | |
1303 | * @lru: index of lru list the page is sitting on | |
1304 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1305 | * |
fa9add64 HD |
1306 | * This function must be called when a page is added to or removed from an |
1307 | * lru list. | |
3f58a829 | 1308 | */ |
fa9add64 HD |
1309 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1310 | int nr_pages) | |
3f58a829 MK |
1311 | { |
1312 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1313 | unsigned long *lru_size; |
3f58a829 MK |
1314 | |
1315 | if (mem_cgroup_disabled()) | |
1316 | return; | |
1317 | ||
fa9add64 HD |
1318 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1319 | lru_size = mz->lru_size + lru; | |
1320 | *lru_size += nr_pages; | |
1321 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1322 | } |
544122e5 | 1323 | |
3e92041d | 1324 | /* |
c0ff4b85 | 1325 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1326 | * hierarchy subtree |
1327 | */ | |
c3ac9a8a JW |
1328 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1329 | struct mem_cgroup *memcg) | |
3e92041d | 1330 | { |
91c63734 JW |
1331 | if (root_memcg == memcg) |
1332 | return true; | |
3a981f48 | 1333 | if (!root_memcg->use_hierarchy || !memcg) |
91c63734 | 1334 | return false; |
c3ac9a8a JW |
1335 | return css_is_ancestor(&memcg->css, &root_memcg->css); |
1336 | } | |
1337 | ||
1338 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, | |
1339 | struct mem_cgroup *memcg) | |
1340 | { | |
1341 | bool ret; | |
1342 | ||
91c63734 | 1343 | rcu_read_lock(); |
c3ac9a8a | 1344 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
91c63734 JW |
1345 | rcu_read_unlock(); |
1346 | return ret; | |
3e92041d MH |
1347 | } |
1348 | ||
ffbdccf5 DR |
1349 | bool task_in_mem_cgroup(struct task_struct *task, |
1350 | const struct mem_cgroup *memcg) | |
4c4a2214 | 1351 | { |
0b7f569e | 1352 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1353 | struct task_struct *p; |
ffbdccf5 | 1354 | bool ret; |
4c4a2214 | 1355 | |
158e0a2d | 1356 | p = find_lock_task_mm(task); |
de077d22 DR |
1357 | if (p) { |
1358 | curr = try_get_mem_cgroup_from_mm(p->mm); | |
1359 | task_unlock(p); | |
1360 | } else { | |
1361 | /* | |
1362 | * All threads may have already detached their mm's, but the oom | |
1363 | * killer still needs to detect if they have already been oom | |
1364 | * killed to prevent needlessly killing additional tasks. | |
1365 | */ | |
ffbdccf5 | 1366 | rcu_read_lock(); |
de077d22 DR |
1367 | curr = mem_cgroup_from_task(task); |
1368 | if (curr) | |
1369 | css_get(&curr->css); | |
ffbdccf5 | 1370 | rcu_read_unlock(); |
de077d22 | 1371 | } |
0b7f569e | 1372 | if (!curr) |
ffbdccf5 | 1373 | return false; |
d31f56db | 1374 | /* |
c0ff4b85 | 1375 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1376 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1377 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1378 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1379 | */ |
c0ff4b85 | 1380 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1381 | css_put(&curr->css); |
4c4a2214 DR |
1382 | return ret; |
1383 | } | |
1384 | ||
c56d5c7d | 1385 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1386 | { |
9b272977 | 1387 | unsigned long inactive_ratio; |
14797e23 | 1388 | unsigned long inactive; |
9b272977 | 1389 | unsigned long active; |
c772be93 | 1390 | unsigned long gb; |
14797e23 | 1391 | |
4d7dcca2 HD |
1392 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1393 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1394 | |
c772be93 KM |
1395 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1396 | if (gb) | |
1397 | inactive_ratio = int_sqrt(10 * gb); | |
1398 | else | |
1399 | inactive_ratio = 1; | |
1400 | ||
9b272977 | 1401 | return inactive * inactive_ratio < active; |
14797e23 KM |
1402 | } |
1403 | ||
6d61ef40 BS |
1404 | #define mem_cgroup_from_res_counter(counter, member) \ |
1405 | container_of(counter, struct mem_cgroup, member) | |
1406 | ||
19942822 | 1407 | /** |
9d11ea9f | 1408 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1409 | * @memcg: the memory cgroup |
19942822 | 1410 | * |
9d11ea9f | 1411 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1412 | * pages. |
19942822 | 1413 | */ |
c0ff4b85 | 1414 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1415 | { |
9d11ea9f JW |
1416 | unsigned long long margin; |
1417 | ||
c0ff4b85 | 1418 | margin = res_counter_margin(&memcg->res); |
9d11ea9f | 1419 | if (do_swap_account) |
c0ff4b85 | 1420 | margin = min(margin, res_counter_margin(&memcg->memsw)); |
7ec99d62 | 1421 | return margin >> PAGE_SHIFT; |
19942822 JW |
1422 | } |
1423 | ||
1f4c025b | 1424 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 | 1425 | { |
a7885eb8 | 1426 | /* root ? */ |
63876986 | 1427 | if (!css_parent(&memcg->css)) |
a7885eb8 KM |
1428 | return vm_swappiness; |
1429 | ||
bf1ff263 | 1430 | return memcg->swappiness; |
a7885eb8 KM |
1431 | } |
1432 | ||
619d094b KH |
1433 | /* |
1434 | * memcg->moving_account is used for checking possibility that some thread is | |
1435 | * calling move_account(). When a thread on CPU-A starts moving pages under | |
1436 | * a memcg, other threads should check memcg->moving_account under | |
1437 | * rcu_read_lock(), like this: | |
1438 | * | |
1439 | * CPU-A CPU-B | |
1440 | * rcu_read_lock() | |
1441 | * memcg->moving_account+1 if (memcg->mocing_account) | |
1442 | * take heavy locks. | |
1443 | * synchronize_rcu() update something. | |
1444 | * rcu_read_unlock() | |
1445 | * start move here. | |
1446 | */ | |
4331f7d3 KH |
1447 | |
1448 | /* for quick checking without looking up memcg */ | |
1449 | atomic_t memcg_moving __read_mostly; | |
1450 | ||
c0ff4b85 | 1451 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a | 1452 | { |
4331f7d3 | 1453 | atomic_inc(&memcg_moving); |
619d094b | 1454 | atomic_inc(&memcg->moving_account); |
32047e2a KH |
1455 | synchronize_rcu(); |
1456 | } | |
1457 | ||
c0ff4b85 | 1458 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a | 1459 | { |
619d094b KH |
1460 | /* |
1461 | * Now, mem_cgroup_clear_mc() may call this function with NULL. | |
1462 | * We check NULL in callee rather than caller. | |
1463 | */ | |
4331f7d3 KH |
1464 | if (memcg) { |
1465 | atomic_dec(&memcg_moving); | |
619d094b | 1466 | atomic_dec(&memcg->moving_account); |
4331f7d3 | 1467 | } |
32047e2a | 1468 | } |
619d094b | 1469 | |
32047e2a KH |
1470 | /* |
1471 | * 2 routines for checking "mem" is under move_account() or not. | |
1472 | * | |
13fd1dd9 AM |
1473 | * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This |
1474 | * is used for avoiding races in accounting. If true, | |
32047e2a KH |
1475 | * pc->mem_cgroup may be overwritten. |
1476 | * | |
1477 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or | |
1478 | * under hierarchy of moving cgroups. This is for | |
1479 | * waiting at hith-memory prressure caused by "move". | |
1480 | */ | |
1481 | ||
13fd1dd9 | 1482 | static bool mem_cgroup_stolen(struct mem_cgroup *memcg) |
32047e2a KH |
1483 | { |
1484 | VM_BUG_ON(!rcu_read_lock_held()); | |
619d094b | 1485 | return atomic_read(&memcg->moving_account) > 0; |
32047e2a | 1486 | } |
4b534334 | 1487 | |
c0ff4b85 | 1488 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1489 | { |
2bd9bb20 KH |
1490 | struct mem_cgroup *from; |
1491 | struct mem_cgroup *to; | |
4b534334 | 1492 | bool ret = false; |
2bd9bb20 KH |
1493 | /* |
1494 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1495 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1496 | */ | |
1497 | spin_lock(&mc.lock); | |
1498 | from = mc.from; | |
1499 | to = mc.to; | |
1500 | if (!from) | |
1501 | goto unlock; | |
3e92041d | 1502 | |
c0ff4b85 R |
1503 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1504 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1505 | unlock: |
1506 | spin_unlock(&mc.lock); | |
4b534334 KH |
1507 | return ret; |
1508 | } | |
1509 | ||
c0ff4b85 | 1510 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1511 | { |
1512 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1513 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1514 | DEFINE_WAIT(wait); |
1515 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1516 | /* moving charge context might have finished. */ | |
1517 | if (mc.moving_task) | |
1518 | schedule(); | |
1519 | finish_wait(&mc.waitq, &wait); | |
1520 | return true; | |
1521 | } | |
1522 | } | |
1523 | return false; | |
1524 | } | |
1525 | ||
312734c0 KH |
1526 | /* |
1527 | * Take this lock when | |
1528 | * - a code tries to modify page's memcg while it's USED. | |
1529 | * - a code tries to modify page state accounting in a memcg. | |
13fd1dd9 | 1530 | * see mem_cgroup_stolen(), too. |
312734c0 KH |
1531 | */ |
1532 | static void move_lock_mem_cgroup(struct mem_cgroup *memcg, | |
1533 | unsigned long *flags) | |
1534 | { | |
1535 | spin_lock_irqsave(&memcg->move_lock, *flags); | |
1536 | } | |
1537 | ||
1538 | static void move_unlock_mem_cgroup(struct mem_cgroup *memcg, | |
1539 | unsigned long *flags) | |
1540 | { | |
1541 | spin_unlock_irqrestore(&memcg->move_lock, *flags); | |
1542 | } | |
1543 | ||
58cf188e | 1544 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1545 | /** |
58cf188e | 1546 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1547 | * @memcg: The memory cgroup that went over limit |
1548 | * @p: Task that is going to be killed | |
1549 | * | |
1550 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1551 | * enabled | |
1552 | */ | |
1553 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1554 | { | |
1555 | struct cgroup *task_cgrp; | |
1556 | struct cgroup *mem_cgrp; | |
1557 | /* | |
1558 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1559 | * on the assumption that OOM is serialized for memory controller. | |
1560 | * If this assumption is broken, revisit this code. | |
1561 | */ | |
1562 | static char memcg_name[PATH_MAX]; | |
1563 | int ret; | |
58cf188e SZ |
1564 | struct mem_cgroup *iter; |
1565 | unsigned int i; | |
e222432b | 1566 | |
58cf188e | 1567 | if (!p) |
e222432b BS |
1568 | return; |
1569 | ||
e222432b BS |
1570 | rcu_read_lock(); |
1571 | ||
1572 | mem_cgrp = memcg->css.cgroup; | |
1573 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1574 | ||
1575 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1576 | if (ret < 0) { | |
1577 | /* | |
1578 | * Unfortunately, we are unable to convert to a useful name | |
1579 | * But we'll still print out the usage information | |
1580 | */ | |
1581 | rcu_read_unlock(); | |
1582 | goto done; | |
1583 | } | |
1584 | rcu_read_unlock(); | |
1585 | ||
d045197f | 1586 | pr_info("Task in %s killed", memcg_name); |
e222432b BS |
1587 | |
1588 | rcu_read_lock(); | |
1589 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1590 | if (ret < 0) { | |
1591 | rcu_read_unlock(); | |
1592 | goto done; | |
1593 | } | |
1594 | rcu_read_unlock(); | |
1595 | ||
1596 | /* | |
1597 | * Continues from above, so we don't need an KERN_ level | |
1598 | */ | |
d045197f | 1599 | pr_cont(" as a result of limit of %s\n", memcg_name); |
e222432b BS |
1600 | done: |
1601 | ||
d045197f | 1602 | pr_info("memory: usage %llukB, limit %llukB, failcnt %llu\n", |
e222432b BS |
1603 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, |
1604 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1605 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
d045197f | 1606 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %llu\n", |
e222432b BS |
1607 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, |
1608 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1609 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
d045197f | 1610 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %llu\n", |
510fc4e1 GC |
1611 | res_counter_read_u64(&memcg->kmem, RES_USAGE) >> 10, |
1612 | res_counter_read_u64(&memcg->kmem, RES_LIMIT) >> 10, | |
1613 | res_counter_read_u64(&memcg->kmem, RES_FAILCNT)); | |
58cf188e SZ |
1614 | |
1615 | for_each_mem_cgroup_tree(iter, memcg) { | |
1616 | pr_info("Memory cgroup stats"); | |
1617 | ||
1618 | rcu_read_lock(); | |
1619 | ret = cgroup_path(iter->css.cgroup, memcg_name, PATH_MAX); | |
1620 | if (!ret) | |
1621 | pr_cont(" for %s", memcg_name); | |
1622 | rcu_read_unlock(); | |
1623 | pr_cont(":"); | |
1624 | ||
1625 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1626 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1627 | continue; | |
1628 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], | |
1629 | K(mem_cgroup_read_stat(iter, i))); | |
1630 | } | |
1631 | ||
1632 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1633 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1634 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1635 | ||
1636 | pr_cont("\n"); | |
1637 | } | |
e222432b BS |
1638 | } |
1639 | ||
81d39c20 KH |
1640 | /* |
1641 | * This function returns the number of memcg under hierarchy tree. Returns | |
1642 | * 1(self count) if no children. | |
1643 | */ | |
c0ff4b85 | 1644 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1645 | { |
1646 | int num = 0; | |
7d74b06f KH |
1647 | struct mem_cgroup *iter; |
1648 | ||
c0ff4b85 | 1649 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1650 | num++; |
81d39c20 KH |
1651 | return num; |
1652 | } | |
1653 | ||
a63d83f4 DR |
1654 | /* |
1655 | * Return the memory (and swap, if configured) limit for a memcg. | |
1656 | */ | |
9cbb78bb | 1657 | static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 DR |
1658 | { |
1659 | u64 limit; | |
a63d83f4 | 1660 | |
f3e8eb70 | 1661 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
f3e8eb70 | 1662 | |
a63d83f4 | 1663 | /* |
9a5a8f19 | 1664 | * Do not consider swap space if we cannot swap due to swappiness |
a63d83f4 | 1665 | */ |
9a5a8f19 MH |
1666 | if (mem_cgroup_swappiness(memcg)) { |
1667 | u64 memsw; | |
1668 | ||
1669 | limit += total_swap_pages << PAGE_SHIFT; | |
1670 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1671 | ||
1672 | /* | |
1673 | * If memsw is finite and limits the amount of swap space | |
1674 | * available to this memcg, return that limit. | |
1675 | */ | |
1676 | limit = min(limit, memsw); | |
1677 | } | |
1678 | ||
1679 | return limit; | |
a63d83f4 DR |
1680 | } |
1681 | ||
19965460 DR |
1682 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1683 | int order) | |
9cbb78bb DR |
1684 | { |
1685 | struct mem_cgroup *iter; | |
1686 | unsigned long chosen_points = 0; | |
1687 | unsigned long totalpages; | |
1688 | unsigned int points = 0; | |
1689 | struct task_struct *chosen = NULL; | |
1690 | ||
876aafbf | 1691 | /* |
465adcf1 DR |
1692 | * If current has a pending SIGKILL or is exiting, then automatically |
1693 | * select it. The goal is to allow it to allocate so that it may | |
1694 | * quickly exit and free its memory. | |
876aafbf | 1695 | */ |
465adcf1 | 1696 | if (fatal_signal_pending(current) || current->flags & PF_EXITING) { |
876aafbf DR |
1697 | set_thread_flag(TIF_MEMDIE); |
1698 | return; | |
1699 | } | |
1700 | ||
1701 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
9cbb78bb DR |
1702 | totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1; |
1703 | for_each_mem_cgroup_tree(iter, memcg) { | |
72ec7029 | 1704 | struct css_task_iter it; |
9cbb78bb DR |
1705 | struct task_struct *task; |
1706 | ||
72ec7029 TH |
1707 | css_task_iter_start(&iter->css, &it); |
1708 | while ((task = css_task_iter_next(&it))) { | |
9cbb78bb DR |
1709 | switch (oom_scan_process_thread(task, totalpages, NULL, |
1710 | false)) { | |
1711 | case OOM_SCAN_SELECT: | |
1712 | if (chosen) | |
1713 | put_task_struct(chosen); | |
1714 | chosen = task; | |
1715 | chosen_points = ULONG_MAX; | |
1716 | get_task_struct(chosen); | |
1717 | /* fall through */ | |
1718 | case OOM_SCAN_CONTINUE: | |
1719 | continue; | |
1720 | case OOM_SCAN_ABORT: | |
72ec7029 | 1721 | css_task_iter_end(&it); |
9cbb78bb DR |
1722 | mem_cgroup_iter_break(memcg, iter); |
1723 | if (chosen) | |
1724 | put_task_struct(chosen); | |
1725 | return; | |
1726 | case OOM_SCAN_OK: | |
1727 | break; | |
1728 | }; | |
1729 | points = oom_badness(task, memcg, NULL, totalpages); | |
1730 | if (points > chosen_points) { | |
1731 | if (chosen) | |
1732 | put_task_struct(chosen); | |
1733 | chosen = task; | |
1734 | chosen_points = points; | |
1735 | get_task_struct(chosen); | |
1736 | } | |
1737 | } | |
72ec7029 | 1738 | css_task_iter_end(&it); |
9cbb78bb DR |
1739 | } |
1740 | ||
1741 | if (!chosen) | |
1742 | return; | |
1743 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1744 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1745 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1746 | } |
1747 | ||
5660048c JW |
1748 | static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg, |
1749 | gfp_t gfp_mask, | |
1750 | unsigned long flags) | |
1751 | { | |
1752 | unsigned long total = 0; | |
1753 | bool noswap = false; | |
1754 | int loop; | |
1755 | ||
1756 | if (flags & MEM_CGROUP_RECLAIM_NOSWAP) | |
1757 | noswap = true; | |
1758 | if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum) | |
1759 | noswap = true; | |
1760 | ||
1761 | for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) { | |
1762 | if (loop) | |
1763 | drain_all_stock_async(memcg); | |
1764 | total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap); | |
1765 | /* | |
1766 | * Allow limit shrinkers, which are triggered directly | |
1767 | * by userspace, to catch signals and stop reclaim | |
1768 | * after minimal progress, regardless of the margin. | |
1769 | */ | |
1770 | if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK)) | |
1771 | break; | |
1772 | if (mem_cgroup_margin(memcg)) | |
1773 | break; | |
1774 | /* | |
1775 | * If nothing was reclaimed after two attempts, there | |
1776 | * may be no reclaimable pages in this hierarchy. | |
1777 | */ | |
1778 | if (loop && !total) | |
1779 | break; | |
1780 | } | |
1781 | return total; | |
1782 | } | |
1783 | ||
e883110a | 1784 | #if MAX_NUMNODES > 1 |
4d0c066d KH |
1785 | /** |
1786 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1787 | * @memcg: the target memcg |
4d0c066d KH |
1788 | * @nid: the node ID to be checked. |
1789 | * @noswap : specify true here if the user wants flle only information. | |
1790 | * | |
1791 | * This function returns whether the specified memcg contains any | |
1792 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1793 | * pages in the node. | |
1794 | */ | |
c0ff4b85 | 1795 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1796 | int nid, bool noswap) |
1797 | { | |
c0ff4b85 | 1798 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1799 | return true; |
1800 | if (noswap || !total_swap_pages) | |
1801 | return false; | |
c0ff4b85 | 1802 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1803 | return true; |
1804 | return false; | |
1805 | ||
1806 | } | |
889976db YH |
1807 | |
1808 | /* | |
1809 | * Always updating the nodemask is not very good - even if we have an empty | |
1810 | * list or the wrong list here, we can start from some node and traverse all | |
1811 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1812 | * | |
1813 | */ | |
c0ff4b85 | 1814 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1815 | { |
1816 | int nid; | |
453a9bf3 KH |
1817 | /* |
1818 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1819 | * pagein/pageout changes since the last update. | |
1820 | */ | |
c0ff4b85 | 1821 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1822 | return; |
c0ff4b85 | 1823 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1824 | return; |
1825 | ||
889976db | 1826 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1827 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1828 | |
31aaea4a | 1829 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1830 | |
c0ff4b85 R |
1831 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1832 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1833 | } |
453a9bf3 | 1834 | |
c0ff4b85 R |
1835 | atomic_set(&memcg->numainfo_events, 0); |
1836 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1837 | } |
1838 | ||
1839 | /* | |
1840 | * Selecting a node where we start reclaim from. Because what we need is just | |
1841 | * reducing usage counter, start from anywhere is O,K. Considering | |
1842 | * memory reclaim from current node, there are pros. and cons. | |
1843 | * | |
1844 | * Freeing memory from current node means freeing memory from a node which | |
1845 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1846 | * hit limits, it will see a contention on a node. But freeing from remote | |
1847 | * node means more costs for memory reclaim because of memory latency. | |
1848 | * | |
1849 | * Now, we use round-robin. Better algorithm is welcomed. | |
1850 | */ | |
c0ff4b85 | 1851 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1852 | { |
1853 | int node; | |
1854 | ||
c0ff4b85 R |
1855 | mem_cgroup_may_update_nodemask(memcg); |
1856 | node = memcg->last_scanned_node; | |
889976db | 1857 | |
c0ff4b85 | 1858 | node = next_node(node, memcg->scan_nodes); |
889976db | 1859 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1860 | node = first_node(memcg->scan_nodes); |
889976db YH |
1861 | /* |
1862 | * We call this when we hit limit, not when pages are added to LRU. | |
1863 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1864 | * memcg is too small and all pages are not on LRU. In that case, | |
1865 | * we use curret node. | |
1866 | */ | |
1867 | if (unlikely(node == MAX_NUMNODES)) | |
1868 | node = numa_node_id(); | |
1869 | ||
c0ff4b85 | 1870 | memcg->last_scanned_node = node; |
889976db YH |
1871 | return node; |
1872 | } | |
1873 | ||
1874 | #else | |
c0ff4b85 | 1875 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1876 | { |
1877 | return 0; | |
1878 | } | |
4d0c066d | 1879 | |
889976db YH |
1880 | #endif |
1881 | ||
3b38722e | 1882 | /* |
a5b7c87f MH |
1883 | * A group is eligible for the soft limit reclaim under the given root |
1884 | * hierarchy if | |
f894ffa8 AM |
1885 | * a) it is over its soft limit |
1886 | * b) any parent up the hierarchy is over its soft limit | |
7d910c05 MH |
1887 | * |
1888 | * If the given group doesn't have any children over the limit then it | |
1889 | * doesn't make any sense to iterate its subtree. | |
3b38722e | 1890 | */ |
de57780d MH |
1891 | enum mem_cgroup_filter_t |
1892 | mem_cgroup_soft_reclaim_eligible(struct mem_cgroup *memcg, | |
a5b7c87f | 1893 | struct mem_cgroup *root) |
3b38722e | 1894 | { |
e839b6a1 MH |
1895 | struct mem_cgroup *parent; |
1896 | ||
1897 | if (!memcg) | |
1898 | memcg = root_mem_cgroup; | |
1899 | parent = memcg; | |
3b38722e MH |
1900 | |
1901 | if (res_counter_soft_limit_excess(&memcg->res)) | |
de57780d | 1902 | return VISIT; |
3b38722e MH |
1903 | |
1904 | /* | |
a5b7c87f MH |
1905 | * If any parent up to the root in the hierarchy is over its soft limit |
1906 | * then we have to obey and reclaim from this group as well. | |
3b38722e | 1907 | */ |
f894ffa8 | 1908 | while ((parent = parent_mem_cgroup(parent))) { |
3b38722e | 1909 | if (res_counter_soft_limit_excess(&parent->res)) |
de57780d | 1910 | return VISIT; |
a5b7c87f MH |
1911 | if (parent == root) |
1912 | break; | |
6d61ef40 | 1913 | } |
3b38722e | 1914 | |
7d910c05 MH |
1915 | if (!atomic_read(&memcg->children_in_excess)) |
1916 | return SKIP_TREE; | |
de57780d | 1917 | return SKIP; |
6d61ef40 BS |
1918 | } |
1919 | ||
fb2a6fc5 JW |
1920 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1921 | ||
867578cb KH |
1922 | /* |
1923 | * Check OOM-Killer is already running under our hierarchy. | |
1924 | * If someone is running, return false. | |
1925 | */ | |
fb2a6fc5 | 1926 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1927 | { |
79dfdacc | 1928 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1929 | |
fb2a6fc5 JW |
1930 | spin_lock(&memcg_oom_lock); |
1931 | ||
9f3a0d09 | 1932 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1933 | if (iter->oom_lock) { |
79dfdacc MH |
1934 | /* |
1935 | * this subtree of our hierarchy is already locked | |
1936 | * so we cannot give a lock. | |
1937 | */ | |
79dfdacc | 1938 | failed = iter; |
9f3a0d09 JW |
1939 | mem_cgroup_iter_break(memcg, iter); |
1940 | break; | |
23751be0 JW |
1941 | } else |
1942 | iter->oom_lock = true; | |
7d74b06f | 1943 | } |
867578cb | 1944 | |
fb2a6fc5 JW |
1945 | if (failed) { |
1946 | /* | |
1947 | * OK, we failed to lock the whole subtree so we have | |
1948 | * to clean up what we set up to the failing subtree | |
1949 | */ | |
1950 | for_each_mem_cgroup_tree(iter, memcg) { | |
1951 | if (iter == failed) { | |
1952 | mem_cgroup_iter_break(memcg, iter); | |
1953 | break; | |
1954 | } | |
1955 | iter->oom_lock = false; | |
79dfdacc | 1956 | } |
79dfdacc | 1957 | } |
fb2a6fc5 JW |
1958 | |
1959 | spin_unlock(&memcg_oom_lock); | |
1960 | ||
1961 | return !failed; | |
a636b327 | 1962 | } |
0b7f569e | 1963 | |
fb2a6fc5 | 1964 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1965 | { |
7d74b06f KH |
1966 | struct mem_cgroup *iter; |
1967 | ||
fb2a6fc5 | 1968 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1969 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1970 | iter->oom_lock = false; |
fb2a6fc5 | 1971 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1972 | } |
1973 | ||
c0ff4b85 | 1974 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1975 | { |
1976 | struct mem_cgroup *iter; | |
1977 | ||
c0ff4b85 | 1978 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1979 | atomic_inc(&iter->under_oom); |
1980 | } | |
1981 | ||
c0ff4b85 | 1982 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1983 | { |
1984 | struct mem_cgroup *iter; | |
1985 | ||
867578cb KH |
1986 | /* |
1987 | * When a new child is created while the hierarchy is under oom, | |
1988 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1989 | * atomic_add_unless() here. | |
1990 | */ | |
c0ff4b85 | 1991 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1992 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
1993 | } |
1994 | ||
867578cb KH |
1995 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1996 | ||
dc98df5a | 1997 | struct oom_wait_info { |
d79154bb | 1998 | struct mem_cgroup *memcg; |
dc98df5a KH |
1999 | wait_queue_t wait; |
2000 | }; | |
2001 | ||
2002 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
2003 | unsigned mode, int sync, void *arg) | |
2004 | { | |
d79154bb HD |
2005 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
2006 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
2007 | struct oom_wait_info *oom_wait_info; |
2008 | ||
2009 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 2010 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 2011 | |
dc98df5a | 2012 | /* |
d79154bb | 2013 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
dc98df5a KH |
2014 | * Then we can use css_is_ancestor without taking care of RCU. |
2015 | */ | |
c0ff4b85 R |
2016 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
2017 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 2018 | return 0; |
dc98df5a KH |
2019 | return autoremove_wake_function(wait, mode, sync, arg); |
2020 | } | |
2021 | ||
c0ff4b85 | 2022 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 2023 | { |
3812c8c8 | 2024 | atomic_inc(&memcg->oom_wakeups); |
c0ff4b85 R |
2025 | /* for filtering, pass "memcg" as argument. */ |
2026 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
2027 | } |
2028 | ||
c0ff4b85 | 2029 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 2030 | { |
c0ff4b85 R |
2031 | if (memcg && atomic_read(&memcg->under_oom)) |
2032 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
2033 | } |
2034 | ||
867578cb | 2035 | /* |
3812c8c8 | 2036 | * try to call OOM killer |
867578cb | 2037 | */ |
3812c8c8 | 2038 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 2039 | { |
fb2a6fc5 | 2040 | bool locked; |
3812c8c8 | 2041 | int wakeups; |
867578cb | 2042 | |
3812c8c8 JW |
2043 | if (!current->memcg_oom.may_oom) |
2044 | return; | |
2045 | ||
2046 | current->memcg_oom.in_memcg_oom = 1; | |
79dfdacc | 2047 | |
867578cb | 2048 | /* |
fb2a6fc5 JW |
2049 | * As with any blocking lock, a contender needs to start |
2050 | * listening for wakeups before attempting the trylock, | |
2051 | * otherwise it can miss the wakeup from the unlock and sleep | |
2052 | * indefinitely. This is just open-coded because our locking | |
2053 | * is so particular to memcg hierarchies. | |
867578cb | 2054 | */ |
3812c8c8 | 2055 | wakeups = atomic_read(&memcg->oom_wakeups); |
fb2a6fc5 JW |
2056 | mem_cgroup_mark_under_oom(memcg); |
2057 | ||
2058 | locked = mem_cgroup_oom_trylock(memcg); | |
2059 | ||
3c11ecf4 | 2060 | if (locked) |
c0ff4b85 | 2061 | mem_cgroup_oom_notify(memcg); |
867578cb | 2062 | |
fb2a6fc5 JW |
2063 | if (locked && !memcg->oom_kill_disable) { |
2064 | mem_cgroup_unmark_under_oom(memcg); | |
e845e199 | 2065 | mem_cgroup_out_of_memory(memcg, mask, order); |
3812c8c8 JW |
2066 | mem_cgroup_oom_unlock(memcg); |
2067 | /* | |
2068 | * There is no guarantee that an OOM-lock contender | |
2069 | * sees the wakeups triggered by the OOM kill | |
2070 | * uncharges. Wake any sleepers explicitely. | |
2071 | */ | |
2072 | memcg_oom_recover(memcg); | |
3c11ecf4 | 2073 | } else { |
3812c8c8 JW |
2074 | /* |
2075 | * A system call can just return -ENOMEM, but if this | |
2076 | * is a page fault and somebody else is handling the | |
2077 | * OOM already, we need to sleep on the OOM waitqueue | |
2078 | * for this memcg until the situation is resolved. | |
2079 | * Which can take some time because it might be | |
2080 | * handled by a userspace task. | |
2081 | * | |
2082 | * However, this is the charge context, which means | |
2083 | * that we may sit on a large call stack and hold | |
2084 | * various filesystem locks, the mmap_sem etc. and we | |
2085 | * don't want the OOM handler to deadlock on them | |
2086 | * while we sit here and wait. Store the current OOM | |
2087 | * context in the task_struct, then return -ENOMEM. | |
2088 | * At the end of the page fault handler, with the | |
2089 | * stack unwound, pagefault_out_of_memory() will check | |
2090 | * back with us by calling | |
2091 | * mem_cgroup_oom_synchronize(), possibly putting the | |
2092 | * task to sleep. | |
2093 | */ | |
2094 | current->memcg_oom.oom_locked = locked; | |
2095 | current->memcg_oom.wakeups = wakeups; | |
2096 | css_get(&memcg->css); | |
2097 | current->memcg_oom.wait_on_memcg = memcg; | |
867578cb | 2098 | } |
3812c8c8 JW |
2099 | } |
2100 | ||
2101 | /** | |
2102 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
2103 | * | |
2104 | * This has to be called at the end of a page fault if the the memcg | |
2105 | * OOM handler was enabled and the fault is returning %VM_FAULT_OOM. | |
2106 | * | |
2107 | * Memcg supports userspace OOM handling, so failed allocations must | |
2108 | * sleep on a waitqueue until the userspace task resolves the | |
2109 | * situation. Sleeping directly in the charge context with all kinds | |
2110 | * of locks held is not a good idea, instead we remember an OOM state | |
2111 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
2112 | * the end of the page fault to put the task to sleep and clean up the | |
2113 | * OOM state. | |
2114 | * | |
2115 | * Returns %true if an ongoing memcg OOM situation was detected and | |
2116 | * finalized, %false otherwise. | |
2117 | */ | |
2118 | bool mem_cgroup_oom_synchronize(void) | |
2119 | { | |
2120 | struct oom_wait_info owait; | |
2121 | struct mem_cgroup *memcg; | |
2122 | ||
2123 | /* OOM is global, do not handle */ | |
2124 | if (!current->memcg_oom.in_memcg_oom) | |
2125 | return false; | |
2126 | ||
2127 | /* | |
2128 | * We invoked the OOM killer but there is a chance that a kill | |
2129 | * did not free up any charges. Everybody else might already | |
2130 | * be sleeping, so restart the fault and keep the rampage | |
2131 | * going until some charges are released. | |
2132 | */ | |
2133 | memcg = current->memcg_oom.wait_on_memcg; | |
2134 | if (!memcg) | |
2135 | goto out; | |
2136 | ||
2137 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) | |
2138 | goto out_memcg; | |
2139 | ||
2140 | owait.memcg = memcg; | |
2141 | owait.wait.flags = 0; | |
2142 | owait.wait.func = memcg_oom_wake_function; | |
2143 | owait.wait.private = current; | |
2144 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 2145 | |
3812c8c8 JW |
2146 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
2147 | /* Only sleep if we didn't miss any wakeups since OOM */ | |
2148 | if (atomic_read(&memcg->oom_wakeups) == current->memcg_oom.wakeups) | |
2149 | schedule(); | |
2150 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2151 | out_memcg: | |
2152 | mem_cgroup_unmark_under_oom(memcg); | |
2153 | if (current->memcg_oom.oom_locked) { | |
fb2a6fc5 JW |
2154 | mem_cgroup_oom_unlock(memcg); |
2155 | /* | |
2156 | * There is no guarantee that an OOM-lock contender | |
2157 | * sees the wakeups triggered by the OOM kill | |
2158 | * uncharges. Wake any sleepers explicitely. | |
2159 | */ | |
2160 | memcg_oom_recover(memcg); | |
2161 | } | |
3812c8c8 JW |
2162 | css_put(&memcg->css); |
2163 | current->memcg_oom.wait_on_memcg = NULL; | |
2164 | out: | |
2165 | current->memcg_oom.in_memcg_oom = 0; | |
867578cb | 2166 | return true; |
0b7f569e KH |
2167 | } |
2168 | ||
d69b042f BS |
2169 | /* |
2170 | * Currently used to update mapped file statistics, but the routine can be | |
2171 | * generalized to update other statistics as well. | |
32047e2a KH |
2172 | * |
2173 | * Notes: Race condition | |
2174 | * | |
2175 | * We usually use page_cgroup_lock() for accessing page_cgroup member but | |
2176 | * it tends to be costly. But considering some conditions, we doesn't need | |
2177 | * to do so _always_. | |
2178 | * | |
2179 | * Considering "charge", lock_page_cgroup() is not required because all | |
2180 | * file-stat operations happen after a page is attached to radix-tree. There | |
2181 | * are no race with "charge". | |
2182 | * | |
2183 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
2184 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
2185 | * if there are race with "uncharge". Statistics itself is properly handled | |
2186 | * by flags. | |
2187 | * | |
2188 | * Considering "move", this is an only case we see a race. To make the race | |
619d094b KH |
2189 | * small, we check mm->moving_account and detect there are possibility of race |
2190 | * If there is, we take a lock. | |
d69b042f | 2191 | */ |
26174efd | 2192 | |
89c06bd5 KH |
2193 | void __mem_cgroup_begin_update_page_stat(struct page *page, |
2194 | bool *locked, unsigned long *flags) | |
2195 | { | |
2196 | struct mem_cgroup *memcg; | |
2197 | struct page_cgroup *pc; | |
2198 | ||
2199 | pc = lookup_page_cgroup(page); | |
2200 | again: | |
2201 | memcg = pc->mem_cgroup; | |
2202 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
2203 | return; | |
2204 | /* | |
2205 | * If this memory cgroup is not under account moving, we don't | |
da92c47d | 2206 | * need to take move_lock_mem_cgroup(). Because we already hold |
89c06bd5 | 2207 | * rcu_read_lock(), any calls to move_account will be delayed until |
13fd1dd9 | 2208 | * rcu_read_unlock() if mem_cgroup_stolen() == true. |
89c06bd5 | 2209 | */ |
13fd1dd9 | 2210 | if (!mem_cgroup_stolen(memcg)) |
89c06bd5 KH |
2211 | return; |
2212 | ||
2213 | move_lock_mem_cgroup(memcg, flags); | |
2214 | if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) { | |
2215 | move_unlock_mem_cgroup(memcg, flags); | |
2216 | goto again; | |
2217 | } | |
2218 | *locked = true; | |
2219 | } | |
2220 | ||
2221 | void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags) | |
2222 | { | |
2223 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
2224 | ||
2225 | /* | |
2226 | * It's guaranteed that pc->mem_cgroup never changes while | |
2227 | * lock is held because a routine modifies pc->mem_cgroup | |
da92c47d | 2228 | * should take move_lock_mem_cgroup(). |
89c06bd5 KH |
2229 | */ |
2230 | move_unlock_mem_cgroup(pc->mem_cgroup, flags); | |
2231 | } | |
2232 | ||
2a7106f2 GT |
2233 | void mem_cgroup_update_page_stat(struct page *page, |
2234 | enum mem_cgroup_page_stat_item idx, int val) | |
d69b042f | 2235 | { |
c0ff4b85 | 2236 | struct mem_cgroup *memcg; |
32047e2a | 2237 | struct page_cgroup *pc = lookup_page_cgroup(page); |
dbd4ea78 | 2238 | unsigned long uninitialized_var(flags); |
d69b042f | 2239 | |
cfa44946 | 2240 | if (mem_cgroup_disabled()) |
d69b042f | 2241 | return; |
89c06bd5 | 2242 | |
c0ff4b85 R |
2243 | memcg = pc->mem_cgroup; |
2244 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
89c06bd5 | 2245 | return; |
26174efd | 2246 | |
26174efd | 2247 | switch (idx) { |
2a7106f2 | 2248 | case MEMCG_NR_FILE_MAPPED: |
2a7106f2 | 2249 | idx = MEM_CGROUP_STAT_FILE_MAPPED; |
26174efd KH |
2250 | break; |
2251 | default: | |
2252 | BUG(); | |
8725d541 | 2253 | } |
d69b042f | 2254 | |
c0ff4b85 | 2255 | this_cpu_add(memcg->stat->count[idx], val); |
d69b042f | 2256 | } |
26174efd | 2257 | |
cdec2e42 KH |
2258 | /* |
2259 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2260 | * TODO: maybe necessary to use big numbers in big irons. | |
2261 | */ | |
7ec99d62 | 2262 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2263 | struct memcg_stock_pcp { |
2264 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2265 | unsigned int nr_pages; |
cdec2e42 | 2266 | struct work_struct work; |
26fe6168 | 2267 | unsigned long flags; |
a0db00fc | 2268 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2269 | }; |
2270 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2271 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2272 | |
a0956d54 SS |
2273 | /** |
2274 | * consume_stock: Try to consume stocked charge on this cpu. | |
2275 | * @memcg: memcg to consume from. | |
2276 | * @nr_pages: how many pages to charge. | |
2277 | * | |
2278 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2279 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2280 | * service an allocation will refill the stock. | |
2281 | * | |
2282 | * returns true if successful, false otherwise. | |
cdec2e42 | 2283 | */ |
a0956d54 | 2284 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2285 | { |
2286 | struct memcg_stock_pcp *stock; | |
2287 | bool ret = true; | |
2288 | ||
a0956d54 SS |
2289 | if (nr_pages > CHARGE_BATCH) |
2290 | return false; | |
2291 | ||
cdec2e42 | 2292 | stock = &get_cpu_var(memcg_stock); |
a0956d54 SS |
2293 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) |
2294 | stock->nr_pages -= nr_pages; | |
cdec2e42 KH |
2295 | else /* need to call res_counter_charge */ |
2296 | ret = false; | |
2297 | put_cpu_var(memcg_stock); | |
2298 | return ret; | |
2299 | } | |
2300 | ||
2301 | /* | |
2302 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
2303 | */ | |
2304 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2305 | { | |
2306 | struct mem_cgroup *old = stock->cached; | |
2307 | ||
11c9ea4e JW |
2308 | if (stock->nr_pages) { |
2309 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
2310 | ||
2311 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 2312 | if (do_swap_account) |
11c9ea4e JW |
2313 | res_counter_uncharge(&old->memsw, bytes); |
2314 | stock->nr_pages = 0; | |
cdec2e42 KH |
2315 | } |
2316 | stock->cached = NULL; | |
cdec2e42 KH |
2317 | } |
2318 | ||
2319 | /* | |
2320 | * This must be called under preempt disabled or must be called by | |
2321 | * a thread which is pinned to local cpu. | |
2322 | */ | |
2323 | static void drain_local_stock(struct work_struct *dummy) | |
2324 | { | |
2325 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
2326 | drain_stock(stock); | |
26fe6168 | 2327 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2328 | } |
2329 | ||
e4777496 MH |
2330 | static void __init memcg_stock_init(void) |
2331 | { | |
2332 | int cpu; | |
2333 | ||
2334 | for_each_possible_cpu(cpu) { | |
2335 | struct memcg_stock_pcp *stock = | |
2336 | &per_cpu(memcg_stock, cpu); | |
2337 | INIT_WORK(&stock->work, drain_local_stock); | |
2338 | } | |
2339 | } | |
2340 | ||
cdec2e42 KH |
2341 | /* |
2342 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 2343 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2344 | */ |
c0ff4b85 | 2345 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2346 | { |
2347 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2348 | ||
c0ff4b85 | 2349 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2350 | drain_stock(stock); |
c0ff4b85 | 2351 | stock->cached = memcg; |
cdec2e42 | 2352 | } |
11c9ea4e | 2353 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2354 | put_cpu_var(memcg_stock); |
2355 | } | |
2356 | ||
2357 | /* | |
c0ff4b85 | 2358 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
d38144b7 MH |
2359 | * of the hierarchy under it. sync flag says whether we should block |
2360 | * until the work is done. | |
cdec2e42 | 2361 | */ |
c0ff4b85 | 2362 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
cdec2e42 | 2363 | { |
26fe6168 | 2364 | int cpu, curcpu; |
d38144b7 | 2365 | |
cdec2e42 | 2366 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2367 | get_online_cpus(); |
5af12d0e | 2368 | curcpu = get_cpu(); |
cdec2e42 KH |
2369 | for_each_online_cpu(cpu) { |
2370 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2371 | struct mem_cgroup *memcg; |
26fe6168 | 2372 | |
c0ff4b85 R |
2373 | memcg = stock->cached; |
2374 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2375 | continue; |
c0ff4b85 | 2376 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2377 | continue; |
d1a05b69 MH |
2378 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2379 | if (cpu == curcpu) | |
2380 | drain_local_stock(&stock->work); | |
2381 | else | |
2382 | schedule_work_on(cpu, &stock->work); | |
2383 | } | |
cdec2e42 | 2384 | } |
5af12d0e | 2385 | put_cpu(); |
d38144b7 MH |
2386 | |
2387 | if (!sync) | |
2388 | goto out; | |
2389 | ||
2390 | for_each_online_cpu(cpu) { | |
2391 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2392 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2393 | flush_work(&stock->work); |
2394 | } | |
2395 | out: | |
f894ffa8 | 2396 | put_online_cpus(); |
d38144b7 MH |
2397 | } |
2398 | ||
2399 | /* | |
2400 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2401 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
2402 | * expects some charges will be back to res_counter later but cannot wait for | |
2403 | * it. | |
2404 | */ | |
c0ff4b85 | 2405 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
d38144b7 | 2406 | { |
9f50fad6 MH |
2407 | /* |
2408 | * If someone calls draining, avoid adding more kworker runs. | |
2409 | */ | |
2410 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2411 | return; | |
c0ff4b85 | 2412 | drain_all_stock(root_memcg, false); |
9f50fad6 | 2413 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2414 | } |
2415 | ||
2416 | /* This is a synchronous drain interface. */ | |
c0ff4b85 | 2417 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
cdec2e42 KH |
2418 | { |
2419 | /* called when force_empty is called */ | |
9f50fad6 | 2420 | mutex_lock(&percpu_charge_mutex); |
c0ff4b85 | 2421 | drain_all_stock(root_memcg, true); |
9f50fad6 | 2422 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2423 | } |
2424 | ||
711d3d2c KH |
2425 | /* |
2426 | * This function drains percpu counter value from DEAD cpu and | |
2427 | * move it to local cpu. Note that this function can be preempted. | |
2428 | */ | |
c0ff4b85 | 2429 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2430 | { |
2431 | int i; | |
2432 | ||
c0ff4b85 | 2433 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2434 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2435 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2436 | |
c0ff4b85 R |
2437 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2438 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2439 | } |
e9f8974f | 2440 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2441 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2442 | |
c0ff4b85 R |
2443 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2444 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2445 | } |
c0ff4b85 | 2446 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2447 | } |
2448 | ||
0db0628d | 2449 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
2450 | unsigned long action, |
2451 | void *hcpu) | |
2452 | { | |
2453 | int cpu = (unsigned long)hcpu; | |
2454 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2455 | struct mem_cgroup *iter; |
cdec2e42 | 2456 | |
619d094b | 2457 | if (action == CPU_ONLINE) |
1489ebad | 2458 | return NOTIFY_OK; |
1489ebad | 2459 | |
d833049b | 2460 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2461 | return NOTIFY_OK; |
711d3d2c | 2462 | |
9f3a0d09 | 2463 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2464 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2465 | ||
cdec2e42 KH |
2466 | stock = &per_cpu(memcg_stock, cpu); |
2467 | drain_stock(stock); | |
2468 | return NOTIFY_OK; | |
2469 | } | |
2470 | ||
4b534334 KH |
2471 | |
2472 | /* See __mem_cgroup_try_charge() for details */ | |
2473 | enum { | |
2474 | CHARGE_OK, /* success */ | |
2475 | CHARGE_RETRY, /* need to retry but retry is not bad */ | |
2476 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ | |
2477 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ | |
4b534334 KH |
2478 | }; |
2479 | ||
c0ff4b85 | 2480 | static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
4c9c5359 | 2481 | unsigned int nr_pages, unsigned int min_pages, |
3812c8c8 | 2482 | bool invoke_oom) |
4b534334 | 2483 | { |
7ec99d62 | 2484 | unsigned long csize = nr_pages * PAGE_SIZE; |
4b534334 KH |
2485 | struct mem_cgroup *mem_over_limit; |
2486 | struct res_counter *fail_res; | |
2487 | unsigned long flags = 0; | |
2488 | int ret; | |
2489 | ||
c0ff4b85 | 2490 | ret = res_counter_charge(&memcg->res, csize, &fail_res); |
4b534334 KH |
2491 | |
2492 | if (likely(!ret)) { | |
2493 | if (!do_swap_account) | |
2494 | return CHARGE_OK; | |
c0ff4b85 | 2495 | ret = res_counter_charge(&memcg->memsw, csize, &fail_res); |
4b534334 KH |
2496 | if (likely(!ret)) |
2497 | return CHARGE_OK; | |
2498 | ||
c0ff4b85 | 2499 | res_counter_uncharge(&memcg->res, csize); |
4b534334 KH |
2500 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
2501 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
2502 | } else | |
2503 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
9221edb7 | 2504 | /* |
9221edb7 JW |
2505 | * Never reclaim on behalf of optional batching, retry with a |
2506 | * single page instead. | |
2507 | */ | |
4c9c5359 | 2508 | if (nr_pages > min_pages) |
4b534334 KH |
2509 | return CHARGE_RETRY; |
2510 | ||
2511 | if (!(gfp_mask & __GFP_WAIT)) | |
2512 | return CHARGE_WOULDBLOCK; | |
2513 | ||
4c9c5359 SS |
2514 | if (gfp_mask & __GFP_NORETRY) |
2515 | return CHARGE_NOMEM; | |
2516 | ||
5660048c | 2517 | ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags); |
7ec99d62 | 2518 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
19942822 | 2519 | return CHARGE_RETRY; |
4b534334 | 2520 | /* |
19942822 JW |
2521 | * Even though the limit is exceeded at this point, reclaim |
2522 | * may have been able to free some pages. Retry the charge | |
2523 | * before killing the task. | |
2524 | * | |
2525 | * Only for regular pages, though: huge pages are rather | |
2526 | * unlikely to succeed so close to the limit, and we fall back | |
2527 | * to regular pages anyway in case of failure. | |
4b534334 | 2528 | */ |
4c9c5359 | 2529 | if (nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER) && ret) |
4b534334 KH |
2530 | return CHARGE_RETRY; |
2531 | ||
2532 | /* | |
2533 | * At task move, charge accounts can be doubly counted. So, it's | |
2534 | * better to wait until the end of task_move if something is going on. | |
2535 | */ | |
2536 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2537 | return CHARGE_RETRY; | |
2538 | ||
3812c8c8 JW |
2539 | if (invoke_oom) |
2540 | mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(csize)); | |
4b534334 | 2541 | |
3812c8c8 | 2542 | return CHARGE_NOMEM; |
4b534334 KH |
2543 | } |
2544 | ||
f817ed48 | 2545 | /* |
38c5d72f KH |
2546 | * __mem_cgroup_try_charge() does |
2547 | * 1. detect memcg to be charged against from passed *mm and *ptr, | |
2548 | * 2. update res_counter | |
2549 | * 3. call memory reclaim if necessary. | |
2550 | * | |
2551 | * In some special case, if the task is fatal, fatal_signal_pending() or | |
2552 | * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup | |
2553 | * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon | |
2554 | * as possible without any hazards. 2: all pages should have a valid | |
2555 | * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg | |
2556 | * pointer, that is treated as a charge to root_mem_cgroup. | |
2557 | * | |
2558 | * So __mem_cgroup_try_charge() will return | |
2559 | * 0 ... on success, filling *ptr with a valid memcg pointer. | |
2560 | * -ENOMEM ... charge failure because of resource limits. | |
2561 | * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup. | |
2562 | * | |
2563 | * Unlike the exported interface, an "oom" parameter is added. if oom==true, | |
2564 | * the oom-killer can be invoked. | |
8a9f3ccd | 2565 | */ |
f817ed48 | 2566 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
ec168510 | 2567 | gfp_t gfp_mask, |
7ec99d62 | 2568 | unsigned int nr_pages, |
c0ff4b85 | 2569 | struct mem_cgroup **ptr, |
7ec99d62 | 2570 | bool oom) |
8a9f3ccd | 2571 | { |
7ec99d62 | 2572 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
4b534334 | 2573 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
c0ff4b85 | 2574 | struct mem_cgroup *memcg = NULL; |
4b534334 | 2575 | int ret; |
a636b327 | 2576 | |
867578cb KH |
2577 | /* |
2578 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
2579 | * in system level. So, allow to go ahead dying process in addition to | |
2580 | * MEMDIE process. | |
2581 | */ | |
2582 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
2583 | || fatal_signal_pending(current))) | |
2584 | goto bypass; | |
a636b327 | 2585 | |
8a9f3ccd | 2586 | /* |
3be91277 HD |
2587 | * We always charge the cgroup the mm_struct belongs to. |
2588 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd | 2589 | * thread group leader migrates. It's possible that mm is not |
24467cac | 2590 | * set, if so charge the root memcg (happens for pagecache usage). |
8a9f3ccd | 2591 | */ |
c0ff4b85 | 2592 | if (!*ptr && !mm) |
38c5d72f | 2593 | *ptr = root_mem_cgroup; |
f75ca962 | 2594 | again: |
c0ff4b85 R |
2595 | if (*ptr) { /* css should be a valid one */ |
2596 | memcg = *ptr; | |
c0ff4b85 | 2597 | if (mem_cgroup_is_root(memcg)) |
f75ca962 | 2598 | goto done; |
a0956d54 | 2599 | if (consume_stock(memcg, nr_pages)) |
f75ca962 | 2600 | goto done; |
c0ff4b85 | 2601 | css_get(&memcg->css); |
4b534334 | 2602 | } else { |
f75ca962 | 2603 | struct task_struct *p; |
54595fe2 | 2604 | |
f75ca962 KH |
2605 | rcu_read_lock(); |
2606 | p = rcu_dereference(mm->owner); | |
f75ca962 | 2607 | /* |
ebb76ce1 | 2608 | * Because we don't have task_lock(), "p" can exit. |
c0ff4b85 | 2609 | * In that case, "memcg" can point to root or p can be NULL with |
ebb76ce1 KH |
2610 | * race with swapoff. Then, we have small risk of mis-accouning. |
2611 | * But such kind of mis-account by race always happens because | |
2612 | * we don't have cgroup_mutex(). It's overkill and we allo that | |
2613 | * small race, here. | |
2614 | * (*) swapoff at el will charge against mm-struct not against | |
2615 | * task-struct. So, mm->owner can be NULL. | |
f75ca962 | 2616 | */ |
c0ff4b85 | 2617 | memcg = mem_cgroup_from_task(p); |
38c5d72f KH |
2618 | if (!memcg) |
2619 | memcg = root_mem_cgroup; | |
2620 | if (mem_cgroup_is_root(memcg)) { | |
f75ca962 KH |
2621 | rcu_read_unlock(); |
2622 | goto done; | |
2623 | } | |
a0956d54 | 2624 | if (consume_stock(memcg, nr_pages)) { |
f75ca962 KH |
2625 | /* |
2626 | * It seems dagerous to access memcg without css_get(). | |
2627 | * But considering how consume_stok works, it's not | |
2628 | * necessary. If consume_stock success, some charges | |
2629 | * from this memcg are cached on this cpu. So, we | |
2630 | * don't need to call css_get()/css_tryget() before | |
2631 | * calling consume_stock(). | |
2632 | */ | |
2633 | rcu_read_unlock(); | |
2634 | goto done; | |
2635 | } | |
2636 | /* after here, we may be blocked. we need to get refcnt */ | |
c0ff4b85 | 2637 | if (!css_tryget(&memcg->css)) { |
f75ca962 KH |
2638 | rcu_read_unlock(); |
2639 | goto again; | |
2640 | } | |
2641 | rcu_read_unlock(); | |
2642 | } | |
8a9f3ccd | 2643 | |
4b534334 | 2644 | do { |
3812c8c8 | 2645 | bool invoke_oom = oom && !nr_oom_retries; |
7a81b88c | 2646 | |
4b534334 | 2647 | /* If killed, bypass charge */ |
f75ca962 | 2648 | if (fatal_signal_pending(current)) { |
c0ff4b85 | 2649 | css_put(&memcg->css); |
4b534334 | 2650 | goto bypass; |
f75ca962 | 2651 | } |
6d61ef40 | 2652 | |
3812c8c8 JW |
2653 | ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, |
2654 | nr_pages, invoke_oom); | |
4b534334 KH |
2655 | switch (ret) { |
2656 | case CHARGE_OK: | |
2657 | break; | |
2658 | case CHARGE_RETRY: /* not in OOM situation but retry */ | |
7ec99d62 | 2659 | batch = nr_pages; |
c0ff4b85 R |
2660 | css_put(&memcg->css); |
2661 | memcg = NULL; | |
f75ca962 | 2662 | goto again; |
4b534334 | 2663 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
c0ff4b85 | 2664 | css_put(&memcg->css); |
4b534334 KH |
2665 | goto nomem; |
2666 | case CHARGE_NOMEM: /* OOM routine works */ | |
3812c8c8 | 2667 | if (!oom || invoke_oom) { |
c0ff4b85 | 2668 | css_put(&memcg->css); |
867578cb | 2669 | goto nomem; |
f75ca962 | 2670 | } |
4b534334 KH |
2671 | nr_oom_retries--; |
2672 | break; | |
66e1707b | 2673 | } |
4b534334 KH |
2674 | } while (ret != CHARGE_OK); |
2675 | ||
7ec99d62 | 2676 | if (batch > nr_pages) |
c0ff4b85 R |
2677 | refill_stock(memcg, batch - nr_pages); |
2678 | css_put(&memcg->css); | |
0c3e73e8 | 2679 | done: |
c0ff4b85 | 2680 | *ptr = memcg; |
7a81b88c KH |
2681 | return 0; |
2682 | nomem: | |
c0ff4b85 | 2683 | *ptr = NULL; |
7a81b88c | 2684 | return -ENOMEM; |
867578cb | 2685 | bypass: |
38c5d72f KH |
2686 | *ptr = root_mem_cgroup; |
2687 | return -EINTR; | |
7a81b88c | 2688 | } |
8a9f3ccd | 2689 | |
a3032a2c DN |
2690 | /* |
2691 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2692 | * This function is for that and do uncharge, put css's refcnt. | |
2693 | * gotten by try_charge(). | |
2694 | */ | |
c0ff4b85 | 2695 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg, |
e7018b8d | 2696 | unsigned int nr_pages) |
a3032a2c | 2697 | { |
c0ff4b85 | 2698 | if (!mem_cgroup_is_root(memcg)) { |
e7018b8d JW |
2699 | unsigned long bytes = nr_pages * PAGE_SIZE; |
2700 | ||
c0ff4b85 | 2701 | res_counter_uncharge(&memcg->res, bytes); |
a3032a2c | 2702 | if (do_swap_account) |
c0ff4b85 | 2703 | res_counter_uncharge(&memcg->memsw, bytes); |
a3032a2c | 2704 | } |
854ffa8d DN |
2705 | } |
2706 | ||
d01dd17f KH |
2707 | /* |
2708 | * Cancel chrages in this cgroup....doesn't propagate to parent cgroup. | |
2709 | * This is useful when moving usage to parent cgroup. | |
2710 | */ | |
2711 | static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg, | |
2712 | unsigned int nr_pages) | |
2713 | { | |
2714 | unsigned long bytes = nr_pages * PAGE_SIZE; | |
2715 | ||
2716 | if (mem_cgroup_is_root(memcg)) | |
2717 | return; | |
2718 | ||
2719 | res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes); | |
2720 | if (do_swap_account) | |
2721 | res_counter_uncharge_until(&memcg->memsw, | |
2722 | memcg->memsw.parent, bytes); | |
2723 | } | |
2724 | ||
a3b2d692 KH |
2725 | /* |
2726 | * A helper function to get mem_cgroup from ID. must be called under | |
e9316080 TH |
2727 | * rcu_read_lock(). The caller is responsible for calling css_tryget if |
2728 | * the mem_cgroup is used for charging. (dropping refcnt from swap can be | |
2729 | * called against removed memcg.) | |
a3b2d692 KH |
2730 | */ |
2731 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2732 | { | |
2733 | struct cgroup_subsys_state *css; | |
2734 | ||
2735 | /* ID 0 is unused ID */ | |
2736 | if (!id) | |
2737 | return NULL; | |
2738 | css = css_lookup(&mem_cgroup_subsys, id); | |
2739 | if (!css) | |
2740 | return NULL; | |
b2145145 | 2741 | return mem_cgroup_from_css(css); |
a3b2d692 KH |
2742 | } |
2743 | ||
e42d9d5d | 2744 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2745 | { |
c0ff4b85 | 2746 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2747 | struct page_cgroup *pc; |
a3b2d692 | 2748 | unsigned short id; |
b5a84319 KH |
2749 | swp_entry_t ent; |
2750 | ||
3c776e64 DN |
2751 | VM_BUG_ON(!PageLocked(page)); |
2752 | ||
3c776e64 | 2753 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2754 | lock_page_cgroup(pc); |
a3b2d692 | 2755 | if (PageCgroupUsed(pc)) { |
c0ff4b85 R |
2756 | memcg = pc->mem_cgroup; |
2757 | if (memcg && !css_tryget(&memcg->css)) | |
2758 | memcg = NULL; | |
e42d9d5d | 2759 | } else if (PageSwapCache(page)) { |
3c776e64 | 2760 | ent.val = page_private(page); |
9fb4b7cc | 2761 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2762 | rcu_read_lock(); |
c0ff4b85 R |
2763 | memcg = mem_cgroup_lookup(id); |
2764 | if (memcg && !css_tryget(&memcg->css)) | |
2765 | memcg = NULL; | |
a3b2d692 | 2766 | rcu_read_unlock(); |
3c776e64 | 2767 | } |
c0bd3f63 | 2768 | unlock_page_cgroup(pc); |
c0ff4b85 | 2769 | return memcg; |
b5a84319 KH |
2770 | } |
2771 | ||
c0ff4b85 | 2772 | static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg, |
5564e88b | 2773 | struct page *page, |
7ec99d62 | 2774 | unsigned int nr_pages, |
9ce70c02 HD |
2775 | enum charge_type ctype, |
2776 | bool lrucare) | |
7a81b88c | 2777 | { |
ce587e65 | 2778 | struct page_cgroup *pc = lookup_page_cgroup(page); |
9ce70c02 | 2779 | struct zone *uninitialized_var(zone); |
fa9add64 | 2780 | struct lruvec *lruvec; |
9ce70c02 | 2781 | bool was_on_lru = false; |
b2402857 | 2782 | bool anon; |
9ce70c02 | 2783 | |
ca3e0214 | 2784 | lock_page_cgroup(pc); |
90deb788 | 2785 | VM_BUG_ON(PageCgroupUsed(pc)); |
ca3e0214 KH |
2786 | /* |
2787 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2788 | * accessed by any other context at this point. | |
2789 | */ | |
9ce70c02 HD |
2790 | |
2791 | /* | |
2792 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2793 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2794 | */ | |
2795 | if (lrucare) { | |
2796 | zone = page_zone(page); | |
2797 | spin_lock_irq(&zone->lru_lock); | |
2798 | if (PageLRU(page)) { | |
fa9add64 | 2799 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 | 2800 | ClearPageLRU(page); |
fa9add64 | 2801 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2802 | was_on_lru = true; |
2803 | } | |
2804 | } | |
2805 | ||
c0ff4b85 | 2806 | pc->mem_cgroup = memcg; |
261fb61a KH |
2807 | /* |
2808 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
2809 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
2810 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
2811 | * before USED bit, we need memory barrier here. | |
2812 | * See mem_cgroup_add_lru_list(), etc. | |
f894ffa8 | 2813 | */ |
08e552c6 | 2814 | smp_wmb(); |
b2402857 | 2815 | SetPageCgroupUsed(pc); |
3be91277 | 2816 | |
9ce70c02 HD |
2817 | if (lrucare) { |
2818 | if (was_on_lru) { | |
fa9add64 | 2819 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 HD |
2820 | VM_BUG_ON(PageLRU(page)); |
2821 | SetPageLRU(page); | |
fa9add64 | 2822 | add_page_to_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2823 | } |
2824 | spin_unlock_irq(&zone->lru_lock); | |
2825 | } | |
2826 | ||
41326c17 | 2827 | if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON) |
b2402857 KH |
2828 | anon = true; |
2829 | else | |
2830 | anon = false; | |
2831 | ||
b070e65c | 2832 | mem_cgroup_charge_statistics(memcg, page, anon, nr_pages); |
52d4b9ac | 2833 | unlock_page_cgroup(pc); |
9ce70c02 | 2834 | |
430e4863 | 2835 | /* |
e883110a | 2836 | * "charge_statistics" updated event counter. |
430e4863 | 2837 | */ |
c0ff4b85 | 2838 | memcg_check_events(memcg, page); |
7a81b88c | 2839 | } |
66e1707b | 2840 | |
7cf27982 GC |
2841 | static DEFINE_MUTEX(set_limit_mutex); |
2842 | ||
7ae1e1d0 GC |
2843 | #ifdef CONFIG_MEMCG_KMEM |
2844 | static inline bool memcg_can_account_kmem(struct mem_cgroup *memcg) | |
2845 | { | |
2846 | return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg) && | |
2847 | (memcg->kmem_account_flags & KMEM_ACCOUNTED_MASK); | |
2848 | } | |
2849 | ||
1f458cbf GC |
2850 | /* |
2851 | * This is a bit cumbersome, but it is rarely used and avoids a backpointer | |
2852 | * in the memcg_cache_params struct. | |
2853 | */ | |
2854 | static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p) | |
2855 | { | |
2856 | struct kmem_cache *cachep; | |
2857 | ||
2858 | VM_BUG_ON(p->is_root_cache); | |
2859 | cachep = p->root_cache; | |
2860 | return cachep->memcg_params->memcg_caches[memcg_cache_id(p->memcg)]; | |
2861 | } | |
2862 | ||
749c5415 | 2863 | #ifdef CONFIG_SLABINFO |
182446d0 TH |
2864 | static int mem_cgroup_slabinfo_read(struct cgroup_subsys_state *css, |
2865 | struct cftype *cft, struct seq_file *m) | |
749c5415 | 2866 | { |
182446d0 | 2867 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
749c5415 GC |
2868 | struct memcg_cache_params *params; |
2869 | ||
2870 | if (!memcg_can_account_kmem(memcg)) | |
2871 | return -EIO; | |
2872 | ||
2873 | print_slabinfo_header(m); | |
2874 | ||
2875 | mutex_lock(&memcg->slab_caches_mutex); | |
2876 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) | |
2877 | cache_show(memcg_params_to_cache(params), m); | |
2878 | mutex_unlock(&memcg->slab_caches_mutex); | |
2879 | ||
2880 | return 0; | |
2881 | } | |
2882 | #endif | |
2883 | ||
7ae1e1d0 GC |
2884 | static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, u64 size) |
2885 | { | |
2886 | struct res_counter *fail_res; | |
2887 | struct mem_cgroup *_memcg; | |
2888 | int ret = 0; | |
2889 | bool may_oom; | |
2890 | ||
2891 | ret = res_counter_charge(&memcg->kmem, size, &fail_res); | |
2892 | if (ret) | |
2893 | return ret; | |
2894 | ||
2895 | /* | |
2896 | * Conditions under which we can wait for the oom_killer. Those are | |
2897 | * the same conditions tested by the core page allocator | |
2898 | */ | |
2899 | may_oom = (gfp & __GFP_FS) && !(gfp & __GFP_NORETRY); | |
2900 | ||
2901 | _memcg = memcg; | |
2902 | ret = __mem_cgroup_try_charge(NULL, gfp, size >> PAGE_SHIFT, | |
2903 | &_memcg, may_oom); | |
2904 | ||
2905 | if (ret == -EINTR) { | |
2906 | /* | |
2907 | * __mem_cgroup_try_charge() chosed to bypass to root due to | |
2908 | * OOM kill or fatal signal. Since our only options are to | |
2909 | * either fail the allocation or charge it to this cgroup, do | |
2910 | * it as a temporary condition. But we can't fail. From a | |
2911 | * kmem/slab perspective, the cache has already been selected, | |
2912 | * by mem_cgroup_kmem_get_cache(), so it is too late to change | |
2913 | * our minds. | |
2914 | * | |
2915 | * This condition will only trigger if the task entered | |
2916 | * memcg_charge_kmem in a sane state, but was OOM-killed during | |
2917 | * __mem_cgroup_try_charge() above. Tasks that were already | |
2918 | * dying when the allocation triggers should have been already | |
2919 | * directed to the root cgroup in memcontrol.h | |
2920 | */ | |
2921 | res_counter_charge_nofail(&memcg->res, size, &fail_res); | |
2922 | if (do_swap_account) | |
2923 | res_counter_charge_nofail(&memcg->memsw, size, | |
2924 | &fail_res); | |
2925 | ret = 0; | |
2926 | } else if (ret) | |
2927 | res_counter_uncharge(&memcg->kmem, size); | |
2928 | ||
2929 | return ret; | |
2930 | } | |
2931 | ||
2932 | static void memcg_uncharge_kmem(struct mem_cgroup *memcg, u64 size) | |
2933 | { | |
7ae1e1d0 GC |
2934 | res_counter_uncharge(&memcg->res, size); |
2935 | if (do_swap_account) | |
2936 | res_counter_uncharge(&memcg->memsw, size); | |
7de37682 GC |
2937 | |
2938 | /* Not down to 0 */ | |
2939 | if (res_counter_uncharge(&memcg->kmem, size)) | |
2940 | return; | |
2941 | ||
10d5ebf4 LZ |
2942 | /* |
2943 | * Releases a reference taken in kmem_cgroup_css_offline in case | |
2944 | * this last uncharge is racing with the offlining code or it is | |
2945 | * outliving the memcg existence. | |
2946 | * | |
2947 | * The memory barrier imposed by test&clear is paired with the | |
2948 | * explicit one in memcg_kmem_mark_dead(). | |
2949 | */ | |
7de37682 | 2950 | if (memcg_kmem_test_and_clear_dead(memcg)) |
10d5ebf4 | 2951 | css_put(&memcg->css); |
7ae1e1d0 GC |
2952 | } |
2953 | ||
2633d7a0 GC |
2954 | void memcg_cache_list_add(struct mem_cgroup *memcg, struct kmem_cache *cachep) |
2955 | { | |
2956 | if (!memcg) | |
2957 | return; | |
2958 | ||
2959 | mutex_lock(&memcg->slab_caches_mutex); | |
2960 | list_add(&cachep->memcg_params->list, &memcg->memcg_slab_caches); | |
2961 | mutex_unlock(&memcg->slab_caches_mutex); | |
2962 | } | |
2963 | ||
2964 | /* | |
2965 | * helper for acessing a memcg's index. It will be used as an index in the | |
2966 | * child cache array in kmem_cache, and also to derive its name. This function | |
2967 | * will return -1 when this is not a kmem-limited memcg. | |
2968 | */ | |
2969 | int memcg_cache_id(struct mem_cgroup *memcg) | |
2970 | { | |
2971 | return memcg ? memcg->kmemcg_id : -1; | |
2972 | } | |
2973 | ||
55007d84 GC |
2974 | /* |
2975 | * This ends up being protected by the set_limit mutex, during normal | |
2976 | * operation, because that is its main call site. | |
2977 | * | |
2978 | * But when we create a new cache, we can call this as well if its parent | |
2979 | * is kmem-limited. That will have to hold set_limit_mutex as well. | |
2980 | */ | |
2981 | int memcg_update_cache_sizes(struct mem_cgroup *memcg) | |
2982 | { | |
2983 | int num, ret; | |
2984 | ||
2985 | num = ida_simple_get(&kmem_limited_groups, | |
2986 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); | |
2987 | if (num < 0) | |
2988 | return num; | |
2989 | /* | |
2990 | * After this point, kmem_accounted (that we test atomically in | |
2991 | * the beginning of this conditional), is no longer 0. This | |
2992 | * guarantees only one process will set the following boolean | |
2993 | * to true. We don't need test_and_set because we're protected | |
2994 | * by the set_limit_mutex anyway. | |
2995 | */ | |
2996 | memcg_kmem_set_activated(memcg); | |
2997 | ||
2998 | ret = memcg_update_all_caches(num+1); | |
2999 | if (ret) { | |
3000 | ida_simple_remove(&kmem_limited_groups, num); | |
3001 | memcg_kmem_clear_activated(memcg); | |
3002 | return ret; | |
3003 | } | |
3004 | ||
3005 | memcg->kmemcg_id = num; | |
3006 | INIT_LIST_HEAD(&memcg->memcg_slab_caches); | |
3007 | mutex_init(&memcg->slab_caches_mutex); | |
3008 | return 0; | |
3009 | } | |
3010 | ||
3011 | static size_t memcg_caches_array_size(int num_groups) | |
3012 | { | |
3013 | ssize_t size; | |
3014 | if (num_groups <= 0) | |
3015 | return 0; | |
3016 | ||
3017 | size = 2 * num_groups; | |
3018 | if (size < MEMCG_CACHES_MIN_SIZE) | |
3019 | size = MEMCG_CACHES_MIN_SIZE; | |
3020 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
3021 | size = MEMCG_CACHES_MAX_SIZE; | |
3022 | ||
3023 | return size; | |
3024 | } | |
3025 | ||
3026 | /* | |
3027 | * We should update the current array size iff all caches updates succeed. This | |
3028 | * can only be done from the slab side. The slab mutex needs to be held when | |
3029 | * calling this. | |
3030 | */ | |
3031 | void memcg_update_array_size(int num) | |
3032 | { | |
3033 | if (num > memcg_limited_groups_array_size) | |
3034 | memcg_limited_groups_array_size = memcg_caches_array_size(num); | |
3035 | } | |
3036 | ||
15cf17d2 KK |
3037 | static void kmem_cache_destroy_work_func(struct work_struct *w); |
3038 | ||
55007d84 GC |
3039 | int memcg_update_cache_size(struct kmem_cache *s, int num_groups) |
3040 | { | |
3041 | struct memcg_cache_params *cur_params = s->memcg_params; | |
3042 | ||
3043 | VM_BUG_ON(s->memcg_params && !s->memcg_params->is_root_cache); | |
3044 | ||
3045 | if (num_groups > memcg_limited_groups_array_size) { | |
3046 | int i; | |
3047 | ssize_t size = memcg_caches_array_size(num_groups); | |
3048 | ||
3049 | size *= sizeof(void *); | |
90c7a79c | 3050 | size += offsetof(struct memcg_cache_params, memcg_caches); |
55007d84 GC |
3051 | |
3052 | s->memcg_params = kzalloc(size, GFP_KERNEL); | |
3053 | if (!s->memcg_params) { | |
3054 | s->memcg_params = cur_params; | |
3055 | return -ENOMEM; | |
3056 | } | |
3057 | ||
3058 | s->memcg_params->is_root_cache = true; | |
3059 | ||
3060 | /* | |
3061 | * There is the chance it will be bigger than | |
3062 | * memcg_limited_groups_array_size, if we failed an allocation | |
3063 | * in a cache, in which case all caches updated before it, will | |
3064 | * have a bigger array. | |
3065 | * | |
3066 | * But if that is the case, the data after | |
3067 | * memcg_limited_groups_array_size is certainly unused | |
3068 | */ | |
3069 | for (i = 0; i < memcg_limited_groups_array_size; i++) { | |
3070 | if (!cur_params->memcg_caches[i]) | |
3071 | continue; | |
3072 | s->memcg_params->memcg_caches[i] = | |
3073 | cur_params->memcg_caches[i]; | |
3074 | } | |
3075 | ||
3076 | /* | |
3077 | * Ideally, we would wait until all caches succeed, and only | |
3078 | * then free the old one. But this is not worth the extra | |
3079 | * pointer per-cache we'd have to have for this. | |
3080 | * | |
3081 | * It is not a big deal if some caches are left with a size | |
3082 | * bigger than the others. And all updates will reset this | |
3083 | * anyway. | |
3084 | */ | |
3085 | kfree(cur_params); | |
3086 | } | |
3087 | return 0; | |
3088 | } | |
3089 | ||
943a451a GC |
3090 | int memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s, |
3091 | struct kmem_cache *root_cache) | |
2633d7a0 | 3092 | { |
90c7a79c | 3093 | size_t size; |
2633d7a0 GC |
3094 | |
3095 | if (!memcg_kmem_enabled()) | |
3096 | return 0; | |
3097 | ||
90c7a79c AV |
3098 | if (!memcg) { |
3099 | size = offsetof(struct memcg_cache_params, memcg_caches); | |
55007d84 | 3100 | size += memcg_limited_groups_array_size * sizeof(void *); |
90c7a79c AV |
3101 | } else |
3102 | size = sizeof(struct memcg_cache_params); | |
55007d84 | 3103 | |
2633d7a0 GC |
3104 | s->memcg_params = kzalloc(size, GFP_KERNEL); |
3105 | if (!s->memcg_params) | |
3106 | return -ENOMEM; | |
3107 | ||
943a451a | 3108 | if (memcg) { |
2633d7a0 | 3109 | s->memcg_params->memcg = memcg; |
943a451a | 3110 | s->memcg_params->root_cache = root_cache; |
3e6b11df AV |
3111 | INIT_WORK(&s->memcg_params->destroy, |
3112 | kmem_cache_destroy_work_func); | |
4ba902b5 GC |
3113 | } else |
3114 | s->memcg_params->is_root_cache = true; | |
3115 | ||
2633d7a0 GC |
3116 | return 0; |
3117 | } | |
3118 | ||
3119 | void memcg_release_cache(struct kmem_cache *s) | |
3120 | { | |
d7f25f8a GC |
3121 | struct kmem_cache *root; |
3122 | struct mem_cgroup *memcg; | |
3123 | int id; | |
3124 | ||
3125 | /* | |
3126 | * This happens, for instance, when a root cache goes away before we | |
3127 | * add any memcg. | |
3128 | */ | |
3129 | if (!s->memcg_params) | |
3130 | return; | |
3131 | ||
3132 | if (s->memcg_params->is_root_cache) | |
3133 | goto out; | |
3134 | ||
3135 | memcg = s->memcg_params->memcg; | |
3136 | id = memcg_cache_id(memcg); | |
3137 | ||
3138 | root = s->memcg_params->root_cache; | |
3139 | root->memcg_params->memcg_caches[id] = NULL; | |
d7f25f8a GC |
3140 | |
3141 | mutex_lock(&memcg->slab_caches_mutex); | |
3142 | list_del(&s->memcg_params->list); | |
3143 | mutex_unlock(&memcg->slab_caches_mutex); | |
3144 | ||
20f05310 | 3145 | css_put(&memcg->css); |
d7f25f8a | 3146 | out: |
2633d7a0 GC |
3147 | kfree(s->memcg_params); |
3148 | } | |
3149 | ||
0e9d92f2 GC |
3150 | /* |
3151 | * During the creation a new cache, we need to disable our accounting mechanism | |
3152 | * altogether. This is true even if we are not creating, but rather just | |
3153 | * enqueing new caches to be created. | |
3154 | * | |
3155 | * This is because that process will trigger allocations; some visible, like | |
3156 | * explicit kmallocs to auxiliary data structures, name strings and internal | |
3157 | * cache structures; some well concealed, like INIT_WORK() that can allocate | |
3158 | * objects during debug. | |
3159 | * | |
3160 | * If any allocation happens during memcg_kmem_get_cache, we will recurse back | |
3161 | * to it. This may not be a bounded recursion: since the first cache creation | |
3162 | * failed to complete (waiting on the allocation), we'll just try to create the | |
3163 | * cache again, failing at the same point. | |
3164 | * | |
3165 | * memcg_kmem_get_cache is prepared to abort after seeing a positive count of | |
3166 | * memcg_kmem_skip_account. So we enclose anything that might allocate memory | |
3167 | * inside the following two functions. | |
3168 | */ | |
3169 | static inline void memcg_stop_kmem_account(void) | |
3170 | { | |
3171 | VM_BUG_ON(!current->mm); | |
3172 | current->memcg_kmem_skip_account++; | |
3173 | } | |
3174 | ||
3175 | static inline void memcg_resume_kmem_account(void) | |
3176 | { | |
3177 | VM_BUG_ON(!current->mm); | |
3178 | current->memcg_kmem_skip_account--; | |
3179 | } | |
3180 | ||
1f458cbf GC |
3181 | static void kmem_cache_destroy_work_func(struct work_struct *w) |
3182 | { | |
3183 | struct kmem_cache *cachep; | |
3184 | struct memcg_cache_params *p; | |
3185 | ||
3186 | p = container_of(w, struct memcg_cache_params, destroy); | |
3187 | ||
3188 | cachep = memcg_params_to_cache(p); | |
3189 | ||
22933152 GC |
3190 | /* |
3191 | * If we get down to 0 after shrink, we could delete right away. | |
3192 | * However, memcg_release_pages() already puts us back in the workqueue | |
3193 | * in that case. If we proceed deleting, we'll get a dangling | |
3194 | * reference, and removing the object from the workqueue in that case | |
3195 | * is unnecessary complication. We are not a fast path. | |
3196 | * | |
3197 | * Note that this case is fundamentally different from racing with | |
3198 | * shrink_slab(): if memcg_cgroup_destroy_cache() is called in | |
3199 | * kmem_cache_shrink, not only we would be reinserting a dead cache | |
3200 | * into the queue, but doing so from inside the worker racing to | |
3201 | * destroy it. | |
3202 | * | |
3203 | * So if we aren't down to zero, we'll just schedule a worker and try | |
3204 | * again | |
3205 | */ | |
3206 | if (atomic_read(&cachep->memcg_params->nr_pages) != 0) { | |
3207 | kmem_cache_shrink(cachep); | |
3208 | if (atomic_read(&cachep->memcg_params->nr_pages) == 0) | |
3209 | return; | |
3210 | } else | |
1f458cbf GC |
3211 | kmem_cache_destroy(cachep); |
3212 | } | |
3213 | ||
3214 | void mem_cgroup_destroy_cache(struct kmem_cache *cachep) | |
3215 | { | |
3216 | if (!cachep->memcg_params->dead) | |
3217 | return; | |
3218 | ||
22933152 GC |
3219 | /* |
3220 | * There are many ways in which we can get here. | |
3221 | * | |
3222 | * We can get to a memory-pressure situation while the delayed work is | |
3223 | * still pending to run. The vmscan shrinkers can then release all | |
3224 | * cache memory and get us to destruction. If this is the case, we'll | |
3225 | * be executed twice, which is a bug (the second time will execute over | |
3226 | * bogus data). In this case, cancelling the work should be fine. | |
3227 | * | |
3228 | * But we can also get here from the worker itself, if | |
3229 | * kmem_cache_shrink is enough to shake all the remaining objects and | |
3230 | * get the page count to 0. In this case, we'll deadlock if we try to | |
3231 | * cancel the work (the worker runs with an internal lock held, which | |
3232 | * is the same lock we would hold for cancel_work_sync().) | |
3233 | * | |
3234 | * Since we can't possibly know who got us here, just refrain from | |
3235 | * running if there is already work pending | |
3236 | */ | |
3237 | if (work_pending(&cachep->memcg_params->destroy)) | |
3238 | return; | |
1f458cbf GC |
3239 | /* |
3240 | * We have to defer the actual destroying to a workqueue, because | |
3241 | * we might currently be in a context that cannot sleep. | |
3242 | */ | |
3243 | schedule_work(&cachep->memcg_params->destroy); | |
3244 | } | |
3245 | ||
d9c10ddd MH |
3246 | /* |
3247 | * This lock protects updaters, not readers. We want readers to be as fast as | |
3248 | * they can, and they will either see NULL or a valid cache value. Our model | |
3249 | * allow them to see NULL, in which case the root memcg will be selected. | |
3250 | * | |
3251 | * We need this lock because multiple allocations to the same cache from a non | |
3252 | * will span more than one worker. Only one of them can create the cache. | |
3253 | */ | |
3254 | static DEFINE_MUTEX(memcg_cache_mutex); | |
d7f25f8a | 3255 | |
d9c10ddd MH |
3256 | /* |
3257 | * Called with memcg_cache_mutex held | |
3258 | */ | |
d7f25f8a GC |
3259 | static struct kmem_cache *kmem_cache_dup(struct mem_cgroup *memcg, |
3260 | struct kmem_cache *s) | |
3261 | { | |
d7f25f8a | 3262 | struct kmem_cache *new; |
d9c10ddd | 3263 | static char *tmp_name = NULL; |
d7f25f8a | 3264 | |
d9c10ddd MH |
3265 | lockdep_assert_held(&memcg_cache_mutex); |
3266 | ||
3267 | /* | |
3268 | * kmem_cache_create_memcg duplicates the given name and | |
3269 | * cgroup_name for this name requires RCU context. | |
3270 | * This static temporary buffer is used to prevent from | |
3271 | * pointless shortliving allocation. | |
3272 | */ | |
3273 | if (!tmp_name) { | |
3274 | tmp_name = kmalloc(PATH_MAX, GFP_KERNEL); | |
3275 | if (!tmp_name) | |
3276 | return NULL; | |
3277 | } | |
3278 | ||
3279 | rcu_read_lock(); | |
3280 | snprintf(tmp_name, PATH_MAX, "%s(%d:%s)", s->name, | |
3281 | memcg_cache_id(memcg), cgroup_name(memcg->css.cgroup)); | |
3282 | rcu_read_unlock(); | |
d7f25f8a | 3283 | |
d9c10ddd | 3284 | new = kmem_cache_create_memcg(memcg, tmp_name, s->object_size, s->align, |
943a451a | 3285 | (s->flags & ~SLAB_PANIC), s->ctor, s); |
d7f25f8a | 3286 | |
d79923fa GC |
3287 | if (new) |
3288 | new->allocflags |= __GFP_KMEMCG; | |
3289 | ||
d7f25f8a GC |
3290 | return new; |
3291 | } | |
3292 | ||
d7f25f8a GC |
3293 | static struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg, |
3294 | struct kmem_cache *cachep) | |
3295 | { | |
3296 | struct kmem_cache *new_cachep; | |
3297 | int idx; | |
3298 | ||
3299 | BUG_ON(!memcg_can_account_kmem(memcg)); | |
3300 | ||
3301 | idx = memcg_cache_id(memcg); | |
3302 | ||
3303 | mutex_lock(&memcg_cache_mutex); | |
3304 | new_cachep = cachep->memcg_params->memcg_caches[idx]; | |
20f05310 LZ |
3305 | if (new_cachep) { |
3306 | css_put(&memcg->css); | |
d7f25f8a | 3307 | goto out; |
20f05310 | 3308 | } |
d7f25f8a GC |
3309 | |
3310 | new_cachep = kmem_cache_dup(memcg, cachep); | |
d7f25f8a GC |
3311 | if (new_cachep == NULL) { |
3312 | new_cachep = cachep; | |
20f05310 | 3313 | css_put(&memcg->css); |
d7f25f8a GC |
3314 | goto out; |
3315 | } | |
3316 | ||
1f458cbf | 3317 | atomic_set(&new_cachep->memcg_params->nr_pages , 0); |
d7f25f8a GC |
3318 | |
3319 | cachep->memcg_params->memcg_caches[idx] = new_cachep; | |
3320 | /* | |
3321 | * the readers won't lock, make sure everybody sees the updated value, | |
3322 | * so they won't put stuff in the queue again for no reason | |
3323 | */ | |
3324 | wmb(); | |
3325 | out: | |
3326 | mutex_unlock(&memcg_cache_mutex); | |
3327 | return new_cachep; | |
3328 | } | |
3329 | ||
7cf27982 GC |
3330 | void kmem_cache_destroy_memcg_children(struct kmem_cache *s) |
3331 | { | |
3332 | struct kmem_cache *c; | |
3333 | int i; | |
3334 | ||
3335 | if (!s->memcg_params) | |
3336 | return; | |
3337 | if (!s->memcg_params->is_root_cache) | |
3338 | return; | |
3339 | ||
3340 | /* | |
3341 | * If the cache is being destroyed, we trust that there is no one else | |
3342 | * requesting objects from it. Even if there are, the sanity checks in | |
3343 | * kmem_cache_destroy should caught this ill-case. | |
3344 | * | |
3345 | * Still, we don't want anyone else freeing memcg_caches under our | |
3346 | * noses, which can happen if a new memcg comes to life. As usual, | |
3347 | * we'll take the set_limit_mutex to protect ourselves against this. | |
3348 | */ | |
3349 | mutex_lock(&set_limit_mutex); | |
3350 | for (i = 0; i < memcg_limited_groups_array_size; i++) { | |
3351 | c = s->memcg_params->memcg_caches[i]; | |
3352 | if (!c) | |
3353 | continue; | |
3354 | ||
3355 | /* | |
3356 | * We will now manually delete the caches, so to avoid races | |
3357 | * we need to cancel all pending destruction workers and | |
3358 | * proceed with destruction ourselves. | |
3359 | * | |
3360 | * kmem_cache_destroy() will call kmem_cache_shrink internally, | |
3361 | * and that could spawn the workers again: it is likely that | |
3362 | * the cache still have active pages until this very moment. | |
3363 | * This would lead us back to mem_cgroup_destroy_cache. | |
3364 | * | |
3365 | * But that will not execute at all if the "dead" flag is not | |
3366 | * set, so flip it down to guarantee we are in control. | |
3367 | */ | |
3368 | c->memcg_params->dead = false; | |
22933152 | 3369 | cancel_work_sync(&c->memcg_params->destroy); |
7cf27982 GC |
3370 | kmem_cache_destroy(c); |
3371 | } | |
3372 | mutex_unlock(&set_limit_mutex); | |
3373 | } | |
3374 | ||
d7f25f8a GC |
3375 | struct create_work { |
3376 | struct mem_cgroup *memcg; | |
3377 | struct kmem_cache *cachep; | |
3378 | struct work_struct work; | |
3379 | }; | |
3380 | ||
1f458cbf GC |
3381 | static void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg) |
3382 | { | |
3383 | struct kmem_cache *cachep; | |
3384 | struct memcg_cache_params *params; | |
3385 | ||
3386 | if (!memcg_kmem_is_active(memcg)) | |
3387 | return; | |
3388 | ||
3389 | mutex_lock(&memcg->slab_caches_mutex); | |
3390 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) { | |
3391 | cachep = memcg_params_to_cache(params); | |
3392 | cachep->memcg_params->dead = true; | |
1f458cbf GC |
3393 | schedule_work(&cachep->memcg_params->destroy); |
3394 | } | |
3395 | mutex_unlock(&memcg->slab_caches_mutex); | |
3396 | } | |
3397 | ||
d7f25f8a GC |
3398 | static void memcg_create_cache_work_func(struct work_struct *w) |
3399 | { | |
3400 | struct create_work *cw; | |
3401 | ||
3402 | cw = container_of(w, struct create_work, work); | |
3403 | memcg_create_kmem_cache(cw->memcg, cw->cachep); | |
d7f25f8a GC |
3404 | kfree(cw); |
3405 | } | |
3406 | ||
3407 | /* | |
3408 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 3409 | */ |
0e9d92f2 GC |
3410 | static void __memcg_create_cache_enqueue(struct mem_cgroup *memcg, |
3411 | struct kmem_cache *cachep) | |
d7f25f8a GC |
3412 | { |
3413 | struct create_work *cw; | |
3414 | ||
3415 | cw = kmalloc(sizeof(struct create_work), GFP_NOWAIT); | |
ca0dde97 LZ |
3416 | if (cw == NULL) { |
3417 | css_put(&memcg->css); | |
d7f25f8a GC |
3418 | return; |
3419 | } | |
3420 | ||
3421 | cw->memcg = memcg; | |
3422 | cw->cachep = cachep; | |
3423 | ||
3424 | INIT_WORK(&cw->work, memcg_create_cache_work_func); | |
3425 | schedule_work(&cw->work); | |
3426 | } | |
3427 | ||
0e9d92f2 GC |
3428 | static void memcg_create_cache_enqueue(struct mem_cgroup *memcg, |
3429 | struct kmem_cache *cachep) | |
3430 | { | |
3431 | /* | |
3432 | * We need to stop accounting when we kmalloc, because if the | |
3433 | * corresponding kmalloc cache is not yet created, the first allocation | |
3434 | * in __memcg_create_cache_enqueue will recurse. | |
3435 | * | |
3436 | * However, it is better to enclose the whole function. Depending on | |
3437 | * the debugging options enabled, INIT_WORK(), for instance, can | |
3438 | * trigger an allocation. This too, will make us recurse. Because at | |
3439 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
3440 | * the safest choice is to do it like this, wrapping the whole function. | |
3441 | */ | |
3442 | memcg_stop_kmem_account(); | |
3443 | __memcg_create_cache_enqueue(memcg, cachep); | |
3444 | memcg_resume_kmem_account(); | |
3445 | } | |
d7f25f8a GC |
3446 | /* |
3447 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
3448 | * We try to use the current memcg's version of the cache. | |
3449 | * | |
3450 | * If the cache does not exist yet, if we are the first user of it, | |
3451 | * we either create it immediately, if possible, or create it asynchronously | |
3452 | * in a workqueue. | |
3453 | * In the latter case, we will let the current allocation go through with | |
3454 | * the original cache. | |
3455 | * | |
3456 | * Can't be called in interrupt context or from kernel threads. | |
3457 | * This function needs to be called with rcu_read_lock() held. | |
3458 | */ | |
3459 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, | |
3460 | gfp_t gfp) | |
3461 | { | |
3462 | struct mem_cgroup *memcg; | |
3463 | int idx; | |
3464 | ||
3465 | VM_BUG_ON(!cachep->memcg_params); | |
3466 | VM_BUG_ON(!cachep->memcg_params->is_root_cache); | |
3467 | ||
0e9d92f2 GC |
3468 | if (!current->mm || current->memcg_kmem_skip_account) |
3469 | return cachep; | |
3470 | ||
d7f25f8a GC |
3471 | rcu_read_lock(); |
3472 | memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner)); | |
d7f25f8a GC |
3473 | |
3474 | if (!memcg_can_account_kmem(memcg)) | |
ca0dde97 | 3475 | goto out; |
d7f25f8a GC |
3476 | |
3477 | idx = memcg_cache_id(memcg); | |
3478 | ||
3479 | /* | |
3480 | * barrier to mare sure we're always seeing the up to date value. The | |
3481 | * code updating memcg_caches will issue a write barrier to match this. | |
3482 | */ | |
3483 | read_barrier_depends(); | |
ca0dde97 LZ |
3484 | if (likely(cachep->memcg_params->memcg_caches[idx])) { |
3485 | cachep = cachep->memcg_params->memcg_caches[idx]; | |
3486 | goto out; | |
d7f25f8a GC |
3487 | } |
3488 | ||
ca0dde97 LZ |
3489 | /* The corresponding put will be done in the workqueue. */ |
3490 | if (!css_tryget(&memcg->css)) | |
3491 | goto out; | |
3492 | rcu_read_unlock(); | |
3493 | ||
3494 | /* | |
3495 | * If we are in a safe context (can wait, and not in interrupt | |
3496 | * context), we could be be predictable and return right away. | |
3497 | * This would guarantee that the allocation being performed | |
3498 | * already belongs in the new cache. | |
3499 | * | |
3500 | * However, there are some clashes that can arrive from locking. | |
3501 | * For instance, because we acquire the slab_mutex while doing | |
3502 | * kmem_cache_dup, this means no further allocation could happen | |
3503 | * with the slab_mutex held. | |
3504 | * | |
3505 | * Also, because cache creation issue get_online_cpus(), this | |
3506 | * creates a lock chain: memcg_slab_mutex -> cpu_hotplug_mutex, | |
3507 | * that ends up reversed during cpu hotplug. (cpuset allocates | |
3508 | * a bunch of GFP_KERNEL memory during cpuup). Due to all that, | |
3509 | * better to defer everything. | |
3510 | */ | |
3511 | memcg_create_cache_enqueue(memcg, cachep); | |
3512 | return cachep; | |
3513 | out: | |
3514 | rcu_read_unlock(); | |
3515 | return cachep; | |
d7f25f8a GC |
3516 | } |
3517 | EXPORT_SYMBOL(__memcg_kmem_get_cache); | |
3518 | ||
7ae1e1d0 GC |
3519 | /* |
3520 | * We need to verify if the allocation against current->mm->owner's memcg is | |
3521 | * possible for the given order. But the page is not allocated yet, so we'll | |
3522 | * need a further commit step to do the final arrangements. | |
3523 | * | |
3524 | * It is possible for the task to switch cgroups in this mean time, so at | |
3525 | * commit time, we can't rely on task conversion any longer. We'll then use | |
3526 | * the handle argument to return to the caller which cgroup we should commit | |
3527 | * against. We could also return the memcg directly and avoid the pointer | |
3528 | * passing, but a boolean return value gives better semantics considering | |
3529 | * the compiled-out case as well. | |
3530 | * | |
3531 | * Returning true means the allocation is possible. | |
3532 | */ | |
3533 | bool | |
3534 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) | |
3535 | { | |
3536 | struct mem_cgroup *memcg; | |
3537 | int ret; | |
3538 | ||
3539 | *_memcg = NULL; | |
6d42c232 GC |
3540 | |
3541 | /* | |
3542 | * Disabling accounting is only relevant for some specific memcg | |
3543 | * internal allocations. Therefore we would initially not have such | |
3544 | * check here, since direct calls to the page allocator that are marked | |
3545 | * with GFP_KMEMCG only happen outside memcg core. We are mostly | |
3546 | * concerned with cache allocations, and by having this test at | |
3547 | * memcg_kmem_get_cache, we are already able to relay the allocation to | |
3548 | * the root cache and bypass the memcg cache altogether. | |
3549 | * | |
3550 | * There is one exception, though: the SLUB allocator does not create | |
3551 | * large order caches, but rather service large kmallocs directly from | |
3552 | * the page allocator. Therefore, the following sequence when backed by | |
3553 | * the SLUB allocator: | |
3554 | * | |
f894ffa8 AM |
3555 | * memcg_stop_kmem_account(); |
3556 | * kmalloc(<large_number>) | |
3557 | * memcg_resume_kmem_account(); | |
6d42c232 GC |
3558 | * |
3559 | * would effectively ignore the fact that we should skip accounting, | |
3560 | * since it will drive us directly to this function without passing | |
3561 | * through the cache selector memcg_kmem_get_cache. Such large | |
3562 | * allocations are extremely rare but can happen, for instance, for the | |
3563 | * cache arrays. We bring this test here. | |
3564 | */ | |
3565 | if (!current->mm || current->memcg_kmem_skip_account) | |
3566 | return true; | |
3567 | ||
7ae1e1d0 GC |
3568 | memcg = try_get_mem_cgroup_from_mm(current->mm); |
3569 | ||
3570 | /* | |
3571 | * very rare case described in mem_cgroup_from_task. Unfortunately there | |
3572 | * isn't much we can do without complicating this too much, and it would | |
3573 | * be gfp-dependent anyway. Just let it go | |
3574 | */ | |
3575 | if (unlikely(!memcg)) | |
3576 | return true; | |
3577 | ||
3578 | if (!memcg_can_account_kmem(memcg)) { | |
3579 | css_put(&memcg->css); | |
3580 | return true; | |
3581 | } | |
3582 | ||
7ae1e1d0 GC |
3583 | ret = memcg_charge_kmem(memcg, gfp, PAGE_SIZE << order); |
3584 | if (!ret) | |
3585 | *_memcg = memcg; | |
7ae1e1d0 GC |
3586 | |
3587 | css_put(&memcg->css); | |
3588 | return (ret == 0); | |
3589 | } | |
3590 | ||
3591 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
3592 | int order) | |
3593 | { | |
3594 | struct page_cgroup *pc; | |
3595 | ||
3596 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
3597 | ||
3598 | /* The page allocation failed. Revert */ | |
3599 | if (!page) { | |
3600 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); | |
7ae1e1d0 GC |
3601 | return; |
3602 | } | |
3603 | ||
3604 | pc = lookup_page_cgroup(page); | |
3605 | lock_page_cgroup(pc); | |
3606 | pc->mem_cgroup = memcg; | |
3607 | SetPageCgroupUsed(pc); | |
3608 | unlock_page_cgroup(pc); | |
3609 | } | |
3610 | ||
3611 | void __memcg_kmem_uncharge_pages(struct page *page, int order) | |
3612 | { | |
3613 | struct mem_cgroup *memcg = NULL; | |
3614 | struct page_cgroup *pc; | |
3615 | ||
3616 | ||
3617 | pc = lookup_page_cgroup(page); | |
3618 | /* | |
3619 | * Fast unlocked return. Theoretically might have changed, have to | |
3620 | * check again after locking. | |
3621 | */ | |
3622 | if (!PageCgroupUsed(pc)) | |
3623 | return; | |
3624 | ||
3625 | lock_page_cgroup(pc); | |
3626 | if (PageCgroupUsed(pc)) { | |
3627 | memcg = pc->mem_cgroup; | |
3628 | ClearPageCgroupUsed(pc); | |
3629 | } | |
3630 | unlock_page_cgroup(pc); | |
3631 | ||
3632 | /* | |
3633 | * We trust that only if there is a memcg associated with the page, it | |
3634 | * is a valid allocation | |
3635 | */ | |
3636 | if (!memcg) | |
3637 | return; | |
3638 | ||
3639 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
3640 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); | |
7ae1e1d0 | 3641 | } |
1f458cbf GC |
3642 | #else |
3643 | static inline void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg) | |
3644 | { | |
3645 | } | |
7ae1e1d0 GC |
3646 | #endif /* CONFIG_MEMCG_KMEM */ |
3647 | ||
ca3e0214 KH |
3648 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3649 | ||
a0db00fc | 3650 | #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION) |
ca3e0214 KH |
3651 | /* |
3652 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
3653 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
3654 | * charge/uncharge will be never happen and move_account() is done under | |
3655 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 3656 | */ |
e94c8a9c | 3657 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 KH |
3658 | { |
3659 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
e94c8a9c | 3660 | struct page_cgroup *pc; |
b070e65c | 3661 | struct mem_cgroup *memcg; |
e94c8a9c | 3662 | int i; |
ca3e0214 | 3663 | |
3d37c4a9 KH |
3664 | if (mem_cgroup_disabled()) |
3665 | return; | |
b070e65c DR |
3666 | |
3667 | memcg = head_pc->mem_cgroup; | |
e94c8a9c KH |
3668 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
3669 | pc = head_pc + i; | |
b070e65c | 3670 | pc->mem_cgroup = memcg; |
e94c8a9c | 3671 | smp_wmb();/* see __commit_charge() */ |
e94c8a9c KH |
3672 | pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
3673 | } | |
b070e65c DR |
3674 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
3675 | HPAGE_PMD_NR); | |
ca3e0214 | 3676 | } |
12d27107 | 3677 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3678 | |
f817ed48 | 3679 | /** |
de3638d9 | 3680 | * mem_cgroup_move_account - move account of the page |
5564e88b | 3681 | * @page: the page |
7ec99d62 | 3682 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
3683 | * @pc: page_cgroup of the page. |
3684 | * @from: mem_cgroup which the page is moved from. | |
3685 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
3686 | * | |
3687 | * The caller must confirm following. | |
08e552c6 | 3688 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 3689 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 3690 | * |
2f3479b1 KH |
3691 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
3692 | * from old cgroup. | |
f817ed48 | 3693 | */ |
7ec99d62 JW |
3694 | static int mem_cgroup_move_account(struct page *page, |
3695 | unsigned int nr_pages, | |
3696 | struct page_cgroup *pc, | |
3697 | struct mem_cgroup *from, | |
2f3479b1 | 3698 | struct mem_cgroup *to) |
f817ed48 | 3699 | { |
de3638d9 JW |
3700 | unsigned long flags; |
3701 | int ret; | |
b2402857 | 3702 | bool anon = PageAnon(page); |
987eba66 | 3703 | |
f817ed48 | 3704 | VM_BUG_ON(from == to); |
5564e88b | 3705 | VM_BUG_ON(PageLRU(page)); |
de3638d9 JW |
3706 | /* |
3707 | * The page is isolated from LRU. So, collapse function | |
3708 | * will not handle this page. But page splitting can happen. | |
3709 | * Do this check under compound_page_lock(). The caller should | |
3710 | * hold it. | |
3711 | */ | |
3712 | ret = -EBUSY; | |
7ec99d62 | 3713 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
3714 | goto out; |
3715 | ||
3716 | lock_page_cgroup(pc); | |
3717 | ||
3718 | ret = -EINVAL; | |
3719 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
3720 | goto unlock; | |
3721 | ||
312734c0 | 3722 | move_lock_mem_cgroup(from, &flags); |
f817ed48 | 3723 | |
2ff76f11 | 3724 | if (!anon && page_mapped(page)) { |
c62b1a3b KH |
3725 | /* Update mapped_file data for mem_cgroup */ |
3726 | preempt_disable(); | |
3727 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
3728 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
3729 | preempt_enable(); | |
d69b042f | 3730 | } |
b070e65c | 3731 | mem_cgroup_charge_statistics(from, page, anon, -nr_pages); |
d69b042f | 3732 | |
854ffa8d | 3733 | /* caller should have done css_get */ |
08e552c6 | 3734 | pc->mem_cgroup = to; |
b070e65c | 3735 | mem_cgroup_charge_statistics(to, page, anon, nr_pages); |
312734c0 | 3736 | move_unlock_mem_cgroup(from, &flags); |
de3638d9 JW |
3737 | ret = 0; |
3738 | unlock: | |
57f9fd7d | 3739 | unlock_page_cgroup(pc); |
d2265e6f KH |
3740 | /* |
3741 | * check events | |
3742 | */ | |
5564e88b JW |
3743 | memcg_check_events(to, page); |
3744 | memcg_check_events(from, page); | |
de3638d9 | 3745 | out: |
f817ed48 KH |
3746 | return ret; |
3747 | } | |
3748 | ||
2ef37d3f MH |
3749 | /** |
3750 | * mem_cgroup_move_parent - moves page to the parent group | |
3751 | * @page: the page to move | |
3752 | * @pc: page_cgroup of the page | |
3753 | * @child: page's cgroup | |
3754 | * | |
3755 | * move charges to its parent or the root cgroup if the group has no | |
3756 | * parent (aka use_hierarchy==0). | |
3757 | * Although this might fail (get_page_unless_zero, isolate_lru_page or | |
3758 | * mem_cgroup_move_account fails) the failure is always temporary and | |
3759 | * it signals a race with a page removal/uncharge or migration. In the | |
3760 | * first case the page is on the way out and it will vanish from the LRU | |
3761 | * on the next attempt and the call should be retried later. | |
3762 | * Isolation from the LRU fails only if page has been isolated from | |
3763 | * the LRU since we looked at it and that usually means either global | |
3764 | * reclaim or migration going on. The page will either get back to the | |
3765 | * LRU or vanish. | |
3766 | * Finaly mem_cgroup_move_account fails only if the page got uncharged | |
3767 | * (!PageCgroupUsed) or moved to a different group. The page will | |
3768 | * disappear in the next attempt. | |
f817ed48 | 3769 | */ |
5564e88b JW |
3770 | static int mem_cgroup_move_parent(struct page *page, |
3771 | struct page_cgroup *pc, | |
6068bf01 | 3772 | struct mem_cgroup *child) |
f817ed48 | 3773 | { |
f817ed48 | 3774 | struct mem_cgroup *parent; |
7ec99d62 | 3775 | unsigned int nr_pages; |
4be4489f | 3776 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
3777 | int ret; |
3778 | ||
d8423011 | 3779 | VM_BUG_ON(mem_cgroup_is_root(child)); |
f817ed48 | 3780 | |
57f9fd7d DN |
3781 | ret = -EBUSY; |
3782 | if (!get_page_unless_zero(page)) | |
3783 | goto out; | |
3784 | if (isolate_lru_page(page)) | |
3785 | goto put; | |
52dbb905 | 3786 | |
7ec99d62 | 3787 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 3788 | |
cc926f78 KH |
3789 | parent = parent_mem_cgroup(child); |
3790 | /* | |
3791 | * If no parent, move charges to root cgroup. | |
3792 | */ | |
3793 | if (!parent) | |
3794 | parent = root_mem_cgroup; | |
f817ed48 | 3795 | |
2ef37d3f MH |
3796 | if (nr_pages > 1) { |
3797 | VM_BUG_ON(!PageTransHuge(page)); | |
987eba66 | 3798 | flags = compound_lock_irqsave(page); |
2ef37d3f | 3799 | } |
987eba66 | 3800 | |
cc926f78 | 3801 | ret = mem_cgroup_move_account(page, nr_pages, |
2f3479b1 | 3802 | pc, child, parent); |
cc926f78 KH |
3803 | if (!ret) |
3804 | __mem_cgroup_cancel_local_charge(child, nr_pages); | |
8dba474f | 3805 | |
7ec99d62 | 3806 | if (nr_pages > 1) |
987eba66 | 3807 | compound_unlock_irqrestore(page, flags); |
08e552c6 | 3808 | putback_lru_page(page); |
57f9fd7d | 3809 | put: |
40d58138 | 3810 | put_page(page); |
57f9fd7d | 3811 | out: |
f817ed48 KH |
3812 | return ret; |
3813 | } | |
3814 | ||
7a81b88c KH |
3815 | /* |
3816 | * Charge the memory controller for page usage. | |
3817 | * Return | |
3818 | * 0 if the charge was successful | |
3819 | * < 0 if the cgroup is over its limit | |
3820 | */ | |
3821 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
73045c47 | 3822 | gfp_t gfp_mask, enum charge_type ctype) |
7a81b88c | 3823 | { |
c0ff4b85 | 3824 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 3825 | unsigned int nr_pages = 1; |
8493ae43 | 3826 | bool oom = true; |
7a81b88c | 3827 | int ret; |
ec168510 | 3828 | |
37c2ac78 | 3829 | if (PageTransHuge(page)) { |
7ec99d62 | 3830 | nr_pages <<= compound_order(page); |
37c2ac78 | 3831 | VM_BUG_ON(!PageTransHuge(page)); |
8493ae43 JW |
3832 | /* |
3833 | * Never OOM-kill a process for a huge page. The | |
3834 | * fault handler will fall back to regular pages. | |
3835 | */ | |
3836 | oom = false; | |
37c2ac78 | 3837 | } |
7a81b88c | 3838 | |
c0ff4b85 | 3839 | ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom); |
38c5d72f | 3840 | if (ret == -ENOMEM) |
7a81b88c | 3841 | return ret; |
ce587e65 | 3842 | __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false); |
8a9f3ccd | 3843 | return 0; |
8a9f3ccd BS |
3844 | } |
3845 | ||
7a81b88c KH |
3846 | int mem_cgroup_newpage_charge(struct page *page, |
3847 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 3848 | { |
f8d66542 | 3849 | if (mem_cgroup_disabled()) |
cede86ac | 3850 | return 0; |
7a0524cf JW |
3851 | VM_BUG_ON(page_mapped(page)); |
3852 | VM_BUG_ON(page->mapping && !PageAnon(page)); | |
3853 | VM_BUG_ON(!mm); | |
217bc319 | 3854 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
41326c17 | 3855 | MEM_CGROUP_CHARGE_TYPE_ANON); |
217bc319 KH |
3856 | } |
3857 | ||
54595fe2 KH |
3858 | /* |
3859 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
3860 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 3861 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
3862 | * "commit()" or removed by "cancel()" |
3863 | */ | |
0435a2fd JW |
3864 | static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
3865 | struct page *page, | |
3866 | gfp_t mask, | |
3867 | struct mem_cgroup **memcgp) | |
8c7c6e34 | 3868 | { |
c0ff4b85 | 3869 | struct mem_cgroup *memcg; |
90deb788 | 3870 | struct page_cgroup *pc; |
54595fe2 | 3871 | int ret; |
8c7c6e34 | 3872 | |
90deb788 JW |
3873 | pc = lookup_page_cgroup(page); |
3874 | /* | |
3875 | * Every swap fault against a single page tries to charge the | |
3876 | * page, bail as early as possible. shmem_unuse() encounters | |
3877 | * already charged pages, too. The USED bit is protected by | |
3878 | * the page lock, which serializes swap cache removal, which | |
3879 | * in turn serializes uncharging. | |
3880 | */ | |
3881 | if (PageCgroupUsed(pc)) | |
3882 | return 0; | |
8c7c6e34 KH |
3883 | if (!do_swap_account) |
3884 | goto charge_cur_mm; | |
c0ff4b85 R |
3885 | memcg = try_get_mem_cgroup_from_page(page); |
3886 | if (!memcg) | |
54595fe2 | 3887 | goto charge_cur_mm; |
72835c86 JW |
3888 | *memcgp = memcg; |
3889 | ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true); | |
c0ff4b85 | 3890 | css_put(&memcg->css); |
38c5d72f KH |
3891 | if (ret == -EINTR) |
3892 | ret = 0; | |
54595fe2 | 3893 | return ret; |
8c7c6e34 | 3894 | charge_cur_mm: |
38c5d72f KH |
3895 | ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true); |
3896 | if (ret == -EINTR) | |
3897 | ret = 0; | |
3898 | return ret; | |
8c7c6e34 KH |
3899 | } |
3900 | ||
0435a2fd JW |
3901 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page, |
3902 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
3903 | { | |
3904 | *memcgp = NULL; | |
3905 | if (mem_cgroup_disabled()) | |
3906 | return 0; | |
bdf4f4d2 JW |
3907 | /* |
3908 | * A racing thread's fault, or swapoff, may have already | |
3909 | * updated the pte, and even removed page from swap cache: in | |
3910 | * those cases unuse_pte()'s pte_same() test will fail; but | |
3911 | * there's also a KSM case which does need to charge the page. | |
3912 | */ | |
3913 | if (!PageSwapCache(page)) { | |
3914 | int ret; | |
3915 | ||
3916 | ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, memcgp, true); | |
3917 | if (ret == -EINTR) | |
3918 | ret = 0; | |
3919 | return ret; | |
3920 | } | |
0435a2fd JW |
3921 | return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp); |
3922 | } | |
3923 | ||
827a03d2 JW |
3924 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg) |
3925 | { | |
3926 | if (mem_cgroup_disabled()) | |
3927 | return; | |
3928 | if (!memcg) | |
3929 | return; | |
3930 | __mem_cgroup_cancel_charge(memcg, 1); | |
3931 | } | |
3932 | ||
83aae4c7 | 3933 | static void |
72835c86 | 3934 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg, |
83aae4c7 | 3935 | enum charge_type ctype) |
7a81b88c | 3936 | { |
f8d66542 | 3937 | if (mem_cgroup_disabled()) |
7a81b88c | 3938 | return; |
72835c86 | 3939 | if (!memcg) |
7a81b88c | 3940 | return; |
5a6475a4 | 3941 | |
ce587e65 | 3942 | __mem_cgroup_commit_charge(memcg, page, 1, ctype, true); |
8c7c6e34 KH |
3943 | /* |
3944 | * Now swap is on-memory. This means this page may be | |
3945 | * counted both as mem and swap....double count. | |
03f3c433 KH |
3946 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
3947 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
3948 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 3949 | */ |
03f3c433 | 3950 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 3951 | swp_entry_t ent = {.val = page_private(page)}; |
86493009 | 3952 | mem_cgroup_uncharge_swap(ent); |
8c7c6e34 | 3953 | } |
7a81b88c KH |
3954 | } |
3955 | ||
72835c86 JW |
3956 | void mem_cgroup_commit_charge_swapin(struct page *page, |
3957 | struct mem_cgroup *memcg) | |
83aae4c7 | 3958 | { |
72835c86 | 3959 | __mem_cgroup_commit_charge_swapin(page, memcg, |
41326c17 | 3960 | MEM_CGROUP_CHARGE_TYPE_ANON); |
83aae4c7 DN |
3961 | } |
3962 | ||
827a03d2 JW |
3963 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
3964 | gfp_t gfp_mask) | |
7a81b88c | 3965 | { |
827a03d2 JW |
3966 | struct mem_cgroup *memcg = NULL; |
3967 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
3968 | int ret; | |
3969 | ||
f8d66542 | 3970 | if (mem_cgroup_disabled()) |
827a03d2 JW |
3971 | return 0; |
3972 | if (PageCompound(page)) | |
3973 | return 0; | |
3974 | ||
827a03d2 JW |
3975 | if (!PageSwapCache(page)) |
3976 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, type); | |
3977 | else { /* page is swapcache/shmem */ | |
0435a2fd JW |
3978 | ret = __mem_cgroup_try_charge_swapin(mm, page, |
3979 | gfp_mask, &memcg); | |
827a03d2 JW |
3980 | if (!ret) |
3981 | __mem_cgroup_commit_charge_swapin(page, memcg, type); | |
3982 | } | |
3983 | return ret; | |
7a81b88c KH |
3984 | } |
3985 | ||
c0ff4b85 | 3986 | static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg, |
7ec99d62 JW |
3987 | unsigned int nr_pages, |
3988 | const enum charge_type ctype) | |
569b846d KH |
3989 | { |
3990 | struct memcg_batch_info *batch = NULL; | |
3991 | bool uncharge_memsw = true; | |
7ec99d62 | 3992 | |
569b846d KH |
3993 | /* If swapout, usage of swap doesn't decrease */ |
3994 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
3995 | uncharge_memsw = false; | |
569b846d KH |
3996 | |
3997 | batch = ¤t->memcg_batch; | |
3998 | /* | |
3999 | * In usual, we do css_get() when we remember memcg pointer. | |
4000 | * But in this case, we keep res->usage until end of a series of | |
4001 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
4002 | */ | |
4003 | if (!batch->memcg) | |
c0ff4b85 | 4004 | batch->memcg = memcg; |
3c11ecf4 KH |
4005 | /* |
4006 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
25985edc | 4007 | * In those cases, all pages freed continuously can be expected to be in |
3c11ecf4 KH |
4008 | * the same cgroup and we have chance to coalesce uncharges. |
4009 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
4010 | * because we want to do uncharge as soon as possible. | |
4011 | */ | |
4012 | ||
4013 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
4014 | goto direct_uncharge; | |
4015 | ||
7ec99d62 | 4016 | if (nr_pages > 1) |
ec168510 AA |
4017 | goto direct_uncharge; |
4018 | ||
569b846d KH |
4019 | /* |
4020 | * In typical case, batch->memcg == mem. This means we can | |
4021 | * merge a series of uncharges to an uncharge of res_counter. | |
4022 | * If not, we uncharge res_counter ony by one. | |
4023 | */ | |
c0ff4b85 | 4024 | if (batch->memcg != memcg) |
569b846d KH |
4025 | goto direct_uncharge; |
4026 | /* remember freed charge and uncharge it later */ | |
7ffd4ca7 | 4027 | batch->nr_pages++; |
569b846d | 4028 | if (uncharge_memsw) |
7ffd4ca7 | 4029 | batch->memsw_nr_pages++; |
569b846d KH |
4030 | return; |
4031 | direct_uncharge: | |
c0ff4b85 | 4032 | res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE); |
569b846d | 4033 | if (uncharge_memsw) |
c0ff4b85 R |
4034 | res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE); |
4035 | if (unlikely(batch->memcg != memcg)) | |
4036 | memcg_oom_recover(memcg); | |
569b846d | 4037 | } |
7a81b88c | 4038 | |
8a9f3ccd | 4039 | /* |
69029cd5 | 4040 | * uncharge if !page_mapped(page) |
8a9f3ccd | 4041 | */ |
8c7c6e34 | 4042 | static struct mem_cgroup * |
0030f535 JW |
4043 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype, |
4044 | bool end_migration) | |
8a9f3ccd | 4045 | { |
c0ff4b85 | 4046 | struct mem_cgroup *memcg = NULL; |
7ec99d62 JW |
4047 | unsigned int nr_pages = 1; |
4048 | struct page_cgroup *pc; | |
b2402857 | 4049 | bool anon; |
8a9f3ccd | 4050 | |
f8d66542 | 4051 | if (mem_cgroup_disabled()) |
8c7c6e34 | 4052 | return NULL; |
4077960e | 4053 | |
37c2ac78 | 4054 | if (PageTransHuge(page)) { |
7ec99d62 | 4055 | nr_pages <<= compound_order(page); |
37c2ac78 AA |
4056 | VM_BUG_ON(!PageTransHuge(page)); |
4057 | } | |
8697d331 | 4058 | /* |
3c541e14 | 4059 | * Check if our page_cgroup is valid |
8697d331 | 4060 | */ |
52d4b9ac | 4061 | pc = lookup_page_cgroup(page); |
cfa44946 | 4062 | if (unlikely(!PageCgroupUsed(pc))) |
8c7c6e34 | 4063 | return NULL; |
b9c565d5 | 4064 | |
52d4b9ac | 4065 | lock_page_cgroup(pc); |
d13d1443 | 4066 | |
c0ff4b85 | 4067 | memcg = pc->mem_cgroup; |
8c7c6e34 | 4068 | |
d13d1443 KH |
4069 | if (!PageCgroupUsed(pc)) |
4070 | goto unlock_out; | |
4071 | ||
b2402857 KH |
4072 | anon = PageAnon(page); |
4073 | ||
d13d1443 | 4074 | switch (ctype) { |
41326c17 | 4075 | case MEM_CGROUP_CHARGE_TYPE_ANON: |
2ff76f11 KH |
4076 | /* |
4077 | * Generally PageAnon tells if it's the anon statistics to be | |
4078 | * updated; but sometimes e.g. mem_cgroup_uncharge_page() is | |
4079 | * used before page reached the stage of being marked PageAnon. | |
4080 | */ | |
b2402857 KH |
4081 | anon = true; |
4082 | /* fallthrough */ | |
8a9478ca | 4083 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c | 4084 | /* See mem_cgroup_prepare_migration() */ |
0030f535 JW |
4085 | if (page_mapped(page)) |
4086 | goto unlock_out; | |
4087 | /* | |
4088 | * Pages under migration may not be uncharged. But | |
4089 | * end_migration() /must/ be the one uncharging the | |
4090 | * unused post-migration page and so it has to call | |
4091 | * here with the migration bit still set. See the | |
4092 | * res_counter handling below. | |
4093 | */ | |
4094 | if (!end_migration && PageCgroupMigration(pc)) | |
d13d1443 KH |
4095 | goto unlock_out; |
4096 | break; | |
4097 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
4098 | if (!PageAnon(page)) { /* Shared memory */ | |
4099 | if (page->mapping && !page_is_file_cache(page)) | |
4100 | goto unlock_out; | |
4101 | } else if (page_mapped(page)) /* Anon */ | |
4102 | goto unlock_out; | |
4103 | break; | |
4104 | default: | |
4105 | break; | |
52d4b9ac | 4106 | } |
d13d1443 | 4107 | |
b070e65c | 4108 | mem_cgroup_charge_statistics(memcg, page, anon, -nr_pages); |
04046e1a | 4109 | |
52d4b9ac | 4110 | ClearPageCgroupUsed(pc); |
544122e5 KH |
4111 | /* |
4112 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
4113 | * freed from LRU. This is safe because uncharged page is expected not | |
4114 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
4115 | * special functions. | |
4116 | */ | |
b9c565d5 | 4117 | |
52d4b9ac | 4118 | unlock_page_cgroup(pc); |
f75ca962 | 4119 | /* |
c0ff4b85 | 4120 | * even after unlock, we have memcg->res.usage here and this memcg |
4050377b | 4121 | * will never be freed, so it's safe to call css_get(). |
f75ca962 | 4122 | */ |
c0ff4b85 | 4123 | memcg_check_events(memcg, page); |
f75ca962 | 4124 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
c0ff4b85 | 4125 | mem_cgroup_swap_statistics(memcg, true); |
4050377b | 4126 | css_get(&memcg->css); |
f75ca962 | 4127 | } |
0030f535 JW |
4128 | /* |
4129 | * Migration does not charge the res_counter for the | |
4130 | * replacement page, so leave it alone when phasing out the | |
4131 | * page that is unused after the migration. | |
4132 | */ | |
4133 | if (!end_migration && !mem_cgroup_is_root(memcg)) | |
c0ff4b85 | 4134 | mem_cgroup_do_uncharge(memcg, nr_pages, ctype); |
6d12e2d8 | 4135 | |
c0ff4b85 | 4136 | return memcg; |
d13d1443 KH |
4137 | |
4138 | unlock_out: | |
4139 | unlock_page_cgroup(pc); | |
8c7c6e34 | 4140 | return NULL; |
3c541e14 BS |
4141 | } |
4142 | ||
69029cd5 KH |
4143 | void mem_cgroup_uncharge_page(struct page *page) |
4144 | { | |
52d4b9ac KH |
4145 | /* early check. */ |
4146 | if (page_mapped(page)) | |
4147 | return; | |
40f23a21 | 4148 | VM_BUG_ON(page->mapping && !PageAnon(page)); |
28ccddf7 JW |
4149 | /* |
4150 | * If the page is in swap cache, uncharge should be deferred | |
4151 | * to the swap path, which also properly accounts swap usage | |
4152 | * and handles memcg lifetime. | |
4153 | * | |
4154 | * Note that this check is not stable and reclaim may add the | |
4155 | * page to swap cache at any time after this. However, if the | |
4156 | * page is not in swap cache by the time page->mapcount hits | |
4157 | * 0, there won't be any page table references to the swap | |
4158 | * slot, and reclaim will free it and not actually write the | |
4159 | * page to disk. | |
4160 | */ | |
0c59b89c JW |
4161 | if (PageSwapCache(page)) |
4162 | return; | |
0030f535 | 4163 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false); |
69029cd5 KH |
4164 | } |
4165 | ||
4166 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
4167 | { | |
4168 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 4169 | VM_BUG_ON(page->mapping); |
0030f535 | 4170 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false); |
69029cd5 KH |
4171 | } |
4172 | ||
569b846d KH |
4173 | /* |
4174 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
4175 | * In that cases, pages are freed continuously and we can expect pages | |
4176 | * are in the same memcg. All these calls itself limits the number of | |
4177 | * pages freed at once, then uncharge_start/end() is called properly. | |
4178 | * This may be called prural(2) times in a context, | |
4179 | */ | |
4180 | ||
4181 | void mem_cgroup_uncharge_start(void) | |
4182 | { | |
4183 | current->memcg_batch.do_batch++; | |
4184 | /* We can do nest. */ | |
4185 | if (current->memcg_batch.do_batch == 1) { | |
4186 | current->memcg_batch.memcg = NULL; | |
7ffd4ca7 JW |
4187 | current->memcg_batch.nr_pages = 0; |
4188 | current->memcg_batch.memsw_nr_pages = 0; | |
569b846d KH |
4189 | } |
4190 | } | |
4191 | ||
4192 | void mem_cgroup_uncharge_end(void) | |
4193 | { | |
4194 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
4195 | ||
4196 | if (!batch->do_batch) | |
4197 | return; | |
4198 | ||
4199 | batch->do_batch--; | |
4200 | if (batch->do_batch) /* If stacked, do nothing. */ | |
4201 | return; | |
4202 | ||
4203 | if (!batch->memcg) | |
4204 | return; | |
4205 | /* | |
4206 | * This "batch->memcg" is valid without any css_get/put etc... | |
4207 | * bacause we hide charges behind us. | |
4208 | */ | |
7ffd4ca7 JW |
4209 | if (batch->nr_pages) |
4210 | res_counter_uncharge(&batch->memcg->res, | |
4211 | batch->nr_pages * PAGE_SIZE); | |
4212 | if (batch->memsw_nr_pages) | |
4213 | res_counter_uncharge(&batch->memcg->memsw, | |
4214 | batch->memsw_nr_pages * PAGE_SIZE); | |
3c11ecf4 | 4215 | memcg_oom_recover(batch->memcg); |
569b846d KH |
4216 | /* forget this pointer (for sanity check) */ |
4217 | batch->memcg = NULL; | |
4218 | } | |
4219 | ||
e767e056 | 4220 | #ifdef CONFIG_SWAP |
8c7c6e34 | 4221 | /* |
e767e056 | 4222 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
4223 | * memcg information is recorded to swap_cgroup of "ent" |
4224 | */ | |
8a9478ca KH |
4225 | void |
4226 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
4227 | { |
4228 | struct mem_cgroup *memcg; | |
8a9478ca KH |
4229 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
4230 | ||
4231 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
4232 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
4233 | ||
0030f535 | 4234 | memcg = __mem_cgroup_uncharge_common(page, ctype, false); |
8c7c6e34 | 4235 | |
f75ca962 KH |
4236 | /* |
4237 | * record memcg information, if swapout && memcg != NULL, | |
4050377b | 4238 | * css_get() was called in uncharge(). |
f75ca962 KH |
4239 | */ |
4240 | if (do_swap_account && swapout && memcg) | |
a3b2d692 | 4241 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 | 4242 | } |
e767e056 | 4243 | #endif |
8c7c6e34 | 4244 | |
c255a458 | 4245 | #ifdef CONFIG_MEMCG_SWAP |
8c7c6e34 KH |
4246 | /* |
4247 | * called from swap_entry_free(). remove record in swap_cgroup and | |
4248 | * uncharge "memsw" account. | |
4249 | */ | |
4250 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 4251 | { |
8c7c6e34 | 4252 | struct mem_cgroup *memcg; |
a3b2d692 | 4253 | unsigned short id; |
8c7c6e34 KH |
4254 | |
4255 | if (!do_swap_account) | |
4256 | return; | |
4257 | ||
a3b2d692 KH |
4258 | id = swap_cgroup_record(ent, 0); |
4259 | rcu_read_lock(); | |
4260 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 4261 | if (memcg) { |
a3b2d692 KH |
4262 | /* |
4263 | * We uncharge this because swap is freed. | |
4264 | * This memcg can be obsolete one. We avoid calling css_tryget | |
4265 | */ | |
0c3e73e8 | 4266 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 4267 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 4268 | mem_cgroup_swap_statistics(memcg, false); |
4050377b | 4269 | css_put(&memcg->css); |
8c7c6e34 | 4270 | } |
a3b2d692 | 4271 | rcu_read_unlock(); |
d13d1443 | 4272 | } |
02491447 DN |
4273 | |
4274 | /** | |
4275 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
4276 | * @entry: swap entry to be moved | |
4277 | * @from: mem_cgroup which the entry is moved from | |
4278 | * @to: mem_cgroup which the entry is moved to | |
4279 | * | |
4280 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
4281 | * as the mem_cgroup's id of @from. | |
4282 | * | |
4283 | * Returns 0 on success, -EINVAL on failure. | |
4284 | * | |
4285 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
4286 | * both res and memsw, and called css_get(). | |
4287 | */ | |
4288 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 4289 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
4290 | { |
4291 | unsigned short old_id, new_id; | |
4292 | ||
4293 | old_id = css_id(&from->css); | |
4294 | new_id = css_id(&to->css); | |
4295 | ||
4296 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 4297 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 4298 | mem_cgroup_swap_statistics(to, true); |
02491447 | 4299 | /* |
483c30b5 DN |
4300 | * This function is only called from task migration context now. |
4301 | * It postpones res_counter and refcount handling till the end | |
4302 | * of task migration(mem_cgroup_clear_mc()) for performance | |
4050377b LZ |
4303 | * improvement. But we cannot postpone css_get(to) because if |
4304 | * the process that has been moved to @to does swap-in, the | |
4305 | * refcount of @to might be decreased to 0. | |
4306 | * | |
4307 | * We are in attach() phase, so the cgroup is guaranteed to be | |
4308 | * alive, so we can just call css_get(). | |
02491447 | 4309 | */ |
4050377b | 4310 | css_get(&to->css); |
02491447 DN |
4311 | return 0; |
4312 | } | |
4313 | return -EINVAL; | |
4314 | } | |
4315 | #else | |
4316 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 4317 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
4318 | { |
4319 | return -EINVAL; | |
4320 | } | |
8c7c6e34 | 4321 | #endif |
d13d1443 | 4322 | |
ae41be37 | 4323 | /* |
01b1ae63 KH |
4324 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
4325 | * page belongs to. | |
ae41be37 | 4326 | */ |
0030f535 JW |
4327 | void mem_cgroup_prepare_migration(struct page *page, struct page *newpage, |
4328 | struct mem_cgroup **memcgp) | |
ae41be37 | 4329 | { |
c0ff4b85 | 4330 | struct mem_cgroup *memcg = NULL; |
b32967ff | 4331 | unsigned int nr_pages = 1; |
7ec99d62 | 4332 | struct page_cgroup *pc; |
ac39cf8c | 4333 | enum charge_type ctype; |
8869b8f6 | 4334 | |
72835c86 | 4335 | *memcgp = NULL; |
56039efa | 4336 | |
f8d66542 | 4337 | if (mem_cgroup_disabled()) |
0030f535 | 4338 | return; |
4077960e | 4339 | |
b32967ff MG |
4340 | if (PageTransHuge(page)) |
4341 | nr_pages <<= compound_order(page); | |
4342 | ||
52d4b9ac KH |
4343 | pc = lookup_page_cgroup(page); |
4344 | lock_page_cgroup(pc); | |
4345 | if (PageCgroupUsed(pc)) { | |
c0ff4b85 R |
4346 | memcg = pc->mem_cgroup; |
4347 | css_get(&memcg->css); | |
ac39cf8c | 4348 | /* |
4349 | * At migrating an anonymous page, its mapcount goes down | |
4350 | * to 0 and uncharge() will be called. But, even if it's fully | |
4351 | * unmapped, migration may fail and this page has to be | |
4352 | * charged again. We set MIGRATION flag here and delay uncharge | |
4353 | * until end_migration() is called | |
4354 | * | |
4355 | * Corner Case Thinking | |
4356 | * A) | |
4357 | * When the old page was mapped as Anon and it's unmap-and-freed | |
4358 | * while migration was ongoing. | |
4359 | * If unmap finds the old page, uncharge() of it will be delayed | |
4360 | * until end_migration(). If unmap finds a new page, it's | |
4361 | * uncharged when it make mapcount to be 1->0. If unmap code | |
4362 | * finds swap_migration_entry, the new page will not be mapped | |
4363 | * and end_migration() will find it(mapcount==0). | |
4364 | * | |
4365 | * B) | |
4366 | * When the old page was mapped but migraion fails, the kernel | |
4367 | * remaps it. A charge for it is kept by MIGRATION flag even | |
4368 | * if mapcount goes down to 0. We can do remap successfully | |
4369 | * without charging it again. | |
4370 | * | |
4371 | * C) | |
4372 | * The "old" page is under lock_page() until the end of | |
4373 | * migration, so, the old page itself will not be swapped-out. | |
4374 | * If the new page is swapped out before end_migraton, our | |
4375 | * hook to usual swap-out path will catch the event. | |
4376 | */ | |
4377 | if (PageAnon(page)) | |
4378 | SetPageCgroupMigration(pc); | |
e8589cc1 | 4379 | } |
52d4b9ac | 4380 | unlock_page_cgroup(pc); |
ac39cf8c | 4381 | /* |
4382 | * If the page is not charged at this point, | |
4383 | * we return here. | |
4384 | */ | |
c0ff4b85 | 4385 | if (!memcg) |
0030f535 | 4386 | return; |
01b1ae63 | 4387 | |
72835c86 | 4388 | *memcgp = memcg; |
ac39cf8c | 4389 | /* |
4390 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
4391 | * is called before end_migration, we can catch all events on this new | |
4392 | * page. In the case new page is migrated but not remapped, new page's | |
4393 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
4394 | */ | |
ac39cf8c | 4395 | if (PageAnon(page)) |
41326c17 | 4396 | ctype = MEM_CGROUP_CHARGE_TYPE_ANON; |
ac39cf8c | 4397 | else |
62ba7442 | 4398 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; |
0030f535 JW |
4399 | /* |
4400 | * The page is committed to the memcg, but it's not actually | |
4401 | * charged to the res_counter since we plan on replacing the | |
4402 | * old one and only one page is going to be left afterwards. | |
4403 | */ | |
b32967ff | 4404 | __mem_cgroup_commit_charge(memcg, newpage, nr_pages, ctype, false); |
ae41be37 | 4405 | } |
8869b8f6 | 4406 | |
69029cd5 | 4407 | /* remove redundant charge if migration failed*/ |
c0ff4b85 | 4408 | void mem_cgroup_end_migration(struct mem_cgroup *memcg, |
50de1dd9 | 4409 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 4410 | { |
ac39cf8c | 4411 | struct page *used, *unused; |
01b1ae63 | 4412 | struct page_cgroup *pc; |
b2402857 | 4413 | bool anon; |
01b1ae63 | 4414 | |
c0ff4b85 | 4415 | if (!memcg) |
01b1ae63 | 4416 | return; |
b25ed609 | 4417 | |
50de1dd9 | 4418 | if (!migration_ok) { |
ac39cf8c | 4419 | used = oldpage; |
4420 | unused = newpage; | |
01b1ae63 | 4421 | } else { |
ac39cf8c | 4422 | used = newpage; |
01b1ae63 KH |
4423 | unused = oldpage; |
4424 | } | |
0030f535 | 4425 | anon = PageAnon(used); |
7d188958 JW |
4426 | __mem_cgroup_uncharge_common(unused, |
4427 | anon ? MEM_CGROUP_CHARGE_TYPE_ANON | |
4428 | : MEM_CGROUP_CHARGE_TYPE_CACHE, | |
4429 | true); | |
0030f535 | 4430 | css_put(&memcg->css); |
69029cd5 | 4431 | /* |
ac39cf8c | 4432 | * We disallowed uncharge of pages under migration because mapcount |
4433 | * of the page goes down to zero, temporarly. | |
4434 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 4435 | */ |
ac39cf8c | 4436 | pc = lookup_page_cgroup(oldpage); |
4437 | lock_page_cgroup(pc); | |
4438 | ClearPageCgroupMigration(pc); | |
4439 | unlock_page_cgroup(pc); | |
ac39cf8c | 4440 | |
01b1ae63 | 4441 | /* |
ac39cf8c | 4442 | * If a page is a file cache, radix-tree replacement is very atomic |
4443 | * and we can skip this check. When it was an Anon page, its mapcount | |
4444 | * goes down to 0. But because we added MIGRATION flage, it's not | |
4445 | * uncharged yet. There are several case but page->mapcount check | |
4446 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
4447 | * check. (see prepare_charge() also) | |
69029cd5 | 4448 | */ |
b2402857 | 4449 | if (anon) |
ac39cf8c | 4450 | mem_cgroup_uncharge_page(used); |
ae41be37 | 4451 | } |
78fb7466 | 4452 | |
ab936cbc KH |
4453 | /* |
4454 | * At replace page cache, newpage is not under any memcg but it's on | |
4455 | * LRU. So, this function doesn't touch res_counter but handles LRU | |
4456 | * in correct way. Both pages are locked so we cannot race with uncharge. | |
4457 | */ | |
4458 | void mem_cgroup_replace_page_cache(struct page *oldpage, | |
4459 | struct page *newpage) | |
4460 | { | |
bde05d1c | 4461 | struct mem_cgroup *memcg = NULL; |
ab936cbc | 4462 | struct page_cgroup *pc; |
ab936cbc | 4463 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
ab936cbc KH |
4464 | |
4465 | if (mem_cgroup_disabled()) | |
4466 | return; | |
4467 | ||
4468 | pc = lookup_page_cgroup(oldpage); | |
4469 | /* fix accounting on old pages */ | |
4470 | lock_page_cgroup(pc); | |
bde05d1c HD |
4471 | if (PageCgroupUsed(pc)) { |
4472 | memcg = pc->mem_cgroup; | |
b070e65c | 4473 | mem_cgroup_charge_statistics(memcg, oldpage, false, -1); |
bde05d1c HD |
4474 | ClearPageCgroupUsed(pc); |
4475 | } | |
ab936cbc KH |
4476 | unlock_page_cgroup(pc); |
4477 | ||
bde05d1c HD |
4478 | /* |
4479 | * When called from shmem_replace_page(), in some cases the | |
4480 | * oldpage has already been charged, and in some cases not. | |
4481 | */ | |
4482 | if (!memcg) | |
4483 | return; | |
ab936cbc KH |
4484 | /* |
4485 | * Even if newpage->mapping was NULL before starting replacement, | |
4486 | * the newpage may be on LRU(or pagevec for LRU) already. We lock | |
4487 | * LRU while we overwrite pc->mem_cgroup. | |
4488 | */ | |
ce587e65 | 4489 | __mem_cgroup_commit_charge(memcg, newpage, 1, type, true); |
ab936cbc KH |
4490 | } |
4491 | ||
f212ad7c DN |
4492 | #ifdef CONFIG_DEBUG_VM |
4493 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
4494 | { | |
4495 | struct page_cgroup *pc; | |
4496 | ||
4497 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
4498 | /* |
4499 | * Can be NULL while feeding pages into the page allocator for | |
4500 | * the first time, i.e. during boot or memory hotplug; | |
4501 | * or when mem_cgroup_disabled(). | |
4502 | */ | |
f212ad7c DN |
4503 | if (likely(pc) && PageCgroupUsed(pc)) |
4504 | return pc; | |
4505 | return NULL; | |
4506 | } | |
4507 | ||
4508 | bool mem_cgroup_bad_page_check(struct page *page) | |
4509 | { | |
4510 | if (mem_cgroup_disabled()) | |
4511 | return false; | |
4512 | ||
4513 | return lookup_page_cgroup_used(page) != NULL; | |
4514 | } | |
4515 | ||
4516 | void mem_cgroup_print_bad_page(struct page *page) | |
4517 | { | |
4518 | struct page_cgroup *pc; | |
4519 | ||
4520 | pc = lookup_page_cgroup_used(page); | |
4521 | if (pc) { | |
d045197f AM |
4522 | pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n", |
4523 | pc, pc->flags, pc->mem_cgroup); | |
f212ad7c DN |
4524 | } |
4525 | } | |
4526 | #endif | |
4527 | ||
d38d2a75 | 4528 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 4529 | unsigned long long val) |
628f4235 | 4530 | { |
81d39c20 | 4531 | int retry_count; |
3c11ecf4 | 4532 | u64 memswlimit, memlimit; |
628f4235 | 4533 | int ret = 0; |
81d39c20 KH |
4534 | int children = mem_cgroup_count_children(memcg); |
4535 | u64 curusage, oldusage; | |
3c11ecf4 | 4536 | int enlarge; |
81d39c20 KH |
4537 | |
4538 | /* | |
4539 | * For keeping hierarchical_reclaim simple, how long we should retry | |
4540 | * is depends on callers. We set our retry-count to be function | |
4541 | * of # of children which we should visit in this loop. | |
4542 | */ | |
4543 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
4544 | ||
4545 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 4546 | |
3c11ecf4 | 4547 | enlarge = 0; |
8c7c6e34 | 4548 | while (retry_count) { |
628f4235 KH |
4549 | if (signal_pending(current)) { |
4550 | ret = -EINTR; | |
4551 | break; | |
4552 | } | |
8c7c6e34 KH |
4553 | /* |
4554 | * Rather than hide all in some function, I do this in | |
4555 | * open coded manner. You see what this really does. | |
aaad153e | 4556 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
4557 | */ |
4558 | mutex_lock(&set_limit_mutex); | |
4559 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
4560 | if (memswlimit < val) { | |
4561 | ret = -EINVAL; | |
4562 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
4563 | break; |
4564 | } | |
3c11ecf4 KH |
4565 | |
4566 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4567 | if (memlimit < val) | |
4568 | enlarge = 1; | |
4569 | ||
8c7c6e34 | 4570 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
4571 | if (!ret) { |
4572 | if (memswlimit == val) | |
4573 | memcg->memsw_is_minimum = true; | |
4574 | else | |
4575 | memcg->memsw_is_minimum = false; | |
4576 | } | |
8c7c6e34 KH |
4577 | mutex_unlock(&set_limit_mutex); |
4578 | ||
4579 | if (!ret) | |
4580 | break; | |
4581 | ||
5660048c JW |
4582 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
4583 | MEM_CGROUP_RECLAIM_SHRINK); | |
81d39c20 KH |
4584 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
4585 | /* Usage is reduced ? */ | |
f894ffa8 | 4586 | if (curusage >= oldusage) |
81d39c20 KH |
4587 | retry_count--; |
4588 | else | |
4589 | oldusage = curusage; | |
8c7c6e34 | 4590 | } |
3c11ecf4 KH |
4591 | if (!ret && enlarge) |
4592 | memcg_oom_recover(memcg); | |
14797e23 | 4593 | |
8c7c6e34 KH |
4594 | return ret; |
4595 | } | |
4596 | ||
338c8431 LZ |
4597 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
4598 | unsigned long long val) | |
8c7c6e34 | 4599 | { |
81d39c20 | 4600 | int retry_count; |
3c11ecf4 | 4601 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
4602 | int children = mem_cgroup_count_children(memcg); |
4603 | int ret = -EBUSY; | |
3c11ecf4 | 4604 | int enlarge = 0; |
8c7c6e34 | 4605 | |
81d39c20 | 4606 | /* see mem_cgroup_resize_res_limit */ |
f894ffa8 | 4607 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; |
81d39c20 | 4608 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
8c7c6e34 KH |
4609 | while (retry_count) { |
4610 | if (signal_pending(current)) { | |
4611 | ret = -EINTR; | |
4612 | break; | |
4613 | } | |
4614 | /* | |
4615 | * Rather than hide all in some function, I do this in | |
4616 | * open coded manner. You see what this really does. | |
aaad153e | 4617 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
4618 | */ |
4619 | mutex_lock(&set_limit_mutex); | |
4620 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4621 | if (memlimit > val) { | |
4622 | ret = -EINVAL; | |
4623 | mutex_unlock(&set_limit_mutex); | |
4624 | break; | |
4625 | } | |
3c11ecf4 KH |
4626 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
4627 | if (memswlimit < val) | |
4628 | enlarge = 1; | |
8c7c6e34 | 4629 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
4630 | if (!ret) { |
4631 | if (memlimit == val) | |
4632 | memcg->memsw_is_minimum = true; | |
4633 | else | |
4634 | memcg->memsw_is_minimum = false; | |
4635 | } | |
8c7c6e34 KH |
4636 | mutex_unlock(&set_limit_mutex); |
4637 | ||
4638 | if (!ret) | |
4639 | break; | |
4640 | ||
5660048c JW |
4641 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
4642 | MEM_CGROUP_RECLAIM_NOSWAP | | |
4643 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 4644 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 4645 | /* Usage is reduced ? */ |
8c7c6e34 | 4646 | if (curusage >= oldusage) |
628f4235 | 4647 | retry_count--; |
81d39c20 KH |
4648 | else |
4649 | oldusage = curusage; | |
628f4235 | 4650 | } |
3c11ecf4 KH |
4651 | if (!ret && enlarge) |
4652 | memcg_oom_recover(memcg); | |
628f4235 KH |
4653 | return ret; |
4654 | } | |
4655 | ||
2ef37d3f MH |
4656 | /** |
4657 | * mem_cgroup_force_empty_list - clears LRU of a group | |
4658 | * @memcg: group to clear | |
4659 | * @node: NUMA node | |
4660 | * @zid: zone id | |
4661 | * @lru: lru to to clear | |
4662 | * | |
3c935d18 | 4663 | * Traverse a specified page_cgroup list and try to drop them all. This doesn't |
2ef37d3f MH |
4664 | * reclaim the pages page themselves - pages are moved to the parent (or root) |
4665 | * group. | |
cc847582 | 4666 | */ |
2ef37d3f | 4667 | static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
08e552c6 | 4668 | int node, int zid, enum lru_list lru) |
cc847582 | 4669 | { |
bea8c150 | 4670 | struct lruvec *lruvec; |
2ef37d3f | 4671 | unsigned long flags; |
072c56c1 | 4672 | struct list_head *list; |
925b7673 JW |
4673 | struct page *busy; |
4674 | struct zone *zone; | |
072c56c1 | 4675 | |
08e552c6 | 4676 | zone = &NODE_DATA(node)->node_zones[zid]; |
bea8c150 HD |
4677 | lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
4678 | list = &lruvec->lists[lru]; | |
cc847582 | 4679 | |
f817ed48 | 4680 | busy = NULL; |
2ef37d3f | 4681 | do { |
925b7673 | 4682 | struct page_cgroup *pc; |
5564e88b JW |
4683 | struct page *page; |
4684 | ||
08e552c6 | 4685 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 4686 | if (list_empty(list)) { |
08e552c6 | 4687 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 4688 | break; |
f817ed48 | 4689 | } |
925b7673 JW |
4690 | page = list_entry(list->prev, struct page, lru); |
4691 | if (busy == page) { | |
4692 | list_move(&page->lru, list); | |
648bcc77 | 4693 | busy = NULL; |
08e552c6 | 4694 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
4695 | continue; |
4696 | } | |
08e552c6 | 4697 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 4698 | |
925b7673 | 4699 | pc = lookup_page_cgroup(page); |
5564e88b | 4700 | |
3c935d18 | 4701 | if (mem_cgroup_move_parent(page, pc, memcg)) { |
f817ed48 | 4702 | /* found lock contention or "pc" is obsolete. */ |
925b7673 | 4703 | busy = page; |
f817ed48 KH |
4704 | cond_resched(); |
4705 | } else | |
4706 | busy = NULL; | |
2ef37d3f | 4707 | } while (!list_empty(list)); |
cc847582 KH |
4708 | } |
4709 | ||
4710 | /* | |
c26251f9 MH |
4711 | * make mem_cgroup's charge to be 0 if there is no task by moving |
4712 | * all the charges and pages to the parent. | |
cc847582 | 4713 | * This enables deleting this mem_cgroup. |
c26251f9 MH |
4714 | * |
4715 | * Caller is responsible for holding css reference on the memcg. | |
cc847582 | 4716 | */ |
ab5196c2 | 4717 | static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg) |
cc847582 | 4718 | { |
c26251f9 | 4719 | int node, zid; |
bea207c8 | 4720 | u64 usage; |
f817ed48 | 4721 | |
fce66477 | 4722 | do { |
52d4b9ac KH |
4723 | /* This is for making all *used* pages to be on LRU. */ |
4724 | lru_add_drain_all(); | |
c0ff4b85 | 4725 | drain_all_stock_sync(memcg); |
c0ff4b85 | 4726 | mem_cgroup_start_move(memcg); |
31aaea4a | 4727 | for_each_node_state(node, N_MEMORY) { |
2ef37d3f | 4728 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
f156ab93 HD |
4729 | enum lru_list lru; |
4730 | for_each_lru(lru) { | |
2ef37d3f | 4731 | mem_cgroup_force_empty_list(memcg, |
f156ab93 | 4732 | node, zid, lru); |
f817ed48 | 4733 | } |
1ecaab2b | 4734 | } |
f817ed48 | 4735 | } |
c0ff4b85 R |
4736 | mem_cgroup_end_move(memcg); |
4737 | memcg_oom_recover(memcg); | |
52d4b9ac | 4738 | cond_resched(); |
f817ed48 | 4739 | |
2ef37d3f | 4740 | /* |
bea207c8 GC |
4741 | * Kernel memory may not necessarily be trackable to a specific |
4742 | * process. So they are not migrated, and therefore we can't | |
4743 | * expect their value to drop to 0 here. | |
4744 | * Having res filled up with kmem only is enough. | |
4745 | * | |
2ef37d3f MH |
4746 | * This is a safety check because mem_cgroup_force_empty_list |
4747 | * could have raced with mem_cgroup_replace_page_cache callers | |
4748 | * so the lru seemed empty but the page could have been added | |
4749 | * right after the check. RES_USAGE should be safe as we always | |
4750 | * charge before adding to the LRU. | |
4751 | */ | |
bea207c8 GC |
4752 | usage = res_counter_read_u64(&memcg->res, RES_USAGE) - |
4753 | res_counter_read_u64(&memcg->kmem, RES_USAGE); | |
4754 | } while (usage > 0); | |
c26251f9 MH |
4755 | } |
4756 | ||
b5f99b53 GC |
4757 | /* |
4758 | * This mainly exists for tests during the setting of set of use_hierarchy. | |
4759 | * Since this is the very setting we are changing, the current hierarchy value | |
4760 | * is meaningless | |
4761 | */ | |
4762 | static inline bool __memcg_has_children(struct mem_cgroup *memcg) | |
4763 | { | |
492eb21b | 4764 | struct cgroup_subsys_state *pos; |
b5f99b53 GC |
4765 | |
4766 | /* bounce at first found */ | |
492eb21b | 4767 | css_for_each_child(pos, &memcg->css) |
b5f99b53 GC |
4768 | return true; |
4769 | return false; | |
4770 | } | |
4771 | ||
4772 | /* | |
0999821b GC |
4773 | * Must be called with memcg_create_mutex held, unless the cgroup is guaranteed |
4774 | * to be already dead (as in mem_cgroup_force_empty, for instance). This is | |
b5f99b53 GC |
4775 | * from mem_cgroup_count_children(), in the sense that we don't really care how |
4776 | * many children we have; we only need to know if we have any. It also counts | |
4777 | * any memcg without hierarchy as infertile. | |
4778 | */ | |
4779 | static inline bool memcg_has_children(struct mem_cgroup *memcg) | |
4780 | { | |
4781 | return memcg->use_hierarchy && __memcg_has_children(memcg); | |
4782 | } | |
4783 | ||
c26251f9 MH |
4784 | /* |
4785 | * Reclaims as many pages from the given memcg as possible and moves | |
4786 | * the rest to the parent. | |
4787 | * | |
4788 | * Caller is responsible for holding css reference for memcg. | |
4789 | */ | |
4790 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
4791 | { | |
4792 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
4793 | struct cgroup *cgrp = memcg->css.cgroup; | |
f817ed48 | 4794 | |
c1e862c1 | 4795 | /* returns EBUSY if there is a task or if we come here twice. */ |
c26251f9 MH |
4796 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
4797 | return -EBUSY; | |
4798 | ||
c1e862c1 KH |
4799 | /* we call try-to-free pages for make this cgroup empty */ |
4800 | lru_add_drain_all(); | |
f817ed48 | 4801 | /* try to free all pages in this cgroup */ |
569530fb | 4802 | while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) { |
f817ed48 | 4803 | int progress; |
c1e862c1 | 4804 | |
c26251f9 MH |
4805 | if (signal_pending(current)) |
4806 | return -EINTR; | |
4807 | ||
c0ff4b85 | 4808 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, |
185efc0f | 4809 | false); |
c1e862c1 | 4810 | if (!progress) { |
f817ed48 | 4811 | nr_retries--; |
c1e862c1 | 4812 | /* maybe some writeback is necessary */ |
8aa7e847 | 4813 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 4814 | } |
f817ed48 KH |
4815 | |
4816 | } | |
08e552c6 | 4817 | lru_add_drain(); |
ab5196c2 MH |
4818 | mem_cgroup_reparent_charges(memcg); |
4819 | ||
4820 | return 0; | |
cc847582 KH |
4821 | } |
4822 | ||
182446d0 TH |
4823 | static int mem_cgroup_force_empty_write(struct cgroup_subsys_state *css, |
4824 | unsigned int event) | |
c1e862c1 | 4825 | { |
182446d0 | 4826 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c26251f9 | 4827 | |
d8423011 MH |
4828 | if (mem_cgroup_is_root(memcg)) |
4829 | return -EINVAL; | |
c33bd835 | 4830 | return mem_cgroup_force_empty(memcg); |
c1e862c1 KH |
4831 | } |
4832 | ||
182446d0 TH |
4833 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
4834 | struct cftype *cft) | |
18f59ea7 | 4835 | { |
182446d0 | 4836 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
4837 | } |
4838 | ||
182446d0 TH |
4839 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
4840 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
4841 | { |
4842 | int retval = 0; | |
182446d0 | 4843 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
63876986 | 4844 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(css_parent(&memcg->css)); |
18f59ea7 | 4845 | |
0999821b | 4846 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
4847 | |
4848 | if (memcg->use_hierarchy == val) | |
4849 | goto out; | |
4850 | ||
18f59ea7 | 4851 | /* |
af901ca1 | 4852 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
4853 | * in the child subtrees. If it is unset, then the change can |
4854 | * occur, provided the current cgroup has no children. | |
4855 | * | |
4856 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
4857 | * set if there are no children. | |
4858 | */ | |
c0ff4b85 | 4859 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 4860 | (val == 1 || val == 0)) { |
b5f99b53 | 4861 | if (!__memcg_has_children(memcg)) |
c0ff4b85 | 4862 | memcg->use_hierarchy = val; |
18f59ea7 BS |
4863 | else |
4864 | retval = -EBUSY; | |
4865 | } else | |
4866 | retval = -EINVAL; | |
567fb435 GC |
4867 | |
4868 | out: | |
0999821b | 4869 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
4870 | |
4871 | return retval; | |
4872 | } | |
4873 | ||
0c3e73e8 | 4874 | |
c0ff4b85 | 4875 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg, |
7a159cc9 | 4876 | enum mem_cgroup_stat_index idx) |
0c3e73e8 | 4877 | { |
7d74b06f | 4878 | struct mem_cgroup *iter; |
7a159cc9 | 4879 | long val = 0; |
0c3e73e8 | 4880 | |
7a159cc9 | 4881 | /* Per-cpu values can be negative, use a signed accumulator */ |
c0ff4b85 | 4882 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f KH |
4883 | val += mem_cgroup_read_stat(iter, idx); |
4884 | ||
4885 | if (val < 0) /* race ? */ | |
4886 | val = 0; | |
4887 | return val; | |
0c3e73e8 BS |
4888 | } |
4889 | ||
c0ff4b85 | 4890 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
104f3928 | 4891 | { |
7d74b06f | 4892 | u64 val; |
104f3928 | 4893 | |
c0ff4b85 | 4894 | if (!mem_cgroup_is_root(memcg)) { |
104f3928 | 4895 | if (!swap) |
65c64ce8 | 4896 | return res_counter_read_u64(&memcg->res, RES_USAGE); |
104f3928 | 4897 | else |
65c64ce8 | 4898 | return res_counter_read_u64(&memcg->memsw, RES_USAGE); |
104f3928 KS |
4899 | } |
4900 | ||
b070e65c DR |
4901 | /* |
4902 | * Transparent hugepages are still accounted for in MEM_CGROUP_STAT_RSS | |
4903 | * as well as in MEM_CGROUP_STAT_RSS_HUGE. | |
4904 | */ | |
c0ff4b85 R |
4905 | val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE); |
4906 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS); | |
104f3928 | 4907 | |
7d74b06f | 4908 | if (swap) |
bff6bb83 | 4909 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP); |
104f3928 KS |
4910 | |
4911 | return val << PAGE_SHIFT; | |
4912 | } | |
4913 | ||
182446d0 TH |
4914 | static ssize_t mem_cgroup_read(struct cgroup_subsys_state *css, |
4915 | struct cftype *cft, struct file *file, | |
4916 | char __user *buf, size_t nbytes, loff_t *ppos) | |
8cdea7c0 | 4917 | { |
182446d0 | 4918 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
af36f906 | 4919 | char str[64]; |
104f3928 | 4920 | u64 val; |
86ae53e1 GC |
4921 | int name, len; |
4922 | enum res_type type; | |
8c7c6e34 KH |
4923 | |
4924 | type = MEMFILE_TYPE(cft->private); | |
4925 | name = MEMFILE_ATTR(cft->private); | |
af36f906 | 4926 | |
8c7c6e34 KH |
4927 | switch (type) { |
4928 | case _MEM: | |
104f3928 | 4929 | if (name == RES_USAGE) |
c0ff4b85 | 4930 | val = mem_cgroup_usage(memcg, false); |
104f3928 | 4931 | else |
c0ff4b85 | 4932 | val = res_counter_read_u64(&memcg->res, name); |
8c7c6e34 KH |
4933 | break; |
4934 | case _MEMSWAP: | |
104f3928 | 4935 | if (name == RES_USAGE) |
c0ff4b85 | 4936 | val = mem_cgroup_usage(memcg, true); |
104f3928 | 4937 | else |
c0ff4b85 | 4938 | val = res_counter_read_u64(&memcg->memsw, name); |
8c7c6e34 | 4939 | break; |
510fc4e1 GC |
4940 | case _KMEM: |
4941 | val = res_counter_read_u64(&memcg->kmem, name); | |
4942 | break; | |
8c7c6e34 KH |
4943 | default: |
4944 | BUG(); | |
8c7c6e34 | 4945 | } |
af36f906 TH |
4946 | |
4947 | len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val); | |
4948 | return simple_read_from_buffer(buf, nbytes, ppos, str, len); | |
8cdea7c0 | 4949 | } |
510fc4e1 | 4950 | |
182446d0 | 4951 | static int memcg_update_kmem_limit(struct cgroup_subsys_state *css, u64 val) |
510fc4e1 GC |
4952 | { |
4953 | int ret = -EINVAL; | |
4954 | #ifdef CONFIG_MEMCG_KMEM | |
182446d0 | 4955 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
510fc4e1 GC |
4956 | /* |
4957 | * For simplicity, we won't allow this to be disabled. It also can't | |
4958 | * be changed if the cgroup has children already, or if tasks had | |
4959 | * already joined. | |
4960 | * | |
4961 | * If tasks join before we set the limit, a person looking at | |
4962 | * kmem.usage_in_bytes will have no way to determine when it took | |
4963 | * place, which makes the value quite meaningless. | |
4964 | * | |
4965 | * After it first became limited, changes in the value of the limit are | |
4966 | * of course permitted. | |
510fc4e1 | 4967 | */ |
0999821b | 4968 | mutex_lock(&memcg_create_mutex); |
510fc4e1 | 4969 | mutex_lock(&set_limit_mutex); |
6de5a8bf | 4970 | if (!memcg->kmem_account_flags && val != RES_COUNTER_MAX) { |
182446d0 | 4971 | if (cgroup_task_count(css->cgroup) || memcg_has_children(memcg)) { |
510fc4e1 GC |
4972 | ret = -EBUSY; |
4973 | goto out; | |
4974 | } | |
4975 | ret = res_counter_set_limit(&memcg->kmem, val); | |
4976 | VM_BUG_ON(ret); | |
4977 | ||
55007d84 GC |
4978 | ret = memcg_update_cache_sizes(memcg); |
4979 | if (ret) { | |
6de5a8bf | 4980 | res_counter_set_limit(&memcg->kmem, RES_COUNTER_MAX); |
55007d84 GC |
4981 | goto out; |
4982 | } | |
692e89ab GC |
4983 | static_key_slow_inc(&memcg_kmem_enabled_key); |
4984 | /* | |
4985 | * setting the active bit after the inc will guarantee no one | |
4986 | * starts accounting before all call sites are patched | |
4987 | */ | |
4988 | memcg_kmem_set_active(memcg); | |
510fc4e1 GC |
4989 | } else |
4990 | ret = res_counter_set_limit(&memcg->kmem, val); | |
4991 | out: | |
4992 | mutex_unlock(&set_limit_mutex); | |
0999821b | 4993 | mutex_unlock(&memcg_create_mutex); |
510fc4e1 GC |
4994 | #endif |
4995 | return ret; | |
4996 | } | |
4997 | ||
6d043990 | 4998 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 | 4999 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 5000 | { |
55007d84 | 5001 | int ret = 0; |
510fc4e1 GC |
5002 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
5003 | if (!parent) | |
55007d84 GC |
5004 | goto out; |
5005 | ||
510fc4e1 | 5006 | memcg->kmem_account_flags = parent->kmem_account_flags; |
a8964b9b GC |
5007 | /* |
5008 | * When that happen, we need to disable the static branch only on those | |
5009 | * memcgs that enabled it. To achieve this, we would be forced to | |
5010 | * complicate the code by keeping track of which memcgs were the ones | |
5011 | * that actually enabled limits, and which ones got it from its | |
5012 | * parents. | |
5013 | * | |
5014 | * It is a lot simpler just to do static_key_slow_inc() on every child | |
5015 | * that is accounted. | |
5016 | */ | |
55007d84 GC |
5017 | if (!memcg_kmem_is_active(memcg)) |
5018 | goto out; | |
5019 | ||
5020 | /* | |
10d5ebf4 LZ |
5021 | * __mem_cgroup_free() will issue static_key_slow_dec() because this |
5022 | * memcg is active already. If the later initialization fails then the | |
5023 | * cgroup core triggers the cleanup so we do not have to do it here. | |
55007d84 | 5024 | */ |
55007d84 GC |
5025 | static_key_slow_inc(&memcg_kmem_enabled_key); |
5026 | ||
5027 | mutex_lock(&set_limit_mutex); | |
425c598d | 5028 | memcg_stop_kmem_account(); |
55007d84 | 5029 | ret = memcg_update_cache_sizes(memcg); |
425c598d | 5030 | memcg_resume_kmem_account(); |
55007d84 | 5031 | mutex_unlock(&set_limit_mutex); |
55007d84 GC |
5032 | out: |
5033 | return ret; | |
510fc4e1 | 5034 | } |
6d043990 | 5035 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 5036 | |
628f4235 KH |
5037 | /* |
5038 | * The user of this function is... | |
5039 | * RES_LIMIT. | |
5040 | */ | |
182446d0 | 5041 | static int mem_cgroup_write(struct cgroup_subsys_state *css, struct cftype *cft, |
856c13aa | 5042 | const char *buffer) |
8cdea7c0 | 5043 | { |
182446d0 | 5044 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
86ae53e1 GC |
5045 | enum res_type type; |
5046 | int name; | |
628f4235 KH |
5047 | unsigned long long val; |
5048 | int ret; | |
5049 | ||
8c7c6e34 KH |
5050 | type = MEMFILE_TYPE(cft->private); |
5051 | name = MEMFILE_ATTR(cft->private); | |
af36f906 | 5052 | |
8c7c6e34 | 5053 | switch (name) { |
628f4235 | 5054 | case RES_LIMIT: |
4b3bde4c BS |
5055 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
5056 | ret = -EINVAL; | |
5057 | break; | |
5058 | } | |
628f4235 KH |
5059 | /* This function does all necessary parse...reuse it */ |
5060 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
5061 | if (ret) |
5062 | break; | |
5063 | if (type == _MEM) | |
628f4235 | 5064 | ret = mem_cgroup_resize_limit(memcg, val); |
510fc4e1 | 5065 | else if (type == _MEMSWAP) |
8c7c6e34 | 5066 | ret = mem_cgroup_resize_memsw_limit(memcg, val); |
510fc4e1 | 5067 | else if (type == _KMEM) |
182446d0 | 5068 | ret = memcg_update_kmem_limit(css, val); |
510fc4e1 GC |
5069 | else |
5070 | return -EINVAL; | |
628f4235 | 5071 | break; |
296c81d8 BS |
5072 | case RES_SOFT_LIMIT: |
5073 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
5074 | if (ret) | |
5075 | break; | |
5076 | /* | |
5077 | * For memsw, soft limits are hard to implement in terms | |
5078 | * of semantics, for now, we support soft limits for | |
5079 | * control without swap | |
5080 | */ | |
5081 | if (type == _MEM) | |
5082 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
5083 | else | |
5084 | ret = -EINVAL; | |
5085 | break; | |
628f4235 KH |
5086 | default: |
5087 | ret = -EINVAL; /* should be BUG() ? */ | |
5088 | break; | |
5089 | } | |
5090 | return ret; | |
8cdea7c0 BS |
5091 | } |
5092 | ||
fee7b548 KH |
5093 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
5094 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
5095 | { | |
fee7b548 KH |
5096 | unsigned long long min_limit, min_memsw_limit, tmp; |
5097 | ||
5098 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
5099 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
fee7b548 KH |
5100 | if (!memcg->use_hierarchy) |
5101 | goto out; | |
5102 | ||
63876986 TH |
5103 | while (css_parent(&memcg->css)) { |
5104 | memcg = mem_cgroup_from_css(css_parent(&memcg->css)); | |
fee7b548 KH |
5105 | if (!memcg->use_hierarchy) |
5106 | break; | |
5107 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
5108 | min_limit = min(min_limit, tmp); | |
5109 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
5110 | min_memsw_limit = min(min_memsw_limit, tmp); | |
5111 | } | |
5112 | out: | |
5113 | *mem_limit = min_limit; | |
5114 | *memsw_limit = min_memsw_limit; | |
fee7b548 KH |
5115 | } |
5116 | ||
182446d0 | 5117 | static int mem_cgroup_reset(struct cgroup_subsys_state *css, unsigned int event) |
c84872e1 | 5118 | { |
182446d0 | 5119 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
86ae53e1 GC |
5120 | int name; |
5121 | enum res_type type; | |
c84872e1 | 5122 | |
8c7c6e34 KH |
5123 | type = MEMFILE_TYPE(event); |
5124 | name = MEMFILE_ATTR(event); | |
af36f906 | 5125 | |
8c7c6e34 | 5126 | switch (name) { |
29f2a4da | 5127 | case RES_MAX_USAGE: |
8c7c6e34 | 5128 | if (type == _MEM) |
c0ff4b85 | 5129 | res_counter_reset_max(&memcg->res); |
510fc4e1 | 5130 | else if (type == _MEMSWAP) |
c0ff4b85 | 5131 | res_counter_reset_max(&memcg->memsw); |
510fc4e1 GC |
5132 | else if (type == _KMEM) |
5133 | res_counter_reset_max(&memcg->kmem); | |
5134 | else | |
5135 | return -EINVAL; | |
29f2a4da PE |
5136 | break; |
5137 | case RES_FAILCNT: | |
8c7c6e34 | 5138 | if (type == _MEM) |
c0ff4b85 | 5139 | res_counter_reset_failcnt(&memcg->res); |
510fc4e1 | 5140 | else if (type == _MEMSWAP) |
c0ff4b85 | 5141 | res_counter_reset_failcnt(&memcg->memsw); |
510fc4e1 GC |
5142 | else if (type == _KMEM) |
5143 | res_counter_reset_failcnt(&memcg->kmem); | |
5144 | else | |
5145 | return -EINVAL; | |
29f2a4da PE |
5146 | break; |
5147 | } | |
f64c3f54 | 5148 | |
85cc59db | 5149 | return 0; |
c84872e1 PE |
5150 | } |
5151 | ||
182446d0 | 5152 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
5153 | struct cftype *cft) |
5154 | { | |
182446d0 | 5155 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
5156 | } |
5157 | ||
02491447 | 5158 | #ifdef CONFIG_MMU |
182446d0 | 5159 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
5160 | struct cftype *cft, u64 val) |
5161 | { | |
182446d0 | 5162 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 DN |
5163 | |
5164 | if (val >= (1 << NR_MOVE_TYPE)) | |
5165 | return -EINVAL; | |
ee5e8472 | 5166 | |
7dc74be0 | 5167 | /* |
ee5e8472 GC |
5168 | * No kind of locking is needed in here, because ->can_attach() will |
5169 | * check this value once in the beginning of the process, and then carry | |
5170 | * on with stale data. This means that changes to this value will only | |
5171 | * affect task migrations starting after the change. | |
7dc74be0 | 5172 | */ |
c0ff4b85 | 5173 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
5174 | return 0; |
5175 | } | |
02491447 | 5176 | #else |
182446d0 | 5177 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
5178 | struct cftype *cft, u64 val) |
5179 | { | |
5180 | return -ENOSYS; | |
5181 | } | |
5182 | #endif | |
7dc74be0 | 5183 | |
406eb0c9 | 5184 | #ifdef CONFIG_NUMA |
182446d0 TH |
5185 | static int memcg_numa_stat_show(struct cgroup_subsys_state *css, |
5186 | struct cftype *cft, struct seq_file *m) | |
406eb0c9 YH |
5187 | { |
5188 | int nid; | |
5189 | unsigned long total_nr, file_nr, anon_nr, unevictable_nr; | |
5190 | unsigned long node_nr; | |
182446d0 | 5191 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
406eb0c9 | 5192 | |
d79154bb | 5193 | total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL); |
406eb0c9 | 5194 | seq_printf(m, "total=%lu", total_nr); |
31aaea4a | 5195 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5196 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL); |
406eb0c9 YH |
5197 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5198 | } | |
5199 | seq_putc(m, '\n'); | |
5200 | ||
d79154bb | 5201 | file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE); |
406eb0c9 | 5202 | seq_printf(m, "file=%lu", file_nr); |
31aaea4a | 5203 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5204 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 5205 | LRU_ALL_FILE); |
406eb0c9 YH |
5206 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5207 | } | |
5208 | seq_putc(m, '\n'); | |
5209 | ||
d79154bb | 5210 | anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON); |
406eb0c9 | 5211 | seq_printf(m, "anon=%lu", anon_nr); |
31aaea4a | 5212 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5213 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 5214 | LRU_ALL_ANON); |
406eb0c9 YH |
5215 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5216 | } | |
5217 | seq_putc(m, '\n'); | |
5218 | ||
d79154bb | 5219 | unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE)); |
406eb0c9 | 5220 | seq_printf(m, "unevictable=%lu", unevictable_nr); |
31aaea4a | 5221 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5222 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 5223 | BIT(LRU_UNEVICTABLE)); |
406eb0c9 YH |
5224 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5225 | } | |
5226 | seq_putc(m, '\n'); | |
5227 | return 0; | |
5228 | } | |
5229 | #endif /* CONFIG_NUMA */ | |
5230 | ||
af7c4b0e JW |
5231 | static inline void mem_cgroup_lru_names_not_uptodate(void) |
5232 | { | |
5233 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); | |
5234 | } | |
5235 | ||
182446d0 | 5236 | static int memcg_stat_show(struct cgroup_subsys_state *css, struct cftype *cft, |
78ccf5b5 | 5237 | struct seq_file *m) |
d2ceb9b7 | 5238 | { |
182446d0 | 5239 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
af7c4b0e JW |
5240 | struct mem_cgroup *mi; |
5241 | unsigned int i; | |
406eb0c9 | 5242 | |
af7c4b0e | 5243 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 5244 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 5245 | continue; |
af7c4b0e JW |
5246 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
5247 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 5248 | } |
7b854121 | 5249 | |
af7c4b0e JW |
5250 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
5251 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
5252 | mem_cgroup_read_events(memcg, i)); | |
5253 | ||
5254 | for (i = 0; i < NR_LRU_LISTS; i++) | |
5255 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
5256 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
5257 | ||
14067bb3 | 5258 | /* Hierarchical information */ |
fee7b548 KH |
5259 | { |
5260 | unsigned long long limit, memsw_limit; | |
d79154bb | 5261 | memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit); |
78ccf5b5 | 5262 | seq_printf(m, "hierarchical_memory_limit %llu\n", limit); |
fee7b548 | 5263 | if (do_swap_account) |
78ccf5b5 JW |
5264 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
5265 | memsw_limit); | |
fee7b548 | 5266 | } |
7f016ee8 | 5267 | |
af7c4b0e JW |
5268 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
5269 | long long val = 0; | |
5270 | ||
bff6bb83 | 5271 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 5272 | continue; |
af7c4b0e JW |
5273 | for_each_mem_cgroup_tree(mi, memcg) |
5274 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
5275 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
5276 | } | |
5277 | ||
5278 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
5279 | unsigned long long val = 0; | |
5280 | ||
5281 | for_each_mem_cgroup_tree(mi, memcg) | |
5282 | val += mem_cgroup_read_events(mi, i); | |
5283 | seq_printf(m, "total_%s %llu\n", | |
5284 | mem_cgroup_events_names[i], val); | |
5285 | } | |
5286 | ||
5287 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
5288 | unsigned long long val = 0; | |
5289 | ||
5290 | for_each_mem_cgroup_tree(mi, memcg) | |
5291 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
5292 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 5293 | } |
14067bb3 | 5294 | |
7f016ee8 | 5295 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
5296 | { |
5297 | int nid, zid; | |
5298 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 5299 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
5300 | unsigned long recent_rotated[2] = {0, 0}; |
5301 | unsigned long recent_scanned[2] = {0, 0}; | |
5302 | ||
5303 | for_each_online_node(nid) | |
5304 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
d79154bb | 5305 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
89abfab1 | 5306 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 5307 | |
89abfab1 HD |
5308 | recent_rotated[0] += rstat->recent_rotated[0]; |
5309 | recent_rotated[1] += rstat->recent_rotated[1]; | |
5310 | recent_scanned[0] += rstat->recent_scanned[0]; | |
5311 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 5312 | } |
78ccf5b5 JW |
5313 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
5314 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
5315 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
5316 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
5317 | } |
5318 | #endif | |
5319 | ||
d2ceb9b7 KH |
5320 | return 0; |
5321 | } | |
5322 | ||
182446d0 TH |
5323 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
5324 | struct cftype *cft) | |
a7885eb8 | 5325 | { |
182446d0 | 5326 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 5327 | |
1f4c025b | 5328 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
5329 | } |
5330 | ||
182446d0 TH |
5331 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
5332 | struct cftype *cft, u64 val) | |
a7885eb8 | 5333 | { |
182446d0 | 5334 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
63876986 | 5335 | struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css)); |
a7885eb8 | 5336 | |
63876986 | 5337 | if (val > 100 || !parent) |
a7885eb8 KM |
5338 | return -EINVAL; |
5339 | ||
0999821b | 5340 | mutex_lock(&memcg_create_mutex); |
068b38c1 | 5341 | |
a7885eb8 | 5342 | /* If under hierarchy, only empty-root can set this value */ |
b5f99b53 | 5343 | if ((parent->use_hierarchy) || memcg_has_children(memcg)) { |
0999821b | 5344 | mutex_unlock(&memcg_create_mutex); |
a7885eb8 | 5345 | return -EINVAL; |
068b38c1 | 5346 | } |
a7885eb8 | 5347 | |
a7885eb8 | 5348 | memcg->swappiness = val; |
a7885eb8 | 5349 | |
0999821b | 5350 | mutex_unlock(&memcg_create_mutex); |
068b38c1 | 5351 | |
a7885eb8 KM |
5352 | return 0; |
5353 | } | |
5354 | ||
2e72b634 KS |
5355 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
5356 | { | |
5357 | struct mem_cgroup_threshold_ary *t; | |
5358 | u64 usage; | |
5359 | int i; | |
5360 | ||
5361 | rcu_read_lock(); | |
5362 | if (!swap) | |
2c488db2 | 5363 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 5364 | else |
2c488db2 | 5365 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
5366 | |
5367 | if (!t) | |
5368 | goto unlock; | |
5369 | ||
5370 | usage = mem_cgroup_usage(memcg, swap); | |
5371 | ||
5372 | /* | |
748dad36 | 5373 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
5374 | * If it's not true, a threshold was crossed after last |
5375 | * call of __mem_cgroup_threshold(). | |
5376 | */ | |
5407a562 | 5377 | i = t->current_threshold; |
2e72b634 KS |
5378 | |
5379 | /* | |
5380 | * Iterate backward over array of thresholds starting from | |
5381 | * current_threshold and check if a threshold is crossed. | |
5382 | * If none of thresholds below usage is crossed, we read | |
5383 | * only one element of the array here. | |
5384 | */ | |
5385 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
5386 | eventfd_signal(t->entries[i].eventfd, 1); | |
5387 | ||
5388 | /* i = current_threshold + 1 */ | |
5389 | i++; | |
5390 | ||
5391 | /* | |
5392 | * Iterate forward over array of thresholds starting from | |
5393 | * current_threshold+1 and check if a threshold is crossed. | |
5394 | * If none of thresholds above usage is crossed, we read | |
5395 | * only one element of the array here. | |
5396 | */ | |
5397 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
5398 | eventfd_signal(t->entries[i].eventfd, 1); | |
5399 | ||
5400 | /* Update current_threshold */ | |
5407a562 | 5401 | t->current_threshold = i - 1; |
2e72b634 KS |
5402 | unlock: |
5403 | rcu_read_unlock(); | |
5404 | } | |
5405 | ||
5406 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
5407 | { | |
ad4ca5f4 KS |
5408 | while (memcg) { |
5409 | __mem_cgroup_threshold(memcg, false); | |
5410 | if (do_swap_account) | |
5411 | __mem_cgroup_threshold(memcg, true); | |
5412 | ||
5413 | memcg = parent_mem_cgroup(memcg); | |
5414 | } | |
2e72b634 KS |
5415 | } |
5416 | ||
5417 | static int compare_thresholds(const void *a, const void *b) | |
5418 | { | |
5419 | const struct mem_cgroup_threshold *_a = a; | |
5420 | const struct mem_cgroup_threshold *_b = b; | |
5421 | ||
2bff24a3 GT |
5422 | if (_a->threshold > _b->threshold) |
5423 | return 1; | |
5424 | ||
5425 | if (_a->threshold < _b->threshold) | |
5426 | return -1; | |
5427 | ||
5428 | return 0; | |
2e72b634 KS |
5429 | } |
5430 | ||
c0ff4b85 | 5431 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
5432 | { |
5433 | struct mem_cgroup_eventfd_list *ev; | |
5434 | ||
c0ff4b85 | 5435 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 KH |
5436 | eventfd_signal(ev->eventfd, 1); |
5437 | return 0; | |
5438 | } | |
5439 | ||
c0ff4b85 | 5440 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 5441 | { |
7d74b06f KH |
5442 | struct mem_cgroup *iter; |
5443 | ||
c0ff4b85 | 5444 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 5445 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
5446 | } |
5447 | ||
81eeaf04 | 5448 | static int mem_cgroup_usage_register_event(struct cgroup_subsys_state *css, |
9490ff27 | 5449 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) |
2e72b634 | 5450 | { |
81eeaf04 | 5451 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
2c488db2 KS |
5452 | struct mem_cgroup_thresholds *thresholds; |
5453 | struct mem_cgroup_threshold_ary *new; | |
86ae53e1 | 5454 | enum res_type type = MEMFILE_TYPE(cft->private); |
2e72b634 | 5455 | u64 threshold, usage; |
2c488db2 | 5456 | int i, size, ret; |
2e72b634 KS |
5457 | |
5458 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
5459 | if (ret) | |
5460 | return ret; | |
5461 | ||
5462 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 5463 | |
2e72b634 | 5464 | if (type == _MEM) |
2c488db2 | 5465 | thresholds = &memcg->thresholds; |
2e72b634 | 5466 | else if (type == _MEMSWAP) |
2c488db2 | 5467 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
5468 | else |
5469 | BUG(); | |
5470 | ||
5471 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
5472 | ||
5473 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 5474 | if (thresholds->primary) |
2e72b634 KS |
5475 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
5476 | ||
2c488db2 | 5477 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
5478 | |
5479 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 5480 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 5481 | GFP_KERNEL); |
2c488db2 | 5482 | if (!new) { |
2e72b634 KS |
5483 | ret = -ENOMEM; |
5484 | goto unlock; | |
5485 | } | |
2c488db2 | 5486 | new->size = size; |
2e72b634 KS |
5487 | |
5488 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
5489 | if (thresholds->primary) { |
5490 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 5491 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
5492 | } |
5493 | ||
2e72b634 | 5494 | /* Add new threshold */ |
2c488db2 KS |
5495 | new->entries[size - 1].eventfd = eventfd; |
5496 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
5497 | |
5498 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 5499 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
5500 | compare_thresholds, NULL); |
5501 | ||
5502 | /* Find current threshold */ | |
2c488db2 | 5503 | new->current_threshold = -1; |
2e72b634 | 5504 | for (i = 0; i < size; i++) { |
748dad36 | 5505 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 5506 | /* |
2c488db2 KS |
5507 | * new->current_threshold will not be used until |
5508 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
5509 | * it here. |
5510 | */ | |
2c488db2 | 5511 | ++new->current_threshold; |
748dad36 SZ |
5512 | } else |
5513 | break; | |
2e72b634 KS |
5514 | } |
5515 | ||
2c488db2 KS |
5516 | /* Free old spare buffer and save old primary buffer as spare */ |
5517 | kfree(thresholds->spare); | |
5518 | thresholds->spare = thresholds->primary; | |
5519 | ||
5520 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 5521 | |
907860ed | 5522 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
5523 | synchronize_rcu(); |
5524 | ||
2e72b634 KS |
5525 | unlock: |
5526 | mutex_unlock(&memcg->thresholds_lock); | |
5527 | ||
5528 | return ret; | |
5529 | } | |
5530 | ||
81eeaf04 | 5531 | static void mem_cgroup_usage_unregister_event(struct cgroup_subsys_state *css, |
9490ff27 | 5532 | struct cftype *cft, struct eventfd_ctx *eventfd) |
2e72b634 | 5533 | { |
81eeaf04 | 5534 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
2c488db2 KS |
5535 | struct mem_cgroup_thresholds *thresholds; |
5536 | struct mem_cgroup_threshold_ary *new; | |
86ae53e1 | 5537 | enum res_type type = MEMFILE_TYPE(cft->private); |
2e72b634 | 5538 | u64 usage; |
2c488db2 | 5539 | int i, j, size; |
2e72b634 KS |
5540 | |
5541 | mutex_lock(&memcg->thresholds_lock); | |
5542 | if (type == _MEM) | |
2c488db2 | 5543 | thresholds = &memcg->thresholds; |
2e72b634 | 5544 | else if (type == _MEMSWAP) |
2c488db2 | 5545 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
5546 | else |
5547 | BUG(); | |
5548 | ||
371528ca AV |
5549 | if (!thresholds->primary) |
5550 | goto unlock; | |
5551 | ||
2e72b634 KS |
5552 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); |
5553 | ||
5554 | /* Check if a threshold crossed before removing */ | |
5555 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
5556 | ||
5557 | /* Calculate new number of threshold */ | |
2c488db2 KS |
5558 | size = 0; |
5559 | for (i = 0; i < thresholds->primary->size; i++) { | |
5560 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
5561 | size++; |
5562 | } | |
5563 | ||
2c488db2 | 5564 | new = thresholds->spare; |
907860ed | 5565 | |
2e72b634 KS |
5566 | /* Set thresholds array to NULL if we don't have thresholds */ |
5567 | if (!size) { | |
2c488db2 KS |
5568 | kfree(new); |
5569 | new = NULL; | |
907860ed | 5570 | goto swap_buffers; |
2e72b634 KS |
5571 | } |
5572 | ||
2c488db2 | 5573 | new->size = size; |
2e72b634 KS |
5574 | |
5575 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
5576 | new->current_threshold = -1; |
5577 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
5578 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
5579 | continue; |
5580 | ||
2c488db2 | 5581 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 5582 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 5583 | /* |
2c488db2 | 5584 | * new->current_threshold will not be used |
2e72b634 KS |
5585 | * until rcu_assign_pointer(), so it's safe to increment |
5586 | * it here. | |
5587 | */ | |
2c488db2 | 5588 | ++new->current_threshold; |
2e72b634 KS |
5589 | } |
5590 | j++; | |
5591 | } | |
5592 | ||
907860ed | 5593 | swap_buffers: |
2c488db2 KS |
5594 | /* Swap primary and spare array */ |
5595 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
5596 | /* If all events are unregistered, free the spare array */ |
5597 | if (!new) { | |
5598 | kfree(thresholds->spare); | |
5599 | thresholds->spare = NULL; | |
5600 | } | |
5601 | ||
2c488db2 | 5602 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 5603 | |
907860ed | 5604 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 5605 | synchronize_rcu(); |
371528ca | 5606 | unlock: |
2e72b634 | 5607 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 5608 | } |
c1e862c1 | 5609 | |
81eeaf04 | 5610 | static int mem_cgroup_oom_register_event(struct cgroup_subsys_state *css, |
9490ff27 KH |
5611 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) |
5612 | { | |
81eeaf04 | 5613 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
9490ff27 | 5614 | struct mem_cgroup_eventfd_list *event; |
86ae53e1 | 5615 | enum res_type type = MEMFILE_TYPE(cft->private); |
9490ff27 KH |
5616 | |
5617 | BUG_ON(type != _OOM_TYPE); | |
5618 | event = kmalloc(sizeof(*event), GFP_KERNEL); | |
5619 | if (!event) | |
5620 | return -ENOMEM; | |
5621 | ||
1af8efe9 | 5622 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
5623 | |
5624 | event->eventfd = eventfd; | |
5625 | list_add(&event->list, &memcg->oom_notify); | |
5626 | ||
5627 | /* already in OOM ? */ | |
79dfdacc | 5628 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 5629 | eventfd_signal(eventfd, 1); |
1af8efe9 | 5630 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
5631 | |
5632 | return 0; | |
5633 | } | |
5634 | ||
81eeaf04 | 5635 | static void mem_cgroup_oom_unregister_event(struct cgroup_subsys_state *css, |
9490ff27 KH |
5636 | struct cftype *cft, struct eventfd_ctx *eventfd) |
5637 | { | |
81eeaf04 | 5638 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
9490ff27 | 5639 | struct mem_cgroup_eventfd_list *ev, *tmp; |
86ae53e1 | 5640 | enum res_type type = MEMFILE_TYPE(cft->private); |
9490ff27 KH |
5641 | |
5642 | BUG_ON(type != _OOM_TYPE); | |
5643 | ||
1af8efe9 | 5644 | spin_lock(&memcg_oom_lock); |
9490ff27 | 5645 | |
c0ff4b85 | 5646 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
5647 | if (ev->eventfd == eventfd) { |
5648 | list_del(&ev->list); | |
5649 | kfree(ev); | |
5650 | } | |
5651 | } | |
5652 | ||
1af8efe9 | 5653 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
5654 | } |
5655 | ||
182446d0 | 5656 | static int mem_cgroup_oom_control_read(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
5657 | struct cftype *cft, struct cgroup_map_cb *cb) |
5658 | { | |
182446d0 | 5659 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 | 5660 | |
c0ff4b85 | 5661 | cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable); |
3c11ecf4 | 5662 | |
c0ff4b85 | 5663 | if (atomic_read(&memcg->under_oom)) |
3c11ecf4 KH |
5664 | cb->fill(cb, "under_oom", 1); |
5665 | else | |
5666 | cb->fill(cb, "under_oom", 0); | |
5667 | return 0; | |
5668 | } | |
5669 | ||
182446d0 | 5670 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
5671 | struct cftype *cft, u64 val) |
5672 | { | |
182446d0 | 5673 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
63876986 | 5674 | struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css)); |
3c11ecf4 KH |
5675 | |
5676 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
63876986 | 5677 | if (!parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
5678 | return -EINVAL; |
5679 | ||
0999821b | 5680 | mutex_lock(&memcg_create_mutex); |
3c11ecf4 | 5681 | /* oom-kill-disable is a flag for subhierarchy. */ |
b5f99b53 | 5682 | if ((parent->use_hierarchy) || memcg_has_children(memcg)) { |
0999821b | 5683 | mutex_unlock(&memcg_create_mutex); |
3c11ecf4 KH |
5684 | return -EINVAL; |
5685 | } | |
c0ff4b85 | 5686 | memcg->oom_kill_disable = val; |
4d845ebf | 5687 | if (!val) |
c0ff4b85 | 5688 | memcg_oom_recover(memcg); |
0999821b | 5689 | mutex_unlock(&memcg_create_mutex); |
3c11ecf4 KH |
5690 | return 0; |
5691 | } | |
5692 | ||
c255a458 | 5693 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 5694 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 5695 | { |
55007d84 GC |
5696 | int ret; |
5697 | ||
2633d7a0 | 5698 | memcg->kmemcg_id = -1; |
55007d84 GC |
5699 | ret = memcg_propagate_kmem(memcg); |
5700 | if (ret) | |
5701 | return ret; | |
2633d7a0 | 5702 | |
1d62e436 | 5703 | return mem_cgroup_sockets_init(memcg, ss); |
573b400d | 5704 | } |
e5671dfa | 5705 | |
10d5ebf4 | 5706 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 5707 | { |
1d62e436 | 5708 | mem_cgroup_sockets_destroy(memcg); |
10d5ebf4 LZ |
5709 | } |
5710 | ||
5711 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) | |
5712 | { | |
5713 | if (!memcg_kmem_is_active(memcg)) | |
5714 | return; | |
5715 | ||
5716 | /* | |
5717 | * kmem charges can outlive the cgroup. In the case of slab | |
5718 | * pages, for instance, a page contain objects from various | |
5719 | * processes. As we prevent from taking a reference for every | |
5720 | * such allocation we have to be careful when doing uncharge | |
5721 | * (see memcg_uncharge_kmem) and here during offlining. | |
5722 | * | |
5723 | * The idea is that that only the _last_ uncharge which sees | |
5724 | * the dead memcg will drop the last reference. An additional | |
5725 | * reference is taken here before the group is marked dead | |
5726 | * which is then paired with css_put during uncharge resp. here. | |
5727 | * | |
5728 | * Although this might sound strange as this path is called from | |
5729 | * css_offline() when the referencemight have dropped down to 0 | |
5730 | * and shouldn't be incremented anymore (css_tryget would fail) | |
5731 | * we do not have other options because of the kmem allocations | |
5732 | * lifetime. | |
5733 | */ | |
5734 | css_get(&memcg->css); | |
7de37682 GC |
5735 | |
5736 | memcg_kmem_mark_dead(memcg); | |
5737 | ||
5738 | if (res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0) | |
5739 | return; | |
5740 | ||
7de37682 | 5741 | if (memcg_kmem_test_and_clear_dead(memcg)) |
10d5ebf4 | 5742 | css_put(&memcg->css); |
d1a4c0b3 | 5743 | } |
e5671dfa | 5744 | #else |
cbe128e3 | 5745 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
5746 | { |
5747 | return 0; | |
5748 | } | |
d1a4c0b3 | 5749 | |
10d5ebf4 LZ |
5750 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
5751 | { | |
5752 | } | |
5753 | ||
5754 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) | |
d1a4c0b3 GC |
5755 | { |
5756 | } | |
e5671dfa GC |
5757 | #endif |
5758 | ||
8cdea7c0 BS |
5759 | static struct cftype mem_cgroup_files[] = { |
5760 | { | |
0eea1030 | 5761 | .name = "usage_in_bytes", |
8c7c6e34 | 5762 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
af36f906 | 5763 | .read = mem_cgroup_read, |
9490ff27 KH |
5764 | .register_event = mem_cgroup_usage_register_event, |
5765 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8cdea7c0 | 5766 | }, |
c84872e1 PE |
5767 | { |
5768 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 5769 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 5770 | .trigger = mem_cgroup_reset, |
af36f906 | 5771 | .read = mem_cgroup_read, |
c84872e1 | 5772 | }, |
8cdea7c0 | 5773 | { |
0eea1030 | 5774 | .name = "limit_in_bytes", |
8c7c6e34 | 5775 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 5776 | .write_string = mem_cgroup_write, |
af36f906 | 5777 | .read = mem_cgroup_read, |
8cdea7c0 | 5778 | }, |
296c81d8 BS |
5779 | { |
5780 | .name = "soft_limit_in_bytes", | |
5781 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
5782 | .write_string = mem_cgroup_write, | |
af36f906 | 5783 | .read = mem_cgroup_read, |
296c81d8 | 5784 | }, |
8cdea7c0 BS |
5785 | { |
5786 | .name = "failcnt", | |
8c7c6e34 | 5787 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 5788 | .trigger = mem_cgroup_reset, |
af36f906 | 5789 | .read = mem_cgroup_read, |
8cdea7c0 | 5790 | }, |
d2ceb9b7 KH |
5791 | { |
5792 | .name = "stat", | |
ab215884 | 5793 | .read_seq_string = memcg_stat_show, |
d2ceb9b7 | 5794 | }, |
c1e862c1 KH |
5795 | { |
5796 | .name = "force_empty", | |
5797 | .trigger = mem_cgroup_force_empty_write, | |
5798 | }, | |
18f59ea7 BS |
5799 | { |
5800 | .name = "use_hierarchy", | |
f00baae7 | 5801 | .flags = CFTYPE_INSANE, |
18f59ea7 BS |
5802 | .write_u64 = mem_cgroup_hierarchy_write, |
5803 | .read_u64 = mem_cgroup_hierarchy_read, | |
5804 | }, | |
a7885eb8 KM |
5805 | { |
5806 | .name = "swappiness", | |
5807 | .read_u64 = mem_cgroup_swappiness_read, | |
5808 | .write_u64 = mem_cgroup_swappiness_write, | |
5809 | }, | |
7dc74be0 DN |
5810 | { |
5811 | .name = "move_charge_at_immigrate", | |
5812 | .read_u64 = mem_cgroup_move_charge_read, | |
5813 | .write_u64 = mem_cgroup_move_charge_write, | |
5814 | }, | |
9490ff27 KH |
5815 | { |
5816 | .name = "oom_control", | |
3c11ecf4 KH |
5817 | .read_map = mem_cgroup_oom_control_read, |
5818 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
5819 | .register_event = mem_cgroup_oom_register_event, |
5820 | .unregister_event = mem_cgroup_oom_unregister_event, | |
5821 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), | |
5822 | }, | |
70ddf637 AV |
5823 | { |
5824 | .name = "pressure_level", | |
5825 | .register_event = vmpressure_register_event, | |
5826 | .unregister_event = vmpressure_unregister_event, | |
5827 | }, | |
406eb0c9 YH |
5828 | #ifdef CONFIG_NUMA |
5829 | { | |
5830 | .name = "numa_stat", | |
ab215884 | 5831 | .read_seq_string = memcg_numa_stat_show, |
406eb0c9 YH |
5832 | }, |
5833 | #endif | |
510fc4e1 GC |
5834 | #ifdef CONFIG_MEMCG_KMEM |
5835 | { | |
5836 | .name = "kmem.limit_in_bytes", | |
5837 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
5838 | .write_string = mem_cgroup_write, | |
5839 | .read = mem_cgroup_read, | |
5840 | }, | |
5841 | { | |
5842 | .name = "kmem.usage_in_bytes", | |
5843 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
5844 | .read = mem_cgroup_read, | |
5845 | }, | |
5846 | { | |
5847 | .name = "kmem.failcnt", | |
5848 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
5849 | .trigger = mem_cgroup_reset, | |
5850 | .read = mem_cgroup_read, | |
5851 | }, | |
5852 | { | |
5853 | .name = "kmem.max_usage_in_bytes", | |
5854 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
5855 | .trigger = mem_cgroup_reset, | |
5856 | .read = mem_cgroup_read, | |
5857 | }, | |
749c5415 GC |
5858 | #ifdef CONFIG_SLABINFO |
5859 | { | |
5860 | .name = "kmem.slabinfo", | |
5861 | .read_seq_string = mem_cgroup_slabinfo_read, | |
5862 | }, | |
5863 | #endif | |
8c7c6e34 | 5864 | #endif |
6bc10349 | 5865 | { }, /* terminate */ |
af36f906 | 5866 | }; |
8c7c6e34 | 5867 | |
2d11085e MH |
5868 | #ifdef CONFIG_MEMCG_SWAP |
5869 | static struct cftype memsw_cgroup_files[] = { | |
5870 | { | |
5871 | .name = "memsw.usage_in_bytes", | |
5872 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
5873 | .read = mem_cgroup_read, | |
5874 | .register_event = mem_cgroup_usage_register_event, | |
5875 | .unregister_event = mem_cgroup_usage_unregister_event, | |
5876 | }, | |
5877 | { | |
5878 | .name = "memsw.max_usage_in_bytes", | |
5879 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
5880 | .trigger = mem_cgroup_reset, | |
5881 | .read = mem_cgroup_read, | |
5882 | }, | |
5883 | { | |
5884 | .name = "memsw.limit_in_bytes", | |
5885 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
5886 | .write_string = mem_cgroup_write, | |
5887 | .read = mem_cgroup_read, | |
5888 | }, | |
5889 | { | |
5890 | .name = "memsw.failcnt", | |
5891 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
5892 | .trigger = mem_cgroup_reset, | |
5893 | .read = mem_cgroup_read, | |
5894 | }, | |
5895 | { }, /* terminate */ | |
5896 | }; | |
5897 | #endif | |
c0ff4b85 | 5898 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
5899 | { |
5900 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 5901 | struct mem_cgroup_per_zone *mz; |
41e3355d | 5902 | int zone, tmp = node; |
1ecaab2b KH |
5903 | /* |
5904 | * This routine is called against possible nodes. | |
5905 | * But it's BUG to call kmalloc() against offline node. | |
5906 | * | |
5907 | * TODO: this routine can waste much memory for nodes which will | |
5908 | * never be onlined. It's better to use memory hotplug callback | |
5909 | * function. | |
5910 | */ | |
41e3355d KH |
5911 | if (!node_state(node, N_NORMAL_MEMORY)) |
5912 | tmp = -1; | |
17295c88 | 5913 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
5914 | if (!pn) |
5915 | return 1; | |
1ecaab2b | 5916 | |
1ecaab2b KH |
5917 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
5918 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 5919 | lruvec_init(&mz->lruvec); |
d79154bb | 5920 | mz->memcg = memcg; |
1ecaab2b | 5921 | } |
54f72fe0 | 5922 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
5923 | return 0; |
5924 | } | |
5925 | ||
c0ff4b85 | 5926 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 5927 | { |
54f72fe0 | 5928 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
5929 | } |
5930 | ||
33327948 KH |
5931 | static struct mem_cgroup *mem_cgroup_alloc(void) |
5932 | { | |
d79154bb | 5933 | struct mem_cgroup *memcg; |
45cf7ebd | 5934 | size_t size = memcg_size(); |
33327948 | 5935 | |
45cf7ebd | 5936 | /* Can be very big if nr_node_ids is very big */ |
c8dad2bb | 5937 | if (size < PAGE_SIZE) |
d79154bb | 5938 | memcg = kzalloc(size, GFP_KERNEL); |
33327948 | 5939 | else |
d79154bb | 5940 | memcg = vzalloc(size); |
33327948 | 5941 | |
d79154bb | 5942 | if (!memcg) |
e7bbcdf3 DC |
5943 | return NULL; |
5944 | ||
d79154bb HD |
5945 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
5946 | if (!memcg->stat) | |
d2e61b8d | 5947 | goto out_free; |
d79154bb HD |
5948 | spin_lock_init(&memcg->pcp_counter_lock); |
5949 | return memcg; | |
d2e61b8d DC |
5950 | |
5951 | out_free: | |
5952 | if (size < PAGE_SIZE) | |
d79154bb | 5953 | kfree(memcg); |
d2e61b8d | 5954 | else |
d79154bb | 5955 | vfree(memcg); |
d2e61b8d | 5956 | return NULL; |
33327948 KH |
5957 | } |
5958 | ||
59927fb9 | 5959 | /* |
c8b2a36f GC |
5960 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
5961 | * (scanning all at force_empty is too costly...) | |
5962 | * | |
5963 | * Instead of clearing all references at force_empty, we remember | |
5964 | * the number of reference from swap_cgroup and free mem_cgroup when | |
5965 | * it goes down to 0. | |
5966 | * | |
5967 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 5968 | */ |
c8b2a36f GC |
5969 | |
5970 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 5971 | { |
c8b2a36f | 5972 | int node; |
45cf7ebd | 5973 | size_t size = memcg_size(); |
59927fb9 | 5974 | |
c8b2a36f GC |
5975 | free_css_id(&mem_cgroup_subsys, &memcg->css); |
5976 | ||
5977 | for_each_node(node) | |
5978 | free_mem_cgroup_per_zone_info(memcg, node); | |
5979 | ||
5980 | free_percpu(memcg->stat); | |
5981 | ||
3f134619 GC |
5982 | /* |
5983 | * We need to make sure that (at least for now), the jump label | |
5984 | * destruction code runs outside of the cgroup lock. This is because | |
5985 | * get_online_cpus(), which is called from the static_branch update, | |
5986 | * can't be called inside the cgroup_lock. cpusets are the ones | |
5987 | * enforcing this dependency, so if they ever change, we might as well. | |
5988 | * | |
5989 | * schedule_work() will guarantee this happens. Be careful if you need | |
5990 | * to move this code around, and make sure it is outside | |
5991 | * the cgroup_lock. | |
5992 | */ | |
a8964b9b | 5993 | disarm_static_keys(memcg); |
3afe36b1 GC |
5994 | if (size < PAGE_SIZE) |
5995 | kfree(memcg); | |
5996 | else | |
5997 | vfree(memcg); | |
59927fb9 | 5998 | } |
3afe36b1 | 5999 | |
7bcc1bb1 DN |
6000 | /* |
6001 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
6002 | */ | |
e1aab161 | 6003 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 6004 | { |
c0ff4b85 | 6005 | if (!memcg->res.parent) |
7bcc1bb1 | 6006 | return NULL; |
c0ff4b85 | 6007 | return mem_cgroup_from_res_counter(memcg->res.parent, res); |
7bcc1bb1 | 6008 | } |
e1aab161 | 6009 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 6010 | |
0eb253e2 | 6011 | static struct cgroup_subsys_state * __ref |
eb95419b | 6012 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 6013 | { |
d142e3e6 | 6014 | struct mem_cgroup *memcg; |
04046e1a | 6015 | long error = -ENOMEM; |
6d12e2d8 | 6016 | int node; |
8cdea7c0 | 6017 | |
c0ff4b85 R |
6018 | memcg = mem_cgroup_alloc(); |
6019 | if (!memcg) | |
04046e1a | 6020 | return ERR_PTR(error); |
78fb7466 | 6021 | |
3ed28fa1 | 6022 | for_each_node(node) |
c0ff4b85 | 6023 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 6024 | goto free_out; |
f64c3f54 | 6025 | |
c077719b | 6026 | /* root ? */ |
eb95419b | 6027 | if (parent_css == NULL) { |
a41c58a6 | 6028 | root_mem_cgroup = memcg; |
d142e3e6 GC |
6029 | res_counter_init(&memcg->res, NULL); |
6030 | res_counter_init(&memcg->memsw, NULL); | |
6031 | res_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 6032 | } |
28dbc4b6 | 6033 | |
d142e3e6 GC |
6034 | memcg->last_scanned_node = MAX_NUMNODES; |
6035 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
6036 | memcg->move_charge_at_immigrate = 0; |
6037 | mutex_init(&memcg->thresholds_lock); | |
6038 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 6039 | vmpressure_init(&memcg->vmpressure); |
7d910c05 | 6040 | spin_lock_init(&memcg->soft_lock); |
d142e3e6 GC |
6041 | |
6042 | return &memcg->css; | |
6043 | ||
6044 | free_out: | |
6045 | __mem_cgroup_free(memcg); | |
6046 | return ERR_PTR(error); | |
6047 | } | |
6048 | ||
6049 | static int | |
eb95419b | 6050 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 6051 | { |
eb95419b TH |
6052 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
6053 | struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(css)); | |
d142e3e6 GC |
6054 | int error = 0; |
6055 | ||
63876986 | 6056 | if (!parent) |
d142e3e6 GC |
6057 | return 0; |
6058 | ||
0999821b | 6059 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
6060 | |
6061 | memcg->use_hierarchy = parent->use_hierarchy; | |
6062 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
6063 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
6064 | ||
6065 | if (parent->use_hierarchy) { | |
c0ff4b85 R |
6066 | res_counter_init(&memcg->res, &parent->res); |
6067 | res_counter_init(&memcg->memsw, &parent->memsw); | |
510fc4e1 | 6068 | res_counter_init(&memcg->kmem, &parent->kmem); |
55007d84 | 6069 | |
7bcc1bb1 | 6070 | /* |
8d76a979 LZ |
6071 | * No need to take a reference to the parent because cgroup |
6072 | * core guarantees its existence. | |
7bcc1bb1 | 6073 | */ |
18f59ea7 | 6074 | } else { |
c0ff4b85 R |
6075 | res_counter_init(&memcg->res, NULL); |
6076 | res_counter_init(&memcg->memsw, NULL); | |
510fc4e1 | 6077 | res_counter_init(&memcg->kmem, NULL); |
8c7f6edb TH |
6078 | /* |
6079 | * Deeper hierachy with use_hierarchy == false doesn't make | |
6080 | * much sense so let cgroup subsystem know about this | |
6081 | * unfortunate state in our controller. | |
6082 | */ | |
d142e3e6 | 6083 | if (parent != root_mem_cgroup) |
8c7f6edb | 6084 | mem_cgroup_subsys.broken_hierarchy = true; |
18f59ea7 | 6085 | } |
cbe128e3 GC |
6086 | |
6087 | error = memcg_init_kmem(memcg, &mem_cgroup_subsys); | |
0999821b | 6088 | mutex_unlock(&memcg_create_mutex); |
d142e3e6 | 6089 | return error; |
8cdea7c0 BS |
6090 | } |
6091 | ||
5f578161 MH |
6092 | /* |
6093 | * Announce all parents that a group from their hierarchy is gone. | |
6094 | */ | |
6095 | static void mem_cgroup_invalidate_reclaim_iterators(struct mem_cgroup *memcg) | |
6096 | { | |
6097 | struct mem_cgroup *parent = memcg; | |
6098 | ||
6099 | while ((parent = parent_mem_cgroup(parent))) | |
519ebea3 | 6100 | mem_cgroup_iter_invalidate(parent); |
5f578161 MH |
6101 | |
6102 | /* | |
6103 | * if the root memcg is not hierarchical we have to check it | |
6104 | * explicitely. | |
6105 | */ | |
6106 | if (!root_mem_cgroup->use_hierarchy) | |
519ebea3 | 6107 | mem_cgroup_iter_invalidate(root_mem_cgroup); |
5f578161 MH |
6108 | } |
6109 | ||
eb95419b | 6110 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 6111 | { |
eb95419b | 6112 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
ec64f515 | 6113 | |
10d5ebf4 LZ |
6114 | kmem_cgroup_css_offline(memcg); |
6115 | ||
5f578161 | 6116 | mem_cgroup_invalidate_reclaim_iterators(memcg); |
ab5196c2 | 6117 | mem_cgroup_reparent_charges(memcg); |
7d910c05 MH |
6118 | if (memcg->soft_contributed) { |
6119 | while ((memcg = parent_mem_cgroup(memcg))) | |
6120 | atomic_dec(&memcg->children_in_excess); | |
1be171d6 MH |
6121 | |
6122 | if (memcg != root_mem_cgroup && !root_mem_cgroup->use_hierarchy) | |
6123 | atomic_dec(&root_mem_cgroup->children_in_excess); | |
7d910c05 | 6124 | } |
1f458cbf | 6125 | mem_cgroup_destroy_all_caches(memcg); |
33cb876e | 6126 | vmpressure_cleanup(&memcg->vmpressure); |
df878fb0 KH |
6127 | } |
6128 | ||
eb95419b | 6129 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 6130 | { |
eb95419b | 6131 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 6132 | |
10d5ebf4 | 6133 | memcg_destroy_kmem(memcg); |
465939a1 | 6134 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
6135 | } |
6136 | ||
02491447 | 6137 | #ifdef CONFIG_MMU |
7dc74be0 | 6138 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
6139 | #define PRECHARGE_COUNT_AT_ONCE 256 |
6140 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 6141 | { |
854ffa8d DN |
6142 | int ret = 0; |
6143 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
c0ff4b85 | 6144 | struct mem_cgroup *memcg = mc.to; |
4ffef5fe | 6145 | |
c0ff4b85 | 6146 | if (mem_cgroup_is_root(memcg)) { |
854ffa8d DN |
6147 | mc.precharge += count; |
6148 | /* we don't need css_get for root */ | |
6149 | return ret; | |
6150 | } | |
6151 | /* try to charge at once */ | |
6152 | if (count > 1) { | |
6153 | struct res_counter *dummy; | |
6154 | /* | |
c0ff4b85 | 6155 | * "memcg" cannot be under rmdir() because we've already checked |
854ffa8d DN |
6156 | * by cgroup_lock_live_cgroup() that it is not removed and we |
6157 | * are still under the same cgroup_mutex. So we can postpone | |
6158 | * css_get(). | |
6159 | */ | |
c0ff4b85 | 6160 | if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy)) |
854ffa8d | 6161 | goto one_by_one; |
c0ff4b85 | 6162 | if (do_swap_account && res_counter_charge(&memcg->memsw, |
854ffa8d | 6163 | PAGE_SIZE * count, &dummy)) { |
c0ff4b85 | 6164 | res_counter_uncharge(&memcg->res, PAGE_SIZE * count); |
854ffa8d DN |
6165 | goto one_by_one; |
6166 | } | |
6167 | mc.precharge += count; | |
854ffa8d DN |
6168 | return ret; |
6169 | } | |
6170 | one_by_one: | |
6171 | /* fall back to one by one charge */ | |
6172 | while (count--) { | |
6173 | if (signal_pending(current)) { | |
6174 | ret = -EINTR; | |
6175 | break; | |
6176 | } | |
6177 | if (!batch_count--) { | |
6178 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
6179 | cond_resched(); | |
6180 | } | |
c0ff4b85 R |
6181 | ret = __mem_cgroup_try_charge(NULL, |
6182 | GFP_KERNEL, 1, &memcg, false); | |
38c5d72f | 6183 | if (ret) |
854ffa8d | 6184 | /* mem_cgroup_clear_mc() will do uncharge later */ |
38c5d72f | 6185 | return ret; |
854ffa8d DN |
6186 | mc.precharge++; |
6187 | } | |
4ffef5fe DN |
6188 | return ret; |
6189 | } | |
6190 | ||
6191 | /** | |
8d32ff84 | 6192 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
6193 | * @vma: the vma the pte to be checked belongs |
6194 | * @addr: the address corresponding to the pte to be checked | |
6195 | * @ptent: the pte to be checked | |
02491447 | 6196 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
6197 | * |
6198 | * Returns | |
6199 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
6200 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
6201 | * move charge. if @target is not NULL, the page is stored in target->page | |
6202 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
6203 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
6204 | * target for charge migration. if @target is not NULL, the entry is stored | |
6205 | * in target->ent. | |
4ffef5fe DN |
6206 | * |
6207 | * Called with pte lock held. | |
6208 | */ | |
4ffef5fe DN |
6209 | union mc_target { |
6210 | struct page *page; | |
02491447 | 6211 | swp_entry_t ent; |
4ffef5fe DN |
6212 | }; |
6213 | ||
4ffef5fe | 6214 | enum mc_target_type { |
8d32ff84 | 6215 | MC_TARGET_NONE = 0, |
4ffef5fe | 6216 | MC_TARGET_PAGE, |
02491447 | 6217 | MC_TARGET_SWAP, |
4ffef5fe DN |
6218 | }; |
6219 | ||
90254a65 DN |
6220 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
6221 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 6222 | { |
90254a65 | 6223 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 6224 | |
90254a65 DN |
6225 | if (!page || !page_mapped(page)) |
6226 | return NULL; | |
6227 | if (PageAnon(page)) { | |
6228 | /* we don't move shared anon */ | |
4b91355e | 6229 | if (!move_anon()) |
90254a65 | 6230 | return NULL; |
87946a72 DN |
6231 | } else if (!move_file()) |
6232 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
6233 | return NULL; |
6234 | if (!get_page_unless_zero(page)) | |
6235 | return NULL; | |
6236 | ||
6237 | return page; | |
6238 | } | |
6239 | ||
4b91355e | 6240 | #ifdef CONFIG_SWAP |
90254a65 DN |
6241 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
6242 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6243 | { | |
90254a65 DN |
6244 | struct page *page = NULL; |
6245 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
6246 | ||
6247 | if (!move_anon() || non_swap_entry(ent)) | |
6248 | return NULL; | |
4b91355e KH |
6249 | /* |
6250 | * Because lookup_swap_cache() updates some statistics counter, | |
6251 | * we call find_get_page() with swapper_space directly. | |
6252 | */ | |
33806f06 | 6253 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
6254 | if (do_swap_account) |
6255 | entry->val = ent.val; | |
6256 | ||
6257 | return page; | |
6258 | } | |
4b91355e KH |
6259 | #else |
6260 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
6261 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6262 | { | |
6263 | return NULL; | |
6264 | } | |
6265 | #endif | |
90254a65 | 6266 | |
87946a72 DN |
6267 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
6268 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6269 | { | |
6270 | struct page *page = NULL; | |
87946a72 DN |
6271 | struct address_space *mapping; |
6272 | pgoff_t pgoff; | |
6273 | ||
6274 | if (!vma->vm_file) /* anonymous vma */ | |
6275 | return NULL; | |
6276 | if (!move_file()) | |
6277 | return NULL; | |
6278 | ||
87946a72 DN |
6279 | mapping = vma->vm_file->f_mapping; |
6280 | if (pte_none(ptent)) | |
6281 | pgoff = linear_page_index(vma, addr); | |
6282 | else /* pte_file(ptent) is true */ | |
6283 | pgoff = pte_to_pgoff(ptent); | |
6284 | ||
6285 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
6286 | page = find_get_page(mapping, pgoff); |
6287 | ||
6288 | #ifdef CONFIG_SWAP | |
6289 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
6290 | if (radix_tree_exceptional_entry(page)) { | |
6291 | swp_entry_t swap = radix_to_swp_entry(page); | |
87946a72 | 6292 | if (do_swap_account) |
aa3b1895 | 6293 | *entry = swap; |
33806f06 | 6294 | page = find_get_page(swap_address_space(swap), swap.val); |
87946a72 | 6295 | } |
aa3b1895 | 6296 | #endif |
87946a72 DN |
6297 | return page; |
6298 | } | |
6299 | ||
8d32ff84 | 6300 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
6301 | unsigned long addr, pte_t ptent, union mc_target *target) |
6302 | { | |
6303 | struct page *page = NULL; | |
6304 | struct page_cgroup *pc; | |
8d32ff84 | 6305 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
6306 | swp_entry_t ent = { .val = 0 }; |
6307 | ||
6308 | if (pte_present(ptent)) | |
6309 | page = mc_handle_present_pte(vma, addr, ptent); | |
6310 | else if (is_swap_pte(ptent)) | |
6311 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
6312 | else if (pte_none(ptent) || pte_file(ptent)) |
6313 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
6314 | |
6315 | if (!page && !ent.val) | |
8d32ff84 | 6316 | return ret; |
02491447 DN |
6317 | if (page) { |
6318 | pc = lookup_page_cgroup(page); | |
6319 | /* | |
6320 | * Do only loose check w/o page_cgroup lock. | |
6321 | * mem_cgroup_move_account() checks the pc is valid or not under | |
6322 | * the lock. | |
6323 | */ | |
6324 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
6325 | ret = MC_TARGET_PAGE; | |
6326 | if (target) | |
6327 | target->page = page; | |
6328 | } | |
6329 | if (!ret || !target) | |
6330 | put_page(page); | |
6331 | } | |
90254a65 DN |
6332 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
6333 | if (ent.val && !ret && | |
9fb4b7cc | 6334 | css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
6335 | ret = MC_TARGET_SWAP; |
6336 | if (target) | |
6337 | target->ent = ent; | |
4ffef5fe | 6338 | } |
4ffef5fe DN |
6339 | return ret; |
6340 | } | |
6341 | ||
12724850 NH |
6342 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
6343 | /* | |
6344 | * We don't consider swapping or file mapped pages because THP does not | |
6345 | * support them for now. | |
6346 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
6347 | */ | |
6348 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6349 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6350 | { | |
6351 | struct page *page = NULL; | |
6352 | struct page_cgroup *pc; | |
6353 | enum mc_target_type ret = MC_TARGET_NONE; | |
6354 | ||
6355 | page = pmd_page(pmd); | |
6356 | VM_BUG_ON(!page || !PageHead(page)); | |
6357 | if (!move_anon()) | |
6358 | return ret; | |
6359 | pc = lookup_page_cgroup(page); | |
6360 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
6361 | ret = MC_TARGET_PAGE; | |
6362 | if (target) { | |
6363 | get_page(page); | |
6364 | target->page = page; | |
6365 | } | |
6366 | } | |
6367 | return ret; | |
6368 | } | |
6369 | #else | |
6370 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6371 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6372 | { | |
6373 | return MC_TARGET_NONE; | |
6374 | } | |
6375 | #endif | |
6376 | ||
4ffef5fe DN |
6377 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
6378 | unsigned long addr, unsigned long end, | |
6379 | struct mm_walk *walk) | |
6380 | { | |
6381 | struct vm_area_struct *vma = walk->private; | |
6382 | pte_t *pte; | |
6383 | spinlock_t *ptl; | |
6384 | ||
12724850 NH |
6385 | if (pmd_trans_huge_lock(pmd, vma) == 1) { |
6386 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) | |
6387 | mc.precharge += HPAGE_PMD_NR; | |
6388 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 6389 | return 0; |
12724850 | 6390 | } |
03319327 | 6391 | |
45f83cef AA |
6392 | if (pmd_trans_unstable(pmd)) |
6393 | return 0; | |
4ffef5fe DN |
6394 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
6395 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 6396 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
6397 | mc.precharge++; /* increment precharge temporarily */ |
6398 | pte_unmap_unlock(pte - 1, ptl); | |
6399 | cond_resched(); | |
6400 | ||
7dc74be0 DN |
6401 | return 0; |
6402 | } | |
6403 | ||
4ffef5fe DN |
6404 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
6405 | { | |
6406 | unsigned long precharge; | |
6407 | struct vm_area_struct *vma; | |
6408 | ||
dfe076b0 | 6409 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
6410 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
6411 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
6412 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
6413 | .mm = mm, | |
6414 | .private = vma, | |
6415 | }; | |
6416 | if (is_vm_hugetlb_page(vma)) | |
6417 | continue; | |
4ffef5fe DN |
6418 | walk_page_range(vma->vm_start, vma->vm_end, |
6419 | &mem_cgroup_count_precharge_walk); | |
6420 | } | |
dfe076b0 | 6421 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
6422 | |
6423 | precharge = mc.precharge; | |
6424 | mc.precharge = 0; | |
6425 | ||
6426 | return precharge; | |
6427 | } | |
6428 | ||
4ffef5fe DN |
6429 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
6430 | { | |
dfe076b0 DN |
6431 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
6432 | ||
6433 | VM_BUG_ON(mc.moving_task); | |
6434 | mc.moving_task = current; | |
6435 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
6436 | } |
6437 | ||
dfe076b0 DN |
6438 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
6439 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 6440 | { |
2bd9bb20 KH |
6441 | struct mem_cgroup *from = mc.from; |
6442 | struct mem_cgroup *to = mc.to; | |
4050377b | 6443 | int i; |
2bd9bb20 | 6444 | |
4ffef5fe | 6445 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
6446 | if (mc.precharge) { |
6447 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
6448 | mc.precharge = 0; | |
6449 | } | |
6450 | /* | |
6451 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
6452 | * we must uncharge here. | |
6453 | */ | |
6454 | if (mc.moved_charge) { | |
6455 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
6456 | mc.moved_charge = 0; | |
4ffef5fe | 6457 | } |
483c30b5 DN |
6458 | /* we must fixup refcnts and charges */ |
6459 | if (mc.moved_swap) { | |
483c30b5 DN |
6460 | /* uncharge swap account from the old cgroup */ |
6461 | if (!mem_cgroup_is_root(mc.from)) | |
6462 | res_counter_uncharge(&mc.from->memsw, | |
6463 | PAGE_SIZE * mc.moved_swap); | |
4050377b LZ |
6464 | |
6465 | for (i = 0; i < mc.moved_swap; i++) | |
6466 | css_put(&mc.from->css); | |
483c30b5 DN |
6467 | |
6468 | if (!mem_cgroup_is_root(mc.to)) { | |
6469 | /* | |
6470 | * we charged both to->res and to->memsw, so we should | |
6471 | * uncharge to->res. | |
6472 | */ | |
6473 | res_counter_uncharge(&mc.to->res, | |
6474 | PAGE_SIZE * mc.moved_swap); | |
483c30b5 | 6475 | } |
4050377b | 6476 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
6477 | mc.moved_swap = 0; |
6478 | } | |
dfe076b0 DN |
6479 | memcg_oom_recover(from); |
6480 | memcg_oom_recover(to); | |
6481 | wake_up_all(&mc.waitq); | |
6482 | } | |
6483 | ||
6484 | static void mem_cgroup_clear_mc(void) | |
6485 | { | |
6486 | struct mem_cgroup *from = mc.from; | |
6487 | ||
6488 | /* | |
6489 | * we must clear moving_task before waking up waiters at the end of | |
6490 | * task migration. | |
6491 | */ | |
6492 | mc.moving_task = NULL; | |
6493 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 6494 | spin_lock(&mc.lock); |
4ffef5fe DN |
6495 | mc.from = NULL; |
6496 | mc.to = NULL; | |
2bd9bb20 | 6497 | spin_unlock(&mc.lock); |
32047e2a | 6498 | mem_cgroup_end_move(from); |
4ffef5fe DN |
6499 | } |
6500 | ||
eb95419b | 6501 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6502 | struct cgroup_taskset *tset) |
7dc74be0 | 6503 | { |
2f7ee569 | 6504 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 6505 | int ret = 0; |
eb95419b | 6506 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
ee5e8472 | 6507 | unsigned long move_charge_at_immigrate; |
7dc74be0 | 6508 | |
ee5e8472 GC |
6509 | /* |
6510 | * We are now commited to this value whatever it is. Changes in this | |
6511 | * tunable will only affect upcoming migrations, not the current one. | |
6512 | * So we need to save it, and keep it going. | |
6513 | */ | |
6514 | move_charge_at_immigrate = memcg->move_charge_at_immigrate; | |
6515 | if (move_charge_at_immigrate) { | |
7dc74be0 DN |
6516 | struct mm_struct *mm; |
6517 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
6518 | ||
c0ff4b85 | 6519 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
6520 | |
6521 | mm = get_task_mm(p); | |
6522 | if (!mm) | |
6523 | return 0; | |
7dc74be0 | 6524 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
6525 | if (mm->owner == p) { |
6526 | VM_BUG_ON(mc.from); | |
6527 | VM_BUG_ON(mc.to); | |
6528 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 6529 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 6530 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 6531 | mem_cgroup_start_move(from); |
2bd9bb20 | 6532 | spin_lock(&mc.lock); |
4ffef5fe | 6533 | mc.from = from; |
c0ff4b85 | 6534 | mc.to = memcg; |
ee5e8472 | 6535 | mc.immigrate_flags = move_charge_at_immigrate; |
2bd9bb20 | 6536 | spin_unlock(&mc.lock); |
dfe076b0 | 6537 | /* We set mc.moving_task later */ |
4ffef5fe DN |
6538 | |
6539 | ret = mem_cgroup_precharge_mc(mm); | |
6540 | if (ret) | |
6541 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
6542 | } |
6543 | mmput(mm); | |
7dc74be0 DN |
6544 | } |
6545 | return ret; | |
6546 | } | |
6547 | ||
eb95419b | 6548 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6549 | struct cgroup_taskset *tset) |
7dc74be0 | 6550 | { |
4ffef5fe | 6551 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
6552 | } |
6553 | ||
4ffef5fe DN |
6554 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
6555 | unsigned long addr, unsigned long end, | |
6556 | struct mm_walk *walk) | |
7dc74be0 | 6557 | { |
4ffef5fe DN |
6558 | int ret = 0; |
6559 | struct vm_area_struct *vma = walk->private; | |
6560 | pte_t *pte; | |
6561 | spinlock_t *ptl; | |
12724850 NH |
6562 | enum mc_target_type target_type; |
6563 | union mc_target target; | |
6564 | struct page *page; | |
6565 | struct page_cgroup *pc; | |
4ffef5fe | 6566 | |
12724850 NH |
6567 | /* |
6568 | * We don't take compound_lock() here but no race with splitting thp | |
6569 | * happens because: | |
6570 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
6571 | * under splitting, which means there's no concurrent thp split, | |
6572 | * - if another thread runs into split_huge_page() just after we | |
6573 | * entered this if-block, the thread must wait for page table lock | |
6574 | * to be unlocked in __split_huge_page_splitting(), where the main | |
6575 | * part of thp split is not executed yet. | |
6576 | */ | |
6577 | if (pmd_trans_huge_lock(pmd, vma) == 1) { | |
62ade86a | 6578 | if (mc.precharge < HPAGE_PMD_NR) { |
12724850 NH |
6579 | spin_unlock(&vma->vm_mm->page_table_lock); |
6580 | return 0; | |
6581 | } | |
6582 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
6583 | if (target_type == MC_TARGET_PAGE) { | |
6584 | page = target.page; | |
6585 | if (!isolate_lru_page(page)) { | |
6586 | pc = lookup_page_cgroup(page); | |
6587 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, | |
2f3479b1 | 6588 | pc, mc.from, mc.to)) { |
12724850 NH |
6589 | mc.precharge -= HPAGE_PMD_NR; |
6590 | mc.moved_charge += HPAGE_PMD_NR; | |
6591 | } | |
6592 | putback_lru_page(page); | |
6593 | } | |
6594 | put_page(page); | |
6595 | } | |
6596 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 6597 | return 0; |
12724850 NH |
6598 | } |
6599 | ||
45f83cef AA |
6600 | if (pmd_trans_unstable(pmd)) |
6601 | return 0; | |
4ffef5fe DN |
6602 | retry: |
6603 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
6604 | for (; addr != end; addr += PAGE_SIZE) { | |
6605 | pte_t ptent = *(pte++); | |
02491447 | 6606 | swp_entry_t ent; |
4ffef5fe DN |
6607 | |
6608 | if (!mc.precharge) | |
6609 | break; | |
6610 | ||
8d32ff84 | 6611 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
6612 | case MC_TARGET_PAGE: |
6613 | page = target.page; | |
6614 | if (isolate_lru_page(page)) | |
6615 | goto put; | |
6616 | pc = lookup_page_cgroup(page); | |
7ec99d62 | 6617 | if (!mem_cgroup_move_account(page, 1, pc, |
2f3479b1 | 6618 | mc.from, mc.to)) { |
4ffef5fe | 6619 | mc.precharge--; |
854ffa8d DN |
6620 | /* we uncharge from mc.from later. */ |
6621 | mc.moved_charge++; | |
4ffef5fe DN |
6622 | } |
6623 | putback_lru_page(page); | |
8d32ff84 | 6624 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
6625 | put_page(page); |
6626 | break; | |
02491447 DN |
6627 | case MC_TARGET_SWAP: |
6628 | ent = target.ent; | |
e91cbb42 | 6629 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6630 | mc.precharge--; |
483c30b5 DN |
6631 | /* we fixup refcnts and charges later. */ |
6632 | mc.moved_swap++; | |
6633 | } | |
02491447 | 6634 | break; |
4ffef5fe DN |
6635 | default: |
6636 | break; | |
6637 | } | |
6638 | } | |
6639 | pte_unmap_unlock(pte - 1, ptl); | |
6640 | cond_resched(); | |
6641 | ||
6642 | if (addr != end) { | |
6643 | /* | |
6644 | * We have consumed all precharges we got in can_attach(). | |
6645 | * We try charge one by one, but don't do any additional | |
6646 | * charges to mc.to if we have failed in charge once in attach() | |
6647 | * phase. | |
6648 | */ | |
854ffa8d | 6649 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6650 | if (!ret) |
6651 | goto retry; | |
6652 | } | |
6653 | ||
6654 | return ret; | |
6655 | } | |
6656 | ||
6657 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
6658 | { | |
6659 | struct vm_area_struct *vma; | |
6660 | ||
6661 | lru_add_drain_all(); | |
dfe076b0 DN |
6662 | retry: |
6663 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
6664 | /* | |
6665 | * Someone who are holding the mmap_sem might be waiting in | |
6666 | * waitq. So we cancel all extra charges, wake up all waiters, | |
6667 | * and retry. Because we cancel precharges, we might not be able | |
6668 | * to move enough charges, but moving charge is a best-effort | |
6669 | * feature anyway, so it wouldn't be a big problem. | |
6670 | */ | |
6671 | __mem_cgroup_clear_mc(); | |
6672 | cond_resched(); | |
6673 | goto retry; | |
6674 | } | |
4ffef5fe DN |
6675 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
6676 | int ret; | |
6677 | struct mm_walk mem_cgroup_move_charge_walk = { | |
6678 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6679 | .mm = mm, | |
6680 | .private = vma, | |
6681 | }; | |
6682 | if (is_vm_hugetlb_page(vma)) | |
6683 | continue; | |
4ffef5fe DN |
6684 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
6685 | &mem_cgroup_move_charge_walk); | |
6686 | if (ret) | |
6687 | /* | |
6688 | * means we have consumed all precharges and failed in | |
6689 | * doing additional charge. Just abandon here. | |
6690 | */ | |
6691 | break; | |
6692 | } | |
dfe076b0 | 6693 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
6694 | } |
6695 | ||
eb95419b | 6696 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 6697 | struct cgroup_taskset *tset) |
67e465a7 | 6698 | { |
2f7ee569 | 6699 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 6700 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 6701 | |
dfe076b0 | 6702 | if (mm) { |
a433658c KM |
6703 | if (mc.to) |
6704 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
6705 | mmput(mm); |
6706 | } | |
a433658c KM |
6707 | if (mc.to) |
6708 | mem_cgroup_clear_mc(); | |
67e465a7 | 6709 | } |
5cfb80a7 | 6710 | #else /* !CONFIG_MMU */ |
eb95419b | 6711 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6712 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
6713 | { |
6714 | return 0; | |
6715 | } | |
eb95419b | 6716 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6717 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
6718 | { |
6719 | } | |
eb95419b | 6720 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 6721 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
6722 | { |
6723 | } | |
6724 | #endif | |
67e465a7 | 6725 | |
f00baae7 TH |
6726 | /* |
6727 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
6728 | * to verify sane_behavior flag on each mount attempt. | |
6729 | */ | |
eb95419b | 6730 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
6731 | { |
6732 | /* | |
6733 | * use_hierarchy is forced with sane_behavior. cgroup core | |
6734 | * guarantees that @root doesn't have any children, so turning it | |
6735 | * on for the root memcg is enough. | |
6736 | */ | |
eb95419b TH |
6737 | if (cgroup_sane_behavior(root_css->cgroup)) |
6738 | mem_cgroup_from_css(root_css)->use_hierarchy = true; | |
f00baae7 TH |
6739 | } |
6740 | ||
8cdea7c0 BS |
6741 | struct cgroup_subsys mem_cgroup_subsys = { |
6742 | .name = "memory", | |
6743 | .subsys_id = mem_cgroup_subsys_id, | |
92fb9748 | 6744 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6745 | .css_online = mem_cgroup_css_online, |
92fb9748 TH |
6746 | .css_offline = mem_cgroup_css_offline, |
6747 | .css_free = mem_cgroup_css_free, | |
7dc74be0 DN |
6748 | .can_attach = mem_cgroup_can_attach, |
6749 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 6750 | .attach = mem_cgroup_move_task, |
f00baae7 | 6751 | .bind = mem_cgroup_bind, |
6bc10349 | 6752 | .base_cftypes = mem_cgroup_files, |
6d12e2d8 | 6753 | .early_init = 0, |
04046e1a | 6754 | .use_id = 1, |
8cdea7c0 | 6755 | }; |
c077719b | 6756 | |
c255a458 | 6757 | #ifdef CONFIG_MEMCG_SWAP |
a42c390c MH |
6758 | static int __init enable_swap_account(char *s) |
6759 | { | |
a2c8990a | 6760 | if (!strcmp(s, "1")) |
a42c390c | 6761 | really_do_swap_account = 1; |
a2c8990a | 6762 | else if (!strcmp(s, "0")) |
a42c390c MH |
6763 | really_do_swap_account = 0; |
6764 | return 1; | |
6765 | } | |
a2c8990a | 6766 | __setup("swapaccount=", enable_swap_account); |
c077719b | 6767 | |
2d11085e MH |
6768 | static void __init memsw_file_init(void) |
6769 | { | |
6acc8b02 MH |
6770 | WARN_ON(cgroup_add_cftypes(&mem_cgroup_subsys, memsw_cgroup_files)); |
6771 | } | |
6772 | ||
6773 | static void __init enable_swap_cgroup(void) | |
6774 | { | |
6775 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
6776 | do_swap_account = 1; | |
6777 | memsw_file_init(); | |
6778 | } | |
2d11085e | 6779 | } |
6acc8b02 | 6780 | |
2d11085e | 6781 | #else |
6acc8b02 | 6782 | static void __init enable_swap_cgroup(void) |
2d11085e MH |
6783 | { |
6784 | } | |
c077719b | 6785 | #endif |
2d11085e MH |
6786 | |
6787 | /* | |
1081312f MH |
6788 | * subsys_initcall() for memory controller. |
6789 | * | |
6790 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
6791 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
6792 | * everything that doesn't depend on a specific mem_cgroup structure should | |
6793 | * be initialized from here. | |
2d11085e MH |
6794 | */ |
6795 | static int __init mem_cgroup_init(void) | |
6796 | { | |
6797 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); | |
6acc8b02 | 6798 | enable_swap_cgroup(); |
e4777496 | 6799 | memcg_stock_init(); |
2d11085e MH |
6800 | return 0; |
6801 | } | |
6802 | subsys_initcall(mem_cgroup_init); |