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> |
bb4cc1a8 | 42 | #include <linux/rbtree.h> |
b6ac57d5 | 43 | #include <linux/slab.h> |
66e1707b | 44 | #include <linux/swap.h> |
02491447 | 45 | #include <linux/swapops.h> |
66e1707b | 46 | #include <linux/spinlock.h> |
2e72b634 | 47 | #include <linux/eventfd.h> |
79bd9814 | 48 | #include <linux/poll.h> |
2e72b634 | 49 | #include <linux/sort.h> |
66e1707b | 50 | #include <linux/fs.h> |
d2ceb9b7 | 51 | #include <linux/seq_file.h> |
33327948 | 52 | #include <linux/vmalloc.h> |
70ddf637 | 53 | #include <linux/vmpressure.h> |
b69408e8 | 54 | #include <linux/mm_inline.h> |
52d4b9ac | 55 | #include <linux/page_cgroup.h> |
cdec2e42 | 56 | #include <linux/cpu.h> |
158e0a2d | 57 | #include <linux/oom.h> |
0056f4e6 | 58 | #include <linux/lockdep.h> |
79bd9814 | 59 | #include <linux/file.h> |
08e552c6 | 60 | #include "internal.h" |
d1a4c0b3 | 61 | #include <net/sock.h> |
4bd2c1ee | 62 | #include <net/ip.h> |
d1a4c0b3 | 63 | #include <net/tcp_memcontrol.h> |
8cdea7c0 | 64 | |
8697d331 BS |
65 | #include <asm/uaccess.h> |
66 | ||
cc8e970c KM |
67 | #include <trace/events/vmscan.h> |
68 | ||
a181b0e8 | 69 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
68ae564b DR |
70 | EXPORT_SYMBOL(mem_cgroup_subsys); |
71 | ||
a181b0e8 | 72 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
6bbda35c | 73 | static struct mem_cgroup *root_mem_cgroup __read_mostly; |
8cdea7c0 | 74 | |
c255a458 | 75 | #ifdef CONFIG_MEMCG_SWAP |
338c8431 | 76 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ |
c077719b | 77 | int do_swap_account __read_mostly; |
a42c390c MH |
78 | |
79 | /* for remember boot option*/ | |
c255a458 | 80 | #ifdef CONFIG_MEMCG_SWAP_ENABLED |
a42c390c MH |
81 | static int really_do_swap_account __initdata = 1; |
82 | #else | |
83 | static int really_do_swap_account __initdata = 0; | |
84 | #endif | |
85 | ||
c077719b | 86 | #else |
a0db00fc | 87 | #define do_swap_account 0 |
c077719b KH |
88 | #endif |
89 | ||
90 | ||
af7c4b0e JW |
91 | static const char * const mem_cgroup_stat_names[] = { |
92 | "cache", | |
93 | "rss", | |
b070e65c | 94 | "rss_huge", |
af7c4b0e | 95 | "mapped_file", |
3ea67d06 | 96 | "writeback", |
af7c4b0e JW |
97 | "swap", |
98 | }; | |
99 | ||
e9f8974f JW |
100 | enum mem_cgroup_events_index { |
101 | MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */ | |
102 | MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */ | |
456f998e YH |
103 | MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */ |
104 | MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */ | |
e9f8974f JW |
105 | MEM_CGROUP_EVENTS_NSTATS, |
106 | }; | |
af7c4b0e JW |
107 | |
108 | static const char * const mem_cgroup_events_names[] = { | |
109 | "pgpgin", | |
110 | "pgpgout", | |
111 | "pgfault", | |
112 | "pgmajfault", | |
113 | }; | |
114 | ||
58cf188e SZ |
115 | static const char * const mem_cgroup_lru_names[] = { |
116 | "inactive_anon", | |
117 | "active_anon", | |
118 | "inactive_file", | |
119 | "active_file", | |
120 | "unevictable", | |
121 | }; | |
122 | ||
7a159cc9 JW |
123 | /* |
124 | * Per memcg event counter is incremented at every pagein/pageout. With THP, | |
125 | * it will be incremated by the number of pages. This counter is used for | |
126 | * for trigger some periodic events. This is straightforward and better | |
127 | * than using jiffies etc. to handle periodic memcg event. | |
128 | */ | |
129 | enum mem_cgroup_events_target { | |
130 | MEM_CGROUP_TARGET_THRESH, | |
bb4cc1a8 | 131 | MEM_CGROUP_TARGET_SOFTLIMIT, |
453a9bf3 | 132 | MEM_CGROUP_TARGET_NUMAINFO, |
7a159cc9 JW |
133 | MEM_CGROUP_NTARGETS, |
134 | }; | |
a0db00fc KS |
135 | #define THRESHOLDS_EVENTS_TARGET 128 |
136 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
137 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 138 | |
d52aa412 | 139 | struct mem_cgroup_stat_cpu { |
7a159cc9 | 140 | long count[MEM_CGROUP_STAT_NSTATS]; |
e9f8974f | 141 | unsigned long events[MEM_CGROUP_EVENTS_NSTATS]; |
13114716 | 142 | unsigned long nr_page_events; |
7a159cc9 | 143 | unsigned long targets[MEM_CGROUP_NTARGETS]; |
d52aa412 KH |
144 | }; |
145 | ||
527a5ec9 | 146 | struct mem_cgroup_reclaim_iter { |
5f578161 MH |
147 | /* |
148 | * last scanned hierarchy member. Valid only if last_dead_count | |
149 | * matches memcg->dead_count of the hierarchy root group. | |
150 | */ | |
542f85f9 | 151 | struct mem_cgroup *last_visited; |
5f578161 MH |
152 | unsigned long last_dead_count; |
153 | ||
527a5ec9 JW |
154 | /* scan generation, increased every round-trip */ |
155 | unsigned int generation; | |
156 | }; | |
157 | ||
6d12e2d8 KH |
158 | /* |
159 | * per-zone information in memory controller. | |
160 | */ | |
6d12e2d8 | 161 | struct mem_cgroup_per_zone { |
6290df54 | 162 | struct lruvec lruvec; |
1eb49272 | 163 | unsigned long lru_size[NR_LRU_LISTS]; |
3e2f41f1 | 164 | |
527a5ec9 JW |
165 | struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1]; |
166 | ||
bb4cc1a8 AM |
167 | struct rb_node tree_node; /* RB tree node */ |
168 | unsigned long long usage_in_excess;/* Set to the value by which */ | |
169 | /* the soft limit is exceeded*/ | |
170 | bool on_tree; | |
d79154bb | 171 | struct mem_cgroup *memcg; /* Back pointer, we cannot */ |
4e416953 | 172 | /* use container_of */ |
6d12e2d8 | 173 | }; |
6d12e2d8 KH |
174 | |
175 | struct mem_cgroup_per_node { | |
176 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
177 | }; | |
178 | ||
bb4cc1a8 AM |
179 | /* |
180 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
181 | * their hierarchy representation | |
182 | */ | |
183 | ||
184 | struct mem_cgroup_tree_per_zone { | |
185 | struct rb_root rb_root; | |
186 | spinlock_t lock; | |
187 | }; | |
188 | ||
189 | struct mem_cgroup_tree_per_node { | |
190 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
191 | }; | |
192 | ||
193 | struct mem_cgroup_tree { | |
194 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
195 | }; | |
196 | ||
197 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
198 | ||
2e72b634 KS |
199 | struct mem_cgroup_threshold { |
200 | struct eventfd_ctx *eventfd; | |
201 | u64 threshold; | |
202 | }; | |
203 | ||
9490ff27 | 204 | /* For threshold */ |
2e72b634 | 205 | struct mem_cgroup_threshold_ary { |
748dad36 | 206 | /* An array index points to threshold just below or equal to usage. */ |
5407a562 | 207 | int current_threshold; |
2e72b634 KS |
208 | /* Size of entries[] */ |
209 | unsigned int size; | |
210 | /* Array of thresholds */ | |
211 | struct mem_cgroup_threshold entries[0]; | |
212 | }; | |
2c488db2 KS |
213 | |
214 | struct mem_cgroup_thresholds { | |
215 | /* Primary thresholds array */ | |
216 | struct mem_cgroup_threshold_ary *primary; | |
217 | /* | |
218 | * Spare threshold array. | |
219 | * This is needed to make mem_cgroup_unregister_event() "never fail". | |
220 | * It must be able to store at least primary->size - 1 entries. | |
221 | */ | |
222 | struct mem_cgroup_threshold_ary *spare; | |
223 | }; | |
224 | ||
9490ff27 KH |
225 | /* for OOM */ |
226 | struct mem_cgroup_eventfd_list { | |
227 | struct list_head list; | |
228 | struct eventfd_ctx *eventfd; | |
229 | }; | |
2e72b634 | 230 | |
79bd9814 TH |
231 | /* |
232 | * cgroup_event represents events which userspace want to receive. | |
233 | */ | |
234 | struct cgroup_event { | |
235 | /* | |
236 | * css which the event belongs to. | |
237 | */ | |
238 | struct cgroup_subsys_state *css; | |
239 | /* | |
240 | * Control file which the event associated. | |
241 | */ | |
242 | struct cftype *cft; | |
243 | /* | |
244 | * eventfd to signal userspace about the event. | |
245 | */ | |
246 | struct eventfd_ctx *eventfd; | |
247 | /* | |
248 | * Each of these stored in a list by the cgroup. | |
249 | */ | |
250 | struct list_head list; | |
251 | /* | |
252 | * All fields below needed to unregister event when | |
253 | * userspace closes eventfd. | |
254 | */ | |
255 | poll_table pt; | |
256 | wait_queue_head_t *wqh; | |
257 | wait_queue_t wait; | |
258 | struct work_struct remove; | |
259 | }; | |
260 | ||
c0ff4b85 R |
261 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
262 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 263 | |
8cdea7c0 BS |
264 | /* |
265 | * The memory controller data structure. The memory controller controls both | |
266 | * page cache and RSS per cgroup. We would eventually like to provide | |
267 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
268 | * to help the administrator determine what knobs to tune. | |
269 | * | |
270 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
271 | * we hit the water mark. May be even add a low water mark, such that |
272 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
273 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
274 | */ |
275 | struct mem_cgroup { | |
276 | struct cgroup_subsys_state css; | |
277 | /* | |
278 | * the counter to account for memory usage | |
279 | */ | |
280 | struct res_counter res; | |
59927fb9 | 281 | |
70ddf637 AV |
282 | /* vmpressure notifications */ |
283 | struct vmpressure vmpressure; | |
284 | ||
465939a1 LZ |
285 | /* |
286 | * the counter to account for mem+swap usage. | |
287 | */ | |
288 | struct res_counter memsw; | |
59927fb9 | 289 | |
510fc4e1 GC |
290 | /* |
291 | * the counter to account for kernel memory usage. | |
292 | */ | |
293 | struct res_counter kmem; | |
18f59ea7 BS |
294 | /* |
295 | * Should the accounting and control be hierarchical, per subtree? | |
296 | */ | |
297 | bool use_hierarchy; | |
510fc4e1 | 298 | unsigned long kmem_account_flags; /* See KMEM_ACCOUNTED_*, below */ |
79dfdacc MH |
299 | |
300 | bool oom_lock; | |
301 | atomic_t under_oom; | |
3812c8c8 | 302 | atomic_t oom_wakeups; |
79dfdacc | 303 | |
1f4c025b | 304 | int swappiness; |
3c11ecf4 KH |
305 | /* OOM-Killer disable */ |
306 | int oom_kill_disable; | |
a7885eb8 | 307 | |
22a668d7 KH |
308 | /* set when res.limit == memsw.limit */ |
309 | bool memsw_is_minimum; | |
310 | ||
2e72b634 KS |
311 | /* protect arrays of thresholds */ |
312 | struct mutex thresholds_lock; | |
313 | ||
314 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 315 | struct mem_cgroup_thresholds thresholds; |
907860ed | 316 | |
2e72b634 | 317 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 318 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 319 | |
9490ff27 KH |
320 | /* For oom notifier event fd */ |
321 | struct list_head oom_notify; | |
185efc0f | 322 | |
7dc74be0 DN |
323 | /* |
324 | * Should we move charges of a task when a task is moved into this | |
325 | * mem_cgroup ? And what type of charges should we move ? | |
326 | */ | |
f894ffa8 | 327 | unsigned long move_charge_at_immigrate; |
619d094b KH |
328 | /* |
329 | * set > 0 if pages under this cgroup are moving to other cgroup. | |
330 | */ | |
331 | atomic_t moving_account; | |
312734c0 KH |
332 | /* taken only while moving_account > 0 */ |
333 | spinlock_t move_lock; | |
d52aa412 | 334 | /* |
c62b1a3b | 335 | * percpu counter. |
d52aa412 | 336 | */ |
3a7951b4 | 337 | struct mem_cgroup_stat_cpu __percpu *stat; |
711d3d2c KH |
338 | /* |
339 | * used when a cpu is offlined or other synchronizations | |
340 | * See mem_cgroup_read_stat(). | |
341 | */ | |
342 | struct mem_cgroup_stat_cpu nocpu_base; | |
343 | spinlock_t pcp_counter_lock; | |
d1a4c0b3 | 344 | |
5f578161 | 345 | atomic_t dead_count; |
4bd2c1ee | 346 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET) |
d1a4c0b3 GC |
347 | struct tcp_memcontrol tcp_mem; |
348 | #endif | |
2633d7a0 GC |
349 | #if defined(CONFIG_MEMCG_KMEM) |
350 | /* analogous to slab_common's slab_caches list. per-memcg */ | |
351 | struct list_head memcg_slab_caches; | |
352 | /* Not a spinlock, we can take a lot of time walking the list */ | |
353 | struct mutex slab_caches_mutex; | |
354 | /* Index in the kmem_cache->memcg_params->memcg_caches array */ | |
355 | int kmemcg_id; | |
356 | #endif | |
45cf7ebd GC |
357 | |
358 | int last_scanned_node; | |
359 | #if MAX_NUMNODES > 1 | |
360 | nodemask_t scan_nodes; | |
361 | atomic_t numainfo_events; | |
362 | atomic_t numainfo_updating; | |
363 | #endif | |
70ddf637 | 364 | |
54f72fe0 JW |
365 | struct mem_cgroup_per_node *nodeinfo[0]; |
366 | /* WARNING: nodeinfo must be the last member here */ | |
8cdea7c0 BS |
367 | }; |
368 | ||
45cf7ebd GC |
369 | static size_t memcg_size(void) |
370 | { | |
371 | return sizeof(struct mem_cgroup) + | |
372 | nr_node_ids * sizeof(struct mem_cgroup_per_node); | |
373 | } | |
374 | ||
510fc4e1 GC |
375 | /* internal only representation about the status of kmem accounting. */ |
376 | enum { | |
377 | KMEM_ACCOUNTED_ACTIVE = 0, /* accounted by this cgroup itself */ | |
a8964b9b | 378 | KMEM_ACCOUNTED_ACTIVATED, /* static key enabled. */ |
7de37682 | 379 | KMEM_ACCOUNTED_DEAD, /* dead memcg with pending kmem charges */ |
510fc4e1 GC |
380 | }; |
381 | ||
a8964b9b GC |
382 | /* We account when limit is on, but only after call sites are patched */ |
383 | #define KMEM_ACCOUNTED_MASK \ | |
384 | ((1 << KMEM_ACCOUNTED_ACTIVE) | (1 << KMEM_ACCOUNTED_ACTIVATED)) | |
510fc4e1 GC |
385 | |
386 | #ifdef CONFIG_MEMCG_KMEM | |
387 | static inline void memcg_kmem_set_active(struct mem_cgroup *memcg) | |
388 | { | |
389 | set_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
390 | } | |
7de37682 GC |
391 | |
392 | static bool memcg_kmem_is_active(struct mem_cgroup *memcg) | |
393 | { | |
394 | return test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
395 | } | |
396 | ||
a8964b9b GC |
397 | static void memcg_kmem_set_activated(struct mem_cgroup *memcg) |
398 | { | |
399 | set_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags); | |
400 | } | |
401 | ||
55007d84 GC |
402 | static void memcg_kmem_clear_activated(struct mem_cgroup *memcg) |
403 | { | |
404 | clear_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags); | |
405 | } | |
406 | ||
7de37682 GC |
407 | static void memcg_kmem_mark_dead(struct mem_cgroup *memcg) |
408 | { | |
10d5ebf4 LZ |
409 | /* |
410 | * Our caller must use css_get() first, because memcg_uncharge_kmem() | |
411 | * will call css_put() if it sees the memcg is dead. | |
412 | */ | |
413 | smp_wmb(); | |
7de37682 GC |
414 | if (test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags)) |
415 | set_bit(KMEM_ACCOUNTED_DEAD, &memcg->kmem_account_flags); | |
416 | } | |
417 | ||
418 | static bool memcg_kmem_test_and_clear_dead(struct mem_cgroup *memcg) | |
419 | { | |
420 | return test_and_clear_bit(KMEM_ACCOUNTED_DEAD, | |
421 | &memcg->kmem_account_flags); | |
422 | } | |
510fc4e1 GC |
423 | #endif |
424 | ||
7dc74be0 DN |
425 | /* Stuffs for move charges at task migration. */ |
426 | /* | |
ee5e8472 GC |
427 | * Types of charges to be moved. "move_charge_at_immitgrate" and |
428 | * "immigrate_flags" are treated as a left-shifted bitmap of these types. | |
7dc74be0 DN |
429 | */ |
430 | enum move_type { | |
4ffef5fe | 431 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 432 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
433 | NR_MOVE_TYPE, |
434 | }; | |
435 | ||
4ffef5fe DN |
436 | /* "mc" and its members are protected by cgroup_mutex */ |
437 | static struct move_charge_struct { | |
b1dd693e | 438 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
439 | struct mem_cgroup *from; |
440 | struct mem_cgroup *to; | |
ee5e8472 | 441 | unsigned long immigrate_flags; |
4ffef5fe | 442 | unsigned long precharge; |
854ffa8d | 443 | unsigned long moved_charge; |
483c30b5 | 444 | unsigned long moved_swap; |
8033b97c DN |
445 | struct task_struct *moving_task; /* a task moving charges */ |
446 | wait_queue_head_t waitq; /* a waitq for other context */ | |
447 | } mc = { | |
2bd9bb20 | 448 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
449 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
450 | }; | |
4ffef5fe | 451 | |
90254a65 DN |
452 | static bool move_anon(void) |
453 | { | |
ee5e8472 | 454 | return test_bit(MOVE_CHARGE_TYPE_ANON, &mc.immigrate_flags); |
90254a65 DN |
455 | } |
456 | ||
87946a72 DN |
457 | static bool move_file(void) |
458 | { | |
ee5e8472 | 459 | return test_bit(MOVE_CHARGE_TYPE_FILE, &mc.immigrate_flags); |
87946a72 DN |
460 | } |
461 | ||
4e416953 BS |
462 | /* |
463 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
464 | * limit reclaim to prevent infinite loops, if they ever occur. | |
465 | */ | |
a0db00fc | 466 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 467 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 468 | |
217bc319 KH |
469 | enum charge_type { |
470 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 471 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 472 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 473 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
474 | NR_CHARGE_TYPE, |
475 | }; | |
476 | ||
8c7c6e34 | 477 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
478 | enum res_type { |
479 | _MEM, | |
480 | _MEMSWAP, | |
481 | _OOM_TYPE, | |
510fc4e1 | 482 | _KMEM, |
86ae53e1 GC |
483 | }; |
484 | ||
a0db00fc KS |
485 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
486 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 487 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
488 | /* Used for OOM nofiier */ |
489 | #define OOM_CONTROL (0) | |
8c7c6e34 | 490 | |
75822b44 BS |
491 | /* |
492 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
493 | */ | |
494 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
495 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
496 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
497 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
498 | ||
0999821b GC |
499 | /* |
500 | * The memcg_create_mutex will be held whenever a new cgroup is created. | |
501 | * As a consequence, any change that needs to protect against new child cgroups | |
502 | * appearing has to hold it as well. | |
503 | */ | |
504 | static DEFINE_MUTEX(memcg_create_mutex); | |
505 | ||
b2145145 WL |
506 | struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s) |
507 | { | |
a7c6d554 | 508 | return s ? container_of(s, struct mem_cgroup, css) : NULL; |
b2145145 WL |
509 | } |
510 | ||
70ddf637 AV |
511 | /* Some nice accessors for the vmpressure. */ |
512 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
513 | { | |
514 | if (!memcg) | |
515 | memcg = root_mem_cgroup; | |
516 | return &memcg->vmpressure; | |
517 | } | |
518 | ||
519 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
520 | { | |
521 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
522 | } | |
523 | ||
524 | struct vmpressure *css_to_vmpressure(struct cgroup_subsys_state *css) | |
525 | { | |
526 | return &mem_cgroup_from_css(css)->vmpressure; | |
527 | } | |
528 | ||
7ffc0edc MH |
529 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
530 | { | |
531 | return (memcg == root_mem_cgroup); | |
532 | } | |
533 | ||
e1aab161 | 534 | /* Writing them here to avoid exposing memcg's inner layout */ |
4bd2c1ee | 535 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
e1aab161 | 536 | |
e1aab161 GC |
537 | void sock_update_memcg(struct sock *sk) |
538 | { | |
376be5ff | 539 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 540 | struct mem_cgroup *memcg; |
3f134619 | 541 | struct cg_proto *cg_proto; |
e1aab161 GC |
542 | |
543 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
544 | ||
f3f511e1 GC |
545 | /* Socket cloning can throw us here with sk_cgrp already |
546 | * filled. It won't however, necessarily happen from | |
547 | * process context. So the test for root memcg given | |
548 | * the current task's memcg won't help us in this case. | |
549 | * | |
550 | * Respecting the original socket's memcg is a better | |
551 | * decision in this case. | |
552 | */ | |
553 | if (sk->sk_cgrp) { | |
554 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
5347e5ae | 555 | css_get(&sk->sk_cgrp->memcg->css); |
f3f511e1 GC |
556 | return; |
557 | } | |
558 | ||
e1aab161 GC |
559 | rcu_read_lock(); |
560 | memcg = mem_cgroup_from_task(current); | |
3f134619 | 561 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
5347e5ae LZ |
562 | if (!mem_cgroup_is_root(memcg) && |
563 | memcg_proto_active(cg_proto) && css_tryget(&memcg->css)) { | |
3f134619 | 564 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
565 | } |
566 | rcu_read_unlock(); | |
567 | } | |
568 | } | |
569 | EXPORT_SYMBOL(sock_update_memcg); | |
570 | ||
571 | void sock_release_memcg(struct sock *sk) | |
572 | { | |
376be5ff | 573 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
574 | struct mem_cgroup *memcg; |
575 | WARN_ON(!sk->sk_cgrp->memcg); | |
576 | memcg = sk->sk_cgrp->memcg; | |
5347e5ae | 577 | css_put(&sk->sk_cgrp->memcg->css); |
e1aab161 GC |
578 | } |
579 | } | |
d1a4c0b3 GC |
580 | |
581 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
582 | { | |
583 | if (!memcg || mem_cgroup_is_root(memcg)) | |
584 | return NULL; | |
585 | ||
586 | return &memcg->tcp_mem.cg_proto; | |
587 | } | |
588 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 589 | |
3f134619 GC |
590 | static void disarm_sock_keys(struct mem_cgroup *memcg) |
591 | { | |
592 | if (!memcg_proto_activated(&memcg->tcp_mem.cg_proto)) | |
593 | return; | |
594 | static_key_slow_dec(&memcg_socket_limit_enabled); | |
595 | } | |
596 | #else | |
597 | static void disarm_sock_keys(struct mem_cgroup *memcg) | |
598 | { | |
599 | } | |
600 | #endif | |
601 | ||
a8964b9b | 602 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 GC |
603 | /* |
604 | * This will be the memcg's index in each cache's ->memcg_params->memcg_caches. | |
605 | * There are two main reasons for not using the css_id for this: | |
606 | * 1) this works better in sparse environments, where we have a lot of memcgs, | |
607 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
608 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
609 | * 200 entry array for that. | |
610 | * | |
611 | * 2) In order not to violate the cgroup API, we would like to do all memory | |
612 | * allocation in ->create(). At that point, we haven't yet allocated the | |
613 | * css_id. Having a separate index prevents us from messing with the cgroup | |
614 | * core for this | |
615 | * | |
616 | * The current size of the caches array is stored in | |
617 | * memcg_limited_groups_array_size. It will double each time we have to | |
618 | * increase it. | |
619 | */ | |
620 | static DEFINE_IDA(kmem_limited_groups); | |
749c5415 GC |
621 | int memcg_limited_groups_array_size; |
622 | ||
55007d84 GC |
623 | /* |
624 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
625 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
626 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
627 | * tunable, but that is strictly not necessary. | |
628 | * | |
629 | * MAX_SIZE should be as large as the number of css_ids. Ideally, we could get | |
630 | * this constant directly from cgroup, but it is understandable that this is | |
631 | * better kept as an internal representation in cgroup.c. In any case, the | |
632 | * css_id space is not getting any smaller, and we don't have to necessarily | |
633 | * increase ours as well if it increases. | |
634 | */ | |
635 | #define MEMCG_CACHES_MIN_SIZE 4 | |
636 | #define MEMCG_CACHES_MAX_SIZE 65535 | |
637 | ||
d7f25f8a GC |
638 | /* |
639 | * A lot of the calls to the cache allocation functions are expected to be | |
640 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
641 | * conditional to this static branch, we'll have to allow modules that does | |
642 | * kmem_cache_alloc and the such to see this symbol as well | |
643 | */ | |
a8964b9b | 644 | struct static_key memcg_kmem_enabled_key; |
d7f25f8a | 645 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b GC |
646 | |
647 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
648 | { | |
55007d84 | 649 | if (memcg_kmem_is_active(memcg)) { |
a8964b9b | 650 | static_key_slow_dec(&memcg_kmem_enabled_key); |
55007d84 GC |
651 | ida_simple_remove(&kmem_limited_groups, memcg->kmemcg_id); |
652 | } | |
bea207c8 GC |
653 | /* |
654 | * This check can't live in kmem destruction function, | |
655 | * since the charges will outlive the cgroup | |
656 | */ | |
657 | WARN_ON(res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0); | |
a8964b9b GC |
658 | } |
659 | #else | |
660 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
661 | { | |
662 | } | |
663 | #endif /* CONFIG_MEMCG_KMEM */ | |
664 | ||
665 | static void disarm_static_keys(struct mem_cgroup *memcg) | |
666 | { | |
667 | disarm_sock_keys(memcg); | |
668 | disarm_kmem_keys(memcg); | |
669 | } | |
670 | ||
c0ff4b85 | 671 | static void drain_all_stock_async(struct mem_cgroup *memcg); |
8c7c6e34 | 672 | |
f64c3f54 | 673 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 674 | mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid) |
f64c3f54 | 675 | { |
45cf7ebd | 676 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
54f72fe0 | 677 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
678 | } |
679 | ||
c0ff4b85 | 680 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 681 | { |
c0ff4b85 | 682 | return &memcg->css; |
d324236b WF |
683 | } |
684 | ||
f64c3f54 | 685 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 686 | page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 687 | { |
97a6c37b JW |
688 | int nid = page_to_nid(page); |
689 | int zid = page_zonenum(page); | |
f64c3f54 | 690 | |
c0ff4b85 | 691 | return mem_cgroup_zoneinfo(memcg, nid, zid); |
f64c3f54 BS |
692 | } |
693 | ||
bb4cc1a8 AM |
694 | static struct mem_cgroup_tree_per_zone * |
695 | soft_limit_tree_node_zone(int nid, int zid) | |
696 | { | |
697 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
698 | } | |
699 | ||
700 | static struct mem_cgroup_tree_per_zone * | |
701 | soft_limit_tree_from_page(struct page *page) | |
702 | { | |
703 | int nid = page_to_nid(page); | |
704 | int zid = page_zonenum(page); | |
705 | ||
706 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
707 | } | |
708 | ||
709 | static void | |
710 | __mem_cgroup_insert_exceeded(struct mem_cgroup *memcg, | |
711 | struct mem_cgroup_per_zone *mz, | |
712 | struct mem_cgroup_tree_per_zone *mctz, | |
713 | unsigned long long new_usage_in_excess) | |
714 | { | |
715 | struct rb_node **p = &mctz->rb_root.rb_node; | |
716 | struct rb_node *parent = NULL; | |
717 | struct mem_cgroup_per_zone *mz_node; | |
718 | ||
719 | if (mz->on_tree) | |
720 | return; | |
721 | ||
722 | mz->usage_in_excess = new_usage_in_excess; | |
723 | if (!mz->usage_in_excess) | |
724 | return; | |
725 | while (*p) { | |
726 | parent = *p; | |
727 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
728 | tree_node); | |
729 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
730 | p = &(*p)->rb_left; | |
731 | /* | |
732 | * We can't avoid mem cgroups that are over their soft | |
733 | * limit by the same amount | |
734 | */ | |
735 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
736 | p = &(*p)->rb_right; | |
737 | } | |
738 | rb_link_node(&mz->tree_node, parent, p); | |
739 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
740 | mz->on_tree = true; | |
741 | } | |
742 | ||
743 | static void | |
744 | __mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, | |
745 | struct mem_cgroup_per_zone *mz, | |
746 | struct mem_cgroup_tree_per_zone *mctz) | |
747 | { | |
748 | if (!mz->on_tree) | |
749 | return; | |
750 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
751 | mz->on_tree = false; | |
752 | } | |
753 | ||
754 | static void | |
755 | mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, | |
756 | struct mem_cgroup_per_zone *mz, | |
757 | struct mem_cgroup_tree_per_zone *mctz) | |
758 | { | |
759 | spin_lock(&mctz->lock); | |
760 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); | |
761 | spin_unlock(&mctz->lock); | |
762 | } | |
763 | ||
764 | ||
765 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
766 | { | |
767 | unsigned long long excess; | |
768 | struct mem_cgroup_per_zone *mz; | |
769 | struct mem_cgroup_tree_per_zone *mctz; | |
770 | int nid = page_to_nid(page); | |
771 | int zid = page_zonenum(page); | |
772 | mctz = soft_limit_tree_from_page(page); | |
773 | ||
774 | /* | |
775 | * Necessary to update all ancestors when hierarchy is used. | |
776 | * because their event counter is not touched. | |
777 | */ | |
778 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
779 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
780 | excess = res_counter_soft_limit_excess(&memcg->res); | |
781 | /* | |
782 | * We have to update the tree if mz is on RB-tree or | |
783 | * mem is over its softlimit. | |
784 | */ | |
785 | if (excess || mz->on_tree) { | |
786 | spin_lock(&mctz->lock); | |
787 | /* if on-tree, remove it */ | |
788 | if (mz->on_tree) | |
789 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); | |
790 | /* | |
791 | * Insert again. mz->usage_in_excess will be updated. | |
792 | * If excess is 0, no tree ops. | |
793 | */ | |
794 | __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess); | |
795 | spin_unlock(&mctz->lock); | |
796 | } | |
797 | } | |
798 | } | |
799 | ||
800 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
801 | { | |
802 | int node, zone; | |
803 | struct mem_cgroup_per_zone *mz; | |
804 | struct mem_cgroup_tree_per_zone *mctz; | |
805 | ||
806 | for_each_node(node) { | |
807 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
808 | mz = mem_cgroup_zoneinfo(memcg, node, zone); | |
809 | mctz = soft_limit_tree_node_zone(node, zone); | |
810 | mem_cgroup_remove_exceeded(memcg, mz, mctz); | |
811 | } | |
812 | } | |
813 | } | |
814 | ||
815 | static struct mem_cgroup_per_zone * | |
816 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
817 | { | |
818 | struct rb_node *rightmost = NULL; | |
819 | struct mem_cgroup_per_zone *mz; | |
820 | ||
821 | retry: | |
822 | mz = NULL; | |
823 | rightmost = rb_last(&mctz->rb_root); | |
824 | if (!rightmost) | |
825 | goto done; /* Nothing to reclaim from */ | |
826 | ||
827 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
828 | /* | |
829 | * Remove the node now but someone else can add it back, | |
830 | * we will to add it back at the end of reclaim to its correct | |
831 | * position in the tree. | |
832 | */ | |
833 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); | |
834 | if (!res_counter_soft_limit_excess(&mz->memcg->res) || | |
835 | !css_tryget(&mz->memcg->css)) | |
836 | goto retry; | |
837 | done: | |
838 | return mz; | |
839 | } | |
840 | ||
841 | static struct mem_cgroup_per_zone * | |
842 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
843 | { | |
844 | struct mem_cgroup_per_zone *mz; | |
845 | ||
846 | spin_lock(&mctz->lock); | |
847 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
848 | spin_unlock(&mctz->lock); | |
849 | return mz; | |
850 | } | |
851 | ||
711d3d2c KH |
852 | /* |
853 | * Implementation Note: reading percpu statistics for memcg. | |
854 | * | |
855 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
856 | * synchronization to implement "quick" read. There are trade-off between | |
857 | * reading cost and precision of value. Then, we may have a chance to implement | |
858 | * a periodic synchronizion of counter in memcg's counter. | |
859 | * | |
860 | * But this _read() function is used for user interface now. The user accounts | |
861 | * memory usage by memory cgroup and he _always_ requires exact value because | |
862 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
863 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
864 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
865 | * | |
866 | * If there are kernel internal actions which can make use of some not-exact | |
867 | * value, and reading all cpu value can be performance bottleneck in some | |
868 | * common workload, threashold and synchonization as vmstat[] should be | |
869 | * implemented. | |
870 | */ | |
c0ff4b85 | 871 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 872 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 873 | { |
7a159cc9 | 874 | long val = 0; |
c62b1a3b | 875 | int cpu; |
c62b1a3b | 876 | |
711d3d2c KH |
877 | get_online_cpus(); |
878 | for_each_online_cpu(cpu) | |
c0ff4b85 | 879 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 880 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
881 | spin_lock(&memcg->pcp_counter_lock); |
882 | val += memcg->nocpu_base.count[idx]; | |
883 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
884 | #endif |
885 | put_online_cpus(); | |
c62b1a3b KH |
886 | return val; |
887 | } | |
888 | ||
c0ff4b85 | 889 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
0c3e73e8 BS |
890 | bool charge) |
891 | { | |
892 | int val = (charge) ? 1 : -1; | |
bff6bb83 | 893 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); |
0c3e73e8 BS |
894 | } |
895 | ||
c0ff4b85 | 896 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
897 | enum mem_cgroup_events_index idx) |
898 | { | |
899 | unsigned long val = 0; | |
900 | int cpu; | |
901 | ||
9c567512 | 902 | get_online_cpus(); |
e9f8974f | 903 | for_each_online_cpu(cpu) |
c0ff4b85 | 904 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 905 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
906 | spin_lock(&memcg->pcp_counter_lock); |
907 | val += memcg->nocpu_base.events[idx]; | |
908 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f | 909 | #endif |
9c567512 | 910 | put_online_cpus(); |
e9f8974f JW |
911 | return val; |
912 | } | |
913 | ||
c0ff4b85 | 914 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 915 | struct page *page, |
b2402857 | 916 | bool anon, int nr_pages) |
d52aa412 | 917 | { |
c62b1a3b KH |
918 | preempt_disable(); |
919 | ||
b2402857 KH |
920 | /* |
921 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
922 | * counted as CACHE even if it's on ANON LRU. | |
923 | */ | |
924 | if (anon) | |
925 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], | |
c0ff4b85 | 926 | nr_pages); |
d52aa412 | 927 | else |
b2402857 | 928 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 929 | nr_pages); |
55e462b0 | 930 | |
b070e65c DR |
931 | if (PageTransHuge(page)) |
932 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], | |
933 | nr_pages); | |
934 | ||
e401f176 KH |
935 | /* pagein of a big page is an event. So, ignore page size */ |
936 | if (nr_pages > 0) | |
c0ff4b85 | 937 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 938 | else { |
c0ff4b85 | 939 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
940 | nr_pages = -nr_pages; /* for event */ |
941 | } | |
e401f176 | 942 | |
13114716 | 943 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
2e72b634 | 944 | |
c62b1a3b | 945 | preempt_enable(); |
6d12e2d8 KH |
946 | } |
947 | ||
bb2a0de9 | 948 | unsigned long |
4d7dcca2 | 949 | mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
950 | { |
951 | struct mem_cgroup_per_zone *mz; | |
952 | ||
953 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
954 | return mz->lru_size[lru]; | |
955 | } | |
956 | ||
957 | static unsigned long | |
c0ff4b85 | 958 | mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid, |
bb2a0de9 | 959 | unsigned int lru_mask) |
889976db YH |
960 | { |
961 | struct mem_cgroup_per_zone *mz; | |
f156ab93 | 962 | enum lru_list lru; |
bb2a0de9 KH |
963 | unsigned long ret = 0; |
964 | ||
c0ff4b85 | 965 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
bb2a0de9 | 966 | |
f156ab93 HD |
967 | for_each_lru(lru) { |
968 | if (BIT(lru) & lru_mask) | |
969 | ret += mz->lru_size[lru]; | |
bb2a0de9 KH |
970 | } |
971 | return ret; | |
972 | } | |
973 | ||
974 | static unsigned long | |
c0ff4b85 | 975 | mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 KH |
976 | int nid, unsigned int lru_mask) |
977 | { | |
889976db YH |
978 | u64 total = 0; |
979 | int zid; | |
980 | ||
bb2a0de9 | 981 | for (zid = 0; zid < MAX_NR_ZONES; zid++) |
c0ff4b85 R |
982 | total += mem_cgroup_zone_nr_lru_pages(memcg, |
983 | nid, zid, lru_mask); | |
bb2a0de9 | 984 | |
889976db YH |
985 | return total; |
986 | } | |
bb2a0de9 | 987 | |
c0ff4b85 | 988 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 989 | unsigned int lru_mask) |
6d12e2d8 | 990 | { |
889976db | 991 | int nid; |
6d12e2d8 KH |
992 | u64 total = 0; |
993 | ||
31aaea4a | 994 | for_each_node_state(nid, N_MEMORY) |
c0ff4b85 | 995 | total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
6d12e2d8 | 996 | return total; |
d52aa412 KH |
997 | } |
998 | ||
f53d7ce3 JW |
999 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
1000 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
1001 | { |
1002 | unsigned long val, next; | |
1003 | ||
13114716 | 1004 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 1005 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 1006 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
1007 | if ((long)next - (long)val < 0) { |
1008 | switch (target) { | |
1009 | case MEM_CGROUP_TARGET_THRESH: | |
1010 | next = val + THRESHOLDS_EVENTS_TARGET; | |
1011 | break; | |
bb4cc1a8 AM |
1012 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
1013 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
1014 | break; | |
f53d7ce3 JW |
1015 | case MEM_CGROUP_TARGET_NUMAINFO: |
1016 | next = val + NUMAINFO_EVENTS_TARGET; | |
1017 | break; | |
1018 | default: | |
1019 | break; | |
1020 | } | |
1021 | __this_cpu_write(memcg->stat->targets[target], next); | |
1022 | return true; | |
7a159cc9 | 1023 | } |
f53d7ce3 | 1024 | return false; |
d2265e6f KH |
1025 | } |
1026 | ||
1027 | /* | |
1028 | * Check events in order. | |
1029 | * | |
1030 | */ | |
c0ff4b85 | 1031 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f | 1032 | { |
4799401f | 1033 | preempt_disable(); |
d2265e6f | 1034 | /* threshold event is triggered in finer grain than soft limit */ |
f53d7ce3 JW |
1035 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
1036 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 1037 | bool do_softlimit; |
82b3f2a7 | 1038 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 1039 | |
bb4cc1a8 AM |
1040 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
1041 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
1042 | #if MAX_NUMNODES > 1 |
1043 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
1044 | MEM_CGROUP_TARGET_NUMAINFO); | |
1045 | #endif | |
1046 | preempt_enable(); | |
1047 | ||
c0ff4b85 | 1048 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
1049 | if (unlikely(do_softlimit)) |
1050 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 1051 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 1052 | if (unlikely(do_numainfo)) |
c0ff4b85 | 1053 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 1054 | #endif |
f53d7ce3 JW |
1055 | } else |
1056 | preempt_enable(); | |
d2265e6f KH |
1057 | } |
1058 | ||
cf475ad2 | 1059 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 1060 | { |
31a78f23 BS |
1061 | /* |
1062 | * mm_update_next_owner() may clear mm->owner to NULL | |
1063 | * if it races with swapoff, page migration, etc. | |
1064 | * So this can be called with p == NULL. | |
1065 | */ | |
1066 | if (unlikely(!p)) | |
1067 | return NULL; | |
1068 | ||
8af01f56 | 1069 | return mem_cgroup_from_css(task_css(p, mem_cgroup_subsys_id)); |
78fb7466 PE |
1070 | } |
1071 | ||
a433658c | 1072 | struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 1073 | { |
c0ff4b85 | 1074 | struct mem_cgroup *memcg = NULL; |
0b7f569e KH |
1075 | |
1076 | if (!mm) | |
1077 | return NULL; | |
54595fe2 KH |
1078 | /* |
1079 | * Because we have no locks, mm->owner's may be being moved to other | |
1080 | * cgroup. We use css_tryget() here even if this looks | |
1081 | * pessimistic (rather than adding locks here). | |
1082 | */ | |
1083 | rcu_read_lock(); | |
1084 | do { | |
c0ff4b85 R |
1085 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1086 | if (unlikely(!memcg)) | |
54595fe2 | 1087 | break; |
c0ff4b85 | 1088 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 1089 | rcu_read_unlock(); |
c0ff4b85 | 1090 | return memcg; |
54595fe2 KH |
1091 | } |
1092 | ||
16248d8f MH |
1093 | /* |
1094 | * Returns a next (in a pre-order walk) alive memcg (with elevated css | |
1095 | * ref. count) or NULL if the whole root's subtree has been visited. | |
1096 | * | |
1097 | * helper function to be used by mem_cgroup_iter | |
1098 | */ | |
1099 | static struct mem_cgroup *__mem_cgroup_iter_next(struct mem_cgroup *root, | |
694fbc0f | 1100 | struct mem_cgroup *last_visited) |
16248d8f | 1101 | { |
492eb21b | 1102 | struct cgroup_subsys_state *prev_css, *next_css; |
16248d8f | 1103 | |
bd8815a6 | 1104 | prev_css = last_visited ? &last_visited->css : NULL; |
16248d8f | 1105 | skip_node: |
492eb21b | 1106 | next_css = css_next_descendant_pre(prev_css, &root->css); |
16248d8f MH |
1107 | |
1108 | /* | |
1109 | * Even if we found a group we have to make sure it is | |
1110 | * alive. css && !memcg means that the groups should be | |
1111 | * skipped and we should continue the tree walk. | |
1112 | * last_visited css is safe to use because it is | |
1113 | * protected by css_get and the tree walk is rcu safe. | |
1114 | */ | |
492eb21b TH |
1115 | if (next_css) { |
1116 | struct mem_cgroup *mem = mem_cgroup_from_css(next_css); | |
1117 | ||
694fbc0f AM |
1118 | if (css_tryget(&mem->css)) |
1119 | return mem; | |
1120 | else { | |
492eb21b | 1121 | prev_css = next_css; |
16248d8f MH |
1122 | goto skip_node; |
1123 | } | |
1124 | } | |
1125 | ||
1126 | return NULL; | |
1127 | } | |
1128 | ||
519ebea3 JW |
1129 | static void mem_cgroup_iter_invalidate(struct mem_cgroup *root) |
1130 | { | |
1131 | /* | |
1132 | * When a group in the hierarchy below root is destroyed, the | |
1133 | * hierarchy iterator can no longer be trusted since it might | |
1134 | * have pointed to the destroyed group. Invalidate it. | |
1135 | */ | |
1136 | atomic_inc(&root->dead_count); | |
1137 | } | |
1138 | ||
1139 | static struct mem_cgroup * | |
1140 | mem_cgroup_iter_load(struct mem_cgroup_reclaim_iter *iter, | |
1141 | struct mem_cgroup *root, | |
1142 | int *sequence) | |
1143 | { | |
1144 | struct mem_cgroup *position = NULL; | |
1145 | /* | |
1146 | * A cgroup destruction happens in two stages: offlining and | |
1147 | * release. They are separated by a RCU grace period. | |
1148 | * | |
1149 | * If the iterator is valid, we may still race with an | |
1150 | * offlining. The RCU lock ensures the object won't be | |
1151 | * released, tryget will fail if we lost the race. | |
1152 | */ | |
1153 | *sequence = atomic_read(&root->dead_count); | |
1154 | if (iter->last_dead_count == *sequence) { | |
1155 | smp_rmb(); | |
1156 | position = iter->last_visited; | |
1157 | if (position && !css_tryget(&position->css)) | |
1158 | position = NULL; | |
1159 | } | |
1160 | return position; | |
1161 | } | |
1162 | ||
1163 | static void mem_cgroup_iter_update(struct mem_cgroup_reclaim_iter *iter, | |
1164 | struct mem_cgroup *last_visited, | |
1165 | struct mem_cgroup *new_position, | |
1166 | int sequence) | |
1167 | { | |
1168 | if (last_visited) | |
1169 | css_put(&last_visited->css); | |
1170 | /* | |
1171 | * We store the sequence count from the time @last_visited was | |
1172 | * loaded successfully instead of rereading it here so that we | |
1173 | * don't lose destruction events in between. We could have | |
1174 | * raced with the destruction of @new_position after all. | |
1175 | */ | |
1176 | iter->last_visited = new_position; | |
1177 | smp_wmb(); | |
1178 | iter->last_dead_count = sequence; | |
1179 | } | |
1180 | ||
5660048c JW |
1181 | /** |
1182 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1183 | * @root: hierarchy root | |
1184 | * @prev: previously returned memcg, NULL on first invocation | |
1185 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1186 | * | |
1187 | * Returns references to children of the hierarchy below @root, or | |
1188 | * @root itself, or %NULL after a full round-trip. | |
1189 | * | |
1190 | * Caller must pass the return value in @prev on subsequent | |
1191 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1192 | * to cancel a hierarchy walk before the round-trip is complete. | |
1193 | * | |
1194 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
1195 | * divide up the memcgs in the hierarchy among all concurrent | |
1196 | * reclaimers operating on the same zone and priority. | |
1197 | */ | |
694fbc0f | 1198 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1199 | struct mem_cgroup *prev, |
694fbc0f | 1200 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1201 | { |
9f3a0d09 | 1202 | struct mem_cgroup *memcg = NULL; |
542f85f9 | 1203 | struct mem_cgroup *last_visited = NULL; |
711d3d2c | 1204 | |
694fbc0f AM |
1205 | if (mem_cgroup_disabled()) |
1206 | return NULL; | |
5660048c | 1207 | |
9f3a0d09 JW |
1208 | if (!root) |
1209 | root = root_mem_cgroup; | |
7d74b06f | 1210 | |
9f3a0d09 | 1211 | if (prev && !reclaim) |
542f85f9 | 1212 | last_visited = prev; |
14067bb3 | 1213 | |
9f3a0d09 JW |
1214 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1215 | if (prev) | |
c40046f3 | 1216 | goto out_css_put; |
694fbc0f | 1217 | return root; |
9f3a0d09 | 1218 | } |
14067bb3 | 1219 | |
542f85f9 | 1220 | rcu_read_lock(); |
9f3a0d09 | 1221 | while (!memcg) { |
527a5ec9 | 1222 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
519ebea3 | 1223 | int uninitialized_var(seq); |
711d3d2c | 1224 | |
527a5ec9 JW |
1225 | if (reclaim) { |
1226 | int nid = zone_to_nid(reclaim->zone); | |
1227 | int zid = zone_idx(reclaim->zone); | |
1228 | struct mem_cgroup_per_zone *mz; | |
1229 | ||
1230 | mz = mem_cgroup_zoneinfo(root, nid, zid); | |
1231 | iter = &mz->reclaim_iter[reclaim->priority]; | |
542f85f9 | 1232 | if (prev && reclaim->generation != iter->generation) { |
5f578161 | 1233 | iter->last_visited = NULL; |
542f85f9 MH |
1234 | goto out_unlock; |
1235 | } | |
5f578161 | 1236 | |
519ebea3 | 1237 | last_visited = mem_cgroup_iter_load(iter, root, &seq); |
527a5ec9 | 1238 | } |
7d74b06f | 1239 | |
694fbc0f | 1240 | memcg = __mem_cgroup_iter_next(root, last_visited); |
14067bb3 | 1241 | |
527a5ec9 | 1242 | if (reclaim) { |
519ebea3 | 1243 | mem_cgroup_iter_update(iter, last_visited, memcg, seq); |
542f85f9 | 1244 | |
19f39402 | 1245 | if (!memcg) |
527a5ec9 JW |
1246 | iter->generation++; |
1247 | else if (!prev && memcg) | |
1248 | reclaim->generation = iter->generation; | |
1249 | } | |
9f3a0d09 | 1250 | |
694fbc0f | 1251 | if (prev && !memcg) |
542f85f9 | 1252 | goto out_unlock; |
9f3a0d09 | 1253 | } |
542f85f9 MH |
1254 | out_unlock: |
1255 | rcu_read_unlock(); | |
c40046f3 MH |
1256 | out_css_put: |
1257 | if (prev && prev != root) | |
1258 | css_put(&prev->css); | |
1259 | ||
9f3a0d09 | 1260 | return memcg; |
14067bb3 | 1261 | } |
7d74b06f | 1262 | |
5660048c JW |
1263 | /** |
1264 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1265 | * @root: hierarchy root | |
1266 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1267 | */ | |
1268 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1269 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1270 | { |
1271 | if (!root) | |
1272 | root = root_mem_cgroup; | |
1273 | if (prev && prev != root) | |
1274 | css_put(&prev->css); | |
1275 | } | |
7d74b06f | 1276 | |
9f3a0d09 JW |
1277 | /* |
1278 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1279 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1280 | * be used for reference counting. | |
1281 | */ | |
1282 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1283 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1284 | iter != NULL; \ |
527a5ec9 | 1285 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1286 | |
9f3a0d09 | 1287 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1288 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1289 | iter != NULL; \ |
527a5ec9 | 1290 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1291 | |
68ae564b | 1292 | void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
456f998e | 1293 | { |
c0ff4b85 | 1294 | struct mem_cgroup *memcg; |
456f998e | 1295 | |
456f998e | 1296 | rcu_read_lock(); |
c0ff4b85 R |
1297 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1298 | if (unlikely(!memcg)) | |
456f998e YH |
1299 | goto out; |
1300 | ||
1301 | switch (idx) { | |
456f998e | 1302 | case PGFAULT: |
0e574a93 JW |
1303 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1304 | break; | |
1305 | case PGMAJFAULT: | |
1306 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1307 | break; |
1308 | default: | |
1309 | BUG(); | |
1310 | } | |
1311 | out: | |
1312 | rcu_read_unlock(); | |
1313 | } | |
68ae564b | 1314 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
456f998e | 1315 | |
925b7673 JW |
1316 | /** |
1317 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1318 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1319 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1320 | * |
1321 | * Returns the lru list vector holding pages for the given @zone and | |
1322 | * @mem. This can be the global zone lruvec, if the memory controller | |
1323 | * is disabled. | |
1324 | */ | |
1325 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1326 | struct mem_cgroup *memcg) | |
1327 | { | |
1328 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1329 | struct lruvec *lruvec; |
925b7673 | 1330 | |
bea8c150 HD |
1331 | if (mem_cgroup_disabled()) { |
1332 | lruvec = &zone->lruvec; | |
1333 | goto out; | |
1334 | } | |
925b7673 JW |
1335 | |
1336 | mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone)); | |
bea8c150 HD |
1337 | lruvec = &mz->lruvec; |
1338 | out: | |
1339 | /* | |
1340 | * Since a node can be onlined after the mem_cgroup was created, | |
1341 | * we have to be prepared to initialize lruvec->zone here; | |
1342 | * and if offlined then reonlined, we need to reinitialize it. | |
1343 | */ | |
1344 | if (unlikely(lruvec->zone != zone)) | |
1345 | lruvec->zone = zone; | |
1346 | return lruvec; | |
925b7673 JW |
1347 | } |
1348 | ||
08e552c6 KH |
1349 | /* |
1350 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
1351 | * Operations are called by routine of global LRU independently from memcg. | |
1352 | * What we have to take care of here is validness of pc->mem_cgroup. | |
1353 | * | |
1354 | * Changes to pc->mem_cgroup happens when | |
1355 | * 1. charge | |
1356 | * 2. moving account | |
1357 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
1358 | * It is added to LRU before charge. | |
1359 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
1360 | * When moving account, the page is not on LRU. It's isolated. | |
1361 | */ | |
4f98a2fe | 1362 | |
925b7673 | 1363 | /** |
fa9add64 | 1364 | * mem_cgroup_page_lruvec - return lruvec for adding an lru page |
925b7673 | 1365 | * @page: the page |
fa9add64 | 1366 | * @zone: zone of the page |
925b7673 | 1367 | */ |
fa9add64 | 1368 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1369 | { |
08e552c6 | 1370 | struct mem_cgroup_per_zone *mz; |
925b7673 JW |
1371 | struct mem_cgroup *memcg; |
1372 | struct page_cgroup *pc; | |
bea8c150 | 1373 | struct lruvec *lruvec; |
6d12e2d8 | 1374 | |
bea8c150 HD |
1375 | if (mem_cgroup_disabled()) { |
1376 | lruvec = &zone->lruvec; | |
1377 | goto out; | |
1378 | } | |
925b7673 | 1379 | |
08e552c6 | 1380 | pc = lookup_page_cgroup(page); |
38c5d72f | 1381 | memcg = pc->mem_cgroup; |
7512102c HD |
1382 | |
1383 | /* | |
fa9add64 | 1384 | * Surreptitiously switch any uncharged offlist page to root: |
7512102c HD |
1385 | * an uncharged page off lru does nothing to secure |
1386 | * its former mem_cgroup from sudden removal. | |
1387 | * | |
1388 | * Our caller holds lru_lock, and PageCgroupUsed is updated | |
1389 | * under page_cgroup lock: between them, they make all uses | |
1390 | * of pc->mem_cgroup safe. | |
1391 | */ | |
fa9add64 | 1392 | if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup) |
7512102c HD |
1393 | pc->mem_cgroup = memcg = root_mem_cgroup; |
1394 | ||
925b7673 | 1395 | mz = page_cgroup_zoneinfo(memcg, page); |
bea8c150 HD |
1396 | lruvec = &mz->lruvec; |
1397 | out: | |
1398 | /* | |
1399 | * Since a node can be onlined after the mem_cgroup was created, | |
1400 | * we have to be prepared to initialize lruvec->zone here; | |
1401 | * and if offlined then reonlined, we need to reinitialize it. | |
1402 | */ | |
1403 | if (unlikely(lruvec->zone != zone)) | |
1404 | lruvec->zone = zone; | |
1405 | return lruvec; | |
08e552c6 | 1406 | } |
b69408e8 | 1407 | |
925b7673 | 1408 | /** |
fa9add64 HD |
1409 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1410 | * @lruvec: mem_cgroup per zone lru vector | |
1411 | * @lru: index of lru list the page is sitting on | |
1412 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1413 | * |
fa9add64 HD |
1414 | * This function must be called when a page is added to or removed from an |
1415 | * lru list. | |
3f58a829 | 1416 | */ |
fa9add64 HD |
1417 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1418 | int nr_pages) | |
3f58a829 MK |
1419 | { |
1420 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1421 | unsigned long *lru_size; |
3f58a829 MK |
1422 | |
1423 | if (mem_cgroup_disabled()) | |
1424 | return; | |
1425 | ||
fa9add64 HD |
1426 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1427 | lru_size = mz->lru_size + lru; | |
1428 | *lru_size += nr_pages; | |
1429 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1430 | } |
544122e5 | 1431 | |
3e92041d | 1432 | /* |
c0ff4b85 | 1433 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1434 | * hierarchy subtree |
1435 | */ | |
c3ac9a8a JW |
1436 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1437 | struct mem_cgroup *memcg) | |
3e92041d | 1438 | { |
91c63734 JW |
1439 | if (root_memcg == memcg) |
1440 | return true; | |
3a981f48 | 1441 | if (!root_memcg->use_hierarchy || !memcg) |
91c63734 | 1442 | return false; |
c3ac9a8a JW |
1443 | return css_is_ancestor(&memcg->css, &root_memcg->css); |
1444 | } | |
1445 | ||
1446 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, | |
1447 | struct mem_cgroup *memcg) | |
1448 | { | |
1449 | bool ret; | |
1450 | ||
91c63734 | 1451 | rcu_read_lock(); |
c3ac9a8a | 1452 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
91c63734 JW |
1453 | rcu_read_unlock(); |
1454 | return ret; | |
3e92041d MH |
1455 | } |
1456 | ||
ffbdccf5 DR |
1457 | bool task_in_mem_cgroup(struct task_struct *task, |
1458 | const struct mem_cgroup *memcg) | |
4c4a2214 | 1459 | { |
0b7f569e | 1460 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1461 | struct task_struct *p; |
ffbdccf5 | 1462 | bool ret; |
4c4a2214 | 1463 | |
158e0a2d | 1464 | p = find_lock_task_mm(task); |
de077d22 DR |
1465 | if (p) { |
1466 | curr = try_get_mem_cgroup_from_mm(p->mm); | |
1467 | task_unlock(p); | |
1468 | } else { | |
1469 | /* | |
1470 | * All threads may have already detached their mm's, but the oom | |
1471 | * killer still needs to detect if they have already been oom | |
1472 | * killed to prevent needlessly killing additional tasks. | |
1473 | */ | |
ffbdccf5 | 1474 | rcu_read_lock(); |
de077d22 DR |
1475 | curr = mem_cgroup_from_task(task); |
1476 | if (curr) | |
1477 | css_get(&curr->css); | |
ffbdccf5 | 1478 | rcu_read_unlock(); |
de077d22 | 1479 | } |
0b7f569e | 1480 | if (!curr) |
ffbdccf5 | 1481 | return false; |
d31f56db | 1482 | /* |
c0ff4b85 | 1483 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1484 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1485 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1486 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1487 | */ |
c0ff4b85 | 1488 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1489 | css_put(&curr->css); |
4c4a2214 DR |
1490 | return ret; |
1491 | } | |
1492 | ||
c56d5c7d | 1493 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1494 | { |
9b272977 | 1495 | unsigned long inactive_ratio; |
14797e23 | 1496 | unsigned long inactive; |
9b272977 | 1497 | unsigned long active; |
c772be93 | 1498 | unsigned long gb; |
14797e23 | 1499 | |
4d7dcca2 HD |
1500 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1501 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1502 | |
c772be93 KM |
1503 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1504 | if (gb) | |
1505 | inactive_ratio = int_sqrt(10 * gb); | |
1506 | else | |
1507 | inactive_ratio = 1; | |
1508 | ||
9b272977 | 1509 | return inactive * inactive_ratio < active; |
14797e23 KM |
1510 | } |
1511 | ||
6d61ef40 BS |
1512 | #define mem_cgroup_from_res_counter(counter, member) \ |
1513 | container_of(counter, struct mem_cgroup, member) | |
1514 | ||
19942822 | 1515 | /** |
9d11ea9f | 1516 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1517 | * @memcg: the memory cgroup |
19942822 | 1518 | * |
9d11ea9f | 1519 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1520 | * pages. |
19942822 | 1521 | */ |
c0ff4b85 | 1522 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1523 | { |
9d11ea9f JW |
1524 | unsigned long long margin; |
1525 | ||
c0ff4b85 | 1526 | margin = res_counter_margin(&memcg->res); |
9d11ea9f | 1527 | if (do_swap_account) |
c0ff4b85 | 1528 | margin = min(margin, res_counter_margin(&memcg->memsw)); |
7ec99d62 | 1529 | return margin >> PAGE_SHIFT; |
19942822 JW |
1530 | } |
1531 | ||
1f4c025b | 1532 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 | 1533 | { |
a7885eb8 | 1534 | /* root ? */ |
63876986 | 1535 | if (!css_parent(&memcg->css)) |
a7885eb8 KM |
1536 | return vm_swappiness; |
1537 | ||
bf1ff263 | 1538 | return memcg->swappiness; |
a7885eb8 KM |
1539 | } |
1540 | ||
619d094b KH |
1541 | /* |
1542 | * memcg->moving_account is used for checking possibility that some thread is | |
1543 | * calling move_account(). When a thread on CPU-A starts moving pages under | |
1544 | * a memcg, other threads should check memcg->moving_account under | |
1545 | * rcu_read_lock(), like this: | |
1546 | * | |
1547 | * CPU-A CPU-B | |
1548 | * rcu_read_lock() | |
1549 | * memcg->moving_account+1 if (memcg->mocing_account) | |
1550 | * take heavy locks. | |
1551 | * synchronize_rcu() update something. | |
1552 | * rcu_read_unlock() | |
1553 | * start move here. | |
1554 | */ | |
4331f7d3 KH |
1555 | |
1556 | /* for quick checking without looking up memcg */ | |
1557 | atomic_t memcg_moving __read_mostly; | |
1558 | ||
c0ff4b85 | 1559 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a | 1560 | { |
4331f7d3 | 1561 | atomic_inc(&memcg_moving); |
619d094b | 1562 | atomic_inc(&memcg->moving_account); |
32047e2a KH |
1563 | synchronize_rcu(); |
1564 | } | |
1565 | ||
c0ff4b85 | 1566 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a | 1567 | { |
619d094b KH |
1568 | /* |
1569 | * Now, mem_cgroup_clear_mc() may call this function with NULL. | |
1570 | * We check NULL in callee rather than caller. | |
1571 | */ | |
4331f7d3 KH |
1572 | if (memcg) { |
1573 | atomic_dec(&memcg_moving); | |
619d094b | 1574 | atomic_dec(&memcg->moving_account); |
4331f7d3 | 1575 | } |
32047e2a | 1576 | } |
619d094b | 1577 | |
32047e2a KH |
1578 | /* |
1579 | * 2 routines for checking "mem" is under move_account() or not. | |
1580 | * | |
13fd1dd9 AM |
1581 | * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This |
1582 | * is used for avoiding races in accounting. If true, | |
32047e2a KH |
1583 | * pc->mem_cgroup may be overwritten. |
1584 | * | |
1585 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or | |
1586 | * under hierarchy of moving cgroups. This is for | |
1587 | * waiting at hith-memory prressure caused by "move". | |
1588 | */ | |
1589 | ||
13fd1dd9 | 1590 | static bool mem_cgroup_stolen(struct mem_cgroup *memcg) |
32047e2a KH |
1591 | { |
1592 | VM_BUG_ON(!rcu_read_lock_held()); | |
619d094b | 1593 | return atomic_read(&memcg->moving_account) > 0; |
32047e2a | 1594 | } |
4b534334 | 1595 | |
c0ff4b85 | 1596 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1597 | { |
2bd9bb20 KH |
1598 | struct mem_cgroup *from; |
1599 | struct mem_cgroup *to; | |
4b534334 | 1600 | bool ret = false; |
2bd9bb20 KH |
1601 | /* |
1602 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1603 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1604 | */ | |
1605 | spin_lock(&mc.lock); | |
1606 | from = mc.from; | |
1607 | to = mc.to; | |
1608 | if (!from) | |
1609 | goto unlock; | |
3e92041d | 1610 | |
c0ff4b85 R |
1611 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1612 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1613 | unlock: |
1614 | spin_unlock(&mc.lock); | |
4b534334 KH |
1615 | return ret; |
1616 | } | |
1617 | ||
c0ff4b85 | 1618 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1619 | { |
1620 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1621 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1622 | DEFINE_WAIT(wait); |
1623 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1624 | /* moving charge context might have finished. */ | |
1625 | if (mc.moving_task) | |
1626 | schedule(); | |
1627 | finish_wait(&mc.waitq, &wait); | |
1628 | return true; | |
1629 | } | |
1630 | } | |
1631 | return false; | |
1632 | } | |
1633 | ||
312734c0 KH |
1634 | /* |
1635 | * Take this lock when | |
1636 | * - a code tries to modify page's memcg while it's USED. | |
1637 | * - a code tries to modify page state accounting in a memcg. | |
13fd1dd9 | 1638 | * see mem_cgroup_stolen(), too. |
312734c0 KH |
1639 | */ |
1640 | static void move_lock_mem_cgroup(struct mem_cgroup *memcg, | |
1641 | unsigned long *flags) | |
1642 | { | |
1643 | spin_lock_irqsave(&memcg->move_lock, *flags); | |
1644 | } | |
1645 | ||
1646 | static void move_unlock_mem_cgroup(struct mem_cgroup *memcg, | |
1647 | unsigned long *flags) | |
1648 | { | |
1649 | spin_unlock_irqrestore(&memcg->move_lock, *flags); | |
1650 | } | |
1651 | ||
58cf188e | 1652 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1653 | /** |
58cf188e | 1654 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1655 | * @memcg: The memory cgroup that went over limit |
1656 | * @p: Task that is going to be killed | |
1657 | * | |
1658 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1659 | * enabled | |
1660 | */ | |
1661 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1662 | { | |
1663 | struct cgroup *task_cgrp; | |
1664 | struct cgroup *mem_cgrp; | |
1665 | /* | |
1666 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1667 | * on the assumption that OOM is serialized for memory controller. | |
1668 | * If this assumption is broken, revisit this code. | |
1669 | */ | |
1670 | static char memcg_name[PATH_MAX]; | |
1671 | int ret; | |
58cf188e SZ |
1672 | struct mem_cgroup *iter; |
1673 | unsigned int i; | |
e222432b | 1674 | |
58cf188e | 1675 | if (!p) |
e222432b BS |
1676 | return; |
1677 | ||
e222432b BS |
1678 | rcu_read_lock(); |
1679 | ||
1680 | mem_cgrp = memcg->css.cgroup; | |
1681 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1682 | ||
1683 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1684 | if (ret < 0) { | |
1685 | /* | |
1686 | * Unfortunately, we are unable to convert to a useful name | |
1687 | * But we'll still print out the usage information | |
1688 | */ | |
1689 | rcu_read_unlock(); | |
1690 | goto done; | |
1691 | } | |
1692 | rcu_read_unlock(); | |
1693 | ||
d045197f | 1694 | pr_info("Task in %s killed", memcg_name); |
e222432b BS |
1695 | |
1696 | rcu_read_lock(); | |
1697 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1698 | if (ret < 0) { | |
1699 | rcu_read_unlock(); | |
1700 | goto done; | |
1701 | } | |
1702 | rcu_read_unlock(); | |
1703 | ||
1704 | /* | |
1705 | * Continues from above, so we don't need an KERN_ level | |
1706 | */ | |
d045197f | 1707 | pr_cont(" as a result of limit of %s\n", memcg_name); |
e222432b BS |
1708 | done: |
1709 | ||
d045197f | 1710 | pr_info("memory: usage %llukB, limit %llukB, failcnt %llu\n", |
e222432b BS |
1711 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, |
1712 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1713 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
d045197f | 1714 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %llu\n", |
e222432b BS |
1715 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, |
1716 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1717 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
d045197f | 1718 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %llu\n", |
510fc4e1 GC |
1719 | res_counter_read_u64(&memcg->kmem, RES_USAGE) >> 10, |
1720 | res_counter_read_u64(&memcg->kmem, RES_LIMIT) >> 10, | |
1721 | res_counter_read_u64(&memcg->kmem, RES_FAILCNT)); | |
58cf188e SZ |
1722 | |
1723 | for_each_mem_cgroup_tree(iter, memcg) { | |
1724 | pr_info("Memory cgroup stats"); | |
1725 | ||
1726 | rcu_read_lock(); | |
1727 | ret = cgroup_path(iter->css.cgroup, memcg_name, PATH_MAX); | |
1728 | if (!ret) | |
1729 | pr_cont(" for %s", memcg_name); | |
1730 | rcu_read_unlock(); | |
1731 | pr_cont(":"); | |
1732 | ||
1733 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1734 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1735 | continue; | |
1736 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], | |
1737 | K(mem_cgroup_read_stat(iter, i))); | |
1738 | } | |
1739 | ||
1740 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1741 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1742 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1743 | ||
1744 | pr_cont("\n"); | |
1745 | } | |
e222432b BS |
1746 | } |
1747 | ||
81d39c20 KH |
1748 | /* |
1749 | * This function returns the number of memcg under hierarchy tree. Returns | |
1750 | * 1(self count) if no children. | |
1751 | */ | |
c0ff4b85 | 1752 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1753 | { |
1754 | int num = 0; | |
7d74b06f KH |
1755 | struct mem_cgroup *iter; |
1756 | ||
c0ff4b85 | 1757 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1758 | num++; |
81d39c20 KH |
1759 | return num; |
1760 | } | |
1761 | ||
a63d83f4 DR |
1762 | /* |
1763 | * Return the memory (and swap, if configured) limit for a memcg. | |
1764 | */ | |
9cbb78bb | 1765 | static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 DR |
1766 | { |
1767 | u64 limit; | |
a63d83f4 | 1768 | |
f3e8eb70 | 1769 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
f3e8eb70 | 1770 | |
a63d83f4 | 1771 | /* |
9a5a8f19 | 1772 | * Do not consider swap space if we cannot swap due to swappiness |
a63d83f4 | 1773 | */ |
9a5a8f19 MH |
1774 | if (mem_cgroup_swappiness(memcg)) { |
1775 | u64 memsw; | |
1776 | ||
1777 | limit += total_swap_pages << PAGE_SHIFT; | |
1778 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1779 | ||
1780 | /* | |
1781 | * If memsw is finite and limits the amount of swap space | |
1782 | * available to this memcg, return that limit. | |
1783 | */ | |
1784 | limit = min(limit, memsw); | |
1785 | } | |
1786 | ||
1787 | return limit; | |
a63d83f4 DR |
1788 | } |
1789 | ||
19965460 DR |
1790 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1791 | int order) | |
9cbb78bb DR |
1792 | { |
1793 | struct mem_cgroup *iter; | |
1794 | unsigned long chosen_points = 0; | |
1795 | unsigned long totalpages; | |
1796 | unsigned int points = 0; | |
1797 | struct task_struct *chosen = NULL; | |
1798 | ||
876aafbf | 1799 | /* |
465adcf1 DR |
1800 | * If current has a pending SIGKILL or is exiting, then automatically |
1801 | * select it. The goal is to allow it to allocate so that it may | |
1802 | * quickly exit and free its memory. | |
876aafbf | 1803 | */ |
465adcf1 | 1804 | if (fatal_signal_pending(current) || current->flags & PF_EXITING) { |
876aafbf DR |
1805 | set_thread_flag(TIF_MEMDIE); |
1806 | return; | |
1807 | } | |
1808 | ||
1809 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
9cbb78bb DR |
1810 | totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1; |
1811 | for_each_mem_cgroup_tree(iter, memcg) { | |
72ec7029 | 1812 | struct css_task_iter it; |
9cbb78bb DR |
1813 | struct task_struct *task; |
1814 | ||
72ec7029 TH |
1815 | css_task_iter_start(&iter->css, &it); |
1816 | while ((task = css_task_iter_next(&it))) { | |
9cbb78bb DR |
1817 | switch (oom_scan_process_thread(task, totalpages, NULL, |
1818 | false)) { | |
1819 | case OOM_SCAN_SELECT: | |
1820 | if (chosen) | |
1821 | put_task_struct(chosen); | |
1822 | chosen = task; | |
1823 | chosen_points = ULONG_MAX; | |
1824 | get_task_struct(chosen); | |
1825 | /* fall through */ | |
1826 | case OOM_SCAN_CONTINUE: | |
1827 | continue; | |
1828 | case OOM_SCAN_ABORT: | |
72ec7029 | 1829 | css_task_iter_end(&it); |
9cbb78bb DR |
1830 | mem_cgroup_iter_break(memcg, iter); |
1831 | if (chosen) | |
1832 | put_task_struct(chosen); | |
1833 | return; | |
1834 | case OOM_SCAN_OK: | |
1835 | break; | |
1836 | }; | |
1837 | points = oom_badness(task, memcg, NULL, totalpages); | |
1838 | if (points > chosen_points) { | |
1839 | if (chosen) | |
1840 | put_task_struct(chosen); | |
1841 | chosen = task; | |
1842 | chosen_points = points; | |
1843 | get_task_struct(chosen); | |
1844 | } | |
1845 | } | |
72ec7029 | 1846 | css_task_iter_end(&it); |
9cbb78bb DR |
1847 | } |
1848 | ||
1849 | if (!chosen) | |
1850 | return; | |
1851 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1852 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1853 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1854 | } |
1855 | ||
5660048c JW |
1856 | static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg, |
1857 | gfp_t gfp_mask, | |
1858 | unsigned long flags) | |
1859 | { | |
1860 | unsigned long total = 0; | |
1861 | bool noswap = false; | |
1862 | int loop; | |
1863 | ||
1864 | if (flags & MEM_CGROUP_RECLAIM_NOSWAP) | |
1865 | noswap = true; | |
1866 | if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum) | |
1867 | noswap = true; | |
1868 | ||
1869 | for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) { | |
1870 | if (loop) | |
1871 | drain_all_stock_async(memcg); | |
1872 | total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap); | |
1873 | /* | |
1874 | * Allow limit shrinkers, which are triggered directly | |
1875 | * by userspace, to catch signals and stop reclaim | |
1876 | * after minimal progress, regardless of the margin. | |
1877 | */ | |
1878 | if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK)) | |
1879 | break; | |
1880 | if (mem_cgroup_margin(memcg)) | |
1881 | break; | |
1882 | /* | |
1883 | * If nothing was reclaimed after two attempts, there | |
1884 | * may be no reclaimable pages in this hierarchy. | |
1885 | */ | |
1886 | if (loop && !total) | |
1887 | break; | |
1888 | } | |
1889 | return total; | |
1890 | } | |
1891 | ||
4d0c066d KH |
1892 | /** |
1893 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1894 | * @memcg: the target memcg |
4d0c066d KH |
1895 | * @nid: the node ID to be checked. |
1896 | * @noswap : specify true here if the user wants flle only information. | |
1897 | * | |
1898 | * This function returns whether the specified memcg contains any | |
1899 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1900 | * pages in the node. | |
1901 | */ | |
c0ff4b85 | 1902 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1903 | int nid, bool noswap) |
1904 | { | |
c0ff4b85 | 1905 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1906 | return true; |
1907 | if (noswap || !total_swap_pages) | |
1908 | return false; | |
c0ff4b85 | 1909 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1910 | return true; |
1911 | return false; | |
1912 | ||
1913 | } | |
bb4cc1a8 | 1914 | #if MAX_NUMNODES > 1 |
889976db YH |
1915 | |
1916 | /* | |
1917 | * Always updating the nodemask is not very good - even if we have an empty | |
1918 | * list or the wrong list here, we can start from some node and traverse all | |
1919 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1920 | * | |
1921 | */ | |
c0ff4b85 | 1922 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1923 | { |
1924 | int nid; | |
453a9bf3 KH |
1925 | /* |
1926 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1927 | * pagein/pageout changes since the last update. | |
1928 | */ | |
c0ff4b85 | 1929 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1930 | return; |
c0ff4b85 | 1931 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1932 | return; |
1933 | ||
889976db | 1934 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1935 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1936 | |
31aaea4a | 1937 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1938 | |
c0ff4b85 R |
1939 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1940 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1941 | } |
453a9bf3 | 1942 | |
c0ff4b85 R |
1943 | atomic_set(&memcg->numainfo_events, 0); |
1944 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1945 | } |
1946 | ||
1947 | /* | |
1948 | * Selecting a node where we start reclaim from. Because what we need is just | |
1949 | * reducing usage counter, start from anywhere is O,K. Considering | |
1950 | * memory reclaim from current node, there are pros. and cons. | |
1951 | * | |
1952 | * Freeing memory from current node means freeing memory from a node which | |
1953 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1954 | * hit limits, it will see a contention on a node. But freeing from remote | |
1955 | * node means more costs for memory reclaim because of memory latency. | |
1956 | * | |
1957 | * Now, we use round-robin. Better algorithm is welcomed. | |
1958 | */ | |
c0ff4b85 | 1959 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1960 | { |
1961 | int node; | |
1962 | ||
c0ff4b85 R |
1963 | mem_cgroup_may_update_nodemask(memcg); |
1964 | node = memcg->last_scanned_node; | |
889976db | 1965 | |
c0ff4b85 | 1966 | node = next_node(node, memcg->scan_nodes); |
889976db | 1967 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1968 | node = first_node(memcg->scan_nodes); |
889976db YH |
1969 | /* |
1970 | * We call this when we hit limit, not when pages are added to LRU. | |
1971 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1972 | * memcg is too small and all pages are not on LRU. In that case, | |
1973 | * we use curret node. | |
1974 | */ | |
1975 | if (unlikely(node == MAX_NUMNODES)) | |
1976 | node = numa_node_id(); | |
1977 | ||
c0ff4b85 | 1978 | memcg->last_scanned_node = node; |
889976db YH |
1979 | return node; |
1980 | } | |
1981 | ||
bb4cc1a8 AM |
1982 | /* |
1983 | * Check all nodes whether it contains reclaimable pages or not. | |
1984 | * For quick scan, we make use of scan_nodes. This will allow us to skip | |
1985 | * unused nodes. But scan_nodes is lazily updated and may not cotain | |
1986 | * enough new information. We need to do double check. | |
1987 | */ | |
1988 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) | |
1989 | { | |
1990 | int nid; | |
1991 | ||
1992 | /* | |
1993 | * quick check...making use of scan_node. | |
1994 | * We can skip unused nodes. | |
1995 | */ | |
1996 | if (!nodes_empty(memcg->scan_nodes)) { | |
1997 | for (nid = first_node(memcg->scan_nodes); | |
1998 | nid < MAX_NUMNODES; | |
1999 | nid = next_node(nid, memcg->scan_nodes)) { | |
2000 | ||
2001 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) | |
2002 | return true; | |
2003 | } | |
2004 | } | |
2005 | /* | |
2006 | * Check rest of nodes. | |
2007 | */ | |
2008 | for_each_node_state(nid, N_MEMORY) { | |
2009 | if (node_isset(nid, memcg->scan_nodes)) | |
2010 | continue; | |
2011 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) | |
2012 | return true; | |
2013 | } | |
2014 | return false; | |
2015 | } | |
2016 | ||
889976db | 2017 | #else |
c0ff4b85 | 2018 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
2019 | { |
2020 | return 0; | |
2021 | } | |
4d0c066d | 2022 | |
bb4cc1a8 AM |
2023 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
2024 | { | |
2025 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); | |
2026 | } | |
889976db YH |
2027 | #endif |
2028 | ||
0608f43d AM |
2029 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
2030 | struct zone *zone, | |
2031 | gfp_t gfp_mask, | |
2032 | unsigned long *total_scanned) | |
2033 | { | |
2034 | struct mem_cgroup *victim = NULL; | |
2035 | int total = 0; | |
2036 | int loop = 0; | |
2037 | unsigned long excess; | |
2038 | unsigned long nr_scanned; | |
2039 | struct mem_cgroup_reclaim_cookie reclaim = { | |
2040 | .zone = zone, | |
2041 | .priority = 0, | |
2042 | }; | |
2043 | ||
2044 | excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT; | |
2045 | ||
2046 | while (1) { | |
2047 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
2048 | if (!victim) { | |
2049 | loop++; | |
2050 | if (loop >= 2) { | |
2051 | /* | |
2052 | * If we have not been able to reclaim | |
2053 | * anything, it might because there are | |
2054 | * no reclaimable pages under this hierarchy | |
2055 | */ | |
2056 | if (!total) | |
2057 | break; | |
2058 | /* | |
2059 | * We want to do more targeted reclaim. | |
2060 | * excess >> 2 is not to excessive so as to | |
2061 | * reclaim too much, nor too less that we keep | |
2062 | * coming back to reclaim from this cgroup | |
2063 | */ | |
2064 | if (total >= (excess >> 2) || | |
2065 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
2066 | break; | |
2067 | } | |
2068 | continue; | |
2069 | } | |
2070 | if (!mem_cgroup_reclaimable(victim, false)) | |
2071 | continue; | |
2072 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, | |
2073 | zone, &nr_scanned); | |
2074 | *total_scanned += nr_scanned; | |
2075 | if (!res_counter_soft_limit_excess(&root_memcg->res)) | |
2076 | break; | |
6d61ef40 | 2077 | } |
0608f43d AM |
2078 | mem_cgroup_iter_break(root_memcg, victim); |
2079 | return total; | |
6d61ef40 BS |
2080 | } |
2081 | ||
0056f4e6 JW |
2082 | #ifdef CONFIG_LOCKDEP |
2083 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
2084 | .name = "memcg_oom_lock", | |
2085 | }; | |
2086 | #endif | |
2087 | ||
fb2a6fc5 JW |
2088 | static DEFINE_SPINLOCK(memcg_oom_lock); |
2089 | ||
867578cb KH |
2090 | /* |
2091 | * Check OOM-Killer is already running under our hierarchy. | |
2092 | * If someone is running, return false. | |
2093 | */ | |
fb2a6fc5 | 2094 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 2095 | { |
79dfdacc | 2096 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 2097 | |
fb2a6fc5 JW |
2098 | spin_lock(&memcg_oom_lock); |
2099 | ||
9f3a0d09 | 2100 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 2101 | if (iter->oom_lock) { |
79dfdacc MH |
2102 | /* |
2103 | * this subtree of our hierarchy is already locked | |
2104 | * so we cannot give a lock. | |
2105 | */ | |
79dfdacc | 2106 | failed = iter; |
9f3a0d09 JW |
2107 | mem_cgroup_iter_break(memcg, iter); |
2108 | break; | |
23751be0 JW |
2109 | } else |
2110 | iter->oom_lock = true; | |
7d74b06f | 2111 | } |
867578cb | 2112 | |
fb2a6fc5 JW |
2113 | if (failed) { |
2114 | /* | |
2115 | * OK, we failed to lock the whole subtree so we have | |
2116 | * to clean up what we set up to the failing subtree | |
2117 | */ | |
2118 | for_each_mem_cgroup_tree(iter, memcg) { | |
2119 | if (iter == failed) { | |
2120 | mem_cgroup_iter_break(memcg, iter); | |
2121 | break; | |
2122 | } | |
2123 | iter->oom_lock = false; | |
79dfdacc | 2124 | } |
0056f4e6 JW |
2125 | } else |
2126 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
2127 | |
2128 | spin_unlock(&memcg_oom_lock); | |
2129 | ||
2130 | return !failed; | |
a636b327 | 2131 | } |
0b7f569e | 2132 | |
fb2a6fc5 | 2133 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 2134 | { |
7d74b06f KH |
2135 | struct mem_cgroup *iter; |
2136 | ||
fb2a6fc5 | 2137 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 2138 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 2139 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 2140 | iter->oom_lock = false; |
fb2a6fc5 | 2141 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
2142 | } |
2143 | ||
c0ff4b85 | 2144 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2145 | { |
2146 | struct mem_cgroup *iter; | |
2147 | ||
c0ff4b85 | 2148 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
2149 | atomic_inc(&iter->under_oom); |
2150 | } | |
2151 | ||
c0ff4b85 | 2152 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2153 | { |
2154 | struct mem_cgroup *iter; | |
2155 | ||
867578cb KH |
2156 | /* |
2157 | * When a new child is created while the hierarchy is under oom, | |
2158 | * mem_cgroup_oom_lock() may not be called. We have to use | |
2159 | * atomic_add_unless() here. | |
2160 | */ | |
c0ff4b85 | 2161 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 2162 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
2163 | } |
2164 | ||
867578cb KH |
2165 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
2166 | ||
dc98df5a | 2167 | struct oom_wait_info { |
d79154bb | 2168 | struct mem_cgroup *memcg; |
dc98df5a KH |
2169 | wait_queue_t wait; |
2170 | }; | |
2171 | ||
2172 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
2173 | unsigned mode, int sync, void *arg) | |
2174 | { | |
d79154bb HD |
2175 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
2176 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
2177 | struct oom_wait_info *oom_wait_info; |
2178 | ||
2179 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 2180 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 2181 | |
dc98df5a | 2182 | /* |
d79154bb | 2183 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
dc98df5a KH |
2184 | * Then we can use css_is_ancestor without taking care of RCU. |
2185 | */ | |
c0ff4b85 R |
2186 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
2187 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 2188 | return 0; |
dc98df5a KH |
2189 | return autoremove_wake_function(wait, mode, sync, arg); |
2190 | } | |
2191 | ||
c0ff4b85 | 2192 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 2193 | { |
3812c8c8 | 2194 | atomic_inc(&memcg->oom_wakeups); |
c0ff4b85 R |
2195 | /* for filtering, pass "memcg" as argument. */ |
2196 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
2197 | } |
2198 | ||
c0ff4b85 | 2199 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 2200 | { |
c0ff4b85 R |
2201 | if (memcg && atomic_read(&memcg->under_oom)) |
2202 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
2203 | } |
2204 | ||
3812c8c8 | 2205 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 2206 | { |
3812c8c8 JW |
2207 | if (!current->memcg_oom.may_oom) |
2208 | return; | |
867578cb | 2209 | /* |
49426420 JW |
2210 | * We are in the middle of the charge context here, so we |
2211 | * don't want to block when potentially sitting on a callstack | |
2212 | * that holds all kinds of filesystem and mm locks. | |
2213 | * | |
2214 | * Also, the caller may handle a failed allocation gracefully | |
2215 | * (like optional page cache readahead) and so an OOM killer | |
2216 | * invocation might not even be necessary. | |
2217 | * | |
2218 | * That's why we don't do anything here except remember the | |
2219 | * OOM context and then deal with it at the end of the page | |
2220 | * fault when the stack is unwound, the locks are released, | |
2221 | * and when we know whether the fault was overall successful. | |
867578cb | 2222 | */ |
49426420 JW |
2223 | css_get(&memcg->css); |
2224 | current->memcg_oom.memcg = memcg; | |
2225 | current->memcg_oom.gfp_mask = mask; | |
2226 | current->memcg_oom.order = order; | |
3812c8c8 JW |
2227 | } |
2228 | ||
2229 | /** | |
2230 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 2231 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 2232 | * |
49426420 JW |
2233 | * This has to be called at the end of a page fault if the memcg OOM |
2234 | * handler was enabled. | |
3812c8c8 | 2235 | * |
49426420 | 2236 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
2237 | * sleep on a waitqueue until the userspace task resolves the |
2238 | * situation. Sleeping directly in the charge context with all kinds | |
2239 | * of locks held is not a good idea, instead we remember an OOM state | |
2240 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 2241 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
2242 | * |
2243 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 2244 | * completed, %false otherwise. |
3812c8c8 | 2245 | */ |
49426420 | 2246 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 2247 | { |
49426420 | 2248 | struct mem_cgroup *memcg = current->memcg_oom.memcg; |
3812c8c8 | 2249 | struct oom_wait_info owait; |
49426420 | 2250 | bool locked; |
3812c8c8 JW |
2251 | |
2252 | /* OOM is global, do not handle */ | |
3812c8c8 | 2253 | if (!memcg) |
49426420 | 2254 | return false; |
3812c8c8 | 2255 | |
49426420 JW |
2256 | if (!handle) |
2257 | goto cleanup; | |
3812c8c8 JW |
2258 | |
2259 | owait.memcg = memcg; | |
2260 | owait.wait.flags = 0; | |
2261 | owait.wait.func = memcg_oom_wake_function; | |
2262 | owait.wait.private = current; | |
2263 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 2264 | |
3812c8c8 | 2265 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
2266 | mem_cgroup_mark_under_oom(memcg); |
2267 | ||
2268 | locked = mem_cgroup_oom_trylock(memcg); | |
2269 | ||
2270 | if (locked) | |
2271 | mem_cgroup_oom_notify(memcg); | |
2272 | ||
2273 | if (locked && !memcg->oom_kill_disable) { | |
2274 | mem_cgroup_unmark_under_oom(memcg); | |
2275 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2276 | mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask, | |
2277 | current->memcg_oom.order); | |
2278 | } else { | |
3812c8c8 | 2279 | schedule(); |
49426420 JW |
2280 | mem_cgroup_unmark_under_oom(memcg); |
2281 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2282 | } | |
2283 | ||
2284 | if (locked) { | |
fb2a6fc5 JW |
2285 | mem_cgroup_oom_unlock(memcg); |
2286 | /* | |
2287 | * There is no guarantee that an OOM-lock contender | |
2288 | * sees the wakeups triggered by the OOM kill | |
2289 | * uncharges. Wake any sleepers explicitely. | |
2290 | */ | |
2291 | memcg_oom_recover(memcg); | |
2292 | } | |
49426420 JW |
2293 | cleanup: |
2294 | current->memcg_oom.memcg = NULL; | |
3812c8c8 | 2295 | css_put(&memcg->css); |
867578cb | 2296 | return true; |
0b7f569e KH |
2297 | } |
2298 | ||
d69b042f BS |
2299 | /* |
2300 | * Currently used to update mapped file statistics, but the routine can be | |
2301 | * generalized to update other statistics as well. | |
32047e2a KH |
2302 | * |
2303 | * Notes: Race condition | |
2304 | * | |
2305 | * We usually use page_cgroup_lock() for accessing page_cgroup member but | |
2306 | * it tends to be costly. But considering some conditions, we doesn't need | |
2307 | * to do so _always_. | |
2308 | * | |
2309 | * Considering "charge", lock_page_cgroup() is not required because all | |
2310 | * file-stat operations happen after a page is attached to radix-tree. There | |
2311 | * are no race with "charge". | |
2312 | * | |
2313 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
2314 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
2315 | * if there are race with "uncharge". Statistics itself is properly handled | |
2316 | * by flags. | |
2317 | * | |
2318 | * Considering "move", this is an only case we see a race. To make the race | |
619d094b KH |
2319 | * small, we check mm->moving_account and detect there are possibility of race |
2320 | * If there is, we take a lock. | |
d69b042f | 2321 | */ |
26174efd | 2322 | |
89c06bd5 KH |
2323 | void __mem_cgroup_begin_update_page_stat(struct page *page, |
2324 | bool *locked, unsigned long *flags) | |
2325 | { | |
2326 | struct mem_cgroup *memcg; | |
2327 | struct page_cgroup *pc; | |
2328 | ||
2329 | pc = lookup_page_cgroup(page); | |
2330 | again: | |
2331 | memcg = pc->mem_cgroup; | |
2332 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
2333 | return; | |
2334 | /* | |
2335 | * If this memory cgroup is not under account moving, we don't | |
da92c47d | 2336 | * need to take move_lock_mem_cgroup(). Because we already hold |
89c06bd5 | 2337 | * rcu_read_lock(), any calls to move_account will be delayed until |
13fd1dd9 | 2338 | * rcu_read_unlock() if mem_cgroup_stolen() == true. |
89c06bd5 | 2339 | */ |
13fd1dd9 | 2340 | if (!mem_cgroup_stolen(memcg)) |
89c06bd5 KH |
2341 | return; |
2342 | ||
2343 | move_lock_mem_cgroup(memcg, flags); | |
2344 | if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) { | |
2345 | move_unlock_mem_cgroup(memcg, flags); | |
2346 | goto again; | |
2347 | } | |
2348 | *locked = true; | |
2349 | } | |
2350 | ||
2351 | void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags) | |
2352 | { | |
2353 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
2354 | ||
2355 | /* | |
2356 | * It's guaranteed that pc->mem_cgroup never changes while | |
2357 | * lock is held because a routine modifies pc->mem_cgroup | |
da92c47d | 2358 | * should take move_lock_mem_cgroup(). |
89c06bd5 KH |
2359 | */ |
2360 | move_unlock_mem_cgroup(pc->mem_cgroup, flags); | |
2361 | } | |
2362 | ||
2a7106f2 | 2363 | void mem_cgroup_update_page_stat(struct page *page, |
68b4876d | 2364 | enum mem_cgroup_stat_index idx, int val) |
d69b042f | 2365 | { |
c0ff4b85 | 2366 | struct mem_cgroup *memcg; |
32047e2a | 2367 | struct page_cgroup *pc = lookup_page_cgroup(page); |
dbd4ea78 | 2368 | unsigned long uninitialized_var(flags); |
d69b042f | 2369 | |
cfa44946 | 2370 | if (mem_cgroup_disabled()) |
d69b042f | 2371 | return; |
89c06bd5 | 2372 | |
658b72c5 | 2373 | VM_BUG_ON(!rcu_read_lock_held()); |
c0ff4b85 R |
2374 | memcg = pc->mem_cgroup; |
2375 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
89c06bd5 | 2376 | return; |
26174efd | 2377 | |
c0ff4b85 | 2378 | this_cpu_add(memcg->stat->count[idx], val); |
d69b042f | 2379 | } |
26174efd | 2380 | |
cdec2e42 KH |
2381 | /* |
2382 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2383 | * TODO: maybe necessary to use big numbers in big irons. | |
2384 | */ | |
7ec99d62 | 2385 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2386 | struct memcg_stock_pcp { |
2387 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2388 | unsigned int nr_pages; |
cdec2e42 | 2389 | struct work_struct work; |
26fe6168 | 2390 | unsigned long flags; |
a0db00fc | 2391 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2392 | }; |
2393 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2394 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2395 | |
a0956d54 SS |
2396 | /** |
2397 | * consume_stock: Try to consume stocked charge on this cpu. | |
2398 | * @memcg: memcg to consume from. | |
2399 | * @nr_pages: how many pages to charge. | |
2400 | * | |
2401 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2402 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2403 | * service an allocation will refill the stock. | |
2404 | * | |
2405 | * returns true if successful, false otherwise. | |
cdec2e42 | 2406 | */ |
a0956d54 | 2407 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2408 | { |
2409 | struct memcg_stock_pcp *stock; | |
2410 | bool ret = true; | |
2411 | ||
a0956d54 SS |
2412 | if (nr_pages > CHARGE_BATCH) |
2413 | return false; | |
2414 | ||
cdec2e42 | 2415 | stock = &get_cpu_var(memcg_stock); |
a0956d54 SS |
2416 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) |
2417 | stock->nr_pages -= nr_pages; | |
cdec2e42 KH |
2418 | else /* need to call res_counter_charge */ |
2419 | ret = false; | |
2420 | put_cpu_var(memcg_stock); | |
2421 | return ret; | |
2422 | } | |
2423 | ||
2424 | /* | |
2425 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
2426 | */ | |
2427 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2428 | { | |
2429 | struct mem_cgroup *old = stock->cached; | |
2430 | ||
11c9ea4e JW |
2431 | if (stock->nr_pages) { |
2432 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
2433 | ||
2434 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 2435 | if (do_swap_account) |
11c9ea4e JW |
2436 | res_counter_uncharge(&old->memsw, bytes); |
2437 | stock->nr_pages = 0; | |
cdec2e42 KH |
2438 | } |
2439 | stock->cached = NULL; | |
cdec2e42 KH |
2440 | } |
2441 | ||
2442 | /* | |
2443 | * This must be called under preempt disabled or must be called by | |
2444 | * a thread which is pinned to local cpu. | |
2445 | */ | |
2446 | static void drain_local_stock(struct work_struct *dummy) | |
2447 | { | |
2448 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
2449 | drain_stock(stock); | |
26fe6168 | 2450 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2451 | } |
2452 | ||
e4777496 MH |
2453 | static void __init memcg_stock_init(void) |
2454 | { | |
2455 | int cpu; | |
2456 | ||
2457 | for_each_possible_cpu(cpu) { | |
2458 | struct memcg_stock_pcp *stock = | |
2459 | &per_cpu(memcg_stock, cpu); | |
2460 | INIT_WORK(&stock->work, drain_local_stock); | |
2461 | } | |
2462 | } | |
2463 | ||
cdec2e42 KH |
2464 | /* |
2465 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 2466 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2467 | */ |
c0ff4b85 | 2468 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2469 | { |
2470 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2471 | ||
c0ff4b85 | 2472 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2473 | drain_stock(stock); |
c0ff4b85 | 2474 | stock->cached = memcg; |
cdec2e42 | 2475 | } |
11c9ea4e | 2476 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2477 | put_cpu_var(memcg_stock); |
2478 | } | |
2479 | ||
2480 | /* | |
c0ff4b85 | 2481 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
d38144b7 MH |
2482 | * of the hierarchy under it. sync flag says whether we should block |
2483 | * until the work is done. | |
cdec2e42 | 2484 | */ |
c0ff4b85 | 2485 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
cdec2e42 | 2486 | { |
26fe6168 | 2487 | int cpu, curcpu; |
d38144b7 | 2488 | |
cdec2e42 | 2489 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2490 | get_online_cpus(); |
5af12d0e | 2491 | curcpu = get_cpu(); |
cdec2e42 KH |
2492 | for_each_online_cpu(cpu) { |
2493 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2494 | struct mem_cgroup *memcg; |
26fe6168 | 2495 | |
c0ff4b85 R |
2496 | memcg = stock->cached; |
2497 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2498 | continue; |
c0ff4b85 | 2499 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2500 | continue; |
d1a05b69 MH |
2501 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2502 | if (cpu == curcpu) | |
2503 | drain_local_stock(&stock->work); | |
2504 | else | |
2505 | schedule_work_on(cpu, &stock->work); | |
2506 | } | |
cdec2e42 | 2507 | } |
5af12d0e | 2508 | put_cpu(); |
d38144b7 MH |
2509 | |
2510 | if (!sync) | |
2511 | goto out; | |
2512 | ||
2513 | for_each_online_cpu(cpu) { | |
2514 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2515 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2516 | flush_work(&stock->work); |
2517 | } | |
2518 | out: | |
f894ffa8 | 2519 | put_online_cpus(); |
d38144b7 MH |
2520 | } |
2521 | ||
2522 | /* | |
2523 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2524 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
2525 | * expects some charges will be back to res_counter later but cannot wait for | |
2526 | * it. | |
2527 | */ | |
c0ff4b85 | 2528 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
d38144b7 | 2529 | { |
9f50fad6 MH |
2530 | /* |
2531 | * If someone calls draining, avoid adding more kworker runs. | |
2532 | */ | |
2533 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2534 | return; | |
c0ff4b85 | 2535 | drain_all_stock(root_memcg, false); |
9f50fad6 | 2536 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2537 | } |
2538 | ||
2539 | /* This is a synchronous drain interface. */ | |
c0ff4b85 | 2540 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
cdec2e42 KH |
2541 | { |
2542 | /* called when force_empty is called */ | |
9f50fad6 | 2543 | mutex_lock(&percpu_charge_mutex); |
c0ff4b85 | 2544 | drain_all_stock(root_memcg, true); |
9f50fad6 | 2545 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2546 | } |
2547 | ||
711d3d2c KH |
2548 | /* |
2549 | * This function drains percpu counter value from DEAD cpu and | |
2550 | * move it to local cpu. Note that this function can be preempted. | |
2551 | */ | |
c0ff4b85 | 2552 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2553 | { |
2554 | int i; | |
2555 | ||
c0ff4b85 | 2556 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2557 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2558 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2559 | |
c0ff4b85 R |
2560 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2561 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2562 | } |
e9f8974f | 2563 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2564 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2565 | |
c0ff4b85 R |
2566 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2567 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2568 | } |
c0ff4b85 | 2569 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2570 | } |
2571 | ||
0db0628d | 2572 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
2573 | unsigned long action, |
2574 | void *hcpu) | |
2575 | { | |
2576 | int cpu = (unsigned long)hcpu; | |
2577 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2578 | struct mem_cgroup *iter; |
cdec2e42 | 2579 | |
619d094b | 2580 | if (action == CPU_ONLINE) |
1489ebad | 2581 | return NOTIFY_OK; |
1489ebad | 2582 | |
d833049b | 2583 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2584 | return NOTIFY_OK; |
711d3d2c | 2585 | |
9f3a0d09 | 2586 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2587 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2588 | ||
cdec2e42 KH |
2589 | stock = &per_cpu(memcg_stock, cpu); |
2590 | drain_stock(stock); | |
2591 | return NOTIFY_OK; | |
2592 | } | |
2593 | ||
4b534334 KH |
2594 | |
2595 | /* See __mem_cgroup_try_charge() for details */ | |
2596 | enum { | |
2597 | CHARGE_OK, /* success */ | |
2598 | CHARGE_RETRY, /* need to retry but retry is not bad */ | |
2599 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ | |
2600 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ | |
4b534334 KH |
2601 | }; |
2602 | ||
c0ff4b85 | 2603 | static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
4c9c5359 | 2604 | unsigned int nr_pages, unsigned int min_pages, |
3812c8c8 | 2605 | bool invoke_oom) |
4b534334 | 2606 | { |
7ec99d62 | 2607 | unsigned long csize = nr_pages * PAGE_SIZE; |
4b534334 KH |
2608 | struct mem_cgroup *mem_over_limit; |
2609 | struct res_counter *fail_res; | |
2610 | unsigned long flags = 0; | |
2611 | int ret; | |
2612 | ||
c0ff4b85 | 2613 | ret = res_counter_charge(&memcg->res, csize, &fail_res); |
4b534334 KH |
2614 | |
2615 | if (likely(!ret)) { | |
2616 | if (!do_swap_account) | |
2617 | return CHARGE_OK; | |
c0ff4b85 | 2618 | ret = res_counter_charge(&memcg->memsw, csize, &fail_res); |
4b534334 KH |
2619 | if (likely(!ret)) |
2620 | return CHARGE_OK; | |
2621 | ||
c0ff4b85 | 2622 | res_counter_uncharge(&memcg->res, csize); |
4b534334 KH |
2623 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
2624 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
2625 | } else | |
2626 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
9221edb7 | 2627 | /* |
9221edb7 JW |
2628 | * Never reclaim on behalf of optional batching, retry with a |
2629 | * single page instead. | |
2630 | */ | |
4c9c5359 | 2631 | if (nr_pages > min_pages) |
4b534334 KH |
2632 | return CHARGE_RETRY; |
2633 | ||
2634 | if (!(gfp_mask & __GFP_WAIT)) | |
2635 | return CHARGE_WOULDBLOCK; | |
2636 | ||
4c9c5359 SS |
2637 | if (gfp_mask & __GFP_NORETRY) |
2638 | return CHARGE_NOMEM; | |
2639 | ||
5660048c | 2640 | ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags); |
7ec99d62 | 2641 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
19942822 | 2642 | return CHARGE_RETRY; |
4b534334 | 2643 | /* |
19942822 JW |
2644 | * Even though the limit is exceeded at this point, reclaim |
2645 | * may have been able to free some pages. Retry the charge | |
2646 | * before killing the task. | |
2647 | * | |
2648 | * Only for regular pages, though: huge pages are rather | |
2649 | * unlikely to succeed so close to the limit, and we fall back | |
2650 | * to regular pages anyway in case of failure. | |
4b534334 | 2651 | */ |
4c9c5359 | 2652 | if (nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER) && ret) |
4b534334 KH |
2653 | return CHARGE_RETRY; |
2654 | ||
2655 | /* | |
2656 | * At task move, charge accounts can be doubly counted. So, it's | |
2657 | * better to wait until the end of task_move if something is going on. | |
2658 | */ | |
2659 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2660 | return CHARGE_RETRY; | |
2661 | ||
3812c8c8 JW |
2662 | if (invoke_oom) |
2663 | mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(csize)); | |
4b534334 | 2664 | |
3812c8c8 | 2665 | return CHARGE_NOMEM; |
4b534334 KH |
2666 | } |
2667 | ||
f817ed48 | 2668 | /* |
38c5d72f KH |
2669 | * __mem_cgroup_try_charge() does |
2670 | * 1. detect memcg to be charged against from passed *mm and *ptr, | |
2671 | * 2. update res_counter | |
2672 | * 3. call memory reclaim if necessary. | |
2673 | * | |
2674 | * In some special case, if the task is fatal, fatal_signal_pending() or | |
2675 | * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup | |
2676 | * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon | |
2677 | * as possible without any hazards. 2: all pages should have a valid | |
2678 | * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg | |
2679 | * pointer, that is treated as a charge to root_mem_cgroup. | |
2680 | * | |
2681 | * So __mem_cgroup_try_charge() will return | |
2682 | * 0 ... on success, filling *ptr with a valid memcg pointer. | |
2683 | * -ENOMEM ... charge failure because of resource limits. | |
2684 | * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup. | |
2685 | * | |
2686 | * Unlike the exported interface, an "oom" parameter is added. if oom==true, | |
2687 | * the oom-killer can be invoked. | |
8a9f3ccd | 2688 | */ |
f817ed48 | 2689 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
ec168510 | 2690 | gfp_t gfp_mask, |
7ec99d62 | 2691 | unsigned int nr_pages, |
c0ff4b85 | 2692 | struct mem_cgroup **ptr, |
7ec99d62 | 2693 | bool oom) |
8a9f3ccd | 2694 | { |
7ec99d62 | 2695 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
4b534334 | 2696 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
c0ff4b85 | 2697 | struct mem_cgroup *memcg = NULL; |
4b534334 | 2698 | int ret; |
a636b327 | 2699 | |
867578cb KH |
2700 | /* |
2701 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
2702 | * in system level. So, allow to go ahead dying process in addition to | |
2703 | * MEMDIE process. | |
2704 | */ | |
2705 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
2706 | || fatal_signal_pending(current))) | |
2707 | goto bypass; | |
a636b327 | 2708 | |
49426420 JW |
2709 | if (unlikely(task_in_memcg_oom(current))) |
2710 | goto bypass; | |
2711 | ||
8a9f3ccd | 2712 | /* |
3be91277 HD |
2713 | * We always charge the cgroup the mm_struct belongs to. |
2714 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd | 2715 | * thread group leader migrates. It's possible that mm is not |
24467cac | 2716 | * set, if so charge the root memcg (happens for pagecache usage). |
8a9f3ccd | 2717 | */ |
c0ff4b85 | 2718 | if (!*ptr && !mm) |
38c5d72f | 2719 | *ptr = root_mem_cgroup; |
f75ca962 | 2720 | again: |
c0ff4b85 R |
2721 | if (*ptr) { /* css should be a valid one */ |
2722 | memcg = *ptr; | |
c0ff4b85 | 2723 | if (mem_cgroup_is_root(memcg)) |
f75ca962 | 2724 | goto done; |
a0956d54 | 2725 | if (consume_stock(memcg, nr_pages)) |
f75ca962 | 2726 | goto done; |
c0ff4b85 | 2727 | css_get(&memcg->css); |
4b534334 | 2728 | } else { |
f75ca962 | 2729 | struct task_struct *p; |
54595fe2 | 2730 | |
f75ca962 KH |
2731 | rcu_read_lock(); |
2732 | p = rcu_dereference(mm->owner); | |
f75ca962 | 2733 | /* |
ebb76ce1 | 2734 | * Because we don't have task_lock(), "p" can exit. |
c0ff4b85 | 2735 | * In that case, "memcg" can point to root or p can be NULL with |
ebb76ce1 KH |
2736 | * race with swapoff. Then, we have small risk of mis-accouning. |
2737 | * But such kind of mis-account by race always happens because | |
2738 | * we don't have cgroup_mutex(). It's overkill and we allo that | |
2739 | * small race, here. | |
2740 | * (*) swapoff at el will charge against mm-struct not against | |
2741 | * task-struct. So, mm->owner can be NULL. | |
f75ca962 | 2742 | */ |
c0ff4b85 | 2743 | memcg = mem_cgroup_from_task(p); |
38c5d72f KH |
2744 | if (!memcg) |
2745 | memcg = root_mem_cgroup; | |
2746 | if (mem_cgroup_is_root(memcg)) { | |
f75ca962 KH |
2747 | rcu_read_unlock(); |
2748 | goto done; | |
2749 | } | |
a0956d54 | 2750 | if (consume_stock(memcg, nr_pages)) { |
f75ca962 KH |
2751 | /* |
2752 | * It seems dagerous to access memcg without css_get(). | |
2753 | * But considering how consume_stok works, it's not | |
2754 | * necessary. If consume_stock success, some charges | |
2755 | * from this memcg are cached on this cpu. So, we | |
2756 | * don't need to call css_get()/css_tryget() before | |
2757 | * calling consume_stock(). | |
2758 | */ | |
2759 | rcu_read_unlock(); | |
2760 | goto done; | |
2761 | } | |
2762 | /* after here, we may be blocked. we need to get refcnt */ | |
c0ff4b85 | 2763 | if (!css_tryget(&memcg->css)) { |
f75ca962 KH |
2764 | rcu_read_unlock(); |
2765 | goto again; | |
2766 | } | |
2767 | rcu_read_unlock(); | |
2768 | } | |
8a9f3ccd | 2769 | |
4b534334 | 2770 | do { |
3812c8c8 | 2771 | bool invoke_oom = oom && !nr_oom_retries; |
7a81b88c | 2772 | |
4b534334 | 2773 | /* If killed, bypass charge */ |
f75ca962 | 2774 | if (fatal_signal_pending(current)) { |
c0ff4b85 | 2775 | css_put(&memcg->css); |
4b534334 | 2776 | goto bypass; |
f75ca962 | 2777 | } |
6d61ef40 | 2778 | |
3812c8c8 JW |
2779 | ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, |
2780 | nr_pages, invoke_oom); | |
4b534334 KH |
2781 | switch (ret) { |
2782 | case CHARGE_OK: | |
2783 | break; | |
2784 | case CHARGE_RETRY: /* not in OOM situation but retry */ | |
7ec99d62 | 2785 | batch = nr_pages; |
c0ff4b85 R |
2786 | css_put(&memcg->css); |
2787 | memcg = NULL; | |
f75ca962 | 2788 | goto again; |
4b534334 | 2789 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
c0ff4b85 | 2790 | css_put(&memcg->css); |
4b534334 KH |
2791 | goto nomem; |
2792 | case CHARGE_NOMEM: /* OOM routine works */ | |
3812c8c8 | 2793 | if (!oom || invoke_oom) { |
c0ff4b85 | 2794 | css_put(&memcg->css); |
867578cb | 2795 | goto nomem; |
f75ca962 | 2796 | } |
4b534334 KH |
2797 | nr_oom_retries--; |
2798 | break; | |
66e1707b | 2799 | } |
4b534334 KH |
2800 | } while (ret != CHARGE_OK); |
2801 | ||
7ec99d62 | 2802 | if (batch > nr_pages) |
c0ff4b85 R |
2803 | refill_stock(memcg, batch - nr_pages); |
2804 | css_put(&memcg->css); | |
0c3e73e8 | 2805 | done: |
c0ff4b85 | 2806 | *ptr = memcg; |
7a81b88c KH |
2807 | return 0; |
2808 | nomem: | |
3168ecbe JW |
2809 | if (!(gfp_mask & __GFP_NOFAIL)) { |
2810 | *ptr = NULL; | |
2811 | return -ENOMEM; | |
2812 | } | |
867578cb | 2813 | bypass: |
38c5d72f KH |
2814 | *ptr = root_mem_cgroup; |
2815 | return -EINTR; | |
7a81b88c | 2816 | } |
8a9f3ccd | 2817 | |
a3032a2c DN |
2818 | /* |
2819 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2820 | * This function is for that and do uncharge, put css's refcnt. | |
2821 | * gotten by try_charge(). | |
2822 | */ | |
c0ff4b85 | 2823 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg, |
e7018b8d | 2824 | unsigned int nr_pages) |
a3032a2c | 2825 | { |
c0ff4b85 | 2826 | if (!mem_cgroup_is_root(memcg)) { |
e7018b8d JW |
2827 | unsigned long bytes = nr_pages * PAGE_SIZE; |
2828 | ||
c0ff4b85 | 2829 | res_counter_uncharge(&memcg->res, bytes); |
a3032a2c | 2830 | if (do_swap_account) |
c0ff4b85 | 2831 | res_counter_uncharge(&memcg->memsw, bytes); |
a3032a2c | 2832 | } |
854ffa8d DN |
2833 | } |
2834 | ||
d01dd17f KH |
2835 | /* |
2836 | * Cancel chrages in this cgroup....doesn't propagate to parent cgroup. | |
2837 | * This is useful when moving usage to parent cgroup. | |
2838 | */ | |
2839 | static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg, | |
2840 | unsigned int nr_pages) | |
2841 | { | |
2842 | unsigned long bytes = nr_pages * PAGE_SIZE; | |
2843 | ||
2844 | if (mem_cgroup_is_root(memcg)) | |
2845 | return; | |
2846 | ||
2847 | res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes); | |
2848 | if (do_swap_account) | |
2849 | res_counter_uncharge_until(&memcg->memsw, | |
2850 | memcg->memsw.parent, bytes); | |
2851 | } | |
2852 | ||
a3b2d692 KH |
2853 | /* |
2854 | * A helper function to get mem_cgroup from ID. must be called under | |
e9316080 TH |
2855 | * rcu_read_lock(). The caller is responsible for calling css_tryget if |
2856 | * the mem_cgroup is used for charging. (dropping refcnt from swap can be | |
2857 | * called against removed memcg.) | |
a3b2d692 KH |
2858 | */ |
2859 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2860 | { | |
2861 | struct cgroup_subsys_state *css; | |
2862 | ||
2863 | /* ID 0 is unused ID */ | |
2864 | if (!id) | |
2865 | return NULL; | |
2866 | css = css_lookup(&mem_cgroup_subsys, id); | |
2867 | if (!css) | |
2868 | return NULL; | |
b2145145 | 2869 | return mem_cgroup_from_css(css); |
a3b2d692 KH |
2870 | } |
2871 | ||
e42d9d5d | 2872 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2873 | { |
c0ff4b85 | 2874 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2875 | struct page_cgroup *pc; |
a3b2d692 | 2876 | unsigned short id; |
b5a84319 KH |
2877 | swp_entry_t ent; |
2878 | ||
3c776e64 DN |
2879 | VM_BUG_ON(!PageLocked(page)); |
2880 | ||
3c776e64 | 2881 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2882 | lock_page_cgroup(pc); |
a3b2d692 | 2883 | if (PageCgroupUsed(pc)) { |
c0ff4b85 R |
2884 | memcg = pc->mem_cgroup; |
2885 | if (memcg && !css_tryget(&memcg->css)) | |
2886 | memcg = NULL; | |
e42d9d5d | 2887 | } else if (PageSwapCache(page)) { |
3c776e64 | 2888 | ent.val = page_private(page); |
9fb4b7cc | 2889 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2890 | rcu_read_lock(); |
c0ff4b85 R |
2891 | memcg = mem_cgroup_lookup(id); |
2892 | if (memcg && !css_tryget(&memcg->css)) | |
2893 | memcg = NULL; | |
a3b2d692 | 2894 | rcu_read_unlock(); |
3c776e64 | 2895 | } |
c0bd3f63 | 2896 | unlock_page_cgroup(pc); |
c0ff4b85 | 2897 | return memcg; |
b5a84319 KH |
2898 | } |
2899 | ||
c0ff4b85 | 2900 | static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg, |
5564e88b | 2901 | struct page *page, |
7ec99d62 | 2902 | unsigned int nr_pages, |
9ce70c02 HD |
2903 | enum charge_type ctype, |
2904 | bool lrucare) | |
7a81b88c | 2905 | { |
ce587e65 | 2906 | struct page_cgroup *pc = lookup_page_cgroup(page); |
9ce70c02 | 2907 | struct zone *uninitialized_var(zone); |
fa9add64 | 2908 | struct lruvec *lruvec; |
9ce70c02 | 2909 | bool was_on_lru = false; |
b2402857 | 2910 | bool anon; |
9ce70c02 | 2911 | |
ca3e0214 | 2912 | lock_page_cgroup(pc); |
90deb788 | 2913 | VM_BUG_ON(PageCgroupUsed(pc)); |
ca3e0214 KH |
2914 | /* |
2915 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2916 | * accessed by any other context at this point. | |
2917 | */ | |
9ce70c02 HD |
2918 | |
2919 | /* | |
2920 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2921 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2922 | */ | |
2923 | if (lrucare) { | |
2924 | zone = page_zone(page); | |
2925 | spin_lock_irq(&zone->lru_lock); | |
2926 | if (PageLRU(page)) { | |
fa9add64 | 2927 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 | 2928 | ClearPageLRU(page); |
fa9add64 | 2929 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2930 | was_on_lru = true; |
2931 | } | |
2932 | } | |
2933 | ||
c0ff4b85 | 2934 | pc->mem_cgroup = memcg; |
261fb61a KH |
2935 | /* |
2936 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
2937 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
2938 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
2939 | * before USED bit, we need memory barrier here. | |
2940 | * See mem_cgroup_add_lru_list(), etc. | |
f894ffa8 | 2941 | */ |
08e552c6 | 2942 | smp_wmb(); |
b2402857 | 2943 | SetPageCgroupUsed(pc); |
3be91277 | 2944 | |
9ce70c02 HD |
2945 | if (lrucare) { |
2946 | if (was_on_lru) { | |
fa9add64 | 2947 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 HD |
2948 | VM_BUG_ON(PageLRU(page)); |
2949 | SetPageLRU(page); | |
fa9add64 | 2950 | add_page_to_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2951 | } |
2952 | spin_unlock_irq(&zone->lru_lock); | |
2953 | } | |
2954 | ||
41326c17 | 2955 | if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON) |
b2402857 KH |
2956 | anon = true; |
2957 | else | |
2958 | anon = false; | |
2959 | ||
b070e65c | 2960 | mem_cgroup_charge_statistics(memcg, page, anon, nr_pages); |
52d4b9ac | 2961 | unlock_page_cgroup(pc); |
9ce70c02 | 2962 | |
430e4863 | 2963 | /* |
bb4cc1a8 AM |
2964 | * "charge_statistics" updated event counter. Then, check it. |
2965 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
2966 | * if they exceeds softlimit. | |
430e4863 | 2967 | */ |
c0ff4b85 | 2968 | memcg_check_events(memcg, page); |
7a81b88c | 2969 | } |
66e1707b | 2970 | |
7cf27982 GC |
2971 | static DEFINE_MUTEX(set_limit_mutex); |
2972 | ||
7ae1e1d0 GC |
2973 | #ifdef CONFIG_MEMCG_KMEM |
2974 | static inline bool memcg_can_account_kmem(struct mem_cgroup *memcg) | |
2975 | { | |
2976 | return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg) && | |
2977 | (memcg->kmem_account_flags & KMEM_ACCOUNTED_MASK); | |
2978 | } | |
2979 | ||
1f458cbf GC |
2980 | /* |
2981 | * This is a bit cumbersome, but it is rarely used and avoids a backpointer | |
2982 | * in the memcg_cache_params struct. | |
2983 | */ | |
2984 | static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p) | |
2985 | { | |
2986 | struct kmem_cache *cachep; | |
2987 | ||
2988 | VM_BUG_ON(p->is_root_cache); | |
2989 | cachep = p->root_cache; | |
2990 | return cachep->memcg_params->memcg_caches[memcg_cache_id(p->memcg)]; | |
2991 | } | |
2992 | ||
749c5415 | 2993 | #ifdef CONFIG_SLABINFO |
182446d0 TH |
2994 | static int mem_cgroup_slabinfo_read(struct cgroup_subsys_state *css, |
2995 | struct cftype *cft, struct seq_file *m) | |
749c5415 | 2996 | { |
182446d0 | 2997 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
749c5415 GC |
2998 | struct memcg_cache_params *params; |
2999 | ||
3000 | if (!memcg_can_account_kmem(memcg)) | |
3001 | return -EIO; | |
3002 | ||
3003 | print_slabinfo_header(m); | |
3004 | ||
3005 | mutex_lock(&memcg->slab_caches_mutex); | |
3006 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) | |
3007 | cache_show(memcg_params_to_cache(params), m); | |
3008 | mutex_unlock(&memcg->slab_caches_mutex); | |
3009 | ||
3010 | return 0; | |
3011 | } | |
3012 | #endif | |
3013 | ||
7ae1e1d0 GC |
3014 | static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, u64 size) |
3015 | { | |
3016 | struct res_counter *fail_res; | |
3017 | struct mem_cgroup *_memcg; | |
3018 | int ret = 0; | |
3019 | bool may_oom; | |
3020 | ||
3021 | ret = res_counter_charge(&memcg->kmem, size, &fail_res); | |
3022 | if (ret) | |
3023 | return ret; | |
3024 | ||
3025 | /* | |
3026 | * Conditions under which we can wait for the oom_killer. Those are | |
3027 | * the same conditions tested by the core page allocator | |
3028 | */ | |
3029 | may_oom = (gfp & __GFP_FS) && !(gfp & __GFP_NORETRY); | |
3030 | ||
3031 | _memcg = memcg; | |
3032 | ret = __mem_cgroup_try_charge(NULL, gfp, size >> PAGE_SHIFT, | |
3033 | &_memcg, may_oom); | |
3034 | ||
3035 | if (ret == -EINTR) { | |
3036 | /* | |
3037 | * __mem_cgroup_try_charge() chosed to bypass to root due to | |
3038 | * OOM kill or fatal signal. Since our only options are to | |
3039 | * either fail the allocation or charge it to this cgroup, do | |
3040 | * it as a temporary condition. But we can't fail. From a | |
3041 | * kmem/slab perspective, the cache has already been selected, | |
3042 | * by mem_cgroup_kmem_get_cache(), so it is too late to change | |
3043 | * our minds. | |
3044 | * | |
3045 | * This condition will only trigger if the task entered | |
3046 | * memcg_charge_kmem in a sane state, but was OOM-killed during | |
3047 | * __mem_cgroup_try_charge() above. Tasks that were already | |
3048 | * dying when the allocation triggers should have been already | |
3049 | * directed to the root cgroup in memcontrol.h | |
3050 | */ | |
3051 | res_counter_charge_nofail(&memcg->res, size, &fail_res); | |
3052 | if (do_swap_account) | |
3053 | res_counter_charge_nofail(&memcg->memsw, size, | |
3054 | &fail_res); | |
3055 | ret = 0; | |
3056 | } else if (ret) | |
3057 | res_counter_uncharge(&memcg->kmem, size); | |
3058 | ||
3059 | return ret; | |
3060 | } | |
3061 | ||
3062 | static void memcg_uncharge_kmem(struct mem_cgroup *memcg, u64 size) | |
3063 | { | |
7ae1e1d0 GC |
3064 | res_counter_uncharge(&memcg->res, size); |
3065 | if (do_swap_account) | |
3066 | res_counter_uncharge(&memcg->memsw, size); | |
7de37682 GC |
3067 | |
3068 | /* Not down to 0 */ | |
3069 | if (res_counter_uncharge(&memcg->kmem, size)) | |
3070 | return; | |
3071 | ||
10d5ebf4 LZ |
3072 | /* |
3073 | * Releases a reference taken in kmem_cgroup_css_offline in case | |
3074 | * this last uncharge is racing with the offlining code or it is | |
3075 | * outliving the memcg existence. | |
3076 | * | |
3077 | * The memory barrier imposed by test&clear is paired with the | |
3078 | * explicit one in memcg_kmem_mark_dead(). | |
3079 | */ | |
7de37682 | 3080 | if (memcg_kmem_test_and_clear_dead(memcg)) |
10d5ebf4 | 3081 | css_put(&memcg->css); |
7ae1e1d0 GC |
3082 | } |
3083 | ||
2633d7a0 GC |
3084 | void memcg_cache_list_add(struct mem_cgroup *memcg, struct kmem_cache *cachep) |
3085 | { | |
3086 | if (!memcg) | |
3087 | return; | |
3088 | ||
3089 | mutex_lock(&memcg->slab_caches_mutex); | |
3090 | list_add(&cachep->memcg_params->list, &memcg->memcg_slab_caches); | |
3091 | mutex_unlock(&memcg->slab_caches_mutex); | |
3092 | } | |
3093 | ||
3094 | /* | |
3095 | * helper for acessing a memcg's index. It will be used as an index in the | |
3096 | * child cache array in kmem_cache, and also to derive its name. This function | |
3097 | * will return -1 when this is not a kmem-limited memcg. | |
3098 | */ | |
3099 | int memcg_cache_id(struct mem_cgroup *memcg) | |
3100 | { | |
3101 | return memcg ? memcg->kmemcg_id : -1; | |
3102 | } | |
3103 | ||
55007d84 GC |
3104 | /* |
3105 | * This ends up being protected by the set_limit mutex, during normal | |
3106 | * operation, because that is its main call site. | |
3107 | * | |
3108 | * But when we create a new cache, we can call this as well if its parent | |
3109 | * is kmem-limited. That will have to hold set_limit_mutex as well. | |
3110 | */ | |
3111 | int memcg_update_cache_sizes(struct mem_cgroup *memcg) | |
3112 | { | |
3113 | int num, ret; | |
3114 | ||
3115 | num = ida_simple_get(&kmem_limited_groups, | |
3116 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); | |
3117 | if (num < 0) | |
3118 | return num; | |
3119 | /* | |
3120 | * After this point, kmem_accounted (that we test atomically in | |
3121 | * the beginning of this conditional), is no longer 0. This | |
3122 | * guarantees only one process will set the following boolean | |
3123 | * to true. We don't need test_and_set because we're protected | |
3124 | * by the set_limit_mutex anyway. | |
3125 | */ | |
3126 | memcg_kmem_set_activated(memcg); | |
3127 | ||
3128 | ret = memcg_update_all_caches(num+1); | |
3129 | if (ret) { | |
3130 | ida_simple_remove(&kmem_limited_groups, num); | |
3131 | memcg_kmem_clear_activated(memcg); | |
3132 | return ret; | |
3133 | } | |
3134 | ||
3135 | memcg->kmemcg_id = num; | |
3136 | INIT_LIST_HEAD(&memcg->memcg_slab_caches); | |
3137 | mutex_init(&memcg->slab_caches_mutex); | |
3138 | return 0; | |
3139 | } | |
3140 | ||
3141 | static size_t memcg_caches_array_size(int num_groups) | |
3142 | { | |
3143 | ssize_t size; | |
3144 | if (num_groups <= 0) | |
3145 | return 0; | |
3146 | ||
3147 | size = 2 * num_groups; | |
3148 | if (size < MEMCG_CACHES_MIN_SIZE) | |
3149 | size = MEMCG_CACHES_MIN_SIZE; | |
3150 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
3151 | size = MEMCG_CACHES_MAX_SIZE; | |
3152 | ||
3153 | return size; | |
3154 | } | |
3155 | ||
3156 | /* | |
3157 | * We should update the current array size iff all caches updates succeed. This | |
3158 | * can only be done from the slab side. The slab mutex needs to be held when | |
3159 | * calling this. | |
3160 | */ | |
3161 | void memcg_update_array_size(int num) | |
3162 | { | |
3163 | if (num > memcg_limited_groups_array_size) | |
3164 | memcg_limited_groups_array_size = memcg_caches_array_size(num); | |
3165 | } | |
3166 | ||
15cf17d2 KK |
3167 | static void kmem_cache_destroy_work_func(struct work_struct *w); |
3168 | ||
55007d84 GC |
3169 | int memcg_update_cache_size(struct kmem_cache *s, int num_groups) |
3170 | { | |
3171 | struct memcg_cache_params *cur_params = s->memcg_params; | |
3172 | ||
3173 | VM_BUG_ON(s->memcg_params && !s->memcg_params->is_root_cache); | |
3174 | ||
3175 | if (num_groups > memcg_limited_groups_array_size) { | |
3176 | int i; | |
3177 | ssize_t size = memcg_caches_array_size(num_groups); | |
3178 | ||
3179 | size *= sizeof(void *); | |
90c7a79c | 3180 | size += offsetof(struct memcg_cache_params, memcg_caches); |
55007d84 GC |
3181 | |
3182 | s->memcg_params = kzalloc(size, GFP_KERNEL); | |
3183 | if (!s->memcg_params) { | |
3184 | s->memcg_params = cur_params; | |
3185 | return -ENOMEM; | |
3186 | } | |
3187 | ||
3188 | s->memcg_params->is_root_cache = true; | |
3189 | ||
3190 | /* | |
3191 | * There is the chance it will be bigger than | |
3192 | * memcg_limited_groups_array_size, if we failed an allocation | |
3193 | * in a cache, in which case all caches updated before it, will | |
3194 | * have a bigger array. | |
3195 | * | |
3196 | * But if that is the case, the data after | |
3197 | * memcg_limited_groups_array_size is certainly unused | |
3198 | */ | |
3199 | for (i = 0; i < memcg_limited_groups_array_size; i++) { | |
3200 | if (!cur_params->memcg_caches[i]) | |
3201 | continue; | |
3202 | s->memcg_params->memcg_caches[i] = | |
3203 | cur_params->memcg_caches[i]; | |
3204 | } | |
3205 | ||
3206 | /* | |
3207 | * Ideally, we would wait until all caches succeed, and only | |
3208 | * then free the old one. But this is not worth the extra | |
3209 | * pointer per-cache we'd have to have for this. | |
3210 | * | |
3211 | * It is not a big deal if some caches are left with a size | |
3212 | * bigger than the others. And all updates will reset this | |
3213 | * anyway. | |
3214 | */ | |
3215 | kfree(cur_params); | |
3216 | } | |
3217 | return 0; | |
3218 | } | |
3219 | ||
943a451a GC |
3220 | int memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s, |
3221 | struct kmem_cache *root_cache) | |
2633d7a0 | 3222 | { |
90c7a79c | 3223 | size_t size; |
2633d7a0 GC |
3224 | |
3225 | if (!memcg_kmem_enabled()) | |
3226 | return 0; | |
3227 | ||
90c7a79c AV |
3228 | if (!memcg) { |
3229 | size = offsetof(struct memcg_cache_params, memcg_caches); | |
55007d84 | 3230 | size += memcg_limited_groups_array_size * sizeof(void *); |
90c7a79c AV |
3231 | } else |
3232 | size = sizeof(struct memcg_cache_params); | |
55007d84 | 3233 | |
2633d7a0 GC |
3234 | s->memcg_params = kzalloc(size, GFP_KERNEL); |
3235 | if (!s->memcg_params) | |
3236 | return -ENOMEM; | |
3237 | ||
943a451a | 3238 | if (memcg) { |
2633d7a0 | 3239 | s->memcg_params->memcg = memcg; |
943a451a | 3240 | s->memcg_params->root_cache = root_cache; |
3e6b11df AV |
3241 | INIT_WORK(&s->memcg_params->destroy, |
3242 | kmem_cache_destroy_work_func); | |
4ba902b5 GC |
3243 | } else |
3244 | s->memcg_params->is_root_cache = true; | |
3245 | ||
2633d7a0 GC |
3246 | return 0; |
3247 | } | |
3248 | ||
3249 | void memcg_release_cache(struct kmem_cache *s) | |
3250 | { | |
d7f25f8a GC |
3251 | struct kmem_cache *root; |
3252 | struct mem_cgroup *memcg; | |
3253 | int id; | |
3254 | ||
3255 | /* | |
3256 | * This happens, for instance, when a root cache goes away before we | |
3257 | * add any memcg. | |
3258 | */ | |
3259 | if (!s->memcg_params) | |
3260 | return; | |
3261 | ||
3262 | if (s->memcg_params->is_root_cache) | |
3263 | goto out; | |
3264 | ||
3265 | memcg = s->memcg_params->memcg; | |
3266 | id = memcg_cache_id(memcg); | |
3267 | ||
3268 | root = s->memcg_params->root_cache; | |
3269 | root->memcg_params->memcg_caches[id] = NULL; | |
d7f25f8a GC |
3270 | |
3271 | mutex_lock(&memcg->slab_caches_mutex); | |
3272 | list_del(&s->memcg_params->list); | |
3273 | mutex_unlock(&memcg->slab_caches_mutex); | |
3274 | ||
20f05310 | 3275 | css_put(&memcg->css); |
d7f25f8a | 3276 | out: |
2633d7a0 GC |
3277 | kfree(s->memcg_params); |
3278 | } | |
3279 | ||
0e9d92f2 GC |
3280 | /* |
3281 | * During the creation a new cache, we need to disable our accounting mechanism | |
3282 | * altogether. This is true even if we are not creating, but rather just | |
3283 | * enqueing new caches to be created. | |
3284 | * | |
3285 | * This is because that process will trigger allocations; some visible, like | |
3286 | * explicit kmallocs to auxiliary data structures, name strings and internal | |
3287 | * cache structures; some well concealed, like INIT_WORK() that can allocate | |
3288 | * objects during debug. | |
3289 | * | |
3290 | * If any allocation happens during memcg_kmem_get_cache, we will recurse back | |
3291 | * to it. This may not be a bounded recursion: since the first cache creation | |
3292 | * failed to complete (waiting on the allocation), we'll just try to create the | |
3293 | * cache again, failing at the same point. | |
3294 | * | |
3295 | * memcg_kmem_get_cache is prepared to abort after seeing a positive count of | |
3296 | * memcg_kmem_skip_account. So we enclose anything that might allocate memory | |
3297 | * inside the following two functions. | |
3298 | */ | |
3299 | static inline void memcg_stop_kmem_account(void) | |
3300 | { | |
3301 | VM_BUG_ON(!current->mm); | |
3302 | current->memcg_kmem_skip_account++; | |
3303 | } | |
3304 | ||
3305 | static inline void memcg_resume_kmem_account(void) | |
3306 | { | |
3307 | VM_BUG_ON(!current->mm); | |
3308 | current->memcg_kmem_skip_account--; | |
3309 | } | |
3310 | ||
1f458cbf GC |
3311 | static void kmem_cache_destroy_work_func(struct work_struct *w) |
3312 | { | |
3313 | struct kmem_cache *cachep; | |
3314 | struct memcg_cache_params *p; | |
3315 | ||
3316 | p = container_of(w, struct memcg_cache_params, destroy); | |
3317 | ||
3318 | cachep = memcg_params_to_cache(p); | |
3319 | ||
22933152 GC |
3320 | /* |
3321 | * If we get down to 0 after shrink, we could delete right away. | |
3322 | * However, memcg_release_pages() already puts us back in the workqueue | |
3323 | * in that case. If we proceed deleting, we'll get a dangling | |
3324 | * reference, and removing the object from the workqueue in that case | |
3325 | * is unnecessary complication. We are not a fast path. | |
3326 | * | |
3327 | * Note that this case is fundamentally different from racing with | |
3328 | * shrink_slab(): if memcg_cgroup_destroy_cache() is called in | |
3329 | * kmem_cache_shrink, not only we would be reinserting a dead cache | |
3330 | * into the queue, but doing so from inside the worker racing to | |
3331 | * destroy it. | |
3332 | * | |
3333 | * So if we aren't down to zero, we'll just schedule a worker and try | |
3334 | * again | |
3335 | */ | |
3336 | if (atomic_read(&cachep->memcg_params->nr_pages) != 0) { | |
3337 | kmem_cache_shrink(cachep); | |
3338 | if (atomic_read(&cachep->memcg_params->nr_pages) == 0) | |
3339 | return; | |
3340 | } else | |
1f458cbf GC |
3341 | kmem_cache_destroy(cachep); |
3342 | } | |
3343 | ||
3344 | void mem_cgroup_destroy_cache(struct kmem_cache *cachep) | |
3345 | { | |
3346 | if (!cachep->memcg_params->dead) | |
3347 | return; | |
3348 | ||
22933152 GC |
3349 | /* |
3350 | * There are many ways in which we can get here. | |
3351 | * | |
3352 | * We can get to a memory-pressure situation while the delayed work is | |
3353 | * still pending to run. The vmscan shrinkers can then release all | |
3354 | * cache memory and get us to destruction. If this is the case, we'll | |
3355 | * be executed twice, which is a bug (the second time will execute over | |
3356 | * bogus data). In this case, cancelling the work should be fine. | |
3357 | * | |
3358 | * But we can also get here from the worker itself, if | |
3359 | * kmem_cache_shrink is enough to shake all the remaining objects and | |
3360 | * get the page count to 0. In this case, we'll deadlock if we try to | |
3361 | * cancel the work (the worker runs with an internal lock held, which | |
3362 | * is the same lock we would hold for cancel_work_sync().) | |
3363 | * | |
3364 | * Since we can't possibly know who got us here, just refrain from | |
3365 | * running if there is already work pending | |
3366 | */ | |
3367 | if (work_pending(&cachep->memcg_params->destroy)) | |
3368 | return; | |
1f458cbf GC |
3369 | /* |
3370 | * We have to defer the actual destroying to a workqueue, because | |
3371 | * we might currently be in a context that cannot sleep. | |
3372 | */ | |
3373 | schedule_work(&cachep->memcg_params->destroy); | |
3374 | } | |
3375 | ||
d9c10ddd MH |
3376 | /* |
3377 | * This lock protects updaters, not readers. We want readers to be as fast as | |
3378 | * they can, and they will either see NULL or a valid cache value. Our model | |
3379 | * allow them to see NULL, in which case the root memcg will be selected. | |
3380 | * | |
3381 | * We need this lock because multiple allocations to the same cache from a non | |
3382 | * will span more than one worker. Only one of them can create the cache. | |
3383 | */ | |
3384 | static DEFINE_MUTEX(memcg_cache_mutex); | |
d7f25f8a | 3385 | |
d9c10ddd MH |
3386 | /* |
3387 | * Called with memcg_cache_mutex held | |
3388 | */ | |
d7f25f8a GC |
3389 | static struct kmem_cache *kmem_cache_dup(struct mem_cgroup *memcg, |
3390 | struct kmem_cache *s) | |
3391 | { | |
d7f25f8a | 3392 | struct kmem_cache *new; |
d9c10ddd | 3393 | static char *tmp_name = NULL; |
d7f25f8a | 3394 | |
d9c10ddd MH |
3395 | lockdep_assert_held(&memcg_cache_mutex); |
3396 | ||
3397 | /* | |
3398 | * kmem_cache_create_memcg duplicates the given name and | |
3399 | * cgroup_name for this name requires RCU context. | |
3400 | * This static temporary buffer is used to prevent from | |
3401 | * pointless shortliving allocation. | |
3402 | */ | |
3403 | if (!tmp_name) { | |
3404 | tmp_name = kmalloc(PATH_MAX, GFP_KERNEL); | |
3405 | if (!tmp_name) | |
3406 | return NULL; | |
3407 | } | |
3408 | ||
3409 | rcu_read_lock(); | |
3410 | snprintf(tmp_name, PATH_MAX, "%s(%d:%s)", s->name, | |
3411 | memcg_cache_id(memcg), cgroup_name(memcg->css.cgroup)); | |
3412 | rcu_read_unlock(); | |
d7f25f8a | 3413 | |
d9c10ddd | 3414 | new = kmem_cache_create_memcg(memcg, tmp_name, s->object_size, s->align, |
943a451a | 3415 | (s->flags & ~SLAB_PANIC), s->ctor, s); |
d7f25f8a | 3416 | |
d79923fa GC |
3417 | if (new) |
3418 | new->allocflags |= __GFP_KMEMCG; | |
3419 | ||
d7f25f8a GC |
3420 | return new; |
3421 | } | |
3422 | ||
d7f25f8a GC |
3423 | static struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg, |
3424 | struct kmem_cache *cachep) | |
3425 | { | |
3426 | struct kmem_cache *new_cachep; | |
3427 | int idx; | |
3428 | ||
3429 | BUG_ON(!memcg_can_account_kmem(memcg)); | |
3430 | ||
3431 | idx = memcg_cache_id(memcg); | |
3432 | ||
3433 | mutex_lock(&memcg_cache_mutex); | |
3434 | new_cachep = cachep->memcg_params->memcg_caches[idx]; | |
20f05310 LZ |
3435 | if (new_cachep) { |
3436 | css_put(&memcg->css); | |
d7f25f8a | 3437 | goto out; |
20f05310 | 3438 | } |
d7f25f8a GC |
3439 | |
3440 | new_cachep = kmem_cache_dup(memcg, cachep); | |
d7f25f8a GC |
3441 | if (new_cachep == NULL) { |
3442 | new_cachep = cachep; | |
20f05310 | 3443 | css_put(&memcg->css); |
d7f25f8a GC |
3444 | goto out; |
3445 | } | |
3446 | ||
1f458cbf | 3447 | atomic_set(&new_cachep->memcg_params->nr_pages , 0); |
d7f25f8a GC |
3448 | |
3449 | cachep->memcg_params->memcg_caches[idx] = new_cachep; | |
3450 | /* | |
3451 | * the readers won't lock, make sure everybody sees the updated value, | |
3452 | * so they won't put stuff in the queue again for no reason | |
3453 | */ | |
3454 | wmb(); | |
3455 | out: | |
3456 | mutex_unlock(&memcg_cache_mutex); | |
3457 | return new_cachep; | |
3458 | } | |
3459 | ||
7cf27982 GC |
3460 | void kmem_cache_destroy_memcg_children(struct kmem_cache *s) |
3461 | { | |
3462 | struct kmem_cache *c; | |
3463 | int i; | |
3464 | ||
3465 | if (!s->memcg_params) | |
3466 | return; | |
3467 | if (!s->memcg_params->is_root_cache) | |
3468 | return; | |
3469 | ||
3470 | /* | |
3471 | * If the cache is being destroyed, we trust that there is no one else | |
3472 | * requesting objects from it. Even if there are, the sanity checks in | |
3473 | * kmem_cache_destroy should caught this ill-case. | |
3474 | * | |
3475 | * Still, we don't want anyone else freeing memcg_caches under our | |
3476 | * noses, which can happen if a new memcg comes to life. As usual, | |
3477 | * we'll take the set_limit_mutex to protect ourselves against this. | |
3478 | */ | |
3479 | mutex_lock(&set_limit_mutex); | |
3480 | for (i = 0; i < memcg_limited_groups_array_size; i++) { | |
3481 | c = s->memcg_params->memcg_caches[i]; | |
3482 | if (!c) | |
3483 | continue; | |
3484 | ||
3485 | /* | |
3486 | * We will now manually delete the caches, so to avoid races | |
3487 | * we need to cancel all pending destruction workers and | |
3488 | * proceed with destruction ourselves. | |
3489 | * | |
3490 | * kmem_cache_destroy() will call kmem_cache_shrink internally, | |
3491 | * and that could spawn the workers again: it is likely that | |
3492 | * the cache still have active pages until this very moment. | |
3493 | * This would lead us back to mem_cgroup_destroy_cache. | |
3494 | * | |
3495 | * But that will not execute at all if the "dead" flag is not | |
3496 | * set, so flip it down to guarantee we are in control. | |
3497 | */ | |
3498 | c->memcg_params->dead = false; | |
22933152 | 3499 | cancel_work_sync(&c->memcg_params->destroy); |
7cf27982 GC |
3500 | kmem_cache_destroy(c); |
3501 | } | |
3502 | mutex_unlock(&set_limit_mutex); | |
3503 | } | |
3504 | ||
d7f25f8a GC |
3505 | struct create_work { |
3506 | struct mem_cgroup *memcg; | |
3507 | struct kmem_cache *cachep; | |
3508 | struct work_struct work; | |
3509 | }; | |
3510 | ||
1f458cbf GC |
3511 | static void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg) |
3512 | { | |
3513 | struct kmem_cache *cachep; | |
3514 | struct memcg_cache_params *params; | |
3515 | ||
3516 | if (!memcg_kmem_is_active(memcg)) | |
3517 | return; | |
3518 | ||
3519 | mutex_lock(&memcg->slab_caches_mutex); | |
3520 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) { | |
3521 | cachep = memcg_params_to_cache(params); | |
3522 | cachep->memcg_params->dead = true; | |
1f458cbf GC |
3523 | schedule_work(&cachep->memcg_params->destroy); |
3524 | } | |
3525 | mutex_unlock(&memcg->slab_caches_mutex); | |
3526 | } | |
3527 | ||
d7f25f8a GC |
3528 | static void memcg_create_cache_work_func(struct work_struct *w) |
3529 | { | |
3530 | struct create_work *cw; | |
3531 | ||
3532 | cw = container_of(w, struct create_work, work); | |
3533 | memcg_create_kmem_cache(cw->memcg, cw->cachep); | |
d7f25f8a GC |
3534 | kfree(cw); |
3535 | } | |
3536 | ||
3537 | /* | |
3538 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 3539 | */ |
0e9d92f2 GC |
3540 | static void __memcg_create_cache_enqueue(struct mem_cgroup *memcg, |
3541 | struct kmem_cache *cachep) | |
d7f25f8a GC |
3542 | { |
3543 | struct create_work *cw; | |
3544 | ||
3545 | cw = kmalloc(sizeof(struct create_work), GFP_NOWAIT); | |
ca0dde97 LZ |
3546 | if (cw == NULL) { |
3547 | css_put(&memcg->css); | |
d7f25f8a GC |
3548 | return; |
3549 | } | |
3550 | ||
3551 | cw->memcg = memcg; | |
3552 | cw->cachep = cachep; | |
3553 | ||
3554 | INIT_WORK(&cw->work, memcg_create_cache_work_func); | |
3555 | schedule_work(&cw->work); | |
3556 | } | |
3557 | ||
0e9d92f2 GC |
3558 | static void memcg_create_cache_enqueue(struct mem_cgroup *memcg, |
3559 | struct kmem_cache *cachep) | |
3560 | { | |
3561 | /* | |
3562 | * We need to stop accounting when we kmalloc, because if the | |
3563 | * corresponding kmalloc cache is not yet created, the first allocation | |
3564 | * in __memcg_create_cache_enqueue will recurse. | |
3565 | * | |
3566 | * However, it is better to enclose the whole function. Depending on | |
3567 | * the debugging options enabled, INIT_WORK(), for instance, can | |
3568 | * trigger an allocation. This too, will make us recurse. Because at | |
3569 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
3570 | * the safest choice is to do it like this, wrapping the whole function. | |
3571 | */ | |
3572 | memcg_stop_kmem_account(); | |
3573 | __memcg_create_cache_enqueue(memcg, cachep); | |
3574 | memcg_resume_kmem_account(); | |
3575 | } | |
d7f25f8a GC |
3576 | /* |
3577 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
3578 | * We try to use the current memcg's version of the cache. | |
3579 | * | |
3580 | * If the cache does not exist yet, if we are the first user of it, | |
3581 | * we either create it immediately, if possible, or create it asynchronously | |
3582 | * in a workqueue. | |
3583 | * In the latter case, we will let the current allocation go through with | |
3584 | * the original cache. | |
3585 | * | |
3586 | * Can't be called in interrupt context or from kernel threads. | |
3587 | * This function needs to be called with rcu_read_lock() held. | |
3588 | */ | |
3589 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, | |
3590 | gfp_t gfp) | |
3591 | { | |
3592 | struct mem_cgroup *memcg; | |
3593 | int idx; | |
3594 | ||
3595 | VM_BUG_ON(!cachep->memcg_params); | |
3596 | VM_BUG_ON(!cachep->memcg_params->is_root_cache); | |
3597 | ||
0e9d92f2 GC |
3598 | if (!current->mm || current->memcg_kmem_skip_account) |
3599 | return cachep; | |
3600 | ||
d7f25f8a GC |
3601 | rcu_read_lock(); |
3602 | memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner)); | |
d7f25f8a GC |
3603 | |
3604 | if (!memcg_can_account_kmem(memcg)) | |
ca0dde97 | 3605 | goto out; |
d7f25f8a GC |
3606 | |
3607 | idx = memcg_cache_id(memcg); | |
3608 | ||
3609 | /* | |
3610 | * barrier to mare sure we're always seeing the up to date value. The | |
3611 | * code updating memcg_caches will issue a write barrier to match this. | |
3612 | */ | |
3613 | read_barrier_depends(); | |
ca0dde97 LZ |
3614 | if (likely(cachep->memcg_params->memcg_caches[idx])) { |
3615 | cachep = cachep->memcg_params->memcg_caches[idx]; | |
3616 | goto out; | |
d7f25f8a GC |
3617 | } |
3618 | ||
ca0dde97 LZ |
3619 | /* The corresponding put will be done in the workqueue. */ |
3620 | if (!css_tryget(&memcg->css)) | |
3621 | goto out; | |
3622 | rcu_read_unlock(); | |
3623 | ||
3624 | /* | |
3625 | * If we are in a safe context (can wait, and not in interrupt | |
3626 | * context), we could be be predictable and return right away. | |
3627 | * This would guarantee that the allocation being performed | |
3628 | * already belongs in the new cache. | |
3629 | * | |
3630 | * However, there are some clashes that can arrive from locking. | |
3631 | * For instance, because we acquire the slab_mutex while doing | |
3632 | * kmem_cache_dup, this means no further allocation could happen | |
3633 | * with the slab_mutex held. | |
3634 | * | |
3635 | * Also, because cache creation issue get_online_cpus(), this | |
3636 | * creates a lock chain: memcg_slab_mutex -> cpu_hotplug_mutex, | |
3637 | * that ends up reversed during cpu hotplug. (cpuset allocates | |
3638 | * a bunch of GFP_KERNEL memory during cpuup). Due to all that, | |
3639 | * better to defer everything. | |
3640 | */ | |
3641 | memcg_create_cache_enqueue(memcg, cachep); | |
3642 | return cachep; | |
3643 | out: | |
3644 | rcu_read_unlock(); | |
3645 | return cachep; | |
d7f25f8a GC |
3646 | } |
3647 | EXPORT_SYMBOL(__memcg_kmem_get_cache); | |
3648 | ||
7ae1e1d0 GC |
3649 | /* |
3650 | * We need to verify if the allocation against current->mm->owner's memcg is | |
3651 | * possible for the given order. But the page is not allocated yet, so we'll | |
3652 | * need a further commit step to do the final arrangements. | |
3653 | * | |
3654 | * It is possible for the task to switch cgroups in this mean time, so at | |
3655 | * commit time, we can't rely on task conversion any longer. We'll then use | |
3656 | * the handle argument to return to the caller which cgroup we should commit | |
3657 | * against. We could also return the memcg directly and avoid the pointer | |
3658 | * passing, but a boolean return value gives better semantics considering | |
3659 | * the compiled-out case as well. | |
3660 | * | |
3661 | * Returning true means the allocation is possible. | |
3662 | */ | |
3663 | bool | |
3664 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) | |
3665 | { | |
3666 | struct mem_cgroup *memcg; | |
3667 | int ret; | |
3668 | ||
3669 | *_memcg = NULL; | |
6d42c232 GC |
3670 | |
3671 | /* | |
3672 | * Disabling accounting is only relevant for some specific memcg | |
3673 | * internal allocations. Therefore we would initially not have such | |
3674 | * check here, since direct calls to the page allocator that are marked | |
3675 | * with GFP_KMEMCG only happen outside memcg core. We are mostly | |
3676 | * concerned with cache allocations, and by having this test at | |
3677 | * memcg_kmem_get_cache, we are already able to relay the allocation to | |
3678 | * the root cache and bypass the memcg cache altogether. | |
3679 | * | |
3680 | * There is one exception, though: the SLUB allocator does not create | |
3681 | * large order caches, but rather service large kmallocs directly from | |
3682 | * the page allocator. Therefore, the following sequence when backed by | |
3683 | * the SLUB allocator: | |
3684 | * | |
f894ffa8 AM |
3685 | * memcg_stop_kmem_account(); |
3686 | * kmalloc(<large_number>) | |
3687 | * memcg_resume_kmem_account(); | |
6d42c232 GC |
3688 | * |
3689 | * would effectively ignore the fact that we should skip accounting, | |
3690 | * since it will drive us directly to this function without passing | |
3691 | * through the cache selector memcg_kmem_get_cache. Such large | |
3692 | * allocations are extremely rare but can happen, for instance, for the | |
3693 | * cache arrays. We bring this test here. | |
3694 | */ | |
3695 | if (!current->mm || current->memcg_kmem_skip_account) | |
3696 | return true; | |
3697 | ||
7ae1e1d0 GC |
3698 | memcg = try_get_mem_cgroup_from_mm(current->mm); |
3699 | ||
3700 | /* | |
3701 | * very rare case described in mem_cgroup_from_task. Unfortunately there | |
3702 | * isn't much we can do without complicating this too much, and it would | |
3703 | * be gfp-dependent anyway. Just let it go | |
3704 | */ | |
3705 | if (unlikely(!memcg)) | |
3706 | return true; | |
3707 | ||
3708 | if (!memcg_can_account_kmem(memcg)) { | |
3709 | css_put(&memcg->css); | |
3710 | return true; | |
3711 | } | |
3712 | ||
7ae1e1d0 GC |
3713 | ret = memcg_charge_kmem(memcg, gfp, PAGE_SIZE << order); |
3714 | if (!ret) | |
3715 | *_memcg = memcg; | |
7ae1e1d0 GC |
3716 | |
3717 | css_put(&memcg->css); | |
3718 | return (ret == 0); | |
3719 | } | |
3720 | ||
3721 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
3722 | int order) | |
3723 | { | |
3724 | struct page_cgroup *pc; | |
3725 | ||
3726 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
3727 | ||
3728 | /* The page allocation failed. Revert */ | |
3729 | if (!page) { | |
3730 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); | |
7ae1e1d0 GC |
3731 | return; |
3732 | } | |
3733 | ||
3734 | pc = lookup_page_cgroup(page); | |
3735 | lock_page_cgroup(pc); | |
3736 | pc->mem_cgroup = memcg; | |
3737 | SetPageCgroupUsed(pc); | |
3738 | unlock_page_cgroup(pc); | |
3739 | } | |
3740 | ||
3741 | void __memcg_kmem_uncharge_pages(struct page *page, int order) | |
3742 | { | |
3743 | struct mem_cgroup *memcg = NULL; | |
3744 | struct page_cgroup *pc; | |
3745 | ||
3746 | ||
3747 | pc = lookup_page_cgroup(page); | |
3748 | /* | |
3749 | * Fast unlocked return. Theoretically might have changed, have to | |
3750 | * check again after locking. | |
3751 | */ | |
3752 | if (!PageCgroupUsed(pc)) | |
3753 | return; | |
3754 | ||
3755 | lock_page_cgroup(pc); | |
3756 | if (PageCgroupUsed(pc)) { | |
3757 | memcg = pc->mem_cgroup; | |
3758 | ClearPageCgroupUsed(pc); | |
3759 | } | |
3760 | unlock_page_cgroup(pc); | |
3761 | ||
3762 | /* | |
3763 | * We trust that only if there is a memcg associated with the page, it | |
3764 | * is a valid allocation | |
3765 | */ | |
3766 | if (!memcg) | |
3767 | return; | |
3768 | ||
3769 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
3770 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); | |
7ae1e1d0 | 3771 | } |
1f458cbf GC |
3772 | #else |
3773 | static inline void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg) | |
3774 | { | |
3775 | } | |
7ae1e1d0 GC |
3776 | #endif /* CONFIG_MEMCG_KMEM */ |
3777 | ||
ca3e0214 KH |
3778 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3779 | ||
a0db00fc | 3780 | #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION) |
ca3e0214 KH |
3781 | /* |
3782 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
3783 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
3784 | * charge/uncharge will be never happen and move_account() is done under | |
3785 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 3786 | */ |
e94c8a9c | 3787 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 KH |
3788 | { |
3789 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
e94c8a9c | 3790 | struct page_cgroup *pc; |
b070e65c | 3791 | struct mem_cgroup *memcg; |
e94c8a9c | 3792 | int i; |
ca3e0214 | 3793 | |
3d37c4a9 KH |
3794 | if (mem_cgroup_disabled()) |
3795 | return; | |
b070e65c DR |
3796 | |
3797 | memcg = head_pc->mem_cgroup; | |
e94c8a9c KH |
3798 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
3799 | pc = head_pc + i; | |
b070e65c | 3800 | pc->mem_cgroup = memcg; |
e94c8a9c | 3801 | smp_wmb();/* see __commit_charge() */ |
e94c8a9c KH |
3802 | pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
3803 | } | |
b070e65c DR |
3804 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
3805 | HPAGE_PMD_NR); | |
ca3e0214 | 3806 | } |
12d27107 | 3807 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3808 | |
3ea67d06 SZ |
3809 | static inline |
3810 | void mem_cgroup_move_account_page_stat(struct mem_cgroup *from, | |
3811 | struct mem_cgroup *to, | |
3812 | unsigned int nr_pages, | |
3813 | enum mem_cgroup_stat_index idx) | |
3814 | { | |
3815 | /* Update stat data for mem_cgroup */ | |
3816 | preempt_disable(); | |
5e8cfc3c | 3817 | __this_cpu_sub(from->stat->count[idx], nr_pages); |
3ea67d06 SZ |
3818 | __this_cpu_add(to->stat->count[idx], nr_pages); |
3819 | preempt_enable(); | |
3820 | } | |
3821 | ||
f817ed48 | 3822 | /** |
de3638d9 | 3823 | * mem_cgroup_move_account - move account of the page |
5564e88b | 3824 | * @page: the page |
7ec99d62 | 3825 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
3826 | * @pc: page_cgroup of the page. |
3827 | * @from: mem_cgroup which the page is moved from. | |
3828 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
3829 | * | |
3830 | * The caller must confirm following. | |
08e552c6 | 3831 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 3832 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 3833 | * |
2f3479b1 KH |
3834 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
3835 | * from old cgroup. | |
f817ed48 | 3836 | */ |
7ec99d62 JW |
3837 | static int mem_cgroup_move_account(struct page *page, |
3838 | unsigned int nr_pages, | |
3839 | struct page_cgroup *pc, | |
3840 | struct mem_cgroup *from, | |
2f3479b1 | 3841 | struct mem_cgroup *to) |
f817ed48 | 3842 | { |
de3638d9 JW |
3843 | unsigned long flags; |
3844 | int ret; | |
b2402857 | 3845 | bool anon = PageAnon(page); |
987eba66 | 3846 | |
f817ed48 | 3847 | VM_BUG_ON(from == to); |
5564e88b | 3848 | VM_BUG_ON(PageLRU(page)); |
de3638d9 JW |
3849 | /* |
3850 | * The page is isolated from LRU. So, collapse function | |
3851 | * will not handle this page. But page splitting can happen. | |
3852 | * Do this check under compound_page_lock(). The caller should | |
3853 | * hold it. | |
3854 | */ | |
3855 | ret = -EBUSY; | |
7ec99d62 | 3856 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
3857 | goto out; |
3858 | ||
3859 | lock_page_cgroup(pc); | |
3860 | ||
3861 | ret = -EINVAL; | |
3862 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
3863 | goto unlock; | |
3864 | ||
312734c0 | 3865 | move_lock_mem_cgroup(from, &flags); |
f817ed48 | 3866 | |
3ea67d06 SZ |
3867 | if (!anon && page_mapped(page)) |
3868 | mem_cgroup_move_account_page_stat(from, to, nr_pages, | |
3869 | MEM_CGROUP_STAT_FILE_MAPPED); | |
3870 | ||
3871 | if (PageWriteback(page)) | |
3872 | mem_cgroup_move_account_page_stat(from, to, nr_pages, | |
3873 | MEM_CGROUP_STAT_WRITEBACK); | |
3874 | ||
b070e65c | 3875 | mem_cgroup_charge_statistics(from, page, anon, -nr_pages); |
d69b042f | 3876 | |
854ffa8d | 3877 | /* caller should have done css_get */ |
08e552c6 | 3878 | pc->mem_cgroup = to; |
b070e65c | 3879 | mem_cgroup_charge_statistics(to, page, anon, nr_pages); |
312734c0 | 3880 | move_unlock_mem_cgroup(from, &flags); |
de3638d9 JW |
3881 | ret = 0; |
3882 | unlock: | |
57f9fd7d | 3883 | unlock_page_cgroup(pc); |
d2265e6f KH |
3884 | /* |
3885 | * check events | |
3886 | */ | |
5564e88b JW |
3887 | memcg_check_events(to, page); |
3888 | memcg_check_events(from, page); | |
de3638d9 | 3889 | out: |
f817ed48 KH |
3890 | return ret; |
3891 | } | |
3892 | ||
2ef37d3f MH |
3893 | /** |
3894 | * mem_cgroup_move_parent - moves page to the parent group | |
3895 | * @page: the page to move | |
3896 | * @pc: page_cgroup of the page | |
3897 | * @child: page's cgroup | |
3898 | * | |
3899 | * move charges to its parent or the root cgroup if the group has no | |
3900 | * parent (aka use_hierarchy==0). | |
3901 | * Although this might fail (get_page_unless_zero, isolate_lru_page or | |
3902 | * mem_cgroup_move_account fails) the failure is always temporary and | |
3903 | * it signals a race with a page removal/uncharge or migration. In the | |
3904 | * first case the page is on the way out and it will vanish from the LRU | |
3905 | * on the next attempt and the call should be retried later. | |
3906 | * Isolation from the LRU fails only if page has been isolated from | |
3907 | * the LRU since we looked at it and that usually means either global | |
3908 | * reclaim or migration going on. The page will either get back to the | |
3909 | * LRU or vanish. | |
3910 | * Finaly mem_cgroup_move_account fails only if the page got uncharged | |
3911 | * (!PageCgroupUsed) or moved to a different group. The page will | |
3912 | * disappear in the next attempt. | |
f817ed48 | 3913 | */ |
5564e88b JW |
3914 | static int mem_cgroup_move_parent(struct page *page, |
3915 | struct page_cgroup *pc, | |
6068bf01 | 3916 | struct mem_cgroup *child) |
f817ed48 | 3917 | { |
f817ed48 | 3918 | struct mem_cgroup *parent; |
7ec99d62 | 3919 | unsigned int nr_pages; |
4be4489f | 3920 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
3921 | int ret; |
3922 | ||
d8423011 | 3923 | VM_BUG_ON(mem_cgroup_is_root(child)); |
f817ed48 | 3924 | |
57f9fd7d DN |
3925 | ret = -EBUSY; |
3926 | if (!get_page_unless_zero(page)) | |
3927 | goto out; | |
3928 | if (isolate_lru_page(page)) | |
3929 | goto put; | |
52dbb905 | 3930 | |
7ec99d62 | 3931 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 3932 | |
cc926f78 KH |
3933 | parent = parent_mem_cgroup(child); |
3934 | /* | |
3935 | * If no parent, move charges to root cgroup. | |
3936 | */ | |
3937 | if (!parent) | |
3938 | parent = root_mem_cgroup; | |
f817ed48 | 3939 | |
2ef37d3f MH |
3940 | if (nr_pages > 1) { |
3941 | VM_BUG_ON(!PageTransHuge(page)); | |
987eba66 | 3942 | flags = compound_lock_irqsave(page); |
2ef37d3f | 3943 | } |
987eba66 | 3944 | |
cc926f78 | 3945 | ret = mem_cgroup_move_account(page, nr_pages, |
2f3479b1 | 3946 | pc, child, parent); |
cc926f78 KH |
3947 | if (!ret) |
3948 | __mem_cgroup_cancel_local_charge(child, nr_pages); | |
8dba474f | 3949 | |
7ec99d62 | 3950 | if (nr_pages > 1) |
987eba66 | 3951 | compound_unlock_irqrestore(page, flags); |
08e552c6 | 3952 | putback_lru_page(page); |
57f9fd7d | 3953 | put: |
40d58138 | 3954 | put_page(page); |
57f9fd7d | 3955 | out: |
f817ed48 KH |
3956 | return ret; |
3957 | } | |
3958 | ||
7a81b88c KH |
3959 | /* |
3960 | * Charge the memory controller for page usage. | |
3961 | * Return | |
3962 | * 0 if the charge was successful | |
3963 | * < 0 if the cgroup is over its limit | |
3964 | */ | |
3965 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
73045c47 | 3966 | gfp_t gfp_mask, enum charge_type ctype) |
7a81b88c | 3967 | { |
c0ff4b85 | 3968 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 3969 | unsigned int nr_pages = 1; |
8493ae43 | 3970 | bool oom = true; |
7a81b88c | 3971 | int ret; |
ec168510 | 3972 | |
37c2ac78 | 3973 | if (PageTransHuge(page)) { |
7ec99d62 | 3974 | nr_pages <<= compound_order(page); |
37c2ac78 | 3975 | VM_BUG_ON(!PageTransHuge(page)); |
8493ae43 JW |
3976 | /* |
3977 | * Never OOM-kill a process for a huge page. The | |
3978 | * fault handler will fall back to regular pages. | |
3979 | */ | |
3980 | oom = false; | |
37c2ac78 | 3981 | } |
7a81b88c | 3982 | |
c0ff4b85 | 3983 | ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom); |
38c5d72f | 3984 | if (ret == -ENOMEM) |
7a81b88c | 3985 | return ret; |
ce587e65 | 3986 | __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false); |
8a9f3ccd | 3987 | return 0; |
8a9f3ccd BS |
3988 | } |
3989 | ||
7a81b88c KH |
3990 | int mem_cgroup_newpage_charge(struct page *page, |
3991 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 3992 | { |
f8d66542 | 3993 | if (mem_cgroup_disabled()) |
cede86ac | 3994 | return 0; |
7a0524cf JW |
3995 | VM_BUG_ON(page_mapped(page)); |
3996 | VM_BUG_ON(page->mapping && !PageAnon(page)); | |
3997 | VM_BUG_ON(!mm); | |
217bc319 | 3998 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
41326c17 | 3999 | MEM_CGROUP_CHARGE_TYPE_ANON); |
217bc319 KH |
4000 | } |
4001 | ||
54595fe2 KH |
4002 | /* |
4003 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
4004 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 4005 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
4006 | * "commit()" or removed by "cancel()" |
4007 | */ | |
0435a2fd JW |
4008 | static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
4009 | struct page *page, | |
4010 | gfp_t mask, | |
4011 | struct mem_cgroup **memcgp) | |
8c7c6e34 | 4012 | { |
c0ff4b85 | 4013 | struct mem_cgroup *memcg; |
90deb788 | 4014 | struct page_cgroup *pc; |
54595fe2 | 4015 | int ret; |
8c7c6e34 | 4016 | |
90deb788 JW |
4017 | pc = lookup_page_cgroup(page); |
4018 | /* | |
4019 | * Every swap fault against a single page tries to charge the | |
4020 | * page, bail as early as possible. shmem_unuse() encounters | |
4021 | * already charged pages, too. The USED bit is protected by | |
4022 | * the page lock, which serializes swap cache removal, which | |
4023 | * in turn serializes uncharging. | |
4024 | */ | |
4025 | if (PageCgroupUsed(pc)) | |
4026 | return 0; | |
8c7c6e34 KH |
4027 | if (!do_swap_account) |
4028 | goto charge_cur_mm; | |
c0ff4b85 R |
4029 | memcg = try_get_mem_cgroup_from_page(page); |
4030 | if (!memcg) | |
54595fe2 | 4031 | goto charge_cur_mm; |
72835c86 JW |
4032 | *memcgp = memcg; |
4033 | ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true); | |
c0ff4b85 | 4034 | css_put(&memcg->css); |
38c5d72f KH |
4035 | if (ret == -EINTR) |
4036 | ret = 0; | |
54595fe2 | 4037 | return ret; |
8c7c6e34 | 4038 | charge_cur_mm: |
38c5d72f KH |
4039 | ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true); |
4040 | if (ret == -EINTR) | |
4041 | ret = 0; | |
4042 | return ret; | |
8c7c6e34 KH |
4043 | } |
4044 | ||
0435a2fd JW |
4045 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page, |
4046 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
4047 | { | |
4048 | *memcgp = NULL; | |
4049 | if (mem_cgroup_disabled()) | |
4050 | return 0; | |
bdf4f4d2 JW |
4051 | /* |
4052 | * A racing thread's fault, or swapoff, may have already | |
4053 | * updated the pte, and even removed page from swap cache: in | |
4054 | * those cases unuse_pte()'s pte_same() test will fail; but | |
4055 | * there's also a KSM case which does need to charge the page. | |
4056 | */ | |
4057 | if (!PageSwapCache(page)) { | |
4058 | int ret; | |
4059 | ||
4060 | ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, memcgp, true); | |
4061 | if (ret == -EINTR) | |
4062 | ret = 0; | |
4063 | return ret; | |
4064 | } | |
0435a2fd JW |
4065 | return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp); |
4066 | } | |
4067 | ||
827a03d2 JW |
4068 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg) |
4069 | { | |
4070 | if (mem_cgroup_disabled()) | |
4071 | return; | |
4072 | if (!memcg) | |
4073 | return; | |
4074 | __mem_cgroup_cancel_charge(memcg, 1); | |
4075 | } | |
4076 | ||
83aae4c7 | 4077 | static void |
72835c86 | 4078 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg, |
83aae4c7 | 4079 | enum charge_type ctype) |
7a81b88c | 4080 | { |
f8d66542 | 4081 | if (mem_cgroup_disabled()) |
7a81b88c | 4082 | return; |
72835c86 | 4083 | if (!memcg) |
7a81b88c | 4084 | return; |
5a6475a4 | 4085 | |
ce587e65 | 4086 | __mem_cgroup_commit_charge(memcg, page, 1, ctype, true); |
8c7c6e34 KH |
4087 | /* |
4088 | * Now swap is on-memory. This means this page may be | |
4089 | * counted both as mem and swap....double count. | |
03f3c433 KH |
4090 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
4091 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
4092 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 4093 | */ |
03f3c433 | 4094 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 4095 | swp_entry_t ent = {.val = page_private(page)}; |
86493009 | 4096 | mem_cgroup_uncharge_swap(ent); |
8c7c6e34 | 4097 | } |
7a81b88c KH |
4098 | } |
4099 | ||
72835c86 JW |
4100 | void mem_cgroup_commit_charge_swapin(struct page *page, |
4101 | struct mem_cgroup *memcg) | |
83aae4c7 | 4102 | { |
72835c86 | 4103 | __mem_cgroup_commit_charge_swapin(page, memcg, |
41326c17 | 4104 | MEM_CGROUP_CHARGE_TYPE_ANON); |
83aae4c7 DN |
4105 | } |
4106 | ||
827a03d2 JW |
4107 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
4108 | gfp_t gfp_mask) | |
7a81b88c | 4109 | { |
827a03d2 JW |
4110 | struct mem_cgroup *memcg = NULL; |
4111 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
4112 | int ret; | |
4113 | ||
f8d66542 | 4114 | if (mem_cgroup_disabled()) |
827a03d2 JW |
4115 | return 0; |
4116 | if (PageCompound(page)) | |
4117 | return 0; | |
4118 | ||
827a03d2 JW |
4119 | if (!PageSwapCache(page)) |
4120 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, type); | |
4121 | else { /* page is swapcache/shmem */ | |
0435a2fd JW |
4122 | ret = __mem_cgroup_try_charge_swapin(mm, page, |
4123 | gfp_mask, &memcg); | |
827a03d2 JW |
4124 | if (!ret) |
4125 | __mem_cgroup_commit_charge_swapin(page, memcg, type); | |
4126 | } | |
4127 | return ret; | |
7a81b88c KH |
4128 | } |
4129 | ||
c0ff4b85 | 4130 | static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg, |
7ec99d62 JW |
4131 | unsigned int nr_pages, |
4132 | const enum charge_type ctype) | |
569b846d KH |
4133 | { |
4134 | struct memcg_batch_info *batch = NULL; | |
4135 | bool uncharge_memsw = true; | |
7ec99d62 | 4136 | |
569b846d KH |
4137 | /* If swapout, usage of swap doesn't decrease */ |
4138 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
4139 | uncharge_memsw = false; | |
569b846d KH |
4140 | |
4141 | batch = ¤t->memcg_batch; | |
4142 | /* | |
4143 | * In usual, we do css_get() when we remember memcg pointer. | |
4144 | * But in this case, we keep res->usage until end of a series of | |
4145 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
4146 | */ | |
4147 | if (!batch->memcg) | |
c0ff4b85 | 4148 | batch->memcg = memcg; |
3c11ecf4 KH |
4149 | /* |
4150 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
25985edc | 4151 | * In those cases, all pages freed continuously can be expected to be in |
3c11ecf4 KH |
4152 | * the same cgroup and we have chance to coalesce uncharges. |
4153 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
4154 | * because we want to do uncharge as soon as possible. | |
4155 | */ | |
4156 | ||
4157 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
4158 | goto direct_uncharge; | |
4159 | ||
7ec99d62 | 4160 | if (nr_pages > 1) |
ec168510 AA |
4161 | goto direct_uncharge; |
4162 | ||
569b846d KH |
4163 | /* |
4164 | * In typical case, batch->memcg == mem. This means we can | |
4165 | * merge a series of uncharges to an uncharge of res_counter. | |
4166 | * If not, we uncharge res_counter ony by one. | |
4167 | */ | |
c0ff4b85 | 4168 | if (batch->memcg != memcg) |
569b846d KH |
4169 | goto direct_uncharge; |
4170 | /* remember freed charge and uncharge it later */ | |
7ffd4ca7 | 4171 | batch->nr_pages++; |
569b846d | 4172 | if (uncharge_memsw) |
7ffd4ca7 | 4173 | batch->memsw_nr_pages++; |
569b846d KH |
4174 | return; |
4175 | direct_uncharge: | |
c0ff4b85 | 4176 | res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE); |
569b846d | 4177 | if (uncharge_memsw) |
c0ff4b85 R |
4178 | res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE); |
4179 | if (unlikely(batch->memcg != memcg)) | |
4180 | memcg_oom_recover(memcg); | |
569b846d | 4181 | } |
7a81b88c | 4182 | |
8a9f3ccd | 4183 | /* |
69029cd5 | 4184 | * uncharge if !page_mapped(page) |
8a9f3ccd | 4185 | */ |
8c7c6e34 | 4186 | static struct mem_cgroup * |
0030f535 JW |
4187 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype, |
4188 | bool end_migration) | |
8a9f3ccd | 4189 | { |
c0ff4b85 | 4190 | struct mem_cgroup *memcg = NULL; |
7ec99d62 JW |
4191 | unsigned int nr_pages = 1; |
4192 | struct page_cgroup *pc; | |
b2402857 | 4193 | bool anon; |
8a9f3ccd | 4194 | |
f8d66542 | 4195 | if (mem_cgroup_disabled()) |
8c7c6e34 | 4196 | return NULL; |
4077960e | 4197 | |
37c2ac78 | 4198 | if (PageTransHuge(page)) { |
7ec99d62 | 4199 | nr_pages <<= compound_order(page); |
37c2ac78 AA |
4200 | VM_BUG_ON(!PageTransHuge(page)); |
4201 | } | |
8697d331 | 4202 | /* |
3c541e14 | 4203 | * Check if our page_cgroup is valid |
8697d331 | 4204 | */ |
52d4b9ac | 4205 | pc = lookup_page_cgroup(page); |
cfa44946 | 4206 | if (unlikely(!PageCgroupUsed(pc))) |
8c7c6e34 | 4207 | return NULL; |
b9c565d5 | 4208 | |
52d4b9ac | 4209 | lock_page_cgroup(pc); |
d13d1443 | 4210 | |
c0ff4b85 | 4211 | memcg = pc->mem_cgroup; |
8c7c6e34 | 4212 | |
d13d1443 KH |
4213 | if (!PageCgroupUsed(pc)) |
4214 | goto unlock_out; | |
4215 | ||
b2402857 KH |
4216 | anon = PageAnon(page); |
4217 | ||
d13d1443 | 4218 | switch (ctype) { |
41326c17 | 4219 | case MEM_CGROUP_CHARGE_TYPE_ANON: |
2ff76f11 KH |
4220 | /* |
4221 | * Generally PageAnon tells if it's the anon statistics to be | |
4222 | * updated; but sometimes e.g. mem_cgroup_uncharge_page() is | |
4223 | * used before page reached the stage of being marked PageAnon. | |
4224 | */ | |
b2402857 KH |
4225 | anon = true; |
4226 | /* fallthrough */ | |
8a9478ca | 4227 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c | 4228 | /* See mem_cgroup_prepare_migration() */ |
0030f535 JW |
4229 | if (page_mapped(page)) |
4230 | goto unlock_out; | |
4231 | /* | |
4232 | * Pages under migration may not be uncharged. But | |
4233 | * end_migration() /must/ be the one uncharging the | |
4234 | * unused post-migration page and so it has to call | |
4235 | * here with the migration bit still set. See the | |
4236 | * res_counter handling below. | |
4237 | */ | |
4238 | if (!end_migration && PageCgroupMigration(pc)) | |
d13d1443 KH |
4239 | goto unlock_out; |
4240 | break; | |
4241 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
4242 | if (!PageAnon(page)) { /* Shared memory */ | |
4243 | if (page->mapping && !page_is_file_cache(page)) | |
4244 | goto unlock_out; | |
4245 | } else if (page_mapped(page)) /* Anon */ | |
4246 | goto unlock_out; | |
4247 | break; | |
4248 | default: | |
4249 | break; | |
52d4b9ac | 4250 | } |
d13d1443 | 4251 | |
b070e65c | 4252 | mem_cgroup_charge_statistics(memcg, page, anon, -nr_pages); |
04046e1a | 4253 | |
52d4b9ac | 4254 | ClearPageCgroupUsed(pc); |
544122e5 KH |
4255 | /* |
4256 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
4257 | * freed from LRU. This is safe because uncharged page is expected not | |
4258 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
4259 | * special functions. | |
4260 | */ | |
b9c565d5 | 4261 | |
52d4b9ac | 4262 | unlock_page_cgroup(pc); |
f75ca962 | 4263 | /* |
c0ff4b85 | 4264 | * even after unlock, we have memcg->res.usage here and this memcg |
4050377b | 4265 | * will never be freed, so it's safe to call css_get(). |
f75ca962 | 4266 | */ |
c0ff4b85 | 4267 | memcg_check_events(memcg, page); |
f75ca962 | 4268 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
c0ff4b85 | 4269 | mem_cgroup_swap_statistics(memcg, true); |
4050377b | 4270 | css_get(&memcg->css); |
f75ca962 | 4271 | } |
0030f535 JW |
4272 | /* |
4273 | * Migration does not charge the res_counter for the | |
4274 | * replacement page, so leave it alone when phasing out the | |
4275 | * page that is unused after the migration. | |
4276 | */ | |
4277 | if (!end_migration && !mem_cgroup_is_root(memcg)) | |
c0ff4b85 | 4278 | mem_cgroup_do_uncharge(memcg, nr_pages, ctype); |
6d12e2d8 | 4279 | |
c0ff4b85 | 4280 | return memcg; |
d13d1443 KH |
4281 | |
4282 | unlock_out: | |
4283 | unlock_page_cgroup(pc); | |
8c7c6e34 | 4284 | return NULL; |
3c541e14 BS |
4285 | } |
4286 | ||
69029cd5 KH |
4287 | void mem_cgroup_uncharge_page(struct page *page) |
4288 | { | |
52d4b9ac KH |
4289 | /* early check. */ |
4290 | if (page_mapped(page)) | |
4291 | return; | |
40f23a21 | 4292 | VM_BUG_ON(page->mapping && !PageAnon(page)); |
28ccddf7 JW |
4293 | /* |
4294 | * If the page is in swap cache, uncharge should be deferred | |
4295 | * to the swap path, which also properly accounts swap usage | |
4296 | * and handles memcg lifetime. | |
4297 | * | |
4298 | * Note that this check is not stable and reclaim may add the | |
4299 | * page to swap cache at any time after this. However, if the | |
4300 | * page is not in swap cache by the time page->mapcount hits | |
4301 | * 0, there won't be any page table references to the swap | |
4302 | * slot, and reclaim will free it and not actually write the | |
4303 | * page to disk. | |
4304 | */ | |
0c59b89c JW |
4305 | if (PageSwapCache(page)) |
4306 | return; | |
0030f535 | 4307 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false); |
69029cd5 KH |
4308 | } |
4309 | ||
4310 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
4311 | { | |
4312 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 4313 | VM_BUG_ON(page->mapping); |
0030f535 | 4314 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false); |
69029cd5 KH |
4315 | } |
4316 | ||
569b846d KH |
4317 | /* |
4318 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
4319 | * In that cases, pages are freed continuously and we can expect pages | |
4320 | * are in the same memcg. All these calls itself limits the number of | |
4321 | * pages freed at once, then uncharge_start/end() is called properly. | |
4322 | * This may be called prural(2) times in a context, | |
4323 | */ | |
4324 | ||
4325 | void mem_cgroup_uncharge_start(void) | |
4326 | { | |
4327 | current->memcg_batch.do_batch++; | |
4328 | /* We can do nest. */ | |
4329 | if (current->memcg_batch.do_batch == 1) { | |
4330 | current->memcg_batch.memcg = NULL; | |
7ffd4ca7 JW |
4331 | current->memcg_batch.nr_pages = 0; |
4332 | current->memcg_batch.memsw_nr_pages = 0; | |
569b846d KH |
4333 | } |
4334 | } | |
4335 | ||
4336 | void mem_cgroup_uncharge_end(void) | |
4337 | { | |
4338 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
4339 | ||
4340 | if (!batch->do_batch) | |
4341 | return; | |
4342 | ||
4343 | batch->do_batch--; | |
4344 | if (batch->do_batch) /* If stacked, do nothing. */ | |
4345 | return; | |
4346 | ||
4347 | if (!batch->memcg) | |
4348 | return; | |
4349 | /* | |
4350 | * This "batch->memcg" is valid without any css_get/put etc... | |
4351 | * bacause we hide charges behind us. | |
4352 | */ | |
7ffd4ca7 JW |
4353 | if (batch->nr_pages) |
4354 | res_counter_uncharge(&batch->memcg->res, | |
4355 | batch->nr_pages * PAGE_SIZE); | |
4356 | if (batch->memsw_nr_pages) | |
4357 | res_counter_uncharge(&batch->memcg->memsw, | |
4358 | batch->memsw_nr_pages * PAGE_SIZE); | |
3c11ecf4 | 4359 | memcg_oom_recover(batch->memcg); |
569b846d KH |
4360 | /* forget this pointer (for sanity check) */ |
4361 | batch->memcg = NULL; | |
4362 | } | |
4363 | ||
e767e056 | 4364 | #ifdef CONFIG_SWAP |
8c7c6e34 | 4365 | /* |
e767e056 | 4366 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
4367 | * memcg information is recorded to swap_cgroup of "ent" |
4368 | */ | |
8a9478ca KH |
4369 | void |
4370 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
4371 | { |
4372 | struct mem_cgroup *memcg; | |
8a9478ca KH |
4373 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
4374 | ||
4375 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
4376 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
4377 | ||
0030f535 | 4378 | memcg = __mem_cgroup_uncharge_common(page, ctype, false); |
8c7c6e34 | 4379 | |
f75ca962 KH |
4380 | /* |
4381 | * record memcg information, if swapout && memcg != NULL, | |
4050377b | 4382 | * css_get() was called in uncharge(). |
f75ca962 KH |
4383 | */ |
4384 | if (do_swap_account && swapout && memcg) | |
a3b2d692 | 4385 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 | 4386 | } |
e767e056 | 4387 | #endif |
8c7c6e34 | 4388 | |
c255a458 | 4389 | #ifdef CONFIG_MEMCG_SWAP |
8c7c6e34 KH |
4390 | /* |
4391 | * called from swap_entry_free(). remove record in swap_cgroup and | |
4392 | * uncharge "memsw" account. | |
4393 | */ | |
4394 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 4395 | { |
8c7c6e34 | 4396 | struct mem_cgroup *memcg; |
a3b2d692 | 4397 | unsigned short id; |
8c7c6e34 KH |
4398 | |
4399 | if (!do_swap_account) | |
4400 | return; | |
4401 | ||
a3b2d692 KH |
4402 | id = swap_cgroup_record(ent, 0); |
4403 | rcu_read_lock(); | |
4404 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 4405 | if (memcg) { |
a3b2d692 KH |
4406 | /* |
4407 | * We uncharge this because swap is freed. | |
4408 | * This memcg can be obsolete one. We avoid calling css_tryget | |
4409 | */ | |
0c3e73e8 | 4410 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 4411 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 4412 | mem_cgroup_swap_statistics(memcg, false); |
4050377b | 4413 | css_put(&memcg->css); |
8c7c6e34 | 4414 | } |
a3b2d692 | 4415 | rcu_read_unlock(); |
d13d1443 | 4416 | } |
02491447 DN |
4417 | |
4418 | /** | |
4419 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
4420 | * @entry: swap entry to be moved | |
4421 | * @from: mem_cgroup which the entry is moved from | |
4422 | * @to: mem_cgroup which the entry is moved to | |
4423 | * | |
4424 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
4425 | * as the mem_cgroup's id of @from. | |
4426 | * | |
4427 | * Returns 0 on success, -EINVAL on failure. | |
4428 | * | |
4429 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
4430 | * both res and memsw, and called css_get(). | |
4431 | */ | |
4432 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 4433 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
4434 | { |
4435 | unsigned short old_id, new_id; | |
4436 | ||
4437 | old_id = css_id(&from->css); | |
4438 | new_id = css_id(&to->css); | |
4439 | ||
4440 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 4441 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 4442 | mem_cgroup_swap_statistics(to, true); |
02491447 | 4443 | /* |
483c30b5 DN |
4444 | * This function is only called from task migration context now. |
4445 | * It postpones res_counter and refcount handling till the end | |
4446 | * of task migration(mem_cgroup_clear_mc()) for performance | |
4050377b LZ |
4447 | * improvement. But we cannot postpone css_get(to) because if |
4448 | * the process that has been moved to @to does swap-in, the | |
4449 | * refcount of @to might be decreased to 0. | |
4450 | * | |
4451 | * We are in attach() phase, so the cgroup is guaranteed to be | |
4452 | * alive, so we can just call css_get(). | |
02491447 | 4453 | */ |
4050377b | 4454 | css_get(&to->css); |
02491447 DN |
4455 | return 0; |
4456 | } | |
4457 | return -EINVAL; | |
4458 | } | |
4459 | #else | |
4460 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 4461 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
4462 | { |
4463 | return -EINVAL; | |
4464 | } | |
8c7c6e34 | 4465 | #endif |
d13d1443 | 4466 | |
ae41be37 | 4467 | /* |
01b1ae63 KH |
4468 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
4469 | * page belongs to. | |
ae41be37 | 4470 | */ |
0030f535 JW |
4471 | void mem_cgroup_prepare_migration(struct page *page, struct page *newpage, |
4472 | struct mem_cgroup **memcgp) | |
ae41be37 | 4473 | { |
c0ff4b85 | 4474 | struct mem_cgroup *memcg = NULL; |
b32967ff | 4475 | unsigned int nr_pages = 1; |
7ec99d62 | 4476 | struct page_cgroup *pc; |
ac39cf8c | 4477 | enum charge_type ctype; |
8869b8f6 | 4478 | |
72835c86 | 4479 | *memcgp = NULL; |
56039efa | 4480 | |
f8d66542 | 4481 | if (mem_cgroup_disabled()) |
0030f535 | 4482 | return; |
4077960e | 4483 | |
b32967ff MG |
4484 | if (PageTransHuge(page)) |
4485 | nr_pages <<= compound_order(page); | |
4486 | ||
52d4b9ac KH |
4487 | pc = lookup_page_cgroup(page); |
4488 | lock_page_cgroup(pc); | |
4489 | if (PageCgroupUsed(pc)) { | |
c0ff4b85 R |
4490 | memcg = pc->mem_cgroup; |
4491 | css_get(&memcg->css); | |
ac39cf8c | 4492 | /* |
4493 | * At migrating an anonymous page, its mapcount goes down | |
4494 | * to 0 and uncharge() will be called. But, even if it's fully | |
4495 | * unmapped, migration may fail and this page has to be | |
4496 | * charged again. We set MIGRATION flag here and delay uncharge | |
4497 | * until end_migration() is called | |
4498 | * | |
4499 | * Corner Case Thinking | |
4500 | * A) | |
4501 | * When the old page was mapped as Anon and it's unmap-and-freed | |
4502 | * while migration was ongoing. | |
4503 | * If unmap finds the old page, uncharge() of it will be delayed | |
4504 | * until end_migration(). If unmap finds a new page, it's | |
4505 | * uncharged when it make mapcount to be 1->0. If unmap code | |
4506 | * finds swap_migration_entry, the new page will not be mapped | |
4507 | * and end_migration() will find it(mapcount==0). | |
4508 | * | |
4509 | * B) | |
4510 | * When the old page was mapped but migraion fails, the kernel | |
4511 | * remaps it. A charge for it is kept by MIGRATION flag even | |
4512 | * if mapcount goes down to 0. We can do remap successfully | |
4513 | * without charging it again. | |
4514 | * | |
4515 | * C) | |
4516 | * The "old" page is under lock_page() until the end of | |
4517 | * migration, so, the old page itself will not be swapped-out. | |
4518 | * If the new page is swapped out before end_migraton, our | |
4519 | * hook to usual swap-out path will catch the event. | |
4520 | */ | |
4521 | if (PageAnon(page)) | |
4522 | SetPageCgroupMigration(pc); | |
e8589cc1 | 4523 | } |
52d4b9ac | 4524 | unlock_page_cgroup(pc); |
ac39cf8c | 4525 | /* |
4526 | * If the page is not charged at this point, | |
4527 | * we return here. | |
4528 | */ | |
c0ff4b85 | 4529 | if (!memcg) |
0030f535 | 4530 | return; |
01b1ae63 | 4531 | |
72835c86 | 4532 | *memcgp = memcg; |
ac39cf8c | 4533 | /* |
4534 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
4535 | * is called before end_migration, we can catch all events on this new | |
4536 | * page. In the case new page is migrated but not remapped, new page's | |
4537 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
4538 | */ | |
ac39cf8c | 4539 | if (PageAnon(page)) |
41326c17 | 4540 | ctype = MEM_CGROUP_CHARGE_TYPE_ANON; |
ac39cf8c | 4541 | else |
62ba7442 | 4542 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; |
0030f535 JW |
4543 | /* |
4544 | * The page is committed to the memcg, but it's not actually | |
4545 | * charged to the res_counter since we plan on replacing the | |
4546 | * old one and only one page is going to be left afterwards. | |
4547 | */ | |
b32967ff | 4548 | __mem_cgroup_commit_charge(memcg, newpage, nr_pages, ctype, false); |
ae41be37 | 4549 | } |
8869b8f6 | 4550 | |
69029cd5 | 4551 | /* remove redundant charge if migration failed*/ |
c0ff4b85 | 4552 | void mem_cgroup_end_migration(struct mem_cgroup *memcg, |
50de1dd9 | 4553 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 4554 | { |
ac39cf8c | 4555 | struct page *used, *unused; |
01b1ae63 | 4556 | struct page_cgroup *pc; |
b2402857 | 4557 | bool anon; |
01b1ae63 | 4558 | |
c0ff4b85 | 4559 | if (!memcg) |
01b1ae63 | 4560 | return; |
b25ed609 | 4561 | |
50de1dd9 | 4562 | if (!migration_ok) { |
ac39cf8c | 4563 | used = oldpage; |
4564 | unused = newpage; | |
01b1ae63 | 4565 | } else { |
ac39cf8c | 4566 | used = newpage; |
01b1ae63 KH |
4567 | unused = oldpage; |
4568 | } | |
0030f535 | 4569 | anon = PageAnon(used); |
7d188958 JW |
4570 | __mem_cgroup_uncharge_common(unused, |
4571 | anon ? MEM_CGROUP_CHARGE_TYPE_ANON | |
4572 | : MEM_CGROUP_CHARGE_TYPE_CACHE, | |
4573 | true); | |
0030f535 | 4574 | css_put(&memcg->css); |
69029cd5 | 4575 | /* |
ac39cf8c | 4576 | * We disallowed uncharge of pages under migration because mapcount |
4577 | * of the page goes down to zero, temporarly. | |
4578 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 4579 | */ |
ac39cf8c | 4580 | pc = lookup_page_cgroup(oldpage); |
4581 | lock_page_cgroup(pc); | |
4582 | ClearPageCgroupMigration(pc); | |
4583 | unlock_page_cgroup(pc); | |
ac39cf8c | 4584 | |
01b1ae63 | 4585 | /* |
ac39cf8c | 4586 | * If a page is a file cache, radix-tree replacement is very atomic |
4587 | * and we can skip this check. When it was an Anon page, its mapcount | |
4588 | * goes down to 0. But because we added MIGRATION flage, it's not | |
4589 | * uncharged yet. There are several case but page->mapcount check | |
4590 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
4591 | * check. (see prepare_charge() also) | |
69029cd5 | 4592 | */ |
b2402857 | 4593 | if (anon) |
ac39cf8c | 4594 | mem_cgroup_uncharge_page(used); |
ae41be37 | 4595 | } |
78fb7466 | 4596 | |
ab936cbc KH |
4597 | /* |
4598 | * At replace page cache, newpage is not under any memcg but it's on | |
4599 | * LRU. So, this function doesn't touch res_counter but handles LRU | |
4600 | * in correct way. Both pages are locked so we cannot race with uncharge. | |
4601 | */ | |
4602 | void mem_cgroup_replace_page_cache(struct page *oldpage, | |
4603 | struct page *newpage) | |
4604 | { | |
bde05d1c | 4605 | struct mem_cgroup *memcg = NULL; |
ab936cbc | 4606 | struct page_cgroup *pc; |
ab936cbc | 4607 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
ab936cbc KH |
4608 | |
4609 | if (mem_cgroup_disabled()) | |
4610 | return; | |
4611 | ||
4612 | pc = lookup_page_cgroup(oldpage); | |
4613 | /* fix accounting on old pages */ | |
4614 | lock_page_cgroup(pc); | |
bde05d1c HD |
4615 | if (PageCgroupUsed(pc)) { |
4616 | memcg = pc->mem_cgroup; | |
b070e65c | 4617 | mem_cgroup_charge_statistics(memcg, oldpage, false, -1); |
bde05d1c HD |
4618 | ClearPageCgroupUsed(pc); |
4619 | } | |
ab936cbc KH |
4620 | unlock_page_cgroup(pc); |
4621 | ||
bde05d1c HD |
4622 | /* |
4623 | * When called from shmem_replace_page(), in some cases the | |
4624 | * oldpage has already been charged, and in some cases not. | |
4625 | */ | |
4626 | if (!memcg) | |
4627 | return; | |
ab936cbc KH |
4628 | /* |
4629 | * Even if newpage->mapping was NULL before starting replacement, | |
4630 | * the newpage may be on LRU(or pagevec for LRU) already. We lock | |
4631 | * LRU while we overwrite pc->mem_cgroup. | |
4632 | */ | |
ce587e65 | 4633 | __mem_cgroup_commit_charge(memcg, newpage, 1, type, true); |
ab936cbc KH |
4634 | } |
4635 | ||
f212ad7c DN |
4636 | #ifdef CONFIG_DEBUG_VM |
4637 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
4638 | { | |
4639 | struct page_cgroup *pc; | |
4640 | ||
4641 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
4642 | /* |
4643 | * Can be NULL while feeding pages into the page allocator for | |
4644 | * the first time, i.e. during boot or memory hotplug; | |
4645 | * or when mem_cgroup_disabled(). | |
4646 | */ | |
f212ad7c DN |
4647 | if (likely(pc) && PageCgroupUsed(pc)) |
4648 | return pc; | |
4649 | return NULL; | |
4650 | } | |
4651 | ||
4652 | bool mem_cgroup_bad_page_check(struct page *page) | |
4653 | { | |
4654 | if (mem_cgroup_disabled()) | |
4655 | return false; | |
4656 | ||
4657 | return lookup_page_cgroup_used(page) != NULL; | |
4658 | } | |
4659 | ||
4660 | void mem_cgroup_print_bad_page(struct page *page) | |
4661 | { | |
4662 | struct page_cgroup *pc; | |
4663 | ||
4664 | pc = lookup_page_cgroup_used(page); | |
4665 | if (pc) { | |
d045197f AM |
4666 | pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n", |
4667 | pc, pc->flags, pc->mem_cgroup); | |
f212ad7c DN |
4668 | } |
4669 | } | |
4670 | #endif | |
4671 | ||
d38d2a75 | 4672 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 4673 | unsigned long long val) |
628f4235 | 4674 | { |
81d39c20 | 4675 | int retry_count; |
3c11ecf4 | 4676 | u64 memswlimit, memlimit; |
628f4235 | 4677 | int ret = 0; |
81d39c20 KH |
4678 | int children = mem_cgroup_count_children(memcg); |
4679 | u64 curusage, oldusage; | |
3c11ecf4 | 4680 | int enlarge; |
81d39c20 KH |
4681 | |
4682 | /* | |
4683 | * For keeping hierarchical_reclaim simple, how long we should retry | |
4684 | * is depends on callers. We set our retry-count to be function | |
4685 | * of # of children which we should visit in this loop. | |
4686 | */ | |
4687 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
4688 | ||
4689 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 4690 | |
3c11ecf4 | 4691 | enlarge = 0; |
8c7c6e34 | 4692 | while (retry_count) { |
628f4235 KH |
4693 | if (signal_pending(current)) { |
4694 | ret = -EINTR; | |
4695 | break; | |
4696 | } | |
8c7c6e34 KH |
4697 | /* |
4698 | * Rather than hide all in some function, I do this in | |
4699 | * open coded manner. You see what this really does. | |
aaad153e | 4700 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
4701 | */ |
4702 | mutex_lock(&set_limit_mutex); | |
4703 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
4704 | if (memswlimit < val) { | |
4705 | ret = -EINVAL; | |
4706 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
4707 | break; |
4708 | } | |
3c11ecf4 KH |
4709 | |
4710 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4711 | if (memlimit < val) | |
4712 | enlarge = 1; | |
4713 | ||
8c7c6e34 | 4714 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
4715 | if (!ret) { |
4716 | if (memswlimit == val) | |
4717 | memcg->memsw_is_minimum = true; | |
4718 | else | |
4719 | memcg->memsw_is_minimum = false; | |
4720 | } | |
8c7c6e34 KH |
4721 | mutex_unlock(&set_limit_mutex); |
4722 | ||
4723 | if (!ret) | |
4724 | break; | |
4725 | ||
5660048c JW |
4726 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
4727 | MEM_CGROUP_RECLAIM_SHRINK); | |
81d39c20 KH |
4728 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
4729 | /* Usage is reduced ? */ | |
f894ffa8 | 4730 | if (curusage >= oldusage) |
81d39c20 KH |
4731 | retry_count--; |
4732 | else | |
4733 | oldusage = curusage; | |
8c7c6e34 | 4734 | } |
3c11ecf4 KH |
4735 | if (!ret && enlarge) |
4736 | memcg_oom_recover(memcg); | |
14797e23 | 4737 | |
8c7c6e34 KH |
4738 | return ret; |
4739 | } | |
4740 | ||
338c8431 LZ |
4741 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
4742 | unsigned long long val) | |
8c7c6e34 | 4743 | { |
81d39c20 | 4744 | int retry_count; |
3c11ecf4 | 4745 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
4746 | int children = mem_cgroup_count_children(memcg); |
4747 | int ret = -EBUSY; | |
3c11ecf4 | 4748 | int enlarge = 0; |
8c7c6e34 | 4749 | |
81d39c20 | 4750 | /* see mem_cgroup_resize_res_limit */ |
f894ffa8 | 4751 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; |
81d39c20 | 4752 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
8c7c6e34 KH |
4753 | while (retry_count) { |
4754 | if (signal_pending(current)) { | |
4755 | ret = -EINTR; | |
4756 | break; | |
4757 | } | |
4758 | /* | |
4759 | * Rather than hide all in some function, I do this in | |
4760 | * open coded manner. You see what this really does. | |
aaad153e | 4761 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
4762 | */ |
4763 | mutex_lock(&set_limit_mutex); | |
4764 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4765 | if (memlimit > val) { | |
4766 | ret = -EINVAL; | |
4767 | mutex_unlock(&set_limit_mutex); | |
4768 | break; | |
4769 | } | |
3c11ecf4 KH |
4770 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
4771 | if (memswlimit < val) | |
4772 | enlarge = 1; | |
8c7c6e34 | 4773 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
4774 | if (!ret) { |
4775 | if (memlimit == val) | |
4776 | memcg->memsw_is_minimum = true; | |
4777 | else | |
4778 | memcg->memsw_is_minimum = false; | |
4779 | } | |
8c7c6e34 KH |
4780 | mutex_unlock(&set_limit_mutex); |
4781 | ||
4782 | if (!ret) | |
4783 | break; | |
4784 | ||
5660048c JW |
4785 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
4786 | MEM_CGROUP_RECLAIM_NOSWAP | | |
4787 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 4788 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 4789 | /* Usage is reduced ? */ |
8c7c6e34 | 4790 | if (curusage >= oldusage) |
628f4235 | 4791 | retry_count--; |
81d39c20 KH |
4792 | else |
4793 | oldusage = curusage; | |
628f4235 | 4794 | } |
3c11ecf4 KH |
4795 | if (!ret && enlarge) |
4796 | memcg_oom_recover(memcg); | |
628f4235 KH |
4797 | return ret; |
4798 | } | |
4799 | ||
0608f43d AM |
4800 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
4801 | gfp_t gfp_mask, | |
4802 | unsigned long *total_scanned) | |
4803 | { | |
4804 | unsigned long nr_reclaimed = 0; | |
4805 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
4806 | unsigned long reclaimed; | |
4807 | int loop = 0; | |
4808 | struct mem_cgroup_tree_per_zone *mctz; | |
4809 | unsigned long long excess; | |
4810 | unsigned long nr_scanned; | |
4811 | ||
4812 | if (order > 0) | |
4813 | return 0; | |
4814 | ||
4815 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
4816 | /* | |
4817 | * This loop can run a while, specially if mem_cgroup's continuously | |
4818 | * keep exceeding their soft limit and putting the system under | |
4819 | * pressure | |
4820 | */ | |
4821 | do { | |
4822 | if (next_mz) | |
4823 | mz = next_mz; | |
4824 | else | |
4825 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
4826 | if (!mz) | |
4827 | break; | |
4828 | ||
4829 | nr_scanned = 0; | |
4830 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
4831 | gfp_mask, &nr_scanned); | |
4832 | nr_reclaimed += reclaimed; | |
4833 | *total_scanned += nr_scanned; | |
4834 | spin_lock(&mctz->lock); | |
4835 | ||
4836 | /* | |
4837 | * If we failed to reclaim anything from this memory cgroup | |
4838 | * it is time to move on to the next cgroup | |
4839 | */ | |
4840 | next_mz = NULL; | |
4841 | if (!reclaimed) { | |
4842 | do { | |
4843 | /* | |
4844 | * Loop until we find yet another one. | |
4845 | * | |
4846 | * By the time we get the soft_limit lock | |
4847 | * again, someone might have aded the | |
4848 | * group back on the RB tree. Iterate to | |
4849 | * make sure we get a different mem. | |
4850 | * mem_cgroup_largest_soft_limit_node returns | |
4851 | * NULL if no other cgroup is present on | |
4852 | * the tree | |
4853 | */ | |
4854 | next_mz = | |
4855 | __mem_cgroup_largest_soft_limit_node(mctz); | |
4856 | if (next_mz == mz) | |
4857 | css_put(&next_mz->memcg->css); | |
4858 | else /* next_mz == NULL or other memcg */ | |
4859 | break; | |
4860 | } while (1); | |
4861 | } | |
4862 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); | |
4863 | excess = res_counter_soft_limit_excess(&mz->memcg->res); | |
4864 | /* | |
4865 | * One school of thought says that we should not add | |
4866 | * back the node to the tree if reclaim returns 0. | |
4867 | * But our reclaim could return 0, simply because due | |
4868 | * to priority we are exposing a smaller subset of | |
4869 | * memory to reclaim from. Consider this as a longer | |
4870 | * term TODO. | |
4871 | */ | |
4872 | /* If excess == 0, no tree ops */ | |
4873 | __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess); | |
4874 | spin_unlock(&mctz->lock); | |
4875 | css_put(&mz->memcg->css); | |
4876 | loop++; | |
4877 | /* | |
4878 | * Could not reclaim anything and there are no more | |
4879 | * mem cgroups to try or we seem to be looping without | |
4880 | * reclaiming anything. | |
4881 | */ | |
4882 | if (!nr_reclaimed && | |
4883 | (next_mz == NULL || | |
4884 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
4885 | break; | |
4886 | } while (!nr_reclaimed); | |
4887 | if (next_mz) | |
4888 | css_put(&next_mz->memcg->css); | |
4889 | return nr_reclaimed; | |
4890 | } | |
4891 | ||
2ef37d3f MH |
4892 | /** |
4893 | * mem_cgroup_force_empty_list - clears LRU of a group | |
4894 | * @memcg: group to clear | |
4895 | * @node: NUMA node | |
4896 | * @zid: zone id | |
4897 | * @lru: lru to to clear | |
4898 | * | |
3c935d18 | 4899 | * Traverse a specified page_cgroup list and try to drop them all. This doesn't |
2ef37d3f MH |
4900 | * reclaim the pages page themselves - pages are moved to the parent (or root) |
4901 | * group. | |
cc847582 | 4902 | */ |
2ef37d3f | 4903 | static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
08e552c6 | 4904 | int node, int zid, enum lru_list lru) |
cc847582 | 4905 | { |
bea8c150 | 4906 | struct lruvec *lruvec; |
2ef37d3f | 4907 | unsigned long flags; |
072c56c1 | 4908 | struct list_head *list; |
925b7673 JW |
4909 | struct page *busy; |
4910 | struct zone *zone; | |
072c56c1 | 4911 | |
08e552c6 | 4912 | zone = &NODE_DATA(node)->node_zones[zid]; |
bea8c150 HD |
4913 | lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
4914 | list = &lruvec->lists[lru]; | |
cc847582 | 4915 | |
f817ed48 | 4916 | busy = NULL; |
2ef37d3f | 4917 | do { |
925b7673 | 4918 | struct page_cgroup *pc; |
5564e88b JW |
4919 | struct page *page; |
4920 | ||
08e552c6 | 4921 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 4922 | if (list_empty(list)) { |
08e552c6 | 4923 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 4924 | break; |
f817ed48 | 4925 | } |
925b7673 JW |
4926 | page = list_entry(list->prev, struct page, lru); |
4927 | if (busy == page) { | |
4928 | list_move(&page->lru, list); | |
648bcc77 | 4929 | busy = NULL; |
08e552c6 | 4930 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
4931 | continue; |
4932 | } | |
08e552c6 | 4933 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 4934 | |
925b7673 | 4935 | pc = lookup_page_cgroup(page); |
5564e88b | 4936 | |
3c935d18 | 4937 | if (mem_cgroup_move_parent(page, pc, memcg)) { |
f817ed48 | 4938 | /* found lock contention or "pc" is obsolete. */ |
925b7673 | 4939 | busy = page; |
f817ed48 KH |
4940 | cond_resched(); |
4941 | } else | |
4942 | busy = NULL; | |
2ef37d3f | 4943 | } while (!list_empty(list)); |
cc847582 KH |
4944 | } |
4945 | ||
4946 | /* | |
c26251f9 MH |
4947 | * make mem_cgroup's charge to be 0 if there is no task by moving |
4948 | * all the charges and pages to the parent. | |
cc847582 | 4949 | * This enables deleting this mem_cgroup. |
c26251f9 MH |
4950 | * |
4951 | * Caller is responsible for holding css reference on the memcg. | |
cc847582 | 4952 | */ |
ab5196c2 | 4953 | static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg) |
cc847582 | 4954 | { |
c26251f9 | 4955 | int node, zid; |
bea207c8 | 4956 | u64 usage; |
f817ed48 | 4957 | |
fce66477 | 4958 | do { |
52d4b9ac KH |
4959 | /* This is for making all *used* pages to be on LRU. */ |
4960 | lru_add_drain_all(); | |
c0ff4b85 | 4961 | drain_all_stock_sync(memcg); |
c0ff4b85 | 4962 | mem_cgroup_start_move(memcg); |
31aaea4a | 4963 | for_each_node_state(node, N_MEMORY) { |
2ef37d3f | 4964 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
f156ab93 HD |
4965 | enum lru_list lru; |
4966 | for_each_lru(lru) { | |
2ef37d3f | 4967 | mem_cgroup_force_empty_list(memcg, |
f156ab93 | 4968 | node, zid, lru); |
f817ed48 | 4969 | } |
1ecaab2b | 4970 | } |
f817ed48 | 4971 | } |
c0ff4b85 R |
4972 | mem_cgroup_end_move(memcg); |
4973 | memcg_oom_recover(memcg); | |
52d4b9ac | 4974 | cond_resched(); |
f817ed48 | 4975 | |
2ef37d3f | 4976 | /* |
bea207c8 GC |
4977 | * Kernel memory may not necessarily be trackable to a specific |
4978 | * process. So they are not migrated, and therefore we can't | |
4979 | * expect their value to drop to 0 here. | |
4980 | * Having res filled up with kmem only is enough. | |
4981 | * | |
2ef37d3f MH |
4982 | * This is a safety check because mem_cgroup_force_empty_list |
4983 | * could have raced with mem_cgroup_replace_page_cache callers | |
4984 | * so the lru seemed empty but the page could have been added | |
4985 | * right after the check. RES_USAGE should be safe as we always | |
4986 | * charge before adding to the LRU. | |
4987 | */ | |
bea207c8 GC |
4988 | usage = res_counter_read_u64(&memcg->res, RES_USAGE) - |
4989 | res_counter_read_u64(&memcg->kmem, RES_USAGE); | |
4990 | } while (usage > 0); | |
c26251f9 MH |
4991 | } |
4992 | ||
b5f99b53 GC |
4993 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
4994 | { | |
696ac172 JW |
4995 | lockdep_assert_held(&memcg_create_mutex); |
4996 | /* | |
4997 | * The lock does not prevent addition or deletion to the list | |
4998 | * of children, but it prevents a new child from being | |
4999 | * initialized based on this parent in css_online(), so it's | |
5000 | * enough to decide whether hierarchically inherited | |
5001 | * attributes can still be changed or not. | |
5002 | */ | |
5003 | return memcg->use_hierarchy && | |
5004 | !list_empty(&memcg->css.cgroup->children); | |
b5f99b53 GC |
5005 | } |
5006 | ||
c26251f9 MH |
5007 | /* |
5008 | * Reclaims as many pages from the given memcg as possible and moves | |
5009 | * the rest to the parent. | |
5010 | * | |
5011 | * Caller is responsible for holding css reference for memcg. | |
5012 | */ | |
5013 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
5014 | { | |
5015 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
5016 | struct cgroup *cgrp = memcg->css.cgroup; | |
f817ed48 | 5017 | |
c1e862c1 | 5018 | /* returns EBUSY if there is a task or if we come here twice. */ |
c26251f9 MH |
5019 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
5020 | return -EBUSY; | |
5021 | ||
c1e862c1 KH |
5022 | /* we call try-to-free pages for make this cgroup empty */ |
5023 | lru_add_drain_all(); | |
f817ed48 | 5024 | /* try to free all pages in this cgroup */ |
569530fb | 5025 | while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) { |
f817ed48 | 5026 | int progress; |
c1e862c1 | 5027 | |
c26251f9 MH |
5028 | if (signal_pending(current)) |
5029 | return -EINTR; | |
5030 | ||
c0ff4b85 | 5031 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, |
185efc0f | 5032 | false); |
c1e862c1 | 5033 | if (!progress) { |
f817ed48 | 5034 | nr_retries--; |
c1e862c1 | 5035 | /* maybe some writeback is necessary */ |
8aa7e847 | 5036 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 5037 | } |
f817ed48 KH |
5038 | |
5039 | } | |
08e552c6 | 5040 | lru_add_drain(); |
ab5196c2 MH |
5041 | mem_cgroup_reparent_charges(memcg); |
5042 | ||
5043 | return 0; | |
cc847582 KH |
5044 | } |
5045 | ||
182446d0 TH |
5046 | static int mem_cgroup_force_empty_write(struct cgroup_subsys_state *css, |
5047 | unsigned int event) | |
c1e862c1 | 5048 | { |
182446d0 | 5049 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c26251f9 | 5050 | |
d8423011 MH |
5051 | if (mem_cgroup_is_root(memcg)) |
5052 | return -EINVAL; | |
c33bd835 | 5053 | return mem_cgroup_force_empty(memcg); |
c1e862c1 KH |
5054 | } |
5055 | ||
182446d0 TH |
5056 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
5057 | struct cftype *cft) | |
18f59ea7 | 5058 | { |
182446d0 | 5059 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
5060 | } |
5061 | ||
182446d0 TH |
5062 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
5063 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
5064 | { |
5065 | int retval = 0; | |
182446d0 | 5066 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
63876986 | 5067 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(css_parent(&memcg->css)); |
18f59ea7 | 5068 | |
0999821b | 5069 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
5070 | |
5071 | if (memcg->use_hierarchy == val) | |
5072 | goto out; | |
5073 | ||
18f59ea7 | 5074 | /* |
af901ca1 | 5075 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
5076 | * in the child subtrees. If it is unset, then the change can |
5077 | * occur, provided the current cgroup has no children. | |
5078 | * | |
5079 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
5080 | * set if there are no children. | |
5081 | */ | |
c0ff4b85 | 5082 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 5083 | (val == 1 || val == 0)) { |
696ac172 | 5084 | if (list_empty(&memcg->css.cgroup->children)) |
c0ff4b85 | 5085 | memcg->use_hierarchy = val; |
18f59ea7 BS |
5086 | else |
5087 | retval = -EBUSY; | |
5088 | } else | |
5089 | retval = -EINVAL; | |
567fb435 GC |
5090 | |
5091 | out: | |
0999821b | 5092 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
5093 | |
5094 | return retval; | |
5095 | } | |
5096 | ||
0c3e73e8 | 5097 | |
c0ff4b85 | 5098 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg, |
7a159cc9 | 5099 | enum mem_cgroup_stat_index idx) |
0c3e73e8 | 5100 | { |
7d74b06f | 5101 | struct mem_cgroup *iter; |
7a159cc9 | 5102 | long val = 0; |
0c3e73e8 | 5103 | |
7a159cc9 | 5104 | /* Per-cpu values can be negative, use a signed accumulator */ |
c0ff4b85 | 5105 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f KH |
5106 | val += mem_cgroup_read_stat(iter, idx); |
5107 | ||
5108 | if (val < 0) /* race ? */ | |
5109 | val = 0; | |
5110 | return val; | |
0c3e73e8 BS |
5111 | } |
5112 | ||
c0ff4b85 | 5113 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
104f3928 | 5114 | { |
7d74b06f | 5115 | u64 val; |
104f3928 | 5116 | |
c0ff4b85 | 5117 | if (!mem_cgroup_is_root(memcg)) { |
104f3928 | 5118 | if (!swap) |
65c64ce8 | 5119 | return res_counter_read_u64(&memcg->res, RES_USAGE); |
104f3928 | 5120 | else |
65c64ce8 | 5121 | return res_counter_read_u64(&memcg->memsw, RES_USAGE); |
104f3928 KS |
5122 | } |
5123 | ||
b070e65c DR |
5124 | /* |
5125 | * Transparent hugepages are still accounted for in MEM_CGROUP_STAT_RSS | |
5126 | * as well as in MEM_CGROUP_STAT_RSS_HUGE. | |
5127 | */ | |
c0ff4b85 R |
5128 | val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE); |
5129 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS); | |
104f3928 | 5130 | |
7d74b06f | 5131 | if (swap) |
bff6bb83 | 5132 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP); |
104f3928 KS |
5133 | |
5134 | return val << PAGE_SHIFT; | |
5135 | } | |
5136 | ||
182446d0 TH |
5137 | static ssize_t mem_cgroup_read(struct cgroup_subsys_state *css, |
5138 | struct cftype *cft, struct file *file, | |
5139 | char __user *buf, size_t nbytes, loff_t *ppos) | |
8cdea7c0 | 5140 | { |
182446d0 | 5141 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
af36f906 | 5142 | char str[64]; |
104f3928 | 5143 | u64 val; |
86ae53e1 GC |
5144 | int name, len; |
5145 | enum res_type type; | |
8c7c6e34 KH |
5146 | |
5147 | type = MEMFILE_TYPE(cft->private); | |
5148 | name = MEMFILE_ATTR(cft->private); | |
af36f906 | 5149 | |
8c7c6e34 KH |
5150 | switch (type) { |
5151 | case _MEM: | |
104f3928 | 5152 | if (name == RES_USAGE) |
c0ff4b85 | 5153 | val = mem_cgroup_usage(memcg, false); |
104f3928 | 5154 | else |
c0ff4b85 | 5155 | val = res_counter_read_u64(&memcg->res, name); |
8c7c6e34 KH |
5156 | break; |
5157 | case _MEMSWAP: | |
104f3928 | 5158 | if (name == RES_USAGE) |
c0ff4b85 | 5159 | val = mem_cgroup_usage(memcg, true); |
104f3928 | 5160 | else |
c0ff4b85 | 5161 | val = res_counter_read_u64(&memcg->memsw, name); |
8c7c6e34 | 5162 | break; |
510fc4e1 GC |
5163 | case _KMEM: |
5164 | val = res_counter_read_u64(&memcg->kmem, name); | |
5165 | break; | |
8c7c6e34 KH |
5166 | default: |
5167 | BUG(); | |
8c7c6e34 | 5168 | } |
af36f906 TH |
5169 | |
5170 | len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val); | |
5171 | return simple_read_from_buffer(buf, nbytes, ppos, str, len); | |
8cdea7c0 | 5172 | } |
510fc4e1 | 5173 | |
182446d0 | 5174 | static int memcg_update_kmem_limit(struct cgroup_subsys_state *css, u64 val) |
510fc4e1 GC |
5175 | { |
5176 | int ret = -EINVAL; | |
5177 | #ifdef CONFIG_MEMCG_KMEM | |
182446d0 | 5178 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
510fc4e1 GC |
5179 | /* |
5180 | * For simplicity, we won't allow this to be disabled. It also can't | |
5181 | * be changed if the cgroup has children already, or if tasks had | |
5182 | * already joined. | |
5183 | * | |
5184 | * If tasks join before we set the limit, a person looking at | |
5185 | * kmem.usage_in_bytes will have no way to determine when it took | |
5186 | * place, which makes the value quite meaningless. | |
5187 | * | |
5188 | * After it first became limited, changes in the value of the limit are | |
5189 | * of course permitted. | |
510fc4e1 | 5190 | */ |
0999821b | 5191 | mutex_lock(&memcg_create_mutex); |
510fc4e1 | 5192 | mutex_lock(&set_limit_mutex); |
6de5a8bf | 5193 | if (!memcg->kmem_account_flags && val != RES_COUNTER_MAX) { |
182446d0 | 5194 | if (cgroup_task_count(css->cgroup) || memcg_has_children(memcg)) { |
510fc4e1 GC |
5195 | ret = -EBUSY; |
5196 | goto out; | |
5197 | } | |
5198 | ret = res_counter_set_limit(&memcg->kmem, val); | |
5199 | VM_BUG_ON(ret); | |
5200 | ||
55007d84 GC |
5201 | ret = memcg_update_cache_sizes(memcg); |
5202 | if (ret) { | |
6de5a8bf | 5203 | res_counter_set_limit(&memcg->kmem, RES_COUNTER_MAX); |
55007d84 GC |
5204 | goto out; |
5205 | } | |
692e89ab GC |
5206 | static_key_slow_inc(&memcg_kmem_enabled_key); |
5207 | /* | |
5208 | * setting the active bit after the inc will guarantee no one | |
5209 | * starts accounting before all call sites are patched | |
5210 | */ | |
5211 | memcg_kmem_set_active(memcg); | |
510fc4e1 GC |
5212 | } else |
5213 | ret = res_counter_set_limit(&memcg->kmem, val); | |
5214 | out: | |
5215 | mutex_unlock(&set_limit_mutex); | |
0999821b | 5216 | mutex_unlock(&memcg_create_mutex); |
510fc4e1 GC |
5217 | #endif |
5218 | return ret; | |
5219 | } | |
5220 | ||
6d043990 | 5221 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 | 5222 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 5223 | { |
55007d84 | 5224 | int ret = 0; |
510fc4e1 GC |
5225 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
5226 | if (!parent) | |
55007d84 GC |
5227 | goto out; |
5228 | ||
510fc4e1 | 5229 | memcg->kmem_account_flags = parent->kmem_account_flags; |
a8964b9b GC |
5230 | /* |
5231 | * When that happen, we need to disable the static branch only on those | |
5232 | * memcgs that enabled it. To achieve this, we would be forced to | |
5233 | * complicate the code by keeping track of which memcgs were the ones | |
5234 | * that actually enabled limits, and which ones got it from its | |
5235 | * parents. | |
5236 | * | |
5237 | * It is a lot simpler just to do static_key_slow_inc() on every child | |
5238 | * that is accounted. | |
5239 | */ | |
55007d84 GC |
5240 | if (!memcg_kmem_is_active(memcg)) |
5241 | goto out; | |
5242 | ||
5243 | /* | |
10d5ebf4 LZ |
5244 | * __mem_cgroup_free() will issue static_key_slow_dec() because this |
5245 | * memcg is active already. If the later initialization fails then the | |
5246 | * cgroup core triggers the cleanup so we do not have to do it here. | |
55007d84 | 5247 | */ |
55007d84 GC |
5248 | static_key_slow_inc(&memcg_kmem_enabled_key); |
5249 | ||
5250 | mutex_lock(&set_limit_mutex); | |
425c598d | 5251 | memcg_stop_kmem_account(); |
55007d84 | 5252 | ret = memcg_update_cache_sizes(memcg); |
425c598d | 5253 | memcg_resume_kmem_account(); |
55007d84 | 5254 | mutex_unlock(&set_limit_mutex); |
55007d84 GC |
5255 | out: |
5256 | return ret; | |
510fc4e1 | 5257 | } |
6d043990 | 5258 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 5259 | |
628f4235 KH |
5260 | /* |
5261 | * The user of this function is... | |
5262 | * RES_LIMIT. | |
5263 | */ | |
182446d0 | 5264 | static int mem_cgroup_write(struct cgroup_subsys_state *css, struct cftype *cft, |
856c13aa | 5265 | const char *buffer) |
8cdea7c0 | 5266 | { |
182446d0 | 5267 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
86ae53e1 GC |
5268 | enum res_type type; |
5269 | int name; | |
628f4235 KH |
5270 | unsigned long long val; |
5271 | int ret; | |
5272 | ||
8c7c6e34 KH |
5273 | type = MEMFILE_TYPE(cft->private); |
5274 | name = MEMFILE_ATTR(cft->private); | |
af36f906 | 5275 | |
8c7c6e34 | 5276 | switch (name) { |
628f4235 | 5277 | case RES_LIMIT: |
4b3bde4c BS |
5278 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
5279 | ret = -EINVAL; | |
5280 | break; | |
5281 | } | |
628f4235 KH |
5282 | /* This function does all necessary parse...reuse it */ |
5283 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
5284 | if (ret) |
5285 | break; | |
5286 | if (type == _MEM) | |
628f4235 | 5287 | ret = mem_cgroup_resize_limit(memcg, val); |
510fc4e1 | 5288 | else if (type == _MEMSWAP) |
8c7c6e34 | 5289 | ret = mem_cgroup_resize_memsw_limit(memcg, val); |
510fc4e1 | 5290 | else if (type == _KMEM) |
182446d0 | 5291 | ret = memcg_update_kmem_limit(css, val); |
510fc4e1 GC |
5292 | else |
5293 | return -EINVAL; | |
628f4235 | 5294 | break; |
296c81d8 BS |
5295 | case RES_SOFT_LIMIT: |
5296 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
5297 | if (ret) | |
5298 | break; | |
5299 | /* | |
5300 | * For memsw, soft limits are hard to implement in terms | |
5301 | * of semantics, for now, we support soft limits for | |
5302 | * control without swap | |
5303 | */ | |
5304 | if (type == _MEM) | |
5305 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
5306 | else | |
5307 | ret = -EINVAL; | |
5308 | break; | |
628f4235 KH |
5309 | default: |
5310 | ret = -EINVAL; /* should be BUG() ? */ | |
5311 | break; | |
5312 | } | |
5313 | return ret; | |
8cdea7c0 BS |
5314 | } |
5315 | ||
fee7b548 KH |
5316 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
5317 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
5318 | { | |
fee7b548 KH |
5319 | unsigned long long min_limit, min_memsw_limit, tmp; |
5320 | ||
5321 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
5322 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
fee7b548 KH |
5323 | if (!memcg->use_hierarchy) |
5324 | goto out; | |
5325 | ||
63876986 TH |
5326 | while (css_parent(&memcg->css)) { |
5327 | memcg = mem_cgroup_from_css(css_parent(&memcg->css)); | |
fee7b548 KH |
5328 | if (!memcg->use_hierarchy) |
5329 | break; | |
5330 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
5331 | min_limit = min(min_limit, tmp); | |
5332 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
5333 | min_memsw_limit = min(min_memsw_limit, tmp); | |
5334 | } | |
5335 | out: | |
5336 | *mem_limit = min_limit; | |
5337 | *memsw_limit = min_memsw_limit; | |
fee7b548 KH |
5338 | } |
5339 | ||
182446d0 | 5340 | static int mem_cgroup_reset(struct cgroup_subsys_state *css, unsigned int event) |
c84872e1 | 5341 | { |
182446d0 | 5342 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
86ae53e1 GC |
5343 | int name; |
5344 | enum res_type type; | |
c84872e1 | 5345 | |
8c7c6e34 KH |
5346 | type = MEMFILE_TYPE(event); |
5347 | name = MEMFILE_ATTR(event); | |
af36f906 | 5348 | |
8c7c6e34 | 5349 | switch (name) { |
29f2a4da | 5350 | case RES_MAX_USAGE: |
8c7c6e34 | 5351 | if (type == _MEM) |
c0ff4b85 | 5352 | res_counter_reset_max(&memcg->res); |
510fc4e1 | 5353 | else if (type == _MEMSWAP) |
c0ff4b85 | 5354 | res_counter_reset_max(&memcg->memsw); |
510fc4e1 GC |
5355 | else if (type == _KMEM) |
5356 | res_counter_reset_max(&memcg->kmem); | |
5357 | else | |
5358 | return -EINVAL; | |
29f2a4da PE |
5359 | break; |
5360 | case RES_FAILCNT: | |
8c7c6e34 | 5361 | if (type == _MEM) |
c0ff4b85 | 5362 | res_counter_reset_failcnt(&memcg->res); |
510fc4e1 | 5363 | else if (type == _MEMSWAP) |
c0ff4b85 | 5364 | res_counter_reset_failcnt(&memcg->memsw); |
510fc4e1 GC |
5365 | else if (type == _KMEM) |
5366 | res_counter_reset_failcnt(&memcg->kmem); | |
5367 | else | |
5368 | return -EINVAL; | |
29f2a4da PE |
5369 | break; |
5370 | } | |
f64c3f54 | 5371 | |
85cc59db | 5372 | return 0; |
c84872e1 PE |
5373 | } |
5374 | ||
182446d0 | 5375 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
5376 | struct cftype *cft) |
5377 | { | |
182446d0 | 5378 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
5379 | } |
5380 | ||
02491447 | 5381 | #ifdef CONFIG_MMU |
182446d0 | 5382 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
5383 | struct cftype *cft, u64 val) |
5384 | { | |
182446d0 | 5385 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 DN |
5386 | |
5387 | if (val >= (1 << NR_MOVE_TYPE)) | |
5388 | return -EINVAL; | |
ee5e8472 | 5389 | |
7dc74be0 | 5390 | /* |
ee5e8472 GC |
5391 | * No kind of locking is needed in here, because ->can_attach() will |
5392 | * check this value once in the beginning of the process, and then carry | |
5393 | * on with stale data. This means that changes to this value will only | |
5394 | * affect task migrations starting after the change. | |
7dc74be0 | 5395 | */ |
c0ff4b85 | 5396 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
5397 | return 0; |
5398 | } | |
02491447 | 5399 | #else |
182446d0 | 5400 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
5401 | struct cftype *cft, u64 val) |
5402 | { | |
5403 | return -ENOSYS; | |
5404 | } | |
5405 | #endif | |
7dc74be0 | 5406 | |
406eb0c9 | 5407 | #ifdef CONFIG_NUMA |
182446d0 TH |
5408 | static int memcg_numa_stat_show(struct cgroup_subsys_state *css, |
5409 | struct cftype *cft, struct seq_file *m) | |
406eb0c9 YH |
5410 | { |
5411 | int nid; | |
5412 | unsigned long total_nr, file_nr, anon_nr, unevictable_nr; | |
5413 | unsigned long node_nr; | |
182446d0 | 5414 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
406eb0c9 | 5415 | |
d79154bb | 5416 | total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL); |
406eb0c9 | 5417 | seq_printf(m, "total=%lu", total_nr); |
31aaea4a | 5418 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5419 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL); |
406eb0c9 YH |
5420 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5421 | } | |
5422 | seq_putc(m, '\n'); | |
5423 | ||
d79154bb | 5424 | file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE); |
406eb0c9 | 5425 | seq_printf(m, "file=%lu", file_nr); |
31aaea4a | 5426 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5427 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 5428 | LRU_ALL_FILE); |
406eb0c9 YH |
5429 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5430 | } | |
5431 | seq_putc(m, '\n'); | |
5432 | ||
d79154bb | 5433 | anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON); |
406eb0c9 | 5434 | seq_printf(m, "anon=%lu", anon_nr); |
31aaea4a | 5435 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5436 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 5437 | LRU_ALL_ANON); |
406eb0c9 YH |
5438 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5439 | } | |
5440 | seq_putc(m, '\n'); | |
5441 | ||
d79154bb | 5442 | unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE)); |
406eb0c9 | 5443 | seq_printf(m, "unevictable=%lu", unevictable_nr); |
31aaea4a | 5444 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5445 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 5446 | BIT(LRU_UNEVICTABLE)); |
406eb0c9 YH |
5447 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5448 | } | |
5449 | seq_putc(m, '\n'); | |
5450 | return 0; | |
5451 | } | |
5452 | #endif /* CONFIG_NUMA */ | |
5453 | ||
af7c4b0e JW |
5454 | static inline void mem_cgroup_lru_names_not_uptodate(void) |
5455 | { | |
5456 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); | |
5457 | } | |
5458 | ||
182446d0 | 5459 | static int memcg_stat_show(struct cgroup_subsys_state *css, struct cftype *cft, |
78ccf5b5 | 5460 | struct seq_file *m) |
d2ceb9b7 | 5461 | { |
182446d0 | 5462 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
af7c4b0e JW |
5463 | struct mem_cgroup *mi; |
5464 | unsigned int i; | |
406eb0c9 | 5465 | |
af7c4b0e | 5466 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 5467 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 5468 | continue; |
af7c4b0e JW |
5469 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
5470 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 5471 | } |
7b854121 | 5472 | |
af7c4b0e JW |
5473 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
5474 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
5475 | mem_cgroup_read_events(memcg, i)); | |
5476 | ||
5477 | for (i = 0; i < NR_LRU_LISTS; i++) | |
5478 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
5479 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
5480 | ||
14067bb3 | 5481 | /* Hierarchical information */ |
fee7b548 KH |
5482 | { |
5483 | unsigned long long limit, memsw_limit; | |
d79154bb | 5484 | memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit); |
78ccf5b5 | 5485 | seq_printf(m, "hierarchical_memory_limit %llu\n", limit); |
fee7b548 | 5486 | if (do_swap_account) |
78ccf5b5 JW |
5487 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
5488 | memsw_limit); | |
fee7b548 | 5489 | } |
7f016ee8 | 5490 | |
af7c4b0e JW |
5491 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
5492 | long long val = 0; | |
5493 | ||
bff6bb83 | 5494 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 5495 | continue; |
af7c4b0e JW |
5496 | for_each_mem_cgroup_tree(mi, memcg) |
5497 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
5498 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
5499 | } | |
5500 | ||
5501 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
5502 | unsigned long long val = 0; | |
5503 | ||
5504 | for_each_mem_cgroup_tree(mi, memcg) | |
5505 | val += mem_cgroup_read_events(mi, i); | |
5506 | seq_printf(m, "total_%s %llu\n", | |
5507 | mem_cgroup_events_names[i], val); | |
5508 | } | |
5509 | ||
5510 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
5511 | unsigned long long val = 0; | |
5512 | ||
5513 | for_each_mem_cgroup_tree(mi, memcg) | |
5514 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
5515 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 5516 | } |
14067bb3 | 5517 | |
7f016ee8 | 5518 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
5519 | { |
5520 | int nid, zid; | |
5521 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 5522 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
5523 | unsigned long recent_rotated[2] = {0, 0}; |
5524 | unsigned long recent_scanned[2] = {0, 0}; | |
5525 | ||
5526 | for_each_online_node(nid) | |
5527 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
d79154bb | 5528 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
89abfab1 | 5529 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 5530 | |
89abfab1 HD |
5531 | recent_rotated[0] += rstat->recent_rotated[0]; |
5532 | recent_rotated[1] += rstat->recent_rotated[1]; | |
5533 | recent_scanned[0] += rstat->recent_scanned[0]; | |
5534 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 5535 | } |
78ccf5b5 JW |
5536 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
5537 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
5538 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
5539 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
5540 | } |
5541 | #endif | |
5542 | ||
d2ceb9b7 KH |
5543 | return 0; |
5544 | } | |
5545 | ||
182446d0 TH |
5546 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
5547 | struct cftype *cft) | |
a7885eb8 | 5548 | { |
182446d0 | 5549 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 5550 | |
1f4c025b | 5551 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
5552 | } |
5553 | ||
182446d0 TH |
5554 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
5555 | struct cftype *cft, u64 val) | |
a7885eb8 | 5556 | { |
182446d0 | 5557 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
63876986 | 5558 | struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css)); |
a7885eb8 | 5559 | |
63876986 | 5560 | if (val > 100 || !parent) |
a7885eb8 KM |
5561 | return -EINVAL; |
5562 | ||
0999821b | 5563 | mutex_lock(&memcg_create_mutex); |
068b38c1 | 5564 | |
a7885eb8 | 5565 | /* If under hierarchy, only empty-root can set this value */ |
b5f99b53 | 5566 | if ((parent->use_hierarchy) || memcg_has_children(memcg)) { |
0999821b | 5567 | mutex_unlock(&memcg_create_mutex); |
a7885eb8 | 5568 | return -EINVAL; |
068b38c1 | 5569 | } |
a7885eb8 | 5570 | |
a7885eb8 | 5571 | memcg->swappiness = val; |
a7885eb8 | 5572 | |
0999821b | 5573 | mutex_unlock(&memcg_create_mutex); |
068b38c1 | 5574 | |
a7885eb8 KM |
5575 | return 0; |
5576 | } | |
5577 | ||
2e72b634 KS |
5578 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
5579 | { | |
5580 | struct mem_cgroup_threshold_ary *t; | |
5581 | u64 usage; | |
5582 | int i; | |
5583 | ||
5584 | rcu_read_lock(); | |
5585 | if (!swap) | |
2c488db2 | 5586 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 5587 | else |
2c488db2 | 5588 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
5589 | |
5590 | if (!t) | |
5591 | goto unlock; | |
5592 | ||
5593 | usage = mem_cgroup_usage(memcg, swap); | |
5594 | ||
5595 | /* | |
748dad36 | 5596 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
5597 | * If it's not true, a threshold was crossed after last |
5598 | * call of __mem_cgroup_threshold(). | |
5599 | */ | |
5407a562 | 5600 | i = t->current_threshold; |
2e72b634 KS |
5601 | |
5602 | /* | |
5603 | * Iterate backward over array of thresholds starting from | |
5604 | * current_threshold and check if a threshold is crossed. | |
5605 | * If none of thresholds below usage is crossed, we read | |
5606 | * only one element of the array here. | |
5607 | */ | |
5608 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
5609 | eventfd_signal(t->entries[i].eventfd, 1); | |
5610 | ||
5611 | /* i = current_threshold + 1 */ | |
5612 | i++; | |
5613 | ||
5614 | /* | |
5615 | * Iterate forward over array of thresholds starting from | |
5616 | * current_threshold+1 and check if a threshold is crossed. | |
5617 | * If none of thresholds above usage is crossed, we read | |
5618 | * only one element of the array here. | |
5619 | */ | |
5620 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
5621 | eventfd_signal(t->entries[i].eventfd, 1); | |
5622 | ||
5623 | /* Update current_threshold */ | |
5407a562 | 5624 | t->current_threshold = i - 1; |
2e72b634 KS |
5625 | unlock: |
5626 | rcu_read_unlock(); | |
5627 | } | |
5628 | ||
5629 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
5630 | { | |
ad4ca5f4 KS |
5631 | while (memcg) { |
5632 | __mem_cgroup_threshold(memcg, false); | |
5633 | if (do_swap_account) | |
5634 | __mem_cgroup_threshold(memcg, true); | |
5635 | ||
5636 | memcg = parent_mem_cgroup(memcg); | |
5637 | } | |
2e72b634 KS |
5638 | } |
5639 | ||
5640 | static int compare_thresholds(const void *a, const void *b) | |
5641 | { | |
5642 | const struct mem_cgroup_threshold *_a = a; | |
5643 | const struct mem_cgroup_threshold *_b = b; | |
5644 | ||
2bff24a3 GT |
5645 | if (_a->threshold > _b->threshold) |
5646 | return 1; | |
5647 | ||
5648 | if (_a->threshold < _b->threshold) | |
5649 | return -1; | |
5650 | ||
5651 | return 0; | |
2e72b634 KS |
5652 | } |
5653 | ||
c0ff4b85 | 5654 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
5655 | { |
5656 | struct mem_cgroup_eventfd_list *ev; | |
5657 | ||
c0ff4b85 | 5658 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 KH |
5659 | eventfd_signal(ev->eventfd, 1); |
5660 | return 0; | |
5661 | } | |
5662 | ||
c0ff4b85 | 5663 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 5664 | { |
7d74b06f KH |
5665 | struct mem_cgroup *iter; |
5666 | ||
c0ff4b85 | 5667 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 5668 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
5669 | } |
5670 | ||
81eeaf04 | 5671 | static int mem_cgroup_usage_register_event(struct cgroup_subsys_state *css, |
9490ff27 | 5672 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) |
2e72b634 | 5673 | { |
81eeaf04 | 5674 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
2c488db2 KS |
5675 | struct mem_cgroup_thresholds *thresholds; |
5676 | struct mem_cgroup_threshold_ary *new; | |
86ae53e1 | 5677 | enum res_type type = MEMFILE_TYPE(cft->private); |
2e72b634 | 5678 | u64 threshold, usage; |
2c488db2 | 5679 | int i, size, ret; |
2e72b634 KS |
5680 | |
5681 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
5682 | if (ret) | |
5683 | return ret; | |
5684 | ||
5685 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 5686 | |
2e72b634 | 5687 | if (type == _MEM) |
2c488db2 | 5688 | thresholds = &memcg->thresholds; |
2e72b634 | 5689 | else if (type == _MEMSWAP) |
2c488db2 | 5690 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
5691 | else |
5692 | BUG(); | |
5693 | ||
5694 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
5695 | ||
5696 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 5697 | if (thresholds->primary) |
2e72b634 KS |
5698 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
5699 | ||
2c488db2 | 5700 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
5701 | |
5702 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 5703 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 5704 | GFP_KERNEL); |
2c488db2 | 5705 | if (!new) { |
2e72b634 KS |
5706 | ret = -ENOMEM; |
5707 | goto unlock; | |
5708 | } | |
2c488db2 | 5709 | new->size = size; |
2e72b634 KS |
5710 | |
5711 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
5712 | if (thresholds->primary) { |
5713 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 5714 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
5715 | } |
5716 | ||
2e72b634 | 5717 | /* Add new threshold */ |
2c488db2 KS |
5718 | new->entries[size - 1].eventfd = eventfd; |
5719 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
5720 | |
5721 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 5722 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
5723 | compare_thresholds, NULL); |
5724 | ||
5725 | /* Find current threshold */ | |
2c488db2 | 5726 | new->current_threshold = -1; |
2e72b634 | 5727 | for (i = 0; i < size; i++) { |
748dad36 | 5728 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 5729 | /* |
2c488db2 KS |
5730 | * new->current_threshold will not be used until |
5731 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
5732 | * it here. |
5733 | */ | |
2c488db2 | 5734 | ++new->current_threshold; |
748dad36 SZ |
5735 | } else |
5736 | break; | |
2e72b634 KS |
5737 | } |
5738 | ||
2c488db2 KS |
5739 | /* Free old spare buffer and save old primary buffer as spare */ |
5740 | kfree(thresholds->spare); | |
5741 | thresholds->spare = thresholds->primary; | |
5742 | ||
5743 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 5744 | |
907860ed | 5745 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
5746 | synchronize_rcu(); |
5747 | ||
2e72b634 KS |
5748 | unlock: |
5749 | mutex_unlock(&memcg->thresholds_lock); | |
5750 | ||
5751 | return ret; | |
5752 | } | |
5753 | ||
81eeaf04 | 5754 | static void mem_cgroup_usage_unregister_event(struct cgroup_subsys_state *css, |
9490ff27 | 5755 | struct cftype *cft, struct eventfd_ctx *eventfd) |
2e72b634 | 5756 | { |
81eeaf04 | 5757 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
2c488db2 KS |
5758 | struct mem_cgroup_thresholds *thresholds; |
5759 | struct mem_cgroup_threshold_ary *new; | |
86ae53e1 | 5760 | enum res_type type = MEMFILE_TYPE(cft->private); |
2e72b634 | 5761 | u64 usage; |
2c488db2 | 5762 | int i, j, size; |
2e72b634 KS |
5763 | |
5764 | mutex_lock(&memcg->thresholds_lock); | |
5765 | if (type == _MEM) | |
2c488db2 | 5766 | thresholds = &memcg->thresholds; |
2e72b634 | 5767 | else if (type == _MEMSWAP) |
2c488db2 | 5768 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
5769 | else |
5770 | BUG(); | |
5771 | ||
371528ca AV |
5772 | if (!thresholds->primary) |
5773 | goto unlock; | |
5774 | ||
2e72b634 KS |
5775 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); |
5776 | ||
5777 | /* Check if a threshold crossed before removing */ | |
5778 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
5779 | ||
5780 | /* Calculate new number of threshold */ | |
2c488db2 KS |
5781 | size = 0; |
5782 | for (i = 0; i < thresholds->primary->size; i++) { | |
5783 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
5784 | size++; |
5785 | } | |
5786 | ||
2c488db2 | 5787 | new = thresholds->spare; |
907860ed | 5788 | |
2e72b634 KS |
5789 | /* Set thresholds array to NULL if we don't have thresholds */ |
5790 | if (!size) { | |
2c488db2 KS |
5791 | kfree(new); |
5792 | new = NULL; | |
907860ed | 5793 | goto swap_buffers; |
2e72b634 KS |
5794 | } |
5795 | ||
2c488db2 | 5796 | new->size = size; |
2e72b634 KS |
5797 | |
5798 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
5799 | new->current_threshold = -1; |
5800 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
5801 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
5802 | continue; |
5803 | ||
2c488db2 | 5804 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 5805 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 5806 | /* |
2c488db2 | 5807 | * new->current_threshold will not be used |
2e72b634 KS |
5808 | * until rcu_assign_pointer(), so it's safe to increment |
5809 | * it here. | |
5810 | */ | |
2c488db2 | 5811 | ++new->current_threshold; |
2e72b634 KS |
5812 | } |
5813 | j++; | |
5814 | } | |
5815 | ||
907860ed | 5816 | swap_buffers: |
2c488db2 KS |
5817 | /* Swap primary and spare array */ |
5818 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
5819 | /* If all events are unregistered, free the spare array */ |
5820 | if (!new) { | |
5821 | kfree(thresholds->spare); | |
5822 | thresholds->spare = NULL; | |
5823 | } | |
5824 | ||
2c488db2 | 5825 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 5826 | |
907860ed | 5827 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 5828 | synchronize_rcu(); |
371528ca | 5829 | unlock: |
2e72b634 | 5830 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 5831 | } |
c1e862c1 | 5832 | |
81eeaf04 | 5833 | static int mem_cgroup_oom_register_event(struct cgroup_subsys_state *css, |
9490ff27 KH |
5834 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) |
5835 | { | |
81eeaf04 | 5836 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
9490ff27 | 5837 | struct mem_cgroup_eventfd_list *event; |
86ae53e1 | 5838 | enum res_type type = MEMFILE_TYPE(cft->private); |
9490ff27 KH |
5839 | |
5840 | BUG_ON(type != _OOM_TYPE); | |
5841 | event = kmalloc(sizeof(*event), GFP_KERNEL); | |
5842 | if (!event) | |
5843 | return -ENOMEM; | |
5844 | ||
1af8efe9 | 5845 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
5846 | |
5847 | event->eventfd = eventfd; | |
5848 | list_add(&event->list, &memcg->oom_notify); | |
5849 | ||
5850 | /* already in OOM ? */ | |
79dfdacc | 5851 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 5852 | eventfd_signal(eventfd, 1); |
1af8efe9 | 5853 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
5854 | |
5855 | return 0; | |
5856 | } | |
5857 | ||
81eeaf04 | 5858 | static void mem_cgroup_oom_unregister_event(struct cgroup_subsys_state *css, |
9490ff27 KH |
5859 | struct cftype *cft, struct eventfd_ctx *eventfd) |
5860 | { | |
81eeaf04 | 5861 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
9490ff27 | 5862 | struct mem_cgroup_eventfd_list *ev, *tmp; |
86ae53e1 | 5863 | enum res_type type = MEMFILE_TYPE(cft->private); |
9490ff27 KH |
5864 | |
5865 | BUG_ON(type != _OOM_TYPE); | |
5866 | ||
1af8efe9 | 5867 | spin_lock(&memcg_oom_lock); |
9490ff27 | 5868 | |
c0ff4b85 | 5869 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
5870 | if (ev->eventfd == eventfd) { |
5871 | list_del(&ev->list); | |
5872 | kfree(ev); | |
5873 | } | |
5874 | } | |
5875 | ||
1af8efe9 | 5876 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
5877 | } |
5878 | ||
182446d0 | 5879 | static int mem_cgroup_oom_control_read(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
5880 | struct cftype *cft, struct cgroup_map_cb *cb) |
5881 | { | |
182446d0 | 5882 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 | 5883 | |
c0ff4b85 | 5884 | cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable); |
3c11ecf4 | 5885 | |
c0ff4b85 | 5886 | if (atomic_read(&memcg->under_oom)) |
3c11ecf4 KH |
5887 | cb->fill(cb, "under_oom", 1); |
5888 | else | |
5889 | cb->fill(cb, "under_oom", 0); | |
5890 | return 0; | |
5891 | } | |
5892 | ||
182446d0 | 5893 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
5894 | struct cftype *cft, u64 val) |
5895 | { | |
182446d0 | 5896 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
63876986 | 5897 | struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css)); |
3c11ecf4 KH |
5898 | |
5899 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
63876986 | 5900 | if (!parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
5901 | return -EINVAL; |
5902 | ||
0999821b | 5903 | mutex_lock(&memcg_create_mutex); |
3c11ecf4 | 5904 | /* oom-kill-disable is a flag for subhierarchy. */ |
b5f99b53 | 5905 | if ((parent->use_hierarchy) || memcg_has_children(memcg)) { |
0999821b | 5906 | mutex_unlock(&memcg_create_mutex); |
3c11ecf4 KH |
5907 | return -EINVAL; |
5908 | } | |
c0ff4b85 | 5909 | memcg->oom_kill_disable = val; |
4d845ebf | 5910 | if (!val) |
c0ff4b85 | 5911 | memcg_oom_recover(memcg); |
0999821b | 5912 | mutex_unlock(&memcg_create_mutex); |
3c11ecf4 KH |
5913 | return 0; |
5914 | } | |
5915 | ||
c255a458 | 5916 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 5917 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 5918 | { |
55007d84 GC |
5919 | int ret; |
5920 | ||
2633d7a0 | 5921 | memcg->kmemcg_id = -1; |
55007d84 GC |
5922 | ret = memcg_propagate_kmem(memcg); |
5923 | if (ret) | |
5924 | return ret; | |
2633d7a0 | 5925 | |
1d62e436 | 5926 | return mem_cgroup_sockets_init(memcg, ss); |
573b400d | 5927 | } |
e5671dfa | 5928 | |
10d5ebf4 | 5929 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 5930 | { |
1d62e436 | 5931 | mem_cgroup_sockets_destroy(memcg); |
10d5ebf4 LZ |
5932 | } |
5933 | ||
5934 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) | |
5935 | { | |
5936 | if (!memcg_kmem_is_active(memcg)) | |
5937 | return; | |
5938 | ||
5939 | /* | |
5940 | * kmem charges can outlive the cgroup. In the case of slab | |
5941 | * pages, for instance, a page contain objects from various | |
5942 | * processes. As we prevent from taking a reference for every | |
5943 | * such allocation we have to be careful when doing uncharge | |
5944 | * (see memcg_uncharge_kmem) and here during offlining. | |
5945 | * | |
5946 | * The idea is that that only the _last_ uncharge which sees | |
5947 | * the dead memcg will drop the last reference. An additional | |
5948 | * reference is taken here before the group is marked dead | |
5949 | * which is then paired with css_put during uncharge resp. here. | |
5950 | * | |
5951 | * Although this might sound strange as this path is called from | |
5952 | * css_offline() when the referencemight have dropped down to 0 | |
5953 | * and shouldn't be incremented anymore (css_tryget would fail) | |
5954 | * we do not have other options because of the kmem allocations | |
5955 | * lifetime. | |
5956 | */ | |
5957 | css_get(&memcg->css); | |
7de37682 GC |
5958 | |
5959 | memcg_kmem_mark_dead(memcg); | |
5960 | ||
5961 | if (res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0) | |
5962 | return; | |
5963 | ||
7de37682 | 5964 | if (memcg_kmem_test_and_clear_dead(memcg)) |
10d5ebf4 | 5965 | css_put(&memcg->css); |
d1a4c0b3 | 5966 | } |
e5671dfa | 5967 | #else |
cbe128e3 | 5968 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
5969 | { |
5970 | return 0; | |
5971 | } | |
d1a4c0b3 | 5972 | |
10d5ebf4 LZ |
5973 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
5974 | { | |
5975 | } | |
5976 | ||
5977 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) | |
d1a4c0b3 GC |
5978 | { |
5979 | } | |
e5671dfa GC |
5980 | #endif |
5981 | ||
79bd9814 TH |
5982 | /* |
5983 | * Unregister event and free resources. | |
5984 | * | |
5985 | * Gets called from workqueue. | |
5986 | */ | |
5987 | static void cgroup_event_remove(struct work_struct *work) | |
5988 | { | |
5989 | struct cgroup_event *event = container_of(work, struct cgroup_event, | |
5990 | remove); | |
5991 | struct cgroup_subsys_state *css = event->css; | |
5992 | ||
5993 | remove_wait_queue(event->wqh, &event->wait); | |
5994 | ||
5995 | event->cft->unregister_event(css, event->cft, event->eventfd); | |
5996 | ||
5997 | /* Notify userspace the event is going away. */ | |
5998 | eventfd_signal(event->eventfd, 1); | |
5999 | ||
6000 | eventfd_ctx_put(event->eventfd); | |
6001 | kfree(event); | |
6002 | css_put(css); | |
6003 | } | |
6004 | ||
6005 | /* | |
6006 | * Gets called on POLLHUP on eventfd when user closes it. | |
6007 | * | |
6008 | * Called with wqh->lock held and interrupts disabled. | |
6009 | */ | |
6010 | static int cgroup_event_wake(wait_queue_t *wait, unsigned mode, | |
6011 | int sync, void *key) | |
6012 | { | |
6013 | struct cgroup_event *event = container_of(wait, | |
6014 | struct cgroup_event, wait); | |
6015 | struct cgroup *cgrp = event->css->cgroup; | |
6016 | unsigned long flags = (unsigned long)key; | |
6017 | ||
6018 | if (flags & POLLHUP) { | |
6019 | /* | |
6020 | * If the event has been detached at cgroup removal, we | |
6021 | * can simply return knowing the other side will cleanup | |
6022 | * for us. | |
6023 | * | |
6024 | * We can't race against event freeing since the other | |
6025 | * side will require wqh->lock via remove_wait_queue(), | |
6026 | * which we hold. | |
6027 | */ | |
6028 | spin_lock(&cgrp->event_list_lock); | |
6029 | if (!list_empty(&event->list)) { | |
6030 | list_del_init(&event->list); | |
6031 | /* | |
6032 | * We are in atomic context, but cgroup_event_remove() | |
6033 | * may sleep, so we have to call it in workqueue. | |
6034 | */ | |
6035 | schedule_work(&event->remove); | |
6036 | } | |
6037 | spin_unlock(&cgrp->event_list_lock); | |
6038 | } | |
6039 | ||
6040 | return 0; | |
6041 | } | |
6042 | ||
6043 | static void cgroup_event_ptable_queue_proc(struct file *file, | |
6044 | wait_queue_head_t *wqh, poll_table *pt) | |
6045 | { | |
6046 | struct cgroup_event *event = container_of(pt, | |
6047 | struct cgroup_event, pt); | |
6048 | ||
6049 | event->wqh = wqh; | |
6050 | add_wait_queue(wqh, &event->wait); | |
6051 | } | |
6052 | ||
6053 | /* | |
6054 | * Parse input and register new cgroup event handler. | |
6055 | * | |
6056 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
6057 | * Interpretation of args is defined by control file implementation. | |
6058 | */ | |
6059 | static int cgroup_write_event_control(struct cgroup_subsys_state *dummy_css, | |
6060 | struct cftype *cft, const char *buffer) | |
6061 | { | |
6062 | struct cgroup *cgrp = dummy_css->cgroup; | |
6063 | struct cgroup_event *event; | |
6064 | struct cgroup_subsys_state *cfile_css; | |
6065 | unsigned int efd, cfd; | |
6066 | struct fd efile; | |
6067 | struct fd cfile; | |
6068 | char *endp; | |
6069 | int ret; | |
6070 | ||
6071 | efd = simple_strtoul(buffer, &endp, 10); | |
6072 | if (*endp != ' ') | |
6073 | return -EINVAL; | |
6074 | buffer = endp + 1; | |
6075 | ||
6076 | cfd = simple_strtoul(buffer, &endp, 10); | |
6077 | if ((*endp != ' ') && (*endp != '\0')) | |
6078 | return -EINVAL; | |
6079 | buffer = endp + 1; | |
6080 | ||
6081 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
6082 | if (!event) | |
6083 | return -ENOMEM; | |
6084 | ||
6085 | INIT_LIST_HEAD(&event->list); | |
6086 | init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc); | |
6087 | init_waitqueue_func_entry(&event->wait, cgroup_event_wake); | |
6088 | INIT_WORK(&event->remove, cgroup_event_remove); | |
6089 | ||
6090 | efile = fdget(efd); | |
6091 | if (!efile.file) { | |
6092 | ret = -EBADF; | |
6093 | goto out_kfree; | |
6094 | } | |
6095 | ||
6096 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
6097 | if (IS_ERR(event->eventfd)) { | |
6098 | ret = PTR_ERR(event->eventfd); | |
6099 | goto out_put_efile; | |
6100 | } | |
6101 | ||
6102 | cfile = fdget(cfd); | |
6103 | if (!cfile.file) { | |
6104 | ret = -EBADF; | |
6105 | goto out_put_eventfd; | |
6106 | } | |
6107 | ||
6108 | /* the process need read permission on control file */ | |
6109 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
6110 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
6111 | if (ret < 0) | |
6112 | goto out_put_cfile; | |
6113 | ||
6114 | event->cft = __file_cft(cfile.file); | |
6115 | if (IS_ERR(event->cft)) { | |
6116 | ret = PTR_ERR(event->cft); | |
6117 | goto out_put_cfile; | |
6118 | } | |
6119 | ||
6120 | if (!event->cft->ss) { | |
6121 | ret = -EBADF; | |
6122 | goto out_put_cfile; | |
6123 | } | |
6124 | ||
6125 | /* | |
6126 | * Determine the css of @cfile, verify it belongs to the same | |
6127 | * cgroup as cgroup.event_control, and associate @event with it. | |
6128 | * Remaining events are automatically removed on cgroup destruction | |
6129 | * but the removal is asynchronous, so take an extra ref. | |
6130 | */ | |
6131 | rcu_read_lock(); | |
6132 | ||
6133 | ret = -EINVAL; | |
6134 | event->css = cgroup_css(cgrp, event->cft->ss); | |
6135 | cfile_css = css_from_dir(cfile.file->f_dentry->d_parent, event->cft->ss); | |
6136 | if (event->css && event->css == cfile_css && css_tryget(event->css)) | |
6137 | ret = 0; | |
6138 | ||
6139 | rcu_read_unlock(); | |
6140 | if (ret) | |
6141 | goto out_put_cfile; | |
6142 | ||
6143 | if (!event->cft->register_event || !event->cft->unregister_event) { | |
6144 | ret = -EINVAL; | |
6145 | goto out_put_css; | |
6146 | } | |
6147 | ||
6148 | ret = event->cft->register_event(event->css, event->cft, | |
6149 | event->eventfd, buffer); | |
6150 | if (ret) | |
6151 | goto out_put_css; | |
6152 | ||
6153 | efile.file->f_op->poll(efile.file, &event->pt); | |
6154 | ||
6155 | spin_lock(&cgrp->event_list_lock); | |
6156 | list_add(&event->list, &cgrp->event_list); | |
6157 | spin_unlock(&cgrp->event_list_lock); | |
6158 | ||
6159 | fdput(cfile); | |
6160 | fdput(efile); | |
6161 | ||
6162 | return 0; | |
6163 | ||
6164 | out_put_css: | |
6165 | css_put(event->css); | |
6166 | out_put_cfile: | |
6167 | fdput(cfile); | |
6168 | out_put_eventfd: | |
6169 | eventfd_ctx_put(event->eventfd); | |
6170 | out_put_efile: | |
6171 | fdput(efile); | |
6172 | out_kfree: | |
6173 | kfree(event); | |
6174 | ||
6175 | return ret; | |
6176 | } | |
6177 | ||
8cdea7c0 BS |
6178 | static struct cftype mem_cgroup_files[] = { |
6179 | { | |
0eea1030 | 6180 | .name = "usage_in_bytes", |
8c7c6e34 | 6181 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
af36f906 | 6182 | .read = mem_cgroup_read, |
9490ff27 KH |
6183 | .register_event = mem_cgroup_usage_register_event, |
6184 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8cdea7c0 | 6185 | }, |
c84872e1 PE |
6186 | { |
6187 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 6188 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 6189 | .trigger = mem_cgroup_reset, |
af36f906 | 6190 | .read = mem_cgroup_read, |
c84872e1 | 6191 | }, |
8cdea7c0 | 6192 | { |
0eea1030 | 6193 | .name = "limit_in_bytes", |
8c7c6e34 | 6194 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 6195 | .write_string = mem_cgroup_write, |
af36f906 | 6196 | .read = mem_cgroup_read, |
8cdea7c0 | 6197 | }, |
296c81d8 BS |
6198 | { |
6199 | .name = "soft_limit_in_bytes", | |
6200 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
6201 | .write_string = mem_cgroup_write, | |
af36f906 | 6202 | .read = mem_cgroup_read, |
296c81d8 | 6203 | }, |
8cdea7c0 BS |
6204 | { |
6205 | .name = "failcnt", | |
8c7c6e34 | 6206 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 6207 | .trigger = mem_cgroup_reset, |
af36f906 | 6208 | .read = mem_cgroup_read, |
8cdea7c0 | 6209 | }, |
d2ceb9b7 KH |
6210 | { |
6211 | .name = "stat", | |
ab215884 | 6212 | .read_seq_string = memcg_stat_show, |
d2ceb9b7 | 6213 | }, |
c1e862c1 KH |
6214 | { |
6215 | .name = "force_empty", | |
6216 | .trigger = mem_cgroup_force_empty_write, | |
6217 | }, | |
18f59ea7 BS |
6218 | { |
6219 | .name = "use_hierarchy", | |
f00baae7 | 6220 | .flags = CFTYPE_INSANE, |
18f59ea7 BS |
6221 | .write_u64 = mem_cgroup_hierarchy_write, |
6222 | .read_u64 = mem_cgroup_hierarchy_read, | |
6223 | }, | |
79bd9814 TH |
6224 | { |
6225 | .name = "cgroup.event_control", | |
6226 | .write_string = cgroup_write_event_control, | |
6227 | .flags = CFTYPE_NO_PREFIX, | |
6228 | .mode = S_IWUGO, | |
6229 | }, | |
a7885eb8 KM |
6230 | { |
6231 | .name = "swappiness", | |
6232 | .read_u64 = mem_cgroup_swappiness_read, | |
6233 | .write_u64 = mem_cgroup_swappiness_write, | |
6234 | }, | |
7dc74be0 DN |
6235 | { |
6236 | .name = "move_charge_at_immigrate", | |
6237 | .read_u64 = mem_cgroup_move_charge_read, | |
6238 | .write_u64 = mem_cgroup_move_charge_write, | |
6239 | }, | |
9490ff27 KH |
6240 | { |
6241 | .name = "oom_control", | |
3c11ecf4 KH |
6242 | .read_map = mem_cgroup_oom_control_read, |
6243 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
6244 | .register_event = mem_cgroup_oom_register_event, |
6245 | .unregister_event = mem_cgroup_oom_unregister_event, | |
6246 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), | |
6247 | }, | |
70ddf637 AV |
6248 | { |
6249 | .name = "pressure_level", | |
6250 | .register_event = vmpressure_register_event, | |
6251 | .unregister_event = vmpressure_unregister_event, | |
6252 | }, | |
406eb0c9 YH |
6253 | #ifdef CONFIG_NUMA |
6254 | { | |
6255 | .name = "numa_stat", | |
ab215884 | 6256 | .read_seq_string = memcg_numa_stat_show, |
406eb0c9 YH |
6257 | }, |
6258 | #endif | |
510fc4e1 GC |
6259 | #ifdef CONFIG_MEMCG_KMEM |
6260 | { | |
6261 | .name = "kmem.limit_in_bytes", | |
6262 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
6263 | .write_string = mem_cgroup_write, | |
6264 | .read = mem_cgroup_read, | |
6265 | }, | |
6266 | { | |
6267 | .name = "kmem.usage_in_bytes", | |
6268 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
6269 | .read = mem_cgroup_read, | |
6270 | }, | |
6271 | { | |
6272 | .name = "kmem.failcnt", | |
6273 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6274 | .trigger = mem_cgroup_reset, | |
6275 | .read = mem_cgroup_read, | |
6276 | }, | |
6277 | { | |
6278 | .name = "kmem.max_usage_in_bytes", | |
6279 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6280 | .trigger = mem_cgroup_reset, | |
6281 | .read = mem_cgroup_read, | |
6282 | }, | |
749c5415 GC |
6283 | #ifdef CONFIG_SLABINFO |
6284 | { | |
6285 | .name = "kmem.slabinfo", | |
6286 | .read_seq_string = mem_cgroup_slabinfo_read, | |
6287 | }, | |
6288 | #endif | |
8c7c6e34 | 6289 | #endif |
6bc10349 | 6290 | { }, /* terminate */ |
af36f906 | 6291 | }; |
8c7c6e34 | 6292 | |
2d11085e MH |
6293 | #ifdef CONFIG_MEMCG_SWAP |
6294 | static struct cftype memsw_cgroup_files[] = { | |
6295 | { | |
6296 | .name = "memsw.usage_in_bytes", | |
6297 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
6298 | .read = mem_cgroup_read, | |
6299 | .register_event = mem_cgroup_usage_register_event, | |
6300 | .unregister_event = mem_cgroup_usage_unregister_event, | |
6301 | }, | |
6302 | { | |
6303 | .name = "memsw.max_usage_in_bytes", | |
6304 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6305 | .trigger = mem_cgroup_reset, | |
6306 | .read = mem_cgroup_read, | |
6307 | }, | |
6308 | { | |
6309 | .name = "memsw.limit_in_bytes", | |
6310 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
6311 | .write_string = mem_cgroup_write, | |
6312 | .read = mem_cgroup_read, | |
6313 | }, | |
6314 | { | |
6315 | .name = "memsw.failcnt", | |
6316 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6317 | .trigger = mem_cgroup_reset, | |
6318 | .read = mem_cgroup_read, | |
6319 | }, | |
6320 | { }, /* terminate */ | |
6321 | }; | |
6322 | #endif | |
c0ff4b85 | 6323 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
6324 | { |
6325 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 6326 | struct mem_cgroup_per_zone *mz; |
41e3355d | 6327 | int zone, tmp = node; |
1ecaab2b KH |
6328 | /* |
6329 | * This routine is called against possible nodes. | |
6330 | * But it's BUG to call kmalloc() against offline node. | |
6331 | * | |
6332 | * TODO: this routine can waste much memory for nodes which will | |
6333 | * never be onlined. It's better to use memory hotplug callback | |
6334 | * function. | |
6335 | */ | |
41e3355d KH |
6336 | if (!node_state(node, N_NORMAL_MEMORY)) |
6337 | tmp = -1; | |
17295c88 | 6338 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
6339 | if (!pn) |
6340 | return 1; | |
1ecaab2b | 6341 | |
1ecaab2b KH |
6342 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
6343 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 6344 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
6345 | mz->usage_in_excess = 0; |
6346 | mz->on_tree = false; | |
d79154bb | 6347 | mz->memcg = memcg; |
1ecaab2b | 6348 | } |
54f72fe0 | 6349 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
6350 | return 0; |
6351 | } | |
6352 | ||
c0ff4b85 | 6353 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 6354 | { |
54f72fe0 | 6355 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
6356 | } |
6357 | ||
33327948 KH |
6358 | static struct mem_cgroup *mem_cgroup_alloc(void) |
6359 | { | |
d79154bb | 6360 | struct mem_cgroup *memcg; |
45cf7ebd | 6361 | size_t size = memcg_size(); |
33327948 | 6362 | |
45cf7ebd | 6363 | /* Can be very big if nr_node_ids is very big */ |
c8dad2bb | 6364 | if (size < PAGE_SIZE) |
d79154bb | 6365 | memcg = kzalloc(size, GFP_KERNEL); |
33327948 | 6366 | else |
d79154bb | 6367 | memcg = vzalloc(size); |
33327948 | 6368 | |
d79154bb | 6369 | if (!memcg) |
e7bbcdf3 DC |
6370 | return NULL; |
6371 | ||
d79154bb HD |
6372 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
6373 | if (!memcg->stat) | |
d2e61b8d | 6374 | goto out_free; |
d79154bb HD |
6375 | spin_lock_init(&memcg->pcp_counter_lock); |
6376 | return memcg; | |
d2e61b8d DC |
6377 | |
6378 | out_free: | |
6379 | if (size < PAGE_SIZE) | |
d79154bb | 6380 | kfree(memcg); |
d2e61b8d | 6381 | else |
d79154bb | 6382 | vfree(memcg); |
d2e61b8d | 6383 | return NULL; |
33327948 KH |
6384 | } |
6385 | ||
59927fb9 | 6386 | /* |
c8b2a36f GC |
6387 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
6388 | * (scanning all at force_empty is too costly...) | |
6389 | * | |
6390 | * Instead of clearing all references at force_empty, we remember | |
6391 | * the number of reference from swap_cgroup and free mem_cgroup when | |
6392 | * it goes down to 0. | |
6393 | * | |
6394 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 6395 | */ |
c8b2a36f GC |
6396 | |
6397 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 6398 | { |
c8b2a36f | 6399 | int node; |
45cf7ebd | 6400 | size_t size = memcg_size(); |
59927fb9 | 6401 | |
bb4cc1a8 | 6402 | mem_cgroup_remove_from_trees(memcg); |
c8b2a36f GC |
6403 | free_css_id(&mem_cgroup_subsys, &memcg->css); |
6404 | ||
6405 | for_each_node(node) | |
6406 | free_mem_cgroup_per_zone_info(memcg, node); | |
6407 | ||
6408 | free_percpu(memcg->stat); | |
6409 | ||
3f134619 GC |
6410 | /* |
6411 | * We need to make sure that (at least for now), the jump label | |
6412 | * destruction code runs outside of the cgroup lock. This is because | |
6413 | * get_online_cpus(), which is called from the static_branch update, | |
6414 | * can't be called inside the cgroup_lock. cpusets are the ones | |
6415 | * enforcing this dependency, so if they ever change, we might as well. | |
6416 | * | |
6417 | * schedule_work() will guarantee this happens. Be careful if you need | |
6418 | * to move this code around, and make sure it is outside | |
6419 | * the cgroup_lock. | |
6420 | */ | |
a8964b9b | 6421 | disarm_static_keys(memcg); |
3afe36b1 GC |
6422 | if (size < PAGE_SIZE) |
6423 | kfree(memcg); | |
6424 | else | |
6425 | vfree(memcg); | |
59927fb9 | 6426 | } |
3afe36b1 | 6427 | |
7bcc1bb1 DN |
6428 | /* |
6429 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
6430 | */ | |
e1aab161 | 6431 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 6432 | { |
c0ff4b85 | 6433 | if (!memcg->res.parent) |
7bcc1bb1 | 6434 | return NULL; |
c0ff4b85 | 6435 | return mem_cgroup_from_res_counter(memcg->res.parent, res); |
7bcc1bb1 | 6436 | } |
e1aab161 | 6437 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 6438 | |
bb4cc1a8 AM |
6439 | static void __init mem_cgroup_soft_limit_tree_init(void) |
6440 | { | |
6441 | struct mem_cgroup_tree_per_node *rtpn; | |
6442 | struct mem_cgroup_tree_per_zone *rtpz; | |
6443 | int tmp, node, zone; | |
6444 | ||
6445 | for_each_node(node) { | |
6446 | tmp = node; | |
6447 | if (!node_state(node, N_NORMAL_MEMORY)) | |
6448 | tmp = -1; | |
6449 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
6450 | BUG_ON(!rtpn); | |
6451 | ||
6452 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
6453 | ||
6454 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
6455 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
6456 | rtpz->rb_root = RB_ROOT; | |
6457 | spin_lock_init(&rtpz->lock); | |
6458 | } | |
6459 | } | |
6460 | } | |
6461 | ||
0eb253e2 | 6462 | static struct cgroup_subsys_state * __ref |
eb95419b | 6463 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 6464 | { |
d142e3e6 | 6465 | struct mem_cgroup *memcg; |
04046e1a | 6466 | long error = -ENOMEM; |
6d12e2d8 | 6467 | int node; |
8cdea7c0 | 6468 | |
c0ff4b85 R |
6469 | memcg = mem_cgroup_alloc(); |
6470 | if (!memcg) | |
04046e1a | 6471 | return ERR_PTR(error); |
78fb7466 | 6472 | |
3ed28fa1 | 6473 | for_each_node(node) |
c0ff4b85 | 6474 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 6475 | goto free_out; |
f64c3f54 | 6476 | |
c077719b | 6477 | /* root ? */ |
eb95419b | 6478 | if (parent_css == NULL) { |
a41c58a6 | 6479 | root_mem_cgroup = memcg; |
d142e3e6 GC |
6480 | res_counter_init(&memcg->res, NULL); |
6481 | res_counter_init(&memcg->memsw, NULL); | |
6482 | res_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 6483 | } |
28dbc4b6 | 6484 | |
d142e3e6 GC |
6485 | memcg->last_scanned_node = MAX_NUMNODES; |
6486 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
6487 | memcg->move_charge_at_immigrate = 0; |
6488 | mutex_init(&memcg->thresholds_lock); | |
6489 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 6490 | vmpressure_init(&memcg->vmpressure); |
d142e3e6 GC |
6491 | |
6492 | return &memcg->css; | |
6493 | ||
6494 | free_out: | |
6495 | __mem_cgroup_free(memcg); | |
6496 | return ERR_PTR(error); | |
6497 | } | |
6498 | ||
6499 | static int | |
eb95419b | 6500 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 6501 | { |
eb95419b TH |
6502 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
6503 | struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(css)); | |
d142e3e6 GC |
6504 | int error = 0; |
6505 | ||
63876986 | 6506 | if (!parent) |
d142e3e6 GC |
6507 | return 0; |
6508 | ||
0999821b | 6509 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
6510 | |
6511 | memcg->use_hierarchy = parent->use_hierarchy; | |
6512 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
6513 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
6514 | ||
6515 | if (parent->use_hierarchy) { | |
c0ff4b85 R |
6516 | res_counter_init(&memcg->res, &parent->res); |
6517 | res_counter_init(&memcg->memsw, &parent->memsw); | |
510fc4e1 | 6518 | res_counter_init(&memcg->kmem, &parent->kmem); |
55007d84 | 6519 | |
7bcc1bb1 | 6520 | /* |
8d76a979 LZ |
6521 | * No need to take a reference to the parent because cgroup |
6522 | * core guarantees its existence. | |
7bcc1bb1 | 6523 | */ |
18f59ea7 | 6524 | } else { |
c0ff4b85 R |
6525 | res_counter_init(&memcg->res, NULL); |
6526 | res_counter_init(&memcg->memsw, NULL); | |
510fc4e1 | 6527 | res_counter_init(&memcg->kmem, NULL); |
8c7f6edb TH |
6528 | /* |
6529 | * Deeper hierachy with use_hierarchy == false doesn't make | |
6530 | * much sense so let cgroup subsystem know about this | |
6531 | * unfortunate state in our controller. | |
6532 | */ | |
d142e3e6 | 6533 | if (parent != root_mem_cgroup) |
8c7f6edb | 6534 | mem_cgroup_subsys.broken_hierarchy = true; |
18f59ea7 | 6535 | } |
cbe128e3 GC |
6536 | |
6537 | error = memcg_init_kmem(memcg, &mem_cgroup_subsys); | |
0999821b | 6538 | mutex_unlock(&memcg_create_mutex); |
d142e3e6 | 6539 | return error; |
8cdea7c0 BS |
6540 | } |
6541 | ||
5f578161 MH |
6542 | /* |
6543 | * Announce all parents that a group from their hierarchy is gone. | |
6544 | */ | |
6545 | static void mem_cgroup_invalidate_reclaim_iterators(struct mem_cgroup *memcg) | |
6546 | { | |
6547 | struct mem_cgroup *parent = memcg; | |
6548 | ||
6549 | while ((parent = parent_mem_cgroup(parent))) | |
519ebea3 | 6550 | mem_cgroup_iter_invalidate(parent); |
5f578161 MH |
6551 | |
6552 | /* | |
6553 | * if the root memcg is not hierarchical we have to check it | |
6554 | * explicitely. | |
6555 | */ | |
6556 | if (!root_mem_cgroup->use_hierarchy) | |
519ebea3 | 6557 | mem_cgroup_iter_invalidate(root_mem_cgroup); |
5f578161 MH |
6558 | } |
6559 | ||
eb95419b | 6560 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 6561 | { |
eb95419b | 6562 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
79bd9814 TH |
6563 | struct cgroup *cgrp = css->cgroup; |
6564 | struct cgroup_event *event, *tmp; | |
6565 | ||
6566 | /* | |
6567 | * Unregister events and notify userspace. | |
6568 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
6569 | * directory to avoid race between userspace and kernelspace. | |
6570 | */ | |
6571 | spin_lock(&cgrp->event_list_lock); | |
6572 | list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) { | |
6573 | list_del_init(&event->list); | |
6574 | schedule_work(&event->remove); | |
6575 | } | |
6576 | spin_unlock(&cgrp->event_list_lock); | |
ec64f515 | 6577 | |
10d5ebf4 LZ |
6578 | kmem_cgroup_css_offline(memcg); |
6579 | ||
5f578161 | 6580 | mem_cgroup_invalidate_reclaim_iterators(memcg); |
ab5196c2 | 6581 | mem_cgroup_reparent_charges(memcg); |
1f458cbf | 6582 | mem_cgroup_destroy_all_caches(memcg); |
33cb876e | 6583 | vmpressure_cleanup(&memcg->vmpressure); |
df878fb0 KH |
6584 | } |
6585 | ||
eb95419b | 6586 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 6587 | { |
eb95419b | 6588 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 6589 | |
10d5ebf4 | 6590 | memcg_destroy_kmem(memcg); |
465939a1 | 6591 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
6592 | } |
6593 | ||
02491447 | 6594 | #ifdef CONFIG_MMU |
7dc74be0 | 6595 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
6596 | #define PRECHARGE_COUNT_AT_ONCE 256 |
6597 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 6598 | { |
854ffa8d DN |
6599 | int ret = 0; |
6600 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
c0ff4b85 | 6601 | struct mem_cgroup *memcg = mc.to; |
4ffef5fe | 6602 | |
c0ff4b85 | 6603 | if (mem_cgroup_is_root(memcg)) { |
854ffa8d DN |
6604 | mc.precharge += count; |
6605 | /* we don't need css_get for root */ | |
6606 | return ret; | |
6607 | } | |
6608 | /* try to charge at once */ | |
6609 | if (count > 1) { | |
6610 | struct res_counter *dummy; | |
6611 | /* | |
c0ff4b85 | 6612 | * "memcg" cannot be under rmdir() because we've already checked |
854ffa8d DN |
6613 | * by cgroup_lock_live_cgroup() that it is not removed and we |
6614 | * are still under the same cgroup_mutex. So we can postpone | |
6615 | * css_get(). | |
6616 | */ | |
c0ff4b85 | 6617 | if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy)) |
854ffa8d | 6618 | goto one_by_one; |
c0ff4b85 | 6619 | if (do_swap_account && res_counter_charge(&memcg->memsw, |
854ffa8d | 6620 | PAGE_SIZE * count, &dummy)) { |
c0ff4b85 | 6621 | res_counter_uncharge(&memcg->res, PAGE_SIZE * count); |
854ffa8d DN |
6622 | goto one_by_one; |
6623 | } | |
6624 | mc.precharge += count; | |
854ffa8d DN |
6625 | return ret; |
6626 | } | |
6627 | one_by_one: | |
6628 | /* fall back to one by one charge */ | |
6629 | while (count--) { | |
6630 | if (signal_pending(current)) { | |
6631 | ret = -EINTR; | |
6632 | break; | |
6633 | } | |
6634 | if (!batch_count--) { | |
6635 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
6636 | cond_resched(); | |
6637 | } | |
c0ff4b85 R |
6638 | ret = __mem_cgroup_try_charge(NULL, |
6639 | GFP_KERNEL, 1, &memcg, false); | |
38c5d72f | 6640 | if (ret) |
854ffa8d | 6641 | /* mem_cgroup_clear_mc() will do uncharge later */ |
38c5d72f | 6642 | return ret; |
854ffa8d DN |
6643 | mc.precharge++; |
6644 | } | |
4ffef5fe DN |
6645 | return ret; |
6646 | } | |
6647 | ||
6648 | /** | |
8d32ff84 | 6649 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
6650 | * @vma: the vma the pte to be checked belongs |
6651 | * @addr: the address corresponding to the pte to be checked | |
6652 | * @ptent: the pte to be checked | |
02491447 | 6653 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
6654 | * |
6655 | * Returns | |
6656 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
6657 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
6658 | * move charge. if @target is not NULL, the page is stored in target->page | |
6659 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
6660 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
6661 | * target for charge migration. if @target is not NULL, the entry is stored | |
6662 | * in target->ent. | |
4ffef5fe DN |
6663 | * |
6664 | * Called with pte lock held. | |
6665 | */ | |
4ffef5fe DN |
6666 | union mc_target { |
6667 | struct page *page; | |
02491447 | 6668 | swp_entry_t ent; |
4ffef5fe DN |
6669 | }; |
6670 | ||
4ffef5fe | 6671 | enum mc_target_type { |
8d32ff84 | 6672 | MC_TARGET_NONE = 0, |
4ffef5fe | 6673 | MC_TARGET_PAGE, |
02491447 | 6674 | MC_TARGET_SWAP, |
4ffef5fe DN |
6675 | }; |
6676 | ||
90254a65 DN |
6677 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
6678 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 6679 | { |
90254a65 | 6680 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 6681 | |
90254a65 DN |
6682 | if (!page || !page_mapped(page)) |
6683 | return NULL; | |
6684 | if (PageAnon(page)) { | |
6685 | /* we don't move shared anon */ | |
4b91355e | 6686 | if (!move_anon()) |
90254a65 | 6687 | return NULL; |
87946a72 DN |
6688 | } else if (!move_file()) |
6689 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
6690 | return NULL; |
6691 | if (!get_page_unless_zero(page)) | |
6692 | return NULL; | |
6693 | ||
6694 | return page; | |
6695 | } | |
6696 | ||
4b91355e | 6697 | #ifdef CONFIG_SWAP |
90254a65 DN |
6698 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
6699 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6700 | { | |
90254a65 DN |
6701 | struct page *page = NULL; |
6702 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
6703 | ||
6704 | if (!move_anon() || non_swap_entry(ent)) | |
6705 | return NULL; | |
4b91355e KH |
6706 | /* |
6707 | * Because lookup_swap_cache() updates some statistics counter, | |
6708 | * we call find_get_page() with swapper_space directly. | |
6709 | */ | |
33806f06 | 6710 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
6711 | if (do_swap_account) |
6712 | entry->val = ent.val; | |
6713 | ||
6714 | return page; | |
6715 | } | |
4b91355e KH |
6716 | #else |
6717 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
6718 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6719 | { | |
6720 | return NULL; | |
6721 | } | |
6722 | #endif | |
90254a65 | 6723 | |
87946a72 DN |
6724 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
6725 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6726 | { | |
6727 | struct page *page = NULL; | |
87946a72 DN |
6728 | struct address_space *mapping; |
6729 | pgoff_t pgoff; | |
6730 | ||
6731 | if (!vma->vm_file) /* anonymous vma */ | |
6732 | return NULL; | |
6733 | if (!move_file()) | |
6734 | return NULL; | |
6735 | ||
87946a72 DN |
6736 | mapping = vma->vm_file->f_mapping; |
6737 | if (pte_none(ptent)) | |
6738 | pgoff = linear_page_index(vma, addr); | |
6739 | else /* pte_file(ptent) is true */ | |
6740 | pgoff = pte_to_pgoff(ptent); | |
6741 | ||
6742 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
6743 | page = find_get_page(mapping, pgoff); |
6744 | ||
6745 | #ifdef CONFIG_SWAP | |
6746 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
6747 | if (radix_tree_exceptional_entry(page)) { | |
6748 | swp_entry_t swap = radix_to_swp_entry(page); | |
87946a72 | 6749 | if (do_swap_account) |
aa3b1895 | 6750 | *entry = swap; |
33806f06 | 6751 | page = find_get_page(swap_address_space(swap), swap.val); |
87946a72 | 6752 | } |
aa3b1895 | 6753 | #endif |
87946a72 DN |
6754 | return page; |
6755 | } | |
6756 | ||
8d32ff84 | 6757 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
6758 | unsigned long addr, pte_t ptent, union mc_target *target) |
6759 | { | |
6760 | struct page *page = NULL; | |
6761 | struct page_cgroup *pc; | |
8d32ff84 | 6762 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
6763 | swp_entry_t ent = { .val = 0 }; |
6764 | ||
6765 | if (pte_present(ptent)) | |
6766 | page = mc_handle_present_pte(vma, addr, ptent); | |
6767 | else if (is_swap_pte(ptent)) | |
6768 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
6769 | else if (pte_none(ptent) || pte_file(ptent)) |
6770 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
6771 | |
6772 | if (!page && !ent.val) | |
8d32ff84 | 6773 | return ret; |
02491447 DN |
6774 | if (page) { |
6775 | pc = lookup_page_cgroup(page); | |
6776 | /* | |
6777 | * Do only loose check w/o page_cgroup lock. | |
6778 | * mem_cgroup_move_account() checks the pc is valid or not under | |
6779 | * the lock. | |
6780 | */ | |
6781 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
6782 | ret = MC_TARGET_PAGE; | |
6783 | if (target) | |
6784 | target->page = page; | |
6785 | } | |
6786 | if (!ret || !target) | |
6787 | put_page(page); | |
6788 | } | |
90254a65 DN |
6789 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
6790 | if (ent.val && !ret && | |
9fb4b7cc | 6791 | css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
6792 | ret = MC_TARGET_SWAP; |
6793 | if (target) | |
6794 | target->ent = ent; | |
4ffef5fe | 6795 | } |
4ffef5fe DN |
6796 | return ret; |
6797 | } | |
6798 | ||
12724850 NH |
6799 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
6800 | /* | |
6801 | * We don't consider swapping or file mapped pages because THP does not | |
6802 | * support them for now. | |
6803 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
6804 | */ | |
6805 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6806 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6807 | { | |
6808 | struct page *page = NULL; | |
6809 | struct page_cgroup *pc; | |
6810 | enum mc_target_type ret = MC_TARGET_NONE; | |
6811 | ||
6812 | page = pmd_page(pmd); | |
6813 | VM_BUG_ON(!page || !PageHead(page)); | |
6814 | if (!move_anon()) | |
6815 | return ret; | |
6816 | pc = lookup_page_cgroup(page); | |
6817 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
6818 | ret = MC_TARGET_PAGE; | |
6819 | if (target) { | |
6820 | get_page(page); | |
6821 | target->page = page; | |
6822 | } | |
6823 | } | |
6824 | return ret; | |
6825 | } | |
6826 | #else | |
6827 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6828 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6829 | { | |
6830 | return MC_TARGET_NONE; | |
6831 | } | |
6832 | #endif | |
6833 | ||
4ffef5fe DN |
6834 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
6835 | unsigned long addr, unsigned long end, | |
6836 | struct mm_walk *walk) | |
6837 | { | |
6838 | struct vm_area_struct *vma = walk->private; | |
6839 | pte_t *pte; | |
6840 | spinlock_t *ptl; | |
6841 | ||
12724850 NH |
6842 | if (pmd_trans_huge_lock(pmd, vma) == 1) { |
6843 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) | |
6844 | mc.precharge += HPAGE_PMD_NR; | |
6845 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 6846 | return 0; |
12724850 | 6847 | } |
03319327 | 6848 | |
45f83cef AA |
6849 | if (pmd_trans_unstable(pmd)) |
6850 | return 0; | |
4ffef5fe DN |
6851 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
6852 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 6853 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
6854 | mc.precharge++; /* increment precharge temporarily */ |
6855 | pte_unmap_unlock(pte - 1, ptl); | |
6856 | cond_resched(); | |
6857 | ||
7dc74be0 DN |
6858 | return 0; |
6859 | } | |
6860 | ||
4ffef5fe DN |
6861 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
6862 | { | |
6863 | unsigned long precharge; | |
6864 | struct vm_area_struct *vma; | |
6865 | ||
dfe076b0 | 6866 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
6867 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
6868 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
6869 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
6870 | .mm = mm, | |
6871 | .private = vma, | |
6872 | }; | |
6873 | if (is_vm_hugetlb_page(vma)) | |
6874 | continue; | |
4ffef5fe DN |
6875 | walk_page_range(vma->vm_start, vma->vm_end, |
6876 | &mem_cgroup_count_precharge_walk); | |
6877 | } | |
dfe076b0 | 6878 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
6879 | |
6880 | precharge = mc.precharge; | |
6881 | mc.precharge = 0; | |
6882 | ||
6883 | return precharge; | |
6884 | } | |
6885 | ||
4ffef5fe DN |
6886 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
6887 | { | |
dfe076b0 DN |
6888 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
6889 | ||
6890 | VM_BUG_ON(mc.moving_task); | |
6891 | mc.moving_task = current; | |
6892 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
6893 | } |
6894 | ||
dfe076b0 DN |
6895 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
6896 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 6897 | { |
2bd9bb20 KH |
6898 | struct mem_cgroup *from = mc.from; |
6899 | struct mem_cgroup *to = mc.to; | |
4050377b | 6900 | int i; |
2bd9bb20 | 6901 | |
4ffef5fe | 6902 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
6903 | if (mc.precharge) { |
6904 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
6905 | mc.precharge = 0; | |
6906 | } | |
6907 | /* | |
6908 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
6909 | * we must uncharge here. | |
6910 | */ | |
6911 | if (mc.moved_charge) { | |
6912 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
6913 | mc.moved_charge = 0; | |
4ffef5fe | 6914 | } |
483c30b5 DN |
6915 | /* we must fixup refcnts and charges */ |
6916 | if (mc.moved_swap) { | |
483c30b5 DN |
6917 | /* uncharge swap account from the old cgroup */ |
6918 | if (!mem_cgroup_is_root(mc.from)) | |
6919 | res_counter_uncharge(&mc.from->memsw, | |
6920 | PAGE_SIZE * mc.moved_swap); | |
4050377b LZ |
6921 | |
6922 | for (i = 0; i < mc.moved_swap; i++) | |
6923 | css_put(&mc.from->css); | |
483c30b5 DN |
6924 | |
6925 | if (!mem_cgroup_is_root(mc.to)) { | |
6926 | /* | |
6927 | * we charged both to->res and to->memsw, so we should | |
6928 | * uncharge to->res. | |
6929 | */ | |
6930 | res_counter_uncharge(&mc.to->res, | |
6931 | PAGE_SIZE * mc.moved_swap); | |
483c30b5 | 6932 | } |
4050377b | 6933 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
6934 | mc.moved_swap = 0; |
6935 | } | |
dfe076b0 DN |
6936 | memcg_oom_recover(from); |
6937 | memcg_oom_recover(to); | |
6938 | wake_up_all(&mc.waitq); | |
6939 | } | |
6940 | ||
6941 | static void mem_cgroup_clear_mc(void) | |
6942 | { | |
6943 | struct mem_cgroup *from = mc.from; | |
6944 | ||
6945 | /* | |
6946 | * we must clear moving_task before waking up waiters at the end of | |
6947 | * task migration. | |
6948 | */ | |
6949 | mc.moving_task = NULL; | |
6950 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 6951 | spin_lock(&mc.lock); |
4ffef5fe DN |
6952 | mc.from = NULL; |
6953 | mc.to = NULL; | |
2bd9bb20 | 6954 | spin_unlock(&mc.lock); |
32047e2a | 6955 | mem_cgroup_end_move(from); |
4ffef5fe DN |
6956 | } |
6957 | ||
eb95419b | 6958 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6959 | struct cgroup_taskset *tset) |
7dc74be0 | 6960 | { |
2f7ee569 | 6961 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 6962 | int ret = 0; |
eb95419b | 6963 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
ee5e8472 | 6964 | unsigned long move_charge_at_immigrate; |
7dc74be0 | 6965 | |
ee5e8472 GC |
6966 | /* |
6967 | * We are now commited to this value whatever it is. Changes in this | |
6968 | * tunable will only affect upcoming migrations, not the current one. | |
6969 | * So we need to save it, and keep it going. | |
6970 | */ | |
6971 | move_charge_at_immigrate = memcg->move_charge_at_immigrate; | |
6972 | if (move_charge_at_immigrate) { | |
7dc74be0 DN |
6973 | struct mm_struct *mm; |
6974 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
6975 | ||
c0ff4b85 | 6976 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
6977 | |
6978 | mm = get_task_mm(p); | |
6979 | if (!mm) | |
6980 | return 0; | |
7dc74be0 | 6981 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
6982 | if (mm->owner == p) { |
6983 | VM_BUG_ON(mc.from); | |
6984 | VM_BUG_ON(mc.to); | |
6985 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 6986 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 6987 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 6988 | mem_cgroup_start_move(from); |
2bd9bb20 | 6989 | spin_lock(&mc.lock); |
4ffef5fe | 6990 | mc.from = from; |
c0ff4b85 | 6991 | mc.to = memcg; |
ee5e8472 | 6992 | mc.immigrate_flags = move_charge_at_immigrate; |
2bd9bb20 | 6993 | spin_unlock(&mc.lock); |
dfe076b0 | 6994 | /* We set mc.moving_task later */ |
4ffef5fe DN |
6995 | |
6996 | ret = mem_cgroup_precharge_mc(mm); | |
6997 | if (ret) | |
6998 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
6999 | } |
7000 | mmput(mm); | |
7dc74be0 DN |
7001 | } |
7002 | return ret; | |
7003 | } | |
7004 | ||
eb95419b | 7005 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 7006 | struct cgroup_taskset *tset) |
7dc74be0 | 7007 | { |
4ffef5fe | 7008 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
7009 | } |
7010 | ||
4ffef5fe DN |
7011 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
7012 | unsigned long addr, unsigned long end, | |
7013 | struct mm_walk *walk) | |
7dc74be0 | 7014 | { |
4ffef5fe DN |
7015 | int ret = 0; |
7016 | struct vm_area_struct *vma = walk->private; | |
7017 | pte_t *pte; | |
7018 | spinlock_t *ptl; | |
12724850 NH |
7019 | enum mc_target_type target_type; |
7020 | union mc_target target; | |
7021 | struct page *page; | |
7022 | struct page_cgroup *pc; | |
4ffef5fe | 7023 | |
12724850 NH |
7024 | /* |
7025 | * We don't take compound_lock() here but no race with splitting thp | |
7026 | * happens because: | |
7027 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
7028 | * under splitting, which means there's no concurrent thp split, | |
7029 | * - if another thread runs into split_huge_page() just after we | |
7030 | * entered this if-block, the thread must wait for page table lock | |
7031 | * to be unlocked in __split_huge_page_splitting(), where the main | |
7032 | * part of thp split is not executed yet. | |
7033 | */ | |
7034 | if (pmd_trans_huge_lock(pmd, vma) == 1) { | |
62ade86a | 7035 | if (mc.precharge < HPAGE_PMD_NR) { |
12724850 NH |
7036 | spin_unlock(&vma->vm_mm->page_table_lock); |
7037 | return 0; | |
7038 | } | |
7039 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
7040 | if (target_type == MC_TARGET_PAGE) { | |
7041 | page = target.page; | |
7042 | if (!isolate_lru_page(page)) { | |
7043 | pc = lookup_page_cgroup(page); | |
7044 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, | |
2f3479b1 | 7045 | pc, mc.from, mc.to)) { |
12724850 NH |
7046 | mc.precharge -= HPAGE_PMD_NR; |
7047 | mc.moved_charge += HPAGE_PMD_NR; | |
7048 | } | |
7049 | putback_lru_page(page); | |
7050 | } | |
7051 | put_page(page); | |
7052 | } | |
7053 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 7054 | return 0; |
12724850 NH |
7055 | } |
7056 | ||
45f83cef AA |
7057 | if (pmd_trans_unstable(pmd)) |
7058 | return 0; | |
4ffef5fe DN |
7059 | retry: |
7060 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
7061 | for (; addr != end; addr += PAGE_SIZE) { | |
7062 | pte_t ptent = *(pte++); | |
02491447 | 7063 | swp_entry_t ent; |
4ffef5fe DN |
7064 | |
7065 | if (!mc.precharge) | |
7066 | break; | |
7067 | ||
8d32ff84 | 7068 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
7069 | case MC_TARGET_PAGE: |
7070 | page = target.page; | |
7071 | if (isolate_lru_page(page)) | |
7072 | goto put; | |
7073 | pc = lookup_page_cgroup(page); | |
7ec99d62 | 7074 | if (!mem_cgroup_move_account(page, 1, pc, |
2f3479b1 | 7075 | mc.from, mc.to)) { |
4ffef5fe | 7076 | mc.precharge--; |
854ffa8d DN |
7077 | /* we uncharge from mc.from later. */ |
7078 | mc.moved_charge++; | |
4ffef5fe DN |
7079 | } |
7080 | putback_lru_page(page); | |
8d32ff84 | 7081 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
7082 | put_page(page); |
7083 | break; | |
02491447 DN |
7084 | case MC_TARGET_SWAP: |
7085 | ent = target.ent; | |
e91cbb42 | 7086 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 7087 | mc.precharge--; |
483c30b5 DN |
7088 | /* we fixup refcnts and charges later. */ |
7089 | mc.moved_swap++; | |
7090 | } | |
02491447 | 7091 | break; |
4ffef5fe DN |
7092 | default: |
7093 | break; | |
7094 | } | |
7095 | } | |
7096 | pte_unmap_unlock(pte - 1, ptl); | |
7097 | cond_resched(); | |
7098 | ||
7099 | if (addr != end) { | |
7100 | /* | |
7101 | * We have consumed all precharges we got in can_attach(). | |
7102 | * We try charge one by one, but don't do any additional | |
7103 | * charges to mc.to if we have failed in charge once in attach() | |
7104 | * phase. | |
7105 | */ | |
854ffa8d | 7106 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
7107 | if (!ret) |
7108 | goto retry; | |
7109 | } | |
7110 | ||
7111 | return ret; | |
7112 | } | |
7113 | ||
7114 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
7115 | { | |
7116 | struct vm_area_struct *vma; | |
7117 | ||
7118 | lru_add_drain_all(); | |
dfe076b0 DN |
7119 | retry: |
7120 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
7121 | /* | |
7122 | * Someone who are holding the mmap_sem might be waiting in | |
7123 | * waitq. So we cancel all extra charges, wake up all waiters, | |
7124 | * and retry. Because we cancel precharges, we might not be able | |
7125 | * to move enough charges, but moving charge is a best-effort | |
7126 | * feature anyway, so it wouldn't be a big problem. | |
7127 | */ | |
7128 | __mem_cgroup_clear_mc(); | |
7129 | cond_resched(); | |
7130 | goto retry; | |
7131 | } | |
4ffef5fe DN |
7132 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
7133 | int ret; | |
7134 | struct mm_walk mem_cgroup_move_charge_walk = { | |
7135 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
7136 | .mm = mm, | |
7137 | .private = vma, | |
7138 | }; | |
7139 | if (is_vm_hugetlb_page(vma)) | |
7140 | continue; | |
4ffef5fe DN |
7141 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
7142 | &mem_cgroup_move_charge_walk); | |
7143 | if (ret) | |
7144 | /* | |
7145 | * means we have consumed all precharges and failed in | |
7146 | * doing additional charge. Just abandon here. | |
7147 | */ | |
7148 | break; | |
7149 | } | |
dfe076b0 | 7150 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
7151 | } |
7152 | ||
eb95419b | 7153 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 7154 | struct cgroup_taskset *tset) |
67e465a7 | 7155 | { |
2f7ee569 | 7156 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 7157 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 7158 | |
dfe076b0 | 7159 | if (mm) { |
a433658c KM |
7160 | if (mc.to) |
7161 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
7162 | mmput(mm); |
7163 | } | |
a433658c KM |
7164 | if (mc.to) |
7165 | mem_cgroup_clear_mc(); | |
67e465a7 | 7166 | } |
5cfb80a7 | 7167 | #else /* !CONFIG_MMU */ |
eb95419b | 7168 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 7169 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
7170 | { |
7171 | return 0; | |
7172 | } | |
eb95419b | 7173 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 7174 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
7175 | { |
7176 | } | |
eb95419b | 7177 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 7178 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
7179 | { |
7180 | } | |
7181 | #endif | |
67e465a7 | 7182 | |
f00baae7 TH |
7183 | /* |
7184 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
7185 | * to verify sane_behavior flag on each mount attempt. | |
7186 | */ | |
eb95419b | 7187 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
7188 | { |
7189 | /* | |
7190 | * use_hierarchy is forced with sane_behavior. cgroup core | |
7191 | * guarantees that @root doesn't have any children, so turning it | |
7192 | * on for the root memcg is enough. | |
7193 | */ | |
eb95419b TH |
7194 | if (cgroup_sane_behavior(root_css->cgroup)) |
7195 | mem_cgroup_from_css(root_css)->use_hierarchy = true; | |
f00baae7 TH |
7196 | } |
7197 | ||
8cdea7c0 BS |
7198 | struct cgroup_subsys mem_cgroup_subsys = { |
7199 | .name = "memory", | |
7200 | .subsys_id = mem_cgroup_subsys_id, | |
92fb9748 | 7201 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 7202 | .css_online = mem_cgroup_css_online, |
92fb9748 TH |
7203 | .css_offline = mem_cgroup_css_offline, |
7204 | .css_free = mem_cgroup_css_free, | |
7dc74be0 DN |
7205 | .can_attach = mem_cgroup_can_attach, |
7206 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 7207 | .attach = mem_cgroup_move_task, |
f00baae7 | 7208 | .bind = mem_cgroup_bind, |
6bc10349 | 7209 | .base_cftypes = mem_cgroup_files, |
6d12e2d8 | 7210 | .early_init = 0, |
04046e1a | 7211 | .use_id = 1, |
8cdea7c0 | 7212 | }; |
c077719b | 7213 | |
c255a458 | 7214 | #ifdef CONFIG_MEMCG_SWAP |
a42c390c MH |
7215 | static int __init enable_swap_account(char *s) |
7216 | { | |
a2c8990a | 7217 | if (!strcmp(s, "1")) |
a42c390c | 7218 | really_do_swap_account = 1; |
a2c8990a | 7219 | else if (!strcmp(s, "0")) |
a42c390c MH |
7220 | really_do_swap_account = 0; |
7221 | return 1; | |
7222 | } | |
a2c8990a | 7223 | __setup("swapaccount=", enable_swap_account); |
c077719b | 7224 | |
2d11085e MH |
7225 | static void __init memsw_file_init(void) |
7226 | { | |
6acc8b02 MH |
7227 | WARN_ON(cgroup_add_cftypes(&mem_cgroup_subsys, memsw_cgroup_files)); |
7228 | } | |
7229 | ||
7230 | static void __init enable_swap_cgroup(void) | |
7231 | { | |
7232 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
7233 | do_swap_account = 1; | |
7234 | memsw_file_init(); | |
7235 | } | |
2d11085e | 7236 | } |
6acc8b02 | 7237 | |
2d11085e | 7238 | #else |
6acc8b02 | 7239 | static void __init enable_swap_cgroup(void) |
2d11085e MH |
7240 | { |
7241 | } | |
c077719b | 7242 | #endif |
2d11085e MH |
7243 | |
7244 | /* | |
1081312f MH |
7245 | * subsys_initcall() for memory controller. |
7246 | * | |
7247 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
7248 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
7249 | * everything that doesn't depend on a specific mem_cgroup structure should | |
7250 | * be initialized from here. | |
2d11085e MH |
7251 | */ |
7252 | static int __init mem_cgroup_init(void) | |
7253 | { | |
7254 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); | |
6acc8b02 | 7255 | enable_swap_cgroup(); |
bb4cc1a8 | 7256 | mem_cgroup_soft_limit_tree_init(); |
e4777496 | 7257 | memcg_stock_init(); |
2d11085e MH |
7258 | return 0; |
7259 | } | |
7260 | subsys_initcall(mem_cgroup_init); |