]>
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; | |
79bd9814 TH |
239 | /* |
240 | * eventfd to signal userspace about the event. | |
241 | */ | |
242 | struct eventfd_ctx *eventfd; | |
243 | /* | |
244 | * Each of these stored in a list by the cgroup. | |
245 | */ | |
246 | struct list_head list; | |
fba94807 TH |
247 | /* |
248 | * register_event() callback will be used to add new userspace | |
249 | * waiter for changes related to this event. Use eventfd_signal() | |
250 | * on eventfd to send notification to userspace. | |
251 | */ | |
252 | int (*register_event)(struct cgroup_subsys_state *css, | |
347c4a87 | 253 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
254 | /* |
255 | * unregister_event() callback will be called when userspace closes | |
256 | * the eventfd or on cgroup removing. This callback must be set, | |
257 | * if you want provide notification functionality. | |
258 | */ | |
259 | void (*unregister_event)(struct cgroup_subsys_state *css, | |
fba94807 | 260 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
261 | /* |
262 | * All fields below needed to unregister event when | |
263 | * userspace closes eventfd. | |
264 | */ | |
265 | poll_table pt; | |
266 | wait_queue_head_t *wqh; | |
267 | wait_queue_t wait; | |
268 | struct work_struct remove; | |
269 | }; | |
270 | ||
c0ff4b85 R |
271 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
272 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 273 | |
8cdea7c0 BS |
274 | /* |
275 | * The memory controller data structure. The memory controller controls both | |
276 | * page cache and RSS per cgroup. We would eventually like to provide | |
277 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
278 | * to help the administrator determine what knobs to tune. | |
279 | * | |
280 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
281 | * we hit the water mark. May be even add a low water mark, such that |
282 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
283 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
284 | */ |
285 | struct mem_cgroup { | |
286 | struct cgroup_subsys_state css; | |
287 | /* | |
288 | * the counter to account for memory usage | |
289 | */ | |
290 | struct res_counter res; | |
59927fb9 | 291 | |
70ddf637 AV |
292 | /* vmpressure notifications */ |
293 | struct vmpressure vmpressure; | |
294 | ||
465939a1 LZ |
295 | /* |
296 | * the counter to account for mem+swap usage. | |
297 | */ | |
298 | struct res_counter memsw; | |
59927fb9 | 299 | |
510fc4e1 GC |
300 | /* |
301 | * the counter to account for kernel memory usage. | |
302 | */ | |
303 | struct res_counter kmem; | |
18f59ea7 BS |
304 | /* |
305 | * Should the accounting and control be hierarchical, per subtree? | |
306 | */ | |
307 | bool use_hierarchy; | |
510fc4e1 | 308 | unsigned long kmem_account_flags; /* See KMEM_ACCOUNTED_*, below */ |
79dfdacc MH |
309 | |
310 | bool oom_lock; | |
311 | atomic_t under_oom; | |
3812c8c8 | 312 | atomic_t oom_wakeups; |
79dfdacc | 313 | |
1f4c025b | 314 | int swappiness; |
3c11ecf4 KH |
315 | /* OOM-Killer disable */ |
316 | int oom_kill_disable; | |
a7885eb8 | 317 | |
22a668d7 KH |
318 | /* set when res.limit == memsw.limit */ |
319 | bool memsw_is_minimum; | |
320 | ||
2e72b634 KS |
321 | /* protect arrays of thresholds */ |
322 | struct mutex thresholds_lock; | |
323 | ||
324 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 325 | struct mem_cgroup_thresholds thresholds; |
907860ed | 326 | |
2e72b634 | 327 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 328 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 329 | |
9490ff27 KH |
330 | /* For oom notifier event fd */ |
331 | struct list_head oom_notify; | |
185efc0f | 332 | |
7dc74be0 DN |
333 | /* |
334 | * Should we move charges of a task when a task is moved into this | |
335 | * mem_cgroup ? And what type of charges should we move ? | |
336 | */ | |
f894ffa8 | 337 | unsigned long move_charge_at_immigrate; |
619d094b KH |
338 | /* |
339 | * set > 0 if pages under this cgroup are moving to other cgroup. | |
340 | */ | |
341 | atomic_t moving_account; | |
312734c0 KH |
342 | /* taken only while moving_account > 0 */ |
343 | spinlock_t move_lock; | |
d52aa412 | 344 | /* |
c62b1a3b | 345 | * percpu counter. |
d52aa412 | 346 | */ |
3a7951b4 | 347 | struct mem_cgroup_stat_cpu __percpu *stat; |
711d3d2c KH |
348 | /* |
349 | * used when a cpu is offlined or other synchronizations | |
350 | * See mem_cgroup_read_stat(). | |
351 | */ | |
352 | struct mem_cgroup_stat_cpu nocpu_base; | |
353 | spinlock_t pcp_counter_lock; | |
d1a4c0b3 | 354 | |
5f578161 | 355 | atomic_t dead_count; |
4bd2c1ee | 356 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET) |
d1a4c0b3 GC |
357 | struct tcp_memcontrol tcp_mem; |
358 | #endif | |
2633d7a0 GC |
359 | #if defined(CONFIG_MEMCG_KMEM) |
360 | /* analogous to slab_common's slab_caches list. per-memcg */ | |
361 | struct list_head memcg_slab_caches; | |
362 | /* Not a spinlock, we can take a lot of time walking the list */ | |
363 | struct mutex slab_caches_mutex; | |
364 | /* Index in the kmem_cache->memcg_params->memcg_caches array */ | |
365 | int kmemcg_id; | |
366 | #endif | |
45cf7ebd GC |
367 | |
368 | int last_scanned_node; | |
369 | #if MAX_NUMNODES > 1 | |
370 | nodemask_t scan_nodes; | |
371 | atomic_t numainfo_events; | |
372 | atomic_t numainfo_updating; | |
373 | #endif | |
70ddf637 | 374 | |
fba94807 TH |
375 | /* List of events which userspace want to receive */ |
376 | struct list_head event_list; | |
377 | spinlock_t event_list_lock; | |
378 | ||
54f72fe0 JW |
379 | struct mem_cgroup_per_node *nodeinfo[0]; |
380 | /* WARNING: nodeinfo must be the last member here */ | |
8cdea7c0 BS |
381 | }; |
382 | ||
45cf7ebd GC |
383 | static size_t memcg_size(void) |
384 | { | |
385 | return sizeof(struct mem_cgroup) + | |
386 | nr_node_ids * sizeof(struct mem_cgroup_per_node); | |
387 | } | |
388 | ||
510fc4e1 GC |
389 | /* internal only representation about the status of kmem accounting. */ |
390 | enum { | |
391 | KMEM_ACCOUNTED_ACTIVE = 0, /* accounted by this cgroup itself */ | |
a8964b9b | 392 | KMEM_ACCOUNTED_ACTIVATED, /* static key enabled. */ |
7de37682 | 393 | KMEM_ACCOUNTED_DEAD, /* dead memcg with pending kmem charges */ |
510fc4e1 GC |
394 | }; |
395 | ||
a8964b9b GC |
396 | /* We account when limit is on, but only after call sites are patched */ |
397 | #define KMEM_ACCOUNTED_MASK \ | |
398 | ((1 << KMEM_ACCOUNTED_ACTIVE) | (1 << KMEM_ACCOUNTED_ACTIVATED)) | |
510fc4e1 GC |
399 | |
400 | #ifdef CONFIG_MEMCG_KMEM | |
401 | static inline void memcg_kmem_set_active(struct mem_cgroup *memcg) | |
402 | { | |
403 | set_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
404 | } | |
7de37682 GC |
405 | |
406 | static bool memcg_kmem_is_active(struct mem_cgroup *memcg) | |
407 | { | |
408 | return test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
409 | } | |
410 | ||
a8964b9b GC |
411 | static void memcg_kmem_set_activated(struct mem_cgroup *memcg) |
412 | { | |
413 | set_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags); | |
414 | } | |
415 | ||
55007d84 GC |
416 | static void memcg_kmem_clear_activated(struct mem_cgroup *memcg) |
417 | { | |
418 | clear_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags); | |
419 | } | |
420 | ||
7de37682 GC |
421 | static void memcg_kmem_mark_dead(struct mem_cgroup *memcg) |
422 | { | |
10d5ebf4 LZ |
423 | /* |
424 | * Our caller must use css_get() first, because memcg_uncharge_kmem() | |
425 | * will call css_put() if it sees the memcg is dead. | |
426 | */ | |
427 | smp_wmb(); | |
7de37682 GC |
428 | if (test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags)) |
429 | set_bit(KMEM_ACCOUNTED_DEAD, &memcg->kmem_account_flags); | |
430 | } | |
431 | ||
432 | static bool memcg_kmem_test_and_clear_dead(struct mem_cgroup *memcg) | |
433 | { | |
434 | return test_and_clear_bit(KMEM_ACCOUNTED_DEAD, | |
435 | &memcg->kmem_account_flags); | |
436 | } | |
510fc4e1 GC |
437 | #endif |
438 | ||
7dc74be0 DN |
439 | /* Stuffs for move charges at task migration. */ |
440 | /* | |
ee5e8472 GC |
441 | * Types of charges to be moved. "move_charge_at_immitgrate" and |
442 | * "immigrate_flags" are treated as a left-shifted bitmap of these types. | |
7dc74be0 DN |
443 | */ |
444 | enum move_type { | |
4ffef5fe | 445 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 446 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
447 | NR_MOVE_TYPE, |
448 | }; | |
449 | ||
4ffef5fe DN |
450 | /* "mc" and its members are protected by cgroup_mutex */ |
451 | static struct move_charge_struct { | |
b1dd693e | 452 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
453 | struct mem_cgroup *from; |
454 | struct mem_cgroup *to; | |
ee5e8472 | 455 | unsigned long immigrate_flags; |
4ffef5fe | 456 | unsigned long precharge; |
854ffa8d | 457 | unsigned long moved_charge; |
483c30b5 | 458 | unsigned long moved_swap; |
8033b97c DN |
459 | struct task_struct *moving_task; /* a task moving charges */ |
460 | wait_queue_head_t waitq; /* a waitq for other context */ | |
461 | } mc = { | |
2bd9bb20 | 462 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
463 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
464 | }; | |
4ffef5fe | 465 | |
90254a65 DN |
466 | static bool move_anon(void) |
467 | { | |
ee5e8472 | 468 | return test_bit(MOVE_CHARGE_TYPE_ANON, &mc.immigrate_flags); |
90254a65 DN |
469 | } |
470 | ||
87946a72 DN |
471 | static bool move_file(void) |
472 | { | |
ee5e8472 | 473 | return test_bit(MOVE_CHARGE_TYPE_FILE, &mc.immigrate_flags); |
87946a72 DN |
474 | } |
475 | ||
4e416953 BS |
476 | /* |
477 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
478 | * limit reclaim to prevent infinite loops, if they ever occur. | |
479 | */ | |
a0db00fc | 480 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 481 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 482 | |
217bc319 KH |
483 | enum charge_type { |
484 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 485 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 486 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 487 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
488 | NR_CHARGE_TYPE, |
489 | }; | |
490 | ||
8c7c6e34 | 491 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
492 | enum res_type { |
493 | _MEM, | |
494 | _MEMSWAP, | |
495 | _OOM_TYPE, | |
510fc4e1 | 496 | _KMEM, |
86ae53e1 GC |
497 | }; |
498 | ||
a0db00fc KS |
499 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
500 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 501 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
502 | /* Used for OOM nofiier */ |
503 | #define OOM_CONTROL (0) | |
8c7c6e34 | 504 | |
75822b44 BS |
505 | /* |
506 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
507 | */ | |
508 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
509 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
510 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
511 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
512 | ||
0999821b GC |
513 | /* |
514 | * The memcg_create_mutex will be held whenever a new cgroup is created. | |
515 | * As a consequence, any change that needs to protect against new child cgroups | |
516 | * appearing has to hold it as well. | |
517 | */ | |
518 | static DEFINE_MUTEX(memcg_create_mutex); | |
519 | ||
b2145145 WL |
520 | struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s) |
521 | { | |
a7c6d554 | 522 | return s ? container_of(s, struct mem_cgroup, css) : NULL; |
b2145145 WL |
523 | } |
524 | ||
70ddf637 AV |
525 | /* Some nice accessors for the vmpressure. */ |
526 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
527 | { | |
528 | if (!memcg) | |
529 | memcg = root_mem_cgroup; | |
530 | return &memcg->vmpressure; | |
531 | } | |
532 | ||
533 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
534 | { | |
535 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
536 | } | |
537 | ||
538 | struct vmpressure *css_to_vmpressure(struct cgroup_subsys_state *css) | |
539 | { | |
540 | return &mem_cgroup_from_css(css)->vmpressure; | |
541 | } | |
542 | ||
7ffc0edc MH |
543 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
544 | { | |
545 | return (memcg == root_mem_cgroup); | |
546 | } | |
547 | ||
e1aab161 | 548 | /* Writing them here to avoid exposing memcg's inner layout */ |
4bd2c1ee | 549 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
e1aab161 | 550 | |
e1aab161 GC |
551 | void sock_update_memcg(struct sock *sk) |
552 | { | |
376be5ff | 553 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 554 | struct mem_cgroup *memcg; |
3f134619 | 555 | struct cg_proto *cg_proto; |
e1aab161 GC |
556 | |
557 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
558 | ||
f3f511e1 GC |
559 | /* Socket cloning can throw us here with sk_cgrp already |
560 | * filled. It won't however, necessarily happen from | |
561 | * process context. So the test for root memcg given | |
562 | * the current task's memcg won't help us in this case. | |
563 | * | |
564 | * Respecting the original socket's memcg is a better | |
565 | * decision in this case. | |
566 | */ | |
567 | if (sk->sk_cgrp) { | |
568 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
5347e5ae | 569 | css_get(&sk->sk_cgrp->memcg->css); |
f3f511e1 GC |
570 | return; |
571 | } | |
572 | ||
e1aab161 GC |
573 | rcu_read_lock(); |
574 | memcg = mem_cgroup_from_task(current); | |
3f134619 | 575 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
5347e5ae LZ |
576 | if (!mem_cgroup_is_root(memcg) && |
577 | memcg_proto_active(cg_proto) && css_tryget(&memcg->css)) { | |
3f134619 | 578 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
579 | } |
580 | rcu_read_unlock(); | |
581 | } | |
582 | } | |
583 | EXPORT_SYMBOL(sock_update_memcg); | |
584 | ||
585 | void sock_release_memcg(struct sock *sk) | |
586 | { | |
376be5ff | 587 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
588 | struct mem_cgroup *memcg; |
589 | WARN_ON(!sk->sk_cgrp->memcg); | |
590 | memcg = sk->sk_cgrp->memcg; | |
5347e5ae | 591 | css_put(&sk->sk_cgrp->memcg->css); |
e1aab161 GC |
592 | } |
593 | } | |
d1a4c0b3 GC |
594 | |
595 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
596 | { | |
597 | if (!memcg || mem_cgroup_is_root(memcg)) | |
598 | return NULL; | |
599 | ||
600 | return &memcg->tcp_mem.cg_proto; | |
601 | } | |
602 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 603 | |
3f134619 GC |
604 | static void disarm_sock_keys(struct mem_cgroup *memcg) |
605 | { | |
606 | if (!memcg_proto_activated(&memcg->tcp_mem.cg_proto)) | |
607 | return; | |
608 | static_key_slow_dec(&memcg_socket_limit_enabled); | |
609 | } | |
610 | #else | |
611 | static void disarm_sock_keys(struct mem_cgroup *memcg) | |
612 | { | |
613 | } | |
614 | #endif | |
615 | ||
a8964b9b | 616 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 GC |
617 | /* |
618 | * This will be the memcg's index in each cache's ->memcg_params->memcg_caches. | |
619 | * There are two main reasons for not using the css_id for this: | |
620 | * 1) this works better in sparse environments, where we have a lot of memcgs, | |
621 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
622 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
623 | * 200 entry array for that. | |
624 | * | |
625 | * 2) In order not to violate the cgroup API, we would like to do all memory | |
626 | * allocation in ->create(). At that point, we haven't yet allocated the | |
627 | * css_id. Having a separate index prevents us from messing with the cgroup | |
628 | * core for this | |
629 | * | |
630 | * The current size of the caches array is stored in | |
631 | * memcg_limited_groups_array_size. It will double each time we have to | |
632 | * increase it. | |
633 | */ | |
634 | static DEFINE_IDA(kmem_limited_groups); | |
749c5415 GC |
635 | int memcg_limited_groups_array_size; |
636 | ||
55007d84 GC |
637 | /* |
638 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
639 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
640 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
641 | * tunable, but that is strictly not necessary. | |
642 | * | |
643 | * MAX_SIZE should be as large as the number of css_ids. Ideally, we could get | |
644 | * this constant directly from cgroup, but it is understandable that this is | |
645 | * better kept as an internal representation in cgroup.c. In any case, the | |
646 | * css_id space is not getting any smaller, and we don't have to necessarily | |
647 | * increase ours as well if it increases. | |
648 | */ | |
649 | #define MEMCG_CACHES_MIN_SIZE 4 | |
650 | #define MEMCG_CACHES_MAX_SIZE 65535 | |
651 | ||
d7f25f8a GC |
652 | /* |
653 | * A lot of the calls to the cache allocation functions are expected to be | |
654 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
655 | * conditional to this static branch, we'll have to allow modules that does | |
656 | * kmem_cache_alloc and the such to see this symbol as well | |
657 | */ | |
a8964b9b | 658 | struct static_key memcg_kmem_enabled_key; |
d7f25f8a | 659 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b GC |
660 | |
661 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
662 | { | |
55007d84 | 663 | if (memcg_kmem_is_active(memcg)) { |
a8964b9b | 664 | static_key_slow_dec(&memcg_kmem_enabled_key); |
55007d84 GC |
665 | ida_simple_remove(&kmem_limited_groups, memcg->kmemcg_id); |
666 | } | |
bea207c8 GC |
667 | /* |
668 | * This check can't live in kmem destruction function, | |
669 | * since the charges will outlive the cgroup | |
670 | */ | |
671 | WARN_ON(res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0); | |
a8964b9b GC |
672 | } |
673 | #else | |
674 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
675 | { | |
676 | } | |
677 | #endif /* CONFIG_MEMCG_KMEM */ | |
678 | ||
679 | static void disarm_static_keys(struct mem_cgroup *memcg) | |
680 | { | |
681 | disarm_sock_keys(memcg); | |
682 | disarm_kmem_keys(memcg); | |
683 | } | |
684 | ||
c0ff4b85 | 685 | static void drain_all_stock_async(struct mem_cgroup *memcg); |
8c7c6e34 | 686 | |
f64c3f54 | 687 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 688 | mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid) |
f64c3f54 | 689 | { |
45cf7ebd | 690 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
54f72fe0 | 691 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
692 | } |
693 | ||
c0ff4b85 | 694 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 695 | { |
c0ff4b85 | 696 | return &memcg->css; |
d324236b WF |
697 | } |
698 | ||
f64c3f54 | 699 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 700 | page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 701 | { |
97a6c37b JW |
702 | int nid = page_to_nid(page); |
703 | int zid = page_zonenum(page); | |
f64c3f54 | 704 | |
c0ff4b85 | 705 | return mem_cgroup_zoneinfo(memcg, nid, zid); |
f64c3f54 BS |
706 | } |
707 | ||
bb4cc1a8 AM |
708 | static struct mem_cgroup_tree_per_zone * |
709 | soft_limit_tree_node_zone(int nid, int zid) | |
710 | { | |
711 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
712 | } | |
713 | ||
714 | static struct mem_cgroup_tree_per_zone * | |
715 | soft_limit_tree_from_page(struct page *page) | |
716 | { | |
717 | int nid = page_to_nid(page); | |
718 | int zid = page_zonenum(page); | |
719 | ||
720 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
721 | } | |
722 | ||
723 | static void | |
724 | __mem_cgroup_insert_exceeded(struct mem_cgroup *memcg, | |
725 | struct mem_cgroup_per_zone *mz, | |
726 | struct mem_cgroup_tree_per_zone *mctz, | |
727 | unsigned long long new_usage_in_excess) | |
728 | { | |
729 | struct rb_node **p = &mctz->rb_root.rb_node; | |
730 | struct rb_node *parent = NULL; | |
731 | struct mem_cgroup_per_zone *mz_node; | |
732 | ||
733 | if (mz->on_tree) | |
734 | return; | |
735 | ||
736 | mz->usage_in_excess = new_usage_in_excess; | |
737 | if (!mz->usage_in_excess) | |
738 | return; | |
739 | while (*p) { | |
740 | parent = *p; | |
741 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
742 | tree_node); | |
743 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
744 | p = &(*p)->rb_left; | |
745 | /* | |
746 | * We can't avoid mem cgroups that are over their soft | |
747 | * limit by the same amount | |
748 | */ | |
749 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
750 | p = &(*p)->rb_right; | |
751 | } | |
752 | rb_link_node(&mz->tree_node, parent, p); | |
753 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
754 | mz->on_tree = true; | |
755 | } | |
756 | ||
757 | static void | |
758 | __mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, | |
759 | struct mem_cgroup_per_zone *mz, | |
760 | struct mem_cgroup_tree_per_zone *mctz) | |
761 | { | |
762 | if (!mz->on_tree) | |
763 | return; | |
764 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
765 | mz->on_tree = false; | |
766 | } | |
767 | ||
768 | static void | |
769 | mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, | |
770 | struct mem_cgroup_per_zone *mz, | |
771 | struct mem_cgroup_tree_per_zone *mctz) | |
772 | { | |
773 | spin_lock(&mctz->lock); | |
774 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); | |
775 | spin_unlock(&mctz->lock); | |
776 | } | |
777 | ||
778 | ||
779 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
780 | { | |
781 | unsigned long long excess; | |
782 | struct mem_cgroup_per_zone *mz; | |
783 | struct mem_cgroup_tree_per_zone *mctz; | |
784 | int nid = page_to_nid(page); | |
785 | int zid = page_zonenum(page); | |
786 | mctz = soft_limit_tree_from_page(page); | |
787 | ||
788 | /* | |
789 | * Necessary to update all ancestors when hierarchy is used. | |
790 | * because their event counter is not touched. | |
791 | */ | |
792 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
793 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
794 | excess = res_counter_soft_limit_excess(&memcg->res); | |
795 | /* | |
796 | * We have to update the tree if mz is on RB-tree or | |
797 | * mem is over its softlimit. | |
798 | */ | |
799 | if (excess || mz->on_tree) { | |
800 | spin_lock(&mctz->lock); | |
801 | /* if on-tree, remove it */ | |
802 | if (mz->on_tree) | |
803 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); | |
804 | /* | |
805 | * Insert again. mz->usage_in_excess will be updated. | |
806 | * If excess is 0, no tree ops. | |
807 | */ | |
808 | __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess); | |
809 | spin_unlock(&mctz->lock); | |
810 | } | |
811 | } | |
812 | } | |
813 | ||
814 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
815 | { | |
816 | int node, zone; | |
817 | struct mem_cgroup_per_zone *mz; | |
818 | struct mem_cgroup_tree_per_zone *mctz; | |
819 | ||
820 | for_each_node(node) { | |
821 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
822 | mz = mem_cgroup_zoneinfo(memcg, node, zone); | |
823 | mctz = soft_limit_tree_node_zone(node, zone); | |
824 | mem_cgroup_remove_exceeded(memcg, mz, mctz); | |
825 | } | |
826 | } | |
827 | } | |
828 | ||
829 | static struct mem_cgroup_per_zone * | |
830 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
831 | { | |
832 | struct rb_node *rightmost = NULL; | |
833 | struct mem_cgroup_per_zone *mz; | |
834 | ||
835 | retry: | |
836 | mz = NULL; | |
837 | rightmost = rb_last(&mctz->rb_root); | |
838 | if (!rightmost) | |
839 | goto done; /* Nothing to reclaim from */ | |
840 | ||
841 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
842 | /* | |
843 | * Remove the node now but someone else can add it back, | |
844 | * we will to add it back at the end of reclaim to its correct | |
845 | * position in the tree. | |
846 | */ | |
847 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); | |
848 | if (!res_counter_soft_limit_excess(&mz->memcg->res) || | |
849 | !css_tryget(&mz->memcg->css)) | |
850 | goto retry; | |
851 | done: | |
852 | return mz; | |
853 | } | |
854 | ||
855 | static struct mem_cgroup_per_zone * | |
856 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
857 | { | |
858 | struct mem_cgroup_per_zone *mz; | |
859 | ||
860 | spin_lock(&mctz->lock); | |
861 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
862 | spin_unlock(&mctz->lock); | |
863 | return mz; | |
864 | } | |
865 | ||
711d3d2c KH |
866 | /* |
867 | * Implementation Note: reading percpu statistics for memcg. | |
868 | * | |
869 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
870 | * synchronization to implement "quick" read. There are trade-off between | |
871 | * reading cost and precision of value. Then, we may have a chance to implement | |
872 | * a periodic synchronizion of counter in memcg's counter. | |
873 | * | |
874 | * But this _read() function is used for user interface now. The user accounts | |
875 | * memory usage by memory cgroup and he _always_ requires exact value because | |
876 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
877 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
878 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
879 | * | |
880 | * If there are kernel internal actions which can make use of some not-exact | |
881 | * value, and reading all cpu value can be performance bottleneck in some | |
882 | * common workload, threashold and synchonization as vmstat[] should be | |
883 | * implemented. | |
884 | */ | |
c0ff4b85 | 885 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 886 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 887 | { |
7a159cc9 | 888 | long val = 0; |
c62b1a3b | 889 | int cpu; |
c62b1a3b | 890 | |
711d3d2c KH |
891 | get_online_cpus(); |
892 | for_each_online_cpu(cpu) | |
c0ff4b85 | 893 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 894 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
895 | spin_lock(&memcg->pcp_counter_lock); |
896 | val += memcg->nocpu_base.count[idx]; | |
897 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
898 | #endif |
899 | put_online_cpus(); | |
c62b1a3b KH |
900 | return val; |
901 | } | |
902 | ||
c0ff4b85 | 903 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
0c3e73e8 BS |
904 | bool charge) |
905 | { | |
906 | int val = (charge) ? 1 : -1; | |
bff6bb83 | 907 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); |
0c3e73e8 BS |
908 | } |
909 | ||
c0ff4b85 | 910 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
911 | enum mem_cgroup_events_index idx) |
912 | { | |
913 | unsigned long val = 0; | |
914 | int cpu; | |
915 | ||
9c567512 | 916 | get_online_cpus(); |
e9f8974f | 917 | for_each_online_cpu(cpu) |
c0ff4b85 | 918 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 919 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
920 | spin_lock(&memcg->pcp_counter_lock); |
921 | val += memcg->nocpu_base.events[idx]; | |
922 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f | 923 | #endif |
9c567512 | 924 | put_online_cpus(); |
e9f8974f JW |
925 | return val; |
926 | } | |
927 | ||
c0ff4b85 | 928 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 929 | struct page *page, |
b2402857 | 930 | bool anon, int nr_pages) |
d52aa412 | 931 | { |
c62b1a3b KH |
932 | preempt_disable(); |
933 | ||
b2402857 KH |
934 | /* |
935 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
936 | * counted as CACHE even if it's on ANON LRU. | |
937 | */ | |
938 | if (anon) | |
939 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], | |
c0ff4b85 | 940 | nr_pages); |
d52aa412 | 941 | else |
b2402857 | 942 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 943 | nr_pages); |
55e462b0 | 944 | |
b070e65c DR |
945 | if (PageTransHuge(page)) |
946 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], | |
947 | nr_pages); | |
948 | ||
e401f176 KH |
949 | /* pagein of a big page is an event. So, ignore page size */ |
950 | if (nr_pages > 0) | |
c0ff4b85 | 951 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 952 | else { |
c0ff4b85 | 953 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
954 | nr_pages = -nr_pages; /* for event */ |
955 | } | |
e401f176 | 956 | |
13114716 | 957 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
2e72b634 | 958 | |
c62b1a3b | 959 | preempt_enable(); |
6d12e2d8 KH |
960 | } |
961 | ||
bb2a0de9 | 962 | unsigned long |
4d7dcca2 | 963 | mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
964 | { |
965 | struct mem_cgroup_per_zone *mz; | |
966 | ||
967 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
968 | return mz->lru_size[lru]; | |
969 | } | |
970 | ||
971 | static unsigned long | |
c0ff4b85 | 972 | mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid, |
bb2a0de9 | 973 | unsigned int lru_mask) |
889976db YH |
974 | { |
975 | struct mem_cgroup_per_zone *mz; | |
f156ab93 | 976 | enum lru_list lru; |
bb2a0de9 KH |
977 | unsigned long ret = 0; |
978 | ||
c0ff4b85 | 979 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
bb2a0de9 | 980 | |
f156ab93 HD |
981 | for_each_lru(lru) { |
982 | if (BIT(lru) & lru_mask) | |
983 | ret += mz->lru_size[lru]; | |
bb2a0de9 KH |
984 | } |
985 | return ret; | |
986 | } | |
987 | ||
988 | static unsigned long | |
c0ff4b85 | 989 | mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 KH |
990 | int nid, unsigned int lru_mask) |
991 | { | |
889976db YH |
992 | u64 total = 0; |
993 | int zid; | |
994 | ||
bb2a0de9 | 995 | for (zid = 0; zid < MAX_NR_ZONES; zid++) |
c0ff4b85 R |
996 | total += mem_cgroup_zone_nr_lru_pages(memcg, |
997 | nid, zid, lru_mask); | |
bb2a0de9 | 998 | |
889976db YH |
999 | return total; |
1000 | } | |
bb2a0de9 | 1001 | |
c0ff4b85 | 1002 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 1003 | unsigned int lru_mask) |
6d12e2d8 | 1004 | { |
889976db | 1005 | int nid; |
6d12e2d8 KH |
1006 | u64 total = 0; |
1007 | ||
31aaea4a | 1008 | for_each_node_state(nid, N_MEMORY) |
c0ff4b85 | 1009 | total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
6d12e2d8 | 1010 | return total; |
d52aa412 KH |
1011 | } |
1012 | ||
f53d7ce3 JW |
1013 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
1014 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
1015 | { |
1016 | unsigned long val, next; | |
1017 | ||
13114716 | 1018 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 1019 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 1020 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
1021 | if ((long)next - (long)val < 0) { |
1022 | switch (target) { | |
1023 | case MEM_CGROUP_TARGET_THRESH: | |
1024 | next = val + THRESHOLDS_EVENTS_TARGET; | |
1025 | break; | |
bb4cc1a8 AM |
1026 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
1027 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
1028 | break; | |
f53d7ce3 JW |
1029 | case MEM_CGROUP_TARGET_NUMAINFO: |
1030 | next = val + NUMAINFO_EVENTS_TARGET; | |
1031 | break; | |
1032 | default: | |
1033 | break; | |
1034 | } | |
1035 | __this_cpu_write(memcg->stat->targets[target], next); | |
1036 | return true; | |
7a159cc9 | 1037 | } |
f53d7ce3 | 1038 | return false; |
d2265e6f KH |
1039 | } |
1040 | ||
1041 | /* | |
1042 | * Check events in order. | |
1043 | * | |
1044 | */ | |
c0ff4b85 | 1045 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f | 1046 | { |
4799401f | 1047 | preempt_disable(); |
d2265e6f | 1048 | /* threshold event is triggered in finer grain than soft limit */ |
f53d7ce3 JW |
1049 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
1050 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 1051 | bool do_softlimit; |
82b3f2a7 | 1052 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 1053 | |
bb4cc1a8 AM |
1054 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
1055 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
1056 | #if MAX_NUMNODES > 1 |
1057 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
1058 | MEM_CGROUP_TARGET_NUMAINFO); | |
1059 | #endif | |
1060 | preempt_enable(); | |
1061 | ||
c0ff4b85 | 1062 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
1063 | if (unlikely(do_softlimit)) |
1064 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 1065 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 1066 | if (unlikely(do_numainfo)) |
c0ff4b85 | 1067 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 1068 | #endif |
f53d7ce3 JW |
1069 | } else |
1070 | preempt_enable(); | |
d2265e6f KH |
1071 | } |
1072 | ||
cf475ad2 | 1073 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 1074 | { |
31a78f23 BS |
1075 | /* |
1076 | * mm_update_next_owner() may clear mm->owner to NULL | |
1077 | * if it races with swapoff, page migration, etc. | |
1078 | * So this can be called with p == NULL. | |
1079 | */ | |
1080 | if (unlikely(!p)) | |
1081 | return NULL; | |
1082 | ||
8af01f56 | 1083 | return mem_cgroup_from_css(task_css(p, mem_cgroup_subsys_id)); |
78fb7466 PE |
1084 | } |
1085 | ||
a433658c | 1086 | struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 1087 | { |
c0ff4b85 | 1088 | struct mem_cgroup *memcg = NULL; |
0b7f569e KH |
1089 | |
1090 | if (!mm) | |
1091 | return NULL; | |
54595fe2 KH |
1092 | /* |
1093 | * Because we have no locks, mm->owner's may be being moved to other | |
1094 | * cgroup. We use css_tryget() here even if this looks | |
1095 | * pessimistic (rather than adding locks here). | |
1096 | */ | |
1097 | rcu_read_lock(); | |
1098 | do { | |
c0ff4b85 R |
1099 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1100 | if (unlikely(!memcg)) | |
54595fe2 | 1101 | break; |
c0ff4b85 | 1102 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 1103 | rcu_read_unlock(); |
c0ff4b85 | 1104 | return memcg; |
54595fe2 KH |
1105 | } |
1106 | ||
16248d8f MH |
1107 | /* |
1108 | * Returns a next (in a pre-order walk) alive memcg (with elevated css | |
1109 | * ref. count) or NULL if the whole root's subtree has been visited. | |
1110 | * | |
1111 | * helper function to be used by mem_cgroup_iter | |
1112 | */ | |
1113 | static struct mem_cgroup *__mem_cgroup_iter_next(struct mem_cgroup *root, | |
694fbc0f | 1114 | struct mem_cgroup *last_visited) |
16248d8f | 1115 | { |
492eb21b | 1116 | struct cgroup_subsys_state *prev_css, *next_css; |
16248d8f | 1117 | |
bd8815a6 | 1118 | prev_css = last_visited ? &last_visited->css : NULL; |
16248d8f | 1119 | skip_node: |
492eb21b | 1120 | next_css = css_next_descendant_pre(prev_css, &root->css); |
16248d8f MH |
1121 | |
1122 | /* | |
1123 | * Even if we found a group we have to make sure it is | |
1124 | * alive. css && !memcg means that the groups should be | |
1125 | * skipped and we should continue the tree walk. | |
1126 | * last_visited css is safe to use because it is | |
1127 | * protected by css_get and the tree walk is rcu safe. | |
1128 | */ | |
492eb21b TH |
1129 | if (next_css) { |
1130 | struct mem_cgroup *mem = mem_cgroup_from_css(next_css); | |
1131 | ||
694fbc0f AM |
1132 | if (css_tryget(&mem->css)) |
1133 | return mem; | |
1134 | else { | |
492eb21b | 1135 | prev_css = next_css; |
16248d8f MH |
1136 | goto skip_node; |
1137 | } | |
1138 | } | |
1139 | ||
1140 | return NULL; | |
1141 | } | |
1142 | ||
519ebea3 JW |
1143 | static void mem_cgroup_iter_invalidate(struct mem_cgroup *root) |
1144 | { | |
1145 | /* | |
1146 | * When a group in the hierarchy below root is destroyed, the | |
1147 | * hierarchy iterator can no longer be trusted since it might | |
1148 | * have pointed to the destroyed group. Invalidate it. | |
1149 | */ | |
1150 | atomic_inc(&root->dead_count); | |
1151 | } | |
1152 | ||
1153 | static struct mem_cgroup * | |
1154 | mem_cgroup_iter_load(struct mem_cgroup_reclaim_iter *iter, | |
1155 | struct mem_cgroup *root, | |
1156 | int *sequence) | |
1157 | { | |
1158 | struct mem_cgroup *position = NULL; | |
1159 | /* | |
1160 | * A cgroup destruction happens in two stages: offlining and | |
1161 | * release. They are separated by a RCU grace period. | |
1162 | * | |
1163 | * If the iterator is valid, we may still race with an | |
1164 | * offlining. The RCU lock ensures the object won't be | |
1165 | * released, tryget will fail if we lost the race. | |
1166 | */ | |
1167 | *sequence = atomic_read(&root->dead_count); | |
1168 | if (iter->last_dead_count == *sequence) { | |
1169 | smp_rmb(); | |
1170 | position = iter->last_visited; | |
1171 | if (position && !css_tryget(&position->css)) | |
1172 | position = NULL; | |
1173 | } | |
1174 | return position; | |
1175 | } | |
1176 | ||
1177 | static void mem_cgroup_iter_update(struct mem_cgroup_reclaim_iter *iter, | |
1178 | struct mem_cgroup *last_visited, | |
1179 | struct mem_cgroup *new_position, | |
1180 | int sequence) | |
1181 | { | |
1182 | if (last_visited) | |
1183 | css_put(&last_visited->css); | |
1184 | /* | |
1185 | * We store the sequence count from the time @last_visited was | |
1186 | * loaded successfully instead of rereading it here so that we | |
1187 | * don't lose destruction events in between. We could have | |
1188 | * raced with the destruction of @new_position after all. | |
1189 | */ | |
1190 | iter->last_visited = new_position; | |
1191 | smp_wmb(); | |
1192 | iter->last_dead_count = sequence; | |
1193 | } | |
1194 | ||
5660048c JW |
1195 | /** |
1196 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1197 | * @root: hierarchy root | |
1198 | * @prev: previously returned memcg, NULL on first invocation | |
1199 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1200 | * | |
1201 | * Returns references to children of the hierarchy below @root, or | |
1202 | * @root itself, or %NULL after a full round-trip. | |
1203 | * | |
1204 | * Caller must pass the return value in @prev on subsequent | |
1205 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1206 | * to cancel a hierarchy walk before the round-trip is complete. | |
1207 | * | |
1208 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
1209 | * divide up the memcgs in the hierarchy among all concurrent | |
1210 | * reclaimers operating on the same zone and priority. | |
1211 | */ | |
694fbc0f | 1212 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1213 | struct mem_cgroup *prev, |
694fbc0f | 1214 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1215 | { |
9f3a0d09 | 1216 | struct mem_cgroup *memcg = NULL; |
542f85f9 | 1217 | struct mem_cgroup *last_visited = NULL; |
711d3d2c | 1218 | |
694fbc0f AM |
1219 | if (mem_cgroup_disabled()) |
1220 | return NULL; | |
5660048c | 1221 | |
9f3a0d09 JW |
1222 | if (!root) |
1223 | root = root_mem_cgroup; | |
7d74b06f | 1224 | |
9f3a0d09 | 1225 | if (prev && !reclaim) |
542f85f9 | 1226 | last_visited = prev; |
14067bb3 | 1227 | |
9f3a0d09 JW |
1228 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1229 | if (prev) | |
c40046f3 | 1230 | goto out_css_put; |
694fbc0f | 1231 | return root; |
9f3a0d09 | 1232 | } |
14067bb3 | 1233 | |
542f85f9 | 1234 | rcu_read_lock(); |
9f3a0d09 | 1235 | while (!memcg) { |
527a5ec9 | 1236 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
519ebea3 | 1237 | int uninitialized_var(seq); |
711d3d2c | 1238 | |
527a5ec9 JW |
1239 | if (reclaim) { |
1240 | int nid = zone_to_nid(reclaim->zone); | |
1241 | int zid = zone_idx(reclaim->zone); | |
1242 | struct mem_cgroup_per_zone *mz; | |
1243 | ||
1244 | mz = mem_cgroup_zoneinfo(root, nid, zid); | |
1245 | iter = &mz->reclaim_iter[reclaim->priority]; | |
542f85f9 | 1246 | if (prev && reclaim->generation != iter->generation) { |
5f578161 | 1247 | iter->last_visited = NULL; |
542f85f9 MH |
1248 | goto out_unlock; |
1249 | } | |
5f578161 | 1250 | |
519ebea3 | 1251 | last_visited = mem_cgroup_iter_load(iter, root, &seq); |
527a5ec9 | 1252 | } |
7d74b06f | 1253 | |
694fbc0f | 1254 | memcg = __mem_cgroup_iter_next(root, last_visited); |
14067bb3 | 1255 | |
527a5ec9 | 1256 | if (reclaim) { |
519ebea3 | 1257 | mem_cgroup_iter_update(iter, last_visited, memcg, seq); |
542f85f9 | 1258 | |
19f39402 | 1259 | if (!memcg) |
527a5ec9 JW |
1260 | iter->generation++; |
1261 | else if (!prev && memcg) | |
1262 | reclaim->generation = iter->generation; | |
1263 | } | |
9f3a0d09 | 1264 | |
694fbc0f | 1265 | if (prev && !memcg) |
542f85f9 | 1266 | goto out_unlock; |
9f3a0d09 | 1267 | } |
542f85f9 MH |
1268 | out_unlock: |
1269 | rcu_read_unlock(); | |
c40046f3 MH |
1270 | out_css_put: |
1271 | if (prev && prev != root) | |
1272 | css_put(&prev->css); | |
1273 | ||
9f3a0d09 | 1274 | return memcg; |
14067bb3 | 1275 | } |
7d74b06f | 1276 | |
5660048c JW |
1277 | /** |
1278 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1279 | * @root: hierarchy root | |
1280 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1281 | */ | |
1282 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1283 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1284 | { |
1285 | if (!root) | |
1286 | root = root_mem_cgroup; | |
1287 | if (prev && prev != root) | |
1288 | css_put(&prev->css); | |
1289 | } | |
7d74b06f | 1290 | |
9f3a0d09 JW |
1291 | /* |
1292 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1293 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1294 | * be used for reference counting. | |
1295 | */ | |
1296 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1297 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1298 | iter != NULL; \ |
527a5ec9 | 1299 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1300 | |
9f3a0d09 | 1301 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1302 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1303 | iter != NULL; \ |
527a5ec9 | 1304 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1305 | |
68ae564b | 1306 | void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
456f998e | 1307 | { |
c0ff4b85 | 1308 | struct mem_cgroup *memcg; |
456f998e | 1309 | |
456f998e | 1310 | rcu_read_lock(); |
c0ff4b85 R |
1311 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1312 | if (unlikely(!memcg)) | |
456f998e YH |
1313 | goto out; |
1314 | ||
1315 | switch (idx) { | |
456f998e | 1316 | case PGFAULT: |
0e574a93 JW |
1317 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1318 | break; | |
1319 | case PGMAJFAULT: | |
1320 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1321 | break; |
1322 | default: | |
1323 | BUG(); | |
1324 | } | |
1325 | out: | |
1326 | rcu_read_unlock(); | |
1327 | } | |
68ae564b | 1328 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
456f998e | 1329 | |
925b7673 JW |
1330 | /** |
1331 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1332 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1333 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1334 | * |
1335 | * Returns the lru list vector holding pages for the given @zone and | |
1336 | * @mem. This can be the global zone lruvec, if the memory controller | |
1337 | * is disabled. | |
1338 | */ | |
1339 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1340 | struct mem_cgroup *memcg) | |
1341 | { | |
1342 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1343 | struct lruvec *lruvec; |
925b7673 | 1344 | |
bea8c150 HD |
1345 | if (mem_cgroup_disabled()) { |
1346 | lruvec = &zone->lruvec; | |
1347 | goto out; | |
1348 | } | |
925b7673 JW |
1349 | |
1350 | mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone)); | |
bea8c150 HD |
1351 | lruvec = &mz->lruvec; |
1352 | out: | |
1353 | /* | |
1354 | * Since a node can be onlined after the mem_cgroup was created, | |
1355 | * we have to be prepared to initialize lruvec->zone here; | |
1356 | * and if offlined then reonlined, we need to reinitialize it. | |
1357 | */ | |
1358 | if (unlikely(lruvec->zone != zone)) | |
1359 | lruvec->zone = zone; | |
1360 | return lruvec; | |
925b7673 JW |
1361 | } |
1362 | ||
08e552c6 KH |
1363 | /* |
1364 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
1365 | * Operations are called by routine of global LRU independently from memcg. | |
1366 | * What we have to take care of here is validness of pc->mem_cgroup. | |
1367 | * | |
1368 | * Changes to pc->mem_cgroup happens when | |
1369 | * 1. charge | |
1370 | * 2. moving account | |
1371 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
1372 | * It is added to LRU before charge. | |
1373 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
1374 | * When moving account, the page is not on LRU. It's isolated. | |
1375 | */ | |
4f98a2fe | 1376 | |
925b7673 | 1377 | /** |
fa9add64 | 1378 | * mem_cgroup_page_lruvec - return lruvec for adding an lru page |
925b7673 | 1379 | * @page: the page |
fa9add64 | 1380 | * @zone: zone of the page |
925b7673 | 1381 | */ |
fa9add64 | 1382 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1383 | { |
08e552c6 | 1384 | struct mem_cgroup_per_zone *mz; |
925b7673 JW |
1385 | struct mem_cgroup *memcg; |
1386 | struct page_cgroup *pc; | |
bea8c150 | 1387 | struct lruvec *lruvec; |
6d12e2d8 | 1388 | |
bea8c150 HD |
1389 | if (mem_cgroup_disabled()) { |
1390 | lruvec = &zone->lruvec; | |
1391 | goto out; | |
1392 | } | |
925b7673 | 1393 | |
08e552c6 | 1394 | pc = lookup_page_cgroup(page); |
38c5d72f | 1395 | memcg = pc->mem_cgroup; |
7512102c HD |
1396 | |
1397 | /* | |
fa9add64 | 1398 | * Surreptitiously switch any uncharged offlist page to root: |
7512102c HD |
1399 | * an uncharged page off lru does nothing to secure |
1400 | * its former mem_cgroup from sudden removal. | |
1401 | * | |
1402 | * Our caller holds lru_lock, and PageCgroupUsed is updated | |
1403 | * under page_cgroup lock: between them, they make all uses | |
1404 | * of pc->mem_cgroup safe. | |
1405 | */ | |
fa9add64 | 1406 | if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup) |
7512102c HD |
1407 | pc->mem_cgroup = memcg = root_mem_cgroup; |
1408 | ||
925b7673 | 1409 | mz = page_cgroup_zoneinfo(memcg, page); |
bea8c150 HD |
1410 | lruvec = &mz->lruvec; |
1411 | out: | |
1412 | /* | |
1413 | * Since a node can be onlined after the mem_cgroup was created, | |
1414 | * we have to be prepared to initialize lruvec->zone here; | |
1415 | * and if offlined then reonlined, we need to reinitialize it. | |
1416 | */ | |
1417 | if (unlikely(lruvec->zone != zone)) | |
1418 | lruvec->zone = zone; | |
1419 | return lruvec; | |
08e552c6 | 1420 | } |
b69408e8 | 1421 | |
925b7673 | 1422 | /** |
fa9add64 HD |
1423 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1424 | * @lruvec: mem_cgroup per zone lru vector | |
1425 | * @lru: index of lru list the page is sitting on | |
1426 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1427 | * |
fa9add64 HD |
1428 | * This function must be called when a page is added to or removed from an |
1429 | * lru list. | |
3f58a829 | 1430 | */ |
fa9add64 HD |
1431 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1432 | int nr_pages) | |
3f58a829 MK |
1433 | { |
1434 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1435 | unsigned long *lru_size; |
3f58a829 MK |
1436 | |
1437 | if (mem_cgroup_disabled()) | |
1438 | return; | |
1439 | ||
fa9add64 HD |
1440 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1441 | lru_size = mz->lru_size + lru; | |
1442 | *lru_size += nr_pages; | |
1443 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1444 | } |
544122e5 | 1445 | |
3e92041d | 1446 | /* |
c0ff4b85 | 1447 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1448 | * hierarchy subtree |
1449 | */ | |
c3ac9a8a JW |
1450 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1451 | struct mem_cgroup *memcg) | |
3e92041d | 1452 | { |
91c63734 JW |
1453 | if (root_memcg == memcg) |
1454 | return true; | |
3a981f48 | 1455 | if (!root_memcg->use_hierarchy || !memcg) |
91c63734 | 1456 | return false; |
c3ac9a8a JW |
1457 | return css_is_ancestor(&memcg->css, &root_memcg->css); |
1458 | } | |
1459 | ||
1460 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, | |
1461 | struct mem_cgroup *memcg) | |
1462 | { | |
1463 | bool ret; | |
1464 | ||
91c63734 | 1465 | rcu_read_lock(); |
c3ac9a8a | 1466 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
91c63734 JW |
1467 | rcu_read_unlock(); |
1468 | return ret; | |
3e92041d MH |
1469 | } |
1470 | ||
ffbdccf5 DR |
1471 | bool task_in_mem_cgroup(struct task_struct *task, |
1472 | const struct mem_cgroup *memcg) | |
4c4a2214 | 1473 | { |
0b7f569e | 1474 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1475 | struct task_struct *p; |
ffbdccf5 | 1476 | bool ret; |
4c4a2214 | 1477 | |
158e0a2d | 1478 | p = find_lock_task_mm(task); |
de077d22 DR |
1479 | if (p) { |
1480 | curr = try_get_mem_cgroup_from_mm(p->mm); | |
1481 | task_unlock(p); | |
1482 | } else { | |
1483 | /* | |
1484 | * All threads may have already detached their mm's, but the oom | |
1485 | * killer still needs to detect if they have already been oom | |
1486 | * killed to prevent needlessly killing additional tasks. | |
1487 | */ | |
ffbdccf5 | 1488 | rcu_read_lock(); |
de077d22 DR |
1489 | curr = mem_cgroup_from_task(task); |
1490 | if (curr) | |
1491 | css_get(&curr->css); | |
ffbdccf5 | 1492 | rcu_read_unlock(); |
de077d22 | 1493 | } |
0b7f569e | 1494 | if (!curr) |
ffbdccf5 | 1495 | return false; |
d31f56db | 1496 | /* |
c0ff4b85 | 1497 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1498 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1499 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1500 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1501 | */ |
c0ff4b85 | 1502 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1503 | css_put(&curr->css); |
4c4a2214 DR |
1504 | return ret; |
1505 | } | |
1506 | ||
c56d5c7d | 1507 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1508 | { |
9b272977 | 1509 | unsigned long inactive_ratio; |
14797e23 | 1510 | unsigned long inactive; |
9b272977 | 1511 | unsigned long active; |
c772be93 | 1512 | unsigned long gb; |
14797e23 | 1513 | |
4d7dcca2 HD |
1514 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1515 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1516 | |
c772be93 KM |
1517 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1518 | if (gb) | |
1519 | inactive_ratio = int_sqrt(10 * gb); | |
1520 | else | |
1521 | inactive_ratio = 1; | |
1522 | ||
9b272977 | 1523 | return inactive * inactive_ratio < active; |
14797e23 KM |
1524 | } |
1525 | ||
6d61ef40 BS |
1526 | #define mem_cgroup_from_res_counter(counter, member) \ |
1527 | container_of(counter, struct mem_cgroup, member) | |
1528 | ||
19942822 | 1529 | /** |
9d11ea9f | 1530 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1531 | * @memcg: the memory cgroup |
19942822 | 1532 | * |
9d11ea9f | 1533 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1534 | * pages. |
19942822 | 1535 | */ |
c0ff4b85 | 1536 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1537 | { |
9d11ea9f JW |
1538 | unsigned long long margin; |
1539 | ||
c0ff4b85 | 1540 | margin = res_counter_margin(&memcg->res); |
9d11ea9f | 1541 | if (do_swap_account) |
c0ff4b85 | 1542 | margin = min(margin, res_counter_margin(&memcg->memsw)); |
7ec99d62 | 1543 | return margin >> PAGE_SHIFT; |
19942822 JW |
1544 | } |
1545 | ||
1f4c025b | 1546 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 | 1547 | { |
a7885eb8 | 1548 | /* root ? */ |
63876986 | 1549 | if (!css_parent(&memcg->css)) |
a7885eb8 KM |
1550 | return vm_swappiness; |
1551 | ||
bf1ff263 | 1552 | return memcg->swappiness; |
a7885eb8 KM |
1553 | } |
1554 | ||
619d094b KH |
1555 | /* |
1556 | * memcg->moving_account is used for checking possibility that some thread is | |
1557 | * calling move_account(). When a thread on CPU-A starts moving pages under | |
1558 | * a memcg, other threads should check memcg->moving_account under | |
1559 | * rcu_read_lock(), like this: | |
1560 | * | |
1561 | * CPU-A CPU-B | |
1562 | * rcu_read_lock() | |
1563 | * memcg->moving_account+1 if (memcg->mocing_account) | |
1564 | * take heavy locks. | |
1565 | * synchronize_rcu() update something. | |
1566 | * rcu_read_unlock() | |
1567 | * start move here. | |
1568 | */ | |
4331f7d3 KH |
1569 | |
1570 | /* for quick checking without looking up memcg */ | |
1571 | atomic_t memcg_moving __read_mostly; | |
1572 | ||
c0ff4b85 | 1573 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a | 1574 | { |
4331f7d3 | 1575 | atomic_inc(&memcg_moving); |
619d094b | 1576 | atomic_inc(&memcg->moving_account); |
32047e2a KH |
1577 | synchronize_rcu(); |
1578 | } | |
1579 | ||
c0ff4b85 | 1580 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a | 1581 | { |
619d094b KH |
1582 | /* |
1583 | * Now, mem_cgroup_clear_mc() may call this function with NULL. | |
1584 | * We check NULL in callee rather than caller. | |
1585 | */ | |
4331f7d3 KH |
1586 | if (memcg) { |
1587 | atomic_dec(&memcg_moving); | |
619d094b | 1588 | atomic_dec(&memcg->moving_account); |
4331f7d3 | 1589 | } |
32047e2a | 1590 | } |
619d094b | 1591 | |
32047e2a KH |
1592 | /* |
1593 | * 2 routines for checking "mem" is under move_account() or not. | |
1594 | * | |
13fd1dd9 AM |
1595 | * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This |
1596 | * is used for avoiding races in accounting. If true, | |
32047e2a KH |
1597 | * pc->mem_cgroup may be overwritten. |
1598 | * | |
1599 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or | |
1600 | * under hierarchy of moving cgroups. This is for | |
1601 | * waiting at hith-memory prressure caused by "move". | |
1602 | */ | |
1603 | ||
13fd1dd9 | 1604 | static bool mem_cgroup_stolen(struct mem_cgroup *memcg) |
32047e2a KH |
1605 | { |
1606 | VM_BUG_ON(!rcu_read_lock_held()); | |
619d094b | 1607 | return atomic_read(&memcg->moving_account) > 0; |
32047e2a | 1608 | } |
4b534334 | 1609 | |
c0ff4b85 | 1610 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1611 | { |
2bd9bb20 KH |
1612 | struct mem_cgroup *from; |
1613 | struct mem_cgroup *to; | |
4b534334 | 1614 | bool ret = false; |
2bd9bb20 KH |
1615 | /* |
1616 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1617 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1618 | */ | |
1619 | spin_lock(&mc.lock); | |
1620 | from = mc.from; | |
1621 | to = mc.to; | |
1622 | if (!from) | |
1623 | goto unlock; | |
3e92041d | 1624 | |
c0ff4b85 R |
1625 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1626 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1627 | unlock: |
1628 | spin_unlock(&mc.lock); | |
4b534334 KH |
1629 | return ret; |
1630 | } | |
1631 | ||
c0ff4b85 | 1632 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1633 | { |
1634 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1635 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1636 | DEFINE_WAIT(wait); |
1637 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1638 | /* moving charge context might have finished. */ | |
1639 | if (mc.moving_task) | |
1640 | schedule(); | |
1641 | finish_wait(&mc.waitq, &wait); | |
1642 | return true; | |
1643 | } | |
1644 | } | |
1645 | return false; | |
1646 | } | |
1647 | ||
312734c0 KH |
1648 | /* |
1649 | * Take this lock when | |
1650 | * - a code tries to modify page's memcg while it's USED. | |
1651 | * - a code tries to modify page state accounting in a memcg. | |
13fd1dd9 | 1652 | * see mem_cgroup_stolen(), too. |
312734c0 KH |
1653 | */ |
1654 | static void move_lock_mem_cgroup(struct mem_cgroup *memcg, | |
1655 | unsigned long *flags) | |
1656 | { | |
1657 | spin_lock_irqsave(&memcg->move_lock, *flags); | |
1658 | } | |
1659 | ||
1660 | static void move_unlock_mem_cgroup(struct mem_cgroup *memcg, | |
1661 | unsigned long *flags) | |
1662 | { | |
1663 | spin_unlock_irqrestore(&memcg->move_lock, *flags); | |
1664 | } | |
1665 | ||
58cf188e | 1666 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1667 | /** |
58cf188e | 1668 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1669 | * @memcg: The memory cgroup that went over limit |
1670 | * @p: Task that is going to be killed | |
1671 | * | |
1672 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1673 | * enabled | |
1674 | */ | |
1675 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1676 | { | |
1677 | struct cgroup *task_cgrp; | |
1678 | struct cgroup *mem_cgrp; | |
1679 | /* | |
1680 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1681 | * on the assumption that OOM is serialized for memory controller. | |
1682 | * If this assumption is broken, revisit this code. | |
1683 | */ | |
1684 | static char memcg_name[PATH_MAX]; | |
1685 | int ret; | |
58cf188e SZ |
1686 | struct mem_cgroup *iter; |
1687 | unsigned int i; | |
e222432b | 1688 | |
58cf188e | 1689 | if (!p) |
e222432b BS |
1690 | return; |
1691 | ||
e222432b BS |
1692 | rcu_read_lock(); |
1693 | ||
1694 | mem_cgrp = memcg->css.cgroup; | |
1695 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1696 | ||
1697 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1698 | if (ret < 0) { | |
1699 | /* | |
1700 | * Unfortunately, we are unable to convert to a useful name | |
1701 | * But we'll still print out the usage information | |
1702 | */ | |
1703 | rcu_read_unlock(); | |
1704 | goto done; | |
1705 | } | |
1706 | rcu_read_unlock(); | |
1707 | ||
d045197f | 1708 | pr_info("Task in %s killed", memcg_name); |
e222432b BS |
1709 | |
1710 | rcu_read_lock(); | |
1711 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1712 | if (ret < 0) { | |
1713 | rcu_read_unlock(); | |
1714 | goto done; | |
1715 | } | |
1716 | rcu_read_unlock(); | |
1717 | ||
1718 | /* | |
1719 | * Continues from above, so we don't need an KERN_ level | |
1720 | */ | |
d045197f | 1721 | pr_cont(" as a result of limit of %s\n", memcg_name); |
e222432b BS |
1722 | done: |
1723 | ||
d045197f | 1724 | pr_info("memory: usage %llukB, limit %llukB, failcnt %llu\n", |
e222432b BS |
1725 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, |
1726 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1727 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
d045197f | 1728 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %llu\n", |
e222432b BS |
1729 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, |
1730 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1731 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
d045197f | 1732 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %llu\n", |
510fc4e1 GC |
1733 | res_counter_read_u64(&memcg->kmem, RES_USAGE) >> 10, |
1734 | res_counter_read_u64(&memcg->kmem, RES_LIMIT) >> 10, | |
1735 | res_counter_read_u64(&memcg->kmem, RES_FAILCNT)); | |
58cf188e SZ |
1736 | |
1737 | for_each_mem_cgroup_tree(iter, memcg) { | |
1738 | pr_info("Memory cgroup stats"); | |
1739 | ||
1740 | rcu_read_lock(); | |
1741 | ret = cgroup_path(iter->css.cgroup, memcg_name, PATH_MAX); | |
1742 | if (!ret) | |
1743 | pr_cont(" for %s", memcg_name); | |
1744 | rcu_read_unlock(); | |
1745 | pr_cont(":"); | |
1746 | ||
1747 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1748 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1749 | continue; | |
1750 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], | |
1751 | K(mem_cgroup_read_stat(iter, i))); | |
1752 | } | |
1753 | ||
1754 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1755 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1756 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1757 | ||
1758 | pr_cont("\n"); | |
1759 | } | |
e222432b BS |
1760 | } |
1761 | ||
81d39c20 KH |
1762 | /* |
1763 | * This function returns the number of memcg under hierarchy tree. Returns | |
1764 | * 1(self count) if no children. | |
1765 | */ | |
c0ff4b85 | 1766 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1767 | { |
1768 | int num = 0; | |
7d74b06f KH |
1769 | struct mem_cgroup *iter; |
1770 | ||
c0ff4b85 | 1771 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1772 | num++; |
81d39c20 KH |
1773 | return num; |
1774 | } | |
1775 | ||
a63d83f4 DR |
1776 | /* |
1777 | * Return the memory (and swap, if configured) limit for a memcg. | |
1778 | */ | |
9cbb78bb | 1779 | static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 DR |
1780 | { |
1781 | u64 limit; | |
a63d83f4 | 1782 | |
f3e8eb70 | 1783 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
f3e8eb70 | 1784 | |
a63d83f4 | 1785 | /* |
9a5a8f19 | 1786 | * Do not consider swap space if we cannot swap due to swappiness |
a63d83f4 | 1787 | */ |
9a5a8f19 MH |
1788 | if (mem_cgroup_swappiness(memcg)) { |
1789 | u64 memsw; | |
1790 | ||
1791 | limit += total_swap_pages << PAGE_SHIFT; | |
1792 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1793 | ||
1794 | /* | |
1795 | * If memsw is finite and limits the amount of swap space | |
1796 | * available to this memcg, return that limit. | |
1797 | */ | |
1798 | limit = min(limit, memsw); | |
1799 | } | |
1800 | ||
1801 | return limit; | |
a63d83f4 DR |
1802 | } |
1803 | ||
19965460 DR |
1804 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1805 | int order) | |
9cbb78bb DR |
1806 | { |
1807 | struct mem_cgroup *iter; | |
1808 | unsigned long chosen_points = 0; | |
1809 | unsigned long totalpages; | |
1810 | unsigned int points = 0; | |
1811 | struct task_struct *chosen = NULL; | |
1812 | ||
876aafbf | 1813 | /* |
465adcf1 DR |
1814 | * If current has a pending SIGKILL or is exiting, then automatically |
1815 | * select it. The goal is to allow it to allocate so that it may | |
1816 | * quickly exit and free its memory. | |
876aafbf | 1817 | */ |
465adcf1 | 1818 | if (fatal_signal_pending(current) || current->flags & PF_EXITING) { |
876aafbf DR |
1819 | set_thread_flag(TIF_MEMDIE); |
1820 | return; | |
1821 | } | |
1822 | ||
1823 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
9cbb78bb DR |
1824 | totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1; |
1825 | for_each_mem_cgroup_tree(iter, memcg) { | |
72ec7029 | 1826 | struct css_task_iter it; |
9cbb78bb DR |
1827 | struct task_struct *task; |
1828 | ||
72ec7029 TH |
1829 | css_task_iter_start(&iter->css, &it); |
1830 | while ((task = css_task_iter_next(&it))) { | |
9cbb78bb DR |
1831 | switch (oom_scan_process_thread(task, totalpages, NULL, |
1832 | false)) { | |
1833 | case OOM_SCAN_SELECT: | |
1834 | if (chosen) | |
1835 | put_task_struct(chosen); | |
1836 | chosen = task; | |
1837 | chosen_points = ULONG_MAX; | |
1838 | get_task_struct(chosen); | |
1839 | /* fall through */ | |
1840 | case OOM_SCAN_CONTINUE: | |
1841 | continue; | |
1842 | case OOM_SCAN_ABORT: | |
72ec7029 | 1843 | css_task_iter_end(&it); |
9cbb78bb DR |
1844 | mem_cgroup_iter_break(memcg, iter); |
1845 | if (chosen) | |
1846 | put_task_struct(chosen); | |
1847 | return; | |
1848 | case OOM_SCAN_OK: | |
1849 | break; | |
1850 | }; | |
1851 | points = oom_badness(task, memcg, NULL, totalpages); | |
1852 | if (points > chosen_points) { | |
1853 | if (chosen) | |
1854 | put_task_struct(chosen); | |
1855 | chosen = task; | |
1856 | chosen_points = points; | |
1857 | get_task_struct(chosen); | |
1858 | } | |
1859 | } | |
72ec7029 | 1860 | css_task_iter_end(&it); |
9cbb78bb DR |
1861 | } |
1862 | ||
1863 | if (!chosen) | |
1864 | return; | |
1865 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1866 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1867 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1868 | } |
1869 | ||
5660048c JW |
1870 | static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg, |
1871 | gfp_t gfp_mask, | |
1872 | unsigned long flags) | |
1873 | { | |
1874 | unsigned long total = 0; | |
1875 | bool noswap = false; | |
1876 | int loop; | |
1877 | ||
1878 | if (flags & MEM_CGROUP_RECLAIM_NOSWAP) | |
1879 | noswap = true; | |
1880 | if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum) | |
1881 | noswap = true; | |
1882 | ||
1883 | for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) { | |
1884 | if (loop) | |
1885 | drain_all_stock_async(memcg); | |
1886 | total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap); | |
1887 | /* | |
1888 | * Allow limit shrinkers, which are triggered directly | |
1889 | * by userspace, to catch signals and stop reclaim | |
1890 | * after minimal progress, regardless of the margin. | |
1891 | */ | |
1892 | if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK)) | |
1893 | break; | |
1894 | if (mem_cgroup_margin(memcg)) | |
1895 | break; | |
1896 | /* | |
1897 | * If nothing was reclaimed after two attempts, there | |
1898 | * may be no reclaimable pages in this hierarchy. | |
1899 | */ | |
1900 | if (loop && !total) | |
1901 | break; | |
1902 | } | |
1903 | return total; | |
1904 | } | |
1905 | ||
4d0c066d KH |
1906 | /** |
1907 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1908 | * @memcg: the target memcg |
4d0c066d KH |
1909 | * @nid: the node ID to be checked. |
1910 | * @noswap : specify true here if the user wants flle only information. | |
1911 | * | |
1912 | * This function returns whether the specified memcg contains any | |
1913 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1914 | * pages in the node. | |
1915 | */ | |
c0ff4b85 | 1916 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1917 | int nid, bool noswap) |
1918 | { | |
c0ff4b85 | 1919 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1920 | return true; |
1921 | if (noswap || !total_swap_pages) | |
1922 | return false; | |
c0ff4b85 | 1923 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1924 | return true; |
1925 | return false; | |
1926 | ||
1927 | } | |
bb4cc1a8 | 1928 | #if MAX_NUMNODES > 1 |
889976db YH |
1929 | |
1930 | /* | |
1931 | * Always updating the nodemask is not very good - even if we have an empty | |
1932 | * list or the wrong list here, we can start from some node and traverse all | |
1933 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1934 | * | |
1935 | */ | |
c0ff4b85 | 1936 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1937 | { |
1938 | int nid; | |
453a9bf3 KH |
1939 | /* |
1940 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1941 | * pagein/pageout changes since the last update. | |
1942 | */ | |
c0ff4b85 | 1943 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1944 | return; |
c0ff4b85 | 1945 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1946 | return; |
1947 | ||
889976db | 1948 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1949 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1950 | |
31aaea4a | 1951 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1952 | |
c0ff4b85 R |
1953 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1954 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1955 | } |
453a9bf3 | 1956 | |
c0ff4b85 R |
1957 | atomic_set(&memcg->numainfo_events, 0); |
1958 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1959 | } |
1960 | ||
1961 | /* | |
1962 | * Selecting a node where we start reclaim from. Because what we need is just | |
1963 | * reducing usage counter, start from anywhere is O,K. Considering | |
1964 | * memory reclaim from current node, there are pros. and cons. | |
1965 | * | |
1966 | * Freeing memory from current node means freeing memory from a node which | |
1967 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1968 | * hit limits, it will see a contention on a node. But freeing from remote | |
1969 | * node means more costs for memory reclaim because of memory latency. | |
1970 | * | |
1971 | * Now, we use round-robin. Better algorithm is welcomed. | |
1972 | */ | |
c0ff4b85 | 1973 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1974 | { |
1975 | int node; | |
1976 | ||
c0ff4b85 R |
1977 | mem_cgroup_may_update_nodemask(memcg); |
1978 | node = memcg->last_scanned_node; | |
889976db | 1979 | |
c0ff4b85 | 1980 | node = next_node(node, memcg->scan_nodes); |
889976db | 1981 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1982 | node = first_node(memcg->scan_nodes); |
889976db YH |
1983 | /* |
1984 | * We call this when we hit limit, not when pages are added to LRU. | |
1985 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1986 | * memcg is too small and all pages are not on LRU. In that case, | |
1987 | * we use curret node. | |
1988 | */ | |
1989 | if (unlikely(node == MAX_NUMNODES)) | |
1990 | node = numa_node_id(); | |
1991 | ||
c0ff4b85 | 1992 | memcg->last_scanned_node = node; |
889976db YH |
1993 | return node; |
1994 | } | |
1995 | ||
bb4cc1a8 AM |
1996 | /* |
1997 | * Check all nodes whether it contains reclaimable pages or not. | |
1998 | * For quick scan, we make use of scan_nodes. This will allow us to skip | |
1999 | * unused nodes. But scan_nodes is lazily updated and may not cotain | |
2000 | * enough new information. We need to do double check. | |
2001 | */ | |
2002 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) | |
2003 | { | |
2004 | int nid; | |
2005 | ||
2006 | /* | |
2007 | * quick check...making use of scan_node. | |
2008 | * We can skip unused nodes. | |
2009 | */ | |
2010 | if (!nodes_empty(memcg->scan_nodes)) { | |
2011 | for (nid = first_node(memcg->scan_nodes); | |
2012 | nid < MAX_NUMNODES; | |
2013 | nid = next_node(nid, memcg->scan_nodes)) { | |
2014 | ||
2015 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) | |
2016 | return true; | |
2017 | } | |
2018 | } | |
2019 | /* | |
2020 | * Check rest of nodes. | |
2021 | */ | |
2022 | for_each_node_state(nid, N_MEMORY) { | |
2023 | if (node_isset(nid, memcg->scan_nodes)) | |
2024 | continue; | |
2025 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) | |
2026 | return true; | |
2027 | } | |
2028 | return false; | |
2029 | } | |
2030 | ||
889976db | 2031 | #else |
c0ff4b85 | 2032 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
2033 | { |
2034 | return 0; | |
2035 | } | |
4d0c066d | 2036 | |
bb4cc1a8 AM |
2037 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
2038 | { | |
2039 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); | |
2040 | } | |
889976db YH |
2041 | #endif |
2042 | ||
0608f43d AM |
2043 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
2044 | struct zone *zone, | |
2045 | gfp_t gfp_mask, | |
2046 | unsigned long *total_scanned) | |
2047 | { | |
2048 | struct mem_cgroup *victim = NULL; | |
2049 | int total = 0; | |
2050 | int loop = 0; | |
2051 | unsigned long excess; | |
2052 | unsigned long nr_scanned; | |
2053 | struct mem_cgroup_reclaim_cookie reclaim = { | |
2054 | .zone = zone, | |
2055 | .priority = 0, | |
2056 | }; | |
2057 | ||
2058 | excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT; | |
2059 | ||
2060 | while (1) { | |
2061 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
2062 | if (!victim) { | |
2063 | loop++; | |
2064 | if (loop >= 2) { | |
2065 | /* | |
2066 | * If we have not been able to reclaim | |
2067 | * anything, it might because there are | |
2068 | * no reclaimable pages under this hierarchy | |
2069 | */ | |
2070 | if (!total) | |
2071 | break; | |
2072 | /* | |
2073 | * We want to do more targeted reclaim. | |
2074 | * excess >> 2 is not to excessive so as to | |
2075 | * reclaim too much, nor too less that we keep | |
2076 | * coming back to reclaim from this cgroup | |
2077 | */ | |
2078 | if (total >= (excess >> 2) || | |
2079 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
2080 | break; | |
2081 | } | |
2082 | continue; | |
2083 | } | |
2084 | if (!mem_cgroup_reclaimable(victim, false)) | |
2085 | continue; | |
2086 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, | |
2087 | zone, &nr_scanned); | |
2088 | *total_scanned += nr_scanned; | |
2089 | if (!res_counter_soft_limit_excess(&root_memcg->res)) | |
2090 | break; | |
6d61ef40 | 2091 | } |
0608f43d AM |
2092 | mem_cgroup_iter_break(root_memcg, victim); |
2093 | return total; | |
6d61ef40 BS |
2094 | } |
2095 | ||
0056f4e6 JW |
2096 | #ifdef CONFIG_LOCKDEP |
2097 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
2098 | .name = "memcg_oom_lock", | |
2099 | }; | |
2100 | #endif | |
2101 | ||
fb2a6fc5 JW |
2102 | static DEFINE_SPINLOCK(memcg_oom_lock); |
2103 | ||
867578cb KH |
2104 | /* |
2105 | * Check OOM-Killer is already running under our hierarchy. | |
2106 | * If someone is running, return false. | |
2107 | */ | |
fb2a6fc5 | 2108 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 2109 | { |
79dfdacc | 2110 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 2111 | |
fb2a6fc5 JW |
2112 | spin_lock(&memcg_oom_lock); |
2113 | ||
9f3a0d09 | 2114 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 2115 | if (iter->oom_lock) { |
79dfdacc MH |
2116 | /* |
2117 | * this subtree of our hierarchy is already locked | |
2118 | * so we cannot give a lock. | |
2119 | */ | |
79dfdacc | 2120 | failed = iter; |
9f3a0d09 JW |
2121 | mem_cgroup_iter_break(memcg, iter); |
2122 | break; | |
23751be0 JW |
2123 | } else |
2124 | iter->oom_lock = true; | |
7d74b06f | 2125 | } |
867578cb | 2126 | |
fb2a6fc5 JW |
2127 | if (failed) { |
2128 | /* | |
2129 | * OK, we failed to lock the whole subtree so we have | |
2130 | * to clean up what we set up to the failing subtree | |
2131 | */ | |
2132 | for_each_mem_cgroup_tree(iter, memcg) { | |
2133 | if (iter == failed) { | |
2134 | mem_cgroup_iter_break(memcg, iter); | |
2135 | break; | |
2136 | } | |
2137 | iter->oom_lock = false; | |
79dfdacc | 2138 | } |
0056f4e6 JW |
2139 | } else |
2140 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
2141 | |
2142 | spin_unlock(&memcg_oom_lock); | |
2143 | ||
2144 | return !failed; | |
a636b327 | 2145 | } |
0b7f569e | 2146 | |
fb2a6fc5 | 2147 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 2148 | { |
7d74b06f KH |
2149 | struct mem_cgroup *iter; |
2150 | ||
fb2a6fc5 | 2151 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 2152 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 2153 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 2154 | iter->oom_lock = false; |
fb2a6fc5 | 2155 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
2156 | } |
2157 | ||
c0ff4b85 | 2158 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2159 | { |
2160 | struct mem_cgroup *iter; | |
2161 | ||
c0ff4b85 | 2162 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
2163 | atomic_inc(&iter->under_oom); |
2164 | } | |
2165 | ||
c0ff4b85 | 2166 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2167 | { |
2168 | struct mem_cgroup *iter; | |
2169 | ||
867578cb KH |
2170 | /* |
2171 | * When a new child is created while the hierarchy is under oom, | |
2172 | * mem_cgroup_oom_lock() may not be called. We have to use | |
2173 | * atomic_add_unless() here. | |
2174 | */ | |
c0ff4b85 | 2175 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 2176 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
2177 | } |
2178 | ||
867578cb KH |
2179 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
2180 | ||
dc98df5a | 2181 | struct oom_wait_info { |
d79154bb | 2182 | struct mem_cgroup *memcg; |
dc98df5a KH |
2183 | wait_queue_t wait; |
2184 | }; | |
2185 | ||
2186 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
2187 | unsigned mode, int sync, void *arg) | |
2188 | { | |
d79154bb HD |
2189 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
2190 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
2191 | struct oom_wait_info *oom_wait_info; |
2192 | ||
2193 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 2194 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 2195 | |
dc98df5a | 2196 | /* |
d79154bb | 2197 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
dc98df5a KH |
2198 | * Then we can use css_is_ancestor without taking care of RCU. |
2199 | */ | |
c0ff4b85 R |
2200 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
2201 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 2202 | return 0; |
dc98df5a KH |
2203 | return autoremove_wake_function(wait, mode, sync, arg); |
2204 | } | |
2205 | ||
c0ff4b85 | 2206 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 2207 | { |
3812c8c8 | 2208 | atomic_inc(&memcg->oom_wakeups); |
c0ff4b85 R |
2209 | /* for filtering, pass "memcg" as argument. */ |
2210 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
2211 | } |
2212 | ||
c0ff4b85 | 2213 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 2214 | { |
c0ff4b85 R |
2215 | if (memcg && atomic_read(&memcg->under_oom)) |
2216 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
2217 | } |
2218 | ||
3812c8c8 | 2219 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 2220 | { |
3812c8c8 JW |
2221 | if (!current->memcg_oom.may_oom) |
2222 | return; | |
867578cb | 2223 | /* |
49426420 JW |
2224 | * We are in the middle of the charge context here, so we |
2225 | * don't want to block when potentially sitting on a callstack | |
2226 | * that holds all kinds of filesystem and mm locks. | |
2227 | * | |
2228 | * Also, the caller may handle a failed allocation gracefully | |
2229 | * (like optional page cache readahead) and so an OOM killer | |
2230 | * invocation might not even be necessary. | |
2231 | * | |
2232 | * That's why we don't do anything here except remember the | |
2233 | * OOM context and then deal with it at the end of the page | |
2234 | * fault when the stack is unwound, the locks are released, | |
2235 | * and when we know whether the fault was overall successful. | |
867578cb | 2236 | */ |
49426420 JW |
2237 | css_get(&memcg->css); |
2238 | current->memcg_oom.memcg = memcg; | |
2239 | current->memcg_oom.gfp_mask = mask; | |
2240 | current->memcg_oom.order = order; | |
3812c8c8 JW |
2241 | } |
2242 | ||
2243 | /** | |
2244 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 2245 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 2246 | * |
49426420 JW |
2247 | * This has to be called at the end of a page fault if the memcg OOM |
2248 | * handler was enabled. | |
3812c8c8 | 2249 | * |
49426420 | 2250 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
2251 | * sleep on a waitqueue until the userspace task resolves the |
2252 | * situation. Sleeping directly in the charge context with all kinds | |
2253 | * of locks held is not a good idea, instead we remember an OOM state | |
2254 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 2255 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
2256 | * |
2257 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 2258 | * completed, %false otherwise. |
3812c8c8 | 2259 | */ |
49426420 | 2260 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 2261 | { |
49426420 | 2262 | struct mem_cgroup *memcg = current->memcg_oom.memcg; |
3812c8c8 | 2263 | struct oom_wait_info owait; |
49426420 | 2264 | bool locked; |
3812c8c8 JW |
2265 | |
2266 | /* OOM is global, do not handle */ | |
3812c8c8 | 2267 | if (!memcg) |
49426420 | 2268 | return false; |
3812c8c8 | 2269 | |
49426420 JW |
2270 | if (!handle) |
2271 | goto cleanup; | |
3812c8c8 JW |
2272 | |
2273 | owait.memcg = memcg; | |
2274 | owait.wait.flags = 0; | |
2275 | owait.wait.func = memcg_oom_wake_function; | |
2276 | owait.wait.private = current; | |
2277 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 2278 | |
3812c8c8 | 2279 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
2280 | mem_cgroup_mark_under_oom(memcg); |
2281 | ||
2282 | locked = mem_cgroup_oom_trylock(memcg); | |
2283 | ||
2284 | if (locked) | |
2285 | mem_cgroup_oom_notify(memcg); | |
2286 | ||
2287 | if (locked && !memcg->oom_kill_disable) { | |
2288 | mem_cgroup_unmark_under_oom(memcg); | |
2289 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2290 | mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask, | |
2291 | current->memcg_oom.order); | |
2292 | } else { | |
3812c8c8 | 2293 | schedule(); |
49426420 JW |
2294 | mem_cgroup_unmark_under_oom(memcg); |
2295 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2296 | } | |
2297 | ||
2298 | if (locked) { | |
fb2a6fc5 JW |
2299 | mem_cgroup_oom_unlock(memcg); |
2300 | /* | |
2301 | * There is no guarantee that an OOM-lock contender | |
2302 | * sees the wakeups triggered by the OOM kill | |
2303 | * uncharges. Wake any sleepers explicitely. | |
2304 | */ | |
2305 | memcg_oom_recover(memcg); | |
2306 | } | |
49426420 JW |
2307 | cleanup: |
2308 | current->memcg_oom.memcg = NULL; | |
3812c8c8 | 2309 | css_put(&memcg->css); |
867578cb | 2310 | return true; |
0b7f569e KH |
2311 | } |
2312 | ||
d69b042f BS |
2313 | /* |
2314 | * Currently used to update mapped file statistics, but the routine can be | |
2315 | * generalized to update other statistics as well. | |
32047e2a KH |
2316 | * |
2317 | * Notes: Race condition | |
2318 | * | |
2319 | * We usually use page_cgroup_lock() for accessing page_cgroup member but | |
2320 | * it tends to be costly. But considering some conditions, we doesn't need | |
2321 | * to do so _always_. | |
2322 | * | |
2323 | * Considering "charge", lock_page_cgroup() is not required because all | |
2324 | * file-stat operations happen after a page is attached to radix-tree. There | |
2325 | * are no race with "charge". | |
2326 | * | |
2327 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
2328 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
2329 | * if there are race with "uncharge". Statistics itself is properly handled | |
2330 | * by flags. | |
2331 | * | |
2332 | * Considering "move", this is an only case we see a race. To make the race | |
619d094b KH |
2333 | * small, we check mm->moving_account and detect there are possibility of race |
2334 | * If there is, we take a lock. | |
d69b042f | 2335 | */ |
26174efd | 2336 | |
89c06bd5 KH |
2337 | void __mem_cgroup_begin_update_page_stat(struct page *page, |
2338 | bool *locked, unsigned long *flags) | |
2339 | { | |
2340 | struct mem_cgroup *memcg; | |
2341 | struct page_cgroup *pc; | |
2342 | ||
2343 | pc = lookup_page_cgroup(page); | |
2344 | again: | |
2345 | memcg = pc->mem_cgroup; | |
2346 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
2347 | return; | |
2348 | /* | |
2349 | * If this memory cgroup is not under account moving, we don't | |
da92c47d | 2350 | * need to take move_lock_mem_cgroup(). Because we already hold |
89c06bd5 | 2351 | * rcu_read_lock(), any calls to move_account will be delayed until |
13fd1dd9 | 2352 | * rcu_read_unlock() if mem_cgroup_stolen() == true. |
89c06bd5 | 2353 | */ |
13fd1dd9 | 2354 | if (!mem_cgroup_stolen(memcg)) |
89c06bd5 KH |
2355 | return; |
2356 | ||
2357 | move_lock_mem_cgroup(memcg, flags); | |
2358 | if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) { | |
2359 | move_unlock_mem_cgroup(memcg, flags); | |
2360 | goto again; | |
2361 | } | |
2362 | *locked = true; | |
2363 | } | |
2364 | ||
2365 | void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags) | |
2366 | { | |
2367 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
2368 | ||
2369 | /* | |
2370 | * It's guaranteed that pc->mem_cgroup never changes while | |
2371 | * lock is held because a routine modifies pc->mem_cgroup | |
da92c47d | 2372 | * should take move_lock_mem_cgroup(). |
89c06bd5 KH |
2373 | */ |
2374 | move_unlock_mem_cgroup(pc->mem_cgroup, flags); | |
2375 | } | |
2376 | ||
2a7106f2 | 2377 | void mem_cgroup_update_page_stat(struct page *page, |
68b4876d | 2378 | enum mem_cgroup_stat_index idx, int val) |
d69b042f | 2379 | { |
c0ff4b85 | 2380 | struct mem_cgroup *memcg; |
32047e2a | 2381 | struct page_cgroup *pc = lookup_page_cgroup(page); |
dbd4ea78 | 2382 | unsigned long uninitialized_var(flags); |
d69b042f | 2383 | |
cfa44946 | 2384 | if (mem_cgroup_disabled()) |
d69b042f | 2385 | return; |
89c06bd5 | 2386 | |
658b72c5 | 2387 | VM_BUG_ON(!rcu_read_lock_held()); |
c0ff4b85 R |
2388 | memcg = pc->mem_cgroup; |
2389 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
89c06bd5 | 2390 | return; |
26174efd | 2391 | |
c0ff4b85 | 2392 | this_cpu_add(memcg->stat->count[idx], val); |
d69b042f | 2393 | } |
26174efd | 2394 | |
cdec2e42 KH |
2395 | /* |
2396 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2397 | * TODO: maybe necessary to use big numbers in big irons. | |
2398 | */ | |
7ec99d62 | 2399 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2400 | struct memcg_stock_pcp { |
2401 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2402 | unsigned int nr_pages; |
cdec2e42 | 2403 | struct work_struct work; |
26fe6168 | 2404 | unsigned long flags; |
a0db00fc | 2405 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2406 | }; |
2407 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2408 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2409 | |
a0956d54 SS |
2410 | /** |
2411 | * consume_stock: Try to consume stocked charge on this cpu. | |
2412 | * @memcg: memcg to consume from. | |
2413 | * @nr_pages: how many pages to charge. | |
2414 | * | |
2415 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2416 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2417 | * service an allocation will refill the stock. | |
2418 | * | |
2419 | * returns true if successful, false otherwise. | |
cdec2e42 | 2420 | */ |
a0956d54 | 2421 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2422 | { |
2423 | struct memcg_stock_pcp *stock; | |
2424 | bool ret = true; | |
2425 | ||
a0956d54 SS |
2426 | if (nr_pages > CHARGE_BATCH) |
2427 | return false; | |
2428 | ||
cdec2e42 | 2429 | stock = &get_cpu_var(memcg_stock); |
a0956d54 SS |
2430 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) |
2431 | stock->nr_pages -= nr_pages; | |
cdec2e42 KH |
2432 | else /* need to call res_counter_charge */ |
2433 | ret = false; | |
2434 | put_cpu_var(memcg_stock); | |
2435 | return ret; | |
2436 | } | |
2437 | ||
2438 | /* | |
2439 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
2440 | */ | |
2441 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2442 | { | |
2443 | struct mem_cgroup *old = stock->cached; | |
2444 | ||
11c9ea4e JW |
2445 | if (stock->nr_pages) { |
2446 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
2447 | ||
2448 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 2449 | if (do_swap_account) |
11c9ea4e JW |
2450 | res_counter_uncharge(&old->memsw, bytes); |
2451 | stock->nr_pages = 0; | |
cdec2e42 KH |
2452 | } |
2453 | stock->cached = NULL; | |
cdec2e42 KH |
2454 | } |
2455 | ||
2456 | /* | |
2457 | * This must be called under preempt disabled or must be called by | |
2458 | * a thread which is pinned to local cpu. | |
2459 | */ | |
2460 | static void drain_local_stock(struct work_struct *dummy) | |
2461 | { | |
2462 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
2463 | drain_stock(stock); | |
26fe6168 | 2464 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2465 | } |
2466 | ||
e4777496 MH |
2467 | static void __init memcg_stock_init(void) |
2468 | { | |
2469 | int cpu; | |
2470 | ||
2471 | for_each_possible_cpu(cpu) { | |
2472 | struct memcg_stock_pcp *stock = | |
2473 | &per_cpu(memcg_stock, cpu); | |
2474 | INIT_WORK(&stock->work, drain_local_stock); | |
2475 | } | |
2476 | } | |
2477 | ||
cdec2e42 KH |
2478 | /* |
2479 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 2480 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2481 | */ |
c0ff4b85 | 2482 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2483 | { |
2484 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2485 | ||
c0ff4b85 | 2486 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2487 | drain_stock(stock); |
c0ff4b85 | 2488 | stock->cached = memcg; |
cdec2e42 | 2489 | } |
11c9ea4e | 2490 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2491 | put_cpu_var(memcg_stock); |
2492 | } | |
2493 | ||
2494 | /* | |
c0ff4b85 | 2495 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
d38144b7 MH |
2496 | * of the hierarchy under it. sync flag says whether we should block |
2497 | * until the work is done. | |
cdec2e42 | 2498 | */ |
c0ff4b85 | 2499 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
cdec2e42 | 2500 | { |
26fe6168 | 2501 | int cpu, curcpu; |
d38144b7 | 2502 | |
cdec2e42 | 2503 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2504 | get_online_cpus(); |
5af12d0e | 2505 | curcpu = get_cpu(); |
cdec2e42 KH |
2506 | for_each_online_cpu(cpu) { |
2507 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2508 | struct mem_cgroup *memcg; |
26fe6168 | 2509 | |
c0ff4b85 R |
2510 | memcg = stock->cached; |
2511 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2512 | continue; |
c0ff4b85 | 2513 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2514 | continue; |
d1a05b69 MH |
2515 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2516 | if (cpu == curcpu) | |
2517 | drain_local_stock(&stock->work); | |
2518 | else | |
2519 | schedule_work_on(cpu, &stock->work); | |
2520 | } | |
cdec2e42 | 2521 | } |
5af12d0e | 2522 | put_cpu(); |
d38144b7 MH |
2523 | |
2524 | if (!sync) | |
2525 | goto out; | |
2526 | ||
2527 | for_each_online_cpu(cpu) { | |
2528 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2529 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2530 | flush_work(&stock->work); |
2531 | } | |
2532 | out: | |
f894ffa8 | 2533 | put_online_cpus(); |
d38144b7 MH |
2534 | } |
2535 | ||
2536 | /* | |
2537 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2538 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
2539 | * expects some charges will be back to res_counter later but cannot wait for | |
2540 | * it. | |
2541 | */ | |
c0ff4b85 | 2542 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
d38144b7 | 2543 | { |
9f50fad6 MH |
2544 | /* |
2545 | * If someone calls draining, avoid adding more kworker runs. | |
2546 | */ | |
2547 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2548 | return; | |
c0ff4b85 | 2549 | drain_all_stock(root_memcg, false); |
9f50fad6 | 2550 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2551 | } |
2552 | ||
2553 | /* This is a synchronous drain interface. */ | |
c0ff4b85 | 2554 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
cdec2e42 KH |
2555 | { |
2556 | /* called when force_empty is called */ | |
9f50fad6 | 2557 | mutex_lock(&percpu_charge_mutex); |
c0ff4b85 | 2558 | drain_all_stock(root_memcg, true); |
9f50fad6 | 2559 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2560 | } |
2561 | ||
711d3d2c KH |
2562 | /* |
2563 | * This function drains percpu counter value from DEAD cpu and | |
2564 | * move it to local cpu. Note that this function can be preempted. | |
2565 | */ | |
c0ff4b85 | 2566 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2567 | { |
2568 | int i; | |
2569 | ||
c0ff4b85 | 2570 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2571 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2572 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2573 | |
c0ff4b85 R |
2574 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2575 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2576 | } |
e9f8974f | 2577 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2578 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2579 | |
c0ff4b85 R |
2580 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2581 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2582 | } |
c0ff4b85 | 2583 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2584 | } |
2585 | ||
0db0628d | 2586 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
2587 | unsigned long action, |
2588 | void *hcpu) | |
2589 | { | |
2590 | int cpu = (unsigned long)hcpu; | |
2591 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2592 | struct mem_cgroup *iter; |
cdec2e42 | 2593 | |
619d094b | 2594 | if (action == CPU_ONLINE) |
1489ebad | 2595 | return NOTIFY_OK; |
1489ebad | 2596 | |
d833049b | 2597 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2598 | return NOTIFY_OK; |
711d3d2c | 2599 | |
9f3a0d09 | 2600 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2601 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2602 | ||
cdec2e42 KH |
2603 | stock = &per_cpu(memcg_stock, cpu); |
2604 | drain_stock(stock); | |
2605 | return NOTIFY_OK; | |
2606 | } | |
2607 | ||
4b534334 KH |
2608 | |
2609 | /* See __mem_cgroup_try_charge() for details */ | |
2610 | enum { | |
2611 | CHARGE_OK, /* success */ | |
2612 | CHARGE_RETRY, /* need to retry but retry is not bad */ | |
2613 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ | |
2614 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ | |
4b534334 KH |
2615 | }; |
2616 | ||
c0ff4b85 | 2617 | static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
4c9c5359 | 2618 | unsigned int nr_pages, unsigned int min_pages, |
3812c8c8 | 2619 | bool invoke_oom) |
4b534334 | 2620 | { |
7ec99d62 | 2621 | unsigned long csize = nr_pages * PAGE_SIZE; |
4b534334 KH |
2622 | struct mem_cgroup *mem_over_limit; |
2623 | struct res_counter *fail_res; | |
2624 | unsigned long flags = 0; | |
2625 | int ret; | |
2626 | ||
c0ff4b85 | 2627 | ret = res_counter_charge(&memcg->res, csize, &fail_res); |
4b534334 KH |
2628 | |
2629 | if (likely(!ret)) { | |
2630 | if (!do_swap_account) | |
2631 | return CHARGE_OK; | |
c0ff4b85 | 2632 | ret = res_counter_charge(&memcg->memsw, csize, &fail_res); |
4b534334 KH |
2633 | if (likely(!ret)) |
2634 | return CHARGE_OK; | |
2635 | ||
c0ff4b85 | 2636 | res_counter_uncharge(&memcg->res, csize); |
4b534334 KH |
2637 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
2638 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
2639 | } else | |
2640 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
9221edb7 | 2641 | /* |
9221edb7 JW |
2642 | * Never reclaim on behalf of optional batching, retry with a |
2643 | * single page instead. | |
2644 | */ | |
4c9c5359 | 2645 | if (nr_pages > min_pages) |
4b534334 KH |
2646 | return CHARGE_RETRY; |
2647 | ||
2648 | if (!(gfp_mask & __GFP_WAIT)) | |
2649 | return CHARGE_WOULDBLOCK; | |
2650 | ||
4c9c5359 SS |
2651 | if (gfp_mask & __GFP_NORETRY) |
2652 | return CHARGE_NOMEM; | |
2653 | ||
5660048c | 2654 | ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags); |
7ec99d62 | 2655 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
19942822 | 2656 | return CHARGE_RETRY; |
4b534334 | 2657 | /* |
19942822 JW |
2658 | * Even though the limit is exceeded at this point, reclaim |
2659 | * may have been able to free some pages. Retry the charge | |
2660 | * before killing the task. | |
2661 | * | |
2662 | * Only for regular pages, though: huge pages are rather | |
2663 | * unlikely to succeed so close to the limit, and we fall back | |
2664 | * to regular pages anyway in case of failure. | |
4b534334 | 2665 | */ |
4c9c5359 | 2666 | if (nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER) && ret) |
4b534334 KH |
2667 | return CHARGE_RETRY; |
2668 | ||
2669 | /* | |
2670 | * At task move, charge accounts can be doubly counted. So, it's | |
2671 | * better to wait until the end of task_move if something is going on. | |
2672 | */ | |
2673 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2674 | return CHARGE_RETRY; | |
2675 | ||
3812c8c8 JW |
2676 | if (invoke_oom) |
2677 | mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(csize)); | |
4b534334 | 2678 | |
3812c8c8 | 2679 | return CHARGE_NOMEM; |
4b534334 KH |
2680 | } |
2681 | ||
f817ed48 | 2682 | /* |
38c5d72f KH |
2683 | * __mem_cgroup_try_charge() does |
2684 | * 1. detect memcg to be charged against from passed *mm and *ptr, | |
2685 | * 2. update res_counter | |
2686 | * 3. call memory reclaim if necessary. | |
2687 | * | |
2688 | * In some special case, if the task is fatal, fatal_signal_pending() or | |
2689 | * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup | |
2690 | * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon | |
2691 | * as possible without any hazards. 2: all pages should have a valid | |
2692 | * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg | |
2693 | * pointer, that is treated as a charge to root_mem_cgroup. | |
2694 | * | |
2695 | * So __mem_cgroup_try_charge() will return | |
2696 | * 0 ... on success, filling *ptr with a valid memcg pointer. | |
2697 | * -ENOMEM ... charge failure because of resource limits. | |
2698 | * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup. | |
2699 | * | |
2700 | * Unlike the exported interface, an "oom" parameter is added. if oom==true, | |
2701 | * the oom-killer can be invoked. | |
8a9f3ccd | 2702 | */ |
f817ed48 | 2703 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
ec168510 | 2704 | gfp_t gfp_mask, |
7ec99d62 | 2705 | unsigned int nr_pages, |
c0ff4b85 | 2706 | struct mem_cgroup **ptr, |
7ec99d62 | 2707 | bool oom) |
8a9f3ccd | 2708 | { |
7ec99d62 | 2709 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
4b534334 | 2710 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
c0ff4b85 | 2711 | struct mem_cgroup *memcg = NULL; |
4b534334 | 2712 | int ret; |
a636b327 | 2713 | |
867578cb KH |
2714 | /* |
2715 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
2716 | * in system level. So, allow to go ahead dying process in addition to | |
2717 | * MEMDIE process. | |
2718 | */ | |
2719 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
2720 | || fatal_signal_pending(current))) | |
2721 | goto bypass; | |
a636b327 | 2722 | |
49426420 JW |
2723 | if (unlikely(task_in_memcg_oom(current))) |
2724 | goto bypass; | |
2725 | ||
8a9f3ccd | 2726 | /* |
3be91277 HD |
2727 | * We always charge the cgroup the mm_struct belongs to. |
2728 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd | 2729 | * thread group leader migrates. It's possible that mm is not |
24467cac | 2730 | * set, if so charge the root memcg (happens for pagecache usage). |
8a9f3ccd | 2731 | */ |
c0ff4b85 | 2732 | if (!*ptr && !mm) |
38c5d72f | 2733 | *ptr = root_mem_cgroup; |
f75ca962 | 2734 | again: |
c0ff4b85 R |
2735 | if (*ptr) { /* css should be a valid one */ |
2736 | memcg = *ptr; | |
c0ff4b85 | 2737 | if (mem_cgroup_is_root(memcg)) |
f75ca962 | 2738 | goto done; |
a0956d54 | 2739 | if (consume_stock(memcg, nr_pages)) |
f75ca962 | 2740 | goto done; |
c0ff4b85 | 2741 | css_get(&memcg->css); |
4b534334 | 2742 | } else { |
f75ca962 | 2743 | struct task_struct *p; |
54595fe2 | 2744 | |
f75ca962 KH |
2745 | rcu_read_lock(); |
2746 | p = rcu_dereference(mm->owner); | |
f75ca962 | 2747 | /* |
ebb76ce1 | 2748 | * Because we don't have task_lock(), "p" can exit. |
c0ff4b85 | 2749 | * In that case, "memcg" can point to root or p can be NULL with |
ebb76ce1 KH |
2750 | * race with swapoff. Then, we have small risk of mis-accouning. |
2751 | * But such kind of mis-account by race always happens because | |
2752 | * we don't have cgroup_mutex(). It's overkill and we allo that | |
2753 | * small race, here. | |
2754 | * (*) swapoff at el will charge against mm-struct not against | |
2755 | * task-struct. So, mm->owner can be NULL. | |
f75ca962 | 2756 | */ |
c0ff4b85 | 2757 | memcg = mem_cgroup_from_task(p); |
38c5d72f KH |
2758 | if (!memcg) |
2759 | memcg = root_mem_cgroup; | |
2760 | if (mem_cgroup_is_root(memcg)) { | |
f75ca962 KH |
2761 | rcu_read_unlock(); |
2762 | goto done; | |
2763 | } | |
a0956d54 | 2764 | if (consume_stock(memcg, nr_pages)) { |
f75ca962 KH |
2765 | /* |
2766 | * It seems dagerous to access memcg without css_get(). | |
2767 | * But considering how consume_stok works, it's not | |
2768 | * necessary. If consume_stock success, some charges | |
2769 | * from this memcg are cached on this cpu. So, we | |
2770 | * don't need to call css_get()/css_tryget() before | |
2771 | * calling consume_stock(). | |
2772 | */ | |
2773 | rcu_read_unlock(); | |
2774 | goto done; | |
2775 | } | |
2776 | /* after here, we may be blocked. we need to get refcnt */ | |
c0ff4b85 | 2777 | if (!css_tryget(&memcg->css)) { |
f75ca962 KH |
2778 | rcu_read_unlock(); |
2779 | goto again; | |
2780 | } | |
2781 | rcu_read_unlock(); | |
2782 | } | |
8a9f3ccd | 2783 | |
4b534334 | 2784 | do { |
3812c8c8 | 2785 | bool invoke_oom = oom && !nr_oom_retries; |
7a81b88c | 2786 | |
4b534334 | 2787 | /* If killed, bypass charge */ |
f75ca962 | 2788 | if (fatal_signal_pending(current)) { |
c0ff4b85 | 2789 | css_put(&memcg->css); |
4b534334 | 2790 | goto bypass; |
f75ca962 | 2791 | } |
6d61ef40 | 2792 | |
3812c8c8 JW |
2793 | ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, |
2794 | nr_pages, invoke_oom); | |
4b534334 KH |
2795 | switch (ret) { |
2796 | case CHARGE_OK: | |
2797 | break; | |
2798 | case CHARGE_RETRY: /* not in OOM situation but retry */ | |
7ec99d62 | 2799 | batch = nr_pages; |
c0ff4b85 R |
2800 | css_put(&memcg->css); |
2801 | memcg = NULL; | |
f75ca962 | 2802 | goto again; |
4b534334 | 2803 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
c0ff4b85 | 2804 | css_put(&memcg->css); |
4b534334 KH |
2805 | goto nomem; |
2806 | case CHARGE_NOMEM: /* OOM routine works */ | |
3812c8c8 | 2807 | if (!oom || invoke_oom) { |
c0ff4b85 | 2808 | css_put(&memcg->css); |
867578cb | 2809 | goto nomem; |
f75ca962 | 2810 | } |
4b534334 KH |
2811 | nr_oom_retries--; |
2812 | break; | |
66e1707b | 2813 | } |
4b534334 KH |
2814 | } while (ret != CHARGE_OK); |
2815 | ||
7ec99d62 | 2816 | if (batch > nr_pages) |
c0ff4b85 R |
2817 | refill_stock(memcg, batch - nr_pages); |
2818 | css_put(&memcg->css); | |
0c3e73e8 | 2819 | done: |
c0ff4b85 | 2820 | *ptr = memcg; |
7a81b88c KH |
2821 | return 0; |
2822 | nomem: | |
3168ecbe JW |
2823 | if (!(gfp_mask & __GFP_NOFAIL)) { |
2824 | *ptr = NULL; | |
2825 | return -ENOMEM; | |
2826 | } | |
867578cb | 2827 | bypass: |
38c5d72f KH |
2828 | *ptr = root_mem_cgroup; |
2829 | return -EINTR; | |
7a81b88c | 2830 | } |
8a9f3ccd | 2831 | |
a3032a2c DN |
2832 | /* |
2833 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2834 | * This function is for that and do uncharge, put css's refcnt. | |
2835 | * gotten by try_charge(). | |
2836 | */ | |
c0ff4b85 | 2837 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg, |
e7018b8d | 2838 | unsigned int nr_pages) |
a3032a2c | 2839 | { |
c0ff4b85 | 2840 | if (!mem_cgroup_is_root(memcg)) { |
e7018b8d JW |
2841 | unsigned long bytes = nr_pages * PAGE_SIZE; |
2842 | ||
c0ff4b85 | 2843 | res_counter_uncharge(&memcg->res, bytes); |
a3032a2c | 2844 | if (do_swap_account) |
c0ff4b85 | 2845 | res_counter_uncharge(&memcg->memsw, bytes); |
a3032a2c | 2846 | } |
854ffa8d DN |
2847 | } |
2848 | ||
d01dd17f KH |
2849 | /* |
2850 | * Cancel chrages in this cgroup....doesn't propagate to parent cgroup. | |
2851 | * This is useful when moving usage to parent cgroup. | |
2852 | */ | |
2853 | static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg, | |
2854 | unsigned int nr_pages) | |
2855 | { | |
2856 | unsigned long bytes = nr_pages * PAGE_SIZE; | |
2857 | ||
2858 | if (mem_cgroup_is_root(memcg)) | |
2859 | return; | |
2860 | ||
2861 | res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes); | |
2862 | if (do_swap_account) | |
2863 | res_counter_uncharge_until(&memcg->memsw, | |
2864 | memcg->memsw.parent, bytes); | |
2865 | } | |
2866 | ||
a3b2d692 KH |
2867 | /* |
2868 | * A helper function to get mem_cgroup from ID. must be called under | |
e9316080 TH |
2869 | * rcu_read_lock(). The caller is responsible for calling css_tryget if |
2870 | * the mem_cgroup is used for charging. (dropping refcnt from swap can be | |
2871 | * called against removed memcg.) | |
a3b2d692 KH |
2872 | */ |
2873 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2874 | { | |
2875 | struct cgroup_subsys_state *css; | |
2876 | ||
2877 | /* ID 0 is unused ID */ | |
2878 | if (!id) | |
2879 | return NULL; | |
2880 | css = css_lookup(&mem_cgroup_subsys, id); | |
2881 | if (!css) | |
2882 | return NULL; | |
b2145145 | 2883 | return mem_cgroup_from_css(css); |
a3b2d692 KH |
2884 | } |
2885 | ||
e42d9d5d | 2886 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2887 | { |
c0ff4b85 | 2888 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2889 | struct page_cgroup *pc; |
a3b2d692 | 2890 | unsigned short id; |
b5a84319 KH |
2891 | swp_entry_t ent; |
2892 | ||
3c776e64 DN |
2893 | VM_BUG_ON(!PageLocked(page)); |
2894 | ||
3c776e64 | 2895 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2896 | lock_page_cgroup(pc); |
a3b2d692 | 2897 | if (PageCgroupUsed(pc)) { |
c0ff4b85 R |
2898 | memcg = pc->mem_cgroup; |
2899 | if (memcg && !css_tryget(&memcg->css)) | |
2900 | memcg = NULL; | |
e42d9d5d | 2901 | } else if (PageSwapCache(page)) { |
3c776e64 | 2902 | ent.val = page_private(page); |
9fb4b7cc | 2903 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2904 | rcu_read_lock(); |
c0ff4b85 R |
2905 | memcg = mem_cgroup_lookup(id); |
2906 | if (memcg && !css_tryget(&memcg->css)) | |
2907 | memcg = NULL; | |
a3b2d692 | 2908 | rcu_read_unlock(); |
3c776e64 | 2909 | } |
c0bd3f63 | 2910 | unlock_page_cgroup(pc); |
c0ff4b85 | 2911 | return memcg; |
b5a84319 KH |
2912 | } |
2913 | ||
c0ff4b85 | 2914 | static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg, |
5564e88b | 2915 | struct page *page, |
7ec99d62 | 2916 | unsigned int nr_pages, |
9ce70c02 HD |
2917 | enum charge_type ctype, |
2918 | bool lrucare) | |
7a81b88c | 2919 | { |
ce587e65 | 2920 | struct page_cgroup *pc = lookup_page_cgroup(page); |
9ce70c02 | 2921 | struct zone *uninitialized_var(zone); |
fa9add64 | 2922 | struct lruvec *lruvec; |
9ce70c02 | 2923 | bool was_on_lru = false; |
b2402857 | 2924 | bool anon; |
9ce70c02 | 2925 | |
ca3e0214 | 2926 | lock_page_cgroup(pc); |
90deb788 | 2927 | VM_BUG_ON(PageCgroupUsed(pc)); |
ca3e0214 KH |
2928 | /* |
2929 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2930 | * accessed by any other context at this point. | |
2931 | */ | |
9ce70c02 HD |
2932 | |
2933 | /* | |
2934 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2935 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2936 | */ | |
2937 | if (lrucare) { | |
2938 | zone = page_zone(page); | |
2939 | spin_lock_irq(&zone->lru_lock); | |
2940 | if (PageLRU(page)) { | |
fa9add64 | 2941 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 | 2942 | ClearPageLRU(page); |
fa9add64 | 2943 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2944 | was_on_lru = true; |
2945 | } | |
2946 | } | |
2947 | ||
c0ff4b85 | 2948 | pc->mem_cgroup = memcg; |
261fb61a KH |
2949 | /* |
2950 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
2951 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
2952 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
2953 | * before USED bit, we need memory barrier here. | |
2954 | * See mem_cgroup_add_lru_list(), etc. | |
f894ffa8 | 2955 | */ |
08e552c6 | 2956 | smp_wmb(); |
b2402857 | 2957 | SetPageCgroupUsed(pc); |
3be91277 | 2958 | |
9ce70c02 HD |
2959 | if (lrucare) { |
2960 | if (was_on_lru) { | |
fa9add64 | 2961 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 HD |
2962 | VM_BUG_ON(PageLRU(page)); |
2963 | SetPageLRU(page); | |
fa9add64 | 2964 | add_page_to_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2965 | } |
2966 | spin_unlock_irq(&zone->lru_lock); | |
2967 | } | |
2968 | ||
41326c17 | 2969 | if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON) |
b2402857 KH |
2970 | anon = true; |
2971 | else | |
2972 | anon = false; | |
2973 | ||
b070e65c | 2974 | mem_cgroup_charge_statistics(memcg, page, anon, nr_pages); |
52d4b9ac | 2975 | unlock_page_cgroup(pc); |
9ce70c02 | 2976 | |
430e4863 | 2977 | /* |
bb4cc1a8 AM |
2978 | * "charge_statistics" updated event counter. Then, check it. |
2979 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
2980 | * if they exceeds softlimit. | |
430e4863 | 2981 | */ |
c0ff4b85 | 2982 | memcg_check_events(memcg, page); |
7a81b88c | 2983 | } |
66e1707b | 2984 | |
7cf27982 GC |
2985 | static DEFINE_MUTEX(set_limit_mutex); |
2986 | ||
7ae1e1d0 GC |
2987 | #ifdef CONFIG_MEMCG_KMEM |
2988 | static inline bool memcg_can_account_kmem(struct mem_cgroup *memcg) | |
2989 | { | |
2990 | return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg) && | |
2991 | (memcg->kmem_account_flags & KMEM_ACCOUNTED_MASK); | |
2992 | } | |
2993 | ||
1f458cbf GC |
2994 | /* |
2995 | * This is a bit cumbersome, but it is rarely used and avoids a backpointer | |
2996 | * in the memcg_cache_params struct. | |
2997 | */ | |
2998 | static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p) | |
2999 | { | |
3000 | struct kmem_cache *cachep; | |
3001 | ||
3002 | VM_BUG_ON(p->is_root_cache); | |
3003 | cachep = p->root_cache; | |
3004 | return cachep->memcg_params->memcg_caches[memcg_cache_id(p->memcg)]; | |
3005 | } | |
3006 | ||
749c5415 | 3007 | #ifdef CONFIG_SLABINFO |
182446d0 TH |
3008 | static int mem_cgroup_slabinfo_read(struct cgroup_subsys_state *css, |
3009 | struct cftype *cft, struct seq_file *m) | |
749c5415 | 3010 | { |
182446d0 | 3011 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
749c5415 GC |
3012 | struct memcg_cache_params *params; |
3013 | ||
3014 | if (!memcg_can_account_kmem(memcg)) | |
3015 | return -EIO; | |
3016 | ||
3017 | print_slabinfo_header(m); | |
3018 | ||
3019 | mutex_lock(&memcg->slab_caches_mutex); | |
3020 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) | |
3021 | cache_show(memcg_params_to_cache(params), m); | |
3022 | mutex_unlock(&memcg->slab_caches_mutex); | |
3023 | ||
3024 | return 0; | |
3025 | } | |
3026 | #endif | |
3027 | ||
7ae1e1d0 GC |
3028 | static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, u64 size) |
3029 | { | |
3030 | struct res_counter *fail_res; | |
3031 | struct mem_cgroup *_memcg; | |
3032 | int ret = 0; | |
3033 | bool may_oom; | |
3034 | ||
3035 | ret = res_counter_charge(&memcg->kmem, size, &fail_res); | |
3036 | if (ret) | |
3037 | return ret; | |
3038 | ||
3039 | /* | |
3040 | * Conditions under which we can wait for the oom_killer. Those are | |
3041 | * the same conditions tested by the core page allocator | |
3042 | */ | |
3043 | may_oom = (gfp & __GFP_FS) && !(gfp & __GFP_NORETRY); | |
3044 | ||
3045 | _memcg = memcg; | |
3046 | ret = __mem_cgroup_try_charge(NULL, gfp, size >> PAGE_SHIFT, | |
3047 | &_memcg, may_oom); | |
3048 | ||
3049 | if (ret == -EINTR) { | |
3050 | /* | |
3051 | * __mem_cgroup_try_charge() chosed to bypass to root due to | |
3052 | * OOM kill or fatal signal. Since our only options are to | |
3053 | * either fail the allocation or charge it to this cgroup, do | |
3054 | * it as a temporary condition. But we can't fail. From a | |
3055 | * kmem/slab perspective, the cache has already been selected, | |
3056 | * by mem_cgroup_kmem_get_cache(), so it is too late to change | |
3057 | * our minds. | |
3058 | * | |
3059 | * This condition will only trigger if the task entered | |
3060 | * memcg_charge_kmem in a sane state, but was OOM-killed during | |
3061 | * __mem_cgroup_try_charge() above. Tasks that were already | |
3062 | * dying when the allocation triggers should have been already | |
3063 | * directed to the root cgroup in memcontrol.h | |
3064 | */ | |
3065 | res_counter_charge_nofail(&memcg->res, size, &fail_res); | |
3066 | if (do_swap_account) | |
3067 | res_counter_charge_nofail(&memcg->memsw, size, | |
3068 | &fail_res); | |
3069 | ret = 0; | |
3070 | } else if (ret) | |
3071 | res_counter_uncharge(&memcg->kmem, size); | |
3072 | ||
3073 | return ret; | |
3074 | } | |
3075 | ||
3076 | static void memcg_uncharge_kmem(struct mem_cgroup *memcg, u64 size) | |
3077 | { | |
7ae1e1d0 GC |
3078 | res_counter_uncharge(&memcg->res, size); |
3079 | if (do_swap_account) | |
3080 | res_counter_uncharge(&memcg->memsw, size); | |
7de37682 GC |
3081 | |
3082 | /* Not down to 0 */ | |
3083 | if (res_counter_uncharge(&memcg->kmem, size)) | |
3084 | return; | |
3085 | ||
10d5ebf4 LZ |
3086 | /* |
3087 | * Releases a reference taken in kmem_cgroup_css_offline in case | |
3088 | * this last uncharge is racing with the offlining code or it is | |
3089 | * outliving the memcg existence. | |
3090 | * | |
3091 | * The memory barrier imposed by test&clear is paired with the | |
3092 | * explicit one in memcg_kmem_mark_dead(). | |
3093 | */ | |
7de37682 | 3094 | if (memcg_kmem_test_and_clear_dead(memcg)) |
10d5ebf4 | 3095 | css_put(&memcg->css); |
7ae1e1d0 GC |
3096 | } |
3097 | ||
2633d7a0 GC |
3098 | void memcg_cache_list_add(struct mem_cgroup *memcg, struct kmem_cache *cachep) |
3099 | { | |
3100 | if (!memcg) | |
3101 | return; | |
3102 | ||
3103 | mutex_lock(&memcg->slab_caches_mutex); | |
3104 | list_add(&cachep->memcg_params->list, &memcg->memcg_slab_caches); | |
3105 | mutex_unlock(&memcg->slab_caches_mutex); | |
3106 | } | |
3107 | ||
3108 | /* | |
3109 | * helper for acessing a memcg's index. It will be used as an index in the | |
3110 | * child cache array in kmem_cache, and also to derive its name. This function | |
3111 | * will return -1 when this is not a kmem-limited memcg. | |
3112 | */ | |
3113 | int memcg_cache_id(struct mem_cgroup *memcg) | |
3114 | { | |
3115 | return memcg ? memcg->kmemcg_id : -1; | |
3116 | } | |
3117 | ||
55007d84 GC |
3118 | /* |
3119 | * This ends up being protected by the set_limit mutex, during normal | |
3120 | * operation, because that is its main call site. | |
3121 | * | |
3122 | * But when we create a new cache, we can call this as well if its parent | |
3123 | * is kmem-limited. That will have to hold set_limit_mutex as well. | |
3124 | */ | |
3125 | int memcg_update_cache_sizes(struct mem_cgroup *memcg) | |
3126 | { | |
3127 | int num, ret; | |
3128 | ||
3129 | num = ida_simple_get(&kmem_limited_groups, | |
3130 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); | |
3131 | if (num < 0) | |
3132 | return num; | |
3133 | /* | |
3134 | * After this point, kmem_accounted (that we test atomically in | |
3135 | * the beginning of this conditional), is no longer 0. This | |
3136 | * guarantees only one process will set the following boolean | |
3137 | * to true. We don't need test_and_set because we're protected | |
3138 | * by the set_limit_mutex anyway. | |
3139 | */ | |
3140 | memcg_kmem_set_activated(memcg); | |
3141 | ||
3142 | ret = memcg_update_all_caches(num+1); | |
3143 | if (ret) { | |
3144 | ida_simple_remove(&kmem_limited_groups, num); | |
3145 | memcg_kmem_clear_activated(memcg); | |
3146 | return ret; | |
3147 | } | |
3148 | ||
3149 | memcg->kmemcg_id = num; | |
3150 | INIT_LIST_HEAD(&memcg->memcg_slab_caches); | |
3151 | mutex_init(&memcg->slab_caches_mutex); | |
3152 | return 0; | |
3153 | } | |
3154 | ||
3155 | static size_t memcg_caches_array_size(int num_groups) | |
3156 | { | |
3157 | ssize_t size; | |
3158 | if (num_groups <= 0) | |
3159 | return 0; | |
3160 | ||
3161 | size = 2 * num_groups; | |
3162 | if (size < MEMCG_CACHES_MIN_SIZE) | |
3163 | size = MEMCG_CACHES_MIN_SIZE; | |
3164 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
3165 | size = MEMCG_CACHES_MAX_SIZE; | |
3166 | ||
3167 | return size; | |
3168 | } | |
3169 | ||
3170 | /* | |
3171 | * We should update the current array size iff all caches updates succeed. This | |
3172 | * can only be done from the slab side. The slab mutex needs to be held when | |
3173 | * calling this. | |
3174 | */ | |
3175 | void memcg_update_array_size(int num) | |
3176 | { | |
3177 | if (num > memcg_limited_groups_array_size) | |
3178 | memcg_limited_groups_array_size = memcg_caches_array_size(num); | |
3179 | } | |
3180 | ||
15cf17d2 KK |
3181 | static void kmem_cache_destroy_work_func(struct work_struct *w); |
3182 | ||
55007d84 GC |
3183 | int memcg_update_cache_size(struct kmem_cache *s, int num_groups) |
3184 | { | |
3185 | struct memcg_cache_params *cur_params = s->memcg_params; | |
3186 | ||
3187 | VM_BUG_ON(s->memcg_params && !s->memcg_params->is_root_cache); | |
3188 | ||
3189 | if (num_groups > memcg_limited_groups_array_size) { | |
3190 | int i; | |
3191 | ssize_t size = memcg_caches_array_size(num_groups); | |
3192 | ||
3193 | size *= sizeof(void *); | |
90c7a79c | 3194 | size += offsetof(struct memcg_cache_params, memcg_caches); |
55007d84 GC |
3195 | |
3196 | s->memcg_params = kzalloc(size, GFP_KERNEL); | |
3197 | if (!s->memcg_params) { | |
3198 | s->memcg_params = cur_params; | |
3199 | return -ENOMEM; | |
3200 | } | |
3201 | ||
3202 | s->memcg_params->is_root_cache = true; | |
3203 | ||
3204 | /* | |
3205 | * There is the chance it will be bigger than | |
3206 | * memcg_limited_groups_array_size, if we failed an allocation | |
3207 | * in a cache, in which case all caches updated before it, will | |
3208 | * have a bigger array. | |
3209 | * | |
3210 | * But if that is the case, the data after | |
3211 | * memcg_limited_groups_array_size is certainly unused | |
3212 | */ | |
3213 | for (i = 0; i < memcg_limited_groups_array_size; i++) { | |
3214 | if (!cur_params->memcg_caches[i]) | |
3215 | continue; | |
3216 | s->memcg_params->memcg_caches[i] = | |
3217 | cur_params->memcg_caches[i]; | |
3218 | } | |
3219 | ||
3220 | /* | |
3221 | * Ideally, we would wait until all caches succeed, and only | |
3222 | * then free the old one. But this is not worth the extra | |
3223 | * pointer per-cache we'd have to have for this. | |
3224 | * | |
3225 | * It is not a big deal if some caches are left with a size | |
3226 | * bigger than the others. And all updates will reset this | |
3227 | * anyway. | |
3228 | */ | |
3229 | kfree(cur_params); | |
3230 | } | |
3231 | return 0; | |
3232 | } | |
3233 | ||
943a451a GC |
3234 | int memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s, |
3235 | struct kmem_cache *root_cache) | |
2633d7a0 | 3236 | { |
90c7a79c | 3237 | size_t size; |
2633d7a0 GC |
3238 | |
3239 | if (!memcg_kmem_enabled()) | |
3240 | return 0; | |
3241 | ||
90c7a79c AV |
3242 | if (!memcg) { |
3243 | size = offsetof(struct memcg_cache_params, memcg_caches); | |
55007d84 | 3244 | size += memcg_limited_groups_array_size * sizeof(void *); |
90c7a79c AV |
3245 | } else |
3246 | size = sizeof(struct memcg_cache_params); | |
55007d84 | 3247 | |
2633d7a0 GC |
3248 | s->memcg_params = kzalloc(size, GFP_KERNEL); |
3249 | if (!s->memcg_params) | |
3250 | return -ENOMEM; | |
3251 | ||
943a451a | 3252 | if (memcg) { |
2633d7a0 | 3253 | s->memcg_params->memcg = memcg; |
943a451a | 3254 | s->memcg_params->root_cache = root_cache; |
3e6b11df AV |
3255 | INIT_WORK(&s->memcg_params->destroy, |
3256 | kmem_cache_destroy_work_func); | |
4ba902b5 GC |
3257 | } else |
3258 | s->memcg_params->is_root_cache = true; | |
3259 | ||
2633d7a0 GC |
3260 | return 0; |
3261 | } | |
3262 | ||
3263 | void memcg_release_cache(struct kmem_cache *s) | |
3264 | { | |
d7f25f8a GC |
3265 | struct kmem_cache *root; |
3266 | struct mem_cgroup *memcg; | |
3267 | int id; | |
3268 | ||
3269 | /* | |
3270 | * This happens, for instance, when a root cache goes away before we | |
3271 | * add any memcg. | |
3272 | */ | |
3273 | if (!s->memcg_params) | |
3274 | return; | |
3275 | ||
3276 | if (s->memcg_params->is_root_cache) | |
3277 | goto out; | |
3278 | ||
3279 | memcg = s->memcg_params->memcg; | |
3280 | id = memcg_cache_id(memcg); | |
3281 | ||
3282 | root = s->memcg_params->root_cache; | |
3283 | root->memcg_params->memcg_caches[id] = NULL; | |
d7f25f8a GC |
3284 | |
3285 | mutex_lock(&memcg->slab_caches_mutex); | |
3286 | list_del(&s->memcg_params->list); | |
3287 | mutex_unlock(&memcg->slab_caches_mutex); | |
3288 | ||
20f05310 | 3289 | css_put(&memcg->css); |
d7f25f8a | 3290 | out: |
2633d7a0 GC |
3291 | kfree(s->memcg_params); |
3292 | } | |
3293 | ||
0e9d92f2 GC |
3294 | /* |
3295 | * During the creation a new cache, we need to disable our accounting mechanism | |
3296 | * altogether. This is true even if we are not creating, but rather just | |
3297 | * enqueing new caches to be created. | |
3298 | * | |
3299 | * This is because that process will trigger allocations; some visible, like | |
3300 | * explicit kmallocs to auxiliary data structures, name strings and internal | |
3301 | * cache structures; some well concealed, like INIT_WORK() that can allocate | |
3302 | * objects during debug. | |
3303 | * | |
3304 | * If any allocation happens during memcg_kmem_get_cache, we will recurse back | |
3305 | * to it. This may not be a bounded recursion: since the first cache creation | |
3306 | * failed to complete (waiting on the allocation), we'll just try to create the | |
3307 | * cache again, failing at the same point. | |
3308 | * | |
3309 | * memcg_kmem_get_cache is prepared to abort after seeing a positive count of | |
3310 | * memcg_kmem_skip_account. So we enclose anything that might allocate memory | |
3311 | * inside the following two functions. | |
3312 | */ | |
3313 | static inline void memcg_stop_kmem_account(void) | |
3314 | { | |
3315 | VM_BUG_ON(!current->mm); | |
3316 | current->memcg_kmem_skip_account++; | |
3317 | } | |
3318 | ||
3319 | static inline void memcg_resume_kmem_account(void) | |
3320 | { | |
3321 | VM_BUG_ON(!current->mm); | |
3322 | current->memcg_kmem_skip_account--; | |
3323 | } | |
3324 | ||
1f458cbf GC |
3325 | static void kmem_cache_destroy_work_func(struct work_struct *w) |
3326 | { | |
3327 | struct kmem_cache *cachep; | |
3328 | struct memcg_cache_params *p; | |
3329 | ||
3330 | p = container_of(w, struct memcg_cache_params, destroy); | |
3331 | ||
3332 | cachep = memcg_params_to_cache(p); | |
3333 | ||
22933152 GC |
3334 | /* |
3335 | * If we get down to 0 after shrink, we could delete right away. | |
3336 | * However, memcg_release_pages() already puts us back in the workqueue | |
3337 | * in that case. If we proceed deleting, we'll get a dangling | |
3338 | * reference, and removing the object from the workqueue in that case | |
3339 | * is unnecessary complication. We are not a fast path. | |
3340 | * | |
3341 | * Note that this case is fundamentally different from racing with | |
3342 | * shrink_slab(): if memcg_cgroup_destroy_cache() is called in | |
3343 | * kmem_cache_shrink, not only we would be reinserting a dead cache | |
3344 | * into the queue, but doing so from inside the worker racing to | |
3345 | * destroy it. | |
3346 | * | |
3347 | * So if we aren't down to zero, we'll just schedule a worker and try | |
3348 | * again | |
3349 | */ | |
3350 | if (atomic_read(&cachep->memcg_params->nr_pages) != 0) { | |
3351 | kmem_cache_shrink(cachep); | |
3352 | if (atomic_read(&cachep->memcg_params->nr_pages) == 0) | |
3353 | return; | |
3354 | } else | |
1f458cbf GC |
3355 | kmem_cache_destroy(cachep); |
3356 | } | |
3357 | ||
3358 | void mem_cgroup_destroy_cache(struct kmem_cache *cachep) | |
3359 | { | |
3360 | if (!cachep->memcg_params->dead) | |
3361 | return; | |
3362 | ||
22933152 GC |
3363 | /* |
3364 | * There are many ways in which we can get here. | |
3365 | * | |
3366 | * We can get to a memory-pressure situation while the delayed work is | |
3367 | * still pending to run. The vmscan shrinkers can then release all | |
3368 | * cache memory and get us to destruction. If this is the case, we'll | |
3369 | * be executed twice, which is a bug (the second time will execute over | |
3370 | * bogus data). In this case, cancelling the work should be fine. | |
3371 | * | |
3372 | * But we can also get here from the worker itself, if | |
3373 | * kmem_cache_shrink is enough to shake all the remaining objects and | |
3374 | * get the page count to 0. In this case, we'll deadlock if we try to | |
3375 | * cancel the work (the worker runs with an internal lock held, which | |
3376 | * is the same lock we would hold for cancel_work_sync().) | |
3377 | * | |
3378 | * Since we can't possibly know who got us here, just refrain from | |
3379 | * running if there is already work pending | |
3380 | */ | |
3381 | if (work_pending(&cachep->memcg_params->destroy)) | |
3382 | return; | |
1f458cbf GC |
3383 | /* |
3384 | * We have to defer the actual destroying to a workqueue, because | |
3385 | * we might currently be in a context that cannot sleep. | |
3386 | */ | |
3387 | schedule_work(&cachep->memcg_params->destroy); | |
3388 | } | |
3389 | ||
d9c10ddd MH |
3390 | /* |
3391 | * This lock protects updaters, not readers. We want readers to be as fast as | |
3392 | * they can, and they will either see NULL or a valid cache value. Our model | |
3393 | * allow them to see NULL, in which case the root memcg will be selected. | |
3394 | * | |
3395 | * We need this lock because multiple allocations to the same cache from a non | |
3396 | * will span more than one worker. Only one of them can create the cache. | |
3397 | */ | |
3398 | static DEFINE_MUTEX(memcg_cache_mutex); | |
d7f25f8a | 3399 | |
d9c10ddd MH |
3400 | /* |
3401 | * Called with memcg_cache_mutex held | |
3402 | */ | |
d7f25f8a GC |
3403 | static struct kmem_cache *kmem_cache_dup(struct mem_cgroup *memcg, |
3404 | struct kmem_cache *s) | |
3405 | { | |
d7f25f8a | 3406 | struct kmem_cache *new; |
d9c10ddd | 3407 | static char *tmp_name = NULL; |
d7f25f8a | 3408 | |
d9c10ddd MH |
3409 | lockdep_assert_held(&memcg_cache_mutex); |
3410 | ||
3411 | /* | |
3412 | * kmem_cache_create_memcg duplicates the given name and | |
3413 | * cgroup_name for this name requires RCU context. | |
3414 | * This static temporary buffer is used to prevent from | |
3415 | * pointless shortliving allocation. | |
3416 | */ | |
3417 | if (!tmp_name) { | |
3418 | tmp_name = kmalloc(PATH_MAX, GFP_KERNEL); | |
3419 | if (!tmp_name) | |
3420 | return NULL; | |
3421 | } | |
3422 | ||
3423 | rcu_read_lock(); | |
3424 | snprintf(tmp_name, PATH_MAX, "%s(%d:%s)", s->name, | |
3425 | memcg_cache_id(memcg), cgroup_name(memcg->css.cgroup)); | |
3426 | rcu_read_unlock(); | |
d7f25f8a | 3427 | |
d9c10ddd | 3428 | new = kmem_cache_create_memcg(memcg, tmp_name, s->object_size, s->align, |
943a451a | 3429 | (s->flags & ~SLAB_PANIC), s->ctor, s); |
d7f25f8a | 3430 | |
d79923fa GC |
3431 | if (new) |
3432 | new->allocflags |= __GFP_KMEMCG; | |
3433 | ||
d7f25f8a GC |
3434 | return new; |
3435 | } | |
3436 | ||
d7f25f8a GC |
3437 | static struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg, |
3438 | struct kmem_cache *cachep) | |
3439 | { | |
3440 | struct kmem_cache *new_cachep; | |
3441 | int idx; | |
3442 | ||
3443 | BUG_ON(!memcg_can_account_kmem(memcg)); | |
3444 | ||
3445 | idx = memcg_cache_id(memcg); | |
3446 | ||
3447 | mutex_lock(&memcg_cache_mutex); | |
3448 | new_cachep = cachep->memcg_params->memcg_caches[idx]; | |
20f05310 LZ |
3449 | if (new_cachep) { |
3450 | css_put(&memcg->css); | |
d7f25f8a | 3451 | goto out; |
20f05310 | 3452 | } |
d7f25f8a GC |
3453 | |
3454 | new_cachep = kmem_cache_dup(memcg, cachep); | |
d7f25f8a GC |
3455 | if (new_cachep == NULL) { |
3456 | new_cachep = cachep; | |
20f05310 | 3457 | css_put(&memcg->css); |
d7f25f8a GC |
3458 | goto out; |
3459 | } | |
3460 | ||
1f458cbf | 3461 | atomic_set(&new_cachep->memcg_params->nr_pages , 0); |
d7f25f8a GC |
3462 | |
3463 | cachep->memcg_params->memcg_caches[idx] = new_cachep; | |
3464 | /* | |
3465 | * the readers won't lock, make sure everybody sees the updated value, | |
3466 | * so they won't put stuff in the queue again for no reason | |
3467 | */ | |
3468 | wmb(); | |
3469 | out: | |
3470 | mutex_unlock(&memcg_cache_mutex); | |
3471 | return new_cachep; | |
3472 | } | |
3473 | ||
7cf27982 GC |
3474 | void kmem_cache_destroy_memcg_children(struct kmem_cache *s) |
3475 | { | |
3476 | struct kmem_cache *c; | |
3477 | int i; | |
3478 | ||
3479 | if (!s->memcg_params) | |
3480 | return; | |
3481 | if (!s->memcg_params->is_root_cache) | |
3482 | return; | |
3483 | ||
3484 | /* | |
3485 | * If the cache is being destroyed, we trust that there is no one else | |
3486 | * requesting objects from it. Even if there are, the sanity checks in | |
3487 | * kmem_cache_destroy should caught this ill-case. | |
3488 | * | |
3489 | * Still, we don't want anyone else freeing memcg_caches under our | |
3490 | * noses, which can happen if a new memcg comes to life. As usual, | |
3491 | * we'll take the set_limit_mutex to protect ourselves against this. | |
3492 | */ | |
3493 | mutex_lock(&set_limit_mutex); | |
3494 | for (i = 0; i < memcg_limited_groups_array_size; i++) { | |
3495 | c = s->memcg_params->memcg_caches[i]; | |
3496 | if (!c) | |
3497 | continue; | |
3498 | ||
3499 | /* | |
3500 | * We will now manually delete the caches, so to avoid races | |
3501 | * we need to cancel all pending destruction workers and | |
3502 | * proceed with destruction ourselves. | |
3503 | * | |
3504 | * kmem_cache_destroy() will call kmem_cache_shrink internally, | |
3505 | * and that could spawn the workers again: it is likely that | |
3506 | * the cache still have active pages until this very moment. | |
3507 | * This would lead us back to mem_cgroup_destroy_cache. | |
3508 | * | |
3509 | * But that will not execute at all if the "dead" flag is not | |
3510 | * set, so flip it down to guarantee we are in control. | |
3511 | */ | |
3512 | c->memcg_params->dead = false; | |
22933152 | 3513 | cancel_work_sync(&c->memcg_params->destroy); |
7cf27982 GC |
3514 | kmem_cache_destroy(c); |
3515 | } | |
3516 | mutex_unlock(&set_limit_mutex); | |
3517 | } | |
3518 | ||
d7f25f8a GC |
3519 | struct create_work { |
3520 | struct mem_cgroup *memcg; | |
3521 | struct kmem_cache *cachep; | |
3522 | struct work_struct work; | |
3523 | }; | |
3524 | ||
1f458cbf GC |
3525 | static void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg) |
3526 | { | |
3527 | struct kmem_cache *cachep; | |
3528 | struct memcg_cache_params *params; | |
3529 | ||
3530 | if (!memcg_kmem_is_active(memcg)) | |
3531 | return; | |
3532 | ||
3533 | mutex_lock(&memcg->slab_caches_mutex); | |
3534 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) { | |
3535 | cachep = memcg_params_to_cache(params); | |
3536 | cachep->memcg_params->dead = true; | |
1f458cbf GC |
3537 | schedule_work(&cachep->memcg_params->destroy); |
3538 | } | |
3539 | mutex_unlock(&memcg->slab_caches_mutex); | |
3540 | } | |
3541 | ||
d7f25f8a GC |
3542 | static void memcg_create_cache_work_func(struct work_struct *w) |
3543 | { | |
3544 | struct create_work *cw; | |
3545 | ||
3546 | cw = container_of(w, struct create_work, work); | |
3547 | memcg_create_kmem_cache(cw->memcg, cw->cachep); | |
d7f25f8a GC |
3548 | kfree(cw); |
3549 | } | |
3550 | ||
3551 | /* | |
3552 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 3553 | */ |
0e9d92f2 GC |
3554 | static void __memcg_create_cache_enqueue(struct mem_cgroup *memcg, |
3555 | struct kmem_cache *cachep) | |
d7f25f8a GC |
3556 | { |
3557 | struct create_work *cw; | |
3558 | ||
3559 | cw = kmalloc(sizeof(struct create_work), GFP_NOWAIT); | |
ca0dde97 LZ |
3560 | if (cw == NULL) { |
3561 | css_put(&memcg->css); | |
d7f25f8a GC |
3562 | return; |
3563 | } | |
3564 | ||
3565 | cw->memcg = memcg; | |
3566 | cw->cachep = cachep; | |
3567 | ||
3568 | INIT_WORK(&cw->work, memcg_create_cache_work_func); | |
3569 | schedule_work(&cw->work); | |
3570 | } | |
3571 | ||
0e9d92f2 GC |
3572 | static void memcg_create_cache_enqueue(struct mem_cgroup *memcg, |
3573 | struct kmem_cache *cachep) | |
3574 | { | |
3575 | /* | |
3576 | * We need to stop accounting when we kmalloc, because if the | |
3577 | * corresponding kmalloc cache is not yet created, the first allocation | |
3578 | * in __memcg_create_cache_enqueue will recurse. | |
3579 | * | |
3580 | * However, it is better to enclose the whole function. Depending on | |
3581 | * the debugging options enabled, INIT_WORK(), for instance, can | |
3582 | * trigger an allocation. This too, will make us recurse. Because at | |
3583 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
3584 | * the safest choice is to do it like this, wrapping the whole function. | |
3585 | */ | |
3586 | memcg_stop_kmem_account(); | |
3587 | __memcg_create_cache_enqueue(memcg, cachep); | |
3588 | memcg_resume_kmem_account(); | |
3589 | } | |
d7f25f8a GC |
3590 | /* |
3591 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
3592 | * We try to use the current memcg's version of the cache. | |
3593 | * | |
3594 | * If the cache does not exist yet, if we are the first user of it, | |
3595 | * we either create it immediately, if possible, or create it asynchronously | |
3596 | * in a workqueue. | |
3597 | * In the latter case, we will let the current allocation go through with | |
3598 | * the original cache. | |
3599 | * | |
3600 | * Can't be called in interrupt context or from kernel threads. | |
3601 | * This function needs to be called with rcu_read_lock() held. | |
3602 | */ | |
3603 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, | |
3604 | gfp_t gfp) | |
3605 | { | |
3606 | struct mem_cgroup *memcg; | |
3607 | int idx; | |
3608 | ||
3609 | VM_BUG_ON(!cachep->memcg_params); | |
3610 | VM_BUG_ON(!cachep->memcg_params->is_root_cache); | |
3611 | ||
0e9d92f2 GC |
3612 | if (!current->mm || current->memcg_kmem_skip_account) |
3613 | return cachep; | |
3614 | ||
d7f25f8a GC |
3615 | rcu_read_lock(); |
3616 | memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner)); | |
d7f25f8a GC |
3617 | |
3618 | if (!memcg_can_account_kmem(memcg)) | |
ca0dde97 | 3619 | goto out; |
d7f25f8a GC |
3620 | |
3621 | idx = memcg_cache_id(memcg); | |
3622 | ||
3623 | /* | |
3624 | * barrier to mare sure we're always seeing the up to date value. The | |
3625 | * code updating memcg_caches will issue a write barrier to match this. | |
3626 | */ | |
3627 | read_barrier_depends(); | |
ca0dde97 LZ |
3628 | if (likely(cachep->memcg_params->memcg_caches[idx])) { |
3629 | cachep = cachep->memcg_params->memcg_caches[idx]; | |
3630 | goto out; | |
d7f25f8a GC |
3631 | } |
3632 | ||
ca0dde97 LZ |
3633 | /* The corresponding put will be done in the workqueue. */ |
3634 | if (!css_tryget(&memcg->css)) | |
3635 | goto out; | |
3636 | rcu_read_unlock(); | |
3637 | ||
3638 | /* | |
3639 | * If we are in a safe context (can wait, and not in interrupt | |
3640 | * context), we could be be predictable and return right away. | |
3641 | * This would guarantee that the allocation being performed | |
3642 | * already belongs in the new cache. | |
3643 | * | |
3644 | * However, there are some clashes that can arrive from locking. | |
3645 | * For instance, because we acquire the slab_mutex while doing | |
3646 | * kmem_cache_dup, this means no further allocation could happen | |
3647 | * with the slab_mutex held. | |
3648 | * | |
3649 | * Also, because cache creation issue get_online_cpus(), this | |
3650 | * creates a lock chain: memcg_slab_mutex -> cpu_hotplug_mutex, | |
3651 | * that ends up reversed during cpu hotplug. (cpuset allocates | |
3652 | * a bunch of GFP_KERNEL memory during cpuup). Due to all that, | |
3653 | * better to defer everything. | |
3654 | */ | |
3655 | memcg_create_cache_enqueue(memcg, cachep); | |
3656 | return cachep; | |
3657 | out: | |
3658 | rcu_read_unlock(); | |
3659 | return cachep; | |
d7f25f8a GC |
3660 | } |
3661 | EXPORT_SYMBOL(__memcg_kmem_get_cache); | |
3662 | ||
7ae1e1d0 GC |
3663 | /* |
3664 | * We need to verify if the allocation against current->mm->owner's memcg is | |
3665 | * possible for the given order. But the page is not allocated yet, so we'll | |
3666 | * need a further commit step to do the final arrangements. | |
3667 | * | |
3668 | * It is possible for the task to switch cgroups in this mean time, so at | |
3669 | * commit time, we can't rely on task conversion any longer. We'll then use | |
3670 | * the handle argument to return to the caller which cgroup we should commit | |
3671 | * against. We could also return the memcg directly and avoid the pointer | |
3672 | * passing, but a boolean return value gives better semantics considering | |
3673 | * the compiled-out case as well. | |
3674 | * | |
3675 | * Returning true means the allocation is possible. | |
3676 | */ | |
3677 | bool | |
3678 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) | |
3679 | { | |
3680 | struct mem_cgroup *memcg; | |
3681 | int ret; | |
3682 | ||
3683 | *_memcg = NULL; | |
6d42c232 GC |
3684 | |
3685 | /* | |
3686 | * Disabling accounting is only relevant for some specific memcg | |
3687 | * internal allocations. Therefore we would initially not have such | |
3688 | * check here, since direct calls to the page allocator that are marked | |
3689 | * with GFP_KMEMCG only happen outside memcg core. We are mostly | |
3690 | * concerned with cache allocations, and by having this test at | |
3691 | * memcg_kmem_get_cache, we are already able to relay the allocation to | |
3692 | * the root cache and bypass the memcg cache altogether. | |
3693 | * | |
3694 | * There is one exception, though: the SLUB allocator does not create | |
3695 | * large order caches, but rather service large kmallocs directly from | |
3696 | * the page allocator. Therefore, the following sequence when backed by | |
3697 | * the SLUB allocator: | |
3698 | * | |
f894ffa8 AM |
3699 | * memcg_stop_kmem_account(); |
3700 | * kmalloc(<large_number>) | |
3701 | * memcg_resume_kmem_account(); | |
6d42c232 GC |
3702 | * |
3703 | * would effectively ignore the fact that we should skip accounting, | |
3704 | * since it will drive us directly to this function without passing | |
3705 | * through the cache selector memcg_kmem_get_cache. Such large | |
3706 | * allocations are extremely rare but can happen, for instance, for the | |
3707 | * cache arrays. We bring this test here. | |
3708 | */ | |
3709 | if (!current->mm || current->memcg_kmem_skip_account) | |
3710 | return true; | |
3711 | ||
7ae1e1d0 GC |
3712 | memcg = try_get_mem_cgroup_from_mm(current->mm); |
3713 | ||
3714 | /* | |
3715 | * very rare case described in mem_cgroup_from_task. Unfortunately there | |
3716 | * isn't much we can do without complicating this too much, and it would | |
3717 | * be gfp-dependent anyway. Just let it go | |
3718 | */ | |
3719 | if (unlikely(!memcg)) | |
3720 | return true; | |
3721 | ||
3722 | if (!memcg_can_account_kmem(memcg)) { | |
3723 | css_put(&memcg->css); | |
3724 | return true; | |
3725 | } | |
3726 | ||
7ae1e1d0 GC |
3727 | ret = memcg_charge_kmem(memcg, gfp, PAGE_SIZE << order); |
3728 | if (!ret) | |
3729 | *_memcg = memcg; | |
7ae1e1d0 GC |
3730 | |
3731 | css_put(&memcg->css); | |
3732 | return (ret == 0); | |
3733 | } | |
3734 | ||
3735 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
3736 | int order) | |
3737 | { | |
3738 | struct page_cgroup *pc; | |
3739 | ||
3740 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
3741 | ||
3742 | /* The page allocation failed. Revert */ | |
3743 | if (!page) { | |
3744 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); | |
7ae1e1d0 GC |
3745 | return; |
3746 | } | |
3747 | ||
3748 | pc = lookup_page_cgroup(page); | |
3749 | lock_page_cgroup(pc); | |
3750 | pc->mem_cgroup = memcg; | |
3751 | SetPageCgroupUsed(pc); | |
3752 | unlock_page_cgroup(pc); | |
3753 | } | |
3754 | ||
3755 | void __memcg_kmem_uncharge_pages(struct page *page, int order) | |
3756 | { | |
3757 | struct mem_cgroup *memcg = NULL; | |
3758 | struct page_cgroup *pc; | |
3759 | ||
3760 | ||
3761 | pc = lookup_page_cgroup(page); | |
3762 | /* | |
3763 | * Fast unlocked return. Theoretically might have changed, have to | |
3764 | * check again after locking. | |
3765 | */ | |
3766 | if (!PageCgroupUsed(pc)) | |
3767 | return; | |
3768 | ||
3769 | lock_page_cgroup(pc); | |
3770 | if (PageCgroupUsed(pc)) { | |
3771 | memcg = pc->mem_cgroup; | |
3772 | ClearPageCgroupUsed(pc); | |
3773 | } | |
3774 | unlock_page_cgroup(pc); | |
3775 | ||
3776 | /* | |
3777 | * We trust that only if there is a memcg associated with the page, it | |
3778 | * is a valid allocation | |
3779 | */ | |
3780 | if (!memcg) | |
3781 | return; | |
3782 | ||
3783 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
3784 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); | |
7ae1e1d0 | 3785 | } |
1f458cbf GC |
3786 | #else |
3787 | static inline void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg) | |
3788 | { | |
3789 | } | |
7ae1e1d0 GC |
3790 | #endif /* CONFIG_MEMCG_KMEM */ |
3791 | ||
ca3e0214 KH |
3792 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3793 | ||
a0db00fc | 3794 | #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION) |
ca3e0214 KH |
3795 | /* |
3796 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
3797 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
3798 | * charge/uncharge will be never happen and move_account() is done under | |
3799 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 3800 | */ |
e94c8a9c | 3801 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 KH |
3802 | { |
3803 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
e94c8a9c | 3804 | struct page_cgroup *pc; |
b070e65c | 3805 | struct mem_cgroup *memcg; |
e94c8a9c | 3806 | int i; |
ca3e0214 | 3807 | |
3d37c4a9 KH |
3808 | if (mem_cgroup_disabled()) |
3809 | return; | |
b070e65c DR |
3810 | |
3811 | memcg = head_pc->mem_cgroup; | |
e94c8a9c KH |
3812 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
3813 | pc = head_pc + i; | |
b070e65c | 3814 | pc->mem_cgroup = memcg; |
e94c8a9c | 3815 | smp_wmb();/* see __commit_charge() */ |
e94c8a9c KH |
3816 | pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
3817 | } | |
b070e65c DR |
3818 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
3819 | HPAGE_PMD_NR); | |
ca3e0214 | 3820 | } |
12d27107 | 3821 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3822 | |
3ea67d06 SZ |
3823 | static inline |
3824 | void mem_cgroup_move_account_page_stat(struct mem_cgroup *from, | |
3825 | struct mem_cgroup *to, | |
3826 | unsigned int nr_pages, | |
3827 | enum mem_cgroup_stat_index idx) | |
3828 | { | |
3829 | /* Update stat data for mem_cgroup */ | |
3830 | preempt_disable(); | |
5e8cfc3c | 3831 | __this_cpu_sub(from->stat->count[idx], nr_pages); |
3ea67d06 SZ |
3832 | __this_cpu_add(to->stat->count[idx], nr_pages); |
3833 | preempt_enable(); | |
3834 | } | |
3835 | ||
f817ed48 | 3836 | /** |
de3638d9 | 3837 | * mem_cgroup_move_account - move account of the page |
5564e88b | 3838 | * @page: the page |
7ec99d62 | 3839 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
3840 | * @pc: page_cgroup of the page. |
3841 | * @from: mem_cgroup which the page is moved from. | |
3842 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
3843 | * | |
3844 | * The caller must confirm following. | |
08e552c6 | 3845 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 3846 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 3847 | * |
2f3479b1 KH |
3848 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
3849 | * from old cgroup. | |
f817ed48 | 3850 | */ |
7ec99d62 JW |
3851 | static int mem_cgroup_move_account(struct page *page, |
3852 | unsigned int nr_pages, | |
3853 | struct page_cgroup *pc, | |
3854 | struct mem_cgroup *from, | |
2f3479b1 | 3855 | struct mem_cgroup *to) |
f817ed48 | 3856 | { |
de3638d9 JW |
3857 | unsigned long flags; |
3858 | int ret; | |
b2402857 | 3859 | bool anon = PageAnon(page); |
987eba66 | 3860 | |
f817ed48 | 3861 | VM_BUG_ON(from == to); |
5564e88b | 3862 | VM_BUG_ON(PageLRU(page)); |
de3638d9 JW |
3863 | /* |
3864 | * The page is isolated from LRU. So, collapse function | |
3865 | * will not handle this page. But page splitting can happen. | |
3866 | * Do this check under compound_page_lock(). The caller should | |
3867 | * hold it. | |
3868 | */ | |
3869 | ret = -EBUSY; | |
7ec99d62 | 3870 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
3871 | goto out; |
3872 | ||
3873 | lock_page_cgroup(pc); | |
3874 | ||
3875 | ret = -EINVAL; | |
3876 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
3877 | goto unlock; | |
3878 | ||
312734c0 | 3879 | move_lock_mem_cgroup(from, &flags); |
f817ed48 | 3880 | |
3ea67d06 SZ |
3881 | if (!anon && page_mapped(page)) |
3882 | mem_cgroup_move_account_page_stat(from, to, nr_pages, | |
3883 | MEM_CGROUP_STAT_FILE_MAPPED); | |
3884 | ||
3885 | if (PageWriteback(page)) | |
3886 | mem_cgroup_move_account_page_stat(from, to, nr_pages, | |
3887 | MEM_CGROUP_STAT_WRITEBACK); | |
3888 | ||
b070e65c | 3889 | mem_cgroup_charge_statistics(from, page, anon, -nr_pages); |
d69b042f | 3890 | |
854ffa8d | 3891 | /* caller should have done css_get */ |
08e552c6 | 3892 | pc->mem_cgroup = to; |
b070e65c | 3893 | mem_cgroup_charge_statistics(to, page, anon, nr_pages); |
312734c0 | 3894 | move_unlock_mem_cgroup(from, &flags); |
de3638d9 JW |
3895 | ret = 0; |
3896 | unlock: | |
57f9fd7d | 3897 | unlock_page_cgroup(pc); |
d2265e6f KH |
3898 | /* |
3899 | * check events | |
3900 | */ | |
5564e88b JW |
3901 | memcg_check_events(to, page); |
3902 | memcg_check_events(from, page); | |
de3638d9 | 3903 | out: |
f817ed48 KH |
3904 | return ret; |
3905 | } | |
3906 | ||
2ef37d3f MH |
3907 | /** |
3908 | * mem_cgroup_move_parent - moves page to the parent group | |
3909 | * @page: the page to move | |
3910 | * @pc: page_cgroup of the page | |
3911 | * @child: page's cgroup | |
3912 | * | |
3913 | * move charges to its parent or the root cgroup if the group has no | |
3914 | * parent (aka use_hierarchy==0). | |
3915 | * Although this might fail (get_page_unless_zero, isolate_lru_page or | |
3916 | * mem_cgroup_move_account fails) the failure is always temporary and | |
3917 | * it signals a race with a page removal/uncharge or migration. In the | |
3918 | * first case the page is on the way out and it will vanish from the LRU | |
3919 | * on the next attempt and the call should be retried later. | |
3920 | * Isolation from the LRU fails only if page has been isolated from | |
3921 | * the LRU since we looked at it and that usually means either global | |
3922 | * reclaim or migration going on. The page will either get back to the | |
3923 | * LRU or vanish. | |
3924 | * Finaly mem_cgroup_move_account fails only if the page got uncharged | |
3925 | * (!PageCgroupUsed) or moved to a different group. The page will | |
3926 | * disappear in the next attempt. | |
f817ed48 | 3927 | */ |
5564e88b JW |
3928 | static int mem_cgroup_move_parent(struct page *page, |
3929 | struct page_cgroup *pc, | |
6068bf01 | 3930 | struct mem_cgroup *child) |
f817ed48 | 3931 | { |
f817ed48 | 3932 | struct mem_cgroup *parent; |
7ec99d62 | 3933 | unsigned int nr_pages; |
4be4489f | 3934 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
3935 | int ret; |
3936 | ||
d8423011 | 3937 | VM_BUG_ON(mem_cgroup_is_root(child)); |
f817ed48 | 3938 | |
57f9fd7d DN |
3939 | ret = -EBUSY; |
3940 | if (!get_page_unless_zero(page)) | |
3941 | goto out; | |
3942 | if (isolate_lru_page(page)) | |
3943 | goto put; | |
52dbb905 | 3944 | |
7ec99d62 | 3945 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 3946 | |
cc926f78 KH |
3947 | parent = parent_mem_cgroup(child); |
3948 | /* | |
3949 | * If no parent, move charges to root cgroup. | |
3950 | */ | |
3951 | if (!parent) | |
3952 | parent = root_mem_cgroup; | |
f817ed48 | 3953 | |
2ef37d3f MH |
3954 | if (nr_pages > 1) { |
3955 | VM_BUG_ON(!PageTransHuge(page)); | |
987eba66 | 3956 | flags = compound_lock_irqsave(page); |
2ef37d3f | 3957 | } |
987eba66 | 3958 | |
cc926f78 | 3959 | ret = mem_cgroup_move_account(page, nr_pages, |
2f3479b1 | 3960 | pc, child, parent); |
cc926f78 KH |
3961 | if (!ret) |
3962 | __mem_cgroup_cancel_local_charge(child, nr_pages); | |
8dba474f | 3963 | |
7ec99d62 | 3964 | if (nr_pages > 1) |
987eba66 | 3965 | compound_unlock_irqrestore(page, flags); |
08e552c6 | 3966 | putback_lru_page(page); |
57f9fd7d | 3967 | put: |
40d58138 | 3968 | put_page(page); |
57f9fd7d | 3969 | out: |
f817ed48 KH |
3970 | return ret; |
3971 | } | |
3972 | ||
7a81b88c KH |
3973 | /* |
3974 | * Charge the memory controller for page usage. | |
3975 | * Return | |
3976 | * 0 if the charge was successful | |
3977 | * < 0 if the cgroup is over its limit | |
3978 | */ | |
3979 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
73045c47 | 3980 | gfp_t gfp_mask, enum charge_type ctype) |
7a81b88c | 3981 | { |
c0ff4b85 | 3982 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 3983 | unsigned int nr_pages = 1; |
8493ae43 | 3984 | bool oom = true; |
7a81b88c | 3985 | int ret; |
ec168510 | 3986 | |
37c2ac78 | 3987 | if (PageTransHuge(page)) { |
7ec99d62 | 3988 | nr_pages <<= compound_order(page); |
37c2ac78 | 3989 | VM_BUG_ON(!PageTransHuge(page)); |
8493ae43 JW |
3990 | /* |
3991 | * Never OOM-kill a process for a huge page. The | |
3992 | * fault handler will fall back to regular pages. | |
3993 | */ | |
3994 | oom = false; | |
37c2ac78 | 3995 | } |
7a81b88c | 3996 | |
c0ff4b85 | 3997 | ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom); |
38c5d72f | 3998 | if (ret == -ENOMEM) |
7a81b88c | 3999 | return ret; |
ce587e65 | 4000 | __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false); |
8a9f3ccd | 4001 | return 0; |
8a9f3ccd BS |
4002 | } |
4003 | ||
7a81b88c KH |
4004 | int mem_cgroup_newpage_charge(struct page *page, |
4005 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 4006 | { |
f8d66542 | 4007 | if (mem_cgroup_disabled()) |
cede86ac | 4008 | return 0; |
7a0524cf JW |
4009 | VM_BUG_ON(page_mapped(page)); |
4010 | VM_BUG_ON(page->mapping && !PageAnon(page)); | |
4011 | VM_BUG_ON(!mm); | |
217bc319 | 4012 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
41326c17 | 4013 | MEM_CGROUP_CHARGE_TYPE_ANON); |
217bc319 KH |
4014 | } |
4015 | ||
54595fe2 KH |
4016 | /* |
4017 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
4018 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 4019 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
4020 | * "commit()" or removed by "cancel()" |
4021 | */ | |
0435a2fd JW |
4022 | static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
4023 | struct page *page, | |
4024 | gfp_t mask, | |
4025 | struct mem_cgroup **memcgp) | |
8c7c6e34 | 4026 | { |
c0ff4b85 | 4027 | struct mem_cgroup *memcg; |
90deb788 | 4028 | struct page_cgroup *pc; |
54595fe2 | 4029 | int ret; |
8c7c6e34 | 4030 | |
90deb788 JW |
4031 | pc = lookup_page_cgroup(page); |
4032 | /* | |
4033 | * Every swap fault against a single page tries to charge the | |
4034 | * page, bail as early as possible. shmem_unuse() encounters | |
4035 | * already charged pages, too. The USED bit is protected by | |
4036 | * the page lock, which serializes swap cache removal, which | |
4037 | * in turn serializes uncharging. | |
4038 | */ | |
4039 | if (PageCgroupUsed(pc)) | |
4040 | return 0; | |
8c7c6e34 KH |
4041 | if (!do_swap_account) |
4042 | goto charge_cur_mm; | |
c0ff4b85 R |
4043 | memcg = try_get_mem_cgroup_from_page(page); |
4044 | if (!memcg) | |
54595fe2 | 4045 | goto charge_cur_mm; |
72835c86 JW |
4046 | *memcgp = memcg; |
4047 | ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true); | |
c0ff4b85 | 4048 | css_put(&memcg->css); |
38c5d72f KH |
4049 | if (ret == -EINTR) |
4050 | ret = 0; | |
54595fe2 | 4051 | return ret; |
8c7c6e34 | 4052 | charge_cur_mm: |
38c5d72f KH |
4053 | ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true); |
4054 | if (ret == -EINTR) | |
4055 | ret = 0; | |
4056 | return ret; | |
8c7c6e34 KH |
4057 | } |
4058 | ||
0435a2fd JW |
4059 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page, |
4060 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
4061 | { | |
4062 | *memcgp = NULL; | |
4063 | if (mem_cgroup_disabled()) | |
4064 | return 0; | |
bdf4f4d2 JW |
4065 | /* |
4066 | * A racing thread's fault, or swapoff, may have already | |
4067 | * updated the pte, and even removed page from swap cache: in | |
4068 | * those cases unuse_pte()'s pte_same() test will fail; but | |
4069 | * there's also a KSM case which does need to charge the page. | |
4070 | */ | |
4071 | if (!PageSwapCache(page)) { | |
4072 | int ret; | |
4073 | ||
4074 | ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, memcgp, true); | |
4075 | if (ret == -EINTR) | |
4076 | ret = 0; | |
4077 | return ret; | |
4078 | } | |
0435a2fd JW |
4079 | return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp); |
4080 | } | |
4081 | ||
827a03d2 JW |
4082 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg) |
4083 | { | |
4084 | if (mem_cgroup_disabled()) | |
4085 | return; | |
4086 | if (!memcg) | |
4087 | return; | |
4088 | __mem_cgroup_cancel_charge(memcg, 1); | |
4089 | } | |
4090 | ||
83aae4c7 | 4091 | static void |
72835c86 | 4092 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg, |
83aae4c7 | 4093 | enum charge_type ctype) |
7a81b88c | 4094 | { |
f8d66542 | 4095 | if (mem_cgroup_disabled()) |
7a81b88c | 4096 | return; |
72835c86 | 4097 | if (!memcg) |
7a81b88c | 4098 | return; |
5a6475a4 | 4099 | |
ce587e65 | 4100 | __mem_cgroup_commit_charge(memcg, page, 1, ctype, true); |
8c7c6e34 KH |
4101 | /* |
4102 | * Now swap is on-memory. This means this page may be | |
4103 | * counted both as mem and swap....double count. | |
03f3c433 KH |
4104 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
4105 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
4106 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 4107 | */ |
03f3c433 | 4108 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 4109 | swp_entry_t ent = {.val = page_private(page)}; |
86493009 | 4110 | mem_cgroup_uncharge_swap(ent); |
8c7c6e34 | 4111 | } |
7a81b88c KH |
4112 | } |
4113 | ||
72835c86 JW |
4114 | void mem_cgroup_commit_charge_swapin(struct page *page, |
4115 | struct mem_cgroup *memcg) | |
83aae4c7 | 4116 | { |
72835c86 | 4117 | __mem_cgroup_commit_charge_swapin(page, memcg, |
41326c17 | 4118 | MEM_CGROUP_CHARGE_TYPE_ANON); |
83aae4c7 DN |
4119 | } |
4120 | ||
827a03d2 JW |
4121 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
4122 | gfp_t gfp_mask) | |
7a81b88c | 4123 | { |
827a03d2 JW |
4124 | struct mem_cgroup *memcg = NULL; |
4125 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
4126 | int ret; | |
4127 | ||
f8d66542 | 4128 | if (mem_cgroup_disabled()) |
827a03d2 JW |
4129 | return 0; |
4130 | if (PageCompound(page)) | |
4131 | return 0; | |
4132 | ||
827a03d2 JW |
4133 | if (!PageSwapCache(page)) |
4134 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, type); | |
4135 | else { /* page is swapcache/shmem */ | |
0435a2fd JW |
4136 | ret = __mem_cgroup_try_charge_swapin(mm, page, |
4137 | gfp_mask, &memcg); | |
827a03d2 JW |
4138 | if (!ret) |
4139 | __mem_cgroup_commit_charge_swapin(page, memcg, type); | |
4140 | } | |
4141 | return ret; | |
7a81b88c KH |
4142 | } |
4143 | ||
c0ff4b85 | 4144 | static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg, |
7ec99d62 JW |
4145 | unsigned int nr_pages, |
4146 | const enum charge_type ctype) | |
569b846d KH |
4147 | { |
4148 | struct memcg_batch_info *batch = NULL; | |
4149 | bool uncharge_memsw = true; | |
7ec99d62 | 4150 | |
569b846d KH |
4151 | /* If swapout, usage of swap doesn't decrease */ |
4152 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
4153 | uncharge_memsw = false; | |
569b846d KH |
4154 | |
4155 | batch = ¤t->memcg_batch; | |
4156 | /* | |
4157 | * In usual, we do css_get() when we remember memcg pointer. | |
4158 | * But in this case, we keep res->usage until end of a series of | |
4159 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
4160 | */ | |
4161 | if (!batch->memcg) | |
c0ff4b85 | 4162 | batch->memcg = memcg; |
3c11ecf4 KH |
4163 | /* |
4164 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
25985edc | 4165 | * In those cases, all pages freed continuously can be expected to be in |
3c11ecf4 KH |
4166 | * the same cgroup and we have chance to coalesce uncharges. |
4167 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
4168 | * because we want to do uncharge as soon as possible. | |
4169 | */ | |
4170 | ||
4171 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
4172 | goto direct_uncharge; | |
4173 | ||
7ec99d62 | 4174 | if (nr_pages > 1) |
ec168510 AA |
4175 | goto direct_uncharge; |
4176 | ||
569b846d KH |
4177 | /* |
4178 | * In typical case, batch->memcg == mem. This means we can | |
4179 | * merge a series of uncharges to an uncharge of res_counter. | |
4180 | * If not, we uncharge res_counter ony by one. | |
4181 | */ | |
c0ff4b85 | 4182 | if (batch->memcg != memcg) |
569b846d KH |
4183 | goto direct_uncharge; |
4184 | /* remember freed charge and uncharge it later */ | |
7ffd4ca7 | 4185 | batch->nr_pages++; |
569b846d | 4186 | if (uncharge_memsw) |
7ffd4ca7 | 4187 | batch->memsw_nr_pages++; |
569b846d KH |
4188 | return; |
4189 | direct_uncharge: | |
c0ff4b85 | 4190 | res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE); |
569b846d | 4191 | if (uncharge_memsw) |
c0ff4b85 R |
4192 | res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE); |
4193 | if (unlikely(batch->memcg != memcg)) | |
4194 | memcg_oom_recover(memcg); | |
569b846d | 4195 | } |
7a81b88c | 4196 | |
8a9f3ccd | 4197 | /* |
69029cd5 | 4198 | * uncharge if !page_mapped(page) |
8a9f3ccd | 4199 | */ |
8c7c6e34 | 4200 | static struct mem_cgroup * |
0030f535 JW |
4201 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype, |
4202 | bool end_migration) | |
8a9f3ccd | 4203 | { |
c0ff4b85 | 4204 | struct mem_cgroup *memcg = NULL; |
7ec99d62 JW |
4205 | unsigned int nr_pages = 1; |
4206 | struct page_cgroup *pc; | |
b2402857 | 4207 | bool anon; |
8a9f3ccd | 4208 | |
f8d66542 | 4209 | if (mem_cgroup_disabled()) |
8c7c6e34 | 4210 | return NULL; |
4077960e | 4211 | |
37c2ac78 | 4212 | if (PageTransHuge(page)) { |
7ec99d62 | 4213 | nr_pages <<= compound_order(page); |
37c2ac78 AA |
4214 | VM_BUG_ON(!PageTransHuge(page)); |
4215 | } | |
8697d331 | 4216 | /* |
3c541e14 | 4217 | * Check if our page_cgroup is valid |
8697d331 | 4218 | */ |
52d4b9ac | 4219 | pc = lookup_page_cgroup(page); |
cfa44946 | 4220 | if (unlikely(!PageCgroupUsed(pc))) |
8c7c6e34 | 4221 | return NULL; |
b9c565d5 | 4222 | |
52d4b9ac | 4223 | lock_page_cgroup(pc); |
d13d1443 | 4224 | |
c0ff4b85 | 4225 | memcg = pc->mem_cgroup; |
8c7c6e34 | 4226 | |
d13d1443 KH |
4227 | if (!PageCgroupUsed(pc)) |
4228 | goto unlock_out; | |
4229 | ||
b2402857 KH |
4230 | anon = PageAnon(page); |
4231 | ||
d13d1443 | 4232 | switch (ctype) { |
41326c17 | 4233 | case MEM_CGROUP_CHARGE_TYPE_ANON: |
2ff76f11 KH |
4234 | /* |
4235 | * Generally PageAnon tells if it's the anon statistics to be | |
4236 | * updated; but sometimes e.g. mem_cgroup_uncharge_page() is | |
4237 | * used before page reached the stage of being marked PageAnon. | |
4238 | */ | |
b2402857 KH |
4239 | anon = true; |
4240 | /* fallthrough */ | |
8a9478ca | 4241 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c | 4242 | /* See mem_cgroup_prepare_migration() */ |
0030f535 JW |
4243 | if (page_mapped(page)) |
4244 | goto unlock_out; | |
4245 | /* | |
4246 | * Pages under migration may not be uncharged. But | |
4247 | * end_migration() /must/ be the one uncharging the | |
4248 | * unused post-migration page and so it has to call | |
4249 | * here with the migration bit still set. See the | |
4250 | * res_counter handling below. | |
4251 | */ | |
4252 | if (!end_migration && PageCgroupMigration(pc)) | |
d13d1443 KH |
4253 | goto unlock_out; |
4254 | break; | |
4255 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
4256 | if (!PageAnon(page)) { /* Shared memory */ | |
4257 | if (page->mapping && !page_is_file_cache(page)) | |
4258 | goto unlock_out; | |
4259 | } else if (page_mapped(page)) /* Anon */ | |
4260 | goto unlock_out; | |
4261 | break; | |
4262 | default: | |
4263 | break; | |
52d4b9ac | 4264 | } |
d13d1443 | 4265 | |
b070e65c | 4266 | mem_cgroup_charge_statistics(memcg, page, anon, -nr_pages); |
04046e1a | 4267 | |
52d4b9ac | 4268 | ClearPageCgroupUsed(pc); |
544122e5 KH |
4269 | /* |
4270 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
4271 | * freed from LRU. This is safe because uncharged page is expected not | |
4272 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
4273 | * special functions. | |
4274 | */ | |
b9c565d5 | 4275 | |
52d4b9ac | 4276 | unlock_page_cgroup(pc); |
f75ca962 | 4277 | /* |
c0ff4b85 | 4278 | * even after unlock, we have memcg->res.usage here and this memcg |
4050377b | 4279 | * will never be freed, so it's safe to call css_get(). |
f75ca962 | 4280 | */ |
c0ff4b85 | 4281 | memcg_check_events(memcg, page); |
f75ca962 | 4282 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
c0ff4b85 | 4283 | mem_cgroup_swap_statistics(memcg, true); |
4050377b | 4284 | css_get(&memcg->css); |
f75ca962 | 4285 | } |
0030f535 JW |
4286 | /* |
4287 | * Migration does not charge the res_counter for the | |
4288 | * replacement page, so leave it alone when phasing out the | |
4289 | * page that is unused after the migration. | |
4290 | */ | |
4291 | if (!end_migration && !mem_cgroup_is_root(memcg)) | |
c0ff4b85 | 4292 | mem_cgroup_do_uncharge(memcg, nr_pages, ctype); |
6d12e2d8 | 4293 | |
c0ff4b85 | 4294 | return memcg; |
d13d1443 KH |
4295 | |
4296 | unlock_out: | |
4297 | unlock_page_cgroup(pc); | |
8c7c6e34 | 4298 | return NULL; |
3c541e14 BS |
4299 | } |
4300 | ||
69029cd5 KH |
4301 | void mem_cgroup_uncharge_page(struct page *page) |
4302 | { | |
52d4b9ac KH |
4303 | /* early check. */ |
4304 | if (page_mapped(page)) | |
4305 | return; | |
40f23a21 | 4306 | VM_BUG_ON(page->mapping && !PageAnon(page)); |
28ccddf7 JW |
4307 | /* |
4308 | * If the page is in swap cache, uncharge should be deferred | |
4309 | * to the swap path, which also properly accounts swap usage | |
4310 | * and handles memcg lifetime. | |
4311 | * | |
4312 | * Note that this check is not stable and reclaim may add the | |
4313 | * page to swap cache at any time after this. However, if the | |
4314 | * page is not in swap cache by the time page->mapcount hits | |
4315 | * 0, there won't be any page table references to the swap | |
4316 | * slot, and reclaim will free it and not actually write the | |
4317 | * page to disk. | |
4318 | */ | |
0c59b89c JW |
4319 | if (PageSwapCache(page)) |
4320 | return; | |
0030f535 | 4321 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false); |
69029cd5 KH |
4322 | } |
4323 | ||
4324 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
4325 | { | |
4326 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 4327 | VM_BUG_ON(page->mapping); |
0030f535 | 4328 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false); |
69029cd5 KH |
4329 | } |
4330 | ||
569b846d KH |
4331 | /* |
4332 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
4333 | * In that cases, pages are freed continuously and we can expect pages | |
4334 | * are in the same memcg. All these calls itself limits the number of | |
4335 | * pages freed at once, then uncharge_start/end() is called properly. | |
4336 | * This may be called prural(2) times in a context, | |
4337 | */ | |
4338 | ||
4339 | void mem_cgroup_uncharge_start(void) | |
4340 | { | |
4341 | current->memcg_batch.do_batch++; | |
4342 | /* We can do nest. */ | |
4343 | if (current->memcg_batch.do_batch == 1) { | |
4344 | current->memcg_batch.memcg = NULL; | |
7ffd4ca7 JW |
4345 | current->memcg_batch.nr_pages = 0; |
4346 | current->memcg_batch.memsw_nr_pages = 0; | |
569b846d KH |
4347 | } |
4348 | } | |
4349 | ||
4350 | void mem_cgroup_uncharge_end(void) | |
4351 | { | |
4352 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
4353 | ||
4354 | if (!batch->do_batch) | |
4355 | return; | |
4356 | ||
4357 | batch->do_batch--; | |
4358 | if (batch->do_batch) /* If stacked, do nothing. */ | |
4359 | return; | |
4360 | ||
4361 | if (!batch->memcg) | |
4362 | return; | |
4363 | /* | |
4364 | * This "batch->memcg" is valid without any css_get/put etc... | |
4365 | * bacause we hide charges behind us. | |
4366 | */ | |
7ffd4ca7 JW |
4367 | if (batch->nr_pages) |
4368 | res_counter_uncharge(&batch->memcg->res, | |
4369 | batch->nr_pages * PAGE_SIZE); | |
4370 | if (batch->memsw_nr_pages) | |
4371 | res_counter_uncharge(&batch->memcg->memsw, | |
4372 | batch->memsw_nr_pages * PAGE_SIZE); | |
3c11ecf4 | 4373 | memcg_oom_recover(batch->memcg); |
569b846d KH |
4374 | /* forget this pointer (for sanity check) */ |
4375 | batch->memcg = NULL; | |
4376 | } | |
4377 | ||
e767e056 | 4378 | #ifdef CONFIG_SWAP |
8c7c6e34 | 4379 | /* |
e767e056 | 4380 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
4381 | * memcg information is recorded to swap_cgroup of "ent" |
4382 | */ | |
8a9478ca KH |
4383 | void |
4384 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
4385 | { |
4386 | struct mem_cgroup *memcg; | |
8a9478ca KH |
4387 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
4388 | ||
4389 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
4390 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
4391 | ||
0030f535 | 4392 | memcg = __mem_cgroup_uncharge_common(page, ctype, false); |
8c7c6e34 | 4393 | |
f75ca962 KH |
4394 | /* |
4395 | * record memcg information, if swapout && memcg != NULL, | |
4050377b | 4396 | * css_get() was called in uncharge(). |
f75ca962 KH |
4397 | */ |
4398 | if (do_swap_account && swapout && memcg) | |
a3b2d692 | 4399 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 | 4400 | } |
e767e056 | 4401 | #endif |
8c7c6e34 | 4402 | |
c255a458 | 4403 | #ifdef CONFIG_MEMCG_SWAP |
8c7c6e34 KH |
4404 | /* |
4405 | * called from swap_entry_free(). remove record in swap_cgroup and | |
4406 | * uncharge "memsw" account. | |
4407 | */ | |
4408 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 4409 | { |
8c7c6e34 | 4410 | struct mem_cgroup *memcg; |
a3b2d692 | 4411 | unsigned short id; |
8c7c6e34 KH |
4412 | |
4413 | if (!do_swap_account) | |
4414 | return; | |
4415 | ||
a3b2d692 KH |
4416 | id = swap_cgroup_record(ent, 0); |
4417 | rcu_read_lock(); | |
4418 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 4419 | if (memcg) { |
a3b2d692 KH |
4420 | /* |
4421 | * We uncharge this because swap is freed. | |
4422 | * This memcg can be obsolete one. We avoid calling css_tryget | |
4423 | */ | |
0c3e73e8 | 4424 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 4425 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 4426 | mem_cgroup_swap_statistics(memcg, false); |
4050377b | 4427 | css_put(&memcg->css); |
8c7c6e34 | 4428 | } |
a3b2d692 | 4429 | rcu_read_unlock(); |
d13d1443 | 4430 | } |
02491447 DN |
4431 | |
4432 | /** | |
4433 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
4434 | * @entry: swap entry to be moved | |
4435 | * @from: mem_cgroup which the entry is moved from | |
4436 | * @to: mem_cgroup which the entry is moved to | |
4437 | * | |
4438 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
4439 | * as the mem_cgroup's id of @from. | |
4440 | * | |
4441 | * Returns 0 on success, -EINVAL on failure. | |
4442 | * | |
4443 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
4444 | * both res and memsw, and called css_get(). | |
4445 | */ | |
4446 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 4447 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
4448 | { |
4449 | unsigned short old_id, new_id; | |
4450 | ||
4451 | old_id = css_id(&from->css); | |
4452 | new_id = css_id(&to->css); | |
4453 | ||
4454 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 4455 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 4456 | mem_cgroup_swap_statistics(to, true); |
02491447 | 4457 | /* |
483c30b5 DN |
4458 | * This function is only called from task migration context now. |
4459 | * It postpones res_counter and refcount handling till the end | |
4460 | * of task migration(mem_cgroup_clear_mc()) for performance | |
4050377b LZ |
4461 | * improvement. But we cannot postpone css_get(to) because if |
4462 | * the process that has been moved to @to does swap-in, the | |
4463 | * refcount of @to might be decreased to 0. | |
4464 | * | |
4465 | * We are in attach() phase, so the cgroup is guaranteed to be | |
4466 | * alive, so we can just call css_get(). | |
02491447 | 4467 | */ |
4050377b | 4468 | css_get(&to->css); |
02491447 DN |
4469 | return 0; |
4470 | } | |
4471 | return -EINVAL; | |
4472 | } | |
4473 | #else | |
4474 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 4475 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
4476 | { |
4477 | return -EINVAL; | |
4478 | } | |
8c7c6e34 | 4479 | #endif |
d13d1443 | 4480 | |
ae41be37 | 4481 | /* |
01b1ae63 KH |
4482 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
4483 | * page belongs to. | |
ae41be37 | 4484 | */ |
0030f535 JW |
4485 | void mem_cgroup_prepare_migration(struct page *page, struct page *newpage, |
4486 | struct mem_cgroup **memcgp) | |
ae41be37 | 4487 | { |
c0ff4b85 | 4488 | struct mem_cgroup *memcg = NULL; |
b32967ff | 4489 | unsigned int nr_pages = 1; |
7ec99d62 | 4490 | struct page_cgroup *pc; |
ac39cf8c | 4491 | enum charge_type ctype; |
8869b8f6 | 4492 | |
72835c86 | 4493 | *memcgp = NULL; |
56039efa | 4494 | |
f8d66542 | 4495 | if (mem_cgroup_disabled()) |
0030f535 | 4496 | return; |
4077960e | 4497 | |
b32967ff MG |
4498 | if (PageTransHuge(page)) |
4499 | nr_pages <<= compound_order(page); | |
4500 | ||
52d4b9ac KH |
4501 | pc = lookup_page_cgroup(page); |
4502 | lock_page_cgroup(pc); | |
4503 | if (PageCgroupUsed(pc)) { | |
c0ff4b85 R |
4504 | memcg = pc->mem_cgroup; |
4505 | css_get(&memcg->css); | |
ac39cf8c AM |
4506 | /* |
4507 | * At migrating an anonymous page, its mapcount goes down | |
4508 | * to 0 and uncharge() will be called. But, even if it's fully | |
4509 | * unmapped, migration may fail and this page has to be | |
4510 | * charged again. We set MIGRATION flag here and delay uncharge | |
4511 | * until end_migration() is called | |
4512 | * | |
4513 | * Corner Case Thinking | |
4514 | * A) | |
4515 | * When the old page was mapped as Anon and it's unmap-and-freed | |
4516 | * while migration was ongoing. | |
4517 | * If unmap finds the old page, uncharge() of it will be delayed | |
4518 | * until end_migration(). If unmap finds a new page, it's | |
4519 | * uncharged when it make mapcount to be 1->0. If unmap code | |
4520 | * finds swap_migration_entry, the new page will not be mapped | |
4521 | * and end_migration() will find it(mapcount==0). | |
4522 | * | |
4523 | * B) | |
4524 | * When the old page was mapped but migraion fails, the kernel | |
4525 | * remaps it. A charge for it is kept by MIGRATION flag even | |
4526 | * if mapcount goes down to 0. We can do remap successfully | |
4527 | * without charging it again. | |
4528 | * | |
4529 | * C) | |
4530 | * The "old" page is under lock_page() until the end of | |
4531 | * migration, so, the old page itself will not be swapped-out. | |
4532 | * If the new page is swapped out before end_migraton, our | |
4533 | * hook to usual swap-out path will catch the event. | |
4534 | */ | |
4535 | if (PageAnon(page)) | |
4536 | SetPageCgroupMigration(pc); | |
e8589cc1 | 4537 | } |
52d4b9ac | 4538 | unlock_page_cgroup(pc); |
ac39cf8c AM |
4539 | /* |
4540 | * If the page is not charged at this point, | |
4541 | * we return here. | |
4542 | */ | |
c0ff4b85 | 4543 | if (!memcg) |
0030f535 | 4544 | return; |
01b1ae63 | 4545 | |
72835c86 | 4546 | *memcgp = memcg; |
ac39cf8c AM |
4547 | /* |
4548 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
4549 | * is called before end_migration, we can catch all events on this new | |
4550 | * page. In the case new page is migrated but not remapped, new page's | |
4551 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
4552 | */ | |
ac39cf8c | 4553 | if (PageAnon(page)) |
41326c17 | 4554 | ctype = MEM_CGROUP_CHARGE_TYPE_ANON; |
ac39cf8c | 4555 | else |
62ba7442 | 4556 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; |
0030f535 JW |
4557 | /* |
4558 | * The page is committed to the memcg, but it's not actually | |
4559 | * charged to the res_counter since we plan on replacing the | |
4560 | * old one and only one page is going to be left afterwards. | |
4561 | */ | |
b32967ff | 4562 | __mem_cgroup_commit_charge(memcg, newpage, nr_pages, ctype, false); |
ae41be37 | 4563 | } |
8869b8f6 | 4564 | |
69029cd5 | 4565 | /* remove redundant charge if migration failed*/ |
c0ff4b85 | 4566 | void mem_cgroup_end_migration(struct mem_cgroup *memcg, |
50de1dd9 | 4567 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 4568 | { |
ac39cf8c | 4569 | struct page *used, *unused; |
01b1ae63 | 4570 | struct page_cgroup *pc; |
b2402857 | 4571 | bool anon; |
01b1ae63 | 4572 | |
c0ff4b85 | 4573 | if (!memcg) |
01b1ae63 | 4574 | return; |
b25ed609 | 4575 | |
50de1dd9 | 4576 | if (!migration_ok) { |
ac39cf8c AM |
4577 | used = oldpage; |
4578 | unused = newpage; | |
01b1ae63 | 4579 | } else { |
ac39cf8c | 4580 | used = newpage; |
01b1ae63 KH |
4581 | unused = oldpage; |
4582 | } | |
0030f535 | 4583 | anon = PageAnon(used); |
7d188958 JW |
4584 | __mem_cgroup_uncharge_common(unused, |
4585 | anon ? MEM_CGROUP_CHARGE_TYPE_ANON | |
4586 | : MEM_CGROUP_CHARGE_TYPE_CACHE, | |
4587 | true); | |
0030f535 | 4588 | css_put(&memcg->css); |
69029cd5 | 4589 | /* |
ac39cf8c AM |
4590 | * We disallowed uncharge of pages under migration because mapcount |
4591 | * of the page goes down to zero, temporarly. | |
4592 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 4593 | */ |
ac39cf8c AM |
4594 | pc = lookup_page_cgroup(oldpage); |
4595 | lock_page_cgroup(pc); | |
4596 | ClearPageCgroupMigration(pc); | |
4597 | unlock_page_cgroup(pc); | |
ac39cf8c | 4598 | |
01b1ae63 | 4599 | /* |
ac39cf8c AM |
4600 | * If a page is a file cache, radix-tree replacement is very atomic |
4601 | * and we can skip this check. When it was an Anon page, its mapcount | |
4602 | * goes down to 0. But because we added MIGRATION flage, it's not | |
4603 | * uncharged yet. There are several case but page->mapcount check | |
4604 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
4605 | * check. (see prepare_charge() also) | |
69029cd5 | 4606 | */ |
b2402857 | 4607 | if (anon) |
ac39cf8c | 4608 | mem_cgroup_uncharge_page(used); |
ae41be37 | 4609 | } |
78fb7466 | 4610 | |
ab936cbc KH |
4611 | /* |
4612 | * At replace page cache, newpage is not under any memcg but it's on | |
4613 | * LRU. So, this function doesn't touch res_counter but handles LRU | |
4614 | * in correct way. Both pages are locked so we cannot race with uncharge. | |
4615 | */ | |
4616 | void mem_cgroup_replace_page_cache(struct page *oldpage, | |
4617 | struct page *newpage) | |
4618 | { | |
bde05d1c | 4619 | struct mem_cgroup *memcg = NULL; |
ab936cbc | 4620 | struct page_cgroup *pc; |
ab936cbc | 4621 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
ab936cbc KH |
4622 | |
4623 | if (mem_cgroup_disabled()) | |
4624 | return; | |
4625 | ||
4626 | pc = lookup_page_cgroup(oldpage); | |
4627 | /* fix accounting on old pages */ | |
4628 | lock_page_cgroup(pc); | |
bde05d1c HD |
4629 | if (PageCgroupUsed(pc)) { |
4630 | memcg = pc->mem_cgroup; | |
b070e65c | 4631 | mem_cgroup_charge_statistics(memcg, oldpage, false, -1); |
bde05d1c HD |
4632 | ClearPageCgroupUsed(pc); |
4633 | } | |
ab936cbc KH |
4634 | unlock_page_cgroup(pc); |
4635 | ||
bde05d1c HD |
4636 | /* |
4637 | * When called from shmem_replace_page(), in some cases the | |
4638 | * oldpage has already been charged, and in some cases not. | |
4639 | */ | |
4640 | if (!memcg) | |
4641 | return; | |
ab936cbc KH |
4642 | /* |
4643 | * Even if newpage->mapping was NULL before starting replacement, | |
4644 | * the newpage may be on LRU(or pagevec for LRU) already. We lock | |
4645 | * LRU while we overwrite pc->mem_cgroup. | |
4646 | */ | |
ce587e65 | 4647 | __mem_cgroup_commit_charge(memcg, newpage, 1, type, true); |
ab936cbc KH |
4648 | } |
4649 | ||
f212ad7c DN |
4650 | #ifdef CONFIG_DEBUG_VM |
4651 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
4652 | { | |
4653 | struct page_cgroup *pc; | |
4654 | ||
4655 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
4656 | /* |
4657 | * Can be NULL while feeding pages into the page allocator for | |
4658 | * the first time, i.e. during boot or memory hotplug; | |
4659 | * or when mem_cgroup_disabled(). | |
4660 | */ | |
f212ad7c DN |
4661 | if (likely(pc) && PageCgroupUsed(pc)) |
4662 | return pc; | |
4663 | return NULL; | |
4664 | } | |
4665 | ||
4666 | bool mem_cgroup_bad_page_check(struct page *page) | |
4667 | { | |
4668 | if (mem_cgroup_disabled()) | |
4669 | return false; | |
4670 | ||
4671 | return lookup_page_cgroup_used(page) != NULL; | |
4672 | } | |
4673 | ||
4674 | void mem_cgroup_print_bad_page(struct page *page) | |
4675 | { | |
4676 | struct page_cgroup *pc; | |
4677 | ||
4678 | pc = lookup_page_cgroup_used(page); | |
4679 | if (pc) { | |
d045197f AM |
4680 | pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n", |
4681 | pc, pc->flags, pc->mem_cgroup); | |
f212ad7c DN |
4682 | } |
4683 | } | |
4684 | #endif | |
4685 | ||
d38d2a75 | 4686 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 4687 | unsigned long long val) |
628f4235 | 4688 | { |
81d39c20 | 4689 | int retry_count; |
3c11ecf4 | 4690 | u64 memswlimit, memlimit; |
628f4235 | 4691 | int ret = 0; |
81d39c20 KH |
4692 | int children = mem_cgroup_count_children(memcg); |
4693 | u64 curusage, oldusage; | |
3c11ecf4 | 4694 | int enlarge; |
81d39c20 KH |
4695 | |
4696 | /* | |
4697 | * For keeping hierarchical_reclaim simple, how long we should retry | |
4698 | * is depends on callers. We set our retry-count to be function | |
4699 | * of # of children which we should visit in this loop. | |
4700 | */ | |
4701 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
4702 | ||
4703 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 4704 | |
3c11ecf4 | 4705 | enlarge = 0; |
8c7c6e34 | 4706 | while (retry_count) { |
628f4235 KH |
4707 | if (signal_pending(current)) { |
4708 | ret = -EINTR; | |
4709 | break; | |
4710 | } | |
8c7c6e34 KH |
4711 | /* |
4712 | * Rather than hide all in some function, I do this in | |
4713 | * open coded manner. You see what this really does. | |
aaad153e | 4714 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
4715 | */ |
4716 | mutex_lock(&set_limit_mutex); | |
4717 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
4718 | if (memswlimit < val) { | |
4719 | ret = -EINVAL; | |
4720 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
4721 | break; |
4722 | } | |
3c11ecf4 KH |
4723 | |
4724 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4725 | if (memlimit < val) | |
4726 | enlarge = 1; | |
4727 | ||
8c7c6e34 | 4728 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
4729 | if (!ret) { |
4730 | if (memswlimit == val) | |
4731 | memcg->memsw_is_minimum = true; | |
4732 | else | |
4733 | memcg->memsw_is_minimum = false; | |
4734 | } | |
8c7c6e34 KH |
4735 | mutex_unlock(&set_limit_mutex); |
4736 | ||
4737 | if (!ret) | |
4738 | break; | |
4739 | ||
5660048c JW |
4740 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
4741 | MEM_CGROUP_RECLAIM_SHRINK); | |
81d39c20 KH |
4742 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
4743 | /* Usage is reduced ? */ | |
f894ffa8 | 4744 | if (curusage >= oldusage) |
81d39c20 KH |
4745 | retry_count--; |
4746 | else | |
4747 | oldusage = curusage; | |
8c7c6e34 | 4748 | } |
3c11ecf4 KH |
4749 | if (!ret && enlarge) |
4750 | memcg_oom_recover(memcg); | |
14797e23 | 4751 | |
8c7c6e34 KH |
4752 | return ret; |
4753 | } | |
4754 | ||
338c8431 LZ |
4755 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
4756 | unsigned long long val) | |
8c7c6e34 | 4757 | { |
81d39c20 | 4758 | int retry_count; |
3c11ecf4 | 4759 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
4760 | int children = mem_cgroup_count_children(memcg); |
4761 | int ret = -EBUSY; | |
3c11ecf4 | 4762 | int enlarge = 0; |
8c7c6e34 | 4763 | |
81d39c20 | 4764 | /* see mem_cgroup_resize_res_limit */ |
f894ffa8 | 4765 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; |
81d39c20 | 4766 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
8c7c6e34 KH |
4767 | while (retry_count) { |
4768 | if (signal_pending(current)) { | |
4769 | ret = -EINTR; | |
4770 | break; | |
4771 | } | |
4772 | /* | |
4773 | * Rather than hide all in some function, I do this in | |
4774 | * open coded manner. You see what this really does. | |
aaad153e | 4775 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
4776 | */ |
4777 | mutex_lock(&set_limit_mutex); | |
4778 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4779 | if (memlimit > val) { | |
4780 | ret = -EINVAL; | |
4781 | mutex_unlock(&set_limit_mutex); | |
4782 | break; | |
4783 | } | |
3c11ecf4 KH |
4784 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
4785 | if (memswlimit < val) | |
4786 | enlarge = 1; | |
8c7c6e34 | 4787 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
4788 | if (!ret) { |
4789 | if (memlimit == val) | |
4790 | memcg->memsw_is_minimum = true; | |
4791 | else | |
4792 | memcg->memsw_is_minimum = false; | |
4793 | } | |
8c7c6e34 KH |
4794 | mutex_unlock(&set_limit_mutex); |
4795 | ||
4796 | if (!ret) | |
4797 | break; | |
4798 | ||
5660048c JW |
4799 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
4800 | MEM_CGROUP_RECLAIM_NOSWAP | | |
4801 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 4802 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 4803 | /* Usage is reduced ? */ |
8c7c6e34 | 4804 | if (curusage >= oldusage) |
628f4235 | 4805 | retry_count--; |
81d39c20 KH |
4806 | else |
4807 | oldusage = curusage; | |
628f4235 | 4808 | } |
3c11ecf4 KH |
4809 | if (!ret && enlarge) |
4810 | memcg_oom_recover(memcg); | |
628f4235 KH |
4811 | return ret; |
4812 | } | |
4813 | ||
0608f43d AM |
4814 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
4815 | gfp_t gfp_mask, | |
4816 | unsigned long *total_scanned) | |
4817 | { | |
4818 | unsigned long nr_reclaimed = 0; | |
4819 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
4820 | unsigned long reclaimed; | |
4821 | int loop = 0; | |
4822 | struct mem_cgroup_tree_per_zone *mctz; | |
4823 | unsigned long long excess; | |
4824 | unsigned long nr_scanned; | |
4825 | ||
4826 | if (order > 0) | |
4827 | return 0; | |
4828 | ||
4829 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
4830 | /* | |
4831 | * This loop can run a while, specially if mem_cgroup's continuously | |
4832 | * keep exceeding their soft limit and putting the system under | |
4833 | * pressure | |
4834 | */ | |
4835 | do { | |
4836 | if (next_mz) | |
4837 | mz = next_mz; | |
4838 | else | |
4839 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
4840 | if (!mz) | |
4841 | break; | |
4842 | ||
4843 | nr_scanned = 0; | |
4844 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
4845 | gfp_mask, &nr_scanned); | |
4846 | nr_reclaimed += reclaimed; | |
4847 | *total_scanned += nr_scanned; | |
4848 | spin_lock(&mctz->lock); | |
4849 | ||
4850 | /* | |
4851 | * If we failed to reclaim anything from this memory cgroup | |
4852 | * it is time to move on to the next cgroup | |
4853 | */ | |
4854 | next_mz = NULL; | |
4855 | if (!reclaimed) { | |
4856 | do { | |
4857 | /* | |
4858 | * Loop until we find yet another one. | |
4859 | * | |
4860 | * By the time we get the soft_limit lock | |
4861 | * again, someone might have aded the | |
4862 | * group back on the RB tree. Iterate to | |
4863 | * make sure we get a different mem. | |
4864 | * mem_cgroup_largest_soft_limit_node returns | |
4865 | * NULL if no other cgroup is present on | |
4866 | * the tree | |
4867 | */ | |
4868 | next_mz = | |
4869 | __mem_cgroup_largest_soft_limit_node(mctz); | |
4870 | if (next_mz == mz) | |
4871 | css_put(&next_mz->memcg->css); | |
4872 | else /* next_mz == NULL or other memcg */ | |
4873 | break; | |
4874 | } while (1); | |
4875 | } | |
4876 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); | |
4877 | excess = res_counter_soft_limit_excess(&mz->memcg->res); | |
4878 | /* | |
4879 | * One school of thought says that we should not add | |
4880 | * back the node to the tree if reclaim returns 0. | |
4881 | * But our reclaim could return 0, simply because due | |
4882 | * to priority we are exposing a smaller subset of | |
4883 | * memory to reclaim from. Consider this as a longer | |
4884 | * term TODO. | |
4885 | */ | |
4886 | /* If excess == 0, no tree ops */ | |
4887 | __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess); | |
4888 | spin_unlock(&mctz->lock); | |
4889 | css_put(&mz->memcg->css); | |
4890 | loop++; | |
4891 | /* | |
4892 | * Could not reclaim anything and there are no more | |
4893 | * mem cgroups to try or we seem to be looping without | |
4894 | * reclaiming anything. | |
4895 | */ | |
4896 | if (!nr_reclaimed && | |
4897 | (next_mz == NULL || | |
4898 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
4899 | break; | |
4900 | } while (!nr_reclaimed); | |
4901 | if (next_mz) | |
4902 | css_put(&next_mz->memcg->css); | |
4903 | return nr_reclaimed; | |
4904 | } | |
4905 | ||
2ef37d3f MH |
4906 | /** |
4907 | * mem_cgroup_force_empty_list - clears LRU of a group | |
4908 | * @memcg: group to clear | |
4909 | * @node: NUMA node | |
4910 | * @zid: zone id | |
4911 | * @lru: lru to to clear | |
4912 | * | |
3c935d18 | 4913 | * Traverse a specified page_cgroup list and try to drop them all. This doesn't |
2ef37d3f MH |
4914 | * reclaim the pages page themselves - pages are moved to the parent (or root) |
4915 | * group. | |
cc847582 | 4916 | */ |
2ef37d3f | 4917 | static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
08e552c6 | 4918 | int node, int zid, enum lru_list lru) |
cc847582 | 4919 | { |
bea8c150 | 4920 | struct lruvec *lruvec; |
2ef37d3f | 4921 | unsigned long flags; |
072c56c1 | 4922 | struct list_head *list; |
925b7673 JW |
4923 | struct page *busy; |
4924 | struct zone *zone; | |
072c56c1 | 4925 | |
08e552c6 | 4926 | zone = &NODE_DATA(node)->node_zones[zid]; |
bea8c150 HD |
4927 | lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
4928 | list = &lruvec->lists[lru]; | |
cc847582 | 4929 | |
f817ed48 | 4930 | busy = NULL; |
2ef37d3f | 4931 | do { |
925b7673 | 4932 | struct page_cgroup *pc; |
5564e88b JW |
4933 | struct page *page; |
4934 | ||
08e552c6 | 4935 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 4936 | if (list_empty(list)) { |
08e552c6 | 4937 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 4938 | break; |
f817ed48 | 4939 | } |
925b7673 JW |
4940 | page = list_entry(list->prev, struct page, lru); |
4941 | if (busy == page) { | |
4942 | list_move(&page->lru, list); | |
648bcc77 | 4943 | busy = NULL; |
08e552c6 | 4944 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
4945 | continue; |
4946 | } | |
08e552c6 | 4947 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 4948 | |
925b7673 | 4949 | pc = lookup_page_cgroup(page); |
5564e88b | 4950 | |
3c935d18 | 4951 | if (mem_cgroup_move_parent(page, pc, memcg)) { |
f817ed48 | 4952 | /* found lock contention or "pc" is obsolete. */ |
925b7673 | 4953 | busy = page; |
f817ed48 KH |
4954 | cond_resched(); |
4955 | } else | |
4956 | busy = NULL; | |
2ef37d3f | 4957 | } while (!list_empty(list)); |
cc847582 KH |
4958 | } |
4959 | ||
4960 | /* | |
c26251f9 MH |
4961 | * make mem_cgroup's charge to be 0 if there is no task by moving |
4962 | * all the charges and pages to the parent. | |
cc847582 | 4963 | * This enables deleting this mem_cgroup. |
c26251f9 MH |
4964 | * |
4965 | * Caller is responsible for holding css reference on the memcg. | |
cc847582 | 4966 | */ |
ab5196c2 | 4967 | static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg) |
cc847582 | 4968 | { |
c26251f9 | 4969 | int node, zid; |
bea207c8 | 4970 | u64 usage; |
f817ed48 | 4971 | |
fce66477 | 4972 | do { |
52d4b9ac KH |
4973 | /* This is for making all *used* pages to be on LRU. */ |
4974 | lru_add_drain_all(); | |
c0ff4b85 | 4975 | drain_all_stock_sync(memcg); |
c0ff4b85 | 4976 | mem_cgroup_start_move(memcg); |
31aaea4a | 4977 | for_each_node_state(node, N_MEMORY) { |
2ef37d3f | 4978 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
f156ab93 HD |
4979 | enum lru_list lru; |
4980 | for_each_lru(lru) { | |
2ef37d3f | 4981 | mem_cgroup_force_empty_list(memcg, |
f156ab93 | 4982 | node, zid, lru); |
f817ed48 | 4983 | } |
1ecaab2b | 4984 | } |
f817ed48 | 4985 | } |
c0ff4b85 R |
4986 | mem_cgroup_end_move(memcg); |
4987 | memcg_oom_recover(memcg); | |
52d4b9ac | 4988 | cond_resched(); |
f817ed48 | 4989 | |
2ef37d3f | 4990 | /* |
bea207c8 GC |
4991 | * Kernel memory may not necessarily be trackable to a specific |
4992 | * process. So they are not migrated, and therefore we can't | |
4993 | * expect their value to drop to 0 here. | |
4994 | * Having res filled up with kmem only is enough. | |
4995 | * | |
2ef37d3f MH |
4996 | * This is a safety check because mem_cgroup_force_empty_list |
4997 | * could have raced with mem_cgroup_replace_page_cache callers | |
4998 | * so the lru seemed empty but the page could have been added | |
4999 | * right after the check. RES_USAGE should be safe as we always | |
5000 | * charge before adding to the LRU. | |
5001 | */ | |
bea207c8 GC |
5002 | usage = res_counter_read_u64(&memcg->res, RES_USAGE) - |
5003 | res_counter_read_u64(&memcg->kmem, RES_USAGE); | |
5004 | } while (usage > 0); | |
c26251f9 MH |
5005 | } |
5006 | ||
b5f99b53 GC |
5007 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
5008 | { | |
696ac172 JW |
5009 | lockdep_assert_held(&memcg_create_mutex); |
5010 | /* | |
5011 | * The lock does not prevent addition or deletion to the list | |
5012 | * of children, but it prevents a new child from being | |
5013 | * initialized based on this parent in css_online(), so it's | |
5014 | * enough to decide whether hierarchically inherited | |
5015 | * attributes can still be changed or not. | |
5016 | */ | |
5017 | return memcg->use_hierarchy && | |
5018 | !list_empty(&memcg->css.cgroup->children); | |
b5f99b53 GC |
5019 | } |
5020 | ||
c26251f9 MH |
5021 | /* |
5022 | * Reclaims as many pages from the given memcg as possible and moves | |
5023 | * the rest to the parent. | |
5024 | * | |
5025 | * Caller is responsible for holding css reference for memcg. | |
5026 | */ | |
5027 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
5028 | { | |
5029 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
5030 | struct cgroup *cgrp = memcg->css.cgroup; | |
f817ed48 | 5031 | |
c1e862c1 | 5032 | /* returns EBUSY if there is a task or if we come here twice. */ |
c26251f9 MH |
5033 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
5034 | return -EBUSY; | |
5035 | ||
c1e862c1 KH |
5036 | /* we call try-to-free pages for make this cgroup empty */ |
5037 | lru_add_drain_all(); | |
f817ed48 | 5038 | /* try to free all pages in this cgroup */ |
569530fb | 5039 | while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) { |
f817ed48 | 5040 | int progress; |
c1e862c1 | 5041 | |
c26251f9 MH |
5042 | if (signal_pending(current)) |
5043 | return -EINTR; | |
5044 | ||
c0ff4b85 | 5045 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, |
185efc0f | 5046 | false); |
c1e862c1 | 5047 | if (!progress) { |
f817ed48 | 5048 | nr_retries--; |
c1e862c1 | 5049 | /* maybe some writeback is necessary */ |
8aa7e847 | 5050 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 5051 | } |
f817ed48 KH |
5052 | |
5053 | } | |
08e552c6 | 5054 | lru_add_drain(); |
ab5196c2 MH |
5055 | mem_cgroup_reparent_charges(memcg); |
5056 | ||
5057 | return 0; | |
cc847582 KH |
5058 | } |
5059 | ||
182446d0 TH |
5060 | static int mem_cgroup_force_empty_write(struct cgroup_subsys_state *css, |
5061 | unsigned int event) | |
c1e862c1 | 5062 | { |
182446d0 | 5063 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c26251f9 | 5064 | |
d8423011 MH |
5065 | if (mem_cgroup_is_root(memcg)) |
5066 | return -EINVAL; | |
c33bd835 | 5067 | return mem_cgroup_force_empty(memcg); |
c1e862c1 KH |
5068 | } |
5069 | ||
182446d0 TH |
5070 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
5071 | struct cftype *cft) | |
18f59ea7 | 5072 | { |
182446d0 | 5073 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
5074 | } |
5075 | ||
182446d0 TH |
5076 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
5077 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
5078 | { |
5079 | int retval = 0; | |
182446d0 | 5080 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
63876986 | 5081 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(css_parent(&memcg->css)); |
18f59ea7 | 5082 | |
0999821b | 5083 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
5084 | |
5085 | if (memcg->use_hierarchy == val) | |
5086 | goto out; | |
5087 | ||
18f59ea7 | 5088 | /* |
af901ca1 | 5089 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
5090 | * in the child subtrees. If it is unset, then the change can |
5091 | * occur, provided the current cgroup has no children. | |
5092 | * | |
5093 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
5094 | * set if there are no children. | |
5095 | */ | |
c0ff4b85 | 5096 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 5097 | (val == 1 || val == 0)) { |
696ac172 | 5098 | if (list_empty(&memcg->css.cgroup->children)) |
c0ff4b85 | 5099 | memcg->use_hierarchy = val; |
18f59ea7 BS |
5100 | else |
5101 | retval = -EBUSY; | |
5102 | } else | |
5103 | retval = -EINVAL; | |
567fb435 GC |
5104 | |
5105 | out: | |
0999821b | 5106 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
5107 | |
5108 | return retval; | |
5109 | } | |
5110 | ||
0c3e73e8 | 5111 | |
c0ff4b85 | 5112 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg, |
7a159cc9 | 5113 | enum mem_cgroup_stat_index idx) |
0c3e73e8 | 5114 | { |
7d74b06f | 5115 | struct mem_cgroup *iter; |
7a159cc9 | 5116 | long val = 0; |
0c3e73e8 | 5117 | |
7a159cc9 | 5118 | /* Per-cpu values can be negative, use a signed accumulator */ |
c0ff4b85 | 5119 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f KH |
5120 | val += mem_cgroup_read_stat(iter, idx); |
5121 | ||
5122 | if (val < 0) /* race ? */ | |
5123 | val = 0; | |
5124 | return val; | |
0c3e73e8 BS |
5125 | } |
5126 | ||
c0ff4b85 | 5127 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
104f3928 | 5128 | { |
7d74b06f | 5129 | u64 val; |
104f3928 | 5130 | |
c0ff4b85 | 5131 | if (!mem_cgroup_is_root(memcg)) { |
104f3928 | 5132 | if (!swap) |
65c64ce8 | 5133 | return res_counter_read_u64(&memcg->res, RES_USAGE); |
104f3928 | 5134 | else |
65c64ce8 | 5135 | return res_counter_read_u64(&memcg->memsw, RES_USAGE); |
104f3928 KS |
5136 | } |
5137 | ||
b070e65c DR |
5138 | /* |
5139 | * Transparent hugepages are still accounted for in MEM_CGROUP_STAT_RSS | |
5140 | * as well as in MEM_CGROUP_STAT_RSS_HUGE. | |
5141 | */ | |
c0ff4b85 R |
5142 | val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE); |
5143 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS); | |
104f3928 | 5144 | |
7d74b06f | 5145 | if (swap) |
bff6bb83 | 5146 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP); |
104f3928 KS |
5147 | |
5148 | return val << PAGE_SHIFT; | |
5149 | } | |
5150 | ||
182446d0 TH |
5151 | static ssize_t mem_cgroup_read(struct cgroup_subsys_state *css, |
5152 | struct cftype *cft, struct file *file, | |
5153 | char __user *buf, size_t nbytes, loff_t *ppos) | |
8cdea7c0 | 5154 | { |
182446d0 | 5155 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
af36f906 | 5156 | char str[64]; |
104f3928 | 5157 | u64 val; |
86ae53e1 GC |
5158 | int name, len; |
5159 | enum res_type type; | |
8c7c6e34 KH |
5160 | |
5161 | type = MEMFILE_TYPE(cft->private); | |
5162 | name = MEMFILE_ATTR(cft->private); | |
af36f906 | 5163 | |
8c7c6e34 KH |
5164 | switch (type) { |
5165 | case _MEM: | |
104f3928 | 5166 | if (name == RES_USAGE) |
c0ff4b85 | 5167 | val = mem_cgroup_usage(memcg, false); |
104f3928 | 5168 | else |
c0ff4b85 | 5169 | val = res_counter_read_u64(&memcg->res, name); |
8c7c6e34 KH |
5170 | break; |
5171 | case _MEMSWAP: | |
104f3928 | 5172 | if (name == RES_USAGE) |
c0ff4b85 | 5173 | val = mem_cgroup_usage(memcg, true); |
104f3928 | 5174 | else |
c0ff4b85 | 5175 | val = res_counter_read_u64(&memcg->memsw, name); |
8c7c6e34 | 5176 | break; |
510fc4e1 GC |
5177 | case _KMEM: |
5178 | val = res_counter_read_u64(&memcg->kmem, name); | |
5179 | break; | |
8c7c6e34 KH |
5180 | default: |
5181 | BUG(); | |
8c7c6e34 | 5182 | } |
af36f906 TH |
5183 | |
5184 | len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val); | |
5185 | return simple_read_from_buffer(buf, nbytes, ppos, str, len); | |
8cdea7c0 | 5186 | } |
510fc4e1 | 5187 | |
182446d0 | 5188 | static int memcg_update_kmem_limit(struct cgroup_subsys_state *css, u64 val) |
510fc4e1 GC |
5189 | { |
5190 | int ret = -EINVAL; | |
5191 | #ifdef CONFIG_MEMCG_KMEM | |
182446d0 | 5192 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
510fc4e1 GC |
5193 | /* |
5194 | * For simplicity, we won't allow this to be disabled. It also can't | |
5195 | * be changed if the cgroup has children already, or if tasks had | |
5196 | * already joined. | |
5197 | * | |
5198 | * If tasks join before we set the limit, a person looking at | |
5199 | * kmem.usage_in_bytes will have no way to determine when it took | |
5200 | * place, which makes the value quite meaningless. | |
5201 | * | |
5202 | * After it first became limited, changes in the value of the limit are | |
5203 | * of course permitted. | |
510fc4e1 | 5204 | */ |
0999821b | 5205 | mutex_lock(&memcg_create_mutex); |
510fc4e1 | 5206 | mutex_lock(&set_limit_mutex); |
6de5a8bf | 5207 | if (!memcg->kmem_account_flags && val != RES_COUNTER_MAX) { |
182446d0 | 5208 | if (cgroup_task_count(css->cgroup) || memcg_has_children(memcg)) { |
510fc4e1 GC |
5209 | ret = -EBUSY; |
5210 | goto out; | |
5211 | } | |
5212 | ret = res_counter_set_limit(&memcg->kmem, val); | |
5213 | VM_BUG_ON(ret); | |
5214 | ||
55007d84 GC |
5215 | ret = memcg_update_cache_sizes(memcg); |
5216 | if (ret) { | |
6de5a8bf | 5217 | res_counter_set_limit(&memcg->kmem, RES_COUNTER_MAX); |
55007d84 GC |
5218 | goto out; |
5219 | } | |
692e89ab GC |
5220 | static_key_slow_inc(&memcg_kmem_enabled_key); |
5221 | /* | |
5222 | * setting the active bit after the inc will guarantee no one | |
5223 | * starts accounting before all call sites are patched | |
5224 | */ | |
5225 | memcg_kmem_set_active(memcg); | |
510fc4e1 GC |
5226 | } else |
5227 | ret = res_counter_set_limit(&memcg->kmem, val); | |
5228 | out: | |
5229 | mutex_unlock(&set_limit_mutex); | |
0999821b | 5230 | mutex_unlock(&memcg_create_mutex); |
510fc4e1 GC |
5231 | #endif |
5232 | return ret; | |
5233 | } | |
5234 | ||
6d043990 | 5235 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 | 5236 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 5237 | { |
55007d84 | 5238 | int ret = 0; |
510fc4e1 GC |
5239 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
5240 | if (!parent) | |
55007d84 GC |
5241 | goto out; |
5242 | ||
510fc4e1 | 5243 | memcg->kmem_account_flags = parent->kmem_account_flags; |
a8964b9b GC |
5244 | /* |
5245 | * When that happen, we need to disable the static branch only on those | |
5246 | * memcgs that enabled it. To achieve this, we would be forced to | |
5247 | * complicate the code by keeping track of which memcgs were the ones | |
5248 | * that actually enabled limits, and which ones got it from its | |
5249 | * parents. | |
5250 | * | |
5251 | * It is a lot simpler just to do static_key_slow_inc() on every child | |
5252 | * that is accounted. | |
5253 | */ | |
55007d84 GC |
5254 | if (!memcg_kmem_is_active(memcg)) |
5255 | goto out; | |
5256 | ||
5257 | /* | |
10d5ebf4 LZ |
5258 | * __mem_cgroup_free() will issue static_key_slow_dec() because this |
5259 | * memcg is active already. If the later initialization fails then the | |
5260 | * cgroup core triggers the cleanup so we do not have to do it here. | |
55007d84 | 5261 | */ |
55007d84 GC |
5262 | static_key_slow_inc(&memcg_kmem_enabled_key); |
5263 | ||
5264 | mutex_lock(&set_limit_mutex); | |
425c598d | 5265 | memcg_stop_kmem_account(); |
55007d84 | 5266 | ret = memcg_update_cache_sizes(memcg); |
425c598d | 5267 | memcg_resume_kmem_account(); |
55007d84 | 5268 | mutex_unlock(&set_limit_mutex); |
55007d84 GC |
5269 | out: |
5270 | return ret; | |
510fc4e1 | 5271 | } |
6d043990 | 5272 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 5273 | |
628f4235 KH |
5274 | /* |
5275 | * The user of this function is... | |
5276 | * RES_LIMIT. | |
5277 | */ | |
182446d0 | 5278 | static int mem_cgroup_write(struct cgroup_subsys_state *css, struct cftype *cft, |
856c13aa | 5279 | const char *buffer) |
8cdea7c0 | 5280 | { |
182446d0 | 5281 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
86ae53e1 GC |
5282 | enum res_type type; |
5283 | int name; | |
628f4235 KH |
5284 | unsigned long long val; |
5285 | int ret; | |
5286 | ||
8c7c6e34 KH |
5287 | type = MEMFILE_TYPE(cft->private); |
5288 | name = MEMFILE_ATTR(cft->private); | |
af36f906 | 5289 | |
8c7c6e34 | 5290 | switch (name) { |
628f4235 | 5291 | case RES_LIMIT: |
4b3bde4c BS |
5292 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
5293 | ret = -EINVAL; | |
5294 | break; | |
5295 | } | |
628f4235 KH |
5296 | /* This function does all necessary parse...reuse it */ |
5297 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
5298 | if (ret) |
5299 | break; | |
5300 | if (type == _MEM) | |
628f4235 | 5301 | ret = mem_cgroup_resize_limit(memcg, val); |
510fc4e1 | 5302 | else if (type == _MEMSWAP) |
8c7c6e34 | 5303 | ret = mem_cgroup_resize_memsw_limit(memcg, val); |
510fc4e1 | 5304 | else if (type == _KMEM) |
182446d0 | 5305 | ret = memcg_update_kmem_limit(css, val); |
510fc4e1 GC |
5306 | else |
5307 | return -EINVAL; | |
628f4235 | 5308 | break; |
296c81d8 BS |
5309 | case RES_SOFT_LIMIT: |
5310 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
5311 | if (ret) | |
5312 | break; | |
5313 | /* | |
5314 | * For memsw, soft limits are hard to implement in terms | |
5315 | * of semantics, for now, we support soft limits for | |
5316 | * control without swap | |
5317 | */ | |
5318 | if (type == _MEM) | |
5319 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
5320 | else | |
5321 | ret = -EINVAL; | |
5322 | break; | |
628f4235 KH |
5323 | default: |
5324 | ret = -EINVAL; /* should be BUG() ? */ | |
5325 | break; | |
5326 | } | |
5327 | return ret; | |
8cdea7c0 BS |
5328 | } |
5329 | ||
fee7b548 KH |
5330 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
5331 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
5332 | { | |
fee7b548 KH |
5333 | unsigned long long min_limit, min_memsw_limit, tmp; |
5334 | ||
5335 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
5336 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
fee7b548 KH |
5337 | if (!memcg->use_hierarchy) |
5338 | goto out; | |
5339 | ||
63876986 TH |
5340 | while (css_parent(&memcg->css)) { |
5341 | memcg = mem_cgroup_from_css(css_parent(&memcg->css)); | |
fee7b548 KH |
5342 | if (!memcg->use_hierarchy) |
5343 | break; | |
5344 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
5345 | min_limit = min(min_limit, tmp); | |
5346 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
5347 | min_memsw_limit = min(min_memsw_limit, tmp); | |
5348 | } | |
5349 | out: | |
5350 | *mem_limit = min_limit; | |
5351 | *memsw_limit = min_memsw_limit; | |
fee7b548 KH |
5352 | } |
5353 | ||
182446d0 | 5354 | static int mem_cgroup_reset(struct cgroup_subsys_state *css, unsigned int event) |
c84872e1 | 5355 | { |
182446d0 | 5356 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
86ae53e1 GC |
5357 | int name; |
5358 | enum res_type type; | |
c84872e1 | 5359 | |
8c7c6e34 KH |
5360 | type = MEMFILE_TYPE(event); |
5361 | name = MEMFILE_ATTR(event); | |
af36f906 | 5362 | |
8c7c6e34 | 5363 | switch (name) { |
29f2a4da | 5364 | case RES_MAX_USAGE: |
8c7c6e34 | 5365 | if (type == _MEM) |
c0ff4b85 | 5366 | res_counter_reset_max(&memcg->res); |
510fc4e1 | 5367 | else if (type == _MEMSWAP) |
c0ff4b85 | 5368 | res_counter_reset_max(&memcg->memsw); |
510fc4e1 GC |
5369 | else if (type == _KMEM) |
5370 | res_counter_reset_max(&memcg->kmem); | |
5371 | else | |
5372 | return -EINVAL; | |
29f2a4da PE |
5373 | break; |
5374 | case RES_FAILCNT: | |
8c7c6e34 | 5375 | if (type == _MEM) |
c0ff4b85 | 5376 | res_counter_reset_failcnt(&memcg->res); |
510fc4e1 | 5377 | else if (type == _MEMSWAP) |
c0ff4b85 | 5378 | res_counter_reset_failcnt(&memcg->memsw); |
510fc4e1 GC |
5379 | else if (type == _KMEM) |
5380 | res_counter_reset_failcnt(&memcg->kmem); | |
5381 | else | |
5382 | return -EINVAL; | |
29f2a4da PE |
5383 | break; |
5384 | } | |
f64c3f54 | 5385 | |
85cc59db | 5386 | return 0; |
c84872e1 PE |
5387 | } |
5388 | ||
182446d0 | 5389 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
5390 | struct cftype *cft) |
5391 | { | |
182446d0 | 5392 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
5393 | } |
5394 | ||
02491447 | 5395 | #ifdef CONFIG_MMU |
182446d0 | 5396 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
5397 | struct cftype *cft, u64 val) |
5398 | { | |
182446d0 | 5399 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 DN |
5400 | |
5401 | if (val >= (1 << NR_MOVE_TYPE)) | |
5402 | return -EINVAL; | |
ee5e8472 | 5403 | |
7dc74be0 | 5404 | /* |
ee5e8472 GC |
5405 | * No kind of locking is needed in here, because ->can_attach() will |
5406 | * check this value once in the beginning of the process, and then carry | |
5407 | * on with stale data. This means that changes to this value will only | |
5408 | * affect task migrations starting after the change. | |
7dc74be0 | 5409 | */ |
c0ff4b85 | 5410 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
5411 | return 0; |
5412 | } | |
02491447 | 5413 | #else |
182446d0 | 5414 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
5415 | struct cftype *cft, u64 val) |
5416 | { | |
5417 | return -ENOSYS; | |
5418 | } | |
5419 | #endif | |
7dc74be0 | 5420 | |
406eb0c9 | 5421 | #ifdef CONFIG_NUMA |
182446d0 TH |
5422 | static int memcg_numa_stat_show(struct cgroup_subsys_state *css, |
5423 | struct cftype *cft, struct seq_file *m) | |
406eb0c9 YH |
5424 | { |
5425 | int nid; | |
5426 | unsigned long total_nr, file_nr, anon_nr, unevictable_nr; | |
5427 | unsigned long node_nr; | |
182446d0 | 5428 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
406eb0c9 | 5429 | |
d79154bb | 5430 | total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL); |
406eb0c9 | 5431 | seq_printf(m, "total=%lu", total_nr); |
31aaea4a | 5432 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5433 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL); |
406eb0c9 YH |
5434 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5435 | } | |
5436 | seq_putc(m, '\n'); | |
5437 | ||
d79154bb | 5438 | file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE); |
406eb0c9 | 5439 | seq_printf(m, "file=%lu", file_nr); |
31aaea4a | 5440 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5441 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 5442 | LRU_ALL_FILE); |
406eb0c9 YH |
5443 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5444 | } | |
5445 | seq_putc(m, '\n'); | |
5446 | ||
d79154bb | 5447 | anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON); |
406eb0c9 | 5448 | seq_printf(m, "anon=%lu", anon_nr); |
31aaea4a | 5449 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5450 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 5451 | LRU_ALL_ANON); |
406eb0c9 YH |
5452 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5453 | } | |
5454 | seq_putc(m, '\n'); | |
5455 | ||
d79154bb | 5456 | unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE)); |
406eb0c9 | 5457 | seq_printf(m, "unevictable=%lu", unevictable_nr); |
31aaea4a | 5458 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5459 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 5460 | BIT(LRU_UNEVICTABLE)); |
406eb0c9 YH |
5461 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5462 | } | |
5463 | seq_putc(m, '\n'); | |
5464 | return 0; | |
5465 | } | |
5466 | #endif /* CONFIG_NUMA */ | |
5467 | ||
af7c4b0e JW |
5468 | static inline void mem_cgroup_lru_names_not_uptodate(void) |
5469 | { | |
5470 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); | |
5471 | } | |
5472 | ||
182446d0 | 5473 | static int memcg_stat_show(struct cgroup_subsys_state *css, struct cftype *cft, |
78ccf5b5 | 5474 | struct seq_file *m) |
d2ceb9b7 | 5475 | { |
182446d0 | 5476 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
af7c4b0e JW |
5477 | struct mem_cgroup *mi; |
5478 | unsigned int i; | |
406eb0c9 | 5479 | |
af7c4b0e | 5480 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 5481 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 5482 | continue; |
af7c4b0e JW |
5483 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
5484 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 5485 | } |
7b854121 | 5486 | |
af7c4b0e JW |
5487 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
5488 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
5489 | mem_cgroup_read_events(memcg, i)); | |
5490 | ||
5491 | for (i = 0; i < NR_LRU_LISTS; i++) | |
5492 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
5493 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
5494 | ||
14067bb3 | 5495 | /* Hierarchical information */ |
fee7b548 KH |
5496 | { |
5497 | unsigned long long limit, memsw_limit; | |
d79154bb | 5498 | memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit); |
78ccf5b5 | 5499 | seq_printf(m, "hierarchical_memory_limit %llu\n", limit); |
fee7b548 | 5500 | if (do_swap_account) |
78ccf5b5 JW |
5501 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
5502 | memsw_limit); | |
fee7b548 | 5503 | } |
7f016ee8 | 5504 | |
af7c4b0e JW |
5505 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
5506 | long long val = 0; | |
5507 | ||
bff6bb83 | 5508 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 5509 | continue; |
af7c4b0e JW |
5510 | for_each_mem_cgroup_tree(mi, memcg) |
5511 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
5512 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
5513 | } | |
5514 | ||
5515 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
5516 | unsigned long long val = 0; | |
5517 | ||
5518 | for_each_mem_cgroup_tree(mi, memcg) | |
5519 | val += mem_cgroup_read_events(mi, i); | |
5520 | seq_printf(m, "total_%s %llu\n", | |
5521 | mem_cgroup_events_names[i], val); | |
5522 | } | |
5523 | ||
5524 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
5525 | unsigned long long val = 0; | |
5526 | ||
5527 | for_each_mem_cgroup_tree(mi, memcg) | |
5528 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
5529 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 5530 | } |
14067bb3 | 5531 | |
7f016ee8 | 5532 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
5533 | { |
5534 | int nid, zid; | |
5535 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 5536 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
5537 | unsigned long recent_rotated[2] = {0, 0}; |
5538 | unsigned long recent_scanned[2] = {0, 0}; | |
5539 | ||
5540 | for_each_online_node(nid) | |
5541 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
d79154bb | 5542 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
89abfab1 | 5543 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 5544 | |
89abfab1 HD |
5545 | recent_rotated[0] += rstat->recent_rotated[0]; |
5546 | recent_rotated[1] += rstat->recent_rotated[1]; | |
5547 | recent_scanned[0] += rstat->recent_scanned[0]; | |
5548 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 5549 | } |
78ccf5b5 JW |
5550 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
5551 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
5552 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
5553 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
5554 | } |
5555 | #endif | |
5556 | ||
d2ceb9b7 KH |
5557 | return 0; |
5558 | } | |
5559 | ||
182446d0 TH |
5560 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
5561 | struct cftype *cft) | |
a7885eb8 | 5562 | { |
182446d0 | 5563 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 5564 | |
1f4c025b | 5565 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
5566 | } |
5567 | ||
182446d0 TH |
5568 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
5569 | struct cftype *cft, u64 val) | |
a7885eb8 | 5570 | { |
182446d0 | 5571 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
63876986 | 5572 | struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css)); |
a7885eb8 | 5573 | |
63876986 | 5574 | if (val > 100 || !parent) |
a7885eb8 KM |
5575 | return -EINVAL; |
5576 | ||
0999821b | 5577 | mutex_lock(&memcg_create_mutex); |
068b38c1 | 5578 | |
a7885eb8 | 5579 | /* If under hierarchy, only empty-root can set this value */ |
b5f99b53 | 5580 | if ((parent->use_hierarchy) || memcg_has_children(memcg)) { |
0999821b | 5581 | mutex_unlock(&memcg_create_mutex); |
a7885eb8 | 5582 | return -EINVAL; |
068b38c1 | 5583 | } |
a7885eb8 | 5584 | |
a7885eb8 | 5585 | memcg->swappiness = val; |
a7885eb8 | 5586 | |
0999821b | 5587 | mutex_unlock(&memcg_create_mutex); |
068b38c1 | 5588 | |
a7885eb8 KM |
5589 | return 0; |
5590 | } | |
5591 | ||
2e72b634 KS |
5592 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
5593 | { | |
5594 | struct mem_cgroup_threshold_ary *t; | |
5595 | u64 usage; | |
5596 | int i; | |
5597 | ||
5598 | rcu_read_lock(); | |
5599 | if (!swap) | |
2c488db2 | 5600 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 5601 | else |
2c488db2 | 5602 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
5603 | |
5604 | if (!t) | |
5605 | goto unlock; | |
5606 | ||
5607 | usage = mem_cgroup_usage(memcg, swap); | |
5608 | ||
5609 | /* | |
748dad36 | 5610 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
5611 | * If it's not true, a threshold was crossed after last |
5612 | * call of __mem_cgroup_threshold(). | |
5613 | */ | |
5407a562 | 5614 | i = t->current_threshold; |
2e72b634 KS |
5615 | |
5616 | /* | |
5617 | * Iterate backward over array of thresholds starting from | |
5618 | * current_threshold and check if a threshold is crossed. | |
5619 | * If none of thresholds below usage is crossed, we read | |
5620 | * only one element of the array here. | |
5621 | */ | |
5622 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
5623 | eventfd_signal(t->entries[i].eventfd, 1); | |
5624 | ||
5625 | /* i = current_threshold + 1 */ | |
5626 | i++; | |
5627 | ||
5628 | /* | |
5629 | * Iterate forward over array of thresholds starting from | |
5630 | * current_threshold+1 and check if a threshold is crossed. | |
5631 | * If none of thresholds above usage is crossed, we read | |
5632 | * only one element of the array here. | |
5633 | */ | |
5634 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
5635 | eventfd_signal(t->entries[i].eventfd, 1); | |
5636 | ||
5637 | /* Update current_threshold */ | |
5407a562 | 5638 | t->current_threshold = i - 1; |
2e72b634 KS |
5639 | unlock: |
5640 | rcu_read_unlock(); | |
5641 | } | |
5642 | ||
5643 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
5644 | { | |
ad4ca5f4 KS |
5645 | while (memcg) { |
5646 | __mem_cgroup_threshold(memcg, false); | |
5647 | if (do_swap_account) | |
5648 | __mem_cgroup_threshold(memcg, true); | |
5649 | ||
5650 | memcg = parent_mem_cgroup(memcg); | |
5651 | } | |
2e72b634 KS |
5652 | } |
5653 | ||
5654 | static int compare_thresholds(const void *a, const void *b) | |
5655 | { | |
5656 | const struct mem_cgroup_threshold *_a = a; | |
5657 | const struct mem_cgroup_threshold *_b = b; | |
5658 | ||
2bff24a3 GT |
5659 | if (_a->threshold > _b->threshold) |
5660 | return 1; | |
5661 | ||
5662 | if (_a->threshold < _b->threshold) | |
5663 | return -1; | |
5664 | ||
5665 | return 0; | |
2e72b634 KS |
5666 | } |
5667 | ||
c0ff4b85 | 5668 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
5669 | { |
5670 | struct mem_cgroup_eventfd_list *ev; | |
5671 | ||
c0ff4b85 | 5672 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 KH |
5673 | eventfd_signal(ev->eventfd, 1); |
5674 | return 0; | |
5675 | } | |
5676 | ||
c0ff4b85 | 5677 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 5678 | { |
7d74b06f KH |
5679 | struct mem_cgroup *iter; |
5680 | ||
c0ff4b85 | 5681 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 5682 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
5683 | } |
5684 | ||
347c4a87 TH |
5685 | static int __mem_cgroup_usage_register_event(struct cgroup_subsys_state *css, |
5686 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) | |
2e72b634 | 5687 | { |
81eeaf04 | 5688 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
2c488db2 KS |
5689 | struct mem_cgroup_thresholds *thresholds; |
5690 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 | 5691 | u64 threshold, usage; |
2c488db2 | 5692 | int i, size, ret; |
2e72b634 KS |
5693 | |
5694 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
5695 | if (ret) | |
5696 | return ret; | |
5697 | ||
5698 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 5699 | |
2e72b634 | 5700 | if (type == _MEM) |
2c488db2 | 5701 | thresholds = &memcg->thresholds; |
2e72b634 | 5702 | else if (type == _MEMSWAP) |
2c488db2 | 5703 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
5704 | else |
5705 | BUG(); | |
5706 | ||
5707 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
5708 | ||
5709 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 5710 | if (thresholds->primary) |
2e72b634 KS |
5711 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
5712 | ||
2c488db2 | 5713 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
5714 | |
5715 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 5716 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 5717 | GFP_KERNEL); |
2c488db2 | 5718 | if (!new) { |
2e72b634 KS |
5719 | ret = -ENOMEM; |
5720 | goto unlock; | |
5721 | } | |
2c488db2 | 5722 | new->size = size; |
2e72b634 KS |
5723 | |
5724 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
5725 | if (thresholds->primary) { |
5726 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 5727 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
5728 | } |
5729 | ||
2e72b634 | 5730 | /* Add new threshold */ |
2c488db2 KS |
5731 | new->entries[size - 1].eventfd = eventfd; |
5732 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
5733 | |
5734 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 5735 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
5736 | compare_thresholds, NULL); |
5737 | ||
5738 | /* Find current threshold */ | |
2c488db2 | 5739 | new->current_threshold = -1; |
2e72b634 | 5740 | for (i = 0; i < size; i++) { |
748dad36 | 5741 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 5742 | /* |
2c488db2 KS |
5743 | * new->current_threshold will not be used until |
5744 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
5745 | * it here. |
5746 | */ | |
2c488db2 | 5747 | ++new->current_threshold; |
748dad36 SZ |
5748 | } else |
5749 | break; | |
2e72b634 KS |
5750 | } |
5751 | ||
2c488db2 KS |
5752 | /* Free old spare buffer and save old primary buffer as spare */ |
5753 | kfree(thresholds->spare); | |
5754 | thresholds->spare = thresholds->primary; | |
5755 | ||
5756 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 5757 | |
907860ed | 5758 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
5759 | synchronize_rcu(); |
5760 | ||
2e72b634 KS |
5761 | unlock: |
5762 | mutex_unlock(&memcg->thresholds_lock); | |
5763 | ||
5764 | return ret; | |
5765 | } | |
5766 | ||
347c4a87 TH |
5767 | static int mem_cgroup_usage_register_event(struct cgroup_subsys_state *css, |
5768 | struct eventfd_ctx *eventfd, const char *args) | |
5769 | { | |
5770 | return __mem_cgroup_usage_register_event(css, eventfd, args, _MEM); | |
5771 | } | |
5772 | ||
5773 | static int memsw_cgroup_usage_register_event(struct cgroup_subsys_state *css, | |
5774 | struct eventfd_ctx *eventfd, const char *args) | |
5775 | { | |
5776 | return __mem_cgroup_usage_register_event(css, eventfd, args, _MEMSWAP); | |
5777 | } | |
5778 | ||
5779 | static void __mem_cgroup_usage_unregister_event(struct cgroup_subsys_state *css, | |
5780 | struct eventfd_ctx *eventfd, enum res_type type) | |
2e72b634 | 5781 | { |
81eeaf04 | 5782 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
2c488db2 KS |
5783 | struct mem_cgroup_thresholds *thresholds; |
5784 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 | 5785 | u64 usage; |
2c488db2 | 5786 | int i, j, size; |
2e72b634 KS |
5787 | |
5788 | mutex_lock(&memcg->thresholds_lock); | |
5789 | if (type == _MEM) | |
2c488db2 | 5790 | thresholds = &memcg->thresholds; |
2e72b634 | 5791 | else if (type == _MEMSWAP) |
2c488db2 | 5792 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
5793 | else |
5794 | BUG(); | |
5795 | ||
371528ca AV |
5796 | if (!thresholds->primary) |
5797 | goto unlock; | |
5798 | ||
2e72b634 KS |
5799 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); |
5800 | ||
5801 | /* Check if a threshold crossed before removing */ | |
5802 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
5803 | ||
5804 | /* Calculate new number of threshold */ | |
2c488db2 KS |
5805 | size = 0; |
5806 | for (i = 0; i < thresholds->primary->size; i++) { | |
5807 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
5808 | size++; |
5809 | } | |
5810 | ||
2c488db2 | 5811 | new = thresholds->spare; |
907860ed | 5812 | |
2e72b634 KS |
5813 | /* Set thresholds array to NULL if we don't have thresholds */ |
5814 | if (!size) { | |
2c488db2 KS |
5815 | kfree(new); |
5816 | new = NULL; | |
907860ed | 5817 | goto swap_buffers; |
2e72b634 KS |
5818 | } |
5819 | ||
2c488db2 | 5820 | new->size = size; |
2e72b634 KS |
5821 | |
5822 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
5823 | new->current_threshold = -1; |
5824 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
5825 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
5826 | continue; |
5827 | ||
2c488db2 | 5828 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 5829 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 5830 | /* |
2c488db2 | 5831 | * new->current_threshold will not be used |
2e72b634 KS |
5832 | * until rcu_assign_pointer(), so it's safe to increment |
5833 | * it here. | |
5834 | */ | |
2c488db2 | 5835 | ++new->current_threshold; |
2e72b634 KS |
5836 | } |
5837 | j++; | |
5838 | } | |
5839 | ||
907860ed | 5840 | swap_buffers: |
2c488db2 KS |
5841 | /* Swap primary and spare array */ |
5842 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
5843 | /* If all events are unregistered, free the spare array */ |
5844 | if (!new) { | |
5845 | kfree(thresholds->spare); | |
5846 | thresholds->spare = NULL; | |
5847 | } | |
5848 | ||
2c488db2 | 5849 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 5850 | |
907860ed | 5851 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 5852 | synchronize_rcu(); |
371528ca | 5853 | unlock: |
2e72b634 | 5854 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 5855 | } |
c1e862c1 | 5856 | |
347c4a87 TH |
5857 | static void mem_cgroup_usage_unregister_event(struct cgroup_subsys_state *css, |
5858 | struct eventfd_ctx *eventfd) | |
5859 | { | |
5860 | return __mem_cgroup_usage_unregister_event(css, eventfd, _MEM); | |
5861 | } | |
5862 | ||
5863 | static void memsw_cgroup_usage_unregister_event(struct cgroup_subsys_state *css, | |
5864 | struct eventfd_ctx *eventfd) | |
5865 | { | |
5866 | return __mem_cgroup_usage_unregister_event(css, eventfd, _MEMSWAP); | |
5867 | } | |
5868 | ||
81eeaf04 | 5869 | static int mem_cgroup_oom_register_event(struct cgroup_subsys_state *css, |
347c4a87 | 5870 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 5871 | { |
81eeaf04 | 5872 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
9490ff27 | 5873 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 5874 | |
9490ff27 KH |
5875 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
5876 | if (!event) | |
5877 | return -ENOMEM; | |
5878 | ||
1af8efe9 | 5879 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
5880 | |
5881 | event->eventfd = eventfd; | |
5882 | list_add(&event->list, &memcg->oom_notify); | |
5883 | ||
5884 | /* already in OOM ? */ | |
79dfdacc | 5885 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 5886 | eventfd_signal(eventfd, 1); |
1af8efe9 | 5887 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
5888 | |
5889 | return 0; | |
5890 | } | |
5891 | ||
81eeaf04 | 5892 | static void mem_cgroup_oom_unregister_event(struct cgroup_subsys_state *css, |
347c4a87 | 5893 | struct eventfd_ctx *eventfd) |
9490ff27 | 5894 | { |
81eeaf04 | 5895 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
9490ff27 | 5896 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 5897 | |
1af8efe9 | 5898 | spin_lock(&memcg_oom_lock); |
9490ff27 | 5899 | |
c0ff4b85 | 5900 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
5901 | if (ev->eventfd == eventfd) { |
5902 | list_del(&ev->list); | |
5903 | kfree(ev); | |
5904 | } | |
5905 | } | |
5906 | ||
1af8efe9 | 5907 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
5908 | } |
5909 | ||
182446d0 | 5910 | static int mem_cgroup_oom_control_read(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
5911 | struct cftype *cft, struct cgroup_map_cb *cb) |
5912 | { | |
182446d0 | 5913 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 | 5914 | |
c0ff4b85 | 5915 | cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable); |
3c11ecf4 | 5916 | |
c0ff4b85 | 5917 | if (atomic_read(&memcg->under_oom)) |
3c11ecf4 KH |
5918 | cb->fill(cb, "under_oom", 1); |
5919 | else | |
5920 | cb->fill(cb, "under_oom", 0); | |
5921 | return 0; | |
5922 | } | |
5923 | ||
182446d0 | 5924 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
5925 | struct cftype *cft, u64 val) |
5926 | { | |
182446d0 | 5927 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
63876986 | 5928 | struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css)); |
3c11ecf4 KH |
5929 | |
5930 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
63876986 | 5931 | if (!parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
5932 | return -EINVAL; |
5933 | ||
0999821b | 5934 | mutex_lock(&memcg_create_mutex); |
3c11ecf4 | 5935 | /* oom-kill-disable is a flag for subhierarchy. */ |
b5f99b53 | 5936 | if ((parent->use_hierarchy) || memcg_has_children(memcg)) { |
0999821b | 5937 | mutex_unlock(&memcg_create_mutex); |
3c11ecf4 KH |
5938 | return -EINVAL; |
5939 | } | |
c0ff4b85 | 5940 | memcg->oom_kill_disable = val; |
4d845ebf | 5941 | if (!val) |
c0ff4b85 | 5942 | memcg_oom_recover(memcg); |
0999821b | 5943 | mutex_unlock(&memcg_create_mutex); |
3c11ecf4 KH |
5944 | return 0; |
5945 | } | |
5946 | ||
c255a458 | 5947 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 5948 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 5949 | { |
55007d84 GC |
5950 | int ret; |
5951 | ||
2633d7a0 | 5952 | memcg->kmemcg_id = -1; |
55007d84 GC |
5953 | ret = memcg_propagate_kmem(memcg); |
5954 | if (ret) | |
5955 | return ret; | |
2633d7a0 | 5956 | |
1d62e436 | 5957 | return mem_cgroup_sockets_init(memcg, ss); |
573b400d | 5958 | } |
e5671dfa | 5959 | |
10d5ebf4 | 5960 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 5961 | { |
1d62e436 | 5962 | mem_cgroup_sockets_destroy(memcg); |
10d5ebf4 LZ |
5963 | } |
5964 | ||
5965 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) | |
5966 | { | |
5967 | if (!memcg_kmem_is_active(memcg)) | |
5968 | return; | |
5969 | ||
5970 | /* | |
5971 | * kmem charges can outlive the cgroup. In the case of slab | |
5972 | * pages, for instance, a page contain objects from various | |
5973 | * processes. As we prevent from taking a reference for every | |
5974 | * such allocation we have to be careful when doing uncharge | |
5975 | * (see memcg_uncharge_kmem) and here during offlining. | |
5976 | * | |
5977 | * The idea is that that only the _last_ uncharge which sees | |
5978 | * the dead memcg will drop the last reference. An additional | |
5979 | * reference is taken here before the group is marked dead | |
5980 | * which is then paired with css_put during uncharge resp. here. | |
5981 | * | |
5982 | * Although this might sound strange as this path is called from | |
5983 | * css_offline() when the referencemight have dropped down to 0 | |
5984 | * and shouldn't be incremented anymore (css_tryget would fail) | |
5985 | * we do not have other options because of the kmem allocations | |
5986 | * lifetime. | |
5987 | */ | |
5988 | css_get(&memcg->css); | |
7de37682 GC |
5989 | |
5990 | memcg_kmem_mark_dead(memcg); | |
5991 | ||
5992 | if (res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0) | |
5993 | return; | |
5994 | ||
7de37682 | 5995 | if (memcg_kmem_test_and_clear_dead(memcg)) |
10d5ebf4 | 5996 | css_put(&memcg->css); |
d1a4c0b3 | 5997 | } |
e5671dfa | 5998 | #else |
cbe128e3 | 5999 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
6000 | { |
6001 | return 0; | |
6002 | } | |
d1a4c0b3 | 6003 | |
10d5ebf4 LZ |
6004 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
6005 | { | |
6006 | } | |
6007 | ||
6008 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) | |
d1a4c0b3 GC |
6009 | { |
6010 | } | |
e5671dfa GC |
6011 | #endif |
6012 | ||
79bd9814 TH |
6013 | /* |
6014 | * Unregister event and free resources. | |
6015 | * | |
6016 | * Gets called from workqueue. | |
6017 | */ | |
6018 | static void cgroup_event_remove(struct work_struct *work) | |
6019 | { | |
6020 | struct cgroup_event *event = container_of(work, struct cgroup_event, | |
6021 | remove); | |
6022 | struct cgroup_subsys_state *css = event->css; | |
6023 | ||
6024 | remove_wait_queue(event->wqh, &event->wait); | |
6025 | ||
347c4a87 | 6026 | event->unregister_event(css, event->eventfd); |
79bd9814 TH |
6027 | |
6028 | /* Notify userspace the event is going away. */ | |
6029 | eventfd_signal(event->eventfd, 1); | |
6030 | ||
6031 | eventfd_ctx_put(event->eventfd); | |
6032 | kfree(event); | |
6033 | css_put(css); | |
6034 | } | |
6035 | ||
6036 | /* | |
6037 | * Gets called on POLLHUP on eventfd when user closes it. | |
6038 | * | |
6039 | * Called with wqh->lock held and interrupts disabled. | |
6040 | */ | |
6041 | static int cgroup_event_wake(wait_queue_t *wait, unsigned mode, | |
6042 | int sync, void *key) | |
6043 | { | |
6044 | struct cgroup_event *event = container_of(wait, | |
6045 | struct cgroup_event, wait); | |
fba94807 | 6046 | struct mem_cgroup *memcg = mem_cgroup_from_css(event->css); |
79bd9814 TH |
6047 | unsigned long flags = (unsigned long)key; |
6048 | ||
6049 | if (flags & POLLHUP) { | |
6050 | /* | |
6051 | * If the event has been detached at cgroup removal, we | |
6052 | * can simply return knowing the other side will cleanup | |
6053 | * for us. | |
6054 | * | |
6055 | * We can't race against event freeing since the other | |
6056 | * side will require wqh->lock via remove_wait_queue(), | |
6057 | * which we hold. | |
6058 | */ | |
fba94807 | 6059 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
6060 | if (!list_empty(&event->list)) { |
6061 | list_del_init(&event->list); | |
6062 | /* | |
6063 | * We are in atomic context, but cgroup_event_remove() | |
6064 | * may sleep, so we have to call it in workqueue. | |
6065 | */ | |
6066 | schedule_work(&event->remove); | |
6067 | } | |
fba94807 | 6068 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
6069 | } |
6070 | ||
6071 | return 0; | |
6072 | } | |
6073 | ||
6074 | static void cgroup_event_ptable_queue_proc(struct file *file, | |
6075 | wait_queue_head_t *wqh, poll_table *pt) | |
6076 | { | |
6077 | struct cgroup_event *event = container_of(pt, | |
6078 | struct cgroup_event, pt); | |
6079 | ||
6080 | event->wqh = wqh; | |
6081 | add_wait_queue(wqh, &event->wait); | |
6082 | } | |
6083 | ||
6084 | /* | |
6085 | * Parse input and register new cgroup event handler. | |
6086 | * | |
6087 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
6088 | * Interpretation of args is defined by control file implementation. | |
6089 | */ | |
b5557c4c | 6090 | static int cgroup_write_event_control(struct cgroup_subsys_state *css, |
79bd9814 TH |
6091 | struct cftype *cft, const char *buffer) |
6092 | { | |
fba94807 | 6093 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
79bd9814 TH |
6094 | struct cgroup_event *event; |
6095 | struct cgroup_subsys_state *cfile_css; | |
6096 | unsigned int efd, cfd; | |
6097 | struct fd efile; | |
6098 | struct fd cfile; | |
fba94807 | 6099 | const char *name; |
79bd9814 TH |
6100 | char *endp; |
6101 | int ret; | |
6102 | ||
6103 | efd = simple_strtoul(buffer, &endp, 10); | |
6104 | if (*endp != ' ') | |
6105 | return -EINVAL; | |
6106 | buffer = endp + 1; | |
6107 | ||
6108 | cfd = simple_strtoul(buffer, &endp, 10); | |
6109 | if ((*endp != ' ') && (*endp != '\0')) | |
6110 | return -EINVAL; | |
6111 | buffer = endp + 1; | |
6112 | ||
6113 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
6114 | if (!event) | |
6115 | return -ENOMEM; | |
6116 | ||
b5557c4c | 6117 | event->css = css; |
79bd9814 TH |
6118 | INIT_LIST_HEAD(&event->list); |
6119 | init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc); | |
6120 | init_waitqueue_func_entry(&event->wait, cgroup_event_wake); | |
6121 | INIT_WORK(&event->remove, cgroup_event_remove); | |
6122 | ||
6123 | efile = fdget(efd); | |
6124 | if (!efile.file) { | |
6125 | ret = -EBADF; | |
6126 | goto out_kfree; | |
6127 | } | |
6128 | ||
6129 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
6130 | if (IS_ERR(event->eventfd)) { | |
6131 | ret = PTR_ERR(event->eventfd); | |
6132 | goto out_put_efile; | |
6133 | } | |
6134 | ||
6135 | cfile = fdget(cfd); | |
6136 | if (!cfile.file) { | |
6137 | ret = -EBADF; | |
6138 | goto out_put_eventfd; | |
6139 | } | |
6140 | ||
6141 | /* the process need read permission on control file */ | |
6142 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
6143 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
6144 | if (ret < 0) | |
6145 | goto out_put_cfile; | |
6146 | ||
fba94807 TH |
6147 | /* |
6148 | * Determine the event callbacks and set them in @event. This used | |
6149 | * to be done via struct cftype but cgroup core no longer knows | |
6150 | * about these events. The following is crude but the whole thing | |
6151 | * is for compatibility anyway. | |
6152 | */ | |
6153 | name = cfile.file->f_dentry->d_name.name; | |
6154 | ||
6155 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
6156 | event->register_event = mem_cgroup_usage_register_event; | |
6157 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
6158 | } else if (!strcmp(name, "memory.oom_control")) { | |
6159 | event->register_event = mem_cgroup_oom_register_event; | |
6160 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
6161 | } else if (!strcmp(name, "memory.pressure_level")) { | |
6162 | event->register_event = vmpressure_register_event; | |
6163 | event->unregister_event = vmpressure_unregister_event; | |
6164 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
6165 | event->register_event = memsw_cgroup_usage_register_event; |
6166 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
6167 | } else { |
6168 | ret = -EINVAL; | |
6169 | goto out_put_cfile; | |
6170 | } | |
6171 | ||
79bd9814 | 6172 | /* |
b5557c4c TH |
6173 | * Verify @cfile should belong to @css. Also, remaining events are |
6174 | * automatically removed on cgroup destruction but the removal is | |
6175 | * asynchronous, so take an extra ref on @css. | |
79bd9814 TH |
6176 | */ |
6177 | rcu_read_lock(); | |
6178 | ||
6179 | ret = -EINVAL; | |
b5557c4c TH |
6180 | cfile_css = css_from_dir(cfile.file->f_dentry->d_parent, |
6181 | &mem_cgroup_subsys); | |
6182 | if (cfile_css == css && css_tryget(css)) | |
79bd9814 TH |
6183 | ret = 0; |
6184 | ||
6185 | rcu_read_unlock(); | |
6186 | if (ret) | |
6187 | goto out_put_cfile; | |
6188 | ||
347c4a87 | 6189 | ret = event->register_event(css, event->eventfd, buffer); |
79bd9814 TH |
6190 | if (ret) |
6191 | goto out_put_css; | |
6192 | ||
6193 | efile.file->f_op->poll(efile.file, &event->pt); | |
6194 | ||
fba94807 TH |
6195 | spin_lock(&memcg->event_list_lock); |
6196 | list_add(&event->list, &memcg->event_list); | |
6197 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
6198 | |
6199 | fdput(cfile); | |
6200 | fdput(efile); | |
6201 | ||
6202 | return 0; | |
6203 | ||
6204 | out_put_css: | |
b5557c4c | 6205 | css_put(css); |
79bd9814 TH |
6206 | out_put_cfile: |
6207 | fdput(cfile); | |
6208 | out_put_eventfd: | |
6209 | eventfd_ctx_put(event->eventfd); | |
6210 | out_put_efile: | |
6211 | fdput(efile); | |
6212 | out_kfree: | |
6213 | kfree(event); | |
6214 | ||
6215 | return ret; | |
6216 | } | |
6217 | ||
8cdea7c0 BS |
6218 | static struct cftype mem_cgroup_files[] = { |
6219 | { | |
0eea1030 | 6220 | .name = "usage_in_bytes", |
8c7c6e34 | 6221 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
af36f906 | 6222 | .read = mem_cgroup_read, |
8cdea7c0 | 6223 | }, |
c84872e1 PE |
6224 | { |
6225 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 6226 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 6227 | .trigger = mem_cgroup_reset, |
af36f906 | 6228 | .read = mem_cgroup_read, |
c84872e1 | 6229 | }, |
8cdea7c0 | 6230 | { |
0eea1030 | 6231 | .name = "limit_in_bytes", |
8c7c6e34 | 6232 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 6233 | .write_string = mem_cgroup_write, |
af36f906 | 6234 | .read = mem_cgroup_read, |
8cdea7c0 | 6235 | }, |
296c81d8 BS |
6236 | { |
6237 | .name = "soft_limit_in_bytes", | |
6238 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
6239 | .write_string = mem_cgroup_write, | |
af36f906 | 6240 | .read = mem_cgroup_read, |
296c81d8 | 6241 | }, |
8cdea7c0 BS |
6242 | { |
6243 | .name = "failcnt", | |
8c7c6e34 | 6244 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 6245 | .trigger = mem_cgroup_reset, |
af36f906 | 6246 | .read = mem_cgroup_read, |
8cdea7c0 | 6247 | }, |
d2ceb9b7 KH |
6248 | { |
6249 | .name = "stat", | |
ab215884 | 6250 | .read_seq_string = memcg_stat_show, |
d2ceb9b7 | 6251 | }, |
c1e862c1 KH |
6252 | { |
6253 | .name = "force_empty", | |
6254 | .trigger = mem_cgroup_force_empty_write, | |
6255 | }, | |
18f59ea7 BS |
6256 | { |
6257 | .name = "use_hierarchy", | |
f00baae7 | 6258 | .flags = CFTYPE_INSANE, |
18f59ea7 BS |
6259 | .write_u64 = mem_cgroup_hierarchy_write, |
6260 | .read_u64 = mem_cgroup_hierarchy_read, | |
6261 | }, | |
79bd9814 TH |
6262 | { |
6263 | .name = "cgroup.event_control", | |
6264 | .write_string = cgroup_write_event_control, | |
6265 | .flags = CFTYPE_NO_PREFIX, | |
6266 | .mode = S_IWUGO, | |
6267 | }, | |
a7885eb8 KM |
6268 | { |
6269 | .name = "swappiness", | |
6270 | .read_u64 = mem_cgroup_swappiness_read, | |
6271 | .write_u64 = mem_cgroup_swappiness_write, | |
6272 | }, | |
7dc74be0 DN |
6273 | { |
6274 | .name = "move_charge_at_immigrate", | |
6275 | .read_u64 = mem_cgroup_move_charge_read, | |
6276 | .write_u64 = mem_cgroup_move_charge_write, | |
6277 | }, | |
9490ff27 KH |
6278 | { |
6279 | .name = "oom_control", | |
3c11ecf4 KH |
6280 | .read_map = mem_cgroup_oom_control_read, |
6281 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
6282 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
6283 | }, | |
70ddf637 AV |
6284 | { |
6285 | .name = "pressure_level", | |
70ddf637 | 6286 | }, |
406eb0c9 YH |
6287 | #ifdef CONFIG_NUMA |
6288 | { | |
6289 | .name = "numa_stat", | |
ab215884 | 6290 | .read_seq_string = memcg_numa_stat_show, |
406eb0c9 YH |
6291 | }, |
6292 | #endif | |
510fc4e1 GC |
6293 | #ifdef CONFIG_MEMCG_KMEM |
6294 | { | |
6295 | .name = "kmem.limit_in_bytes", | |
6296 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
6297 | .write_string = mem_cgroup_write, | |
6298 | .read = mem_cgroup_read, | |
6299 | }, | |
6300 | { | |
6301 | .name = "kmem.usage_in_bytes", | |
6302 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
6303 | .read = mem_cgroup_read, | |
6304 | }, | |
6305 | { | |
6306 | .name = "kmem.failcnt", | |
6307 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6308 | .trigger = mem_cgroup_reset, | |
6309 | .read = mem_cgroup_read, | |
6310 | }, | |
6311 | { | |
6312 | .name = "kmem.max_usage_in_bytes", | |
6313 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6314 | .trigger = mem_cgroup_reset, | |
6315 | .read = mem_cgroup_read, | |
6316 | }, | |
749c5415 GC |
6317 | #ifdef CONFIG_SLABINFO |
6318 | { | |
6319 | .name = "kmem.slabinfo", | |
6320 | .read_seq_string = mem_cgroup_slabinfo_read, | |
6321 | }, | |
6322 | #endif | |
8c7c6e34 | 6323 | #endif |
6bc10349 | 6324 | { }, /* terminate */ |
af36f906 | 6325 | }; |
8c7c6e34 | 6326 | |
2d11085e MH |
6327 | #ifdef CONFIG_MEMCG_SWAP |
6328 | static struct cftype memsw_cgroup_files[] = { | |
6329 | { | |
6330 | .name = "memsw.usage_in_bytes", | |
6331 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
6332 | .read = mem_cgroup_read, | |
2d11085e MH |
6333 | }, |
6334 | { | |
6335 | .name = "memsw.max_usage_in_bytes", | |
6336 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6337 | .trigger = mem_cgroup_reset, | |
6338 | .read = mem_cgroup_read, | |
6339 | }, | |
6340 | { | |
6341 | .name = "memsw.limit_in_bytes", | |
6342 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
6343 | .write_string = mem_cgroup_write, | |
6344 | .read = mem_cgroup_read, | |
6345 | }, | |
6346 | { | |
6347 | .name = "memsw.failcnt", | |
6348 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6349 | .trigger = mem_cgroup_reset, | |
6350 | .read = mem_cgroup_read, | |
6351 | }, | |
6352 | { }, /* terminate */ | |
6353 | }; | |
6354 | #endif | |
c0ff4b85 | 6355 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
6356 | { |
6357 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 6358 | struct mem_cgroup_per_zone *mz; |
41e3355d | 6359 | int zone, tmp = node; |
1ecaab2b KH |
6360 | /* |
6361 | * This routine is called against possible nodes. | |
6362 | * But it's BUG to call kmalloc() against offline node. | |
6363 | * | |
6364 | * TODO: this routine can waste much memory for nodes which will | |
6365 | * never be onlined. It's better to use memory hotplug callback | |
6366 | * function. | |
6367 | */ | |
41e3355d KH |
6368 | if (!node_state(node, N_NORMAL_MEMORY)) |
6369 | tmp = -1; | |
17295c88 | 6370 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
6371 | if (!pn) |
6372 | return 1; | |
1ecaab2b | 6373 | |
1ecaab2b KH |
6374 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
6375 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 6376 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
6377 | mz->usage_in_excess = 0; |
6378 | mz->on_tree = false; | |
d79154bb | 6379 | mz->memcg = memcg; |
1ecaab2b | 6380 | } |
54f72fe0 | 6381 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
6382 | return 0; |
6383 | } | |
6384 | ||
c0ff4b85 | 6385 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 6386 | { |
54f72fe0 | 6387 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
6388 | } |
6389 | ||
33327948 KH |
6390 | static struct mem_cgroup *mem_cgroup_alloc(void) |
6391 | { | |
d79154bb | 6392 | struct mem_cgroup *memcg; |
45cf7ebd | 6393 | size_t size = memcg_size(); |
33327948 | 6394 | |
45cf7ebd | 6395 | /* Can be very big if nr_node_ids is very big */ |
c8dad2bb | 6396 | if (size < PAGE_SIZE) |
d79154bb | 6397 | memcg = kzalloc(size, GFP_KERNEL); |
33327948 | 6398 | else |
d79154bb | 6399 | memcg = vzalloc(size); |
33327948 | 6400 | |
d79154bb | 6401 | if (!memcg) |
e7bbcdf3 DC |
6402 | return NULL; |
6403 | ||
d79154bb HD |
6404 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
6405 | if (!memcg->stat) | |
d2e61b8d | 6406 | goto out_free; |
d79154bb HD |
6407 | spin_lock_init(&memcg->pcp_counter_lock); |
6408 | return memcg; | |
d2e61b8d DC |
6409 | |
6410 | out_free: | |
6411 | if (size < PAGE_SIZE) | |
d79154bb | 6412 | kfree(memcg); |
d2e61b8d | 6413 | else |
d79154bb | 6414 | vfree(memcg); |
d2e61b8d | 6415 | return NULL; |
33327948 KH |
6416 | } |
6417 | ||
59927fb9 | 6418 | /* |
c8b2a36f GC |
6419 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
6420 | * (scanning all at force_empty is too costly...) | |
6421 | * | |
6422 | * Instead of clearing all references at force_empty, we remember | |
6423 | * the number of reference from swap_cgroup and free mem_cgroup when | |
6424 | * it goes down to 0. | |
6425 | * | |
6426 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 6427 | */ |
c8b2a36f GC |
6428 | |
6429 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 6430 | { |
c8b2a36f | 6431 | int node; |
45cf7ebd | 6432 | size_t size = memcg_size(); |
59927fb9 | 6433 | |
bb4cc1a8 | 6434 | mem_cgroup_remove_from_trees(memcg); |
c8b2a36f GC |
6435 | free_css_id(&mem_cgroup_subsys, &memcg->css); |
6436 | ||
6437 | for_each_node(node) | |
6438 | free_mem_cgroup_per_zone_info(memcg, node); | |
6439 | ||
6440 | free_percpu(memcg->stat); | |
6441 | ||
3f134619 GC |
6442 | /* |
6443 | * We need to make sure that (at least for now), the jump label | |
6444 | * destruction code runs outside of the cgroup lock. This is because | |
6445 | * get_online_cpus(), which is called from the static_branch update, | |
6446 | * can't be called inside the cgroup_lock. cpusets are the ones | |
6447 | * enforcing this dependency, so if they ever change, we might as well. | |
6448 | * | |
6449 | * schedule_work() will guarantee this happens. Be careful if you need | |
6450 | * to move this code around, and make sure it is outside | |
6451 | * the cgroup_lock. | |
6452 | */ | |
a8964b9b | 6453 | disarm_static_keys(memcg); |
3afe36b1 GC |
6454 | if (size < PAGE_SIZE) |
6455 | kfree(memcg); | |
6456 | else | |
6457 | vfree(memcg); | |
59927fb9 | 6458 | } |
3afe36b1 | 6459 | |
7bcc1bb1 DN |
6460 | /* |
6461 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
6462 | */ | |
e1aab161 | 6463 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 6464 | { |
c0ff4b85 | 6465 | if (!memcg->res.parent) |
7bcc1bb1 | 6466 | return NULL; |
c0ff4b85 | 6467 | return mem_cgroup_from_res_counter(memcg->res.parent, res); |
7bcc1bb1 | 6468 | } |
e1aab161 | 6469 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 6470 | |
bb4cc1a8 AM |
6471 | static void __init mem_cgroup_soft_limit_tree_init(void) |
6472 | { | |
6473 | struct mem_cgroup_tree_per_node *rtpn; | |
6474 | struct mem_cgroup_tree_per_zone *rtpz; | |
6475 | int tmp, node, zone; | |
6476 | ||
6477 | for_each_node(node) { | |
6478 | tmp = node; | |
6479 | if (!node_state(node, N_NORMAL_MEMORY)) | |
6480 | tmp = -1; | |
6481 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
6482 | BUG_ON(!rtpn); | |
6483 | ||
6484 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
6485 | ||
6486 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
6487 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
6488 | rtpz->rb_root = RB_ROOT; | |
6489 | spin_lock_init(&rtpz->lock); | |
6490 | } | |
6491 | } | |
6492 | } | |
6493 | ||
0eb253e2 | 6494 | static struct cgroup_subsys_state * __ref |
eb95419b | 6495 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 6496 | { |
d142e3e6 | 6497 | struct mem_cgroup *memcg; |
04046e1a | 6498 | long error = -ENOMEM; |
6d12e2d8 | 6499 | int node; |
8cdea7c0 | 6500 | |
c0ff4b85 R |
6501 | memcg = mem_cgroup_alloc(); |
6502 | if (!memcg) | |
04046e1a | 6503 | return ERR_PTR(error); |
78fb7466 | 6504 | |
3ed28fa1 | 6505 | for_each_node(node) |
c0ff4b85 | 6506 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 6507 | goto free_out; |
f64c3f54 | 6508 | |
c077719b | 6509 | /* root ? */ |
eb95419b | 6510 | if (parent_css == NULL) { |
a41c58a6 | 6511 | root_mem_cgroup = memcg; |
d142e3e6 GC |
6512 | res_counter_init(&memcg->res, NULL); |
6513 | res_counter_init(&memcg->memsw, NULL); | |
6514 | res_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 6515 | } |
28dbc4b6 | 6516 | |
d142e3e6 GC |
6517 | memcg->last_scanned_node = MAX_NUMNODES; |
6518 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
6519 | memcg->move_charge_at_immigrate = 0; |
6520 | mutex_init(&memcg->thresholds_lock); | |
6521 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 6522 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
6523 | INIT_LIST_HEAD(&memcg->event_list); |
6524 | spin_lock_init(&memcg->event_list_lock); | |
d142e3e6 GC |
6525 | |
6526 | return &memcg->css; | |
6527 | ||
6528 | free_out: | |
6529 | __mem_cgroup_free(memcg); | |
6530 | return ERR_PTR(error); | |
6531 | } | |
6532 | ||
6533 | static int | |
eb95419b | 6534 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 6535 | { |
eb95419b TH |
6536 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
6537 | struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(css)); | |
d142e3e6 GC |
6538 | int error = 0; |
6539 | ||
63876986 | 6540 | if (!parent) |
d142e3e6 GC |
6541 | return 0; |
6542 | ||
0999821b | 6543 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
6544 | |
6545 | memcg->use_hierarchy = parent->use_hierarchy; | |
6546 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
6547 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
6548 | ||
6549 | if (parent->use_hierarchy) { | |
c0ff4b85 R |
6550 | res_counter_init(&memcg->res, &parent->res); |
6551 | res_counter_init(&memcg->memsw, &parent->memsw); | |
510fc4e1 | 6552 | res_counter_init(&memcg->kmem, &parent->kmem); |
55007d84 | 6553 | |
7bcc1bb1 | 6554 | /* |
8d76a979 LZ |
6555 | * No need to take a reference to the parent because cgroup |
6556 | * core guarantees its existence. | |
7bcc1bb1 | 6557 | */ |
18f59ea7 | 6558 | } else { |
c0ff4b85 R |
6559 | res_counter_init(&memcg->res, NULL); |
6560 | res_counter_init(&memcg->memsw, NULL); | |
510fc4e1 | 6561 | res_counter_init(&memcg->kmem, NULL); |
8c7f6edb TH |
6562 | /* |
6563 | * Deeper hierachy with use_hierarchy == false doesn't make | |
6564 | * much sense so let cgroup subsystem know about this | |
6565 | * unfortunate state in our controller. | |
6566 | */ | |
d142e3e6 | 6567 | if (parent != root_mem_cgroup) |
8c7f6edb | 6568 | mem_cgroup_subsys.broken_hierarchy = true; |
18f59ea7 | 6569 | } |
cbe128e3 GC |
6570 | |
6571 | error = memcg_init_kmem(memcg, &mem_cgroup_subsys); | |
0999821b | 6572 | mutex_unlock(&memcg_create_mutex); |
d142e3e6 | 6573 | return error; |
8cdea7c0 BS |
6574 | } |
6575 | ||
5f578161 MH |
6576 | /* |
6577 | * Announce all parents that a group from their hierarchy is gone. | |
6578 | */ | |
6579 | static void mem_cgroup_invalidate_reclaim_iterators(struct mem_cgroup *memcg) | |
6580 | { | |
6581 | struct mem_cgroup *parent = memcg; | |
6582 | ||
6583 | while ((parent = parent_mem_cgroup(parent))) | |
519ebea3 | 6584 | mem_cgroup_iter_invalidate(parent); |
5f578161 MH |
6585 | |
6586 | /* | |
6587 | * if the root memcg is not hierarchical we have to check it | |
6588 | * explicitely. | |
6589 | */ | |
6590 | if (!root_mem_cgroup->use_hierarchy) | |
519ebea3 | 6591 | mem_cgroup_iter_invalidate(root_mem_cgroup); |
5f578161 MH |
6592 | } |
6593 | ||
eb95419b | 6594 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 6595 | { |
eb95419b | 6596 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
79bd9814 TH |
6597 | struct cgroup_event *event, *tmp; |
6598 | ||
6599 | /* | |
6600 | * Unregister events and notify userspace. | |
6601 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
6602 | * directory to avoid race between userspace and kernelspace. | |
6603 | */ | |
fba94807 TH |
6604 | spin_lock(&memcg->event_list_lock); |
6605 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
6606 | list_del_init(&event->list); |
6607 | schedule_work(&event->remove); | |
6608 | } | |
fba94807 | 6609 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 6610 | |
10d5ebf4 LZ |
6611 | kmem_cgroup_css_offline(memcg); |
6612 | ||
5f578161 | 6613 | mem_cgroup_invalidate_reclaim_iterators(memcg); |
ab5196c2 | 6614 | mem_cgroup_reparent_charges(memcg); |
1f458cbf | 6615 | mem_cgroup_destroy_all_caches(memcg); |
33cb876e | 6616 | vmpressure_cleanup(&memcg->vmpressure); |
df878fb0 KH |
6617 | } |
6618 | ||
eb95419b | 6619 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 6620 | { |
eb95419b | 6621 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 6622 | |
10d5ebf4 | 6623 | memcg_destroy_kmem(memcg); |
465939a1 | 6624 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
6625 | } |
6626 | ||
02491447 | 6627 | #ifdef CONFIG_MMU |
7dc74be0 | 6628 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
6629 | #define PRECHARGE_COUNT_AT_ONCE 256 |
6630 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 6631 | { |
854ffa8d DN |
6632 | int ret = 0; |
6633 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
c0ff4b85 | 6634 | struct mem_cgroup *memcg = mc.to; |
4ffef5fe | 6635 | |
c0ff4b85 | 6636 | if (mem_cgroup_is_root(memcg)) { |
854ffa8d DN |
6637 | mc.precharge += count; |
6638 | /* we don't need css_get for root */ | |
6639 | return ret; | |
6640 | } | |
6641 | /* try to charge at once */ | |
6642 | if (count > 1) { | |
6643 | struct res_counter *dummy; | |
6644 | /* | |
c0ff4b85 | 6645 | * "memcg" cannot be under rmdir() because we've already checked |
854ffa8d DN |
6646 | * by cgroup_lock_live_cgroup() that it is not removed and we |
6647 | * are still under the same cgroup_mutex. So we can postpone | |
6648 | * css_get(). | |
6649 | */ | |
c0ff4b85 | 6650 | if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy)) |
854ffa8d | 6651 | goto one_by_one; |
c0ff4b85 | 6652 | if (do_swap_account && res_counter_charge(&memcg->memsw, |
854ffa8d | 6653 | PAGE_SIZE * count, &dummy)) { |
c0ff4b85 | 6654 | res_counter_uncharge(&memcg->res, PAGE_SIZE * count); |
854ffa8d DN |
6655 | goto one_by_one; |
6656 | } | |
6657 | mc.precharge += count; | |
854ffa8d DN |
6658 | return ret; |
6659 | } | |
6660 | one_by_one: | |
6661 | /* fall back to one by one charge */ | |
6662 | while (count--) { | |
6663 | if (signal_pending(current)) { | |
6664 | ret = -EINTR; | |
6665 | break; | |
6666 | } | |
6667 | if (!batch_count--) { | |
6668 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
6669 | cond_resched(); | |
6670 | } | |
c0ff4b85 R |
6671 | ret = __mem_cgroup_try_charge(NULL, |
6672 | GFP_KERNEL, 1, &memcg, false); | |
38c5d72f | 6673 | if (ret) |
854ffa8d | 6674 | /* mem_cgroup_clear_mc() will do uncharge later */ |
38c5d72f | 6675 | return ret; |
854ffa8d DN |
6676 | mc.precharge++; |
6677 | } | |
4ffef5fe DN |
6678 | return ret; |
6679 | } | |
6680 | ||
6681 | /** | |
8d32ff84 | 6682 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
6683 | * @vma: the vma the pte to be checked belongs |
6684 | * @addr: the address corresponding to the pte to be checked | |
6685 | * @ptent: the pte to be checked | |
02491447 | 6686 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
6687 | * |
6688 | * Returns | |
6689 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
6690 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
6691 | * move charge. if @target is not NULL, the page is stored in target->page | |
6692 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
6693 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
6694 | * target for charge migration. if @target is not NULL, the entry is stored | |
6695 | * in target->ent. | |
4ffef5fe DN |
6696 | * |
6697 | * Called with pte lock held. | |
6698 | */ | |
4ffef5fe DN |
6699 | union mc_target { |
6700 | struct page *page; | |
02491447 | 6701 | swp_entry_t ent; |
4ffef5fe DN |
6702 | }; |
6703 | ||
4ffef5fe | 6704 | enum mc_target_type { |
8d32ff84 | 6705 | MC_TARGET_NONE = 0, |
4ffef5fe | 6706 | MC_TARGET_PAGE, |
02491447 | 6707 | MC_TARGET_SWAP, |
4ffef5fe DN |
6708 | }; |
6709 | ||
90254a65 DN |
6710 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
6711 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 6712 | { |
90254a65 | 6713 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 6714 | |
90254a65 DN |
6715 | if (!page || !page_mapped(page)) |
6716 | return NULL; | |
6717 | if (PageAnon(page)) { | |
6718 | /* we don't move shared anon */ | |
4b91355e | 6719 | if (!move_anon()) |
90254a65 | 6720 | return NULL; |
87946a72 DN |
6721 | } else if (!move_file()) |
6722 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
6723 | return NULL; |
6724 | if (!get_page_unless_zero(page)) | |
6725 | return NULL; | |
6726 | ||
6727 | return page; | |
6728 | } | |
6729 | ||
4b91355e | 6730 | #ifdef CONFIG_SWAP |
90254a65 DN |
6731 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
6732 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6733 | { | |
90254a65 DN |
6734 | struct page *page = NULL; |
6735 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
6736 | ||
6737 | if (!move_anon() || non_swap_entry(ent)) | |
6738 | return NULL; | |
4b91355e KH |
6739 | /* |
6740 | * Because lookup_swap_cache() updates some statistics counter, | |
6741 | * we call find_get_page() with swapper_space directly. | |
6742 | */ | |
33806f06 | 6743 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
6744 | if (do_swap_account) |
6745 | entry->val = ent.val; | |
6746 | ||
6747 | return page; | |
6748 | } | |
4b91355e KH |
6749 | #else |
6750 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
6751 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6752 | { | |
6753 | return NULL; | |
6754 | } | |
6755 | #endif | |
90254a65 | 6756 | |
87946a72 DN |
6757 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
6758 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6759 | { | |
6760 | struct page *page = NULL; | |
87946a72 DN |
6761 | struct address_space *mapping; |
6762 | pgoff_t pgoff; | |
6763 | ||
6764 | if (!vma->vm_file) /* anonymous vma */ | |
6765 | return NULL; | |
6766 | if (!move_file()) | |
6767 | return NULL; | |
6768 | ||
87946a72 DN |
6769 | mapping = vma->vm_file->f_mapping; |
6770 | if (pte_none(ptent)) | |
6771 | pgoff = linear_page_index(vma, addr); | |
6772 | else /* pte_file(ptent) is true */ | |
6773 | pgoff = pte_to_pgoff(ptent); | |
6774 | ||
6775 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
6776 | page = find_get_page(mapping, pgoff); |
6777 | ||
6778 | #ifdef CONFIG_SWAP | |
6779 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
6780 | if (radix_tree_exceptional_entry(page)) { | |
6781 | swp_entry_t swap = radix_to_swp_entry(page); | |
87946a72 | 6782 | if (do_swap_account) |
aa3b1895 | 6783 | *entry = swap; |
33806f06 | 6784 | page = find_get_page(swap_address_space(swap), swap.val); |
87946a72 | 6785 | } |
aa3b1895 | 6786 | #endif |
87946a72 DN |
6787 | return page; |
6788 | } | |
6789 | ||
8d32ff84 | 6790 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
6791 | unsigned long addr, pte_t ptent, union mc_target *target) |
6792 | { | |
6793 | struct page *page = NULL; | |
6794 | struct page_cgroup *pc; | |
8d32ff84 | 6795 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
6796 | swp_entry_t ent = { .val = 0 }; |
6797 | ||
6798 | if (pte_present(ptent)) | |
6799 | page = mc_handle_present_pte(vma, addr, ptent); | |
6800 | else if (is_swap_pte(ptent)) | |
6801 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
6802 | else if (pte_none(ptent) || pte_file(ptent)) |
6803 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
6804 | |
6805 | if (!page && !ent.val) | |
8d32ff84 | 6806 | return ret; |
02491447 DN |
6807 | if (page) { |
6808 | pc = lookup_page_cgroup(page); | |
6809 | /* | |
6810 | * Do only loose check w/o page_cgroup lock. | |
6811 | * mem_cgroup_move_account() checks the pc is valid or not under | |
6812 | * the lock. | |
6813 | */ | |
6814 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
6815 | ret = MC_TARGET_PAGE; | |
6816 | if (target) | |
6817 | target->page = page; | |
6818 | } | |
6819 | if (!ret || !target) | |
6820 | put_page(page); | |
6821 | } | |
90254a65 DN |
6822 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
6823 | if (ent.val && !ret && | |
9fb4b7cc | 6824 | css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
6825 | ret = MC_TARGET_SWAP; |
6826 | if (target) | |
6827 | target->ent = ent; | |
4ffef5fe | 6828 | } |
4ffef5fe DN |
6829 | return ret; |
6830 | } | |
6831 | ||
12724850 NH |
6832 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
6833 | /* | |
6834 | * We don't consider swapping or file mapped pages because THP does not | |
6835 | * support them for now. | |
6836 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
6837 | */ | |
6838 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6839 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6840 | { | |
6841 | struct page *page = NULL; | |
6842 | struct page_cgroup *pc; | |
6843 | enum mc_target_type ret = MC_TARGET_NONE; | |
6844 | ||
6845 | page = pmd_page(pmd); | |
6846 | VM_BUG_ON(!page || !PageHead(page)); | |
6847 | if (!move_anon()) | |
6848 | return ret; | |
6849 | pc = lookup_page_cgroup(page); | |
6850 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
6851 | ret = MC_TARGET_PAGE; | |
6852 | if (target) { | |
6853 | get_page(page); | |
6854 | target->page = page; | |
6855 | } | |
6856 | } | |
6857 | return ret; | |
6858 | } | |
6859 | #else | |
6860 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6861 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6862 | { | |
6863 | return MC_TARGET_NONE; | |
6864 | } | |
6865 | #endif | |
6866 | ||
4ffef5fe DN |
6867 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
6868 | unsigned long addr, unsigned long end, | |
6869 | struct mm_walk *walk) | |
6870 | { | |
6871 | struct vm_area_struct *vma = walk->private; | |
6872 | pte_t *pte; | |
6873 | spinlock_t *ptl; | |
6874 | ||
12724850 NH |
6875 | if (pmd_trans_huge_lock(pmd, vma) == 1) { |
6876 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) | |
6877 | mc.precharge += HPAGE_PMD_NR; | |
6878 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 6879 | return 0; |
12724850 | 6880 | } |
03319327 | 6881 | |
45f83cef AA |
6882 | if (pmd_trans_unstable(pmd)) |
6883 | return 0; | |
4ffef5fe DN |
6884 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
6885 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 6886 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
6887 | mc.precharge++; /* increment precharge temporarily */ |
6888 | pte_unmap_unlock(pte - 1, ptl); | |
6889 | cond_resched(); | |
6890 | ||
7dc74be0 DN |
6891 | return 0; |
6892 | } | |
6893 | ||
4ffef5fe DN |
6894 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
6895 | { | |
6896 | unsigned long precharge; | |
6897 | struct vm_area_struct *vma; | |
6898 | ||
dfe076b0 | 6899 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
6900 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
6901 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
6902 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
6903 | .mm = mm, | |
6904 | .private = vma, | |
6905 | }; | |
6906 | if (is_vm_hugetlb_page(vma)) | |
6907 | continue; | |
4ffef5fe DN |
6908 | walk_page_range(vma->vm_start, vma->vm_end, |
6909 | &mem_cgroup_count_precharge_walk); | |
6910 | } | |
dfe076b0 | 6911 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
6912 | |
6913 | precharge = mc.precharge; | |
6914 | mc.precharge = 0; | |
6915 | ||
6916 | return precharge; | |
6917 | } | |
6918 | ||
4ffef5fe DN |
6919 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
6920 | { | |
dfe076b0 DN |
6921 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
6922 | ||
6923 | VM_BUG_ON(mc.moving_task); | |
6924 | mc.moving_task = current; | |
6925 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
6926 | } |
6927 | ||
dfe076b0 DN |
6928 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
6929 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 6930 | { |
2bd9bb20 KH |
6931 | struct mem_cgroup *from = mc.from; |
6932 | struct mem_cgroup *to = mc.to; | |
4050377b | 6933 | int i; |
2bd9bb20 | 6934 | |
4ffef5fe | 6935 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
6936 | if (mc.precharge) { |
6937 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
6938 | mc.precharge = 0; | |
6939 | } | |
6940 | /* | |
6941 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
6942 | * we must uncharge here. | |
6943 | */ | |
6944 | if (mc.moved_charge) { | |
6945 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
6946 | mc.moved_charge = 0; | |
4ffef5fe | 6947 | } |
483c30b5 DN |
6948 | /* we must fixup refcnts and charges */ |
6949 | if (mc.moved_swap) { | |
483c30b5 DN |
6950 | /* uncharge swap account from the old cgroup */ |
6951 | if (!mem_cgroup_is_root(mc.from)) | |
6952 | res_counter_uncharge(&mc.from->memsw, | |
6953 | PAGE_SIZE * mc.moved_swap); | |
4050377b LZ |
6954 | |
6955 | for (i = 0; i < mc.moved_swap; i++) | |
6956 | css_put(&mc.from->css); | |
483c30b5 DN |
6957 | |
6958 | if (!mem_cgroup_is_root(mc.to)) { | |
6959 | /* | |
6960 | * we charged both to->res and to->memsw, so we should | |
6961 | * uncharge to->res. | |
6962 | */ | |
6963 | res_counter_uncharge(&mc.to->res, | |
6964 | PAGE_SIZE * mc.moved_swap); | |
483c30b5 | 6965 | } |
4050377b | 6966 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
6967 | mc.moved_swap = 0; |
6968 | } | |
dfe076b0 DN |
6969 | memcg_oom_recover(from); |
6970 | memcg_oom_recover(to); | |
6971 | wake_up_all(&mc.waitq); | |
6972 | } | |
6973 | ||
6974 | static void mem_cgroup_clear_mc(void) | |
6975 | { | |
6976 | struct mem_cgroup *from = mc.from; | |
6977 | ||
6978 | /* | |
6979 | * we must clear moving_task before waking up waiters at the end of | |
6980 | * task migration. | |
6981 | */ | |
6982 | mc.moving_task = NULL; | |
6983 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 6984 | spin_lock(&mc.lock); |
4ffef5fe DN |
6985 | mc.from = NULL; |
6986 | mc.to = NULL; | |
2bd9bb20 | 6987 | spin_unlock(&mc.lock); |
32047e2a | 6988 | mem_cgroup_end_move(from); |
4ffef5fe DN |
6989 | } |
6990 | ||
eb95419b | 6991 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6992 | struct cgroup_taskset *tset) |
7dc74be0 | 6993 | { |
2f7ee569 | 6994 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 6995 | int ret = 0; |
eb95419b | 6996 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
ee5e8472 | 6997 | unsigned long move_charge_at_immigrate; |
7dc74be0 | 6998 | |
ee5e8472 GC |
6999 | /* |
7000 | * We are now commited to this value whatever it is. Changes in this | |
7001 | * tunable will only affect upcoming migrations, not the current one. | |
7002 | * So we need to save it, and keep it going. | |
7003 | */ | |
7004 | move_charge_at_immigrate = memcg->move_charge_at_immigrate; | |
7005 | if (move_charge_at_immigrate) { | |
7dc74be0 DN |
7006 | struct mm_struct *mm; |
7007 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
7008 | ||
c0ff4b85 | 7009 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
7010 | |
7011 | mm = get_task_mm(p); | |
7012 | if (!mm) | |
7013 | return 0; | |
7dc74be0 | 7014 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
7015 | if (mm->owner == p) { |
7016 | VM_BUG_ON(mc.from); | |
7017 | VM_BUG_ON(mc.to); | |
7018 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 7019 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 7020 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 7021 | mem_cgroup_start_move(from); |
2bd9bb20 | 7022 | spin_lock(&mc.lock); |
4ffef5fe | 7023 | mc.from = from; |
c0ff4b85 | 7024 | mc.to = memcg; |
ee5e8472 | 7025 | mc.immigrate_flags = move_charge_at_immigrate; |
2bd9bb20 | 7026 | spin_unlock(&mc.lock); |
dfe076b0 | 7027 | /* We set mc.moving_task later */ |
4ffef5fe DN |
7028 | |
7029 | ret = mem_cgroup_precharge_mc(mm); | |
7030 | if (ret) | |
7031 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
7032 | } |
7033 | mmput(mm); | |
7dc74be0 DN |
7034 | } |
7035 | return ret; | |
7036 | } | |
7037 | ||
eb95419b | 7038 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 7039 | struct cgroup_taskset *tset) |
7dc74be0 | 7040 | { |
4ffef5fe | 7041 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
7042 | } |
7043 | ||
4ffef5fe DN |
7044 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
7045 | unsigned long addr, unsigned long end, | |
7046 | struct mm_walk *walk) | |
7dc74be0 | 7047 | { |
4ffef5fe DN |
7048 | int ret = 0; |
7049 | struct vm_area_struct *vma = walk->private; | |
7050 | pte_t *pte; | |
7051 | spinlock_t *ptl; | |
12724850 NH |
7052 | enum mc_target_type target_type; |
7053 | union mc_target target; | |
7054 | struct page *page; | |
7055 | struct page_cgroup *pc; | |
4ffef5fe | 7056 | |
12724850 NH |
7057 | /* |
7058 | * We don't take compound_lock() here but no race with splitting thp | |
7059 | * happens because: | |
7060 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
7061 | * under splitting, which means there's no concurrent thp split, | |
7062 | * - if another thread runs into split_huge_page() just after we | |
7063 | * entered this if-block, the thread must wait for page table lock | |
7064 | * to be unlocked in __split_huge_page_splitting(), where the main | |
7065 | * part of thp split is not executed yet. | |
7066 | */ | |
7067 | if (pmd_trans_huge_lock(pmd, vma) == 1) { | |
62ade86a | 7068 | if (mc.precharge < HPAGE_PMD_NR) { |
12724850 NH |
7069 | spin_unlock(&vma->vm_mm->page_table_lock); |
7070 | return 0; | |
7071 | } | |
7072 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
7073 | if (target_type == MC_TARGET_PAGE) { | |
7074 | page = target.page; | |
7075 | if (!isolate_lru_page(page)) { | |
7076 | pc = lookup_page_cgroup(page); | |
7077 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, | |
2f3479b1 | 7078 | pc, mc.from, mc.to)) { |
12724850 NH |
7079 | mc.precharge -= HPAGE_PMD_NR; |
7080 | mc.moved_charge += HPAGE_PMD_NR; | |
7081 | } | |
7082 | putback_lru_page(page); | |
7083 | } | |
7084 | put_page(page); | |
7085 | } | |
7086 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 7087 | return 0; |
12724850 NH |
7088 | } |
7089 | ||
45f83cef AA |
7090 | if (pmd_trans_unstable(pmd)) |
7091 | return 0; | |
4ffef5fe DN |
7092 | retry: |
7093 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
7094 | for (; addr != end; addr += PAGE_SIZE) { | |
7095 | pte_t ptent = *(pte++); | |
02491447 | 7096 | swp_entry_t ent; |
4ffef5fe DN |
7097 | |
7098 | if (!mc.precharge) | |
7099 | break; | |
7100 | ||
8d32ff84 | 7101 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
7102 | case MC_TARGET_PAGE: |
7103 | page = target.page; | |
7104 | if (isolate_lru_page(page)) | |
7105 | goto put; | |
7106 | pc = lookup_page_cgroup(page); | |
7ec99d62 | 7107 | if (!mem_cgroup_move_account(page, 1, pc, |
2f3479b1 | 7108 | mc.from, mc.to)) { |
4ffef5fe | 7109 | mc.precharge--; |
854ffa8d DN |
7110 | /* we uncharge from mc.from later. */ |
7111 | mc.moved_charge++; | |
4ffef5fe DN |
7112 | } |
7113 | putback_lru_page(page); | |
8d32ff84 | 7114 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
7115 | put_page(page); |
7116 | break; | |
02491447 DN |
7117 | case MC_TARGET_SWAP: |
7118 | ent = target.ent; | |
e91cbb42 | 7119 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 7120 | mc.precharge--; |
483c30b5 DN |
7121 | /* we fixup refcnts and charges later. */ |
7122 | mc.moved_swap++; | |
7123 | } | |
02491447 | 7124 | break; |
4ffef5fe DN |
7125 | default: |
7126 | break; | |
7127 | } | |
7128 | } | |
7129 | pte_unmap_unlock(pte - 1, ptl); | |
7130 | cond_resched(); | |
7131 | ||
7132 | if (addr != end) { | |
7133 | /* | |
7134 | * We have consumed all precharges we got in can_attach(). | |
7135 | * We try charge one by one, but don't do any additional | |
7136 | * charges to mc.to if we have failed in charge once in attach() | |
7137 | * phase. | |
7138 | */ | |
854ffa8d | 7139 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
7140 | if (!ret) |
7141 | goto retry; | |
7142 | } | |
7143 | ||
7144 | return ret; | |
7145 | } | |
7146 | ||
7147 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
7148 | { | |
7149 | struct vm_area_struct *vma; | |
7150 | ||
7151 | lru_add_drain_all(); | |
dfe076b0 DN |
7152 | retry: |
7153 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
7154 | /* | |
7155 | * Someone who are holding the mmap_sem might be waiting in | |
7156 | * waitq. So we cancel all extra charges, wake up all waiters, | |
7157 | * and retry. Because we cancel precharges, we might not be able | |
7158 | * to move enough charges, but moving charge is a best-effort | |
7159 | * feature anyway, so it wouldn't be a big problem. | |
7160 | */ | |
7161 | __mem_cgroup_clear_mc(); | |
7162 | cond_resched(); | |
7163 | goto retry; | |
7164 | } | |
4ffef5fe DN |
7165 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
7166 | int ret; | |
7167 | struct mm_walk mem_cgroup_move_charge_walk = { | |
7168 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
7169 | .mm = mm, | |
7170 | .private = vma, | |
7171 | }; | |
7172 | if (is_vm_hugetlb_page(vma)) | |
7173 | continue; | |
4ffef5fe DN |
7174 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
7175 | &mem_cgroup_move_charge_walk); | |
7176 | if (ret) | |
7177 | /* | |
7178 | * means we have consumed all precharges and failed in | |
7179 | * doing additional charge. Just abandon here. | |
7180 | */ | |
7181 | break; | |
7182 | } | |
dfe076b0 | 7183 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
7184 | } |
7185 | ||
eb95419b | 7186 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 7187 | struct cgroup_taskset *tset) |
67e465a7 | 7188 | { |
2f7ee569 | 7189 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 7190 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 7191 | |
dfe076b0 | 7192 | if (mm) { |
a433658c KM |
7193 | if (mc.to) |
7194 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
7195 | mmput(mm); |
7196 | } | |
a433658c KM |
7197 | if (mc.to) |
7198 | mem_cgroup_clear_mc(); | |
67e465a7 | 7199 | } |
5cfb80a7 | 7200 | #else /* !CONFIG_MMU */ |
eb95419b | 7201 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 7202 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
7203 | { |
7204 | return 0; | |
7205 | } | |
eb95419b | 7206 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 7207 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
7208 | { |
7209 | } | |
eb95419b | 7210 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 7211 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
7212 | { |
7213 | } | |
7214 | #endif | |
67e465a7 | 7215 | |
f00baae7 TH |
7216 | /* |
7217 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
7218 | * to verify sane_behavior flag on each mount attempt. | |
7219 | */ | |
eb95419b | 7220 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
7221 | { |
7222 | /* | |
7223 | * use_hierarchy is forced with sane_behavior. cgroup core | |
7224 | * guarantees that @root doesn't have any children, so turning it | |
7225 | * on for the root memcg is enough. | |
7226 | */ | |
eb95419b TH |
7227 | if (cgroup_sane_behavior(root_css->cgroup)) |
7228 | mem_cgroup_from_css(root_css)->use_hierarchy = true; | |
f00baae7 TH |
7229 | } |
7230 | ||
8cdea7c0 BS |
7231 | struct cgroup_subsys mem_cgroup_subsys = { |
7232 | .name = "memory", | |
7233 | .subsys_id = mem_cgroup_subsys_id, | |
92fb9748 | 7234 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 7235 | .css_online = mem_cgroup_css_online, |
92fb9748 TH |
7236 | .css_offline = mem_cgroup_css_offline, |
7237 | .css_free = mem_cgroup_css_free, | |
7dc74be0 DN |
7238 | .can_attach = mem_cgroup_can_attach, |
7239 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 7240 | .attach = mem_cgroup_move_task, |
f00baae7 | 7241 | .bind = mem_cgroup_bind, |
6bc10349 | 7242 | .base_cftypes = mem_cgroup_files, |
6d12e2d8 | 7243 | .early_init = 0, |
04046e1a | 7244 | .use_id = 1, |
8cdea7c0 | 7245 | }; |
c077719b | 7246 | |
c255a458 | 7247 | #ifdef CONFIG_MEMCG_SWAP |
a42c390c MH |
7248 | static int __init enable_swap_account(char *s) |
7249 | { | |
a2c8990a | 7250 | if (!strcmp(s, "1")) |
a42c390c | 7251 | really_do_swap_account = 1; |
a2c8990a | 7252 | else if (!strcmp(s, "0")) |
a42c390c MH |
7253 | really_do_swap_account = 0; |
7254 | return 1; | |
7255 | } | |
a2c8990a | 7256 | __setup("swapaccount=", enable_swap_account); |
c077719b | 7257 | |
2d11085e MH |
7258 | static void __init memsw_file_init(void) |
7259 | { | |
6acc8b02 MH |
7260 | WARN_ON(cgroup_add_cftypes(&mem_cgroup_subsys, memsw_cgroup_files)); |
7261 | } | |
7262 | ||
7263 | static void __init enable_swap_cgroup(void) | |
7264 | { | |
7265 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
7266 | do_swap_account = 1; | |
7267 | memsw_file_init(); | |
7268 | } | |
2d11085e | 7269 | } |
6acc8b02 | 7270 | |
2d11085e | 7271 | #else |
6acc8b02 | 7272 | static void __init enable_swap_cgroup(void) |
2d11085e MH |
7273 | { |
7274 | } | |
c077719b | 7275 | #endif |
2d11085e MH |
7276 | |
7277 | /* | |
1081312f MH |
7278 | * subsys_initcall() for memory controller. |
7279 | * | |
7280 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
7281 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
7282 | * everything that doesn't depend on a specific mem_cgroup structure should | |
7283 | * be initialized from here. | |
2d11085e MH |
7284 | */ |
7285 | static int __init mem_cgroup_init(void) | |
7286 | { | |
7287 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); | |
6acc8b02 | 7288 | enable_swap_cgroup(); |
bb4cc1a8 | 7289 | mem_cgroup_soft_limit_tree_init(); |
e4777496 | 7290 | memcg_stock_init(); |
2d11085e MH |
7291 | return 0; |
7292 | } | |
7293 | subsys_initcall(mem_cgroup_init); |