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> |
08e552c6 | 65 | #include "internal.h" |
d1a4c0b3 | 66 | #include <net/sock.h> |
4bd2c1ee | 67 | #include <net/ip.h> |
d1a4c0b3 | 68 | #include <net/tcp_memcontrol.h> |
f35c3a8e | 69 | #include "slab.h" |
8cdea7c0 | 70 | |
8697d331 BS |
71 | #include <asm/uaccess.h> |
72 | ||
cc8e970c KM |
73 | #include <trace/events/vmscan.h> |
74 | ||
073219e9 TH |
75 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
76 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 77 | |
a181b0e8 | 78 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
6bbda35c | 79 | static struct mem_cgroup *root_mem_cgroup __read_mostly; |
56161634 | 80 | struct cgroup_subsys_state *mem_cgroup_root_css __read_mostly; |
8cdea7c0 | 81 | |
21afa38e | 82 | /* Whether the swap controller is active */ |
c255a458 | 83 | #ifdef CONFIG_MEMCG_SWAP |
c077719b | 84 | int do_swap_account __read_mostly; |
c077719b | 85 | #else |
a0db00fc | 86 | #define do_swap_account 0 |
c077719b KH |
87 | #endif |
88 | ||
af7c4b0e JW |
89 | static const char * const mem_cgroup_stat_names[] = { |
90 | "cache", | |
91 | "rss", | |
b070e65c | 92 | "rss_huge", |
af7c4b0e | 93 | "mapped_file", |
c4843a75 | 94 | "dirty", |
3ea67d06 | 95 | "writeback", |
af7c4b0e JW |
96 | "swap", |
97 | }; | |
98 | ||
af7c4b0e JW |
99 | static const char * const mem_cgroup_events_names[] = { |
100 | "pgpgin", | |
101 | "pgpgout", | |
102 | "pgfault", | |
103 | "pgmajfault", | |
104 | }; | |
105 | ||
58cf188e SZ |
106 | static const char * const mem_cgroup_lru_names[] = { |
107 | "inactive_anon", | |
108 | "active_anon", | |
109 | "inactive_file", | |
110 | "active_file", | |
111 | "unevictable", | |
112 | }; | |
113 | ||
a0db00fc KS |
114 | #define THRESHOLDS_EVENTS_TARGET 128 |
115 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
116 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 117 | |
bb4cc1a8 AM |
118 | /* |
119 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
120 | * their hierarchy representation | |
121 | */ | |
122 | ||
123 | struct mem_cgroup_tree_per_zone { | |
124 | struct rb_root rb_root; | |
125 | spinlock_t lock; | |
126 | }; | |
127 | ||
128 | struct mem_cgroup_tree_per_node { | |
129 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
130 | }; | |
131 | ||
132 | struct mem_cgroup_tree { | |
133 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
134 | }; | |
135 | ||
136 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
137 | ||
9490ff27 KH |
138 | /* for OOM */ |
139 | struct mem_cgroup_eventfd_list { | |
140 | struct list_head list; | |
141 | struct eventfd_ctx *eventfd; | |
142 | }; | |
2e72b634 | 143 | |
79bd9814 TH |
144 | /* |
145 | * cgroup_event represents events which userspace want to receive. | |
146 | */ | |
3bc942f3 | 147 | struct mem_cgroup_event { |
79bd9814 | 148 | /* |
59b6f873 | 149 | * memcg which the event belongs to. |
79bd9814 | 150 | */ |
59b6f873 | 151 | struct mem_cgroup *memcg; |
79bd9814 TH |
152 | /* |
153 | * eventfd to signal userspace about the event. | |
154 | */ | |
155 | struct eventfd_ctx *eventfd; | |
156 | /* | |
157 | * Each of these stored in a list by the cgroup. | |
158 | */ | |
159 | struct list_head list; | |
fba94807 TH |
160 | /* |
161 | * register_event() callback will be used to add new userspace | |
162 | * waiter for changes related to this event. Use eventfd_signal() | |
163 | * on eventfd to send notification to userspace. | |
164 | */ | |
59b6f873 | 165 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 166 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
167 | /* |
168 | * unregister_event() callback will be called when userspace closes | |
169 | * the eventfd or on cgroup removing. This callback must be set, | |
170 | * if you want provide notification functionality. | |
171 | */ | |
59b6f873 | 172 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 173 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
174 | /* |
175 | * All fields below needed to unregister event when | |
176 | * userspace closes eventfd. | |
177 | */ | |
178 | poll_table pt; | |
179 | wait_queue_head_t *wqh; | |
180 | wait_queue_t wait; | |
181 | struct work_struct remove; | |
182 | }; | |
183 | ||
c0ff4b85 R |
184 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
185 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 186 | |
7dc74be0 DN |
187 | /* Stuffs for move charges at task migration. */ |
188 | /* | |
1dfab5ab | 189 | * Types of charges to be moved. |
7dc74be0 | 190 | */ |
1dfab5ab JW |
191 | #define MOVE_ANON 0x1U |
192 | #define MOVE_FILE 0x2U | |
193 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 194 | |
4ffef5fe DN |
195 | /* "mc" and its members are protected by cgroup_mutex */ |
196 | static struct move_charge_struct { | |
b1dd693e | 197 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
198 | struct mem_cgroup *from; |
199 | struct mem_cgroup *to; | |
1dfab5ab | 200 | unsigned long flags; |
4ffef5fe | 201 | unsigned long precharge; |
854ffa8d | 202 | unsigned long moved_charge; |
483c30b5 | 203 | unsigned long moved_swap; |
8033b97c DN |
204 | struct task_struct *moving_task; /* a task moving charges */ |
205 | wait_queue_head_t waitq; /* a waitq for other context */ | |
206 | } mc = { | |
2bd9bb20 | 207 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
208 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
209 | }; | |
4ffef5fe | 210 | |
4e416953 BS |
211 | /* |
212 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
213 | * limit reclaim to prevent infinite loops, if they ever occur. | |
214 | */ | |
a0db00fc | 215 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 216 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 217 | |
217bc319 KH |
218 | enum charge_type { |
219 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 220 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 221 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 222 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
223 | NR_CHARGE_TYPE, |
224 | }; | |
225 | ||
8c7c6e34 | 226 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
227 | enum res_type { |
228 | _MEM, | |
229 | _MEMSWAP, | |
230 | _OOM_TYPE, | |
510fc4e1 | 231 | _KMEM, |
86ae53e1 GC |
232 | }; |
233 | ||
a0db00fc KS |
234 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
235 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 236 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
237 | /* Used for OOM nofiier */ |
238 | #define OOM_CONTROL (0) | |
8c7c6e34 | 239 | |
0999821b GC |
240 | /* |
241 | * The memcg_create_mutex will be held whenever a new cgroup is created. | |
242 | * As a consequence, any change that needs to protect against new child cgroups | |
243 | * appearing has to hold it as well. | |
244 | */ | |
245 | static DEFINE_MUTEX(memcg_create_mutex); | |
246 | ||
70ddf637 AV |
247 | /* Some nice accessors for the vmpressure. */ |
248 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
249 | { | |
250 | if (!memcg) | |
251 | memcg = root_mem_cgroup; | |
252 | return &memcg->vmpressure; | |
253 | } | |
254 | ||
255 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
256 | { | |
257 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
258 | } | |
259 | ||
7ffc0edc MH |
260 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
261 | { | |
262 | return (memcg == root_mem_cgroup); | |
263 | } | |
264 | ||
4219b2da LZ |
265 | /* |
266 | * We restrict the id in the range of [1, 65535], so it can fit into | |
267 | * an unsigned short. | |
268 | */ | |
269 | #define MEM_CGROUP_ID_MAX USHRT_MAX | |
270 | ||
34c00c31 LZ |
271 | static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) |
272 | { | |
15a4c835 | 273 | return memcg->css.id; |
34c00c31 LZ |
274 | } |
275 | ||
adbe427b VD |
276 | /* |
277 | * A helper function to get mem_cgroup from ID. must be called under | |
278 | * rcu_read_lock(). The caller is responsible for calling | |
279 | * css_tryget_online() if the mem_cgroup is used for charging. (dropping | |
280 | * refcnt from swap can be called against removed memcg.) | |
281 | */ | |
34c00c31 LZ |
282 | static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) |
283 | { | |
284 | struct cgroup_subsys_state *css; | |
285 | ||
7d699ddb | 286 | css = css_from_id(id, &memory_cgrp_subsys); |
34c00c31 LZ |
287 | return mem_cgroup_from_css(css); |
288 | } | |
289 | ||
e1aab161 | 290 | /* Writing them here to avoid exposing memcg's inner layout */ |
4bd2c1ee | 291 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
e1aab161 | 292 | |
e1aab161 GC |
293 | void sock_update_memcg(struct sock *sk) |
294 | { | |
376be5ff | 295 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 296 | struct mem_cgroup *memcg; |
3f134619 | 297 | struct cg_proto *cg_proto; |
e1aab161 GC |
298 | |
299 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
300 | ||
f3f511e1 GC |
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 | */ | |
309 | if (sk->sk_cgrp) { | |
310 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
5347e5ae | 311 | css_get(&sk->sk_cgrp->memcg->css); |
f3f511e1 GC |
312 | return; |
313 | } | |
314 | ||
e1aab161 GC |
315 | rcu_read_lock(); |
316 | memcg = mem_cgroup_from_task(current); | |
3f134619 | 317 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
e752eb68 | 318 | if (cg_proto && test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags) && |
ec903c0c | 319 | css_tryget_online(&memcg->css)) { |
3f134619 | 320 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
321 | } |
322 | rcu_read_unlock(); | |
323 | } | |
324 | } | |
325 | EXPORT_SYMBOL(sock_update_memcg); | |
326 | ||
327 | void sock_release_memcg(struct sock *sk) | |
328 | { | |
376be5ff | 329 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
330 | struct mem_cgroup *memcg; |
331 | WARN_ON(!sk->sk_cgrp->memcg); | |
332 | memcg = sk->sk_cgrp->memcg; | |
5347e5ae | 333 | css_put(&sk->sk_cgrp->memcg->css); |
e1aab161 GC |
334 | } |
335 | } | |
d1a4c0b3 GC |
336 | |
337 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
338 | { | |
339 | if (!memcg || mem_cgroup_is_root(memcg)) | |
340 | return NULL; | |
341 | ||
2e685cad | 342 | return &memcg->tcp_mem; |
d1a4c0b3 GC |
343 | } |
344 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 345 | |
3f134619 GC |
346 | #endif |
347 | ||
a8964b9b | 348 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 | 349 | /* |
f7ce3190 | 350 | * This will be the memcg's index in each cache's ->memcg_params.memcg_caches. |
b8627835 LZ |
351 | * The main reason for not using cgroup id for this: |
352 | * this works better in sparse environments, where we have a lot of memcgs, | |
353 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
354 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
355 | * 200 entry array for that. | |
55007d84 | 356 | * |
dbcf73e2 VD |
357 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
358 | * will double each time we have to increase it. | |
55007d84 | 359 | */ |
dbcf73e2 VD |
360 | static DEFINE_IDA(memcg_cache_ida); |
361 | int memcg_nr_cache_ids; | |
749c5415 | 362 | |
05257a1a VD |
363 | /* Protects memcg_nr_cache_ids */ |
364 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
365 | ||
366 | void memcg_get_cache_ids(void) | |
367 | { | |
368 | down_read(&memcg_cache_ids_sem); | |
369 | } | |
370 | ||
371 | void memcg_put_cache_ids(void) | |
372 | { | |
373 | up_read(&memcg_cache_ids_sem); | |
374 | } | |
375 | ||
55007d84 GC |
376 | /* |
377 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
378 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
379 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
380 | * tunable, but that is strictly not necessary. | |
381 | * | |
b8627835 | 382 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
383 | * this constant directly from cgroup, but it is understandable that this is |
384 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 385 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
386 | * increase ours as well if it increases. |
387 | */ | |
388 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 389 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 390 | |
d7f25f8a GC |
391 | /* |
392 | * A lot of the calls to the cache allocation functions are expected to be | |
393 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
394 | * conditional to this static branch, we'll have to allow modules that does | |
395 | * kmem_cache_alloc and the such to see this symbol as well | |
396 | */ | |
a8964b9b | 397 | struct static_key memcg_kmem_enabled_key; |
d7f25f8a | 398 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 399 | |
a8964b9b GC |
400 | #endif /* CONFIG_MEMCG_KMEM */ |
401 | ||
f64c3f54 | 402 | static struct mem_cgroup_per_zone * |
e231875b | 403 | mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone) |
f64c3f54 | 404 | { |
e231875b JZ |
405 | int nid = zone_to_nid(zone); |
406 | int zid = zone_idx(zone); | |
407 | ||
54f72fe0 | 408 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
409 | } |
410 | ||
ad7fa852 TH |
411 | /** |
412 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
413 | * @page: page of interest | |
414 | * | |
415 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
416 | * with @page is returned. The returned css remains associated with @page | |
417 | * until it is released. | |
418 | * | |
419 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
420 | * is returned. | |
421 | * | |
422 | * XXX: The above description of behavior on the default hierarchy isn't | |
423 | * strictly true yet as replace_page_cache_page() can modify the | |
424 | * association before @page is released even on the default hierarchy; | |
425 | * however, the current and planned usages don't mix the the two functions | |
426 | * and replace_page_cache_page() will soon be updated to make the invariant | |
427 | * actually true. | |
428 | */ | |
429 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
430 | { | |
431 | struct mem_cgroup *memcg; | |
432 | ||
433 | rcu_read_lock(); | |
434 | ||
435 | memcg = page->mem_cgroup; | |
436 | ||
437 | if (!memcg || !cgroup_on_dfl(memcg->css.cgroup)) | |
438 | memcg = root_mem_cgroup; | |
439 | ||
440 | rcu_read_unlock(); | |
441 | return &memcg->css; | |
442 | } | |
443 | ||
2fc04524 VD |
444 | /** |
445 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
446 | * @page: the page | |
447 | * | |
448 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
449 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
450 | * is safe to call this function without holding a reference to @page. | |
451 | * | |
452 | * Note, this function is inherently racy, because there is nothing to prevent | |
453 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
454 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
455 | * do not care (such as procfs interfaces). | |
456 | */ | |
457 | ino_t page_cgroup_ino(struct page *page) | |
458 | { | |
459 | struct mem_cgroup *memcg; | |
460 | unsigned long ino = 0; | |
461 | ||
462 | rcu_read_lock(); | |
463 | memcg = READ_ONCE(page->mem_cgroup); | |
464 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) | |
465 | memcg = parent_mem_cgroup(memcg); | |
466 | if (memcg) | |
467 | ino = cgroup_ino(memcg->css.cgroup); | |
468 | rcu_read_unlock(); | |
469 | return ino; | |
470 | } | |
471 | ||
f64c3f54 | 472 | static struct mem_cgroup_per_zone * |
e231875b | 473 | mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 474 | { |
97a6c37b JW |
475 | int nid = page_to_nid(page); |
476 | int zid = page_zonenum(page); | |
f64c3f54 | 477 | |
e231875b | 478 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
479 | } |
480 | ||
bb4cc1a8 AM |
481 | static struct mem_cgroup_tree_per_zone * |
482 | soft_limit_tree_node_zone(int nid, int zid) | |
483 | { | |
484 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
485 | } | |
486 | ||
487 | static struct mem_cgroup_tree_per_zone * | |
488 | soft_limit_tree_from_page(struct page *page) | |
489 | { | |
490 | int nid = page_to_nid(page); | |
491 | int zid = page_zonenum(page); | |
492 | ||
493 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
494 | } | |
495 | ||
cf2c8127 JW |
496 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz, |
497 | struct mem_cgroup_tree_per_zone *mctz, | |
3e32cb2e | 498 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
499 | { |
500 | struct rb_node **p = &mctz->rb_root.rb_node; | |
501 | struct rb_node *parent = NULL; | |
502 | struct mem_cgroup_per_zone *mz_node; | |
503 | ||
504 | if (mz->on_tree) | |
505 | return; | |
506 | ||
507 | mz->usage_in_excess = new_usage_in_excess; | |
508 | if (!mz->usage_in_excess) | |
509 | return; | |
510 | while (*p) { | |
511 | parent = *p; | |
512 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
513 | tree_node); | |
514 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
515 | p = &(*p)->rb_left; | |
516 | /* | |
517 | * We can't avoid mem cgroups that are over their soft | |
518 | * limit by the same amount | |
519 | */ | |
520 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
521 | p = &(*p)->rb_right; | |
522 | } | |
523 | rb_link_node(&mz->tree_node, parent, p); | |
524 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
525 | mz->on_tree = true; | |
526 | } | |
527 | ||
cf2c8127 JW |
528 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
529 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 AM |
530 | { |
531 | if (!mz->on_tree) | |
532 | return; | |
533 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
534 | mz->on_tree = false; | |
535 | } | |
536 | ||
cf2c8127 JW |
537 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
538 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 | 539 | { |
0a31bc97 JW |
540 | unsigned long flags; |
541 | ||
542 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 543 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 544 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
545 | } |
546 | ||
3e32cb2e JW |
547 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
548 | { | |
549 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 550 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
551 | unsigned long excess = 0; |
552 | ||
553 | if (nr_pages > soft_limit) | |
554 | excess = nr_pages - soft_limit; | |
555 | ||
556 | return excess; | |
557 | } | |
bb4cc1a8 AM |
558 | |
559 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
560 | { | |
3e32cb2e | 561 | unsigned long excess; |
bb4cc1a8 AM |
562 | struct mem_cgroup_per_zone *mz; |
563 | struct mem_cgroup_tree_per_zone *mctz; | |
bb4cc1a8 | 564 | |
e231875b | 565 | mctz = soft_limit_tree_from_page(page); |
bb4cc1a8 AM |
566 | /* |
567 | * Necessary to update all ancestors when hierarchy is used. | |
568 | * because their event counter is not touched. | |
569 | */ | |
570 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
e231875b | 571 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
3e32cb2e | 572 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
573 | /* |
574 | * We have to update the tree if mz is on RB-tree or | |
575 | * mem is over its softlimit. | |
576 | */ | |
577 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
578 | unsigned long flags; |
579 | ||
580 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
581 | /* if on-tree, remove it */ |
582 | if (mz->on_tree) | |
cf2c8127 | 583 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
584 | /* |
585 | * Insert again. mz->usage_in_excess will be updated. | |
586 | * If excess is 0, no tree ops. | |
587 | */ | |
cf2c8127 | 588 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 589 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
590 | } |
591 | } | |
592 | } | |
593 | ||
594 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
595 | { | |
bb4cc1a8 | 596 | struct mem_cgroup_tree_per_zone *mctz; |
e231875b JZ |
597 | struct mem_cgroup_per_zone *mz; |
598 | int nid, zid; | |
bb4cc1a8 | 599 | |
e231875b JZ |
600 | for_each_node(nid) { |
601 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
602 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
603 | mctz = soft_limit_tree_node_zone(nid, zid); | |
cf2c8127 | 604 | mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
605 | } |
606 | } | |
607 | } | |
608 | ||
609 | static struct mem_cgroup_per_zone * | |
610 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
611 | { | |
612 | struct rb_node *rightmost = NULL; | |
613 | struct mem_cgroup_per_zone *mz; | |
614 | ||
615 | retry: | |
616 | mz = NULL; | |
617 | rightmost = rb_last(&mctz->rb_root); | |
618 | if (!rightmost) | |
619 | goto done; /* Nothing to reclaim from */ | |
620 | ||
621 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
622 | /* | |
623 | * Remove the node now but someone else can add it back, | |
624 | * we will to add it back at the end of reclaim to its correct | |
625 | * position in the tree. | |
626 | */ | |
cf2c8127 | 627 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 628 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 629 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
630 | goto retry; |
631 | done: | |
632 | return mz; | |
633 | } | |
634 | ||
635 | static struct mem_cgroup_per_zone * | |
636 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
637 | { | |
638 | struct mem_cgroup_per_zone *mz; | |
639 | ||
0a31bc97 | 640 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 641 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 642 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
643 | return mz; |
644 | } | |
645 | ||
711d3d2c KH |
646 | /* |
647 | * Implementation Note: reading percpu statistics for memcg. | |
648 | * | |
649 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
650 | * synchronization to implement "quick" read. There are trade-off between | |
651 | * reading cost and precision of value. Then, we may have a chance to implement | |
652 | * a periodic synchronizion of counter in memcg's counter. | |
653 | * | |
654 | * But this _read() function is used for user interface now. The user accounts | |
655 | * memory usage by memory cgroup and he _always_ requires exact value because | |
656 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
657 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
658 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
659 | * | |
660 | * If there are kernel internal actions which can make use of some not-exact | |
661 | * value, and reading all cpu value can be performance bottleneck in some | |
662 | * common workload, threashold and synchonization as vmstat[] should be | |
663 | * implemented. | |
664 | */ | |
c0ff4b85 | 665 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 666 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 667 | { |
7a159cc9 | 668 | long val = 0; |
c62b1a3b | 669 | int cpu; |
c62b1a3b | 670 | |
733a572e | 671 | for_each_possible_cpu(cpu) |
c0ff4b85 | 672 | val += per_cpu(memcg->stat->count[idx], cpu); |
c62b1a3b KH |
673 | return val; |
674 | } | |
675 | ||
c0ff4b85 | 676 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
677 | enum mem_cgroup_events_index idx) |
678 | { | |
679 | unsigned long val = 0; | |
680 | int cpu; | |
681 | ||
733a572e | 682 | for_each_possible_cpu(cpu) |
c0ff4b85 | 683 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f JW |
684 | return val; |
685 | } | |
686 | ||
c0ff4b85 | 687 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 688 | struct page *page, |
0a31bc97 | 689 | int nr_pages) |
d52aa412 | 690 | { |
b2402857 KH |
691 | /* |
692 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
693 | * counted as CACHE even if it's on ANON LRU. | |
694 | */ | |
0a31bc97 | 695 | if (PageAnon(page)) |
b2402857 | 696 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
c0ff4b85 | 697 | nr_pages); |
d52aa412 | 698 | else |
b2402857 | 699 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 700 | nr_pages); |
55e462b0 | 701 | |
b070e65c DR |
702 | if (PageTransHuge(page)) |
703 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], | |
704 | nr_pages); | |
705 | ||
e401f176 KH |
706 | /* pagein of a big page is an event. So, ignore page size */ |
707 | if (nr_pages > 0) | |
c0ff4b85 | 708 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 709 | else { |
c0ff4b85 | 710 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
711 | nr_pages = -nr_pages; /* for event */ |
712 | } | |
e401f176 | 713 | |
13114716 | 714 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
6d12e2d8 KH |
715 | } |
716 | ||
e231875b JZ |
717 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
718 | int nid, | |
719 | unsigned int lru_mask) | |
bb2a0de9 | 720 | { |
e231875b | 721 | unsigned long nr = 0; |
889976db YH |
722 | int zid; |
723 | ||
e231875b | 724 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 725 | |
e231875b JZ |
726 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
727 | struct mem_cgroup_per_zone *mz; | |
728 | enum lru_list lru; | |
729 | ||
730 | for_each_lru(lru) { | |
731 | if (!(BIT(lru) & lru_mask)) | |
732 | continue; | |
733 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
734 | nr += mz->lru_size[lru]; | |
735 | } | |
736 | } | |
737 | return nr; | |
889976db | 738 | } |
bb2a0de9 | 739 | |
c0ff4b85 | 740 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 741 | unsigned int lru_mask) |
6d12e2d8 | 742 | { |
e231875b | 743 | unsigned long nr = 0; |
889976db | 744 | int nid; |
6d12e2d8 | 745 | |
31aaea4a | 746 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
747 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
748 | return nr; | |
d52aa412 KH |
749 | } |
750 | ||
f53d7ce3 JW |
751 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
752 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
753 | { |
754 | unsigned long val, next; | |
755 | ||
13114716 | 756 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 757 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 758 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
759 | if ((long)next - (long)val < 0) { |
760 | switch (target) { | |
761 | case MEM_CGROUP_TARGET_THRESH: | |
762 | next = val + THRESHOLDS_EVENTS_TARGET; | |
763 | break; | |
bb4cc1a8 AM |
764 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
765 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
766 | break; | |
f53d7ce3 JW |
767 | case MEM_CGROUP_TARGET_NUMAINFO: |
768 | next = val + NUMAINFO_EVENTS_TARGET; | |
769 | break; | |
770 | default: | |
771 | break; | |
772 | } | |
773 | __this_cpu_write(memcg->stat->targets[target], next); | |
774 | return true; | |
7a159cc9 | 775 | } |
f53d7ce3 | 776 | return false; |
d2265e6f KH |
777 | } |
778 | ||
779 | /* | |
780 | * Check events in order. | |
781 | * | |
782 | */ | |
c0ff4b85 | 783 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
784 | { |
785 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
786 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
787 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 788 | bool do_softlimit; |
82b3f2a7 | 789 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 790 | |
bb4cc1a8 AM |
791 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
792 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
793 | #if MAX_NUMNODES > 1 |
794 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
795 | MEM_CGROUP_TARGET_NUMAINFO); | |
796 | #endif | |
c0ff4b85 | 797 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
798 | if (unlikely(do_softlimit)) |
799 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 800 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 801 | if (unlikely(do_numainfo)) |
c0ff4b85 | 802 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 803 | #endif |
0a31bc97 | 804 | } |
d2265e6f KH |
805 | } |
806 | ||
cf475ad2 | 807 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 808 | { |
31a78f23 BS |
809 | /* |
810 | * mm_update_next_owner() may clear mm->owner to NULL | |
811 | * if it races with swapoff, page migration, etc. | |
812 | * So this can be called with p == NULL. | |
813 | */ | |
814 | if (unlikely(!p)) | |
815 | return NULL; | |
816 | ||
073219e9 | 817 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 818 | } |
33398cf2 | 819 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 820 | |
df381975 | 821 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 822 | { |
c0ff4b85 | 823 | struct mem_cgroup *memcg = NULL; |
0b7f569e | 824 | |
54595fe2 KH |
825 | rcu_read_lock(); |
826 | do { | |
6f6acb00 MH |
827 | /* |
828 | * Page cache insertions can happen withou an | |
829 | * actual mm context, e.g. during disk probing | |
830 | * on boot, loopback IO, acct() writes etc. | |
831 | */ | |
832 | if (unlikely(!mm)) | |
df381975 | 833 | memcg = root_mem_cgroup; |
6f6acb00 MH |
834 | else { |
835 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
836 | if (unlikely(!memcg)) | |
837 | memcg = root_mem_cgroup; | |
838 | } | |
ec903c0c | 839 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 840 | rcu_read_unlock(); |
c0ff4b85 | 841 | return memcg; |
54595fe2 KH |
842 | } |
843 | ||
5660048c JW |
844 | /** |
845 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
846 | * @root: hierarchy root | |
847 | * @prev: previously returned memcg, NULL on first invocation | |
848 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
849 | * | |
850 | * Returns references to children of the hierarchy below @root, or | |
851 | * @root itself, or %NULL after a full round-trip. | |
852 | * | |
853 | * Caller must pass the return value in @prev on subsequent | |
854 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
855 | * to cancel a hierarchy walk before the round-trip is complete. | |
856 | * | |
857 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
858 | * divide up the memcgs in the hierarchy among all concurrent | |
859 | * reclaimers operating on the same zone and priority. | |
860 | */ | |
694fbc0f | 861 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 862 | struct mem_cgroup *prev, |
694fbc0f | 863 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 864 | { |
33398cf2 | 865 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
5ac8fb31 | 866 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 867 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 868 | struct mem_cgroup *pos = NULL; |
711d3d2c | 869 | |
694fbc0f AM |
870 | if (mem_cgroup_disabled()) |
871 | return NULL; | |
5660048c | 872 | |
9f3a0d09 JW |
873 | if (!root) |
874 | root = root_mem_cgroup; | |
7d74b06f | 875 | |
9f3a0d09 | 876 | if (prev && !reclaim) |
5ac8fb31 | 877 | pos = prev; |
14067bb3 | 878 | |
9f3a0d09 JW |
879 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
880 | if (prev) | |
5ac8fb31 | 881 | goto out; |
694fbc0f | 882 | return root; |
9f3a0d09 | 883 | } |
14067bb3 | 884 | |
542f85f9 | 885 | rcu_read_lock(); |
5f578161 | 886 | |
5ac8fb31 JW |
887 | if (reclaim) { |
888 | struct mem_cgroup_per_zone *mz; | |
889 | ||
890 | mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone); | |
891 | iter = &mz->iter[reclaim->priority]; | |
892 | ||
893 | if (prev && reclaim->generation != iter->generation) | |
894 | goto out_unlock; | |
895 | ||
896 | do { | |
4db0c3c2 | 897 | pos = READ_ONCE(iter->position); |
5ac8fb31 JW |
898 | /* |
899 | * A racing update may change the position and | |
900 | * put the last reference, hence css_tryget(), | |
901 | * or retry to see the updated position. | |
902 | */ | |
903 | } while (pos && !css_tryget(&pos->css)); | |
904 | } | |
905 | ||
906 | if (pos) | |
907 | css = &pos->css; | |
908 | ||
909 | for (;;) { | |
910 | css = css_next_descendant_pre(css, &root->css); | |
911 | if (!css) { | |
912 | /* | |
913 | * Reclaimers share the hierarchy walk, and a | |
914 | * new one might jump in right at the end of | |
915 | * the hierarchy - make sure they see at least | |
916 | * one group and restart from the beginning. | |
917 | */ | |
918 | if (!prev) | |
919 | continue; | |
920 | break; | |
527a5ec9 | 921 | } |
7d74b06f | 922 | |
5ac8fb31 JW |
923 | /* |
924 | * Verify the css and acquire a reference. The root | |
925 | * is provided by the caller, so we know it's alive | |
926 | * and kicking, and don't take an extra reference. | |
927 | */ | |
928 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 929 | |
5ac8fb31 JW |
930 | if (css == &root->css) |
931 | break; | |
14067bb3 | 932 | |
b2052564 | 933 | if (css_tryget(css)) { |
5ac8fb31 JW |
934 | /* |
935 | * Make sure the memcg is initialized: | |
936 | * mem_cgroup_css_online() orders the the | |
937 | * initialization against setting the flag. | |
938 | */ | |
939 | if (smp_load_acquire(&memcg->initialized)) | |
940 | break; | |
542f85f9 | 941 | |
5ac8fb31 | 942 | css_put(css); |
527a5ec9 | 943 | } |
9f3a0d09 | 944 | |
5ac8fb31 | 945 | memcg = NULL; |
9f3a0d09 | 946 | } |
5ac8fb31 JW |
947 | |
948 | if (reclaim) { | |
949 | if (cmpxchg(&iter->position, pos, memcg) == pos) { | |
950 | if (memcg) | |
951 | css_get(&memcg->css); | |
952 | if (pos) | |
953 | css_put(&pos->css); | |
954 | } | |
955 | ||
956 | /* | |
957 | * pairs with css_tryget when dereferencing iter->position | |
958 | * above. | |
959 | */ | |
960 | if (pos) | |
961 | css_put(&pos->css); | |
962 | ||
963 | if (!memcg) | |
964 | iter->generation++; | |
965 | else if (!prev) | |
966 | reclaim->generation = iter->generation; | |
9f3a0d09 | 967 | } |
5ac8fb31 | 968 | |
542f85f9 MH |
969 | out_unlock: |
970 | rcu_read_unlock(); | |
5ac8fb31 | 971 | out: |
c40046f3 MH |
972 | if (prev && prev != root) |
973 | css_put(&prev->css); | |
974 | ||
9f3a0d09 | 975 | return memcg; |
14067bb3 | 976 | } |
7d74b06f | 977 | |
5660048c JW |
978 | /** |
979 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
980 | * @root: hierarchy root | |
981 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
982 | */ | |
983 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
984 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
985 | { |
986 | if (!root) | |
987 | root = root_mem_cgroup; | |
988 | if (prev && prev != root) | |
989 | css_put(&prev->css); | |
990 | } | |
7d74b06f | 991 | |
9f3a0d09 JW |
992 | /* |
993 | * Iteration constructs for visiting all cgroups (under a tree). If | |
994 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
995 | * be used for reference counting. | |
996 | */ | |
997 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 998 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 999 | iter != NULL; \ |
527a5ec9 | 1000 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1001 | |
9f3a0d09 | 1002 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1003 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1004 | iter != NULL; \ |
527a5ec9 | 1005 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1006 | |
925b7673 JW |
1007 | /** |
1008 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1009 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1010 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1011 | * |
1012 | * Returns the lru list vector holding pages for the given @zone and | |
1013 | * @mem. This can be the global zone lruvec, if the memory controller | |
1014 | * is disabled. | |
1015 | */ | |
1016 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1017 | struct mem_cgroup *memcg) | |
1018 | { | |
1019 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1020 | struct lruvec *lruvec; |
925b7673 | 1021 | |
bea8c150 HD |
1022 | if (mem_cgroup_disabled()) { |
1023 | lruvec = &zone->lruvec; | |
1024 | goto out; | |
1025 | } | |
925b7673 | 1026 | |
e231875b | 1027 | mz = mem_cgroup_zone_zoneinfo(memcg, zone); |
bea8c150 HD |
1028 | lruvec = &mz->lruvec; |
1029 | out: | |
1030 | /* | |
1031 | * Since a node can be onlined after the mem_cgroup was created, | |
1032 | * we have to be prepared to initialize lruvec->zone here; | |
1033 | * and if offlined then reonlined, we need to reinitialize it. | |
1034 | */ | |
1035 | if (unlikely(lruvec->zone != zone)) | |
1036 | lruvec->zone = zone; | |
1037 | return lruvec; | |
925b7673 JW |
1038 | } |
1039 | ||
925b7673 | 1040 | /** |
dfe0e773 | 1041 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1042 | * @page: the page |
fa9add64 | 1043 | * @zone: zone of the page |
dfe0e773 JW |
1044 | * |
1045 | * This function is only safe when following the LRU page isolation | |
1046 | * and putback protocol: the LRU lock must be held, and the page must | |
1047 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 1048 | */ |
fa9add64 | 1049 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1050 | { |
08e552c6 | 1051 | struct mem_cgroup_per_zone *mz; |
925b7673 | 1052 | struct mem_cgroup *memcg; |
bea8c150 | 1053 | struct lruvec *lruvec; |
6d12e2d8 | 1054 | |
bea8c150 HD |
1055 | if (mem_cgroup_disabled()) { |
1056 | lruvec = &zone->lruvec; | |
1057 | goto out; | |
1058 | } | |
925b7673 | 1059 | |
1306a85a | 1060 | memcg = page->mem_cgroup; |
7512102c | 1061 | /* |
dfe0e773 | 1062 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1063 | * possibly migrated - before they are charged. |
7512102c | 1064 | */ |
29833315 JW |
1065 | if (!memcg) |
1066 | memcg = root_mem_cgroup; | |
7512102c | 1067 | |
e231875b | 1068 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
bea8c150 HD |
1069 | lruvec = &mz->lruvec; |
1070 | out: | |
1071 | /* | |
1072 | * Since a node can be onlined after the mem_cgroup was created, | |
1073 | * we have to be prepared to initialize lruvec->zone here; | |
1074 | * and if offlined then reonlined, we need to reinitialize it. | |
1075 | */ | |
1076 | if (unlikely(lruvec->zone != zone)) | |
1077 | lruvec->zone = zone; | |
1078 | return lruvec; | |
08e552c6 | 1079 | } |
b69408e8 | 1080 | |
925b7673 | 1081 | /** |
fa9add64 HD |
1082 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1083 | * @lruvec: mem_cgroup per zone lru vector | |
1084 | * @lru: index of lru list the page is sitting on | |
1085 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1086 | * |
fa9add64 HD |
1087 | * This function must be called when a page is added to or removed from an |
1088 | * lru list. | |
3f58a829 | 1089 | */ |
fa9add64 HD |
1090 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1091 | int nr_pages) | |
3f58a829 MK |
1092 | { |
1093 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1094 | unsigned long *lru_size; |
3f58a829 MK |
1095 | |
1096 | if (mem_cgroup_disabled()) | |
1097 | return; | |
1098 | ||
fa9add64 HD |
1099 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1100 | lru_size = mz->lru_size + lru; | |
1101 | *lru_size += nr_pages; | |
1102 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1103 | } |
544122e5 | 1104 | |
2314b42d | 1105 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
c3ac9a8a | 1106 | { |
2314b42d | 1107 | struct mem_cgroup *task_memcg; |
158e0a2d | 1108 | struct task_struct *p; |
ffbdccf5 | 1109 | bool ret; |
4c4a2214 | 1110 | |
158e0a2d | 1111 | p = find_lock_task_mm(task); |
de077d22 | 1112 | if (p) { |
2314b42d | 1113 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1114 | task_unlock(p); |
1115 | } else { | |
1116 | /* | |
1117 | * All threads may have already detached their mm's, but the oom | |
1118 | * killer still needs to detect if they have already been oom | |
1119 | * killed to prevent needlessly killing additional tasks. | |
1120 | */ | |
ffbdccf5 | 1121 | rcu_read_lock(); |
2314b42d JW |
1122 | task_memcg = mem_cgroup_from_task(task); |
1123 | css_get(&task_memcg->css); | |
ffbdccf5 | 1124 | rcu_read_unlock(); |
de077d22 | 1125 | } |
2314b42d JW |
1126 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1127 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1128 | return ret; |
1129 | } | |
1130 | ||
3e32cb2e | 1131 | #define mem_cgroup_from_counter(counter, member) \ |
6d61ef40 BS |
1132 | container_of(counter, struct mem_cgroup, member) |
1133 | ||
19942822 | 1134 | /** |
9d11ea9f | 1135 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1136 | * @memcg: the memory cgroup |
19942822 | 1137 | * |
9d11ea9f | 1138 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1139 | * pages. |
19942822 | 1140 | */ |
c0ff4b85 | 1141 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1142 | { |
3e32cb2e JW |
1143 | unsigned long margin = 0; |
1144 | unsigned long count; | |
1145 | unsigned long limit; | |
9d11ea9f | 1146 | |
3e32cb2e | 1147 | count = page_counter_read(&memcg->memory); |
4db0c3c2 | 1148 | limit = READ_ONCE(memcg->memory.limit); |
3e32cb2e JW |
1149 | if (count < limit) |
1150 | margin = limit - count; | |
1151 | ||
1152 | if (do_swap_account) { | |
1153 | count = page_counter_read(&memcg->memsw); | |
4db0c3c2 | 1154 | limit = READ_ONCE(memcg->memsw.limit); |
3e32cb2e JW |
1155 | if (count <= limit) |
1156 | margin = min(margin, limit - count); | |
1157 | } | |
1158 | ||
1159 | return margin; | |
19942822 JW |
1160 | } |
1161 | ||
32047e2a | 1162 | /* |
bdcbb659 | 1163 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1164 | * |
bdcbb659 QH |
1165 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1166 | * moving cgroups. This is for waiting at high-memory pressure | |
1167 | * caused by "move". | |
32047e2a | 1168 | */ |
c0ff4b85 | 1169 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1170 | { |
2bd9bb20 KH |
1171 | struct mem_cgroup *from; |
1172 | struct mem_cgroup *to; | |
4b534334 | 1173 | bool ret = false; |
2bd9bb20 KH |
1174 | /* |
1175 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1176 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1177 | */ | |
1178 | spin_lock(&mc.lock); | |
1179 | from = mc.from; | |
1180 | to = mc.to; | |
1181 | if (!from) | |
1182 | goto unlock; | |
3e92041d | 1183 | |
2314b42d JW |
1184 | ret = mem_cgroup_is_descendant(from, memcg) || |
1185 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1186 | unlock: |
1187 | spin_unlock(&mc.lock); | |
4b534334 KH |
1188 | return ret; |
1189 | } | |
1190 | ||
c0ff4b85 | 1191 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1192 | { |
1193 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1194 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1195 | DEFINE_WAIT(wait); |
1196 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1197 | /* moving charge context might have finished. */ | |
1198 | if (mc.moving_task) | |
1199 | schedule(); | |
1200 | finish_wait(&mc.waitq, &wait); | |
1201 | return true; | |
1202 | } | |
1203 | } | |
1204 | return false; | |
1205 | } | |
1206 | ||
58cf188e | 1207 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1208 | /** |
58cf188e | 1209 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1210 | * @memcg: The memory cgroup that went over limit |
1211 | * @p: Task that is going to be killed | |
1212 | * | |
1213 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1214 | * enabled | |
1215 | */ | |
1216 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1217 | { | |
e61734c5 | 1218 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
08088cb9 | 1219 | static DEFINE_MUTEX(oom_info_lock); |
58cf188e SZ |
1220 | struct mem_cgroup *iter; |
1221 | unsigned int i; | |
e222432b | 1222 | |
08088cb9 | 1223 | mutex_lock(&oom_info_lock); |
e222432b BS |
1224 | rcu_read_lock(); |
1225 | ||
2415b9f5 BV |
1226 | if (p) { |
1227 | pr_info("Task in "); | |
1228 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1229 | pr_cont(" killed as a result of limit of "); | |
1230 | } else { | |
1231 | pr_info("Memory limit reached of cgroup "); | |
1232 | } | |
1233 | ||
e61734c5 | 1234 | pr_cont_cgroup_path(memcg->css.cgroup); |
0346dadb | 1235 | pr_cont("\n"); |
e222432b | 1236 | |
e222432b BS |
1237 | rcu_read_unlock(); |
1238 | ||
3e32cb2e JW |
1239 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1240 | K((u64)page_counter_read(&memcg->memory)), | |
1241 | K((u64)memcg->memory.limit), memcg->memory.failcnt); | |
1242 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1243 | K((u64)page_counter_read(&memcg->memsw)), | |
1244 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); | |
1245 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1246 | K((u64)page_counter_read(&memcg->kmem)), | |
1247 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); | |
58cf188e SZ |
1248 | |
1249 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1250 | pr_info("Memory cgroup stats for "); |
1251 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1252 | pr_cont(":"); |
1253 | ||
1254 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1255 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1256 | continue; | |
1257 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], | |
1258 | K(mem_cgroup_read_stat(iter, i))); | |
1259 | } | |
1260 | ||
1261 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1262 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1263 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1264 | ||
1265 | pr_cont("\n"); | |
1266 | } | |
08088cb9 | 1267 | mutex_unlock(&oom_info_lock); |
e222432b BS |
1268 | } |
1269 | ||
81d39c20 KH |
1270 | /* |
1271 | * This function returns the number of memcg under hierarchy tree. Returns | |
1272 | * 1(self count) if no children. | |
1273 | */ | |
c0ff4b85 | 1274 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1275 | { |
1276 | int num = 0; | |
7d74b06f KH |
1277 | struct mem_cgroup *iter; |
1278 | ||
c0ff4b85 | 1279 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1280 | num++; |
81d39c20 KH |
1281 | return num; |
1282 | } | |
1283 | ||
a63d83f4 DR |
1284 | /* |
1285 | * Return the memory (and swap, if configured) limit for a memcg. | |
1286 | */ | |
3e32cb2e | 1287 | static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 | 1288 | { |
3e32cb2e | 1289 | unsigned long limit; |
f3e8eb70 | 1290 | |
3e32cb2e | 1291 | limit = memcg->memory.limit; |
9a5a8f19 | 1292 | if (mem_cgroup_swappiness(memcg)) { |
3e32cb2e | 1293 | unsigned long memsw_limit; |
9a5a8f19 | 1294 | |
3e32cb2e JW |
1295 | memsw_limit = memcg->memsw.limit; |
1296 | limit = min(limit + total_swap_pages, memsw_limit); | |
9a5a8f19 | 1297 | } |
9a5a8f19 | 1298 | return limit; |
a63d83f4 DR |
1299 | } |
1300 | ||
19965460 DR |
1301 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1302 | int order) | |
9cbb78bb | 1303 | { |
6e0fc46d DR |
1304 | struct oom_control oc = { |
1305 | .zonelist = NULL, | |
1306 | .nodemask = NULL, | |
1307 | .gfp_mask = gfp_mask, | |
1308 | .order = order, | |
6e0fc46d | 1309 | }; |
9cbb78bb DR |
1310 | struct mem_cgroup *iter; |
1311 | unsigned long chosen_points = 0; | |
1312 | unsigned long totalpages; | |
1313 | unsigned int points = 0; | |
1314 | struct task_struct *chosen = NULL; | |
1315 | ||
dc56401f JW |
1316 | mutex_lock(&oom_lock); |
1317 | ||
876aafbf | 1318 | /* |
465adcf1 DR |
1319 | * If current has a pending SIGKILL or is exiting, then automatically |
1320 | * select it. The goal is to allow it to allocate so that it may | |
1321 | * quickly exit and free its memory. | |
876aafbf | 1322 | */ |
d003f371 | 1323 | if (fatal_signal_pending(current) || task_will_free_mem(current)) { |
16e95196 | 1324 | mark_oom_victim(current); |
dc56401f | 1325 | goto unlock; |
876aafbf DR |
1326 | } |
1327 | ||
6e0fc46d | 1328 | check_panic_on_oom(&oc, CONSTRAINT_MEMCG, memcg); |
3e32cb2e | 1329 | totalpages = mem_cgroup_get_limit(memcg) ? : 1; |
9cbb78bb | 1330 | for_each_mem_cgroup_tree(iter, memcg) { |
72ec7029 | 1331 | struct css_task_iter it; |
9cbb78bb DR |
1332 | struct task_struct *task; |
1333 | ||
72ec7029 TH |
1334 | css_task_iter_start(&iter->css, &it); |
1335 | while ((task = css_task_iter_next(&it))) { | |
6e0fc46d | 1336 | switch (oom_scan_process_thread(&oc, task, totalpages)) { |
9cbb78bb DR |
1337 | case OOM_SCAN_SELECT: |
1338 | if (chosen) | |
1339 | put_task_struct(chosen); | |
1340 | chosen = task; | |
1341 | chosen_points = ULONG_MAX; | |
1342 | get_task_struct(chosen); | |
1343 | /* fall through */ | |
1344 | case OOM_SCAN_CONTINUE: | |
1345 | continue; | |
1346 | case OOM_SCAN_ABORT: | |
72ec7029 | 1347 | css_task_iter_end(&it); |
9cbb78bb DR |
1348 | mem_cgroup_iter_break(memcg, iter); |
1349 | if (chosen) | |
1350 | put_task_struct(chosen); | |
dc56401f | 1351 | goto unlock; |
9cbb78bb DR |
1352 | case OOM_SCAN_OK: |
1353 | break; | |
1354 | }; | |
1355 | points = oom_badness(task, memcg, NULL, totalpages); | |
d49ad935 DR |
1356 | if (!points || points < chosen_points) |
1357 | continue; | |
1358 | /* Prefer thread group leaders for display purposes */ | |
1359 | if (points == chosen_points && | |
1360 | thread_group_leader(chosen)) | |
1361 | continue; | |
1362 | ||
1363 | if (chosen) | |
1364 | put_task_struct(chosen); | |
1365 | chosen = task; | |
1366 | chosen_points = points; | |
1367 | get_task_struct(chosen); | |
9cbb78bb | 1368 | } |
72ec7029 | 1369 | css_task_iter_end(&it); |
9cbb78bb DR |
1370 | } |
1371 | ||
dc56401f JW |
1372 | if (chosen) { |
1373 | points = chosen_points * 1000 / totalpages; | |
6e0fc46d DR |
1374 | oom_kill_process(&oc, chosen, points, totalpages, memcg, |
1375 | "Memory cgroup out of memory"); | |
dc56401f JW |
1376 | } |
1377 | unlock: | |
1378 | mutex_unlock(&oom_lock); | |
9cbb78bb DR |
1379 | } |
1380 | ||
ae6e71d3 MC |
1381 | #if MAX_NUMNODES > 1 |
1382 | ||
4d0c066d KH |
1383 | /** |
1384 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1385 | * @memcg: the target memcg |
4d0c066d KH |
1386 | * @nid: the node ID to be checked. |
1387 | * @noswap : specify true here if the user wants flle only information. | |
1388 | * | |
1389 | * This function returns whether the specified memcg contains any | |
1390 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1391 | * pages in the node. | |
1392 | */ | |
c0ff4b85 | 1393 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1394 | int nid, bool noswap) |
1395 | { | |
c0ff4b85 | 1396 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1397 | return true; |
1398 | if (noswap || !total_swap_pages) | |
1399 | return false; | |
c0ff4b85 | 1400 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1401 | return true; |
1402 | return false; | |
1403 | ||
1404 | } | |
889976db YH |
1405 | |
1406 | /* | |
1407 | * Always updating the nodemask is not very good - even if we have an empty | |
1408 | * list or the wrong list here, we can start from some node and traverse all | |
1409 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1410 | * | |
1411 | */ | |
c0ff4b85 | 1412 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1413 | { |
1414 | int nid; | |
453a9bf3 KH |
1415 | /* |
1416 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1417 | * pagein/pageout changes since the last update. | |
1418 | */ | |
c0ff4b85 | 1419 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1420 | return; |
c0ff4b85 | 1421 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1422 | return; |
1423 | ||
889976db | 1424 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1425 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1426 | |
31aaea4a | 1427 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1428 | |
c0ff4b85 R |
1429 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1430 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1431 | } |
453a9bf3 | 1432 | |
c0ff4b85 R |
1433 | atomic_set(&memcg->numainfo_events, 0); |
1434 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1435 | } |
1436 | ||
1437 | /* | |
1438 | * Selecting a node where we start reclaim from. Because what we need is just | |
1439 | * reducing usage counter, start from anywhere is O,K. Considering | |
1440 | * memory reclaim from current node, there are pros. and cons. | |
1441 | * | |
1442 | * Freeing memory from current node means freeing memory from a node which | |
1443 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1444 | * hit limits, it will see a contention on a node. But freeing from remote | |
1445 | * node means more costs for memory reclaim because of memory latency. | |
1446 | * | |
1447 | * Now, we use round-robin. Better algorithm is welcomed. | |
1448 | */ | |
c0ff4b85 | 1449 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1450 | { |
1451 | int node; | |
1452 | ||
c0ff4b85 R |
1453 | mem_cgroup_may_update_nodemask(memcg); |
1454 | node = memcg->last_scanned_node; | |
889976db | 1455 | |
c0ff4b85 | 1456 | node = next_node(node, memcg->scan_nodes); |
889976db | 1457 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1458 | node = first_node(memcg->scan_nodes); |
889976db YH |
1459 | /* |
1460 | * We call this when we hit limit, not when pages are added to LRU. | |
1461 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1462 | * memcg is too small and all pages are not on LRU. In that case, | |
1463 | * we use curret node. | |
1464 | */ | |
1465 | if (unlikely(node == MAX_NUMNODES)) | |
1466 | node = numa_node_id(); | |
1467 | ||
c0ff4b85 | 1468 | memcg->last_scanned_node = node; |
889976db YH |
1469 | return node; |
1470 | } | |
889976db | 1471 | #else |
c0ff4b85 | 1472 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1473 | { |
1474 | return 0; | |
1475 | } | |
1476 | #endif | |
1477 | ||
0608f43d AM |
1478 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1479 | struct zone *zone, | |
1480 | gfp_t gfp_mask, | |
1481 | unsigned long *total_scanned) | |
1482 | { | |
1483 | struct mem_cgroup *victim = NULL; | |
1484 | int total = 0; | |
1485 | int loop = 0; | |
1486 | unsigned long excess; | |
1487 | unsigned long nr_scanned; | |
1488 | struct mem_cgroup_reclaim_cookie reclaim = { | |
1489 | .zone = zone, | |
1490 | .priority = 0, | |
1491 | }; | |
1492 | ||
3e32cb2e | 1493 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1494 | |
1495 | while (1) { | |
1496 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1497 | if (!victim) { | |
1498 | loop++; | |
1499 | if (loop >= 2) { | |
1500 | /* | |
1501 | * If we have not been able to reclaim | |
1502 | * anything, it might because there are | |
1503 | * no reclaimable pages under this hierarchy | |
1504 | */ | |
1505 | if (!total) | |
1506 | break; | |
1507 | /* | |
1508 | * We want to do more targeted reclaim. | |
1509 | * excess >> 2 is not to excessive so as to | |
1510 | * reclaim too much, nor too less that we keep | |
1511 | * coming back to reclaim from this cgroup | |
1512 | */ | |
1513 | if (total >= (excess >> 2) || | |
1514 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1515 | break; | |
1516 | } | |
1517 | continue; | |
1518 | } | |
0608f43d AM |
1519 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1520 | zone, &nr_scanned); | |
1521 | *total_scanned += nr_scanned; | |
3e32cb2e | 1522 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1523 | break; |
6d61ef40 | 1524 | } |
0608f43d AM |
1525 | mem_cgroup_iter_break(root_memcg, victim); |
1526 | return total; | |
6d61ef40 BS |
1527 | } |
1528 | ||
0056f4e6 JW |
1529 | #ifdef CONFIG_LOCKDEP |
1530 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1531 | .name = "memcg_oom_lock", | |
1532 | }; | |
1533 | #endif | |
1534 | ||
fb2a6fc5 JW |
1535 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1536 | ||
867578cb KH |
1537 | /* |
1538 | * Check OOM-Killer is already running under our hierarchy. | |
1539 | * If someone is running, return false. | |
1540 | */ | |
fb2a6fc5 | 1541 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1542 | { |
79dfdacc | 1543 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1544 | |
fb2a6fc5 JW |
1545 | spin_lock(&memcg_oom_lock); |
1546 | ||
9f3a0d09 | 1547 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1548 | if (iter->oom_lock) { |
79dfdacc MH |
1549 | /* |
1550 | * this subtree of our hierarchy is already locked | |
1551 | * so we cannot give a lock. | |
1552 | */ | |
79dfdacc | 1553 | failed = iter; |
9f3a0d09 JW |
1554 | mem_cgroup_iter_break(memcg, iter); |
1555 | break; | |
23751be0 JW |
1556 | } else |
1557 | iter->oom_lock = true; | |
7d74b06f | 1558 | } |
867578cb | 1559 | |
fb2a6fc5 JW |
1560 | if (failed) { |
1561 | /* | |
1562 | * OK, we failed to lock the whole subtree so we have | |
1563 | * to clean up what we set up to the failing subtree | |
1564 | */ | |
1565 | for_each_mem_cgroup_tree(iter, memcg) { | |
1566 | if (iter == failed) { | |
1567 | mem_cgroup_iter_break(memcg, iter); | |
1568 | break; | |
1569 | } | |
1570 | iter->oom_lock = false; | |
79dfdacc | 1571 | } |
0056f4e6 JW |
1572 | } else |
1573 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1574 | |
1575 | spin_unlock(&memcg_oom_lock); | |
1576 | ||
1577 | return !failed; | |
a636b327 | 1578 | } |
0b7f569e | 1579 | |
fb2a6fc5 | 1580 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1581 | { |
7d74b06f KH |
1582 | struct mem_cgroup *iter; |
1583 | ||
fb2a6fc5 | 1584 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1585 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1586 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1587 | iter->oom_lock = false; |
fb2a6fc5 | 1588 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1589 | } |
1590 | ||
c0ff4b85 | 1591 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1592 | { |
1593 | struct mem_cgroup *iter; | |
1594 | ||
c2b42d3c | 1595 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1596 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1597 | iter->under_oom++; |
1598 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1599 | } |
1600 | ||
c0ff4b85 | 1601 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1602 | { |
1603 | struct mem_cgroup *iter; | |
1604 | ||
867578cb KH |
1605 | /* |
1606 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1607 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1608 | */ |
c2b42d3c | 1609 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1610 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1611 | if (iter->under_oom > 0) |
1612 | iter->under_oom--; | |
1613 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1614 | } |
1615 | ||
867578cb KH |
1616 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1617 | ||
dc98df5a | 1618 | struct oom_wait_info { |
d79154bb | 1619 | struct mem_cgroup *memcg; |
dc98df5a KH |
1620 | wait_queue_t wait; |
1621 | }; | |
1622 | ||
1623 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1624 | unsigned mode, int sync, void *arg) | |
1625 | { | |
d79154bb HD |
1626 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1627 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1628 | struct oom_wait_info *oom_wait_info; |
1629 | ||
1630 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1631 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1632 | |
2314b42d JW |
1633 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1634 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1635 | return 0; |
dc98df5a KH |
1636 | return autoremove_wake_function(wait, mode, sync, arg); |
1637 | } | |
1638 | ||
c0ff4b85 | 1639 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1640 | { |
c2b42d3c TH |
1641 | /* |
1642 | * For the following lockless ->under_oom test, the only required | |
1643 | * guarantee is that it must see the state asserted by an OOM when | |
1644 | * this function is called as a result of userland actions | |
1645 | * triggered by the notification of the OOM. This is trivially | |
1646 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1647 | * triggering notification. | |
1648 | */ | |
1649 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1650 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1651 | } |
1652 | ||
3812c8c8 | 1653 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 1654 | { |
3812c8c8 JW |
1655 | if (!current->memcg_oom.may_oom) |
1656 | return; | |
867578cb | 1657 | /* |
49426420 JW |
1658 | * We are in the middle of the charge context here, so we |
1659 | * don't want to block when potentially sitting on a callstack | |
1660 | * that holds all kinds of filesystem and mm locks. | |
1661 | * | |
1662 | * Also, the caller may handle a failed allocation gracefully | |
1663 | * (like optional page cache readahead) and so an OOM killer | |
1664 | * invocation might not even be necessary. | |
1665 | * | |
1666 | * That's why we don't do anything here except remember the | |
1667 | * OOM context and then deal with it at the end of the page | |
1668 | * fault when the stack is unwound, the locks are released, | |
1669 | * and when we know whether the fault was overall successful. | |
867578cb | 1670 | */ |
49426420 JW |
1671 | css_get(&memcg->css); |
1672 | current->memcg_oom.memcg = memcg; | |
1673 | current->memcg_oom.gfp_mask = mask; | |
1674 | current->memcg_oom.order = order; | |
3812c8c8 JW |
1675 | } |
1676 | ||
1677 | /** | |
1678 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1679 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1680 | * |
49426420 JW |
1681 | * This has to be called at the end of a page fault if the memcg OOM |
1682 | * handler was enabled. | |
3812c8c8 | 1683 | * |
49426420 | 1684 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1685 | * sleep on a waitqueue until the userspace task resolves the |
1686 | * situation. Sleeping directly in the charge context with all kinds | |
1687 | * of locks held is not a good idea, instead we remember an OOM state | |
1688 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1689 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1690 | * |
1691 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1692 | * completed, %false otherwise. |
3812c8c8 | 1693 | */ |
49426420 | 1694 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1695 | { |
49426420 | 1696 | struct mem_cgroup *memcg = current->memcg_oom.memcg; |
3812c8c8 | 1697 | struct oom_wait_info owait; |
49426420 | 1698 | bool locked; |
3812c8c8 JW |
1699 | |
1700 | /* OOM is global, do not handle */ | |
3812c8c8 | 1701 | if (!memcg) |
49426420 | 1702 | return false; |
3812c8c8 | 1703 | |
c32b3cbe | 1704 | if (!handle || oom_killer_disabled) |
49426420 | 1705 | goto cleanup; |
3812c8c8 JW |
1706 | |
1707 | owait.memcg = memcg; | |
1708 | owait.wait.flags = 0; | |
1709 | owait.wait.func = memcg_oom_wake_function; | |
1710 | owait.wait.private = current; | |
1711 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 1712 | |
3812c8c8 | 1713 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1714 | mem_cgroup_mark_under_oom(memcg); |
1715 | ||
1716 | locked = mem_cgroup_oom_trylock(memcg); | |
1717 | ||
1718 | if (locked) | |
1719 | mem_cgroup_oom_notify(memcg); | |
1720 | ||
1721 | if (locked && !memcg->oom_kill_disable) { | |
1722 | mem_cgroup_unmark_under_oom(memcg); | |
1723 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1724 | mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask, | |
1725 | current->memcg_oom.order); | |
1726 | } else { | |
3812c8c8 | 1727 | schedule(); |
49426420 JW |
1728 | mem_cgroup_unmark_under_oom(memcg); |
1729 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1730 | } | |
1731 | ||
1732 | if (locked) { | |
fb2a6fc5 JW |
1733 | mem_cgroup_oom_unlock(memcg); |
1734 | /* | |
1735 | * There is no guarantee that an OOM-lock contender | |
1736 | * sees the wakeups triggered by the OOM kill | |
1737 | * uncharges. Wake any sleepers explicitely. | |
1738 | */ | |
1739 | memcg_oom_recover(memcg); | |
1740 | } | |
49426420 JW |
1741 | cleanup: |
1742 | current->memcg_oom.memcg = NULL; | |
3812c8c8 | 1743 | css_put(&memcg->css); |
867578cb | 1744 | return true; |
0b7f569e KH |
1745 | } |
1746 | ||
d7365e78 JW |
1747 | /** |
1748 | * mem_cgroup_begin_page_stat - begin a page state statistics transaction | |
1749 | * @page: page that is going to change accounted state | |
32047e2a | 1750 | * |
d7365e78 JW |
1751 | * This function must mark the beginning of an accounted page state |
1752 | * change to prevent double accounting when the page is concurrently | |
1753 | * being moved to another memcg: | |
32047e2a | 1754 | * |
6de22619 | 1755 | * memcg = mem_cgroup_begin_page_stat(page); |
d7365e78 JW |
1756 | * if (TestClearPageState(page)) |
1757 | * mem_cgroup_update_page_stat(memcg, state, -1); | |
6de22619 | 1758 | * mem_cgroup_end_page_stat(memcg); |
d69b042f | 1759 | */ |
6de22619 | 1760 | struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page) |
89c06bd5 KH |
1761 | { |
1762 | struct mem_cgroup *memcg; | |
6de22619 | 1763 | unsigned long flags; |
89c06bd5 | 1764 | |
6de22619 JW |
1765 | /* |
1766 | * The RCU lock is held throughout the transaction. The fast | |
1767 | * path can get away without acquiring the memcg->move_lock | |
1768 | * because page moving starts with an RCU grace period. | |
1769 | * | |
1770 | * The RCU lock also protects the memcg from being freed when | |
1771 | * the page state that is going to change is the only thing | |
1772 | * preventing the page from being uncharged. | |
1773 | * E.g. end-writeback clearing PageWriteback(), which allows | |
1774 | * migration to go ahead and uncharge the page before the | |
1775 | * account transaction might be complete. | |
1776 | */ | |
d7365e78 JW |
1777 | rcu_read_lock(); |
1778 | ||
1779 | if (mem_cgroup_disabled()) | |
1780 | return NULL; | |
89c06bd5 | 1781 | again: |
1306a85a | 1782 | memcg = page->mem_cgroup; |
29833315 | 1783 | if (unlikely(!memcg)) |
d7365e78 JW |
1784 | return NULL; |
1785 | ||
bdcbb659 | 1786 | if (atomic_read(&memcg->moving_account) <= 0) |
d7365e78 | 1787 | return memcg; |
89c06bd5 | 1788 | |
6de22619 | 1789 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 1790 | if (memcg != page->mem_cgroup) { |
6de22619 | 1791 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
1792 | goto again; |
1793 | } | |
6de22619 JW |
1794 | |
1795 | /* | |
1796 | * When charge migration first begins, we can have locked and | |
1797 | * unlocked page stat updates happening concurrently. Track | |
1798 | * the task who has the lock for mem_cgroup_end_page_stat(). | |
1799 | */ | |
1800 | memcg->move_lock_task = current; | |
1801 | memcg->move_lock_flags = flags; | |
d7365e78 JW |
1802 | |
1803 | return memcg; | |
89c06bd5 | 1804 | } |
c4843a75 | 1805 | EXPORT_SYMBOL(mem_cgroup_begin_page_stat); |
89c06bd5 | 1806 | |
d7365e78 JW |
1807 | /** |
1808 | * mem_cgroup_end_page_stat - finish a page state statistics transaction | |
1809 | * @memcg: the memcg that was accounted against | |
d7365e78 | 1810 | */ |
6de22619 | 1811 | void mem_cgroup_end_page_stat(struct mem_cgroup *memcg) |
89c06bd5 | 1812 | { |
6de22619 JW |
1813 | if (memcg && memcg->move_lock_task == current) { |
1814 | unsigned long flags = memcg->move_lock_flags; | |
1815 | ||
1816 | memcg->move_lock_task = NULL; | |
1817 | memcg->move_lock_flags = 0; | |
1818 | ||
1819 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
1820 | } | |
89c06bd5 | 1821 | |
d7365e78 | 1822 | rcu_read_unlock(); |
89c06bd5 | 1823 | } |
c4843a75 | 1824 | EXPORT_SYMBOL(mem_cgroup_end_page_stat); |
89c06bd5 | 1825 | |
cdec2e42 KH |
1826 | /* |
1827 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1828 | * TODO: maybe necessary to use big numbers in big irons. | |
1829 | */ | |
7ec99d62 | 1830 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
1831 | struct memcg_stock_pcp { |
1832 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 1833 | unsigned int nr_pages; |
cdec2e42 | 1834 | struct work_struct work; |
26fe6168 | 1835 | unsigned long flags; |
a0db00fc | 1836 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
1837 | }; |
1838 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 1839 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 1840 | |
a0956d54 SS |
1841 | /** |
1842 | * consume_stock: Try to consume stocked charge on this cpu. | |
1843 | * @memcg: memcg to consume from. | |
1844 | * @nr_pages: how many pages to charge. | |
1845 | * | |
1846 | * The charges will only happen if @memcg matches the current cpu's memcg | |
1847 | * stock, and at least @nr_pages are available in that stock. Failure to | |
1848 | * service an allocation will refill the stock. | |
1849 | * | |
1850 | * returns true if successful, false otherwise. | |
cdec2e42 | 1851 | */ |
a0956d54 | 1852 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1853 | { |
1854 | struct memcg_stock_pcp *stock; | |
3e32cb2e | 1855 | bool ret = false; |
cdec2e42 | 1856 | |
a0956d54 | 1857 | if (nr_pages > CHARGE_BATCH) |
3e32cb2e | 1858 | return ret; |
a0956d54 | 1859 | |
cdec2e42 | 1860 | stock = &get_cpu_var(memcg_stock); |
3e32cb2e | 1861 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 1862 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
1863 | ret = true; |
1864 | } | |
cdec2e42 KH |
1865 | put_cpu_var(memcg_stock); |
1866 | return ret; | |
1867 | } | |
1868 | ||
1869 | /* | |
3e32cb2e | 1870 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
1871 | */ |
1872 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1873 | { | |
1874 | struct mem_cgroup *old = stock->cached; | |
1875 | ||
11c9ea4e | 1876 | if (stock->nr_pages) { |
3e32cb2e | 1877 | page_counter_uncharge(&old->memory, stock->nr_pages); |
cdec2e42 | 1878 | if (do_swap_account) |
3e32cb2e | 1879 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 1880 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 1881 | stock->nr_pages = 0; |
cdec2e42 KH |
1882 | } |
1883 | stock->cached = NULL; | |
cdec2e42 KH |
1884 | } |
1885 | ||
1886 | /* | |
1887 | * This must be called under preempt disabled or must be called by | |
1888 | * a thread which is pinned to local cpu. | |
1889 | */ | |
1890 | static void drain_local_stock(struct work_struct *dummy) | |
1891 | { | |
7c8e0181 | 1892 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
cdec2e42 | 1893 | drain_stock(stock); |
26fe6168 | 1894 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
1895 | } |
1896 | ||
1897 | /* | |
3e32cb2e | 1898 | * Cache charges(val) to local per_cpu area. |
320cc51d | 1899 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 1900 | */ |
c0ff4b85 | 1901 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1902 | { |
1903 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
1904 | ||
c0ff4b85 | 1905 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 1906 | drain_stock(stock); |
c0ff4b85 | 1907 | stock->cached = memcg; |
cdec2e42 | 1908 | } |
11c9ea4e | 1909 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
1910 | put_cpu_var(memcg_stock); |
1911 | } | |
1912 | ||
1913 | /* | |
c0ff4b85 | 1914 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 1915 | * of the hierarchy under it. |
cdec2e42 | 1916 | */ |
6d3d6aa2 | 1917 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 1918 | { |
26fe6168 | 1919 | int cpu, curcpu; |
d38144b7 | 1920 | |
6d3d6aa2 JW |
1921 | /* If someone's already draining, avoid adding running more workers. */ |
1922 | if (!mutex_trylock(&percpu_charge_mutex)) | |
1923 | return; | |
cdec2e42 | 1924 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 1925 | get_online_cpus(); |
5af12d0e | 1926 | curcpu = get_cpu(); |
cdec2e42 KH |
1927 | for_each_online_cpu(cpu) { |
1928 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 1929 | struct mem_cgroup *memcg; |
26fe6168 | 1930 | |
c0ff4b85 R |
1931 | memcg = stock->cached; |
1932 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 1933 | continue; |
2314b42d | 1934 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) |
3e92041d | 1935 | continue; |
d1a05b69 MH |
1936 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
1937 | if (cpu == curcpu) | |
1938 | drain_local_stock(&stock->work); | |
1939 | else | |
1940 | schedule_work_on(cpu, &stock->work); | |
1941 | } | |
cdec2e42 | 1942 | } |
5af12d0e | 1943 | put_cpu(); |
f894ffa8 | 1944 | put_online_cpus(); |
9f50fad6 | 1945 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
1946 | } |
1947 | ||
0db0628d | 1948 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
1949 | unsigned long action, |
1950 | void *hcpu) | |
1951 | { | |
1952 | int cpu = (unsigned long)hcpu; | |
1953 | struct memcg_stock_pcp *stock; | |
1954 | ||
619d094b | 1955 | if (action == CPU_ONLINE) |
1489ebad | 1956 | return NOTIFY_OK; |
1489ebad | 1957 | |
d833049b | 1958 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 1959 | return NOTIFY_OK; |
711d3d2c | 1960 | |
cdec2e42 KH |
1961 | stock = &per_cpu(memcg_stock, cpu); |
1962 | drain_stock(stock); | |
1963 | return NOTIFY_OK; | |
1964 | } | |
1965 | ||
00501b53 JW |
1966 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1967 | unsigned int nr_pages) | |
8a9f3ccd | 1968 | { |
7ec99d62 | 1969 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 1970 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 1971 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 1972 | struct page_counter *counter; |
6539cc05 | 1973 | unsigned long nr_reclaimed; |
b70a2a21 JW |
1974 | bool may_swap = true; |
1975 | bool drained = false; | |
05b84301 | 1976 | int ret = 0; |
a636b327 | 1977 | |
ce00a967 JW |
1978 | if (mem_cgroup_is_root(memcg)) |
1979 | goto done; | |
6539cc05 | 1980 | retry: |
b6b6cc72 MH |
1981 | if (consume_stock(memcg, nr_pages)) |
1982 | goto done; | |
8a9f3ccd | 1983 | |
3fbe7244 | 1984 | if (!do_swap_account || |
3e32cb2e JW |
1985 | !page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
1986 | if (!page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 1987 | goto done_restock; |
3fbe7244 | 1988 | if (do_swap_account) |
3e32cb2e JW |
1989 | page_counter_uncharge(&memcg->memsw, batch); |
1990 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 1991 | } else { |
3e32cb2e | 1992 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 1993 | may_swap = false; |
3fbe7244 | 1994 | } |
7a81b88c | 1995 | |
6539cc05 JW |
1996 | if (batch > nr_pages) { |
1997 | batch = nr_pages; | |
1998 | goto retry; | |
1999 | } | |
6d61ef40 | 2000 | |
06b078fc JW |
2001 | /* |
2002 | * Unlike in global OOM situations, memcg is not in a physical | |
2003 | * memory shortage. Allow dying and OOM-killed tasks to | |
2004 | * bypass the last charges so that they can exit quickly and | |
2005 | * free their memory. | |
2006 | */ | |
2007 | if (unlikely(test_thread_flag(TIF_MEMDIE) || | |
2008 | fatal_signal_pending(current) || | |
2009 | current->flags & PF_EXITING)) | |
2010 | goto bypass; | |
2011 | ||
2012 | if (unlikely(task_in_memcg_oom(current))) | |
2013 | goto nomem; | |
2014 | ||
6539cc05 JW |
2015 | if (!(gfp_mask & __GFP_WAIT)) |
2016 | goto nomem; | |
4b534334 | 2017 | |
241994ed JW |
2018 | mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1); |
2019 | ||
b70a2a21 JW |
2020 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2021 | gfp_mask, may_swap); | |
6539cc05 | 2022 | |
61e02c74 | 2023 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2024 | goto retry; |
28c34c29 | 2025 | |
b70a2a21 | 2026 | if (!drained) { |
6d3d6aa2 | 2027 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2028 | drained = true; |
2029 | goto retry; | |
2030 | } | |
2031 | ||
28c34c29 JW |
2032 | if (gfp_mask & __GFP_NORETRY) |
2033 | goto nomem; | |
6539cc05 JW |
2034 | /* |
2035 | * Even though the limit is exceeded at this point, reclaim | |
2036 | * may have been able to free some pages. Retry the charge | |
2037 | * before killing the task. | |
2038 | * | |
2039 | * Only for regular pages, though: huge pages are rather | |
2040 | * unlikely to succeed so close to the limit, and we fall back | |
2041 | * to regular pages anyway in case of failure. | |
2042 | */ | |
61e02c74 | 2043 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2044 | goto retry; |
2045 | /* | |
2046 | * At task move, charge accounts can be doubly counted. So, it's | |
2047 | * better to wait until the end of task_move if something is going on. | |
2048 | */ | |
2049 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2050 | goto retry; | |
2051 | ||
9b130619 JW |
2052 | if (nr_retries--) |
2053 | goto retry; | |
2054 | ||
06b078fc JW |
2055 | if (gfp_mask & __GFP_NOFAIL) |
2056 | goto bypass; | |
2057 | ||
6539cc05 JW |
2058 | if (fatal_signal_pending(current)) |
2059 | goto bypass; | |
2060 | ||
241994ed JW |
2061 | mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1); |
2062 | ||
61e02c74 | 2063 | mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(nr_pages)); |
7a81b88c | 2064 | nomem: |
6d1fdc48 | 2065 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2066 | return -ENOMEM; |
867578cb | 2067 | bypass: |
ce00a967 | 2068 | return -EINTR; |
6539cc05 JW |
2069 | |
2070 | done_restock: | |
e8ea14cc | 2071 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2072 | if (batch > nr_pages) |
2073 | refill_stock(memcg, batch - nr_pages); | |
7d638093 VD |
2074 | if (!(gfp_mask & __GFP_WAIT)) |
2075 | goto done; | |
241994ed JW |
2076 | /* |
2077 | * If the hierarchy is above the normal consumption range, | |
2078 | * make the charging task trim their excess contribution. | |
2079 | */ | |
2080 | do { | |
2081 | if (page_counter_read(&memcg->memory) <= memcg->high) | |
2082 | continue; | |
2083 | mem_cgroup_events(memcg, MEMCG_HIGH, 1); | |
2084 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); | |
2085 | } while ((memcg = parent_mem_cgroup(memcg))); | |
6539cc05 | 2086 | done: |
05b84301 | 2087 | return ret; |
7a81b88c | 2088 | } |
8a9f3ccd | 2089 | |
00501b53 | 2090 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2091 | { |
ce00a967 JW |
2092 | if (mem_cgroup_is_root(memcg)) |
2093 | return; | |
2094 | ||
3e32cb2e | 2095 | page_counter_uncharge(&memcg->memory, nr_pages); |
05b84301 | 2096 | if (do_swap_account) |
3e32cb2e | 2097 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2098 | |
e8ea14cc | 2099 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2100 | } |
2101 | ||
0a31bc97 JW |
2102 | static void lock_page_lru(struct page *page, int *isolated) |
2103 | { | |
2104 | struct zone *zone = page_zone(page); | |
2105 | ||
2106 | spin_lock_irq(&zone->lru_lock); | |
2107 | if (PageLRU(page)) { | |
2108 | struct lruvec *lruvec; | |
2109 | ||
2110 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2111 | ClearPageLRU(page); | |
2112 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2113 | *isolated = 1; | |
2114 | } else | |
2115 | *isolated = 0; | |
2116 | } | |
2117 | ||
2118 | static void unlock_page_lru(struct page *page, int isolated) | |
2119 | { | |
2120 | struct zone *zone = page_zone(page); | |
2121 | ||
2122 | if (isolated) { | |
2123 | struct lruvec *lruvec; | |
2124 | ||
2125 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2126 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
2127 | SetPageLRU(page); | |
2128 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2129 | } | |
2130 | spin_unlock_irq(&zone->lru_lock); | |
2131 | } | |
2132 | ||
00501b53 | 2133 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2134 | bool lrucare) |
7a81b88c | 2135 | { |
0a31bc97 | 2136 | int isolated; |
9ce70c02 | 2137 | |
1306a85a | 2138 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2139 | |
2140 | /* | |
2141 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2142 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2143 | */ | |
0a31bc97 JW |
2144 | if (lrucare) |
2145 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2146 | |
0a31bc97 JW |
2147 | /* |
2148 | * Nobody should be changing or seriously looking at | |
1306a85a | 2149 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2150 | * |
2151 | * - the page is uncharged | |
2152 | * | |
2153 | * - the page is off-LRU | |
2154 | * | |
2155 | * - an anonymous fault has exclusive page access, except for | |
2156 | * a locked page table | |
2157 | * | |
2158 | * - a page cache insertion, a swapin fault, or a migration | |
2159 | * have the page locked | |
2160 | */ | |
1306a85a | 2161 | page->mem_cgroup = memcg; |
9ce70c02 | 2162 | |
0a31bc97 JW |
2163 | if (lrucare) |
2164 | unlock_page_lru(page, isolated); | |
7a81b88c | 2165 | } |
66e1707b | 2166 | |
7ae1e1d0 | 2167 | #ifdef CONFIG_MEMCG_KMEM |
dbf22eb6 VD |
2168 | int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, |
2169 | unsigned long nr_pages) | |
7ae1e1d0 | 2170 | { |
3e32cb2e | 2171 | struct page_counter *counter; |
7ae1e1d0 | 2172 | int ret = 0; |
7ae1e1d0 | 2173 | |
3e32cb2e JW |
2174 | ret = page_counter_try_charge(&memcg->kmem, nr_pages, &counter); |
2175 | if (ret < 0) | |
7ae1e1d0 GC |
2176 | return ret; |
2177 | ||
3e32cb2e | 2178 | ret = try_charge(memcg, gfp, nr_pages); |
7ae1e1d0 GC |
2179 | if (ret == -EINTR) { |
2180 | /* | |
00501b53 JW |
2181 | * try_charge() chose to bypass to root due to OOM kill or |
2182 | * fatal signal. Since our only options are to either fail | |
2183 | * the allocation or charge it to this cgroup, do it as a | |
2184 | * temporary condition. But we can't fail. From a kmem/slab | |
2185 | * perspective, the cache has already been selected, by | |
2186 | * mem_cgroup_kmem_get_cache(), so it is too late to change | |
7ae1e1d0 GC |
2187 | * our minds. |
2188 | * | |
2189 | * This condition will only trigger if the task entered | |
00501b53 JW |
2190 | * memcg_charge_kmem in a sane state, but was OOM-killed |
2191 | * during try_charge() above. Tasks that were already dying | |
2192 | * when the allocation triggers should have been already | |
7ae1e1d0 GC |
2193 | * directed to the root cgroup in memcontrol.h |
2194 | */ | |
3e32cb2e | 2195 | page_counter_charge(&memcg->memory, nr_pages); |
7ae1e1d0 | 2196 | if (do_swap_account) |
3e32cb2e | 2197 | page_counter_charge(&memcg->memsw, nr_pages); |
e8ea14cc | 2198 | css_get_many(&memcg->css, nr_pages); |
7ae1e1d0 GC |
2199 | ret = 0; |
2200 | } else if (ret) | |
3e32cb2e | 2201 | page_counter_uncharge(&memcg->kmem, nr_pages); |
7ae1e1d0 GC |
2202 | |
2203 | return ret; | |
2204 | } | |
2205 | ||
dbf22eb6 | 2206 | void memcg_uncharge_kmem(struct mem_cgroup *memcg, unsigned long nr_pages) |
7ae1e1d0 | 2207 | { |
3e32cb2e | 2208 | page_counter_uncharge(&memcg->memory, nr_pages); |
7ae1e1d0 | 2209 | if (do_swap_account) |
3e32cb2e | 2210 | page_counter_uncharge(&memcg->memsw, nr_pages); |
7de37682 | 2211 | |
64f21993 | 2212 | page_counter_uncharge(&memcg->kmem, nr_pages); |
7de37682 | 2213 | |
e8ea14cc | 2214 | css_put_many(&memcg->css, nr_pages); |
7ae1e1d0 GC |
2215 | } |
2216 | ||
f3bb3043 | 2217 | static int memcg_alloc_cache_id(void) |
55007d84 | 2218 | { |
f3bb3043 VD |
2219 | int id, size; |
2220 | int err; | |
2221 | ||
dbcf73e2 | 2222 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2223 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2224 | if (id < 0) | |
2225 | return id; | |
55007d84 | 2226 | |
dbcf73e2 | 2227 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2228 | return id; |
2229 | ||
2230 | /* | |
2231 | * There's no space for the new id in memcg_caches arrays, | |
2232 | * so we have to grow them. | |
2233 | */ | |
05257a1a | 2234 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2235 | |
2236 | size = 2 * (id + 1); | |
55007d84 GC |
2237 | if (size < MEMCG_CACHES_MIN_SIZE) |
2238 | size = MEMCG_CACHES_MIN_SIZE; | |
2239 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2240 | size = MEMCG_CACHES_MAX_SIZE; | |
2241 | ||
f3bb3043 | 2242 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2243 | if (!err) |
2244 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2245 | if (!err) |
2246 | memcg_nr_cache_ids = size; | |
2247 | ||
2248 | up_write(&memcg_cache_ids_sem); | |
2249 | ||
f3bb3043 | 2250 | if (err) { |
dbcf73e2 | 2251 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2252 | return err; |
2253 | } | |
2254 | return id; | |
2255 | } | |
2256 | ||
2257 | static void memcg_free_cache_id(int id) | |
2258 | { | |
dbcf73e2 | 2259 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2260 | } |
2261 | ||
d5b3cf71 | 2262 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2263 | struct mem_cgroup *memcg; |
2264 | struct kmem_cache *cachep; | |
2265 | struct work_struct work; | |
2266 | }; | |
2267 | ||
d5b3cf71 | 2268 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2269 | { |
d5b3cf71 VD |
2270 | struct memcg_kmem_cache_create_work *cw = |
2271 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2272 | struct mem_cgroup *memcg = cw->memcg; |
2273 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2274 | |
d5b3cf71 | 2275 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2276 | |
5722d094 | 2277 | css_put(&memcg->css); |
d7f25f8a GC |
2278 | kfree(cw); |
2279 | } | |
2280 | ||
2281 | /* | |
2282 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2283 | */ |
d5b3cf71 VD |
2284 | static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2285 | struct kmem_cache *cachep) | |
d7f25f8a | 2286 | { |
d5b3cf71 | 2287 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2288 | |
776ed0f0 | 2289 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
8135be5a | 2290 | if (!cw) |
d7f25f8a | 2291 | return; |
8135be5a VD |
2292 | |
2293 | css_get(&memcg->css); | |
d7f25f8a GC |
2294 | |
2295 | cw->memcg = memcg; | |
2296 | cw->cachep = cachep; | |
d5b3cf71 | 2297 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2298 | |
d7f25f8a GC |
2299 | schedule_work(&cw->work); |
2300 | } | |
2301 | ||
d5b3cf71 VD |
2302 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2303 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
2304 | { |
2305 | /* | |
2306 | * We need to stop accounting when we kmalloc, because if the | |
2307 | * corresponding kmalloc cache is not yet created, the first allocation | |
d5b3cf71 | 2308 | * in __memcg_schedule_kmem_cache_create will recurse. |
0e9d92f2 GC |
2309 | * |
2310 | * However, it is better to enclose the whole function. Depending on | |
2311 | * the debugging options enabled, INIT_WORK(), for instance, can | |
2312 | * trigger an allocation. This too, will make us recurse. Because at | |
2313 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
2314 | * the safest choice is to do it like this, wrapping the whole function. | |
2315 | */ | |
6f185c29 | 2316 | current->memcg_kmem_skip_account = 1; |
d5b3cf71 | 2317 | __memcg_schedule_kmem_cache_create(memcg, cachep); |
6f185c29 | 2318 | current->memcg_kmem_skip_account = 0; |
0e9d92f2 | 2319 | } |
c67a8a68 | 2320 | |
d7f25f8a GC |
2321 | /* |
2322 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
2323 | * We try to use the current memcg's version of the cache. | |
2324 | * | |
2325 | * If the cache does not exist yet, if we are the first user of it, | |
2326 | * we either create it immediately, if possible, or create it asynchronously | |
2327 | * in a workqueue. | |
2328 | * In the latter case, we will let the current allocation go through with | |
2329 | * the original cache. | |
2330 | * | |
2331 | * Can't be called in interrupt context or from kernel threads. | |
2332 | * This function needs to be called with rcu_read_lock() held. | |
2333 | */ | |
056b7cce | 2334 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep) |
d7f25f8a GC |
2335 | { |
2336 | struct mem_cgroup *memcg; | |
959c8963 | 2337 | struct kmem_cache *memcg_cachep; |
2a4db7eb | 2338 | int kmemcg_id; |
d7f25f8a | 2339 | |
f7ce3190 | 2340 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2341 | |
9d100c5e | 2342 | if (current->memcg_kmem_skip_account) |
0e9d92f2 GC |
2343 | return cachep; |
2344 | ||
8135be5a | 2345 | memcg = get_mem_cgroup_from_mm(current->mm); |
4db0c3c2 | 2346 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
2a4db7eb | 2347 | if (kmemcg_id < 0) |
ca0dde97 | 2348 | goto out; |
d7f25f8a | 2349 | |
2a4db7eb | 2350 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
8135be5a VD |
2351 | if (likely(memcg_cachep)) |
2352 | return memcg_cachep; | |
ca0dde97 LZ |
2353 | |
2354 | /* | |
2355 | * If we are in a safe context (can wait, and not in interrupt | |
2356 | * context), we could be be predictable and return right away. | |
2357 | * This would guarantee that the allocation being performed | |
2358 | * already belongs in the new cache. | |
2359 | * | |
2360 | * However, there are some clashes that can arrive from locking. | |
2361 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2362 | * memcg_create_kmem_cache, this means no further allocation |
2363 | * could happen with the slab_mutex held. So it's better to | |
2364 | * defer everything. | |
ca0dde97 | 2365 | */ |
d5b3cf71 | 2366 | memcg_schedule_kmem_cache_create(memcg, cachep); |
ca0dde97 | 2367 | out: |
8135be5a | 2368 | css_put(&memcg->css); |
ca0dde97 | 2369 | return cachep; |
d7f25f8a | 2370 | } |
d7f25f8a | 2371 | |
8135be5a VD |
2372 | void __memcg_kmem_put_cache(struct kmem_cache *cachep) |
2373 | { | |
2374 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2375 | css_put(&cachep->memcg_params.memcg->css); |
8135be5a VD |
2376 | } |
2377 | ||
7ae1e1d0 GC |
2378 | /* |
2379 | * We need to verify if the allocation against current->mm->owner's memcg is | |
2380 | * possible for the given order. But the page is not allocated yet, so we'll | |
2381 | * need a further commit step to do the final arrangements. | |
2382 | * | |
2383 | * It is possible for the task to switch cgroups in this mean time, so at | |
2384 | * commit time, we can't rely on task conversion any longer. We'll then use | |
2385 | * the handle argument to return to the caller which cgroup we should commit | |
2386 | * against. We could also return the memcg directly and avoid the pointer | |
2387 | * passing, but a boolean return value gives better semantics considering | |
2388 | * the compiled-out case as well. | |
2389 | * | |
2390 | * Returning true means the allocation is possible. | |
2391 | */ | |
2392 | bool | |
2393 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) | |
2394 | { | |
2395 | struct mem_cgroup *memcg; | |
2396 | int ret; | |
2397 | ||
2398 | *_memcg = NULL; | |
6d42c232 | 2399 | |
df381975 | 2400 | memcg = get_mem_cgroup_from_mm(current->mm); |
7ae1e1d0 | 2401 | |
cf2b8fbf | 2402 | if (!memcg_kmem_is_active(memcg)) { |
7ae1e1d0 GC |
2403 | css_put(&memcg->css); |
2404 | return true; | |
2405 | } | |
2406 | ||
3e32cb2e | 2407 | ret = memcg_charge_kmem(memcg, gfp, 1 << order); |
7ae1e1d0 GC |
2408 | if (!ret) |
2409 | *_memcg = memcg; | |
7ae1e1d0 GC |
2410 | |
2411 | css_put(&memcg->css); | |
2412 | return (ret == 0); | |
2413 | } | |
2414 | ||
2415 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
2416 | int order) | |
2417 | { | |
7ae1e1d0 GC |
2418 | VM_BUG_ON(mem_cgroup_is_root(memcg)); |
2419 | ||
2420 | /* The page allocation failed. Revert */ | |
2421 | if (!page) { | |
3e32cb2e | 2422 | memcg_uncharge_kmem(memcg, 1 << order); |
7ae1e1d0 GC |
2423 | return; |
2424 | } | |
1306a85a | 2425 | page->mem_cgroup = memcg; |
7ae1e1d0 GC |
2426 | } |
2427 | ||
2428 | void __memcg_kmem_uncharge_pages(struct page *page, int order) | |
2429 | { | |
1306a85a | 2430 | struct mem_cgroup *memcg = page->mem_cgroup; |
7ae1e1d0 | 2431 | |
7ae1e1d0 GC |
2432 | if (!memcg) |
2433 | return; | |
2434 | ||
309381fe | 2435 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 2436 | |
3e32cb2e | 2437 | memcg_uncharge_kmem(memcg, 1 << order); |
1306a85a | 2438 | page->mem_cgroup = NULL; |
7ae1e1d0 | 2439 | } |
60d3fd32 VD |
2440 | |
2441 | struct mem_cgroup *__mem_cgroup_from_kmem(void *ptr) | |
2442 | { | |
2443 | struct mem_cgroup *memcg = NULL; | |
2444 | struct kmem_cache *cachep; | |
2445 | struct page *page; | |
2446 | ||
2447 | page = virt_to_head_page(ptr); | |
2448 | if (PageSlab(page)) { | |
2449 | cachep = page->slab_cache; | |
2450 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2451 | memcg = cachep->memcg_params.memcg; |
60d3fd32 VD |
2452 | } else |
2453 | /* page allocated by alloc_kmem_pages */ | |
2454 | memcg = page->mem_cgroup; | |
2455 | ||
2456 | return memcg; | |
2457 | } | |
7ae1e1d0 GC |
2458 | #endif /* CONFIG_MEMCG_KMEM */ |
2459 | ||
ca3e0214 KH |
2460 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2461 | ||
ca3e0214 KH |
2462 | /* |
2463 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
2464 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
2465 | * charge/uncharge will be never happen and move_account() is done under | |
2466 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 2467 | */ |
e94c8a9c | 2468 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2469 | { |
e94c8a9c | 2470 | int i; |
ca3e0214 | 2471 | |
3d37c4a9 KH |
2472 | if (mem_cgroup_disabled()) |
2473 | return; | |
b070e65c | 2474 | |
29833315 | 2475 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2476 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2477 | |
1306a85a | 2478 | __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
b070e65c | 2479 | HPAGE_PMD_NR); |
ca3e0214 | 2480 | } |
12d27107 | 2481 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2482 | |
c255a458 | 2483 | #ifdef CONFIG_MEMCG_SWAP |
0a31bc97 JW |
2484 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
2485 | bool charge) | |
d13d1443 | 2486 | { |
0a31bc97 JW |
2487 | int val = (charge) ? 1 : -1; |
2488 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); | |
d13d1443 | 2489 | } |
02491447 DN |
2490 | |
2491 | /** | |
2492 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2493 | * @entry: swap entry to be moved | |
2494 | * @from: mem_cgroup which the entry is moved from | |
2495 | * @to: mem_cgroup which the entry is moved to | |
2496 | * | |
2497 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2498 | * as the mem_cgroup's id of @from. | |
2499 | * | |
2500 | * Returns 0 on success, -EINVAL on failure. | |
2501 | * | |
3e32cb2e | 2502 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2503 | * both res and memsw, and called css_get(). |
2504 | */ | |
2505 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2506 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2507 | { |
2508 | unsigned short old_id, new_id; | |
2509 | ||
34c00c31 LZ |
2510 | old_id = mem_cgroup_id(from); |
2511 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2512 | |
2513 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 2514 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 2515 | mem_cgroup_swap_statistics(to, true); |
02491447 DN |
2516 | return 0; |
2517 | } | |
2518 | return -EINVAL; | |
2519 | } | |
2520 | #else | |
2521 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2522 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2523 | { |
2524 | return -EINVAL; | |
2525 | } | |
8c7c6e34 | 2526 | #endif |
d13d1443 | 2527 | |
3e32cb2e | 2528 | static DEFINE_MUTEX(memcg_limit_mutex); |
f212ad7c | 2529 | |
d38d2a75 | 2530 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2531 | unsigned long limit) |
628f4235 | 2532 | { |
3e32cb2e JW |
2533 | unsigned long curusage; |
2534 | unsigned long oldusage; | |
2535 | bool enlarge = false; | |
81d39c20 | 2536 | int retry_count; |
3e32cb2e | 2537 | int ret; |
81d39c20 KH |
2538 | |
2539 | /* | |
2540 | * For keeping hierarchical_reclaim simple, how long we should retry | |
2541 | * is depends on callers. We set our retry-count to be function | |
2542 | * of # of children which we should visit in this loop. | |
2543 | */ | |
3e32cb2e JW |
2544 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2545 | mem_cgroup_count_children(memcg); | |
81d39c20 | 2546 | |
3e32cb2e | 2547 | oldusage = page_counter_read(&memcg->memory); |
628f4235 | 2548 | |
3e32cb2e | 2549 | do { |
628f4235 KH |
2550 | if (signal_pending(current)) { |
2551 | ret = -EINTR; | |
2552 | break; | |
2553 | } | |
3e32cb2e JW |
2554 | |
2555 | mutex_lock(&memcg_limit_mutex); | |
2556 | if (limit > memcg->memsw.limit) { | |
2557 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2558 | ret = -EINVAL; |
628f4235 KH |
2559 | break; |
2560 | } | |
3e32cb2e JW |
2561 | if (limit > memcg->memory.limit) |
2562 | enlarge = true; | |
2563 | ret = page_counter_limit(&memcg->memory, limit); | |
2564 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2565 | |
2566 | if (!ret) | |
2567 | break; | |
2568 | ||
b70a2a21 JW |
2569 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
2570 | ||
3e32cb2e | 2571 | curusage = page_counter_read(&memcg->memory); |
81d39c20 | 2572 | /* Usage is reduced ? */ |
f894ffa8 | 2573 | if (curusage >= oldusage) |
81d39c20 KH |
2574 | retry_count--; |
2575 | else | |
2576 | oldusage = curusage; | |
3e32cb2e JW |
2577 | } while (retry_count); |
2578 | ||
3c11ecf4 KH |
2579 | if (!ret && enlarge) |
2580 | memcg_oom_recover(memcg); | |
14797e23 | 2581 | |
8c7c6e34 KH |
2582 | return ret; |
2583 | } | |
2584 | ||
338c8431 | 2585 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2586 | unsigned long limit) |
8c7c6e34 | 2587 | { |
3e32cb2e JW |
2588 | unsigned long curusage; |
2589 | unsigned long oldusage; | |
2590 | bool enlarge = false; | |
81d39c20 | 2591 | int retry_count; |
3e32cb2e | 2592 | int ret; |
8c7c6e34 | 2593 | |
81d39c20 | 2594 | /* see mem_cgroup_resize_res_limit */ |
3e32cb2e JW |
2595 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2596 | mem_cgroup_count_children(memcg); | |
2597 | ||
2598 | oldusage = page_counter_read(&memcg->memsw); | |
2599 | ||
2600 | do { | |
8c7c6e34 KH |
2601 | if (signal_pending(current)) { |
2602 | ret = -EINTR; | |
2603 | break; | |
2604 | } | |
3e32cb2e JW |
2605 | |
2606 | mutex_lock(&memcg_limit_mutex); | |
2607 | if (limit < memcg->memory.limit) { | |
2608 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2609 | ret = -EINVAL; |
8c7c6e34 KH |
2610 | break; |
2611 | } | |
3e32cb2e JW |
2612 | if (limit > memcg->memsw.limit) |
2613 | enlarge = true; | |
2614 | ret = page_counter_limit(&memcg->memsw, limit); | |
2615 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2616 | |
2617 | if (!ret) | |
2618 | break; | |
2619 | ||
b70a2a21 JW |
2620 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
2621 | ||
3e32cb2e | 2622 | curusage = page_counter_read(&memcg->memsw); |
81d39c20 | 2623 | /* Usage is reduced ? */ |
8c7c6e34 | 2624 | if (curusage >= oldusage) |
628f4235 | 2625 | retry_count--; |
81d39c20 KH |
2626 | else |
2627 | oldusage = curusage; | |
3e32cb2e JW |
2628 | } while (retry_count); |
2629 | ||
3c11ecf4 KH |
2630 | if (!ret && enlarge) |
2631 | memcg_oom_recover(memcg); | |
3e32cb2e | 2632 | |
628f4235 KH |
2633 | return ret; |
2634 | } | |
2635 | ||
0608f43d AM |
2636 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
2637 | gfp_t gfp_mask, | |
2638 | unsigned long *total_scanned) | |
2639 | { | |
2640 | unsigned long nr_reclaimed = 0; | |
2641 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
2642 | unsigned long reclaimed; | |
2643 | int loop = 0; | |
2644 | struct mem_cgroup_tree_per_zone *mctz; | |
3e32cb2e | 2645 | unsigned long excess; |
0608f43d AM |
2646 | unsigned long nr_scanned; |
2647 | ||
2648 | if (order > 0) | |
2649 | return 0; | |
2650 | ||
2651 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
2652 | /* | |
2653 | * This loop can run a while, specially if mem_cgroup's continuously | |
2654 | * keep exceeding their soft limit and putting the system under | |
2655 | * pressure | |
2656 | */ | |
2657 | do { | |
2658 | if (next_mz) | |
2659 | mz = next_mz; | |
2660 | else | |
2661 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2662 | if (!mz) | |
2663 | break; | |
2664 | ||
2665 | nr_scanned = 0; | |
2666 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
2667 | gfp_mask, &nr_scanned); | |
2668 | nr_reclaimed += reclaimed; | |
2669 | *total_scanned += nr_scanned; | |
0a31bc97 | 2670 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 2671 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
2672 | |
2673 | /* | |
2674 | * If we failed to reclaim anything from this memory cgroup | |
2675 | * it is time to move on to the next cgroup | |
2676 | */ | |
2677 | next_mz = NULL; | |
bc2f2e7f VD |
2678 | if (!reclaimed) |
2679 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
2680 | ||
3e32cb2e | 2681 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
2682 | /* |
2683 | * One school of thought says that we should not add | |
2684 | * back the node to the tree if reclaim returns 0. | |
2685 | * But our reclaim could return 0, simply because due | |
2686 | * to priority we are exposing a smaller subset of | |
2687 | * memory to reclaim from. Consider this as a longer | |
2688 | * term TODO. | |
2689 | */ | |
2690 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 2691 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 2692 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
2693 | css_put(&mz->memcg->css); |
2694 | loop++; | |
2695 | /* | |
2696 | * Could not reclaim anything and there are no more | |
2697 | * mem cgroups to try or we seem to be looping without | |
2698 | * reclaiming anything. | |
2699 | */ | |
2700 | if (!nr_reclaimed && | |
2701 | (next_mz == NULL || | |
2702 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2703 | break; | |
2704 | } while (!nr_reclaimed); | |
2705 | if (next_mz) | |
2706 | css_put(&next_mz->memcg->css); | |
2707 | return nr_reclaimed; | |
2708 | } | |
2709 | ||
ea280e7b TH |
2710 | /* |
2711 | * Test whether @memcg has children, dead or alive. Note that this | |
2712 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
2713 | * returns %true if there are child csses according to the cgroup | |
2714 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
2715 | */ | |
b5f99b53 GC |
2716 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
2717 | { | |
ea280e7b TH |
2718 | bool ret; |
2719 | ||
696ac172 | 2720 | /* |
ea280e7b TH |
2721 | * The lock does not prevent addition or deletion of children, but |
2722 | * it prevents a new child from being initialized based on this | |
2723 | * parent in css_online(), so it's enough to decide whether | |
2724 | * hierarchically inherited attributes can still be changed or not. | |
696ac172 | 2725 | */ |
ea280e7b TH |
2726 | lockdep_assert_held(&memcg_create_mutex); |
2727 | ||
2728 | rcu_read_lock(); | |
2729 | ret = css_next_child(NULL, &memcg->css); | |
2730 | rcu_read_unlock(); | |
2731 | return ret; | |
b5f99b53 GC |
2732 | } |
2733 | ||
c26251f9 MH |
2734 | /* |
2735 | * Reclaims as many pages from the given memcg as possible and moves | |
2736 | * the rest to the parent. | |
2737 | * | |
2738 | * Caller is responsible for holding css reference for memcg. | |
2739 | */ | |
2740 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
2741 | { | |
2742 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 2743 | |
c1e862c1 KH |
2744 | /* we call try-to-free pages for make this cgroup empty */ |
2745 | lru_add_drain_all(); | |
f817ed48 | 2746 | /* try to free all pages in this cgroup */ |
3e32cb2e | 2747 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 2748 | int progress; |
c1e862c1 | 2749 | |
c26251f9 MH |
2750 | if (signal_pending(current)) |
2751 | return -EINTR; | |
2752 | ||
b70a2a21 JW |
2753 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
2754 | GFP_KERNEL, true); | |
c1e862c1 | 2755 | if (!progress) { |
f817ed48 | 2756 | nr_retries--; |
c1e862c1 | 2757 | /* maybe some writeback is necessary */ |
8aa7e847 | 2758 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 2759 | } |
f817ed48 KH |
2760 | |
2761 | } | |
ab5196c2 MH |
2762 | |
2763 | return 0; | |
cc847582 KH |
2764 | } |
2765 | ||
6770c64e TH |
2766 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
2767 | char *buf, size_t nbytes, | |
2768 | loff_t off) | |
c1e862c1 | 2769 | { |
6770c64e | 2770 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 2771 | |
d8423011 MH |
2772 | if (mem_cgroup_is_root(memcg)) |
2773 | return -EINVAL; | |
6770c64e | 2774 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
2775 | } |
2776 | ||
182446d0 TH |
2777 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
2778 | struct cftype *cft) | |
18f59ea7 | 2779 | { |
182446d0 | 2780 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
2781 | } |
2782 | ||
182446d0 TH |
2783 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
2784 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
2785 | { |
2786 | int retval = 0; | |
182446d0 | 2787 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 2788 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 2789 | |
0999821b | 2790 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
2791 | |
2792 | if (memcg->use_hierarchy == val) | |
2793 | goto out; | |
2794 | ||
18f59ea7 | 2795 | /* |
af901ca1 | 2796 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
2797 | * in the child subtrees. If it is unset, then the change can |
2798 | * occur, provided the current cgroup has no children. | |
2799 | * | |
2800 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
2801 | * set if there are no children. | |
2802 | */ | |
c0ff4b85 | 2803 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 2804 | (val == 1 || val == 0)) { |
ea280e7b | 2805 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 2806 | memcg->use_hierarchy = val; |
18f59ea7 BS |
2807 | else |
2808 | retval = -EBUSY; | |
2809 | } else | |
2810 | retval = -EINVAL; | |
567fb435 GC |
2811 | |
2812 | out: | |
0999821b | 2813 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
2814 | |
2815 | return retval; | |
2816 | } | |
2817 | ||
3e32cb2e JW |
2818 | static unsigned long tree_stat(struct mem_cgroup *memcg, |
2819 | enum mem_cgroup_stat_index idx) | |
ce00a967 JW |
2820 | { |
2821 | struct mem_cgroup *iter; | |
2822 | long val = 0; | |
2823 | ||
2824 | /* Per-cpu values can be negative, use a signed accumulator */ | |
2825 | for_each_mem_cgroup_tree(iter, memcg) | |
2826 | val += mem_cgroup_read_stat(iter, idx); | |
2827 | ||
2828 | if (val < 0) /* race ? */ | |
2829 | val = 0; | |
2830 | return val; | |
2831 | } | |
2832 | ||
2833 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) | |
2834 | { | |
2835 | u64 val; | |
2836 | ||
3e32cb2e JW |
2837 | if (mem_cgroup_is_root(memcg)) { |
2838 | val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE); | |
2839 | val += tree_stat(memcg, MEM_CGROUP_STAT_RSS); | |
2840 | if (swap) | |
2841 | val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP); | |
2842 | } else { | |
ce00a967 | 2843 | if (!swap) |
3e32cb2e | 2844 | val = page_counter_read(&memcg->memory); |
ce00a967 | 2845 | else |
3e32cb2e | 2846 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 2847 | } |
ce00a967 JW |
2848 | return val << PAGE_SHIFT; |
2849 | } | |
2850 | ||
3e32cb2e JW |
2851 | enum { |
2852 | RES_USAGE, | |
2853 | RES_LIMIT, | |
2854 | RES_MAX_USAGE, | |
2855 | RES_FAILCNT, | |
2856 | RES_SOFT_LIMIT, | |
2857 | }; | |
ce00a967 | 2858 | |
791badbd | 2859 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 2860 | struct cftype *cft) |
8cdea7c0 | 2861 | { |
182446d0 | 2862 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 2863 | struct page_counter *counter; |
af36f906 | 2864 | |
3e32cb2e | 2865 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 2866 | case _MEM: |
3e32cb2e JW |
2867 | counter = &memcg->memory; |
2868 | break; | |
8c7c6e34 | 2869 | case _MEMSWAP: |
3e32cb2e JW |
2870 | counter = &memcg->memsw; |
2871 | break; | |
510fc4e1 | 2872 | case _KMEM: |
3e32cb2e | 2873 | counter = &memcg->kmem; |
510fc4e1 | 2874 | break; |
8c7c6e34 KH |
2875 | default: |
2876 | BUG(); | |
8c7c6e34 | 2877 | } |
3e32cb2e JW |
2878 | |
2879 | switch (MEMFILE_ATTR(cft->private)) { | |
2880 | case RES_USAGE: | |
2881 | if (counter == &memcg->memory) | |
2882 | return mem_cgroup_usage(memcg, false); | |
2883 | if (counter == &memcg->memsw) | |
2884 | return mem_cgroup_usage(memcg, true); | |
2885 | return (u64)page_counter_read(counter) * PAGE_SIZE; | |
2886 | case RES_LIMIT: | |
2887 | return (u64)counter->limit * PAGE_SIZE; | |
2888 | case RES_MAX_USAGE: | |
2889 | return (u64)counter->watermark * PAGE_SIZE; | |
2890 | case RES_FAILCNT: | |
2891 | return counter->failcnt; | |
2892 | case RES_SOFT_LIMIT: | |
2893 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
2894 | default: | |
2895 | BUG(); | |
2896 | } | |
8cdea7c0 | 2897 | } |
510fc4e1 | 2898 | |
510fc4e1 | 2899 | #ifdef CONFIG_MEMCG_KMEM |
8c0145b6 VD |
2900 | static int memcg_activate_kmem(struct mem_cgroup *memcg, |
2901 | unsigned long nr_pages) | |
d6441637 VD |
2902 | { |
2903 | int err = 0; | |
2904 | int memcg_id; | |
2905 | ||
2a4db7eb | 2906 | BUG_ON(memcg->kmemcg_id >= 0); |
2788cf0c | 2907 | BUG_ON(memcg->kmem_acct_activated); |
2a4db7eb | 2908 | BUG_ON(memcg->kmem_acct_active); |
d6441637 | 2909 | |
510fc4e1 GC |
2910 | /* |
2911 | * For simplicity, we won't allow this to be disabled. It also can't | |
2912 | * be changed if the cgroup has children already, or if tasks had | |
2913 | * already joined. | |
2914 | * | |
2915 | * If tasks join before we set the limit, a person looking at | |
2916 | * kmem.usage_in_bytes will have no way to determine when it took | |
2917 | * place, which makes the value quite meaningless. | |
2918 | * | |
2919 | * After it first became limited, changes in the value of the limit are | |
2920 | * of course permitted. | |
510fc4e1 | 2921 | */ |
0999821b | 2922 | mutex_lock(&memcg_create_mutex); |
ea280e7b TH |
2923 | if (cgroup_has_tasks(memcg->css.cgroup) || |
2924 | (memcg->use_hierarchy && memcg_has_children(memcg))) | |
d6441637 VD |
2925 | err = -EBUSY; |
2926 | mutex_unlock(&memcg_create_mutex); | |
2927 | if (err) | |
2928 | goto out; | |
510fc4e1 | 2929 | |
f3bb3043 | 2930 | memcg_id = memcg_alloc_cache_id(); |
d6441637 VD |
2931 | if (memcg_id < 0) { |
2932 | err = memcg_id; | |
2933 | goto out; | |
2934 | } | |
2935 | ||
d6441637 | 2936 | /* |
900a38f0 VD |
2937 | * We couldn't have accounted to this cgroup, because it hasn't got |
2938 | * activated yet, so this should succeed. | |
d6441637 | 2939 | */ |
3e32cb2e | 2940 | err = page_counter_limit(&memcg->kmem, nr_pages); |
d6441637 VD |
2941 | VM_BUG_ON(err); |
2942 | ||
2943 | static_key_slow_inc(&memcg_kmem_enabled_key); | |
2944 | /* | |
900a38f0 VD |
2945 | * A memory cgroup is considered kmem-active as soon as it gets |
2946 | * kmemcg_id. Setting the id after enabling static branching will | |
d6441637 VD |
2947 | * guarantee no one starts accounting before all call sites are |
2948 | * patched. | |
2949 | */ | |
900a38f0 | 2950 | memcg->kmemcg_id = memcg_id; |
2788cf0c | 2951 | memcg->kmem_acct_activated = true; |
2a4db7eb | 2952 | memcg->kmem_acct_active = true; |
510fc4e1 | 2953 | out: |
d6441637 | 2954 | return err; |
d6441637 VD |
2955 | } |
2956 | ||
d6441637 | 2957 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2958 | unsigned long limit) |
d6441637 VD |
2959 | { |
2960 | int ret; | |
2961 | ||
3e32cb2e | 2962 | mutex_lock(&memcg_limit_mutex); |
d6441637 | 2963 | if (!memcg_kmem_is_active(memcg)) |
3e32cb2e | 2964 | ret = memcg_activate_kmem(memcg, limit); |
d6441637 | 2965 | else |
3e32cb2e JW |
2966 | ret = page_counter_limit(&memcg->kmem, limit); |
2967 | mutex_unlock(&memcg_limit_mutex); | |
510fc4e1 GC |
2968 | return ret; |
2969 | } | |
2970 | ||
55007d84 | 2971 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 2972 | { |
55007d84 | 2973 | int ret = 0; |
510fc4e1 | 2974 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
55007d84 | 2975 | |
d6441637 VD |
2976 | if (!parent) |
2977 | return 0; | |
55007d84 | 2978 | |
8c0145b6 | 2979 | mutex_lock(&memcg_limit_mutex); |
55007d84 | 2980 | /* |
d6441637 VD |
2981 | * If the parent cgroup is not kmem-active now, it cannot be activated |
2982 | * after this point, because it has at least one child already. | |
55007d84 | 2983 | */ |
d6441637 | 2984 | if (memcg_kmem_is_active(parent)) |
8c0145b6 VD |
2985 | ret = memcg_activate_kmem(memcg, PAGE_COUNTER_MAX); |
2986 | mutex_unlock(&memcg_limit_mutex); | |
55007d84 | 2987 | return ret; |
510fc4e1 | 2988 | } |
d6441637 VD |
2989 | #else |
2990 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
3e32cb2e | 2991 | unsigned long limit) |
d6441637 VD |
2992 | { |
2993 | return -EINVAL; | |
2994 | } | |
6d043990 | 2995 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 2996 | |
628f4235 KH |
2997 | /* |
2998 | * The user of this function is... | |
2999 | * RES_LIMIT. | |
3000 | */ | |
451af504 TH |
3001 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3002 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3003 | { |
451af504 | 3004 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3005 | unsigned long nr_pages; |
628f4235 KH |
3006 | int ret; |
3007 | ||
451af504 | 3008 | buf = strstrip(buf); |
650c5e56 | 3009 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3010 | if (ret) |
3011 | return ret; | |
af36f906 | 3012 | |
3e32cb2e | 3013 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3014 | case RES_LIMIT: |
4b3bde4c BS |
3015 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3016 | ret = -EINVAL; | |
3017 | break; | |
3018 | } | |
3e32cb2e JW |
3019 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3020 | case _MEM: | |
3021 | ret = mem_cgroup_resize_limit(memcg, nr_pages); | |
8c7c6e34 | 3022 | break; |
3e32cb2e JW |
3023 | case _MEMSWAP: |
3024 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); | |
296c81d8 | 3025 | break; |
3e32cb2e JW |
3026 | case _KMEM: |
3027 | ret = memcg_update_kmem_limit(memcg, nr_pages); | |
3028 | break; | |
3029 | } | |
296c81d8 | 3030 | break; |
3e32cb2e JW |
3031 | case RES_SOFT_LIMIT: |
3032 | memcg->soft_limit = nr_pages; | |
3033 | ret = 0; | |
628f4235 KH |
3034 | break; |
3035 | } | |
451af504 | 3036 | return ret ?: nbytes; |
8cdea7c0 BS |
3037 | } |
3038 | ||
6770c64e TH |
3039 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3040 | size_t nbytes, loff_t off) | |
c84872e1 | 3041 | { |
6770c64e | 3042 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3043 | struct page_counter *counter; |
c84872e1 | 3044 | |
3e32cb2e JW |
3045 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3046 | case _MEM: | |
3047 | counter = &memcg->memory; | |
3048 | break; | |
3049 | case _MEMSWAP: | |
3050 | counter = &memcg->memsw; | |
3051 | break; | |
3052 | case _KMEM: | |
3053 | counter = &memcg->kmem; | |
3054 | break; | |
3055 | default: | |
3056 | BUG(); | |
3057 | } | |
af36f906 | 3058 | |
3e32cb2e | 3059 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3060 | case RES_MAX_USAGE: |
3e32cb2e | 3061 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3062 | break; |
3063 | case RES_FAILCNT: | |
3e32cb2e | 3064 | counter->failcnt = 0; |
29f2a4da | 3065 | break; |
3e32cb2e JW |
3066 | default: |
3067 | BUG(); | |
29f2a4da | 3068 | } |
f64c3f54 | 3069 | |
6770c64e | 3070 | return nbytes; |
c84872e1 PE |
3071 | } |
3072 | ||
182446d0 | 3073 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3074 | struct cftype *cft) |
3075 | { | |
182446d0 | 3076 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3077 | } |
3078 | ||
02491447 | 3079 | #ifdef CONFIG_MMU |
182446d0 | 3080 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3081 | struct cftype *cft, u64 val) |
3082 | { | |
182446d0 | 3083 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3084 | |
1dfab5ab | 3085 | if (val & ~MOVE_MASK) |
7dc74be0 | 3086 | return -EINVAL; |
ee5e8472 | 3087 | |
7dc74be0 | 3088 | /* |
ee5e8472 GC |
3089 | * No kind of locking is needed in here, because ->can_attach() will |
3090 | * check this value once in the beginning of the process, and then carry | |
3091 | * on with stale data. This means that changes to this value will only | |
3092 | * affect task migrations starting after the change. | |
7dc74be0 | 3093 | */ |
c0ff4b85 | 3094 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3095 | return 0; |
3096 | } | |
02491447 | 3097 | #else |
182446d0 | 3098 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3099 | struct cftype *cft, u64 val) |
3100 | { | |
3101 | return -ENOSYS; | |
3102 | } | |
3103 | #endif | |
7dc74be0 | 3104 | |
406eb0c9 | 3105 | #ifdef CONFIG_NUMA |
2da8ca82 | 3106 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3107 | { |
25485de6 GT |
3108 | struct numa_stat { |
3109 | const char *name; | |
3110 | unsigned int lru_mask; | |
3111 | }; | |
3112 | ||
3113 | static const struct numa_stat stats[] = { | |
3114 | { "total", LRU_ALL }, | |
3115 | { "file", LRU_ALL_FILE }, | |
3116 | { "anon", LRU_ALL_ANON }, | |
3117 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3118 | }; | |
3119 | const struct numa_stat *stat; | |
406eb0c9 | 3120 | int nid; |
25485de6 | 3121 | unsigned long nr; |
2da8ca82 | 3122 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 3123 | |
25485de6 GT |
3124 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3125 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3126 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3127 | for_each_node_state(nid, N_MEMORY) { | |
3128 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3129 | stat->lru_mask); | |
3130 | seq_printf(m, " N%d=%lu", nid, nr); | |
3131 | } | |
3132 | seq_putc(m, '\n'); | |
406eb0c9 | 3133 | } |
406eb0c9 | 3134 | |
071aee13 YH |
3135 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3136 | struct mem_cgroup *iter; | |
3137 | ||
3138 | nr = 0; | |
3139 | for_each_mem_cgroup_tree(iter, memcg) | |
3140 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3141 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3142 | for_each_node_state(nid, N_MEMORY) { | |
3143 | nr = 0; | |
3144 | for_each_mem_cgroup_tree(iter, memcg) | |
3145 | nr += mem_cgroup_node_nr_lru_pages( | |
3146 | iter, nid, stat->lru_mask); | |
3147 | seq_printf(m, " N%d=%lu", nid, nr); | |
3148 | } | |
3149 | seq_putc(m, '\n'); | |
406eb0c9 | 3150 | } |
406eb0c9 | 3151 | |
406eb0c9 YH |
3152 | return 0; |
3153 | } | |
3154 | #endif /* CONFIG_NUMA */ | |
3155 | ||
2da8ca82 | 3156 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3157 | { |
2da8ca82 | 3158 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 3159 | unsigned long memory, memsw; |
af7c4b0e JW |
3160 | struct mem_cgroup *mi; |
3161 | unsigned int i; | |
406eb0c9 | 3162 | |
0ca44b14 GT |
3163 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) != |
3164 | MEM_CGROUP_STAT_NSTATS); | |
3165 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) != | |
3166 | MEM_CGROUP_EVENTS_NSTATS); | |
70bc068c RS |
3167 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3168 | ||
af7c4b0e | 3169 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 3170 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 3171 | continue; |
af7c4b0e JW |
3172 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
3173 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 3174 | } |
7b854121 | 3175 | |
af7c4b0e JW |
3176 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
3177 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
3178 | mem_cgroup_read_events(memcg, i)); | |
3179 | ||
3180 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3181 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
3182 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
3183 | ||
14067bb3 | 3184 | /* Hierarchical information */ |
3e32cb2e JW |
3185 | memory = memsw = PAGE_COUNTER_MAX; |
3186 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
3187 | memory = min(memory, mi->memory.limit); | |
3188 | memsw = min(memsw, mi->memsw.limit); | |
fee7b548 | 3189 | } |
3e32cb2e JW |
3190 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3191 | (u64)memory * PAGE_SIZE); | |
3192 | if (do_swap_account) | |
3193 | seq_printf(m, "hierarchical_memsw_limit %llu\n", | |
3194 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3195 | |
af7c4b0e JW |
3196 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
3197 | long long val = 0; | |
3198 | ||
bff6bb83 | 3199 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 3200 | continue; |
af7c4b0e JW |
3201 | for_each_mem_cgroup_tree(mi, memcg) |
3202 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
3203 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
3204 | } | |
3205 | ||
3206 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
3207 | unsigned long long val = 0; | |
3208 | ||
3209 | for_each_mem_cgroup_tree(mi, memcg) | |
3210 | val += mem_cgroup_read_events(mi, i); | |
3211 | seq_printf(m, "total_%s %llu\n", | |
3212 | mem_cgroup_events_names[i], val); | |
3213 | } | |
3214 | ||
3215 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
3216 | unsigned long long val = 0; | |
3217 | ||
3218 | for_each_mem_cgroup_tree(mi, memcg) | |
3219 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
3220 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 3221 | } |
14067bb3 | 3222 | |
7f016ee8 | 3223 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
3224 | { |
3225 | int nid, zid; | |
3226 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 3227 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3228 | unsigned long recent_rotated[2] = {0, 0}; |
3229 | unsigned long recent_scanned[2] = {0, 0}; | |
3230 | ||
3231 | for_each_online_node(nid) | |
3232 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
e231875b | 3233 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
89abfab1 | 3234 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 3235 | |
89abfab1 HD |
3236 | recent_rotated[0] += rstat->recent_rotated[0]; |
3237 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3238 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3239 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 3240 | } |
78ccf5b5 JW |
3241 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3242 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3243 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3244 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3245 | } |
3246 | #endif | |
3247 | ||
d2ceb9b7 KH |
3248 | return 0; |
3249 | } | |
3250 | ||
182446d0 TH |
3251 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3252 | struct cftype *cft) | |
a7885eb8 | 3253 | { |
182446d0 | 3254 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3255 | |
1f4c025b | 3256 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3257 | } |
3258 | ||
182446d0 TH |
3259 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3260 | struct cftype *cft, u64 val) | |
a7885eb8 | 3261 | { |
182446d0 | 3262 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3263 | |
3dae7fec | 3264 | if (val > 100) |
a7885eb8 KM |
3265 | return -EINVAL; |
3266 | ||
14208b0e | 3267 | if (css->parent) |
3dae7fec JW |
3268 | memcg->swappiness = val; |
3269 | else | |
3270 | vm_swappiness = val; | |
068b38c1 | 3271 | |
a7885eb8 KM |
3272 | return 0; |
3273 | } | |
3274 | ||
2e72b634 KS |
3275 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3276 | { | |
3277 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3278 | unsigned long usage; |
2e72b634 KS |
3279 | int i; |
3280 | ||
3281 | rcu_read_lock(); | |
3282 | if (!swap) | |
2c488db2 | 3283 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3284 | else |
2c488db2 | 3285 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3286 | |
3287 | if (!t) | |
3288 | goto unlock; | |
3289 | ||
ce00a967 | 3290 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3291 | |
3292 | /* | |
748dad36 | 3293 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3294 | * If it's not true, a threshold was crossed after last |
3295 | * call of __mem_cgroup_threshold(). | |
3296 | */ | |
5407a562 | 3297 | i = t->current_threshold; |
2e72b634 KS |
3298 | |
3299 | /* | |
3300 | * Iterate backward over array of thresholds starting from | |
3301 | * current_threshold and check if a threshold is crossed. | |
3302 | * If none of thresholds below usage is crossed, we read | |
3303 | * only one element of the array here. | |
3304 | */ | |
3305 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3306 | eventfd_signal(t->entries[i].eventfd, 1); | |
3307 | ||
3308 | /* i = current_threshold + 1 */ | |
3309 | i++; | |
3310 | ||
3311 | /* | |
3312 | * Iterate forward over array of thresholds starting from | |
3313 | * current_threshold+1 and check if a threshold is crossed. | |
3314 | * If none of thresholds above usage is crossed, we read | |
3315 | * only one element of the array here. | |
3316 | */ | |
3317 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3318 | eventfd_signal(t->entries[i].eventfd, 1); | |
3319 | ||
3320 | /* Update current_threshold */ | |
5407a562 | 3321 | t->current_threshold = i - 1; |
2e72b634 KS |
3322 | unlock: |
3323 | rcu_read_unlock(); | |
3324 | } | |
3325 | ||
3326 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3327 | { | |
ad4ca5f4 KS |
3328 | while (memcg) { |
3329 | __mem_cgroup_threshold(memcg, false); | |
3330 | if (do_swap_account) | |
3331 | __mem_cgroup_threshold(memcg, true); | |
3332 | ||
3333 | memcg = parent_mem_cgroup(memcg); | |
3334 | } | |
2e72b634 KS |
3335 | } |
3336 | ||
3337 | static int compare_thresholds(const void *a, const void *b) | |
3338 | { | |
3339 | const struct mem_cgroup_threshold *_a = a; | |
3340 | const struct mem_cgroup_threshold *_b = b; | |
3341 | ||
2bff24a3 GT |
3342 | if (_a->threshold > _b->threshold) |
3343 | return 1; | |
3344 | ||
3345 | if (_a->threshold < _b->threshold) | |
3346 | return -1; | |
3347 | ||
3348 | return 0; | |
2e72b634 KS |
3349 | } |
3350 | ||
c0ff4b85 | 3351 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3352 | { |
3353 | struct mem_cgroup_eventfd_list *ev; | |
3354 | ||
2bcf2e92 MH |
3355 | spin_lock(&memcg_oom_lock); |
3356 | ||
c0ff4b85 | 3357 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3358 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3359 | |
3360 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3361 | return 0; |
3362 | } | |
3363 | ||
c0ff4b85 | 3364 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3365 | { |
7d74b06f KH |
3366 | struct mem_cgroup *iter; |
3367 | ||
c0ff4b85 | 3368 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3369 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3370 | } |
3371 | ||
59b6f873 | 3372 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3373 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3374 | { |
2c488db2 KS |
3375 | struct mem_cgroup_thresholds *thresholds; |
3376 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3377 | unsigned long threshold; |
3378 | unsigned long usage; | |
2c488db2 | 3379 | int i, size, ret; |
2e72b634 | 3380 | |
650c5e56 | 3381 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
3382 | if (ret) |
3383 | return ret; | |
3384 | ||
3385 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3386 | |
05b84301 | 3387 | if (type == _MEM) { |
2c488db2 | 3388 | thresholds = &memcg->thresholds; |
ce00a967 | 3389 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3390 | } else if (type == _MEMSWAP) { |
2c488db2 | 3391 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3392 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3393 | } else |
2e72b634 KS |
3394 | BUG(); |
3395 | ||
2e72b634 | 3396 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3397 | if (thresholds->primary) |
2e72b634 KS |
3398 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3399 | ||
2c488db2 | 3400 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3401 | |
3402 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 3403 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 3404 | GFP_KERNEL); |
2c488db2 | 3405 | if (!new) { |
2e72b634 KS |
3406 | ret = -ENOMEM; |
3407 | goto unlock; | |
3408 | } | |
2c488db2 | 3409 | new->size = size; |
2e72b634 KS |
3410 | |
3411 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3412 | if (thresholds->primary) { |
3413 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3414 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3415 | } |
3416 | ||
2e72b634 | 3417 | /* Add new threshold */ |
2c488db2 KS |
3418 | new->entries[size - 1].eventfd = eventfd; |
3419 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3420 | |
3421 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3422 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3423 | compare_thresholds, NULL); |
3424 | ||
3425 | /* Find current threshold */ | |
2c488db2 | 3426 | new->current_threshold = -1; |
2e72b634 | 3427 | for (i = 0; i < size; i++) { |
748dad36 | 3428 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3429 | /* |
2c488db2 KS |
3430 | * new->current_threshold will not be used until |
3431 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3432 | * it here. |
3433 | */ | |
2c488db2 | 3434 | ++new->current_threshold; |
748dad36 SZ |
3435 | } else |
3436 | break; | |
2e72b634 KS |
3437 | } |
3438 | ||
2c488db2 KS |
3439 | /* Free old spare buffer and save old primary buffer as spare */ |
3440 | kfree(thresholds->spare); | |
3441 | thresholds->spare = thresholds->primary; | |
3442 | ||
3443 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3444 | |
907860ed | 3445 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3446 | synchronize_rcu(); |
3447 | ||
2e72b634 KS |
3448 | unlock: |
3449 | mutex_unlock(&memcg->thresholds_lock); | |
3450 | ||
3451 | return ret; | |
3452 | } | |
3453 | ||
59b6f873 | 3454 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3455 | struct eventfd_ctx *eventfd, const char *args) |
3456 | { | |
59b6f873 | 3457 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3458 | } |
3459 | ||
59b6f873 | 3460 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3461 | struct eventfd_ctx *eventfd, const char *args) |
3462 | { | |
59b6f873 | 3463 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3464 | } |
3465 | ||
59b6f873 | 3466 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3467 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3468 | { |
2c488db2 KS |
3469 | struct mem_cgroup_thresholds *thresholds; |
3470 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3471 | unsigned long usage; |
2c488db2 | 3472 | int i, j, size; |
2e72b634 KS |
3473 | |
3474 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3475 | |
3476 | if (type == _MEM) { | |
2c488db2 | 3477 | thresholds = &memcg->thresholds; |
ce00a967 | 3478 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3479 | } else if (type == _MEMSWAP) { |
2c488db2 | 3480 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3481 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3482 | } else |
2e72b634 KS |
3483 | BUG(); |
3484 | ||
371528ca AV |
3485 | if (!thresholds->primary) |
3486 | goto unlock; | |
3487 | ||
2e72b634 KS |
3488 | /* Check if a threshold crossed before removing */ |
3489 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3490 | ||
3491 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3492 | size = 0; |
3493 | for (i = 0; i < thresholds->primary->size; i++) { | |
3494 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3495 | size++; |
3496 | } | |
3497 | ||
2c488db2 | 3498 | new = thresholds->spare; |
907860ed | 3499 | |
2e72b634 KS |
3500 | /* Set thresholds array to NULL if we don't have thresholds */ |
3501 | if (!size) { | |
2c488db2 KS |
3502 | kfree(new); |
3503 | new = NULL; | |
907860ed | 3504 | goto swap_buffers; |
2e72b634 KS |
3505 | } |
3506 | ||
2c488db2 | 3507 | new->size = size; |
2e72b634 KS |
3508 | |
3509 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3510 | new->current_threshold = -1; |
3511 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3512 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3513 | continue; |
3514 | ||
2c488db2 | 3515 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3516 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3517 | /* |
2c488db2 | 3518 | * new->current_threshold will not be used |
2e72b634 KS |
3519 | * until rcu_assign_pointer(), so it's safe to increment |
3520 | * it here. | |
3521 | */ | |
2c488db2 | 3522 | ++new->current_threshold; |
2e72b634 KS |
3523 | } |
3524 | j++; | |
3525 | } | |
3526 | ||
907860ed | 3527 | swap_buffers: |
2c488db2 KS |
3528 | /* Swap primary and spare array */ |
3529 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
3530 | /* If all events are unregistered, free the spare array */ |
3531 | if (!new) { | |
3532 | kfree(thresholds->spare); | |
3533 | thresholds->spare = NULL; | |
3534 | } | |
3535 | ||
2c488db2 | 3536 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3537 | |
907860ed | 3538 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3539 | synchronize_rcu(); |
371528ca | 3540 | unlock: |
2e72b634 | 3541 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3542 | } |
c1e862c1 | 3543 | |
59b6f873 | 3544 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3545 | struct eventfd_ctx *eventfd) |
3546 | { | |
59b6f873 | 3547 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
3548 | } |
3549 | ||
59b6f873 | 3550 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3551 | struct eventfd_ctx *eventfd) |
3552 | { | |
59b6f873 | 3553 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
3554 | } |
3555 | ||
59b6f873 | 3556 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3557 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 3558 | { |
9490ff27 | 3559 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 3560 | |
9490ff27 KH |
3561 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
3562 | if (!event) | |
3563 | return -ENOMEM; | |
3564 | ||
1af8efe9 | 3565 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
3566 | |
3567 | event->eventfd = eventfd; | |
3568 | list_add(&event->list, &memcg->oom_notify); | |
3569 | ||
3570 | /* already in OOM ? */ | |
c2b42d3c | 3571 | if (memcg->under_oom) |
9490ff27 | 3572 | eventfd_signal(eventfd, 1); |
1af8efe9 | 3573 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3574 | |
3575 | return 0; | |
3576 | } | |
3577 | ||
59b6f873 | 3578 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3579 | struct eventfd_ctx *eventfd) |
9490ff27 | 3580 | { |
9490ff27 | 3581 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 3582 | |
1af8efe9 | 3583 | spin_lock(&memcg_oom_lock); |
9490ff27 | 3584 | |
c0ff4b85 | 3585 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
3586 | if (ev->eventfd == eventfd) { |
3587 | list_del(&ev->list); | |
3588 | kfree(ev); | |
3589 | } | |
3590 | } | |
3591 | ||
1af8efe9 | 3592 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3593 | } |
3594 | ||
2da8ca82 | 3595 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 3596 | { |
2da8ca82 | 3597 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 3598 | |
791badbd | 3599 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 3600 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
3c11ecf4 KH |
3601 | return 0; |
3602 | } | |
3603 | ||
182446d0 | 3604 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
3605 | struct cftype *cft, u64 val) |
3606 | { | |
182446d0 | 3607 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
3608 | |
3609 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 3610 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
3611 | return -EINVAL; |
3612 | ||
c0ff4b85 | 3613 | memcg->oom_kill_disable = val; |
4d845ebf | 3614 | if (!val) |
c0ff4b85 | 3615 | memcg_oom_recover(memcg); |
3dae7fec | 3616 | |
3c11ecf4 KH |
3617 | return 0; |
3618 | } | |
3619 | ||
c255a458 | 3620 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 3621 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 3622 | { |
55007d84 GC |
3623 | int ret; |
3624 | ||
55007d84 GC |
3625 | ret = memcg_propagate_kmem(memcg); |
3626 | if (ret) | |
3627 | return ret; | |
2633d7a0 | 3628 | |
1d62e436 | 3629 | return mem_cgroup_sockets_init(memcg, ss); |
573b400d | 3630 | } |
e5671dfa | 3631 | |
2a4db7eb VD |
3632 | static void memcg_deactivate_kmem(struct mem_cgroup *memcg) |
3633 | { | |
2788cf0c VD |
3634 | struct cgroup_subsys_state *css; |
3635 | struct mem_cgroup *parent, *child; | |
3636 | int kmemcg_id; | |
3637 | ||
2a4db7eb VD |
3638 | if (!memcg->kmem_acct_active) |
3639 | return; | |
3640 | ||
3641 | /* | |
3642 | * Clear the 'active' flag before clearing memcg_caches arrays entries. | |
3643 | * Since we take the slab_mutex in memcg_deactivate_kmem_caches(), it | |
3644 | * guarantees no cache will be created for this cgroup after we are | |
3645 | * done (see memcg_create_kmem_cache()). | |
3646 | */ | |
3647 | memcg->kmem_acct_active = false; | |
3648 | ||
3649 | memcg_deactivate_kmem_caches(memcg); | |
2788cf0c VD |
3650 | |
3651 | kmemcg_id = memcg->kmemcg_id; | |
3652 | BUG_ON(kmemcg_id < 0); | |
3653 | ||
3654 | parent = parent_mem_cgroup(memcg); | |
3655 | if (!parent) | |
3656 | parent = root_mem_cgroup; | |
3657 | ||
3658 | /* | |
3659 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3660 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3661 | * to ones of the parent. After we have finished, all list_lrus | |
3662 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3663 | * ordering is imposed by list_lru_node->lock taken by | |
3664 | * memcg_drain_all_list_lrus(). | |
3665 | */ | |
3666 | css_for_each_descendant_pre(css, &memcg->css) { | |
3667 | child = mem_cgroup_from_css(css); | |
3668 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3669 | child->kmemcg_id = parent->kmemcg_id; | |
3670 | if (!memcg->use_hierarchy) | |
3671 | break; | |
3672 | } | |
3673 | memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id); | |
3674 | ||
3675 | memcg_free_cache_id(kmemcg_id); | |
2a4db7eb VD |
3676 | } |
3677 | ||
10d5ebf4 | 3678 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 3679 | { |
f48b80a5 VD |
3680 | if (memcg->kmem_acct_activated) { |
3681 | memcg_destroy_kmem_caches(memcg); | |
3682 | static_key_slow_dec(&memcg_kmem_enabled_key); | |
3683 | WARN_ON(page_counter_read(&memcg->kmem)); | |
3684 | } | |
1d62e436 | 3685 | mem_cgroup_sockets_destroy(memcg); |
10d5ebf4 | 3686 | } |
e5671dfa | 3687 | #else |
cbe128e3 | 3688 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
3689 | { |
3690 | return 0; | |
3691 | } | |
d1a4c0b3 | 3692 | |
2a4db7eb VD |
3693 | static void memcg_deactivate_kmem(struct mem_cgroup *memcg) |
3694 | { | |
3695 | } | |
3696 | ||
10d5ebf4 LZ |
3697 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
3698 | { | |
3699 | } | |
e5671dfa GC |
3700 | #endif |
3701 | ||
52ebea74 TH |
3702 | #ifdef CONFIG_CGROUP_WRITEBACK |
3703 | ||
3704 | struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg) | |
3705 | { | |
3706 | return &memcg->cgwb_list; | |
3707 | } | |
3708 | ||
841710aa TH |
3709 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
3710 | { | |
3711 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
3712 | } | |
3713 | ||
3714 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3715 | { | |
3716 | wb_domain_exit(&memcg->cgwb_domain); | |
3717 | } | |
3718 | ||
2529bb3a TH |
3719 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3720 | { | |
3721 | wb_domain_size_changed(&memcg->cgwb_domain); | |
3722 | } | |
3723 | ||
841710aa TH |
3724 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
3725 | { | |
3726 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3727 | ||
3728 | if (!memcg->css.parent) | |
3729 | return NULL; | |
3730 | ||
3731 | return &memcg->cgwb_domain; | |
3732 | } | |
3733 | ||
c2aa723a TH |
3734 | /** |
3735 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
3736 | * @wb: bdi_writeback in question | |
3737 | * @pavail: out parameter for number of available pages | |
3738 | * @pdirty: out parameter for number of dirty pages | |
3739 | * @pwriteback: out parameter for number of pages under writeback | |
3740 | * | |
3741 | * Determine the numbers of available, dirty, and writeback pages in @wb's | |
3742 | * memcg. Dirty and writeback are self-explanatory. Available is a bit | |
3743 | * more involved. | |
3744 | * | |
3745 | * A memcg's headroom is "min(max, high) - used". The available memory is | |
3746 | * calculated as the lowest headroom of itself and the ancestors plus the | |
3747 | * number of pages already being used for file pages. Note that this | |
3748 | * doesn't consider the actual amount of available memory in the system. | |
3749 | * The caller should further cap *@pavail accordingly. | |
3750 | */ | |
3751 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pavail, | |
3752 | unsigned long *pdirty, unsigned long *pwriteback) | |
3753 | { | |
3754 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3755 | struct mem_cgroup *parent; | |
3756 | unsigned long head_room = PAGE_COUNTER_MAX; | |
3757 | unsigned long file_pages; | |
3758 | ||
3759 | *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY); | |
3760 | ||
3761 | /* this should eventually include NR_UNSTABLE_NFS */ | |
3762 | *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK); | |
3763 | ||
3764 | file_pages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) | | |
3765 | (1 << LRU_ACTIVE_FILE)); | |
3766 | while ((parent = parent_mem_cgroup(memcg))) { | |
3767 | unsigned long ceiling = min(memcg->memory.limit, memcg->high); | |
3768 | unsigned long used = page_counter_read(&memcg->memory); | |
3769 | ||
3770 | head_room = min(head_room, ceiling - min(ceiling, used)); | |
3771 | memcg = parent; | |
3772 | } | |
3773 | ||
3774 | *pavail = file_pages + head_room; | |
3775 | } | |
3776 | ||
841710aa TH |
3777 | #else /* CONFIG_CGROUP_WRITEBACK */ |
3778 | ||
3779 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
3780 | { | |
3781 | return 0; | |
3782 | } | |
3783 | ||
3784 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3785 | { | |
3786 | } | |
3787 | ||
2529bb3a TH |
3788 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3789 | { | |
3790 | } | |
3791 | ||
52ebea74 TH |
3792 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
3793 | ||
3bc942f3 TH |
3794 | /* |
3795 | * DO NOT USE IN NEW FILES. | |
3796 | * | |
3797 | * "cgroup.event_control" implementation. | |
3798 | * | |
3799 | * This is way over-engineered. It tries to support fully configurable | |
3800 | * events for each user. Such level of flexibility is completely | |
3801 | * unnecessary especially in the light of the planned unified hierarchy. | |
3802 | * | |
3803 | * Please deprecate this and replace with something simpler if at all | |
3804 | * possible. | |
3805 | */ | |
3806 | ||
79bd9814 TH |
3807 | /* |
3808 | * Unregister event and free resources. | |
3809 | * | |
3810 | * Gets called from workqueue. | |
3811 | */ | |
3bc942f3 | 3812 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 3813 | { |
3bc942f3 TH |
3814 | struct mem_cgroup_event *event = |
3815 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 3816 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3817 | |
3818 | remove_wait_queue(event->wqh, &event->wait); | |
3819 | ||
59b6f873 | 3820 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
3821 | |
3822 | /* Notify userspace the event is going away. */ | |
3823 | eventfd_signal(event->eventfd, 1); | |
3824 | ||
3825 | eventfd_ctx_put(event->eventfd); | |
3826 | kfree(event); | |
59b6f873 | 3827 | css_put(&memcg->css); |
79bd9814 TH |
3828 | } |
3829 | ||
3830 | /* | |
3831 | * Gets called on POLLHUP on eventfd when user closes it. | |
3832 | * | |
3833 | * Called with wqh->lock held and interrupts disabled. | |
3834 | */ | |
3bc942f3 TH |
3835 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
3836 | int sync, void *key) | |
79bd9814 | 3837 | { |
3bc942f3 TH |
3838 | struct mem_cgroup_event *event = |
3839 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 3840 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3841 | unsigned long flags = (unsigned long)key; |
3842 | ||
3843 | if (flags & POLLHUP) { | |
3844 | /* | |
3845 | * If the event has been detached at cgroup removal, we | |
3846 | * can simply return knowing the other side will cleanup | |
3847 | * for us. | |
3848 | * | |
3849 | * We can't race against event freeing since the other | |
3850 | * side will require wqh->lock via remove_wait_queue(), | |
3851 | * which we hold. | |
3852 | */ | |
fba94807 | 3853 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
3854 | if (!list_empty(&event->list)) { |
3855 | list_del_init(&event->list); | |
3856 | /* | |
3857 | * We are in atomic context, but cgroup_event_remove() | |
3858 | * may sleep, so we have to call it in workqueue. | |
3859 | */ | |
3860 | schedule_work(&event->remove); | |
3861 | } | |
fba94807 | 3862 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
3863 | } |
3864 | ||
3865 | return 0; | |
3866 | } | |
3867 | ||
3bc942f3 | 3868 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
3869 | wait_queue_head_t *wqh, poll_table *pt) |
3870 | { | |
3bc942f3 TH |
3871 | struct mem_cgroup_event *event = |
3872 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
3873 | |
3874 | event->wqh = wqh; | |
3875 | add_wait_queue(wqh, &event->wait); | |
3876 | } | |
3877 | ||
3878 | /* | |
3bc942f3 TH |
3879 | * DO NOT USE IN NEW FILES. |
3880 | * | |
79bd9814 TH |
3881 | * Parse input and register new cgroup event handler. |
3882 | * | |
3883 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
3884 | * Interpretation of args is defined by control file implementation. | |
3885 | */ | |
451af504 TH |
3886 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
3887 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 3888 | { |
451af504 | 3889 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 3890 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 3891 | struct mem_cgroup_event *event; |
79bd9814 TH |
3892 | struct cgroup_subsys_state *cfile_css; |
3893 | unsigned int efd, cfd; | |
3894 | struct fd efile; | |
3895 | struct fd cfile; | |
fba94807 | 3896 | const char *name; |
79bd9814 TH |
3897 | char *endp; |
3898 | int ret; | |
3899 | ||
451af504 TH |
3900 | buf = strstrip(buf); |
3901 | ||
3902 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
3903 | if (*endp != ' ') |
3904 | return -EINVAL; | |
451af504 | 3905 | buf = endp + 1; |
79bd9814 | 3906 | |
451af504 | 3907 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
3908 | if ((*endp != ' ') && (*endp != '\0')) |
3909 | return -EINVAL; | |
451af504 | 3910 | buf = endp + 1; |
79bd9814 TH |
3911 | |
3912 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
3913 | if (!event) | |
3914 | return -ENOMEM; | |
3915 | ||
59b6f873 | 3916 | event->memcg = memcg; |
79bd9814 | 3917 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
3918 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
3919 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
3920 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
3921 | |
3922 | efile = fdget(efd); | |
3923 | if (!efile.file) { | |
3924 | ret = -EBADF; | |
3925 | goto out_kfree; | |
3926 | } | |
3927 | ||
3928 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
3929 | if (IS_ERR(event->eventfd)) { | |
3930 | ret = PTR_ERR(event->eventfd); | |
3931 | goto out_put_efile; | |
3932 | } | |
3933 | ||
3934 | cfile = fdget(cfd); | |
3935 | if (!cfile.file) { | |
3936 | ret = -EBADF; | |
3937 | goto out_put_eventfd; | |
3938 | } | |
3939 | ||
3940 | /* the process need read permission on control file */ | |
3941 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
3942 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
3943 | if (ret < 0) | |
3944 | goto out_put_cfile; | |
3945 | ||
fba94807 TH |
3946 | /* |
3947 | * Determine the event callbacks and set them in @event. This used | |
3948 | * to be done via struct cftype but cgroup core no longer knows | |
3949 | * about these events. The following is crude but the whole thing | |
3950 | * is for compatibility anyway. | |
3bc942f3 TH |
3951 | * |
3952 | * DO NOT ADD NEW FILES. | |
fba94807 | 3953 | */ |
b583043e | 3954 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
3955 | |
3956 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
3957 | event->register_event = mem_cgroup_usage_register_event; | |
3958 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
3959 | } else if (!strcmp(name, "memory.oom_control")) { | |
3960 | event->register_event = mem_cgroup_oom_register_event; | |
3961 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
3962 | } else if (!strcmp(name, "memory.pressure_level")) { | |
3963 | event->register_event = vmpressure_register_event; | |
3964 | event->unregister_event = vmpressure_unregister_event; | |
3965 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
3966 | event->register_event = memsw_cgroup_usage_register_event; |
3967 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
3968 | } else { |
3969 | ret = -EINVAL; | |
3970 | goto out_put_cfile; | |
3971 | } | |
3972 | ||
79bd9814 | 3973 | /* |
b5557c4c TH |
3974 | * Verify @cfile should belong to @css. Also, remaining events are |
3975 | * automatically removed on cgroup destruction but the removal is | |
3976 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 3977 | */ |
b583043e | 3978 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 3979 | &memory_cgrp_subsys); |
79bd9814 | 3980 | ret = -EINVAL; |
5a17f543 | 3981 | if (IS_ERR(cfile_css)) |
79bd9814 | 3982 | goto out_put_cfile; |
5a17f543 TH |
3983 | if (cfile_css != css) { |
3984 | css_put(cfile_css); | |
79bd9814 | 3985 | goto out_put_cfile; |
5a17f543 | 3986 | } |
79bd9814 | 3987 | |
451af504 | 3988 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
3989 | if (ret) |
3990 | goto out_put_css; | |
3991 | ||
3992 | efile.file->f_op->poll(efile.file, &event->pt); | |
3993 | ||
fba94807 TH |
3994 | spin_lock(&memcg->event_list_lock); |
3995 | list_add(&event->list, &memcg->event_list); | |
3996 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
3997 | |
3998 | fdput(cfile); | |
3999 | fdput(efile); | |
4000 | ||
451af504 | 4001 | return nbytes; |
79bd9814 TH |
4002 | |
4003 | out_put_css: | |
b5557c4c | 4004 | css_put(css); |
79bd9814 TH |
4005 | out_put_cfile: |
4006 | fdput(cfile); | |
4007 | out_put_eventfd: | |
4008 | eventfd_ctx_put(event->eventfd); | |
4009 | out_put_efile: | |
4010 | fdput(efile); | |
4011 | out_kfree: | |
4012 | kfree(event); | |
4013 | ||
4014 | return ret; | |
4015 | } | |
4016 | ||
241994ed | 4017 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4018 | { |
0eea1030 | 4019 | .name = "usage_in_bytes", |
8c7c6e34 | 4020 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4021 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4022 | }, |
c84872e1 PE |
4023 | { |
4024 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4025 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4026 | .write = mem_cgroup_reset, |
791badbd | 4027 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4028 | }, |
8cdea7c0 | 4029 | { |
0eea1030 | 4030 | .name = "limit_in_bytes", |
8c7c6e34 | 4031 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4032 | .write = mem_cgroup_write, |
791badbd | 4033 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4034 | }, |
296c81d8 BS |
4035 | { |
4036 | .name = "soft_limit_in_bytes", | |
4037 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4038 | .write = mem_cgroup_write, |
791badbd | 4039 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4040 | }, |
8cdea7c0 BS |
4041 | { |
4042 | .name = "failcnt", | |
8c7c6e34 | 4043 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4044 | .write = mem_cgroup_reset, |
791badbd | 4045 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4046 | }, |
d2ceb9b7 KH |
4047 | { |
4048 | .name = "stat", | |
2da8ca82 | 4049 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4050 | }, |
c1e862c1 KH |
4051 | { |
4052 | .name = "force_empty", | |
6770c64e | 4053 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4054 | }, |
18f59ea7 BS |
4055 | { |
4056 | .name = "use_hierarchy", | |
4057 | .write_u64 = mem_cgroup_hierarchy_write, | |
4058 | .read_u64 = mem_cgroup_hierarchy_read, | |
4059 | }, | |
79bd9814 | 4060 | { |
3bc942f3 | 4061 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4062 | .write = memcg_write_event_control, |
79bd9814 TH |
4063 | .flags = CFTYPE_NO_PREFIX, |
4064 | .mode = S_IWUGO, | |
4065 | }, | |
a7885eb8 KM |
4066 | { |
4067 | .name = "swappiness", | |
4068 | .read_u64 = mem_cgroup_swappiness_read, | |
4069 | .write_u64 = mem_cgroup_swappiness_write, | |
4070 | }, | |
7dc74be0 DN |
4071 | { |
4072 | .name = "move_charge_at_immigrate", | |
4073 | .read_u64 = mem_cgroup_move_charge_read, | |
4074 | .write_u64 = mem_cgroup_move_charge_write, | |
4075 | }, | |
9490ff27 KH |
4076 | { |
4077 | .name = "oom_control", | |
2da8ca82 | 4078 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4079 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4080 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4081 | }, | |
70ddf637 AV |
4082 | { |
4083 | .name = "pressure_level", | |
70ddf637 | 4084 | }, |
406eb0c9 YH |
4085 | #ifdef CONFIG_NUMA |
4086 | { | |
4087 | .name = "numa_stat", | |
2da8ca82 | 4088 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4089 | }, |
4090 | #endif | |
510fc4e1 GC |
4091 | #ifdef CONFIG_MEMCG_KMEM |
4092 | { | |
4093 | .name = "kmem.limit_in_bytes", | |
4094 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4095 | .write = mem_cgroup_write, |
791badbd | 4096 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4097 | }, |
4098 | { | |
4099 | .name = "kmem.usage_in_bytes", | |
4100 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4101 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4102 | }, |
4103 | { | |
4104 | .name = "kmem.failcnt", | |
4105 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4106 | .write = mem_cgroup_reset, |
791badbd | 4107 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4108 | }, |
4109 | { | |
4110 | .name = "kmem.max_usage_in_bytes", | |
4111 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4112 | .write = mem_cgroup_reset, |
791badbd | 4113 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4114 | }, |
749c5415 GC |
4115 | #ifdef CONFIG_SLABINFO |
4116 | { | |
4117 | .name = "kmem.slabinfo", | |
b047501c VD |
4118 | .seq_start = slab_start, |
4119 | .seq_next = slab_next, | |
4120 | .seq_stop = slab_stop, | |
4121 | .seq_show = memcg_slab_show, | |
749c5415 GC |
4122 | }, |
4123 | #endif | |
8c7c6e34 | 4124 | #endif |
6bc10349 | 4125 | { }, /* terminate */ |
af36f906 | 4126 | }; |
8c7c6e34 | 4127 | |
c0ff4b85 | 4128 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4129 | { |
4130 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4131 | struct mem_cgroup_per_zone *mz; |
41e3355d | 4132 | int zone, tmp = node; |
1ecaab2b KH |
4133 | /* |
4134 | * This routine is called against possible nodes. | |
4135 | * But it's BUG to call kmalloc() against offline node. | |
4136 | * | |
4137 | * TODO: this routine can waste much memory for nodes which will | |
4138 | * never be onlined. It's better to use memory hotplug callback | |
4139 | * function. | |
4140 | */ | |
41e3355d KH |
4141 | if (!node_state(node, N_NORMAL_MEMORY)) |
4142 | tmp = -1; | |
17295c88 | 4143 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4144 | if (!pn) |
4145 | return 1; | |
1ecaab2b | 4146 | |
1ecaab2b KH |
4147 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4148 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 4149 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
4150 | mz->usage_in_excess = 0; |
4151 | mz->on_tree = false; | |
d79154bb | 4152 | mz->memcg = memcg; |
1ecaab2b | 4153 | } |
54f72fe0 | 4154 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4155 | return 0; |
4156 | } | |
4157 | ||
c0ff4b85 | 4158 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4159 | { |
54f72fe0 | 4160 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
4161 | } |
4162 | ||
33327948 KH |
4163 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4164 | { | |
d79154bb | 4165 | struct mem_cgroup *memcg; |
8ff69e2c | 4166 | size_t size; |
33327948 | 4167 | |
8ff69e2c VD |
4168 | size = sizeof(struct mem_cgroup); |
4169 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
33327948 | 4170 | |
8ff69e2c | 4171 | memcg = kzalloc(size, GFP_KERNEL); |
d79154bb | 4172 | if (!memcg) |
e7bbcdf3 DC |
4173 | return NULL; |
4174 | ||
d79154bb HD |
4175 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
4176 | if (!memcg->stat) | |
d2e61b8d | 4177 | goto out_free; |
841710aa TH |
4178 | |
4179 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) | |
4180 | goto out_free_stat; | |
4181 | ||
d79154bb HD |
4182 | spin_lock_init(&memcg->pcp_counter_lock); |
4183 | return memcg; | |
d2e61b8d | 4184 | |
841710aa TH |
4185 | out_free_stat: |
4186 | free_percpu(memcg->stat); | |
d2e61b8d | 4187 | out_free: |
8ff69e2c | 4188 | kfree(memcg); |
d2e61b8d | 4189 | return NULL; |
33327948 KH |
4190 | } |
4191 | ||
59927fb9 | 4192 | /* |
c8b2a36f GC |
4193 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
4194 | * (scanning all at force_empty is too costly...) | |
4195 | * | |
4196 | * Instead of clearing all references at force_empty, we remember | |
4197 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4198 | * it goes down to 0. | |
4199 | * | |
4200 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 4201 | */ |
c8b2a36f GC |
4202 | |
4203 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 4204 | { |
c8b2a36f | 4205 | int node; |
59927fb9 | 4206 | |
bb4cc1a8 | 4207 | mem_cgroup_remove_from_trees(memcg); |
c8b2a36f GC |
4208 | |
4209 | for_each_node(node) | |
4210 | free_mem_cgroup_per_zone_info(memcg, node); | |
4211 | ||
4212 | free_percpu(memcg->stat); | |
841710aa | 4213 | memcg_wb_domain_exit(memcg); |
8ff69e2c | 4214 | kfree(memcg); |
59927fb9 | 4215 | } |
3afe36b1 | 4216 | |
7bcc1bb1 DN |
4217 | /* |
4218 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4219 | */ | |
e1aab161 | 4220 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 4221 | { |
3e32cb2e | 4222 | if (!memcg->memory.parent) |
7bcc1bb1 | 4223 | return NULL; |
3e32cb2e | 4224 | return mem_cgroup_from_counter(memcg->memory.parent, memory); |
7bcc1bb1 | 4225 | } |
e1aab161 | 4226 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 4227 | |
0eb253e2 | 4228 | static struct cgroup_subsys_state * __ref |
eb95419b | 4229 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 4230 | { |
d142e3e6 | 4231 | struct mem_cgroup *memcg; |
04046e1a | 4232 | long error = -ENOMEM; |
6d12e2d8 | 4233 | int node; |
8cdea7c0 | 4234 | |
c0ff4b85 R |
4235 | memcg = mem_cgroup_alloc(); |
4236 | if (!memcg) | |
04046e1a | 4237 | return ERR_PTR(error); |
78fb7466 | 4238 | |
3ed28fa1 | 4239 | for_each_node(node) |
c0ff4b85 | 4240 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 4241 | goto free_out; |
f64c3f54 | 4242 | |
c077719b | 4243 | /* root ? */ |
eb95419b | 4244 | if (parent_css == NULL) { |
a41c58a6 | 4245 | root_mem_cgroup = memcg; |
56161634 | 4246 | mem_cgroup_root_css = &memcg->css; |
3e32cb2e | 4247 | page_counter_init(&memcg->memory, NULL); |
241994ed | 4248 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4249 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4250 | page_counter_init(&memcg->memsw, NULL); |
4251 | page_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 4252 | } |
28dbc4b6 | 4253 | |
d142e3e6 GC |
4254 | memcg->last_scanned_node = MAX_NUMNODES; |
4255 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4256 | memcg->move_charge_at_immigrate = 0; |
4257 | mutex_init(&memcg->thresholds_lock); | |
4258 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4259 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4260 | INIT_LIST_HEAD(&memcg->event_list); |
4261 | spin_lock_init(&memcg->event_list_lock); | |
900a38f0 VD |
4262 | #ifdef CONFIG_MEMCG_KMEM |
4263 | memcg->kmemcg_id = -1; | |
900a38f0 | 4264 | #endif |
52ebea74 TH |
4265 | #ifdef CONFIG_CGROUP_WRITEBACK |
4266 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
4267 | #endif | |
d142e3e6 GC |
4268 | return &memcg->css; |
4269 | ||
4270 | free_out: | |
4271 | __mem_cgroup_free(memcg); | |
4272 | return ERR_PTR(error); | |
4273 | } | |
4274 | ||
4275 | static int | |
eb95419b | 4276 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 4277 | { |
eb95419b | 4278 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 4279 | struct mem_cgroup *parent = mem_cgroup_from_css(css->parent); |
2f7dd7a4 | 4280 | int ret; |
d142e3e6 | 4281 | |
15a4c835 | 4282 | if (css->id > MEM_CGROUP_ID_MAX) |
4219b2da LZ |
4283 | return -ENOSPC; |
4284 | ||
63876986 | 4285 | if (!parent) |
d142e3e6 GC |
4286 | return 0; |
4287 | ||
0999821b | 4288 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
4289 | |
4290 | memcg->use_hierarchy = parent->use_hierarchy; | |
4291 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4292 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4293 | ||
4294 | if (parent->use_hierarchy) { | |
3e32cb2e | 4295 | page_counter_init(&memcg->memory, &parent->memory); |
241994ed | 4296 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4297 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4298 | page_counter_init(&memcg->memsw, &parent->memsw); |
4299 | page_counter_init(&memcg->kmem, &parent->kmem); | |
55007d84 | 4300 | |
7bcc1bb1 | 4301 | /* |
8d76a979 LZ |
4302 | * No need to take a reference to the parent because cgroup |
4303 | * core guarantees its existence. | |
7bcc1bb1 | 4304 | */ |
18f59ea7 | 4305 | } else { |
3e32cb2e | 4306 | page_counter_init(&memcg->memory, NULL); |
241994ed | 4307 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4308 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4309 | page_counter_init(&memcg->memsw, NULL); |
4310 | page_counter_init(&memcg->kmem, NULL); | |
8c7f6edb TH |
4311 | /* |
4312 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4313 | * much sense so let cgroup subsystem know about this | |
4314 | * unfortunate state in our controller. | |
4315 | */ | |
d142e3e6 | 4316 | if (parent != root_mem_cgroup) |
073219e9 | 4317 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4318 | } |
0999821b | 4319 | mutex_unlock(&memcg_create_mutex); |
d6441637 | 4320 | |
2f7dd7a4 JW |
4321 | ret = memcg_init_kmem(memcg, &memory_cgrp_subsys); |
4322 | if (ret) | |
4323 | return ret; | |
4324 | ||
4325 | /* | |
4326 | * Make sure the memcg is initialized: mem_cgroup_iter() | |
4327 | * orders reading memcg->initialized against its callers | |
4328 | * reading the memcg members. | |
4329 | */ | |
4330 | smp_store_release(&memcg->initialized, 1); | |
4331 | ||
4332 | return 0; | |
8cdea7c0 BS |
4333 | } |
4334 | ||
eb95419b | 4335 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4336 | { |
eb95419b | 4337 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4338 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4339 | |
4340 | /* | |
4341 | * Unregister events and notify userspace. | |
4342 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4343 | * directory to avoid race between userspace and kernelspace. | |
4344 | */ | |
fba94807 TH |
4345 | spin_lock(&memcg->event_list_lock); |
4346 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4347 | list_del_init(&event->list); |
4348 | schedule_work(&event->remove); | |
4349 | } | |
fba94807 | 4350 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4351 | |
33cb876e | 4352 | vmpressure_cleanup(&memcg->vmpressure); |
2a4db7eb VD |
4353 | |
4354 | memcg_deactivate_kmem(memcg); | |
52ebea74 TH |
4355 | |
4356 | wb_memcg_offline(memcg); | |
df878fb0 KH |
4357 | } |
4358 | ||
eb95419b | 4359 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4360 | { |
eb95419b | 4361 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4362 | |
10d5ebf4 | 4363 | memcg_destroy_kmem(memcg); |
465939a1 | 4364 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
4365 | } |
4366 | ||
1ced953b TH |
4367 | /** |
4368 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4369 | * @css: the target css | |
4370 | * | |
4371 | * Reset the states of the mem_cgroup associated with @css. This is | |
4372 | * invoked when the userland requests disabling on the default hierarchy | |
4373 | * but the memcg is pinned through dependency. The memcg should stop | |
4374 | * applying policies and should revert to the vanilla state as it may be | |
4375 | * made visible again. | |
4376 | * | |
4377 | * The current implementation only resets the essential configurations. | |
4378 | * This needs to be expanded to cover all the visible parts. | |
4379 | */ | |
4380 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4381 | { | |
4382 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4383 | ||
3e32cb2e JW |
4384 | mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX); |
4385 | mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX); | |
4386 | memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX); | |
241994ed JW |
4387 | memcg->low = 0; |
4388 | memcg->high = PAGE_COUNTER_MAX; | |
24d404dc | 4389 | memcg->soft_limit = PAGE_COUNTER_MAX; |
2529bb3a | 4390 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
4391 | } |
4392 | ||
02491447 | 4393 | #ifdef CONFIG_MMU |
7dc74be0 | 4394 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4395 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4396 | { |
05b84301 | 4397 | int ret; |
9476db97 JW |
4398 | |
4399 | /* Try a single bulk charge without reclaim first */ | |
00501b53 | 4400 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_WAIT, count); |
9476db97 | 4401 | if (!ret) { |
854ffa8d | 4402 | mc.precharge += count; |
854ffa8d DN |
4403 | return ret; |
4404 | } | |
692e7c45 | 4405 | if (ret == -EINTR) { |
00501b53 | 4406 | cancel_charge(root_mem_cgroup, count); |
692e7c45 JW |
4407 | return ret; |
4408 | } | |
9476db97 JW |
4409 | |
4410 | /* Try charges one by one with reclaim */ | |
854ffa8d | 4411 | while (count--) { |
00501b53 | 4412 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1); |
9476db97 JW |
4413 | /* |
4414 | * In case of failure, any residual charges against | |
4415 | * mc.to will be dropped by mem_cgroup_clear_mc() | |
692e7c45 JW |
4416 | * later on. However, cancel any charges that are |
4417 | * bypassed to root right away or they'll be lost. | |
9476db97 | 4418 | */ |
692e7c45 | 4419 | if (ret == -EINTR) |
00501b53 | 4420 | cancel_charge(root_mem_cgroup, 1); |
38c5d72f | 4421 | if (ret) |
38c5d72f | 4422 | return ret; |
854ffa8d | 4423 | mc.precharge++; |
9476db97 | 4424 | cond_resched(); |
854ffa8d | 4425 | } |
9476db97 | 4426 | return 0; |
4ffef5fe DN |
4427 | } |
4428 | ||
4429 | /** | |
8d32ff84 | 4430 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
4431 | * @vma: the vma the pte to be checked belongs |
4432 | * @addr: the address corresponding to the pte to be checked | |
4433 | * @ptent: the pte to be checked | |
02491447 | 4434 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
4435 | * |
4436 | * Returns | |
4437 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4438 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4439 | * move charge. if @target is not NULL, the page is stored in target->page | |
4440 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
4441 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
4442 | * target for charge migration. if @target is not NULL, the entry is stored | |
4443 | * in target->ent. | |
4ffef5fe DN |
4444 | * |
4445 | * Called with pte lock held. | |
4446 | */ | |
4ffef5fe DN |
4447 | union mc_target { |
4448 | struct page *page; | |
02491447 | 4449 | swp_entry_t ent; |
4ffef5fe DN |
4450 | }; |
4451 | ||
4ffef5fe | 4452 | enum mc_target_type { |
8d32ff84 | 4453 | MC_TARGET_NONE = 0, |
4ffef5fe | 4454 | MC_TARGET_PAGE, |
02491447 | 4455 | MC_TARGET_SWAP, |
4ffef5fe DN |
4456 | }; |
4457 | ||
90254a65 DN |
4458 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4459 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4460 | { |
90254a65 | 4461 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 4462 | |
90254a65 DN |
4463 | if (!page || !page_mapped(page)) |
4464 | return NULL; | |
4465 | if (PageAnon(page)) { | |
1dfab5ab | 4466 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 4467 | return NULL; |
1dfab5ab JW |
4468 | } else { |
4469 | if (!(mc.flags & MOVE_FILE)) | |
4470 | return NULL; | |
4471 | } | |
90254a65 DN |
4472 | if (!get_page_unless_zero(page)) |
4473 | return NULL; | |
4474 | ||
4475 | return page; | |
4476 | } | |
4477 | ||
4b91355e | 4478 | #ifdef CONFIG_SWAP |
90254a65 DN |
4479 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
4480 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4481 | { | |
90254a65 DN |
4482 | struct page *page = NULL; |
4483 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4484 | ||
1dfab5ab | 4485 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 4486 | return NULL; |
4b91355e KH |
4487 | /* |
4488 | * Because lookup_swap_cache() updates some statistics counter, | |
4489 | * we call find_get_page() with swapper_space directly. | |
4490 | */ | |
33806f06 | 4491 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
4492 | if (do_swap_account) |
4493 | entry->val = ent.val; | |
4494 | ||
4495 | return page; | |
4496 | } | |
4b91355e KH |
4497 | #else |
4498 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
4499 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4500 | { | |
4501 | return NULL; | |
4502 | } | |
4503 | #endif | |
90254a65 | 4504 | |
87946a72 DN |
4505 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4506 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4507 | { | |
4508 | struct page *page = NULL; | |
87946a72 DN |
4509 | struct address_space *mapping; |
4510 | pgoff_t pgoff; | |
4511 | ||
4512 | if (!vma->vm_file) /* anonymous vma */ | |
4513 | return NULL; | |
1dfab5ab | 4514 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
4515 | return NULL; |
4516 | ||
87946a72 | 4517 | mapping = vma->vm_file->f_mapping; |
0661a336 | 4518 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
4519 | |
4520 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
4521 | #ifdef CONFIG_SWAP |
4522 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
4523 | if (shmem_mapping(mapping)) { |
4524 | page = find_get_entry(mapping, pgoff); | |
4525 | if (radix_tree_exceptional_entry(page)) { | |
4526 | swp_entry_t swp = radix_to_swp_entry(page); | |
4527 | if (do_swap_account) | |
4528 | *entry = swp; | |
4529 | page = find_get_page(swap_address_space(swp), swp.val); | |
4530 | } | |
4531 | } else | |
4532 | page = find_get_page(mapping, pgoff); | |
4533 | #else | |
4534 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 4535 | #endif |
87946a72 DN |
4536 | return page; |
4537 | } | |
4538 | ||
b1b0deab CG |
4539 | /** |
4540 | * mem_cgroup_move_account - move account of the page | |
4541 | * @page: the page | |
4542 | * @nr_pages: number of regular pages (>1 for huge pages) | |
4543 | * @from: mem_cgroup which the page is moved from. | |
4544 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
4545 | * | |
4546 | * The caller must confirm following. | |
4547 | * - page is not on LRU (isolate_page() is useful.) | |
4548 | * - compound_lock is held when nr_pages > 1 | |
4549 | * | |
4550 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
4551 | * from old cgroup. | |
4552 | */ | |
4553 | static int mem_cgroup_move_account(struct page *page, | |
4554 | unsigned int nr_pages, | |
4555 | struct mem_cgroup *from, | |
4556 | struct mem_cgroup *to) | |
4557 | { | |
4558 | unsigned long flags; | |
4559 | int ret; | |
c4843a75 | 4560 | bool anon; |
b1b0deab CG |
4561 | |
4562 | VM_BUG_ON(from == to); | |
4563 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
4564 | /* | |
4565 | * The page is isolated from LRU. So, collapse function | |
4566 | * will not handle this page. But page splitting can happen. | |
4567 | * Do this check under compound_page_lock(). The caller should | |
4568 | * hold it. | |
4569 | */ | |
4570 | ret = -EBUSY; | |
4571 | if (nr_pages > 1 && !PageTransHuge(page)) | |
4572 | goto out; | |
4573 | ||
4574 | /* | |
4575 | * Prevent mem_cgroup_migrate() from looking at page->mem_cgroup | |
4576 | * of its source page while we change it: page migration takes | |
4577 | * both pages off the LRU, but page cache replacement doesn't. | |
4578 | */ | |
4579 | if (!trylock_page(page)) | |
4580 | goto out; | |
4581 | ||
4582 | ret = -EINVAL; | |
4583 | if (page->mem_cgroup != from) | |
4584 | goto out_unlock; | |
4585 | ||
c4843a75 GT |
4586 | anon = PageAnon(page); |
4587 | ||
b1b0deab CG |
4588 | spin_lock_irqsave(&from->move_lock, flags); |
4589 | ||
c4843a75 | 4590 | if (!anon && page_mapped(page)) { |
b1b0deab CG |
4591 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
4592 | nr_pages); | |
4593 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
4594 | nr_pages); | |
4595 | } | |
4596 | ||
c4843a75 GT |
4597 | /* |
4598 | * move_lock grabbed above and caller set from->moving_account, so | |
4599 | * mem_cgroup_update_page_stat() will serialize updates to PageDirty. | |
4600 | * So mapping should be stable for dirty pages. | |
4601 | */ | |
4602 | if (!anon && PageDirty(page)) { | |
4603 | struct address_space *mapping = page_mapping(page); | |
4604 | ||
4605 | if (mapping_cap_account_dirty(mapping)) { | |
4606 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY], | |
4607 | nr_pages); | |
4608 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY], | |
4609 | nr_pages); | |
4610 | } | |
4611 | } | |
4612 | ||
b1b0deab CG |
4613 | if (PageWriteback(page)) { |
4614 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
4615 | nr_pages); | |
4616 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
4617 | nr_pages); | |
4618 | } | |
4619 | ||
4620 | /* | |
4621 | * It is safe to change page->mem_cgroup here because the page | |
4622 | * is referenced, charged, and isolated - we can't race with | |
4623 | * uncharging, charging, migration, or LRU putback. | |
4624 | */ | |
4625 | ||
4626 | /* caller should have done css_get */ | |
4627 | page->mem_cgroup = to; | |
4628 | spin_unlock_irqrestore(&from->move_lock, flags); | |
4629 | ||
4630 | ret = 0; | |
4631 | ||
4632 | local_irq_disable(); | |
4633 | mem_cgroup_charge_statistics(to, page, nr_pages); | |
4634 | memcg_check_events(to, page); | |
4635 | mem_cgroup_charge_statistics(from, page, -nr_pages); | |
4636 | memcg_check_events(from, page); | |
4637 | local_irq_enable(); | |
4638 | out_unlock: | |
4639 | unlock_page(page); | |
4640 | out: | |
4641 | return ret; | |
4642 | } | |
4643 | ||
8d32ff84 | 4644 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
4645 | unsigned long addr, pte_t ptent, union mc_target *target) |
4646 | { | |
4647 | struct page *page = NULL; | |
8d32ff84 | 4648 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
4649 | swp_entry_t ent = { .val = 0 }; |
4650 | ||
4651 | if (pte_present(ptent)) | |
4652 | page = mc_handle_present_pte(vma, addr, ptent); | |
4653 | else if (is_swap_pte(ptent)) | |
4654 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
0661a336 | 4655 | else if (pte_none(ptent)) |
87946a72 | 4656 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
4657 | |
4658 | if (!page && !ent.val) | |
8d32ff84 | 4659 | return ret; |
02491447 | 4660 | if (page) { |
02491447 | 4661 | /* |
0a31bc97 | 4662 | * Do only loose check w/o serialization. |
1306a85a | 4663 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 4664 | * not under LRU exclusion. |
02491447 | 4665 | */ |
1306a85a | 4666 | if (page->mem_cgroup == mc.from) { |
02491447 DN |
4667 | ret = MC_TARGET_PAGE; |
4668 | if (target) | |
4669 | target->page = page; | |
4670 | } | |
4671 | if (!ret || !target) | |
4672 | put_page(page); | |
4673 | } | |
90254a65 DN |
4674 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
4675 | if (ent.val && !ret && | |
34c00c31 | 4676 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
4677 | ret = MC_TARGET_SWAP; |
4678 | if (target) | |
4679 | target->ent = ent; | |
4ffef5fe | 4680 | } |
4ffef5fe DN |
4681 | return ret; |
4682 | } | |
4683 | ||
12724850 NH |
4684 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4685 | /* | |
4686 | * We don't consider swapping or file mapped pages because THP does not | |
4687 | * support them for now. | |
4688 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
4689 | */ | |
4690 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4691 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4692 | { | |
4693 | struct page *page = NULL; | |
12724850 NH |
4694 | enum mc_target_type ret = MC_TARGET_NONE; |
4695 | ||
4696 | page = pmd_page(pmd); | |
309381fe | 4697 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 4698 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 4699 | return ret; |
1306a85a | 4700 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
4701 | ret = MC_TARGET_PAGE; |
4702 | if (target) { | |
4703 | get_page(page); | |
4704 | target->page = page; | |
4705 | } | |
4706 | } | |
4707 | return ret; | |
4708 | } | |
4709 | #else | |
4710 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4711 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4712 | { | |
4713 | return MC_TARGET_NONE; | |
4714 | } | |
4715 | #endif | |
4716 | ||
4ffef5fe DN |
4717 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
4718 | unsigned long addr, unsigned long end, | |
4719 | struct mm_walk *walk) | |
4720 | { | |
26bcd64a | 4721 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4722 | pte_t *pte; |
4723 | spinlock_t *ptl; | |
4724 | ||
bf929152 | 4725 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
12724850 NH |
4726 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
4727 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 4728 | spin_unlock(ptl); |
1a5a9906 | 4729 | return 0; |
12724850 | 4730 | } |
03319327 | 4731 | |
45f83cef AA |
4732 | if (pmd_trans_unstable(pmd)) |
4733 | return 0; | |
4ffef5fe DN |
4734 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
4735 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 4736 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
4737 | mc.precharge++; /* increment precharge temporarily */ |
4738 | pte_unmap_unlock(pte - 1, ptl); | |
4739 | cond_resched(); | |
4740 | ||
7dc74be0 DN |
4741 | return 0; |
4742 | } | |
4743 | ||
4ffef5fe DN |
4744 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
4745 | { | |
4746 | unsigned long precharge; | |
4ffef5fe | 4747 | |
26bcd64a NH |
4748 | struct mm_walk mem_cgroup_count_precharge_walk = { |
4749 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
4750 | .mm = mm, | |
4751 | }; | |
dfe076b0 | 4752 | down_read(&mm->mmap_sem); |
26bcd64a | 4753 | walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk); |
dfe076b0 | 4754 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
4755 | |
4756 | precharge = mc.precharge; | |
4757 | mc.precharge = 0; | |
4758 | ||
4759 | return precharge; | |
4760 | } | |
4761 | ||
4ffef5fe DN |
4762 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
4763 | { | |
dfe076b0 DN |
4764 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
4765 | ||
4766 | VM_BUG_ON(mc.moving_task); | |
4767 | mc.moving_task = current; | |
4768 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
4769 | } |
4770 | ||
dfe076b0 DN |
4771 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
4772 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 4773 | { |
2bd9bb20 KH |
4774 | struct mem_cgroup *from = mc.from; |
4775 | struct mem_cgroup *to = mc.to; | |
4776 | ||
4ffef5fe | 4777 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 4778 | if (mc.precharge) { |
00501b53 | 4779 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
4780 | mc.precharge = 0; |
4781 | } | |
4782 | /* | |
4783 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
4784 | * we must uncharge here. | |
4785 | */ | |
4786 | if (mc.moved_charge) { | |
00501b53 | 4787 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 4788 | mc.moved_charge = 0; |
4ffef5fe | 4789 | } |
483c30b5 DN |
4790 | /* we must fixup refcnts and charges */ |
4791 | if (mc.moved_swap) { | |
483c30b5 | 4792 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 4793 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 4794 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 4795 | |
05b84301 | 4796 | /* |
3e32cb2e JW |
4797 | * we charged both to->memory and to->memsw, so we |
4798 | * should uncharge to->memory. | |
05b84301 | 4799 | */ |
ce00a967 | 4800 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
4801 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
4802 | ||
e8ea14cc | 4803 | css_put_many(&mc.from->css, mc.moved_swap); |
3e32cb2e | 4804 | |
4050377b | 4805 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
4806 | mc.moved_swap = 0; |
4807 | } | |
dfe076b0 DN |
4808 | memcg_oom_recover(from); |
4809 | memcg_oom_recover(to); | |
4810 | wake_up_all(&mc.waitq); | |
4811 | } | |
4812 | ||
4813 | static void mem_cgroup_clear_mc(void) | |
4814 | { | |
dfe076b0 DN |
4815 | /* |
4816 | * we must clear moving_task before waking up waiters at the end of | |
4817 | * task migration. | |
4818 | */ | |
4819 | mc.moving_task = NULL; | |
4820 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 4821 | spin_lock(&mc.lock); |
4ffef5fe DN |
4822 | mc.from = NULL; |
4823 | mc.to = NULL; | |
2bd9bb20 | 4824 | spin_unlock(&mc.lock); |
4ffef5fe DN |
4825 | } |
4826 | ||
eb95419b | 4827 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 4828 | struct cgroup_taskset *tset) |
7dc74be0 | 4829 | { |
eb95419b | 4830 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
9f2115f9 TH |
4831 | struct mem_cgroup *from; |
4832 | struct task_struct *p; | |
4833 | struct mm_struct *mm; | |
1dfab5ab | 4834 | unsigned long move_flags; |
9f2115f9 | 4835 | int ret = 0; |
7dc74be0 | 4836 | |
ee5e8472 GC |
4837 | /* |
4838 | * We are now commited to this value whatever it is. Changes in this | |
4839 | * tunable will only affect upcoming migrations, not the current one. | |
4840 | * So we need to save it, and keep it going. | |
4841 | */ | |
4db0c3c2 | 4842 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); |
9f2115f9 TH |
4843 | if (!move_flags) |
4844 | return 0; | |
4845 | ||
4846 | p = cgroup_taskset_first(tset); | |
4847 | from = mem_cgroup_from_task(p); | |
4848 | ||
4849 | VM_BUG_ON(from == memcg); | |
4850 | ||
4851 | mm = get_task_mm(p); | |
4852 | if (!mm) | |
4853 | return 0; | |
4854 | /* We move charges only when we move a owner of the mm */ | |
4855 | if (mm->owner == p) { | |
4856 | VM_BUG_ON(mc.from); | |
4857 | VM_BUG_ON(mc.to); | |
4858 | VM_BUG_ON(mc.precharge); | |
4859 | VM_BUG_ON(mc.moved_charge); | |
4860 | VM_BUG_ON(mc.moved_swap); | |
4861 | ||
4862 | spin_lock(&mc.lock); | |
4863 | mc.from = from; | |
4864 | mc.to = memcg; | |
4865 | mc.flags = move_flags; | |
4866 | spin_unlock(&mc.lock); | |
4867 | /* We set mc.moving_task later */ | |
4868 | ||
4869 | ret = mem_cgroup_precharge_mc(mm); | |
4870 | if (ret) | |
4871 | mem_cgroup_clear_mc(); | |
7dc74be0 | 4872 | } |
9f2115f9 | 4873 | mmput(mm); |
7dc74be0 DN |
4874 | return ret; |
4875 | } | |
4876 | ||
eb95419b | 4877 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 4878 | struct cgroup_taskset *tset) |
7dc74be0 | 4879 | { |
4e2f245d JW |
4880 | if (mc.to) |
4881 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
4882 | } |
4883 | ||
4ffef5fe DN |
4884 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
4885 | unsigned long addr, unsigned long end, | |
4886 | struct mm_walk *walk) | |
7dc74be0 | 4887 | { |
4ffef5fe | 4888 | int ret = 0; |
26bcd64a | 4889 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4890 | pte_t *pte; |
4891 | spinlock_t *ptl; | |
12724850 NH |
4892 | enum mc_target_type target_type; |
4893 | union mc_target target; | |
4894 | struct page *page; | |
4ffef5fe | 4895 | |
12724850 NH |
4896 | /* |
4897 | * We don't take compound_lock() here but no race with splitting thp | |
4898 | * happens because: | |
4899 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
4900 | * under splitting, which means there's no concurrent thp split, | |
4901 | * - if another thread runs into split_huge_page() just after we | |
4902 | * entered this if-block, the thread must wait for page table lock | |
4903 | * to be unlocked in __split_huge_page_splitting(), where the main | |
4904 | * part of thp split is not executed yet. | |
4905 | */ | |
bf929152 | 4906 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
62ade86a | 4907 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 4908 | spin_unlock(ptl); |
12724850 NH |
4909 | return 0; |
4910 | } | |
4911 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
4912 | if (target_type == MC_TARGET_PAGE) { | |
4913 | page = target.page; | |
4914 | if (!isolate_lru_page(page)) { | |
12724850 | 4915 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, |
1306a85a | 4916 | mc.from, mc.to)) { |
12724850 NH |
4917 | mc.precharge -= HPAGE_PMD_NR; |
4918 | mc.moved_charge += HPAGE_PMD_NR; | |
4919 | } | |
4920 | putback_lru_page(page); | |
4921 | } | |
4922 | put_page(page); | |
4923 | } | |
bf929152 | 4924 | spin_unlock(ptl); |
1a5a9906 | 4925 | return 0; |
12724850 NH |
4926 | } |
4927 | ||
45f83cef AA |
4928 | if (pmd_trans_unstable(pmd)) |
4929 | return 0; | |
4ffef5fe DN |
4930 | retry: |
4931 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
4932 | for (; addr != end; addr += PAGE_SIZE) { | |
4933 | pte_t ptent = *(pte++); | |
02491447 | 4934 | swp_entry_t ent; |
4ffef5fe DN |
4935 | |
4936 | if (!mc.precharge) | |
4937 | break; | |
4938 | ||
8d32ff84 | 4939 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
4940 | case MC_TARGET_PAGE: |
4941 | page = target.page; | |
4942 | if (isolate_lru_page(page)) | |
4943 | goto put; | |
1306a85a | 4944 | if (!mem_cgroup_move_account(page, 1, mc.from, mc.to)) { |
4ffef5fe | 4945 | mc.precharge--; |
854ffa8d DN |
4946 | /* we uncharge from mc.from later. */ |
4947 | mc.moved_charge++; | |
4ffef5fe DN |
4948 | } |
4949 | putback_lru_page(page); | |
8d32ff84 | 4950 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
4951 | put_page(page); |
4952 | break; | |
02491447 DN |
4953 | case MC_TARGET_SWAP: |
4954 | ent = target.ent; | |
e91cbb42 | 4955 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 4956 | mc.precharge--; |
483c30b5 DN |
4957 | /* we fixup refcnts and charges later. */ |
4958 | mc.moved_swap++; | |
4959 | } | |
02491447 | 4960 | break; |
4ffef5fe DN |
4961 | default: |
4962 | break; | |
4963 | } | |
4964 | } | |
4965 | pte_unmap_unlock(pte - 1, ptl); | |
4966 | cond_resched(); | |
4967 | ||
4968 | if (addr != end) { | |
4969 | /* | |
4970 | * We have consumed all precharges we got in can_attach(). | |
4971 | * We try charge one by one, but don't do any additional | |
4972 | * charges to mc.to if we have failed in charge once in attach() | |
4973 | * phase. | |
4974 | */ | |
854ffa8d | 4975 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
4976 | if (!ret) |
4977 | goto retry; | |
4978 | } | |
4979 | ||
4980 | return ret; | |
4981 | } | |
4982 | ||
4983 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
4984 | { | |
26bcd64a NH |
4985 | struct mm_walk mem_cgroup_move_charge_walk = { |
4986 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
4987 | .mm = mm, | |
4988 | }; | |
4ffef5fe DN |
4989 | |
4990 | lru_add_drain_all(); | |
312722cb JW |
4991 | /* |
4992 | * Signal mem_cgroup_begin_page_stat() to take the memcg's | |
4993 | * move_lock while we're moving its pages to another memcg. | |
4994 | * Then wait for already started RCU-only updates to finish. | |
4995 | */ | |
4996 | atomic_inc(&mc.from->moving_account); | |
4997 | synchronize_rcu(); | |
dfe076b0 DN |
4998 | retry: |
4999 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5000 | /* | |
5001 | * Someone who are holding the mmap_sem might be waiting in | |
5002 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5003 | * and retry. Because we cancel precharges, we might not be able | |
5004 | * to move enough charges, but moving charge is a best-effort | |
5005 | * feature anyway, so it wouldn't be a big problem. | |
5006 | */ | |
5007 | __mem_cgroup_clear_mc(); | |
5008 | cond_resched(); | |
5009 | goto retry; | |
5010 | } | |
26bcd64a NH |
5011 | /* |
5012 | * When we have consumed all precharges and failed in doing | |
5013 | * additional charge, the page walk just aborts. | |
5014 | */ | |
5015 | walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk); | |
dfe076b0 | 5016 | up_read(&mm->mmap_sem); |
312722cb | 5017 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
5018 | } |
5019 | ||
eb95419b | 5020 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 5021 | struct cgroup_taskset *tset) |
67e465a7 | 5022 | { |
2f7ee569 | 5023 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 5024 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 5025 | |
dfe076b0 | 5026 | if (mm) { |
a433658c KM |
5027 | if (mc.to) |
5028 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
5029 | mmput(mm); |
5030 | } | |
a433658c KM |
5031 | if (mc.to) |
5032 | mem_cgroup_clear_mc(); | |
67e465a7 | 5033 | } |
5cfb80a7 | 5034 | #else /* !CONFIG_MMU */ |
eb95419b | 5035 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5036 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5037 | { |
5038 | return 0; | |
5039 | } | |
eb95419b | 5040 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5041 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5042 | { |
5043 | } | |
eb95419b | 5044 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 5045 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5046 | { |
5047 | } | |
5048 | #endif | |
67e465a7 | 5049 | |
f00baae7 TH |
5050 | /* |
5051 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5052 | * to verify whether we're attached to the default hierarchy on each mount |
5053 | * attempt. | |
f00baae7 | 5054 | */ |
eb95419b | 5055 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5056 | { |
5057 | /* | |
aa6ec29b | 5058 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5059 | * guarantees that @root doesn't have any children, so turning it |
5060 | * on for the root memcg is enough. | |
5061 | */ | |
aa6ec29b | 5062 | if (cgroup_on_dfl(root_css->cgroup)) |
7feee590 VD |
5063 | root_mem_cgroup->use_hierarchy = true; |
5064 | else | |
5065 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
5066 | } |
5067 | ||
241994ed JW |
5068 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
5069 | struct cftype *cft) | |
5070 | { | |
5071 | return mem_cgroup_usage(mem_cgroup_from_css(css), false); | |
5072 | } | |
5073 | ||
5074 | static int memory_low_show(struct seq_file *m, void *v) | |
5075 | { | |
5076 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5077 | unsigned long low = READ_ONCE(memcg->low); |
241994ed JW |
5078 | |
5079 | if (low == PAGE_COUNTER_MAX) | |
d2973697 | 5080 | seq_puts(m, "max\n"); |
241994ed JW |
5081 | else |
5082 | seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE); | |
5083 | ||
5084 | return 0; | |
5085 | } | |
5086 | ||
5087 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
5088 | char *buf, size_t nbytes, loff_t off) | |
5089 | { | |
5090 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5091 | unsigned long low; | |
5092 | int err; | |
5093 | ||
5094 | buf = strstrip(buf); | |
d2973697 | 5095 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
5096 | if (err) |
5097 | return err; | |
5098 | ||
5099 | memcg->low = low; | |
5100 | ||
5101 | return nbytes; | |
5102 | } | |
5103 | ||
5104 | static int memory_high_show(struct seq_file *m, void *v) | |
5105 | { | |
5106 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5107 | unsigned long high = READ_ONCE(memcg->high); |
241994ed JW |
5108 | |
5109 | if (high == PAGE_COUNTER_MAX) | |
d2973697 | 5110 | seq_puts(m, "max\n"); |
241994ed JW |
5111 | else |
5112 | seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE); | |
5113 | ||
5114 | return 0; | |
5115 | } | |
5116 | ||
5117 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
5118 | char *buf, size_t nbytes, loff_t off) | |
5119 | { | |
5120 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5121 | unsigned long high; | |
5122 | int err; | |
5123 | ||
5124 | buf = strstrip(buf); | |
d2973697 | 5125 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
5126 | if (err) |
5127 | return err; | |
5128 | ||
5129 | memcg->high = high; | |
5130 | ||
2529bb3a | 5131 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5132 | return nbytes; |
5133 | } | |
5134 | ||
5135 | static int memory_max_show(struct seq_file *m, void *v) | |
5136 | { | |
5137 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5138 | unsigned long max = READ_ONCE(memcg->memory.limit); |
241994ed JW |
5139 | |
5140 | if (max == PAGE_COUNTER_MAX) | |
d2973697 | 5141 | seq_puts(m, "max\n"); |
241994ed JW |
5142 | else |
5143 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
5144 | ||
5145 | return 0; | |
5146 | } | |
5147 | ||
5148 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
5149 | char *buf, size_t nbytes, loff_t off) | |
5150 | { | |
5151 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5152 | unsigned long max; | |
5153 | int err; | |
5154 | ||
5155 | buf = strstrip(buf); | |
d2973697 | 5156 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
5157 | if (err) |
5158 | return err; | |
5159 | ||
5160 | err = mem_cgroup_resize_limit(memcg, max); | |
5161 | if (err) | |
5162 | return err; | |
5163 | ||
2529bb3a | 5164 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5165 | return nbytes; |
5166 | } | |
5167 | ||
5168 | static int memory_events_show(struct seq_file *m, void *v) | |
5169 | { | |
5170 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5171 | ||
5172 | seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW)); | |
5173 | seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH)); | |
5174 | seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX)); | |
5175 | seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM)); | |
5176 | ||
5177 | return 0; | |
5178 | } | |
5179 | ||
5180 | static struct cftype memory_files[] = { | |
5181 | { | |
5182 | .name = "current", | |
5183 | .read_u64 = memory_current_read, | |
5184 | }, | |
5185 | { | |
5186 | .name = "low", | |
5187 | .flags = CFTYPE_NOT_ON_ROOT, | |
5188 | .seq_show = memory_low_show, | |
5189 | .write = memory_low_write, | |
5190 | }, | |
5191 | { | |
5192 | .name = "high", | |
5193 | .flags = CFTYPE_NOT_ON_ROOT, | |
5194 | .seq_show = memory_high_show, | |
5195 | .write = memory_high_write, | |
5196 | }, | |
5197 | { | |
5198 | .name = "max", | |
5199 | .flags = CFTYPE_NOT_ON_ROOT, | |
5200 | .seq_show = memory_max_show, | |
5201 | .write = memory_max_write, | |
5202 | }, | |
5203 | { | |
5204 | .name = "events", | |
5205 | .flags = CFTYPE_NOT_ON_ROOT, | |
5206 | .seq_show = memory_events_show, | |
5207 | }, | |
5208 | { } /* terminate */ | |
5209 | }; | |
5210 | ||
073219e9 | 5211 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5212 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5213 | .css_online = mem_cgroup_css_online, |
92fb9748 TH |
5214 | .css_offline = mem_cgroup_css_offline, |
5215 | .css_free = mem_cgroup_css_free, | |
1ced953b | 5216 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5217 | .can_attach = mem_cgroup_can_attach, |
5218 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5219 | .attach = mem_cgroup_move_task, |
f00baae7 | 5220 | .bind = mem_cgroup_bind, |
241994ed JW |
5221 | .dfl_cftypes = memory_files, |
5222 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 5223 | .early_init = 0, |
8cdea7c0 | 5224 | }; |
c077719b | 5225 | |
241994ed JW |
5226 | /** |
5227 | * mem_cgroup_low - check if memory consumption is below the normal range | |
5228 | * @root: the highest ancestor to consider | |
5229 | * @memcg: the memory cgroup to check | |
5230 | * | |
5231 | * Returns %true if memory consumption of @memcg, and that of all | |
5232 | * configurable ancestors up to @root, is below the normal range. | |
5233 | */ | |
5234 | bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg) | |
5235 | { | |
5236 | if (mem_cgroup_disabled()) | |
5237 | return false; | |
5238 | ||
5239 | /* | |
5240 | * The toplevel group doesn't have a configurable range, so | |
5241 | * it's never low when looked at directly, and it is not | |
5242 | * considered an ancestor when assessing the hierarchy. | |
5243 | */ | |
5244 | ||
5245 | if (memcg == root_mem_cgroup) | |
5246 | return false; | |
5247 | ||
4e54dede | 5248 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5249 | return false; |
5250 | ||
5251 | while (memcg != root) { | |
5252 | memcg = parent_mem_cgroup(memcg); | |
5253 | ||
5254 | if (memcg == root_mem_cgroup) | |
5255 | break; | |
5256 | ||
4e54dede | 5257 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5258 | return false; |
5259 | } | |
5260 | return true; | |
5261 | } | |
5262 | ||
00501b53 JW |
5263 | /** |
5264 | * mem_cgroup_try_charge - try charging a page | |
5265 | * @page: page to charge | |
5266 | * @mm: mm context of the victim | |
5267 | * @gfp_mask: reclaim mode | |
5268 | * @memcgp: charged memcg return | |
5269 | * | |
5270 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5271 | * pages according to @gfp_mask if necessary. | |
5272 | * | |
5273 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5274 | * Otherwise, an error code is returned. | |
5275 | * | |
5276 | * After page->mapping has been set up, the caller must finalize the | |
5277 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5278 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5279 | */ | |
5280 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
5281 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
5282 | { | |
5283 | struct mem_cgroup *memcg = NULL; | |
5284 | unsigned int nr_pages = 1; | |
5285 | int ret = 0; | |
5286 | ||
5287 | if (mem_cgroup_disabled()) | |
5288 | goto out; | |
5289 | ||
5290 | if (PageSwapCache(page)) { | |
00501b53 JW |
5291 | /* |
5292 | * Every swap fault against a single page tries to charge the | |
5293 | * page, bail as early as possible. shmem_unuse() encounters | |
5294 | * already charged pages, too. The USED bit is protected by | |
5295 | * the page lock, which serializes swap cache removal, which | |
5296 | * in turn serializes uncharging. | |
5297 | */ | |
e993d905 | 5298 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1306a85a | 5299 | if (page->mem_cgroup) |
00501b53 | 5300 | goto out; |
e993d905 VD |
5301 | |
5302 | if (do_swap_account) { | |
5303 | swp_entry_t ent = { .val = page_private(page), }; | |
5304 | unsigned short id = lookup_swap_cgroup_id(ent); | |
5305 | ||
5306 | rcu_read_lock(); | |
5307 | memcg = mem_cgroup_from_id(id); | |
5308 | if (memcg && !css_tryget_online(&memcg->css)) | |
5309 | memcg = NULL; | |
5310 | rcu_read_unlock(); | |
5311 | } | |
00501b53 JW |
5312 | } |
5313 | ||
5314 | if (PageTransHuge(page)) { | |
5315 | nr_pages <<= compound_order(page); | |
5316 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5317 | } | |
5318 | ||
00501b53 JW |
5319 | if (!memcg) |
5320 | memcg = get_mem_cgroup_from_mm(mm); | |
5321 | ||
5322 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
5323 | ||
5324 | css_put(&memcg->css); | |
5325 | ||
5326 | if (ret == -EINTR) { | |
5327 | memcg = root_mem_cgroup; | |
5328 | ret = 0; | |
5329 | } | |
5330 | out: | |
5331 | *memcgp = memcg; | |
5332 | return ret; | |
5333 | } | |
5334 | ||
5335 | /** | |
5336 | * mem_cgroup_commit_charge - commit a page charge | |
5337 | * @page: page to charge | |
5338 | * @memcg: memcg to charge the page to | |
5339 | * @lrucare: page might be on LRU already | |
5340 | * | |
5341 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
5342 | * after page->mapping has been set up. This must happen atomically | |
5343 | * as part of the page instantiation, i.e. under the page table lock | |
5344 | * for anonymous pages, under the page lock for page and swap cache. | |
5345 | * | |
5346 | * In addition, the page must not be on the LRU during the commit, to | |
5347 | * prevent racing with task migration. If it might be, use @lrucare. | |
5348 | * | |
5349 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
5350 | */ | |
5351 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
5352 | bool lrucare) | |
5353 | { | |
5354 | unsigned int nr_pages = 1; | |
5355 | ||
5356 | VM_BUG_ON_PAGE(!page->mapping, page); | |
5357 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
5358 | ||
5359 | if (mem_cgroup_disabled()) | |
5360 | return; | |
5361 | /* | |
5362 | * Swap faults will attempt to charge the same page multiple | |
5363 | * times. But reuse_swap_page() might have removed the page | |
5364 | * from swapcache already, so we can't check PageSwapCache(). | |
5365 | */ | |
5366 | if (!memcg) | |
5367 | return; | |
5368 | ||
6abb5a86 JW |
5369 | commit_charge(page, memcg, lrucare); |
5370 | ||
00501b53 JW |
5371 | if (PageTransHuge(page)) { |
5372 | nr_pages <<= compound_order(page); | |
5373 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5374 | } | |
5375 | ||
6abb5a86 JW |
5376 | local_irq_disable(); |
5377 | mem_cgroup_charge_statistics(memcg, page, nr_pages); | |
5378 | memcg_check_events(memcg, page); | |
5379 | local_irq_enable(); | |
00501b53 JW |
5380 | |
5381 | if (do_swap_account && PageSwapCache(page)) { | |
5382 | swp_entry_t entry = { .val = page_private(page) }; | |
5383 | /* | |
5384 | * The swap entry might not get freed for a long time, | |
5385 | * let's not wait for it. The page already received a | |
5386 | * memory+swap charge, drop the swap entry duplicate. | |
5387 | */ | |
5388 | mem_cgroup_uncharge_swap(entry); | |
5389 | } | |
5390 | } | |
5391 | ||
5392 | /** | |
5393 | * mem_cgroup_cancel_charge - cancel a page charge | |
5394 | * @page: page to charge | |
5395 | * @memcg: memcg to charge the page to | |
5396 | * | |
5397 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
5398 | */ | |
5399 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg) | |
5400 | { | |
5401 | unsigned int nr_pages = 1; | |
5402 | ||
5403 | if (mem_cgroup_disabled()) | |
5404 | return; | |
5405 | /* | |
5406 | * Swap faults will attempt to charge the same page multiple | |
5407 | * times. But reuse_swap_page() might have removed the page | |
5408 | * from swapcache already, so we can't check PageSwapCache(). | |
5409 | */ | |
5410 | if (!memcg) | |
5411 | return; | |
5412 | ||
5413 | if (PageTransHuge(page)) { | |
5414 | nr_pages <<= compound_order(page); | |
5415 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5416 | } | |
5417 | ||
5418 | cancel_charge(memcg, nr_pages); | |
5419 | } | |
5420 | ||
747db954 | 5421 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
747db954 JW |
5422 | unsigned long nr_anon, unsigned long nr_file, |
5423 | unsigned long nr_huge, struct page *dummy_page) | |
5424 | { | |
18eca2e6 | 5425 | unsigned long nr_pages = nr_anon + nr_file; |
747db954 JW |
5426 | unsigned long flags; |
5427 | ||
ce00a967 | 5428 | if (!mem_cgroup_is_root(memcg)) { |
18eca2e6 JW |
5429 | page_counter_uncharge(&memcg->memory, nr_pages); |
5430 | if (do_swap_account) | |
5431 | page_counter_uncharge(&memcg->memsw, nr_pages); | |
ce00a967 JW |
5432 | memcg_oom_recover(memcg); |
5433 | } | |
747db954 JW |
5434 | |
5435 | local_irq_save(flags); | |
5436 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); | |
5437 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); | |
5438 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); | |
5439 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); | |
18eca2e6 | 5440 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
747db954 JW |
5441 | memcg_check_events(memcg, dummy_page); |
5442 | local_irq_restore(flags); | |
e8ea14cc JW |
5443 | |
5444 | if (!mem_cgroup_is_root(memcg)) | |
18eca2e6 | 5445 | css_put_many(&memcg->css, nr_pages); |
747db954 JW |
5446 | } |
5447 | ||
5448 | static void uncharge_list(struct list_head *page_list) | |
5449 | { | |
5450 | struct mem_cgroup *memcg = NULL; | |
747db954 JW |
5451 | unsigned long nr_anon = 0; |
5452 | unsigned long nr_file = 0; | |
5453 | unsigned long nr_huge = 0; | |
5454 | unsigned long pgpgout = 0; | |
747db954 JW |
5455 | struct list_head *next; |
5456 | struct page *page; | |
5457 | ||
5458 | next = page_list->next; | |
5459 | do { | |
5460 | unsigned int nr_pages = 1; | |
747db954 JW |
5461 | |
5462 | page = list_entry(next, struct page, lru); | |
5463 | next = page->lru.next; | |
5464 | ||
5465 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5466 | VM_BUG_ON_PAGE(page_count(page), page); | |
5467 | ||
1306a85a | 5468 | if (!page->mem_cgroup) |
747db954 JW |
5469 | continue; |
5470 | ||
5471 | /* | |
5472 | * Nobody should be changing or seriously looking at | |
1306a85a | 5473 | * page->mem_cgroup at this point, we have fully |
29833315 | 5474 | * exclusive access to the page. |
747db954 JW |
5475 | */ |
5476 | ||
1306a85a | 5477 | if (memcg != page->mem_cgroup) { |
747db954 | 5478 | if (memcg) { |
18eca2e6 JW |
5479 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5480 | nr_huge, page); | |
5481 | pgpgout = nr_anon = nr_file = nr_huge = 0; | |
747db954 | 5482 | } |
1306a85a | 5483 | memcg = page->mem_cgroup; |
747db954 JW |
5484 | } |
5485 | ||
5486 | if (PageTransHuge(page)) { | |
5487 | nr_pages <<= compound_order(page); | |
5488 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5489 | nr_huge += nr_pages; | |
5490 | } | |
5491 | ||
5492 | if (PageAnon(page)) | |
5493 | nr_anon += nr_pages; | |
5494 | else | |
5495 | nr_file += nr_pages; | |
5496 | ||
1306a85a | 5497 | page->mem_cgroup = NULL; |
747db954 JW |
5498 | |
5499 | pgpgout++; | |
5500 | } while (next != page_list); | |
5501 | ||
5502 | if (memcg) | |
18eca2e6 JW |
5503 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5504 | nr_huge, page); | |
747db954 JW |
5505 | } |
5506 | ||
0a31bc97 JW |
5507 | /** |
5508 | * mem_cgroup_uncharge - uncharge a page | |
5509 | * @page: page to uncharge | |
5510 | * | |
5511 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
5512 | * mem_cgroup_commit_charge(). | |
5513 | */ | |
5514 | void mem_cgroup_uncharge(struct page *page) | |
5515 | { | |
0a31bc97 JW |
5516 | if (mem_cgroup_disabled()) |
5517 | return; | |
5518 | ||
747db954 | 5519 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 5520 | if (!page->mem_cgroup) |
0a31bc97 JW |
5521 | return; |
5522 | ||
747db954 JW |
5523 | INIT_LIST_HEAD(&page->lru); |
5524 | uncharge_list(&page->lru); | |
5525 | } | |
0a31bc97 | 5526 | |
747db954 JW |
5527 | /** |
5528 | * mem_cgroup_uncharge_list - uncharge a list of page | |
5529 | * @page_list: list of pages to uncharge | |
5530 | * | |
5531 | * Uncharge a list of pages previously charged with | |
5532 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
5533 | */ | |
5534 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
5535 | { | |
5536 | if (mem_cgroup_disabled()) | |
5537 | return; | |
0a31bc97 | 5538 | |
747db954 JW |
5539 | if (!list_empty(page_list)) |
5540 | uncharge_list(page_list); | |
0a31bc97 JW |
5541 | } |
5542 | ||
5543 | /** | |
5544 | * mem_cgroup_migrate - migrate a charge to another page | |
5545 | * @oldpage: currently charged page | |
5546 | * @newpage: page to transfer the charge to | |
f5e03a49 | 5547 | * @lrucare: either or both pages might be on the LRU already |
0a31bc97 JW |
5548 | * |
5549 | * Migrate the charge from @oldpage to @newpage. | |
5550 | * | |
5551 | * Both pages must be locked, @newpage->mapping must be set up. | |
5552 | */ | |
5553 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage, | |
5554 | bool lrucare) | |
5555 | { | |
29833315 | 5556 | struct mem_cgroup *memcg; |
0a31bc97 JW |
5557 | int isolated; |
5558 | ||
5559 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
5560 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
5561 | VM_BUG_ON_PAGE(!lrucare && PageLRU(oldpage), oldpage); | |
5562 | VM_BUG_ON_PAGE(!lrucare && PageLRU(newpage), newpage); | |
5563 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); | |
6abb5a86 JW |
5564 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
5565 | newpage); | |
0a31bc97 JW |
5566 | |
5567 | if (mem_cgroup_disabled()) | |
5568 | return; | |
5569 | ||
5570 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 5571 | if (newpage->mem_cgroup) |
0a31bc97 JW |
5572 | return; |
5573 | ||
7d5e3245 JW |
5574 | /* |
5575 | * Swapcache readahead pages can get migrated before being | |
5576 | * charged, and migration from compaction can happen to an | |
5577 | * uncharged page when the PFN walker finds a page that | |
5578 | * reclaim just put back on the LRU but has not released yet. | |
5579 | */ | |
1306a85a | 5580 | memcg = oldpage->mem_cgroup; |
29833315 | 5581 | if (!memcg) |
0a31bc97 JW |
5582 | return; |
5583 | ||
0a31bc97 JW |
5584 | if (lrucare) |
5585 | lock_page_lru(oldpage, &isolated); | |
5586 | ||
1306a85a | 5587 | oldpage->mem_cgroup = NULL; |
0a31bc97 JW |
5588 | |
5589 | if (lrucare) | |
5590 | unlock_page_lru(oldpage, isolated); | |
5591 | ||
29833315 | 5592 | commit_charge(newpage, memcg, lrucare); |
0a31bc97 JW |
5593 | } |
5594 | ||
2d11085e | 5595 | /* |
1081312f MH |
5596 | * subsys_initcall() for memory controller. |
5597 | * | |
5598 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
5599 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
5600 | * everything that doesn't depend on a specific mem_cgroup structure should | |
5601 | * be initialized from here. | |
2d11085e MH |
5602 | */ |
5603 | static int __init mem_cgroup_init(void) | |
5604 | { | |
95a045f6 JW |
5605 | int cpu, node; |
5606 | ||
2d11085e | 5607 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
95a045f6 JW |
5608 | |
5609 | for_each_possible_cpu(cpu) | |
5610 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
5611 | drain_local_stock); | |
5612 | ||
5613 | for_each_node(node) { | |
5614 | struct mem_cgroup_tree_per_node *rtpn; | |
5615 | int zone; | |
5616 | ||
5617 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
5618 | node_online(node) ? node : NUMA_NO_NODE); | |
5619 | ||
5620 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
5621 | struct mem_cgroup_tree_per_zone *rtpz; | |
5622 | ||
5623 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
5624 | rtpz->rb_root = RB_ROOT; | |
5625 | spin_lock_init(&rtpz->lock); | |
5626 | } | |
5627 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
5628 | } | |
5629 | ||
2d11085e MH |
5630 | return 0; |
5631 | } | |
5632 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
5633 | |
5634 | #ifdef CONFIG_MEMCG_SWAP | |
5635 | /** | |
5636 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
5637 | * @page: page whose memsw charge to transfer | |
5638 | * @entry: swap entry to move the charge to | |
5639 | * | |
5640 | * Transfer the memsw charge of @page to @entry. | |
5641 | */ | |
5642 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
5643 | { | |
5644 | struct mem_cgroup *memcg; | |
5645 | unsigned short oldid; | |
5646 | ||
5647 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5648 | VM_BUG_ON_PAGE(page_count(page), page); | |
5649 | ||
5650 | if (!do_swap_account) | |
5651 | return; | |
5652 | ||
5653 | memcg = page->mem_cgroup; | |
5654 | ||
5655 | /* Readahead page, never charged */ | |
5656 | if (!memcg) | |
5657 | return; | |
5658 | ||
5659 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); | |
5660 | VM_BUG_ON_PAGE(oldid, page); | |
5661 | mem_cgroup_swap_statistics(memcg, true); | |
5662 | ||
5663 | page->mem_cgroup = NULL; | |
5664 | ||
5665 | if (!mem_cgroup_is_root(memcg)) | |
5666 | page_counter_uncharge(&memcg->memory, 1); | |
5667 | ||
ce9ce665 SAS |
5668 | /* |
5669 | * Interrupts should be disabled here because the caller holds the | |
5670 | * mapping->tree_lock lock which is taken with interrupts-off. It is | |
5671 | * important here to have the interrupts disabled because it is the | |
5672 | * only synchronisation we have for udpating the per-CPU variables. | |
5673 | */ | |
5674 | VM_BUG_ON(!irqs_disabled()); | |
21afa38e JW |
5675 | mem_cgroup_charge_statistics(memcg, page, -1); |
5676 | memcg_check_events(memcg, page); | |
5677 | } | |
5678 | ||
5679 | /** | |
5680 | * mem_cgroup_uncharge_swap - uncharge a swap entry | |
5681 | * @entry: swap entry to uncharge | |
5682 | * | |
5683 | * Drop the memsw charge associated with @entry. | |
5684 | */ | |
5685 | void mem_cgroup_uncharge_swap(swp_entry_t entry) | |
5686 | { | |
5687 | struct mem_cgroup *memcg; | |
5688 | unsigned short id; | |
5689 | ||
5690 | if (!do_swap_account) | |
5691 | return; | |
5692 | ||
5693 | id = swap_cgroup_record(entry, 0); | |
5694 | rcu_read_lock(); | |
adbe427b | 5695 | memcg = mem_cgroup_from_id(id); |
21afa38e JW |
5696 | if (memcg) { |
5697 | if (!mem_cgroup_is_root(memcg)) | |
5698 | page_counter_uncharge(&memcg->memsw, 1); | |
5699 | mem_cgroup_swap_statistics(memcg, false); | |
5700 | css_put(&memcg->css); | |
5701 | } | |
5702 | rcu_read_unlock(); | |
5703 | } | |
5704 | ||
5705 | /* for remember boot option*/ | |
5706 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
5707 | static int really_do_swap_account __initdata = 1; | |
5708 | #else | |
5709 | static int really_do_swap_account __initdata; | |
5710 | #endif | |
5711 | ||
5712 | static int __init enable_swap_account(char *s) | |
5713 | { | |
5714 | if (!strcmp(s, "1")) | |
5715 | really_do_swap_account = 1; | |
5716 | else if (!strcmp(s, "0")) | |
5717 | really_do_swap_account = 0; | |
5718 | return 1; | |
5719 | } | |
5720 | __setup("swapaccount=", enable_swap_account); | |
5721 | ||
5722 | static struct cftype memsw_cgroup_files[] = { | |
5723 | { | |
5724 | .name = "memsw.usage_in_bytes", | |
5725 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
5726 | .read_u64 = mem_cgroup_read_u64, | |
5727 | }, | |
5728 | { | |
5729 | .name = "memsw.max_usage_in_bytes", | |
5730 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
5731 | .write = mem_cgroup_reset, | |
5732 | .read_u64 = mem_cgroup_read_u64, | |
5733 | }, | |
5734 | { | |
5735 | .name = "memsw.limit_in_bytes", | |
5736 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
5737 | .write = mem_cgroup_write, | |
5738 | .read_u64 = mem_cgroup_read_u64, | |
5739 | }, | |
5740 | { | |
5741 | .name = "memsw.failcnt", | |
5742 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
5743 | .write = mem_cgroup_reset, | |
5744 | .read_u64 = mem_cgroup_read_u64, | |
5745 | }, | |
5746 | { }, /* terminate */ | |
5747 | }; | |
5748 | ||
5749 | static int __init mem_cgroup_swap_init(void) | |
5750 | { | |
5751 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
5752 | do_swap_account = 1; | |
5753 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, | |
5754 | memsw_cgroup_files)); | |
5755 | } | |
5756 | return 0; | |
5757 | } | |
5758 | subsys_initcall(mem_cgroup_swap_init); | |
5759 | ||
5760 | #endif /* CONFIG_MEMCG_SWAP */ |