Commit | Line | Data |
---|---|---|
8cdea7c0 BS |
1 | /* memcontrol.c - Memory Controller |
2 | * | |
3 | * Copyright IBM Corporation, 2007 | |
4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
5 | * | |
78fb7466 PE |
6 | * Copyright 2007 OpenVZ SWsoft Inc |
7 | * Author: Pavel Emelianov <xemul@openvz.org> | |
8 | * | |
2e72b634 KS |
9 | * Memory thresholds |
10 | * Copyright (C) 2009 Nokia Corporation | |
11 | * Author: Kirill A. Shutemov | |
12 | * | |
7ae1e1d0 GC |
13 | * Kernel Memory Controller |
14 | * Copyright (C) 2012 Parallels Inc. and Google Inc. | |
15 | * Authors: Glauber Costa and Suleiman Souhlal | |
16 | * | |
1575e68b JW |
17 | * Native page reclaim |
18 | * Charge lifetime sanitation | |
19 | * Lockless page tracking & accounting | |
20 | * Unified hierarchy configuration model | |
21 | * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner | |
22 | * | |
8cdea7c0 BS |
23 | * This program is free software; you can redistribute it and/or modify |
24 | * it under the terms of the GNU General Public License as published by | |
25 | * the Free Software Foundation; either version 2 of the License, or | |
26 | * (at your option) any later version. | |
27 | * | |
28 | * This program is distributed in the hope that it will be useful, | |
29 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
30 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
31 | * GNU General Public License for more details. | |
32 | */ | |
33 | ||
3e32cb2e | 34 | #include <linux/page_counter.h> |
8cdea7c0 BS |
35 | #include <linux/memcontrol.h> |
36 | #include <linux/cgroup.h> | |
78fb7466 | 37 | #include <linux/mm.h> |
4ffef5fe | 38 | #include <linux/hugetlb.h> |
d13d1443 | 39 | #include <linux/pagemap.h> |
d52aa412 | 40 | #include <linux/smp.h> |
8a9f3ccd | 41 | #include <linux/page-flags.h> |
66e1707b | 42 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
43 | #include <linux/bit_spinlock.h> |
44 | #include <linux/rcupdate.h> | |
e222432b | 45 | #include <linux/limits.h> |
b9e15baf | 46 | #include <linux/export.h> |
8c7c6e34 | 47 | #include <linux/mutex.h> |
bb4cc1a8 | 48 | #include <linux/rbtree.h> |
b6ac57d5 | 49 | #include <linux/slab.h> |
66e1707b | 50 | #include <linux/swap.h> |
02491447 | 51 | #include <linux/swapops.h> |
66e1707b | 52 | #include <linux/spinlock.h> |
2e72b634 | 53 | #include <linux/eventfd.h> |
79bd9814 | 54 | #include <linux/poll.h> |
2e72b634 | 55 | #include <linux/sort.h> |
66e1707b | 56 | #include <linux/fs.h> |
d2ceb9b7 | 57 | #include <linux/seq_file.h> |
70ddf637 | 58 | #include <linux/vmpressure.h> |
b69408e8 | 59 | #include <linux/mm_inline.h> |
5d1ea48b | 60 | #include <linux/swap_cgroup.h> |
cdec2e42 | 61 | #include <linux/cpu.h> |
158e0a2d | 62 | #include <linux/oom.h> |
0056f4e6 | 63 | #include <linux/lockdep.h> |
79bd9814 | 64 | #include <linux/file.h> |
b23afb93 | 65 | #include <linux/tracehook.h> |
08e552c6 | 66 | #include "internal.h" |
d1a4c0b3 | 67 | #include <net/sock.h> |
4bd2c1ee | 68 | #include <net/ip.h> |
d1a4c0b3 | 69 | #include <net/tcp_memcontrol.h> |
f35c3a8e | 70 | #include "slab.h" |
8cdea7c0 | 71 | |
8697d331 BS |
72 | #include <asm/uaccess.h> |
73 | ||
cc8e970c KM |
74 | #include <trace/events/vmscan.h> |
75 | ||
073219e9 TH |
76 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
77 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 78 | |
a181b0e8 | 79 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
6bbda35c | 80 | static struct mem_cgroup *root_mem_cgroup __read_mostly; |
56161634 | 81 | struct cgroup_subsys_state *mem_cgroup_root_css __read_mostly; |
8cdea7c0 | 82 | |
21afa38e | 83 | /* Whether the swap controller is active */ |
c255a458 | 84 | #ifdef CONFIG_MEMCG_SWAP |
c077719b | 85 | int do_swap_account __read_mostly; |
c077719b | 86 | #else |
a0db00fc | 87 | #define do_swap_account 0 |
c077719b KH |
88 | #endif |
89 | ||
af7c4b0e JW |
90 | static const char * const mem_cgroup_stat_names[] = { |
91 | "cache", | |
92 | "rss", | |
b070e65c | 93 | "rss_huge", |
af7c4b0e | 94 | "mapped_file", |
c4843a75 | 95 | "dirty", |
3ea67d06 | 96 | "writeback", |
af7c4b0e JW |
97 | "swap", |
98 | }; | |
99 | ||
af7c4b0e JW |
100 | static const char * const mem_cgroup_events_names[] = { |
101 | "pgpgin", | |
102 | "pgpgout", | |
103 | "pgfault", | |
104 | "pgmajfault", | |
105 | }; | |
106 | ||
58cf188e SZ |
107 | static const char * const mem_cgroup_lru_names[] = { |
108 | "inactive_anon", | |
109 | "active_anon", | |
110 | "inactive_file", | |
111 | "active_file", | |
112 | "unevictable", | |
113 | }; | |
114 | ||
a0db00fc KS |
115 | #define THRESHOLDS_EVENTS_TARGET 128 |
116 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
117 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 118 | |
bb4cc1a8 AM |
119 | /* |
120 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
121 | * their hierarchy representation | |
122 | */ | |
123 | ||
124 | struct mem_cgroup_tree_per_zone { | |
125 | struct rb_root rb_root; | |
126 | spinlock_t lock; | |
127 | }; | |
128 | ||
129 | struct mem_cgroup_tree_per_node { | |
130 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
131 | }; | |
132 | ||
133 | struct mem_cgroup_tree { | |
134 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
135 | }; | |
136 | ||
137 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
138 | ||
9490ff27 KH |
139 | /* for OOM */ |
140 | struct mem_cgroup_eventfd_list { | |
141 | struct list_head list; | |
142 | struct eventfd_ctx *eventfd; | |
143 | }; | |
2e72b634 | 144 | |
79bd9814 TH |
145 | /* |
146 | * cgroup_event represents events which userspace want to receive. | |
147 | */ | |
3bc942f3 | 148 | struct mem_cgroup_event { |
79bd9814 | 149 | /* |
59b6f873 | 150 | * memcg which the event belongs to. |
79bd9814 | 151 | */ |
59b6f873 | 152 | struct mem_cgroup *memcg; |
79bd9814 TH |
153 | /* |
154 | * eventfd to signal userspace about the event. | |
155 | */ | |
156 | struct eventfd_ctx *eventfd; | |
157 | /* | |
158 | * Each of these stored in a list by the cgroup. | |
159 | */ | |
160 | struct list_head list; | |
fba94807 TH |
161 | /* |
162 | * register_event() callback will be used to add new userspace | |
163 | * waiter for changes related to this event. Use eventfd_signal() | |
164 | * on eventfd to send notification to userspace. | |
165 | */ | |
59b6f873 | 166 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 167 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
168 | /* |
169 | * unregister_event() callback will be called when userspace closes | |
170 | * the eventfd or on cgroup removing. This callback must be set, | |
171 | * if you want provide notification functionality. | |
172 | */ | |
59b6f873 | 173 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 174 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
175 | /* |
176 | * All fields below needed to unregister event when | |
177 | * userspace closes eventfd. | |
178 | */ | |
179 | poll_table pt; | |
180 | wait_queue_head_t *wqh; | |
181 | wait_queue_t wait; | |
182 | struct work_struct remove; | |
183 | }; | |
184 | ||
c0ff4b85 R |
185 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
186 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 187 | |
7dc74be0 DN |
188 | /* Stuffs for move charges at task migration. */ |
189 | /* | |
1dfab5ab | 190 | * Types of charges to be moved. |
7dc74be0 | 191 | */ |
1dfab5ab JW |
192 | #define MOVE_ANON 0x1U |
193 | #define MOVE_FILE 0x2U | |
194 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 195 | |
4ffef5fe DN |
196 | /* "mc" and its members are protected by cgroup_mutex */ |
197 | static struct move_charge_struct { | |
b1dd693e | 198 | spinlock_t lock; /* for from, to */ |
52526076 | 199 | struct mm_struct *mm; |
4ffef5fe DN |
200 | struct mem_cgroup *from; |
201 | struct mem_cgroup *to; | |
1dfab5ab | 202 | unsigned long flags; |
4ffef5fe | 203 | unsigned long precharge; |
854ffa8d | 204 | unsigned long moved_charge; |
483c30b5 | 205 | unsigned long moved_swap; |
8033b97c DN |
206 | struct task_struct *moving_task; /* a task moving charges */ |
207 | wait_queue_head_t waitq; /* a waitq for other context */ | |
208 | } mc = { | |
2bd9bb20 | 209 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
210 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
211 | }; | |
4ffef5fe | 212 | |
4e416953 BS |
213 | /* |
214 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
215 | * limit reclaim to prevent infinite loops, if they ever occur. | |
216 | */ | |
a0db00fc | 217 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 218 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 219 | |
217bc319 KH |
220 | enum charge_type { |
221 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 222 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 223 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 224 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
225 | NR_CHARGE_TYPE, |
226 | }; | |
227 | ||
8c7c6e34 | 228 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
229 | enum res_type { |
230 | _MEM, | |
231 | _MEMSWAP, | |
232 | _OOM_TYPE, | |
510fc4e1 | 233 | _KMEM, |
86ae53e1 GC |
234 | }; |
235 | ||
a0db00fc KS |
236 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
237 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 238 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
239 | /* Used for OOM nofiier */ |
240 | #define OOM_CONTROL (0) | |
8c7c6e34 | 241 | |
0999821b GC |
242 | /* |
243 | * The memcg_create_mutex will be held whenever a new cgroup is created. | |
244 | * As a consequence, any change that needs to protect against new child cgroups | |
245 | * appearing has to hold it as well. | |
246 | */ | |
247 | static DEFINE_MUTEX(memcg_create_mutex); | |
248 | ||
70ddf637 AV |
249 | /* Some nice accessors for the vmpressure. */ |
250 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
251 | { | |
252 | if (!memcg) | |
253 | memcg = root_mem_cgroup; | |
254 | return &memcg->vmpressure; | |
255 | } | |
256 | ||
257 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
258 | { | |
259 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
260 | } | |
261 | ||
7ffc0edc MH |
262 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
263 | { | |
264 | return (memcg == root_mem_cgroup); | |
265 | } | |
266 | ||
4219b2da LZ |
267 | /* |
268 | * We restrict the id in the range of [1, 65535], so it can fit into | |
269 | * an unsigned short. | |
270 | */ | |
271 | #define MEM_CGROUP_ID_MAX USHRT_MAX | |
272 | ||
34c00c31 LZ |
273 | static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) |
274 | { | |
8627c775 | 275 | return memcg->id.id; |
34c00c31 LZ |
276 | } |
277 | ||
e1aab161 | 278 | /* Writing them here to avoid exposing memcg's inner layout */ |
4bd2c1ee | 279 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
e1aab161 | 280 | |
e1aab161 GC |
281 | void sock_update_memcg(struct sock *sk) |
282 | { | |
376be5ff | 283 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 284 | struct mem_cgroup *memcg; |
3f134619 | 285 | struct cg_proto *cg_proto; |
e1aab161 GC |
286 | |
287 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
288 | ||
f3f511e1 GC |
289 | /* Socket cloning can throw us here with sk_cgrp already |
290 | * filled. It won't however, necessarily happen from | |
291 | * process context. So the test for root memcg given | |
292 | * the current task's memcg won't help us in this case. | |
293 | * | |
294 | * Respecting the original socket's memcg is a better | |
295 | * decision in this case. | |
296 | */ | |
297 | if (sk->sk_cgrp) { | |
298 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
5347e5ae | 299 | css_get(&sk->sk_cgrp->memcg->css); |
f3f511e1 GC |
300 | return; |
301 | } | |
302 | ||
e1aab161 GC |
303 | rcu_read_lock(); |
304 | memcg = mem_cgroup_from_task(current); | |
3f134619 | 305 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
e752eb68 | 306 | if (cg_proto && test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags) && |
ec903c0c | 307 | css_tryget_online(&memcg->css)) { |
3f134619 | 308 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
309 | } |
310 | rcu_read_unlock(); | |
311 | } | |
312 | } | |
313 | EXPORT_SYMBOL(sock_update_memcg); | |
314 | ||
315 | void sock_release_memcg(struct sock *sk) | |
316 | { | |
376be5ff | 317 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
318 | struct mem_cgroup *memcg; |
319 | WARN_ON(!sk->sk_cgrp->memcg); | |
320 | memcg = sk->sk_cgrp->memcg; | |
5347e5ae | 321 | css_put(&sk->sk_cgrp->memcg->css); |
e1aab161 GC |
322 | } |
323 | } | |
d1a4c0b3 GC |
324 | |
325 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
326 | { | |
327 | if (!memcg || mem_cgroup_is_root(memcg)) | |
328 | return NULL; | |
329 | ||
2e685cad | 330 | return &memcg->tcp_mem; |
d1a4c0b3 GC |
331 | } |
332 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 333 | |
3f134619 GC |
334 | #endif |
335 | ||
a8964b9b | 336 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 | 337 | /* |
f7ce3190 | 338 | * This will be the memcg's index in each cache's ->memcg_params.memcg_caches. |
b8627835 LZ |
339 | * The main reason for not using cgroup id for this: |
340 | * this works better in sparse environments, where we have a lot of memcgs, | |
341 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
342 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
343 | * 200 entry array for that. | |
55007d84 | 344 | * |
dbcf73e2 VD |
345 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
346 | * will double each time we have to increase it. | |
55007d84 | 347 | */ |
dbcf73e2 VD |
348 | static DEFINE_IDA(memcg_cache_ida); |
349 | int memcg_nr_cache_ids; | |
749c5415 | 350 | |
05257a1a VD |
351 | /* Protects memcg_nr_cache_ids */ |
352 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
353 | ||
354 | void memcg_get_cache_ids(void) | |
355 | { | |
356 | down_read(&memcg_cache_ids_sem); | |
357 | } | |
358 | ||
359 | void memcg_put_cache_ids(void) | |
360 | { | |
361 | up_read(&memcg_cache_ids_sem); | |
362 | } | |
363 | ||
55007d84 GC |
364 | /* |
365 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
366 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
367 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
368 | * tunable, but that is strictly not necessary. | |
369 | * | |
b8627835 | 370 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
371 | * this constant directly from cgroup, but it is understandable that this is |
372 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 373 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
374 | * increase ours as well if it increases. |
375 | */ | |
376 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 377 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 378 | |
d7f25f8a GC |
379 | /* |
380 | * A lot of the calls to the cache allocation functions are expected to be | |
381 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
382 | * conditional to this static branch, we'll have to allow modules that does | |
383 | * kmem_cache_alloc and the such to see this symbol as well | |
384 | */ | |
a8964b9b | 385 | struct static_key memcg_kmem_enabled_key; |
d7f25f8a | 386 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 387 | |
a8964b9b GC |
388 | #endif /* CONFIG_MEMCG_KMEM */ |
389 | ||
f64c3f54 | 390 | static struct mem_cgroup_per_zone * |
e231875b | 391 | mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone) |
f64c3f54 | 392 | { |
e231875b JZ |
393 | int nid = zone_to_nid(zone); |
394 | int zid = zone_idx(zone); | |
395 | ||
54f72fe0 | 396 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
397 | } |
398 | ||
ad7fa852 TH |
399 | /** |
400 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
401 | * @page: page of interest | |
402 | * | |
403 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
404 | * with @page is returned. The returned css remains associated with @page | |
405 | * until it is released. | |
406 | * | |
407 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
408 | * is returned. | |
409 | * | |
410 | * XXX: The above description of behavior on the default hierarchy isn't | |
411 | * strictly true yet as replace_page_cache_page() can modify the | |
412 | * association before @page is released even on the default hierarchy; | |
413 | * however, the current and planned usages don't mix the the two functions | |
414 | * and replace_page_cache_page() will soon be updated to make the invariant | |
415 | * actually true. | |
416 | */ | |
417 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
418 | { | |
419 | struct mem_cgroup *memcg; | |
420 | ||
421 | rcu_read_lock(); | |
422 | ||
423 | memcg = page->mem_cgroup; | |
424 | ||
9e10a130 | 425 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
426 | memcg = root_mem_cgroup; |
427 | ||
428 | rcu_read_unlock(); | |
429 | return &memcg->css; | |
430 | } | |
431 | ||
2fc04524 VD |
432 | /** |
433 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
434 | * @page: the page | |
435 | * | |
436 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
437 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
438 | * is safe to call this function without holding a reference to @page. | |
439 | * | |
440 | * Note, this function is inherently racy, because there is nothing to prevent | |
441 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
442 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
443 | * do not care (such as procfs interfaces). | |
444 | */ | |
445 | ino_t page_cgroup_ino(struct page *page) | |
446 | { | |
447 | struct mem_cgroup *memcg; | |
448 | unsigned long ino = 0; | |
449 | ||
450 | rcu_read_lock(); | |
451 | memcg = READ_ONCE(page->mem_cgroup); | |
452 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) | |
453 | memcg = parent_mem_cgroup(memcg); | |
454 | if (memcg) | |
455 | ino = cgroup_ino(memcg->css.cgroup); | |
456 | rcu_read_unlock(); | |
457 | return ino; | |
458 | } | |
459 | ||
f64c3f54 | 460 | static struct mem_cgroup_per_zone * |
e231875b | 461 | mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 462 | { |
97a6c37b JW |
463 | int nid = page_to_nid(page); |
464 | int zid = page_zonenum(page); | |
f64c3f54 | 465 | |
e231875b | 466 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
467 | } |
468 | ||
bb4cc1a8 AM |
469 | static struct mem_cgroup_tree_per_zone * |
470 | soft_limit_tree_node_zone(int nid, int zid) | |
471 | { | |
472 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
473 | } | |
474 | ||
475 | static struct mem_cgroup_tree_per_zone * | |
476 | soft_limit_tree_from_page(struct page *page) | |
477 | { | |
478 | int nid = page_to_nid(page); | |
479 | int zid = page_zonenum(page); | |
480 | ||
481 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
482 | } | |
483 | ||
cf2c8127 JW |
484 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz, |
485 | struct mem_cgroup_tree_per_zone *mctz, | |
3e32cb2e | 486 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
487 | { |
488 | struct rb_node **p = &mctz->rb_root.rb_node; | |
489 | struct rb_node *parent = NULL; | |
490 | struct mem_cgroup_per_zone *mz_node; | |
491 | ||
492 | if (mz->on_tree) | |
493 | return; | |
494 | ||
495 | mz->usage_in_excess = new_usage_in_excess; | |
496 | if (!mz->usage_in_excess) | |
497 | return; | |
498 | while (*p) { | |
499 | parent = *p; | |
500 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
501 | tree_node); | |
502 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
503 | p = &(*p)->rb_left; | |
504 | /* | |
505 | * We can't avoid mem cgroups that are over their soft | |
506 | * limit by the same amount | |
507 | */ | |
508 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
509 | p = &(*p)->rb_right; | |
510 | } | |
511 | rb_link_node(&mz->tree_node, parent, p); | |
512 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
513 | mz->on_tree = true; | |
514 | } | |
515 | ||
cf2c8127 JW |
516 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
517 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 AM |
518 | { |
519 | if (!mz->on_tree) | |
520 | return; | |
521 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
522 | mz->on_tree = false; | |
523 | } | |
524 | ||
cf2c8127 JW |
525 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
526 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 | 527 | { |
0a31bc97 JW |
528 | unsigned long flags; |
529 | ||
530 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 531 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 532 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
533 | } |
534 | ||
3e32cb2e JW |
535 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
536 | { | |
537 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 538 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
539 | unsigned long excess = 0; |
540 | ||
541 | if (nr_pages > soft_limit) | |
542 | excess = nr_pages - soft_limit; | |
543 | ||
544 | return excess; | |
545 | } | |
bb4cc1a8 AM |
546 | |
547 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
548 | { | |
3e32cb2e | 549 | unsigned long excess; |
bb4cc1a8 AM |
550 | struct mem_cgroup_per_zone *mz; |
551 | struct mem_cgroup_tree_per_zone *mctz; | |
bb4cc1a8 | 552 | |
e231875b | 553 | mctz = soft_limit_tree_from_page(page); |
bb4cc1a8 AM |
554 | /* |
555 | * Necessary to update all ancestors when hierarchy is used. | |
556 | * because their event counter is not touched. | |
557 | */ | |
558 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
e231875b | 559 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
3e32cb2e | 560 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
561 | /* |
562 | * We have to update the tree if mz is on RB-tree or | |
563 | * mem is over its softlimit. | |
564 | */ | |
565 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
566 | unsigned long flags; |
567 | ||
568 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
569 | /* if on-tree, remove it */ |
570 | if (mz->on_tree) | |
cf2c8127 | 571 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
572 | /* |
573 | * Insert again. mz->usage_in_excess will be updated. | |
574 | * If excess is 0, no tree ops. | |
575 | */ | |
cf2c8127 | 576 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 577 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
578 | } |
579 | } | |
580 | } | |
581 | ||
582 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
583 | { | |
bb4cc1a8 | 584 | struct mem_cgroup_tree_per_zone *mctz; |
e231875b JZ |
585 | struct mem_cgroup_per_zone *mz; |
586 | int nid, zid; | |
bb4cc1a8 | 587 | |
e231875b JZ |
588 | for_each_node(nid) { |
589 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
590 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
591 | mctz = soft_limit_tree_node_zone(nid, zid); | |
cf2c8127 | 592 | mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
593 | } |
594 | } | |
595 | } | |
596 | ||
597 | static struct mem_cgroup_per_zone * | |
598 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
599 | { | |
600 | struct rb_node *rightmost = NULL; | |
601 | struct mem_cgroup_per_zone *mz; | |
602 | ||
603 | retry: | |
604 | mz = NULL; | |
605 | rightmost = rb_last(&mctz->rb_root); | |
606 | if (!rightmost) | |
607 | goto done; /* Nothing to reclaim from */ | |
608 | ||
609 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
610 | /* | |
611 | * Remove the node now but someone else can add it back, | |
612 | * we will to add it back at the end of reclaim to its correct | |
613 | * position in the tree. | |
614 | */ | |
cf2c8127 | 615 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 616 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 617 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
618 | goto retry; |
619 | done: | |
620 | return mz; | |
621 | } | |
622 | ||
623 | static struct mem_cgroup_per_zone * | |
624 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
625 | { | |
626 | struct mem_cgroup_per_zone *mz; | |
627 | ||
0a31bc97 | 628 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 629 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 630 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
631 | return mz; |
632 | } | |
633 | ||
711d3d2c | 634 | /* |
484ebb3b GT |
635 | * Return page count for single (non recursive) @memcg. |
636 | * | |
711d3d2c KH |
637 | * Implementation Note: reading percpu statistics for memcg. |
638 | * | |
639 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
640 | * synchronization to implement "quick" read. There are trade-off between | |
641 | * reading cost and precision of value. Then, we may have a chance to implement | |
484ebb3b | 642 | * a periodic synchronization of counter in memcg's counter. |
711d3d2c KH |
643 | * |
644 | * But this _read() function is used for user interface now. The user accounts | |
645 | * memory usage by memory cgroup and he _always_ requires exact value because | |
646 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
647 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
648 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
649 | * | |
650 | * If there are kernel internal actions which can make use of some not-exact | |
651 | * value, and reading all cpu value can be performance bottleneck in some | |
484ebb3b | 652 | * common workload, threshold and synchronization as vmstat[] should be |
711d3d2c KH |
653 | * implemented. |
654 | */ | |
484ebb3b GT |
655 | static unsigned long |
656 | mem_cgroup_read_stat(struct mem_cgroup *memcg, enum mem_cgroup_stat_index idx) | |
c62b1a3b | 657 | { |
7a159cc9 | 658 | long val = 0; |
c62b1a3b | 659 | int cpu; |
c62b1a3b | 660 | |
484ebb3b | 661 | /* Per-cpu values can be negative, use a signed accumulator */ |
733a572e | 662 | for_each_possible_cpu(cpu) |
c0ff4b85 | 663 | val += per_cpu(memcg->stat->count[idx], cpu); |
484ebb3b GT |
664 | /* |
665 | * Summing races with updates, so val may be negative. Avoid exposing | |
666 | * transient negative values. | |
667 | */ | |
668 | if (val < 0) | |
669 | val = 0; | |
c62b1a3b KH |
670 | return val; |
671 | } | |
672 | ||
c0ff4b85 | 673 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
674 | enum mem_cgroup_events_index idx) |
675 | { | |
676 | unsigned long val = 0; | |
677 | int cpu; | |
678 | ||
733a572e | 679 | for_each_possible_cpu(cpu) |
c0ff4b85 | 680 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f JW |
681 | return val; |
682 | } | |
683 | ||
c0ff4b85 | 684 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 685 | struct page *page, |
0a31bc97 | 686 | int nr_pages) |
d52aa412 | 687 | { |
b2402857 KH |
688 | /* |
689 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
690 | * counted as CACHE even if it's on ANON LRU. | |
691 | */ | |
0a31bc97 | 692 | if (PageAnon(page)) |
b2402857 | 693 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
c0ff4b85 | 694 | nr_pages); |
d52aa412 | 695 | else |
b2402857 | 696 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 697 | nr_pages); |
55e462b0 | 698 | |
b070e65c DR |
699 | if (PageTransHuge(page)) |
700 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], | |
701 | nr_pages); | |
702 | ||
e401f176 KH |
703 | /* pagein of a big page is an event. So, ignore page size */ |
704 | if (nr_pages > 0) | |
c0ff4b85 | 705 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 706 | else { |
c0ff4b85 | 707 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
708 | nr_pages = -nr_pages; /* for event */ |
709 | } | |
e401f176 | 710 | |
13114716 | 711 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
6d12e2d8 KH |
712 | } |
713 | ||
e231875b JZ |
714 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
715 | int nid, | |
716 | unsigned int lru_mask) | |
bb2a0de9 | 717 | { |
e231875b | 718 | unsigned long nr = 0; |
889976db YH |
719 | int zid; |
720 | ||
e231875b | 721 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 722 | |
e231875b JZ |
723 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
724 | struct mem_cgroup_per_zone *mz; | |
725 | enum lru_list lru; | |
726 | ||
727 | for_each_lru(lru) { | |
728 | if (!(BIT(lru) & lru_mask)) | |
729 | continue; | |
730 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
731 | nr += mz->lru_size[lru]; | |
732 | } | |
733 | } | |
734 | return nr; | |
889976db | 735 | } |
bb2a0de9 | 736 | |
c0ff4b85 | 737 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 738 | unsigned int lru_mask) |
6d12e2d8 | 739 | { |
e231875b | 740 | unsigned long nr = 0; |
889976db | 741 | int nid; |
6d12e2d8 | 742 | |
31aaea4a | 743 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
744 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
745 | return nr; | |
d52aa412 KH |
746 | } |
747 | ||
f53d7ce3 JW |
748 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
749 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
750 | { |
751 | unsigned long val, next; | |
752 | ||
13114716 | 753 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 754 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 755 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
756 | if ((long)next - (long)val < 0) { |
757 | switch (target) { | |
758 | case MEM_CGROUP_TARGET_THRESH: | |
759 | next = val + THRESHOLDS_EVENTS_TARGET; | |
760 | break; | |
bb4cc1a8 AM |
761 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
762 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
763 | break; | |
f53d7ce3 JW |
764 | case MEM_CGROUP_TARGET_NUMAINFO: |
765 | next = val + NUMAINFO_EVENTS_TARGET; | |
766 | break; | |
767 | default: | |
768 | break; | |
769 | } | |
770 | __this_cpu_write(memcg->stat->targets[target], next); | |
771 | return true; | |
7a159cc9 | 772 | } |
f53d7ce3 | 773 | return false; |
d2265e6f KH |
774 | } |
775 | ||
776 | /* | |
777 | * Check events in order. | |
778 | * | |
779 | */ | |
c0ff4b85 | 780 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
781 | { |
782 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
783 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
784 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 785 | bool do_softlimit; |
82b3f2a7 | 786 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 787 | |
bb4cc1a8 AM |
788 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
789 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
790 | #if MAX_NUMNODES > 1 |
791 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
792 | MEM_CGROUP_TARGET_NUMAINFO); | |
793 | #endif | |
c0ff4b85 | 794 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
795 | if (unlikely(do_softlimit)) |
796 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 797 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 798 | if (unlikely(do_numainfo)) |
c0ff4b85 | 799 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 800 | #endif |
0a31bc97 | 801 | } |
d2265e6f KH |
802 | } |
803 | ||
cf475ad2 | 804 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 805 | { |
31a78f23 BS |
806 | /* |
807 | * mm_update_next_owner() may clear mm->owner to NULL | |
808 | * if it races with swapoff, page migration, etc. | |
809 | * So this can be called with p == NULL. | |
810 | */ | |
811 | if (unlikely(!p)) | |
812 | return NULL; | |
813 | ||
073219e9 | 814 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 815 | } |
33398cf2 | 816 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 817 | |
df381975 | 818 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 819 | { |
c0ff4b85 | 820 | struct mem_cgroup *memcg = NULL; |
0b7f569e | 821 | |
54595fe2 KH |
822 | rcu_read_lock(); |
823 | do { | |
6f6acb00 MH |
824 | /* |
825 | * Page cache insertions can happen withou an | |
826 | * actual mm context, e.g. during disk probing | |
827 | * on boot, loopback IO, acct() writes etc. | |
828 | */ | |
829 | if (unlikely(!mm)) | |
df381975 | 830 | memcg = root_mem_cgroup; |
6f6acb00 MH |
831 | else { |
832 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
833 | if (unlikely(!memcg)) | |
834 | memcg = root_mem_cgroup; | |
835 | } | |
ec903c0c | 836 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 837 | rcu_read_unlock(); |
c0ff4b85 | 838 | return memcg; |
54595fe2 KH |
839 | } |
840 | ||
5660048c JW |
841 | /** |
842 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
843 | * @root: hierarchy root | |
844 | * @prev: previously returned memcg, NULL on first invocation | |
845 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
846 | * | |
847 | * Returns references to children of the hierarchy below @root, or | |
848 | * @root itself, or %NULL after a full round-trip. | |
849 | * | |
850 | * Caller must pass the return value in @prev on subsequent | |
851 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
852 | * to cancel a hierarchy walk before the round-trip is complete. | |
853 | * | |
854 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
855 | * divide up the memcgs in the hierarchy among all concurrent | |
856 | * reclaimers operating on the same zone and priority. | |
857 | */ | |
694fbc0f | 858 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 859 | struct mem_cgroup *prev, |
694fbc0f | 860 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 861 | { |
33398cf2 | 862 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
5ac8fb31 | 863 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 864 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 865 | struct mem_cgroup *pos = NULL; |
711d3d2c | 866 | |
694fbc0f AM |
867 | if (mem_cgroup_disabled()) |
868 | return NULL; | |
5660048c | 869 | |
9f3a0d09 JW |
870 | if (!root) |
871 | root = root_mem_cgroup; | |
7d74b06f | 872 | |
9f3a0d09 | 873 | if (prev && !reclaim) |
5ac8fb31 | 874 | pos = prev; |
14067bb3 | 875 | |
9f3a0d09 JW |
876 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
877 | if (prev) | |
5ac8fb31 | 878 | goto out; |
694fbc0f | 879 | return root; |
9f3a0d09 | 880 | } |
14067bb3 | 881 | |
542f85f9 | 882 | rcu_read_lock(); |
5f578161 | 883 | |
5ac8fb31 JW |
884 | if (reclaim) { |
885 | struct mem_cgroup_per_zone *mz; | |
886 | ||
887 | mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone); | |
888 | iter = &mz->iter[reclaim->priority]; | |
889 | ||
890 | if (prev && reclaim->generation != iter->generation) | |
891 | goto out_unlock; | |
892 | ||
6df38689 | 893 | while (1) { |
4db0c3c2 | 894 | pos = READ_ONCE(iter->position); |
6df38689 VD |
895 | if (!pos || css_tryget(&pos->css)) |
896 | break; | |
5ac8fb31 | 897 | /* |
6df38689 VD |
898 | * css reference reached zero, so iter->position will |
899 | * be cleared by ->css_released. However, we should not | |
900 | * rely on this happening soon, because ->css_released | |
901 | * is called from a work queue, and by busy-waiting we | |
902 | * might block it. So we clear iter->position right | |
903 | * away. | |
5ac8fb31 | 904 | */ |
6df38689 VD |
905 | (void)cmpxchg(&iter->position, pos, NULL); |
906 | } | |
5ac8fb31 JW |
907 | } |
908 | ||
909 | if (pos) | |
910 | css = &pos->css; | |
911 | ||
912 | for (;;) { | |
913 | css = css_next_descendant_pre(css, &root->css); | |
914 | if (!css) { | |
915 | /* | |
916 | * Reclaimers share the hierarchy walk, and a | |
917 | * new one might jump in right at the end of | |
918 | * the hierarchy - make sure they see at least | |
919 | * one group and restart from the beginning. | |
920 | */ | |
921 | if (!prev) | |
922 | continue; | |
923 | break; | |
527a5ec9 | 924 | } |
7d74b06f | 925 | |
5ac8fb31 JW |
926 | /* |
927 | * Verify the css and acquire a reference. The root | |
928 | * is provided by the caller, so we know it's alive | |
929 | * and kicking, and don't take an extra reference. | |
930 | */ | |
931 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 932 | |
5ac8fb31 JW |
933 | if (css == &root->css) |
934 | break; | |
14067bb3 | 935 | |
b2052564 | 936 | if (css_tryget(css)) { |
5ac8fb31 JW |
937 | /* |
938 | * Make sure the memcg is initialized: | |
939 | * mem_cgroup_css_online() orders the the | |
940 | * initialization against setting the flag. | |
941 | */ | |
942 | if (smp_load_acquire(&memcg->initialized)) | |
943 | break; | |
542f85f9 | 944 | |
5ac8fb31 | 945 | css_put(css); |
527a5ec9 | 946 | } |
9f3a0d09 | 947 | |
5ac8fb31 | 948 | memcg = NULL; |
9f3a0d09 | 949 | } |
5ac8fb31 JW |
950 | |
951 | if (reclaim) { | |
5ac8fb31 | 952 | /* |
6df38689 VD |
953 | * The position could have already been updated by a competing |
954 | * thread, so check that the value hasn't changed since we read | |
955 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 956 | */ |
6df38689 VD |
957 | (void)cmpxchg(&iter->position, pos, memcg); |
958 | ||
5ac8fb31 JW |
959 | if (pos) |
960 | css_put(&pos->css); | |
961 | ||
962 | if (!memcg) | |
963 | iter->generation++; | |
964 | else if (!prev) | |
965 | reclaim->generation = iter->generation; | |
9f3a0d09 | 966 | } |
5ac8fb31 | 967 | |
542f85f9 MH |
968 | out_unlock: |
969 | rcu_read_unlock(); | |
5ac8fb31 | 970 | out: |
c40046f3 MH |
971 | if (prev && prev != root) |
972 | css_put(&prev->css); | |
973 | ||
9f3a0d09 | 974 | return memcg; |
14067bb3 | 975 | } |
7d74b06f | 976 | |
5660048c JW |
977 | /** |
978 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
979 | * @root: hierarchy root | |
980 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
981 | */ | |
982 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
983 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
984 | { |
985 | if (!root) | |
986 | root = root_mem_cgroup; | |
987 | if (prev && prev != root) | |
988 | css_put(&prev->css); | |
989 | } | |
7d74b06f | 990 | |
6df38689 VD |
991 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
992 | { | |
993 | struct mem_cgroup *memcg = dead_memcg; | |
994 | struct mem_cgroup_reclaim_iter *iter; | |
995 | struct mem_cgroup_per_zone *mz; | |
996 | int nid, zid; | |
997 | int i; | |
998 | ||
999 | while ((memcg = parent_mem_cgroup(memcg))) { | |
1000 | for_each_node(nid) { | |
1001 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1002 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
1003 | for (i = 0; i <= DEF_PRIORITY; i++) { | |
1004 | iter = &mz->iter[i]; | |
1005 | cmpxchg(&iter->position, | |
1006 | dead_memcg, NULL); | |
1007 | } | |
1008 | } | |
1009 | } | |
1010 | } | |
1011 | } | |
1012 | ||
9f3a0d09 JW |
1013 | /* |
1014 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1015 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1016 | * be used for reference counting. | |
1017 | */ | |
1018 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1019 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1020 | iter != NULL; \ |
527a5ec9 | 1021 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1022 | |
9f3a0d09 | 1023 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1024 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1025 | iter != NULL; \ |
527a5ec9 | 1026 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1027 | |
925b7673 JW |
1028 | /** |
1029 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1030 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1031 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1032 | * |
1033 | * Returns the lru list vector holding pages for the given @zone and | |
1034 | * @mem. This can be the global zone lruvec, if the memory controller | |
1035 | * is disabled. | |
1036 | */ | |
1037 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1038 | struct mem_cgroup *memcg) | |
1039 | { | |
1040 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1041 | struct lruvec *lruvec; |
925b7673 | 1042 | |
bea8c150 HD |
1043 | if (mem_cgroup_disabled()) { |
1044 | lruvec = &zone->lruvec; | |
1045 | goto out; | |
1046 | } | |
925b7673 | 1047 | |
e231875b | 1048 | mz = mem_cgroup_zone_zoneinfo(memcg, zone); |
bea8c150 HD |
1049 | lruvec = &mz->lruvec; |
1050 | out: | |
1051 | /* | |
1052 | * Since a node can be onlined after the mem_cgroup was created, | |
1053 | * we have to be prepared to initialize lruvec->zone here; | |
1054 | * and if offlined then reonlined, we need to reinitialize it. | |
1055 | */ | |
1056 | if (unlikely(lruvec->zone != zone)) | |
1057 | lruvec->zone = zone; | |
1058 | return lruvec; | |
925b7673 JW |
1059 | } |
1060 | ||
925b7673 | 1061 | /** |
dfe0e773 | 1062 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1063 | * @page: the page |
fa9add64 | 1064 | * @zone: zone of the page |
dfe0e773 JW |
1065 | * |
1066 | * This function is only safe when following the LRU page isolation | |
1067 | * and putback protocol: the LRU lock must be held, and the page must | |
1068 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 1069 | */ |
fa9add64 | 1070 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1071 | { |
08e552c6 | 1072 | struct mem_cgroup_per_zone *mz; |
925b7673 | 1073 | struct mem_cgroup *memcg; |
bea8c150 | 1074 | struct lruvec *lruvec; |
6d12e2d8 | 1075 | |
bea8c150 HD |
1076 | if (mem_cgroup_disabled()) { |
1077 | lruvec = &zone->lruvec; | |
1078 | goto out; | |
1079 | } | |
925b7673 | 1080 | |
1306a85a | 1081 | memcg = page->mem_cgroup; |
7512102c | 1082 | /* |
dfe0e773 | 1083 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1084 | * possibly migrated - before they are charged. |
7512102c | 1085 | */ |
29833315 JW |
1086 | if (!memcg) |
1087 | memcg = root_mem_cgroup; | |
7512102c | 1088 | |
e231875b | 1089 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
bea8c150 HD |
1090 | lruvec = &mz->lruvec; |
1091 | out: | |
1092 | /* | |
1093 | * Since a node can be onlined after the mem_cgroup was created, | |
1094 | * we have to be prepared to initialize lruvec->zone here; | |
1095 | * and if offlined then reonlined, we need to reinitialize it. | |
1096 | */ | |
1097 | if (unlikely(lruvec->zone != zone)) | |
1098 | lruvec->zone = zone; | |
1099 | return lruvec; | |
08e552c6 | 1100 | } |
b69408e8 | 1101 | |
925b7673 | 1102 | /** |
fa9add64 HD |
1103 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1104 | * @lruvec: mem_cgroup per zone lru vector | |
1105 | * @lru: index of lru list the page is sitting on | |
1106 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1107 | * |
fa9add64 HD |
1108 | * This function must be called when a page is added to or removed from an |
1109 | * lru list. | |
3f58a829 | 1110 | */ |
fa9add64 HD |
1111 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1112 | int nr_pages) | |
3f58a829 MK |
1113 | { |
1114 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1115 | unsigned long *lru_size; |
3f58a829 MK |
1116 | |
1117 | if (mem_cgroup_disabled()) | |
1118 | return; | |
1119 | ||
fa9add64 HD |
1120 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1121 | lru_size = mz->lru_size + lru; | |
1122 | *lru_size += nr_pages; | |
1123 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1124 | } |
544122e5 | 1125 | |
2314b42d | 1126 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
c3ac9a8a | 1127 | { |
2314b42d | 1128 | struct mem_cgroup *task_memcg; |
158e0a2d | 1129 | struct task_struct *p; |
ffbdccf5 | 1130 | bool ret; |
4c4a2214 | 1131 | |
158e0a2d | 1132 | p = find_lock_task_mm(task); |
de077d22 | 1133 | if (p) { |
2314b42d | 1134 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1135 | task_unlock(p); |
1136 | } else { | |
1137 | /* | |
1138 | * All threads may have already detached their mm's, but the oom | |
1139 | * killer still needs to detect if they have already been oom | |
1140 | * killed to prevent needlessly killing additional tasks. | |
1141 | */ | |
ffbdccf5 | 1142 | rcu_read_lock(); |
2314b42d JW |
1143 | task_memcg = mem_cgroup_from_task(task); |
1144 | css_get(&task_memcg->css); | |
ffbdccf5 | 1145 | rcu_read_unlock(); |
de077d22 | 1146 | } |
2314b42d JW |
1147 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1148 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1149 | return ret; |
1150 | } | |
1151 | ||
3e32cb2e | 1152 | #define mem_cgroup_from_counter(counter, member) \ |
6d61ef40 BS |
1153 | container_of(counter, struct mem_cgroup, member) |
1154 | ||
19942822 | 1155 | /** |
9d11ea9f | 1156 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1157 | * @memcg: the memory cgroup |
19942822 | 1158 | * |
9d11ea9f | 1159 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1160 | * pages. |
19942822 | 1161 | */ |
c0ff4b85 | 1162 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1163 | { |
3e32cb2e JW |
1164 | unsigned long margin = 0; |
1165 | unsigned long count; | |
1166 | unsigned long limit; | |
9d11ea9f | 1167 | |
3e32cb2e | 1168 | count = page_counter_read(&memcg->memory); |
4db0c3c2 | 1169 | limit = READ_ONCE(memcg->memory.limit); |
3e32cb2e JW |
1170 | if (count < limit) |
1171 | margin = limit - count; | |
1172 | ||
1173 | if (do_swap_account) { | |
1174 | count = page_counter_read(&memcg->memsw); | |
4db0c3c2 | 1175 | limit = READ_ONCE(memcg->memsw.limit); |
3e32cb2e JW |
1176 | if (count <= limit) |
1177 | margin = min(margin, limit - count); | |
1178 | } | |
1179 | ||
1180 | return margin; | |
19942822 JW |
1181 | } |
1182 | ||
32047e2a | 1183 | /* |
bdcbb659 | 1184 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1185 | * |
bdcbb659 QH |
1186 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1187 | * moving cgroups. This is for waiting at high-memory pressure | |
1188 | * caused by "move". | |
32047e2a | 1189 | */ |
c0ff4b85 | 1190 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1191 | { |
2bd9bb20 KH |
1192 | struct mem_cgroup *from; |
1193 | struct mem_cgroup *to; | |
4b534334 | 1194 | bool ret = false; |
2bd9bb20 KH |
1195 | /* |
1196 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1197 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1198 | */ | |
1199 | spin_lock(&mc.lock); | |
1200 | from = mc.from; | |
1201 | to = mc.to; | |
1202 | if (!from) | |
1203 | goto unlock; | |
3e92041d | 1204 | |
2314b42d JW |
1205 | ret = mem_cgroup_is_descendant(from, memcg) || |
1206 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1207 | unlock: |
1208 | spin_unlock(&mc.lock); | |
4b534334 KH |
1209 | return ret; |
1210 | } | |
1211 | ||
c0ff4b85 | 1212 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1213 | { |
1214 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1215 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1216 | DEFINE_WAIT(wait); |
1217 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1218 | /* moving charge context might have finished. */ | |
1219 | if (mc.moving_task) | |
1220 | schedule(); | |
1221 | finish_wait(&mc.waitq, &wait); | |
1222 | return true; | |
1223 | } | |
1224 | } | |
1225 | return false; | |
1226 | } | |
1227 | ||
58cf188e | 1228 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1229 | /** |
58cf188e | 1230 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1231 | * @memcg: The memory cgroup that went over limit |
1232 | * @p: Task that is going to be killed | |
1233 | * | |
1234 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1235 | * enabled | |
1236 | */ | |
1237 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1238 | { | |
e61734c5 | 1239 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
08088cb9 | 1240 | static DEFINE_MUTEX(oom_info_lock); |
58cf188e SZ |
1241 | struct mem_cgroup *iter; |
1242 | unsigned int i; | |
e222432b | 1243 | |
08088cb9 | 1244 | mutex_lock(&oom_info_lock); |
e222432b BS |
1245 | rcu_read_lock(); |
1246 | ||
2415b9f5 BV |
1247 | if (p) { |
1248 | pr_info("Task in "); | |
1249 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1250 | pr_cont(" killed as a result of limit of "); | |
1251 | } else { | |
1252 | pr_info("Memory limit reached of cgroup "); | |
1253 | } | |
1254 | ||
e61734c5 | 1255 | pr_cont_cgroup_path(memcg->css.cgroup); |
0346dadb | 1256 | pr_cont("\n"); |
e222432b | 1257 | |
e222432b BS |
1258 | rcu_read_unlock(); |
1259 | ||
3e32cb2e JW |
1260 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1261 | K((u64)page_counter_read(&memcg->memory)), | |
1262 | K((u64)memcg->memory.limit), memcg->memory.failcnt); | |
1263 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1264 | K((u64)page_counter_read(&memcg->memsw)), | |
1265 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); | |
1266 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1267 | K((u64)page_counter_read(&memcg->kmem)), | |
1268 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); | |
58cf188e SZ |
1269 | |
1270 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1271 | pr_info("Memory cgroup stats for "); |
1272 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1273 | pr_cont(":"); |
1274 | ||
1275 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1276 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1277 | continue; | |
484ebb3b | 1278 | pr_cont(" %s:%luKB", mem_cgroup_stat_names[i], |
58cf188e SZ |
1279 | K(mem_cgroup_read_stat(iter, i))); |
1280 | } | |
1281 | ||
1282 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1283 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1284 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1285 | ||
1286 | pr_cont("\n"); | |
1287 | } | |
08088cb9 | 1288 | mutex_unlock(&oom_info_lock); |
e222432b BS |
1289 | } |
1290 | ||
81d39c20 KH |
1291 | /* |
1292 | * This function returns the number of memcg under hierarchy tree. Returns | |
1293 | * 1(self count) if no children. | |
1294 | */ | |
c0ff4b85 | 1295 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1296 | { |
1297 | int num = 0; | |
7d74b06f KH |
1298 | struct mem_cgroup *iter; |
1299 | ||
c0ff4b85 | 1300 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1301 | num++; |
81d39c20 KH |
1302 | return num; |
1303 | } | |
1304 | ||
a63d83f4 DR |
1305 | /* |
1306 | * Return the memory (and swap, if configured) limit for a memcg. | |
1307 | */ | |
3e32cb2e | 1308 | static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 | 1309 | { |
3e32cb2e | 1310 | unsigned long limit; |
f3e8eb70 | 1311 | |
3e32cb2e | 1312 | limit = memcg->memory.limit; |
9a5a8f19 | 1313 | if (mem_cgroup_swappiness(memcg)) { |
3e32cb2e | 1314 | unsigned long memsw_limit; |
9a5a8f19 | 1315 | |
3e32cb2e JW |
1316 | memsw_limit = memcg->memsw.limit; |
1317 | limit = min(limit + total_swap_pages, memsw_limit); | |
9a5a8f19 | 1318 | } |
9a5a8f19 | 1319 | return limit; |
a63d83f4 DR |
1320 | } |
1321 | ||
0ccab5b1 | 1322 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1323 | int order) |
9cbb78bb | 1324 | { |
6e0fc46d DR |
1325 | struct oom_control oc = { |
1326 | .zonelist = NULL, | |
1327 | .nodemask = NULL, | |
1328 | .gfp_mask = gfp_mask, | |
1329 | .order = order, | |
6e0fc46d | 1330 | }; |
9cbb78bb DR |
1331 | struct mem_cgroup *iter; |
1332 | unsigned long chosen_points = 0; | |
1333 | unsigned long totalpages; | |
1334 | unsigned int points = 0; | |
1335 | struct task_struct *chosen = NULL; | |
1336 | ||
dc56401f JW |
1337 | mutex_lock(&oom_lock); |
1338 | ||
876aafbf | 1339 | /* |
465adcf1 DR |
1340 | * If current has a pending SIGKILL or is exiting, then automatically |
1341 | * select it. The goal is to allow it to allocate so that it may | |
1342 | * quickly exit and free its memory. | |
876aafbf | 1343 | */ |
d003f371 | 1344 | if (fatal_signal_pending(current) || task_will_free_mem(current)) { |
16e95196 | 1345 | mark_oom_victim(current); |
dc56401f | 1346 | goto unlock; |
876aafbf DR |
1347 | } |
1348 | ||
6e0fc46d | 1349 | check_panic_on_oom(&oc, CONSTRAINT_MEMCG, memcg); |
3e32cb2e | 1350 | totalpages = mem_cgroup_get_limit(memcg) ? : 1; |
9cbb78bb | 1351 | for_each_mem_cgroup_tree(iter, memcg) { |
72ec7029 | 1352 | struct css_task_iter it; |
9cbb78bb DR |
1353 | struct task_struct *task; |
1354 | ||
72ec7029 TH |
1355 | css_task_iter_start(&iter->css, &it); |
1356 | while ((task = css_task_iter_next(&it))) { | |
6e0fc46d | 1357 | switch (oom_scan_process_thread(&oc, task, totalpages)) { |
9cbb78bb DR |
1358 | case OOM_SCAN_SELECT: |
1359 | if (chosen) | |
1360 | put_task_struct(chosen); | |
1361 | chosen = task; | |
1362 | chosen_points = ULONG_MAX; | |
1363 | get_task_struct(chosen); | |
1364 | /* fall through */ | |
1365 | case OOM_SCAN_CONTINUE: | |
1366 | continue; | |
1367 | case OOM_SCAN_ABORT: | |
72ec7029 | 1368 | css_task_iter_end(&it); |
9cbb78bb DR |
1369 | mem_cgroup_iter_break(memcg, iter); |
1370 | if (chosen) | |
1371 | put_task_struct(chosen); | |
dc56401f | 1372 | goto unlock; |
9cbb78bb DR |
1373 | case OOM_SCAN_OK: |
1374 | break; | |
1375 | }; | |
1376 | points = oom_badness(task, memcg, NULL, totalpages); | |
d49ad935 DR |
1377 | if (!points || points < chosen_points) |
1378 | continue; | |
1379 | /* Prefer thread group leaders for display purposes */ | |
1380 | if (points == chosen_points && | |
1381 | thread_group_leader(chosen)) | |
1382 | continue; | |
1383 | ||
1384 | if (chosen) | |
1385 | put_task_struct(chosen); | |
1386 | chosen = task; | |
1387 | chosen_points = points; | |
1388 | get_task_struct(chosen); | |
9cbb78bb | 1389 | } |
72ec7029 | 1390 | css_task_iter_end(&it); |
9cbb78bb DR |
1391 | } |
1392 | ||
dc56401f JW |
1393 | if (chosen) { |
1394 | points = chosen_points * 1000 / totalpages; | |
6e0fc46d DR |
1395 | oom_kill_process(&oc, chosen, points, totalpages, memcg, |
1396 | "Memory cgroup out of memory"); | |
dc56401f JW |
1397 | } |
1398 | unlock: | |
1399 | mutex_unlock(&oom_lock); | |
0ccab5b1 | 1400 | return chosen; |
9cbb78bb DR |
1401 | } |
1402 | ||
ae6e71d3 MC |
1403 | #if MAX_NUMNODES > 1 |
1404 | ||
4d0c066d KH |
1405 | /** |
1406 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1407 | * @memcg: the target memcg |
4d0c066d KH |
1408 | * @nid: the node ID to be checked. |
1409 | * @noswap : specify true here if the user wants flle only information. | |
1410 | * | |
1411 | * This function returns whether the specified memcg contains any | |
1412 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1413 | * pages in the node. | |
1414 | */ | |
c0ff4b85 | 1415 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1416 | int nid, bool noswap) |
1417 | { | |
c0ff4b85 | 1418 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1419 | return true; |
1420 | if (noswap || !total_swap_pages) | |
1421 | return false; | |
c0ff4b85 | 1422 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1423 | return true; |
1424 | return false; | |
1425 | ||
1426 | } | |
889976db YH |
1427 | |
1428 | /* | |
1429 | * Always updating the nodemask is not very good - even if we have an empty | |
1430 | * list or the wrong list here, we can start from some node and traverse all | |
1431 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1432 | * | |
1433 | */ | |
c0ff4b85 | 1434 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1435 | { |
1436 | int nid; | |
453a9bf3 KH |
1437 | /* |
1438 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1439 | * pagein/pageout changes since the last update. | |
1440 | */ | |
c0ff4b85 | 1441 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1442 | return; |
c0ff4b85 | 1443 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1444 | return; |
1445 | ||
889976db | 1446 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1447 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1448 | |
31aaea4a | 1449 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1450 | |
c0ff4b85 R |
1451 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1452 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1453 | } |
453a9bf3 | 1454 | |
c0ff4b85 R |
1455 | atomic_set(&memcg->numainfo_events, 0); |
1456 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1457 | } |
1458 | ||
1459 | /* | |
1460 | * Selecting a node where we start reclaim from. Because what we need is just | |
1461 | * reducing usage counter, start from anywhere is O,K. Considering | |
1462 | * memory reclaim from current node, there are pros. and cons. | |
1463 | * | |
1464 | * Freeing memory from current node means freeing memory from a node which | |
1465 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1466 | * hit limits, it will see a contention on a node. But freeing from remote | |
1467 | * node means more costs for memory reclaim because of memory latency. | |
1468 | * | |
1469 | * Now, we use round-robin. Better algorithm is welcomed. | |
1470 | */ | |
c0ff4b85 | 1471 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1472 | { |
1473 | int node; | |
1474 | ||
c0ff4b85 R |
1475 | mem_cgroup_may_update_nodemask(memcg); |
1476 | node = memcg->last_scanned_node; | |
889976db | 1477 | |
c0ff4b85 | 1478 | node = next_node(node, memcg->scan_nodes); |
889976db | 1479 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1480 | node = first_node(memcg->scan_nodes); |
889976db YH |
1481 | /* |
1482 | * We call this when we hit limit, not when pages are added to LRU. | |
1483 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1484 | * memcg is too small and all pages are not on LRU. In that case, | |
1485 | * we use curret node. | |
1486 | */ | |
1487 | if (unlikely(node == MAX_NUMNODES)) | |
1488 | node = numa_node_id(); | |
1489 | ||
c0ff4b85 | 1490 | memcg->last_scanned_node = node; |
889976db YH |
1491 | return node; |
1492 | } | |
889976db | 1493 | #else |
c0ff4b85 | 1494 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1495 | { |
1496 | return 0; | |
1497 | } | |
1498 | #endif | |
1499 | ||
0608f43d AM |
1500 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1501 | struct zone *zone, | |
1502 | gfp_t gfp_mask, | |
1503 | unsigned long *total_scanned) | |
1504 | { | |
1505 | struct mem_cgroup *victim = NULL; | |
1506 | int total = 0; | |
1507 | int loop = 0; | |
1508 | unsigned long excess; | |
1509 | unsigned long nr_scanned; | |
1510 | struct mem_cgroup_reclaim_cookie reclaim = { | |
1511 | .zone = zone, | |
1512 | .priority = 0, | |
1513 | }; | |
1514 | ||
3e32cb2e | 1515 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1516 | |
1517 | while (1) { | |
1518 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1519 | if (!victim) { | |
1520 | loop++; | |
1521 | if (loop >= 2) { | |
1522 | /* | |
1523 | * If we have not been able to reclaim | |
1524 | * anything, it might because there are | |
1525 | * no reclaimable pages under this hierarchy | |
1526 | */ | |
1527 | if (!total) | |
1528 | break; | |
1529 | /* | |
1530 | * We want to do more targeted reclaim. | |
1531 | * excess >> 2 is not to excessive so as to | |
1532 | * reclaim too much, nor too less that we keep | |
1533 | * coming back to reclaim from this cgroup | |
1534 | */ | |
1535 | if (total >= (excess >> 2) || | |
1536 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1537 | break; | |
1538 | } | |
1539 | continue; | |
1540 | } | |
0608f43d AM |
1541 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1542 | zone, &nr_scanned); | |
1543 | *total_scanned += nr_scanned; | |
3e32cb2e | 1544 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1545 | break; |
6d61ef40 | 1546 | } |
0608f43d AM |
1547 | mem_cgroup_iter_break(root_memcg, victim); |
1548 | return total; | |
6d61ef40 BS |
1549 | } |
1550 | ||
0056f4e6 JW |
1551 | #ifdef CONFIG_LOCKDEP |
1552 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1553 | .name = "memcg_oom_lock", | |
1554 | }; | |
1555 | #endif | |
1556 | ||
fb2a6fc5 JW |
1557 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1558 | ||
867578cb KH |
1559 | /* |
1560 | * Check OOM-Killer is already running under our hierarchy. | |
1561 | * If someone is running, return false. | |
1562 | */ | |
fb2a6fc5 | 1563 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1564 | { |
79dfdacc | 1565 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1566 | |
fb2a6fc5 JW |
1567 | spin_lock(&memcg_oom_lock); |
1568 | ||
9f3a0d09 | 1569 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1570 | if (iter->oom_lock) { |
79dfdacc MH |
1571 | /* |
1572 | * this subtree of our hierarchy is already locked | |
1573 | * so we cannot give a lock. | |
1574 | */ | |
79dfdacc | 1575 | failed = iter; |
9f3a0d09 JW |
1576 | mem_cgroup_iter_break(memcg, iter); |
1577 | break; | |
23751be0 JW |
1578 | } else |
1579 | iter->oom_lock = true; | |
7d74b06f | 1580 | } |
867578cb | 1581 | |
fb2a6fc5 JW |
1582 | if (failed) { |
1583 | /* | |
1584 | * OK, we failed to lock the whole subtree so we have | |
1585 | * to clean up what we set up to the failing subtree | |
1586 | */ | |
1587 | for_each_mem_cgroup_tree(iter, memcg) { | |
1588 | if (iter == failed) { | |
1589 | mem_cgroup_iter_break(memcg, iter); | |
1590 | break; | |
1591 | } | |
1592 | iter->oom_lock = false; | |
79dfdacc | 1593 | } |
0056f4e6 JW |
1594 | } else |
1595 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1596 | |
1597 | spin_unlock(&memcg_oom_lock); | |
1598 | ||
1599 | return !failed; | |
a636b327 | 1600 | } |
0b7f569e | 1601 | |
fb2a6fc5 | 1602 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1603 | { |
7d74b06f KH |
1604 | struct mem_cgroup *iter; |
1605 | ||
fb2a6fc5 | 1606 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1607 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1608 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1609 | iter->oom_lock = false; |
fb2a6fc5 | 1610 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1611 | } |
1612 | ||
c0ff4b85 | 1613 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1614 | { |
1615 | struct mem_cgroup *iter; | |
1616 | ||
c2b42d3c | 1617 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1618 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1619 | iter->under_oom++; |
1620 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1621 | } |
1622 | ||
c0ff4b85 | 1623 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1624 | { |
1625 | struct mem_cgroup *iter; | |
1626 | ||
867578cb KH |
1627 | /* |
1628 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1629 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1630 | */ |
c2b42d3c | 1631 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1632 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1633 | if (iter->under_oom > 0) |
1634 | iter->under_oom--; | |
1635 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1636 | } |
1637 | ||
867578cb KH |
1638 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1639 | ||
dc98df5a | 1640 | struct oom_wait_info { |
d79154bb | 1641 | struct mem_cgroup *memcg; |
dc98df5a KH |
1642 | wait_queue_t wait; |
1643 | }; | |
1644 | ||
1645 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1646 | unsigned mode, int sync, void *arg) | |
1647 | { | |
d79154bb HD |
1648 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1649 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1650 | struct oom_wait_info *oom_wait_info; |
1651 | ||
1652 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1653 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1654 | |
2314b42d JW |
1655 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1656 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1657 | return 0; |
dc98df5a KH |
1658 | return autoremove_wake_function(wait, mode, sync, arg); |
1659 | } | |
1660 | ||
c0ff4b85 | 1661 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1662 | { |
c2b42d3c TH |
1663 | /* |
1664 | * For the following lockless ->under_oom test, the only required | |
1665 | * guarantee is that it must see the state asserted by an OOM when | |
1666 | * this function is called as a result of userland actions | |
1667 | * triggered by the notification of the OOM. This is trivially | |
1668 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1669 | * triggering notification. | |
1670 | */ | |
1671 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1672 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1673 | } |
1674 | ||
3812c8c8 | 1675 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 1676 | { |
626ebc41 | 1677 | if (!current->memcg_may_oom) |
3812c8c8 | 1678 | return; |
867578cb | 1679 | /* |
49426420 JW |
1680 | * We are in the middle of the charge context here, so we |
1681 | * don't want to block when potentially sitting on a callstack | |
1682 | * that holds all kinds of filesystem and mm locks. | |
1683 | * | |
1684 | * Also, the caller may handle a failed allocation gracefully | |
1685 | * (like optional page cache readahead) and so an OOM killer | |
1686 | * invocation might not even be necessary. | |
1687 | * | |
1688 | * That's why we don't do anything here except remember the | |
1689 | * OOM context and then deal with it at the end of the page | |
1690 | * fault when the stack is unwound, the locks are released, | |
1691 | * and when we know whether the fault was overall successful. | |
867578cb | 1692 | */ |
49426420 | 1693 | css_get(&memcg->css); |
626ebc41 TH |
1694 | current->memcg_in_oom = memcg; |
1695 | current->memcg_oom_gfp_mask = mask; | |
1696 | current->memcg_oom_order = order; | |
3812c8c8 JW |
1697 | } |
1698 | ||
1699 | /** | |
1700 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1701 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1702 | * |
49426420 JW |
1703 | * This has to be called at the end of a page fault if the memcg OOM |
1704 | * handler was enabled. | |
3812c8c8 | 1705 | * |
49426420 | 1706 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1707 | * sleep on a waitqueue until the userspace task resolves the |
1708 | * situation. Sleeping directly in the charge context with all kinds | |
1709 | * of locks held is not a good idea, instead we remember an OOM state | |
1710 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1711 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1712 | * |
1713 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1714 | * completed, %false otherwise. |
3812c8c8 | 1715 | */ |
49426420 | 1716 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1717 | { |
626ebc41 | 1718 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1719 | struct oom_wait_info owait; |
49426420 | 1720 | bool locked; |
3812c8c8 JW |
1721 | |
1722 | /* OOM is global, do not handle */ | |
3812c8c8 | 1723 | if (!memcg) |
49426420 | 1724 | return false; |
3812c8c8 | 1725 | |
c32b3cbe | 1726 | if (!handle || oom_killer_disabled) |
49426420 | 1727 | goto cleanup; |
3812c8c8 JW |
1728 | |
1729 | owait.memcg = memcg; | |
1730 | owait.wait.flags = 0; | |
1731 | owait.wait.func = memcg_oom_wake_function; | |
1732 | owait.wait.private = current; | |
1733 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 1734 | |
3812c8c8 | 1735 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1736 | mem_cgroup_mark_under_oom(memcg); |
1737 | ||
1738 | locked = mem_cgroup_oom_trylock(memcg); | |
1739 | ||
1740 | if (locked) | |
1741 | mem_cgroup_oom_notify(memcg); | |
1742 | ||
1743 | if (locked && !memcg->oom_kill_disable) { | |
1744 | mem_cgroup_unmark_under_oom(memcg); | |
1745 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1746 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1747 | current->memcg_oom_order); | |
49426420 | 1748 | } else { |
3812c8c8 | 1749 | schedule(); |
49426420 JW |
1750 | mem_cgroup_unmark_under_oom(memcg); |
1751 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1752 | } | |
1753 | ||
1754 | if (locked) { | |
fb2a6fc5 JW |
1755 | mem_cgroup_oom_unlock(memcg); |
1756 | /* | |
1757 | * There is no guarantee that an OOM-lock contender | |
1758 | * sees the wakeups triggered by the OOM kill | |
1759 | * uncharges. Wake any sleepers explicitely. | |
1760 | */ | |
1761 | memcg_oom_recover(memcg); | |
1762 | } | |
49426420 | 1763 | cleanup: |
626ebc41 | 1764 | current->memcg_in_oom = NULL; |
3812c8c8 | 1765 | css_put(&memcg->css); |
867578cb | 1766 | return true; |
0b7f569e KH |
1767 | } |
1768 | ||
d7365e78 JW |
1769 | /** |
1770 | * mem_cgroup_begin_page_stat - begin a page state statistics transaction | |
1771 | * @page: page that is going to change accounted state | |
32047e2a | 1772 | * |
d7365e78 JW |
1773 | * This function must mark the beginning of an accounted page state |
1774 | * change to prevent double accounting when the page is concurrently | |
1775 | * being moved to another memcg: | |
32047e2a | 1776 | * |
6de22619 | 1777 | * memcg = mem_cgroup_begin_page_stat(page); |
d7365e78 JW |
1778 | * if (TestClearPageState(page)) |
1779 | * mem_cgroup_update_page_stat(memcg, state, -1); | |
6de22619 | 1780 | * mem_cgroup_end_page_stat(memcg); |
d69b042f | 1781 | */ |
6de22619 | 1782 | struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page) |
89c06bd5 KH |
1783 | { |
1784 | struct mem_cgroup *memcg; | |
6de22619 | 1785 | unsigned long flags; |
89c06bd5 | 1786 | |
6de22619 JW |
1787 | /* |
1788 | * The RCU lock is held throughout the transaction. The fast | |
1789 | * path can get away without acquiring the memcg->move_lock | |
1790 | * because page moving starts with an RCU grace period. | |
1791 | * | |
1792 | * The RCU lock also protects the memcg from being freed when | |
1793 | * the page state that is going to change is the only thing | |
1794 | * preventing the page from being uncharged. | |
1795 | * E.g. end-writeback clearing PageWriteback(), which allows | |
1796 | * migration to go ahead and uncharge the page before the | |
1797 | * account transaction might be complete. | |
1798 | */ | |
d7365e78 JW |
1799 | rcu_read_lock(); |
1800 | ||
1801 | if (mem_cgroup_disabled()) | |
1802 | return NULL; | |
89c06bd5 | 1803 | again: |
1306a85a | 1804 | memcg = page->mem_cgroup; |
29833315 | 1805 | if (unlikely(!memcg)) |
d7365e78 JW |
1806 | return NULL; |
1807 | ||
bdcbb659 | 1808 | if (atomic_read(&memcg->moving_account) <= 0) |
d7365e78 | 1809 | return memcg; |
89c06bd5 | 1810 | |
6de22619 | 1811 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 1812 | if (memcg != page->mem_cgroup) { |
6de22619 | 1813 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
1814 | goto again; |
1815 | } | |
6de22619 JW |
1816 | |
1817 | /* | |
1818 | * When charge migration first begins, we can have locked and | |
1819 | * unlocked page stat updates happening concurrently. Track | |
1820 | * the task who has the lock for mem_cgroup_end_page_stat(). | |
1821 | */ | |
1822 | memcg->move_lock_task = current; | |
1823 | memcg->move_lock_flags = flags; | |
d7365e78 JW |
1824 | |
1825 | return memcg; | |
89c06bd5 | 1826 | } |
c4843a75 | 1827 | EXPORT_SYMBOL(mem_cgroup_begin_page_stat); |
89c06bd5 | 1828 | |
d7365e78 JW |
1829 | /** |
1830 | * mem_cgroup_end_page_stat - finish a page state statistics transaction | |
1831 | * @memcg: the memcg that was accounted against | |
d7365e78 | 1832 | */ |
6de22619 | 1833 | void mem_cgroup_end_page_stat(struct mem_cgroup *memcg) |
89c06bd5 | 1834 | { |
6de22619 JW |
1835 | if (memcg && memcg->move_lock_task == current) { |
1836 | unsigned long flags = memcg->move_lock_flags; | |
1837 | ||
1838 | memcg->move_lock_task = NULL; | |
1839 | memcg->move_lock_flags = 0; | |
1840 | ||
1841 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
1842 | } | |
89c06bd5 | 1843 | |
d7365e78 | 1844 | rcu_read_unlock(); |
89c06bd5 | 1845 | } |
c4843a75 | 1846 | EXPORT_SYMBOL(mem_cgroup_end_page_stat); |
89c06bd5 | 1847 | |
cdec2e42 KH |
1848 | /* |
1849 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1850 | * TODO: maybe necessary to use big numbers in big irons. | |
1851 | */ | |
7ec99d62 | 1852 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
1853 | struct memcg_stock_pcp { |
1854 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 1855 | unsigned int nr_pages; |
cdec2e42 | 1856 | struct work_struct work; |
26fe6168 | 1857 | unsigned long flags; |
a0db00fc | 1858 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
1859 | }; |
1860 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 1861 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 1862 | |
a0956d54 SS |
1863 | /** |
1864 | * consume_stock: Try to consume stocked charge on this cpu. | |
1865 | * @memcg: memcg to consume from. | |
1866 | * @nr_pages: how many pages to charge. | |
1867 | * | |
1868 | * The charges will only happen if @memcg matches the current cpu's memcg | |
1869 | * stock, and at least @nr_pages are available in that stock. Failure to | |
1870 | * service an allocation will refill the stock. | |
1871 | * | |
1872 | * returns true if successful, false otherwise. | |
cdec2e42 | 1873 | */ |
a0956d54 | 1874 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1875 | { |
1876 | struct memcg_stock_pcp *stock; | |
3e32cb2e | 1877 | bool ret = false; |
cdec2e42 | 1878 | |
a0956d54 | 1879 | if (nr_pages > CHARGE_BATCH) |
3e32cb2e | 1880 | return ret; |
a0956d54 | 1881 | |
cdec2e42 | 1882 | stock = &get_cpu_var(memcg_stock); |
3e32cb2e | 1883 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 1884 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
1885 | ret = true; |
1886 | } | |
cdec2e42 KH |
1887 | put_cpu_var(memcg_stock); |
1888 | return ret; | |
1889 | } | |
1890 | ||
1891 | /* | |
3e32cb2e | 1892 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
1893 | */ |
1894 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1895 | { | |
1896 | struct mem_cgroup *old = stock->cached; | |
1897 | ||
11c9ea4e | 1898 | if (stock->nr_pages) { |
3e32cb2e | 1899 | page_counter_uncharge(&old->memory, stock->nr_pages); |
cdec2e42 | 1900 | if (do_swap_account) |
3e32cb2e | 1901 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 1902 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 1903 | stock->nr_pages = 0; |
cdec2e42 KH |
1904 | } |
1905 | stock->cached = NULL; | |
cdec2e42 KH |
1906 | } |
1907 | ||
1908 | /* | |
1909 | * This must be called under preempt disabled or must be called by | |
1910 | * a thread which is pinned to local cpu. | |
1911 | */ | |
1912 | static void drain_local_stock(struct work_struct *dummy) | |
1913 | { | |
7c8e0181 | 1914 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
cdec2e42 | 1915 | drain_stock(stock); |
26fe6168 | 1916 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
1917 | } |
1918 | ||
1919 | /* | |
3e32cb2e | 1920 | * Cache charges(val) to local per_cpu area. |
320cc51d | 1921 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 1922 | */ |
c0ff4b85 | 1923 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1924 | { |
1925 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
1926 | ||
c0ff4b85 | 1927 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 1928 | drain_stock(stock); |
c0ff4b85 | 1929 | stock->cached = memcg; |
cdec2e42 | 1930 | } |
11c9ea4e | 1931 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
1932 | put_cpu_var(memcg_stock); |
1933 | } | |
1934 | ||
1935 | /* | |
c0ff4b85 | 1936 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 1937 | * of the hierarchy under it. |
cdec2e42 | 1938 | */ |
6d3d6aa2 | 1939 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 1940 | { |
26fe6168 | 1941 | int cpu, curcpu; |
d38144b7 | 1942 | |
6d3d6aa2 JW |
1943 | /* If someone's already draining, avoid adding running more workers. */ |
1944 | if (!mutex_trylock(&percpu_charge_mutex)) | |
1945 | return; | |
cdec2e42 | 1946 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 1947 | get_online_cpus(); |
5af12d0e | 1948 | curcpu = get_cpu(); |
cdec2e42 KH |
1949 | for_each_online_cpu(cpu) { |
1950 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 1951 | struct mem_cgroup *memcg; |
26fe6168 | 1952 | |
c0ff4b85 R |
1953 | memcg = stock->cached; |
1954 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 1955 | continue; |
2314b42d | 1956 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) |
3e92041d | 1957 | continue; |
d1a05b69 MH |
1958 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
1959 | if (cpu == curcpu) | |
1960 | drain_local_stock(&stock->work); | |
1961 | else | |
1962 | schedule_work_on(cpu, &stock->work); | |
1963 | } | |
cdec2e42 | 1964 | } |
5af12d0e | 1965 | put_cpu(); |
f894ffa8 | 1966 | put_online_cpus(); |
9f50fad6 | 1967 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
1968 | } |
1969 | ||
0db0628d | 1970 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
1971 | unsigned long action, |
1972 | void *hcpu) | |
1973 | { | |
1974 | int cpu = (unsigned long)hcpu; | |
1975 | struct memcg_stock_pcp *stock; | |
1976 | ||
619d094b | 1977 | if (action == CPU_ONLINE) |
1489ebad | 1978 | return NOTIFY_OK; |
1489ebad | 1979 | |
d833049b | 1980 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 1981 | return NOTIFY_OK; |
711d3d2c | 1982 | |
cdec2e42 KH |
1983 | stock = &per_cpu(memcg_stock, cpu); |
1984 | drain_stock(stock); | |
1985 | return NOTIFY_OK; | |
1986 | } | |
1987 | ||
b23afb93 TH |
1988 | /* |
1989 | * Scheduled by try_charge() to be executed from the userland return path | |
1990 | * and reclaims memory over the high limit. | |
1991 | */ | |
1992 | void mem_cgroup_handle_over_high(void) | |
1993 | { | |
1994 | unsigned int nr_pages = current->memcg_nr_pages_over_high; | |
1995 | struct mem_cgroup *memcg, *pos; | |
1996 | ||
1997 | if (likely(!nr_pages)) | |
1998 | return; | |
1999 | ||
2000 | pos = memcg = get_mem_cgroup_from_mm(current->mm); | |
2001 | ||
2002 | do { | |
2003 | if (page_counter_read(&pos->memory) <= pos->high) | |
2004 | continue; | |
2005 | mem_cgroup_events(pos, MEMCG_HIGH, 1); | |
2006 | try_to_free_mem_cgroup_pages(pos, nr_pages, GFP_KERNEL, true); | |
2007 | } while ((pos = parent_mem_cgroup(pos))); | |
2008 | ||
2009 | css_put(&memcg->css); | |
2010 | current->memcg_nr_pages_over_high = 0; | |
2011 | } | |
2012 | ||
00501b53 JW |
2013 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2014 | unsigned int nr_pages) | |
8a9f3ccd | 2015 | { |
7ec99d62 | 2016 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 2017 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 2018 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2019 | struct page_counter *counter; |
6539cc05 | 2020 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2021 | bool may_swap = true; |
2022 | bool drained = false; | |
a636b327 | 2023 | |
ce00a967 | 2024 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 2025 | return 0; |
6539cc05 | 2026 | retry: |
b6b6cc72 | 2027 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2028 | return 0; |
8a9f3ccd | 2029 | |
3fbe7244 | 2030 | if (!do_swap_account || |
6071ca52 JW |
2031 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2032 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2033 | goto done_restock; |
3fbe7244 | 2034 | if (do_swap_account) |
3e32cb2e JW |
2035 | page_counter_uncharge(&memcg->memsw, batch); |
2036 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2037 | } else { |
3e32cb2e | 2038 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2039 | may_swap = false; |
3fbe7244 | 2040 | } |
7a81b88c | 2041 | |
6539cc05 JW |
2042 | if (batch > nr_pages) { |
2043 | batch = nr_pages; | |
2044 | goto retry; | |
2045 | } | |
6d61ef40 | 2046 | |
06b078fc JW |
2047 | /* |
2048 | * Unlike in global OOM situations, memcg is not in a physical | |
2049 | * memory shortage. Allow dying and OOM-killed tasks to | |
2050 | * bypass the last charges so that they can exit quickly and | |
2051 | * free their memory. | |
2052 | */ | |
2053 | if (unlikely(test_thread_flag(TIF_MEMDIE) || | |
2054 | fatal_signal_pending(current) || | |
2055 | current->flags & PF_EXITING)) | |
10d53c74 | 2056 | goto force; |
06b078fc JW |
2057 | |
2058 | if (unlikely(task_in_memcg_oom(current))) | |
2059 | goto nomem; | |
2060 | ||
d0164adc | 2061 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2062 | goto nomem; |
4b534334 | 2063 | |
241994ed JW |
2064 | mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1); |
2065 | ||
b70a2a21 JW |
2066 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2067 | gfp_mask, may_swap); | |
6539cc05 | 2068 | |
61e02c74 | 2069 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2070 | goto retry; |
28c34c29 | 2071 | |
b70a2a21 | 2072 | if (!drained) { |
6d3d6aa2 | 2073 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2074 | drained = true; |
2075 | goto retry; | |
2076 | } | |
2077 | ||
28c34c29 JW |
2078 | if (gfp_mask & __GFP_NORETRY) |
2079 | goto nomem; | |
6539cc05 JW |
2080 | /* |
2081 | * Even though the limit is exceeded at this point, reclaim | |
2082 | * may have been able to free some pages. Retry the charge | |
2083 | * before killing the task. | |
2084 | * | |
2085 | * Only for regular pages, though: huge pages are rather | |
2086 | * unlikely to succeed so close to the limit, and we fall back | |
2087 | * to regular pages anyway in case of failure. | |
2088 | */ | |
61e02c74 | 2089 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2090 | goto retry; |
2091 | /* | |
2092 | * At task move, charge accounts can be doubly counted. So, it's | |
2093 | * better to wait until the end of task_move if something is going on. | |
2094 | */ | |
2095 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2096 | goto retry; | |
2097 | ||
9b130619 JW |
2098 | if (nr_retries--) |
2099 | goto retry; | |
2100 | ||
06b078fc | 2101 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2102 | goto force; |
06b078fc | 2103 | |
6539cc05 | 2104 | if (fatal_signal_pending(current)) |
10d53c74 | 2105 | goto force; |
6539cc05 | 2106 | |
241994ed JW |
2107 | mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1); |
2108 | ||
3608de07 JM |
2109 | mem_cgroup_oom(mem_over_limit, gfp_mask, |
2110 | get_order(nr_pages * PAGE_SIZE)); | |
7a81b88c | 2111 | nomem: |
6d1fdc48 | 2112 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2113 | return -ENOMEM; |
10d53c74 TH |
2114 | force: |
2115 | /* | |
2116 | * The allocation either can't fail or will lead to more memory | |
2117 | * being freed very soon. Allow memory usage go over the limit | |
2118 | * temporarily by force charging it. | |
2119 | */ | |
2120 | page_counter_charge(&memcg->memory, nr_pages); | |
2121 | if (do_swap_account) | |
2122 | page_counter_charge(&memcg->memsw, nr_pages); | |
2123 | css_get_many(&memcg->css, nr_pages); | |
2124 | ||
2125 | return 0; | |
6539cc05 JW |
2126 | |
2127 | done_restock: | |
e8ea14cc | 2128 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2129 | if (batch > nr_pages) |
2130 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2131 | |
241994ed | 2132 | /* |
b23afb93 TH |
2133 | * If the hierarchy is above the normal consumption range, schedule |
2134 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2135 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2136 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2137 | * not recorded as it most likely matches current's and won't | |
2138 | * change in the meantime. As high limit is checked again before | |
2139 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2140 | */ |
2141 | do { | |
b23afb93 | 2142 | if (page_counter_read(&memcg->memory) > memcg->high) { |
9516a18a | 2143 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2144 | set_notify_resume(current); |
2145 | break; | |
2146 | } | |
241994ed | 2147 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2148 | |
2149 | return 0; | |
7a81b88c | 2150 | } |
8a9f3ccd | 2151 | |
00501b53 | 2152 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2153 | { |
ce00a967 JW |
2154 | if (mem_cgroup_is_root(memcg)) |
2155 | return; | |
2156 | ||
3e32cb2e | 2157 | page_counter_uncharge(&memcg->memory, nr_pages); |
05b84301 | 2158 | if (do_swap_account) |
3e32cb2e | 2159 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2160 | |
e8ea14cc | 2161 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2162 | } |
2163 | ||
0a31bc97 JW |
2164 | static void lock_page_lru(struct page *page, int *isolated) |
2165 | { | |
2166 | struct zone *zone = page_zone(page); | |
2167 | ||
2168 | spin_lock_irq(&zone->lru_lock); | |
2169 | if (PageLRU(page)) { | |
2170 | struct lruvec *lruvec; | |
2171 | ||
2172 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2173 | ClearPageLRU(page); | |
2174 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2175 | *isolated = 1; | |
2176 | } else | |
2177 | *isolated = 0; | |
2178 | } | |
2179 | ||
2180 | static void unlock_page_lru(struct page *page, int isolated) | |
2181 | { | |
2182 | struct zone *zone = page_zone(page); | |
2183 | ||
2184 | if (isolated) { | |
2185 | struct lruvec *lruvec; | |
2186 | ||
2187 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2188 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
2189 | SetPageLRU(page); | |
2190 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2191 | } | |
2192 | spin_unlock_irq(&zone->lru_lock); | |
2193 | } | |
2194 | ||
00501b53 | 2195 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2196 | bool lrucare) |
7a81b88c | 2197 | { |
0a31bc97 | 2198 | int isolated; |
9ce70c02 | 2199 | |
1306a85a | 2200 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2201 | |
2202 | /* | |
2203 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2204 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2205 | */ | |
0a31bc97 JW |
2206 | if (lrucare) |
2207 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2208 | |
0a31bc97 JW |
2209 | /* |
2210 | * Nobody should be changing or seriously looking at | |
1306a85a | 2211 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2212 | * |
2213 | * - the page is uncharged | |
2214 | * | |
2215 | * - the page is off-LRU | |
2216 | * | |
2217 | * - an anonymous fault has exclusive page access, except for | |
2218 | * a locked page table | |
2219 | * | |
2220 | * - a page cache insertion, a swapin fault, or a migration | |
2221 | * have the page locked | |
2222 | */ | |
1306a85a | 2223 | page->mem_cgroup = memcg; |
9ce70c02 | 2224 | |
0a31bc97 JW |
2225 | if (lrucare) |
2226 | unlock_page_lru(page, isolated); | |
7a81b88c | 2227 | } |
66e1707b | 2228 | |
7ae1e1d0 | 2229 | #ifdef CONFIG_MEMCG_KMEM |
f3bb3043 | 2230 | static int memcg_alloc_cache_id(void) |
55007d84 | 2231 | { |
f3bb3043 VD |
2232 | int id, size; |
2233 | int err; | |
2234 | ||
dbcf73e2 | 2235 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2236 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2237 | if (id < 0) | |
2238 | return id; | |
55007d84 | 2239 | |
dbcf73e2 | 2240 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2241 | return id; |
2242 | ||
2243 | /* | |
2244 | * There's no space for the new id in memcg_caches arrays, | |
2245 | * so we have to grow them. | |
2246 | */ | |
05257a1a | 2247 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2248 | |
2249 | size = 2 * (id + 1); | |
55007d84 GC |
2250 | if (size < MEMCG_CACHES_MIN_SIZE) |
2251 | size = MEMCG_CACHES_MIN_SIZE; | |
2252 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2253 | size = MEMCG_CACHES_MAX_SIZE; | |
2254 | ||
f3bb3043 | 2255 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2256 | if (!err) |
2257 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2258 | if (!err) |
2259 | memcg_nr_cache_ids = size; | |
2260 | ||
2261 | up_write(&memcg_cache_ids_sem); | |
2262 | ||
f3bb3043 | 2263 | if (err) { |
dbcf73e2 | 2264 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2265 | return err; |
2266 | } | |
2267 | return id; | |
2268 | } | |
2269 | ||
2270 | static void memcg_free_cache_id(int id) | |
2271 | { | |
dbcf73e2 | 2272 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2273 | } |
2274 | ||
d5b3cf71 | 2275 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2276 | struct mem_cgroup *memcg; |
2277 | struct kmem_cache *cachep; | |
2278 | struct work_struct work; | |
2279 | }; | |
2280 | ||
d5b3cf71 | 2281 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2282 | { |
d5b3cf71 VD |
2283 | struct memcg_kmem_cache_create_work *cw = |
2284 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2285 | struct mem_cgroup *memcg = cw->memcg; |
2286 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2287 | |
d5b3cf71 | 2288 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2289 | |
5722d094 | 2290 | css_put(&memcg->css); |
d7f25f8a GC |
2291 | kfree(cw); |
2292 | } | |
2293 | ||
2294 | /* | |
2295 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2296 | */ |
d5b3cf71 VD |
2297 | static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2298 | struct kmem_cache *cachep) | |
d7f25f8a | 2299 | { |
d5b3cf71 | 2300 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2301 | |
776ed0f0 | 2302 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
8135be5a | 2303 | if (!cw) |
d7f25f8a | 2304 | return; |
8135be5a VD |
2305 | |
2306 | css_get(&memcg->css); | |
d7f25f8a GC |
2307 | |
2308 | cw->memcg = memcg; | |
2309 | cw->cachep = cachep; | |
d5b3cf71 | 2310 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2311 | |
d7f25f8a GC |
2312 | schedule_work(&cw->work); |
2313 | } | |
2314 | ||
d5b3cf71 VD |
2315 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2316 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
2317 | { |
2318 | /* | |
2319 | * We need to stop accounting when we kmalloc, because if the | |
2320 | * corresponding kmalloc cache is not yet created, the first allocation | |
d5b3cf71 | 2321 | * in __memcg_schedule_kmem_cache_create will recurse. |
0e9d92f2 GC |
2322 | * |
2323 | * However, it is better to enclose the whole function. Depending on | |
2324 | * the debugging options enabled, INIT_WORK(), for instance, can | |
2325 | * trigger an allocation. This too, will make us recurse. Because at | |
2326 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
2327 | * the safest choice is to do it like this, wrapping the whole function. | |
2328 | */ | |
6f185c29 | 2329 | current->memcg_kmem_skip_account = 1; |
d5b3cf71 | 2330 | __memcg_schedule_kmem_cache_create(memcg, cachep); |
6f185c29 | 2331 | current->memcg_kmem_skip_account = 0; |
0e9d92f2 | 2332 | } |
c67a8a68 | 2333 | |
d7f25f8a GC |
2334 | /* |
2335 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
2336 | * We try to use the current memcg's version of the cache. | |
2337 | * | |
2338 | * If the cache does not exist yet, if we are the first user of it, | |
2339 | * we either create it immediately, if possible, or create it asynchronously | |
2340 | * in a workqueue. | |
2341 | * In the latter case, we will let the current allocation go through with | |
2342 | * the original cache. | |
2343 | * | |
2344 | * Can't be called in interrupt context or from kernel threads. | |
2345 | * This function needs to be called with rcu_read_lock() held. | |
2346 | */ | |
056b7cce | 2347 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep) |
d7f25f8a GC |
2348 | { |
2349 | struct mem_cgroup *memcg; | |
959c8963 | 2350 | struct kmem_cache *memcg_cachep; |
2a4db7eb | 2351 | int kmemcg_id; |
d7f25f8a | 2352 | |
f7ce3190 | 2353 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2354 | |
9d100c5e | 2355 | if (current->memcg_kmem_skip_account) |
0e9d92f2 GC |
2356 | return cachep; |
2357 | ||
8135be5a | 2358 | memcg = get_mem_cgroup_from_mm(current->mm); |
4db0c3c2 | 2359 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
2a4db7eb | 2360 | if (kmemcg_id < 0) |
ca0dde97 | 2361 | goto out; |
d7f25f8a | 2362 | |
2a4db7eb | 2363 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
8135be5a VD |
2364 | if (likely(memcg_cachep)) |
2365 | return memcg_cachep; | |
ca0dde97 LZ |
2366 | |
2367 | /* | |
2368 | * If we are in a safe context (can wait, and not in interrupt | |
2369 | * context), we could be be predictable and return right away. | |
2370 | * This would guarantee that the allocation being performed | |
2371 | * already belongs in the new cache. | |
2372 | * | |
2373 | * However, there are some clashes that can arrive from locking. | |
2374 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2375 | * memcg_create_kmem_cache, this means no further allocation |
2376 | * could happen with the slab_mutex held. So it's better to | |
2377 | * defer everything. | |
ca0dde97 | 2378 | */ |
d5b3cf71 | 2379 | memcg_schedule_kmem_cache_create(memcg, cachep); |
ca0dde97 | 2380 | out: |
8135be5a | 2381 | css_put(&memcg->css); |
ca0dde97 | 2382 | return cachep; |
d7f25f8a | 2383 | } |
d7f25f8a | 2384 | |
8135be5a VD |
2385 | void __memcg_kmem_put_cache(struct kmem_cache *cachep) |
2386 | { | |
2387 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2388 | css_put(&cachep->memcg_params.memcg->css); |
8135be5a VD |
2389 | } |
2390 | ||
f3ccb2c4 VD |
2391 | int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, |
2392 | struct mem_cgroup *memcg) | |
7ae1e1d0 | 2393 | { |
f3ccb2c4 VD |
2394 | unsigned int nr_pages = 1 << order; |
2395 | struct page_counter *counter; | |
7ae1e1d0 GC |
2396 | int ret; |
2397 | ||
f3ccb2c4 | 2398 | if (!memcg_kmem_is_active(memcg)) |
d05e83a6 | 2399 | return 0; |
6d42c232 | 2400 | |
6071ca52 JW |
2401 | if (!page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) |
2402 | return -ENOMEM; | |
7ae1e1d0 | 2403 | |
f3ccb2c4 VD |
2404 | ret = try_charge(memcg, gfp, nr_pages); |
2405 | if (ret) { | |
2406 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
2407 | return ret; | |
7ae1e1d0 GC |
2408 | } |
2409 | ||
f3ccb2c4 | 2410 | page->mem_cgroup = memcg; |
7ae1e1d0 | 2411 | |
f3ccb2c4 | 2412 | return 0; |
7ae1e1d0 GC |
2413 | } |
2414 | ||
f3ccb2c4 | 2415 | int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 2416 | { |
f3ccb2c4 VD |
2417 | struct mem_cgroup *memcg; |
2418 | int ret; | |
7ae1e1d0 | 2419 | |
f3ccb2c4 VD |
2420 | memcg = get_mem_cgroup_from_mm(current->mm); |
2421 | ret = __memcg_kmem_charge_memcg(page, gfp, order, memcg); | |
7ae1e1d0 | 2422 | css_put(&memcg->css); |
d05e83a6 | 2423 | return ret; |
7ae1e1d0 GC |
2424 | } |
2425 | ||
d05e83a6 | 2426 | void __memcg_kmem_uncharge(struct page *page, int order) |
7ae1e1d0 | 2427 | { |
1306a85a | 2428 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 2429 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 2430 | |
7ae1e1d0 GC |
2431 | if (!memcg) |
2432 | return; | |
2433 | ||
309381fe | 2434 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 2435 | |
f3ccb2c4 VD |
2436 | page_counter_uncharge(&memcg->kmem, nr_pages); |
2437 | page_counter_uncharge(&memcg->memory, nr_pages); | |
2438 | if (do_swap_account) | |
2439 | page_counter_uncharge(&memcg->memsw, nr_pages); | |
60d3fd32 | 2440 | |
1306a85a | 2441 | page->mem_cgroup = NULL; |
f3ccb2c4 | 2442 | css_put_many(&memcg->css, nr_pages); |
60d3fd32 | 2443 | } |
7ae1e1d0 GC |
2444 | #endif /* CONFIG_MEMCG_KMEM */ |
2445 | ||
ca3e0214 KH |
2446 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2447 | ||
ca3e0214 KH |
2448 | /* |
2449 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
2450 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
2451 | * charge/uncharge will be never happen and move_account() is done under | |
2452 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 2453 | */ |
e94c8a9c | 2454 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2455 | { |
e94c8a9c | 2456 | int i; |
ca3e0214 | 2457 | |
3d37c4a9 KH |
2458 | if (mem_cgroup_disabled()) |
2459 | return; | |
b070e65c | 2460 | |
29833315 | 2461 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2462 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2463 | |
1306a85a | 2464 | __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
b070e65c | 2465 | HPAGE_PMD_NR); |
ca3e0214 | 2466 | } |
12d27107 | 2467 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2468 | |
c255a458 | 2469 | #ifdef CONFIG_MEMCG_SWAP |
0a31bc97 JW |
2470 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
2471 | bool charge) | |
d13d1443 | 2472 | { |
0a31bc97 JW |
2473 | int val = (charge) ? 1 : -1; |
2474 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); | |
d13d1443 | 2475 | } |
02491447 DN |
2476 | |
2477 | /** | |
2478 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2479 | * @entry: swap entry to be moved | |
2480 | * @from: mem_cgroup which the entry is moved from | |
2481 | * @to: mem_cgroup which the entry is moved to | |
2482 | * | |
2483 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2484 | * as the mem_cgroup's id of @from. | |
2485 | * | |
2486 | * Returns 0 on success, -EINVAL on failure. | |
2487 | * | |
3e32cb2e | 2488 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2489 | * both res and memsw, and called css_get(). |
2490 | */ | |
2491 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2492 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2493 | { |
2494 | unsigned short old_id, new_id; | |
2495 | ||
34c00c31 LZ |
2496 | old_id = mem_cgroup_id(from); |
2497 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2498 | |
2499 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 2500 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 2501 | mem_cgroup_swap_statistics(to, true); |
02491447 DN |
2502 | return 0; |
2503 | } | |
2504 | return -EINVAL; | |
2505 | } | |
2506 | #else | |
2507 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2508 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2509 | { |
2510 | return -EINVAL; | |
2511 | } | |
8c7c6e34 | 2512 | #endif |
d13d1443 | 2513 | |
3e32cb2e | 2514 | static DEFINE_MUTEX(memcg_limit_mutex); |
f212ad7c | 2515 | |
d38d2a75 | 2516 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2517 | unsigned long limit) |
628f4235 | 2518 | { |
3e32cb2e JW |
2519 | unsigned long curusage; |
2520 | unsigned long oldusage; | |
2521 | bool enlarge = false; | |
81d39c20 | 2522 | int retry_count; |
3e32cb2e | 2523 | int ret; |
81d39c20 KH |
2524 | |
2525 | /* | |
2526 | * For keeping hierarchical_reclaim simple, how long we should retry | |
2527 | * is depends on callers. We set our retry-count to be function | |
2528 | * of # of children which we should visit in this loop. | |
2529 | */ | |
3e32cb2e JW |
2530 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2531 | mem_cgroup_count_children(memcg); | |
81d39c20 | 2532 | |
3e32cb2e | 2533 | oldusage = page_counter_read(&memcg->memory); |
628f4235 | 2534 | |
3e32cb2e | 2535 | do { |
628f4235 KH |
2536 | if (signal_pending(current)) { |
2537 | ret = -EINTR; | |
2538 | break; | |
2539 | } | |
3e32cb2e JW |
2540 | |
2541 | mutex_lock(&memcg_limit_mutex); | |
2542 | if (limit > memcg->memsw.limit) { | |
2543 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2544 | ret = -EINVAL; |
628f4235 KH |
2545 | break; |
2546 | } | |
3e32cb2e JW |
2547 | if (limit > memcg->memory.limit) |
2548 | enlarge = true; | |
2549 | ret = page_counter_limit(&memcg->memory, limit); | |
2550 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2551 | |
2552 | if (!ret) | |
2553 | break; | |
2554 | ||
b70a2a21 JW |
2555 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
2556 | ||
3e32cb2e | 2557 | curusage = page_counter_read(&memcg->memory); |
81d39c20 | 2558 | /* Usage is reduced ? */ |
f894ffa8 | 2559 | if (curusage >= oldusage) |
81d39c20 KH |
2560 | retry_count--; |
2561 | else | |
2562 | oldusage = curusage; | |
3e32cb2e JW |
2563 | } while (retry_count); |
2564 | ||
3c11ecf4 KH |
2565 | if (!ret && enlarge) |
2566 | memcg_oom_recover(memcg); | |
14797e23 | 2567 | |
8c7c6e34 KH |
2568 | return ret; |
2569 | } | |
2570 | ||
338c8431 | 2571 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2572 | unsigned long limit) |
8c7c6e34 | 2573 | { |
3e32cb2e JW |
2574 | unsigned long curusage; |
2575 | unsigned long oldusage; | |
2576 | bool enlarge = false; | |
81d39c20 | 2577 | int retry_count; |
3e32cb2e | 2578 | int ret; |
8c7c6e34 | 2579 | |
81d39c20 | 2580 | /* see mem_cgroup_resize_res_limit */ |
3e32cb2e JW |
2581 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2582 | mem_cgroup_count_children(memcg); | |
2583 | ||
2584 | oldusage = page_counter_read(&memcg->memsw); | |
2585 | ||
2586 | do { | |
8c7c6e34 KH |
2587 | if (signal_pending(current)) { |
2588 | ret = -EINTR; | |
2589 | break; | |
2590 | } | |
3e32cb2e JW |
2591 | |
2592 | mutex_lock(&memcg_limit_mutex); | |
2593 | if (limit < memcg->memory.limit) { | |
2594 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2595 | ret = -EINVAL; |
8c7c6e34 KH |
2596 | break; |
2597 | } | |
3e32cb2e JW |
2598 | if (limit > memcg->memsw.limit) |
2599 | enlarge = true; | |
2600 | ret = page_counter_limit(&memcg->memsw, limit); | |
2601 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2602 | |
2603 | if (!ret) | |
2604 | break; | |
2605 | ||
b70a2a21 JW |
2606 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
2607 | ||
3e32cb2e | 2608 | curusage = page_counter_read(&memcg->memsw); |
81d39c20 | 2609 | /* Usage is reduced ? */ |
8c7c6e34 | 2610 | if (curusage >= oldusage) |
628f4235 | 2611 | retry_count--; |
81d39c20 KH |
2612 | else |
2613 | oldusage = curusage; | |
3e32cb2e JW |
2614 | } while (retry_count); |
2615 | ||
3c11ecf4 KH |
2616 | if (!ret && enlarge) |
2617 | memcg_oom_recover(memcg); | |
3e32cb2e | 2618 | |
628f4235 KH |
2619 | return ret; |
2620 | } | |
2621 | ||
0608f43d AM |
2622 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
2623 | gfp_t gfp_mask, | |
2624 | unsigned long *total_scanned) | |
2625 | { | |
2626 | unsigned long nr_reclaimed = 0; | |
2627 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
2628 | unsigned long reclaimed; | |
2629 | int loop = 0; | |
2630 | struct mem_cgroup_tree_per_zone *mctz; | |
3e32cb2e | 2631 | unsigned long excess; |
0608f43d AM |
2632 | unsigned long nr_scanned; |
2633 | ||
2634 | if (order > 0) | |
2635 | return 0; | |
2636 | ||
2637 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
2638 | /* | |
2639 | * This loop can run a while, specially if mem_cgroup's continuously | |
2640 | * keep exceeding their soft limit and putting the system under | |
2641 | * pressure | |
2642 | */ | |
2643 | do { | |
2644 | if (next_mz) | |
2645 | mz = next_mz; | |
2646 | else | |
2647 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2648 | if (!mz) | |
2649 | break; | |
2650 | ||
2651 | nr_scanned = 0; | |
2652 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
2653 | gfp_mask, &nr_scanned); | |
2654 | nr_reclaimed += reclaimed; | |
2655 | *total_scanned += nr_scanned; | |
0a31bc97 | 2656 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 2657 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
2658 | |
2659 | /* | |
2660 | * If we failed to reclaim anything from this memory cgroup | |
2661 | * it is time to move on to the next cgroup | |
2662 | */ | |
2663 | next_mz = NULL; | |
bc2f2e7f VD |
2664 | if (!reclaimed) |
2665 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
2666 | ||
3e32cb2e | 2667 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
2668 | /* |
2669 | * One school of thought says that we should not add | |
2670 | * back the node to the tree if reclaim returns 0. | |
2671 | * But our reclaim could return 0, simply because due | |
2672 | * to priority we are exposing a smaller subset of | |
2673 | * memory to reclaim from. Consider this as a longer | |
2674 | * term TODO. | |
2675 | */ | |
2676 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 2677 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 2678 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
2679 | css_put(&mz->memcg->css); |
2680 | loop++; | |
2681 | /* | |
2682 | * Could not reclaim anything and there are no more | |
2683 | * mem cgroups to try or we seem to be looping without | |
2684 | * reclaiming anything. | |
2685 | */ | |
2686 | if (!nr_reclaimed && | |
2687 | (next_mz == NULL || | |
2688 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2689 | break; | |
2690 | } while (!nr_reclaimed); | |
2691 | if (next_mz) | |
2692 | css_put(&next_mz->memcg->css); | |
2693 | return nr_reclaimed; | |
2694 | } | |
2695 | ||
ea280e7b TH |
2696 | /* |
2697 | * Test whether @memcg has children, dead or alive. Note that this | |
2698 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
2699 | * returns %true if there are child csses according to the cgroup | |
2700 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
2701 | */ | |
b5f99b53 GC |
2702 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
2703 | { | |
ea280e7b TH |
2704 | bool ret; |
2705 | ||
696ac172 | 2706 | /* |
ea280e7b TH |
2707 | * The lock does not prevent addition or deletion of children, but |
2708 | * it prevents a new child from being initialized based on this | |
2709 | * parent in css_online(), so it's enough to decide whether | |
2710 | * hierarchically inherited attributes can still be changed or not. | |
696ac172 | 2711 | */ |
ea280e7b TH |
2712 | lockdep_assert_held(&memcg_create_mutex); |
2713 | ||
2714 | rcu_read_lock(); | |
2715 | ret = css_next_child(NULL, &memcg->css); | |
2716 | rcu_read_unlock(); | |
2717 | return ret; | |
b5f99b53 GC |
2718 | } |
2719 | ||
c26251f9 MH |
2720 | /* |
2721 | * Reclaims as many pages from the given memcg as possible and moves | |
2722 | * the rest to the parent. | |
2723 | * | |
2724 | * Caller is responsible for holding css reference for memcg. | |
2725 | */ | |
2726 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
2727 | { | |
2728 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 2729 | |
c1e862c1 KH |
2730 | /* we call try-to-free pages for make this cgroup empty */ |
2731 | lru_add_drain_all(); | |
f817ed48 | 2732 | /* try to free all pages in this cgroup */ |
3e32cb2e | 2733 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 2734 | int progress; |
c1e862c1 | 2735 | |
c26251f9 MH |
2736 | if (signal_pending(current)) |
2737 | return -EINTR; | |
2738 | ||
b70a2a21 JW |
2739 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
2740 | GFP_KERNEL, true); | |
c1e862c1 | 2741 | if (!progress) { |
f817ed48 | 2742 | nr_retries--; |
c1e862c1 | 2743 | /* maybe some writeback is necessary */ |
8aa7e847 | 2744 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 2745 | } |
f817ed48 KH |
2746 | |
2747 | } | |
ab5196c2 MH |
2748 | |
2749 | return 0; | |
cc847582 KH |
2750 | } |
2751 | ||
6770c64e TH |
2752 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
2753 | char *buf, size_t nbytes, | |
2754 | loff_t off) | |
c1e862c1 | 2755 | { |
6770c64e | 2756 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 2757 | |
d8423011 MH |
2758 | if (mem_cgroup_is_root(memcg)) |
2759 | return -EINVAL; | |
6770c64e | 2760 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
2761 | } |
2762 | ||
182446d0 TH |
2763 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
2764 | struct cftype *cft) | |
18f59ea7 | 2765 | { |
182446d0 | 2766 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
2767 | } |
2768 | ||
182446d0 TH |
2769 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
2770 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
2771 | { |
2772 | int retval = 0; | |
182446d0 | 2773 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 2774 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 2775 | |
0999821b | 2776 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
2777 | |
2778 | if (memcg->use_hierarchy == val) | |
2779 | goto out; | |
2780 | ||
18f59ea7 | 2781 | /* |
af901ca1 | 2782 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
2783 | * in the child subtrees. If it is unset, then the change can |
2784 | * occur, provided the current cgroup has no children. | |
2785 | * | |
2786 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
2787 | * set if there are no children. | |
2788 | */ | |
c0ff4b85 | 2789 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 2790 | (val == 1 || val == 0)) { |
ea280e7b | 2791 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 2792 | memcg->use_hierarchy = val; |
18f59ea7 BS |
2793 | else |
2794 | retval = -EBUSY; | |
2795 | } else | |
2796 | retval = -EINVAL; | |
567fb435 GC |
2797 | |
2798 | out: | |
0999821b | 2799 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
2800 | |
2801 | return retval; | |
2802 | } | |
2803 | ||
3e32cb2e JW |
2804 | static unsigned long tree_stat(struct mem_cgroup *memcg, |
2805 | enum mem_cgroup_stat_index idx) | |
ce00a967 JW |
2806 | { |
2807 | struct mem_cgroup *iter; | |
484ebb3b | 2808 | unsigned long val = 0; |
ce00a967 | 2809 | |
ce00a967 JW |
2810 | for_each_mem_cgroup_tree(iter, memcg) |
2811 | val += mem_cgroup_read_stat(iter, idx); | |
2812 | ||
ce00a967 JW |
2813 | return val; |
2814 | } | |
2815 | ||
6f646156 | 2816 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 2817 | { |
c12176d3 | 2818 | unsigned long val; |
ce00a967 | 2819 | |
3e32cb2e JW |
2820 | if (mem_cgroup_is_root(memcg)) { |
2821 | val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE); | |
2822 | val += tree_stat(memcg, MEM_CGROUP_STAT_RSS); | |
2823 | if (swap) | |
2824 | val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP); | |
2825 | } else { | |
ce00a967 | 2826 | if (!swap) |
3e32cb2e | 2827 | val = page_counter_read(&memcg->memory); |
ce00a967 | 2828 | else |
3e32cb2e | 2829 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 2830 | } |
c12176d3 | 2831 | return val; |
ce00a967 JW |
2832 | } |
2833 | ||
3e32cb2e JW |
2834 | enum { |
2835 | RES_USAGE, | |
2836 | RES_LIMIT, | |
2837 | RES_MAX_USAGE, | |
2838 | RES_FAILCNT, | |
2839 | RES_SOFT_LIMIT, | |
2840 | }; | |
ce00a967 | 2841 | |
791badbd | 2842 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 2843 | struct cftype *cft) |
8cdea7c0 | 2844 | { |
182446d0 | 2845 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 2846 | struct page_counter *counter; |
af36f906 | 2847 | |
3e32cb2e | 2848 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 2849 | case _MEM: |
3e32cb2e JW |
2850 | counter = &memcg->memory; |
2851 | break; | |
8c7c6e34 | 2852 | case _MEMSWAP: |
3e32cb2e JW |
2853 | counter = &memcg->memsw; |
2854 | break; | |
510fc4e1 | 2855 | case _KMEM: |
3e32cb2e | 2856 | counter = &memcg->kmem; |
510fc4e1 | 2857 | break; |
8c7c6e34 KH |
2858 | default: |
2859 | BUG(); | |
8c7c6e34 | 2860 | } |
3e32cb2e JW |
2861 | |
2862 | switch (MEMFILE_ATTR(cft->private)) { | |
2863 | case RES_USAGE: | |
2864 | if (counter == &memcg->memory) | |
c12176d3 | 2865 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 2866 | if (counter == &memcg->memsw) |
c12176d3 | 2867 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
2868 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
2869 | case RES_LIMIT: | |
2870 | return (u64)counter->limit * PAGE_SIZE; | |
2871 | case RES_MAX_USAGE: | |
2872 | return (u64)counter->watermark * PAGE_SIZE; | |
2873 | case RES_FAILCNT: | |
2874 | return counter->failcnt; | |
2875 | case RES_SOFT_LIMIT: | |
2876 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
2877 | default: | |
2878 | BUG(); | |
2879 | } | |
8cdea7c0 | 2880 | } |
510fc4e1 | 2881 | |
510fc4e1 | 2882 | #ifdef CONFIG_MEMCG_KMEM |
8c0145b6 VD |
2883 | static int memcg_activate_kmem(struct mem_cgroup *memcg, |
2884 | unsigned long nr_pages) | |
d6441637 VD |
2885 | { |
2886 | int err = 0; | |
2887 | int memcg_id; | |
2888 | ||
2a4db7eb | 2889 | BUG_ON(memcg->kmemcg_id >= 0); |
2788cf0c | 2890 | BUG_ON(memcg->kmem_acct_activated); |
2a4db7eb | 2891 | BUG_ON(memcg->kmem_acct_active); |
d6441637 | 2892 | |
510fc4e1 GC |
2893 | /* |
2894 | * For simplicity, we won't allow this to be disabled. It also can't | |
2895 | * be changed if the cgroup has children already, or if tasks had | |
2896 | * already joined. | |
2897 | * | |
2898 | * If tasks join before we set the limit, a person looking at | |
2899 | * kmem.usage_in_bytes will have no way to determine when it took | |
2900 | * place, which makes the value quite meaningless. | |
2901 | * | |
2902 | * After it first became limited, changes in the value of the limit are | |
2903 | * of course permitted. | |
510fc4e1 | 2904 | */ |
0999821b | 2905 | mutex_lock(&memcg_create_mutex); |
27bd4dbb | 2906 | if (cgroup_is_populated(memcg->css.cgroup) || |
ea280e7b | 2907 | (memcg->use_hierarchy && memcg_has_children(memcg))) |
d6441637 VD |
2908 | err = -EBUSY; |
2909 | mutex_unlock(&memcg_create_mutex); | |
2910 | if (err) | |
2911 | goto out; | |
510fc4e1 | 2912 | |
f3bb3043 | 2913 | memcg_id = memcg_alloc_cache_id(); |
d6441637 VD |
2914 | if (memcg_id < 0) { |
2915 | err = memcg_id; | |
2916 | goto out; | |
2917 | } | |
2918 | ||
d6441637 | 2919 | /* |
900a38f0 VD |
2920 | * We couldn't have accounted to this cgroup, because it hasn't got |
2921 | * activated yet, so this should succeed. | |
d6441637 | 2922 | */ |
3e32cb2e | 2923 | err = page_counter_limit(&memcg->kmem, nr_pages); |
d6441637 VD |
2924 | VM_BUG_ON(err); |
2925 | ||
2926 | static_key_slow_inc(&memcg_kmem_enabled_key); | |
2927 | /* | |
900a38f0 VD |
2928 | * A memory cgroup is considered kmem-active as soon as it gets |
2929 | * kmemcg_id. Setting the id after enabling static branching will | |
d6441637 VD |
2930 | * guarantee no one starts accounting before all call sites are |
2931 | * patched. | |
2932 | */ | |
900a38f0 | 2933 | memcg->kmemcg_id = memcg_id; |
2788cf0c | 2934 | memcg->kmem_acct_activated = true; |
2a4db7eb | 2935 | memcg->kmem_acct_active = true; |
510fc4e1 | 2936 | out: |
d6441637 | 2937 | return err; |
d6441637 VD |
2938 | } |
2939 | ||
d6441637 | 2940 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2941 | unsigned long limit) |
d6441637 VD |
2942 | { |
2943 | int ret; | |
2944 | ||
3e32cb2e | 2945 | mutex_lock(&memcg_limit_mutex); |
d6441637 | 2946 | if (!memcg_kmem_is_active(memcg)) |
3e32cb2e | 2947 | ret = memcg_activate_kmem(memcg, limit); |
d6441637 | 2948 | else |
3e32cb2e JW |
2949 | ret = page_counter_limit(&memcg->kmem, limit); |
2950 | mutex_unlock(&memcg_limit_mutex); | |
510fc4e1 GC |
2951 | return ret; |
2952 | } | |
2953 | ||
55007d84 | 2954 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 2955 | { |
55007d84 | 2956 | int ret = 0; |
510fc4e1 | 2957 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
55007d84 | 2958 | |
d6441637 VD |
2959 | if (!parent) |
2960 | return 0; | |
55007d84 | 2961 | |
8c0145b6 | 2962 | mutex_lock(&memcg_limit_mutex); |
55007d84 | 2963 | /* |
d6441637 VD |
2964 | * If the parent cgroup is not kmem-active now, it cannot be activated |
2965 | * after this point, because it has at least one child already. | |
55007d84 | 2966 | */ |
d6441637 | 2967 | if (memcg_kmem_is_active(parent)) |
8c0145b6 VD |
2968 | ret = memcg_activate_kmem(memcg, PAGE_COUNTER_MAX); |
2969 | mutex_unlock(&memcg_limit_mutex); | |
55007d84 | 2970 | return ret; |
510fc4e1 | 2971 | } |
d6441637 VD |
2972 | #else |
2973 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
3e32cb2e | 2974 | unsigned long limit) |
d6441637 VD |
2975 | { |
2976 | return -EINVAL; | |
2977 | } | |
6d043990 | 2978 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 2979 | |
628f4235 KH |
2980 | /* |
2981 | * The user of this function is... | |
2982 | * RES_LIMIT. | |
2983 | */ | |
451af504 TH |
2984 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
2985 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 2986 | { |
451af504 | 2987 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 2988 | unsigned long nr_pages; |
628f4235 KH |
2989 | int ret; |
2990 | ||
451af504 | 2991 | buf = strstrip(buf); |
650c5e56 | 2992 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
2993 | if (ret) |
2994 | return ret; | |
af36f906 | 2995 | |
3e32cb2e | 2996 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 2997 | case RES_LIMIT: |
4b3bde4c BS |
2998 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
2999 | ret = -EINVAL; | |
3000 | break; | |
3001 | } | |
3e32cb2e JW |
3002 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3003 | case _MEM: | |
3004 | ret = mem_cgroup_resize_limit(memcg, nr_pages); | |
8c7c6e34 | 3005 | break; |
3e32cb2e JW |
3006 | case _MEMSWAP: |
3007 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); | |
296c81d8 | 3008 | break; |
3e32cb2e JW |
3009 | case _KMEM: |
3010 | ret = memcg_update_kmem_limit(memcg, nr_pages); | |
3011 | break; | |
3012 | } | |
296c81d8 | 3013 | break; |
3e32cb2e JW |
3014 | case RES_SOFT_LIMIT: |
3015 | memcg->soft_limit = nr_pages; | |
3016 | ret = 0; | |
628f4235 KH |
3017 | break; |
3018 | } | |
451af504 | 3019 | return ret ?: nbytes; |
8cdea7c0 BS |
3020 | } |
3021 | ||
6770c64e TH |
3022 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3023 | size_t nbytes, loff_t off) | |
c84872e1 | 3024 | { |
6770c64e | 3025 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3026 | struct page_counter *counter; |
c84872e1 | 3027 | |
3e32cb2e JW |
3028 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3029 | case _MEM: | |
3030 | counter = &memcg->memory; | |
3031 | break; | |
3032 | case _MEMSWAP: | |
3033 | counter = &memcg->memsw; | |
3034 | break; | |
3035 | case _KMEM: | |
3036 | counter = &memcg->kmem; | |
3037 | break; | |
3038 | default: | |
3039 | BUG(); | |
3040 | } | |
af36f906 | 3041 | |
3e32cb2e | 3042 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3043 | case RES_MAX_USAGE: |
3e32cb2e | 3044 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3045 | break; |
3046 | case RES_FAILCNT: | |
3e32cb2e | 3047 | counter->failcnt = 0; |
29f2a4da | 3048 | break; |
3e32cb2e JW |
3049 | default: |
3050 | BUG(); | |
29f2a4da | 3051 | } |
f64c3f54 | 3052 | |
6770c64e | 3053 | return nbytes; |
c84872e1 PE |
3054 | } |
3055 | ||
182446d0 | 3056 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3057 | struct cftype *cft) |
3058 | { | |
182446d0 | 3059 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3060 | } |
3061 | ||
02491447 | 3062 | #ifdef CONFIG_MMU |
182446d0 | 3063 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3064 | struct cftype *cft, u64 val) |
3065 | { | |
182446d0 | 3066 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3067 | |
1dfab5ab | 3068 | if (val & ~MOVE_MASK) |
7dc74be0 | 3069 | return -EINVAL; |
ee5e8472 | 3070 | |
7dc74be0 | 3071 | /* |
ee5e8472 GC |
3072 | * No kind of locking is needed in here, because ->can_attach() will |
3073 | * check this value once in the beginning of the process, and then carry | |
3074 | * on with stale data. This means that changes to this value will only | |
3075 | * affect task migrations starting after the change. | |
7dc74be0 | 3076 | */ |
c0ff4b85 | 3077 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3078 | return 0; |
3079 | } | |
02491447 | 3080 | #else |
182446d0 | 3081 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3082 | struct cftype *cft, u64 val) |
3083 | { | |
3084 | return -ENOSYS; | |
3085 | } | |
3086 | #endif | |
7dc74be0 | 3087 | |
406eb0c9 | 3088 | #ifdef CONFIG_NUMA |
2da8ca82 | 3089 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3090 | { |
25485de6 GT |
3091 | struct numa_stat { |
3092 | const char *name; | |
3093 | unsigned int lru_mask; | |
3094 | }; | |
3095 | ||
3096 | static const struct numa_stat stats[] = { | |
3097 | { "total", LRU_ALL }, | |
3098 | { "file", LRU_ALL_FILE }, | |
3099 | { "anon", LRU_ALL_ANON }, | |
3100 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3101 | }; | |
3102 | const struct numa_stat *stat; | |
406eb0c9 | 3103 | int nid; |
25485de6 | 3104 | unsigned long nr; |
2da8ca82 | 3105 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 3106 | |
25485de6 GT |
3107 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3108 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3109 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3110 | for_each_node_state(nid, N_MEMORY) { | |
3111 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3112 | stat->lru_mask); | |
3113 | seq_printf(m, " N%d=%lu", nid, nr); | |
3114 | } | |
3115 | seq_putc(m, '\n'); | |
406eb0c9 | 3116 | } |
406eb0c9 | 3117 | |
071aee13 YH |
3118 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3119 | struct mem_cgroup *iter; | |
3120 | ||
3121 | nr = 0; | |
3122 | for_each_mem_cgroup_tree(iter, memcg) | |
3123 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3124 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3125 | for_each_node_state(nid, N_MEMORY) { | |
3126 | nr = 0; | |
3127 | for_each_mem_cgroup_tree(iter, memcg) | |
3128 | nr += mem_cgroup_node_nr_lru_pages( | |
3129 | iter, nid, stat->lru_mask); | |
3130 | seq_printf(m, " N%d=%lu", nid, nr); | |
3131 | } | |
3132 | seq_putc(m, '\n'); | |
406eb0c9 | 3133 | } |
406eb0c9 | 3134 | |
406eb0c9 YH |
3135 | return 0; |
3136 | } | |
3137 | #endif /* CONFIG_NUMA */ | |
3138 | ||
2da8ca82 | 3139 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3140 | { |
2da8ca82 | 3141 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 3142 | unsigned long memory, memsw; |
af7c4b0e JW |
3143 | struct mem_cgroup *mi; |
3144 | unsigned int i; | |
406eb0c9 | 3145 | |
0ca44b14 GT |
3146 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) != |
3147 | MEM_CGROUP_STAT_NSTATS); | |
3148 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) != | |
3149 | MEM_CGROUP_EVENTS_NSTATS); | |
70bc068c RS |
3150 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3151 | ||
af7c4b0e | 3152 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 3153 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 3154 | continue; |
484ebb3b | 3155 | seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i], |
af7c4b0e | 3156 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); |
1dd3a273 | 3157 | } |
7b854121 | 3158 | |
af7c4b0e JW |
3159 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
3160 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
3161 | mem_cgroup_read_events(memcg, i)); | |
3162 | ||
3163 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3164 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
3165 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
3166 | ||
14067bb3 | 3167 | /* Hierarchical information */ |
3e32cb2e JW |
3168 | memory = memsw = PAGE_COUNTER_MAX; |
3169 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
3170 | memory = min(memory, mi->memory.limit); | |
3171 | memsw = min(memsw, mi->memsw.limit); | |
fee7b548 | 3172 | } |
3e32cb2e JW |
3173 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3174 | (u64)memory * PAGE_SIZE); | |
3175 | if (do_swap_account) | |
3176 | seq_printf(m, "hierarchical_memsw_limit %llu\n", | |
3177 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3178 | |
af7c4b0e | 3179 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
484ebb3b | 3180 | unsigned long long val = 0; |
af7c4b0e | 3181 | |
bff6bb83 | 3182 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 3183 | continue; |
af7c4b0e JW |
3184 | for_each_mem_cgroup_tree(mi, memcg) |
3185 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
484ebb3b | 3186 | seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val); |
af7c4b0e JW |
3187 | } |
3188 | ||
3189 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
3190 | unsigned long long val = 0; | |
3191 | ||
3192 | for_each_mem_cgroup_tree(mi, memcg) | |
3193 | val += mem_cgroup_read_events(mi, i); | |
3194 | seq_printf(m, "total_%s %llu\n", | |
3195 | mem_cgroup_events_names[i], val); | |
3196 | } | |
3197 | ||
3198 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
3199 | unsigned long long val = 0; | |
3200 | ||
3201 | for_each_mem_cgroup_tree(mi, memcg) | |
3202 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
3203 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 3204 | } |
14067bb3 | 3205 | |
7f016ee8 | 3206 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
3207 | { |
3208 | int nid, zid; | |
3209 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 3210 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3211 | unsigned long recent_rotated[2] = {0, 0}; |
3212 | unsigned long recent_scanned[2] = {0, 0}; | |
3213 | ||
3214 | for_each_online_node(nid) | |
3215 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
e231875b | 3216 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
89abfab1 | 3217 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 3218 | |
89abfab1 HD |
3219 | recent_rotated[0] += rstat->recent_rotated[0]; |
3220 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3221 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3222 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 3223 | } |
78ccf5b5 JW |
3224 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3225 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3226 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3227 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3228 | } |
3229 | #endif | |
3230 | ||
d2ceb9b7 KH |
3231 | return 0; |
3232 | } | |
3233 | ||
182446d0 TH |
3234 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3235 | struct cftype *cft) | |
a7885eb8 | 3236 | { |
182446d0 | 3237 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3238 | |
1f4c025b | 3239 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3240 | } |
3241 | ||
182446d0 TH |
3242 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3243 | struct cftype *cft, u64 val) | |
a7885eb8 | 3244 | { |
182446d0 | 3245 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3246 | |
3dae7fec | 3247 | if (val > 100) |
a7885eb8 KM |
3248 | return -EINVAL; |
3249 | ||
14208b0e | 3250 | if (css->parent) |
3dae7fec JW |
3251 | memcg->swappiness = val; |
3252 | else | |
3253 | vm_swappiness = val; | |
068b38c1 | 3254 | |
a7885eb8 KM |
3255 | return 0; |
3256 | } | |
3257 | ||
2e72b634 KS |
3258 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3259 | { | |
3260 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3261 | unsigned long usage; |
2e72b634 KS |
3262 | int i; |
3263 | ||
3264 | rcu_read_lock(); | |
3265 | if (!swap) | |
2c488db2 | 3266 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3267 | else |
2c488db2 | 3268 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3269 | |
3270 | if (!t) | |
3271 | goto unlock; | |
3272 | ||
ce00a967 | 3273 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3274 | |
3275 | /* | |
748dad36 | 3276 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3277 | * If it's not true, a threshold was crossed after last |
3278 | * call of __mem_cgroup_threshold(). | |
3279 | */ | |
5407a562 | 3280 | i = t->current_threshold; |
2e72b634 KS |
3281 | |
3282 | /* | |
3283 | * Iterate backward over array of thresholds starting from | |
3284 | * current_threshold and check if a threshold is crossed. | |
3285 | * If none of thresholds below usage is crossed, we read | |
3286 | * only one element of the array here. | |
3287 | */ | |
3288 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3289 | eventfd_signal(t->entries[i].eventfd, 1); | |
3290 | ||
3291 | /* i = current_threshold + 1 */ | |
3292 | i++; | |
3293 | ||
3294 | /* | |
3295 | * Iterate forward over array of thresholds starting from | |
3296 | * current_threshold+1 and check if a threshold is crossed. | |
3297 | * If none of thresholds above usage is crossed, we read | |
3298 | * only one element of the array here. | |
3299 | */ | |
3300 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3301 | eventfd_signal(t->entries[i].eventfd, 1); | |
3302 | ||
3303 | /* Update current_threshold */ | |
5407a562 | 3304 | t->current_threshold = i - 1; |
2e72b634 KS |
3305 | unlock: |
3306 | rcu_read_unlock(); | |
3307 | } | |
3308 | ||
3309 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3310 | { | |
ad4ca5f4 KS |
3311 | while (memcg) { |
3312 | __mem_cgroup_threshold(memcg, false); | |
3313 | if (do_swap_account) | |
3314 | __mem_cgroup_threshold(memcg, true); | |
3315 | ||
3316 | memcg = parent_mem_cgroup(memcg); | |
3317 | } | |
2e72b634 KS |
3318 | } |
3319 | ||
3320 | static int compare_thresholds(const void *a, const void *b) | |
3321 | { | |
3322 | const struct mem_cgroup_threshold *_a = a; | |
3323 | const struct mem_cgroup_threshold *_b = b; | |
3324 | ||
2bff24a3 GT |
3325 | if (_a->threshold > _b->threshold) |
3326 | return 1; | |
3327 | ||
3328 | if (_a->threshold < _b->threshold) | |
3329 | return -1; | |
3330 | ||
3331 | return 0; | |
2e72b634 KS |
3332 | } |
3333 | ||
c0ff4b85 | 3334 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3335 | { |
3336 | struct mem_cgroup_eventfd_list *ev; | |
3337 | ||
2bcf2e92 MH |
3338 | spin_lock(&memcg_oom_lock); |
3339 | ||
c0ff4b85 | 3340 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3341 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3342 | |
3343 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3344 | return 0; |
3345 | } | |
3346 | ||
c0ff4b85 | 3347 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3348 | { |
7d74b06f KH |
3349 | struct mem_cgroup *iter; |
3350 | ||
c0ff4b85 | 3351 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3352 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3353 | } |
3354 | ||
59b6f873 | 3355 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3356 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3357 | { |
2c488db2 KS |
3358 | struct mem_cgroup_thresholds *thresholds; |
3359 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3360 | unsigned long threshold; |
3361 | unsigned long usage; | |
2c488db2 | 3362 | int i, size, ret; |
2e72b634 | 3363 | |
650c5e56 | 3364 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
3365 | if (ret) |
3366 | return ret; | |
3367 | ||
3368 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3369 | |
05b84301 | 3370 | if (type == _MEM) { |
2c488db2 | 3371 | thresholds = &memcg->thresholds; |
ce00a967 | 3372 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3373 | } else if (type == _MEMSWAP) { |
2c488db2 | 3374 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3375 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3376 | } else |
2e72b634 KS |
3377 | BUG(); |
3378 | ||
2e72b634 | 3379 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3380 | if (thresholds->primary) |
2e72b634 KS |
3381 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3382 | ||
2c488db2 | 3383 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3384 | |
3385 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 3386 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 3387 | GFP_KERNEL); |
2c488db2 | 3388 | if (!new) { |
2e72b634 KS |
3389 | ret = -ENOMEM; |
3390 | goto unlock; | |
3391 | } | |
2c488db2 | 3392 | new->size = size; |
2e72b634 KS |
3393 | |
3394 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3395 | if (thresholds->primary) { |
3396 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3397 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3398 | } |
3399 | ||
2e72b634 | 3400 | /* Add new threshold */ |
2c488db2 KS |
3401 | new->entries[size - 1].eventfd = eventfd; |
3402 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3403 | |
3404 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3405 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3406 | compare_thresholds, NULL); |
3407 | ||
3408 | /* Find current threshold */ | |
2c488db2 | 3409 | new->current_threshold = -1; |
2e72b634 | 3410 | for (i = 0; i < size; i++) { |
748dad36 | 3411 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3412 | /* |
2c488db2 KS |
3413 | * new->current_threshold will not be used until |
3414 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3415 | * it here. |
3416 | */ | |
2c488db2 | 3417 | ++new->current_threshold; |
748dad36 SZ |
3418 | } else |
3419 | break; | |
2e72b634 KS |
3420 | } |
3421 | ||
2c488db2 KS |
3422 | /* Free old spare buffer and save old primary buffer as spare */ |
3423 | kfree(thresholds->spare); | |
3424 | thresholds->spare = thresholds->primary; | |
3425 | ||
3426 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3427 | |
907860ed | 3428 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3429 | synchronize_rcu(); |
3430 | ||
2e72b634 KS |
3431 | unlock: |
3432 | mutex_unlock(&memcg->thresholds_lock); | |
3433 | ||
3434 | return ret; | |
3435 | } | |
3436 | ||
59b6f873 | 3437 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3438 | struct eventfd_ctx *eventfd, const char *args) |
3439 | { | |
59b6f873 | 3440 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3441 | } |
3442 | ||
59b6f873 | 3443 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3444 | struct eventfd_ctx *eventfd, const char *args) |
3445 | { | |
59b6f873 | 3446 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3447 | } |
3448 | ||
59b6f873 | 3449 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3450 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3451 | { |
2c488db2 KS |
3452 | struct mem_cgroup_thresholds *thresholds; |
3453 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3454 | unsigned long usage; |
2c488db2 | 3455 | int i, j, size; |
2e72b634 KS |
3456 | |
3457 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3458 | |
3459 | if (type == _MEM) { | |
2c488db2 | 3460 | thresholds = &memcg->thresholds; |
ce00a967 | 3461 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3462 | } else if (type == _MEMSWAP) { |
2c488db2 | 3463 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3464 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3465 | } else |
2e72b634 KS |
3466 | BUG(); |
3467 | ||
371528ca AV |
3468 | if (!thresholds->primary) |
3469 | goto unlock; | |
3470 | ||
2e72b634 KS |
3471 | /* Check if a threshold crossed before removing */ |
3472 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3473 | ||
3474 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3475 | size = 0; |
3476 | for (i = 0; i < thresholds->primary->size; i++) { | |
3477 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3478 | size++; |
3479 | } | |
3480 | ||
2c488db2 | 3481 | new = thresholds->spare; |
907860ed | 3482 | |
2e72b634 KS |
3483 | /* Set thresholds array to NULL if we don't have thresholds */ |
3484 | if (!size) { | |
2c488db2 KS |
3485 | kfree(new); |
3486 | new = NULL; | |
907860ed | 3487 | goto swap_buffers; |
2e72b634 KS |
3488 | } |
3489 | ||
2c488db2 | 3490 | new->size = size; |
2e72b634 KS |
3491 | |
3492 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3493 | new->current_threshold = -1; |
3494 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3495 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3496 | continue; |
3497 | ||
2c488db2 | 3498 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3499 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3500 | /* |
2c488db2 | 3501 | * new->current_threshold will not be used |
2e72b634 KS |
3502 | * until rcu_assign_pointer(), so it's safe to increment |
3503 | * it here. | |
3504 | */ | |
2c488db2 | 3505 | ++new->current_threshold; |
2e72b634 KS |
3506 | } |
3507 | j++; | |
3508 | } | |
3509 | ||
907860ed | 3510 | swap_buffers: |
2c488db2 KS |
3511 | /* Swap primary and spare array */ |
3512 | thresholds->spare = thresholds->primary; | |
8c757763 | 3513 | |
2c488db2 | 3514 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3515 | |
907860ed | 3516 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3517 | synchronize_rcu(); |
ececa3eb MC |
3518 | |
3519 | /* If all events are unregistered, free the spare array */ | |
3520 | if (!new) { | |
3521 | kfree(thresholds->spare); | |
3522 | thresholds->spare = NULL; | |
3523 | } | |
371528ca | 3524 | unlock: |
2e72b634 | 3525 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3526 | } |
c1e862c1 | 3527 | |
59b6f873 | 3528 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3529 | struct eventfd_ctx *eventfd) |
3530 | { | |
59b6f873 | 3531 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
3532 | } |
3533 | ||
59b6f873 | 3534 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3535 | struct eventfd_ctx *eventfd) |
3536 | { | |
59b6f873 | 3537 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
3538 | } |
3539 | ||
59b6f873 | 3540 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3541 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 3542 | { |
9490ff27 | 3543 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 3544 | |
9490ff27 KH |
3545 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
3546 | if (!event) | |
3547 | return -ENOMEM; | |
3548 | ||
1af8efe9 | 3549 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
3550 | |
3551 | event->eventfd = eventfd; | |
3552 | list_add(&event->list, &memcg->oom_notify); | |
3553 | ||
3554 | /* already in OOM ? */ | |
c2b42d3c | 3555 | if (memcg->under_oom) |
9490ff27 | 3556 | eventfd_signal(eventfd, 1); |
1af8efe9 | 3557 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3558 | |
3559 | return 0; | |
3560 | } | |
3561 | ||
59b6f873 | 3562 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3563 | struct eventfd_ctx *eventfd) |
9490ff27 | 3564 | { |
9490ff27 | 3565 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 3566 | |
1af8efe9 | 3567 | spin_lock(&memcg_oom_lock); |
9490ff27 | 3568 | |
c0ff4b85 | 3569 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
3570 | if (ev->eventfd == eventfd) { |
3571 | list_del(&ev->list); | |
3572 | kfree(ev); | |
3573 | } | |
3574 | } | |
3575 | ||
1af8efe9 | 3576 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3577 | } |
3578 | ||
2da8ca82 | 3579 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 3580 | { |
2da8ca82 | 3581 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 3582 | |
791badbd | 3583 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 3584 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
3c11ecf4 KH |
3585 | return 0; |
3586 | } | |
3587 | ||
182446d0 | 3588 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
3589 | struct cftype *cft, u64 val) |
3590 | { | |
182446d0 | 3591 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
3592 | |
3593 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 3594 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
3595 | return -EINVAL; |
3596 | ||
c0ff4b85 | 3597 | memcg->oom_kill_disable = val; |
4d845ebf | 3598 | if (!val) |
c0ff4b85 | 3599 | memcg_oom_recover(memcg); |
3dae7fec | 3600 | |
3c11ecf4 KH |
3601 | return 0; |
3602 | } | |
3603 | ||
c255a458 | 3604 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 3605 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 3606 | { |
55007d84 GC |
3607 | int ret; |
3608 | ||
55007d84 GC |
3609 | ret = memcg_propagate_kmem(memcg); |
3610 | if (ret) | |
3611 | return ret; | |
2633d7a0 | 3612 | |
1d62e436 | 3613 | return mem_cgroup_sockets_init(memcg, ss); |
573b400d | 3614 | } |
e5671dfa | 3615 | |
2a4db7eb VD |
3616 | static void memcg_deactivate_kmem(struct mem_cgroup *memcg) |
3617 | { | |
2788cf0c VD |
3618 | struct cgroup_subsys_state *css; |
3619 | struct mem_cgroup *parent, *child; | |
3620 | int kmemcg_id; | |
3621 | ||
2a4db7eb VD |
3622 | if (!memcg->kmem_acct_active) |
3623 | return; | |
3624 | ||
3625 | /* | |
3626 | * Clear the 'active' flag before clearing memcg_caches arrays entries. | |
3627 | * Since we take the slab_mutex in memcg_deactivate_kmem_caches(), it | |
3628 | * guarantees no cache will be created for this cgroup after we are | |
3629 | * done (see memcg_create_kmem_cache()). | |
3630 | */ | |
3631 | memcg->kmem_acct_active = false; | |
3632 | ||
3633 | memcg_deactivate_kmem_caches(memcg); | |
2788cf0c VD |
3634 | |
3635 | kmemcg_id = memcg->kmemcg_id; | |
3636 | BUG_ON(kmemcg_id < 0); | |
3637 | ||
3638 | parent = parent_mem_cgroup(memcg); | |
3639 | if (!parent) | |
3640 | parent = root_mem_cgroup; | |
3641 | ||
3642 | /* | |
3643 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3644 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3645 | * to ones of the parent. After we have finished, all list_lrus | |
3646 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3647 | * ordering is imposed by list_lru_node->lock taken by | |
3648 | * memcg_drain_all_list_lrus(). | |
3649 | */ | |
d3f97524 | 3650 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
2788cf0c VD |
3651 | css_for_each_descendant_pre(css, &memcg->css) { |
3652 | child = mem_cgroup_from_css(css); | |
3653 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3654 | child->kmemcg_id = parent->kmemcg_id; | |
3655 | if (!memcg->use_hierarchy) | |
3656 | break; | |
3657 | } | |
d3f97524 TH |
3658 | rcu_read_unlock(); |
3659 | ||
2788cf0c VD |
3660 | memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id); |
3661 | ||
3662 | memcg_free_cache_id(kmemcg_id); | |
2a4db7eb VD |
3663 | } |
3664 | ||
10d5ebf4 | 3665 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 3666 | { |
f48b80a5 VD |
3667 | if (memcg->kmem_acct_activated) { |
3668 | memcg_destroy_kmem_caches(memcg); | |
3669 | static_key_slow_dec(&memcg_kmem_enabled_key); | |
3670 | WARN_ON(page_counter_read(&memcg->kmem)); | |
3671 | } | |
1d62e436 | 3672 | mem_cgroup_sockets_destroy(memcg); |
10d5ebf4 | 3673 | } |
e5671dfa | 3674 | #else |
cbe128e3 | 3675 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
3676 | { |
3677 | return 0; | |
3678 | } | |
d1a4c0b3 | 3679 | |
2a4db7eb VD |
3680 | static void memcg_deactivate_kmem(struct mem_cgroup *memcg) |
3681 | { | |
3682 | } | |
3683 | ||
10d5ebf4 LZ |
3684 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
3685 | { | |
3686 | } | |
e5671dfa GC |
3687 | #endif |
3688 | ||
52ebea74 TH |
3689 | #ifdef CONFIG_CGROUP_WRITEBACK |
3690 | ||
3691 | struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg) | |
3692 | { | |
3693 | return &memcg->cgwb_list; | |
3694 | } | |
3695 | ||
841710aa TH |
3696 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
3697 | { | |
3698 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
3699 | } | |
3700 | ||
3701 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3702 | { | |
3703 | wb_domain_exit(&memcg->cgwb_domain); | |
3704 | } | |
3705 | ||
2529bb3a TH |
3706 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3707 | { | |
3708 | wb_domain_size_changed(&memcg->cgwb_domain); | |
3709 | } | |
3710 | ||
841710aa TH |
3711 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
3712 | { | |
3713 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3714 | ||
3715 | if (!memcg->css.parent) | |
3716 | return NULL; | |
3717 | ||
3718 | return &memcg->cgwb_domain; | |
3719 | } | |
3720 | ||
c2aa723a TH |
3721 | /** |
3722 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
3723 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
3724 | * @pfilepages: out parameter for number of file pages |
3725 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
3726 | * @pdirty: out parameter for number of dirty pages |
3727 | * @pwriteback: out parameter for number of pages under writeback | |
3728 | * | |
c5edf9cd TH |
3729 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
3730 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
3731 | * is a bit more involved. | |
c2aa723a | 3732 | * |
c5edf9cd TH |
3733 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
3734 | * headroom is calculated as the lowest headroom of itself and the | |
3735 | * ancestors. Note that this doesn't consider the actual amount of | |
3736 | * available memory in the system. The caller should further cap | |
3737 | * *@pheadroom accordingly. | |
c2aa723a | 3738 | */ |
c5edf9cd TH |
3739 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
3740 | unsigned long *pheadroom, unsigned long *pdirty, | |
3741 | unsigned long *pwriteback) | |
c2aa723a TH |
3742 | { |
3743 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3744 | struct mem_cgroup *parent; | |
c2aa723a TH |
3745 | |
3746 | *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY); | |
3747 | ||
3748 | /* this should eventually include NR_UNSTABLE_NFS */ | |
3749 | *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK); | |
c5edf9cd TH |
3750 | *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) | |
3751 | (1 << LRU_ACTIVE_FILE)); | |
3752 | *pheadroom = PAGE_COUNTER_MAX; | |
c2aa723a | 3753 | |
c2aa723a TH |
3754 | while ((parent = parent_mem_cgroup(memcg))) { |
3755 | unsigned long ceiling = min(memcg->memory.limit, memcg->high); | |
3756 | unsigned long used = page_counter_read(&memcg->memory); | |
3757 | ||
c5edf9cd | 3758 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
3759 | memcg = parent; |
3760 | } | |
c2aa723a TH |
3761 | } |
3762 | ||
841710aa TH |
3763 | #else /* CONFIG_CGROUP_WRITEBACK */ |
3764 | ||
3765 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
3766 | { | |
3767 | return 0; | |
3768 | } | |
3769 | ||
3770 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3771 | { | |
3772 | } | |
3773 | ||
2529bb3a TH |
3774 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3775 | { | |
3776 | } | |
3777 | ||
52ebea74 TH |
3778 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
3779 | ||
3bc942f3 TH |
3780 | /* |
3781 | * DO NOT USE IN NEW FILES. | |
3782 | * | |
3783 | * "cgroup.event_control" implementation. | |
3784 | * | |
3785 | * This is way over-engineered. It tries to support fully configurable | |
3786 | * events for each user. Such level of flexibility is completely | |
3787 | * unnecessary especially in the light of the planned unified hierarchy. | |
3788 | * | |
3789 | * Please deprecate this and replace with something simpler if at all | |
3790 | * possible. | |
3791 | */ | |
3792 | ||
79bd9814 TH |
3793 | /* |
3794 | * Unregister event and free resources. | |
3795 | * | |
3796 | * Gets called from workqueue. | |
3797 | */ | |
3bc942f3 | 3798 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 3799 | { |
3bc942f3 TH |
3800 | struct mem_cgroup_event *event = |
3801 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 3802 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3803 | |
3804 | remove_wait_queue(event->wqh, &event->wait); | |
3805 | ||
59b6f873 | 3806 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
3807 | |
3808 | /* Notify userspace the event is going away. */ | |
3809 | eventfd_signal(event->eventfd, 1); | |
3810 | ||
3811 | eventfd_ctx_put(event->eventfd); | |
3812 | kfree(event); | |
59b6f873 | 3813 | css_put(&memcg->css); |
79bd9814 TH |
3814 | } |
3815 | ||
3816 | /* | |
3817 | * Gets called on POLLHUP on eventfd when user closes it. | |
3818 | * | |
3819 | * Called with wqh->lock held and interrupts disabled. | |
3820 | */ | |
3bc942f3 TH |
3821 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
3822 | int sync, void *key) | |
79bd9814 | 3823 | { |
3bc942f3 TH |
3824 | struct mem_cgroup_event *event = |
3825 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 3826 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3827 | unsigned long flags = (unsigned long)key; |
3828 | ||
3829 | if (flags & POLLHUP) { | |
3830 | /* | |
3831 | * If the event has been detached at cgroup removal, we | |
3832 | * can simply return knowing the other side will cleanup | |
3833 | * for us. | |
3834 | * | |
3835 | * We can't race against event freeing since the other | |
3836 | * side will require wqh->lock via remove_wait_queue(), | |
3837 | * which we hold. | |
3838 | */ | |
fba94807 | 3839 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
3840 | if (!list_empty(&event->list)) { |
3841 | list_del_init(&event->list); | |
3842 | /* | |
3843 | * We are in atomic context, but cgroup_event_remove() | |
3844 | * may sleep, so we have to call it in workqueue. | |
3845 | */ | |
3846 | schedule_work(&event->remove); | |
3847 | } | |
fba94807 | 3848 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
3849 | } |
3850 | ||
3851 | return 0; | |
3852 | } | |
3853 | ||
3bc942f3 | 3854 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
3855 | wait_queue_head_t *wqh, poll_table *pt) |
3856 | { | |
3bc942f3 TH |
3857 | struct mem_cgroup_event *event = |
3858 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
3859 | |
3860 | event->wqh = wqh; | |
3861 | add_wait_queue(wqh, &event->wait); | |
3862 | } | |
3863 | ||
3864 | /* | |
3bc942f3 TH |
3865 | * DO NOT USE IN NEW FILES. |
3866 | * | |
79bd9814 TH |
3867 | * Parse input and register new cgroup event handler. |
3868 | * | |
3869 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
3870 | * Interpretation of args is defined by control file implementation. | |
3871 | */ | |
451af504 TH |
3872 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
3873 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 3874 | { |
451af504 | 3875 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 3876 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 3877 | struct mem_cgroup_event *event; |
79bd9814 TH |
3878 | struct cgroup_subsys_state *cfile_css; |
3879 | unsigned int efd, cfd; | |
3880 | struct fd efile; | |
3881 | struct fd cfile; | |
fba94807 | 3882 | const char *name; |
79bd9814 TH |
3883 | char *endp; |
3884 | int ret; | |
3885 | ||
451af504 TH |
3886 | buf = strstrip(buf); |
3887 | ||
3888 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
3889 | if (*endp != ' ') |
3890 | return -EINVAL; | |
451af504 | 3891 | buf = endp + 1; |
79bd9814 | 3892 | |
451af504 | 3893 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
3894 | if ((*endp != ' ') && (*endp != '\0')) |
3895 | return -EINVAL; | |
451af504 | 3896 | buf = endp + 1; |
79bd9814 TH |
3897 | |
3898 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
3899 | if (!event) | |
3900 | return -ENOMEM; | |
3901 | ||
59b6f873 | 3902 | event->memcg = memcg; |
79bd9814 | 3903 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
3904 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
3905 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
3906 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
3907 | |
3908 | efile = fdget(efd); | |
3909 | if (!efile.file) { | |
3910 | ret = -EBADF; | |
3911 | goto out_kfree; | |
3912 | } | |
3913 | ||
3914 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
3915 | if (IS_ERR(event->eventfd)) { | |
3916 | ret = PTR_ERR(event->eventfd); | |
3917 | goto out_put_efile; | |
3918 | } | |
3919 | ||
3920 | cfile = fdget(cfd); | |
3921 | if (!cfile.file) { | |
3922 | ret = -EBADF; | |
3923 | goto out_put_eventfd; | |
3924 | } | |
3925 | ||
3926 | /* the process need read permission on control file */ | |
3927 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
3928 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
3929 | if (ret < 0) | |
3930 | goto out_put_cfile; | |
3931 | ||
fba94807 TH |
3932 | /* |
3933 | * Determine the event callbacks and set them in @event. This used | |
3934 | * to be done via struct cftype but cgroup core no longer knows | |
3935 | * about these events. The following is crude but the whole thing | |
3936 | * is for compatibility anyway. | |
3bc942f3 TH |
3937 | * |
3938 | * DO NOT ADD NEW FILES. | |
fba94807 | 3939 | */ |
b583043e | 3940 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
3941 | |
3942 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
3943 | event->register_event = mem_cgroup_usage_register_event; | |
3944 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
3945 | } else if (!strcmp(name, "memory.oom_control")) { | |
3946 | event->register_event = mem_cgroup_oom_register_event; | |
3947 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
3948 | } else if (!strcmp(name, "memory.pressure_level")) { | |
3949 | event->register_event = vmpressure_register_event; | |
3950 | event->unregister_event = vmpressure_unregister_event; | |
3951 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
3952 | event->register_event = memsw_cgroup_usage_register_event; |
3953 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
3954 | } else { |
3955 | ret = -EINVAL; | |
3956 | goto out_put_cfile; | |
3957 | } | |
3958 | ||
79bd9814 | 3959 | /* |
b5557c4c TH |
3960 | * Verify @cfile should belong to @css. Also, remaining events are |
3961 | * automatically removed on cgroup destruction but the removal is | |
3962 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 3963 | */ |
b583043e | 3964 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 3965 | &memory_cgrp_subsys); |
79bd9814 | 3966 | ret = -EINVAL; |
5a17f543 | 3967 | if (IS_ERR(cfile_css)) |
79bd9814 | 3968 | goto out_put_cfile; |
5a17f543 TH |
3969 | if (cfile_css != css) { |
3970 | css_put(cfile_css); | |
79bd9814 | 3971 | goto out_put_cfile; |
5a17f543 | 3972 | } |
79bd9814 | 3973 | |
451af504 | 3974 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
3975 | if (ret) |
3976 | goto out_put_css; | |
3977 | ||
3978 | efile.file->f_op->poll(efile.file, &event->pt); | |
3979 | ||
fba94807 TH |
3980 | spin_lock(&memcg->event_list_lock); |
3981 | list_add(&event->list, &memcg->event_list); | |
3982 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
3983 | |
3984 | fdput(cfile); | |
3985 | fdput(efile); | |
3986 | ||
451af504 | 3987 | return nbytes; |
79bd9814 TH |
3988 | |
3989 | out_put_css: | |
b5557c4c | 3990 | css_put(css); |
79bd9814 TH |
3991 | out_put_cfile: |
3992 | fdput(cfile); | |
3993 | out_put_eventfd: | |
3994 | eventfd_ctx_put(event->eventfd); | |
3995 | out_put_efile: | |
3996 | fdput(efile); | |
3997 | out_kfree: | |
3998 | kfree(event); | |
3999 | ||
4000 | return ret; | |
4001 | } | |
4002 | ||
241994ed | 4003 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4004 | { |
0eea1030 | 4005 | .name = "usage_in_bytes", |
8c7c6e34 | 4006 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4007 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4008 | }, |
c84872e1 PE |
4009 | { |
4010 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4011 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4012 | .write = mem_cgroup_reset, |
791badbd | 4013 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4014 | }, |
8cdea7c0 | 4015 | { |
0eea1030 | 4016 | .name = "limit_in_bytes", |
8c7c6e34 | 4017 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4018 | .write = mem_cgroup_write, |
791badbd | 4019 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4020 | }, |
296c81d8 BS |
4021 | { |
4022 | .name = "soft_limit_in_bytes", | |
4023 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4024 | .write = mem_cgroup_write, |
791badbd | 4025 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4026 | }, |
8cdea7c0 BS |
4027 | { |
4028 | .name = "failcnt", | |
8c7c6e34 | 4029 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4030 | .write = mem_cgroup_reset, |
791badbd | 4031 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4032 | }, |
d2ceb9b7 KH |
4033 | { |
4034 | .name = "stat", | |
2da8ca82 | 4035 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4036 | }, |
c1e862c1 KH |
4037 | { |
4038 | .name = "force_empty", | |
6770c64e | 4039 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4040 | }, |
18f59ea7 BS |
4041 | { |
4042 | .name = "use_hierarchy", | |
4043 | .write_u64 = mem_cgroup_hierarchy_write, | |
4044 | .read_u64 = mem_cgroup_hierarchy_read, | |
4045 | }, | |
79bd9814 | 4046 | { |
3bc942f3 | 4047 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4048 | .write = memcg_write_event_control, |
7dbdb199 | 4049 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 4050 | }, |
a7885eb8 KM |
4051 | { |
4052 | .name = "swappiness", | |
4053 | .read_u64 = mem_cgroup_swappiness_read, | |
4054 | .write_u64 = mem_cgroup_swappiness_write, | |
4055 | }, | |
7dc74be0 DN |
4056 | { |
4057 | .name = "move_charge_at_immigrate", | |
4058 | .read_u64 = mem_cgroup_move_charge_read, | |
4059 | .write_u64 = mem_cgroup_move_charge_write, | |
4060 | }, | |
9490ff27 KH |
4061 | { |
4062 | .name = "oom_control", | |
2da8ca82 | 4063 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4064 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4065 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4066 | }, | |
70ddf637 AV |
4067 | { |
4068 | .name = "pressure_level", | |
70ddf637 | 4069 | }, |
406eb0c9 YH |
4070 | #ifdef CONFIG_NUMA |
4071 | { | |
4072 | .name = "numa_stat", | |
2da8ca82 | 4073 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4074 | }, |
4075 | #endif | |
510fc4e1 GC |
4076 | #ifdef CONFIG_MEMCG_KMEM |
4077 | { | |
4078 | .name = "kmem.limit_in_bytes", | |
4079 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4080 | .write = mem_cgroup_write, |
791badbd | 4081 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4082 | }, |
4083 | { | |
4084 | .name = "kmem.usage_in_bytes", | |
4085 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4086 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4087 | }, |
4088 | { | |
4089 | .name = "kmem.failcnt", | |
4090 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4091 | .write = mem_cgroup_reset, |
791badbd | 4092 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4093 | }, |
4094 | { | |
4095 | .name = "kmem.max_usage_in_bytes", | |
4096 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4097 | .write = mem_cgroup_reset, |
791badbd | 4098 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4099 | }, |
749c5415 GC |
4100 | #ifdef CONFIG_SLABINFO |
4101 | { | |
4102 | .name = "kmem.slabinfo", | |
b047501c VD |
4103 | .seq_start = slab_start, |
4104 | .seq_next = slab_next, | |
4105 | .seq_stop = slab_stop, | |
4106 | .seq_show = memcg_slab_show, | |
749c5415 GC |
4107 | }, |
4108 | #endif | |
8c7c6e34 | 4109 | #endif |
6bc10349 | 4110 | { }, /* terminate */ |
af36f906 | 4111 | }; |
8c7c6e34 | 4112 | |
8627c775 JW |
4113 | /* |
4114 | * Private memory cgroup IDR | |
4115 | * | |
4116 | * Swap-out records and page cache shadow entries need to store memcg | |
4117 | * references in constrained space, so we maintain an ID space that is | |
4118 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
4119 | * memory-controlled cgroups to 64k. | |
4120 | * | |
4121 | * However, there usually are many references to the oflline CSS after | |
4122 | * the cgroup has been destroyed, such as page cache or reclaimable | |
4123 | * slab objects, that don't need to hang on to the ID. We want to keep | |
4124 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
4125 | * relatively small ID space and prevent the creation of new cgroups | |
4126 | * even when there are much fewer than 64k cgroups - possibly none. | |
4127 | * | |
4128 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
4129 | * be freed and recycled when it's no longer needed, which is usually | |
4130 | * when the CSS is offlined. | |
4131 | * | |
4132 | * The only exception to that are records of swapped out tmpfs/shmem | |
4133 | * pages that need to be attributed to live ancestors on swapin. But | |
4134 | * those references are manageable from userspace. | |
4135 | */ | |
4136 | ||
4137 | static DEFINE_IDR(mem_cgroup_idr); | |
4138 | ||
eccccb42 | 4139 | static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n) |
8627c775 | 4140 | { |
eccccb42 | 4141 | atomic_add(n, &memcg->id.ref); |
8627c775 JW |
4142 | } |
4143 | ||
a0fddee3 VD |
4144 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
4145 | { | |
4146 | while (!atomic_inc_not_zero(&memcg->id.ref)) { | |
4147 | /* | |
4148 | * The root cgroup cannot be destroyed, so it's refcount must | |
4149 | * always be >= 1. | |
4150 | */ | |
4151 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
4152 | VM_BUG_ON(1); | |
4153 | break; | |
4154 | } | |
4155 | memcg = parent_mem_cgroup(memcg); | |
4156 | if (!memcg) | |
4157 | memcg = root_mem_cgroup; | |
4158 | } | |
4159 | return memcg; | |
4160 | } | |
4161 | ||
eccccb42 | 4162 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
8627c775 | 4163 | { |
eccccb42 | 4164 | if (atomic_sub_and_test(n, &memcg->id.ref)) { |
8627c775 JW |
4165 | idr_remove(&mem_cgroup_idr, memcg->id.id); |
4166 | memcg->id.id = 0; | |
4167 | ||
4168 | /* Memcg ID pins CSS */ | |
4169 | css_put(&memcg->css); | |
4170 | } | |
4171 | } | |
4172 | ||
eccccb42 VD |
4173 | static inline void mem_cgroup_id_get(struct mem_cgroup *memcg) |
4174 | { | |
4175 | mem_cgroup_id_get_many(memcg, 1); | |
4176 | } | |
4177 | ||
4178 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) | |
4179 | { | |
4180 | mem_cgroup_id_put_many(memcg, 1); | |
4181 | } | |
4182 | ||
8627c775 JW |
4183 | /** |
4184 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
4185 | * @id: the memcg id to look up | |
4186 | * | |
4187 | * Caller must hold rcu_read_lock(). | |
4188 | */ | |
4189 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
4190 | { | |
4191 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
4192 | return idr_find(&mem_cgroup_idr, id); | |
4193 | } | |
4194 | ||
c0ff4b85 | 4195 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4196 | { |
4197 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4198 | struct mem_cgroup_per_zone *mz; |
41e3355d | 4199 | int zone, tmp = node; |
1ecaab2b KH |
4200 | /* |
4201 | * This routine is called against possible nodes. | |
4202 | * But it's BUG to call kmalloc() against offline node. | |
4203 | * | |
4204 | * TODO: this routine can waste much memory for nodes which will | |
4205 | * never be onlined. It's better to use memory hotplug callback | |
4206 | * function. | |
4207 | */ | |
41e3355d KH |
4208 | if (!node_state(node, N_NORMAL_MEMORY)) |
4209 | tmp = -1; | |
17295c88 | 4210 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4211 | if (!pn) |
4212 | return 1; | |
1ecaab2b | 4213 | |
1ecaab2b KH |
4214 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4215 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 4216 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
4217 | mz->usage_in_excess = 0; |
4218 | mz->on_tree = false; | |
d79154bb | 4219 | mz->memcg = memcg; |
1ecaab2b | 4220 | } |
54f72fe0 | 4221 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4222 | return 0; |
4223 | } | |
4224 | ||
c0ff4b85 | 4225 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4226 | { |
54f72fe0 | 4227 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
4228 | } |
4229 | ||
33327948 KH |
4230 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4231 | { | |
d79154bb | 4232 | struct mem_cgroup *memcg; |
8ff69e2c | 4233 | size_t size; |
33327948 | 4234 | |
8ff69e2c VD |
4235 | size = sizeof(struct mem_cgroup); |
4236 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
33327948 | 4237 | |
8ff69e2c | 4238 | memcg = kzalloc(size, GFP_KERNEL); |
d79154bb | 4239 | if (!memcg) |
e7bbcdf3 DC |
4240 | return NULL; |
4241 | ||
d79154bb HD |
4242 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
4243 | if (!memcg->stat) | |
d2e61b8d | 4244 | goto out_free; |
841710aa TH |
4245 | |
4246 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) | |
4247 | goto out_free_stat; | |
4248 | ||
8627c775 JW |
4249 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
4250 | 1, MEM_CGROUP_ID_MAX, | |
4251 | GFP_KERNEL); | |
4252 | if (memcg->id.id < 0) | |
4253 | goto out_free_stat; | |
4254 | ||
d79154bb | 4255 | return memcg; |
d2e61b8d | 4256 | |
841710aa TH |
4257 | out_free_stat: |
4258 | free_percpu(memcg->stat); | |
d2e61b8d | 4259 | out_free: |
8ff69e2c | 4260 | kfree(memcg); |
d2e61b8d | 4261 | return NULL; |
33327948 KH |
4262 | } |
4263 | ||
59927fb9 | 4264 | /* |
c8b2a36f GC |
4265 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
4266 | * (scanning all at force_empty is too costly...) | |
4267 | * | |
4268 | * Instead of clearing all references at force_empty, we remember | |
4269 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4270 | * it goes down to 0. | |
4271 | * | |
4272 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 4273 | */ |
c8b2a36f GC |
4274 | |
4275 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 4276 | { |
c8b2a36f | 4277 | int node; |
59927fb9 | 4278 | |
bb4cc1a8 | 4279 | mem_cgroup_remove_from_trees(memcg); |
c8b2a36f GC |
4280 | |
4281 | for_each_node(node) | |
4282 | free_mem_cgroup_per_zone_info(memcg, node); | |
4283 | ||
4284 | free_percpu(memcg->stat); | |
841710aa | 4285 | memcg_wb_domain_exit(memcg); |
8ff69e2c | 4286 | kfree(memcg); |
59927fb9 | 4287 | } |
3afe36b1 | 4288 | |
7bcc1bb1 DN |
4289 | /* |
4290 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4291 | */ | |
e1aab161 | 4292 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 4293 | { |
3e32cb2e | 4294 | if (!memcg->memory.parent) |
7bcc1bb1 | 4295 | return NULL; |
3e32cb2e | 4296 | return mem_cgroup_from_counter(memcg->memory.parent, memory); |
7bcc1bb1 | 4297 | } |
e1aab161 | 4298 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 4299 | |
0eb253e2 | 4300 | static struct cgroup_subsys_state * __ref |
eb95419b | 4301 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 4302 | { |
d142e3e6 | 4303 | struct mem_cgroup *memcg; |
04046e1a | 4304 | long error = -ENOMEM; |
6d12e2d8 | 4305 | int node; |
8cdea7c0 | 4306 | |
c0ff4b85 R |
4307 | memcg = mem_cgroup_alloc(); |
4308 | if (!memcg) | |
04046e1a | 4309 | return ERR_PTR(error); |
78fb7466 | 4310 | |
3ed28fa1 | 4311 | for_each_node(node) |
c0ff4b85 | 4312 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 4313 | goto free_out; |
f64c3f54 | 4314 | |
c077719b | 4315 | /* root ? */ |
eb95419b | 4316 | if (parent_css == NULL) { |
a41c58a6 | 4317 | root_mem_cgroup = memcg; |
56161634 | 4318 | mem_cgroup_root_css = &memcg->css; |
3e32cb2e | 4319 | page_counter_init(&memcg->memory, NULL); |
241994ed | 4320 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4321 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4322 | page_counter_init(&memcg->memsw, NULL); |
4323 | page_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 4324 | } |
28dbc4b6 | 4325 | |
d142e3e6 GC |
4326 | memcg->last_scanned_node = MAX_NUMNODES; |
4327 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4328 | memcg->move_charge_at_immigrate = 0; |
4329 | mutex_init(&memcg->thresholds_lock); | |
4330 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4331 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4332 | INIT_LIST_HEAD(&memcg->event_list); |
4333 | spin_lock_init(&memcg->event_list_lock); | |
900a38f0 VD |
4334 | #ifdef CONFIG_MEMCG_KMEM |
4335 | memcg->kmemcg_id = -1; | |
900a38f0 | 4336 | #endif |
52ebea74 TH |
4337 | #ifdef CONFIG_CGROUP_WRITEBACK |
4338 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
4339 | #endif | |
8627c775 | 4340 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
d142e3e6 GC |
4341 | return &memcg->css; |
4342 | ||
4343 | free_out: | |
8627c775 | 4344 | idr_remove(&mem_cgroup_idr, memcg->id.id); |
d142e3e6 GC |
4345 | __mem_cgroup_free(memcg); |
4346 | return ERR_PTR(error); | |
4347 | } | |
4348 | ||
4349 | static int | |
eb95419b | 4350 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 4351 | { |
eb95419b | 4352 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 4353 | struct mem_cgroup *parent = mem_cgroup_from_css(css->parent); |
2f7dd7a4 | 4354 | int ret; |
d142e3e6 | 4355 | |
8627c775 JW |
4356 | /* Online state pins memcg ID, memcg ID pins CSS */ |
4357 | mem_cgroup_id_get(mem_cgroup_from_css(css)); | |
4358 | css_get(css); | |
4219b2da | 4359 | |
63876986 | 4360 | if (!parent) |
d142e3e6 GC |
4361 | return 0; |
4362 | ||
0999821b | 4363 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
4364 | |
4365 | memcg->use_hierarchy = parent->use_hierarchy; | |
4366 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4367 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4368 | ||
4369 | if (parent->use_hierarchy) { | |
3e32cb2e | 4370 | page_counter_init(&memcg->memory, &parent->memory); |
241994ed | 4371 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4372 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4373 | page_counter_init(&memcg->memsw, &parent->memsw); |
4374 | page_counter_init(&memcg->kmem, &parent->kmem); | |
55007d84 | 4375 | |
7bcc1bb1 | 4376 | /* |
8d76a979 LZ |
4377 | * No need to take a reference to the parent because cgroup |
4378 | * core guarantees its existence. | |
7bcc1bb1 | 4379 | */ |
18f59ea7 | 4380 | } else { |
3e32cb2e | 4381 | page_counter_init(&memcg->memory, NULL); |
241994ed | 4382 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4383 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4384 | page_counter_init(&memcg->memsw, NULL); |
4385 | page_counter_init(&memcg->kmem, NULL); | |
8c7f6edb TH |
4386 | /* |
4387 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4388 | * much sense so let cgroup subsystem know about this | |
4389 | * unfortunate state in our controller. | |
4390 | */ | |
d142e3e6 | 4391 | if (parent != root_mem_cgroup) |
073219e9 | 4392 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4393 | } |
0999821b | 4394 | mutex_unlock(&memcg_create_mutex); |
d6441637 | 4395 | |
2f7dd7a4 JW |
4396 | ret = memcg_init_kmem(memcg, &memory_cgrp_subsys); |
4397 | if (ret) | |
4398 | return ret; | |
4399 | ||
4400 | /* | |
4401 | * Make sure the memcg is initialized: mem_cgroup_iter() | |
4402 | * orders reading memcg->initialized against its callers | |
4403 | * reading the memcg members. | |
4404 | */ | |
4405 | smp_store_release(&memcg->initialized, 1); | |
4406 | ||
4407 | return 0; | |
8cdea7c0 BS |
4408 | } |
4409 | ||
eb95419b | 4410 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4411 | { |
eb95419b | 4412 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4413 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4414 | |
4415 | /* | |
4416 | * Unregister events and notify userspace. | |
4417 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4418 | * directory to avoid race between userspace and kernelspace. | |
4419 | */ | |
fba94807 TH |
4420 | spin_lock(&memcg->event_list_lock); |
4421 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4422 | list_del_init(&event->list); |
4423 | schedule_work(&event->remove); | |
4424 | } | |
fba94807 | 4425 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4426 | |
33cb876e | 4427 | vmpressure_cleanup(&memcg->vmpressure); |
2a4db7eb VD |
4428 | |
4429 | memcg_deactivate_kmem(memcg); | |
52ebea74 TH |
4430 | |
4431 | wb_memcg_offline(memcg); | |
8627c775 JW |
4432 | |
4433 | mem_cgroup_id_put(memcg); | |
df878fb0 KH |
4434 | } |
4435 | ||
6df38689 VD |
4436 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
4437 | { | |
4438 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4439 | ||
4440 | invalidate_reclaim_iterators(memcg); | |
4441 | } | |
4442 | ||
eb95419b | 4443 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4444 | { |
eb95419b | 4445 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4446 | |
10d5ebf4 | 4447 | memcg_destroy_kmem(memcg); |
465939a1 | 4448 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
4449 | } |
4450 | ||
1ced953b TH |
4451 | /** |
4452 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4453 | * @css: the target css | |
4454 | * | |
4455 | * Reset the states of the mem_cgroup associated with @css. This is | |
4456 | * invoked when the userland requests disabling on the default hierarchy | |
4457 | * but the memcg is pinned through dependency. The memcg should stop | |
4458 | * applying policies and should revert to the vanilla state as it may be | |
4459 | * made visible again. | |
4460 | * | |
4461 | * The current implementation only resets the essential configurations. | |
4462 | * This needs to be expanded to cover all the visible parts. | |
4463 | */ | |
4464 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4465 | { | |
4466 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4467 | ||
3e32cb2e JW |
4468 | mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX); |
4469 | mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX); | |
4470 | memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX); | |
241994ed JW |
4471 | memcg->low = 0; |
4472 | memcg->high = PAGE_COUNTER_MAX; | |
24d404dc | 4473 | memcg->soft_limit = PAGE_COUNTER_MAX; |
2529bb3a | 4474 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
4475 | } |
4476 | ||
02491447 | 4477 | #ifdef CONFIG_MMU |
7dc74be0 | 4478 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4479 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4480 | { |
05b84301 | 4481 | int ret; |
9476db97 | 4482 | |
d0164adc MG |
4483 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
4484 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 4485 | if (!ret) { |
854ffa8d | 4486 | mc.precharge += count; |
854ffa8d DN |
4487 | return ret; |
4488 | } | |
9476db97 JW |
4489 | |
4490 | /* Try charges one by one with reclaim */ | |
854ffa8d | 4491 | while (count--) { |
00501b53 | 4492 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1); |
38c5d72f | 4493 | if (ret) |
38c5d72f | 4494 | return ret; |
854ffa8d | 4495 | mc.precharge++; |
9476db97 | 4496 | cond_resched(); |
854ffa8d | 4497 | } |
9476db97 | 4498 | return 0; |
4ffef5fe DN |
4499 | } |
4500 | ||
4501 | /** | |
8d32ff84 | 4502 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
4503 | * @vma: the vma the pte to be checked belongs |
4504 | * @addr: the address corresponding to the pte to be checked | |
4505 | * @ptent: the pte to be checked | |
02491447 | 4506 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
4507 | * |
4508 | * Returns | |
4509 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4510 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4511 | * move charge. if @target is not NULL, the page is stored in target->page | |
4512 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
4513 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
4514 | * target for charge migration. if @target is not NULL, the entry is stored | |
4515 | * in target->ent. | |
4ffef5fe DN |
4516 | * |
4517 | * Called with pte lock held. | |
4518 | */ | |
4ffef5fe DN |
4519 | union mc_target { |
4520 | struct page *page; | |
02491447 | 4521 | swp_entry_t ent; |
4ffef5fe DN |
4522 | }; |
4523 | ||
4ffef5fe | 4524 | enum mc_target_type { |
8d32ff84 | 4525 | MC_TARGET_NONE = 0, |
4ffef5fe | 4526 | MC_TARGET_PAGE, |
02491447 | 4527 | MC_TARGET_SWAP, |
4ffef5fe DN |
4528 | }; |
4529 | ||
90254a65 DN |
4530 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4531 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4532 | { |
90254a65 | 4533 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 4534 | |
90254a65 DN |
4535 | if (!page || !page_mapped(page)) |
4536 | return NULL; | |
4537 | if (PageAnon(page)) { | |
1dfab5ab | 4538 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 4539 | return NULL; |
1dfab5ab JW |
4540 | } else { |
4541 | if (!(mc.flags & MOVE_FILE)) | |
4542 | return NULL; | |
4543 | } | |
90254a65 DN |
4544 | if (!get_page_unless_zero(page)) |
4545 | return NULL; | |
4546 | ||
4547 | return page; | |
4548 | } | |
4549 | ||
4b91355e | 4550 | #ifdef CONFIG_SWAP |
90254a65 DN |
4551 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
4552 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4553 | { | |
90254a65 DN |
4554 | struct page *page = NULL; |
4555 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4556 | ||
1dfab5ab | 4557 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 4558 | return NULL; |
4b91355e KH |
4559 | /* |
4560 | * Because lookup_swap_cache() updates some statistics counter, | |
4561 | * we call find_get_page() with swapper_space directly. | |
4562 | */ | |
33806f06 | 4563 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
4564 | if (do_swap_account) |
4565 | entry->val = ent.val; | |
4566 | ||
4567 | return page; | |
4568 | } | |
4b91355e KH |
4569 | #else |
4570 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
4571 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4572 | { | |
4573 | return NULL; | |
4574 | } | |
4575 | #endif | |
90254a65 | 4576 | |
87946a72 DN |
4577 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4578 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4579 | { | |
4580 | struct page *page = NULL; | |
87946a72 DN |
4581 | struct address_space *mapping; |
4582 | pgoff_t pgoff; | |
4583 | ||
4584 | if (!vma->vm_file) /* anonymous vma */ | |
4585 | return NULL; | |
1dfab5ab | 4586 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
4587 | return NULL; |
4588 | ||
87946a72 | 4589 | mapping = vma->vm_file->f_mapping; |
0661a336 | 4590 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
4591 | |
4592 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
4593 | #ifdef CONFIG_SWAP |
4594 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
4595 | if (shmem_mapping(mapping)) { |
4596 | page = find_get_entry(mapping, pgoff); | |
4597 | if (radix_tree_exceptional_entry(page)) { | |
4598 | swp_entry_t swp = radix_to_swp_entry(page); | |
4599 | if (do_swap_account) | |
4600 | *entry = swp; | |
4601 | page = find_get_page(swap_address_space(swp), swp.val); | |
4602 | } | |
4603 | } else | |
4604 | page = find_get_page(mapping, pgoff); | |
4605 | #else | |
4606 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 4607 | #endif |
87946a72 DN |
4608 | return page; |
4609 | } | |
4610 | ||
b1b0deab CG |
4611 | /** |
4612 | * mem_cgroup_move_account - move account of the page | |
4613 | * @page: the page | |
4614 | * @nr_pages: number of regular pages (>1 for huge pages) | |
4615 | * @from: mem_cgroup which the page is moved from. | |
4616 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
4617 | * | |
4618 | * The caller must confirm following. | |
4619 | * - page is not on LRU (isolate_page() is useful.) | |
4620 | * - compound_lock is held when nr_pages > 1 | |
4621 | * | |
4622 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
4623 | * from old cgroup. | |
4624 | */ | |
4625 | static int mem_cgroup_move_account(struct page *page, | |
4626 | unsigned int nr_pages, | |
4627 | struct mem_cgroup *from, | |
4628 | struct mem_cgroup *to) | |
4629 | { | |
4630 | unsigned long flags; | |
4631 | int ret; | |
c4843a75 | 4632 | bool anon; |
b1b0deab CG |
4633 | |
4634 | VM_BUG_ON(from == to); | |
4635 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
4636 | /* | |
4637 | * The page is isolated from LRU. So, collapse function | |
4638 | * will not handle this page. But page splitting can happen. | |
4639 | * Do this check under compound_page_lock(). The caller should | |
4640 | * hold it. | |
4641 | */ | |
4642 | ret = -EBUSY; | |
4643 | if (nr_pages > 1 && !PageTransHuge(page)) | |
4644 | goto out; | |
4645 | ||
4646 | /* | |
45637bab HD |
4647 | * Prevent mem_cgroup_replace_page() from looking at |
4648 | * page->mem_cgroup of its source page while we change it. | |
b1b0deab CG |
4649 | */ |
4650 | if (!trylock_page(page)) | |
4651 | goto out; | |
4652 | ||
4653 | ret = -EINVAL; | |
4654 | if (page->mem_cgroup != from) | |
4655 | goto out_unlock; | |
4656 | ||
c4843a75 GT |
4657 | anon = PageAnon(page); |
4658 | ||
b1b0deab CG |
4659 | spin_lock_irqsave(&from->move_lock, flags); |
4660 | ||
c4843a75 | 4661 | if (!anon && page_mapped(page)) { |
b1b0deab CG |
4662 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
4663 | nr_pages); | |
4664 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
4665 | nr_pages); | |
4666 | } | |
4667 | ||
c4843a75 GT |
4668 | /* |
4669 | * move_lock grabbed above and caller set from->moving_account, so | |
4670 | * mem_cgroup_update_page_stat() will serialize updates to PageDirty. | |
4671 | * So mapping should be stable for dirty pages. | |
4672 | */ | |
4673 | if (!anon && PageDirty(page)) { | |
4674 | struct address_space *mapping = page_mapping(page); | |
4675 | ||
4676 | if (mapping_cap_account_dirty(mapping)) { | |
4677 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY], | |
4678 | nr_pages); | |
4679 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY], | |
4680 | nr_pages); | |
4681 | } | |
4682 | } | |
4683 | ||
b1b0deab CG |
4684 | if (PageWriteback(page)) { |
4685 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
4686 | nr_pages); | |
4687 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
4688 | nr_pages); | |
4689 | } | |
4690 | ||
4691 | /* | |
4692 | * It is safe to change page->mem_cgroup here because the page | |
4693 | * is referenced, charged, and isolated - we can't race with | |
4694 | * uncharging, charging, migration, or LRU putback. | |
4695 | */ | |
4696 | ||
4697 | /* caller should have done css_get */ | |
4698 | page->mem_cgroup = to; | |
4699 | spin_unlock_irqrestore(&from->move_lock, flags); | |
4700 | ||
4701 | ret = 0; | |
4702 | ||
4703 | local_irq_disable(); | |
4704 | mem_cgroup_charge_statistics(to, page, nr_pages); | |
4705 | memcg_check_events(to, page); | |
4706 | mem_cgroup_charge_statistics(from, page, -nr_pages); | |
4707 | memcg_check_events(from, page); | |
4708 | local_irq_enable(); | |
4709 | out_unlock: | |
4710 | unlock_page(page); | |
4711 | out: | |
4712 | return ret; | |
4713 | } | |
4714 | ||
8d32ff84 | 4715 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
4716 | unsigned long addr, pte_t ptent, union mc_target *target) |
4717 | { | |
4718 | struct page *page = NULL; | |
8d32ff84 | 4719 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
4720 | swp_entry_t ent = { .val = 0 }; |
4721 | ||
4722 | if (pte_present(ptent)) | |
4723 | page = mc_handle_present_pte(vma, addr, ptent); | |
4724 | else if (is_swap_pte(ptent)) | |
4725 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
0661a336 | 4726 | else if (pte_none(ptent)) |
87946a72 | 4727 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
4728 | |
4729 | if (!page && !ent.val) | |
8d32ff84 | 4730 | return ret; |
02491447 | 4731 | if (page) { |
02491447 | 4732 | /* |
0a31bc97 | 4733 | * Do only loose check w/o serialization. |
1306a85a | 4734 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 4735 | * not under LRU exclusion. |
02491447 | 4736 | */ |
1306a85a | 4737 | if (page->mem_cgroup == mc.from) { |
02491447 DN |
4738 | ret = MC_TARGET_PAGE; |
4739 | if (target) | |
4740 | target->page = page; | |
4741 | } | |
4742 | if (!ret || !target) | |
4743 | put_page(page); | |
4744 | } | |
90254a65 DN |
4745 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
4746 | if (ent.val && !ret && | |
34c00c31 | 4747 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
4748 | ret = MC_TARGET_SWAP; |
4749 | if (target) | |
4750 | target->ent = ent; | |
4ffef5fe | 4751 | } |
4ffef5fe DN |
4752 | return ret; |
4753 | } | |
4754 | ||
12724850 NH |
4755 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4756 | /* | |
4757 | * We don't consider swapping or file mapped pages because THP does not | |
4758 | * support them for now. | |
4759 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
4760 | */ | |
4761 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4762 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4763 | { | |
4764 | struct page *page = NULL; | |
12724850 NH |
4765 | enum mc_target_type ret = MC_TARGET_NONE; |
4766 | ||
4767 | page = pmd_page(pmd); | |
309381fe | 4768 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 4769 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 4770 | return ret; |
1306a85a | 4771 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
4772 | ret = MC_TARGET_PAGE; |
4773 | if (target) { | |
4774 | get_page(page); | |
4775 | target->page = page; | |
4776 | } | |
4777 | } | |
4778 | return ret; | |
4779 | } | |
4780 | #else | |
4781 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4782 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4783 | { | |
4784 | return MC_TARGET_NONE; | |
4785 | } | |
4786 | #endif | |
4787 | ||
4ffef5fe DN |
4788 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
4789 | unsigned long addr, unsigned long end, | |
4790 | struct mm_walk *walk) | |
4791 | { | |
26bcd64a | 4792 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4793 | pte_t *pte; |
4794 | spinlock_t *ptl; | |
4795 | ||
bf929152 | 4796 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
12724850 NH |
4797 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
4798 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 4799 | spin_unlock(ptl); |
1a5a9906 | 4800 | return 0; |
12724850 | 4801 | } |
03319327 | 4802 | |
45f83cef AA |
4803 | if (pmd_trans_unstable(pmd)) |
4804 | return 0; | |
4ffef5fe DN |
4805 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
4806 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 4807 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
4808 | mc.precharge++; /* increment precharge temporarily */ |
4809 | pte_unmap_unlock(pte - 1, ptl); | |
4810 | cond_resched(); | |
4811 | ||
7dc74be0 DN |
4812 | return 0; |
4813 | } | |
4814 | ||
4ffef5fe DN |
4815 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
4816 | { | |
4817 | unsigned long precharge; | |
4ffef5fe | 4818 | |
26bcd64a NH |
4819 | struct mm_walk mem_cgroup_count_precharge_walk = { |
4820 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
4821 | .mm = mm, | |
4822 | }; | |
dfe076b0 | 4823 | down_read(&mm->mmap_sem); |
26bcd64a | 4824 | walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk); |
dfe076b0 | 4825 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
4826 | |
4827 | precharge = mc.precharge; | |
4828 | mc.precharge = 0; | |
4829 | ||
4830 | return precharge; | |
4831 | } | |
4832 | ||
4ffef5fe DN |
4833 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
4834 | { | |
dfe076b0 DN |
4835 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
4836 | ||
4837 | VM_BUG_ON(mc.moving_task); | |
4838 | mc.moving_task = current; | |
4839 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
4840 | } |
4841 | ||
dfe076b0 DN |
4842 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
4843 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 4844 | { |
2bd9bb20 KH |
4845 | struct mem_cgroup *from = mc.from; |
4846 | struct mem_cgroup *to = mc.to; | |
4847 | ||
4ffef5fe | 4848 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 4849 | if (mc.precharge) { |
00501b53 | 4850 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
4851 | mc.precharge = 0; |
4852 | } | |
4853 | /* | |
4854 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
4855 | * we must uncharge here. | |
4856 | */ | |
4857 | if (mc.moved_charge) { | |
00501b53 | 4858 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 4859 | mc.moved_charge = 0; |
4ffef5fe | 4860 | } |
483c30b5 DN |
4861 | /* we must fixup refcnts and charges */ |
4862 | if (mc.moved_swap) { | |
483c30b5 | 4863 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 4864 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 4865 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 4866 | |
eccccb42 VD |
4867 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
4868 | ||
05b84301 | 4869 | /* |
3e32cb2e JW |
4870 | * we charged both to->memory and to->memsw, so we |
4871 | * should uncharge to->memory. | |
05b84301 | 4872 | */ |
ce00a967 | 4873 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
4874 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
4875 | ||
eccccb42 VD |
4876 | mem_cgroup_id_get_many(mc.to, mc.moved_swap); |
4877 | css_put_many(&mc.to->css, mc.moved_swap); | |
3e32cb2e | 4878 | |
483c30b5 DN |
4879 | mc.moved_swap = 0; |
4880 | } | |
dfe076b0 DN |
4881 | memcg_oom_recover(from); |
4882 | memcg_oom_recover(to); | |
4883 | wake_up_all(&mc.waitq); | |
4884 | } | |
4885 | ||
4886 | static void mem_cgroup_clear_mc(void) | |
4887 | { | |
52526076 TH |
4888 | struct mm_struct *mm = mc.mm; |
4889 | ||
dfe076b0 DN |
4890 | /* |
4891 | * we must clear moving_task before waking up waiters at the end of | |
4892 | * task migration. | |
4893 | */ | |
4894 | mc.moving_task = NULL; | |
4895 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 4896 | spin_lock(&mc.lock); |
4ffef5fe DN |
4897 | mc.from = NULL; |
4898 | mc.to = NULL; | |
52526076 | 4899 | mc.mm = NULL; |
2bd9bb20 | 4900 | spin_unlock(&mc.lock); |
52526076 TH |
4901 | |
4902 | mmput(mm); | |
4ffef5fe DN |
4903 | } |
4904 | ||
1f7dd3e5 | 4905 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 4906 | { |
1f7dd3e5 TH |
4907 | struct cgroup_subsys_state *css; |
4908 | struct mem_cgroup *memcg; | |
9f2115f9 | 4909 | struct mem_cgroup *from; |
4530eddb | 4910 | struct task_struct *leader, *p; |
9f2115f9 | 4911 | struct mm_struct *mm; |
1dfab5ab | 4912 | unsigned long move_flags; |
9f2115f9 | 4913 | int ret = 0; |
7dc74be0 | 4914 | |
1f7dd3e5 TH |
4915 | /* charge immigration isn't supported on the default hierarchy */ |
4916 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
4917 | return 0; |
4918 | ||
4530eddb TH |
4919 | /* |
4920 | * Multi-process migrations only happen on the default hierarchy | |
4921 | * where charge immigration is not used. Perform charge | |
4922 | * immigration if @tset contains a leader and whine if there are | |
4923 | * multiple. | |
4924 | */ | |
4925 | p = NULL; | |
1f7dd3e5 | 4926 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
4927 | WARN_ON_ONCE(p); |
4928 | p = leader; | |
1f7dd3e5 | 4929 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
4930 | } |
4931 | if (!p) | |
4932 | return 0; | |
4933 | ||
1f7dd3e5 TH |
4934 | /* |
4935 | * We are now commited to this value whatever it is. Changes in this | |
4936 | * tunable will only affect upcoming migrations, not the current one. | |
4937 | * So we need to save it, and keep it going. | |
4938 | */ | |
4939 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
4940 | if (!move_flags) | |
4941 | return 0; | |
4942 | ||
9f2115f9 TH |
4943 | from = mem_cgroup_from_task(p); |
4944 | ||
4945 | VM_BUG_ON(from == memcg); | |
4946 | ||
4947 | mm = get_task_mm(p); | |
4948 | if (!mm) | |
4949 | return 0; | |
4950 | /* We move charges only when we move a owner of the mm */ | |
4951 | if (mm->owner == p) { | |
4952 | VM_BUG_ON(mc.from); | |
4953 | VM_BUG_ON(mc.to); | |
4954 | VM_BUG_ON(mc.precharge); | |
4955 | VM_BUG_ON(mc.moved_charge); | |
4956 | VM_BUG_ON(mc.moved_swap); | |
4957 | ||
4958 | spin_lock(&mc.lock); | |
52526076 | 4959 | mc.mm = mm; |
9f2115f9 TH |
4960 | mc.from = from; |
4961 | mc.to = memcg; | |
4962 | mc.flags = move_flags; | |
4963 | spin_unlock(&mc.lock); | |
4964 | /* We set mc.moving_task later */ | |
4965 | ||
4966 | ret = mem_cgroup_precharge_mc(mm); | |
4967 | if (ret) | |
4968 | mem_cgroup_clear_mc(); | |
52526076 TH |
4969 | } else { |
4970 | mmput(mm); | |
7dc74be0 DN |
4971 | } |
4972 | return ret; | |
4973 | } | |
4974 | ||
1f7dd3e5 | 4975 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 4976 | { |
4e2f245d JW |
4977 | if (mc.to) |
4978 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
4979 | } |
4980 | ||
4ffef5fe DN |
4981 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
4982 | unsigned long addr, unsigned long end, | |
4983 | struct mm_walk *walk) | |
7dc74be0 | 4984 | { |
4ffef5fe | 4985 | int ret = 0; |
26bcd64a | 4986 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4987 | pte_t *pte; |
4988 | spinlock_t *ptl; | |
12724850 NH |
4989 | enum mc_target_type target_type; |
4990 | union mc_target target; | |
4991 | struct page *page; | |
4ffef5fe | 4992 | |
12724850 NH |
4993 | /* |
4994 | * We don't take compound_lock() here but no race with splitting thp | |
4995 | * happens because: | |
4996 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
4997 | * under splitting, which means there's no concurrent thp split, | |
4998 | * - if another thread runs into split_huge_page() just after we | |
4999 | * entered this if-block, the thread must wait for page table lock | |
5000 | * to be unlocked in __split_huge_page_splitting(), where the main | |
5001 | * part of thp split is not executed yet. | |
5002 | */ | |
bf929152 | 5003 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
62ade86a | 5004 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5005 | spin_unlock(ptl); |
12724850 NH |
5006 | return 0; |
5007 | } | |
5008 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5009 | if (target_type == MC_TARGET_PAGE) { | |
5010 | page = target.page; | |
5011 | if (!isolate_lru_page(page)) { | |
12724850 | 5012 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, |
1306a85a | 5013 | mc.from, mc.to)) { |
12724850 NH |
5014 | mc.precharge -= HPAGE_PMD_NR; |
5015 | mc.moved_charge += HPAGE_PMD_NR; | |
5016 | } | |
5017 | putback_lru_page(page); | |
5018 | } | |
5019 | put_page(page); | |
5020 | } | |
bf929152 | 5021 | spin_unlock(ptl); |
1a5a9906 | 5022 | return 0; |
12724850 NH |
5023 | } |
5024 | ||
45f83cef AA |
5025 | if (pmd_trans_unstable(pmd)) |
5026 | return 0; | |
4ffef5fe DN |
5027 | retry: |
5028 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5029 | for (; addr != end; addr += PAGE_SIZE) { | |
5030 | pte_t ptent = *(pte++); | |
02491447 | 5031 | swp_entry_t ent; |
4ffef5fe DN |
5032 | |
5033 | if (!mc.precharge) | |
5034 | break; | |
5035 | ||
8d32ff84 | 5036 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
5037 | case MC_TARGET_PAGE: |
5038 | page = target.page; | |
5039 | if (isolate_lru_page(page)) | |
5040 | goto put; | |
1306a85a | 5041 | if (!mem_cgroup_move_account(page, 1, mc.from, mc.to)) { |
4ffef5fe | 5042 | mc.precharge--; |
854ffa8d DN |
5043 | /* we uncharge from mc.from later. */ |
5044 | mc.moved_charge++; | |
4ffef5fe DN |
5045 | } |
5046 | putback_lru_page(page); | |
8d32ff84 | 5047 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5048 | put_page(page); |
5049 | break; | |
02491447 DN |
5050 | case MC_TARGET_SWAP: |
5051 | ent = target.ent; | |
e91cbb42 | 5052 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5053 | mc.precharge--; |
483c30b5 DN |
5054 | /* we fixup refcnts and charges later. */ |
5055 | mc.moved_swap++; | |
5056 | } | |
02491447 | 5057 | break; |
4ffef5fe DN |
5058 | default: |
5059 | break; | |
5060 | } | |
5061 | } | |
5062 | pte_unmap_unlock(pte - 1, ptl); | |
5063 | cond_resched(); | |
5064 | ||
5065 | if (addr != end) { | |
5066 | /* | |
5067 | * We have consumed all precharges we got in can_attach(). | |
5068 | * We try charge one by one, but don't do any additional | |
5069 | * charges to mc.to if we have failed in charge once in attach() | |
5070 | * phase. | |
5071 | */ | |
854ffa8d | 5072 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5073 | if (!ret) |
5074 | goto retry; | |
5075 | } | |
5076 | ||
5077 | return ret; | |
5078 | } | |
5079 | ||
52526076 | 5080 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 5081 | { |
26bcd64a NH |
5082 | struct mm_walk mem_cgroup_move_charge_walk = { |
5083 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
52526076 | 5084 | .mm = mc.mm, |
26bcd64a | 5085 | }; |
4ffef5fe DN |
5086 | |
5087 | lru_add_drain_all(); | |
312722cb JW |
5088 | /* |
5089 | * Signal mem_cgroup_begin_page_stat() to take the memcg's | |
5090 | * move_lock while we're moving its pages to another memcg. | |
5091 | * Then wait for already started RCU-only updates to finish. | |
5092 | */ | |
5093 | atomic_inc(&mc.from->moving_account); | |
5094 | synchronize_rcu(); | |
dfe076b0 | 5095 | retry: |
52526076 | 5096 | if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) { |
dfe076b0 DN |
5097 | /* |
5098 | * Someone who are holding the mmap_sem might be waiting in | |
5099 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5100 | * and retry. Because we cancel precharges, we might not be able | |
5101 | * to move enough charges, but moving charge is a best-effort | |
5102 | * feature anyway, so it wouldn't be a big problem. | |
5103 | */ | |
5104 | __mem_cgroup_clear_mc(); | |
5105 | cond_resched(); | |
5106 | goto retry; | |
5107 | } | |
26bcd64a NH |
5108 | /* |
5109 | * When we have consumed all precharges and failed in doing | |
5110 | * additional charge, the page walk just aborts. | |
5111 | */ | |
5112 | walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk); | |
52526076 | 5113 | up_read(&mc.mm->mmap_sem); |
312722cb | 5114 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
5115 | } |
5116 | ||
52526076 | 5117 | static void mem_cgroup_move_task(void) |
67e465a7 | 5118 | { |
52526076 TH |
5119 | if (mc.to) { |
5120 | mem_cgroup_move_charge(); | |
a433658c | 5121 | mem_cgroup_clear_mc(); |
52526076 | 5122 | } |
67e465a7 | 5123 | } |
5cfb80a7 | 5124 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 5125 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5126 | { |
5127 | return 0; | |
5128 | } | |
1f7dd3e5 | 5129 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5130 | { |
5131 | } | |
52526076 | 5132 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
5133 | { |
5134 | } | |
5135 | #endif | |
67e465a7 | 5136 | |
f00baae7 TH |
5137 | /* |
5138 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5139 | * to verify whether we're attached to the default hierarchy on each mount |
5140 | * attempt. | |
f00baae7 | 5141 | */ |
eb95419b | 5142 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5143 | { |
5144 | /* | |
aa6ec29b | 5145 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5146 | * guarantees that @root doesn't have any children, so turning it |
5147 | * on for the root memcg is enough. | |
5148 | */ | |
9e10a130 | 5149 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
5150 | root_mem_cgroup->use_hierarchy = true; |
5151 | else | |
5152 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
5153 | } |
5154 | ||
241994ed JW |
5155 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
5156 | struct cftype *cft) | |
5157 | { | |
f5fc3c5d JW |
5158 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5159 | ||
5160 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
5161 | } |
5162 | ||
5163 | static int memory_low_show(struct seq_file *m, void *v) | |
5164 | { | |
5165 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5166 | unsigned long low = READ_ONCE(memcg->low); |
241994ed JW |
5167 | |
5168 | if (low == PAGE_COUNTER_MAX) | |
d2973697 | 5169 | seq_puts(m, "max\n"); |
241994ed JW |
5170 | else |
5171 | seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE); | |
5172 | ||
5173 | return 0; | |
5174 | } | |
5175 | ||
5176 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
5177 | char *buf, size_t nbytes, loff_t off) | |
5178 | { | |
5179 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5180 | unsigned long low; | |
5181 | int err; | |
5182 | ||
5183 | buf = strstrip(buf); | |
d2973697 | 5184 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
5185 | if (err) |
5186 | return err; | |
5187 | ||
5188 | memcg->low = low; | |
5189 | ||
5190 | return nbytes; | |
5191 | } | |
5192 | ||
5193 | static int memory_high_show(struct seq_file *m, void *v) | |
5194 | { | |
5195 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5196 | unsigned long high = READ_ONCE(memcg->high); |
241994ed JW |
5197 | |
5198 | if (high == PAGE_COUNTER_MAX) | |
d2973697 | 5199 | seq_puts(m, "max\n"); |
241994ed JW |
5200 | else |
5201 | seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE); | |
5202 | ||
5203 | return 0; | |
5204 | } | |
5205 | ||
5206 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
5207 | char *buf, size_t nbytes, loff_t off) | |
5208 | { | |
5209 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
8b42fc47 | 5210 | unsigned long nr_pages; |
241994ed JW |
5211 | unsigned long high; |
5212 | int err; | |
5213 | ||
5214 | buf = strstrip(buf); | |
d2973697 | 5215 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
5216 | if (err) |
5217 | return err; | |
5218 | ||
5219 | memcg->high = high; | |
5220 | ||
8b42fc47 JW |
5221 | nr_pages = page_counter_read(&memcg->memory); |
5222 | if (nr_pages > high) | |
5223 | try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
5224 | GFP_KERNEL, true); | |
5225 | ||
2529bb3a | 5226 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5227 | return nbytes; |
5228 | } | |
5229 | ||
5230 | static int memory_max_show(struct seq_file *m, void *v) | |
5231 | { | |
5232 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5233 | unsigned long max = READ_ONCE(memcg->memory.limit); |
241994ed JW |
5234 | |
5235 | if (max == PAGE_COUNTER_MAX) | |
d2973697 | 5236 | seq_puts(m, "max\n"); |
241994ed JW |
5237 | else |
5238 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
5239 | ||
5240 | return 0; | |
5241 | } | |
5242 | ||
5243 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
5244 | char *buf, size_t nbytes, loff_t off) | |
5245 | { | |
5246 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
0ccab5b1 JW |
5247 | unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES; |
5248 | bool drained = false; | |
241994ed JW |
5249 | unsigned long max; |
5250 | int err; | |
5251 | ||
5252 | buf = strstrip(buf); | |
d2973697 | 5253 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
5254 | if (err) |
5255 | return err; | |
5256 | ||
0ccab5b1 JW |
5257 | xchg(&memcg->memory.limit, max); |
5258 | ||
5259 | for (;;) { | |
5260 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
5261 | ||
5262 | if (nr_pages <= max) | |
5263 | break; | |
5264 | ||
5265 | if (signal_pending(current)) { | |
5266 | err = -EINTR; | |
5267 | break; | |
5268 | } | |
5269 | ||
5270 | if (!drained) { | |
5271 | drain_all_stock(memcg); | |
5272 | drained = true; | |
5273 | continue; | |
5274 | } | |
5275 | ||
5276 | if (nr_reclaims) { | |
5277 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
5278 | GFP_KERNEL, true)) | |
5279 | nr_reclaims--; | |
5280 | continue; | |
5281 | } | |
5282 | ||
5283 | mem_cgroup_events(memcg, MEMCG_OOM, 1); | |
5284 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) | |
5285 | break; | |
5286 | } | |
241994ed | 5287 | |
2529bb3a | 5288 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5289 | return nbytes; |
5290 | } | |
5291 | ||
5292 | static int memory_events_show(struct seq_file *m, void *v) | |
5293 | { | |
5294 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5295 | ||
5296 | seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW)); | |
5297 | seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH)); | |
5298 | seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX)); | |
5299 | seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM)); | |
5300 | ||
5301 | return 0; | |
5302 | } | |
5303 | ||
5304 | static struct cftype memory_files[] = { | |
5305 | { | |
5306 | .name = "current", | |
f5fc3c5d | 5307 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
5308 | .read_u64 = memory_current_read, |
5309 | }, | |
5310 | { | |
5311 | .name = "low", | |
5312 | .flags = CFTYPE_NOT_ON_ROOT, | |
5313 | .seq_show = memory_low_show, | |
5314 | .write = memory_low_write, | |
5315 | }, | |
5316 | { | |
5317 | .name = "high", | |
5318 | .flags = CFTYPE_NOT_ON_ROOT, | |
5319 | .seq_show = memory_high_show, | |
5320 | .write = memory_high_write, | |
5321 | }, | |
5322 | { | |
5323 | .name = "max", | |
5324 | .flags = CFTYPE_NOT_ON_ROOT, | |
5325 | .seq_show = memory_max_show, | |
5326 | .write = memory_max_write, | |
5327 | }, | |
5328 | { | |
5329 | .name = "events", | |
5330 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 5331 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
5332 | .seq_show = memory_events_show, |
5333 | }, | |
5334 | { } /* terminate */ | |
5335 | }; | |
5336 | ||
073219e9 | 5337 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5338 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5339 | .css_online = mem_cgroup_css_online, |
92fb9748 | 5340 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 5341 | .css_released = mem_cgroup_css_released, |
92fb9748 | 5342 | .css_free = mem_cgroup_css_free, |
1ced953b | 5343 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5344 | .can_attach = mem_cgroup_can_attach, |
5345 | .cancel_attach = mem_cgroup_cancel_attach, | |
52526076 | 5346 | .post_attach = mem_cgroup_move_task, |
f00baae7 | 5347 | .bind = mem_cgroup_bind, |
241994ed JW |
5348 | .dfl_cftypes = memory_files, |
5349 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 5350 | .early_init = 0, |
8cdea7c0 | 5351 | }; |
c077719b | 5352 | |
241994ed JW |
5353 | /** |
5354 | * mem_cgroup_low - check if memory consumption is below the normal range | |
5355 | * @root: the highest ancestor to consider | |
5356 | * @memcg: the memory cgroup to check | |
5357 | * | |
5358 | * Returns %true if memory consumption of @memcg, and that of all | |
5359 | * configurable ancestors up to @root, is below the normal range. | |
5360 | */ | |
5361 | bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg) | |
5362 | { | |
5363 | if (mem_cgroup_disabled()) | |
5364 | return false; | |
5365 | ||
5366 | /* | |
5367 | * The toplevel group doesn't have a configurable range, so | |
5368 | * it's never low when looked at directly, and it is not | |
5369 | * considered an ancestor when assessing the hierarchy. | |
5370 | */ | |
5371 | ||
5372 | if (memcg == root_mem_cgroup) | |
5373 | return false; | |
5374 | ||
4e54dede | 5375 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5376 | return false; |
5377 | ||
5378 | while (memcg != root) { | |
5379 | memcg = parent_mem_cgroup(memcg); | |
5380 | ||
5381 | if (memcg == root_mem_cgroup) | |
5382 | break; | |
5383 | ||
4e54dede | 5384 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5385 | return false; |
5386 | } | |
5387 | return true; | |
5388 | } | |
5389 | ||
00501b53 JW |
5390 | /** |
5391 | * mem_cgroup_try_charge - try charging a page | |
5392 | * @page: page to charge | |
5393 | * @mm: mm context of the victim | |
5394 | * @gfp_mask: reclaim mode | |
5395 | * @memcgp: charged memcg return | |
5396 | * | |
5397 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5398 | * pages according to @gfp_mask if necessary. | |
5399 | * | |
5400 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5401 | * Otherwise, an error code is returned. | |
5402 | * | |
5403 | * After page->mapping has been set up, the caller must finalize the | |
5404 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5405 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5406 | */ | |
5407 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
5408 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
5409 | { | |
5410 | struct mem_cgroup *memcg = NULL; | |
5411 | unsigned int nr_pages = 1; | |
5412 | int ret = 0; | |
5413 | ||
5414 | if (mem_cgroup_disabled()) | |
5415 | goto out; | |
5416 | ||
5417 | if (PageSwapCache(page)) { | |
00501b53 JW |
5418 | /* |
5419 | * Every swap fault against a single page tries to charge the | |
5420 | * page, bail as early as possible. shmem_unuse() encounters | |
5421 | * already charged pages, too. The USED bit is protected by | |
5422 | * the page lock, which serializes swap cache removal, which | |
5423 | * in turn serializes uncharging. | |
5424 | */ | |
e993d905 | 5425 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1306a85a | 5426 | if (page->mem_cgroup) |
00501b53 | 5427 | goto out; |
e993d905 VD |
5428 | |
5429 | if (do_swap_account) { | |
5430 | swp_entry_t ent = { .val = page_private(page), }; | |
5431 | unsigned short id = lookup_swap_cgroup_id(ent); | |
5432 | ||
5433 | rcu_read_lock(); | |
5434 | memcg = mem_cgroup_from_id(id); | |
5435 | if (memcg && !css_tryget_online(&memcg->css)) | |
5436 | memcg = NULL; | |
5437 | rcu_read_unlock(); | |
5438 | } | |
00501b53 JW |
5439 | } |
5440 | ||
5441 | if (PageTransHuge(page)) { | |
5442 | nr_pages <<= compound_order(page); | |
5443 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5444 | } | |
5445 | ||
00501b53 JW |
5446 | if (!memcg) |
5447 | memcg = get_mem_cgroup_from_mm(mm); | |
5448 | ||
5449 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
5450 | ||
5451 | css_put(&memcg->css); | |
00501b53 JW |
5452 | out: |
5453 | *memcgp = memcg; | |
5454 | return ret; | |
5455 | } | |
5456 | ||
5457 | /** | |
5458 | * mem_cgroup_commit_charge - commit a page charge | |
5459 | * @page: page to charge | |
5460 | * @memcg: memcg to charge the page to | |
5461 | * @lrucare: page might be on LRU already | |
5462 | * | |
5463 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
5464 | * after page->mapping has been set up. This must happen atomically | |
5465 | * as part of the page instantiation, i.e. under the page table lock | |
5466 | * for anonymous pages, under the page lock for page and swap cache. | |
5467 | * | |
5468 | * In addition, the page must not be on the LRU during the commit, to | |
5469 | * prevent racing with task migration. If it might be, use @lrucare. | |
5470 | * | |
5471 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
5472 | */ | |
5473 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
5474 | bool lrucare) | |
5475 | { | |
5476 | unsigned int nr_pages = 1; | |
5477 | ||
5478 | VM_BUG_ON_PAGE(!page->mapping, page); | |
5479 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
5480 | ||
5481 | if (mem_cgroup_disabled()) | |
5482 | return; | |
5483 | /* | |
5484 | * Swap faults will attempt to charge the same page multiple | |
5485 | * times. But reuse_swap_page() might have removed the page | |
5486 | * from swapcache already, so we can't check PageSwapCache(). | |
5487 | */ | |
5488 | if (!memcg) | |
5489 | return; | |
5490 | ||
6abb5a86 JW |
5491 | commit_charge(page, memcg, lrucare); |
5492 | ||
00501b53 JW |
5493 | if (PageTransHuge(page)) { |
5494 | nr_pages <<= compound_order(page); | |
5495 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5496 | } | |
5497 | ||
6abb5a86 JW |
5498 | local_irq_disable(); |
5499 | mem_cgroup_charge_statistics(memcg, page, nr_pages); | |
5500 | memcg_check_events(memcg, page); | |
5501 | local_irq_enable(); | |
00501b53 JW |
5502 | |
5503 | if (do_swap_account && PageSwapCache(page)) { | |
5504 | swp_entry_t entry = { .val = page_private(page) }; | |
5505 | /* | |
5506 | * The swap entry might not get freed for a long time, | |
5507 | * let's not wait for it. The page already received a | |
5508 | * memory+swap charge, drop the swap entry duplicate. | |
5509 | */ | |
5510 | mem_cgroup_uncharge_swap(entry); | |
5511 | } | |
5512 | } | |
5513 | ||
5514 | /** | |
5515 | * mem_cgroup_cancel_charge - cancel a page charge | |
5516 | * @page: page to charge | |
5517 | * @memcg: memcg to charge the page to | |
5518 | * | |
5519 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
5520 | */ | |
5521 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg) | |
5522 | { | |
5523 | unsigned int nr_pages = 1; | |
5524 | ||
5525 | if (mem_cgroup_disabled()) | |
5526 | return; | |
5527 | /* | |
5528 | * Swap faults will attempt to charge the same page multiple | |
5529 | * times. But reuse_swap_page() might have removed the page | |
5530 | * from swapcache already, so we can't check PageSwapCache(). | |
5531 | */ | |
5532 | if (!memcg) | |
5533 | return; | |
5534 | ||
5535 | if (PageTransHuge(page)) { | |
5536 | nr_pages <<= compound_order(page); | |
5537 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5538 | } | |
5539 | ||
5540 | cancel_charge(memcg, nr_pages); | |
5541 | } | |
5542 | ||
747db954 | 5543 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
747db954 JW |
5544 | unsigned long nr_anon, unsigned long nr_file, |
5545 | unsigned long nr_huge, struct page *dummy_page) | |
5546 | { | |
18eca2e6 | 5547 | unsigned long nr_pages = nr_anon + nr_file; |
747db954 JW |
5548 | unsigned long flags; |
5549 | ||
ce00a967 | 5550 | if (!mem_cgroup_is_root(memcg)) { |
18eca2e6 JW |
5551 | page_counter_uncharge(&memcg->memory, nr_pages); |
5552 | if (do_swap_account) | |
5553 | page_counter_uncharge(&memcg->memsw, nr_pages); | |
ce00a967 JW |
5554 | memcg_oom_recover(memcg); |
5555 | } | |
747db954 JW |
5556 | |
5557 | local_irq_save(flags); | |
5558 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); | |
5559 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); | |
5560 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); | |
5561 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); | |
18eca2e6 | 5562 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
747db954 JW |
5563 | memcg_check_events(memcg, dummy_page); |
5564 | local_irq_restore(flags); | |
e8ea14cc JW |
5565 | |
5566 | if (!mem_cgroup_is_root(memcg)) | |
18eca2e6 | 5567 | css_put_many(&memcg->css, nr_pages); |
747db954 JW |
5568 | } |
5569 | ||
5570 | static void uncharge_list(struct list_head *page_list) | |
5571 | { | |
5572 | struct mem_cgroup *memcg = NULL; | |
747db954 JW |
5573 | unsigned long nr_anon = 0; |
5574 | unsigned long nr_file = 0; | |
5575 | unsigned long nr_huge = 0; | |
5576 | unsigned long pgpgout = 0; | |
747db954 JW |
5577 | struct list_head *next; |
5578 | struct page *page; | |
5579 | ||
5580 | next = page_list->next; | |
5581 | do { | |
5582 | unsigned int nr_pages = 1; | |
747db954 JW |
5583 | |
5584 | page = list_entry(next, struct page, lru); | |
5585 | next = page->lru.next; | |
5586 | ||
5587 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5588 | VM_BUG_ON_PAGE(page_count(page), page); | |
5589 | ||
1306a85a | 5590 | if (!page->mem_cgroup) |
747db954 JW |
5591 | continue; |
5592 | ||
5593 | /* | |
5594 | * Nobody should be changing or seriously looking at | |
1306a85a | 5595 | * page->mem_cgroup at this point, we have fully |
29833315 | 5596 | * exclusive access to the page. |
747db954 JW |
5597 | */ |
5598 | ||
1306a85a | 5599 | if (memcg != page->mem_cgroup) { |
747db954 | 5600 | if (memcg) { |
18eca2e6 JW |
5601 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5602 | nr_huge, page); | |
5603 | pgpgout = nr_anon = nr_file = nr_huge = 0; | |
747db954 | 5604 | } |
1306a85a | 5605 | memcg = page->mem_cgroup; |
747db954 JW |
5606 | } |
5607 | ||
5608 | if (PageTransHuge(page)) { | |
5609 | nr_pages <<= compound_order(page); | |
5610 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5611 | nr_huge += nr_pages; | |
5612 | } | |
5613 | ||
5614 | if (PageAnon(page)) | |
5615 | nr_anon += nr_pages; | |
5616 | else | |
5617 | nr_file += nr_pages; | |
5618 | ||
1306a85a | 5619 | page->mem_cgroup = NULL; |
747db954 JW |
5620 | |
5621 | pgpgout++; | |
5622 | } while (next != page_list); | |
5623 | ||
5624 | if (memcg) | |
18eca2e6 JW |
5625 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5626 | nr_huge, page); | |
747db954 JW |
5627 | } |
5628 | ||
0a31bc97 JW |
5629 | /** |
5630 | * mem_cgroup_uncharge - uncharge a page | |
5631 | * @page: page to uncharge | |
5632 | * | |
5633 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
5634 | * mem_cgroup_commit_charge(). | |
5635 | */ | |
5636 | void mem_cgroup_uncharge(struct page *page) | |
5637 | { | |
0a31bc97 JW |
5638 | if (mem_cgroup_disabled()) |
5639 | return; | |
5640 | ||
747db954 | 5641 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 5642 | if (!page->mem_cgroup) |
0a31bc97 JW |
5643 | return; |
5644 | ||
747db954 JW |
5645 | INIT_LIST_HEAD(&page->lru); |
5646 | uncharge_list(&page->lru); | |
5647 | } | |
0a31bc97 | 5648 | |
747db954 JW |
5649 | /** |
5650 | * mem_cgroup_uncharge_list - uncharge a list of page | |
5651 | * @page_list: list of pages to uncharge | |
5652 | * | |
5653 | * Uncharge a list of pages previously charged with | |
5654 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
5655 | */ | |
5656 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
5657 | { | |
5658 | if (mem_cgroup_disabled()) | |
5659 | return; | |
0a31bc97 | 5660 | |
747db954 JW |
5661 | if (!list_empty(page_list)) |
5662 | uncharge_list(page_list); | |
0a31bc97 JW |
5663 | } |
5664 | ||
5665 | /** | |
45637bab | 5666 | * mem_cgroup_replace_page - migrate a charge to another page |
0a31bc97 JW |
5667 | * @oldpage: currently charged page |
5668 | * @newpage: page to transfer the charge to | |
0a31bc97 JW |
5669 | * |
5670 | * Migrate the charge from @oldpage to @newpage. | |
5671 | * | |
5672 | * Both pages must be locked, @newpage->mapping must be set up. | |
25be6a65 | 5673 | * Either or both pages might be on the LRU already. |
0a31bc97 | 5674 | */ |
45637bab | 5675 | void mem_cgroup_replace_page(struct page *oldpage, struct page *newpage) |
0a31bc97 | 5676 | { |
29833315 | 5677 | struct mem_cgroup *memcg; |
0a31bc97 JW |
5678 | int isolated; |
5679 | ||
5680 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
5681 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 5682 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
5683 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
5684 | newpage); | |
0a31bc97 JW |
5685 | |
5686 | if (mem_cgroup_disabled()) | |
5687 | return; | |
5688 | ||
5689 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 5690 | if (newpage->mem_cgroup) |
0a31bc97 JW |
5691 | return; |
5692 | ||
45637bab | 5693 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 5694 | memcg = oldpage->mem_cgroup; |
29833315 | 5695 | if (!memcg) |
0a31bc97 JW |
5696 | return; |
5697 | ||
45637bab | 5698 | lock_page_lru(oldpage, &isolated); |
1306a85a | 5699 | oldpage->mem_cgroup = NULL; |
45637bab | 5700 | unlock_page_lru(oldpage, isolated); |
0a31bc97 | 5701 | |
45637bab | 5702 | commit_charge(newpage, memcg, true); |
0a31bc97 JW |
5703 | } |
5704 | ||
2d11085e | 5705 | /* |
1081312f MH |
5706 | * subsys_initcall() for memory controller. |
5707 | * | |
5708 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
5709 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
5710 | * everything that doesn't depend on a specific mem_cgroup structure should | |
5711 | * be initialized from here. | |
2d11085e MH |
5712 | */ |
5713 | static int __init mem_cgroup_init(void) | |
5714 | { | |
95a045f6 JW |
5715 | int cpu, node; |
5716 | ||
2d11085e | 5717 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
95a045f6 JW |
5718 | |
5719 | for_each_possible_cpu(cpu) | |
5720 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
5721 | drain_local_stock); | |
5722 | ||
5723 | for_each_node(node) { | |
5724 | struct mem_cgroup_tree_per_node *rtpn; | |
5725 | int zone; | |
5726 | ||
5727 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
5728 | node_online(node) ? node : NUMA_NO_NODE); | |
5729 | ||
5730 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
5731 | struct mem_cgroup_tree_per_zone *rtpz; | |
5732 | ||
5733 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
5734 | rtpz->rb_root = RB_ROOT; | |
5735 | spin_lock_init(&rtpz->lock); | |
5736 | } | |
5737 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
5738 | } | |
5739 | ||
2d11085e MH |
5740 | return 0; |
5741 | } | |
5742 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
5743 | |
5744 | #ifdef CONFIG_MEMCG_SWAP | |
5745 | /** | |
5746 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
5747 | * @page: page whose memsw charge to transfer | |
5748 | * @entry: swap entry to move the charge to | |
5749 | * | |
5750 | * Transfer the memsw charge of @page to @entry. | |
5751 | */ | |
5752 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
5753 | { | |
a0fddee3 | 5754 | struct mem_cgroup *memcg, *swap_memcg; |
21afa38e JW |
5755 | unsigned short oldid; |
5756 | ||
5757 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5758 | VM_BUG_ON_PAGE(page_count(page), page); | |
5759 | ||
5760 | if (!do_swap_account) | |
5761 | return; | |
5762 | ||
5763 | memcg = page->mem_cgroup; | |
5764 | ||
5765 | /* Readahead page, never charged */ | |
5766 | if (!memcg) | |
5767 | return; | |
5768 | ||
a0fddee3 VD |
5769 | /* |
5770 | * In case the memcg owning these pages has been offlined and doesn't | |
5771 | * have an ID allocated to it anymore, charge the closest online | |
5772 | * ancestor for the swap instead and transfer the memory+swap charge. | |
5773 | */ | |
5774 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
5775 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg)); | |
21afa38e | 5776 | VM_BUG_ON_PAGE(oldid, page); |
a0fddee3 | 5777 | mem_cgroup_swap_statistics(swap_memcg, true); |
21afa38e JW |
5778 | |
5779 | page->mem_cgroup = NULL; | |
5780 | ||
5781 | if (!mem_cgroup_is_root(memcg)) | |
5782 | page_counter_uncharge(&memcg->memory, 1); | |
5783 | ||
a0fddee3 VD |
5784 | if (memcg != swap_memcg) { |
5785 | if (!mem_cgroup_is_root(swap_memcg)) | |
5786 | page_counter_charge(&swap_memcg->memsw, 1); | |
5787 | page_counter_uncharge(&memcg->memsw, 1); | |
5788 | } | |
5789 | ||
ce9ce665 SAS |
5790 | /* |
5791 | * Interrupts should be disabled here because the caller holds the | |
5792 | * mapping->tree_lock lock which is taken with interrupts-off. It is | |
5793 | * important here to have the interrupts disabled because it is the | |
5794 | * only synchronisation we have for udpating the per-CPU variables. | |
5795 | */ | |
5796 | VM_BUG_ON(!irqs_disabled()); | |
21afa38e JW |
5797 | mem_cgroup_charge_statistics(memcg, page, -1); |
5798 | memcg_check_events(memcg, page); | |
8627c775 JW |
5799 | |
5800 | if (!mem_cgroup_is_root(memcg)) | |
5801 | css_put(&memcg->css); | |
21afa38e JW |
5802 | } |
5803 | ||
5804 | /** | |
5805 | * mem_cgroup_uncharge_swap - uncharge a swap entry | |
5806 | * @entry: swap entry to uncharge | |
5807 | * | |
5808 | * Drop the memsw charge associated with @entry. | |
5809 | */ | |
5810 | void mem_cgroup_uncharge_swap(swp_entry_t entry) | |
5811 | { | |
5812 | struct mem_cgroup *memcg; | |
5813 | unsigned short id; | |
5814 | ||
5815 | if (!do_swap_account) | |
5816 | return; | |
5817 | ||
5818 | id = swap_cgroup_record(entry, 0); | |
5819 | rcu_read_lock(); | |
adbe427b | 5820 | memcg = mem_cgroup_from_id(id); |
21afa38e JW |
5821 | if (memcg) { |
5822 | if (!mem_cgroup_is_root(memcg)) | |
5823 | page_counter_uncharge(&memcg->memsw, 1); | |
5824 | mem_cgroup_swap_statistics(memcg, false); | |
8627c775 | 5825 | mem_cgroup_id_put(memcg); |
21afa38e JW |
5826 | } |
5827 | rcu_read_unlock(); | |
5828 | } | |
5829 | ||
5830 | /* for remember boot option*/ | |
5831 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
5832 | static int really_do_swap_account __initdata = 1; | |
5833 | #else | |
5834 | static int really_do_swap_account __initdata; | |
5835 | #endif | |
5836 | ||
5837 | static int __init enable_swap_account(char *s) | |
5838 | { | |
5839 | if (!strcmp(s, "1")) | |
5840 | really_do_swap_account = 1; | |
5841 | else if (!strcmp(s, "0")) | |
5842 | really_do_swap_account = 0; | |
5843 | return 1; | |
5844 | } | |
5845 | __setup("swapaccount=", enable_swap_account); | |
5846 | ||
5847 | static struct cftype memsw_cgroup_files[] = { | |
5848 | { | |
5849 | .name = "memsw.usage_in_bytes", | |
5850 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
5851 | .read_u64 = mem_cgroup_read_u64, | |
5852 | }, | |
5853 | { | |
5854 | .name = "memsw.max_usage_in_bytes", | |
5855 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
5856 | .write = mem_cgroup_reset, | |
5857 | .read_u64 = mem_cgroup_read_u64, | |
5858 | }, | |
5859 | { | |
5860 | .name = "memsw.limit_in_bytes", | |
5861 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
5862 | .write = mem_cgroup_write, | |
5863 | .read_u64 = mem_cgroup_read_u64, | |
5864 | }, | |
5865 | { | |
5866 | .name = "memsw.failcnt", | |
5867 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
5868 | .write = mem_cgroup_reset, | |
5869 | .read_u64 = mem_cgroup_read_u64, | |
5870 | }, | |
5871 | { }, /* terminate */ | |
5872 | }; | |
5873 | ||
5874 | static int __init mem_cgroup_swap_init(void) | |
5875 | { | |
5876 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
5877 | do_swap_account = 1; | |
5878 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, | |
5879 | memsw_cgroup_files)); | |
5880 | } | |
5881 | return 0; | |
5882 | } | |
5883 | subsys_initcall(mem_cgroup_swap_init); | |
5884 | ||
5885 | #endif /* CONFIG_MEMCG_SWAP */ |