]>
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 | * | |
1575e68b JW |
17 | * Native page reclaim |
18 | * Charge lifetime sanitation | |
19 | * Lockless page tracking & accounting | |
20 | * Unified hierarchy configuration model | |
21 | * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner | |
22 | * | |
8cdea7c0 BS |
23 | * This program is free software; you can redistribute it and/or modify |
24 | * it under the terms of the GNU General Public License as published by | |
25 | * the Free Software Foundation; either version 2 of the License, or | |
26 | * (at your option) any later version. | |
27 | * | |
28 | * This program is distributed in the hope that it will be useful, | |
29 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
30 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
31 | * GNU General Public License for more details. | |
32 | */ | |
33 | ||
3e32cb2e | 34 | #include <linux/page_counter.h> |
8cdea7c0 BS |
35 | #include <linux/memcontrol.h> |
36 | #include <linux/cgroup.h> | |
78fb7466 | 37 | #include <linux/mm.h> |
4ffef5fe | 38 | #include <linux/hugetlb.h> |
d13d1443 | 39 | #include <linux/pagemap.h> |
d52aa412 | 40 | #include <linux/smp.h> |
8a9f3ccd | 41 | #include <linux/page-flags.h> |
66e1707b | 42 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
43 | #include <linux/bit_spinlock.h> |
44 | #include <linux/rcupdate.h> | |
e222432b | 45 | #include <linux/limits.h> |
b9e15baf | 46 | #include <linux/export.h> |
8c7c6e34 | 47 | #include <linux/mutex.h> |
bb4cc1a8 | 48 | #include <linux/rbtree.h> |
b6ac57d5 | 49 | #include <linux/slab.h> |
66e1707b | 50 | #include <linux/swap.h> |
02491447 | 51 | #include <linux/swapops.h> |
66e1707b | 52 | #include <linux/spinlock.h> |
2e72b634 | 53 | #include <linux/eventfd.h> |
79bd9814 | 54 | #include <linux/poll.h> |
2e72b634 | 55 | #include <linux/sort.h> |
66e1707b | 56 | #include <linux/fs.h> |
d2ceb9b7 | 57 | #include <linux/seq_file.h> |
70ddf637 | 58 | #include <linux/vmpressure.h> |
b69408e8 | 59 | #include <linux/mm_inline.h> |
5d1ea48b | 60 | #include <linux/swap_cgroup.h> |
cdec2e42 | 61 | #include <linux/cpu.h> |
158e0a2d | 62 | #include <linux/oom.h> |
0056f4e6 | 63 | #include <linux/lockdep.h> |
79bd9814 | 64 | #include <linux/file.h> |
b23afb93 | 65 | #include <linux/tracehook.h> |
08e552c6 | 66 | #include "internal.h" |
d1a4c0b3 | 67 | #include <net/sock.h> |
4bd2c1ee | 68 | #include <net/ip.h> |
d1a4c0b3 | 69 | #include <net/tcp_memcontrol.h> |
f35c3a8e | 70 | #include "slab.h" |
8cdea7c0 | 71 | |
8697d331 BS |
72 | #include <asm/uaccess.h> |
73 | ||
cc8e970c KM |
74 | #include <trace/events/vmscan.h> |
75 | ||
073219e9 TH |
76 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
77 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 78 | |
7d828602 JW |
79 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
80 | ||
a181b0e8 | 81 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
8cdea7c0 | 82 | |
21afa38e | 83 | /* Whether the swap controller is active */ |
c255a458 | 84 | #ifdef CONFIG_MEMCG_SWAP |
c077719b | 85 | int do_swap_account __read_mostly; |
c077719b | 86 | #else |
a0db00fc | 87 | #define do_swap_account 0 |
c077719b KH |
88 | #endif |
89 | ||
af7c4b0e JW |
90 | static const char * const mem_cgroup_stat_names[] = { |
91 | "cache", | |
92 | "rss", | |
b070e65c | 93 | "rss_huge", |
af7c4b0e | 94 | "mapped_file", |
c4843a75 | 95 | "dirty", |
3ea67d06 | 96 | "writeback", |
af7c4b0e JW |
97 | "swap", |
98 | }; | |
99 | ||
af7c4b0e JW |
100 | static const char * const mem_cgroup_events_names[] = { |
101 | "pgpgin", | |
102 | "pgpgout", | |
103 | "pgfault", | |
104 | "pgmajfault", | |
105 | }; | |
106 | ||
58cf188e SZ |
107 | static const char * const mem_cgroup_lru_names[] = { |
108 | "inactive_anon", | |
109 | "active_anon", | |
110 | "inactive_file", | |
111 | "active_file", | |
112 | "unevictable", | |
113 | }; | |
114 | ||
a0db00fc KS |
115 | #define THRESHOLDS_EVENTS_TARGET 128 |
116 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
117 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 118 | |
bb4cc1a8 AM |
119 | /* |
120 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
121 | * their hierarchy representation | |
122 | */ | |
123 | ||
124 | struct mem_cgroup_tree_per_zone { | |
125 | struct rb_root rb_root; | |
126 | spinlock_t lock; | |
127 | }; | |
128 | ||
129 | struct mem_cgroup_tree_per_node { | |
130 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
131 | }; | |
132 | ||
133 | struct mem_cgroup_tree { | |
134 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
135 | }; | |
136 | ||
137 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
138 | ||
9490ff27 KH |
139 | /* for OOM */ |
140 | struct mem_cgroup_eventfd_list { | |
141 | struct list_head list; | |
142 | struct eventfd_ctx *eventfd; | |
143 | }; | |
2e72b634 | 144 | |
79bd9814 TH |
145 | /* |
146 | * cgroup_event represents events which userspace want to receive. | |
147 | */ | |
3bc942f3 | 148 | struct mem_cgroup_event { |
79bd9814 | 149 | /* |
59b6f873 | 150 | * memcg which the event belongs to. |
79bd9814 | 151 | */ |
59b6f873 | 152 | struct mem_cgroup *memcg; |
79bd9814 TH |
153 | /* |
154 | * eventfd to signal userspace about the event. | |
155 | */ | |
156 | struct eventfd_ctx *eventfd; | |
157 | /* | |
158 | * Each of these stored in a list by the cgroup. | |
159 | */ | |
160 | struct list_head list; | |
fba94807 TH |
161 | /* |
162 | * register_event() callback will be used to add new userspace | |
163 | * waiter for changes related to this event. Use eventfd_signal() | |
164 | * on eventfd to send notification to userspace. | |
165 | */ | |
59b6f873 | 166 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 167 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
168 | /* |
169 | * unregister_event() callback will be called when userspace closes | |
170 | * the eventfd or on cgroup removing. This callback must be set, | |
171 | * if you want provide notification functionality. | |
172 | */ | |
59b6f873 | 173 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 174 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
175 | /* |
176 | * All fields below needed to unregister event when | |
177 | * userspace closes eventfd. | |
178 | */ | |
179 | poll_table pt; | |
180 | wait_queue_head_t *wqh; | |
181 | wait_queue_t wait; | |
182 | struct work_struct remove; | |
183 | }; | |
184 | ||
c0ff4b85 R |
185 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
186 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 187 | |
7dc74be0 DN |
188 | /* Stuffs for move charges at task migration. */ |
189 | /* | |
1dfab5ab | 190 | * Types of charges to be moved. |
7dc74be0 | 191 | */ |
1dfab5ab JW |
192 | #define MOVE_ANON 0x1U |
193 | #define MOVE_FILE 0x2U | |
194 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 195 | |
4ffef5fe DN |
196 | /* "mc" and its members are protected by cgroup_mutex */ |
197 | static struct move_charge_struct { | |
b1dd693e | 198 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
199 | struct mem_cgroup *from; |
200 | struct mem_cgroup *to; | |
1dfab5ab | 201 | unsigned long flags; |
4ffef5fe | 202 | unsigned long precharge; |
854ffa8d | 203 | unsigned long moved_charge; |
483c30b5 | 204 | unsigned long moved_swap; |
8033b97c DN |
205 | struct task_struct *moving_task; /* a task moving charges */ |
206 | wait_queue_head_t waitq; /* a waitq for other context */ | |
207 | } mc = { | |
2bd9bb20 | 208 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
209 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
210 | }; | |
4ffef5fe | 211 | |
4e416953 BS |
212 | /* |
213 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
214 | * limit reclaim to prevent infinite loops, if they ever occur. | |
215 | */ | |
a0db00fc | 216 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 217 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 218 | |
217bc319 KH |
219 | enum charge_type { |
220 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 221 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 222 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 223 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
224 | NR_CHARGE_TYPE, |
225 | }; | |
226 | ||
8c7c6e34 | 227 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
228 | enum res_type { |
229 | _MEM, | |
230 | _MEMSWAP, | |
231 | _OOM_TYPE, | |
510fc4e1 | 232 | _KMEM, |
86ae53e1 GC |
233 | }; |
234 | ||
a0db00fc KS |
235 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
236 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 237 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
238 | /* Used for OOM nofiier */ |
239 | #define OOM_CONTROL (0) | |
8c7c6e34 | 240 | |
0999821b GC |
241 | /* |
242 | * The memcg_create_mutex will be held whenever a new cgroup is created. | |
243 | * As a consequence, any change that needs to protect against new child cgroups | |
244 | * appearing has to hold it as well. | |
245 | */ | |
246 | static DEFINE_MUTEX(memcg_create_mutex); | |
247 | ||
70ddf637 AV |
248 | /* Some nice accessors for the vmpressure. */ |
249 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
250 | { | |
251 | if (!memcg) | |
252 | memcg = root_mem_cgroup; | |
253 | return &memcg->vmpressure; | |
254 | } | |
255 | ||
256 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
257 | { | |
258 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
259 | } | |
260 | ||
7ffc0edc MH |
261 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
262 | { | |
263 | return (memcg == root_mem_cgroup); | |
264 | } | |
265 | ||
4219b2da LZ |
266 | /* |
267 | * We restrict the id in the range of [1, 65535], so it can fit into | |
268 | * an unsigned short. | |
269 | */ | |
270 | #define MEM_CGROUP_ID_MAX USHRT_MAX | |
271 | ||
34c00c31 LZ |
272 | static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) |
273 | { | |
15a4c835 | 274 | return memcg->css.id; |
34c00c31 LZ |
275 | } |
276 | ||
adbe427b VD |
277 | /* |
278 | * A helper function to get mem_cgroup from ID. must be called under | |
279 | * rcu_read_lock(). The caller is responsible for calling | |
280 | * css_tryget_online() if the mem_cgroup is used for charging. (dropping | |
281 | * refcnt from swap can be called against removed memcg.) | |
282 | */ | |
34c00c31 LZ |
283 | static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) |
284 | { | |
285 | struct cgroup_subsys_state *css; | |
286 | ||
7d699ddb | 287 | css = css_from_id(id, &memory_cgrp_subsys); |
34c00c31 LZ |
288 | return mem_cgroup_from_css(css); |
289 | } | |
290 | ||
e1aab161 | 291 | /* Writing them here to avoid exposing memcg's inner layout */ |
4bd2c1ee | 292 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
e1aab161 | 293 | |
80e95fe0 JW |
294 | struct static_key memcg_sockets_enabled_key; |
295 | EXPORT_SYMBOL(memcg_sockets_enabled_key); | |
296 | ||
e1aab161 GC |
297 | void sock_update_memcg(struct sock *sk) |
298 | { | |
3d596f7b | 299 | struct mem_cgroup *memcg; |
3d596f7b JW |
300 | |
301 | /* Socket cloning can throw us here with sk_cgrp already | |
302 | * filled. It won't however, necessarily happen from | |
303 | * process context. So the test for root memcg given | |
304 | * the current task's memcg won't help us in this case. | |
305 | * | |
306 | * Respecting the original socket's memcg is a better | |
307 | * decision in this case. | |
308 | */ | |
baac50bb JW |
309 | if (sk->sk_memcg) { |
310 | BUG_ON(mem_cgroup_is_root(sk->sk_memcg)); | |
311 | css_get(&sk->sk_memcg->css); | |
3d596f7b | 312 | return; |
e1aab161 | 313 | } |
3d596f7b JW |
314 | |
315 | rcu_read_lock(); | |
316 | memcg = mem_cgroup_from_task(current); | |
baac50bb JW |
317 | if (memcg != root_mem_cgroup && |
318 | memcg->tcp_mem.active && | |
319 | css_tryget_online(&memcg->css)) | |
320 | sk->sk_memcg = memcg; | |
3d596f7b | 321 | rcu_read_unlock(); |
e1aab161 GC |
322 | } |
323 | EXPORT_SYMBOL(sock_update_memcg); | |
324 | ||
325 | void sock_release_memcg(struct sock *sk) | |
326 | { | |
baac50bb JW |
327 | WARN_ON(!sk->sk_memcg); |
328 | css_put(&sk->sk_memcg->css); | |
d1a4c0b3 | 329 | } |
e1aab161 | 330 | |
e805605c JW |
331 | /** |
332 | * mem_cgroup_charge_skmem - charge socket memory | |
baac50bb | 333 | * @memcg: memcg to charge |
e805605c JW |
334 | * @nr_pages: number of pages to charge |
335 | * | |
baac50bb JW |
336 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within |
337 | * @memcg's configured limit, %false if the charge had to be forced. | |
e805605c | 338 | */ |
baac50bb | 339 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) |
e805605c JW |
340 | { |
341 | struct page_counter *counter; | |
342 | ||
baac50bb | 343 | if (page_counter_try_charge(&memcg->tcp_mem.memory_allocated, |
e805605c | 344 | nr_pages, &counter)) { |
baac50bb | 345 | memcg->tcp_mem.memory_pressure = 0; |
e805605c JW |
346 | return true; |
347 | } | |
baac50bb JW |
348 | page_counter_charge(&memcg->tcp_mem.memory_allocated, nr_pages); |
349 | memcg->tcp_mem.memory_pressure = 1; | |
e805605c JW |
350 | return false; |
351 | } | |
352 | ||
353 | /** | |
354 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
baac50bb | 355 | * @memcg - memcg to uncharge |
e805605c JW |
356 | * @nr_pages - number of pages to uncharge |
357 | */ | |
baac50bb | 358 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) |
e805605c | 359 | { |
baac50bb | 360 | page_counter_uncharge(&memcg->tcp_mem.memory_allocated, nr_pages); |
e805605c JW |
361 | } |
362 | ||
3f134619 GC |
363 | #endif |
364 | ||
a8964b9b | 365 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 | 366 | /* |
f7ce3190 | 367 | * This will be the memcg's index in each cache's ->memcg_params.memcg_caches. |
b8627835 LZ |
368 | * The main reason for not using cgroup id for this: |
369 | * this works better in sparse environments, where we have a lot of memcgs, | |
370 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
371 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
372 | * 200 entry array for that. | |
55007d84 | 373 | * |
dbcf73e2 VD |
374 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
375 | * will double each time we have to increase it. | |
55007d84 | 376 | */ |
dbcf73e2 VD |
377 | static DEFINE_IDA(memcg_cache_ida); |
378 | int memcg_nr_cache_ids; | |
749c5415 | 379 | |
05257a1a VD |
380 | /* Protects memcg_nr_cache_ids */ |
381 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
382 | ||
383 | void memcg_get_cache_ids(void) | |
384 | { | |
385 | down_read(&memcg_cache_ids_sem); | |
386 | } | |
387 | ||
388 | void memcg_put_cache_ids(void) | |
389 | { | |
390 | up_read(&memcg_cache_ids_sem); | |
391 | } | |
392 | ||
55007d84 GC |
393 | /* |
394 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
395 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
396 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
397 | * tunable, but that is strictly not necessary. | |
398 | * | |
b8627835 | 399 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
400 | * this constant directly from cgroup, but it is understandable that this is |
401 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 402 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
403 | * increase ours as well if it increases. |
404 | */ | |
405 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 406 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 407 | |
d7f25f8a GC |
408 | /* |
409 | * A lot of the calls to the cache allocation functions are expected to be | |
410 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
411 | * conditional to this static branch, we'll have to allow modules that does | |
412 | * kmem_cache_alloc and the such to see this symbol as well | |
413 | */ | |
a8964b9b | 414 | struct static_key memcg_kmem_enabled_key; |
d7f25f8a | 415 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 416 | |
a8964b9b GC |
417 | #endif /* CONFIG_MEMCG_KMEM */ |
418 | ||
f64c3f54 | 419 | static struct mem_cgroup_per_zone * |
e231875b | 420 | mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone) |
f64c3f54 | 421 | { |
e231875b JZ |
422 | int nid = zone_to_nid(zone); |
423 | int zid = zone_idx(zone); | |
424 | ||
54f72fe0 | 425 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
426 | } |
427 | ||
ad7fa852 TH |
428 | /** |
429 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
430 | * @page: page of interest | |
431 | * | |
432 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
433 | * with @page is returned. The returned css remains associated with @page | |
434 | * until it is released. | |
435 | * | |
436 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
437 | * is returned. | |
438 | * | |
439 | * XXX: The above description of behavior on the default hierarchy isn't | |
440 | * strictly true yet as replace_page_cache_page() can modify the | |
441 | * association before @page is released even on the default hierarchy; | |
442 | * however, the current and planned usages don't mix the the two functions | |
443 | * and replace_page_cache_page() will soon be updated to make the invariant | |
444 | * actually true. | |
445 | */ | |
446 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
447 | { | |
448 | struct mem_cgroup *memcg; | |
449 | ||
450 | rcu_read_lock(); | |
451 | ||
452 | memcg = page->mem_cgroup; | |
453 | ||
9e10a130 | 454 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
455 | memcg = root_mem_cgroup; |
456 | ||
457 | rcu_read_unlock(); | |
458 | return &memcg->css; | |
459 | } | |
460 | ||
2fc04524 VD |
461 | /** |
462 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
463 | * @page: the page | |
464 | * | |
465 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
466 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
467 | * is safe to call this function without holding a reference to @page. | |
468 | * | |
469 | * Note, this function is inherently racy, because there is nothing to prevent | |
470 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
471 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
472 | * do not care (such as procfs interfaces). | |
473 | */ | |
474 | ino_t page_cgroup_ino(struct page *page) | |
475 | { | |
476 | struct mem_cgroup *memcg; | |
477 | unsigned long ino = 0; | |
478 | ||
479 | rcu_read_lock(); | |
480 | memcg = READ_ONCE(page->mem_cgroup); | |
481 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) | |
482 | memcg = parent_mem_cgroup(memcg); | |
483 | if (memcg) | |
484 | ino = cgroup_ino(memcg->css.cgroup); | |
485 | rcu_read_unlock(); | |
486 | return ino; | |
487 | } | |
488 | ||
f64c3f54 | 489 | static struct mem_cgroup_per_zone * |
e231875b | 490 | mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 491 | { |
97a6c37b JW |
492 | int nid = page_to_nid(page); |
493 | int zid = page_zonenum(page); | |
f64c3f54 | 494 | |
e231875b | 495 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
496 | } |
497 | ||
bb4cc1a8 AM |
498 | static struct mem_cgroup_tree_per_zone * |
499 | soft_limit_tree_node_zone(int nid, int zid) | |
500 | { | |
501 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
502 | } | |
503 | ||
504 | static struct mem_cgroup_tree_per_zone * | |
505 | soft_limit_tree_from_page(struct page *page) | |
506 | { | |
507 | int nid = page_to_nid(page); | |
508 | int zid = page_zonenum(page); | |
509 | ||
510 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
511 | } | |
512 | ||
cf2c8127 JW |
513 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz, |
514 | struct mem_cgroup_tree_per_zone *mctz, | |
3e32cb2e | 515 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
516 | { |
517 | struct rb_node **p = &mctz->rb_root.rb_node; | |
518 | struct rb_node *parent = NULL; | |
519 | struct mem_cgroup_per_zone *mz_node; | |
520 | ||
521 | if (mz->on_tree) | |
522 | return; | |
523 | ||
524 | mz->usage_in_excess = new_usage_in_excess; | |
525 | if (!mz->usage_in_excess) | |
526 | return; | |
527 | while (*p) { | |
528 | parent = *p; | |
529 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
530 | tree_node); | |
531 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
532 | p = &(*p)->rb_left; | |
533 | /* | |
534 | * We can't avoid mem cgroups that are over their soft | |
535 | * limit by the same amount | |
536 | */ | |
537 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
538 | p = &(*p)->rb_right; | |
539 | } | |
540 | rb_link_node(&mz->tree_node, parent, p); | |
541 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
542 | mz->on_tree = true; | |
543 | } | |
544 | ||
cf2c8127 JW |
545 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
546 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 AM |
547 | { |
548 | if (!mz->on_tree) | |
549 | return; | |
550 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
551 | mz->on_tree = false; | |
552 | } | |
553 | ||
cf2c8127 JW |
554 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
555 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 | 556 | { |
0a31bc97 JW |
557 | unsigned long flags; |
558 | ||
559 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 560 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 561 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
562 | } |
563 | ||
3e32cb2e JW |
564 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
565 | { | |
566 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 567 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
568 | unsigned long excess = 0; |
569 | ||
570 | if (nr_pages > soft_limit) | |
571 | excess = nr_pages - soft_limit; | |
572 | ||
573 | return excess; | |
574 | } | |
bb4cc1a8 AM |
575 | |
576 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
577 | { | |
3e32cb2e | 578 | unsigned long excess; |
bb4cc1a8 AM |
579 | struct mem_cgroup_per_zone *mz; |
580 | struct mem_cgroup_tree_per_zone *mctz; | |
bb4cc1a8 | 581 | |
e231875b | 582 | mctz = soft_limit_tree_from_page(page); |
bb4cc1a8 AM |
583 | /* |
584 | * Necessary to update all ancestors when hierarchy is used. | |
585 | * because their event counter is not touched. | |
586 | */ | |
587 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
e231875b | 588 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
3e32cb2e | 589 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
590 | /* |
591 | * We have to update the tree if mz is on RB-tree or | |
592 | * mem is over its softlimit. | |
593 | */ | |
594 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
595 | unsigned long flags; |
596 | ||
597 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
598 | /* if on-tree, remove it */ |
599 | if (mz->on_tree) | |
cf2c8127 | 600 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
601 | /* |
602 | * Insert again. mz->usage_in_excess will be updated. | |
603 | * If excess is 0, no tree ops. | |
604 | */ | |
cf2c8127 | 605 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 606 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
607 | } |
608 | } | |
609 | } | |
610 | ||
611 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
612 | { | |
bb4cc1a8 | 613 | struct mem_cgroup_tree_per_zone *mctz; |
e231875b JZ |
614 | struct mem_cgroup_per_zone *mz; |
615 | int nid, zid; | |
bb4cc1a8 | 616 | |
e231875b JZ |
617 | for_each_node(nid) { |
618 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
619 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
620 | mctz = soft_limit_tree_node_zone(nid, zid); | |
cf2c8127 | 621 | mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
622 | } |
623 | } | |
624 | } | |
625 | ||
626 | static struct mem_cgroup_per_zone * | |
627 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
628 | { | |
629 | struct rb_node *rightmost = NULL; | |
630 | struct mem_cgroup_per_zone *mz; | |
631 | ||
632 | retry: | |
633 | mz = NULL; | |
634 | rightmost = rb_last(&mctz->rb_root); | |
635 | if (!rightmost) | |
636 | goto done; /* Nothing to reclaim from */ | |
637 | ||
638 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
639 | /* | |
640 | * Remove the node now but someone else can add it back, | |
641 | * we will to add it back at the end of reclaim to its correct | |
642 | * position in the tree. | |
643 | */ | |
cf2c8127 | 644 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 645 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 646 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
647 | goto retry; |
648 | done: | |
649 | return mz; | |
650 | } | |
651 | ||
652 | static struct mem_cgroup_per_zone * | |
653 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
654 | { | |
655 | struct mem_cgroup_per_zone *mz; | |
656 | ||
0a31bc97 | 657 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 658 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 659 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
660 | return mz; |
661 | } | |
662 | ||
711d3d2c | 663 | /* |
484ebb3b GT |
664 | * Return page count for single (non recursive) @memcg. |
665 | * | |
711d3d2c KH |
666 | * Implementation Note: reading percpu statistics for memcg. |
667 | * | |
668 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
669 | * synchronization to implement "quick" read. There are trade-off between | |
670 | * reading cost and precision of value. Then, we may have a chance to implement | |
484ebb3b | 671 | * a periodic synchronization of counter in memcg's counter. |
711d3d2c KH |
672 | * |
673 | * But this _read() function is used for user interface now. The user accounts | |
674 | * memory usage by memory cgroup and he _always_ requires exact value because | |
675 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
676 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
677 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
678 | * | |
679 | * If there are kernel internal actions which can make use of some not-exact | |
680 | * value, and reading all cpu value can be performance bottleneck in some | |
484ebb3b | 681 | * common workload, threshold and synchronization as vmstat[] should be |
711d3d2c KH |
682 | * implemented. |
683 | */ | |
484ebb3b GT |
684 | static unsigned long |
685 | mem_cgroup_read_stat(struct mem_cgroup *memcg, enum mem_cgroup_stat_index idx) | |
c62b1a3b | 686 | { |
7a159cc9 | 687 | long val = 0; |
c62b1a3b | 688 | int cpu; |
c62b1a3b | 689 | |
484ebb3b | 690 | /* Per-cpu values can be negative, use a signed accumulator */ |
733a572e | 691 | for_each_possible_cpu(cpu) |
c0ff4b85 | 692 | val += per_cpu(memcg->stat->count[idx], cpu); |
484ebb3b GT |
693 | /* |
694 | * Summing races with updates, so val may be negative. Avoid exposing | |
695 | * transient negative values. | |
696 | */ | |
697 | if (val < 0) | |
698 | val = 0; | |
c62b1a3b KH |
699 | return val; |
700 | } | |
701 | ||
c0ff4b85 | 702 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
703 | enum mem_cgroup_events_index idx) |
704 | { | |
705 | unsigned long val = 0; | |
706 | int cpu; | |
707 | ||
733a572e | 708 | for_each_possible_cpu(cpu) |
c0ff4b85 | 709 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f JW |
710 | return val; |
711 | } | |
712 | ||
c0ff4b85 | 713 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 714 | struct page *page, |
0a31bc97 | 715 | int nr_pages) |
d52aa412 | 716 | { |
b2402857 KH |
717 | /* |
718 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
719 | * counted as CACHE even if it's on ANON LRU. | |
720 | */ | |
0a31bc97 | 721 | if (PageAnon(page)) |
b2402857 | 722 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
c0ff4b85 | 723 | nr_pages); |
d52aa412 | 724 | else |
b2402857 | 725 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 726 | nr_pages); |
55e462b0 | 727 | |
b070e65c DR |
728 | if (PageTransHuge(page)) |
729 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], | |
730 | nr_pages); | |
731 | ||
e401f176 KH |
732 | /* pagein of a big page is an event. So, ignore page size */ |
733 | if (nr_pages > 0) | |
c0ff4b85 | 734 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 735 | else { |
c0ff4b85 | 736 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
737 | nr_pages = -nr_pages; /* for event */ |
738 | } | |
e401f176 | 739 | |
13114716 | 740 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
6d12e2d8 KH |
741 | } |
742 | ||
e231875b JZ |
743 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
744 | int nid, | |
745 | unsigned int lru_mask) | |
bb2a0de9 | 746 | { |
e231875b | 747 | unsigned long nr = 0; |
889976db YH |
748 | int zid; |
749 | ||
e231875b | 750 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 751 | |
e231875b JZ |
752 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
753 | struct mem_cgroup_per_zone *mz; | |
754 | enum lru_list lru; | |
755 | ||
756 | for_each_lru(lru) { | |
757 | if (!(BIT(lru) & lru_mask)) | |
758 | continue; | |
759 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
760 | nr += mz->lru_size[lru]; | |
761 | } | |
762 | } | |
763 | return nr; | |
889976db | 764 | } |
bb2a0de9 | 765 | |
c0ff4b85 | 766 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 767 | unsigned int lru_mask) |
6d12e2d8 | 768 | { |
e231875b | 769 | unsigned long nr = 0; |
889976db | 770 | int nid; |
6d12e2d8 | 771 | |
31aaea4a | 772 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
773 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
774 | return nr; | |
d52aa412 KH |
775 | } |
776 | ||
f53d7ce3 JW |
777 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
778 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
779 | { |
780 | unsigned long val, next; | |
781 | ||
13114716 | 782 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 783 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 784 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
785 | if ((long)next - (long)val < 0) { |
786 | switch (target) { | |
787 | case MEM_CGROUP_TARGET_THRESH: | |
788 | next = val + THRESHOLDS_EVENTS_TARGET; | |
789 | break; | |
bb4cc1a8 AM |
790 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
791 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
792 | break; | |
f53d7ce3 JW |
793 | case MEM_CGROUP_TARGET_NUMAINFO: |
794 | next = val + NUMAINFO_EVENTS_TARGET; | |
795 | break; | |
796 | default: | |
797 | break; | |
798 | } | |
799 | __this_cpu_write(memcg->stat->targets[target], next); | |
800 | return true; | |
7a159cc9 | 801 | } |
f53d7ce3 | 802 | return false; |
d2265e6f KH |
803 | } |
804 | ||
805 | /* | |
806 | * Check events in order. | |
807 | * | |
808 | */ | |
c0ff4b85 | 809 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
810 | { |
811 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
812 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
813 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 814 | bool do_softlimit; |
82b3f2a7 | 815 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 816 | |
bb4cc1a8 AM |
817 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
818 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
819 | #if MAX_NUMNODES > 1 |
820 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
821 | MEM_CGROUP_TARGET_NUMAINFO); | |
822 | #endif | |
c0ff4b85 | 823 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
824 | if (unlikely(do_softlimit)) |
825 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 826 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 827 | if (unlikely(do_numainfo)) |
c0ff4b85 | 828 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 829 | #endif |
0a31bc97 | 830 | } |
d2265e6f KH |
831 | } |
832 | ||
cf475ad2 | 833 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 834 | { |
31a78f23 BS |
835 | /* |
836 | * mm_update_next_owner() may clear mm->owner to NULL | |
837 | * if it races with swapoff, page migration, etc. | |
838 | * So this can be called with p == NULL. | |
839 | */ | |
840 | if (unlikely(!p)) | |
841 | return NULL; | |
842 | ||
073219e9 | 843 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 844 | } |
33398cf2 | 845 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 846 | |
df381975 | 847 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 848 | { |
c0ff4b85 | 849 | struct mem_cgroup *memcg = NULL; |
0b7f569e | 850 | |
54595fe2 KH |
851 | rcu_read_lock(); |
852 | do { | |
6f6acb00 MH |
853 | /* |
854 | * Page cache insertions can happen withou an | |
855 | * actual mm context, e.g. during disk probing | |
856 | * on boot, loopback IO, acct() writes etc. | |
857 | */ | |
858 | if (unlikely(!mm)) | |
df381975 | 859 | memcg = root_mem_cgroup; |
6f6acb00 MH |
860 | else { |
861 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
862 | if (unlikely(!memcg)) | |
863 | memcg = root_mem_cgroup; | |
864 | } | |
ec903c0c | 865 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 866 | rcu_read_unlock(); |
c0ff4b85 | 867 | return memcg; |
54595fe2 KH |
868 | } |
869 | ||
5660048c JW |
870 | /** |
871 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
872 | * @root: hierarchy root | |
873 | * @prev: previously returned memcg, NULL on first invocation | |
874 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
875 | * | |
876 | * Returns references to children of the hierarchy below @root, or | |
877 | * @root itself, or %NULL after a full round-trip. | |
878 | * | |
879 | * Caller must pass the return value in @prev on subsequent | |
880 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
881 | * to cancel a hierarchy walk before the round-trip is complete. | |
882 | * | |
883 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
884 | * divide up the memcgs in the hierarchy among all concurrent | |
885 | * reclaimers operating on the same zone and priority. | |
886 | */ | |
694fbc0f | 887 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 888 | struct mem_cgroup *prev, |
694fbc0f | 889 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 890 | { |
33398cf2 | 891 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
5ac8fb31 | 892 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 893 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 894 | struct mem_cgroup *pos = NULL; |
711d3d2c | 895 | |
694fbc0f AM |
896 | if (mem_cgroup_disabled()) |
897 | return NULL; | |
5660048c | 898 | |
9f3a0d09 JW |
899 | if (!root) |
900 | root = root_mem_cgroup; | |
7d74b06f | 901 | |
9f3a0d09 | 902 | if (prev && !reclaim) |
5ac8fb31 | 903 | pos = prev; |
14067bb3 | 904 | |
9f3a0d09 JW |
905 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
906 | if (prev) | |
5ac8fb31 | 907 | goto out; |
694fbc0f | 908 | return root; |
9f3a0d09 | 909 | } |
14067bb3 | 910 | |
542f85f9 | 911 | rcu_read_lock(); |
5f578161 | 912 | |
5ac8fb31 JW |
913 | if (reclaim) { |
914 | struct mem_cgroup_per_zone *mz; | |
915 | ||
916 | mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone); | |
917 | iter = &mz->iter[reclaim->priority]; | |
918 | ||
919 | if (prev && reclaim->generation != iter->generation) | |
920 | goto out_unlock; | |
921 | ||
6df38689 | 922 | while (1) { |
4db0c3c2 | 923 | pos = READ_ONCE(iter->position); |
6df38689 VD |
924 | if (!pos || css_tryget(&pos->css)) |
925 | break; | |
5ac8fb31 | 926 | /* |
6df38689 VD |
927 | * css reference reached zero, so iter->position will |
928 | * be cleared by ->css_released. However, we should not | |
929 | * rely on this happening soon, because ->css_released | |
930 | * is called from a work queue, and by busy-waiting we | |
931 | * might block it. So we clear iter->position right | |
932 | * away. | |
5ac8fb31 | 933 | */ |
6df38689 VD |
934 | (void)cmpxchg(&iter->position, pos, NULL); |
935 | } | |
5ac8fb31 JW |
936 | } |
937 | ||
938 | if (pos) | |
939 | css = &pos->css; | |
940 | ||
941 | for (;;) { | |
942 | css = css_next_descendant_pre(css, &root->css); | |
943 | if (!css) { | |
944 | /* | |
945 | * Reclaimers share the hierarchy walk, and a | |
946 | * new one might jump in right at the end of | |
947 | * the hierarchy - make sure they see at least | |
948 | * one group and restart from the beginning. | |
949 | */ | |
950 | if (!prev) | |
951 | continue; | |
952 | break; | |
527a5ec9 | 953 | } |
7d74b06f | 954 | |
5ac8fb31 JW |
955 | /* |
956 | * Verify the css and acquire a reference. The root | |
957 | * is provided by the caller, so we know it's alive | |
958 | * and kicking, and don't take an extra reference. | |
959 | */ | |
960 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 961 | |
5ac8fb31 JW |
962 | if (css == &root->css) |
963 | break; | |
14067bb3 | 964 | |
b2052564 | 965 | if (css_tryget(css)) { |
5ac8fb31 JW |
966 | /* |
967 | * Make sure the memcg is initialized: | |
968 | * mem_cgroup_css_online() orders the the | |
969 | * initialization against setting the flag. | |
970 | */ | |
971 | if (smp_load_acquire(&memcg->initialized)) | |
972 | break; | |
542f85f9 | 973 | |
5ac8fb31 | 974 | css_put(css); |
527a5ec9 | 975 | } |
9f3a0d09 | 976 | |
5ac8fb31 | 977 | memcg = NULL; |
9f3a0d09 | 978 | } |
5ac8fb31 JW |
979 | |
980 | if (reclaim) { | |
5ac8fb31 | 981 | /* |
6df38689 VD |
982 | * The position could have already been updated by a competing |
983 | * thread, so check that the value hasn't changed since we read | |
984 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 985 | */ |
6df38689 VD |
986 | (void)cmpxchg(&iter->position, pos, memcg); |
987 | ||
5ac8fb31 JW |
988 | if (pos) |
989 | css_put(&pos->css); | |
990 | ||
991 | if (!memcg) | |
992 | iter->generation++; | |
993 | else if (!prev) | |
994 | reclaim->generation = iter->generation; | |
9f3a0d09 | 995 | } |
5ac8fb31 | 996 | |
542f85f9 MH |
997 | out_unlock: |
998 | rcu_read_unlock(); | |
5ac8fb31 | 999 | out: |
c40046f3 MH |
1000 | if (prev && prev != root) |
1001 | css_put(&prev->css); | |
1002 | ||
9f3a0d09 | 1003 | return memcg; |
14067bb3 | 1004 | } |
7d74b06f | 1005 | |
5660048c JW |
1006 | /** |
1007 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1008 | * @root: hierarchy root | |
1009 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1010 | */ | |
1011 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1012 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1013 | { |
1014 | if (!root) | |
1015 | root = root_mem_cgroup; | |
1016 | if (prev && prev != root) | |
1017 | css_put(&prev->css); | |
1018 | } | |
7d74b06f | 1019 | |
6df38689 VD |
1020 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1021 | { | |
1022 | struct mem_cgroup *memcg = dead_memcg; | |
1023 | struct mem_cgroup_reclaim_iter *iter; | |
1024 | struct mem_cgroup_per_zone *mz; | |
1025 | int nid, zid; | |
1026 | int i; | |
1027 | ||
1028 | while ((memcg = parent_mem_cgroup(memcg))) { | |
1029 | for_each_node(nid) { | |
1030 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1031 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
1032 | for (i = 0; i <= DEF_PRIORITY; i++) { | |
1033 | iter = &mz->iter[i]; | |
1034 | cmpxchg(&iter->position, | |
1035 | dead_memcg, NULL); | |
1036 | } | |
1037 | } | |
1038 | } | |
1039 | } | |
1040 | } | |
1041 | ||
9f3a0d09 JW |
1042 | /* |
1043 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1044 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1045 | * be used for reference counting. | |
1046 | */ | |
1047 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1048 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1049 | iter != NULL; \ |
527a5ec9 | 1050 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1051 | |
9f3a0d09 | 1052 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1053 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1054 | iter != NULL; \ |
527a5ec9 | 1055 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1056 | |
925b7673 JW |
1057 | /** |
1058 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1059 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1060 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1061 | * |
1062 | * Returns the lru list vector holding pages for the given @zone and | |
1063 | * @mem. This can be the global zone lruvec, if the memory controller | |
1064 | * is disabled. | |
1065 | */ | |
1066 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1067 | struct mem_cgroup *memcg) | |
1068 | { | |
1069 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1070 | struct lruvec *lruvec; |
925b7673 | 1071 | |
bea8c150 HD |
1072 | if (mem_cgroup_disabled()) { |
1073 | lruvec = &zone->lruvec; | |
1074 | goto out; | |
1075 | } | |
925b7673 | 1076 | |
e231875b | 1077 | mz = mem_cgroup_zone_zoneinfo(memcg, zone); |
bea8c150 HD |
1078 | lruvec = &mz->lruvec; |
1079 | out: | |
1080 | /* | |
1081 | * Since a node can be onlined after the mem_cgroup was created, | |
1082 | * we have to be prepared to initialize lruvec->zone here; | |
1083 | * and if offlined then reonlined, we need to reinitialize it. | |
1084 | */ | |
1085 | if (unlikely(lruvec->zone != zone)) | |
1086 | lruvec->zone = zone; | |
1087 | return lruvec; | |
925b7673 JW |
1088 | } |
1089 | ||
925b7673 | 1090 | /** |
dfe0e773 | 1091 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1092 | * @page: the page |
fa9add64 | 1093 | * @zone: zone of the page |
dfe0e773 JW |
1094 | * |
1095 | * This function is only safe when following the LRU page isolation | |
1096 | * and putback protocol: the LRU lock must be held, and the page must | |
1097 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 1098 | */ |
fa9add64 | 1099 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1100 | { |
08e552c6 | 1101 | struct mem_cgroup_per_zone *mz; |
925b7673 | 1102 | struct mem_cgroup *memcg; |
bea8c150 | 1103 | struct lruvec *lruvec; |
6d12e2d8 | 1104 | |
bea8c150 HD |
1105 | if (mem_cgroup_disabled()) { |
1106 | lruvec = &zone->lruvec; | |
1107 | goto out; | |
1108 | } | |
925b7673 | 1109 | |
1306a85a | 1110 | memcg = page->mem_cgroup; |
7512102c | 1111 | /* |
dfe0e773 | 1112 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1113 | * possibly migrated - before they are charged. |
7512102c | 1114 | */ |
29833315 JW |
1115 | if (!memcg) |
1116 | memcg = root_mem_cgroup; | |
7512102c | 1117 | |
e231875b | 1118 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
bea8c150 HD |
1119 | lruvec = &mz->lruvec; |
1120 | out: | |
1121 | /* | |
1122 | * Since a node can be onlined after the mem_cgroup was created, | |
1123 | * we have to be prepared to initialize lruvec->zone here; | |
1124 | * and if offlined then reonlined, we need to reinitialize it. | |
1125 | */ | |
1126 | if (unlikely(lruvec->zone != zone)) | |
1127 | lruvec->zone = zone; | |
1128 | return lruvec; | |
08e552c6 | 1129 | } |
b69408e8 | 1130 | |
925b7673 | 1131 | /** |
fa9add64 HD |
1132 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1133 | * @lruvec: mem_cgroup per zone lru vector | |
1134 | * @lru: index of lru list the page is sitting on | |
1135 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1136 | * |
fa9add64 HD |
1137 | * This function must be called when a page is added to or removed from an |
1138 | * lru list. | |
3f58a829 | 1139 | */ |
fa9add64 HD |
1140 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1141 | int nr_pages) | |
3f58a829 MK |
1142 | { |
1143 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1144 | unsigned long *lru_size; |
3f58a829 MK |
1145 | |
1146 | if (mem_cgroup_disabled()) | |
1147 | return; | |
1148 | ||
fa9add64 HD |
1149 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1150 | lru_size = mz->lru_size + lru; | |
1151 | *lru_size += nr_pages; | |
1152 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1153 | } |
544122e5 | 1154 | |
2314b42d | 1155 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
c3ac9a8a | 1156 | { |
2314b42d | 1157 | struct mem_cgroup *task_memcg; |
158e0a2d | 1158 | struct task_struct *p; |
ffbdccf5 | 1159 | bool ret; |
4c4a2214 | 1160 | |
158e0a2d | 1161 | p = find_lock_task_mm(task); |
de077d22 | 1162 | if (p) { |
2314b42d | 1163 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1164 | task_unlock(p); |
1165 | } else { | |
1166 | /* | |
1167 | * All threads may have already detached their mm's, but the oom | |
1168 | * killer still needs to detect if they have already been oom | |
1169 | * killed to prevent needlessly killing additional tasks. | |
1170 | */ | |
ffbdccf5 | 1171 | rcu_read_lock(); |
2314b42d JW |
1172 | task_memcg = mem_cgroup_from_task(task); |
1173 | css_get(&task_memcg->css); | |
ffbdccf5 | 1174 | rcu_read_unlock(); |
de077d22 | 1175 | } |
2314b42d JW |
1176 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1177 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1178 | return ret; |
1179 | } | |
1180 | ||
3e32cb2e | 1181 | #define mem_cgroup_from_counter(counter, member) \ |
6d61ef40 BS |
1182 | container_of(counter, struct mem_cgroup, member) |
1183 | ||
19942822 | 1184 | /** |
9d11ea9f | 1185 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1186 | * @memcg: the memory cgroup |
19942822 | 1187 | * |
9d11ea9f | 1188 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1189 | * pages. |
19942822 | 1190 | */ |
c0ff4b85 | 1191 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1192 | { |
3e32cb2e JW |
1193 | unsigned long margin = 0; |
1194 | unsigned long count; | |
1195 | unsigned long limit; | |
9d11ea9f | 1196 | |
3e32cb2e | 1197 | count = page_counter_read(&memcg->memory); |
4db0c3c2 | 1198 | limit = READ_ONCE(memcg->memory.limit); |
3e32cb2e JW |
1199 | if (count < limit) |
1200 | margin = limit - count; | |
1201 | ||
1202 | if (do_swap_account) { | |
1203 | count = page_counter_read(&memcg->memsw); | |
4db0c3c2 | 1204 | limit = READ_ONCE(memcg->memsw.limit); |
3e32cb2e JW |
1205 | if (count <= limit) |
1206 | margin = min(margin, limit - count); | |
1207 | } | |
1208 | ||
1209 | return margin; | |
19942822 JW |
1210 | } |
1211 | ||
32047e2a | 1212 | /* |
bdcbb659 | 1213 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1214 | * |
bdcbb659 QH |
1215 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1216 | * moving cgroups. This is for waiting at high-memory pressure | |
1217 | * caused by "move". | |
32047e2a | 1218 | */ |
c0ff4b85 | 1219 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1220 | { |
2bd9bb20 KH |
1221 | struct mem_cgroup *from; |
1222 | struct mem_cgroup *to; | |
4b534334 | 1223 | bool ret = false; |
2bd9bb20 KH |
1224 | /* |
1225 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1226 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1227 | */ | |
1228 | spin_lock(&mc.lock); | |
1229 | from = mc.from; | |
1230 | to = mc.to; | |
1231 | if (!from) | |
1232 | goto unlock; | |
3e92041d | 1233 | |
2314b42d JW |
1234 | ret = mem_cgroup_is_descendant(from, memcg) || |
1235 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1236 | unlock: |
1237 | spin_unlock(&mc.lock); | |
4b534334 KH |
1238 | return ret; |
1239 | } | |
1240 | ||
c0ff4b85 | 1241 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1242 | { |
1243 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1244 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1245 | DEFINE_WAIT(wait); |
1246 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1247 | /* moving charge context might have finished. */ | |
1248 | if (mc.moving_task) | |
1249 | schedule(); | |
1250 | finish_wait(&mc.waitq, &wait); | |
1251 | return true; | |
1252 | } | |
1253 | } | |
1254 | return false; | |
1255 | } | |
1256 | ||
58cf188e | 1257 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1258 | /** |
58cf188e | 1259 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1260 | * @memcg: The memory cgroup that went over limit |
1261 | * @p: Task that is going to be killed | |
1262 | * | |
1263 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1264 | * enabled | |
1265 | */ | |
1266 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1267 | { | |
e61734c5 | 1268 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
08088cb9 | 1269 | static DEFINE_MUTEX(oom_info_lock); |
58cf188e SZ |
1270 | struct mem_cgroup *iter; |
1271 | unsigned int i; | |
e222432b | 1272 | |
08088cb9 | 1273 | mutex_lock(&oom_info_lock); |
e222432b BS |
1274 | rcu_read_lock(); |
1275 | ||
2415b9f5 BV |
1276 | if (p) { |
1277 | pr_info("Task in "); | |
1278 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1279 | pr_cont(" killed as a result of limit of "); | |
1280 | } else { | |
1281 | pr_info("Memory limit reached of cgroup "); | |
1282 | } | |
1283 | ||
e61734c5 | 1284 | pr_cont_cgroup_path(memcg->css.cgroup); |
0346dadb | 1285 | pr_cont("\n"); |
e222432b | 1286 | |
e222432b BS |
1287 | rcu_read_unlock(); |
1288 | ||
3e32cb2e JW |
1289 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1290 | K((u64)page_counter_read(&memcg->memory)), | |
1291 | K((u64)memcg->memory.limit), memcg->memory.failcnt); | |
1292 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1293 | K((u64)page_counter_read(&memcg->memsw)), | |
1294 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); | |
1295 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1296 | K((u64)page_counter_read(&memcg->kmem)), | |
1297 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); | |
58cf188e SZ |
1298 | |
1299 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1300 | pr_info("Memory cgroup stats for "); |
1301 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1302 | pr_cont(":"); |
1303 | ||
1304 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1305 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1306 | continue; | |
484ebb3b | 1307 | pr_cont(" %s:%luKB", mem_cgroup_stat_names[i], |
58cf188e SZ |
1308 | K(mem_cgroup_read_stat(iter, i))); |
1309 | } | |
1310 | ||
1311 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1312 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1313 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1314 | ||
1315 | pr_cont("\n"); | |
1316 | } | |
08088cb9 | 1317 | mutex_unlock(&oom_info_lock); |
e222432b BS |
1318 | } |
1319 | ||
81d39c20 KH |
1320 | /* |
1321 | * This function returns the number of memcg under hierarchy tree. Returns | |
1322 | * 1(self count) if no children. | |
1323 | */ | |
c0ff4b85 | 1324 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1325 | { |
1326 | int num = 0; | |
7d74b06f KH |
1327 | struct mem_cgroup *iter; |
1328 | ||
c0ff4b85 | 1329 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1330 | num++; |
81d39c20 KH |
1331 | return num; |
1332 | } | |
1333 | ||
a63d83f4 DR |
1334 | /* |
1335 | * Return the memory (and swap, if configured) limit for a memcg. | |
1336 | */ | |
3e32cb2e | 1337 | static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 | 1338 | { |
3e32cb2e | 1339 | unsigned long limit; |
f3e8eb70 | 1340 | |
3e32cb2e | 1341 | limit = memcg->memory.limit; |
9a5a8f19 | 1342 | if (mem_cgroup_swappiness(memcg)) { |
3e32cb2e | 1343 | unsigned long memsw_limit; |
9a5a8f19 | 1344 | |
3e32cb2e JW |
1345 | memsw_limit = memcg->memsw.limit; |
1346 | limit = min(limit + total_swap_pages, memsw_limit); | |
9a5a8f19 | 1347 | } |
9a5a8f19 | 1348 | return limit; |
a63d83f4 DR |
1349 | } |
1350 | ||
19965460 DR |
1351 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1352 | int order) | |
9cbb78bb | 1353 | { |
6e0fc46d DR |
1354 | struct oom_control oc = { |
1355 | .zonelist = NULL, | |
1356 | .nodemask = NULL, | |
1357 | .gfp_mask = gfp_mask, | |
1358 | .order = order, | |
6e0fc46d | 1359 | }; |
9cbb78bb DR |
1360 | struct mem_cgroup *iter; |
1361 | unsigned long chosen_points = 0; | |
1362 | unsigned long totalpages; | |
1363 | unsigned int points = 0; | |
1364 | struct task_struct *chosen = NULL; | |
1365 | ||
dc56401f JW |
1366 | mutex_lock(&oom_lock); |
1367 | ||
876aafbf | 1368 | /* |
465adcf1 DR |
1369 | * If current has a pending SIGKILL or is exiting, then automatically |
1370 | * select it. The goal is to allow it to allocate so that it may | |
1371 | * quickly exit and free its memory. | |
876aafbf | 1372 | */ |
d003f371 | 1373 | if (fatal_signal_pending(current) || task_will_free_mem(current)) { |
16e95196 | 1374 | mark_oom_victim(current); |
dc56401f | 1375 | goto unlock; |
876aafbf DR |
1376 | } |
1377 | ||
6e0fc46d | 1378 | check_panic_on_oom(&oc, CONSTRAINT_MEMCG, memcg); |
3e32cb2e | 1379 | totalpages = mem_cgroup_get_limit(memcg) ? : 1; |
9cbb78bb | 1380 | for_each_mem_cgroup_tree(iter, memcg) { |
72ec7029 | 1381 | struct css_task_iter it; |
9cbb78bb DR |
1382 | struct task_struct *task; |
1383 | ||
72ec7029 TH |
1384 | css_task_iter_start(&iter->css, &it); |
1385 | while ((task = css_task_iter_next(&it))) { | |
6e0fc46d | 1386 | switch (oom_scan_process_thread(&oc, task, totalpages)) { |
9cbb78bb DR |
1387 | case OOM_SCAN_SELECT: |
1388 | if (chosen) | |
1389 | put_task_struct(chosen); | |
1390 | chosen = task; | |
1391 | chosen_points = ULONG_MAX; | |
1392 | get_task_struct(chosen); | |
1393 | /* fall through */ | |
1394 | case OOM_SCAN_CONTINUE: | |
1395 | continue; | |
1396 | case OOM_SCAN_ABORT: | |
72ec7029 | 1397 | css_task_iter_end(&it); |
9cbb78bb DR |
1398 | mem_cgroup_iter_break(memcg, iter); |
1399 | if (chosen) | |
1400 | put_task_struct(chosen); | |
dc56401f | 1401 | goto unlock; |
9cbb78bb DR |
1402 | case OOM_SCAN_OK: |
1403 | break; | |
1404 | }; | |
1405 | points = oom_badness(task, memcg, NULL, totalpages); | |
d49ad935 DR |
1406 | if (!points || points < chosen_points) |
1407 | continue; | |
1408 | /* Prefer thread group leaders for display purposes */ | |
1409 | if (points == chosen_points && | |
1410 | thread_group_leader(chosen)) | |
1411 | continue; | |
1412 | ||
1413 | if (chosen) | |
1414 | put_task_struct(chosen); | |
1415 | chosen = task; | |
1416 | chosen_points = points; | |
1417 | get_task_struct(chosen); | |
9cbb78bb | 1418 | } |
72ec7029 | 1419 | css_task_iter_end(&it); |
9cbb78bb DR |
1420 | } |
1421 | ||
dc56401f JW |
1422 | if (chosen) { |
1423 | points = chosen_points * 1000 / totalpages; | |
6e0fc46d DR |
1424 | oom_kill_process(&oc, chosen, points, totalpages, memcg, |
1425 | "Memory cgroup out of memory"); | |
dc56401f JW |
1426 | } |
1427 | unlock: | |
1428 | mutex_unlock(&oom_lock); | |
9cbb78bb DR |
1429 | } |
1430 | ||
ae6e71d3 MC |
1431 | #if MAX_NUMNODES > 1 |
1432 | ||
4d0c066d KH |
1433 | /** |
1434 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1435 | * @memcg: the target memcg |
4d0c066d KH |
1436 | * @nid: the node ID to be checked. |
1437 | * @noswap : specify true here if the user wants flle only information. | |
1438 | * | |
1439 | * This function returns whether the specified memcg contains any | |
1440 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1441 | * pages in the node. | |
1442 | */ | |
c0ff4b85 | 1443 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1444 | int nid, bool noswap) |
1445 | { | |
c0ff4b85 | 1446 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1447 | return true; |
1448 | if (noswap || !total_swap_pages) | |
1449 | return false; | |
c0ff4b85 | 1450 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1451 | return true; |
1452 | return false; | |
1453 | ||
1454 | } | |
889976db YH |
1455 | |
1456 | /* | |
1457 | * Always updating the nodemask is not very good - even if we have an empty | |
1458 | * list or the wrong list here, we can start from some node and traverse all | |
1459 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1460 | * | |
1461 | */ | |
c0ff4b85 | 1462 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1463 | { |
1464 | int nid; | |
453a9bf3 KH |
1465 | /* |
1466 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1467 | * pagein/pageout changes since the last update. | |
1468 | */ | |
c0ff4b85 | 1469 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1470 | return; |
c0ff4b85 | 1471 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1472 | return; |
1473 | ||
889976db | 1474 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1475 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1476 | |
31aaea4a | 1477 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1478 | |
c0ff4b85 R |
1479 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1480 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1481 | } |
453a9bf3 | 1482 | |
c0ff4b85 R |
1483 | atomic_set(&memcg->numainfo_events, 0); |
1484 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1485 | } |
1486 | ||
1487 | /* | |
1488 | * Selecting a node where we start reclaim from. Because what we need is just | |
1489 | * reducing usage counter, start from anywhere is O,K. Considering | |
1490 | * memory reclaim from current node, there are pros. and cons. | |
1491 | * | |
1492 | * Freeing memory from current node means freeing memory from a node which | |
1493 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1494 | * hit limits, it will see a contention on a node. But freeing from remote | |
1495 | * node means more costs for memory reclaim because of memory latency. | |
1496 | * | |
1497 | * Now, we use round-robin. Better algorithm is welcomed. | |
1498 | */ | |
c0ff4b85 | 1499 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1500 | { |
1501 | int node; | |
1502 | ||
c0ff4b85 R |
1503 | mem_cgroup_may_update_nodemask(memcg); |
1504 | node = memcg->last_scanned_node; | |
889976db | 1505 | |
c0ff4b85 | 1506 | node = next_node(node, memcg->scan_nodes); |
889976db | 1507 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1508 | node = first_node(memcg->scan_nodes); |
889976db YH |
1509 | /* |
1510 | * We call this when we hit limit, not when pages are added to LRU. | |
1511 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1512 | * memcg is too small and all pages are not on LRU. In that case, | |
1513 | * we use curret node. | |
1514 | */ | |
1515 | if (unlikely(node == MAX_NUMNODES)) | |
1516 | node = numa_node_id(); | |
1517 | ||
c0ff4b85 | 1518 | memcg->last_scanned_node = node; |
889976db YH |
1519 | return node; |
1520 | } | |
889976db | 1521 | #else |
c0ff4b85 | 1522 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1523 | { |
1524 | return 0; | |
1525 | } | |
1526 | #endif | |
1527 | ||
0608f43d AM |
1528 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1529 | struct zone *zone, | |
1530 | gfp_t gfp_mask, | |
1531 | unsigned long *total_scanned) | |
1532 | { | |
1533 | struct mem_cgroup *victim = NULL; | |
1534 | int total = 0; | |
1535 | int loop = 0; | |
1536 | unsigned long excess; | |
1537 | unsigned long nr_scanned; | |
1538 | struct mem_cgroup_reclaim_cookie reclaim = { | |
1539 | .zone = zone, | |
1540 | .priority = 0, | |
1541 | }; | |
1542 | ||
3e32cb2e | 1543 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1544 | |
1545 | while (1) { | |
1546 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1547 | if (!victim) { | |
1548 | loop++; | |
1549 | if (loop >= 2) { | |
1550 | /* | |
1551 | * If we have not been able to reclaim | |
1552 | * anything, it might because there are | |
1553 | * no reclaimable pages under this hierarchy | |
1554 | */ | |
1555 | if (!total) | |
1556 | break; | |
1557 | /* | |
1558 | * We want to do more targeted reclaim. | |
1559 | * excess >> 2 is not to excessive so as to | |
1560 | * reclaim too much, nor too less that we keep | |
1561 | * coming back to reclaim from this cgroup | |
1562 | */ | |
1563 | if (total >= (excess >> 2) || | |
1564 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1565 | break; | |
1566 | } | |
1567 | continue; | |
1568 | } | |
0608f43d AM |
1569 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1570 | zone, &nr_scanned); | |
1571 | *total_scanned += nr_scanned; | |
3e32cb2e | 1572 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1573 | break; |
6d61ef40 | 1574 | } |
0608f43d AM |
1575 | mem_cgroup_iter_break(root_memcg, victim); |
1576 | return total; | |
6d61ef40 BS |
1577 | } |
1578 | ||
0056f4e6 JW |
1579 | #ifdef CONFIG_LOCKDEP |
1580 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1581 | .name = "memcg_oom_lock", | |
1582 | }; | |
1583 | #endif | |
1584 | ||
fb2a6fc5 JW |
1585 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1586 | ||
867578cb KH |
1587 | /* |
1588 | * Check OOM-Killer is already running under our hierarchy. | |
1589 | * If someone is running, return false. | |
1590 | */ | |
fb2a6fc5 | 1591 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1592 | { |
79dfdacc | 1593 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1594 | |
fb2a6fc5 JW |
1595 | spin_lock(&memcg_oom_lock); |
1596 | ||
9f3a0d09 | 1597 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1598 | if (iter->oom_lock) { |
79dfdacc MH |
1599 | /* |
1600 | * this subtree of our hierarchy is already locked | |
1601 | * so we cannot give a lock. | |
1602 | */ | |
79dfdacc | 1603 | failed = iter; |
9f3a0d09 JW |
1604 | mem_cgroup_iter_break(memcg, iter); |
1605 | break; | |
23751be0 JW |
1606 | } else |
1607 | iter->oom_lock = true; | |
7d74b06f | 1608 | } |
867578cb | 1609 | |
fb2a6fc5 JW |
1610 | if (failed) { |
1611 | /* | |
1612 | * OK, we failed to lock the whole subtree so we have | |
1613 | * to clean up what we set up to the failing subtree | |
1614 | */ | |
1615 | for_each_mem_cgroup_tree(iter, memcg) { | |
1616 | if (iter == failed) { | |
1617 | mem_cgroup_iter_break(memcg, iter); | |
1618 | break; | |
1619 | } | |
1620 | iter->oom_lock = false; | |
79dfdacc | 1621 | } |
0056f4e6 JW |
1622 | } else |
1623 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1624 | |
1625 | spin_unlock(&memcg_oom_lock); | |
1626 | ||
1627 | return !failed; | |
a636b327 | 1628 | } |
0b7f569e | 1629 | |
fb2a6fc5 | 1630 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1631 | { |
7d74b06f KH |
1632 | struct mem_cgroup *iter; |
1633 | ||
fb2a6fc5 | 1634 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1635 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1636 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1637 | iter->oom_lock = false; |
fb2a6fc5 | 1638 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1639 | } |
1640 | ||
c0ff4b85 | 1641 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1642 | { |
1643 | struct mem_cgroup *iter; | |
1644 | ||
c2b42d3c | 1645 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1646 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1647 | iter->under_oom++; |
1648 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1649 | } |
1650 | ||
c0ff4b85 | 1651 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1652 | { |
1653 | struct mem_cgroup *iter; | |
1654 | ||
867578cb KH |
1655 | /* |
1656 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1657 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1658 | */ |
c2b42d3c | 1659 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1660 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1661 | if (iter->under_oom > 0) |
1662 | iter->under_oom--; | |
1663 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1664 | } |
1665 | ||
867578cb KH |
1666 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1667 | ||
dc98df5a | 1668 | struct oom_wait_info { |
d79154bb | 1669 | struct mem_cgroup *memcg; |
dc98df5a KH |
1670 | wait_queue_t wait; |
1671 | }; | |
1672 | ||
1673 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1674 | unsigned mode, int sync, void *arg) | |
1675 | { | |
d79154bb HD |
1676 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1677 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1678 | struct oom_wait_info *oom_wait_info; |
1679 | ||
1680 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1681 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1682 | |
2314b42d JW |
1683 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1684 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1685 | return 0; |
dc98df5a KH |
1686 | return autoremove_wake_function(wait, mode, sync, arg); |
1687 | } | |
1688 | ||
c0ff4b85 | 1689 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1690 | { |
c2b42d3c TH |
1691 | /* |
1692 | * For the following lockless ->under_oom test, the only required | |
1693 | * guarantee is that it must see the state asserted by an OOM when | |
1694 | * this function is called as a result of userland actions | |
1695 | * triggered by the notification of the OOM. This is trivially | |
1696 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1697 | * triggering notification. | |
1698 | */ | |
1699 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1700 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1701 | } |
1702 | ||
3812c8c8 | 1703 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 1704 | { |
626ebc41 | 1705 | if (!current->memcg_may_oom) |
3812c8c8 | 1706 | return; |
867578cb | 1707 | /* |
49426420 JW |
1708 | * We are in the middle of the charge context here, so we |
1709 | * don't want to block when potentially sitting on a callstack | |
1710 | * that holds all kinds of filesystem and mm locks. | |
1711 | * | |
1712 | * Also, the caller may handle a failed allocation gracefully | |
1713 | * (like optional page cache readahead) and so an OOM killer | |
1714 | * invocation might not even be necessary. | |
1715 | * | |
1716 | * That's why we don't do anything here except remember the | |
1717 | * OOM context and then deal with it at the end of the page | |
1718 | * fault when the stack is unwound, the locks are released, | |
1719 | * and when we know whether the fault was overall successful. | |
867578cb | 1720 | */ |
49426420 | 1721 | css_get(&memcg->css); |
626ebc41 TH |
1722 | current->memcg_in_oom = memcg; |
1723 | current->memcg_oom_gfp_mask = mask; | |
1724 | current->memcg_oom_order = order; | |
3812c8c8 JW |
1725 | } |
1726 | ||
1727 | /** | |
1728 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1729 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1730 | * |
49426420 JW |
1731 | * This has to be called at the end of a page fault if the memcg OOM |
1732 | * handler was enabled. | |
3812c8c8 | 1733 | * |
49426420 | 1734 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1735 | * sleep on a waitqueue until the userspace task resolves the |
1736 | * situation. Sleeping directly in the charge context with all kinds | |
1737 | * of locks held is not a good idea, instead we remember an OOM state | |
1738 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1739 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1740 | * |
1741 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1742 | * completed, %false otherwise. |
3812c8c8 | 1743 | */ |
49426420 | 1744 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1745 | { |
626ebc41 | 1746 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1747 | struct oom_wait_info owait; |
49426420 | 1748 | bool locked; |
3812c8c8 JW |
1749 | |
1750 | /* OOM is global, do not handle */ | |
3812c8c8 | 1751 | if (!memcg) |
49426420 | 1752 | return false; |
3812c8c8 | 1753 | |
c32b3cbe | 1754 | if (!handle || oom_killer_disabled) |
49426420 | 1755 | goto cleanup; |
3812c8c8 JW |
1756 | |
1757 | owait.memcg = memcg; | |
1758 | owait.wait.flags = 0; | |
1759 | owait.wait.func = memcg_oom_wake_function; | |
1760 | owait.wait.private = current; | |
1761 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 1762 | |
3812c8c8 | 1763 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1764 | mem_cgroup_mark_under_oom(memcg); |
1765 | ||
1766 | locked = mem_cgroup_oom_trylock(memcg); | |
1767 | ||
1768 | if (locked) | |
1769 | mem_cgroup_oom_notify(memcg); | |
1770 | ||
1771 | if (locked && !memcg->oom_kill_disable) { | |
1772 | mem_cgroup_unmark_under_oom(memcg); | |
1773 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1774 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1775 | current->memcg_oom_order); | |
49426420 | 1776 | } else { |
3812c8c8 | 1777 | schedule(); |
49426420 JW |
1778 | mem_cgroup_unmark_under_oom(memcg); |
1779 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1780 | } | |
1781 | ||
1782 | if (locked) { | |
fb2a6fc5 JW |
1783 | mem_cgroup_oom_unlock(memcg); |
1784 | /* | |
1785 | * There is no guarantee that an OOM-lock contender | |
1786 | * sees the wakeups triggered by the OOM kill | |
1787 | * uncharges. Wake any sleepers explicitely. | |
1788 | */ | |
1789 | memcg_oom_recover(memcg); | |
1790 | } | |
49426420 | 1791 | cleanup: |
626ebc41 | 1792 | current->memcg_in_oom = NULL; |
3812c8c8 | 1793 | css_put(&memcg->css); |
867578cb | 1794 | return true; |
0b7f569e KH |
1795 | } |
1796 | ||
d7365e78 JW |
1797 | /** |
1798 | * mem_cgroup_begin_page_stat - begin a page state statistics transaction | |
1799 | * @page: page that is going to change accounted state | |
32047e2a | 1800 | * |
d7365e78 JW |
1801 | * This function must mark the beginning of an accounted page state |
1802 | * change to prevent double accounting when the page is concurrently | |
1803 | * being moved to another memcg: | |
32047e2a | 1804 | * |
6de22619 | 1805 | * memcg = mem_cgroup_begin_page_stat(page); |
d7365e78 JW |
1806 | * if (TestClearPageState(page)) |
1807 | * mem_cgroup_update_page_stat(memcg, state, -1); | |
6de22619 | 1808 | * mem_cgroup_end_page_stat(memcg); |
d69b042f | 1809 | */ |
6de22619 | 1810 | struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page) |
89c06bd5 KH |
1811 | { |
1812 | struct mem_cgroup *memcg; | |
6de22619 | 1813 | unsigned long flags; |
89c06bd5 | 1814 | |
6de22619 JW |
1815 | /* |
1816 | * The RCU lock is held throughout the transaction. The fast | |
1817 | * path can get away without acquiring the memcg->move_lock | |
1818 | * because page moving starts with an RCU grace period. | |
1819 | * | |
1820 | * The RCU lock also protects the memcg from being freed when | |
1821 | * the page state that is going to change is the only thing | |
1822 | * preventing the page from being uncharged. | |
1823 | * E.g. end-writeback clearing PageWriteback(), which allows | |
1824 | * migration to go ahead and uncharge the page before the | |
1825 | * account transaction might be complete. | |
1826 | */ | |
d7365e78 JW |
1827 | rcu_read_lock(); |
1828 | ||
1829 | if (mem_cgroup_disabled()) | |
1830 | return NULL; | |
89c06bd5 | 1831 | again: |
1306a85a | 1832 | memcg = page->mem_cgroup; |
29833315 | 1833 | if (unlikely(!memcg)) |
d7365e78 JW |
1834 | return NULL; |
1835 | ||
bdcbb659 | 1836 | if (atomic_read(&memcg->moving_account) <= 0) |
d7365e78 | 1837 | return memcg; |
89c06bd5 | 1838 | |
6de22619 | 1839 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 1840 | if (memcg != page->mem_cgroup) { |
6de22619 | 1841 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
1842 | goto again; |
1843 | } | |
6de22619 JW |
1844 | |
1845 | /* | |
1846 | * When charge migration first begins, we can have locked and | |
1847 | * unlocked page stat updates happening concurrently. Track | |
1848 | * the task who has the lock for mem_cgroup_end_page_stat(). | |
1849 | */ | |
1850 | memcg->move_lock_task = current; | |
1851 | memcg->move_lock_flags = flags; | |
d7365e78 JW |
1852 | |
1853 | return memcg; | |
89c06bd5 | 1854 | } |
c4843a75 | 1855 | EXPORT_SYMBOL(mem_cgroup_begin_page_stat); |
89c06bd5 | 1856 | |
d7365e78 JW |
1857 | /** |
1858 | * mem_cgroup_end_page_stat - finish a page state statistics transaction | |
1859 | * @memcg: the memcg that was accounted against | |
d7365e78 | 1860 | */ |
6de22619 | 1861 | void mem_cgroup_end_page_stat(struct mem_cgroup *memcg) |
89c06bd5 | 1862 | { |
6de22619 JW |
1863 | if (memcg && memcg->move_lock_task == current) { |
1864 | unsigned long flags = memcg->move_lock_flags; | |
1865 | ||
1866 | memcg->move_lock_task = NULL; | |
1867 | memcg->move_lock_flags = 0; | |
1868 | ||
1869 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
1870 | } | |
89c06bd5 | 1871 | |
d7365e78 | 1872 | rcu_read_unlock(); |
89c06bd5 | 1873 | } |
c4843a75 | 1874 | EXPORT_SYMBOL(mem_cgroup_end_page_stat); |
89c06bd5 | 1875 | |
cdec2e42 KH |
1876 | /* |
1877 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1878 | * TODO: maybe necessary to use big numbers in big irons. | |
1879 | */ | |
7ec99d62 | 1880 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
1881 | struct memcg_stock_pcp { |
1882 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 1883 | unsigned int nr_pages; |
cdec2e42 | 1884 | struct work_struct work; |
26fe6168 | 1885 | unsigned long flags; |
a0db00fc | 1886 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
1887 | }; |
1888 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 1889 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 1890 | |
a0956d54 SS |
1891 | /** |
1892 | * consume_stock: Try to consume stocked charge on this cpu. | |
1893 | * @memcg: memcg to consume from. | |
1894 | * @nr_pages: how many pages to charge. | |
1895 | * | |
1896 | * The charges will only happen if @memcg matches the current cpu's memcg | |
1897 | * stock, and at least @nr_pages are available in that stock. Failure to | |
1898 | * service an allocation will refill the stock. | |
1899 | * | |
1900 | * returns true if successful, false otherwise. | |
cdec2e42 | 1901 | */ |
a0956d54 | 1902 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1903 | { |
1904 | struct memcg_stock_pcp *stock; | |
3e32cb2e | 1905 | bool ret = false; |
cdec2e42 | 1906 | |
a0956d54 | 1907 | if (nr_pages > CHARGE_BATCH) |
3e32cb2e | 1908 | return ret; |
a0956d54 | 1909 | |
cdec2e42 | 1910 | stock = &get_cpu_var(memcg_stock); |
3e32cb2e | 1911 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 1912 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
1913 | ret = true; |
1914 | } | |
cdec2e42 KH |
1915 | put_cpu_var(memcg_stock); |
1916 | return ret; | |
1917 | } | |
1918 | ||
1919 | /* | |
3e32cb2e | 1920 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
1921 | */ |
1922 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1923 | { | |
1924 | struct mem_cgroup *old = stock->cached; | |
1925 | ||
11c9ea4e | 1926 | if (stock->nr_pages) { |
3e32cb2e | 1927 | page_counter_uncharge(&old->memory, stock->nr_pages); |
cdec2e42 | 1928 | if (do_swap_account) |
3e32cb2e | 1929 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 1930 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 1931 | stock->nr_pages = 0; |
cdec2e42 KH |
1932 | } |
1933 | stock->cached = NULL; | |
cdec2e42 KH |
1934 | } |
1935 | ||
1936 | /* | |
1937 | * This must be called under preempt disabled or must be called by | |
1938 | * a thread which is pinned to local cpu. | |
1939 | */ | |
1940 | static void drain_local_stock(struct work_struct *dummy) | |
1941 | { | |
7c8e0181 | 1942 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
cdec2e42 | 1943 | drain_stock(stock); |
26fe6168 | 1944 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
1945 | } |
1946 | ||
1947 | /* | |
3e32cb2e | 1948 | * Cache charges(val) to local per_cpu area. |
320cc51d | 1949 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 1950 | */ |
c0ff4b85 | 1951 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1952 | { |
1953 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
1954 | ||
c0ff4b85 | 1955 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 1956 | drain_stock(stock); |
c0ff4b85 | 1957 | stock->cached = memcg; |
cdec2e42 | 1958 | } |
11c9ea4e | 1959 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
1960 | put_cpu_var(memcg_stock); |
1961 | } | |
1962 | ||
1963 | /* | |
c0ff4b85 | 1964 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 1965 | * of the hierarchy under it. |
cdec2e42 | 1966 | */ |
6d3d6aa2 | 1967 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 1968 | { |
26fe6168 | 1969 | int cpu, curcpu; |
d38144b7 | 1970 | |
6d3d6aa2 JW |
1971 | /* If someone's already draining, avoid adding running more workers. */ |
1972 | if (!mutex_trylock(&percpu_charge_mutex)) | |
1973 | return; | |
cdec2e42 | 1974 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 1975 | get_online_cpus(); |
5af12d0e | 1976 | curcpu = get_cpu(); |
cdec2e42 KH |
1977 | for_each_online_cpu(cpu) { |
1978 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 1979 | struct mem_cgroup *memcg; |
26fe6168 | 1980 | |
c0ff4b85 R |
1981 | memcg = stock->cached; |
1982 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 1983 | continue; |
2314b42d | 1984 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) |
3e92041d | 1985 | continue; |
d1a05b69 MH |
1986 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
1987 | if (cpu == curcpu) | |
1988 | drain_local_stock(&stock->work); | |
1989 | else | |
1990 | schedule_work_on(cpu, &stock->work); | |
1991 | } | |
cdec2e42 | 1992 | } |
5af12d0e | 1993 | put_cpu(); |
f894ffa8 | 1994 | put_online_cpus(); |
9f50fad6 | 1995 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
1996 | } |
1997 | ||
0db0628d | 1998 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
1999 | unsigned long action, |
2000 | void *hcpu) | |
2001 | { | |
2002 | int cpu = (unsigned long)hcpu; | |
2003 | struct memcg_stock_pcp *stock; | |
2004 | ||
619d094b | 2005 | if (action == CPU_ONLINE) |
1489ebad | 2006 | return NOTIFY_OK; |
1489ebad | 2007 | |
d833049b | 2008 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2009 | return NOTIFY_OK; |
711d3d2c | 2010 | |
cdec2e42 KH |
2011 | stock = &per_cpu(memcg_stock, cpu); |
2012 | drain_stock(stock); | |
2013 | return NOTIFY_OK; | |
2014 | } | |
2015 | ||
b23afb93 TH |
2016 | /* |
2017 | * Scheduled by try_charge() to be executed from the userland return path | |
2018 | * and reclaims memory over the high limit. | |
2019 | */ | |
2020 | void mem_cgroup_handle_over_high(void) | |
2021 | { | |
2022 | unsigned int nr_pages = current->memcg_nr_pages_over_high; | |
2023 | struct mem_cgroup *memcg, *pos; | |
2024 | ||
2025 | if (likely(!nr_pages)) | |
2026 | return; | |
2027 | ||
2028 | pos = memcg = get_mem_cgroup_from_mm(current->mm); | |
2029 | ||
2030 | do { | |
2031 | if (page_counter_read(&pos->memory) <= pos->high) | |
2032 | continue; | |
2033 | mem_cgroup_events(pos, MEMCG_HIGH, 1); | |
2034 | try_to_free_mem_cgroup_pages(pos, nr_pages, GFP_KERNEL, true); | |
2035 | } while ((pos = parent_mem_cgroup(pos))); | |
2036 | ||
2037 | css_put(&memcg->css); | |
2038 | current->memcg_nr_pages_over_high = 0; | |
2039 | } | |
2040 | ||
00501b53 JW |
2041 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2042 | unsigned int nr_pages) | |
8a9f3ccd | 2043 | { |
7ec99d62 | 2044 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 2045 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 2046 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2047 | struct page_counter *counter; |
6539cc05 | 2048 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2049 | bool may_swap = true; |
2050 | bool drained = false; | |
a636b327 | 2051 | |
ce00a967 | 2052 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 2053 | return 0; |
6539cc05 | 2054 | retry: |
b6b6cc72 | 2055 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2056 | return 0; |
8a9f3ccd | 2057 | |
3fbe7244 | 2058 | if (!do_swap_account || |
6071ca52 JW |
2059 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2060 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2061 | goto done_restock; |
3fbe7244 | 2062 | if (do_swap_account) |
3e32cb2e JW |
2063 | page_counter_uncharge(&memcg->memsw, batch); |
2064 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2065 | } else { |
3e32cb2e | 2066 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2067 | may_swap = false; |
3fbe7244 | 2068 | } |
7a81b88c | 2069 | |
6539cc05 JW |
2070 | if (batch > nr_pages) { |
2071 | batch = nr_pages; | |
2072 | goto retry; | |
2073 | } | |
6d61ef40 | 2074 | |
06b078fc JW |
2075 | /* |
2076 | * Unlike in global OOM situations, memcg is not in a physical | |
2077 | * memory shortage. Allow dying and OOM-killed tasks to | |
2078 | * bypass the last charges so that they can exit quickly and | |
2079 | * free their memory. | |
2080 | */ | |
2081 | if (unlikely(test_thread_flag(TIF_MEMDIE) || | |
2082 | fatal_signal_pending(current) || | |
2083 | current->flags & PF_EXITING)) | |
10d53c74 | 2084 | goto force; |
06b078fc JW |
2085 | |
2086 | if (unlikely(task_in_memcg_oom(current))) | |
2087 | goto nomem; | |
2088 | ||
d0164adc | 2089 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2090 | goto nomem; |
4b534334 | 2091 | |
241994ed JW |
2092 | mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1); |
2093 | ||
b70a2a21 JW |
2094 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2095 | gfp_mask, may_swap); | |
6539cc05 | 2096 | |
61e02c74 | 2097 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2098 | goto retry; |
28c34c29 | 2099 | |
b70a2a21 | 2100 | if (!drained) { |
6d3d6aa2 | 2101 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2102 | drained = true; |
2103 | goto retry; | |
2104 | } | |
2105 | ||
28c34c29 JW |
2106 | if (gfp_mask & __GFP_NORETRY) |
2107 | goto nomem; | |
6539cc05 JW |
2108 | /* |
2109 | * Even though the limit is exceeded at this point, reclaim | |
2110 | * may have been able to free some pages. Retry the charge | |
2111 | * before killing the task. | |
2112 | * | |
2113 | * Only for regular pages, though: huge pages are rather | |
2114 | * unlikely to succeed so close to the limit, and we fall back | |
2115 | * to regular pages anyway in case of failure. | |
2116 | */ | |
61e02c74 | 2117 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2118 | goto retry; |
2119 | /* | |
2120 | * At task move, charge accounts can be doubly counted. So, it's | |
2121 | * better to wait until the end of task_move if something is going on. | |
2122 | */ | |
2123 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2124 | goto retry; | |
2125 | ||
9b130619 JW |
2126 | if (nr_retries--) |
2127 | goto retry; | |
2128 | ||
06b078fc | 2129 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2130 | goto force; |
06b078fc | 2131 | |
6539cc05 | 2132 | if (fatal_signal_pending(current)) |
10d53c74 | 2133 | goto force; |
6539cc05 | 2134 | |
241994ed JW |
2135 | mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1); |
2136 | ||
3608de07 JM |
2137 | mem_cgroup_oom(mem_over_limit, gfp_mask, |
2138 | get_order(nr_pages * PAGE_SIZE)); | |
7a81b88c | 2139 | nomem: |
6d1fdc48 | 2140 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2141 | return -ENOMEM; |
10d53c74 TH |
2142 | force: |
2143 | /* | |
2144 | * The allocation either can't fail or will lead to more memory | |
2145 | * being freed very soon. Allow memory usage go over the limit | |
2146 | * temporarily by force charging it. | |
2147 | */ | |
2148 | page_counter_charge(&memcg->memory, nr_pages); | |
2149 | if (do_swap_account) | |
2150 | page_counter_charge(&memcg->memsw, nr_pages); | |
2151 | css_get_many(&memcg->css, nr_pages); | |
2152 | ||
2153 | return 0; | |
6539cc05 JW |
2154 | |
2155 | done_restock: | |
e8ea14cc | 2156 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2157 | if (batch > nr_pages) |
2158 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2159 | |
241994ed | 2160 | /* |
b23afb93 TH |
2161 | * If the hierarchy is above the normal consumption range, schedule |
2162 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2163 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2164 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2165 | * not recorded as it most likely matches current's and won't | |
2166 | * change in the meantime. As high limit is checked again before | |
2167 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2168 | */ |
2169 | do { | |
b23afb93 | 2170 | if (page_counter_read(&memcg->memory) > memcg->high) { |
9516a18a | 2171 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2172 | set_notify_resume(current); |
2173 | break; | |
2174 | } | |
241994ed | 2175 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2176 | |
2177 | return 0; | |
7a81b88c | 2178 | } |
8a9f3ccd | 2179 | |
00501b53 | 2180 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2181 | { |
ce00a967 JW |
2182 | if (mem_cgroup_is_root(memcg)) |
2183 | return; | |
2184 | ||
3e32cb2e | 2185 | page_counter_uncharge(&memcg->memory, nr_pages); |
05b84301 | 2186 | if (do_swap_account) |
3e32cb2e | 2187 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2188 | |
e8ea14cc | 2189 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2190 | } |
2191 | ||
0a31bc97 JW |
2192 | static void lock_page_lru(struct page *page, int *isolated) |
2193 | { | |
2194 | struct zone *zone = page_zone(page); | |
2195 | ||
2196 | spin_lock_irq(&zone->lru_lock); | |
2197 | if (PageLRU(page)) { | |
2198 | struct lruvec *lruvec; | |
2199 | ||
2200 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2201 | ClearPageLRU(page); | |
2202 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2203 | *isolated = 1; | |
2204 | } else | |
2205 | *isolated = 0; | |
2206 | } | |
2207 | ||
2208 | static void unlock_page_lru(struct page *page, int isolated) | |
2209 | { | |
2210 | struct zone *zone = page_zone(page); | |
2211 | ||
2212 | if (isolated) { | |
2213 | struct lruvec *lruvec; | |
2214 | ||
2215 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2216 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
2217 | SetPageLRU(page); | |
2218 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2219 | } | |
2220 | spin_unlock_irq(&zone->lru_lock); | |
2221 | } | |
2222 | ||
00501b53 | 2223 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2224 | bool lrucare) |
7a81b88c | 2225 | { |
0a31bc97 | 2226 | int isolated; |
9ce70c02 | 2227 | |
1306a85a | 2228 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2229 | |
2230 | /* | |
2231 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2232 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2233 | */ | |
0a31bc97 JW |
2234 | if (lrucare) |
2235 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2236 | |
0a31bc97 JW |
2237 | /* |
2238 | * Nobody should be changing or seriously looking at | |
1306a85a | 2239 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2240 | * |
2241 | * - the page is uncharged | |
2242 | * | |
2243 | * - the page is off-LRU | |
2244 | * | |
2245 | * - an anonymous fault has exclusive page access, except for | |
2246 | * a locked page table | |
2247 | * | |
2248 | * - a page cache insertion, a swapin fault, or a migration | |
2249 | * have the page locked | |
2250 | */ | |
1306a85a | 2251 | page->mem_cgroup = memcg; |
9ce70c02 | 2252 | |
0a31bc97 JW |
2253 | if (lrucare) |
2254 | unlock_page_lru(page, isolated); | |
7a81b88c | 2255 | } |
66e1707b | 2256 | |
7ae1e1d0 | 2257 | #ifdef CONFIG_MEMCG_KMEM |
f3bb3043 | 2258 | static int memcg_alloc_cache_id(void) |
55007d84 | 2259 | { |
f3bb3043 VD |
2260 | int id, size; |
2261 | int err; | |
2262 | ||
dbcf73e2 | 2263 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2264 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2265 | if (id < 0) | |
2266 | return id; | |
55007d84 | 2267 | |
dbcf73e2 | 2268 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2269 | return id; |
2270 | ||
2271 | /* | |
2272 | * There's no space for the new id in memcg_caches arrays, | |
2273 | * so we have to grow them. | |
2274 | */ | |
05257a1a | 2275 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2276 | |
2277 | size = 2 * (id + 1); | |
55007d84 GC |
2278 | if (size < MEMCG_CACHES_MIN_SIZE) |
2279 | size = MEMCG_CACHES_MIN_SIZE; | |
2280 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2281 | size = MEMCG_CACHES_MAX_SIZE; | |
2282 | ||
f3bb3043 | 2283 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2284 | if (!err) |
2285 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2286 | if (!err) |
2287 | memcg_nr_cache_ids = size; | |
2288 | ||
2289 | up_write(&memcg_cache_ids_sem); | |
2290 | ||
f3bb3043 | 2291 | if (err) { |
dbcf73e2 | 2292 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2293 | return err; |
2294 | } | |
2295 | return id; | |
2296 | } | |
2297 | ||
2298 | static void memcg_free_cache_id(int id) | |
2299 | { | |
dbcf73e2 | 2300 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2301 | } |
2302 | ||
d5b3cf71 | 2303 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2304 | struct mem_cgroup *memcg; |
2305 | struct kmem_cache *cachep; | |
2306 | struct work_struct work; | |
2307 | }; | |
2308 | ||
d5b3cf71 | 2309 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2310 | { |
d5b3cf71 VD |
2311 | struct memcg_kmem_cache_create_work *cw = |
2312 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2313 | struct mem_cgroup *memcg = cw->memcg; |
2314 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2315 | |
d5b3cf71 | 2316 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2317 | |
5722d094 | 2318 | css_put(&memcg->css); |
d7f25f8a GC |
2319 | kfree(cw); |
2320 | } | |
2321 | ||
2322 | /* | |
2323 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2324 | */ |
d5b3cf71 VD |
2325 | static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2326 | struct kmem_cache *cachep) | |
d7f25f8a | 2327 | { |
d5b3cf71 | 2328 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2329 | |
776ed0f0 | 2330 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
8135be5a | 2331 | if (!cw) |
d7f25f8a | 2332 | return; |
8135be5a VD |
2333 | |
2334 | css_get(&memcg->css); | |
d7f25f8a GC |
2335 | |
2336 | cw->memcg = memcg; | |
2337 | cw->cachep = cachep; | |
d5b3cf71 | 2338 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2339 | |
d7f25f8a GC |
2340 | schedule_work(&cw->work); |
2341 | } | |
2342 | ||
d5b3cf71 VD |
2343 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2344 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
2345 | { |
2346 | /* | |
2347 | * We need to stop accounting when we kmalloc, because if the | |
2348 | * corresponding kmalloc cache is not yet created, the first allocation | |
d5b3cf71 | 2349 | * in __memcg_schedule_kmem_cache_create will recurse. |
0e9d92f2 GC |
2350 | * |
2351 | * However, it is better to enclose the whole function. Depending on | |
2352 | * the debugging options enabled, INIT_WORK(), for instance, can | |
2353 | * trigger an allocation. This too, will make us recurse. Because at | |
2354 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
2355 | * the safest choice is to do it like this, wrapping the whole function. | |
2356 | */ | |
6f185c29 | 2357 | current->memcg_kmem_skip_account = 1; |
d5b3cf71 | 2358 | __memcg_schedule_kmem_cache_create(memcg, cachep); |
6f185c29 | 2359 | current->memcg_kmem_skip_account = 0; |
0e9d92f2 | 2360 | } |
c67a8a68 | 2361 | |
d7f25f8a GC |
2362 | /* |
2363 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
2364 | * We try to use the current memcg's version of the cache. | |
2365 | * | |
2366 | * If the cache does not exist yet, if we are the first user of it, | |
2367 | * we either create it immediately, if possible, or create it asynchronously | |
2368 | * in a workqueue. | |
2369 | * In the latter case, we will let the current allocation go through with | |
2370 | * the original cache. | |
2371 | * | |
2372 | * Can't be called in interrupt context or from kernel threads. | |
2373 | * This function needs to be called with rcu_read_lock() held. | |
2374 | */ | |
230e9fc2 | 2375 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp) |
d7f25f8a GC |
2376 | { |
2377 | struct mem_cgroup *memcg; | |
959c8963 | 2378 | struct kmem_cache *memcg_cachep; |
2a4db7eb | 2379 | int kmemcg_id; |
d7f25f8a | 2380 | |
f7ce3190 | 2381 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2382 | |
230e9fc2 VD |
2383 | if (cachep->flags & SLAB_ACCOUNT) |
2384 | gfp |= __GFP_ACCOUNT; | |
2385 | ||
2386 | if (!(gfp & __GFP_ACCOUNT)) | |
2387 | return cachep; | |
2388 | ||
9d100c5e | 2389 | if (current->memcg_kmem_skip_account) |
0e9d92f2 GC |
2390 | return cachep; |
2391 | ||
8135be5a | 2392 | memcg = get_mem_cgroup_from_mm(current->mm); |
4db0c3c2 | 2393 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
2a4db7eb | 2394 | if (kmemcg_id < 0) |
ca0dde97 | 2395 | goto out; |
d7f25f8a | 2396 | |
2a4db7eb | 2397 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
8135be5a VD |
2398 | if (likely(memcg_cachep)) |
2399 | return memcg_cachep; | |
ca0dde97 LZ |
2400 | |
2401 | /* | |
2402 | * If we are in a safe context (can wait, and not in interrupt | |
2403 | * context), we could be be predictable and return right away. | |
2404 | * This would guarantee that the allocation being performed | |
2405 | * already belongs in the new cache. | |
2406 | * | |
2407 | * However, there are some clashes that can arrive from locking. | |
2408 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2409 | * memcg_create_kmem_cache, this means no further allocation |
2410 | * could happen with the slab_mutex held. So it's better to | |
2411 | * defer everything. | |
ca0dde97 | 2412 | */ |
d5b3cf71 | 2413 | memcg_schedule_kmem_cache_create(memcg, cachep); |
ca0dde97 | 2414 | out: |
8135be5a | 2415 | css_put(&memcg->css); |
ca0dde97 | 2416 | return cachep; |
d7f25f8a | 2417 | } |
d7f25f8a | 2418 | |
8135be5a VD |
2419 | void __memcg_kmem_put_cache(struct kmem_cache *cachep) |
2420 | { | |
2421 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2422 | css_put(&cachep->memcg_params.memcg->css); |
8135be5a VD |
2423 | } |
2424 | ||
f3ccb2c4 VD |
2425 | int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, |
2426 | struct mem_cgroup *memcg) | |
7ae1e1d0 | 2427 | { |
f3ccb2c4 VD |
2428 | unsigned int nr_pages = 1 << order; |
2429 | struct page_counter *counter; | |
7ae1e1d0 GC |
2430 | int ret; |
2431 | ||
f3ccb2c4 | 2432 | if (!memcg_kmem_is_active(memcg)) |
d05e83a6 | 2433 | return 0; |
6d42c232 | 2434 | |
6071ca52 JW |
2435 | if (!page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) |
2436 | return -ENOMEM; | |
7ae1e1d0 | 2437 | |
f3ccb2c4 VD |
2438 | ret = try_charge(memcg, gfp, nr_pages); |
2439 | if (ret) { | |
2440 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
2441 | return ret; | |
7ae1e1d0 GC |
2442 | } |
2443 | ||
f3ccb2c4 | 2444 | page->mem_cgroup = memcg; |
7ae1e1d0 | 2445 | |
f3ccb2c4 | 2446 | return 0; |
7ae1e1d0 GC |
2447 | } |
2448 | ||
f3ccb2c4 | 2449 | int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 2450 | { |
f3ccb2c4 VD |
2451 | struct mem_cgroup *memcg; |
2452 | int ret; | |
7ae1e1d0 | 2453 | |
f3ccb2c4 VD |
2454 | memcg = get_mem_cgroup_from_mm(current->mm); |
2455 | ret = __memcg_kmem_charge_memcg(page, gfp, order, memcg); | |
7ae1e1d0 | 2456 | css_put(&memcg->css); |
d05e83a6 | 2457 | return ret; |
7ae1e1d0 GC |
2458 | } |
2459 | ||
d05e83a6 | 2460 | void __memcg_kmem_uncharge(struct page *page, int order) |
7ae1e1d0 | 2461 | { |
1306a85a | 2462 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 2463 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 2464 | |
7ae1e1d0 GC |
2465 | if (!memcg) |
2466 | return; | |
2467 | ||
309381fe | 2468 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 2469 | |
f3ccb2c4 VD |
2470 | page_counter_uncharge(&memcg->kmem, nr_pages); |
2471 | page_counter_uncharge(&memcg->memory, nr_pages); | |
2472 | if (do_swap_account) | |
2473 | page_counter_uncharge(&memcg->memsw, nr_pages); | |
60d3fd32 | 2474 | |
1306a85a | 2475 | page->mem_cgroup = NULL; |
f3ccb2c4 | 2476 | css_put_many(&memcg->css, nr_pages); |
60d3fd32 | 2477 | } |
7ae1e1d0 GC |
2478 | #endif /* CONFIG_MEMCG_KMEM */ |
2479 | ||
ca3e0214 KH |
2480 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2481 | ||
ca3e0214 KH |
2482 | /* |
2483 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
2484 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
2485 | * charge/uncharge will be never happen and move_account() is done under | |
2486 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 2487 | */ |
e94c8a9c | 2488 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2489 | { |
e94c8a9c | 2490 | int i; |
ca3e0214 | 2491 | |
3d37c4a9 KH |
2492 | if (mem_cgroup_disabled()) |
2493 | return; | |
b070e65c | 2494 | |
29833315 | 2495 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2496 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2497 | |
1306a85a | 2498 | __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
b070e65c | 2499 | HPAGE_PMD_NR); |
ca3e0214 | 2500 | } |
12d27107 | 2501 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2502 | |
c255a458 | 2503 | #ifdef CONFIG_MEMCG_SWAP |
0a31bc97 JW |
2504 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
2505 | bool charge) | |
d13d1443 | 2506 | { |
0a31bc97 JW |
2507 | int val = (charge) ? 1 : -1; |
2508 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); | |
d13d1443 | 2509 | } |
02491447 DN |
2510 | |
2511 | /** | |
2512 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2513 | * @entry: swap entry to be moved | |
2514 | * @from: mem_cgroup which the entry is moved from | |
2515 | * @to: mem_cgroup which the entry is moved to | |
2516 | * | |
2517 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2518 | * as the mem_cgroup's id of @from. | |
2519 | * | |
2520 | * Returns 0 on success, -EINVAL on failure. | |
2521 | * | |
3e32cb2e | 2522 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2523 | * both res and memsw, and called css_get(). |
2524 | */ | |
2525 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2526 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2527 | { |
2528 | unsigned short old_id, new_id; | |
2529 | ||
34c00c31 LZ |
2530 | old_id = mem_cgroup_id(from); |
2531 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2532 | |
2533 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 2534 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 2535 | mem_cgroup_swap_statistics(to, true); |
02491447 DN |
2536 | return 0; |
2537 | } | |
2538 | return -EINVAL; | |
2539 | } | |
2540 | #else | |
2541 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2542 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2543 | { |
2544 | return -EINVAL; | |
2545 | } | |
8c7c6e34 | 2546 | #endif |
d13d1443 | 2547 | |
3e32cb2e | 2548 | static DEFINE_MUTEX(memcg_limit_mutex); |
f212ad7c | 2549 | |
d38d2a75 | 2550 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2551 | unsigned long limit) |
628f4235 | 2552 | { |
3e32cb2e JW |
2553 | unsigned long curusage; |
2554 | unsigned long oldusage; | |
2555 | bool enlarge = false; | |
81d39c20 | 2556 | int retry_count; |
3e32cb2e | 2557 | int ret; |
81d39c20 KH |
2558 | |
2559 | /* | |
2560 | * For keeping hierarchical_reclaim simple, how long we should retry | |
2561 | * is depends on callers. We set our retry-count to be function | |
2562 | * of # of children which we should visit in this loop. | |
2563 | */ | |
3e32cb2e JW |
2564 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2565 | mem_cgroup_count_children(memcg); | |
81d39c20 | 2566 | |
3e32cb2e | 2567 | oldusage = page_counter_read(&memcg->memory); |
628f4235 | 2568 | |
3e32cb2e | 2569 | do { |
628f4235 KH |
2570 | if (signal_pending(current)) { |
2571 | ret = -EINTR; | |
2572 | break; | |
2573 | } | |
3e32cb2e JW |
2574 | |
2575 | mutex_lock(&memcg_limit_mutex); | |
2576 | if (limit > memcg->memsw.limit) { | |
2577 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2578 | ret = -EINVAL; |
628f4235 KH |
2579 | break; |
2580 | } | |
3e32cb2e JW |
2581 | if (limit > memcg->memory.limit) |
2582 | enlarge = true; | |
2583 | ret = page_counter_limit(&memcg->memory, limit); | |
2584 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2585 | |
2586 | if (!ret) | |
2587 | break; | |
2588 | ||
b70a2a21 JW |
2589 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
2590 | ||
3e32cb2e | 2591 | curusage = page_counter_read(&memcg->memory); |
81d39c20 | 2592 | /* Usage is reduced ? */ |
f894ffa8 | 2593 | if (curusage >= oldusage) |
81d39c20 KH |
2594 | retry_count--; |
2595 | else | |
2596 | oldusage = curusage; | |
3e32cb2e JW |
2597 | } while (retry_count); |
2598 | ||
3c11ecf4 KH |
2599 | if (!ret && enlarge) |
2600 | memcg_oom_recover(memcg); | |
14797e23 | 2601 | |
8c7c6e34 KH |
2602 | return ret; |
2603 | } | |
2604 | ||
338c8431 | 2605 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2606 | unsigned long limit) |
8c7c6e34 | 2607 | { |
3e32cb2e JW |
2608 | unsigned long curusage; |
2609 | unsigned long oldusage; | |
2610 | bool enlarge = false; | |
81d39c20 | 2611 | int retry_count; |
3e32cb2e | 2612 | int ret; |
8c7c6e34 | 2613 | |
81d39c20 | 2614 | /* see mem_cgroup_resize_res_limit */ |
3e32cb2e JW |
2615 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2616 | mem_cgroup_count_children(memcg); | |
2617 | ||
2618 | oldusage = page_counter_read(&memcg->memsw); | |
2619 | ||
2620 | do { | |
8c7c6e34 KH |
2621 | if (signal_pending(current)) { |
2622 | ret = -EINTR; | |
2623 | break; | |
2624 | } | |
3e32cb2e JW |
2625 | |
2626 | mutex_lock(&memcg_limit_mutex); | |
2627 | if (limit < memcg->memory.limit) { | |
2628 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2629 | ret = -EINVAL; |
8c7c6e34 KH |
2630 | break; |
2631 | } | |
3e32cb2e JW |
2632 | if (limit > memcg->memsw.limit) |
2633 | enlarge = true; | |
2634 | ret = page_counter_limit(&memcg->memsw, limit); | |
2635 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2636 | |
2637 | if (!ret) | |
2638 | break; | |
2639 | ||
b70a2a21 JW |
2640 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
2641 | ||
3e32cb2e | 2642 | curusage = page_counter_read(&memcg->memsw); |
81d39c20 | 2643 | /* Usage is reduced ? */ |
8c7c6e34 | 2644 | if (curusage >= oldusage) |
628f4235 | 2645 | retry_count--; |
81d39c20 KH |
2646 | else |
2647 | oldusage = curusage; | |
3e32cb2e JW |
2648 | } while (retry_count); |
2649 | ||
3c11ecf4 KH |
2650 | if (!ret && enlarge) |
2651 | memcg_oom_recover(memcg); | |
3e32cb2e | 2652 | |
628f4235 KH |
2653 | return ret; |
2654 | } | |
2655 | ||
0608f43d AM |
2656 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
2657 | gfp_t gfp_mask, | |
2658 | unsigned long *total_scanned) | |
2659 | { | |
2660 | unsigned long nr_reclaimed = 0; | |
2661 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
2662 | unsigned long reclaimed; | |
2663 | int loop = 0; | |
2664 | struct mem_cgroup_tree_per_zone *mctz; | |
3e32cb2e | 2665 | unsigned long excess; |
0608f43d AM |
2666 | unsigned long nr_scanned; |
2667 | ||
2668 | if (order > 0) | |
2669 | return 0; | |
2670 | ||
2671 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
2672 | /* | |
2673 | * This loop can run a while, specially if mem_cgroup's continuously | |
2674 | * keep exceeding their soft limit and putting the system under | |
2675 | * pressure | |
2676 | */ | |
2677 | do { | |
2678 | if (next_mz) | |
2679 | mz = next_mz; | |
2680 | else | |
2681 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2682 | if (!mz) | |
2683 | break; | |
2684 | ||
2685 | nr_scanned = 0; | |
2686 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
2687 | gfp_mask, &nr_scanned); | |
2688 | nr_reclaimed += reclaimed; | |
2689 | *total_scanned += nr_scanned; | |
0a31bc97 | 2690 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 2691 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
2692 | |
2693 | /* | |
2694 | * If we failed to reclaim anything from this memory cgroup | |
2695 | * it is time to move on to the next cgroup | |
2696 | */ | |
2697 | next_mz = NULL; | |
bc2f2e7f VD |
2698 | if (!reclaimed) |
2699 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
2700 | ||
3e32cb2e | 2701 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
2702 | /* |
2703 | * One school of thought says that we should not add | |
2704 | * back the node to the tree if reclaim returns 0. | |
2705 | * But our reclaim could return 0, simply because due | |
2706 | * to priority we are exposing a smaller subset of | |
2707 | * memory to reclaim from. Consider this as a longer | |
2708 | * term TODO. | |
2709 | */ | |
2710 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 2711 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 2712 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
2713 | css_put(&mz->memcg->css); |
2714 | loop++; | |
2715 | /* | |
2716 | * Could not reclaim anything and there are no more | |
2717 | * mem cgroups to try or we seem to be looping without | |
2718 | * reclaiming anything. | |
2719 | */ | |
2720 | if (!nr_reclaimed && | |
2721 | (next_mz == NULL || | |
2722 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2723 | break; | |
2724 | } while (!nr_reclaimed); | |
2725 | if (next_mz) | |
2726 | css_put(&next_mz->memcg->css); | |
2727 | return nr_reclaimed; | |
2728 | } | |
2729 | ||
ea280e7b TH |
2730 | /* |
2731 | * Test whether @memcg has children, dead or alive. Note that this | |
2732 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
2733 | * returns %true if there are child csses according to the cgroup | |
2734 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
2735 | */ | |
b5f99b53 GC |
2736 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
2737 | { | |
ea280e7b TH |
2738 | bool ret; |
2739 | ||
696ac172 | 2740 | /* |
ea280e7b TH |
2741 | * The lock does not prevent addition or deletion of children, but |
2742 | * it prevents a new child from being initialized based on this | |
2743 | * parent in css_online(), so it's enough to decide whether | |
2744 | * hierarchically inherited attributes can still be changed or not. | |
696ac172 | 2745 | */ |
ea280e7b TH |
2746 | lockdep_assert_held(&memcg_create_mutex); |
2747 | ||
2748 | rcu_read_lock(); | |
2749 | ret = css_next_child(NULL, &memcg->css); | |
2750 | rcu_read_unlock(); | |
2751 | return ret; | |
b5f99b53 GC |
2752 | } |
2753 | ||
c26251f9 MH |
2754 | /* |
2755 | * Reclaims as many pages from the given memcg as possible and moves | |
2756 | * the rest to the parent. | |
2757 | * | |
2758 | * Caller is responsible for holding css reference for memcg. | |
2759 | */ | |
2760 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
2761 | { | |
2762 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 2763 | |
c1e862c1 KH |
2764 | /* we call try-to-free pages for make this cgroup empty */ |
2765 | lru_add_drain_all(); | |
f817ed48 | 2766 | /* try to free all pages in this cgroup */ |
3e32cb2e | 2767 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 2768 | int progress; |
c1e862c1 | 2769 | |
c26251f9 MH |
2770 | if (signal_pending(current)) |
2771 | return -EINTR; | |
2772 | ||
b70a2a21 JW |
2773 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
2774 | GFP_KERNEL, true); | |
c1e862c1 | 2775 | if (!progress) { |
f817ed48 | 2776 | nr_retries--; |
c1e862c1 | 2777 | /* maybe some writeback is necessary */ |
8aa7e847 | 2778 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 2779 | } |
f817ed48 KH |
2780 | |
2781 | } | |
ab5196c2 MH |
2782 | |
2783 | return 0; | |
cc847582 KH |
2784 | } |
2785 | ||
6770c64e TH |
2786 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
2787 | char *buf, size_t nbytes, | |
2788 | loff_t off) | |
c1e862c1 | 2789 | { |
6770c64e | 2790 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 2791 | |
d8423011 MH |
2792 | if (mem_cgroup_is_root(memcg)) |
2793 | return -EINVAL; | |
6770c64e | 2794 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
2795 | } |
2796 | ||
182446d0 TH |
2797 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
2798 | struct cftype *cft) | |
18f59ea7 | 2799 | { |
182446d0 | 2800 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
2801 | } |
2802 | ||
182446d0 TH |
2803 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
2804 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
2805 | { |
2806 | int retval = 0; | |
182446d0 | 2807 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 2808 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 2809 | |
0999821b | 2810 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
2811 | |
2812 | if (memcg->use_hierarchy == val) | |
2813 | goto out; | |
2814 | ||
18f59ea7 | 2815 | /* |
af901ca1 | 2816 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
2817 | * in the child subtrees. If it is unset, then the change can |
2818 | * occur, provided the current cgroup has no children. | |
2819 | * | |
2820 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
2821 | * set if there are no children. | |
2822 | */ | |
c0ff4b85 | 2823 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 2824 | (val == 1 || val == 0)) { |
ea280e7b | 2825 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 2826 | memcg->use_hierarchy = val; |
18f59ea7 BS |
2827 | else |
2828 | retval = -EBUSY; | |
2829 | } else | |
2830 | retval = -EINVAL; | |
567fb435 GC |
2831 | |
2832 | out: | |
0999821b | 2833 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
2834 | |
2835 | return retval; | |
2836 | } | |
2837 | ||
3e32cb2e JW |
2838 | static unsigned long tree_stat(struct mem_cgroup *memcg, |
2839 | enum mem_cgroup_stat_index idx) | |
ce00a967 JW |
2840 | { |
2841 | struct mem_cgroup *iter; | |
484ebb3b | 2842 | unsigned long val = 0; |
ce00a967 | 2843 | |
ce00a967 JW |
2844 | for_each_mem_cgroup_tree(iter, memcg) |
2845 | val += mem_cgroup_read_stat(iter, idx); | |
2846 | ||
ce00a967 JW |
2847 | return val; |
2848 | } | |
2849 | ||
6f646156 | 2850 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 2851 | { |
c12176d3 | 2852 | unsigned long val; |
ce00a967 | 2853 | |
3e32cb2e JW |
2854 | if (mem_cgroup_is_root(memcg)) { |
2855 | val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE); | |
2856 | val += tree_stat(memcg, MEM_CGROUP_STAT_RSS); | |
2857 | if (swap) | |
2858 | val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP); | |
2859 | } else { | |
ce00a967 | 2860 | if (!swap) |
3e32cb2e | 2861 | val = page_counter_read(&memcg->memory); |
ce00a967 | 2862 | else |
3e32cb2e | 2863 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 2864 | } |
c12176d3 | 2865 | return val; |
ce00a967 JW |
2866 | } |
2867 | ||
3e32cb2e JW |
2868 | enum { |
2869 | RES_USAGE, | |
2870 | RES_LIMIT, | |
2871 | RES_MAX_USAGE, | |
2872 | RES_FAILCNT, | |
2873 | RES_SOFT_LIMIT, | |
2874 | }; | |
ce00a967 | 2875 | |
791badbd | 2876 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 2877 | struct cftype *cft) |
8cdea7c0 | 2878 | { |
182446d0 | 2879 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 2880 | struct page_counter *counter; |
af36f906 | 2881 | |
3e32cb2e | 2882 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 2883 | case _MEM: |
3e32cb2e JW |
2884 | counter = &memcg->memory; |
2885 | break; | |
8c7c6e34 | 2886 | case _MEMSWAP: |
3e32cb2e JW |
2887 | counter = &memcg->memsw; |
2888 | break; | |
510fc4e1 | 2889 | case _KMEM: |
3e32cb2e | 2890 | counter = &memcg->kmem; |
510fc4e1 | 2891 | break; |
8c7c6e34 KH |
2892 | default: |
2893 | BUG(); | |
8c7c6e34 | 2894 | } |
3e32cb2e JW |
2895 | |
2896 | switch (MEMFILE_ATTR(cft->private)) { | |
2897 | case RES_USAGE: | |
2898 | if (counter == &memcg->memory) | |
c12176d3 | 2899 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 2900 | if (counter == &memcg->memsw) |
c12176d3 | 2901 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
2902 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
2903 | case RES_LIMIT: | |
2904 | return (u64)counter->limit * PAGE_SIZE; | |
2905 | case RES_MAX_USAGE: | |
2906 | return (u64)counter->watermark * PAGE_SIZE; | |
2907 | case RES_FAILCNT: | |
2908 | return counter->failcnt; | |
2909 | case RES_SOFT_LIMIT: | |
2910 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
2911 | default: | |
2912 | BUG(); | |
2913 | } | |
8cdea7c0 | 2914 | } |
510fc4e1 | 2915 | |
510fc4e1 | 2916 | #ifdef CONFIG_MEMCG_KMEM |
8c0145b6 VD |
2917 | static int memcg_activate_kmem(struct mem_cgroup *memcg, |
2918 | unsigned long nr_pages) | |
d6441637 VD |
2919 | { |
2920 | int err = 0; | |
2921 | int memcg_id; | |
2922 | ||
2a4db7eb | 2923 | BUG_ON(memcg->kmemcg_id >= 0); |
2788cf0c | 2924 | BUG_ON(memcg->kmem_acct_activated); |
2a4db7eb | 2925 | BUG_ON(memcg->kmem_acct_active); |
d6441637 | 2926 | |
510fc4e1 GC |
2927 | /* |
2928 | * For simplicity, we won't allow this to be disabled. It also can't | |
2929 | * be changed if the cgroup has children already, or if tasks had | |
2930 | * already joined. | |
2931 | * | |
2932 | * If tasks join before we set the limit, a person looking at | |
2933 | * kmem.usage_in_bytes will have no way to determine when it took | |
2934 | * place, which makes the value quite meaningless. | |
2935 | * | |
2936 | * After it first became limited, changes in the value of the limit are | |
2937 | * of course permitted. | |
510fc4e1 | 2938 | */ |
0999821b | 2939 | mutex_lock(&memcg_create_mutex); |
27bd4dbb | 2940 | if (cgroup_is_populated(memcg->css.cgroup) || |
ea280e7b | 2941 | (memcg->use_hierarchy && memcg_has_children(memcg))) |
d6441637 VD |
2942 | err = -EBUSY; |
2943 | mutex_unlock(&memcg_create_mutex); | |
2944 | if (err) | |
2945 | goto out; | |
510fc4e1 | 2946 | |
f3bb3043 | 2947 | memcg_id = memcg_alloc_cache_id(); |
d6441637 VD |
2948 | if (memcg_id < 0) { |
2949 | err = memcg_id; | |
2950 | goto out; | |
2951 | } | |
2952 | ||
d6441637 | 2953 | /* |
900a38f0 VD |
2954 | * We couldn't have accounted to this cgroup, because it hasn't got |
2955 | * activated yet, so this should succeed. | |
d6441637 | 2956 | */ |
3e32cb2e | 2957 | err = page_counter_limit(&memcg->kmem, nr_pages); |
d6441637 VD |
2958 | VM_BUG_ON(err); |
2959 | ||
2960 | static_key_slow_inc(&memcg_kmem_enabled_key); | |
2961 | /* | |
900a38f0 VD |
2962 | * A memory cgroup is considered kmem-active as soon as it gets |
2963 | * kmemcg_id. Setting the id after enabling static branching will | |
d6441637 VD |
2964 | * guarantee no one starts accounting before all call sites are |
2965 | * patched. | |
2966 | */ | |
900a38f0 | 2967 | memcg->kmemcg_id = memcg_id; |
2788cf0c | 2968 | memcg->kmem_acct_activated = true; |
2a4db7eb | 2969 | memcg->kmem_acct_active = true; |
510fc4e1 | 2970 | out: |
d6441637 | 2971 | return err; |
d6441637 VD |
2972 | } |
2973 | ||
d6441637 | 2974 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2975 | unsigned long limit) |
d6441637 VD |
2976 | { |
2977 | int ret; | |
2978 | ||
3e32cb2e | 2979 | mutex_lock(&memcg_limit_mutex); |
d6441637 | 2980 | if (!memcg_kmem_is_active(memcg)) |
3e32cb2e | 2981 | ret = memcg_activate_kmem(memcg, limit); |
d6441637 | 2982 | else |
3e32cb2e JW |
2983 | ret = page_counter_limit(&memcg->kmem, limit); |
2984 | mutex_unlock(&memcg_limit_mutex); | |
510fc4e1 GC |
2985 | return ret; |
2986 | } | |
2987 | ||
55007d84 | 2988 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 2989 | { |
55007d84 | 2990 | int ret = 0; |
510fc4e1 | 2991 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
55007d84 | 2992 | |
d6441637 VD |
2993 | if (!parent) |
2994 | return 0; | |
55007d84 | 2995 | |
8c0145b6 | 2996 | mutex_lock(&memcg_limit_mutex); |
55007d84 | 2997 | /* |
d6441637 VD |
2998 | * If the parent cgroup is not kmem-active now, it cannot be activated |
2999 | * after this point, because it has at least one child already. | |
55007d84 | 3000 | */ |
d6441637 | 3001 | if (memcg_kmem_is_active(parent)) |
8c0145b6 VD |
3002 | ret = memcg_activate_kmem(memcg, PAGE_COUNTER_MAX); |
3003 | mutex_unlock(&memcg_limit_mutex); | |
55007d84 | 3004 | return ret; |
510fc4e1 | 3005 | } |
d6441637 VD |
3006 | #else |
3007 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
3e32cb2e | 3008 | unsigned long limit) |
d6441637 VD |
3009 | { |
3010 | return -EINVAL; | |
3011 | } | |
6d043990 | 3012 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 3013 | |
628f4235 KH |
3014 | /* |
3015 | * The user of this function is... | |
3016 | * RES_LIMIT. | |
3017 | */ | |
451af504 TH |
3018 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3019 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3020 | { |
451af504 | 3021 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3022 | unsigned long nr_pages; |
628f4235 KH |
3023 | int ret; |
3024 | ||
451af504 | 3025 | buf = strstrip(buf); |
650c5e56 | 3026 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3027 | if (ret) |
3028 | return ret; | |
af36f906 | 3029 | |
3e32cb2e | 3030 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3031 | case RES_LIMIT: |
4b3bde4c BS |
3032 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3033 | ret = -EINVAL; | |
3034 | break; | |
3035 | } | |
3e32cb2e JW |
3036 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3037 | case _MEM: | |
3038 | ret = mem_cgroup_resize_limit(memcg, nr_pages); | |
8c7c6e34 | 3039 | break; |
3e32cb2e JW |
3040 | case _MEMSWAP: |
3041 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); | |
296c81d8 | 3042 | break; |
3e32cb2e JW |
3043 | case _KMEM: |
3044 | ret = memcg_update_kmem_limit(memcg, nr_pages); | |
3045 | break; | |
3046 | } | |
296c81d8 | 3047 | break; |
3e32cb2e JW |
3048 | case RES_SOFT_LIMIT: |
3049 | memcg->soft_limit = nr_pages; | |
3050 | ret = 0; | |
628f4235 KH |
3051 | break; |
3052 | } | |
451af504 | 3053 | return ret ?: nbytes; |
8cdea7c0 BS |
3054 | } |
3055 | ||
6770c64e TH |
3056 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3057 | size_t nbytes, loff_t off) | |
c84872e1 | 3058 | { |
6770c64e | 3059 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3060 | struct page_counter *counter; |
c84872e1 | 3061 | |
3e32cb2e JW |
3062 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3063 | case _MEM: | |
3064 | counter = &memcg->memory; | |
3065 | break; | |
3066 | case _MEMSWAP: | |
3067 | counter = &memcg->memsw; | |
3068 | break; | |
3069 | case _KMEM: | |
3070 | counter = &memcg->kmem; | |
3071 | break; | |
3072 | default: | |
3073 | BUG(); | |
3074 | } | |
af36f906 | 3075 | |
3e32cb2e | 3076 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3077 | case RES_MAX_USAGE: |
3e32cb2e | 3078 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3079 | break; |
3080 | case RES_FAILCNT: | |
3e32cb2e | 3081 | counter->failcnt = 0; |
29f2a4da | 3082 | break; |
3e32cb2e JW |
3083 | default: |
3084 | BUG(); | |
29f2a4da | 3085 | } |
f64c3f54 | 3086 | |
6770c64e | 3087 | return nbytes; |
c84872e1 PE |
3088 | } |
3089 | ||
182446d0 | 3090 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3091 | struct cftype *cft) |
3092 | { | |
182446d0 | 3093 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3094 | } |
3095 | ||
02491447 | 3096 | #ifdef CONFIG_MMU |
182446d0 | 3097 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3098 | struct cftype *cft, u64 val) |
3099 | { | |
182446d0 | 3100 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3101 | |
1dfab5ab | 3102 | if (val & ~MOVE_MASK) |
7dc74be0 | 3103 | return -EINVAL; |
ee5e8472 | 3104 | |
7dc74be0 | 3105 | /* |
ee5e8472 GC |
3106 | * No kind of locking is needed in here, because ->can_attach() will |
3107 | * check this value once in the beginning of the process, and then carry | |
3108 | * on with stale data. This means that changes to this value will only | |
3109 | * affect task migrations starting after the change. | |
7dc74be0 | 3110 | */ |
c0ff4b85 | 3111 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3112 | return 0; |
3113 | } | |
02491447 | 3114 | #else |
182446d0 | 3115 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3116 | struct cftype *cft, u64 val) |
3117 | { | |
3118 | return -ENOSYS; | |
3119 | } | |
3120 | #endif | |
7dc74be0 | 3121 | |
406eb0c9 | 3122 | #ifdef CONFIG_NUMA |
2da8ca82 | 3123 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3124 | { |
25485de6 GT |
3125 | struct numa_stat { |
3126 | const char *name; | |
3127 | unsigned int lru_mask; | |
3128 | }; | |
3129 | ||
3130 | static const struct numa_stat stats[] = { | |
3131 | { "total", LRU_ALL }, | |
3132 | { "file", LRU_ALL_FILE }, | |
3133 | { "anon", LRU_ALL_ANON }, | |
3134 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3135 | }; | |
3136 | const struct numa_stat *stat; | |
406eb0c9 | 3137 | int nid; |
25485de6 | 3138 | unsigned long nr; |
2da8ca82 | 3139 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 3140 | |
25485de6 GT |
3141 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3142 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3143 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3144 | for_each_node_state(nid, N_MEMORY) { | |
3145 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3146 | stat->lru_mask); | |
3147 | seq_printf(m, " N%d=%lu", nid, nr); | |
3148 | } | |
3149 | seq_putc(m, '\n'); | |
406eb0c9 | 3150 | } |
406eb0c9 | 3151 | |
071aee13 YH |
3152 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3153 | struct mem_cgroup *iter; | |
3154 | ||
3155 | nr = 0; | |
3156 | for_each_mem_cgroup_tree(iter, memcg) | |
3157 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3158 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3159 | for_each_node_state(nid, N_MEMORY) { | |
3160 | nr = 0; | |
3161 | for_each_mem_cgroup_tree(iter, memcg) | |
3162 | nr += mem_cgroup_node_nr_lru_pages( | |
3163 | iter, nid, stat->lru_mask); | |
3164 | seq_printf(m, " N%d=%lu", nid, nr); | |
3165 | } | |
3166 | seq_putc(m, '\n'); | |
406eb0c9 | 3167 | } |
406eb0c9 | 3168 | |
406eb0c9 YH |
3169 | return 0; |
3170 | } | |
3171 | #endif /* CONFIG_NUMA */ | |
3172 | ||
2da8ca82 | 3173 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3174 | { |
2da8ca82 | 3175 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 3176 | unsigned long memory, memsw; |
af7c4b0e JW |
3177 | struct mem_cgroup *mi; |
3178 | unsigned int i; | |
406eb0c9 | 3179 | |
0ca44b14 GT |
3180 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) != |
3181 | MEM_CGROUP_STAT_NSTATS); | |
3182 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) != | |
3183 | MEM_CGROUP_EVENTS_NSTATS); | |
70bc068c RS |
3184 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3185 | ||
af7c4b0e | 3186 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 3187 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 3188 | continue; |
484ebb3b | 3189 | seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i], |
af7c4b0e | 3190 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); |
1dd3a273 | 3191 | } |
7b854121 | 3192 | |
af7c4b0e JW |
3193 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
3194 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
3195 | mem_cgroup_read_events(memcg, i)); | |
3196 | ||
3197 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3198 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
3199 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
3200 | ||
14067bb3 | 3201 | /* Hierarchical information */ |
3e32cb2e JW |
3202 | memory = memsw = PAGE_COUNTER_MAX; |
3203 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
3204 | memory = min(memory, mi->memory.limit); | |
3205 | memsw = min(memsw, mi->memsw.limit); | |
fee7b548 | 3206 | } |
3e32cb2e JW |
3207 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3208 | (u64)memory * PAGE_SIZE); | |
3209 | if (do_swap_account) | |
3210 | seq_printf(m, "hierarchical_memsw_limit %llu\n", | |
3211 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3212 | |
af7c4b0e | 3213 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
484ebb3b | 3214 | unsigned long long val = 0; |
af7c4b0e | 3215 | |
bff6bb83 | 3216 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 3217 | continue; |
af7c4b0e JW |
3218 | for_each_mem_cgroup_tree(mi, memcg) |
3219 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
484ebb3b | 3220 | seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val); |
af7c4b0e JW |
3221 | } |
3222 | ||
3223 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
3224 | unsigned long long val = 0; | |
3225 | ||
3226 | for_each_mem_cgroup_tree(mi, memcg) | |
3227 | val += mem_cgroup_read_events(mi, i); | |
3228 | seq_printf(m, "total_%s %llu\n", | |
3229 | mem_cgroup_events_names[i], val); | |
3230 | } | |
3231 | ||
3232 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
3233 | unsigned long long val = 0; | |
3234 | ||
3235 | for_each_mem_cgroup_tree(mi, memcg) | |
3236 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
3237 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 3238 | } |
14067bb3 | 3239 | |
7f016ee8 | 3240 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
3241 | { |
3242 | int nid, zid; | |
3243 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 3244 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3245 | unsigned long recent_rotated[2] = {0, 0}; |
3246 | unsigned long recent_scanned[2] = {0, 0}; | |
3247 | ||
3248 | for_each_online_node(nid) | |
3249 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
e231875b | 3250 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
89abfab1 | 3251 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 3252 | |
89abfab1 HD |
3253 | recent_rotated[0] += rstat->recent_rotated[0]; |
3254 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3255 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3256 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 3257 | } |
78ccf5b5 JW |
3258 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3259 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3260 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3261 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3262 | } |
3263 | #endif | |
3264 | ||
d2ceb9b7 KH |
3265 | return 0; |
3266 | } | |
3267 | ||
182446d0 TH |
3268 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3269 | struct cftype *cft) | |
a7885eb8 | 3270 | { |
182446d0 | 3271 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3272 | |
1f4c025b | 3273 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3274 | } |
3275 | ||
182446d0 TH |
3276 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3277 | struct cftype *cft, u64 val) | |
a7885eb8 | 3278 | { |
182446d0 | 3279 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3280 | |
3dae7fec | 3281 | if (val > 100) |
a7885eb8 KM |
3282 | return -EINVAL; |
3283 | ||
14208b0e | 3284 | if (css->parent) |
3dae7fec JW |
3285 | memcg->swappiness = val; |
3286 | else | |
3287 | vm_swappiness = val; | |
068b38c1 | 3288 | |
a7885eb8 KM |
3289 | return 0; |
3290 | } | |
3291 | ||
2e72b634 KS |
3292 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3293 | { | |
3294 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3295 | unsigned long usage; |
2e72b634 KS |
3296 | int i; |
3297 | ||
3298 | rcu_read_lock(); | |
3299 | if (!swap) | |
2c488db2 | 3300 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3301 | else |
2c488db2 | 3302 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3303 | |
3304 | if (!t) | |
3305 | goto unlock; | |
3306 | ||
ce00a967 | 3307 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3308 | |
3309 | /* | |
748dad36 | 3310 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3311 | * If it's not true, a threshold was crossed after last |
3312 | * call of __mem_cgroup_threshold(). | |
3313 | */ | |
5407a562 | 3314 | i = t->current_threshold; |
2e72b634 KS |
3315 | |
3316 | /* | |
3317 | * Iterate backward over array of thresholds starting from | |
3318 | * current_threshold and check if a threshold is crossed. | |
3319 | * If none of thresholds below usage is crossed, we read | |
3320 | * only one element of the array here. | |
3321 | */ | |
3322 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3323 | eventfd_signal(t->entries[i].eventfd, 1); | |
3324 | ||
3325 | /* i = current_threshold + 1 */ | |
3326 | i++; | |
3327 | ||
3328 | /* | |
3329 | * Iterate forward over array of thresholds starting from | |
3330 | * current_threshold+1 and check if a threshold is crossed. | |
3331 | * If none of thresholds above usage is crossed, we read | |
3332 | * only one element of the array here. | |
3333 | */ | |
3334 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3335 | eventfd_signal(t->entries[i].eventfd, 1); | |
3336 | ||
3337 | /* Update current_threshold */ | |
5407a562 | 3338 | t->current_threshold = i - 1; |
2e72b634 KS |
3339 | unlock: |
3340 | rcu_read_unlock(); | |
3341 | } | |
3342 | ||
3343 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3344 | { | |
ad4ca5f4 KS |
3345 | while (memcg) { |
3346 | __mem_cgroup_threshold(memcg, false); | |
3347 | if (do_swap_account) | |
3348 | __mem_cgroup_threshold(memcg, true); | |
3349 | ||
3350 | memcg = parent_mem_cgroup(memcg); | |
3351 | } | |
2e72b634 KS |
3352 | } |
3353 | ||
3354 | static int compare_thresholds(const void *a, const void *b) | |
3355 | { | |
3356 | const struct mem_cgroup_threshold *_a = a; | |
3357 | const struct mem_cgroup_threshold *_b = b; | |
3358 | ||
2bff24a3 GT |
3359 | if (_a->threshold > _b->threshold) |
3360 | return 1; | |
3361 | ||
3362 | if (_a->threshold < _b->threshold) | |
3363 | return -1; | |
3364 | ||
3365 | return 0; | |
2e72b634 KS |
3366 | } |
3367 | ||
c0ff4b85 | 3368 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3369 | { |
3370 | struct mem_cgroup_eventfd_list *ev; | |
3371 | ||
2bcf2e92 MH |
3372 | spin_lock(&memcg_oom_lock); |
3373 | ||
c0ff4b85 | 3374 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3375 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3376 | |
3377 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3378 | return 0; |
3379 | } | |
3380 | ||
c0ff4b85 | 3381 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3382 | { |
7d74b06f KH |
3383 | struct mem_cgroup *iter; |
3384 | ||
c0ff4b85 | 3385 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3386 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3387 | } |
3388 | ||
59b6f873 | 3389 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3390 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3391 | { |
2c488db2 KS |
3392 | struct mem_cgroup_thresholds *thresholds; |
3393 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3394 | unsigned long threshold; |
3395 | unsigned long usage; | |
2c488db2 | 3396 | int i, size, ret; |
2e72b634 | 3397 | |
650c5e56 | 3398 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
3399 | if (ret) |
3400 | return ret; | |
3401 | ||
3402 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3403 | |
05b84301 | 3404 | if (type == _MEM) { |
2c488db2 | 3405 | thresholds = &memcg->thresholds; |
ce00a967 | 3406 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3407 | } else if (type == _MEMSWAP) { |
2c488db2 | 3408 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3409 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3410 | } else |
2e72b634 KS |
3411 | BUG(); |
3412 | ||
2e72b634 | 3413 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3414 | if (thresholds->primary) |
2e72b634 KS |
3415 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3416 | ||
2c488db2 | 3417 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3418 | |
3419 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 3420 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 3421 | GFP_KERNEL); |
2c488db2 | 3422 | if (!new) { |
2e72b634 KS |
3423 | ret = -ENOMEM; |
3424 | goto unlock; | |
3425 | } | |
2c488db2 | 3426 | new->size = size; |
2e72b634 KS |
3427 | |
3428 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3429 | if (thresholds->primary) { |
3430 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3431 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3432 | } |
3433 | ||
2e72b634 | 3434 | /* Add new threshold */ |
2c488db2 KS |
3435 | new->entries[size - 1].eventfd = eventfd; |
3436 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3437 | |
3438 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3439 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3440 | compare_thresholds, NULL); |
3441 | ||
3442 | /* Find current threshold */ | |
2c488db2 | 3443 | new->current_threshold = -1; |
2e72b634 | 3444 | for (i = 0; i < size; i++) { |
748dad36 | 3445 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3446 | /* |
2c488db2 KS |
3447 | * new->current_threshold will not be used until |
3448 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3449 | * it here. |
3450 | */ | |
2c488db2 | 3451 | ++new->current_threshold; |
748dad36 SZ |
3452 | } else |
3453 | break; | |
2e72b634 KS |
3454 | } |
3455 | ||
2c488db2 KS |
3456 | /* Free old spare buffer and save old primary buffer as spare */ |
3457 | kfree(thresholds->spare); | |
3458 | thresholds->spare = thresholds->primary; | |
3459 | ||
3460 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3461 | |
907860ed | 3462 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3463 | synchronize_rcu(); |
3464 | ||
2e72b634 KS |
3465 | unlock: |
3466 | mutex_unlock(&memcg->thresholds_lock); | |
3467 | ||
3468 | return ret; | |
3469 | } | |
3470 | ||
59b6f873 | 3471 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3472 | struct eventfd_ctx *eventfd, const char *args) |
3473 | { | |
59b6f873 | 3474 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3475 | } |
3476 | ||
59b6f873 | 3477 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3478 | struct eventfd_ctx *eventfd, const char *args) |
3479 | { | |
59b6f873 | 3480 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3481 | } |
3482 | ||
59b6f873 | 3483 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3484 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3485 | { |
2c488db2 KS |
3486 | struct mem_cgroup_thresholds *thresholds; |
3487 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3488 | unsigned long usage; |
2c488db2 | 3489 | int i, j, size; |
2e72b634 KS |
3490 | |
3491 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3492 | |
3493 | if (type == _MEM) { | |
2c488db2 | 3494 | thresholds = &memcg->thresholds; |
ce00a967 | 3495 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3496 | } else if (type == _MEMSWAP) { |
2c488db2 | 3497 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3498 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3499 | } else |
2e72b634 KS |
3500 | BUG(); |
3501 | ||
371528ca AV |
3502 | if (!thresholds->primary) |
3503 | goto unlock; | |
3504 | ||
2e72b634 KS |
3505 | /* Check if a threshold crossed before removing */ |
3506 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3507 | ||
3508 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3509 | size = 0; |
3510 | for (i = 0; i < thresholds->primary->size; i++) { | |
3511 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3512 | size++; |
3513 | } | |
3514 | ||
2c488db2 | 3515 | new = thresholds->spare; |
907860ed | 3516 | |
2e72b634 KS |
3517 | /* Set thresholds array to NULL if we don't have thresholds */ |
3518 | if (!size) { | |
2c488db2 KS |
3519 | kfree(new); |
3520 | new = NULL; | |
907860ed | 3521 | goto swap_buffers; |
2e72b634 KS |
3522 | } |
3523 | ||
2c488db2 | 3524 | new->size = size; |
2e72b634 KS |
3525 | |
3526 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3527 | new->current_threshold = -1; |
3528 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3529 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3530 | continue; |
3531 | ||
2c488db2 | 3532 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3533 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3534 | /* |
2c488db2 | 3535 | * new->current_threshold will not be used |
2e72b634 KS |
3536 | * until rcu_assign_pointer(), so it's safe to increment |
3537 | * it here. | |
3538 | */ | |
2c488db2 | 3539 | ++new->current_threshold; |
2e72b634 KS |
3540 | } |
3541 | j++; | |
3542 | } | |
3543 | ||
907860ed | 3544 | swap_buffers: |
2c488db2 KS |
3545 | /* Swap primary and spare array */ |
3546 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
3547 | /* If all events are unregistered, free the spare array */ |
3548 | if (!new) { | |
3549 | kfree(thresholds->spare); | |
3550 | thresholds->spare = NULL; | |
3551 | } | |
3552 | ||
2c488db2 | 3553 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3554 | |
907860ed | 3555 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3556 | synchronize_rcu(); |
371528ca | 3557 | unlock: |
2e72b634 | 3558 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3559 | } |
c1e862c1 | 3560 | |
59b6f873 | 3561 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3562 | struct eventfd_ctx *eventfd) |
3563 | { | |
59b6f873 | 3564 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
3565 | } |
3566 | ||
59b6f873 | 3567 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3568 | struct eventfd_ctx *eventfd) |
3569 | { | |
59b6f873 | 3570 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
3571 | } |
3572 | ||
59b6f873 | 3573 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3574 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 3575 | { |
9490ff27 | 3576 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 3577 | |
9490ff27 KH |
3578 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
3579 | if (!event) | |
3580 | return -ENOMEM; | |
3581 | ||
1af8efe9 | 3582 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
3583 | |
3584 | event->eventfd = eventfd; | |
3585 | list_add(&event->list, &memcg->oom_notify); | |
3586 | ||
3587 | /* already in OOM ? */ | |
c2b42d3c | 3588 | if (memcg->under_oom) |
9490ff27 | 3589 | eventfd_signal(eventfd, 1); |
1af8efe9 | 3590 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3591 | |
3592 | return 0; | |
3593 | } | |
3594 | ||
59b6f873 | 3595 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3596 | struct eventfd_ctx *eventfd) |
9490ff27 | 3597 | { |
9490ff27 | 3598 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 3599 | |
1af8efe9 | 3600 | spin_lock(&memcg_oom_lock); |
9490ff27 | 3601 | |
c0ff4b85 | 3602 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
3603 | if (ev->eventfd == eventfd) { |
3604 | list_del(&ev->list); | |
3605 | kfree(ev); | |
3606 | } | |
3607 | } | |
3608 | ||
1af8efe9 | 3609 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3610 | } |
3611 | ||
2da8ca82 | 3612 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 3613 | { |
2da8ca82 | 3614 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 3615 | |
791badbd | 3616 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 3617 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
3c11ecf4 KH |
3618 | return 0; |
3619 | } | |
3620 | ||
182446d0 | 3621 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
3622 | struct cftype *cft, u64 val) |
3623 | { | |
182446d0 | 3624 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
3625 | |
3626 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 3627 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
3628 | return -EINVAL; |
3629 | ||
c0ff4b85 | 3630 | memcg->oom_kill_disable = val; |
4d845ebf | 3631 | if (!val) |
c0ff4b85 | 3632 | memcg_oom_recover(memcg); |
3dae7fec | 3633 | |
3c11ecf4 KH |
3634 | return 0; |
3635 | } | |
3636 | ||
c255a458 | 3637 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 3638 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 3639 | { |
55007d84 GC |
3640 | int ret; |
3641 | ||
55007d84 GC |
3642 | ret = memcg_propagate_kmem(memcg); |
3643 | if (ret) | |
3644 | return ret; | |
2633d7a0 | 3645 | |
baac50bb | 3646 | return tcp_init_cgroup(memcg, ss); |
573b400d | 3647 | } |
e5671dfa | 3648 | |
2a4db7eb VD |
3649 | static void memcg_deactivate_kmem(struct mem_cgroup *memcg) |
3650 | { | |
2788cf0c VD |
3651 | struct cgroup_subsys_state *css; |
3652 | struct mem_cgroup *parent, *child; | |
3653 | int kmemcg_id; | |
3654 | ||
2a4db7eb VD |
3655 | if (!memcg->kmem_acct_active) |
3656 | return; | |
3657 | ||
3658 | /* | |
3659 | * Clear the 'active' flag before clearing memcg_caches arrays entries. | |
3660 | * Since we take the slab_mutex in memcg_deactivate_kmem_caches(), it | |
3661 | * guarantees no cache will be created for this cgroup after we are | |
3662 | * done (see memcg_create_kmem_cache()). | |
3663 | */ | |
3664 | memcg->kmem_acct_active = false; | |
3665 | ||
3666 | memcg_deactivate_kmem_caches(memcg); | |
2788cf0c VD |
3667 | |
3668 | kmemcg_id = memcg->kmemcg_id; | |
3669 | BUG_ON(kmemcg_id < 0); | |
3670 | ||
3671 | parent = parent_mem_cgroup(memcg); | |
3672 | if (!parent) | |
3673 | parent = root_mem_cgroup; | |
3674 | ||
3675 | /* | |
3676 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3677 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3678 | * to ones of the parent. After we have finished, all list_lrus | |
3679 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3680 | * ordering is imposed by list_lru_node->lock taken by | |
3681 | * memcg_drain_all_list_lrus(). | |
3682 | */ | |
3683 | css_for_each_descendant_pre(css, &memcg->css) { | |
3684 | child = mem_cgroup_from_css(css); | |
3685 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3686 | child->kmemcg_id = parent->kmemcg_id; | |
3687 | if (!memcg->use_hierarchy) | |
3688 | break; | |
3689 | } | |
3690 | memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id); | |
3691 | ||
3692 | memcg_free_cache_id(kmemcg_id); | |
2a4db7eb VD |
3693 | } |
3694 | ||
10d5ebf4 | 3695 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 3696 | { |
f48b80a5 VD |
3697 | if (memcg->kmem_acct_activated) { |
3698 | memcg_destroy_kmem_caches(memcg); | |
3699 | static_key_slow_dec(&memcg_kmem_enabled_key); | |
3700 | WARN_ON(page_counter_read(&memcg->kmem)); | |
3701 | } | |
baac50bb | 3702 | tcp_destroy_cgroup(memcg); |
10d5ebf4 | 3703 | } |
e5671dfa | 3704 | #else |
cbe128e3 | 3705 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
3706 | { |
3707 | return 0; | |
3708 | } | |
d1a4c0b3 | 3709 | |
2a4db7eb VD |
3710 | static void memcg_deactivate_kmem(struct mem_cgroup *memcg) |
3711 | { | |
3712 | } | |
3713 | ||
10d5ebf4 LZ |
3714 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
3715 | { | |
3716 | } | |
e5671dfa GC |
3717 | #endif |
3718 | ||
52ebea74 TH |
3719 | #ifdef CONFIG_CGROUP_WRITEBACK |
3720 | ||
3721 | struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg) | |
3722 | { | |
3723 | return &memcg->cgwb_list; | |
3724 | } | |
3725 | ||
841710aa TH |
3726 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
3727 | { | |
3728 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
3729 | } | |
3730 | ||
3731 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3732 | { | |
3733 | wb_domain_exit(&memcg->cgwb_domain); | |
3734 | } | |
3735 | ||
2529bb3a TH |
3736 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3737 | { | |
3738 | wb_domain_size_changed(&memcg->cgwb_domain); | |
3739 | } | |
3740 | ||
841710aa TH |
3741 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
3742 | { | |
3743 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3744 | ||
3745 | if (!memcg->css.parent) | |
3746 | return NULL; | |
3747 | ||
3748 | return &memcg->cgwb_domain; | |
3749 | } | |
3750 | ||
c2aa723a TH |
3751 | /** |
3752 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
3753 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
3754 | * @pfilepages: out parameter for number of file pages |
3755 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
3756 | * @pdirty: out parameter for number of dirty pages |
3757 | * @pwriteback: out parameter for number of pages under writeback | |
3758 | * | |
c5edf9cd TH |
3759 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
3760 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
3761 | * is a bit more involved. | |
c2aa723a | 3762 | * |
c5edf9cd TH |
3763 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
3764 | * headroom is calculated as the lowest headroom of itself and the | |
3765 | * ancestors. Note that this doesn't consider the actual amount of | |
3766 | * available memory in the system. The caller should further cap | |
3767 | * *@pheadroom accordingly. | |
c2aa723a | 3768 | */ |
c5edf9cd TH |
3769 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
3770 | unsigned long *pheadroom, unsigned long *pdirty, | |
3771 | unsigned long *pwriteback) | |
c2aa723a TH |
3772 | { |
3773 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3774 | struct mem_cgroup *parent; | |
c2aa723a TH |
3775 | |
3776 | *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY); | |
3777 | ||
3778 | /* this should eventually include NR_UNSTABLE_NFS */ | |
3779 | *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK); | |
c5edf9cd TH |
3780 | *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) | |
3781 | (1 << LRU_ACTIVE_FILE)); | |
3782 | *pheadroom = PAGE_COUNTER_MAX; | |
c2aa723a | 3783 | |
c2aa723a TH |
3784 | while ((parent = parent_mem_cgroup(memcg))) { |
3785 | unsigned long ceiling = min(memcg->memory.limit, memcg->high); | |
3786 | unsigned long used = page_counter_read(&memcg->memory); | |
3787 | ||
c5edf9cd | 3788 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
3789 | memcg = parent; |
3790 | } | |
c2aa723a TH |
3791 | } |
3792 | ||
841710aa TH |
3793 | #else /* CONFIG_CGROUP_WRITEBACK */ |
3794 | ||
3795 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
3796 | { | |
3797 | return 0; | |
3798 | } | |
3799 | ||
3800 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3801 | { | |
3802 | } | |
3803 | ||
2529bb3a TH |
3804 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3805 | { | |
3806 | } | |
3807 | ||
52ebea74 TH |
3808 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
3809 | ||
3bc942f3 TH |
3810 | /* |
3811 | * DO NOT USE IN NEW FILES. | |
3812 | * | |
3813 | * "cgroup.event_control" implementation. | |
3814 | * | |
3815 | * This is way over-engineered. It tries to support fully configurable | |
3816 | * events for each user. Such level of flexibility is completely | |
3817 | * unnecessary especially in the light of the planned unified hierarchy. | |
3818 | * | |
3819 | * Please deprecate this and replace with something simpler if at all | |
3820 | * possible. | |
3821 | */ | |
3822 | ||
79bd9814 TH |
3823 | /* |
3824 | * Unregister event and free resources. | |
3825 | * | |
3826 | * Gets called from workqueue. | |
3827 | */ | |
3bc942f3 | 3828 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 3829 | { |
3bc942f3 TH |
3830 | struct mem_cgroup_event *event = |
3831 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 3832 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3833 | |
3834 | remove_wait_queue(event->wqh, &event->wait); | |
3835 | ||
59b6f873 | 3836 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
3837 | |
3838 | /* Notify userspace the event is going away. */ | |
3839 | eventfd_signal(event->eventfd, 1); | |
3840 | ||
3841 | eventfd_ctx_put(event->eventfd); | |
3842 | kfree(event); | |
59b6f873 | 3843 | css_put(&memcg->css); |
79bd9814 TH |
3844 | } |
3845 | ||
3846 | /* | |
3847 | * Gets called on POLLHUP on eventfd when user closes it. | |
3848 | * | |
3849 | * Called with wqh->lock held and interrupts disabled. | |
3850 | */ | |
3bc942f3 TH |
3851 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
3852 | int sync, void *key) | |
79bd9814 | 3853 | { |
3bc942f3 TH |
3854 | struct mem_cgroup_event *event = |
3855 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 3856 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3857 | unsigned long flags = (unsigned long)key; |
3858 | ||
3859 | if (flags & POLLHUP) { | |
3860 | /* | |
3861 | * If the event has been detached at cgroup removal, we | |
3862 | * can simply return knowing the other side will cleanup | |
3863 | * for us. | |
3864 | * | |
3865 | * We can't race against event freeing since the other | |
3866 | * side will require wqh->lock via remove_wait_queue(), | |
3867 | * which we hold. | |
3868 | */ | |
fba94807 | 3869 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
3870 | if (!list_empty(&event->list)) { |
3871 | list_del_init(&event->list); | |
3872 | /* | |
3873 | * We are in atomic context, but cgroup_event_remove() | |
3874 | * may sleep, so we have to call it in workqueue. | |
3875 | */ | |
3876 | schedule_work(&event->remove); | |
3877 | } | |
fba94807 | 3878 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
3879 | } |
3880 | ||
3881 | return 0; | |
3882 | } | |
3883 | ||
3bc942f3 | 3884 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
3885 | wait_queue_head_t *wqh, poll_table *pt) |
3886 | { | |
3bc942f3 TH |
3887 | struct mem_cgroup_event *event = |
3888 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
3889 | |
3890 | event->wqh = wqh; | |
3891 | add_wait_queue(wqh, &event->wait); | |
3892 | } | |
3893 | ||
3894 | /* | |
3bc942f3 TH |
3895 | * DO NOT USE IN NEW FILES. |
3896 | * | |
79bd9814 TH |
3897 | * Parse input and register new cgroup event handler. |
3898 | * | |
3899 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
3900 | * Interpretation of args is defined by control file implementation. | |
3901 | */ | |
451af504 TH |
3902 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
3903 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 3904 | { |
451af504 | 3905 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 3906 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 3907 | struct mem_cgroup_event *event; |
79bd9814 TH |
3908 | struct cgroup_subsys_state *cfile_css; |
3909 | unsigned int efd, cfd; | |
3910 | struct fd efile; | |
3911 | struct fd cfile; | |
fba94807 | 3912 | const char *name; |
79bd9814 TH |
3913 | char *endp; |
3914 | int ret; | |
3915 | ||
451af504 TH |
3916 | buf = strstrip(buf); |
3917 | ||
3918 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
3919 | if (*endp != ' ') |
3920 | return -EINVAL; | |
451af504 | 3921 | buf = endp + 1; |
79bd9814 | 3922 | |
451af504 | 3923 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
3924 | if ((*endp != ' ') && (*endp != '\0')) |
3925 | return -EINVAL; | |
451af504 | 3926 | buf = endp + 1; |
79bd9814 TH |
3927 | |
3928 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
3929 | if (!event) | |
3930 | return -ENOMEM; | |
3931 | ||
59b6f873 | 3932 | event->memcg = memcg; |
79bd9814 | 3933 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
3934 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
3935 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
3936 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
3937 | |
3938 | efile = fdget(efd); | |
3939 | if (!efile.file) { | |
3940 | ret = -EBADF; | |
3941 | goto out_kfree; | |
3942 | } | |
3943 | ||
3944 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
3945 | if (IS_ERR(event->eventfd)) { | |
3946 | ret = PTR_ERR(event->eventfd); | |
3947 | goto out_put_efile; | |
3948 | } | |
3949 | ||
3950 | cfile = fdget(cfd); | |
3951 | if (!cfile.file) { | |
3952 | ret = -EBADF; | |
3953 | goto out_put_eventfd; | |
3954 | } | |
3955 | ||
3956 | /* the process need read permission on control file */ | |
3957 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
3958 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
3959 | if (ret < 0) | |
3960 | goto out_put_cfile; | |
3961 | ||
fba94807 TH |
3962 | /* |
3963 | * Determine the event callbacks and set them in @event. This used | |
3964 | * to be done via struct cftype but cgroup core no longer knows | |
3965 | * about these events. The following is crude but the whole thing | |
3966 | * is for compatibility anyway. | |
3bc942f3 TH |
3967 | * |
3968 | * DO NOT ADD NEW FILES. | |
fba94807 | 3969 | */ |
b583043e | 3970 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
3971 | |
3972 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
3973 | event->register_event = mem_cgroup_usage_register_event; | |
3974 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
3975 | } else if (!strcmp(name, "memory.oom_control")) { | |
3976 | event->register_event = mem_cgroup_oom_register_event; | |
3977 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
3978 | } else if (!strcmp(name, "memory.pressure_level")) { | |
3979 | event->register_event = vmpressure_register_event; | |
3980 | event->unregister_event = vmpressure_unregister_event; | |
3981 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
3982 | event->register_event = memsw_cgroup_usage_register_event; |
3983 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
3984 | } else { |
3985 | ret = -EINVAL; | |
3986 | goto out_put_cfile; | |
3987 | } | |
3988 | ||
79bd9814 | 3989 | /* |
b5557c4c TH |
3990 | * Verify @cfile should belong to @css. Also, remaining events are |
3991 | * automatically removed on cgroup destruction but the removal is | |
3992 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 3993 | */ |
b583043e | 3994 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 3995 | &memory_cgrp_subsys); |
79bd9814 | 3996 | ret = -EINVAL; |
5a17f543 | 3997 | if (IS_ERR(cfile_css)) |
79bd9814 | 3998 | goto out_put_cfile; |
5a17f543 TH |
3999 | if (cfile_css != css) { |
4000 | css_put(cfile_css); | |
79bd9814 | 4001 | goto out_put_cfile; |
5a17f543 | 4002 | } |
79bd9814 | 4003 | |
451af504 | 4004 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4005 | if (ret) |
4006 | goto out_put_css; | |
4007 | ||
4008 | efile.file->f_op->poll(efile.file, &event->pt); | |
4009 | ||
fba94807 TH |
4010 | spin_lock(&memcg->event_list_lock); |
4011 | list_add(&event->list, &memcg->event_list); | |
4012 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4013 | |
4014 | fdput(cfile); | |
4015 | fdput(efile); | |
4016 | ||
451af504 | 4017 | return nbytes; |
79bd9814 TH |
4018 | |
4019 | out_put_css: | |
b5557c4c | 4020 | css_put(css); |
79bd9814 TH |
4021 | out_put_cfile: |
4022 | fdput(cfile); | |
4023 | out_put_eventfd: | |
4024 | eventfd_ctx_put(event->eventfd); | |
4025 | out_put_efile: | |
4026 | fdput(efile); | |
4027 | out_kfree: | |
4028 | kfree(event); | |
4029 | ||
4030 | return ret; | |
4031 | } | |
4032 | ||
241994ed | 4033 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4034 | { |
0eea1030 | 4035 | .name = "usage_in_bytes", |
8c7c6e34 | 4036 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4037 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4038 | }, |
c84872e1 PE |
4039 | { |
4040 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4041 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4042 | .write = mem_cgroup_reset, |
791badbd | 4043 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4044 | }, |
8cdea7c0 | 4045 | { |
0eea1030 | 4046 | .name = "limit_in_bytes", |
8c7c6e34 | 4047 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4048 | .write = mem_cgroup_write, |
791badbd | 4049 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4050 | }, |
296c81d8 BS |
4051 | { |
4052 | .name = "soft_limit_in_bytes", | |
4053 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4054 | .write = mem_cgroup_write, |
791badbd | 4055 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4056 | }, |
8cdea7c0 BS |
4057 | { |
4058 | .name = "failcnt", | |
8c7c6e34 | 4059 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4060 | .write = mem_cgroup_reset, |
791badbd | 4061 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4062 | }, |
d2ceb9b7 KH |
4063 | { |
4064 | .name = "stat", | |
2da8ca82 | 4065 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4066 | }, |
c1e862c1 KH |
4067 | { |
4068 | .name = "force_empty", | |
6770c64e | 4069 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4070 | }, |
18f59ea7 BS |
4071 | { |
4072 | .name = "use_hierarchy", | |
4073 | .write_u64 = mem_cgroup_hierarchy_write, | |
4074 | .read_u64 = mem_cgroup_hierarchy_read, | |
4075 | }, | |
79bd9814 | 4076 | { |
3bc942f3 | 4077 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4078 | .write = memcg_write_event_control, |
7dbdb199 | 4079 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 4080 | }, |
a7885eb8 KM |
4081 | { |
4082 | .name = "swappiness", | |
4083 | .read_u64 = mem_cgroup_swappiness_read, | |
4084 | .write_u64 = mem_cgroup_swappiness_write, | |
4085 | }, | |
7dc74be0 DN |
4086 | { |
4087 | .name = "move_charge_at_immigrate", | |
4088 | .read_u64 = mem_cgroup_move_charge_read, | |
4089 | .write_u64 = mem_cgroup_move_charge_write, | |
4090 | }, | |
9490ff27 KH |
4091 | { |
4092 | .name = "oom_control", | |
2da8ca82 | 4093 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4094 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4095 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4096 | }, | |
70ddf637 AV |
4097 | { |
4098 | .name = "pressure_level", | |
70ddf637 | 4099 | }, |
406eb0c9 YH |
4100 | #ifdef CONFIG_NUMA |
4101 | { | |
4102 | .name = "numa_stat", | |
2da8ca82 | 4103 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4104 | }, |
4105 | #endif | |
510fc4e1 GC |
4106 | #ifdef CONFIG_MEMCG_KMEM |
4107 | { | |
4108 | .name = "kmem.limit_in_bytes", | |
4109 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4110 | .write = mem_cgroup_write, |
791badbd | 4111 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4112 | }, |
4113 | { | |
4114 | .name = "kmem.usage_in_bytes", | |
4115 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4116 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4117 | }, |
4118 | { | |
4119 | .name = "kmem.failcnt", | |
4120 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4121 | .write = mem_cgroup_reset, |
791badbd | 4122 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4123 | }, |
4124 | { | |
4125 | .name = "kmem.max_usage_in_bytes", | |
4126 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4127 | .write = mem_cgroup_reset, |
791badbd | 4128 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4129 | }, |
749c5415 GC |
4130 | #ifdef CONFIG_SLABINFO |
4131 | { | |
4132 | .name = "kmem.slabinfo", | |
b047501c VD |
4133 | .seq_start = slab_start, |
4134 | .seq_next = slab_next, | |
4135 | .seq_stop = slab_stop, | |
4136 | .seq_show = memcg_slab_show, | |
749c5415 GC |
4137 | }, |
4138 | #endif | |
8c7c6e34 | 4139 | #endif |
6bc10349 | 4140 | { }, /* terminate */ |
af36f906 | 4141 | }; |
8c7c6e34 | 4142 | |
c0ff4b85 | 4143 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4144 | { |
4145 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4146 | struct mem_cgroup_per_zone *mz; |
41e3355d | 4147 | int zone, tmp = node; |
1ecaab2b KH |
4148 | /* |
4149 | * This routine is called against possible nodes. | |
4150 | * But it's BUG to call kmalloc() against offline node. | |
4151 | * | |
4152 | * TODO: this routine can waste much memory for nodes which will | |
4153 | * never be onlined. It's better to use memory hotplug callback | |
4154 | * function. | |
4155 | */ | |
41e3355d KH |
4156 | if (!node_state(node, N_NORMAL_MEMORY)) |
4157 | tmp = -1; | |
17295c88 | 4158 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4159 | if (!pn) |
4160 | return 1; | |
1ecaab2b | 4161 | |
1ecaab2b KH |
4162 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4163 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 4164 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
4165 | mz->usage_in_excess = 0; |
4166 | mz->on_tree = false; | |
d79154bb | 4167 | mz->memcg = memcg; |
1ecaab2b | 4168 | } |
54f72fe0 | 4169 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4170 | return 0; |
4171 | } | |
4172 | ||
c0ff4b85 | 4173 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4174 | { |
54f72fe0 | 4175 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
4176 | } |
4177 | ||
33327948 KH |
4178 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4179 | { | |
d79154bb | 4180 | struct mem_cgroup *memcg; |
8ff69e2c | 4181 | size_t size; |
33327948 | 4182 | |
8ff69e2c VD |
4183 | size = sizeof(struct mem_cgroup); |
4184 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
33327948 | 4185 | |
8ff69e2c | 4186 | memcg = kzalloc(size, GFP_KERNEL); |
d79154bb | 4187 | if (!memcg) |
e7bbcdf3 DC |
4188 | return NULL; |
4189 | ||
d79154bb HD |
4190 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
4191 | if (!memcg->stat) | |
d2e61b8d | 4192 | goto out_free; |
841710aa TH |
4193 | |
4194 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) | |
4195 | goto out_free_stat; | |
4196 | ||
d79154bb | 4197 | return memcg; |
d2e61b8d | 4198 | |
841710aa TH |
4199 | out_free_stat: |
4200 | free_percpu(memcg->stat); | |
d2e61b8d | 4201 | out_free: |
8ff69e2c | 4202 | kfree(memcg); |
d2e61b8d | 4203 | return NULL; |
33327948 KH |
4204 | } |
4205 | ||
59927fb9 | 4206 | /* |
c8b2a36f GC |
4207 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
4208 | * (scanning all at force_empty is too costly...) | |
4209 | * | |
4210 | * Instead of clearing all references at force_empty, we remember | |
4211 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4212 | * it goes down to 0. | |
4213 | * | |
4214 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 4215 | */ |
c8b2a36f GC |
4216 | |
4217 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 4218 | { |
c8b2a36f | 4219 | int node; |
59927fb9 | 4220 | |
bb4cc1a8 | 4221 | mem_cgroup_remove_from_trees(memcg); |
c8b2a36f GC |
4222 | |
4223 | for_each_node(node) | |
4224 | free_mem_cgroup_per_zone_info(memcg, node); | |
4225 | ||
4226 | free_percpu(memcg->stat); | |
841710aa | 4227 | memcg_wb_domain_exit(memcg); |
8ff69e2c | 4228 | kfree(memcg); |
59927fb9 | 4229 | } |
3afe36b1 | 4230 | |
7bcc1bb1 DN |
4231 | /* |
4232 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4233 | */ | |
e1aab161 | 4234 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 4235 | { |
3e32cb2e | 4236 | if (!memcg->memory.parent) |
7bcc1bb1 | 4237 | return NULL; |
3e32cb2e | 4238 | return mem_cgroup_from_counter(memcg->memory.parent, memory); |
7bcc1bb1 | 4239 | } |
e1aab161 | 4240 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 4241 | |
0eb253e2 | 4242 | static struct cgroup_subsys_state * __ref |
eb95419b | 4243 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 4244 | { |
d142e3e6 | 4245 | struct mem_cgroup *memcg; |
04046e1a | 4246 | long error = -ENOMEM; |
6d12e2d8 | 4247 | int node; |
8cdea7c0 | 4248 | |
c0ff4b85 R |
4249 | memcg = mem_cgroup_alloc(); |
4250 | if (!memcg) | |
04046e1a | 4251 | return ERR_PTR(error); |
78fb7466 | 4252 | |
3ed28fa1 | 4253 | for_each_node(node) |
c0ff4b85 | 4254 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 4255 | goto free_out; |
f64c3f54 | 4256 | |
c077719b | 4257 | /* root ? */ |
eb95419b | 4258 | if (parent_css == NULL) { |
a41c58a6 | 4259 | root_mem_cgroup = memcg; |
3e32cb2e | 4260 | page_counter_init(&memcg->memory, NULL); |
241994ed | 4261 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4262 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4263 | page_counter_init(&memcg->memsw, NULL); |
4264 | page_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 4265 | } |
28dbc4b6 | 4266 | |
d142e3e6 GC |
4267 | memcg->last_scanned_node = MAX_NUMNODES; |
4268 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4269 | memcg->move_charge_at_immigrate = 0; |
4270 | mutex_init(&memcg->thresholds_lock); | |
4271 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4272 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4273 | INIT_LIST_HEAD(&memcg->event_list); |
4274 | spin_lock_init(&memcg->event_list_lock); | |
900a38f0 VD |
4275 | #ifdef CONFIG_MEMCG_KMEM |
4276 | memcg->kmemcg_id = -1; | |
900a38f0 | 4277 | #endif |
52ebea74 TH |
4278 | #ifdef CONFIG_CGROUP_WRITEBACK |
4279 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
4280 | #endif | |
d142e3e6 GC |
4281 | return &memcg->css; |
4282 | ||
4283 | free_out: | |
4284 | __mem_cgroup_free(memcg); | |
4285 | return ERR_PTR(error); | |
4286 | } | |
4287 | ||
4288 | static int | |
eb95419b | 4289 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 4290 | { |
eb95419b | 4291 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 4292 | struct mem_cgroup *parent = mem_cgroup_from_css(css->parent); |
2f7dd7a4 | 4293 | int ret; |
d142e3e6 | 4294 | |
15a4c835 | 4295 | if (css->id > MEM_CGROUP_ID_MAX) |
4219b2da LZ |
4296 | return -ENOSPC; |
4297 | ||
63876986 | 4298 | if (!parent) |
d142e3e6 GC |
4299 | return 0; |
4300 | ||
0999821b | 4301 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
4302 | |
4303 | memcg->use_hierarchy = parent->use_hierarchy; | |
4304 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4305 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4306 | ||
4307 | if (parent->use_hierarchy) { | |
3e32cb2e | 4308 | page_counter_init(&memcg->memory, &parent->memory); |
241994ed | 4309 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4310 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4311 | page_counter_init(&memcg->memsw, &parent->memsw); |
4312 | page_counter_init(&memcg->kmem, &parent->kmem); | |
55007d84 | 4313 | |
7bcc1bb1 | 4314 | /* |
8d76a979 LZ |
4315 | * No need to take a reference to the parent because cgroup |
4316 | * core guarantees its existence. | |
7bcc1bb1 | 4317 | */ |
18f59ea7 | 4318 | } else { |
3e32cb2e | 4319 | page_counter_init(&memcg->memory, NULL); |
241994ed | 4320 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4321 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4322 | page_counter_init(&memcg->memsw, NULL); |
4323 | page_counter_init(&memcg->kmem, NULL); | |
8c7f6edb TH |
4324 | /* |
4325 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4326 | * much sense so let cgroup subsystem know about this | |
4327 | * unfortunate state in our controller. | |
4328 | */ | |
d142e3e6 | 4329 | if (parent != root_mem_cgroup) |
073219e9 | 4330 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4331 | } |
0999821b | 4332 | mutex_unlock(&memcg_create_mutex); |
d6441637 | 4333 | |
2f7dd7a4 JW |
4334 | ret = memcg_init_kmem(memcg, &memory_cgrp_subsys); |
4335 | if (ret) | |
4336 | return ret; | |
4337 | ||
4338 | /* | |
4339 | * Make sure the memcg is initialized: mem_cgroup_iter() | |
4340 | * orders reading memcg->initialized against its callers | |
4341 | * reading the memcg members. | |
4342 | */ | |
4343 | smp_store_release(&memcg->initialized, 1); | |
4344 | ||
4345 | return 0; | |
8cdea7c0 BS |
4346 | } |
4347 | ||
eb95419b | 4348 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4349 | { |
eb95419b | 4350 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4351 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4352 | |
4353 | /* | |
4354 | * Unregister events and notify userspace. | |
4355 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4356 | * directory to avoid race between userspace and kernelspace. | |
4357 | */ | |
fba94807 TH |
4358 | spin_lock(&memcg->event_list_lock); |
4359 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4360 | list_del_init(&event->list); |
4361 | schedule_work(&event->remove); | |
4362 | } | |
fba94807 | 4363 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4364 | |
33cb876e | 4365 | vmpressure_cleanup(&memcg->vmpressure); |
2a4db7eb VD |
4366 | |
4367 | memcg_deactivate_kmem(memcg); | |
52ebea74 TH |
4368 | |
4369 | wb_memcg_offline(memcg); | |
df878fb0 KH |
4370 | } |
4371 | ||
6df38689 VD |
4372 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
4373 | { | |
4374 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4375 | ||
4376 | invalidate_reclaim_iterators(memcg); | |
4377 | } | |
4378 | ||
eb95419b | 4379 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4380 | { |
eb95419b | 4381 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4382 | |
10d5ebf4 | 4383 | memcg_destroy_kmem(memcg); |
465939a1 | 4384 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
4385 | } |
4386 | ||
1ced953b TH |
4387 | /** |
4388 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4389 | * @css: the target css | |
4390 | * | |
4391 | * Reset the states of the mem_cgroup associated with @css. This is | |
4392 | * invoked when the userland requests disabling on the default hierarchy | |
4393 | * but the memcg is pinned through dependency. The memcg should stop | |
4394 | * applying policies and should revert to the vanilla state as it may be | |
4395 | * made visible again. | |
4396 | * | |
4397 | * The current implementation only resets the essential configurations. | |
4398 | * This needs to be expanded to cover all the visible parts. | |
4399 | */ | |
4400 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4401 | { | |
4402 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4403 | ||
3e32cb2e JW |
4404 | mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX); |
4405 | mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX); | |
4406 | memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX); | |
241994ed JW |
4407 | memcg->low = 0; |
4408 | memcg->high = PAGE_COUNTER_MAX; | |
24d404dc | 4409 | memcg->soft_limit = PAGE_COUNTER_MAX; |
2529bb3a | 4410 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
4411 | } |
4412 | ||
02491447 | 4413 | #ifdef CONFIG_MMU |
7dc74be0 | 4414 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4415 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4416 | { |
05b84301 | 4417 | int ret; |
9476db97 | 4418 | |
d0164adc MG |
4419 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
4420 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 4421 | if (!ret) { |
854ffa8d | 4422 | mc.precharge += count; |
854ffa8d DN |
4423 | return ret; |
4424 | } | |
9476db97 JW |
4425 | |
4426 | /* Try charges one by one with reclaim */ | |
854ffa8d | 4427 | while (count--) { |
00501b53 | 4428 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1); |
38c5d72f | 4429 | if (ret) |
38c5d72f | 4430 | return ret; |
854ffa8d | 4431 | mc.precharge++; |
9476db97 | 4432 | cond_resched(); |
854ffa8d | 4433 | } |
9476db97 | 4434 | return 0; |
4ffef5fe DN |
4435 | } |
4436 | ||
4437 | /** | |
8d32ff84 | 4438 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
4439 | * @vma: the vma the pte to be checked belongs |
4440 | * @addr: the address corresponding to the pte to be checked | |
4441 | * @ptent: the pte to be checked | |
02491447 | 4442 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
4443 | * |
4444 | * Returns | |
4445 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4446 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4447 | * move charge. if @target is not NULL, the page is stored in target->page | |
4448 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
4449 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
4450 | * target for charge migration. if @target is not NULL, the entry is stored | |
4451 | * in target->ent. | |
4ffef5fe DN |
4452 | * |
4453 | * Called with pte lock held. | |
4454 | */ | |
4ffef5fe DN |
4455 | union mc_target { |
4456 | struct page *page; | |
02491447 | 4457 | swp_entry_t ent; |
4ffef5fe DN |
4458 | }; |
4459 | ||
4ffef5fe | 4460 | enum mc_target_type { |
8d32ff84 | 4461 | MC_TARGET_NONE = 0, |
4ffef5fe | 4462 | MC_TARGET_PAGE, |
02491447 | 4463 | MC_TARGET_SWAP, |
4ffef5fe DN |
4464 | }; |
4465 | ||
90254a65 DN |
4466 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4467 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4468 | { |
90254a65 | 4469 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 4470 | |
90254a65 DN |
4471 | if (!page || !page_mapped(page)) |
4472 | return NULL; | |
4473 | if (PageAnon(page)) { | |
1dfab5ab | 4474 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 4475 | return NULL; |
1dfab5ab JW |
4476 | } else { |
4477 | if (!(mc.flags & MOVE_FILE)) | |
4478 | return NULL; | |
4479 | } | |
90254a65 DN |
4480 | if (!get_page_unless_zero(page)) |
4481 | return NULL; | |
4482 | ||
4483 | return page; | |
4484 | } | |
4485 | ||
4b91355e | 4486 | #ifdef CONFIG_SWAP |
90254a65 DN |
4487 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
4488 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4489 | { | |
90254a65 DN |
4490 | struct page *page = NULL; |
4491 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4492 | ||
1dfab5ab | 4493 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 4494 | return NULL; |
4b91355e KH |
4495 | /* |
4496 | * Because lookup_swap_cache() updates some statistics counter, | |
4497 | * we call find_get_page() with swapper_space directly. | |
4498 | */ | |
33806f06 | 4499 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
4500 | if (do_swap_account) |
4501 | entry->val = ent.val; | |
4502 | ||
4503 | return page; | |
4504 | } | |
4b91355e KH |
4505 | #else |
4506 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
4507 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4508 | { | |
4509 | return NULL; | |
4510 | } | |
4511 | #endif | |
90254a65 | 4512 | |
87946a72 DN |
4513 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4514 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4515 | { | |
4516 | struct page *page = NULL; | |
87946a72 DN |
4517 | struct address_space *mapping; |
4518 | pgoff_t pgoff; | |
4519 | ||
4520 | if (!vma->vm_file) /* anonymous vma */ | |
4521 | return NULL; | |
1dfab5ab | 4522 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
4523 | return NULL; |
4524 | ||
87946a72 | 4525 | mapping = vma->vm_file->f_mapping; |
0661a336 | 4526 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
4527 | |
4528 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
4529 | #ifdef CONFIG_SWAP |
4530 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
4531 | if (shmem_mapping(mapping)) { |
4532 | page = find_get_entry(mapping, pgoff); | |
4533 | if (radix_tree_exceptional_entry(page)) { | |
4534 | swp_entry_t swp = radix_to_swp_entry(page); | |
4535 | if (do_swap_account) | |
4536 | *entry = swp; | |
4537 | page = find_get_page(swap_address_space(swp), swp.val); | |
4538 | } | |
4539 | } else | |
4540 | page = find_get_page(mapping, pgoff); | |
4541 | #else | |
4542 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 4543 | #endif |
87946a72 DN |
4544 | return page; |
4545 | } | |
4546 | ||
b1b0deab CG |
4547 | /** |
4548 | * mem_cgroup_move_account - move account of the page | |
4549 | * @page: the page | |
4550 | * @nr_pages: number of regular pages (>1 for huge pages) | |
4551 | * @from: mem_cgroup which the page is moved from. | |
4552 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
4553 | * | |
4554 | * The caller must confirm following. | |
4555 | * - page is not on LRU (isolate_page() is useful.) | |
4556 | * - compound_lock is held when nr_pages > 1 | |
4557 | * | |
4558 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
4559 | * from old cgroup. | |
4560 | */ | |
4561 | static int mem_cgroup_move_account(struct page *page, | |
4562 | unsigned int nr_pages, | |
4563 | struct mem_cgroup *from, | |
4564 | struct mem_cgroup *to) | |
4565 | { | |
4566 | unsigned long flags; | |
4567 | int ret; | |
c4843a75 | 4568 | bool anon; |
b1b0deab CG |
4569 | |
4570 | VM_BUG_ON(from == to); | |
4571 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
4572 | /* | |
4573 | * The page is isolated from LRU. So, collapse function | |
4574 | * will not handle this page. But page splitting can happen. | |
4575 | * Do this check under compound_page_lock(). The caller should | |
4576 | * hold it. | |
4577 | */ | |
4578 | ret = -EBUSY; | |
4579 | if (nr_pages > 1 && !PageTransHuge(page)) | |
4580 | goto out; | |
4581 | ||
4582 | /* | |
45637bab HD |
4583 | * Prevent mem_cgroup_replace_page() from looking at |
4584 | * page->mem_cgroup of its source page while we change it. | |
b1b0deab CG |
4585 | */ |
4586 | if (!trylock_page(page)) | |
4587 | goto out; | |
4588 | ||
4589 | ret = -EINVAL; | |
4590 | if (page->mem_cgroup != from) | |
4591 | goto out_unlock; | |
4592 | ||
c4843a75 GT |
4593 | anon = PageAnon(page); |
4594 | ||
b1b0deab CG |
4595 | spin_lock_irqsave(&from->move_lock, flags); |
4596 | ||
c4843a75 | 4597 | if (!anon && page_mapped(page)) { |
b1b0deab CG |
4598 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
4599 | nr_pages); | |
4600 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
4601 | nr_pages); | |
4602 | } | |
4603 | ||
c4843a75 GT |
4604 | /* |
4605 | * move_lock grabbed above and caller set from->moving_account, so | |
4606 | * mem_cgroup_update_page_stat() will serialize updates to PageDirty. | |
4607 | * So mapping should be stable for dirty pages. | |
4608 | */ | |
4609 | if (!anon && PageDirty(page)) { | |
4610 | struct address_space *mapping = page_mapping(page); | |
4611 | ||
4612 | if (mapping_cap_account_dirty(mapping)) { | |
4613 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY], | |
4614 | nr_pages); | |
4615 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY], | |
4616 | nr_pages); | |
4617 | } | |
4618 | } | |
4619 | ||
b1b0deab CG |
4620 | if (PageWriteback(page)) { |
4621 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
4622 | nr_pages); | |
4623 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
4624 | nr_pages); | |
4625 | } | |
4626 | ||
4627 | /* | |
4628 | * It is safe to change page->mem_cgroup here because the page | |
4629 | * is referenced, charged, and isolated - we can't race with | |
4630 | * uncharging, charging, migration, or LRU putback. | |
4631 | */ | |
4632 | ||
4633 | /* caller should have done css_get */ | |
4634 | page->mem_cgroup = to; | |
4635 | spin_unlock_irqrestore(&from->move_lock, flags); | |
4636 | ||
4637 | ret = 0; | |
4638 | ||
4639 | local_irq_disable(); | |
4640 | mem_cgroup_charge_statistics(to, page, nr_pages); | |
4641 | memcg_check_events(to, page); | |
4642 | mem_cgroup_charge_statistics(from, page, -nr_pages); | |
4643 | memcg_check_events(from, page); | |
4644 | local_irq_enable(); | |
4645 | out_unlock: | |
4646 | unlock_page(page); | |
4647 | out: | |
4648 | return ret; | |
4649 | } | |
4650 | ||
8d32ff84 | 4651 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
4652 | unsigned long addr, pte_t ptent, union mc_target *target) |
4653 | { | |
4654 | struct page *page = NULL; | |
8d32ff84 | 4655 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
4656 | swp_entry_t ent = { .val = 0 }; |
4657 | ||
4658 | if (pte_present(ptent)) | |
4659 | page = mc_handle_present_pte(vma, addr, ptent); | |
4660 | else if (is_swap_pte(ptent)) | |
4661 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
0661a336 | 4662 | else if (pte_none(ptent)) |
87946a72 | 4663 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
4664 | |
4665 | if (!page && !ent.val) | |
8d32ff84 | 4666 | return ret; |
02491447 | 4667 | if (page) { |
02491447 | 4668 | /* |
0a31bc97 | 4669 | * Do only loose check w/o serialization. |
1306a85a | 4670 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 4671 | * not under LRU exclusion. |
02491447 | 4672 | */ |
1306a85a | 4673 | if (page->mem_cgroup == mc.from) { |
02491447 DN |
4674 | ret = MC_TARGET_PAGE; |
4675 | if (target) | |
4676 | target->page = page; | |
4677 | } | |
4678 | if (!ret || !target) | |
4679 | put_page(page); | |
4680 | } | |
90254a65 DN |
4681 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
4682 | if (ent.val && !ret && | |
34c00c31 | 4683 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
4684 | ret = MC_TARGET_SWAP; |
4685 | if (target) | |
4686 | target->ent = ent; | |
4ffef5fe | 4687 | } |
4ffef5fe DN |
4688 | return ret; |
4689 | } | |
4690 | ||
12724850 NH |
4691 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4692 | /* | |
4693 | * We don't consider swapping or file mapped pages because THP does not | |
4694 | * support them for now. | |
4695 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
4696 | */ | |
4697 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4698 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4699 | { | |
4700 | struct page *page = NULL; | |
12724850 NH |
4701 | enum mc_target_type ret = MC_TARGET_NONE; |
4702 | ||
4703 | page = pmd_page(pmd); | |
309381fe | 4704 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 4705 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 4706 | return ret; |
1306a85a | 4707 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
4708 | ret = MC_TARGET_PAGE; |
4709 | if (target) { | |
4710 | get_page(page); | |
4711 | target->page = page; | |
4712 | } | |
4713 | } | |
4714 | return ret; | |
4715 | } | |
4716 | #else | |
4717 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4718 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4719 | { | |
4720 | return MC_TARGET_NONE; | |
4721 | } | |
4722 | #endif | |
4723 | ||
4ffef5fe DN |
4724 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
4725 | unsigned long addr, unsigned long end, | |
4726 | struct mm_walk *walk) | |
4727 | { | |
26bcd64a | 4728 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4729 | pte_t *pte; |
4730 | spinlock_t *ptl; | |
4731 | ||
bf929152 | 4732 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
12724850 NH |
4733 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
4734 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 4735 | spin_unlock(ptl); |
1a5a9906 | 4736 | return 0; |
12724850 | 4737 | } |
03319327 | 4738 | |
45f83cef AA |
4739 | if (pmd_trans_unstable(pmd)) |
4740 | return 0; | |
4ffef5fe DN |
4741 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
4742 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 4743 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
4744 | mc.precharge++; /* increment precharge temporarily */ |
4745 | pte_unmap_unlock(pte - 1, ptl); | |
4746 | cond_resched(); | |
4747 | ||
7dc74be0 DN |
4748 | return 0; |
4749 | } | |
4750 | ||
4ffef5fe DN |
4751 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
4752 | { | |
4753 | unsigned long precharge; | |
4ffef5fe | 4754 | |
26bcd64a NH |
4755 | struct mm_walk mem_cgroup_count_precharge_walk = { |
4756 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
4757 | .mm = mm, | |
4758 | }; | |
dfe076b0 | 4759 | down_read(&mm->mmap_sem); |
26bcd64a | 4760 | walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk); |
dfe076b0 | 4761 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
4762 | |
4763 | precharge = mc.precharge; | |
4764 | mc.precharge = 0; | |
4765 | ||
4766 | return precharge; | |
4767 | } | |
4768 | ||
4ffef5fe DN |
4769 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
4770 | { | |
dfe076b0 DN |
4771 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
4772 | ||
4773 | VM_BUG_ON(mc.moving_task); | |
4774 | mc.moving_task = current; | |
4775 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
4776 | } |
4777 | ||
dfe076b0 DN |
4778 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
4779 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 4780 | { |
2bd9bb20 KH |
4781 | struct mem_cgroup *from = mc.from; |
4782 | struct mem_cgroup *to = mc.to; | |
4783 | ||
4ffef5fe | 4784 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 4785 | if (mc.precharge) { |
00501b53 | 4786 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
4787 | mc.precharge = 0; |
4788 | } | |
4789 | /* | |
4790 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
4791 | * we must uncharge here. | |
4792 | */ | |
4793 | if (mc.moved_charge) { | |
00501b53 | 4794 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 4795 | mc.moved_charge = 0; |
4ffef5fe | 4796 | } |
483c30b5 DN |
4797 | /* we must fixup refcnts and charges */ |
4798 | if (mc.moved_swap) { | |
483c30b5 | 4799 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 4800 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 4801 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 4802 | |
05b84301 | 4803 | /* |
3e32cb2e JW |
4804 | * we charged both to->memory and to->memsw, so we |
4805 | * should uncharge to->memory. | |
05b84301 | 4806 | */ |
ce00a967 | 4807 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
4808 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
4809 | ||
e8ea14cc | 4810 | css_put_many(&mc.from->css, mc.moved_swap); |
3e32cb2e | 4811 | |
4050377b | 4812 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
4813 | mc.moved_swap = 0; |
4814 | } | |
dfe076b0 DN |
4815 | memcg_oom_recover(from); |
4816 | memcg_oom_recover(to); | |
4817 | wake_up_all(&mc.waitq); | |
4818 | } | |
4819 | ||
4820 | static void mem_cgroup_clear_mc(void) | |
4821 | { | |
dfe076b0 DN |
4822 | /* |
4823 | * we must clear moving_task before waking up waiters at the end of | |
4824 | * task migration. | |
4825 | */ | |
4826 | mc.moving_task = NULL; | |
4827 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 4828 | spin_lock(&mc.lock); |
4ffef5fe DN |
4829 | mc.from = NULL; |
4830 | mc.to = NULL; | |
2bd9bb20 | 4831 | spin_unlock(&mc.lock); |
4ffef5fe DN |
4832 | } |
4833 | ||
1f7dd3e5 | 4834 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 4835 | { |
1f7dd3e5 | 4836 | struct cgroup_subsys_state *css; |
eed67d75 | 4837 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 4838 | struct mem_cgroup *from; |
4530eddb | 4839 | struct task_struct *leader, *p; |
9f2115f9 | 4840 | struct mm_struct *mm; |
1dfab5ab | 4841 | unsigned long move_flags; |
9f2115f9 | 4842 | int ret = 0; |
7dc74be0 | 4843 | |
1f7dd3e5 TH |
4844 | /* charge immigration isn't supported on the default hierarchy */ |
4845 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
4846 | return 0; |
4847 | ||
4530eddb TH |
4848 | /* |
4849 | * Multi-process migrations only happen on the default hierarchy | |
4850 | * where charge immigration is not used. Perform charge | |
4851 | * immigration if @tset contains a leader and whine if there are | |
4852 | * multiple. | |
4853 | */ | |
4854 | p = NULL; | |
1f7dd3e5 | 4855 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
4856 | WARN_ON_ONCE(p); |
4857 | p = leader; | |
1f7dd3e5 | 4858 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
4859 | } |
4860 | if (!p) | |
4861 | return 0; | |
4862 | ||
1f7dd3e5 TH |
4863 | /* |
4864 | * We are now commited to this value whatever it is. Changes in this | |
4865 | * tunable will only affect upcoming migrations, not the current one. | |
4866 | * So we need to save it, and keep it going. | |
4867 | */ | |
4868 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
4869 | if (!move_flags) | |
4870 | return 0; | |
4871 | ||
9f2115f9 TH |
4872 | from = mem_cgroup_from_task(p); |
4873 | ||
4874 | VM_BUG_ON(from == memcg); | |
4875 | ||
4876 | mm = get_task_mm(p); | |
4877 | if (!mm) | |
4878 | return 0; | |
4879 | /* We move charges only when we move a owner of the mm */ | |
4880 | if (mm->owner == p) { | |
4881 | VM_BUG_ON(mc.from); | |
4882 | VM_BUG_ON(mc.to); | |
4883 | VM_BUG_ON(mc.precharge); | |
4884 | VM_BUG_ON(mc.moved_charge); | |
4885 | VM_BUG_ON(mc.moved_swap); | |
4886 | ||
4887 | spin_lock(&mc.lock); | |
4888 | mc.from = from; | |
4889 | mc.to = memcg; | |
4890 | mc.flags = move_flags; | |
4891 | spin_unlock(&mc.lock); | |
4892 | /* We set mc.moving_task later */ | |
4893 | ||
4894 | ret = mem_cgroup_precharge_mc(mm); | |
4895 | if (ret) | |
4896 | mem_cgroup_clear_mc(); | |
7dc74be0 | 4897 | } |
9f2115f9 | 4898 | mmput(mm); |
7dc74be0 DN |
4899 | return ret; |
4900 | } | |
4901 | ||
1f7dd3e5 | 4902 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 4903 | { |
4e2f245d JW |
4904 | if (mc.to) |
4905 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
4906 | } |
4907 | ||
4ffef5fe DN |
4908 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
4909 | unsigned long addr, unsigned long end, | |
4910 | struct mm_walk *walk) | |
7dc74be0 | 4911 | { |
4ffef5fe | 4912 | int ret = 0; |
26bcd64a | 4913 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4914 | pte_t *pte; |
4915 | spinlock_t *ptl; | |
12724850 NH |
4916 | enum mc_target_type target_type; |
4917 | union mc_target target; | |
4918 | struct page *page; | |
4ffef5fe | 4919 | |
12724850 NH |
4920 | /* |
4921 | * We don't take compound_lock() here but no race with splitting thp | |
4922 | * happens because: | |
4923 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
4924 | * under splitting, which means there's no concurrent thp split, | |
4925 | * - if another thread runs into split_huge_page() just after we | |
4926 | * entered this if-block, the thread must wait for page table lock | |
4927 | * to be unlocked in __split_huge_page_splitting(), where the main | |
4928 | * part of thp split is not executed yet. | |
4929 | */ | |
bf929152 | 4930 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
62ade86a | 4931 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 4932 | spin_unlock(ptl); |
12724850 NH |
4933 | return 0; |
4934 | } | |
4935 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
4936 | if (target_type == MC_TARGET_PAGE) { | |
4937 | page = target.page; | |
4938 | if (!isolate_lru_page(page)) { | |
12724850 | 4939 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, |
1306a85a | 4940 | mc.from, mc.to)) { |
12724850 NH |
4941 | mc.precharge -= HPAGE_PMD_NR; |
4942 | mc.moved_charge += HPAGE_PMD_NR; | |
4943 | } | |
4944 | putback_lru_page(page); | |
4945 | } | |
4946 | put_page(page); | |
4947 | } | |
bf929152 | 4948 | spin_unlock(ptl); |
1a5a9906 | 4949 | return 0; |
12724850 NH |
4950 | } |
4951 | ||
45f83cef AA |
4952 | if (pmd_trans_unstable(pmd)) |
4953 | return 0; | |
4ffef5fe DN |
4954 | retry: |
4955 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
4956 | for (; addr != end; addr += PAGE_SIZE) { | |
4957 | pte_t ptent = *(pte++); | |
02491447 | 4958 | swp_entry_t ent; |
4ffef5fe DN |
4959 | |
4960 | if (!mc.precharge) | |
4961 | break; | |
4962 | ||
8d32ff84 | 4963 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
4964 | case MC_TARGET_PAGE: |
4965 | page = target.page; | |
4966 | if (isolate_lru_page(page)) | |
4967 | goto put; | |
1306a85a | 4968 | if (!mem_cgroup_move_account(page, 1, mc.from, mc.to)) { |
4ffef5fe | 4969 | mc.precharge--; |
854ffa8d DN |
4970 | /* we uncharge from mc.from later. */ |
4971 | mc.moved_charge++; | |
4ffef5fe DN |
4972 | } |
4973 | putback_lru_page(page); | |
8d32ff84 | 4974 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
4975 | put_page(page); |
4976 | break; | |
02491447 DN |
4977 | case MC_TARGET_SWAP: |
4978 | ent = target.ent; | |
e91cbb42 | 4979 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 4980 | mc.precharge--; |
483c30b5 DN |
4981 | /* we fixup refcnts and charges later. */ |
4982 | mc.moved_swap++; | |
4983 | } | |
02491447 | 4984 | break; |
4ffef5fe DN |
4985 | default: |
4986 | break; | |
4987 | } | |
4988 | } | |
4989 | pte_unmap_unlock(pte - 1, ptl); | |
4990 | cond_resched(); | |
4991 | ||
4992 | if (addr != end) { | |
4993 | /* | |
4994 | * We have consumed all precharges we got in can_attach(). | |
4995 | * We try charge one by one, but don't do any additional | |
4996 | * charges to mc.to if we have failed in charge once in attach() | |
4997 | * phase. | |
4998 | */ | |
854ffa8d | 4999 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5000 | if (!ret) |
5001 | goto retry; | |
5002 | } | |
5003 | ||
5004 | return ret; | |
5005 | } | |
5006 | ||
5007 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
5008 | { | |
26bcd64a NH |
5009 | struct mm_walk mem_cgroup_move_charge_walk = { |
5010 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
5011 | .mm = mm, | |
5012 | }; | |
4ffef5fe DN |
5013 | |
5014 | lru_add_drain_all(); | |
312722cb JW |
5015 | /* |
5016 | * Signal mem_cgroup_begin_page_stat() to take the memcg's | |
5017 | * move_lock while we're moving its pages to another memcg. | |
5018 | * Then wait for already started RCU-only updates to finish. | |
5019 | */ | |
5020 | atomic_inc(&mc.from->moving_account); | |
5021 | synchronize_rcu(); | |
dfe076b0 DN |
5022 | retry: |
5023 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5024 | /* | |
5025 | * Someone who are holding the mmap_sem might be waiting in | |
5026 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5027 | * and retry. Because we cancel precharges, we might not be able | |
5028 | * to move enough charges, but moving charge is a best-effort | |
5029 | * feature anyway, so it wouldn't be a big problem. | |
5030 | */ | |
5031 | __mem_cgroup_clear_mc(); | |
5032 | cond_resched(); | |
5033 | goto retry; | |
5034 | } | |
26bcd64a NH |
5035 | /* |
5036 | * When we have consumed all precharges and failed in doing | |
5037 | * additional charge, the page walk just aborts. | |
5038 | */ | |
5039 | walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk); | |
dfe076b0 | 5040 | up_read(&mm->mmap_sem); |
312722cb | 5041 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
5042 | } |
5043 | ||
1f7dd3e5 | 5044 | static void mem_cgroup_move_task(struct cgroup_taskset *tset) |
67e465a7 | 5045 | { |
1f7dd3e5 TH |
5046 | struct cgroup_subsys_state *css; |
5047 | struct task_struct *p = cgroup_taskset_first(tset, &css); | |
a433658c | 5048 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 5049 | |
dfe076b0 | 5050 | if (mm) { |
a433658c KM |
5051 | if (mc.to) |
5052 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
5053 | mmput(mm); |
5054 | } | |
a433658c KM |
5055 | if (mc.to) |
5056 | mem_cgroup_clear_mc(); | |
67e465a7 | 5057 | } |
5cfb80a7 | 5058 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 5059 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5060 | { |
5061 | return 0; | |
5062 | } | |
1f7dd3e5 | 5063 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5064 | { |
5065 | } | |
1f7dd3e5 | 5066 | static void mem_cgroup_move_task(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5067 | { |
5068 | } | |
5069 | #endif | |
67e465a7 | 5070 | |
f00baae7 TH |
5071 | /* |
5072 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5073 | * to verify whether we're attached to the default hierarchy on each mount |
5074 | * attempt. | |
f00baae7 | 5075 | */ |
eb95419b | 5076 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5077 | { |
5078 | /* | |
aa6ec29b | 5079 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5080 | * guarantees that @root doesn't have any children, so turning it |
5081 | * on for the root memcg is enough. | |
5082 | */ | |
9e10a130 | 5083 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
5084 | root_mem_cgroup->use_hierarchy = true; |
5085 | else | |
5086 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
5087 | } |
5088 | ||
241994ed JW |
5089 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
5090 | struct cftype *cft) | |
5091 | { | |
f5fc3c5d JW |
5092 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5093 | ||
5094 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
5095 | } |
5096 | ||
5097 | static int memory_low_show(struct seq_file *m, void *v) | |
5098 | { | |
5099 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5100 | unsigned long low = READ_ONCE(memcg->low); |
241994ed JW |
5101 | |
5102 | if (low == PAGE_COUNTER_MAX) | |
d2973697 | 5103 | seq_puts(m, "max\n"); |
241994ed JW |
5104 | else |
5105 | seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE); | |
5106 | ||
5107 | return 0; | |
5108 | } | |
5109 | ||
5110 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
5111 | char *buf, size_t nbytes, loff_t off) | |
5112 | { | |
5113 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5114 | unsigned long low; | |
5115 | int err; | |
5116 | ||
5117 | buf = strstrip(buf); | |
d2973697 | 5118 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
5119 | if (err) |
5120 | return err; | |
5121 | ||
5122 | memcg->low = low; | |
5123 | ||
5124 | return nbytes; | |
5125 | } | |
5126 | ||
5127 | static int memory_high_show(struct seq_file *m, void *v) | |
5128 | { | |
5129 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5130 | unsigned long high = READ_ONCE(memcg->high); |
241994ed JW |
5131 | |
5132 | if (high == PAGE_COUNTER_MAX) | |
d2973697 | 5133 | seq_puts(m, "max\n"); |
241994ed JW |
5134 | else |
5135 | seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE); | |
5136 | ||
5137 | return 0; | |
5138 | } | |
5139 | ||
5140 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
5141 | char *buf, size_t nbytes, loff_t off) | |
5142 | { | |
5143 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5144 | unsigned long high; | |
5145 | int err; | |
5146 | ||
5147 | buf = strstrip(buf); | |
d2973697 | 5148 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
5149 | if (err) |
5150 | return err; | |
5151 | ||
5152 | memcg->high = high; | |
5153 | ||
2529bb3a | 5154 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5155 | return nbytes; |
5156 | } | |
5157 | ||
5158 | static int memory_max_show(struct seq_file *m, void *v) | |
5159 | { | |
5160 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5161 | unsigned long max = READ_ONCE(memcg->memory.limit); |
241994ed JW |
5162 | |
5163 | if (max == PAGE_COUNTER_MAX) | |
d2973697 | 5164 | seq_puts(m, "max\n"); |
241994ed JW |
5165 | else |
5166 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
5167 | ||
5168 | return 0; | |
5169 | } | |
5170 | ||
5171 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
5172 | char *buf, size_t nbytes, loff_t off) | |
5173 | { | |
5174 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5175 | unsigned long max; | |
5176 | int err; | |
5177 | ||
5178 | buf = strstrip(buf); | |
d2973697 | 5179 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
5180 | if (err) |
5181 | return err; | |
5182 | ||
5183 | err = mem_cgroup_resize_limit(memcg, max); | |
5184 | if (err) | |
5185 | return err; | |
5186 | ||
2529bb3a | 5187 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5188 | return nbytes; |
5189 | } | |
5190 | ||
5191 | static int memory_events_show(struct seq_file *m, void *v) | |
5192 | { | |
5193 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5194 | ||
5195 | seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW)); | |
5196 | seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH)); | |
5197 | seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX)); | |
5198 | seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM)); | |
5199 | ||
5200 | return 0; | |
5201 | } | |
5202 | ||
5203 | static struct cftype memory_files[] = { | |
5204 | { | |
5205 | .name = "current", | |
f5fc3c5d | 5206 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
5207 | .read_u64 = memory_current_read, |
5208 | }, | |
5209 | { | |
5210 | .name = "low", | |
5211 | .flags = CFTYPE_NOT_ON_ROOT, | |
5212 | .seq_show = memory_low_show, | |
5213 | .write = memory_low_write, | |
5214 | }, | |
5215 | { | |
5216 | .name = "high", | |
5217 | .flags = CFTYPE_NOT_ON_ROOT, | |
5218 | .seq_show = memory_high_show, | |
5219 | .write = memory_high_write, | |
5220 | }, | |
5221 | { | |
5222 | .name = "max", | |
5223 | .flags = CFTYPE_NOT_ON_ROOT, | |
5224 | .seq_show = memory_max_show, | |
5225 | .write = memory_max_write, | |
5226 | }, | |
5227 | { | |
5228 | .name = "events", | |
5229 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 5230 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
5231 | .seq_show = memory_events_show, |
5232 | }, | |
5233 | { } /* terminate */ | |
5234 | }; | |
5235 | ||
073219e9 | 5236 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5237 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5238 | .css_online = mem_cgroup_css_online, |
92fb9748 | 5239 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 5240 | .css_released = mem_cgroup_css_released, |
92fb9748 | 5241 | .css_free = mem_cgroup_css_free, |
1ced953b | 5242 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5243 | .can_attach = mem_cgroup_can_attach, |
5244 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5245 | .attach = mem_cgroup_move_task, |
f00baae7 | 5246 | .bind = mem_cgroup_bind, |
241994ed JW |
5247 | .dfl_cftypes = memory_files, |
5248 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 5249 | .early_init = 0, |
8cdea7c0 | 5250 | }; |
c077719b | 5251 | |
241994ed JW |
5252 | /** |
5253 | * mem_cgroup_low - check if memory consumption is below the normal range | |
5254 | * @root: the highest ancestor to consider | |
5255 | * @memcg: the memory cgroup to check | |
5256 | * | |
5257 | * Returns %true if memory consumption of @memcg, and that of all | |
5258 | * configurable ancestors up to @root, is below the normal range. | |
5259 | */ | |
5260 | bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg) | |
5261 | { | |
5262 | if (mem_cgroup_disabled()) | |
5263 | return false; | |
5264 | ||
5265 | /* | |
5266 | * The toplevel group doesn't have a configurable range, so | |
5267 | * it's never low when looked at directly, and it is not | |
5268 | * considered an ancestor when assessing the hierarchy. | |
5269 | */ | |
5270 | ||
5271 | if (memcg == root_mem_cgroup) | |
5272 | return false; | |
5273 | ||
4e54dede | 5274 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5275 | return false; |
5276 | ||
5277 | while (memcg != root) { | |
5278 | memcg = parent_mem_cgroup(memcg); | |
5279 | ||
5280 | if (memcg == root_mem_cgroup) | |
5281 | break; | |
5282 | ||
4e54dede | 5283 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5284 | return false; |
5285 | } | |
5286 | return true; | |
5287 | } | |
5288 | ||
00501b53 JW |
5289 | /** |
5290 | * mem_cgroup_try_charge - try charging a page | |
5291 | * @page: page to charge | |
5292 | * @mm: mm context of the victim | |
5293 | * @gfp_mask: reclaim mode | |
5294 | * @memcgp: charged memcg return | |
5295 | * | |
5296 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5297 | * pages according to @gfp_mask if necessary. | |
5298 | * | |
5299 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5300 | * Otherwise, an error code is returned. | |
5301 | * | |
5302 | * After page->mapping has been set up, the caller must finalize the | |
5303 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5304 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5305 | */ | |
5306 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
5307 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
5308 | { | |
5309 | struct mem_cgroup *memcg = NULL; | |
5310 | unsigned int nr_pages = 1; | |
5311 | int ret = 0; | |
5312 | ||
5313 | if (mem_cgroup_disabled()) | |
5314 | goto out; | |
5315 | ||
5316 | if (PageSwapCache(page)) { | |
00501b53 JW |
5317 | /* |
5318 | * Every swap fault against a single page tries to charge the | |
5319 | * page, bail as early as possible. shmem_unuse() encounters | |
5320 | * already charged pages, too. The USED bit is protected by | |
5321 | * the page lock, which serializes swap cache removal, which | |
5322 | * in turn serializes uncharging. | |
5323 | */ | |
e993d905 | 5324 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1306a85a | 5325 | if (page->mem_cgroup) |
00501b53 | 5326 | goto out; |
e993d905 VD |
5327 | |
5328 | if (do_swap_account) { | |
5329 | swp_entry_t ent = { .val = page_private(page), }; | |
5330 | unsigned short id = lookup_swap_cgroup_id(ent); | |
5331 | ||
5332 | rcu_read_lock(); | |
5333 | memcg = mem_cgroup_from_id(id); | |
5334 | if (memcg && !css_tryget_online(&memcg->css)) | |
5335 | memcg = NULL; | |
5336 | rcu_read_unlock(); | |
5337 | } | |
00501b53 JW |
5338 | } |
5339 | ||
5340 | if (PageTransHuge(page)) { | |
5341 | nr_pages <<= compound_order(page); | |
5342 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5343 | } | |
5344 | ||
00501b53 JW |
5345 | if (!memcg) |
5346 | memcg = get_mem_cgroup_from_mm(mm); | |
5347 | ||
5348 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
5349 | ||
5350 | css_put(&memcg->css); | |
00501b53 JW |
5351 | out: |
5352 | *memcgp = memcg; | |
5353 | return ret; | |
5354 | } | |
5355 | ||
5356 | /** | |
5357 | * mem_cgroup_commit_charge - commit a page charge | |
5358 | * @page: page to charge | |
5359 | * @memcg: memcg to charge the page to | |
5360 | * @lrucare: page might be on LRU already | |
5361 | * | |
5362 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
5363 | * after page->mapping has been set up. This must happen atomically | |
5364 | * as part of the page instantiation, i.e. under the page table lock | |
5365 | * for anonymous pages, under the page lock for page and swap cache. | |
5366 | * | |
5367 | * In addition, the page must not be on the LRU during the commit, to | |
5368 | * prevent racing with task migration. If it might be, use @lrucare. | |
5369 | * | |
5370 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
5371 | */ | |
5372 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
5373 | bool lrucare) | |
5374 | { | |
5375 | unsigned int nr_pages = 1; | |
5376 | ||
5377 | VM_BUG_ON_PAGE(!page->mapping, page); | |
5378 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
5379 | ||
5380 | if (mem_cgroup_disabled()) | |
5381 | return; | |
5382 | /* | |
5383 | * Swap faults will attempt to charge the same page multiple | |
5384 | * times. But reuse_swap_page() might have removed the page | |
5385 | * from swapcache already, so we can't check PageSwapCache(). | |
5386 | */ | |
5387 | if (!memcg) | |
5388 | return; | |
5389 | ||
6abb5a86 JW |
5390 | commit_charge(page, memcg, lrucare); |
5391 | ||
00501b53 JW |
5392 | if (PageTransHuge(page)) { |
5393 | nr_pages <<= compound_order(page); | |
5394 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5395 | } | |
5396 | ||
6abb5a86 JW |
5397 | local_irq_disable(); |
5398 | mem_cgroup_charge_statistics(memcg, page, nr_pages); | |
5399 | memcg_check_events(memcg, page); | |
5400 | local_irq_enable(); | |
00501b53 JW |
5401 | |
5402 | if (do_swap_account && PageSwapCache(page)) { | |
5403 | swp_entry_t entry = { .val = page_private(page) }; | |
5404 | /* | |
5405 | * The swap entry might not get freed for a long time, | |
5406 | * let's not wait for it. The page already received a | |
5407 | * memory+swap charge, drop the swap entry duplicate. | |
5408 | */ | |
5409 | mem_cgroup_uncharge_swap(entry); | |
5410 | } | |
5411 | } | |
5412 | ||
5413 | /** | |
5414 | * mem_cgroup_cancel_charge - cancel a page charge | |
5415 | * @page: page to charge | |
5416 | * @memcg: memcg to charge the page to | |
5417 | * | |
5418 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
5419 | */ | |
5420 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg) | |
5421 | { | |
5422 | unsigned int nr_pages = 1; | |
5423 | ||
5424 | if (mem_cgroup_disabled()) | |
5425 | return; | |
5426 | /* | |
5427 | * Swap faults will attempt to charge the same page multiple | |
5428 | * times. But reuse_swap_page() might have removed the page | |
5429 | * from swapcache already, so we can't check PageSwapCache(). | |
5430 | */ | |
5431 | if (!memcg) | |
5432 | return; | |
5433 | ||
5434 | if (PageTransHuge(page)) { | |
5435 | nr_pages <<= compound_order(page); | |
5436 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5437 | } | |
5438 | ||
5439 | cancel_charge(memcg, nr_pages); | |
5440 | } | |
5441 | ||
747db954 | 5442 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
747db954 JW |
5443 | unsigned long nr_anon, unsigned long nr_file, |
5444 | unsigned long nr_huge, struct page *dummy_page) | |
5445 | { | |
18eca2e6 | 5446 | unsigned long nr_pages = nr_anon + nr_file; |
747db954 JW |
5447 | unsigned long flags; |
5448 | ||
ce00a967 | 5449 | if (!mem_cgroup_is_root(memcg)) { |
18eca2e6 JW |
5450 | page_counter_uncharge(&memcg->memory, nr_pages); |
5451 | if (do_swap_account) | |
5452 | page_counter_uncharge(&memcg->memsw, nr_pages); | |
ce00a967 JW |
5453 | memcg_oom_recover(memcg); |
5454 | } | |
747db954 JW |
5455 | |
5456 | local_irq_save(flags); | |
5457 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); | |
5458 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); | |
5459 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); | |
5460 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); | |
18eca2e6 | 5461 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
747db954 JW |
5462 | memcg_check_events(memcg, dummy_page); |
5463 | local_irq_restore(flags); | |
e8ea14cc JW |
5464 | |
5465 | if (!mem_cgroup_is_root(memcg)) | |
18eca2e6 | 5466 | css_put_many(&memcg->css, nr_pages); |
747db954 JW |
5467 | } |
5468 | ||
5469 | static void uncharge_list(struct list_head *page_list) | |
5470 | { | |
5471 | struct mem_cgroup *memcg = NULL; | |
747db954 JW |
5472 | unsigned long nr_anon = 0; |
5473 | unsigned long nr_file = 0; | |
5474 | unsigned long nr_huge = 0; | |
5475 | unsigned long pgpgout = 0; | |
747db954 JW |
5476 | struct list_head *next; |
5477 | struct page *page; | |
5478 | ||
5479 | next = page_list->next; | |
5480 | do { | |
5481 | unsigned int nr_pages = 1; | |
747db954 JW |
5482 | |
5483 | page = list_entry(next, struct page, lru); | |
5484 | next = page->lru.next; | |
5485 | ||
5486 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5487 | VM_BUG_ON_PAGE(page_count(page), page); | |
5488 | ||
1306a85a | 5489 | if (!page->mem_cgroup) |
747db954 JW |
5490 | continue; |
5491 | ||
5492 | /* | |
5493 | * Nobody should be changing or seriously looking at | |
1306a85a | 5494 | * page->mem_cgroup at this point, we have fully |
29833315 | 5495 | * exclusive access to the page. |
747db954 JW |
5496 | */ |
5497 | ||
1306a85a | 5498 | if (memcg != page->mem_cgroup) { |
747db954 | 5499 | if (memcg) { |
18eca2e6 JW |
5500 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5501 | nr_huge, page); | |
5502 | pgpgout = nr_anon = nr_file = nr_huge = 0; | |
747db954 | 5503 | } |
1306a85a | 5504 | memcg = page->mem_cgroup; |
747db954 JW |
5505 | } |
5506 | ||
5507 | if (PageTransHuge(page)) { | |
5508 | nr_pages <<= compound_order(page); | |
5509 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5510 | nr_huge += nr_pages; | |
5511 | } | |
5512 | ||
5513 | if (PageAnon(page)) | |
5514 | nr_anon += nr_pages; | |
5515 | else | |
5516 | nr_file += nr_pages; | |
5517 | ||
1306a85a | 5518 | page->mem_cgroup = NULL; |
747db954 JW |
5519 | |
5520 | pgpgout++; | |
5521 | } while (next != page_list); | |
5522 | ||
5523 | if (memcg) | |
18eca2e6 JW |
5524 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5525 | nr_huge, page); | |
747db954 JW |
5526 | } |
5527 | ||
0a31bc97 JW |
5528 | /** |
5529 | * mem_cgroup_uncharge - uncharge a page | |
5530 | * @page: page to uncharge | |
5531 | * | |
5532 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
5533 | * mem_cgroup_commit_charge(). | |
5534 | */ | |
5535 | void mem_cgroup_uncharge(struct page *page) | |
5536 | { | |
0a31bc97 JW |
5537 | if (mem_cgroup_disabled()) |
5538 | return; | |
5539 | ||
747db954 | 5540 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 5541 | if (!page->mem_cgroup) |
0a31bc97 JW |
5542 | return; |
5543 | ||
747db954 JW |
5544 | INIT_LIST_HEAD(&page->lru); |
5545 | uncharge_list(&page->lru); | |
5546 | } | |
0a31bc97 | 5547 | |
747db954 JW |
5548 | /** |
5549 | * mem_cgroup_uncharge_list - uncharge a list of page | |
5550 | * @page_list: list of pages to uncharge | |
5551 | * | |
5552 | * Uncharge a list of pages previously charged with | |
5553 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
5554 | */ | |
5555 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
5556 | { | |
5557 | if (mem_cgroup_disabled()) | |
5558 | return; | |
0a31bc97 | 5559 | |
747db954 JW |
5560 | if (!list_empty(page_list)) |
5561 | uncharge_list(page_list); | |
0a31bc97 JW |
5562 | } |
5563 | ||
5564 | /** | |
45637bab | 5565 | * mem_cgroup_replace_page - migrate a charge to another page |
0a31bc97 JW |
5566 | * @oldpage: currently charged page |
5567 | * @newpage: page to transfer the charge to | |
0a31bc97 JW |
5568 | * |
5569 | * Migrate the charge from @oldpage to @newpage. | |
5570 | * | |
5571 | * Both pages must be locked, @newpage->mapping must be set up. | |
25be6a65 | 5572 | * Either or both pages might be on the LRU already. |
0a31bc97 | 5573 | */ |
45637bab | 5574 | void mem_cgroup_replace_page(struct page *oldpage, struct page *newpage) |
0a31bc97 | 5575 | { |
29833315 | 5576 | struct mem_cgroup *memcg; |
0a31bc97 JW |
5577 | int isolated; |
5578 | ||
5579 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
5580 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 5581 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
5582 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
5583 | newpage); | |
0a31bc97 JW |
5584 | |
5585 | if (mem_cgroup_disabled()) | |
5586 | return; | |
5587 | ||
5588 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 5589 | if (newpage->mem_cgroup) |
0a31bc97 JW |
5590 | return; |
5591 | ||
45637bab | 5592 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 5593 | memcg = oldpage->mem_cgroup; |
29833315 | 5594 | if (!memcg) |
0a31bc97 JW |
5595 | return; |
5596 | ||
45637bab | 5597 | lock_page_lru(oldpage, &isolated); |
1306a85a | 5598 | oldpage->mem_cgroup = NULL; |
45637bab | 5599 | unlock_page_lru(oldpage, isolated); |
0a31bc97 | 5600 | |
45637bab | 5601 | commit_charge(newpage, memcg, true); |
0a31bc97 JW |
5602 | } |
5603 | ||
2d11085e | 5604 | /* |
1081312f MH |
5605 | * subsys_initcall() for memory controller. |
5606 | * | |
5607 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
5608 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
5609 | * everything that doesn't depend on a specific mem_cgroup structure should | |
5610 | * be initialized from here. | |
2d11085e MH |
5611 | */ |
5612 | static int __init mem_cgroup_init(void) | |
5613 | { | |
95a045f6 JW |
5614 | int cpu, node; |
5615 | ||
2d11085e | 5616 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
95a045f6 JW |
5617 | |
5618 | for_each_possible_cpu(cpu) | |
5619 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
5620 | drain_local_stock); | |
5621 | ||
5622 | for_each_node(node) { | |
5623 | struct mem_cgroup_tree_per_node *rtpn; | |
5624 | int zone; | |
5625 | ||
5626 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
5627 | node_online(node) ? node : NUMA_NO_NODE); | |
5628 | ||
5629 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
5630 | struct mem_cgroup_tree_per_zone *rtpz; | |
5631 | ||
5632 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
5633 | rtpz->rb_root = RB_ROOT; | |
5634 | spin_lock_init(&rtpz->lock); | |
5635 | } | |
5636 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
5637 | } | |
5638 | ||
2d11085e MH |
5639 | return 0; |
5640 | } | |
5641 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
5642 | |
5643 | #ifdef CONFIG_MEMCG_SWAP | |
5644 | /** | |
5645 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
5646 | * @page: page whose memsw charge to transfer | |
5647 | * @entry: swap entry to move the charge to | |
5648 | * | |
5649 | * Transfer the memsw charge of @page to @entry. | |
5650 | */ | |
5651 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
5652 | { | |
5653 | struct mem_cgroup *memcg; | |
5654 | unsigned short oldid; | |
5655 | ||
5656 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5657 | VM_BUG_ON_PAGE(page_count(page), page); | |
5658 | ||
5659 | if (!do_swap_account) | |
5660 | return; | |
5661 | ||
5662 | memcg = page->mem_cgroup; | |
5663 | ||
5664 | /* Readahead page, never charged */ | |
5665 | if (!memcg) | |
5666 | return; | |
5667 | ||
5668 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); | |
5669 | VM_BUG_ON_PAGE(oldid, page); | |
5670 | mem_cgroup_swap_statistics(memcg, true); | |
5671 | ||
5672 | page->mem_cgroup = NULL; | |
5673 | ||
5674 | if (!mem_cgroup_is_root(memcg)) | |
5675 | page_counter_uncharge(&memcg->memory, 1); | |
5676 | ||
ce9ce665 SAS |
5677 | /* |
5678 | * Interrupts should be disabled here because the caller holds the | |
5679 | * mapping->tree_lock lock which is taken with interrupts-off. It is | |
5680 | * important here to have the interrupts disabled because it is the | |
5681 | * only synchronisation we have for udpating the per-CPU variables. | |
5682 | */ | |
5683 | VM_BUG_ON(!irqs_disabled()); | |
21afa38e JW |
5684 | mem_cgroup_charge_statistics(memcg, page, -1); |
5685 | memcg_check_events(memcg, page); | |
5686 | } | |
5687 | ||
5688 | /** | |
5689 | * mem_cgroup_uncharge_swap - uncharge a swap entry | |
5690 | * @entry: swap entry to uncharge | |
5691 | * | |
5692 | * Drop the memsw charge associated with @entry. | |
5693 | */ | |
5694 | void mem_cgroup_uncharge_swap(swp_entry_t entry) | |
5695 | { | |
5696 | struct mem_cgroup *memcg; | |
5697 | unsigned short id; | |
5698 | ||
5699 | if (!do_swap_account) | |
5700 | return; | |
5701 | ||
5702 | id = swap_cgroup_record(entry, 0); | |
5703 | rcu_read_lock(); | |
adbe427b | 5704 | memcg = mem_cgroup_from_id(id); |
21afa38e JW |
5705 | if (memcg) { |
5706 | if (!mem_cgroup_is_root(memcg)) | |
5707 | page_counter_uncharge(&memcg->memsw, 1); | |
5708 | mem_cgroup_swap_statistics(memcg, false); | |
5709 | css_put(&memcg->css); | |
5710 | } | |
5711 | rcu_read_unlock(); | |
5712 | } | |
5713 | ||
5714 | /* for remember boot option*/ | |
5715 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
5716 | static int really_do_swap_account __initdata = 1; | |
5717 | #else | |
5718 | static int really_do_swap_account __initdata; | |
5719 | #endif | |
5720 | ||
5721 | static int __init enable_swap_account(char *s) | |
5722 | { | |
5723 | if (!strcmp(s, "1")) | |
5724 | really_do_swap_account = 1; | |
5725 | else if (!strcmp(s, "0")) | |
5726 | really_do_swap_account = 0; | |
5727 | return 1; | |
5728 | } | |
5729 | __setup("swapaccount=", enable_swap_account); | |
5730 | ||
5731 | static struct cftype memsw_cgroup_files[] = { | |
5732 | { | |
5733 | .name = "memsw.usage_in_bytes", | |
5734 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
5735 | .read_u64 = mem_cgroup_read_u64, | |
5736 | }, | |
5737 | { | |
5738 | .name = "memsw.max_usage_in_bytes", | |
5739 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
5740 | .write = mem_cgroup_reset, | |
5741 | .read_u64 = mem_cgroup_read_u64, | |
5742 | }, | |
5743 | { | |
5744 | .name = "memsw.limit_in_bytes", | |
5745 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
5746 | .write = mem_cgroup_write, | |
5747 | .read_u64 = mem_cgroup_read_u64, | |
5748 | }, | |
5749 | { | |
5750 | .name = "memsw.failcnt", | |
5751 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
5752 | .write = mem_cgroup_reset, | |
5753 | .read_u64 = mem_cgroup_read_u64, | |
5754 | }, | |
5755 | { }, /* terminate */ | |
5756 | }; | |
5757 | ||
5758 | static int __init mem_cgroup_swap_init(void) | |
5759 | { | |
5760 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
5761 | do_swap_account = 1; | |
5762 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, | |
5763 | memsw_cgroup_files)); | |
5764 | } | |
5765 | return 0; | |
5766 | } | |
5767 | subsys_initcall(mem_cgroup_swap_init); | |
5768 | ||
5769 | #endif /* CONFIG_MEMCG_SWAP */ |