]>
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 KS |
47 | #include <linux/eventfd.h> |
48 | #include <linux/sort.h> | |
66e1707b | 49 | #include <linux/fs.h> |
d2ceb9b7 | 50 | #include <linux/seq_file.h> |
33327948 | 51 | #include <linux/vmalloc.h> |
70ddf637 | 52 | #include <linux/vmpressure.h> |
b69408e8 | 53 | #include <linux/mm_inline.h> |
52d4b9ac | 54 | #include <linux/page_cgroup.h> |
cdec2e42 | 55 | #include <linux/cpu.h> |
158e0a2d | 56 | #include <linux/oom.h> |
0056f4e6 | 57 | #include <linux/lockdep.h> |
08e552c6 | 58 | #include "internal.h" |
d1a4c0b3 | 59 | #include <net/sock.h> |
4bd2c1ee | 60 | #include <net/ip.h> |
d1a4c0b3 | 61 | #include <net/tcp_memcontrol.h> |
8cdea7c0 | 62 | |
8697d331 BS |
63 | #include <asm/uaccess.h> |
64 | ||
cc8e970c KM |
65 | #include <trace/events/vmscan.h> |
66 | ||
a181b0e8 | 67 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
68ae564b DR |
68 | EXPORT_SYMBOL(mem_cgroup_subsys); |
69 | ||
a181b0e8 | 70 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
6bbda35c | 71 | static struct mem_cgroup *root_mem_cgroup __read_mostly; |
8cdea7c0 | 72 | |
c255a458 | 73 | #ifdef CONFIG_MEMCG_SWAP |
338c8431 | 74 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ |
c077719b | 75 | int do_swap_account __read_mostly; |
a42c390c MH |
76 | |
77 | /* for remember boot option*/ | |
c255a458 | 78 | #ifdef CONFIG_MEMCG_SWAP_ENABLED |
a42c390c MH |
79 | static int really_do_swap_account __initdata = 1; |
80 | #else | |
81 | static int really_do_swap_account __initdata = 0; | |
82 | #endif | |
83 | ||
c077719b | 84 | #else |
a0db00fc | 85 | #define do_swap_account 0 |
c077719b KH |
86 | #endif |
87 | ||
88 | ||
af7c4b0e JW |
89 | static const char * const mem_cgroup_stat_names[] = { |
90 | "cache", | |
91 | "rss", | |
b070e65c | 92 | "rss_huge", |
af7c4b0e | 93 | "mapped_file", |
3ea67d06 | 94 | "writeback", |
af7c4b0e JW |
95 | "swap", |
96 | }; | |
97 | ||
e9f8974f JW |
98 | enum mem_cgroup_events_index { |
99 | MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */ | |
100 | MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */ | |
456f998e YH |
101 | MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */ |
102 | MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */ | |
e9f8974f JW |
103 | MEM_CGROUP_EVENTS_NSTATS, |
104 | }; | |
af7c4b0e JW |
105 | |
106 | static const char * const mem_cgroup_events_names[] = { | |
107 | "pgpgin", | |
108 | "pgpgout", | |
109 | "pgfault", | |
110 | "pgmajfault", | |
111 | }; | |
112 | ||
58cf188e SZ |
113 | static const char * const mem_cgroup_lru_names[] = { |
114 | "inactive_anon", | |
115 | "active_anon", | |
116 | "inactive_file", | |
117 | "active_file", | |
118 | "unevictable", | |
119 | }; | |
120 | ||
7a159cc9 JW |
121 | /* |
122 | * Per memcg event counter is incremented at every pagein/pageout. With THP, | |
123 | * it will be incremated by the number of pages. This counter is used for | |
124 | * for trigger some periodic events. This is straightforward and better | |
125 | * than using jiffies etc. to handle periodic memcg event. | |
126 | */ | |
127 | enum mem_cgroup_events_target { | |
128 | MEM_CGROUP_TARGET_THRESH, | |
bb4cc1a8 | 129 | MEM_CGROUP_TARGET_SOFTLIMIT, |
453a9bf3 | 130 | MEM_CGROUP_TARGET_NUMAINFO, |
7a159cc9 JW |
131 | MEM_CGROUP_NTARGETS, |
132 | }; | |
a0db00fc KS |
133 | #define THRESHOLDS_EVENTS_TARGET 128 |
134 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
135 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 136 | |
d52aa412 | 137 | struct mem_cgroup_stat_cpu { |
7a159cc9 | 138 | long count[MEM_CGROUP_STAT_NSTATS]; |
e9f8974f | 139 | unsigned long events[MEM_CGROUP_EVENTS_NSTATS]; |
13114716 | 140 | unsigned long nr_page_events; |
7a159cc9 | 141 | unsigned long targets[MEM_CGROUP_NTARGETS]; |
d52aa412 KH |
142 | }; |
143 | ||
527a5ec9 | 144 | struct mem_cgroup_reclaim_iter { |
5f578161 MH |
145 | /* |
146 | * last scanned hierarchy member. Valid only if last_dead_count | |
147 | * matches memcg->dead_count of the hierarchy root group. | |
148 | */ | |
542f85f9 | 149 | struct mem_cgroup *last_visited; |
5f578161 MH |
150 | unsigned long last_dead_count; |
151 | ||
527a5ec9 JW |
152 | /* scan generation, increased every round-trip */ |
153 | unsigned int generation; | |
154 | }; | |
155 | ||
6d12e2d8 KH |
156 | /* |
157 | * per-zone information in memory controller. | |
158 | */ | |
6d12e2d8 | 159 | struct mem_cgroup_per_zone { |
6290df54 | 160 | struct lruvec lruvec; |
1eb49272 | 161 | unsigned long lru_size[NR_LRU_LISTS]; |
3e2f41f1 | 162 | |
527a5ec9 JW |
163 | struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1]; |
164 | ||
bb4cc1a8 AM |
165 | struct rb_node tree_node; /* RB tree node */ |
166 | unsigned long long usage_in_excess;/* Set to the value by which */ | |
167 | /* the soft limit is exceeded*/ | |
168 | bool on_tree; | |
d79154bb | 169 | struct mem_cgroup *memcg; /* Back pointer, we cannot */ |
4e416953 | 170 | /* use container_of */ |
6d12e2d8 | 171 | }; |
6d12e2d8 KH |
172 | |
173 | struct mem_cgroup_per_node { | |
174 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
175 | }; | |
176 | ||
bb4cc1a8 AM |
177 | /* |
178 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
179 | * their hierarchy representation | |
180 | */ | |
181 | ||
182 | struct mem_cgroup_tree_per_zone { | |
183 | struct rb_root rb_root; | |
184 | spinlock_t lock; | |
185 | }; | |
186 | ||
187 | struct mem_cgroup_tree_per_node { | |
188 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
189 | }; | |
190 | ||
191 | struct mem_cgroup_tree { | |
192 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
193 | }; | |
194 | ||
195 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
196 | ||
2e72b634 KS |
197 | struct mem_cgroup_threshold { |
198 | struct eventfd_ctx *eventfd; | |
199 | u64 threshold; | |
200 | }; | |
201 | ||
9490ff27 | 202 | /* For threshold */ |
2e72b634 | 203 | struct mem_cgroup_threshold_ary { |
748dad36 | 204 | /* An array index points to threshold just below or equal to usage. */ |
5407a562 | 205 | int current_threshold; |
2e72b634 KS |
206 | /* Size of entries[] */ |
207 | unsigned int size; | |
208 | /* Array of thresholds */ | |
209 | struct mem_cgroup_threshold entries[0]; | |
210 | }; | |
2c488db2 KS |
211 | |
212 | struct mem_cgroup_thresholds { | |
213 | /* Primary thresholds array */ | |
214 | struct mem_cgroup_threshold_ary *primary; | |
215 | /* | |
216 | * Spare threshold array. | |
217 | * This is needed to make mem_cgroup_unregister_event() "never fail". | |
218 | * It must be able to store at least primary->size - 1 entries. | |
219 | */ | |
220 | struct mem_cgroup_threshold_ary *spare; | |
221 | }; | |
222 | ||
9490ff27 KH |
223 | /* for OOM */ |
224 | struct mem_cgroup_eventfd_list { | |
225 | struct list_head list; | |
226 | struct eventfd_ctx *eventfd; | |
227 | }; | |
2e72b634 | 228 | |
c0ff4b85 R |
229 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
230 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 231 | |
8cdea7c0 BS |
232 | /* |
233 | * The memory controller data structure. The memory controller controls both | |
234 | * page cache and RSS per cgroup. We would eventually like to provide | |
235 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
236 | * to help the administrator determine what knobs to tune. | |
237 | * | |
238 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
239 | * we hit the water mark. May be even add a low water mark, such that |
240 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
241 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
242 | */ |
243 | struct mem_cgroup { | |
244 | struct cgroup_subsys_state css; | |
245 | /* | |
246 | * the counter to account for memory usage | |
247 | */ | |
248 | struct res_counter res; | |
59927fb9 | 249 | |
70ddf637 AV |
250 | /* vmpressure notifications */ |
251 | struct vmpressure vmpressure; | |
252 | ||
465939a1 LZ |
253 | /* |
254 | * the counter to account for mem+swap usage. | |
255 | */ | |
256 | struct res_counter memsw; | |
59927fb9 | 257 | |
510fc4e1 GC |
258 | /* |
259 | * the counter to account for kernel memory usage. | |
260 | */ | |
261 | struct res_counter kmem; | |
18f59ea7 BS |
262 | /* |
263 | * Should the accounting and control be hierarchical, per subtree? | |
264 | */ | |
265 | bool use_hierarchy; | |
510fc4e1 | 266 | unsigned long kmem_account_flags; /* See KMEM_ACCOUNTED_*, below */ |
79dfdacc MH |
267 | |
268 | bool oom_lock; | |
269 | atomic_t under_oom; | |
3812c8c8 | 270 | atomic_t oom_wakeups; |
79dfdacc | 271 | |
1f4c025b | 272 | int swappiness; |
3c11ecf4 KH |
273 | /* OOM-Killer disable */ |
274 | int oom_kill_disable; | |
a7885eb8 | 275 | |
22a668d7 KH |
276 | /* set when res.limit == memsw.limit */ |
277 | bool memsw_is_minimum; | |
278 | ||
2e72b634 KS |
279 | /* protect arrays of thresholds */ |
280 | struct mutex thresholds_lock; | |
281 | ||
282 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 283 | struct mem_cgroup_thresholds thresholds; |
907860ed | 284 | |
2e72b634 | 285 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 286 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 287 | |
9490ff27 KH |
288 | /* For oom notifier event fd */ |
289 | struct list_head oom_notify; | |
185efc0f | 290 | |
7dc74be0 DN |
291 | /* |
292 | * Should we move charges of a task when a task is moved into this | |
293 | * mem_cgroup ? And what type of charges should we move ? | |
294 | */ | |
f894ffa8 | 295 | unsigned long move_charge_at_immigrate; |
619d094b KH |
296 | /* |
297 | * set > 0 if pages under this cgroup are moving to other cgroup. | |
298 | */ | |
299 | atomic_t moving_account; | |
312734c0 KH |
300 | /* taken only while moving_account > 0 */ |
301 | spinlock_t move_lock; | |
d52aa412 | 302 | /* |
c62b1a3b | 303 | * percpu counter. |
d52aa412 | 304 | */ |
3a7951b4 | 305 | struct mem_cgroup_stat_cpu __percpu *stat; |
711d3d2c KH |
306 | /* |
307 | * used when a cpu is offlined or other synchronizations | |
308 | * See mem_cgroup_read_stat(). | |
309 | */ | |
310 | struct mem_cgroup_stat_cpu nocpu_base; | |
311 | spinlock_t pcp_counter_lock; | |
d1a4c0b3 | 312 | |
5f578161 | 313 | atomic_t dead_count; |
4bd2c1ee | 314 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET) |
d1a4c0b3 GC |
315 | struct tcp_memcontrol tcp_mem; |
316 | #endif | |
2633d7a0 GC |
317 | #if defined(CONFIG_MEMCG_KMEM) |
318 | /* analogous to slab_common's slab_caches list. per-memcg */ | |
319 | struct list_head memcg_slab_caches; | |
320 | /* Not a spinlock, we can take a lot of time walking the list */ | |
321 | struct mutex slab_caches_mutex; | |
322 | /* Index in the kmem_cache->memcg_params->memcg_caches array */ | |
323 | int kmemcg_id; | |
324 | #endif | |
45cf7ebd GC |
325 | |
326 | int last_scanned_node; | |
327 | #if MAX_NUMNODES > 1 | |
328 | nodemask_t scan_nodes; | |
329 | atomic_t numainfo_events; | |
330 | atomic_t numainfo_updating; | |
331 | #endif | |
70ddf637 | 332 | |
54f72fe0 JW |
333 | struct mem_cgroup_per_node *nodeinfo[0]; |
334 | /* WARNING: nodeinfo must be the last member here */ | |
8cdea7c0 BS |
335 | }; |
336 | ||
45cf7ebd GC |
337 | static size_t memcg_size(void) |
338 | { | |
339 | return sizeof(struct mem_cgroup) + | |
340 | nr_node_ids * sizeof(struct mem_cgroup_per_node); | |
341 | } | |
342 | ||
510fc4e1 GC |
343 | /* internal only representation about the status of kmem accounting. */ |
344 | enum { | |
345 | KMEM_ACCOUNTED_ACTIVE = 0, /* accounted by this cgroup itself */ | |
a8964b9b | 346 | KMEM_ACCOUNTED_ACTIVATED, /* static key enabled. */ |
7de37682 | 347 | KMEM_ACCOUNTED_DEAD, /* dead memcg with pending kmem charges */ |
510fc4e1 GC |
348 | }; |
349 | ||
a8964b9b GC |
350 | /* We account when limit is on, but only after call sites are patched */ |
351 | #define KMEM_ACCOUNTED_MASK \ | |
352 | ((1 << KMEM_ACCOUNTED_ACTIVE) | (1 << KMEM_ACCOUNTED_ACTIVATED)) | |
510fc4e1 GC |
353 | |
354 | #ifdef CONFIG_MEMCG_KMEM | |
355 | static inline void memcg_kmem_set_active(struct mem_cgroup *memcg) | |
356 | { | |
357 | set_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
358 | } | |
7de37682 GC |
359 | |
360 | static bool memcg_kmem_is_active(struct mem_cgroup *memcg) | |
361 | { | |
362 | return test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
363 | } | |
364 | ||
a8964b9b GC |
365 | static void memcg_kmem_set_activated(struct mem_cgroup *memcg) |
366 | { | |
367 | set_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags); | |
368 | } | |
369 | ||
55007d84 GC |
370 | static void memcg_kmem_clear_activated(struct mem_cgroup *memcg) |
371 | { | |
372 | clear_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags); | |
373 | } | |
374 | ||
7de37682 GC |
375 | static void memcg_kmem_mark_dead(struct mem_cgroup *memcg) |
376 | { | |
10d5ebf4 LZ |
377 | /* |
378 | * Our caller must use css_get() first, because memcg_uncharge_kmem() | |
379 | * will call css_put() if it sees the memcg is dead. | |
380 | */ | |
381 | smp_wmb(); | |
7de37682 GC |
382 | if (test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags)) |
383 | set_bit(KMEM_ACCOUNTED_DEAD, &memcg->kmem_account_flags); | |
384 | } | |
385 | ||
386 | static bool memcg_kmem_test_and_clear_dead(struct mem_cgroup *memcg) | |
387 | { | |
388 | return test_and_clear_bit(KMEM_ACCOUNTED_DEAD, | |
389 | &memcg->kmem_account_flags); | |
390 | } | |
510fc4e1 GC |
391 | #endif |
392 | ||
7dc74be0 DN |
393 | /* Stuffs for move charges at task migration. */ |
394 | /* | |
ee5e8472 GC |
395 | * Types of charges to be moved. "move_charge_at_immitgrate" and |
396 | * "immigrate_flags" are treated as a left-shifted bitmap of these types. | |
7dc74be0 DN |
397 | */ |
398 | enum move_type { | |
4ffef5fe | 399 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 400 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
401 | NR_MOVE_TYPE, |
402 | }; | |
403 | ||
4ffef5fe DN |
404 | /* "mc" and its members are protected by cgroup_mutex */ |
405 | static struct move_charge_struct { | |
b1dd693e | 406 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
407 | struct mem_cgroup *from; |
408 | struct mem_cgroup *to; | |
ee5e8472 | 409 | unsigned long immigrate_flags; |
4ffef5fe | 410 | unsigned long precharge; |
854ffa8d | 411 | unsigned long moved_charge; |
483c30b5 | 412 | unsigned long moved_swap; |
8033b97c DN |
413 | struct task_struct *moving_task; /* a task moving charges */ |
414 | wait_queue_head_t waitq; /* a waitq for other context */ | |
415 | } mc = { | |
2bd9bb20 | 416 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
417 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
418 | }; | |
4ffef5fe | 419 | |
90254a65 DN |
420 | static bool move_anon(void) |
421 | { | |
ee5e8472 | 422 | return test_bit(MOVE_CHARGE_TYPE_ANON, &mc.immigrate_flags); |
90254a65 DN |
423 | } |
424 | ||
87946a72 DN |
425 | static bool move_file(void) |
426 | { | |
ee5e8472 | 427 | return test_bit(MOVE_CHARGE_TYPE_FILE, &mc.immigrate_flags); |
87946a72 DN |
428 | } |
429 | ||
4e416953 BS |
430 | /* |
431 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
432 | * limit reclaim to prevent infinite loops, if they ever occur. | |
433 | */ | |
a0db00fc | 434 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 435 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 436 | |
217bc319 KH |
437 | enum charge_type { |
438 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 439 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 440 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 441 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
442 | NR_CHARGE_TYPE, |
443 | }; | |
444 | ||
8c7c6e34 | 445 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
446 | enum res_type { |
447 | _MEM, | |
448 | _MEMSWAP, | |
449 | _OOM_TYPE, | |
510fc4e1 | 450 | _KMEM, |
86ae53e1 GC |
451 | }; |
452 | ||
a0db00fc KS |
453 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
454 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 455 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
456 | /* Used for OOM nofiier */ |
457 | #define OOM_CONTROL (0) | |
8c7c6e34 | 458 | |
75822b44 BS |
459 | /* |
460 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
461 | */ | |
462 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
463 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
464 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
465 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
466 | ||
0999821b GC |
467 | /* |
468 | * The memcg_create_mutex will be held whenever a new cgroup is created. | |
469 | * As a consequence, any change that needs to protect against new child cgroups | |
470 | * appearing has to hold it as well. | |
471 | */ | |
472 | static DEFINE_MUTEX(memcg_create_mutex); | |
473 | ||
b2145145 WL |
474 | struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s) |
475 | { | |
a7c6d554 | 476 | return s ? container_of(s, struct mem_cgroup, css) : NULL; |
b2145145 WL |
477 | } |
478 | ||
70ddf637 AV |
479 | /* Some nice accessors for the vmpressure. */ |
480 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
481 | { | |
482 | if (!memcg) | |
483 | memcg = root_mem_cgroup; | |
484 | return &memcg->vmpressure; | |
485 | } | |
486 | ||
487 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
488 | { | |
489 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
490 | } | |
491 | ||
492 | struct vmpressure *css_to_vmpressure(struct cgroup_subsys_state *css) | |
493 | { | |
494 | return &mem_cgroup_from_css(css)->vmpressure; | |
495 | } | |
496 | ||
7ffc0edc MH |
497 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
498 | { | |
499 | return (memcg == root_mem_cgroup); | |
500 | } | |
501 | ||
e1aab161 | 502 | /* Writing them here to avoid exposing memcg's inner layout */ |
4bd2c1ee | 503 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
e1aab161 | 504 | |
e1aab161 GC |
505 | void sock_update_memcg(struct sock *sk) |
506 | { | |
376be5ff | 507 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 508 | struct mem_cgroup *memcg; |
3f134619 | 509 | struct cg_proto *cg_proto; |
e1aab161 GC |
510 | |
511 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
512 | ||
f3f511e1 GC |
513 | /* Socket cloning can throw us here with sk_cgrp already |
514 | * filled. It won't however, necessarily happen from | |
515 | * process context. So the test for root memcg given | |
516 | * the current task's memcg won't help us in this case. | |
517 | * | |
518 | * Respecting the original socket's memcg is a better | |
519 | * decision in this case. | |
520 | */ | |
521 | if (sk->sk_cgrp) { | |
522 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
5347e5ae | 523 | css_get(&sk->sk_cgrp->memcg->css); |
f3f511e1 GC |
524 | return; |
525 | } | |
526 | ||
e1aab161 GC |
527 | rcu_read_lock(); |
528 | memcg = mem_cgroup_from_task(current); | |
3f134619 | 529 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
5347e5ae LZ |
530 | if (!mem_cgroup_is_root(memcg) && |
531 | memcg_proto_active(cg_proto) && css_tryget(&memcg->css)) { | |
3f134619 | 532 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
533 | } |
534 | rcu_read_unlock(); | |
535 | } | |
536 | } | |
537 | EXPORT_SYMBOL(sock_update_memcg); | |
538 | ||
539 | void sock_release_memcg(struct sock *sk) | |
540 | { | |
376be5ff | 541 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
542 | struct mem_cgroup *memcg; |
543 | WARN_ON(!sk->sk_cgrp->memcg); | |
544 | memcg = sk->sk_cgrp->memcg; | |
5347e5ae | 545 | css_put(&sk->sk_cgrp->memcg->css); |
e1aab161 GC |
546 | } |
547 | } | |
d1a4c0b3 GC |
548 | |
549 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
550 | { | |
551 | if (!memcg || mem_cgroup_is_root(memcg)) | |
552 | return NULL; | |
553 | ||
554 | return &memcg->tcp_mem.cg_proto; | |
555 | } | |
556 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 557 | |
3f134619 GC |
558 | static void disarm_sock_keys(struct mem_cgroup *memcg) |
559 | { | |
560 | if (!memcg_proto_activated(&memcg->tcp_mem.cg_proto)) | |
561 | return; | |
562 | static_key_slow_dec(&memcg_socket_limit_enabled); | |
563 | } | |
564 | #else | |
565 | static void disarm_sock_keys(struct mem_cgroup *memcg) | |
566 | { | |
567 | } | |
568 | #endif | |
569 | ||
a8964b9b | 570 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 GC |
571 | /* |
572 | * This will be the memcg's index in each cache's ->memcg_params->memcg_caches. | |
573 | * There are two main reasons for not using the css_id for this: | |
574 | * 1) this works better in sparse environments, where we have a lot of memcgs, | |
575 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
576 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
577 | * 200 entry array for that. | |
578 | * | |
579 | * 2) In order not to violate the cgroup API, we would like to do all memory | |
580 | * allocation in ->create(). At that point, we haven't yet allocated the | |
581 | * css_id. Having a separate index prevents us from messing with the cgroup | |
582 | * core for this | |
583 | * | |
584 | * The current size of the caches array is stored in | |
585 | * memcg_limited_groups_array_size. It will double each time we have to | |
586 | * increase it. | |
587 | */ | |
588 | static DEFINE_IDA(kmem_limited_groups); | |
749c5415 GC |
589 | int memcg_limited_groups_array_size; |
590 | ||
55007d84 GC |
591 | /* |
592 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
593 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
594 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
595 | * tunable, but that is strictly not necessary. | |
596 | * | |
597 | * MAX_SIZE should be as large as the number of css_ids. Ideally, we could get | |
598 | * this constant directly from cgroup, but it is understandable that this is | |
599 | * better kept as an internal representation in cgroup.c. In any case, the | |
600 | * css_id space is not getting any smaller, and we don't have to necessarily | |
601 | * increase ours as well if it increases. | |
602 | */ | |
603 | #define MEMCG_CACHES_MIN_SIZE 4 | |
604 | #define MEMCG_CACHES_MAX_SIZE 65535 | |
605 | ||
d7f25f8a GC |
606 | /* |
607 | * A lot of the calls to the cache allocation functions are expected to be | |
608 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
609 | * conditional to this static branch, we'll have to allow modules that does | |
610 | * kmem_cache_alloc and the such to see this symbol as well | |
611 | */ | |
a8964b9b | 612 | struct static_key memcg_kmem_enabled_key; |
d7f25f8a | 613 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b GC |
614 | |
615 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
616 | { | |
55007d84 | 617 | if (memcg_kmem_is_active(memcg)) { |
a8964b9b | 618 | static_key_slow_dec(&memcg_kmem_enabled_key); |
55007d84 GC |
619 | ida_simple_remove(&kmem_limited_groups, memcg->kmemcg_id); |
620 | } | |
bea207c8 GC |
621 | /* |
622 | * This check can't live in kmem destruction function, | |
623 | * since the charges will outlive the cgroup | |
624 | */ | |
625 | WARN_ON(res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0); | |
a8964b9b GC |
626 | } |
627 | #else | |
628 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
629 | { | |
630 | } | |
631 | #endif /* CONFIG_MEMCG_KMEM */ | |
632 | ||
633 | static void disarm_static_keys(struct mem_cgroup *memcg) | |
634 | { | |
635 | disarm_sock_keys(memcg); | |
636 | disarm_kmem_keys(memcg); | |
637 | } | |
638 | ||
c0ff4b85 | 639 | static void drain_all_stock_async(struct mem_cgroup *memcg); |
8c7c6e34 | 640 | |
f64c3f54 | 641 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 642 | mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid) |
f64c3f54 | 643 | { |
45cf7ebd | 644 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
54f72fe0 | 645 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
646 | } |
647 | ||
c0ff4b85 | 648 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 649 | { |
c0ff4b85 | 650 | return &memcg->css; |
d324236b WF |
651 | } |
652 | ||
f64c3f54 | 653 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 654 | page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 655 | { |
97a6c37b JW |
656 | int nid = page_to_nid(page); |
657 | int zid = page_zonenum(page); | |
f64c3f54 | 658 | |
c0ff4b85 | 659 | return mem_cgroup_zoneinfo(memcg, nid, zid); |
f64c3f54 BS |
660 | } |
661 | ||
bb4cc1a8 AM |
662 | static struct mem_cgroup_tree_per_zone * |
663 | soft_limit_tree_node_zone(int nid, int zid) | |
664 | { | |
665 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
666 | } | |
667 | ||
668 | static struct mem_cgroup_tree_per_zone * | |
669 | soft_limit_tree_from_page(struct page *page) | |
670 | { | |
671 | int nid = page_to_nid(page); | |
672 | int zid = page_zonenum(page); | |
673 | ||
674 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
675 | } | |
676 | ||
677 | static void | |
678 | __mem_cgroup_insert_exceeded(struct mem_cgroup *memcg, | |
679 | struct mem_cgroup_per_zone *mz, | |
680 | struct mem_cgroup_tree_per_zone *mctz, | |
681 | unsigned long long new_usage_in_excess) | |
682 | { | |
683 | struct rb_node **p = &mctz->rb_root.rb_node; | |
684 | struct rb_node *parent = NULL; | |
685 | struct mem_cgroup_per_zone *mz_node; | |
686 | ||
687 | if (mz->on_tree) | |
688 | return; | |
689 | ||
690 | mz->usage_in_excess = new_usage_in_excess; | |
691 | if (!mz->usage_in_excess) | |
692 | return; | |
693 | while (*p) { | |
694 | parent = *p; | |
695 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
696 | tree_node); | |
697 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
698 | p = &(*p)->rb_left; | |
699 | /* | |
700 | * We can't avoid mem cgroups that are over their soft | |
701 | * limit by the same amount | |
702 | */ | |
703 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
704 | p = &(*p)->rb_right; | |
705 | } | |
706 | rb_link_node(&mz->tree_node, parent, p); | |
707 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
708 | mz->on_tree = true; | |
709 | } | |
710 | ||
711 | static void | |
712 | __mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, | |
713 | struct mem_cgroup_per_zone *mz, | |
714 | struct mem_cgroup_tree_per_zone *mctz) | |
715 | { | |
716 | if (!mz->on_tree) | |
717 | return; | |
718 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
719 | mz->on_tree = false; | |
720 | } | |
721 | ||
722 | static void | |
723 | mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, | |
724 | struct mem_cgroup_per_zone *mz, | |
725 | struct mem_cgroup_tree_per_zone *mctz) | |
726 | { | |
727 | spin_lock(&mctz->lock); | |
728 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); | |
729 | spin_unlock(&mctz->lock); | |
730 | } | |
731 | ||
732 | ||
733 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
734 | { | |
735 | unsigned long long excess; | |
736 | struct mem_cgroup_per_zone *mz; | |
737 | struct mem_cgroup_tree_per_zone *mctz; | |
738 | int nid = page_to_nid(page); | |
739 | int zid = page_zonenum(page); | |
740 | mctz = soft_limit_tree_from_page(page); | |
741 | ||
742 | /* | |
743 | * Necessary to update all ancestors when hierarchy is used. | |
744 | * because their event counter is not touched. | |
745 | */ | |
746 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
747 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
748 | excess = res_counter_soft_limit_excess(&memcg->res); | |
749 | /* | |
750 | * We have to update the tree if mz is on RB-tree or | |
751 | * mem is over its softlimit. | |
752 | */ | |
753 | if (excess || mz->on_tree) { | |
754 | spin_lock(&mctz->lock); | |
755 | /* if on-tree, remove it */ | |
756 | if (mz->on_tree) | |
757 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); | |
758 | /* | |
759 | * Insert again. mz->usage_in_excess will be updated. | |
760 | * If excess is 0, no tree ops. | |
761 | */ | |
762 | __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess); | |
763 | spin_unlock(&mctz->lock); | |
764 | } | |
765 | } | |
766 | } | |
767 | ||
768 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
769 | { | |
770 | int node, zone; | |
771 | struct mem_cgroup_per_zone *mz; | |
772 | struct mem_cgroup_tree_per_zone *mctz; | |
773 | ||
774 | for_each_node(node) { | |
775 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
776 | mz = mem_cgroup_zoneinfo(memcg, node, zone); | |
777 | mctz = soft_limit_tree_node_zone(node, zone); | |
778 | mem_cgroup_remove_exceeded(memcg, mz, mctz); | |
779 | } | |
780 | } | |
781 | } | |
782 | ||
783 | static struct mem_cgroup_per_zone * | |
784 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
785 | { | |
786 | struct rb_node *rightmost = NULL; | |
787 | struct mem_cgroup_per_zone *mz; | |
788 | ||
789 | retry: | |
790 | mz = NULL; | |
791 | rightmost = rb_last(&mctz->rb_root); | |
792 | if (!rightmost) | |
793 | goto done; /* Nothing to reclaim from */ | |
794 | ||
795 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
796 | /* | |
797 | * Remove the node now but someone else can add it back, | |
798 | * we will to add it back at the end of reclaim to its correct | |
799 | * position in the tree. | |
800 | */ | |
801 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); | |
802 | if (!res_counter_soft_limit_excess(&mz->memcg->res) || | |
803 | !css_tryget(&mz->memcg->css)) | |
804 | goto retry; | |
805 | done: | |
806 | return mz; | |
807 | } | |
808 | ||
809 | static struct mem_cgroup_per_zone * | |
810 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
811 | { | |
812 | struct mem_cgroup_per_zone *mz; | |
813 | ||
814 | spin_lock(&mctz->lock); | |
815 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
816 | spin_unlock(&mctz->lock); | |
817 | return mz; | |
818 | } | |
819 | ||
711d3d2c KH |
820 | /* |
821 | * Implementation Note: reading percpu statistics for memcg. | |
822 | * | |
823 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
824 | * synchronization to implement "quick" read. There are trade-off between | |
825 | * reading cost and precision of value. Then, we may have a chance to implement | |
826 | * a periodic synchronizion of counter in memcg's counter. | |
827 | * | |
828 | * But this _read() function is used for user interface now. The user accounts | |
829 | * memory usage by memory cgroup and he _always_ requires exact value because | |
830 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
831 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
832 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
833 | * | |
834 | * If there are kernel internal actions which can make use of some not-exact | |
835 | * value, and reading all cpu value can be performance bottleneck in some | |
836 | * common workload, threashold and synchonization as vmstat[] should be | |
837 | * implemented. | |
838 | */ | |
c0ff4b85 | 839 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 840 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 841 | { |
7a159cc9 | 842 | long val = 0; |
c62b1a3b | 843 | int cpu; |
c62b1a3b | 844 | |
711d3d2c KH |
845 | get_online_cpus(); |
846 | for_each_online_cpu(cpu) | |
c0ff4b85 | 847 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 848 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
849 | spin_lock(&memcg->pcp_counter_lock); |
850 | val += memcg->nocpu_base.count[idx]; | |
851 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
852 | #endif |
853 | put_online_cpus(); | |
c62b1a3b KH |
854 | return val; |
855 | } | |
856 | ||
c0ff4b85 | 857 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
0c3e73e8 BS |
858 | bool charge) |
859 | { | |
860 | int val = (charge) ? 1 : -1; | |
bff6bb83 | 861 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); |
0c3e73e8 BS |
862 | } |
863 | ||
c0ff4b85 | 864 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
865 | enum mem_cgroup_events_index idx) |
866 | { | |
867 | unsigned long val = 0; | |
868 | int cpu; | |
869 | ||
9c567512 | 870 | get_online_cpus(); |
e9f8974f | 871 | for_each_online_cpu(cpu) |
c0ff4b85 | 872 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 873 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
874 | spin_lock(&memcg->pcp_counter_lock); |
875 | val += memcg->nocpu_base.events[idx]; | |
876 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f | 877 | #endif |
9c567512 | 878 | put_online_cpus(); |
e9f8974f JW |
879 | return val; |
880 | } | |
881 | ||
c0ff4b85 | 882 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 883 | struct page *page, |
b2402857 | 884 | bool anon, int nr_pages) |
d52aa412 | 885 | { |
c62b1a3b KH |
886 | preempt_disable(); |
887 | ||
b2402857 KH |
888 | /* |
889 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
890 | * counted as CACHE even if it's on ANON LRU. | |
891 | */ | |
892 | if (anon) | |
893 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], | |
c0ff4b85 | 894 | nr_pages); |
d52aa412 | 895 | else |
b2402857 | 896 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 897 | nr_pages); |
55e462b0 | 898 | |
b070e65c DR |
899 | if (PageTransHuge(page)) |
900 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], | |
901 | nr_pages); | |
902 | ||
e401f176 KH |
903 | /* pagein of a big page is an event. So, ignore page size */ |
904 | if (nr_pages > 0) | |
c0ff4b85 | 905 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 906 | else { |
c0ff4b85 | 907 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
908 | nr_pages = -nr_pages; /* for event */ |
909 | } | |
e401f176 | 910 | |
13114716 | 911 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
2e72b634 | 912 | |
c62b1a3b | 913 | preempt_enable(); |
6d12e2d8 KH |
914 | } |
915 | ||
bb2a0de9 | 916 | unsigned long |
4d7dcca2 | 917 | mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
918 | { |
919 | struct mem_cgroup_per_zone *mz; | |
920 | ||
921 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
922 | return mz->lru_size[lru]; | |
923 | } | |
924 | ||
925 | static unsigned long | |
c0ff4b85 | 926 | mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid, |
bb2a0de9 | 927 | unsigned int lru_mask) |
889976db YH |
928 | { |
929 | struct mem_cgroup_per_zone *mz; | |
f156ab93 | 930 | enum lru_list lru; |
bb2a0de9 KH |
931 | unsigned long ret = 0; |
932 | ||
c0ff4b85 | 933 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
bb2a0de9 | 934 | |
f156ab93 HD |
935 | for_each_lru(lru) { |
936 | if (BIT(lru) & lru_mask) | |
937 | ret += mz->lru_size[lru]; | |
bb2a0de9 KH |
938 | } |
939 | return ret; | |
940 | } | |
941 | ||
942 | static unsigned long | |
c0ff4b85 | 943 | mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 KH |
944 | int nid, unsigned int lru_mask) |
945 | { | |
889976db YH |
946 | u64 total = 0; |
947 | int zid; | |
948 | ||
bb2a0de9 | 949 | for (zid = 0; zid < MAX_NR_ZONES; zid++) |
c0ff4b85 R |
950 | total += mem_cgroup_zone_nr_lru_pages(memcg, |
951 | nid, zid, lru_mask); | |
bb2a0de9 | 952 | |
889976db YH |
953 | return total; |
954 | } | |
bb2a0de9 | 955 | |
c0ff4b85 | 956 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 957 | unsigned int lru_mask) |
6d12e2d8 | 958 | { |
889976db | 959 | int nid; |
6d12e2d8 KH |
960 | u64 total = 0; |
961 | ||
31aaea4a | 962 | for_each_node_state(nid, N_MEMORY) |
c0ff4b85 | 963 | total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
6d12e2d8 | 964 | return total; |
d52aa412 KH |
965 | } |
966 | ||
f53d7ce3 JW |
967 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
968 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
969 | { |
970 | unsigned long val, next; | |
971 | ||
13114716 | 972 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 973 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 974 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
975 | if ((long)next - (long)val < 0) { |
976 | switch (target) { | |
977 | case MEM_CGROUP_TARGET_THRESH: | |
978 | next = val + THRESHOLDS_EVENTS_TARGET; | |
979 | break; | |
bb4cc1a8 AM |
980 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
981 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
982 | break; | |
f53d7ce3 JW |
983 | case MEM_CGROUP_TARGET_NUMAINFO: |
984 | next = val + NUMAINFO_EVENTS_TARGET; | |
985 | break; | |
986 | default: | |
987 | break; | |
988 | } | |
989 | __this_cpu_write(memcg->stat->targets[target], next); | |
990 | return true; | |
7a159cc9 | 991 | } |
f53d7ce3 | 992 | return false; |
d2265e6f KH |
993 | } |
994 | ||
995 | /* | |
996 | * Check events in order. | |
997 | * | |
998 | */ | |
c0ff4b85 | 999 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f | 1000 | { |
4799401f | 1001 | preempt_disable(); |
d2265e6f | 1002 | /* threshold event is triggered in finer grain than soft limit */ |
f53d7ce3 JW |
1003 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
1004 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 1005 | bool do_softlimit; |
82b3f2a7 | 1006 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 1007 | |
bb4cc1a8 AM |
1008 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
1009 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
1010 | #if MAX_NUMNODES > 1 |
1011 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
1012 | MEM_CGROUP_TARGET_NUMAINFO); | |
1013 | #endif | |
1014 | preempt_enable(); | |
1015 | ||
c0ff4b85 | 1016 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
1017 | if (unlikely(do_softlimit)) |
1018 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 1019 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 1020 | if (unlikely(do_numainfo)) |
c0ff4b85 | 1021 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 1022 | #endif |
f53d7ce3 JW |
1023 | } else |
1024 | preempt_enable(); | |
d2265e6f KH |
1025 | } |
1026 | ||
cf475ad2 | 1027 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 1028 | { |
31a78f23 BS |
1029 | /* |
1030 | * mm_update_next_owner() may clear mm->owner to NULL | |
1031 | * if it races with swapoff, page migration, etc. | |
1032 | * So this can be called with p == NULL. | |
1033 | */ | |
1034 | if (unlikely(!p)) | |
1035 | return NULL; | |
1036 | ||
8af01f56 | 1037 | return mem_cgroup_from_css(task_css(p, mem_cgroup_subsys_id)); |
78fb7466 PE |
1038 | } |
1039 | ||
a433658c | 1040 | struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 1041 | { |
c0ff4b85 | 1042 | struct mem_cgroup *memcg = NULL; |
0b7f569e KH |
1043 | |
1044 | if (!mm) | |
1045 | return NULL; | |
54595fe2 KH |
1046 | /* |
1047 | * Because we have no locks, mm->owner's may be being moved to other | |
1048 | * cgroup. We use css_tryget() here even if this looks | |
1049 | * pessimistic (rather than adding locks here). | |
1050 | */ | |
1051 | rcu_read_lock(); | |
1052 | do { | |
c0ff4b85 R |
1053 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1054 | if (unlikely(!memcg)) | |
54595fe2 | 1055 | break; |
c0ff4b85 | 1056 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 1057 | rcu_read_unlock(); |
c0ff4b85 | 1058 | return memcg; |
54595fe2 KH |
1059 | } |
1060 | ||
16248d8f MH |
1061 | /* |
1062 | * Returns a next (in a pre-order walk) alive memcg (with elevated css | |
1063 | * ref. count) or NULL if the whole root's subtree has been visited. | |
1064 | * | |
1065 | * helper function to be used by mem_cgroup_iter | |
1066 | */ | |
1067 | static struct mem_cgroup *__mem_cgroup_iter_next(struct mem_cgroup *root, | |
694fbc0f | 1068 | struct mem_cgroup *last_visited) |
16248d8f | 1069 | { |
492eb21b | 1070 | struct cgroup_subsys_state *prev_css, *next_css; |
16248d8f | 1071 | |
bd8815a6 | 1072 | prev_css = last_visited ? &last_visited->css : NULL; |
16248d8f | 1073 | skip_node: |
492eb21b | 1074 | next_css = css_next_descendant_pre(prev_css, &root->css); |
16248d8f MH |
1075 | |
1076 | /* | |
1077 | * Even if we found a group we have to make sure it is | |
1078 | * alive. css && !memcg means that the groups should be | |
1079 | * skipped and we should continue the tree walk. | |
1080 | * last_visited css is safe to use because it is | |
1081 | * protected by css_get and the tree walk is rcu safe. | |
1082 | */ | |
492eb21b TH |
1083 | if (next_css) { |
1084 | struct mem_cgroup *mem = mem_cgroup_from_css(next_css); | |
1085 | ||
694fbc0f AM |
1086 | if (css_tryget(&mem->css)) |
1087 | return mem; | |
1088 | else { | |
492eb21b | 1089 | prev_css = next_css; |
16248d8f MH |
1090 | goto skip_node; |
1091 | } | |
1092 | } | |
1093 | ||
1094 | return NULL; | |
1095 | } | |
1096 | ||
519ebea3 JW |
1097 | static void mem_cgroup_iter_invalidate(struct mem_cgroup *root) |
1098 | { | |
1099 | /* | |
1100 | * When a group in the hierarchy below root is destroyed, the | |
1101 | * hierarchy iterator can no longer be trusted since it might | |
1102 | * have pointed to the destroyed group. Invalidate it. | |
1103 | */ | |
1104 | atomic_inc(&root->dead_count); | |
1105 | } | |
1106 | ||
1107 | static struct mem_cgroup * | |
1108 | mem_cgroup_iter_load(struct mem_cgroup_reclaim_iter *iter, | |
1109 | struct mem_cgroup *root, | |
1110 | int *sequence) | |
1111 | { | |
1112 | struct mem_cgroup *position = NULL; | |
1113 | /* | |
1114 | * A cgroup destruction happens in two stages: offlining and | |
1115 | * release. They are separated by a RCU grace period. | |
1116 | * | |
1117 | * If the iterator is valid, we may still race with an | |
1118 | * offlining. The RCU lock ensures the object won't be | |
1119 | * released, tryget will fail if we lost the race. | |
1120 | */ | |
1121 | *sequence = atomic_read(&root->dead_count); | |
1122 | if (iter->last_dead_count == *sequence) { | |
1123 | smp_rmb(); | |
1124 | position = iter->last_visited; | |
1125 | if (position && !css_tryget(&position->css)) | |
1126 | position = NULL; | |
1127 | } | |
1128 | return position; | |
1129 | } | |
1130 | ||
1131 | static void mem_cgroup_iter_update(struct mem_cgroup_reclaim_iter *iter, | |
1132 | struct mem_cgroup *last_visited, | |
1133 | struct mem_cgroup *new_position, | |
1134 | int sequence) | |
1135 | { | |
1136 | if (last_visited) | |
1137 | css_put(&last_visited->css); | |
1138 | /* | |
1139 | * We store the sequence count from the time @last_visited was | |
1140 | * loaded successfully instead of rereading it here so that we | |
1141 | * don't lose destruction events in between. We could have | |
1142 | * raced with the destruction of @new_position after all. | |
1143 | */ | |
1144 | iter->last_visited = new_position; | |
1145 | smp_wmb(); | |
1146 | iter->last_dead_count = sequence; | |
1147 | } | |
1148 | ||
5660048c JW |
1149 | /** |
1150 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1151 | * @root: hierarchy root | |
1152 | * @prev: previously returned memcg, NULL on first invocation | |
1153 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1154 | * | |
1155 | * Returns references to children of the hierarchy below @root, or | |
1156 | * @root itself, or %NULL after a full round-trip. | |
1157 | * | |
1158 | * Caller must pass the return value in @prev on subsequent | |
1159 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1160 | * to cancel a hierarchy walk before the round-trip is complete. | |
1161 | * | |
1162 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
1163 | * divide up the memcgs in the hierarchy among all concurrent | |
1164 | * reclaimers operating on the same zone and priority. | |
1165 | */ | |
694fbc0f | 1166 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1167 | struct mem_cgroup *prev, |
694fbc0f | 1168 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1169 | { |
9f3a0d09 | 1170 | struct mem_cgroup *memcg = NULL; |
542f85f9 | 1171 | struct mem_cgroup *last_visited = NULL; |
711d3d2c | 1172 | |
694fbc0f AM |
1173 | if (mem_cgroup_disabled()) |
1174 | return NULL; | |
5660048c | 1175 | |
9f3a0d09 JW |
1176 | if (!root) |
1177 | root = root_mem_cgroup; | |
7d74b06f | 1178 | |
9f3a0d09 | 1179 | if (prev && !reclaim) |
542f85f9 | 1180 | last_visited = prev; |
14067bb3 | 1181 | |
9f3a0d09 JW |
1182 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1183 | if (prev) | |
c40046f3 | 1184 | goto out_css_put; |
694fbc0f | 1185 | return root; |
9f3a0d09 | 1186 | } |
14067bb3 | 1187 | |
542f85f9 | 1188 | rcu_read_lock(); |
9f3a0d09 | 1189 | while (!memcg) { |
527a5ec9 | 1190 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
519ebea3 | 1191 | int uninitialized_var(seq); |
711d3d2c | 1192 | |
527a5ec9 JW |
1193 | if (reclaim) { |
1194 | int nid = zone_to_nid(reclaim->zone); | |
1195 | int zid = zone_idx(reclaim->zone); | |
1196 | struct mem_cgroup_per_zone *mz; | |
1197 | ||
1198 | mz = mem_cgroup_zoneinfo(root, nid, zid); | |
1199 | iter = &mz->reclaim_iter[reclaim->priority]; | |
542f85f9 | 1200 | if (prev && reclaim->generation != iter->generation) { |
5f578161 | 1201 | iter->last_visited = NULL; |
542f85f9 MH |
1202 | goto out_unlock; |
1203 | } | |
5f578161 | 1204 | |
519ebea3 | 1205 | last_visited = mem_cgroup_iter_load(iter, root, &seq); |
527a5ec9 | 1206 | } |
7d74b06f | 1207 | |
694fbc0f | 1208 | memcg = __mem_cgroup_iter_next(root, last_visited); |
14067bb3 | 1209 | |
527a5ec9 | 1210 | if (reclaim) { |
519ebea3 | 1211 | mem_cgroup_iter_update(iter, last_visited, memcg, seq); |
542f85f9 | 1212 | |
19f39402 | 1213 | if (!memcg) |
527a5ec9 JW |
1214 | iter->generation++; |
1215 | else if (!prev && memcg) | |
1216 | reclaim->generation = iter->generation; | |
1217 | } | |
9f3a0d09 | 1218 | |
694fbc0f | 1219 | if (prev && !memcg) |
542f85f9 | 1220 | goto out_unlock; |
9f3a0d09 | 1221 | } |
542f85f9 MH |
1222 | out_unlock: |
1223 | rcu_read_unlock(); | |
c40046f3 MH |
1224 | out_css_put: |
1225 | if (prev && prev != root) | |
1226 | css_put(&prev->css); | |
1227 | ||
9f3a0d09 | 1228 | return memcg; |
14067bb3 | 1229 | } |
7d74b06f | 1230 | |
5660048c JW |
1231 | /** |
1232 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1233 | * @root: hierarchy root | |
1234 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1235 | */ | |
1236 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1237 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1238 | { |
1239 | if (!root) | |
1240 | root = root_mem_cgroup; | |
1241 | if (prev && prev != root) | |
1242 | css_put(&prev->css); | |
1243 | } | |
7d74b06f | 1244 | |
9f3a0d09 JW |
1245 | /* |
1246 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1247 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1248 | * be used for reference counting. | |
1249 | */ | |
1250 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1251 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1252 | iter != NULL; \ |
527a5ec9 | 1253 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1254 | |
9f3a0d09 | 1255 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1256 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1257 | iter != NULL; \ |
527a5ec9 | 1258 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1259 | |
68ae564b | 1260 | void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
456f998e | 1261 | { |
c0ff4b85 | 1262 | struct mem_cgroup *memcg; |
456f998e | 1263 | |
456f998e | 1264 | rcu_read_lock(); |
c0ff4b85 R |
1265 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1266 | if (unlikely(!memcg)) | |
456f998e YH |
1267 | goto out; |
1268 | ||
1269 | switch (idx) { | |
456f998e | 1270 | case PGFAULT: |
0e574a93 JW |
1271 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1272 | break; | |
1273 | case PGMAJFAULT: | |
1274 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1275 | break; |
1276 | default: | |
1277 | BUG(); | |
1278 | } | |
1279 | out: | |
1280 | rcu_read_unlock(); | |
1281 | } | |
68ae564b | 1282 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
456f998e | 1283 | |
925b7673 JW |
1284 | /** |
1285 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1286 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1287 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1288 | * |
1289 | * Returns the lru list vector holding pages for the given @zone and | |
1290 | * @mem. This can be the global zone lruvec, if the memory controller | |
1291 | * is disabled. | |
1292 | */ | |
1293 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1294 | struct mem_cgroup *memcg) | |
1295 | { | |
1296 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1297 | struct lruvec *lruvec; |
925b7673 | 1298 | |
bea8c150 HD |
1299 | if (mem_cgroup_disabled()) { |
1300 | lruvec = &zone->lruvec; | |
1301 | goto out; | |
1302 | } | |
925b7673 JW |
1303 | |
1304 | mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone)); | |
bea8c150 HD |
1305 | lruvec = &mz->lruvec; |
1306 | out: | |
1307 | /* | |
1308 | * Since a node can be onlined after the mem_cgroup was created, | |
1309 | * we have to be prepared to initialize lruvec->zone here; | |
1310 | * and if offlined then reonlined, we need to reinitialize it. | |
1311 | */ | |
1312 | if (unlikely(lruvec->zone != zone)) | |
1313 | lruvec->zone = zone; | |
1314 | return lruvec; | |
925b7673 JW |
1315 | } |
1316 | ||
08e552c6 KH |
1317 | /* |
1318 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
1319 | * Operations are called by routine of global LRU independently from memcg. | |
1320 | * What we have to take care of here is validness of pc->mem_cgroup. | |
1321 | * | |
1322 | * Changes to pc->mem_cgroup happens when | |
1323 | * 1. charge | |
1324 | * 2. moving account | |
1325 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
1326 | * It is added to LRU before charge. | |
1327 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
1328 | * When moving account, the page is not on LRU. It's isolated. | |
1329 | */ | |
4f98a2fe | 1330 | |
925b7673 | 1331 | /** |
fa9add64 | 1332 | * mem_cgroup_page_lruvec - return lruvec for adding an lru page |
925b7673 | 1333 | * @page: the page |
fa9add64 | 1334 | * @zone: zone of the page |
925b7673 | 1335 | */ |
fa9add64 | 1336 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1337 | { |
08e552c6 | 1338 | struct mem_cgroup_per_zone *mz; |
925b7673 JW |
1339 | struct mem_cgroup *memcg; |
1340 | struct page_cgroup *pc; | |
bea8c150 | 1341 | struct lruvec *lruvec; |
6d12e2d8 | 1342 | |
bea8c150 HD |
1343 | if (mem_cgroup_disabled()) { |
1344 | lruvec = &zone->lruvec; | |
1345 | goto out; | |
1346 | } | |
925b7673 | 1347 | |
08e552c6 | 1348 | pc = lookup_page_cgroup(page); |
38c5d72f | 1349 | memcg = pc->mem_cgroup; |
7512102c HD |
1350 | |
1351 | /* | |
fa9add64 | 1352 | * Surreptitiously switch any uncharged offlist page to root: |
7512102c HD |
1353 | * an uncharged page off lru does nothing to secure |
1354 | * its former mem_cgroup from sudden removal. | |
1355 | * | |
1356 | * Our caller holds lru_lock, and PageCgroupUsed is updated | |
1357 | * under page_cgroup lock: between them, they make all uses | |
1358 | * of pc->mem_cgroup safe. | |
1359 | */ | |
fa9add64 | 1360 | if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup) |
7512102c HD |
1361 | pc->mem_cgroup = memcg = root_mem_cgroup; |
1362 | ||
925b7673 | 1363 | mz = page_cgroup_zoneinfo(memcg, page); |
bea8c150 HD |
1364 | lruvec = &mz->lruvec; |
1365 | out: | |
1366 | /* | |
1367 | * Since a node can be onlined after the mem_cgroup was created, | |
1368 | * we have to be prepared to initialize lruvec->zone here; | |
1369 | * and if offlined then reonlined, we need to reinitialize it. | |
1370 | */ | |
1371 | if (unlikely(lruvec->zone != zone)) | |
1372 | lruvec->zone = zone; | |
1373 | return lruvec; | |
08e552c6 | 1374 | } |
b69408e8 | 1375 | |
925b7673 | 1376 | /** |
fa9add64 HD |
1377 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1378 | * @lruvec: mem_cgroup per zone lru vector | |
1379 | * @lru: index of lru list the page is sitting on | |
1380 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1381 | * |
fa9add64 HD |
1382 | * This function must be called when a page is added to or removed from an |
1383 | * lru list. | |
3f58a829 | 1384 | */ |
fa9add64 HD |
1385 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1386 | int nr_pages) | |
3f58a829 MK |
1387 | { |
1388 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1389 | unsigned long *lru_size; |
3f58a829 MK |
1390 | |
1391 | if (mem_cgroup_disabled()) | |
1392 | return; | |
1393 | ||
fa9add64 HD |
1394 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1395 | lru_size = mz->lru_size + lru; | |
1396 | *lru_size += nr_pages; | |
1397 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1398 | } |
544122e5 | 1399 | |
3e92041d | 1400 | /* |
c0ff4b85 | 1401 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1402 | * hierarchy subtree |
1403 | */ | |
c3ac9a8a JW |
1404 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1405 | struct mem_cgroup *memcg) | |
3e92041d | 1406 | { |
91c63734 JW |
1407 | if (root_memcg == memcg) |
1408 | return true; | |
3a981f48 | 1409 | if (!root_memcg->use_hierarchy || !memcg) |
91c63734 | 1410 | return false; |
c3ac9a8a JW |
1411 | return css_is_ancestor(&memcg->css, &root_memcg->css); |
1412 | } | |
1413 | ||
1414 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, | |
1415 | struct mem_cgroup *memcg) | |
1416 | { | |
1417 | bool ret; | |
1418 | ||
91c63734 | 1419 | rcu_read_lock(); |
c3ac9a8a | 1420 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
91c63734 JW |
1421 | rcu_read_unlock(); |
1422 | return ret; | |
3e92041d MH |
1423 | } |
1424 | ||
ffbdccf5 DR |
1425 | bool task_in_mem_cgroup(struct task_struct *task, |
1426 | const struct mem_cgroup *memcg) | |
4c4a2214 | 1427 | { |
0b7f569e | 1428 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1429 | struct task_struct *p; |
ffbdccf5 | 1430 | bool ret; |
4c4a2214 | 1431 | |
158e0a2d | 1432 | p = find_lock_task_mm(task); |
de077d22 DR |
1433 | if (p) { |
1434 | curr = try_get_mem_cgroup_from_mm(p->mm); | |
1435 | task_unlock(p); | |
1436 | } else { | |
1437 | /* | |
1438 | * All threads may have already detached their mm's, but the oom | |
1439 | * killer still needs to detect if they have already been oom | |
1440 | * killed to prevent needlessly killing additional tasks. | |
1441 | */ | |
ffbdccf5 | 1442 | rcu_read_lock(); |
de077d22 DR |
1443 | curr = mem_cgroup_from_task(task); |
1444 | if (curr) | |
1445 | css_get(&curr->css); | |
ffbdccf5 | 1446 | rcu_read_unlock(); |
de077d22 | 1447 | } |
0b7f569e | 1448 | if (!curr) |
ffbdccf5 | 1449 | return false; |
d31f56db | 1450 | /* |
c0ff4b85 | 1451 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1452 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1453 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1454 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1455 | */ |
c0ff4b85 | 1456 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1457 | css_put(&curr->css); |
4c4a2214 DR |
1458 | return ret; |
1459 | } | |
1460 | ||
c56d5c7d | 1461 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1462 | { |
9b272977 | 1463 | unsigned long inactive_ratio; |
14797e23 | 1464 | unsigned long inactive; |
9b272977 | 1465 | unsigned long active; |
c772be93 | 1466 | unsigned long gb; |
14797e23 | 1467 | |
4d7dcca2 HD |
1468 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1469 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1470 | |
c772be93 KM |
1471 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1472 | if (gb) | |
1473 | inactive_ratio = int_sqrt(10 * gb); | |
1474 | else | |
1475 | inactive_ratio = 1; | |
1476 | ||
9b272977 | 1477 | return inactive * inactive_ratio < active; |
14797e23 KM |
1478 | } |
1479 | ||
6d61ef40 BS |
1480 | #define mem_cgroup_from_res_counter(counter, member) \ |
1481 | container_of(counter, struct mem_cgroup, member) | |
1482 | ||
19942822 | 1483 | /** |
9d11ea9f | 1484 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1485 | * @memcg: the memory cgroup |
19942822 | 1486 | * |
9d11ea9f | 1487 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1488 | * pages. |
19942822 | 1489 | */ |
c0ff4b85 | 1490 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1491 | { |
9d11ea9f JW |
1492 | unsigned long long margin; |
1493 | ||
c0ff4b85 | 1494 | margin = res_counter_margin(&memcg->res); |
9d11ea9f | 1495 | if (do_swap_account) |
c0ff4b85 | 1496 | margin = min(margin, res_counter_margin(&memcg->memsw)); |
7ec99d62 | 1497 | return margin >> PAGE_SHIFT; |
19942822 JW |
1498 | } |
1499 | ||
1f4c025b | 1500 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 | 1501 | { |
a7885eb8 | 1502 | /* root ? */ |
63876986 | 1503 | if (!css_parent(&memcg->css)) |
a7885eb8 KM |
1504 | return vm_swappiness; |
1505 | ||
bf1ff263 | 1506 | return memcg->swappiness; |
a7885eb8 KM |
1507 | } |
1508 | ||
619d094b KH |
1509 | /* |
1510 | * memcg->moving_account is used for checking possibility that some thread is | |
1511 | * calling move_account(). When a thread on CPU-A starts moving pages under | |
1512 | * a memcg, other threads should check memcg->moving_account under | |
1513 | * rcu_read_lock(), like this: | |
1514 | * | |
1515 | * CPU-A CPU-B | |
1516 | * rcu_read_lock() | |
1517 | * memcg->moving_account+1 if (memcg->mocing_account) | |
1518 | * take heavy locks. | |
1519 | * synchronize_rcu() update something. | |
1520 | * rcu_read_unlock() | |
1521 | * start move here. | |
1522 | */ | |
4331f7d3 KH |
1523 | |
1524 | /* for quick checking without looking up memcg */ | |
1525 | atomic_t memcg_moving __read_mostly; | |
1526 | ||
c0ff4b85 | 1527 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a | 1528 | { |
4331f7d3 | 1529 | atomic_inc(&memcg_moving); |
619d094b | 1530 | atomic_inc(&memcg->moving_account); |
32047e2a KH |
1531 | synchronize_rcu(); |
1532 | } | |
1533 | ||
c0ff4b85 | 1534 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a | 1535 | { |
619d094b KH |
1536 | /* |
1537 | * Now, mem_cgroup_clear_mc() may call this function with NULL. | |
1538 | * We check NULL in callee rather than caller. | |
1539 | */ | |
4331f7d3 KH |
1540 | if (memcg) { |
1541 | atomic_dec(&memcg_moving); | |
619d094b | 1542 | atomic_dec(&memcg->moving_account); |
4331f7d3 | 1543 | } |
32047e2a | 1544 | } |
619d094b | 1545 | |
32047e2a KH |
1546 | /* |
1547 | * 2 routines for checking "mem" is under move_account() or not. | |
1548 | * | |
13fd1dd9 AM |
1549 | * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This |
1550 | * is used for avoiding races in accounting. If true, | |
32047e2a KH |
1551 | * pc->mem_cgroup may be overwritten. |
1552 | * | |
1553 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or | |
1554 | * under hierarchy of moving cgroups. This is for | |
1555 | * waiting at hith-memory prressure caused by "move". | |
1556 | */ | |
1557 | ||
13fd1dd9 | 1558 | static bool mem_cgroup_stolen(struct mem_cgroup *memcg) |
32047e2a KH |
1559 | { |
1560 | VM_BUG_ON(!rcu_read_lock_held()); | |
619d094b | 1561 | return atomic_read(&memcg->moving_account) > 0; |
32047e2a | 1562 | } |
4b534334 | 1563 | |
c0ff4b85 | 1564 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1565 | { |
2bd9bb20 KH |
1566 | struct mem_cgroup *from; |
1567 | struct mem_cgroup *to; | |
4b534334 | 1568 | bool ret = false; |
2bd9bb20 KH |
1569 | /* |
1570 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1571 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1572 | */ | |
1573 | spin_lock(&mc.lock); | |
1574 | from = mc.from; | |
1575 | to = mc.to; | |
1576 | if (!from) | |
1577 | goto unlock; | |
3e92041d | 1578 | |
c0ff4b85 R |
1579 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1580 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1581 | unlock: |
1582 | spin_unlock(&mc.lock); | |
4b534334 KH |
1583 | return ret; |
1584 | } | |
1585 | ||
c0ff4b85 | 1586 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1587 | { |
1588 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1589 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1590 | DEFINE_WAIT(wait); |
1591 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1592 | /* moving charge context might have finished. */ | |
1593 | if (mc.moving_task) | |
1594 | schedule(); | |
1595 | finish_wait(&mc.waitq, &wait); | |
1596 | return true; | |
1597 | } | |
1598 | } | |
1599 | return false; | |
1600 | } | |
1601 | ||
312734c0 KH |
1602 | /* |
1603 | * Take this lock when | |
1604 | * - a code tries to modify page's memcg while it's USED. | |
1605 | * - a code tries to modify page state accounting in a memcg. | |
13fd1dd9 | 1606 | * see mem_cgroup_stolen(), too. |
312734c0 KH |
1607 | */ |
1608 | static void move_lock_mem_cgroup(struct mem_cgroup *memcg, | |
1609 | unsigned long *flags) | |
1610 | { | |
1611 | spin_lock_irqsave(&memcg->move_lock, *flags); | |
1612 | } | |
1613 | ||
1614 | static void move_unlock_mem_cgroup(struct mem_cgroup *memcg, | |
1615 | unsigned long *flags) | |
1616 | { | |
1617 | spin_unlock_irqrestore(&memcg->move_lock, *flags); | |
1618 | } | |
1619 | ||
58cf188e | 1620 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1621 | /** |
58cf188e | 1622 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1623 | * @memcg: The memory cgroup that went over limit |
1624 | * @p: Task that is going to be killed | |
1625 | * | |
1626 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1627 | * enabled | |
1628 | */ | |
1629 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1630 | { | |
1631 | struct cgroup *task_cgrp; | |
1632 | struct cgroup *mem_cgrp; | |
1633 | /* | |
1634 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1635 | * on the assumption that OOM is serialized for memory controller. | |
1636 | * If this assumption is broken, revisit this code. | |
1637 | */ | |
1638 | static char memcg_name[PATH_MAX]; | |
1639 | int ret; | |
58cf188e SZ |
1640 | struct mem_cgroup *iter; |
1641 | unsigned int i; | |
e222432b | 1642 | |
58cf188e | 1643 | if (!p) |
e222432b BS |
1644 | return; |
1645 | ||
e222432b BS |
1646 | rcu_read_lock(); |
1647 | ||
1648 | mem_cgrp = memcg->css.cgroup; | |
1649 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1650 | ||
1651 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1652 | if (ret < 0) { | |
1653 | /* | |
1654 | * Unfortunately, we are unable to convert to a useful name | |
1655 | * But we'll still print out the usage information | |
1656 | */ | |
1657 | rcu_read_unlock(); | |
1658 | goto done; | |
1659 | } | |
1660 | rcu_read_unlock(); | |
1661 | ||
d045197f | 1662 | pr_info("Task in %s killed", memcg_name); |
e222432b BS |
1663 | |
1664 | rcu_read_lock(); | |
1665 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1666 | if (ret < 0) { | |
1667 | rcu_read_unlock(); | |
1668 | goto done; | |
1669 | } | |
1670 | rcu_read_unlock(); | |
1671 | ||
1672 | /* | |
1673 | * Continues from above, so we don't need an KERN_ level | |
1674 | */ | |
d045197f | 1675 | pr_cont(" as a result of limit of %s\n", memcg_name); |
e222432b BS |
1676 | done: |
1677 | ||
d045197f | 1678 | pr_info("memory: usage %llukB, limit %llukB, failcnt %llu\n", |
e222432b BS |
1679 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, |
1680 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1681 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
d045197f | 1682 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %llu\n", |
e222432b BS |
1683 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, |
1684 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1685 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
d045197f | 1686 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %llu\n", |
510fc4e1 GC |
1687 | res_counter_read_u64(&memcg->kmem, RES_USAGE) >> 10, |
1688 | res_counter_read_u64(&memcg->kmem, RES_LIMIT) >> 10, | |
1689 | res_counter_read_u64(&memcg->kmem, RES_FAILCNT)); | |
58cf188e SZ |
1690 | |
1691 | for_each_mem_cgroup_tree(iter, memcg) { | |
1692 | pr_info("Memory cgroup stats"); | |
1693 | ||
1694 | rcu_read_lock(); | |
1695 | ret = cgroup_path(iter->css.cgroup, memcg_name, PATH_MAX); | |
1696 | if (!ret) | |
1697 | pr_cont(" for %s", memcg_name); | |
1698 | rcu_read_unlock(); | |
1699 | pr_cont(":"); | |
1700 | ||
1701 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1702 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1703 | continue; | |
1704 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], | |
1705 | K(mem_cgroup_read_stat(iter, i))); | |
1706 | } | |
1707 | ||
1708 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1709 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1710 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1711 | ||
1712 | pr_cont("\n"); | |
1713 | } | |
e222432b BS |
1714 | } |
1715 | ||
81d39c20 KH |
1716 | /* |
1717 | * This function returns the number of memcg under hierarchy tree. Returns | |
1718 | * 1(self count) if no children. | |
1719 | */ | |
c0ff4b85 | 1720 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1721 | { |
1722 | int num = 0; | |
7d74b06f KH |
1723 | struct mem_cgroup *iter; |
1724 | ||
c0ff4b85 | 1725 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1726 | num++; |
81d39c20 KH |
1727 | return num; |
1728 | } | |
1729 | ||
a63d83f4 DR |
1730 | /* |
1731 | * Return the memory (and swap, if configured) limit for a memcg. | |
1732 | */ | |
9cbb78bb | 1733 | static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 DR |
1734 | { |
1735 | u64 limit; | |
a63d83f4 | 1736 | |
f3e8eb70 | 1737 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
f3e8eb70 | 1738 | |
a63d83f4 | 1739 | /* |
9a5a8f19 | 1740 | * Do not consider swap space if we cannot swap due to swappiness |
a63d83f4 | 1741 | */ |
9a5a8f19 MH |
1742 | if (mem_cgroup_swappiness(memcg)) { |
1743 | u64 memsw; | |
1744 | ||
1745 | limit += total_swap_pages << PAGE_SHIFT; | |
1746 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1747 | ||
1748 | /* | |
1749 | * If memsw is finite and limits the amount of swap space | |
1750 | * available to this memcg, return that limit. | |
1751 | */ | |
1752 | limit = min(limit, memsw); | |
1753 | } | |
1754 | ||
1755 | return limit; | |
a63d83f4 DR |
1756 | } |
1757 | ||
19965460 DR |
1758 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1759 | int order) | |
9cbb78bb DR |
1760 | { |
1761 | struct mem_cgroup *iter; | |
1762 | unsigned long chosen_points = 0; | |
1763 | unsigned long totalpages; | |
1764 | unsigned int points = 0; | |
1765 | struct task_struct *chosen = NULL; | |
1766 | ||
876aafbf | 1767 | /* |
465adcf1 DR |
1768 | * If current has a pending SIGKILL or is exiting, then automatically |
1769 | * select it. The goal is to allow it to allocate so that it may | |
1770 | * quickly exit and free its memory. | |
876aafbf | 1771 | */ |
465adcf1 | 1772 | if (fatal_signal_pending(current) || current->flags & PF_EXITING) { |
876aafbf DR |
1773 | set_thread_flag(TIF_MEMDIE); |
1774 | return; | |
1775 | } | |
1776 | ||
1777 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
9cbb78bb DR |
1778 | totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1; |
1779 | for_each_mem_cgroup_tree(iter, memcg) { | |
72ec7029 | 1780 | struct css_task_iter it; |
9cbb78bb DR |
1781 | struct task_struct *task; |
1782 | ||
72ec7029 TH |
1783 | css_task_iter_start(&iter->css, &it); |
1784 | while ((task = css_task_iter_next(&it))) { | |
9cbb78bb DR |
1785 | switch (oom_scan_process_thread(task, totalpages, NULL, |
1786 | false)) { | |
1787 | case OOM_SCAN_SELECT: | |
1788 | if (chosen) | |
1789 | put_task_struct(chosen); | |
1790 | chosen = task; | |
1791 | chosen_points = ULONG_MAX; | |
1792 | get_task_struct(chosen); | |
1793 | /* fall through */ | |
1794 | case OOM_SCAN_CONTINUE: | |
1795 | continue; | |
1796 | case OOM_SCAN_ABORT: | |
72ec7029 | 1797 | css_task_iter_end(&it); |
9cbb78bb DR |
1798 | mem_cgroup_iter_break(memcg, iter); |
1799 | if (chosen) | |
1800 | put_task_struct(chosen); | |
1801 | return; | |
1802 | case OOM_SCAN_OK: | |
1803 | break; | |
1804 | }; | |
1805 | points = oom_badness(task, memcg, NULL, totalpages); | |
1806 | if (points > chosen_points) { | |
1807 | if (chosen) | |
1808 | put_task_struct(chosen); | |
1809 | chosen = task; | |
1810 | chosen_points = points; | |
1811 | get_task_struct(chosen); | |
1812 | } | |
1813 | } | |
72ec7029 | 1814 | css_task_iter_end(&it); |
9cbb78bb DR |
1815 | } |
1816 | ||
1817 | if (!chosen) | |
1818 | return; | |
1819 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1820 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1821 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1822 | } |
1823 | ||
5660048c JW |
1824 | static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg, |
1825 | gfp_t gfp_mask, | |
1826 | unsigned long flags) | |
1827 | { | |
1828 | unsigned long total = 0; | |
1829 | bool noswap = false; | |
1830 | int loop; | |
1831 | ||
1832 | if (flags & MEM_CGROUP_RECLAIM_NOSWAP) | |
1833 | noswap = true; | |
1834 | if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum) | |
1835 | noswap = true; | |
1836 | ||
1837 | for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) { | |
1838 | if (loop) | |
1839 | drain_all_stock_async(memcg); | |
1840 | total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap); | |
1841 | /* | |
1842 | * Allow limit shrinkers, which are triggered directly | |
1843 | * by userspace, to catch signals and stop reclaim | |
1844 | * after minimal progress, regardless of the margin. | |
1845 | */ | |
1846 | if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK)) | |
1847 | break; | |
1848 | if (mem_cgroup_margin(memcg)) | |
1849 | break; | |
1850 | /* | |
1851 | * If nothing was reclaimed after two attempts, there | |
1852 | * may be no reclaimable pages in this hierarchy. | |
1853 | */ | |
1854 | if (loop && !total) | |
1855 | break; | |
1856 | } | |
1857 | return total; | |
1858 | } | |
1859 | ||
4d0c066d KH |
1860 | /** |
1861 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1862 | * @memcg: the target memcg |
4d0c066d KH |
1863 | * @nid: the node ID to be checked. |
1864 | * @noswap : specify true here if the user wants flle only information. | |
1865 | * | |
1866 | * This function returns whether the specified memcg contains any | |
1867 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1868 | * pages in the node. | |
1869 | */ | |
c0ff4b85 | 1870 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1871 | int nid, bool noswap) |
1872 | { | |
c0ff4b85 | 1873 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1874 | return true; |
1875 | if (noswap || !total_swap_pages) | |
1876 | return false; | |
c0ff4b85 | 1877 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1878 | return true; |
1879 | return false; | |
1880 | ||
1881 | } | |
bb4cc1a8 | 1882 | #if MAX_NUMNODES > 1 |
889976db YH |
1883 | |
1884 | /* | |
1885 | * Always updating the nodemask is not very good - even if we have an empty | |
1886 | * list or the wrong list here, we can start from some node and traverse all | |
1887 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1888 | * | |
1889 | */ | |
c0ff4b85 | 1890 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1891 | { |
1892 | int nid; | |
453a9bf3 KH |
1893 | /* |
1894 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1895 | * pagein/pageout changes since the last update. | |
1896 | */ | |
c0ff4b85 | 1897 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1898 | return; |
c0ff4b85 | 1899 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1900 | return; |
1901 | ||
889976db | 1902 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1903 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1904 | |
31aaea4a | 1905 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1906 | |
c0ff4b85 R |
1907 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1908 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1909 | } |
453a9bf3 | 1910 | |
c0ff4b85 R |
1911 | atomic_set(&memcg->numainfo_events, 0); |
1912 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1913 | } |
1914 | ||
1915 | /* | |
1916 | * Selecting a node where we start reclaim from. Because what we need is just | |
1917 | * reducing usage counter, start from anywhere is O,K. Considering | |
1918 | * memory reclaim from current node, there are pros. and cons. | |
1919 | * | |
1920 | * Freeing memory from current node means freeing memory from a node which | |
1921 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1922 | * hit limits, it will see a contention on a node. But freeing from remote | |
1923 | * node means more costs for memory reclaim because of memory latency. | |
1924 | * | |
1925 | * Now, we use round-robin. Better algorithm is welcomed. | |
1926 | */ | |
c0ff4b85 | 1927 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1928 | { |
1929 | int node; | |
1930 | ||
c0ff4b85 R |
1931 | mem_cgroup_may_update_nodemask(memcg); |
1932 | node = memcg->last_scanned_node; | |
889976db | 1933 | |
c0ff4b85 | 1934 | node = next_node(node, memcg->scan_nodes); |
889976db | 1935 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1936 | node = first_node(memcg->scan_nodes); |
889976db YH |
1937 | /* |
1938 | * We call this when we hit limit, not when pages are added to LRU. | |
1939 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1940 | * memcg is too small and all pages are not on LRU. In that case, | |
1941 | * we use curret node. | |
1942 | */ | |
1943 | if (unlikely(node == MAX_NUMNODES)) | |
1944 | node = numa_node_id(); | |
1945 | ||
c0ff4b85 | 1946 | memcg->last_scanned_node = node; |
889976db YH |
1947 | return node; |
1948 | } | |
1949 | ||
bb4cc1a8 AM |
1950 | /* |
1951 | * Check all nodes whether it contains reclaimable pages or not. | |
1952 | * For quick scan, we make use of scan_nodes. This will allow us to skip | |
1953 | * unused nodes. But scan_nodes is lazily updated and may not cotain | |
1954 | * enough new information. We need to do double check. | |
1955 | */ | |
1956 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) | |
1957 | { | |
1958 | int nid; | |
1959 | ||
1960 | /* | |
1961 | * quick check...making use of scan_node. | |
1962 | * We can skip unused nodes. | |
1963 | */ | |
1964 | if (!nodes_empty(memcg->scan_nodes)) { | |
1965 | for (nid = first_node(memcg->scan_nodes); | |
1966 | nid < MAX_NUMNODES; | |
1967 | nid = next_node(nid, memcg->scan_nodes)) { | |
1968 | ||
1969 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) | |
1970 | return true; | |
1971 | } | |
1972 | } | |
1973 | /* | |
1974 | * Check rest of nodes. | |
1975 | */ | |
1976 | for_each_node_state(nid, N_MEMORY) { | |
1977 | if (node_isset(nid, memcg->scan_nodes)) | |
1978 | continue; | |
1979 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) | |
1980 | return true; | |
1981 | } | |
1982 | return false; | |
1983 | } | |
1984 | ||
889976db | 1985 | #else |
c0ff4b85 | 1986 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1987 | { |
1988 | return 0; | |
1989 | } | |
4d0c066d | 1990 | |
bb4cc1a8 AM |
1991 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
1992 | { | |
1993 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); | |
1994 | } | |
889976db YH |
1995 | #endif |
1996 | ||
0608f43d AM |
1997 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1998 | struct zone *zone, | |
1999 | gfp_t gfp_mask, | |
2000 | unsigned long *total_scanned) | |
2001 | { | |
2002 | struct mem_cgroup *victim = NULL; | |
2003 | int total = 0; | |
2004 | int loop = 0; | |
2005 | unsigned long excess; | |
2006 | unsigned long nr_scanned; | |
2007 | struct mem_cgroup_reclaim_cookie reclaim = { | |
2008 | .zone = zone, | |
2009 | .priority = 0, | |
2010 | }; | |
2011 | ||
2012 | excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT; | |
2013 | ||
2014 | while (1) { | |
2015 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
2016 | if (!victim) { | |
2017 | loop++; | |
2018 | if (loop >= 2) { | |
2019 | /* | |
2020 | * If we have not been able to reclaim | |
2021 | * anything, it might because there are | |
2022 | * no reclaimable pages under this hierarchy | |
2023 | */ | |
2024 | if (!total) | |
2025 | break; | |
2026 | /* | |
2027 | * We want to do more targeted reclaim. | |
2028 | * excess >> 2 is not to excessive so as to | |
2029 | * reclaim too much, nor too less that we keep | |
2030 | * coming back to reclaim from this cgroup | |
2031 | */ | |
2032 | if (total >= (excess >> 2) || | |
2033 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
2034 | break; | |
2035 | } | |
2036 | continue; | |
2037 | } | |
2038 | if (!mem_cgroup_reclaimable(victim, false)) | |
2039 | continue; | |
2040 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, | |
2041 | zone, &nr_scanned); | |
2042 | *total_scanned += nr_scanned; | |
2043 | if (!res_counter_soft_limit_excess(&root_memcg->res)) | |
2044 | break; | |
6d61ef40 | 2045 | } |
0608f43d AM |
2046 | mem_cgroup_iter_break(root_memcg, victim); |
2047 | return total; | |
6d61ef40 BS |
2048 | } |
2049 | ||
0056f4e6 JW |
2050 | #ifdef CONFIG_LOCKDEP |
2051 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
2052 | .name = "memcg_oom_lock", | |
2053 | }; | |
2054 | #endif | |
2055 | ||
fb2a6fc5 JW |
2056 | static DEFINE_SPINLOCK(memcg_oom_lock); |
2057 | ||
867578cb KH |
2058 | /* |
2059 | * Check OOM-Killer is already running under our hierarchy. | |
2060 | * If someone is running, return false. | |
2061 | */ | |
fb2a6fc5 | 2062 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 2063 | { |
79dfdacc | 2064 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 2065 | |
fb2a6fc5 JW |
2066 | spin_lock(&memcg_oom_lock); |
2067 | ||
9f3a0d09 | 2068 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 2069 | if (iter->oom_lock) { |
79dfdacc MH |
2070 | /* |
2071 | * this subtree of our hierarchy is already locked | |
2072 | * so we cannot give a lock. | |
2073 | */ | |
79dfdacc | 2074 | failed = iter; |
9f3a0d09 JW |
2075 | mem_cgroup_iter_break(memcg, iter); |
2076 | break; | |
23751be0 JW |
2077 | } else |
2078 | iter->oom_lock = true; | |
7d74b06f | 2079 | } |
867578cb | 2080 | |
fb2a6fc5 JW |
2081 | if (failed) { |
2082 | /* | |
2083 | * OK, we failed to lock the whole subtree so we have | |
2084 | * to clean up what we set up to the failing subtree | |
2085 | */ | |
2086 | for_each_mem_cgroup_tree(iter, memcg) { | |
2087 | if (iter == failed) { | |
2088 | mem_cgroup_iter_break(memcg, iter); | |
2089 | break; | |
2090 | } | |
2091 | iter->oom_lock = false; | |
79dfdacc | 2092 | } |
0056f4e6 JW |
2093 | } else |
2094 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
2095 | |
2096 | spin_unlock(&memcg_oom_lock); | |
2097 | ||
2098 | return !failed; | |
a636b327 | 2099 | } |
0b7f569e | 2100 | |
fb2a6fc5 | 2101 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 2102 | { |
7d74b06f KH |
2103 | struct mem_cgroup *iter; |
2104 | ||
fb2a6fc5 | 2105 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 2106 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 2107 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 2108 | iter->oom_lock = false; |
fb2a6fc5 | 2109 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
2110 | } |
2111 | ||
c0ff4b85 | 2112 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2113 | { |
2114 | struct mem_cgroup *iter; | |
2115 | ||
c0ff4b85 | 2116 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
2117 | atomic_inc(&iter->under_oom); |
2118 | } | |
2119 | ||
c0ff4b85 | 2120 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2121 | { |
2122 | struct mem_cgroup *iter; | |
2123 | ||
867578cb KH |
2124 | /* |
2125 | * When a new child is created while the hierarchy is under oom, | |
2126 | * mem_cgroup_oom_lock() may not be called. We have to use | |
2127 | * atomic_add_unless() here. | |
2128 | */ | |
c0ff4b85 | 2129 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 2130 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
2131 | } |
2132 | ||
867578cb KH |
2133 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
2134 | ||
dc98df5a | 2135 | struct oom_wait_info { |
d79154bb | 2136 | struct mem_cgroup *memcg; |
dc98df5a KH |
2137 | wait_queue_t wait; |
2138 | }; | |
2139 | ||
2140 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
2141 | unsigned mode, int sync, void *arg) | |
2142 | { | |
d79154bb HD |
2143 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
2144 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
2145 | struct oom_wait_info *oom_wait_info; |
2146 | ||
2147 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 2148 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 2149 | |
dc98df5a | 2150 | /* |
d79154bb | 2151 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
dc98df5a KH |
2152 | * Then we can use css_is_ancestor without taking care of RCU. |
2153 | */ | |
c0ff4b85 R |
2154 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
2155 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 2156 | return 0; |
dc98df5a KH |
2157 | return autoremove_wake_function(wait, mode, sync, arg); |
2158 | } | |
2159 | ||
c0ff4b85 | 2160 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 2161 | { |
3812c8c8 | 2162 | atomic_inc(&memcg->oom_wakeups); |
c0ff4b85 R |
2163 | /* for filtering, pass "memcg" as argument. */ |
2164 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
2165 | } |
2166 | ||
c0ff4b85 | 2167 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 2168 | { |
c0ff4b85 R |
2169 | if (memcg && atomic_read(&memcg->under_oom)) |
2170 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
2171 | } |
2172 | ||
3812c8c8 | 2173 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 2174 | { |
3812c8c8 JW |
2175 | if (!current->memcg_oom.may_oom) |
2176 | return; | |
867578cb | 2177 | /* |
49426420 JW |
2178 | * We are in the middle of the charge context here, so we |
2179 | * don't want to block when potentially sitting on a callstack | |
2180 | * that holds all kinds of filesystem and mm locks. | |
2181 | * | |
2182 | * Also, the caller may handle a failed allocation gracefully | |
2183 | * (like optional page cache readahead) and so an OOM killer | |
2184 | * invocation might not even be necessary. | |
2185 | * | |
2186 | * That's why we don't do anything here except remember the | |
2187 | * OOM context and then deal with it at the end of the page | |
2188 | * fault when the stack is unwound, the locks are released, | |
2189 | * and when we know whether the fault was overall successful. | |
867578cb | 2190 | */ |
49426420 JW |
2191 | css_get(&memcg->css); |
2192 | current->memcg_oom.memcg = memcg; | |
2193 | current->memcg_oom.gfp_mask = mask; | |
2194 | current->memcg_oom.order = order; | |
3812c8c8 JW |
2195 | } |
2196 | ||
2197 | /** | |
2198 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 2199 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 2200 | * |
49426420 JW |
2201 | * This has to be called at the end of a page fault if the memcg OOM |
2202 | * handler was enabled. | |
3812c8c8 | 2203 | * |
49426420 | 2204 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
2205 | * sleep on a waitqueue until the userspace task resolves the |
2206 | * situation. Sleeping directly in the charge context with all kinds | |
2207 | * of locks held is not a good idea, instead we remember an OOM state | |
2208 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 2209 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
2210 | * |
2211 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 2212 | * completed, %false otherwise. |
3812c8c8 | 2213 | */ |
49426420 | 2214 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 2215 | { |
49426420 | 2216 | struct mem_cgroup *memcg = current->memcg_oom.memcg; |
3812c8c8 | 2217 | struct oom_wait_info owait; |
49426420 | 2218 | bool locked; |
3812c8c8 JW |
2219 | |
2220 | /* OOM is global, do not handle */ | |
3812c8c8 | 2221 | if (!memcg) |
49426420 | 2222 | return false; |
3812c8c8 | 2223 | |
49426420 JW |
2224 | if (!handle) |
2225 | goto cleanup; | |
3812c8c8 JW |
2226 | |
2227 | owait.memcg = memcg; | |
2228 | owait.wait.flags = 0; | |
2229 | owait.wait.func = memcg_oom_wake_function; | |
2230 | owait.wait.private = current; | |
2231 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 2232 | |
3812c8c8 | 2233 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
2234 | mem_cgroup_mark_under_oom(memcg); |
2235 | ||
2236 | locked = mem_cgroup_oom_trylock(memcg); | |
2237 | ||
2238 | if (locked) | |
2239 | mem_cgroup_oom_notify(memcg); | |
2240 | ||
2241 | if (locked && !memcg->oom_kill_disable) { | |
2242 | mem_cgroup_unmark_under_oom(memcg); | |
2243 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2244 | mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask, | |
2245 | current->memcg_oom.order); | |
2246 | } else { | |
3812c8c8 | 2247 | schedule(); |
49426420 JW |
2248 | mem_cgroup_unmark_under_oom(memcg); |
2249 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2250 | } | |
2251 | ||
2252 | if (locked) { | |
fb2a6fc5 JW |
2253 | mem_cgroup_oom_unlock(memcg); |
2254 | /* | |
2255 | * There is no guarantee that an OOM-lock contender | |
2256 | * sees the wakeups triggered by the OOM kill | |
2257 | * uncharges. Wake any sleepers explicitely. | |
2258 | */ | |
2259 | memcg_oom_recover(memcg); | |
2260 | } | |
49426420 JW |
2261 | cleanup: |
2262 | current->memcg_oom.memcg = NULL; | |
3812c8c8 | 2263 | css_put(&memcg->css); |
867578cb | 2264 | return true; |
0b7f569e KH |
2265 | } |
2266 | ||
d69b042f BS |
2267 | /* |
2268 | * Currently used to update mapped file statistics, but the routine can be | |
2269 | * generalized to update other statistics as well. | |
32047e2a KH |
2270 | * |
2271 | * Notes: Race condition | |
2272 | * | |
2273 | * We usually use page_cgroup_lock() for accessing page_cgroup member but | |
2274 | * it tends to be costly. But considering some conditions, we doesn't need | |
2275 | * to do so _always_. | |
2276 | * | |
2277 | * Considering "charge", lock_page_cgroup() is not required because all | |
2278 | * file-stat operations happen after a page is attached to radix-tree. There | |
2279 | * are no race with "charge". | |
2280 | * | |
2281 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
2282 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
2283 | * if there are race with "uncharge". Statistics itself is properly handled | |
2284 | * by flags. | |
2285 | * | |
2286 | * Considering "move", this is an only case we see a race. To make the race | |
619d094b KH |
2287 | * small, we check mm->moving_account and detect there are possibility of race |
2288 | * If there is, we take a lock. | |
d69b042f | 2289 | */ |
26174efd | 2290 | |
89c06bd5 KH |
2291 | void __mem_cgroup_begin_update_page_stat(struct page *page, |
2292 | bool *locked, unsigned long *flags) | |
2293 | { | |
2294 | struct mem_cgroup *memcg; | |
2295 | struct page_cgroup *pc; | |
2296 | ||
2297 | pc = lookup_page_cgroup(page); | |
2298 | again: | |
2299 | memcg = pc->mem_cgroup; | |
2300 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
2301 | return; | |
2302 | /* | |
2303 | * If this memory cgroup is not under account moving, we don't | |
da92c47d | 2304 | * need to take move_lock_mem_cgroup(). Because we already hold |
89c06bd5 | 2305 | * rcu_read_lock(), any calls to move_account will be delayed until |
13fd1dd9 | 2306 | * rcu_read_unlock() if mem_cgroup_stolen() == true. |
89c06bd5 | 2307 | */ |
13fd1dd9 | 2308 | if (!mem_cgroup_stolen(memcg)) |
89c06bd5 KH |
2309 | return; |
2310 | ||
2311 | move_lock_mem_cgroup(memcg, flags); | |
2312 | if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) { | |
2313 | move_unlock_mem_cgroup(memcg, flags); | |
2314 | goto again; | |
2315 | } | |
2316 | *locked = true; | |
2317 | } | |
2318 | ||
2319 | void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags) | |
2320 | { | |
2321 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
2322 | ||
2323 | /* | |
2324 | * It's guaranteed that pc->mem_cgroup never changes while | |
2325 | * lock is held because a routine modifies pc->mem_cgroup | |
da92c47d | 2326 | * should take move_lock_mem_cgroup(). |
89c06bd5 KH |
2327 | */ |
2328 | move_unlock_mem_cgroup(pc->mem_cgroup, flags); | |
2329 | } | |
2330 | ||
2a7106f2 | 2331 | void mem_cgroup_update_page_stat(struct page *page, |
68b4876d | 2332 | enum mem_cgroup_stat_index idx, int val) |
d69b042f | 2333 | { |
c0ff4b85 | 2334 | struct mem_cgroup *memcg; |
32047e2a | 2335 | struct page_cgroup *pc = lookup_page_cgroup(page); |
dbd4ea78 | 2336 | unsigned long uninitialized_var(flags); |
d69b042f | 2337 | |
cfa44946 | 2338 | if (mem_cgroup_disabled()) |
d69b042f | 2339 | return; |
89c06bd5 | 2340 | |
658b72c5 | 2341 | VM_BUG_ON(!rcu_read_lock_held()); |
c0ff4b85 R |
2342 | memcg = pc->mem_cgroup; |
2343 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
89c06bd5 | 2344 | return; |
26174efd | 2345 | |
c0ff4b85 | 2346 | this_cpu_add(memcg->stat->count[idx], val); |
d69b042f | 2347 | } |
26174efd | 2348 | |
cdec2e42 KH |
2349 | /* |
2350 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2351 | * TODO: maybe necessary to use big numbers in big irons. | |
2352 | */ | |
7ec99d62 | 2353 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2354 | struct memcg_stock_pcp { |
2355 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2356 | unsigned int nr_pages; |
cdec2e42 | 2357 | struct work_struct work; |
26fe6168 | 2358 | unsigned long flags; |
a0db00fc | 2359 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2360 | }; |
2361 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2362 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2363 | |
a0956d54 SS |
2364 | /** |
2365 | * consume_stock: Try to consume stocked charge on this cpu. | |
2366 | * @memcg: memcg to consume from. | |
2367 | * @nr_pages: how many pages to charge. | |
2368 | * | |
2369 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2370 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2371 | * service an allocation will refill the stock. | |
2372 | * | |
2373 | * returns true if successful, false otherwise. | |
cdec2e42 | 2374 | */ |
a0956d54 | 2375 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2376 | { |
2377 | struct memcg_stock_pcp *stock; | |
2378 | bool ret = true; | |
2379 | ||
a0956d54 SS |
2380 | if (nr_pages > CHARGE_BATCH) |
2381 | return false; | |
2382 | ||
cdec2e42 | 2383 | stock = &get_cpu_var(memcg_stock); |
a0956d54 SS |
2384 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) |
2385 | stock->nr_pages -= nr_pages; | |
cdec2e42 KH |
2386 | else /* need to call res_counter_charge */ |
2387 | ret = false; | |
2388 | put_cpu_var(memcg_stock); | |
2389 | return ret; | |
2390 | } | |
2391 | ||
2392 | /* | |
2393 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
2394 | */ | |
2395 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2396 | { | |
2397 | struct mem_cgroup *old = stock->cached; | |
2398 | ||
11c9ea4e JW |
2399 | if (stock->nr_pages) { |
2400 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
2401 | ||
2402 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 2403 | if (do_swap_account) |
11c9ea4e JW |
2404 | res_counter_uncharge(&old->memsw, bytes); |
2405 | stock->nr_pages = 0; | |
cdec2e42 KH |
2406 | } |
2407 | stock->cached = NULL; | |
cdec2e42 KH |
2408 | } |
2409 | ||
2410 | /* | |
2411 | * This must be called under preempt disabled or must be called by | |
2412 | * a thread which is pinned to local cpu. | |
2413 | */ | |
2414 | static void drain_local_stock(struct work_struct *dummy) | |
2415 | { | |
2416 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
2417 | drain_stock(stock); | |
26fe6168 | 2418 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2419 | } |
2420 | ||
e4777496 MH |
2421 | static void __init memcg_stock_init(void) |
2422 | { | |
2423 | int cpu; | |
2424 | ||
2425 | for_each_possible_cpu(cpu) { | |
2426 | struct memcg_stock_pcp *stock = | |
2427 | &per_cpu(memcg_stock, cpu); | |
2428 | INIT_WORK(&stock->work, drain_local_stock); | |
2429 | } | |
2430 | } | |
2431 | ||
cdec2e42 KH |
2432 | /* |
2433 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 2434 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2435 | */ |
c0ff4b85 | 2436 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2437 | { |
2438 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2439 | ||
c0ff4b85 | 2440 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2441 | drain_stock(stock); |
c0ff4b85 | 2442 | stock->cached = memcg; |
cdec2e42 | 2443 | } |
11c9ea4e | 2444 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2445 | put_cpu_var(memcg_stock); |
2446 | } | |
2447 | ||
2448 | /* | |
c0ff4b85 | 2449 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
d38144b7 MH |
2450 | * of the hierarchy under it. sync flag says whether we should block |
2451 | * until the work is done. | |
cdec2e42 | 2452 | */ |
c0ff4b85 | 2453 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
cdec2e42 | 2454 | { |
26fe6168 | 2455 | int cpu, curcpu; |
d38144b7 | 2456 | |
cdec2e42 | 2457 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2458 | get_online_cpus(); |
5af12d0e | 2459 | curcpu = get_cpu(); |
cdec2e42 KH |
2460 | for_each_online_cpu(cpu) { |
2461 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2462 | struct mem_cgroup *memcg; |
26fe6168 | 2463 | |
c0ff4b85 R |
2464 | memcg = stock->cached; |
2465 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2466 | continue; |
c0ff4b85 | 2467 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2468 | continue; |
d1a05b69 MH |
2469 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2470 | if (cpu == curcpu) | |
2471 | drain_local_stock(&stock->work); | |
2472 | else | |
2473 | schedule_work_on(cpu, &stock->work); | |
2474 | } | |
cdec2e42 | 2475 | } |
5af12d0e | 2476 | put_cpu(); |
d38144b7 MH |
2477 | |
2478 | if (!sync) | |
2479 | goto out; | |
2480 | ||
2481 | for_each_online_cpu(cpu) { | |
2482 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2483 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2484 | flush_work(&stock->work); |
2485 | } | |
2486 | out: | |
f894ffa8 | 2487 | put_online_cpus(); |
d38144b7 MH |
2488 | } |
2489 | ||
2490 | /* | |
2491 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2492 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
2493 | * expects some charges will be back to res_counter later but cannot wait for | |
2494 | * it. | |
2495 | */ | |
c0ff4b85 | 2496 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
d38144b7 | 2497 | { |
9f50fad6 MH |
2498 | /* |
2499 | * If someone calls draining, avoid adding more kworker runs. | |
2500 | */ | |
2501 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2502 | return; | |
c0ff4b85 | 2503 | drain_all_stock(root_memcg, false); |
9f50fad6 | 2504 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2505 | } |
2506 | ||
2507 | /* This is a synchronous drain interface. */ | |
c0ff4b85 | 2508 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
cdec2e42 KH |
2509 | { |
2510 | /* called when force_empty is called */ | |
9f50fad6 | 2511 | mutex_lock(&percpu_charge_mutex); |
c0ff4b85 | 2512 | drain_all_stock(root_memcg, true); |
9f50fad6 | 2513 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2514 | } |
2515 | ||
711d3d2c KH |
2516 | /* |
2517 | * This function drains percpu counter value from DEAD cpu and | |
2518 | * move it to local cpu. Note that this function can be preempted. | |
2519 | */ | |
c0ff4b85 | 2520 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2521 | { |
2522 | int i; | |
2523 | ||
c0ff4b85 | 2524 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2525 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2526 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2527 | |
c0ff4b85 R |
2528 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2529 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2530 | } |
e9f8974f | 2531 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2532 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2533 | |
c0ff4b85 R |
2534 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2535 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2536 | } |
c0ff4b85 | 2537 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2538 | } |
2539 | ||
0db0628d | 2540 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
2541 | unsigned long action, |
2542 | void *hcpu) | |
2543 | { | |
2544 | int cpu = (unsigned long)hcpu; | |
2545 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2546 | struct mem_cgroup *iter; |
cdec2e42 | 2547 | |
619d094b | 2548 | if (action == CPU_ONLINE) |
1489ebad | 2549 | return NOTIFY_OK; |
1489ebad | 2550 | |
d833049b | 2551 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2552 | return NOTIFY_OK; |
711d3d2c | 2553 | |
9f3a0d09 | 2554 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2555 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2556 | ||
cdec2e42 KH |
2557 | stock = &per_cpu(memcg_stock, cpu); |
2558 | drain_stock(stock); | |
2559 | return NOTIFY_OK; | |
2560 | } | |
2561 | ||
4b534334 KH |
2562 | |
2563 | /* See __mem_cgroup_try_charge() for details */ | |
2564 | enum { | |
2565 | CHARGE_OK, /* success */ | |
2566 | CHARGE_RETRY, /* need to retry but retry is not bad */ | |
2567 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ | |
2568 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ | |
4b534334 KH |
2569 | }; |
2570 | ||
c0ff4b85 | 2571 | static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
4c9c5359 | 2572 | unsigned int nr_pages, unsigned int min_pages, |
3812c8c8 | 2573 | bool invoke_oom) |
4b534334 | 2574 | { |
7ec99d62 | 2575 | unsigned long csize = nr_pages * PAGE_SIZE; |
4b534334 KH |
2576 | struct mem_cgroup *mem_over_limit; |
2577 | struct res_counter *fail_res; | |
2578 | unsigned long flags = 0; | |
2579 | int ret; | |
2580 | ||
c0ff4b85 | 2581 | ret = res_counter_charge(&memcg->res, csize, &fail_res); |
4b534334 KH |
2582 | |
2583 | if (likely(!ret)) { | |
2584 | if (!do_swap_account) | |
2585 | return CHARGE_OK; | |
c0ff4b85 | 2586 | ret = res_counter_charge(&memcg->memsw, csize, &fail_res); |
4b534334 KH |
2587 | if (likely(!ret)) |
2588 | return CHARGE_OK; | |
2589 | ||
c0ff4b85 | 2590 | res_counter_uncharge(&memcg->res, csize); |
4b534334 KH |
2591 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
2592 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
2593 | } else | |
2594 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
9221edb7 | 2595 | /* |
9221edb7 JW |
2596 | * Never reclaim on behalf of optional batching, retry with a |
2597 | * single page instead. | |
2598 | */ | |
4c9c5359 | 2599 | if (nr_pages > min_pages) |
4b534334 KH |
2600 | return CHARGE_RETRY; |
2601 | ||
2602 | if (!(gfp_mask & __GFP_WAIT)) | |
2603 | return CHARGE_WOULDBLOCK; | |
2604 | ||
4c9c5359 SS |
2605 | if (gfp_mask & __GFP_NORETRY) |
2606 | return CHARGE_NOMEM; | |
2607 | ||
5660048c | 2608 | ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags); |
7ec99d62 | 2609 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
19942822 | 2610 | return CHARGE_RETRY; |
4b534334 | 2611 | /* |
19942822 JW |
2612 | * Even though the limit is exceeded at this point, reclaim |
2613 | * may have been able to free some pages. Retry the charge | |
2614 | * before killing the task. | |
2615 | * | |
2616 | * Only for regular pages, though: huge pages are rather | |
2617 | * unlikely to succeed so close to the limit, and we fall back | |
2618 | * to regular pages anyway in case of failure. | |
4b534334 | 2619 | */ |
4c9c5359 | 2620 | if (nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER) && ret) |
4b534334 KH |
2621 | return CHARGE_RETRY; |
2622 | ||
2623 | /* | |
2624 | * At task move, charge accounts can be doubly counted. So, it's | |
2625 | * better to wait until the end of task_move if something is going on. | |
2626 | */ | |
2627 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2628 | return CHARGE_RETRY; | |
2629 | ||
3812c8c8 JW |
2630 | if (invoke_oom) |
2631 | mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(csize)); | |
4b534334 | 2632 | |
3812c8c8 | 2633 | return CHARGE_NOMEM; |
4b534334 KH |
2634 | } |
2635 | ||
f817ed48 | 2636 | /* |
38c5d72f KH |
2637 | * __mem_cgroup_try_charge() does |
2638 | * 1. detect memcg to be charged against from passed *mm and *ptr, | |
2639 | * 2. update res_counter | |
2640 | * 3. call memory reclaim if necessary. | |
2641 | * | |
2642 | * In some special case, if the task is fatal, fatal_signal_pending() or | |
2643 | * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup | |
2644 | * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon | |
2645 | * as possible without any hazards. 2: all pages should have a valid | |
2646 | * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg | |
2647 | * pointer, that is treated as a charge to root_mem_cgroup. | |
2648 | * | |
2649 | * So __mem_cgroup_try_charge() will return | |
2650 | * 0 ... on success, filling *ptr with a valid memcg pointer. | |
2651 | * -ENOMEM ... charge failure because of resource limits. | |
2652 | * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup. | |
2653 | * | |
2654 | * Unlike the exported interface, an "oom" parameter is added. if oom==true, | |
2655 | * the oom-killer can be invoked. | |
8a9f3ccd | 2656 | */ |
f817ed48 | 2657 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
ec168510 | 2658 | gfp_t gfp_mask, |
7ec99d62 | 2659 | unsigned int nr_pages, |
c0ff4b85 | 2660 | struct mem_cgroup **ptr, |
7ec99d62 | 2661 | bool oom) |
8a9f3ccd | 2662 | { |
7ec99d62 | 2663 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
4b534334 | 2664 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
c0ff4b85 | 2665 | struct mem_cgroup *memcg = NULL; |
4b534334 | 2666 | int ret; |
a636b327 | 2667 | |
867578cb KH |
2668 | /* |
2669 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
2670 | * in system level. So, allow to go ahead dying process in addition to | |
2671 | * MEMDIE process. | |
2672 | */ | |
2673 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
2674 | || fatal_signal_pending(current))) | |
2675 | goto bypass; | |
a636b327 | 2676 | |
49426420 JW |
2677 | if (unlikely(task_in_memcg_oom(current))) |
2678 | goto bypass; | |
2679 | ||
8a9f3ccd | 2680 | /* |
3be91277 HD |
2681 | * We always charge the cgroup the mm_struct belongs to. |
2682 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd | 2683 | * thread group leader migrates. It's possible that mm is not |
24467cac | 2684 | * set, if so charge the root memcg (happens for pagecache usage). |
8a9f3ccd | 2685 | */ |
c0ff4b85 | 2686 | if (!*ptr && !mm) |
38c5d72f | 2687 | *ptr = root_mem_cgroup; |
f75ca962 | 2688 | again: |
c0ff4b85 R |
2689 | if (*ptr) { /* css should be a valid one */ |
2690 | memcg = *ptr; | |
c0ff4b85 | 2691 | if (mem_cgroup_is_root(memcg)) |
f75ca962 | 2692 | goto done; |
a0956d54 | 2693 | if (consume_stock(memcg, nr_pages)) |
f75ca962 | 2694 | goto done; |
c0ff4b85 | 2695 | css_get(&memcg->css); |
4b534334 | 2696 | } else { |
f75ca962 | 2697 | struct task_struct *p; |
54595fe2 | 2698 | |
f75ca962 KH |
2699 | rcu_read_lock(); |
2700 | p = rcu_dereference(mm->owner); | |
f75ca962 | 2701 | /* |
ebb76ce1 | 2702 | * Because we don't have task_lock(), "p" can exit. |
c0ff4b85 | 2703 | * In that case, "memcg" can point to root or p can be NULL with |
ebb76ce1 KH |
2704 | * race with swapoff. Then, we have small risk of mis-accouning. |
2705 | * But such kind of mis-account by race always happens because | |
2706 | * we don't have cgroup_mutex(). It's overkill and we allo that | |
2707 | * small race, here. | |
2708 | * (*) swapoff at el will charge against mm-struct not against | |
2709 | * task-struct. So, mm->owner can be NULL. | |
f75ca962 | 2710 | */ |
c0ff4b85 | 2711 | memcg = mem_cgroup_from_task(p); |
38c5d72f KH |
2712 | if (!memcg) |
2713 | memcg = root_mem_cgroup; | |
2714 | if (mem_cgroup_is_root(memcg)) { | |
f75ca962 KH |
2715 | rcu_read_unlock(); |
2716 | goto done; | |
2717 | } | |
a0956d54 | 2718 | if (consume_stock(memcg, nr_pages)) { |
f75ca962 KH |
2719 | /* |
2720 | * It seems dagerous to access memcg without css_get(). | |
2721 | * But considering how consume_stok works, it's not | |
2722 | * necessary. If consume_stock success, some charges | |
2723 | * from this memcg are cached on this cpu. So, we | |
2724 | * don't need to call css_get()/css_tryget() before | |
2725 | * calling consume_stock(). | |
2726 | */ | |
2727 | rcu_read_unlock(); | |
2728 | goto done; | |
2729 | } | |
2730 | /* after here, we may be blocked. we need to get refcnt */ | |
c0ff4b85 | 2731 | if (!css_tryget(&memcg->css)) { |
f75ca962 KH |
2732 | rcu_read_unlock(); |
2733 | goto again; | |
2734 | } | |
2735 | rcu_read_unlock(); | |
2736 | } | |
8a9f3ccd | 2737 | |
4b534334 | 2738 | do { |
3812c8c8 | 2739 | bool invoke_oom = oom && !nr_oom_retries; |
7a81b88c | 2740 | |
4b534334 | 2741 | /* If killed, bypass charge */ |
f75ca962 | 2742 | if (fatal_signal_pending(current)) { |
c0ff4b85 | 2743 | css_put(&memcg->css); |
4b534334 | 2744 | goto bypass; |
f75ca962 | 2745 | } |
6d61ef40 | 2746 | |
3812c8c8 JW |
2747 | ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, |
2748 | nr_pages, invoke_oom); | |
4b534334 KH |
2749 | switch (ret) { |
2750 | case CHARGE_OK: | |
2751 | break; | |
2752 | case CHARGE_RETRY: /* not in OOM situation but retry */ | |
7ec99d62 | 2753 | batch = nr_pages; |
c0ff4b85 R |
2754 | css_put(&memcg->css); |
2755 | memcg = NULL; | |
f75ca962 | 2756 | goto again; |
4b534334 | 2757 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
c0ff4b85 | 2758 | css_put(&memcg->css); |
4b534334 KH |
2759 | goto nomem; |
2760 | case CHARGE_NOMEM: /* OOM routine works */ | |
3812c8c8 | 2761 | if (!oom || invoke_oom) { |
c0ff4b85 | 2762 | css_put(&memcg->css); |
867578cb | 2763 | goto nomem; |
f75ca962 | 2764 | } |
4b534334 KH |
2765 | nr_oom_retries--; |
2766 | break; | |
66e1707b | 2767 | } |
4b534334 KH |
2768 | } while (ret != CHARGE_OK); |
2769 | ||
7ec99d62 | 2770 | if (batch > nr_pages) |
c0ff4b85 R |
2771 | refill_stock(memcg, batch - nr_pages); |
2772 | css_put(&memcg->css); | |
0c3e73e8 | 2773 | done: |
c0ff4b85 | 2774 | *ptr = memcg; |
7a81b88c KH |
2775 | return 0; |
2776 | nomem: | |
3168ecbe JW |
2777 | if (!(gfp_mask & __GFP_NOFAIL)) { |
2778 | *ptr = NULL; | |
2779 | return -ENOMEM; | |
2780 | } | |
867578cb | 2781 | bypass: |
38c5d72f KH |
2782 | *ptr = root_mem_cgroup; |
2783 | return -EINTR; | |
7a81b88c | 2784 | } |
8a9f3ccd | 2785 | |
a3032a2c DN |
2786 | /* |
2787 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2788 | * This function is for that and do uncharge, put css's refcnt. | |
2789 | * gotten by try_charge(). | |
2790 | */ | |
c0ff4b85 | 2791 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg, |
e7018b8d | 2792 | unsigned int nr_pages) |
a3032a2c | 2793 | { |
c0ff4b85 | 2794 | if (!mem_cgroup_is_root(memcg)) { |
e7018b8d JW |
2795 | unsigned long bytes = nr_pages * PAGE_SIZE; |
2796 | ||
c0ff4b85 | 2797 | res_counter_uncharge(&memcg->res, bytes); |
a3032a2c | 2798 | if (do_swap_account) |
c0ff4b85 | 2799 | res_counter_uncharge(&memcg->memsw, bytes); |
a3032a2c | 2800 | } |
854ffa8d DN |
2801 | } |
2802 | ||
d01dd17f KH |
2803 | /* |
2804 | * Cancel chrages in this cgroup....doesn't propagate to parent cgroup. | |
2805 | * This is useful when moving usage to parent cgroup. | |
2806 | */ | |
2807 | static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg, | |
2808 | unsigned int nr_pages) | |
2809 | { | |
2810 | unsigned long bytes = nr_pages * PAGE_SIZE; | |
2811 | ||
2812 | if (mem_cgroup_is_root(memcg)) | |
2813 | return; | |
2814 | ||
2815 | res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes); | |
2816 | if (do_swap_account) | |
2817 | res_counter_uncharge_until(&memcg->memsw, | |
2818 | memcg->memsw.parent, bytes); | |
2819 | } | |
2820 | ||
a3b2d692 KH |
2821 | /* |
2822 | * A helper function to get mem_cgroup from ID. must be called under | |
e9316080 TH |
2823 | * rcu_read_lock(). The caller is responsible for calling css_tryget if |
2824 | * the mem_cgroup is used for charging. (dropping refcnt from swap can be | |
2825 | * called against removed memcg.) | |
a3b2d692 KH |
2826 | */ |
2827 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2828 | { | |
2829 | struct cgroup_subsys_state *css; | |
2830 | ||
2831 | /* ID 0 is unused ID */ | |
2832 | if (!id) | |
2833 | return NULL; | |
2834 | css = css_lookup(&mem_cgroup_subsys, id); | |
2835 | if (!css) | |
2836 | return NULL; | |
b2145145 | 2837 | return mem_cgroup_from_css(css); |
a3b2d692 KH |
2838 | } |
2839 | ||
e42d9d5d | 2840 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2841 | { |
c0ff4b85 | 2842 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2843 | struct page_cgroup *pc; |
a3b2d692 | 2844 | unsigned short id; |
b5a84319 KH |
2845 | swp_entry_t ent; |
2846 | ||
3c776e64 DN |
2847 | VM_BUG_ON(!PageLocked(page)); |
2848 | ||
3c776e64 | 2849 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2850 | lock_page_cgroup(pc); |
a3b2d692 | 2851 | if (PageCgroupUsed(pc)) { |
c0ff4b85 R |
2852 | memcg = pc->mem_cgroup; |
2853 | if (memcg && !css_tryget(&memcg->css)) | |
2854 | memcg = NULL; | |
e42d9d5d | 2855 | } else if (PageSwapCache(page)) { |
3c776e64 | 2856 | ent.val = page_private(page); |
9fb4b7cc | 2857 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2858 | rcu_read_lock(); |
c0ff4b85 R |
2859 | memcg = mem_cgroup_lookup(id); |
2860 | if (memcg && !css_tryget(&memcg->css)) | |
2861 | memcg = NULL; | |
a3b2d692 | 2862 | rcu_read_unlock(); |
3c776e64 | 2863 | } |
c0bd3f63 | 2864 | unlock_page_cgroup(pc); |
c0ff4b85 | 2865 | return memcg; |
b5a84319 KH |
2866 | } |
2867 | ||
c0ff4b85 | 2868 | static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg, |
5564e88b | 2869 | struct page *page, |
7ec99d62 | 2870 | unsigned int nr_pages, |
9ce70c02 HD |
2871 | enum charge_type ctype, |
2872 | bool lrucare) | |
7a81b88c | 2873 | { |
ce587e65 | 2874 | struct page_cgroup *pc = lookup_page_cgroup(page); |
9ce70c02 | 2875 | struct zone *uninitialized_var(zone); |
fa9add64 | 2876 | struct lruvec *lruvec; |
9ce70c02 | 2877 | bool was_on_lru = false; |
b2402857 | 2878 | bool anon; |
9ce70c02 | 2879 | |
ca3e0214 | 2880 | lock_page_cgroup(pc); |
90deb788 | 2881 | VM_BUG_ON(PageCgroupUsed(pc)); |
ca3e0214 KH |
2882 | /* |
2883 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2884 | * accessed by any other context at this point. | |
2885 | */ | |
9ce70c02 HD |
2886 | |
2887 | /* | |
2888 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2889 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2890 | */ | |
2891 | if (lrucare) { | |
2892 | zone = page_zone(page); | |
2893 | spin_lock_irq(&zone->lru_lock); | |
2894 | if (PageLRU(page)) { | |
fa9add64 | 2895 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 | 2896 | ClearPageLRU(page); |
fa9add64 | 2897 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2898 | was_on_lru = true; |
2899 | } | |
2900 | } | |
2901 | ||
c0ff4b85 | 2902 | pc->mem_cgroup = memcg; |
261fb61a KH |
2903 | /* |
2904 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
2905 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
2906 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
2907 | * before USED bit, we need memory barrier here. | |
2908 | * See mem_cgroup_add_lru_list(), etc. | |
f894ffa8 | 2909 | */ |
08e552c6 | 2910 | smp_wmb(); |
b2402857 | 2911 | SetPageCgroupUsed(pc); |
3be91277 | 2912 | |
9ce70c02 HD |
2913 | if (lrucare) { |
2914 | if (was_on_lru) { | |
fa9add64 | 2915 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 HD |
2916 | VM_BUG_ON(PageLRU(page)); |
2917 | SetPageLRU(page); | |
fa9add64 | 2918 | add_page_to_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2919 | } |
2920 | spin_unlock_irq(&zone->lru_lock); | |
2921 | } | |
2922 | ||
41326c17 | 2923 | if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON) |
b2402857 KH |
2924 | anon = true; |
2925 | else | |
2926 | anon = false; | |
2927 | ||
b070e65c | 2928 | mem_cgroup_charge_statistics(memcg, page, anon, nr_pages); |
52d4b9ac | 2929 | unlock_page_cgroup(pc); |
9ce70c02 | 2930 | |
430e4863 | 2931 | /* |
bb4cc1a8 AM |
2932 | * "charge_statistics" updated event counter. Then, check it. |
2933 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
2934 | * if they exceeds softlimit. | |
430e4863 | 2935 | */ |
c0ff4b85 | 2936 | memcg_check_events(memcg, page); |
7a81b88c | 2937 | } |
66e1707b | 2938 | |
7cf27982 GC |
2939 | static DEFINE_MUTEX(set_limit_mutex); |
2940 | ||
7ae1e1d0 GC |
2941 | #ifdef CONFIG_MEMCG_KMEM |
2942 | static inline bool memcg_can_account_kmem(struct mem_cgroup *memcg) | |
2943 | { | |
2944 | return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg) && | |
2945 | (memcg->kmem_account_flags & KMEM_ACCOUNTED_MASK); | |
2946 | } | |
2947 | ||
1f458cbf GC |
2948 | /* |
2949 | * This is a bit cumbersome, but it is rarely used and avoids a backpointer | |
2950 | * in the memcg_cache_params struct. | |
2951 | */ | |
2952 | static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p) | |
2953 | { | |
2954 | struct kmem_cache *cachep; | |
2955 | ||
2956 | VM_BUG_ON(p->is_root_cache); | |
2957 | cachep = p->root_cache; | |
2958 | return cachep->memcg_params->memcg_caches[memcg_cache_id(p->memcg)]; | |
2959 | } | |
2960 | ||
749c5415 | 2961 | #ifdef CONFIG_SLABINFO |
182446d0 TH |
2962 | static int mem_cgroup_slabinfo_read(struct cgroup_subsys_state *css, |
2963 | struct cftype *cft, struct seq_file *m) | |
749c5415 | 2964 | { |
182446d0 | 2965 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
749c5415 GC |
2966 | struct memcg_cache_params *params; |
2967 | ||
2968 | if (!memcg_can_account_kmem(memcg)) | |
2969 | return -EIO; | |
2970 | ||
2971 | print_slabinfo_header(m); | |
2972 | ||
2973 | mutex_lock(&memcg->slab_caches_mutex); | |
2974 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) | |
2975 | cache_show(memcg_params_to_cache(params), m); | |
2976 | mutex_unlock(&memcg->slab_caches_mutex); | |
2977 | ||
2978 | return 0; | |
2979 | } | |
2980 | #endif | |
2981 | ||
7ae1e1d0 GC |
2982 | static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, u64 size) |
2983 | { | |
2984 | struct res_counter *fail_res; | |
2985 | struct mem_cgroup *_memcg; | |
2986 | int ret = 0; | |
7ae1e1d0 GC |
2987 | |
2988 | ret = res_counter_charge(&memcg->kmem, size, &fail_res); | |
2989 | if (ret) | |
2990 | return ret; | |
2991 | ||
7ae1e1d0 GC |
2992 | _memcg = memcg; |
2993 | ret = __mem_cgroup_try_charge(NULL, gfp, size >> PAGE_SHIFT, | |
b9921ecd | 2994 | &_memcg, oom_gfp_allowed(gfp)); |
7ae1e1d0 GC |
2995 | |
2996 | if (ret == -EINTR) { | |
2997 | /* | |
2998 | * __mem_cgroup_try_charge() chosed to bypass to root due to | |
2999 | * OOM kill or fatal signal. Since our only options are to | |
3000 | * either fail the allocation or charge it to this cgroup, do | |
3001 | * it as a temporary condition. But we can't fail. From a | |
3002 | * kmem/slab perspective, the cache has already been selected, | |
3003 | * by mem_cgroup_kmem_get_cache(), so it is too late to change | |
3004 | * our minds. | |
3005 | * | |
3006 | * This condition will only trigger if the task entered | |
3007 | * memcg_charge_kmem in a sane state, but was OOM-killed during | |
3008 | * __mem_cgroup_try_charge() above. Tasks that were already | |
3009 | * dying when the allocation triggers should have been already | |
3010 | * directed to the root cgroup in memcontrol.h | |
3011 | */ | |
3012 | res_counter_charge_nofail(&memcg->res, size, &fail_res); | |
3013 | if (do_swap_account) | |
3014 | res_counter_charge_nofail(&memcg->memsw, size, | |
3015 | &fail_res); | |
3016 | ret = 0; | |
3017 | } else if (ret) | |
3018 | res_counter_uncharge(&memcg->kmem, size); | |
3019 | ||
3020 | return ret; | |
3021 | } | |
3022 | ||
3023 | static void memcg_uncharge_kmem(struct mem_cgroup *memcg, u64 size) | |
3024 | { | |
7ae1e1d0 GC |
3025 | res_counter_uncharge(&memcg->res, size); |
3026 | if (do_swap_account) | |
3027 | res_counter_uncharge(&memcg->memsw, size); | |
7de37682 GC |
3028 | |
3029 | /* Not down to 0 */ | |
3030 | if (res_counter_uncharge(&memcg->kmem, size)) | |
3031 | return; | |
3032 | ||
10d5ebf4 LZ |
3033 | /* |
3034 | * Releases a reference taken in kmem_cgroup_css_offline in case | |
3035 | * this last uncharge is racing with the offlining code or it is | |
3036 | * outliving the memcg existence. | |
3037 | * | |
3038 | * The memory barrier imposed by test&clear is paired with the | |
3039 | * explicit one in memcg_kmem_mark_dead(). | |
3040 | */ | |
7de37682 | 3041 | if (memcg_kmem_test_and_clear_dead(memcg)) |
10d5ebf4 | 3042 | css_put(&memcg->css); |
7ae1e1d0 GC |
3043 | } |
3044 | ||
2633d7a0 GC |
3045 | void memcg_cache_list_add(struct mem_cgroup *memcg, struct kmem_cache *cachep) |
3046 | { | |
3047 | if (!memcg) | |
3048 | return; | |
3049 | ||
3050 | mutex_lock(&memcg->slab_caches_mutex); | |
3051 | list_add(&cachep->memcg_params->list, &memcg->memcg_slab_caches); | |
3052 | mutex_unlock(&memcg->slab_caches_mutex); | |
3053 | } | |
3054 | ||
3055 | /* | |
3056 | * helper for acessing a memcg's index. It will be used as an index in the | |
3057 | * child cache array in kmem_cache, and also to derive its name. This function | |
3058 | * will return -1 when this is not a kmem-limited memcg. | |
3059 | */ | |
3060 | int memcg_cache_id(struct mem_cgroup *memcg) | |
3061 | { | |
3062 | return memcg ? memcg->kmemcg_id : -1; | |
3063 | } | |
3064 | ||
55007d84 GC |
3065 | /* |
3066 | * This ends up being protected by the set_limit mutex, during normal | |
3067 | * operation, because that is its main call site. | |
3068 | * | |
3069 | * But when we create a new cache, we can call this as well if its parent | |
3070 | * is kmem-limited. That will have to hold set_limit_mutex as well. | |
3071 | */ | |
3072 | int memcg_update_cache_sizes(struct mem_cgroup *memcg) | |
3073 | { | |
3074 | int num, ret; | |
3075 | ||
3076 | num = ida_simple_get(&kmem_limited_groups, | |
3077 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); | |
3078 | if (num < 0) | |
3079 | return num; | |
3080 | /* | |
3081 | * After this point, kmem_accounted (that we test atomically in | |
3082 | * the beginning of this conditional), is no longer 0. This | |
3083 | * guarantees only one process will set the following boolean | |
3084 | * to true. We don't need test_and_set because we're protected | |
3085 | * by the set_limit_mutex anyway. | |
3086 | */ | |
3087 | memcg_kmem_set_activated(memcg); | |
3088 | ||
3089 | ret = memcg_update_all_caches(num+1); | |
3090 | if (ret) { | |
3091 | ida_simple_remove(&kmem_limited_groups, num); | |
3092 | memcg_kmem_clear_activated(memcg); | |
3093 | return ret; | |
3094 | } | |
3095 | ||
3096 | memcg->kmemcg_id = num; | |
3097 | INIT_LIST_HEAD(&memcg->memcg_slab_caches); | |
3098 | mutex_init(&memcg->slab_caches_mutex); | |
3099 | return 0; | |
3100 | } | |
3101 | ||
3102 | static size_t memcg_caches_array_size(int num_groups) | |
3103 | { | |
3104 | ssize_t size; | |
3105 | if (num_groups <= 0) | |
3106 | return 0; | |
3107 | ||
3108 | size = 2 * num_groups; | |
3109 | if (size < MEMCG_CACHES_MIN_SIZE) | |
3110 | size = MEMCG_CACHES_MIN_SIZE; | |
3111 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
3112 | size = MEMCG_CACHES_MAX_SIZE; | |
3113 | ||
3114 | return size; | |
3115 | } | |
3116 | ||
3117 | /* | |
3118 | * We should update the current array size iff all caches updates succeed. This | |
3119 | * can only be done from the slab side. The slab mutex needs to be held when | |
3120 | * calling this. | |
3121 | */ | |
3122 | void memcg_update_array_size(int num) | |
3123 | { | |
3124 | if (num > memcg_limited_groups_array_size) | |
3125 | memcg_limited_groups_array_size = memcg_caches_array_size(num); | |
3126 | } | |
3127 | ||
15cf17d2 KK |
3128 | static void kmem_cache_destroy_work_func(struct work_struct *w); |
3129 | ||
55007d84 GC |
3130 | int memcg_update_cache_size(struct kmem_cache *s, int num_groups) |
3131 | { | |
3132 | struct memcg_cache_params *cur_params = s->memcg_params; | |
3133 | ||
3134 | VM_BUG_ON(s->memcg_params && !s->memcg_params->is_root_cache); | |
3135 | ||
3136 | if (num_groups > memcg_limited_groups_array_size) { | |
3137 | int i; | |
3138 | ssize_t size = memcg_caches_array_size(num_groups); | |
3139 | ||
3140 | size *= sizeof(void *); | |
90c7a79c | 3141 | size += offsetof(struct memcg_cache_params, memcg_caches); |
55007d84 GC |
3142 | |
3143 | s->memcg_params = kzalloc(size, GFP_KERNEL); | |
3144 | if (!s->memcg_params) { | |
3145 | s->memcg_params = cur_params; | |
3146 | return -ENOMEM; | |
3147 | } | |
3148 | ||
3149 | s->memcg_params->is_root_cache = true; | |
3150 | ||
3151 | /* | |
3152 | * There is the chance it will be bigger than | |
3153 | * memcg_limited_groups_array_size, if we failed an allocation | |
3154 | * in a cache, in which case all caches updated before it, will | |
3155 | * have a bigger array. | |
3156 | * | |
3157 | * But if that is the case, the data after | |
3158 | * memcg_limited_groups_array_size is certainly unused | |
3159 | */ | |
3160 | for (i = 0; i < memcg_limited_groups_array_size; i++) { | |
3161 | if (!cur_params->memcg_caches[i]) | |
3162 | continue; | |
3163 | s->memcg_params->memcg_caches[i] = | |
3164 | cur_params->memcg_caches[i]; | |
3165 | } | |
3166 | ||
3167 | /* | |
3168 | * Ideally, we would wait until all caches succeed, and only | |
3169 | * then free the old one. But this is not worth the extra | |
3170 | * pointer per-cache we'd have to have for this. | |
3171 | * | |
3172 | * It is not a big deal if some caches are left with a size | |
3173 | * bigger than the others. And all updates will reset this | |
3174 | * anyway. | |
3175 | */ | |
3176 | kfree(cur_params); | |
3177 | } | |
3178 | return 0; | |
3179 | } | |
3180 | ||
943a451a GC |
3181 | int memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s, |
3182 | struct kmem_cache *root_cache) | |
2633d7a0 | 3183 | { |
90c7a79c | 3184 | size_t size; |
2633d7a0 GC |
3185 | |
3186 | if (!memcg_kmem_enabled()) | |
3187 | return 0; | |
3188 | ||
90c7a79c AV |
3189 | if (!memcg) { |
3190 | size = offsetof(struct memcg_cache_params, memcg_caches); | |
55007d84 | 3191 | size += memcg_limited_groups_array_size * sizeof(void *); |
90c7a79c AV |
3192 | } else |
3193 | size = sizeof(struct memcg_cache_params); | |
55007d84 | 3194 | |
2633d7a0 GC |
3195 | s->memcg_params = kzalloc(size, GFP_KERNEL); |
3196 | if (!s->memcg_params) | |
3197 | return -ENOMEM; | |
3198 | ||
943a451a | 3199 | if (memcg) { |
2633d7a0 | 3200 | s->memcg_params->memcg = memcg; |
943a451a | 3201 | s->memcg_params->root_cache = root_cache; |
3e6b11df AV |
3202 | INIT_WORK(&s->memcg_params->destroy, |
3203 | kmem_cache_destroy_work_func); | |
4ba902b5 GC |
3204 | } else |
3205 | s->memcg_params->is_root_cache = true; | |
3206 | ||
2633d7a0 GC |
3207 | return 0; |
3208 | } | |
3209 | ||
3210 | void memcg_release_cache(struct kmem_cache *s) | |
3211 | { | |
d7f25f8a GC |
3212 | struct kmem_cache *root; |
3213 | struct mem_cgroup *memcg; | |
3214 | int id; | |
3215 | ||
3216 | /* | |
3217 | * This happens, for instance, when a root cache goes away before we | |
3218 | * add any memcg. | |
3219 | */ | |
3220 | if (!s->memcg_params) | |
3221 | return; | |
3222 | ||
3223 | if (s->memcg_params->is_root_cache) | |
3224 | goto out; | |
3225 | ||
3226 | memcg = s->memcg_params->memcg; | |
3227 | id = memcg_cache_id(memcg); | |
3228 | ||
3229 | root = s->memcg_params->root_cache; | |
3230 | root->memcg_params->memcg_caches[id] = NULL; | |
d7f25f8a GC |
3231 | |
3232 | mutex_lock(&memcg->slab_caches_mutex); | |
3233 | list_del(&s->memcg_params->list); | |
3234 | mutex_unlock(&memcg->slab_caches_mutex); | |
3235 | ||
20f05310 | 3236 | css_put(&memcg->css); |
d7f25f8a | 3237 | out: |
2633d7a0 GC |
3238 | kfree(s->memcg_params); |
3239 | } | |
3240 | ||
0e9d92f2 GC |
3241 | /* |
3242 | * During the creation a new cache, we need to disable our accounting mechanism | |
3243 | * altogether. This is true even if we are not creating, but rather just | |
3244 | * enqueing new caches to be created. | |
3245 | * | |
3246 | * This is because that process will trigger allocations; some visible, like | |
3247 | * explicit kmallocs to auxiliary data structures, name strings and internal | |
3248 | * cache structures; some well concealed, like INIT_WORK() that can allocate | |
3249 | * objects during debug. | |
3250 | * | |
3251 | * If any allocation happens during memcg_kmem_get_cache, we will recurse back | |
3252 | * to it. This may not be a bounded recursion: since the first cache creation | |
3253 | * failed to complete (waiting on the allocation), we'll just try to create the | |
3254 | * cache again, failing at the same point. | |
3255 | * | |
3256 | * memcg_kmem_get_cache is prepared to abort after seeing a positive count of | |
3257 | * memcg_kmem_skip_account. So we enclose anything that might allocate memory | |
3258 | * inside the following two functions. | |
3259 | */ | |
3260 | static inline void memcg_stop_kmem_account(void) | |
3261 | { | |
3262 | VM_BUG_ON(!current->mm); | |
3263 | current->memcg_kmem_skip_account++; | |
3264 | } | |
3265 | ||
3266 | static inline void memcg_resume_kmem_account(void) | |
3267 | { | |
3268 | VM_BUG_ON(!current->mm); | |
3269 | current->memcg_kmem_skip_account--; | |
3270 | } | |
3271 | ||
1f458cbf GC |
3272 | static void kmem_cache_destroy_work_func(struct work_struct *w) |
3273 | { | |
3274 | struct kmem_cache *cachep; | |
3275 | struct memcg_cache_params *p; | |
3276 | ||
3277 | p = container_of(w, struct memcg_cache_params, destroy); | |
3278 | ||
3279 | cachep = memcg_params_to_cache(p); | |
3280 | ||
22933152 GC |
3281 | /* |
3282 | * If we get down to 0 after shrink, we could delete right away. | |
3283 | * However, memcg_release_pages() already puts us back in the workqueue | |
3284 | * in that case. If we proceed deleting, we'll get a dangling | |
3285 | * reference, and removing the object from the workqueue in that case | |
3286 | * is unnecessary complication. We are not a fast path. | |
3287 | * | |
3288 | * Note that this case is fundamentally different from racing with | |
3289 | * shrink_slab(): if memcg_cgroup_destroy_cache() is called in | |
3290 | * kmem_cache_shrink, not only we would be reinserting a dead cache | |
3291 | * into the queue, but doing so from inside the worker racing to | |
3292 | * destroy it. | |
3293 | * | |
3294 | * So if we aren't down to zero, we'll just schedule a worker and try | |
3295 | * again | |
3296 | */ | |
3297 | if (atomic_read(&cachep->memcg_params->nr_pages) != 0) { | |
3298 | kmem_cache_shrink(cachep); | |
3299 | if (atomic_read(&cachep->memcg_params->nr_pages) == 0) | |
3300 | return; | |
3301 | } else | |
1f458cbf GC |
3302 | kmem_cache_destroy(cachep); |
3303 | } | |
3304 | ||
3305 | void mem_cgroup_destroy_cache(struct kmem_cache *cachep) | |
3306 | { | |
3307 | if (!cachep->memcg_params->dead) | |
3308 | return; | |
3309 | ||
22933152 GC |
3310 | /* |
3311 | * There are many ways in which we can get here. | |
3312 | * | |
3313 | * We can get to a memory-pressure situation while the delayed work is | |
3314 | * still pending to run. The vmscan shrinkers can then release all | |
3315 | * cache memory and get us to destruction. If this is the case, we'll | |
3316 | * be executed twice, which is a bug (the second time will execute over | |
3317 | * bogus data). In this case, cancelling the work should be fine. | |
3318 | * | |
3319 | * But we can also get here from the worker itself, if | |
3320 | * kmem_cache_shrink is enough to shake all the remaining objects and | |
3321 | * get the page count to 0. In this case, we'll deadlock if we try to | |
3322 | * cancel the work (the worker runs with an internal lock held, which | |
3323 | * is the same lock we would hold for cancel_work_sync().) | |
3324 | * | |
3325 | * Since we can't possibly know who got us here, just refrain from | |
3326 | * running if there is already work pending | |
3327 | */ | |
3328 | if (work_pending(&cachep->memcg_params->destroy)) | |
3329 | return; | |
1f458cbf GC |
3330 | /* |
3331 | * We have to defer the actual destroying to a workqueue, because | |
3332 | * we might currently be in a context that cannot sleep. | |
3333 | */ | |
3334 | schedule_work(&cachep->memcg_params->destroy); | |
3335 | } | |
3336 | ||
d9c10ddd MH |
3337 | /* |
3338 | * This lock protects updaters, not readers. We want readers to be as fast as | |
3339 | * they can, and they will either see NULL or a valid cache value. Our model | |
3340 | * allow them to see NULL, in which case the root memcg will be selected. | |
3341 | * | |
3342 | * We need this lock because multiple allocations to the same cache from a non | |
3343 | * will span more than one worker. Only one of them can create the cache. | |
3344 | */ | |
3345 | static DEFINE_MUTEX(memcg_cache_mutex); | |
d7f25f8a | 3346 | |
d9c10ddd MH |
3347 | /* |
3348 | * Called with memcg_cache_mutex held | |
3349 | */ | |
d7f25f8a GC |
3350 | static struct kmem_cache *kmem_cache_dup(struct mem_cgroup *memcg, |
3351 | struct kmem_cache *s) | |
3352 | { | |
d7f25f8a | 3353 | struct kmem_cache *new; |
d9c10ddd | 3354 | static char *tmp_name = NULL; |
d7f25f8a | 3355 | |
d9c10ddd MH |
3356 | lockdep_assert_held(&memcg_cache_mutex); |
3357 | ||
3358 | /* | |
3359 | * kmem_cache_create_memcg duplicates the given name and | |
3360 | * cgroup_name for this name requires RCU context. | |
3361 | * This static temporary buffer is used to prevent from | |
3362 | * pointless shortliving allocation. | |
3363 | */ | |
3364 | if (!tmp_name) { | |
3365 | tmp_name = kmalloc(PATH_MAX, GFP_KERNEL); | |
3366 | if (!tmp_name) | |
3367 | return NULL; | |
3368 | } | |
3369 | ||
3370 | rcu_read_lock(); | |
3371 | snprintf(tmp_name, PATH_MAX, "%s(%d:%s)", s->name, | |
3372 | memcg_cache_id(memcg), cgroup_name(memcg->css.cgroup)); | |
3373 | rcu_read_unlock(); | |
d7f25f8a | 3374 | |
d9c10ddd | 3375 | new = kmem_cache_create_memcg(memcg, tmp_name, s->object_size, s->align, |
943a451a | 3376 | (s->flags & ~SLAB_PANIC), s->ctor, s); |
d7f25f8a | 3377 | |
d79923fa GC |
3378 | if (new) |
3379 | new->allocflags |= __GFP_KMEMCG; | |
3380 | ||
d7f25f8a GC |
3381 | return new; |
3382 | } | |
3383 | ||
d7f25f8a GC |
3384 | static struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg, |
3385 | struct kmem_cache *cachep) | |
3386 | { | |
3387 | struct kmem_cache *new_cachep; | |
3388 | int idx; | |
3389 | ||
3390 | BUG_ON(!memcg_can_account_kmem(memcg)); | |
3391 | ||
3392 | idx = memcg_cache_id(memcg); | |
3393 | ||
3394 | mutex_lock(&memcg_cache_mutex); | |
3395 | new_cachep = cachep->memcg_params->memcg_caches[idx]; | |
20f05310 LZ |
3396 | if (new_cachep) { |
3397 | css_put(&memcg->css); | |
d7f25f8a | 3398 | goto out; |
20f05310 | 3399 | } |
d7f25f8a GC |
3400 | |
3401 | new_cachep = kmem_cache_dup(memcg, cachep); | |
d7f25f8a GC |
3402 | if (new_cachep == NULL) { |
3403 | new_cachep = cachep; | |
20f05310 | 3404 | css_put(&memcg->css); |
d7f25f8a GC |
3405 | goto out; |
3406 | } | |
3407 | ||
1f458cbf | 3408 | atomic_set(&new_cachep->memcg_params->nr_pages , 0); |
d7f25f8a GC |
3409 | |
3410 | cachep->memcg_params->memcg_caches[idx] = new_cachep; | |
3411 | /* | |
3412 | * the readers won't lock, make sure everybody sees the updated value, | |
3413 | * so they won't put stuff in the queue again for no reason | |
3414 | */ | |
3415 | wmb(); | |
3416 | out: | |
3417 | mutex_unlock(&memcg_cache_mutex); | |
3418 | return new_cachep; | |
3419 | } | |
3420 | ||
7cf27982 GC |
3421 | void kmem_cache_destroy_memcg_children(struct kmem_cache *s) |
3422 | { | |
3423 | struct kmem_cache *c; | |
3424 | int i; | |
3425 | ||
3426 | if (!s->memcg_params) | |
3427 | return; | |
3428 | if (!s->memcg_params->is_root_cache) | |
3429 | return; | |
3430 | ||
3431 | /* | |
3432 | * If the cache is being destroyed, we trust that there is no one else | |
3433 | * requesting objects from it. Even if there are, the sanity checks in | |
3434 | * kmem_cache_destroy should caught this ill-case. | |
3435 | * | |
3436 | * Still, we don't want anyone else freeing memcg_caches under our | |
3437 | * noses, which can happen if a new memcg comes to life. As usual, | |
3438 | * we'll take the set_limit_mutex to protect ourselves against this. | |
3439 | */ | |
3440 | mutex_lock(&set_limit_mutex); | |
3441 | for (i = 0; i < memcg_limited_groups_array_size; i++) { | |
3442 | c = s->memcg_params->memcg_caches[i]; | |
3443 | if (!c) | |
3444 | continue; | |
3445 | ||
3446 | /* | |
3447 | * We will now manually delete the caches, so to avoid races | |
3448 | * we need to cancel all pending destruction workers and | |
3449 | * proceed with destruction ourselves. | |
3450 | * | |
3451 | * kmem_cache_destroy() will call kmem_cache_shrink internally, | |
3452 | * and that could spawn the workers again: it is likely that | |
3453 | * the cache still have active pages until this very moment. | |
3454 | * This would lead us back to mem_cgroup_destroy_cache. | |
3455 | * | |
3456 | * But that will not execute at all if the "dead" flag is not | |
3457 | * set, so flip it down to guarantee we are in control. | |
3458 | */ | |
3459 | c->memcg_params->dead = false; | |
22933152 | 3460 | cancel_work_sync(&c->memcg_params->destroy); |
7cf27982 GC |
3461 | kmem_cache_destroy(c); |
3462 | } | |
3463 | mutex_unlock(&set_limit_mutex); | |
3464 | } | |
3465 | ||
d7f25f8a GC |
3466 | struct create_work { |
3467 | struct mem_cgroup *memcg; | |
3468 | struct kmem_cache *cachep; | |
3469 | struct work_struct work; | |
3470 | }; | |
3471 | ||
1f458cbf GC |
3472 | static void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg) |
3473 | { | |
3474 | struct kmem_cache *cachep; | |
3475 | struct memcg_cache_params *params; | |
3476 | ||
3477 | if (!memcg_kmem_is_active(memcg)) | |
3478 | return; | |
3479 | ||
3480 | mutex_lock(&memcg->slab_caches_mutex); | |
3481 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) { | |
3482 | cachep = memcg_params_to_cache(params); | |
3483 | cachep->memcg_params->dead = true; | |
1f458cbf GC |
3484 | schedule_work(&cachep->memcg_params->destroy); |
3485 | } | |
3486 | mutex_unlock(&memcg->slab_caches_mutex); | |
3487 | } | |
3488 | ||
d7f25f8a GC |
3489 | static void memcg_create_cache_work_func(struct work_struct *w) |
3490 | { | |
3491 | struct create_work *cw; | |
3492 | ||
3493 | cw = container_of(w, struct create_work, work); | |
3494 | memcg_create_kmem_cache(cw->memcg, cw->cachep); | |
d7f25f8a GC |
3495 | kfree(cw); |
3496 | } | |
3497 | ||
3498 | /* | |
3499 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 3500 | */ |
0e9d92f2 GC |
3501 | static void __memcg_create_cache_enqueue(struct mem_cgroup *memcg, |
3502 | struct kmem_cache *cachep) | |
d7f25f8a GC |
3503 | { |
3504 | struct create_work *cw; | |
3505 | ||
3506 | cw = kmalloc(sizeof(struct create_work), GFP_NOWAIT); | |
ca0dde97 LZ |
3507 | if (cw == NULL) { |
3508 | css_put(&memcg->css); | |
d7f25f8a GC |
3509 | return; |
3510 | } | |
3511 | ||
3512 | cw->memcg = memcg; | |
3513 | cw->cachep = cachep; | |
3514 | ||
3515 | INIT_WORK(&cw->work, memcg_create_cache_work_func); | |
3516 | schedule_work(&cw->work); | |
3517 | } | |
3518 | ||
0e9d92f2 GC |
3519 | static void memcg_create_cache_enqueue(struct mem_cgroup *memcg, |
3520 | struct kmem_cache *cachep) | |
3521 | { | |
3522 | /* | |
3523 | * We need to stop accounting when we kmalloc, because if the | |
3524 | * corresponding kmalloc cache is not yet created, the first allocation | |
3525 | * in __memcg_create_cache_enqueue will recurse. | |
3526 | * | |
3527 | * However, it is better to enclose the whole function. Depending on | |
3528 | * the debugging options enabled, INIT_WORK(), for instance, can | |
3529 | * trigger an allocation. This too, will make us recurse. Because at | |
3530 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
3531 | * the safest choice is to do it like this, wrapping the whole function. | |
3532 | */ | |
3533 | memcg_stop_kmem_account(); | |
3534 | __memcg_create_cache_enqueue(memcg, cachep); | |
3535 | memcg_resume_kmem_account(); | |
3536 | } | |
d7f25f8a GC |
3537 | /* |
3538 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
3539 | * We try to use the current memcg's version of the cache. | |
3540 | * | |
3541 | * If the cache does not exist yet, if we are the first user of it, | |
3542 | * we either create it immediately, if possible, or create it asynchronously | |
3543 | * in a workqueue. | |
3544 | * In the latter case, we will let the current allocation go through with | |
3545 | * the original cache. | |
3546 | * | |
3547 | * Can't be called in interrupt context or from kernel threads. | |
3548 | * This function needs to be called with rcu_read_lock() held. | |
3549 | */ | |
3550 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, | |
3551 | gfp_t gfp) | |
3552 | { | |
3553 | struct mem_cgroup *memcg; | |
3554 | int idx; | |
3555 | ||
3556 | VM_BUG_ON(!cachep->memcg_params); | |
3557 | VM_BUG_ON(!cachep->memcg_params->is_root_cache); | |
3558 | ||
0e9d92f2 GC |
3559 | if (!current->mm || current->memcg_kmem_skip_account) |
3560 | return cachep; | |
3561 | ||
d7f25f8a GC |
3562 | rcu_read_lock(); |
3563 | memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner)); | |
d7f25f8a GC |
3564 | |
3565 | if (!memcg_can_account_kmem(memcg)) | |
ca0dde97 | 3566 | goto out; |
d7f25f8a GC |
3567 | |
3568 | idx = memcg_cache_id(memcg); | |
3569 | ||
3570 | /* | |
3571 | * barrier to mare sure we're always seeing the up to date value. The | |
3572 | * code updating memcg_caches will issue a write barrier to match this. | |
3573 | */ | |
3574 | read_barrier_depends(); | |
ca0dde97 LZ |
3575 | if (likely(cachep->memcg_params->memcg_caches[idx])) { |
3576 | cachep = cachep->memcg_params->memcg_caches[idx]; | |
3577 | goto out; | |
d7f25f8a GC |
3578 | } |
3579 | ||
ca0dde97 LZ |
3580 | /* The corresponding put will be done in the workqueue. */ |
3581 | if (!css_tryget(&memcg->css)) | |
3582 | goto out; | |
3583 | rcu_read_unlock(); | |
3584 | ||
3585 | /* | |
3586 | * If we are in a safe context (can wait, and not in interrupt | |
3587 | * context), we could be be predictable and return right away. | |
3588 | * This would guarantee that the allocation being performed | |
3589 | * already belongs in the new cache. | |
3590 | * | |
3591 | * However, there are some clashes that can arrive from locking. | |
3592 | * For instance, because we acquire the slab_mutex while doing | |
3593 | * kmem_cache_dup, this means no further allocation could happen | |
3594 | * with the slab_mutex held. | |
3595 | * | |
3596 | * Also, because cache creation issue get_online_cpus(), this | |
3597 | * creates a lock chain: memcg_slab_mutex -> cpu_hotplug_mutex, | |
3598 | * that ends up reversed during cpu hotplug. (cpuset allocates | |
3599 | * a bunch of GFP_KERNEL memory during cpuup). Due to all that, | |
3600 | * better to defer everything. | |
3601 | */ | |
3602 | memcg_create_cache_enqueue(memcg, cachep); | |
3603 | return cachep; | |
3604 | out: | |
3605 | rcu_read_unlock(); | |
3606 | return cachep; | |
d7f25f8a GC |
3607 | } |
3608 | EXPORT_SYMBOL(__memcg_kmem_get_cache); | |
3609 | ||
7ae1e1d0 GC |
3610 | /* |
3611 | * We need to verify if the allocation against current->mm->owner's memcg is | |
3612 | * possible for the given order. But the page is not allocated yet, so we'll | |
3613 | * need a further commit step to do the final arrangements. | |
3614 | * | |
3615 | * It is possible for the task to switch cgroups in this mean time, so at | |
3616 | * commit time, we can't rely on task conversion any longer. We'll then use | |
3617 | * the handle argument to return to the caller which cgroup we should commit | |
3618 | * against. We could also return the memcg directly and avoid the pointer | |
3619 | * passing, but a boolean return value gives better semantics considering | |
3620 | * the compiled-out case as well. | |
3621 | * | |
3622 | * Returning true means the allocation is possible. | |
3623 | */ | |
3624 | bool | |
3625 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) | |
3626 | { | |
3627 | struct mem_cgroup *memcg; | |
3628 | int ret; | |
3629 | ||
3630 | *_memcg = NULL; | |
6d42c232 GC |
3631 | |
3632 | /* | |
3633 | * Disabling accounting is only relevant for some specific memcg | |
3634 | * internal allocations. Therefore we would initially not have such | |
3635 | * check here, since direct calls to the page allocator that are marked | |
3636 | * with GFP_KMEMCG only happen outside memcg core. We are mostly | |
3637 | * concerned with cache allocations, and by having this test at | |
3638 | * memcg_kmem_get_cache, we are already able to relay the allocation to | |
3639 | * the root cache and bypass the memcg cache altogether. | |
3640 | * | |
3641 | * There is one exception, though: the SLUB allocator does not create | |
3642 | * large order caches, but rather service large kmallocs directly from | |
3643 | * the page allocator. Therefore, the following sequence when backed by | |
3644 | * the SLUB allocator: | |
3645 | * | |
f894ffa8 AM |
3646 | * memcg_stop_kmem_account(); |
3647 | * kmalloc(<large_number>) | |
3648 | * memcg_resume_kmem_account(); | |
6d42c232 GC |
3649 | * |
3650 | * would effectively ignore the fact that we should skip accounting, | |
3651 | * since it will drive us directly to this function without passing | |
3652 | * through the cache selector memcg_kmem_get_cache. Such large | |
3653 | * allocations are extremely rare but can happen, for instance, for the | |
3654 | * cache arrays. We bring this test here. | |
3655 | */ | |
3656 | if (!current->mm || current->memcg_kmem_skip_account) | |
3657 | return true; | |
3658 | ||
7ae1e1d0 GC |
3659 | memcg = try_get_mem_cgroup_from_mm(current->mm); |
3660 | ||
3661 | /* | |
3662 | * very rare case described in mem_cgroup_from_task. Unfortunately there | |
3663 | * isn't much we can do without complicating this too much, and it would | |
3664 | * be gfp-dependent anyway. Just let it go | |
3665 | */ | |
3666 | if (unlikely(!memcg)) | |
3667 | return true; | |
3668 | ||
3669 | if (!memcg_can_account_kmem(memcg)) { | |
3670 | css_put(&memcg->css); | |
3671 | return true; | |
3672 | } | |
3673 | ||
7ae1e1d0 GC |
3674 | ret = memcg_charge_kmem(memcg, gfp, PAGE_SIZE << order); |
3675 | if (!ret) | |
3676 | *_memcg = memcg; | |
7ae1e1d0 GC |
3677 | |
3678 | css_put(&memcg->css); | |
3679 | return (ret == 0); | |
3680 | } | |
3681 | ||
3682 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
3683 | int order) | |
3684 | { | |
3685 | struct page_cgroup *pc; | |
3686 | ||
3687 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
3688 | ||
3689 | /* The page allocation failed. Revert */ | |
3690 | if (!page) { | |
3691 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); | |
7ae1e1d0 GC |
3692 | return; |
3693 | } | |
3694 | ||
3695 | pc = lookup_page_cgroup(page); | |
3696 | lock_page_cgroup(pc); | |
3697 | pc->mem_cgroup = memcg; | |
3698 | SetPageCgroupUsed(pc); | |
3699 | unlock_page_cgroup(pc); | |
3700 | } | |
3701 | ||
3702 | void __memcg_kmem_uncharge_pages(struct page *page, int order) | |
3703 | { | |
3704 | struct mem_cgroup *memcg = NULL; | |
3705 | struct page_cgroup *pc; | |
3706 | ||
3707 | ||
3708 | pc = lookup_page_cgroup(page); | |
3709 | /* | |
3710 | * Fast unlocked return. Theoretically might have changed, have to | |
3711 | * check again after locking. | |
3712 | */ | |
3713 | if (!PageCgroupUsed(pc)) | |
3714 | return; | |
3715 | ||
3716 | lock_page_cgroup(pc); | |
3717 | if (PageCgroupUsed(pc)) { | |
3718 | memcg = pc->mem_cgroup; | |
3719 | ClearPageCgroupUsed(pc); | |
3720 | } | |
3721 | unlock_page_cgroup(pc); | |
3722 | ||
3723 | /* | |
3724 | * We trust that only if there is a memcg associated with the page, it | |
3725 | * is a valid allocation | |
3726 | */ | |
3727 | if (!memcg) | |
3728 | return; | |
3729 | ||
3730 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
3731 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); | |
7ae1e1d0 | 3732 | } |
1f458cbf GC |
3733 | #else |
3734 | static inline void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg) | |
3735 | { | |
3736 | } | |
7ae1e1d0 GC |
3737 | #endif /* CONFIG_MEMCG_KMEM */ |
3738 | ||
ca3e0214 KH |
3739 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3740 | ||
a0db00fc | 3741 | #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION) |
ca3e0214 KH |
3742 | /* |
3743 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
3744 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
3745 | * charge/uncharge will be never happen and move_account() is done under | |
3746 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 3747 | */ |
e94c8a9c | 3748 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 KH |
3749 | { |
3750 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
e94c8a9c | 3751 | struct page_cgroup *pc; |
b070e65c | 3752 | struct mem_cgroup *memcg; |
e94c8a9c | 3753 | int i; |
ca3e0214 | 3754 | |
3d37c4a9 KH |
3755 | if (mem_cgroup_disabled()) |
3756 | return; | |
b070e65c DR |
3757 | |
3758 | memcg = head_pc->mem_cgroup; | |
e94c8a9c KH |
3759 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
3760 | pc = head_pc + i; | |
b070e65c | 3761 | pc->mem_cgroup = memcg; |
e94c8a9c | 3762 | smp_wmb();/* see __commit_charge() */ |
e94c8a9c KH |
3763 | pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
3764 | } | |
b070e65c DR |
3765 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
3766 | HPAGE_PMD_NR); | |
ca3e0214 | 3767 | } |
12d27107 | 3768 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3769 | |
3ea67d06 SZ |
3770 | static inline |
3771 | void mem_cgroup_move_account_page_stat(struct mem_cgroup *from, | |
3772 | struct mem_cgroup *to, | |
3773 | unsigned int nr_pages, | |
3774 | enum mem_cgroup_stat_index idx) | |
3775 | { | |
3776 | /* Update stat data for mem_cgroup */ | |
3777 | preempt_disable(); | |
5e8cfc3c | 3778 | __this_cpu_sub(from->stat->count[idx], nr_pages); |
3ea67d06 SZ |
3779 | __this_cpu_add(to->stat->count[idx], nr_pages); |
3780 | preempt_enable(); | |
3781 | } | |
3782 | ||
f817ed48 | 3783 | /** |
de3638d9 | 3784 | * mem_cgroup_move_account - move account of the page |
5564e88b | 3785 | * @page: the page |
7ec99d62 | 3786 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
3787 | * @pc: page_cgroup of the page. |
3788 | * @from: mem_cgroup which the page is moved from. | |
3789 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
3790 | * | |
3791 | * The caller must confirm following. | |
08e552c6 | 3792 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 3793 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 3794 | * |
2f3479b1 KH |
3795 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
3796 | * from old cgroup. | |
f817ed48 | 3797 | */ |
7ec99d62 JW |
3798 | static int mem_cgroup_move_account(struct page *page, |
3799 | unsigned int nr_pages, | |
3800 | struct page_cgroup *pc, | |
3801 | struct mem_cgroup *from, | |
2f3479b1 | 3802 | struct mem_cgroup *to) |
f817ed48 | 3803 | { |
de3638d9 JW |
3804 | unsigned long flags; |
3805 | int ret; | |
b2402857 | 3806 | bool anon = PageAnon(page); |
987eba66 | 3807 | |
f817ed48 | 3808 | VM_BUG_ON(from == to); |
5564e88b | 3809 | VM_BUG_ON(PageLRU(page)); |
de3638d9 JW |
3810 | /* |
3811 | * The page is isolated from LRU. So, collapse function | |
3812 | * will not handle this page. But page splitting can happen. | |
3813 | * Do this check under compound_page_lock(). The caller should | |
3814 | * hold it. | |
3815 | */ | |
3816 | ret = -EBUSY; | |
7ec99d62 | 3817 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
3818 | goto out; |
3819 | ||
3820 | lock_page_cgroup(pc); | |
3821 | ||
3822 | ret = -EINVAL; | |
3823 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
3824 | goto unlock; | |
3825 | ||
312734c0 | 3826 | move_lock_mem_cgroup(from, &flags); |
f817ed48 | 3827 | |
3ea67d06 SZ |
3828 | if (!anon && page_mapped(page)) |
3829 | mem_cgroup_move_account_page_stat(from, to, nr_pages, | |
3830 | MEM_CGROUP_STAT_FILE_MAPPED); | |
3831 | ||
3832 | if (PageWriteback(page)) | |
3833 | mem_cgroup_move_account_page_stat(from, to, nr_pages, | |
3834 | MEM_CGROUP_STAT_WRITEBACK); | |
3835 | ||
b070e65c | 3836 | mem_cgroup_charge_statistics(from, page, anon, -nr_pages); |
d69b042f | 3837 | |
854ffa8d | 3838 | /* caller should have done css_get */ |
08e552c6 | 3839 | pc->mem_cgroup = to; |
b070e65c | 3840 | mem_cgroup_charge_statistics(to, page, anon, nr_pages); |
312734c0 | 3841 | move_unlock_mem_cgroup(from, &flags); |
de3638d9 JW |
3842 | ret = 0; |
3843 | unlock: | |
57f9fd7d | 3844 | unlock_page_cgroup(pc); |
d2265e6f KH |
3845 | /* |
3846 | * check events | |
3847 | */ | |
5564e88b JW |
3848 | memcg_check_events(to, page); |
3849 | memcg_check_events(from, page); | |
de3638d9 | 3850 | out: |
f817ed48 KH |
3851 | return ret; |
3852 | } | |
3853 | ||
2ef37d3f MH |
3854 | /** |
3855 | * mem_cgroup_move_parent - moves page to the parent group | |
3856 | * @page: the page to move | |
3857 | * @pc: page_cgroup of the page | |
3858 | * @child: page's cgroup | |
3859 | * | |
3860 | * move charges to its parent or the root cgroup if the group has no | |
3861 | * parent (aka use_hierarchy==0). | |
3862 | * Although this might fail (get_page_unless_zero, isolate_lru_page or | |
3863 | * mem_cgroup_move_account fails) the failure is always temporary and | |
3864 | * it signals a race with a page removal/uncharge or migration. In the | |
3865 | * first case the page is on the way out and it will vanish from the LRU | |
3866 | * on the next attempt and the call should be retried later. | |
3867 | * Isolation from the LRU fails only if page has been isolated from | |
3868 | * the LRU since we looked at it and that usually means either global | |
3869 | * reclaim or migration going on. The page will either get back to the | |
3870 | * LRU or vanish. | |
3871 | * Finaly mem_cgroup_move_account fails only if the page got uncharged | |
3872 | * (!PageCgroupUsed) or moved to a different group. The page will | |
3873 | * disappear in the next attempt. | |
f817ed48 | 3874 | */ |
5564e88b JW |
3875 | static int mem_cgroup_move_parent(struct page *page, |
3876 | struct page_cgroup *pc, | |
6068bf01 | 3877 | struct mem_cgroup *child) |
f817ed48 | 3878 | { |
f817ed48 | 3879 | struct mem_cgroup *parent; |
7ec99d62 | 3880 | unsigned int nr_pages; |
4be4489f | 3881 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
3882 | int ret; |
3883 | ||
d8423011 | 3884 | VM_BUG_ON(mem_cgroup_is_root(child)); |
f817ed48 | 3885 | |
57f9fd7d DN |
3886 | ret = -EBUSY; |
3887 | if (!get_page_unless_zero(page)) | |
3888 | goto out; | |
3889 | if (isolate_lru_page(page)) | |
3890 | goto put; | |
52dbb905 | 3891 | |
7ec99d62 | 3892 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 3893 | |
cc926f78 KH |
3894 | parent = parent_mem_cgroup(child); |
3895 | /* | |
3896 | * If no parent, move charges to root cgroup. | |
3897 | */ | |
3898 | if (!parent) | |
3899 | parent = root_mem_cgroup; | |
f817ed48 | 3900 | |
2ef37d3f MH |
3901 | if (nr_pages > 1) { |
3902 | VM_BUG_ON(!PageTransHuge(page)); | |
987eba66 | 3903 | flags = compound_lock_irqsave(page); |
2ef37d3f | 3904 | } |
987eba66 | 3905 | |
cc926f78 | 3906 | ret = mem_cgroup_move_account(page, nr_pages, |
2f3479b1 | 3907 | pc, child, parent); |
cc926f78 KH |
3908 | if (!ret) |
3909 | __mem_cgroup_cancel_local_charge(child, nr_pages); | |
8dba474f | 3910 | |
7ec99d62 | 3911 | if (nr_pages > 1) |
987eba66 | 3912 | compound_unlock_irqrestore(page, flags); |
08e552c6 | 3913 | putback_lru_page(page); |
57f9fd7d | 3914 | put: |
40d58138 | 3915 | put_page(page); |
57f9fd7d | 3916 | out: |
f817ed48 KH |
3917 | return ret; |
3918 | } | |
3919 | ||
7a81b88c KH |
3920 | /* |
3921 | * Charge the memory controller for page usage. | |
3922 | * Return | |
3923 | * 0 if the charge was successful | |
3924 | * < 0 if the cgroup is over its limit | |
3925 | */ | |
3926 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
73045c47 | 3927 | gfp_t gfp_mask, enum charge_type ctype) |
7a81b88c | 3928 | { |
c0ff4b85 | 3929 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 3930 | unsigned int nr_pages = 1; |
8493ae43 | 3931 | bool oom = true; |
7a81b88c | 3932 | int ret; |
ec168510 | 3933 | |
37c2ac78 | 3934 | if (PageTransHuge(page)) { |
7ec99d62 | 3935 | nr_pages <<= compound_order(page); |
37c2ac78 | 3936 | VM_BUG_ON(!PageTransHuge(page)); |
8493ae43 JW |
3937 | /* |
3938 | * Never OOM-kill a process for a huge page. The | |
3939 | * fault handler will fall back to regular pages. | |
3940 | */ | |
3941 | oom = false; | |
37c2ac78 | 3942 | } |
7a81b88c | 3943 | |
c0ff4b85 | 3944 | ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom); |
38c5d72f | 3945 | if (ret == -ENOMEM) |
7a81b88c | 3946 | return ret; |
ce587e65 | 3947 | __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false); |
8a9f3ccd | 3948 | return 0; |
8a9f3ccd BS |
3949 | } |
3950 | ||
7a81b88c KH |
3951 | int mem_cgroup_newpage_charge(struct page *page, |
3952 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 3953 | { |
f8d66542 | 3954 | if (mem_cgroup_disabled()) |
cede86ac | 3955 | return 0; |
7a0524cf JW |
3956 | VM_BUG_ON(page_mapped(page)); |
3957 | VM_BUG_ON(page->mapping && !PageAnon(page)); | |
3958 | VM_BUG_ON(!mm); | |
217bc319 | 3959 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
41326c17 | 3960 | MEM_CGROUP_CHARGE_TYPE_ANON); |
217bc319 KH |
3961 | } |
3962 | ||
54595fe2 KH |
3963 | /* |
3964 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
3965 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 3966 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
3967 | * "commit()" or removed by "cancel()" |
3968 | */ | |
0435a2fd JW |
3969 | static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
3970 | struct page *page, | |
3971 | gfp_t mask, | |
3972 | struct mem_cgroup **memcgp) | |
8c7c6e34 | 3973 | { |
c0ff4b85 | 3974 | struct mem_cgroup *memcg; |
90deb788 | 3975 | struct page_cgroup *pc; |
54595fe2 | 3976 | int ret; |
8c7c6e34 | 3977 | |
90deb788 JW |
3978 | pc = lookup_page_cgroup(page); |
3979 | /* | |
3980 | * Every swap fault against a single page tries to charge the | |
3981 | * page, bail as early as possible. shmem_unuse() encounters | |
3982 | * already charged pages, too. The USED bit is protected by | |
3983 | * the page lock, which serializes swap cache removal, which | |
3984 | * in turn serializes uncharging. | |
3985 | */ | |
3986 | if (PageCgroupUsed(pc)) | |
3987 | return 0; | |
8c7c6e34 KH |
3988 | if (!do_swap_account) |
3989 | goto charge_cur_mm; | |
c0ff4b85 R |
3990 | memcg = try_get_mem_cgroup_from_page(page); |
3991 | if (!memcg) | |
54595fe2 | 3992 | goto charge_cur_mm; |
72835c86 JW |
3993 | *memcgp = memcg; |
3994 | ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true); | |
c0ff4b85 | 3995 | css_put(&memcg->css); |
38c5d72f KH |
3996 | if (ret == -EINTR) |
3997 | ret = 0; | |
54595fe2 | 3998 | return ret; |
8c7c6e34 | 3999 | charge_cur_mm: |
38c5d72f KH |
4000 | ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true); |
4001 | if (ret == -EINTR) | |
4002 | ret = 0; | |
4003 | return ret; | |
8c7c6e34 KH |
4004 | } |
4005 | ||
0435a2fd JW |
4006 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page, |
4007 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
4008 | { | |
4009 | *memcgp = NULL; | |
4010 | if (mem_cgroup_disabled()) | |
4011 | return 0; | |
bdf4f4d2 JW |
4012 | /* |
4013 | * A racing thread's fault, or swapoff, may have already | |
4014 | * updated the pte, and even removed page from swap cache: in | |
4015 | * those cases unuse_pte()'s pte_same() test will fail; but | |
4016 | * there's also a KSM case which does need to charge the page. | |
4017 | */ | |
4018 | if (!PageSwapCache(page)) { | |
4019 | int ret; | |
4020 | ||
4021 | ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, memcgp, true); | |
4022 | if (ret == -EINTR) | |
4023 | ret = 0; | |
4024 | return ret; | |
4025 | } | |
0435a2fd JW |
4026 | return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp); |
4027 | } | |
4028 | ||
827a03d2 JW |
4029 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg) |
4030 | { | |
4031 | if (mem_cgroup_disabled()) | |
4032 | return; | |
4033 | if (!memcg) | |
4034 | return; | |
4035 | __mem_cgroup_cancel_charge(memcg, 1); | |
4036 | } | |
4037 | ||
83aae4c7 | 4038 | static void |
72835c86 | 4039 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg, |
83aae4c7 | 4040 | enum charge_type ctype) |
7a81b88c | 4041 | { |
f8d66542 | 4042 | if (mem_cgroup_disabled()) |
7a81b88c | 4043 | return; |
72835c86 | 4044 | if (!memcg) |
7a81b88c | 4045 | return; |
5a6475a4 | 4046 | |
ce587e65 | 4047 | __mem_cgroup_commit_charge(memcg, page, 1, ctype, true); |
8c7c6e34 KH |
4048 | /* |
4049 | * Now swap is on-memory. This means this page may be | |
4050 | * counted both as mem and swap....double count. | |
03f3c433 KH |
4051 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
4052 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
4053 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 4054 | */ |
03f3c433 | 4055 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 4056 | swp_entry_t ent = {.val = page_private(page)}; |
86493009 | 4057 | mem_cgroup_uncharge_swap(ent); |
8c7c6e34 | 4058 | } |
7a81b88c KH |
4059 | } |
4060 | ||
72835c86 JW |
4061 | void mem_cgroup_commit_charge_swapin(struct page *page, |
4062 | struct mem_cgroup *memcg) | |
83aae4c7 | 4063 | { |
72835c86 | 4064 | __mem_cgroup_commit_charge_swapin(page, memcg, |
41326c17 | 4065 | MEM_CGROUP_CHARGE_TYPE_ANON); |
83aae4c7 DN |
4066 | } |
4067 | ||
827a03d2 JW |
4068 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
4069 | gfp_t gfp_mask) | |
7a81b88c | 4070 | { |
827a03d2 JW |
4071 | struct mem_cgroup *memcg = NULL; |
4072 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
4073 | int ret; | |
4074 | ||
f8d66542 | 4075 | if (mem_cgroup_disabled()) |
827a03d2 JW |
4076 | return 0; |
4077 | if (PageCompound(page)) | |
4078 | return 0; | |
4079 | ||
827a03d2 JW |
4080 | if (!PageSwapCache(page)) |
4081 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, type); | |
4082 | else { /* page is swapcache/shmem */ | |
0435a2fd JW |
4083 | ret = __mem_cgroup_try_charge_swapin(mm, page, |
4084 | gfp_mask, &memcg); | |
827a03d2 JW |
4085 | if (!ret) |
4086 | __mem_cgroup_commit_charge_swapin(page, memcg, type); | |
4087 | } | |
4088 | return ret; | |
7a81b88c KH |
4089 | } |
4090 | ||
c0ff4b85 | 4091 | static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg, |
7ec99d62 JW |
4092 | unsigned int nr_pages, |
4093 | const enum charge_type ctype) | |
569b846d KH |
4094 | { |
4095 | struct memcg_batch_info *batch = NULL; | |
4096 | bool uncharge_memsw = true; | |
7ec99d62 | 4097 | |
569b846d KH |
4098 | /* If swapout, usage of swap doesn't decrease */ |
4099 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
4100 | uncharge_memsw = false; | |
569b846d KH |
4101 | |
4102 | batch = ¤t->memcg_batch; | |
4103 | /* | |
4104 | * In usual, we do css_get() when we remember memcg pointer. | |
4105 | * But in this case, we keep res->usage until end of a series of | |
4106 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
4107 | */ | |
4108 | if (!batch->memcg) | |
c0ff4b85 | 4109 | batch->memcg = memcg; |
3c11ecf4 KH |
4110 | /* |
4111 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
25985edc | 4112 | * In those cases, all pages freed continuously can be expected to be in |
3c11ecf4 KH |
4113 | * the same cgroup and we have chance to coalesce uncharges. |
4114 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
4115 | * because we want to do uncharge as soon as possible. | |
4116 | */ | |
4117 | ||
4118 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
4119 | goto direct_uncharge; | |
4120 | ||
7ec99d62 | 4121 | if (nr_pages > 1) |
ec168510 AA |
4122 | goto direct_uncharge; |
4123 | ||
569b846d KH |
4124 | /* |
4125 | * In typical case, batch->memcg == mem. This means we can | |
4126 | * merge a series of uncharges to an uncharge of res_counter. | |
4127 | * If not, we uncharge res_counter ony by one. | |
4128 | */ | |
c0ff4b85 | 4129 | if (batch->memcg != memcg) |
569b846d KH |
4130 | goto direct_uncharge; |
4131 | /* remember freed charge and uncharge it later */ | |
7ffd4ca7 | 4132 | batch->nr_pages++; |
569b846d | 4133 | if (uncharge_memsw) |
7ffd4ca7 | 4134 | batch->memsw_nr_pages++; |
569b846d KH |
4135 | return; |
4136 | direct_uncharge: | |
c0ff4b85 | 4137 | res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE); |
569b846d | 4138 | if (uncharge_memsw) |
c0ff4b85 R |
4139 | res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE); |
4140 | if (unlikely(batch->memcg != memcg)) | |
4141 | memcg_oom_recover(memcg); | |
569b846d | 4142 | } |
7a81b88c | 4143 | |
8a9f3ccd | 4144 | /* |
69029cd5 | 4145 | * uncharge if !page_mapped(page) |
8a9f3ccd | 4146 | */ |
8c7c6e34 | 4147 | static struct mem_cgroup * |
0030f535 JW |
4148 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype, |
4149 | bool end_migration) | |
8a9f3ccd | 4150 | { |
c0ff4b85 | 4151 | struct mem_cgroup *memcg = NULL; |
7ec99d62 JW |
4152 | unsigned int nr_pages = 1; |
4153 | struct page_cgroup *pc; | |
b2402857 | 4154 | bool anon; |
8a9f3ccd | 4155 | |
f8d66542 | 4156 | if (mem_cgroup_disabled()) |
8c7c6e34 | 4157 | return NULL; |
4077960e | 4158 | |
37c2ac78 | 4159 | if (PageTransHuge(page)) { |
7ec99d62 | 4160 | nr_pages <<= compound_order(page); |
37c2ac78 AA |
4161 | VM_BUG_ON(!PageTransHuge(page)); |
4162 | } | |
8697d331 | 4163 | /* |
3c541e14 | 4164 | * Check if our page_cgroup is valid |
8697d331 | 4165 | */ |
52d4b9ac | 4166 | pc = lookup_page_cgroup(page); |
cfa44946 | 4167 | if (unlikely(!PageCgroupUsed(pc))) |
8c7c6e34 | 4168 | return NULL; |
b9c565d5 | 4169 | |
52d4b9ac | 4170 | lock_page_cgroup(pc); |
d13d1443 | 4171 | |
c0ff4b85 | 4172 | memcg = pc->mem_cgroup; |
8c7c6e34 | 4173 | |
d13d1443 KH |
4174 | if (!PageCgroupUsed(pc)) |
4175 | goto unlock_out; | |
4176 | ||
b2402857 KH |
4177 | anon = PageAnon(page); |
4178 | ||
d13d1443 | 4179 | switch (ctype) { |
41326c17 | 4180 | case MEM_CGROUP_CHARGE_TYPE_ANON: |
2ff76f11 KH |
4181 | /* |
4182 | * Generally PageAnon tells if it's the anon statistics to be | |
4183 | * updated; but sometimes e.g. mem_cgroup_uncharge_page() is | |
4184 | * used before page reached the stage of being marked PageAnon. | |
4185 | */ | |
b2402857 KH |
4186 | anon = true; |
4187 | /* fallthrough */ | |
8a9478ca | 4188 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c | 4189 | /* See mem_cgroup_prepare_migration() */ |
0030f535 JW |
4190 | if (page_mapped(page)) |
4191 | goto unlock_out; | |
4192 | /* | |
4193 | * Pages under migration may not be uncharged. But | |
4194 | * end_migration() /must/ be the one uncharging the | |
4195 | * unused post-migration page and so it has to call | |
4196 | * here with the migration bit still set. See the | |
4197 | * res_counter handling below. | |
4198 | */ | |
4199 | if (!end_migration && PageCgroupMigration(pc)) | |
d13d1443 KH |
4200 | goto unlock_out; |
4201 | break; | |
4202 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
4203 | if (!PageAnon(page)) { /* Shared memory */ | |
4204 | if (page->mapping && !page_is_file_cache(page)) | |
4205 | goto unlock_out; | |
4206 | } else if (page_mapped(page)) /* Anon */ | |
4207 | goto unlock_out; | |
4208 | break; | |
4209 | default: | |
4210 | break; | |
52d4b9ac | 4211 | } |
d13d1443 | 4212 | |
b070e65c | 4213 | mem_cgroup_charge_statistics(memcg, page, anon, -nr_pages); |
04046e1a | 4214 | |
52d4b9ac | 4215 | ClearPageCgroupUsed(pc); |
544122e5 KH |
4216 | /* |
4217 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
4218 | * freed from LRU. This is safe because uncharged page is expected not | |
4219 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
4220 | * special functions. | |
4221 | */ | |
b9c565d5 | 4222 | |
52d4b9ac | 4223 | unlock_page_cgroup(pc); |
f75ca962 | 4224 | /* |
c0ff4b85 | 4225 | * even after unlock, we have memcg->res.usage here and this memcg |
4050377b | 4226 | * will never be freed, so it's safe to call css_get(). |
f75ca962 | 4227 | */ |
c0ff4b85 | 4228 | memcg_check_events(memcg, page); |
f75ca962 | 4229 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
c0ff4b85 | 4230 | mem_cgroup_swap_statistics(memcg, true); |
4050377b | 4231 | css_get(&memcg->css); |
f75ca962 | 4232 | } |
0030f535 JW |
4233 | /* |
4234 | * Migration does not charge the res_counter for the | |
4235 | * replacement page, so leave it alone when phasing out the | |
4236 | * page that is unused after the migration. | |
4237 | */ | |
4238 | if (!end_migration && !mem_cgroup_is_root(memcg)) | |
c0ff4b85 | 4239 | mem_cgroup_do_uncharge(memcg, nr_pages, ctype); |
6d12e2d8 | 4240 | |
c0ff4b85 | 4241 | return memcg; |
d13d1443 KH |
4242 | |
4243 | unlock_out: | |
4244 | unlock_page_cgroup(pc); | |
8c7c6e34 | 4245 | return NULL; |
3c541e14 BS |
4246 | } |
4247 | ||
69029cd5 KH |
4248 | void mem_cgroup_uncharge_page(struct page *page) |
4249 | { | |
52d4b9ac KH |
4250 | /* early check. */ |
4251 | if (page_mapped(page)) | |
4252 | return; | |
40f23a21 | 4253 | VM_BUG_ON(page->mapping && !PageAnon(page)); |
28ccddf7 JW |
4254 | /* |
4255 | * If the page is in swap cache, uncharge should be deferred | |
4256 | * to the swap path, which also properly accounts swap usage | |
4257 | * and handles memcg lifetime. | |
4258 | * | |
4259 | * Note that this check is not stable and reclaim may add the | |
4260 | * page to swap cache at any time after this. However, if the | |
4261 | * page is not in swap cache by the time page->mapcount hits | |
4262 | * 0, there won't be any page table references to the swap | |
4263 | * slot, and reclaim will free it and not actually write the | |
4264 | * page to disk. | |
4265 | */ | |
0c59b89c JW |
4266 | if (PageSwapCache(page)) |
4267 | return; | |
0030f535 | 4268 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false); |
69029cd5 KH |
4269 | } |
4270 | ||
4271 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
4272 | { | |
4273 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 4274 | VM_BUG_ON(page->mapping); |
0030f535 | 4275 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false); |
69029cd5 KH |
4276 | } |
4277 | ||
569b846d KH |
4278 | /* |
4279 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
4280 | * In that cases, pages are freed continuously and we can expect pages | |
4281 | * are in the same memcg. All these calls itself limits the number of | |
4282 | * pages freed at once, then uncharge_start/end() is called properly. | |
4283 | * This may be called prural(2) times in a context, | |
4284 | */ | |
4285 | ||
4286 | void mem_cgroup_uncharge_start(void) | |
4287 | { | |
4288 | current->memcg_batch.do_batch++; | |
4289 | /* We can do nest. */ | |
4290 | if (current->memcg_batch.do_batch == 1) { | |
4291 | current->memcg_batch.memcg = NULL; | |
7ffd4ca7 JW |
4292 | current->memcg_batch.nr_pages = 0; |
4293 | current->memcg_batch.memsw_nr_pages = 0; | |
569b846d KH |
4294 | } |
4295 | } | |
4296 | ||
4297 | void mem_cgroup_uncharge_end(void) | |
4298 | { | |
4299 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
4300 | ||
4301 | if (!batch->do_batch) | |
4302 | return; | |
4303 | ||
4304 | batch->do_batch--; | |
4305 | if (batch->do_batch) /* If stacked, do nothing. */ | |
4306 | return; | |
4307 | ||
4308 | if (!batch->memcg) | |
4309 | return; | |
4310 | /* | |
4311 | * This "batch->memcg" is valid without any css_get/put etc... | |
4312 | * bacause we hide charges behind us. | |
4313 | */ | |
7ffd4ca7 JW |
4314 | if (batch->nr_pages) |
4315 | res_counter_uncharge(&batch->memcg->res, | |
4316 | batch->nr_pages * PAGE_SIZE); | |
4317 | if (batch->memsw_nr_pages) | |
4318 | res_counter_uncharge(&batch->memcg->memsw, | |
4319 | batch->memsw_nr_pages * PAGE_SIZE); | |
3c11ecf4 | 4320 | memcg_oom_recover(batch->memcg); |
569b846d KH |
4321 | /* forget this pointer (for sanity check) */ |
4322 | batch->memcg = NULL; | |
4323 | } | |
4324 | ||
e767e056 | 4325 | #ifdef CONFIG_SWAP |
8c7c6e34 | 4326 | /* |
e767e056 | 4327 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
4328 | * memcg information is recorded to swap_cgroup of "ent" |
4329 | */ | |
8a9478ca KH |
4330 | void |
4331 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
4332 | { |
4333 | struct mem_cgroup *memcg; | |
8a9478ca KH |
4334 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
4335 | ||
4336 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
4337 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
4338 | ||
0030f535 | 4339 | memcg = __mem_cgroup_uncharge_common(page, ctype, false); |
8c7c6e34 | 4340 | |
f75ca962 KH |
4341 | /* |
4342 | * record memcg information, if swapout && memcg != NULL, | |
4050377b | 4343 | * css_get() was called in uncharge(). |
f75ca962 KH |
4344 | */ |
4345 | if (do_swap_account && swapout && memcg) | |
a3b2d692 | 4346 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 | 4347 | } |
e767e056 | 4348 | #endif |
8c7c6e34 | 4349 | |
c255a458 | 4350 | #ifdef CONFIG_MEMCG_SWAP |
8c7c6e34 KH |
4351 | /* |
4352 | * called from swap_entry_free(). remove record in swap_cgroup and | |
4353 | * uncharge "memsw" account. | |
4354 | */ | |
4355 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 4356 | { |
8c7c6e34 | 4357 | struct mem_cgroup *memcg; |
a3b2d692 | 4358 | unsigned short id; |
8c7c6e34 KH |
4359 | |
4360 | if (!do_swap_account) | |
4361 | return; | |
4362 | ||
a3b2d692 KH |
4363 | id = swap_cgroup_record(ent, 0); |
4364 | rcu_read_lock(); | |
4365 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 4366 | if (memcg) { |
a3b2d692 KH |
4367 | /* |
4368 | * We uncharge this because swap is freed. | |
4369 | * This memcg can be obsolete one. We avoid calling css_tryget | |
4370 | */ | |
0c3e73e8 | 4371 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 4372 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 4373 | mem_cgroup_swap_statistics(memcg, false); |
4050377b | 4374 | css_put(&memcg->css); |
8c7c6e34 | 4375 | } |
a3b2d692 | 4376 | rcu_read_unlock(); |
d13d1443 | 4377 | } |
02491447 DN |
4378 | |
4379 | /** | |
4380 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
4381 | * @entry: swap entry to be moved | |
4382 | * @from: mem_cgroup which the entry is moved from | |
4383 | * @to: mem_cgroup which the entry is moved to | |
4384 | * | |
4385 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
4386 | * as the mem_cgroup's id of @from. | |
4387 | * | |
4388 | * Returns 0 on success, -EINVAL on failure. | |
4389 | * | |
4390 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
4391 | * both res and memsw, and called css_get(). | |
4392 | */ | |
4393 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 4394 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
4395 | { |
4396 | unsigned short old_id, new_id; | |
4397 | ||
4398 | old_id = css_id(&from->css); | |
4399 | new_id = css_id(&to->css); | |
4400 | ||
4401 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 4402 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 4403 | mem_cgroup_swap_statistics(to, true); |
02491447 | 4404 | /* |
483c30b5 DN |
4405 | * This function is only called from task migration context now. |
4406 | * It postpones res_counter and refcount handling till the end | |
4407 | * of task migration(mem_cgroup_clear_mc()) for performance | |
4050377b LZ |
4408 | * improvement. But we cannot postpone css_get(to) because if |
4409 | * the process that has been moved to @to does swap-in, the | |
4410 | * refcount of @to might be decreased to 0. | |
4411 | * | |
4412 | * We are in attach() phase, so the cgroup is guaranteed to be | |
4413 | * alive, so we can just call css_get(). | |
02491447 | 4414 | */ |
4050377b | 4415 | css_get(&to->css); |
02491447 DN |
4416 | return 0; |
4417 | } | |
4418 | return -EINVAL; | |
4419 | } | |
4420 | #else | |
4421 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 4422 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
4423 | { |
4424 | return -EINVAL; | |
4425 | } | |
8c7c6e34 | 4426 | #endif |
d13d1443 | 4427 | |
ae41be37 | 4428 | /* |
01b1ae63 KH |
4429 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
4430 | * page belongs to. | |
ae41be37 | 4431 | */ |
0030f535 JW |
4432 | void mem_cgroup_prepare_migration(struct page *page, struct page *newpage, |
4433 | struct mem_cgroup **memcgp) | |
ae41be37 | 4434 | { |
c0ff4b85 | 4435 | struct mem_cgroup *memcg = NULL; |
b32967ff | 4436 | unsigned int nr_pages = 1; |
7ec99d62 | 4437 | struct page_cgroup *pc; |
ac39cf8c | 4438 | enum charge_type ctype; |
8869b8f6 | 4439 | |
72835c86 | 4440 | *memcgp = NULL; |
56039efa | 4441 | |
f8d66542 | 4442 | if (mem_cgroup_disabled()) |
0030f535 | 4443 | return; |
4077960e | 4444 | |
b32967ff MG |
4445 | if (PageTransHuge(page)) |
4446 | nr_pages <<= compound_order(page); | |
4447 | ||
52d4b9ac KH |
4448 | pc = lookup_page_cgroup(page); |
4449 | lock_page_cgroup(pc); | |
4450 | if (PageCgroupUsed(pc)) { | |
c0ff4b85 R |
4451 | memcg = pc->mem_cgroup; |
4452 | css_get(&memcg->css); | |
ac39cf8c AM |
4453 | /* |
4454 | * At migrating an anonymous page, its mapcount goes down | |
4455 | * to 0 and uncharge() will be called. But, even if it's fully | |
4456 | * unmapped, migration may fail and this page has to be | |
4457 | * charged again. We set MIGRATION flag here and delay uncharge | |
4458 | * until end_migration() is called | |
4459 | * | |
4460 | * Corner Case Thinking | |
4461 | * A) | |
4462 | * When the old page was mapped as Anon and it's unmap-and-freed | |
4463 | * while migration was ongoing. | |
4464 | * If unmap finds the old page, uncharge() of it will be delayed | |
4465 | * until end_migration(). If unmap finds a new page, it's | |
4466 | * uncharged when it make mapcount to be 1->0. If unmap code | |
4467 | * finds swap_migration_entry, the new page will not be mapped | |
4468 | * and end_migration() will find it(mapcount==0). | |
4469 | * | |
4470 | * B) | |
4471 | * When the old page was mapped but migraion fails, the kernel | |
4472 | * remaps it. A charge for it is kept by MIGRATION flag even | |
4473 | * if mapcount goes down to 0. We can do remap successfully | |
4474 | * without charging it again. | |
4475 | * | |
4476 | * C) | |
4477 | * The "old" page is under lock_page() until the end of | |
4478 | * migration, so, the old page itself will not be swapped-out. | |
4479 | * If the new page is swapped out before end_migraton, our | |
4480 | * hook to usual swap-out path will catch the event. | |
4481 | */ | |
4482 | if (PageAnon(page)) | |
4483 | SetPageCgroupMigration(pc); | |
e8589cc1 | 4484 | } |
52d4b9ac | 4485 | unlock_page_cgroup(pc); |
ac39cf8c AM |
4486 | /* |
4487 | * If the page is not charged at this point, | |
4488 | * we return here. | |
4489 | */ | |
c0ff4b85 | 4490 | if (!memcg) |
0030f535 | 4491 | return; |
01b1ae63 | 4492 | |
72835c86 | 4493 | *memcgp = memcg; |
ac39cf8c AM |
4494 | /* |
4495 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
4496 | * is called before end_migration, we can catch all events on this new | |
4497 | * page. In the case new page is migrated but not remapped, new page's | |
4498 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
4499 | */ | |
ac39cf8c | 4500 | if (PageAnon(page)) |
41326c17 | 4501 | ctype = MEM_CGROUP_CHARGE_TYPE_ANON; |
ac39cf8c | 4502 | else |
62ba7442 | 4503 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; |
0030f535 JW |
4504 | /* |
4505 | * The page is committed to the memcg, but it's not actually | |
4506 | * charged to the res_counter since we plan on replacing the | |
4507 | * old one and only one page is going to be left afterwards. | |
4508 | */ | |
b32967ff | 4509 | __mem_cgroup_commit_charge(memcg, newpage, nr_pages, ctype, false); |
ae41be37 | 4510 | } |
8869b8f6 | 4511 | |
69029cd5 | 4512 | /* remove redundant charge if migration failed*/ |
c0ff4b85 | 4513 | void mem_cgroup_end_migration(struct mem_cgroup *memcg, |
50de1dd9 | 4514 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 4515 | { |
ac39cf8c | 4516 | struct page *used, *unused; |
01b1ae63 | 4517 | struct page_cgroup *pc; |
b2402857 | 4518 | bool anon; |
01b1ae63 | 4519 | |
c0ff4b85 | 4520 | if (!memcg) |
01b1ae63 | 4521 | return; |
b25ed609 | 4522 | |
50de1dd9 | 4523 | if (!migration_ok) { |
ac39cf8c AM |
4524 | used = oldpage; |
4525 | unused = newpage; | |
01b1ae63 | 4526 | } else { |
ac39cf8c | 4527 | used = newpage; |
01b1ae63 KH |
4528 | unused = oldpage; |
4529 | } | |
0030f535 | 4530 | anon = PageAnon(used); |
7d188958 JW |
4531 | __mem_cgroup_uncharge_common(unused, |
4532 | anon ? MEM_CGROUP_CHARGE_TYPE_ANON | |
4533 | : MEM_CGROUP_CHARGE_TYPE_CACHE, | |
4534 | true); | |
0030f535 | 4535 | css_put(&memcg->css); |
69029cd5 | 4536 | /* |
ac39cf8c AM |
4537 | * We disallowed uncharge of pages under migration because mapcount |
4538 | * of the page goes down to zero, temporarly. | |
4539 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 4540 | */ |
ac39cf8c AM |
4541 | pc = lookup_page_cgroup(oldpage); |
4542 | lock_page_cgroup(pc); | |
4543 | ClearPageCgroupMigration(pc); | |
4544 | unlock_page_cgroup(pc); | |
ac39cf8c | 4545 | |
01b1ae63 | 4546 | /* |
ac39cf8c AM |
4547 | * If a page is a file cache, radix-tree replacement is very atomic |
4548 | * and we can skip this check. When it was an Anon page, its mapcount | |
4549 | * goes down to 0. But because we added MIGRATION flage, it's not | |
4550 | * uncharged yet. There are several case but page->mapcount check | |
4551 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
4552 | * check. (see prepare_charge() also) | |
69029cd5 | 4553 | */ |
b2402857 | 4554 | if (anon) |
ac39cf8c | 4555 | mem_cgroup_uncharge_page(used); |
ae41be37 | 4556 | } |
78fb7466 | 4557 | |
ab936cbc KH |
4558 | /* |
4559 | * At replace page cache, newpage is not under any memcg but it's on | |
4560 | * LRU. So, this function doesn't touch res_counter but handles LRU | |
4561 | * in correct way. Both pages are locked so we cannot race with uncharge. | |
4562 | */ | |
4563 | void mem_cgroup_replace_page_cache(struct page *oldpage, | |
4564 | struct page *newpage) | |
4565 | { | |
bde05d1c | 4566 | struct mem_cgroup *memcg = NULL; |
ab936cbc | 4567 | struct page_cgroup *pc; |
ab936cbc | 4568 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
ab936cbc KH |
4569 | |
4570 | if (mem_cgroup_disabled()) | |
4571 | return; | |
4572 | ||
4573 | pc = lookup_page_cgroup(oldpage); | |
4574 | /* fix accounting on old pages */ | |
4575 | lock_page_cgroup(pc); | |
bde05d1c HD |
4576 | if (PageCgroupUsed(pc)) { |
4577 | memcg = pc->mem_cgroup; | |
b070e65c | 4578 | mem_cgroup_charge_statistics(memcg, oldpage, false, -1); |
bde05d1c HD |
4579 | ClearPageCgroupUsed(pc); |
4580 | } | |
ab936cbc KH |
4581 | unlock_page_cgroup(pc); |
4582 | ||
bde05d1c HD |
4583 | /* |
4584 | * When called from shmem_replace_page(), in some cases the | |
4585 | * oldpage has already been charged, and in some cases not. | |
4586 | */ | |
4587 | if (!memcg) | |
4588 | return; | |
ab936cbc KH |
4589 | /* |
4590 | * Even if newpage->mapping was NULL before starting replacement, | |
4591 | * the newpage may be on LRU(or pagevec for LRU) already. We lock | |
4592 | * LRU while we overwrite pc->mem_cgroup. | |
4593 | */ | |
ce587e65 | 4594 | __mem_cgroup_commit_charge(memcg, newpage, 1, type, true); |
ab936cbc KH |
4595 | } |
4596 | ||
f212ad7c DN |
4597 | #ifdef CONFIG_DEBUG_VM |
4598 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
4599 | { | |
4600 | struct page_cgroup *pc; | |
4601 | ||
4602 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
4603 | /* |
4604 | * Can be NULL while feeding pages into the page allocator for | |
4605 | * the first time, i.e. during boot or memory hotplug; | |
4606 | * or when mem_cgroup_disabled(). | |
4607 | */ | |
f212ad7c DN |
4608 | if (likely(pc) && PageCgroupUsed(pc)) |
4609 | return pc; | |
4610 | return NULL; | |
4611 | } | |
4612 | ||
4613 | bool mem_cgroup_bad_page_check(struct page *page) | |
4614 | { | |
4615 | if (mem_cgroup_disabled()) | |
4616 | return false; | |
4617 | ||
4618 | return lookup_page_cgroup_used(page) != NULL; | |
4619 | } | |
4620 | ||
4621 | void mem_cgroup_print_bad_page(struct page *page) | |
4622 | { | |
4623 | struct page_cgroup *pc; | |
4624 | ||
4625 | pc = lookup_page_cgroup_used(page); | |
4626 | if (pc) { | |
d045197f AM |
4627 | pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n", |
4628 | pc, pc->flags, pc->mem_cgroup); | |
f212ad7c DN |
4629 | } |
4630 | } | |
4631 | #endif | |
4632 | ||
d38d2a75 | 4633 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 4634 | unsigned long long val) |
628f4235 | 4635 | { |
81d39c20 | 4636 | int retry_count; |
3c11ecf4 | 4637 | u64 memswlimit, memlimit; |
628f4235 | 4638 | int ret = 0; |
81d39c20 KH |
4639 | int children = mem_cgroup_count_children(memcg); |
4640 | u64 curusage, oldusage; | |
3c11ecf4 | 4641 | int enlarge; |
81d39c20 KH |
4642 | |
4643 | /* | |
4644 | * For keeping hierarchical_reclaim simple, how long we should retry | |
4645 | * is depends on callers. We set our retry-count to be function | |
4646 | * of # of children which we should visit in this loop. | |
4647 | */ | |
4648 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
4649 | ||
4650 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 4651 | |
3c11ecf4 | 4652 | enlarge = 0; |
8c7c6e34 | 4653 | while (retry_count) { |
628f4235 KH |
4654 | if (signal_pending(current)) { |
4655 | ret = -EINTR; | |
4656 | break; | |
4657 | } | |
8c7c6e34 KH |
4658 | /* |
4659 | * Rather than hide all in some function, I do this in | |
4660 | * open coded manner. You see what this really does. | |
aaad153e | 4661 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
4662 | */ |
4663 | mutex_lock(&set_limit_mutex); | |
4664 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
4665 | if (memswlimit < val) { | |
4666 | ret = -EINVAL; | |
4667 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
4668 | break; |
4669 | } | |
3c11ecf4 KH |
4670 | |
4671 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4672 | if (memlimit < val) | |
4673 | enlarge = 1; | |
4674 | ||
8c7c6e34 | 4675 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
4676 | if (!ret) { |
4677 | if (memswlimit == val) | |
4678 | memcg->memsw_is_minimum = true; | |
4679 | else | |
4680 | memcg->memsw_is_minimum = false; | |
4681 | } | |
8c7c6e34 KH |
4682 | mutex_unlock(&set_limit_mutex); |
4683 | ||
4684 | if (!ret) | |
4685 | break; | |
4686 | ||
5660048c JW |
4687 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
4688 | MEM_CGROUP_RECLAIM_SHRINK); | |
81d39c20 KH |
4689 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
4690 | /* Usage is reduced ? */ | |
f894ffa8 | 4691 | if (curusage >= oldusage) |
81d39c20 KH |
4692 | retry_count--; |
4693 | else | |
4694 | oldusage = curusage; | |
8c7c6e34 | 4695 | } |
3c11ecf4 KH |
4696 | if (!ret && enlarge) |
4697 | memcg_oom_recover(memcg); | |
14797e23 | 4698 | |
8c7c6e34 KH |
4699 | return ret; |
4700 | } | |
4701 | ||
338c8431 LZ |
4702 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
4703 | unsigned long long val) | |
8c7c6e34 | 4704 | { |
81d39c20 | 4705 | int retry_count; |
3c11ecf4 | 4706 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
4707 | int children = mem_cgroup_count_children(memcg); |
4708 | int ret = -EBUSY; | |
3c11ecf4 | 4709 | int enlarge = 0; |
8c7c6e34 | 4710 | |
81d39c20 | 4711 | /* see mem_cgroup_resize_res_limit */ |
f894ffa8 | 4712 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; |
81d39c20 | 4713 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
8c7c6e34 KH |
4714 | while (retry_count) { |
4715 | if (signal_pending(current)) { | |
4716 | ret = -EINTR; | |
4717 | break; | |
4718 | } | |
4719 | /* | |
4720 | * Rather than hide all in some function, I do this in | |
4721 | * open coded manner. You see what this really does. | |
aaad153e | 4722 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
4723 | */ |
4724 | mutex_lock(&set_limit_mutex); | |
4725 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4726 | if (memlimit > val) { | |
4727 | ret = -EINVAL; | |
4728 | mutex_unlock(&set_limit_mutex); | |
4729 | break; | |
4730 | } | |
3c11ecf4 KH |
4731 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
4732 | if (memswlimit < val) | |
4733 | enlarge = 1; | |
8c7c6e34 | 4734 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
4735 | if (!ret) { |
4736 | if (memlimit == val) | |
4737 | memcg->memsw_is_minimum = true; | |
4738 | else | |
4739 | memcg->memsw_is_minimum = false; | |
4740 | } | |
8c7c6e34 KH |
4741 | mutex_unlock(&set_limit_mutex); |
4742 | ||
4743 | if (!ret) | |
4744 | break; | |
4745 | ||
5660048c JW |
4746 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
4747 | MEM_CGROUP_RECLAIM_NOSWAP | | |
4748 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 4749 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 4750 | /* Usage is reduced ? */ |
8c7c6e34 | 4751 | if (curusage >= oldusage) |
628f4235 | 4752 | retry_count--; |
81d39c20 KH |
4753 | else |
4754 | oldusage = curusage; | |
628f4235 | 4755 | } |
3c11ecf4 KH |
4756 | if (!ret && enlarge) |
4757 | memcg_oom_recover(memcg); | |
628f4235 KH |
4758 | return ret; |
4759 | } | |
4760 | ||
0608f43d AM |
4761 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
4762 | gfp_t gfp_mask, | |
4763 | unsigned long *total_scanned) | |
4764 | { | |
4765 | unsigned long nr_reclaimed = 0; | |
4766 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
4767 | unsigned long reclaimed; | |
4768 | int loop = 0; | |
4769 | struct mem_cgroup_tree_per_zone *mctz; | |
4770 | unsigned long long excess; | |
4771 | unsigned long nr_scanned; | |
4772 | ||
4773 | if (order > 0) | |
4774 | return 0; | |
4775 | ||
4776 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
4777 | /* | |
4778 | * This loop can run a while, specially if mem_cgroup's continuously | |
4779 | * keep exceeding their soft limit and putting the system under | |
4780 | * pressure | |
4781 | */ | |
4782 | do { | |
4783 | if (next_mz) | |
4784 | mz = next_mz; | |
4785 | else | |
4786 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
4787 | if (!mz) | |
4788 | break; | |
4789 | ||
4790 | nr_scanned = 0; | |
4791 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
4792 | gfp_mask, &nr_scanned); | |
4793 | nr_reclaimed += reclaimed; | |
4794 | *total_scanned += nr_scanned; | |
4795 | spin_lock(&mctz->lock); | |
4796 | ||
4797 | /* | |
4798 | * If we failed to reclaim anything from this memory cgroup | |
4799 | * it is time to move on to the next cgroup | |
4800 | */ | |
4801 | next_mz = NULL; | |
4802 | if (!reclaimed) { | |
4803 | do { | |
4804 | /* | |
4805 | * Loop until we find yet another one. | |
4806 | * | |
4807 | * By the time we get the soft_limit lock | |
4808 | * again, someone might have aded the | |
4809 | * group back on the RB tree. Iterate to | |
4810 | * make sure we get a different mem. | |
4811 | * mem_cgroup_largest_soft_limit_node returns | |
4812 | * NULL if no other cgroup is present on | |
4813 | * the tree | |
4814 | */ | |
4815 | next_mz = | |
4816 | __mem_cgroup_largest_soft_limit_node(mctz); | |
4817 | if (next_mz == mz) | |
4818 | css_put(&next_mz->memcg->css); | |
4819 | else /* next_mz == NULL or other memcg */ | |
4820 | break; | |
4821 | } while (1); | |
4822 | } | |
4823 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); | |
4824 | excess = res_counter_soft_limit_excess(&mz->memcg->res); | |
4825 | /* | |
4826 | * One school of thought says that we should not add | |
4827 | * back the node to the tree if reclaim returns 0. | |
4828 | * But our reclaim could return 0, simply because due | |
4829 | * to priority we are exposing a smaller subset of | |
4830 | * memory to reclaim from. Consider this as a longer | |
4831 | * term TODO. | |
4832 | */ | |
4833 | /* If excess == 0, no tree ops */ | |
4834 | __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess); | |
4835 | spin_unlock(&mctz->lock); | |
4836 | css_put(&mz->memcg->css); | |
4837 | loop++; | |
4838 | /* | |
4839 | * Could not reclaim anything and there are no more | |
4840 | * mem cgroups to try or we seem to be looping without | |
4841 | * reclaiming anything. | |
4842 | */ | |
4843 | if (!nr_reclaimed && | |
4844 | (next_mz == NULL || | |
4845 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
4846 | break; | |
4847 | } while (!nr_reclaimed); | |
4848 | if (next_mz) | |
4849 | css_put(&next_mz->memcg->css); | |
4850 | return nr_reclaimed; | |
4851 | } | |
4852 | ||
2ef37d3f MH |
4853 | /** |
4854 | * mem_cgroup_force_empty_list - clears LRU of a group | |
4855 | * @memcg: group to clear | |
4856 | * @node: NUMA node | |
4857 | * @zid: zone id | |
4858 | * @lru: lru to to clear | |
4859 | * | |
3c935d18 | 4860 | * Traverse a specified page_cgroup list and try to drop them all. This doesn't |
2ef37d3f MH |
4861 | * reclaim the pages page themselves - pages are moved to the parent (or root) |
4862 | * group. | |
cc847582 | 4863 | */ |
2ef37d3f | 4864 | static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
08e552c6 | 4865 | int node, int zid, enum lru_list lru) |
cc847582 | 4866 | { |
bea8c150 | 4867 | struct lruvec *lruvec; |
2ef37d3f | 4868 | unsigned long flags; |
072c56c1 | 4869 | struct list_head *list; |
925b7673 JW |
4870 | struct page *busy; |
4871 | struct zone *zone; | |
072c56c1 | 4872 | |
08e552c6 | 4873 | zone = &NODE_DATA(node)->node_zones[zid]; |
bea8c150 HD |
4874 | lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
4875 | list = &lruvec->lists[lru]; | |
cc847582 | 4876 | |
f817ed48 | 4877 | busy = NULL; |
2ef37d3f | 4878 | do { |
925b7673 | 4879 | struct page_cgroup *pc; |
5564e88b JW |
4880 | struct page *page; |
4881 | ||
08e552c6 | 4882 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 4883 | if (list_empty(list)) { |
08e552c6 | 4884 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 4885 | break; |
f817ed48 | 4886 | } |
925b7673 JW |
4887 | page = list_entry(list->prev, struct page, lru); |
4888 | if (busy == page) { | |
4889 | list_move(&page->lru, list); | |
648bcc77 | 4890 | busy = NULL; |
08e552c6 | 4891 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
4892 | continue; |
4893 | } | |
08e552c6 | 4894 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 4895 | |
925b7673 | 4896 | pc = lookup_page_cgroup(page); |
5564e88b | 4897 | |
3c935d18 | 4898 | if (mem_cgroup_move_parent(page, pc, memcg)) { |
f817ed48 | 4899 | /* found lock contention or "pc" is obsolete. */ |
925b7673 | 4900 | busy = page; |
f817ed48 KH |
4901 | cond_resched(); |
4902 | } else | |
4903 | busy = NULL; | |
2ef37d3f | 4904 | } while (!list_empty(list)); |
cc847582 KH |
4905 | } |
4906 | ||
4907 | /* | |
c26251f9 MH |
4908 | * make mem_cgroup's charge to be 0 if there is no task by moving |
4909 | * all the charges and pages to the parent. | |
cc847582 | 4910 | * This enables deleting this mem_cgroup. |
c26251f9 MH |
4911 | * |
4912 | * Caller is responsible for holding css reference on the memcg. | |
cc847582 | 4913 | */ |
ab5196c2 | 4914 | static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg) |
cc847582 | 4915 | { |
c26251f9 | 4916 | int node, zid; |
bea207c8 | 4917 | u64 usage; |
f817ed48 | 4918 | |
fce66477 | 4919 | do { |
52d4b9ac KH |
4920 | /* This is for making all *used* pages to be on LRU. */ |
4921 | lru_add_drain_all(); | |
c0ff4b85 | 4922 | drain_all_stock_sync(memcg); |
c0ff4b85 | 4923 | mem_cgroup_start_move(memcg); |
31aaea4a | 4924 | for_each_node_state(node, N_MEMORY) { |
2ef37d3f | 4925 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
f156ab93 HD |
4926 | enum lru_list lru; |
4927 | for_each_lru(lru) { | |
2ef37d3f | 4928 | mem_cgroup_force_empty_list(memcg, |
f156ab93 | 4929 | node, zid, lru); |
f817ed48 | 4930 | } |
1ecaab2b | 4931 | } |
f817ed48 | 4932 | } |
c0ff4b85 R |
4933 | mem_cgroup_end_move(memcg); |
4934 | memcg_oom_recover(memcg); | |
52d4b9ac | 4935 | cond_resched(); |
f817ed48 | 4936 | |
2ef37d3f | 4937 | /* |
bea207c8 GC |
4938 | * Kernel memory may not necessarily be trackable to a specific |
4939 | * process. So they are not migrated, and therefore we can't | |
4940 | * expect their value to drop to 0 here. | |
4941 | * Having res filled up with kmem only is enough. | |
4942 | * | |
2ef37d3f MH |
4943 | * This is a safety check because mem_cgroup_force_empty_list |
4944 | * could have raced with mem_cgroup_replace_page_cache callers | |
4945 | * so the lru seemed empty but the page could have been added | |
4946 | * right after the check. RES_USAGE should be safe as we always | |
4947 | * charge before adding to the LRU. | |
4948 | */ | |
bea207c8 GC |
4949 | usage = res_counter_read_u64(&memcg->res, RES_USAGE) - |
4950 | res_counter_read_u64(&memcg->kmem, RES_USAGE); | |
4951 | } while (usage > 0); | |
c26251f9 MH |
4952 | } |
4953 | ||
b5f99b53 GC |
4954 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
4955 | { | |
696ac172 JW |
4956 | lockdep_assert_held(&memcg_create_mutex); |
4957 | /* | |
4958 | * The lock does not prevent addition or deletion to the list | |
4959 | * of children, but it prevents a new child from being | |
4960 | * initialized based on this parent in css_online(), so it's | |
4961 | * enough to decide whether hierarchically inherited | |
4962 | * attributes can still be changed or not. | |
4963 | */ | |
4964 | return memcg->use_hierarchy && | |
4965 | !list_empty(&memcg->css.cgroup->children); | |
b5f99b53 GC |
4966 | } |
4967 | ||
c26251f9 MH |
4968 | /* |
4969 | * Reclaims as many pages from the given memcg as possible and moves | |
4970 | * the rest to the parent. | |
4971 | * | |
4972 | * Caller is responsible for holding css reference for memcg. | |
4973 | */ | |
4974 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
4975 | { | |
4976 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
4977 | struct cgroup *cgrp = memcg->css.cgroup; | |
f817ed48 | 4978 | |
c1e862c1 | 4979 | /* returns EBUSY if there is a task or if we come here twice. */ |
c26251f9 MH |
4980 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
4981 | return -EBUSY; | |
4982 | ||
c1e862c1 KH |
4983 | /* we call try-to-free pages for make this cgroup empty */ |
4984 | lru_add_drain_all(); | |
f817ed48 | 4985 | /* try to free all pages in this cgroup */ |
569530fb | 4986 | while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) { |
f817ed48 | 4987 | int progress; |
c1e862c1 | 4988 | |
c26251f9 MH |
4989 | if (signal_pending(current)) |
4990 | return -EINTR; | |
4991 | ||
c0ff4b85 | 4992 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, |
185efc0f | 4993 | false); |
c1e862c1 | 4994 | if (!progress) { |
f817ed48 | 4995 | nr_retries--; |
c1e862c1 | 4996 | /* maybe some writeback is necessary */ |
8aa7e847 | 4997 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 4998 | } |
f817ed48 KH |
4999 | |
5000 | } | |
08e552c6 | 5001 | lru_add_drain(); |
ab5196c2 MH |
5002 | mem_cgroup_reparent_charges(memcg); |
5003 | ||
5004 | return 0; | |
cc847582 KH |
5005 | } |
5006 | ||
182446d0 TH |
5007 | static int mem_cgroup_force_empty_write(struct cgroup_subsys_state *css, |
5008 | unsigned int event) | |
c1e862c1 | 5009 | { |
182446d0 | 5010 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c26251f9 | 5011 | |
d8423011 MH |
5012 | if (mem_cgroup_is_root(memcg)) |
5013 | return -EINVAL; | |
c33bd835 | 5014 | return mem_cgroup_force_empty(memcg); |
c1e862c1 KH |
5015 | } |
5016 | ||
182446d0 TH |
5017 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
5018 | struct cftype *cft) | |
18f59ea7 | 5019 | { |
182446d0 | 5020 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
5021 | } |
5022 | ||
182446d0 TH |
5023 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
5024 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
5025 | { |
5026 | int retval = 0; | |
182446d0 | 5027 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
63876986 | 5028 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(css_parent(&memcg->css)); |
18f59ea7 | 5029 | |
0999821b | 5030 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
5031 | |
5032 | if (memcg->use_hierarchy == val) | |
5033 | goto out; | |
5034 | ||
18f59ea7 | 5035 | /* |
af901ca1 | 5036 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
5037 | * in the child subtrees. If it is unset, then the change can |
5038 | * occur, provided the current cgroup has no children. | |
5039 | * | |
5040 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
5041 | * set if there are no children. | |
5042 | */ | |
c0ff4b85 | 5043 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 5044 | (val == 1 || val == 0)) { |
696ac172 | 5045 | if (list_empty(&memcg->css.cgroup->children)) |
c0ff4b85 | 5046 | memcg->use_hierarchy = val; |
18f59ea7 BS |
5047 | else |
5048 | retval = -EBUSY; | |
5049 | } else | |
5050 | retval = -EINVAL; | |
567fb435 GC |
5051 | |
5052 | out: | |
0999821b | 5053 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
5054 | |
5055 | return retval; | |
5056 | } | |
5057 | ||
0c3e73e8 | 5058 | |
c0ff4b85 | 5059 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg, |
7a159cc9 | 5060 | enum mem_cgroup_stat_index idx) |
0c3e73e8 | 5061 | { |
7d74b06f | 5062 | struct mem_cgroup *iter; |
7a159cc9 | 5063 | long val = 0; |
0c3e73e8 | 5064 | |
7a159cc9 | 5065 | /* Per-cpu values can be negative, use a signed accumulator */ |
c0ff4b85 | 5066 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f KH |
5067 | val += mem_cgroup_read_stat(iter, idx); |
5068 | ||
5069 | if (val < 0) /* race ? */ | |
5070 | val = 0; | |
5071 | return val; | |
0c3e73e8 BS |
5072 | } |
5073 | ||
c0ff4b85 | 5074 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
104f3928 | 5075 | { |
7d74b06f | 5076 | u64 val; |
104f3928 | 5077 | |
c0ff4b85 | 5078 | if (!mem_cgroup_is_root(memcg)) { |
104f3928 | 5079 | if (!swap) |
65c64ce8 | 5080 | return res_counter_read_u64(&memcg->res, RES_USAGE); |
104f3928 | 5081 | else |
65c64ce8 | 5082 | return res_counter_read_u64(&memcg->memsw, RES_USAGE); |
104f3928 KS |
5083 | } |
5084 | ||
b070e65c DR |
5085 | /* |
5086 | * Transparent hugepages are still accounted for in MEM_CGROUP_STAT_RSS | |
5087 | * as well as in MEM_CGROUP_STAT_RSS_HUGE. | |
5088 | */ | |
c0ff4b85 R |
5089 | val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE); |
5090 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS); | |
104f3928 | 5091 | |
7d74b06f | 5092 | if (swap) |
bff6bb83 | 5093 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP); |
104f3928 KS |
5094 | |
5095 | return val << PAGE_SHIFT; | |
5096 | } | |
5097 | ||
182446d0 TH |
5098 | static ssize_t mem_cgroup_read(struct cgroup_subsys_state *css, |
5099 | struct cftype *cft, struct file *file, | |
5100 | char __user *buf, size_t nbytes, loff_t *ppos) | |
8cdea7c0 | 5101 | { |
182446d0 | 5102 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
af36f906 | 5103 | char str[64]; |
104f3928 | 5104 | u64 val; |
86ae53e1 GC |
5105 | int name, len; |
5106 | enum res_type type; | |
8c7c6e34 KH |
5107 | |
5108 | type = MEMFILE_TYPE(cft->private); | |
5109 | name = MEMFILE_ATTR(cft->private); | |
af36f906 | 5110 | |
8c7c6e34 KH |
5111 | switch (type) { |
5112 | case _MEM: | |
104f3928 | 5113 | if (name == RES_USAGE) |
c0ff4b85 | 5114 | val = mem_cgroup_usage(memcg, false); |
104f3928 | 5115 | else |
c0ff4b85 | 5116 | val = res_counter_read_u64(&memcg->res, name); |
8c7c6e34 KH |
5117 | break; |
5118 | case _MEMSWAP: | |
104f3928 | 5119 | if (name == RES_USAGE) |
c0ff4b85 | 5120 | val = mem_cgroup_usage(memcg, true); |
104f3928 | 5121 | else |
c0ff4b85 | 5122 | val = res_counter_read_u64(&memcg->memsw, name); |
8c7c6e34 | 5123 | break; |
510fc4e1 GC |
5124 | case _KMEM: |
5125 | val = res_counter_read_u64(&memcg->kmem, name); | |
5126 | break; | |
8c7c6e34 KH |
5127 | default: |
5128 | BUG(); | |
8c7c6e34 | 5129 | } |
af36f906 TH |
5130 | |
5131 | len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val); | |
5132 | return simple_read_from_buffer(buf, nbytes, ppos, str, len); | |
8cdea7c0 | 5133 | } |
510fc4e1 | 5134 | |
182446d0 | 5135 | static int memcg_update_kmem_limit(struct cgroup_subsys_state *css, u64 val) |
510fc4e1 GC |
5136 | { |
5137 | int ret = -EINVAL; | |
5138 | #ifdef CONFIG_MEMCG_KMEM | |
182446d0 | 5139 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
510fc4e1 GC |
5140 | /* |
5141 | * For simplicity, we won't allow this to be disabled. It also can't | |
5142 | * be changed if the cgroup has children already, or if tasks had | |
5143 | * already joined. | |
5144 | * | |
5145 | * If tasks join before we set the limit, a person looking at | |
5146 | * kmem.usage_in_bytes will have no way to determine when it took | |
5147 | * place, which makes the value quite meaningless. | |
5148 | * | |
5149 | * After it first became limited, changes in the value of the limit are | |
5150 | * of course permitted. | |
510fc4e1 | 5151 | */ |
0999821b | 5152 | mutex_lock(&memcg_create_mutex); |
510fc4e1 | 5153 | mutex_lock(&set_limit_mutex); |
6de5a8bf | 5154 | if (!memcg->kmem_account_flags && val != RES_COUNTER_MAX) { |
182446d0 | 5155 | if (cgroup_task_count(css->cgroup) || memcg_has_children(memcg)) { |
510fc4e1 GC |
5156 | ret = -EBUSY; |
5157 | goto out; | |
5158 | } | |
5159 | ret = res_counter_set_limit(&memcg->kmem, val); | |
5160 | VM_BUG_ON(ret); | |
5161 | ||
55007d84 GC |
5162 | ret = memcg_update_cache_sizes(memcg); |
5163 | if (ret) { | |
6de5a8bf | 5164 | res_counter_set_limit(&memcg->kmem, RES_COUNTER_MAX); |
55007d84 GC |
5165 | goto out; |
5166 | } | |
692e89ab GC |
5167 | static_key_slow_inc(&memcg_kmem_enabled_key); |
5168 | /* | |
5169 | * setting the active bit after the inc will guarantee no one | |
5170 | * starts accounting before all call sites are patched | |
5171 | */ | |
5172 | memcg_kmem_set_active(memcg); | |
510fc4e1 GC |
5173 | } else |
5174 | ret = res_counter_set_limit(&memcg->kmem, val); | |
5175 | out: | |
5176 | mutex_unlock(&set_limit_mutex); | |
0999821b | 5177 | mutex_unlock(&memcg_create_mutex); |
510fc4e1 GC |
5178 | #endif |
5179 | return ret; | |
5180 | } | |
5181 | ||
6d043990 | 5182 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 | 5183 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 5184 | { |
55007d84 | 5185 | int ret = 0; |
510fc4e1 GC |
5186 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
5187 | if (!parent) | |
55007d84 GC |
5188 | goto out; |
5189 | ||
510fc4e1 | 5190 | memcg->kmem_account_flags = parent->kmem_account_flags; |
a8964b9b GC |
5191 | /* |
5192 | * When that happen, we need to disable the static branch only on those | |
5193 | * memcgs that enabled it. To achieve this, we would be forced to | |
5194 | * complicate the code by keeping track of which memcgs were the ones | |
5195 | * that actually enabled limits, and which ones got it from its | |
5196 | * parents. | |
5197 | * | |
5198 | * It is a lot simpler just to do static_key_slow_inc() on every child | |
5199 | * that is accounted. | |
5200 | */ | |
55007d84 GC |
5201 | if (!memcg_kmem_is_active(memcg)) |
5202 | goto out; | |
5203 | ||
5204 | /* | |
10d5ebf4 LZ |
5205 | * __mem_cgroup_free() will issue static_key_slow_dec() because this |
5206 | * memcg is active already. If the later initialization fails then the | |
5207 | * cgroup core triggers the cleanup so we do not have to do it here. | |
55007d84 | 5208 | */ |
55007d84 GC |
5209 | static_key_slow_inc(&memcg_kmem_enabled_key); |
5210 | ||
5211 | mutex_lock(&set_limit_mutex); | |
425c598d | 5212 | memcg_stop_kmem_account(); |
55007d84 | 5213 | ret = memcg_update_cache_sizes(memcg); |
425c598d | 5214 | memcg_resume_kmem_account(); |
55007d84 | 5215 | mutex_unlock(&set_limit_mutex); |
55007d84 GC |
5216 | out: |
5217 | return ret; | |
510fc4e1 | 5218 | } |
6d043990 | 5219 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 5220 | |
628f4235 KH |
5221 | /* |
5222 | * The user of this function is... | |
5223 | * RES_LIMIT. | |
5224 | */ | |
182446d0 | 5225 | static int mem_cgroup_write(struct cgroup_subsys_state *css, struct cftype *cft, |
856c13aa | 5226 | const char *buffer) |
8cdea7c0 | 5227 | { |
182446d0 | 5228 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
86ae53e1 GC |
5229 | enum res_type type; |
5230 | int name; | |
628f4235 KH |
5231 | unsigned long long val; |
5232 | int ret; | |
5233 | ||
8c7c6e34 KH |
5234 | type = MEMFILE_TYPE(cft->private); |
5235 | name = MEMFILE_ATTR(cft->private); | |
af36f906 | 5236 | |
8c7c6e34 | 5237 | switch (name) { |
628f4235 | 5238 | case RES_LIMIT: |
4b3bde4c BS |
5239 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
5240 | ret = -EINVAL; | |
5241 | break; | |
5242 | } | |
628f4235 KH |
5243 | /* This function does all necessary parse...reuse it */ |
5244 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
5245 | if (ret) |
5246 | break; | |
5247 | if (type == _MEM) | |
628f4235 | 5248 | ret = mem_cgroup_resize_limit(memcg, val); |
510fc4e1 | 5249 | else if (type == _MEMSWAP) |
8c7c6e34 | 5250 | ret = mem_cgroup_resize_memsw_limit(memcg, val); |
510fc4e1 | 5251 | else if (type == _KMEM) |
182446d0 | 5252 | ret = memcg_update_kmem_limit(css, val); |
510fc4e1 GC |
5253 | else |
5254 | return -EINVAL; | |
628f4235 | 5255 | break; |
296c81d8 BS |
5256 | case RES_SOFT_LIMIT: |
5257 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
5258 | if (ret) | |
5259 | break; | |
5260 | /* | |
5261 | * For memsw, soft limits are hard to implement in terms | |
5262 | * of semantics, for now, we support soft limits for | |
5263 | * control without swap | |
5264 | */ | |
5265 | if (type == _MEM) | |
5266 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
5267 | else | |
5268 | ret = -EINVAL; | |
5269 | break; | |
628f4235 KH |
5270 | default: |
5271 | ret = -EINVAL; /* should be BUG() ? */ | |
5272 | break; | |
5273 | } | |
5274 | return ret; | |
8cdea7c0 BS |
5275 | } |
5276 | ||
fee7b548 KH |
5277 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
5278 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
5279 | { | |
fee7b548 KH |
5280 | unsigned long long min_limit, min_memsw_limit, tmp; |
5281 | ||
5282 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
5283 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
fee7b548 KH |
5284 | if (!memcg->use_hierarchy) |
5285 | goto out; | |
5286 | ||
63876986 TH |
5287 | while (css_parent(&memcg->css)) { |
5288 | memcg = mem_cgroup_from_css(css_parent(&memcg->css)); | |
fee7b548 KH |
5289 | if (!memcg->use_hierarchy) |
5290 | break; | |
5291 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
5292 | min_limit = min(min_limit, tmp); | |
5293 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
5294 | min_memsw_limit = min(min_memsw_limit, tmp); | |
5295 | } | |
5296 | out: | |
5297 | *mem_limit = min_limit; | |
5298 | *memsw_limit = min_memsw_limit; | |
fee7b548 KH |
5299 | } |
5300 | ||
182446d0 | 5301 | static int mem_cgroup_reset(struct cgroup_subsys_state *css, unsigned int event) |
c84872e1 | 5302 | { |
182446d0 | 5303 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
86ae53e1 GC |
5304 | int name; |
5305 | enum res_type type; | |
c84872e1 | 5306 | |
8c7c6e34 KH |
5307 | type = MEMFILE_TYPE(event); |
5308 | name = MEMFILE_ATTR(event); | |
af36f906 | 5309 | |
8c7c6e34 | 5310 | switch (name) { |
29f2a4da | 5311 | case RES_MAX_USAGE: |
8c7c6e34 | 5312 | if (type == _MEM) |
c0ff4b85 | 5313 | res_counter_reset_max(&memcg->res); |
510fc4e1 | 5314 | else if (type == _MEMSWAP) |
c0ff4b85 | 5315 | res_counter_reset_max(&memcg->memsw); |
510fc4e1 GC |
5316 | else if (type == _KMEM) |
5317 | res_counter_reset_max(&memcg->kmem); | |
5318 | else | |
5319 | return -EINVAL; | |
29f2a4da PE |
5320 | break; |
5321 | case RES_FAILCNT: | |
8c7c6e34 | 5322 | if (type == _MEM) |
c0ff4b85 | 5323 | res_counter_reset_failcnt(&memcg->res); |
510fc4e1 | 5324 | else if (type == _MEMSWAP) |
c0ff4b85 | 5325 | res_counter_reset_failcnt(&memcg->memsw); |
510fc4e1 GC |
5326 | else if (type == _KMEM) |
5327 | res_counter_reset_failcnt(&memcg->kmem); | |
5328 | else | |
5329 | return -EINVAL; | |
29f2a4da PE |
5330 | break; |
5331 | } | |
f64c3f54 | 5332 | |
85cc59db | 5333 | return 0; |
c84872e1 PE |
5334 | } |
5335 | ||
182446d0 | 5336 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
5337 | struct cftype *cft) |
5338 | { | |
182446d0 | 5339 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
5340 | } |
5341 | ||
02491447 | 5342 | #ifdef CONFIG_MMU |
182446d0 | 5343 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
5344 | struct cftype *cft, u64 val) |
5345 | { | |
182446d0 | 5346 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 DN |
5347 | |
5348 | if (val >= (1 << NR_MOVE_TYPE)) | |
5349 | return -EINVAL; | |
ee5e8472 | 5350 | |
7dc74be0 | 5351 | /* |
ee5e8472 GC |
5352 | * No kind of locking is needed in here, because ->can_attach() will |
5353 | * check this value once in the beginning of the process, and then carry | |
5354 | * on with stale data. This means that changes to this value will only | |
5355 | * affect task migrations starting after the change. | |
7dc74be0 | 5356 | */ |
c0ff4b85 | 5357 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
5358 | return 0; |
5359 | } | |
02491447 | 5360 | #else |
182446d0 | 5361 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
5362 | struct cftype *cft, u64 val) |
5363 | { | |
5364 | return -ENOSYS; | |
5365 | } | |
5366 | #endif | |
7dc74be0 | 5367 | |
406eb0c9 | 5368 | #ifdef CONFIG_NUMA |
182446d0 TH |
5369 | static int memcg_numa_stat_show(struct cgroup_subsys_state *css, |
5370 | struct cftype *cft, struct seq_file *m) | |
406eb0c9 YH |
5371 | { |
5372 | int nid; | |
5373 | unsigned long total_nr, file_nr, anon_nr, unevictable_nr; | |
5374 | unsigned long node_nr; | |
182446d0 | 5375 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
406eb0c9 | 5376 | |
d79154bb | 5377 | total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL); |
406eb0c9 | 5378 | seq_printf(m, "total=%lu", total_nr); |
31aaea4a | 5379 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5380 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL); |
406eb0c9 YH |
5381 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5382 | } | |
5383 | seq_putc(m, '\n'); | |
5384 | ||
d79154bb | 5385 | file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE); |
406eb0c9 | 5386 | seq_printf(m, "file=%lu", file_nr); |
31aaea4a | 5387 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5388 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 5389 | LRU_ALL_FILE); |
406eb0c9 YH |
5390 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5391 | } | |
5392 | seq_putc(m, '\n'); | |
5393 | ||
d79154bb | 5394 | anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON); |
406eb0c9 | 5395 | seq_printf(m, "anon=%lu", anon_nr); |
31aaea4a | 5396 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5397 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 5398 | LRU_ALL_ANON); |
406eb0c9 YH |
5399 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5400 | } | |
5401 | seq_putc(m, '\n'); | |
5402 | ||
d79154bb | 5403 | unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE)); |
406eb0c9 | 5404 | seq_printf(m, "unevictable=%lu", unevictable_nr); |
31aaea4a | 5405 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5406 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 5407 | BIT(LRU_UNEVICTABLE)); |
406eb0c9 YH |
5408 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5409 | } | |
5410 | seq_putc(m, '\n'); | |
5411 | return 0; | |
5412 | } | |
5413 | #endif /* CONFIG_NUMA */ | |
5414 | ||
af7c4b0e JW |
5415 | static inline void mem_cgroup_lru_names_not_uptodate(void) |
5416 | { | |
5417 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); | |
5418 | } | |
5419 | ||
182446d0 | 5420 | static int memcg_stat_show(struct cgroup_subsys_state *css, struct cftype *cft, |
78ccf5b5 | 5421 | struct seq_file *m) |
d2ceb9b7 | 5422 | { |
182446d0 | 5423 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
af7c4b0e JW |
5424 | struct mem_cgroup *mi; |
5425 | unsigned int i; | |
406eb0c9 | 5426 | |
af7c4b0e | 5427 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 5428 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 5429 | continue; |
af7c4b0e JW |
5430 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
5431 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 5432 | } |
7b854121 | 5433 | |
af7c4b0e JW |
5434 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
5435 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
5436 | mem_cgroup_read_events(memcg, i)); | |
5437 | ||
5438 | for (i = 0; i < NR_LRU_LISTS; i++) | |
5439 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
5440 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
5441 | ||
14067bb3 | 5442 | /* Hierarchical information */ |
fee7b548 KH |
5443 | { |
5444 | unsigned long long limit, memsw_limit; | |
d79154bb | 5445 | memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit); |
78ccf5b5 | 5446 | seq_printf(m, "hierarchical_memory_limit %llu\n", limit); |
fee7b548 | 5447 | if (do_swap_account) |
78ccf5b5 JW |
5448 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
5449 | memsw_limit); | |
fee7b548 | 5450 | } |
7f016ee8 | 5451 | |
af7c4b0e JW |
5452 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
5453 | long long val = 0; | |
5454 | ||
bff6bb83 | 5455 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 5456 | continue; |
af7c4b0e JW |
5457 | for_each_mem_cgroup_tree(mi, memcg) |
5458 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
5459 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
5460 | } | |
5461 | ||
5462 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
5463 | unsigned long long val = 0; | |
5464 | ||
5465 | for_each_mem_cgroup_tree(mi, memcg) | |
5466 | val += mem_cgroup_read_events(mi, i); | |
5467 | seq_printf(m, "total_%s %llu\n", | |
5468 | mem_cgroup_events_names[i], val); | |
5469 | } | |
5470 | ||
5471 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
5472 | unsigned long long val = 0; | |
5473 | ||
5474 | for_each_mem_cgroup_tree(mi, memcg) | |
5475 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
5476 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 5477 | } |
14067bb3 | 5478 | |
7f016ee8 | 5479 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
5480 | { |
5481 | int nid, zid; | |
5482 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 5483 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
5484 | unsigned long recent_rotated[2] = {0, 0}; |
5485 | unsigned long recent_scanned[2] = {0, 0}; | |
5486 | ||
5487 | for_each_online_node(nid) | |
5488 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
d79154bb | 5489 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
89abfab1 | 5490 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 5491 | |
89abfab1 HD |
5492 | recent_rotated[0] += rstat->recent_rotated[0]; |
5493 | recent_rotated[1] += rstat->recent_rotated[1]; | |
5494 | recent_scanned[0] += rstat->recent_scanned[0]; | |
5495 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 5496 | } |
78ccf5b5 JW |
5497 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
5498 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
5499 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
5500 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
5501 | } |
5502 | #endif | |
5503 | ||
d2ceb9b7 KH |
5504 | return 0; |
5505 | } | |
5506 | ||
182446d0 TH |
5507 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
5508 | struct cftype *cft) | |
a7885eb8 | 5509 | { |
182446d0 | 5510 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 5511 | |
1f4c025b | 5512 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
5513 | } |
5514 | ||
182446d0 TH |
5515 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
5516 | struct cftype *cft, u64 val) | |
a7885eb8 | 5517 | { |
182446d0 | 5518 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
63876986 | 5519 | struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css)); |
a7885eb8 | 5520 | |
63876986 | 5521 | if (val > 100 || !parent) |
a7885eb8 KM |
5522 | return -EINVAL; |
5523 | ||
0999821b | 5524 | mutex_lock(&memcg_create_mutex); |
068b38c1 | 5525 | |
a7885eb8 | 5526 | /* If under hierarchy, only empty-root can set this value */ |
b5f99b53 | 5527 | if ((parent->use_hierarchy) || memcg_has_children(memcg)) { |
0999821b | 5528 | mutex_unlock(&memcg_create_mutex); |
a7885eb8 | 5529 | return -EINVAL; |
068b38c1 | 5530 | } |
a7885eb8 | 5531 | |
a7885eb8 | 5532 | memcg->swappiness = val; |
a7885eb8 | 5533 | |
0999821b | 5534 | mutex_unlock(&memcg_create_mutex); |
068b38c1 | 5535 | |
a7885eb8 KM |
5536 | return 0; |
5537 | } | |
5538 | ||
2e72b634 KS |
5539 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
5540 | { | |
5541 | struct mem_cgroup_threshold_ary *t; | |
5542 | u64 usage; | |
5543 | int i; | |
5544 | ||
5545 | rcu_read_lock(); | |
5546 | if (!swap) | |
2c488db2 | 5547 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 5548 | else |
2c488db2 | 5549 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
5550 | |
5551 | if (!t) | |
5552 | goto unlock; | |
5553 | ||
5554 | usage = mem_cgroup_usage(memcg, swap); | |
5555 | ||
5556 | /* | |
748dad36 | 5557 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
5558 | * If it's not true, a threshold was crossed after last |
5559 | * call of __mem_cgroup_threshold(). | |
5560 | */ | |
5407a562 | 5561 | i = t->current_threshold; |
2e72b634 KS |
5562 | |
5563 | /* | |
5564 | * Iterate backward over array of thresholds starting from | |
5565 | * current_threshold and check if a threshold is crossed. | |
5566 | * If none of thresholds below usage is crossed, we read | |
5567 | * only one element of the array here. | |
5568 | */ | |
5569 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
5570 | eventfd_signal(t->entries[i].eventfd, 1); | |
5571 | ||
5572 | /* i = current_threshold + 1 */ | |
5573 | i++; | |
5574 | ||
5575 | /* | |
5576 | * Iterate forward over array of thresholds starting from | |
5577 | * current_threshold+1 and check if a threshold is crossed. | |
5578 | * If none of thresholds above usage is crossed, we read | |
5579 | * only one element of the array here. | |
5580 | */ | |
5581 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
5582 | eventfd_signal(t->entries[i].eventfd, 1); | |
5583 | ||
5584 | /* Update current_threshold */ | |
5407a562 | 5585 | t->current_threshold = i - 1; |
2e72b634 KS |
5586 | unlock: |
5587 | rcu_read_unlock(); | |
5588 | } | |
5589 | ||
5590 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
5591 | { | |
ad4ca5f4 KS |
5592 | while (memcg) { |
5593 | __mem_cgroup_threshold(memcg, false); | |
5594 | if (do_swap_account) | |
5595 | __mem_cgroup_threshold(memcg, true); | |
5596 | ||
5597 | memcg = parent_mem_cgroup(memcg); | |
5598 | } | |
2e72b634 KS |
5599 | } |
5600 | ||
5601 | static int compare_thresholds(const void *a, const void *b) | |
5602 | { | |
5603 | const struct mem_cgroup_threshold *_a = a; | |
5604 | const struct mem_cgroup_threshold *_b = b; | |
5605 | ||
2bff24a3 GT |
5606 | if (_a->threshold > _b->threshold) |
5607 | return 1; | |
5608 | ||
5609 | if (_a->threshold < _b->threshold) | |
5610 | return -1; | |
5611 | ||
5612 | return 0; | |
2e72b634 KS |
5613 | } |
5614 | ||
c0ff4b85 | 5615 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
5616 | { |
5617 | struct mem_cgroup_eventfd_list *ev; | |
5618 | ||
c0ff4b85 | 5619 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 KH |
5620 | eventfd_signal(ev->eventfd, 1); |
5621 | return 0; | |
5622 | } | |
5623 | ||
c0ff4b85 | 5624 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 5625 | { |
7d74b06f KH |
5626 | struct mem_cgroup *iter; |
5627 | ||
c0ff4b85 | 5628 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 5629 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
5630 | } |
5631 | ||
81eeaf04 | 5632 | static int mem_cgroup_usage_register_event(struct cgroup_subsys_state *css, |
9490ff27 | 5633 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) |
2e72b634 | 5634 | { |
81eeaf04 | 5635 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
2c488db2 KS |
5636 | struct mem_cgroup_thresholds *thresholds; |
5637 | struct mem_cgroup_threshold_ary *new; | |
86ae53e1 | 5638 | enum res_type type = MEMFILE_TYPE(cft->private); |
2e72b634 | 5639 | u64 threshold, usage; |
2c488db2 | 5640 | int i, size, ret; |
2e72b634 KS |
5641 | |
5642 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
5643 | if (ret) | |
5644 | return ret; | |
5645 | ||
5646 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 5647 | |
2e72b634 | 5648 | if (type == _MEM) |
2c488db2 | 5649 | thresholds = &memcg->thresholds; |
2e72b634 | 5650 | else if (type == _MEMSWAP) |
2c488db2 | 5651 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
5652 | else |
5653 | BUG(); | |
5654 | ||
5655 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
5656 | ||
5657 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 5658 | if (thresholds->primary) |
2e72b634 KS |
5659 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
5660 | ||
2c488db2 | 5661 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
5662 | |
5663 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 5664 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 5665 | GFP_KERNEL); |
2c488db2 | 5666 | if (!new) { |
2e72b634 KS |
5667 | ret = -ENOMEM; |
5668 | goto unlock; | |
5669 | } | |
2c488db2 | 5670 | new->size = size; |
2e72b634 KS |
5671 | |
5672 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
5673 | if (thresholds->primary) { |
5674 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 5675 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
5676 | } |
5677 | ||
2e72b634 | 5678 | /* Add new threshold */ |
2c488db2 KS |
5679 | new->entries[size - 1].eventfd = eventfd; |
5680 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
5681 | |
5682 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 5683 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
5684 | compare_thresholds, NULL); |
5685 | ||
5686 | /* Find current threshold */ | |
2c488db2 | 5687 | new->current_threshold = -1; |
2e72b634 | 5688 | for (i = 0; i < size; i++) { |
748dad36 | 5689 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 5690 | /* |
2c488db2 KS |
5691 | * new->current_threshold will not be used until |
5692 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
5693 | * it here. |
5694 | */ | |
2c488db2 | 5695 | ++new->current_threshold; |
748dad36 SZ |
5696 | } else |
5697 | break; | |
2e72b634 KS |
5698 | } |
5699 | ||
2c488db2 KS |
5700 | /* Free old spare buffer and save old primary buffer as spare */ |
5701 | kfree(thresholds->spare); | |
5702 | thresholds->spare = thresholds->primary; | |
5703 | ||
5704 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 5705 | |
907860ed | 5706 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
5707 | synchronize_rcu(); |
5708 | ||
2e72b634 KS |
5709 | unlock: |
5710 | mutex_unlock(&memcg->thresholds_lock); | |
5711 | ||
5712 | return ret; | |
5713 | } | |
5714 | ||
81eeaf04 | 5715 | static void mem_cgroup_usage_unregister_event(struct cgroup_subsys_state *css, |
9490ff27 | 5716 | struct cftype *cft, struct eventfd_ctx *eventfd) |
2e72b634 | 5717 | { |
81eeaf04 | 5718 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
2c488db2 KS |
5719 | struct mem_cgroup_thresholds *thresholds; |
5720 | struct mem_cgroup_threshold_ary *new; | |
86ae53e1 | 5721 | enum res_type type = MEMFILE_TYPE(cft->private); |
2e72b634 | 5722 | u64 usage; |
2c488db2 | 5723 | int i, j, size; |
2e72b634 KS |
5724 | |
5725 | mutex_lock(&memcg->thresholds_lock); | |
5726 | if (type == _MEM) | |
2c488db2 | 5727 | thresholds = &memcg->thresholds; |
2e72b634 | 5728 | else if (type == _MEMSWAP) |
2c488db2 | 5729 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
5730 | else |
5731 | BUG(); | |
5732 | ||
371528ca AV |
5733 | if (!thresholds->primary) |
5734 | goto unlock; | |
5735 | ||
2e72b634 KS |
5736 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); |
5737 | ||
5738 | /* Check if a threshold crossed before removing */ | |
5739 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
5740 | ||
5741 | /* Calculate new number of threshold */ | |
2c488db2 KS |
5742 | size = 0; |
5743 | for (i = 0; i < thresholds->primary->size; i++) { | |
5744 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
5745 | size++; |
5746 | } | |
5747 | ||
2c488db2 | 5748 | new = thresholds->spare; |
907860ed | 5749 | |
2e72b634 KS |
5750 | /* Set thresholds array to NULL if we don't have thresholds */ |
5751 | if (!size) { | |
2c488db2 KS |
5752 | kfree(new); |
5753 | new = NULL; | |
907860ed | 5754 | goto swap_buffers; |
2e72b634 KS |
5755 | } |
5756 | ||
2c488db2 | 5757 | new->size = size; |
2e72b634 KS |
5758 | |
5759 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
5760 | new->current_threshold = -1; |
5761 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
5762 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
5763 | continue; |
5764 | ||
2c488db2 | 5765 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 5766 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 5767 | /* |
2c488db2 | 5768 | * new->current_threshold will not be used |
2e72b634 KS |
5769 | * until rcu_assign_pointer(), so it's safe to increment |
5770 | * it here. | |
5771 | */ | |
2c488db2 | 5772 | ++new->current_threshold; |
2e72b634 KS |
5773 | } |
5774 | j++; | |
5775 | } | |
5776 | ||
907860ed | 5777 | swap_buffers: |
2c488db2 KS |
5778 | /* Swap primary and spare array */ |
5779 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
5780 | /* If all events are unregistered, free the spare array */ |
5781 | if (!new) { | |
5782 | kfree(thresholds->spare); | |
5783 | thresholds->spare = NULL; | |
5784 | } | |
5785 | ||
2c488db2 | 5786 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 5787 | |
907860ed | 5788 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 5789 | synchronize_rcu(); |
371528ca | 5790 | unlock: |
2e72b634 | 5791 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 5792 | } |
c1e862c1 | 5793 | |
81eeaf04 | 5794 | static int mem_cgroup_oom_register_event(struct cgroup_subsys_state *css, |
9490ff27 KH |
5795 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) |
5796 | { | |
81eeaf04 | 5797 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
9490ff27 | 5798 | struct mem_cgroup_eventfd_list *event; |
86ae53e1 | 5799 | enum res_type type = MEMFILE_TYPE(cft->private); |
9490ff27 KH |
5800 | |
5801 | BUG_ON(type != _OOM_TYPE); | |
5802 | event = kmalloc(sizeof(*event), GFP_KERNEL); | |
5803 | if (!event) | |
5804 | return -ENOMEM; | |
5805 | ||
1af8efe9 | 5806 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
5807 | |
5808 | event->eventfd = eventfd; | |
5809 | list_add(&event->list, &memcg->oom_notify); | |
5810 | ||
5811 | /* already in OOM ? */ | |
79dfdacc | 5812 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 5813 | eventfd_signal(eventfd, 1); |
1af8efe9 | 5814 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
5815 | |
5816 | return 0; | |
5817 | } | |
5818 | ||
81eeaf04 | 5819 | static void mem_cgroup_oom_unregister_event(struct cgroup_subsys_state *css, |
9490ff27 KH |
5820 | struct cftype *cft, struct eventfd_ctx *eventfd) |
5821 | { | |
81eeaf04 | 5822 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
9490ff27 | 5823 | struct mem_cgroup_eventfd_list *ev, *tmp; |
86ae53e1 | 5824 | enum res_type type = MEMFILE_TYPE(cft->private); |
9490ff27 KH |
5825 | |
5826 | BUG_ON(type != _OOM_TYPE); | |
5827 | ||
1af8efe9 | 5828 | spin_lock(&memcg_oom_lock); |
9490ff27 | 5829 | |
c0ff4b85 | 5830 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
5831 | if (ev->eventfd == eventfd) { |
5832 | list_del(&ev->list); | |
5833 | kfree(ev); | |
5834 | } | |
5835 | } | |
5836 | ||
1af8efe9 | 5837 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
5838 | } |
5839 | ||
182446d0 | 5840 | static int mem_cgroup_oom_control_read(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
5841 | struct cftype *cft, struct cgroup_map_cb *cb) |
5842 | { | |
182446d0 | 5843 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 | 5844 | |
c0ff4b85 | 5845 | cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable); |
3c11ecf4 | 5846 | |
c0ff4b85 | 5847 | if (atomic_read(&memcg->under_oom)) |
3c11ecf4 KH |
5848 | cb->fill(cb, "under_oom", 1); |
5849 | else | |
5850 | cb->fill(cb, "under_oom", 0); | |
5851 | return 0; | |
5852 | } | |
5853 | ||
182446d0 | 5854 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
5855 | struct cftype *cft, u64 val) |
5856 | { | |
182446d0 | 5857 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
63876986 | 5858 | struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css)); |
3c11ecf4 KH |
5859 | |
5860 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
63876986 | 5861 | if (!parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
5862 | return -EINVAL; |
5863 | ||
0999821b | 5864 | mutex_lock(&memcg_create_mutex); |
3c11ecf4 | 5865 | /* oom-kill-disable is a flag for subhierarchy. */ |
b5f99b53 | 5866 | if ((parent->use_hierarchy) || memcg_has_children(memcg)) { |
0999821b | 5867 | mutex_unlock(&memcg_create_mutex); |
3c11ecf4 KH |
5868 | return -EINVAL; |
5869 | } | |
c0ff4b85 | 5870 | memcg->oom_kill_disable = val; |
4d845ebf | 5871 | if (!val) |
c0ff4b85 | 5872 | memcg_oom_recover(memcg); |
0999821b | 5873 | mutex_unlock(&memcg_create_mutex); |
3c11ecf4 KH |
5874 | return 0; |
5875 | } | |
5876 | ||
c255a458 | 5877 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 5878 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 5879 | { |
55007d84 GC |
5880 | int ret; |
5881 | ||
2633d7a0 | 5882 | memcg->kmemcg_id = -1; |
55007d84 GC |
5883 | ret = memcg_propagate_kmem(memcg); |
5884 | if (ret) | |
5885 | return ret; | |
2633d7a0 | 5886 | |
1d62e436 | 5887 | return mem_cgroup_sockets_init(memcg, ss); |
573b400d | 5888 | } |
e5671dfa | 5889 | |
10d5ebf4 | 5890 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 5891 | { |
1d62e436 | 5892 | mem_cgroup_sockets_destroy(memcg); |
10d5ebf4 LZ |
5893 | } |
5894 | ||
5895 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) | |
5896 | { | |
5897 | if (!memcg_kmem_is_active(memcg)) | |
5898 | return; | |
5899 | ||
5900 | /* | |
5901 | * kmem charges can outlive the cgroup. In the case of slab | |
5902 | * pages, for instance, a page contain objects from various | |
5903 | * processes. As we prevent from taking a reference for every | |
5904 | * such allocation we have to be careful when doing uncharge | |
5905 | * (see memcg_uncharge_kmem) and here during offlining. | |
5906 | * | |
5907 | * The idea is that that only the _last_ uncharge which sees | |
5908 | * the dead memcg will drop the last reference. An additional | |
5909 | * reference is taken here before the group is marked dead | |
5910 | * which is then paired with css_put during uncharge resp. here. | |
5911 | * | |
5912 | * Although this might sound strange as this path is called from | |
5913 | * css_offline() when the referencemight have dropped down to 0 | |
5914 | * and shouldn't be incremented anymore (css_tryget would fail) | |
5915 | * we do not have other options because of the kmem allocations | |
5916 | * lifetime. | |
5917 | */ | |
5918 | css_get(&memcg->css); | |
7de37682 GC |
5919 | |
5920 | memcg_kmem_mark_dead(memcg); | |
5921 | ||
5922 | if (res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0) | |
5923 | return; | |
5924 | ||
7de37682 | 5925 | if (memcg_kmem_test_and_clear_dead(memcg)) |
10d5ebf4 | 5926 | css_put(&memcg->css); |
d1a4c0b3 | 5927 | } |
e5671dfa | 5928 | #else |
cbe128e3 | 5929 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
5930 | { |
5931 | return 0; | |
5932 | } | |
d1a4c0b3 | 5933 | |
10d5ebf4 LZ |
5934 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
5935 | { | |
5936 | } | |
5937 | ||
5938 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) | |
d1a4c0b3 GC |
5939 | { |
5940 | } | |
e5671dfa GC |
5941 | #endif |
5942 | ||
8cdea7c0 BS |
5943 | static struct cftype mem_cgroup_files[] = { |
5944 | { | |
0eea1030 | 5945 | .name = "usage_in_bytes", |
8c7c6e34 | 5946 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
af36f906 | 5947 | .read = mem_cgroup_read, |
9490ff27 KH |
5948 | .register_event = mem_cgroup_usage_register_event, |
5949 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8cdea7c0 | 5950 | }, |
c84872e1 PE |
5951 | { |
5952 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 5953 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 5954 | .trigger = mem_cgroup_reset, |
af36f906 | 5955 | .read = mem_cgroup_read, |
c84872e1 | 5956 | }, |
8cdea7c0 | 5957 | { |
0eea1030 | 5958 | .name = "limit_in_bytes", |
8c7c6e34 | 5959 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 5960 | .write_string = mem_cgroup_write, |
af36f906 | 5961 | .read = mem_cgroup_read, |
8cdea7c0 | 5962 | }, |
296c81d8 BS |
5963 | { |
5964 | .name = "soft_limit_in_bytes", | |
5965 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
5966 | .write_string = mem_cgroup_write, | |
af36f906 | 5967 | .read = mem_cgroup_read, |
296c81d8 | 5968 | }, |
8cdea7c0 BS |
5969 | { |
5970 | .name = "failcnt", | |
8c7c6e34 | 5971 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 5972 | .trigger = mem_cgroup_reset, |
af36f906 | 5973 | .read = mem_cgroup_read, |
8cdea7c0 | 5974 | }, |
d2ceb9b7 KH |
5975 | { |
5976 | .name = "stat", | |
ab215884 | 5977 | .read_seq_string = memcg_stat_show, |
d2ceb9b7 | 5978 | }, |
c1e862c1 KH |
5979 | { |
5980 | .name = "force_empty", | |
5981 | .trigger = mem_cgroup_force_empty_write, | |
5982 | }, | |
18f59ea7 BS |
5983 | { |
5984 | .name = "use_hierarchy", | |
f00baae7 | 5985 | .flags = CFTYPE_INSANE, |
18f59ea7 BS |
5986 | .write_u64 = mem_cgroup_hierarchy_write, |
5987 | .read_u64 = mem_cgroup_hierarchy_read, | |
5988 | }, | |
a7885eb8 KM |
5989 | { |
5990 | .name = "swappiness", | |
5991 | .read_u64 = mem_cgroup_swappiness_read, | |
5992 | .write_u64 = mem_cgroup_swappiness_write, | |
5993 | }, | |
7dc74be0 DN |
5994 | { |
5995 | .name = "move_charge_at_immigrate", | |
5996 | .read_u64 = mem_cgroup_move_charge_read, | |
5997 | .write_u64 = mem_cgroup_move_charge_write, | |
5998 | }, | |
9490ff27 KH |
5999 | { |
6000 | .name = "oom_control", | |
3c11ecf4 KH |
6001 | .read_map = mem_cgroup_oom_control_read, |
6002 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
6003 | .register_event = mem_cgroup_oom_register_event, |
6004 | .unregister_event = mem_cgroup_oom_unregister_event, | |
6005 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), | |
6006 | }, | |
70ddf637 AV |
6007 | { |
6008 | .name = "pressure_level", | |
6009 | .register_event = vmpressure_register_event, | |
6010 | .unregister_event = vmpressure_unregister_event, | |
6011 | }, | |
406eb0c9 YH |
6012 | #ifdef CONFIG_NUMA |
6013 | { | |
6014 | .name = "numa_stat", | |
ab215884 | 6015 | .read_seq_string = memcg_numa_stat_show, |
406eb0c9 YH |
6016 | }, |
6017 | #endif | |
510fc4e1 GC |
6018 | #ifdef CONFIG_MEMCG_KMEM |
6019 | { | |
6020 | .name = "kmem.limit_in_bytes", | |
6021 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
6022 | .write_string = mem_cgroup_write, | |
6023 | .read = mem_cgroup_read, | |
6024 | }, | |
6025 | { | |
6026 | .name = "kmem.usage_in_bytes", | |
6027 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
6028 | .read = mem_cgroup_read, | |
6029 | }, | |
6030 | { | |
6031 | .name = "kmem.failcnt", | |
6032 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6033 | .trigger = mem_cgroup_reset, | |
6034 | .read = mem_cgroup_read, | |
6035 | }, | |
6036 | { | |
6037 | .name = "kmem.max_usage_in_bytes", | |
6038 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6039 | .trigger = mem_cgroup_reset, | |
6040 | .read = mem_cgroup_read, | |
6041 | }, | |
749c5415 GC |
6042 | #ifdef CONFIG_SLABINFO |
6043 | { | |
6044 | .name = "kmem.slabinfo", | |
6045 | .read_seq_string = mem_cgroup_slabinfo_read, | |
6046 | }, | |
6047 | #endif | |
8c7c6e34 | 6048 | #endif |
6bc10349 | 6049 | { }, /* terminate */ |
af36f906 | 6050 | }; |
8c7c6e34 | 6051 | |
2d11085e MH |
6052 | #ifdef CONFIG_MEMCG_SWAP |
6053 | static struct cftype memsw_cgroup_files[] = { | |
6054 | { | |
6055 | .name = "memsw.usage_in_bytes", | |
6056 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
6057 | .read = mem_cgroup_read, | |
6058 | .register_event = mem_cgroup_usage_register_event, | |
6059 | .unregister_event = mem_cgroup_usage_unregister_event, | |
6060 | }, | |
6061 | { | |
6062 | .name = "memsw.max_usage_in_bytes", | |
6063 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6064 | .trigger = mem_cgroup_reset, | |
6065 | .read = mem_cgroup_read, | |
6066 | }, | |
6067 | { | |
6068 | .name = "memsw.limit_in_bytes", | |
6069 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
6070 | .write_string = mem_cgroup_write, | |
6071 | .read = mem_cgroup_read, | |
6072 | }, | |
6073 | { | |
6074 | .name = "memsw.failcnt", | |
6075 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6076 | .trigger = mem_cgroup_reset, | |
6077 | .read = mem_cgroup_read, | |
6078 | }, | |
6079 | { }, /* terminate */ | |
6080 | }; | |
6081 | #endif | |
c0ff4b85 | 6082 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
6083 | { |
6084 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 6085 | struct mem_cgroup_per_zone *mz; |
41e3355d | 6086 | int zone, tmp = node; |
1ecaab2b KH |
6087 | /* |
6088 | * This routine is called against possible nodes. | |
6089 | * But it's BUG to call kmalloc() against offline node. | |
6090 | * | |
6091 | * TODO: this routine can waste much memory for nodes which will | |
6092 | * never be onlined. It's better to use memory hotplug callback | |
6093 | * function. | |
6094 | */ | |
41e3355d KH |
6095 | if (!node_state(node, N_NORMAL_MEMORY)) |
6096 | tmp = -1; | |
17295c88 | 6097 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
6098 | if (!pn) |
6099 | return 1; | |
1ecaab2b | 6100 | |
1ecaab2b KH |
6101 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
6102 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 6103 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
6104 | mz->usage_in_excess = 0; |
6105 | mz->on_tree = false; | |
d79154bb | 6106 | mz->memcg = memcg; |
1ecaab2b | 6107 | } |
54f72fe0 | 6108 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
6109 | return 0; |
6110 | } | |
6111 | ||
c0ff4b85 | 6112 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 6113 | { |
54f72fe0 | 6114 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
6115 | } |
6116 | ||
33327948 KH |
6117 | static struct mem_cgroup *mem_cgroup_alloc(void) |
6118 | { | |
d79154bb | 6119 | struct mem_cgroup *memcg; |
45cf7ebd | 6120 | size_t size = memcg_size(); |
33327948 | 6121 | |
45cf7ebd | 6122 | /* Can be very big if nr_node_ids is very big */ |
c8dad2bb | 6123 | if (size < PAGE_SIZE) |
d79154bb | 6124 | memcg = kzalloc(size, GFP_KERNEL); |
33327948 | 6125 | else |
d79154bb | 6126 | memcg = vzalloc(size); |
33327948 | 6127 | |
d79154bb | 6128 | if (!memcg) |
e7bbcdf3 DC |
6129 | return NULL; |
6130 | ||
d79154bb HD |
6131 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
6132 | if (!memcg->stat) | |
d2e61b8d | 6133 | goto out_free; |
d79154bb HD |
6134 | spin_lock_init(&memcg->pcp_counter_lock); |
6135 | return memcg; | |
d2e61b8d DC |
6136 | |
6137 | out_free: | |
6138 | if (size < PAGE_SIZE) | |
d79154bb | 6139 | kfree(memcg); |
d2e61b8d | 6140 | else |
d79154bb | 6141 | vfree(memcg); |
d2e61b8d | 6142 | return NULL; |
33327948 KH |
6143 | } |
6144 | ||
59927fb9 | 6145 | /* |
c8b2a36f GC |
6146 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
6147 | * (scanning all at force_empty is too costly...) | |
6148 | * | |
6149 | * Instead of clearing all references at force_empty, we remember | |
6150 | * the number of reference from swap_cgroup and free mem_cgroup when | |
6151 | * it goes down to 0. | |
6152 | * | |
6153 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 6154 | */ |
c8b2a36f GC |
6155 | |
6156 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 6157 | { |
c8b2a36f | 6158 | int node; |
45cf7ebd | 6159 | size_t size = memcg_size(); |
59927fb9 | 6160 | |
bb4cc1a8 | 6161 | mem_cgroup_remove_from_trees(memcg); |
c8b2a36f GC |
6162 | free_css_id(&mem_cgroup_subsys, &memcg->css); |
6163 | ||
6164 | for_each_node(node) | |
6165 | free_mem_cgroup_per_zone_info(memcg, node); | |
6166 | ||
6167 | free_percpu(memcg->stat); | |
6168 | ||
3f134619 GC |
6169 | /* |
6170 | * We need to make sure that (at least for now), the jump label | |
6171 | * destruction code runs outside of the cgroup lock. This is because | |
6172 | * get_online_cpus(), which is called from the static_branch update, | |
6173 | * can't be called inside the cgroup_lock. cpusets are the ones | |
6174 | * enforcing this dependency, so if they ever change, we might as well. | |
6175 | * | |
6176 | * schedule_work() will guarantee this happens. Be careful if you need | |
6177 | * to move this code around, and make sure it is outside | |
6178 | * the cgroup_lock. | |
6179 | */ | |
a8964b9b | 6180 | disarm_static_keys(memcg); |
3afe36b1 GC |
6181 | if (size < PAGE_SIZE) |
6182 | kfree(memcg); | |
6183 | else | |
6184 | vfree(memcg); | |
59927fb9 | 6185 | } |
3afe36b1 | 6186 | |
7bcc1bb1 DN |
6187 | /* |
6188 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
6189 | */ | |
e1aab161 | 6190 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 6191 | { |
c0ff4b85 | 6192 | if (!memcg->res.parent) |
7bcc1bb1 | 6193 | return NULL; |
c0ff4b85 | 6194 | return mem_cgroup_from_res_counter(memcg->res.parent, res); |
7bcc1bb1 | 6195 | } |
e1aab161 | 6196 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 6197 | |
bb4cc1a8 AM |
6198 | static void __init mem_cgroup_soft_limit_tree_init(void) |
6199 | { | |
6200 | struct mem_cgroup_tree_per_node *rtpn; | |
6201 | struct mem_cgroup_tree_per_zone *rtpz; | |
6202 | int tmp, node, zone; | |
6203 | ||
6204 | for_each_node(node) { | |
6205 | tmp = node; | |
6206 | if (!node_state(node, N_NORMAL_MEMORY)) | |
6207 | tmp = -1; | |
6208 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
6209 | BUG_ON(!rtpn); | |
6210 | ||
6211 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
6212 | ||
6213 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
6214 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
6215 | rtpz->rb_root = RB_ROOT; | |
6216 | spin_lock_init(&rtpz->lock); | |
6217 | } | |
6218 | } | |
6219 | } | |
6220 | ||
0eb253e2 | 6221 | static struct cgroup_subsys_state * __ref |
eb95419b | 6222 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 6223 | { |
d142e3e6 | 6224 | struct mem_cgroup *memcg; |
04046e1a | 6225 | long error = -ENOMEM; |
6d12e2d8 | 6226 | int node; |
8cdea7c0 | 6227 | |
c0ff4b85 R |
6228 | memcg = mem_cgroup_alloc(); |
6229 | if (!memcg) | |
04046e1a | 6230 | return ERR_PTR(error); |
78fb7466 | 6231 | |
3ed28fa1 | 6232 | for_each_node(node) |
c0ff4b85 | 6233 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 6234 | goto free_out; |
f64c3f54 | 6235 | |
c077719b | 6236 | /* root ? */ |
eb95419b | 6237 | if (parent_css == NULL) { |
a41c58a6 | 6238 | root_mem_cgroup = memcg; |
d142e3e6 GC |
6239 | res_counter_init(&memcg->res, NULL); |
6240 | res_counter_init(&memcg->memsw, NULL); | |
6241 | res_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 6242 | } |
28dbc4b6 | 6243 | |
d142e3e6 GC |
6244 | memcg->last_scanned_node = MAX_NUMNODES; |
6245 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
6246 | memcg->move_charge_at_immigrate = 0; |
6247 | mutex_init(&memcg->thresholds_lock); | |
6248 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 6249 | vmpressure_init(&memcg->vmpressure); |
d142e3e6 GC |
6250 | |
6251 | return &memcg->css; | |
6252 | ||
6253 | free_out: | |
6254 | __mem_cgroup_free(memcg); | |
6255 | return ERR_PTR(error); | |
6256 | } | |
6257 | ||
6258 | static int | |
eb95419b | 6259 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 6260 | { |
eb95419b TH |
6261 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
6262 | struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(css)); | |
d142e3e6 GC |
6263 | int error = 0; |
6264 | ||
63876986 | 6265 | if (!parent) |
d142e3e6 GC |
6266 | return 0; |
6267 | ||
0999821b | 6268 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
6269 | |
6270 | memcg->use_hierarchy = parent->use_hierarchy; | |
6271 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
6272 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
6273 | ||
6274 | if (parent->use_hierarchy) { | |
c0ff4b85 R |
6275 | res_counter_init(&memcg->res, &parent->res); |
6276 | res_counter_init(&memcg->memsw, &parent->memsw); | |
510fc4e1 | 6277 | res_counter_init(&memcg->kmem, &parent->kmem); |
55007d84 | 6278 | |
7bcc1bb1 | 6279 | /* |
8d76a979 LZ |
6280 | * No need to take a reference to the parent because cgroup |
6281 | * core guarantees its existence. | |
7bcc1bb1 | 6282 | */ |
18f59ea7 | 6283 | } else { |
c0ff4b85 R |
6284 | res_counter_init(&memcg->res, NULL); |
6285 | res_counter_init(&memcg->memsw, NULL); | |
510fc4e1 | 6286 | res_counter_init(&memcg->kmem, NULL); |
8c7f6edb TH |
6287 | /* |
6288 | * Deeper hierachy with use_hierarchy == false doesn't make | |
6289 | * much sense so let cgroup subsystem know about this | |
6290 | * unfortunate state in our controller. | |
6291 | */ | |
d142e3e6 | 6292 | if (parent != root_mem_cgroup) |
8c7f6edb | 6293 | mem_cgroup_subsys.broken_hierarchy = true; |
18f59ea7 | 6294 | } |
cbe128e3 GC |
6295 | |
6296 | error = memcg_init_kmem(memcg, &mem_cgroup_subsys); | |
0999821b | 6297 | mutex_unlock(&memcg_create_mutex); |
d142e3e6 | 6298 | return error; |
8cdea7c0 BS |
6299 | } |
6300 | ||
5f578161 MH |
6301 | /* |
6302 | * Announce all parents that a group from their hierarchy is gone. | |
6303 | */ | |
6304 | static void mem_cgroup_invalidate_reclaim_iterators(struct mem_cgroup *memcg) | |
6305 | { | |
6306 | struct mem_cgroup *parent = memcg; | |
6307 | ||
6308 | while ((parent = parent_mem_cgroup(parent))) | |
519ebea3 | 6309 | mem_cgroup_iter_invalidate(parent); |
5f578161 MH |
6310 | |
6311 | /* | |
6312 | * if the root memcg is not hierarchical we have to check it | |
6313 | * explicitely. | |
6314 | */ | |
6315 | if (!root_mem_cgroup->use_hierarchy) | |
519ebea3 | 6316 | mem_cgroup_iter_invalidate(root_mem_cgroup); |
5f578161 MH |
6317 | } |
6318 | ||
eb95419b | 6319 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 6320 | { |
eb95419b | 6321 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
ec64f515 | 6322 | |
10d5ebf4 LZ |
6323 | kmem_cgroup_css_offline(memcg); |
6324 | ||
5f578161 | 6325 | mem_cgroup_invalidate_reclaim_iterators(memcg); |
ab5196c2 | 6326 | mem_cgroup_reparent_charges(memcg); |
1f458cbf | 6327 | mem_cgroup_destroy_all_caches(memcg); |
33cb876e | 6328 | vmpressure_cleanup(&memcg->vmpressure); |
df878fb0 KH |
6329 | } |
6330 | ||
eb95419b | 6331 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 6332 | { |
eb95419b | 6333 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 6334 | |
10d5ebf4 | 6335 | memcg_destroy_kmem(memcg); |
465939a1 | 6336 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
6337 | } |
6338 | ||
02491447 | 6339 | #ifdef CONFIG_MMU |
7dc74be0 | 6340 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
6341 | #define PRECHARGE_COUNT_AT_ONCE 256 |
6342 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 6343 | { |
854ffa8d DN |
6344 | int ret = 0; |
6345 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
c0ff4b85 | 6346 | struct mem_cgroup *memcg = mc.to; |
4ffef5fe | 6347 | |
c0ff4b85 | 6348 | if (mem_cgroup_is_root(memcg)) { |
854ffa8d DN |
6349 | mc.precharge += count; |
6350 | /* we don't need css_get for root */ | |
6351 | return ret; | |
6352 | } | |
6353 | /* try to charge at once */ | |
6354 | if (count > 1) { | |
6355 | struct res_counter *dummy; | |
6356 | /* | |
c0ff4b85 | 6357 | * "memcg" cannot be under rmdir() because we've already checked |
854ffa8d DN |
6358 | * by cgroup_lock_live_cgroup() that it is not removed and we |
6359 | * are still under the same cgroup_mutex. So we can postpone | |
6360 | * css_get(). | |
6361 | */ | |
c0ff4b85 | 6362 | if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy)) |
854ffa8d | 6363 | goto one_by_one; |
c0ff4b85 | 6364 | if (do_swap_account && res_counter_charge(&memcg->memsw, |
854ffa8d | 6365 | PAGE_SIZE * count, &dummy)) { |
c0ff4b85 | 6366 | res_counter_uncharge(&memcg->res, PAGE_SIZE * count); |
854ffa8d DN |
6367 | goto one_by_one; |
6368 | } | |
6369 | mc.precharge += count; | |
854ffa8d DN |
6370 | return ret; |
6371 | } | |
6372 | one_by_one: | |
6373 | /* fall back to one by one charge */ | |
6374 | while (count--) { | |
6375 | if (signal_pending(current)) { | |
6376 | ret = -EINTR; | |
6377 | break; | |
6378 | } | |
6379 | if (!batch_count--) { | |
6380 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
6381 | cond_resched(); | |
6382 | } | |
c0ff4b85 R |
6383 | ret = __mem_cgroup_try_charge(NULL, |
6384 | GFP_KERNEL, 1, &memcg, false); | |
38c5d72f | 6385 | if (ret) |
854ffa8d | 6386 | /* mem_cgroup_clear_mc() will do uncharge later */ |
38c5d72f | 6387 | return ret; |
854ffa8d DN |
6388 | mc.precharge++; |
6389 | } | |
4ffef5fe DN |
6390 | return ret; |
6391 | } | |
6392 | ||
6393 | /** | |
8d32ff84 | 6394 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
6395 | * @vma: the vma the pte to be checked belongs |
6396 | * @addr: the address corresponding to the pte to be checked | |
6397 | * @ptent: the pte to be checked | |
02491447 | 6398 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
6399 | * |
6400 | * Returns | |
6401 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
6402 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
6403 | * move charge. if @target is not NULL, the page is stored in target->page | |
6404 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
6405 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
6406 | * target for charge migration. if @target is not NULL, the entry is stored | |
6407 | * in target->ent. | |
4ffef5fe DN |
6408 | * |
6409 | * Called with pte lock held. | |
6410 | */ | |
4ffef5fe DN |
6411 | union mc_target { |
6412 | struct page *page; | |
02491447 | 6413 | swp_entry_t ent; |
4ffef5fe DN |
6414 | }; |
6415 | ||
4ffef5fe | 6416 | enum mc_target_type { |
8d32ff84 | 6417 | MC_TARGET_NONE = 0, |
4ffef5fe | 6418 | MC_TARGET_PAGE, |
02491447 | 6419 | MC_TARGET_SWAP, |
4ffef5fe DN |
6420 | }; |
6421 | ||
90254a65 DN |
6422 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
6423 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 6424 | { |
90254a65 | 6425 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 6426 | |
90254a65 DN |
6427 | if (!page || !page_mapped(page)) |
6428 | return NULL; | |
6429 | if (PageAnon(page)) { | |
6430 | /* we don't move shared anon */ | |
4b91355e | 6431 | if (!move_anon()) |
90254a65 | 6432 | return NULL; |
87946a72 DN |
6433 | } else if (!move_file()) |
6434 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
6435 | return NULL; |
6436 | if (!get_page_unless_zero(page)) | |
6437 | return NULL; | |
6438 | ||
6439 | return page; | |
6440 | } | |
6441 | ||
4b91355e | 6442 | #ifdef CONFIG_SWAP |
90254a65 DN |
6443 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
6444 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6445 | { | |
90254a65 DN |
6446 | struct page *page = NULL; |
6447 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
6448 | ||
6449 | if (!move_anon() || non_swap_entry(ent)) | |
6450 | return NULL; | |
4b91355e KH |
6451 | /* |
6452 | * Because lookup_swap_cache() updates some statistics counter, | |
6453 | * we call find_get_page() with swapper_space directly. | |
6454 | */ | |
33806f06 | 6455 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
6456 | if (do_swap_account) |
6457 | entry->val = ent.val; | |
6458 | ||
6459 | return page; | |
6460 | } | |
4b91355e KH |
6461 | #else |
6462 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
6463 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6464 | { | |
6465 | return NULL; | |
6466 | } | |
6467 | #endif | |
90254a65 | 6468 | |
87946a72 DN |
6469 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
6470 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6471 | { | |
6472 | struct page *page = NULL; | |
87946a72 DN |
6473 | struct address_space *mapping; |
6474 | pgoff_t pgoff; | |
6475 | ||
6476 | if (!vma->vm_file) /* anonymous vma */ | |
6477 | return NULL; | |
6478 | if (!move_file()) | |
6479 | return NULL; | |
6480 | ||
87946a72 DN |
6481 | mapping = vma->vm_file->f_mapping; |
6482 | if (pte_none(ptent)) | |
6483 | pgoff = linear_page_index(vma, addr); | |
6484 | else /* pte_file(ptent) is true */ | |
6485 | pgoff = pte_to_pgoff(ptent); | |
6486 | ||
6487 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
6488 | page = find_get_page(mapping, pgoff); |
6489 | ||
6490 | #ifdef CONFIG_SWAP | |
6491 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
6492 | if (radix_tree_exceptional_entry(page)) { | |
6493 | swp_entry_t swap = radix_to_swp_entry(page); | |
87946a72 | 6494 | if (do_swap_account) |
aa3b1895 | 6495 | *entry = swap; |
33806f06 | 6496 | page = find_get_page(swap_address_space(swap), swap.val); |
87946a72 | 6497 | } |
aa3b1895 | 6498 | #endif |
87946a72 DN |
6499 | return page; |
6500 | } | |
6501 | ||
8d32ff84 | 6502 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
6503 | unsigned long addr, pte_t ptent, union mc_target *target) |
6504 | { | |
6505 | struct page *page = NULL; | |
6506 | struct page_cgroup *pc; | |
8d32ff84 | 6507 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
6508 | swp_entry_t ent = { .val = 0 }; |
6509 | ||
6510 | if (pte_present(ptent)) | |
6511 | page = mc_handle_present_pte(vma, addr, ptent); | |
6512 | else if (is_swap_pte(ptent)) | |
6513 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
6514 | else if (pte_none(ptent) || pte_file(ptent)) |
6515 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
6516 | |
6517 | if (!page && !ent.val) | |
8d32ff84 | 6518 | return ret; |
02491447 DN |
6519 | if (page) { |
6520 | pc = lookup_page_cgroup(page); | |
6521 | /* | |
6522 | * Do only loose check w/o page_cgroup lock. | |
6523 | * mem_cgroup_move_account() checks the pc is valid or not under | |
6524 | * the lock. | |
6525 | */ | |
6526 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
6527 | ret = MC_TARGET_PAGE; | |
6528 | if (target) | |
6529 | target->page = page; | |
6530 | } | |
6531 | if (!ret || !target) | |
6532 | put_page(page); | |
6533 | } | |
90254a65 DN |
6534 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
6535 | if (ent.val && !ret && | |
9fb4b7cc | 6536 | css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
6537 | ret = MC_TARGET_SWAP; |
6538 | if (target) | |
6539 | target->ent = ent; | |
4ffef5fe | 6540 | } |
4ffef5fe DN |
6541 | return ret; |
6542 | } | |
6543 | ||
12724850 NH |
6544 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
6545 | /* | |
6546 | * We don't consider swapping or file mapped pages because THP does not | |
6547 | * support them for now. | |
6548 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
6549 | */ | |
6550 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6551 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6552 | { | |
6553 | struct page *page = NULL; | |
6554 | struct page_cgroup *pc; | |
6555 | enum mc_target_type ret = MC_TARGET_NONE; | |
6556 | ||
6557 | page = pmd_page(pmd); | |
6558 | VM_BUG_ON(!page || !PageHead(page)); | |
6559 | if (!move_anon()) | |
6560 | return ret; | |
6561 | pc = lookup_page_cgroup(page); | |
6562 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
6563 | ret = MC_TARGET_PAGE; | |
6564 | if (target) { | |
6565 | get_page(page); | |
6566 | target->page = page; | |
6567 | } | |
6568 | } | |
6569 | return ret; | |
6570 | } | |
6571 | #else | |
6572 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6573 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6574 | { | |
6575 | return MC_TARGET_NONE; | |
6576 | } | |
6577 | #endif | |
6578 | ||
4ffef5fe DN |
6579 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
6580 | unsigned long addr, unsigned long end, | |
6581 | struct mm_walk *walk) | |
6582 | { | |
6583 | struct vm_area_struct *vma = walk->private; | |
6584 | pte_t *pte; | |
6585 | spinlock_t *ptl; | |
6586 | ||
12724850 NH |
6587 | if (pmd_trans_huge_lock(pmd, vma) == 1) { |
6588 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) | |
6589 | mc.precharge += HPAGE_PMD_NR; | |
6590 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 6591 | return 0; |
12724850 | 6592 | } |
03319327 | 6593 | |
45f83cef AA |
6594 | if (pmd_trans_unstable(pmd)) |
6595 | return 0; | |
4ffef5fe DN |
6596 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
6597 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 6598 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
6599 | mc.precharge++; /* increment precharge temporarily */ |
6600 | pte_unmap_unlock(pte - 1, ptl); | |
6601 | cond_resched(); | |
6602 | ||
7dc74be0 DN |
6603 | return 0; |
6604 | } | |
6605 | ||
4ffef5fe DN |
6606 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
6607 | { | |
6608 | unsigned long precharge; | |
6609 | struct vm_area_struct *vma; | |
6610 | ||
dfe076b0 | 6611 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
6612 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
6613 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
6614 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
6615 | .mm = mm, | |
6616 | .private = vma, | |
6617 | }; | |
6618 | if (is_vm_hugetlb_page(vma)) | |
6619 | continue; | |
4ffef5fe DN |
6620 | walk_page_range(vma->vm_start, vma->vm_end, |
6621 | &mem_cgroup_count_precharge_walk); | |
6622 | } | |
dfe076b0 | 6623 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
6624 | |
6625 | precharge = mc.precharge; | |
6626 | mc.precharge = 0; | |
6627 | ||
6628 | return precharge; | |
6629 | } | |
6630 | ||
4ffef5fe DN |
6631 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
6632 | { | |
dfe076b0 DN |
6633 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
6634 | ||
6635 | VM_BUG_ON(mc.moving_task); | |
6636 | mc.moving_task = current; | |
6637 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
6638 | } |
6639 | ||
dfe076b0 DN |
6640 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
6641 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 6642 | { |
2bd9bb20 KH |
6643 | struct mem_cgroup *from = mc.from; |
6644 | struct mem_cgroup *to = mc.to; | |
4050377b | 6645 | int i; |
2bd9bb20 | 6646 | |
4ffef5fe | 6647 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
6648 | if (mc.precharge) { |
6649 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
6650 | mc.precharge = 0; | |
6651 | } | |
6652 | /* | |
6653 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
6654 | * we must uncharge here. | |
6655 | */ | |
6656 | if (mc.moved_charge) { | |
6657 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
6658 | mc.moved_charge = 0; | |
4ffef5fe | 6659 | } |
483c30b5 DN |
6660 | /* we must fixup refcnts and charges */ |
6661 | if (mc.moved_swap) { | |
483c30b5 DN |
6662 | /* uncharge swap account from the old cgroup */ |
6663 | if (!mem_cgroup_is_root(mc.from)) | |
6664 | res_counter_uncharge(&mc.from->memsw, | |
6665 | PAGE_SIZE * mc.moved_swap); | |
4050377b LZ |
6666 | |
6667 | for (i = 0; i < mc.moved_swap; i++) | |
6668 | css_put(&mc.from->css); | |
483c30b5 DN |
6669 | |
6670 | if (!mem_cgroup_is_root(mc.to)) { | |
6671 | /* | |
6672 | * we charged both to->res and to->memsw, so we should | |
6673 | * uncharge to->res. | |
6674 | */ | |
6675 | res_counter_uncharge(&mc.to->res, | |
6676 | PAGE_SIZE * mc.moved_swap); | |
483c30b5 | 6677 | } |
4050377b | 6678 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
6679 | mc.moved_swap = 0; |
6680 | } | |
dfe076b0 DN |
6681 | memcg_oom_recover(from); |
6682 | memcg_oom_recover(to); | |
6683 | wake_up_all(&mc.waitq); | |
6684 | } | |
6685 | ||
6686 | static void mem_cgroup_clear_mc(void) | |
6687 | { | |
6688 | struct mem_cgroup *from = mc.from; | |
6689 | ||
6690 | /* | |
6691 | * we must clear moving_task before waking up waiters at the end of | |
6692 | * task migration. | |
6693 | */ | |
6694 | mc.moving_task = NULL; | |
6695 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 6696 | spin_lock(&mc.lock); |
4ffef5fe DN |
6697 | mc.from = NULL; |
6698 | mc.to = NULL; | |
2bd9bb20 | 6699 | spin_unlock(&mc.lock); |
32047e2a | 6700 | mem_cgroup_end_move(from); |
4ffef5fe DN |
6701 | } |
6702 | ||
eb95419b | 6703 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6704 | struct cgroup_taskset *tset) |
7dc74be0 | 6705 | { |
2f7ee569 | 6706 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 6707 | int ret = 0; |
eb95419b | 6708 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
ee5e8472 | 6709 | unsigned long move_charge_at_immigrate; |
7dc74be0 | 6710 | |
ee5e8472 GC |
6711 | /* |
6712 | * We are now commited to this value whatever it is. Changes in this | |
6713 | * tunable will only affect upcoming migrations, not the current one. | |
6714 | * So we need to save it, and keep it going. | |
6715 | */ | |
6716 | move_charge_at_immigrate = memcg->move_charge_at_immigrate; | |
6717 | if (move_charge_at_immigrate) { | |
7dc74be0 DN |
6718 | struct mm_struct *mm; |
6719 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
6720 | ||
c0ff4b85 | 6721 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
6722 | |
6723 | mm = get_task_mm(p); | |
6724 | if (!mm) | |
6725 | return 0; | |
7dc74be0 | 6726 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
6727 | if (mm->owner == p) { |
6728 | VM_BUG_ON(mc.from); | |
6729 | VM_BUG_ON(mc.to); | |
6730 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 6731 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 6732 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 6733 | mem_cgroup_start_move(from); |
2bd9bb20 | 6734 | spin_lock(&mc.lock); |
4ffef5fe | 6735 | mc.from = from; |
c0ff4b85 | 6736 | mc.to = memcg; |
ee5e8472 | 6737 | mc.immigrate_flags = move_charge_at_immigrate; |
2bd9bb20 | 6738 | spin_unlock(&mc.lock); |
dfe076b0 | 6739 | /* We set mc.moving_task later */ |
4ffef5fe DN |
6740 | |
6741 | ret = mem_cgroup_precharge_mc(mm); | |
6742 | if (ret) | |
6743 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
6744 | } |
6745 | mmput(mm); | |
7dc74be0 DN |
6746 | } |
6747 | return ret; | |
6748 | } | |
6749 | ||
eb95419b | 6750 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6751 | struct cgroup_taskset *tset) |
7dc74be0 | 6752 | { |
4ffef5fe | 6753 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
6754 | } |
6755 | ||
4ffef5fe DN |
6756 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
6757 | unsigned long addr, unsigned long end, | |
6758 | struct mm_walk *walk) | |
7dc74be0 | 6759 | { |
4ffef5fe DN |
6760 | int ret = 0; |
6761 | struct vm_area_struct *vma = walk->private; | |
6762 | pte_t *pte; | |
6763 | spinlock_t *ptl; | |
12724850 NH |
6764 | enum mc_target_type target_type; |
6765 | union mc_target target; | |
6766 | struct page *page; | |
6767 | struct page_cgroup *pc; | |
4ffef5fe | 6768 | |
12724850 NH |
6769 | /* |
6770 | * We don't take compound_lock() here but no race with splitting thp | |
6771 | * happens because: | |
6772 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
6773 | * under splitting, which means there's no concurrent thp split, | |
6774 | * - if another thread runs into split_huge_page() just after we | |
6775 | * entered this if-block, the thread must wait for page table lock | |
6776 | * to be unlocked in __split_huge_page_splitting(), where the main | |
6777 | * part of thp split is not executed yet. | |
6778 | */ | |
6779 | if (pmd_trans_huge_lock(pmd, vma) == 1) { | |
62ade86a | 6780 | if (mc.precharge < HPAGE_PMD_NR) { |
12724850 NH |
6781 | spin_unlock(&vma->vm_mm->page_table_lock); |
6782 | return 0; | |
6783 | } | |
6784 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
6785 | if (target_type == MC_TARGET_PAGE) { | |
6786 | page = target.page; | |
6787 | if (!isolate_lru_page(page)) { | |
6788 | pc = lookup_page_cgroup(page); | |
6789 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, | |
2f3479b1 | 6790 | pc, mc.from, mc.to)) { |
12724850 NH |
6791 | mc.precharge -= HPAGE_PMD_NR; |
6792 | mc.moved_charge += HPAGE_PMD_NR; | |
6793 | } | |
6794 | putback_lru_page(page); | |
6795 | } | |
6796 | put_page(page); | |
6797 | } | |
6798 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 6799 | return 0; |
12724850 NH |
6800 | } |
6801 | ||
45f83cef AA |
6802 | if (pmd_trans_unstable(pmd)) |
6803 | return 0; | |
4ffef5fe DN |
6804 | retry: |
6805 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
6806 | for (; addr != end; addr += PAGE_SIZE) { | |
6807 | pte_t ptent = *(pte++); | |
02491447 | 6808 | swp_entry_t ent; |
4ffef5fe DN |
6809 | |
6810 | if (!mc.precharge) | |
6811 | break; | |
6812 | ||
8d32ff84 | 6813 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
6814 | case MC_TARGET_PAGE: |
6815 | page = target.page; | |
6816 | if (isolate_lru_page(page)) | |
6817 | goto put; | |
6818 | pc = lookup_page_cgroup(page); | |
7ec99d62 | 6819 | if (!mem_cgroup_move_account(page, 1, pc, |
2f3479b1 | 6820 | mc.from, mc.to)) { |
4ffef5fe | 6821 | mc.precharge--; |
854ffa8d DN |
6822 | /* we uncharge from mc.from later. */ |
6823 | mc.moved_charge++; | |
4ffef5fe DN |
6824 | } |
6825 | putback_lru_page(page); | |
8d32ff84 | 6826 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
6827 | put_page(page); |
6828 | break; | |
02491447 DN |
6829 | case MC_TARGET_SWAP: |
6830 | ent = target.ent; | |
e91cbb42 | 6831 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6832 | mc.precharge--; |
483c30b5 DN |
6833 | /* we fixup refcnts and charges later. */ |
6834 | mc.moved_swap++; | |
6835 | } | |
02491447 | 6836 | break; |
4ffef5fe DN |
6837 | default: |
6838 | break; | |
6839 | } | |
6840 | } | |
6841 | pte_unmap_unlock(pte - 1, ptl); | |
6842 | cond_resched(); | |
6843 | ||
6844 | if (addr != end) { | |
6845 | /* | |
6846 | * We have consumed all precharges we got in can_attach(). | |
6847 | * We try charge one by one, but don't do any additional | |
6848 | * charges to mc.to if we have failed in charge once in attach() | |
6849 | * phase. | |
6850 | */ | |
854ffa8d | 6851 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6852 | if (!ret) |
6853 | goto retry; | |
6854 | } | |
6855 | ||
6856 | return ret; | |
6857 | } | |
6858 | ||
6859 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
6860 | { | |
6861 | struct vm_area_struct *vma; | |
6862 | ||
6863 | lru_add_drain_all(); | |
dfe076b0 DN |
6864 | retry: |
6865 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
6866 | /* | |
6867 | * Someone who are holding the mmap_sem might be waiting in | |
6868 | * waitq. So we cancel all extra charges, wake up all waiters, | |
6869 | * and retry. Because we cancel precharges, we might not be able | |
6870 | * to move enough charges, but moving charge is a best-effort | |
6871 | * feature anyway, so it wouldn't be a big problem. | |
6872 | */ | |
6873 | __mem_cgroup_clear_mc(); | |
6874 | cond_resched(); | |
6875 | goto retry; | |
6876 | } | |
4ffef5fe DN |
6877 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
6878 | int ret; | |
6879 | struct mm_walk mem_cgroup_move_charge_walk = { | |
6880 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6881 | .mm = mm, | |
6882 | .private = vma, | |
6883 | }; | |
6884 | if (is_vm_hugetlb_page(vma)) | |
6885 | continue; | |
4ffef5fe DN |
6886 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
6887 | &mem_cgroup_move_charge_walk); | |
6888 | if (ret) | |
6889 | /* | |
6890 | * means we have consumed all precharges and failed in | |
6891 | * doing additional charge. Just abandon here. | |
6892 | */ | |
6893 | break; | |
6894 | } | |
dfe076b0 | 6895 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
6896 | } |
6897 | ||
eb95419b | 6898 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 6899 | struct cgroup_taskset *tset) |
67e465a7 | 6900 | { |
2f7ee569 | 6901 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 6902 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 6903 | |
dfe076b0 | 6904 | if (mm) { |
a433658c KM |
6905 | if (mc.to) |
6906 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
6907 | mmput(mm); |
6908 | } | |
a433658c KM |
6909 | if (mc.to) |
6910 | mem_cgroup_clear_mc(); | |
67e465a7 | 6911 | } |
5cfb80a7 | 6912 | #else /* !CONFIG_MMU */ |
eb95419b | 6913 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6914 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
6915 | { |
6916 | return 0; | |
6917 | } | |
eb95419b | 6918 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6919 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
6920 | { |
6921 | } | |
eb95419b | 6922 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 6923 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
6924 | { |
6925 | } | |
6926 | #endif | |
67e465a7 | 6927 | |
f00baae7 TH |
6928 | /* |
6929 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
6930 | * to verify sane_behavior flag on each mount attempt. | |
6931 | */ | |
eb95419b | 6932 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
6933 | { |
6934 | /* | |
6935 | * use_hierarchy is forced with sane_behavior. cgroup core | |
6936 | * guarantees that @root doesn't have any children, so turning it | |
6937 | * on for the root memcg is enough. | |
6938 | */ | |
eb95419b TH |
6939 | if (cgroup_sane_behavior(root_css->cgroup)) |
6940 | mem_cgroup_from_css(root_css)->use_hierarchy = true; | |
f00baae7 TH |
6941 | } |
6942 | ||
8cdea7c0 BS |
6943 | struct cgroup_subsys mem_cgroup_subsys = { |
6944 | .name = "memory", | |
6945 | .subsys_id = mem_cgroup_subsys_id, | |
92fb9748 | 6946 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6947 | .css_online = mem_cgroup_css_online, |
92fb9748 TH |
6948 | .css_offline = mem_cgroup_css_offline, |
6949 | .css_free = mem_cgroup_css_free, | |
7dc74be0 DN |
6950 | .can_attach = mem_cgroup_can_attach, |
6951 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 6952 | .attach = mem_cgroup_move_task, |
f00baae7 | 6953 | .bind = mem_cgroup_bind, |
6bc10349 | 6954 | .base_cftypes = mem_cgroup_files, |
6d12e2d8 | 6955 | .early_init = 0, |
04046e1a | 6956 | .use_id = 1, |
8cdea7c0 | 6957 | }; |
c077719b | 6958 | |
c255a458 | 6959 | #ifdef CONFIG_MEMCG_SWAP |
a42c390c MH |
6960 | static int __init enable_swap_account(char *s) |
6961 | { | |
a2c8990a | 6962 | if (!strcmp(s, "1")) |
a42c390c | 6963 | really_do_swap_account = 1; |
a2c8990a | 6964 | else if (!strcmp(s, "0")) |
a42c390c MH |
6965 | really_do_swap_account = 0; |
6966 | return 1; | |
6967 | } | |
a2c8990a | 6968 | __setup("swapaccount=", enable_swap_account); |
c077719b | 6969 | |
2d11085e MH |
6970 | static void __init memsw_file_init(void) |
6971 | { | |
6acc8b02 MH |
6972 | WARN_ON(cgroup_add_cftypes(&mem_cgroup_subsys, memsw_cgroup_files)); |
6973 | } | |
6974 | ||
6975 | static void __init enable_swap_cgroup(void) | |
6976 | { | |
6977 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
6978 | do_swap_account = 1; | |
6979 | memsw_file_init(); | |
6980 | } | |
2d11085e | 6981 | } |
6acc8b02 | 6982 | |
2d11085e | 6983 | #else |
6acc8b02 | 6984 | static void __init enable_swap_cgroup(void) |
2d11085e MH |
6985 | { |
6986 | } | |
c077719b | 6987 | #endif |
2d11085e MH |
6988 | |
6989 | /* | |
1081312f MH |
6990 | * subsys_initcall() for memory controller. |
6991 | * | |
6992 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
6993 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
6994 | * everything that doesn't depend on a specific mem_cgroup structure should | |
6995 | * be initialized from here. | |
2d11085e MH |
6996 | */ |
6997 | static int __init mem_cgroup_init(void) | |
6998 | { | |
6999 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); | |
6acc8b02 | 7000 | enable_swap_cgroup(); |
bb4cc1a8 | 7001 | mem_cgroup_soft_limit_tree_init(); |
e4777496 | 7002 | memcg_stock_init(); |
2d11085e MH |
7003 | return 0; |
7004 | } | |
7005 | subsys_initcall(mem_cgroup_init); |