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 */ |
4ffef5fe DN |
199 | struct mem_cgroup *from; |
200 | struct mem_cgroup *to; | |
1dfab5ab | 201 | unsigned long flags; |
4ffef5fe | 202 | unsigned long precharge; |
854ffa8d | 203 | unsigned long moved_charge; |
483c30b5 | 204 | unsigned long moved_swap; |
8033b97c DN |
205 | struct task_struct *moving_task; /* a task moving charges */ |
206 | wait_queue_head_t waitq; /* a waitq for other context */ | |
207 | } mc = { | |
2bd9bb20 | 208 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
209 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
210 | }; | |
4ffef5fe | 211 | |
4e416953 BS |
212 | /* |
213 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
214 | * limit reclaim to prevent infinite loops, if they ever occur. | |
215 | */ | |
a0db00fc | 216 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 217 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 218 | |
217bc319 KH |
219 | enum charge_type { |
220 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 221 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 222 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 223 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
224 | NR_CHARGE_TYPE, |
225 | }; | |
226 | ||
8c7c6e34 | 227 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
228 | enum res_type { |
229 | _MEM, | |
230 | _MEMSWAP, | |
231 | _OOM_TYPE, | |
510fc4e1 | 232 | _KMEM, |
86ae53e1 GC |
233 | }; |
234 | ||
a0db00fc KS |
235 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
236 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 237 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
238 | /* Used for OOM nofiier */ |
239 | #define OOM_CONTROL (0) | |
8c7c6e34 | 240 | |
0999821b GC |
241 | /* |
242 | * The memcg_create_mutex will be held whenever a new cgroup is created. | |
243 | * As a consequence, any change that needs to protect against new child cgroups | |
244 | * appearing has to hold it as well. | |
245 | */ | |
246 | static DEFINE_MUTEX(memcg_create_mutex); | |
247 | ||
70ddf637 AV |
248 | /* Some nice accessors for the vmpressure. */ |
249 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
250 | { | |
251 | if (!memcg) | |
252 | memcg = root_mem_cgroup; | |
253 | return &memcg->vmpressure; | |
254 | } | |
255 | ||
256 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
257 | { | |
258 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
259 | } | |
260 | ||
7ffc0edc MH |
261 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
262 | { | |
263 | return (memcg == root_mem_cgroup); | |
264 | } | |
265 | ||
4219b2da LZ |
266 | /* |
267 | * We restrict the id in the range of [1, 65535], so it can fit into | |
268 | * an unsigned short. | |
269 | */ | |
270 | #define MEM_CGROUP_ID_MAX USHRT_MAX | |
271 | ||
34c00c31 LZ |
272 | static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) |
273 | { | |
15a4c835 | 274 | return memcg->css.id; |
34c00c31 LZ |
275 | } |
276 | ||
adbe427b VD |
277 | /* |
278 | * A helper function to get mem_cgroup from ID. must be called under | |
279 | * rcu_read_lock(). The caller is responsible for calling | |
280 | * css_tryget_online() if the mem_cgroup is used for charging. (dropping | |
281 | * refcnt from swap can be called against removed memcg.) | |
282 | */ | |
34c00c31 LZ |
283 | static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) |
284 | { | |
285 | struct cgroup_subsys_state *css; | |
286 | ||
7d699ddb | 287 | css = css_from_id(id, &memory_cgrp_subsys); |
34c00c31 LZ |
288 | return mem_cgroup_from_css(css); |
289 | } | |
290 | ||
e1aab161 | 291 | /* Writing them here to avoid exposing memcg's inner layout */ |
4bd2c1ee | 292 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
e1aab161 | 293 | |
e1aab161 GC |
294 | void sock_update_memcg(struct sock *sk) |
295 | { | |
376be5ff | 296 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 297 | struct mem_cgroup *memcg; |
3f134619 | 298 | struct cg_proto *cg_proto; |
e1aab161 GC |
299 | |
300 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
301 | ||
f3f511e1 GC |
302 | /* Socket cloning can throw us here with sk_cgrp already |
303 | * filled. It won't however, necessarily happen from | |
304 | * process context. So the test for root memcg given | |
305 | * the current task's memcg won't help us in this case. | |
306 | * | |
307 | * Respecting the original socket's memcg is a better | |
308 | * decision in this case. | |
309 | */ | |
310 | if (sk->sk_cgrp) { | |
311 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
5347e5ae | 312 | css_get(&sk->sk_cgrp->memcg->css); |
f3f511e1 GC |
313 | return; |
314 | } | |
315 | ||
e1aab161 GC |
316 | rcu_read_lock(); |
317 | memcg = mem_cgroup_from_task(current); | |
3f134619 | 318 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
e752eb68 | 319 | if (cg_proto && test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags) && |
ec903c0c | 320 | css_tryget_online(&memcg->css)) { |
3f134619 | 321 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
322 | } |
323 | rcu_read_unlock(); | |
324 | } | |
325 | } | |
326 | EXPORT_SYMBOL(sock_update_memcg); | |
327 | ||
328 | void sock_release_memcg(struct sock *sk) | |
329 | { | |
376be5ff | 330 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
331 | struct mem_cgroup *memcg; |
332 | WARN_ON(!sk->sk_cgrp->memcg); | |
333 | memcg = sk->sk_cgrp->memcg; | |
5347e5ae | 334 | css_put(&sk->sk_cgrp->memcg->css); |
e1aab161 GC |
335 | } |
336 | } | |
d1a4c0b3 GC |
337 | |
338 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
339 | { | |
340 | if (!memcg || mem_cgroup_is_root(memcg)) | |
341 | return NULL; | |
342 | ||
2e685cad | 343 | return &memcg->tcp_mem; |
d1a4c0b3 GC |
344 | } |
345 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 346 | |
3f134619 GC |
347 | #endif |
348 | ||
a8964b9b | 349 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 | 350 | /* |
f7ce3190 | 351 | * This will be the memcg's index in each cache's ->memcg_params.memcg_caches. |
b8627835 LZ |
352 | * The main reason for not using cgroup id for this: |
353 | * this works better in sparse environments, where we have a lot of memcgs, | |
354 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
355 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
356 | * 200 entry array for that. | |
55007d84 | 357 | * |
dbcf73e2 VD |
358 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
359 | * will double each time we have to increase it. | |
55007d84 | 360 | */ |
dbcf73e2 VD |
361 | static DEFINE_IDA(memcg_cache_ida); |
362 | int memcg_nr_cache_ids; | |
749c5415 | 363 | |
05257a1a VD |
364 | /* Protects memcg_nr_cache_ids */ |
365 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
366 | ||
367 | void memcg_get_cache_ids(void) | |
368 | { | |
369 | down_read(&memcg_cache_ids_sem); | |
370 | } | |
371 | ||
372 | void memcg_put_cache_ids(void) | |
373 | { | |
374 | up_read(&memcg_cache_ids_sem); | |
375 | } | |
376 | ||
55007d84 GC |
377 | /* |
378 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
379 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
380 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
381 | * tunable, but that is strictly not necessary. | |
382 | * | |
b8627835 | 383 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
384 | * this constant directly from cgroup, but it is understandable that this is |
385 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 386 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
387 | * increase ours as well if it increases. |
388 | */ | |
389 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 390 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 391 | |
d7f25f8a GC |
392 | /* |
393 | * A lot of the calls to the cache allocation functions are expected to be | |
394 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
395 | * conditional to this static branch, we'll have to allow modules that does | |
396 | * kmem_cache_alloc and the such to see this symbol as well | |
397 | */ | |
a8964b9b | 398 | struct static_key memcg_kmem_enabled_key; |
d7f25f8a | 399 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 400 | |
a8964b9b GC |
401 | #endif /* CONFIG_MEMCG_KMEM */ |
402 | ||
f64c3f54 | 403 | static struct mem_cgroup_per_zone * |
e231875b | 404 | mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone) |
f64c3f54 | 405 | { |
e231875b JZ |
406 | int nid = zone_to_nid(zone); |
407 | int zid = zone_idx(zone); | |
408 | ||
54f72fe0 | 409 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
410 | } |
411 | ||
ad7fa852 TH |
412 | /** |
413 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
414 | * @page: page of interest | |
415 | * | |
416 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
417 | * with @page is returned. The returned css remains associated with @page | |
418 | * until it is released. | |
419 | * | |
420 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
421 | * is returned. | |
422 | * | |
423 | * XXX: The above description of behavior on the default hierarchy isn't | |
424 | * strictly true yet as replace_page_cache_page() can modify the | |
425 | * association before @page is released even on the default hierarchy; | |
426 | * however, the current and planned usages don't mix the the two functions | |
427 | * and replace_page_cache_page() will soon be updated to make the invariant | |
428 | * actually true. | |
429 | */ | |
430 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
431 | { | |
432 | struct mem_cgroup *memcg; | |
433 | ||
434 | rcu_read_lock(); | |
435 | ||
436 | memcg = page->mem_cgroup; | |
437 | ||
9e10a130 | 438 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
439 | memcg = root_mem_cgroup; |
440 | ||
441 | rcu_read_unlock(); | |
442 | return &memcg->css; | |
443 | } | |
444 | ||
2fc04524 VD |
445 | /** |
446 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
447 | * @page: the page | |
448 | * | |
449 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
450 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
451 | * is safe to call this function without holding a reference to @page. | |
452 | * | |
453 | * Note, this function is inherently racy, because there is nothing to prevent | |
454 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
455 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
456 | * do not care (such as procfs interfaces). | |
457 | */ | |
458 | ino_t page_cgroup_ino(struct page *page) | |
459 | { | |
460 | struct mem_cgroup *memcg; | |
461 | unsigned long ino = 0; | |
462 | ||
463 | rcu_read_lock(); | |
464 | memcg = READ_ONCE(page->mem_cgroup); | |
465 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) | |
466 | memcg = parent_mem_cgroup(memcg); | |
467 | if (memcg) | |
468 | ino = cgroup_ino(memcg->css.cgroup); | |
469 | rcu_read_unlock(); | |
470 | return ino; | |
471 | } | |
472 | ||
f64c3f54 | 473 | static struct mem_cgroup_per_zone * |
e231875b | 474 | mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 475 | { |
97a6c37b JW |
476 | int nid = page_to_nid(page); |
477 | int zid = page_zonenum(page); | |
f64c3f54 | 478 | |
e231875b | 479 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
480 | } |
481 | ||
bb4cc1a8 AM |
482 | static struct mem_cgroup_tree_per_zone * |
483 | soft_limit_tree_node_zone(int nid, int zid) | |
484 | { | |
485 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
486 | } | |
487 | ||
488 | static struct mem_cgroup_tree_per_zone * | |
489 | soft_limit_tree_from_page(struct page *page) | |
490 | { | |
491 | int nid = page_to_nid(page); | |
492 | int zid = page_zonenum(page); | |
493 | ||
494 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
495 | } | |
496 | ||
cf2c8127 JW |
497 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz, |
498 | struct mem_cgroup_tree_per_zone *mctz, | |
3e32cb2e | 499 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
500 | { |
501 | struct rb_node **p = &mctz->rb_root.rb_node; | |
502 | struct rb_node *parent = NULL; | |
503 | struct mem_cgroup_per_zone *mz_node; | |
504 | ||
505 | if (mz->on_tree) | |
506 | return; | |
507 | ||
508 | mz->usage_in_excess = new_usage_in_excess; | |
509 | if (!mz->usage_in_excess) | |
510 | return; | |
511 | while (*p) { | |
512 | parent = *p; | |
513 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
514 | tree_node); | |
515 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
516 | p = &(*p)->rb_left; | |
517 | /* | |
518 | * We can't avoid mem cgroups that are over their soft | |
519 | * limit by the same amount | |
520 | */ | |
521 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
522 | p = &(*p)->rb_right; | |
523 | } | |
524 | rb_link_node(&mz->tree_node, parent, p); | |
525 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
526 | mz->on_tree = true; | |
527 | } | |
528 | ||
cf2c8127 JW |
529 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
530 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 AM |
531 | { |
532 | if (!mz->on_tree) | |
533 | return; | |
534 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
535 | mz->on_tree = false; | |
536 | } | |
537 | ||
cf2c8127 JW |
538 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
539 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 | 540 | { |
0a31bc97 JW |
541 | unsigned long flags; |
542 | ||
543 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 544 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 545 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
546 | } |
547 | ||
3e32cb2e JW |
548 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
549 | { | |
550 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 551 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
552 | unsigned long excess = 0; |
553 | ||
554 | if (nr_pages > soft_limit) | |
555 | excess = nr_pages - soft_limit; | |
556 | ||
557 | return excess; | |
558 | } | |
bb4cc1a8 AM |
559 | |
560 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
561 | { | |
3e32cb2e | 562 | unsigned long excess; |
bb4cc1a8 AM |
563 | struct mem_cgroup_per_zone *mz; |
564 | struct mem_cgroup_tree_per_zone *mctz; | |
bb4cc1a8 | 565 | |
e231875b | 566 | mctz = soft_limit_tree_from_page(page); |
bb4cc1a8 AM |
567 | /* |
568 | * Necessary to update all ancestors when hierarchy is used. | |
569 | * because their event counter is not touched. | |
570 | */ | |
571 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
e231875b | 572 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
3e32cb2e | 573 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
574 | /* |
575 | * We have to update the tree if mz is on RB-tree or | |
576 | * mem is over its softlimit. | |
577 | */ | |
578 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
579 | unsigned long flags; |
580 | ||
581 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
582 | /* if on-tree, remove it */ |
583 | if (mz->on_tree) | |
cf2c8127 | 584 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
585 | /* |
586 | * Insert again. mz->usage_in_excess will be updated. | |
587 | * If excess is 0, no tree ops. | |
588 | */ | |
cf2c8127 | 589 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 590 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
591 | } |
592 | } | |
593 | } | |
594 | ||
595 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
596 | { | |
bb4cc1a8 | 597 | struct mem_cgroup_tree_per_zone *mctz; |
e231875b JZ |
598 | struct mem_cgroup_per_zone *mz; |
599 | int nid, zid; | |
bb4cc1a8 | 600 | |
e231875b JZ |
601 | for_each_node(nid) { |
602 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
603 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
604 | mctz = soft_limit_tree_node_zone(nid, zid); | |
cf2c8127 | 605 | mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
606 | } |
607 | } | |
608 | } | |
609 | ||
610 | static struct mem_cgroup_per_zone * | |
611 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
612 | { | |
613 | struct rb_node *rightmost = NULL; | |
614 | struct mem_cgroup_per_zone *mz; | |
615 | ||
616 | retry: | |
617 | mz = NULL; | |
618 | rightmost = rb_last(&mctz->rb_root); | |
619 | if (!rightmost) | |
620 | goto done; /* Nothing to reclaim from */ | |
621 | ||
622 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
623 | /* | |
624 | * Remove the node now but someone else can add it back, | |
625 | * we will to add it back at the end of reclaim to its correct | |
626 | * position in the tree. | |
627 | */ | |
cf2c8127 | 628 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 629 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 630 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
631 | goto retry; |
632 | done: | |
633 | return mz; | |
634 | } | |
635 | ||
636 | static struct mem_cgroup_per_zone * | |
637 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
638 | { | |
639 | struct mem_cgroup_per_zone *mz; | |
640 | ||
0a31bc97 | 641 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 642 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 643 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
644 | return mz; |
645 | } | |
646 | ||
711d3d2c | 647 | /* |
484ebb3b GT |
648 | * Return page count for single (non recursive) @memcg. |
649 | * | |
711d3d2c KH |
650 | * Implementation Note: reading percpu statistics for memcg. |
651 | * | |
652 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
653 | * synchronization to implement "quick" read. There are trade-off between | |
654 | * reading cost and precision of value. Then, we may have a chance to implement | |
484ebb3b | 655 | * a periodic synchronization of counter in memcg's counter. |
711d3d2c KH |
656 | * |
657 | * But this _read() function is used for user interface now. The user accounts | |
658 | * memory usage by memory cgroup and he _always_ requires exact value because | |
659 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
660 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
661 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
662 | * | |
663 | * If there are kernel internal actions which can make use of some not-exact | |
664 | * value, and reading all cpu value can be performance bottleneck in some | |
484ebb3b | 665 | * common workload, threshold and synchronization as vmstat[] should be |
711d3d2c KH |
666 | * implemented. |
667 | */ | |
484ebb3b GT |
668 | static unsigned long |
669 | mem_cgroup_read_stat(struct mem_cgroup *memcg, enum mem_cgroup_stat_index idx) | |
c62b1a3b | 670 | { |
7a159cc9 | 671 | long val = 0; |
c62b1a3b | 672 | int cpu; |
c62b1a3b | 673 | |
484ebb3b | 674 | /* Per-cpu values can be negative, use a signed accumulator */ |
733a572e | 675 | for_each_possible_cpu(cpu) |
c0ff4b85 | 676 | val += per_cpu(memcg->stat->count[idx], cpu); |
484ebb3b GT |
677 | /* |
678 | * Summing races with updates, so val may be negative. Avoid exposing | |
679 | * transient negative values. | |
680 | */ | |
681 | if (val < 0) | |
682 | val = 0; | |
c62b1a3b KH |
683 | return val; |
684 | } | |
685 | ||
c0ff4b85 | 686 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
687 | enum mem_cgroup_events_index idx) |
688 | { | |
689 | unsigned long val = 0; | |
690 | int cpu; | |
691 | ||
733a572e | 692 | for_each_possible_cpu(cpu) |
c0ff4b85 | 693 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f JW |
694 | return val; |
695 | } | |
696 | ||
c0ff4b85 | 697 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 698 | struct page *page, |
0a31bc97 | 699 | int nr_pages) |
d52aa412 | 700 | { |
b2402857 KH |
701 | /* |
702 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
703 | * counted as CACHE even if it's on ANON LRU. | |
704 | */ | |
0a31bc97 | 705 | if (PageAnon(page)) |
b2402857 | 706 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
c0ff4b85 | 707 | nr_pages); |
d52aa412 | 708 | else |
b2402857 | 709 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 710 | nr_pages); |
55e462b0 | 711 | |
b070e65c DR |
712 | if (PageTransHuge(page)) |
713 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], | |
714 | nr_pages); | |
715 | ||
e401f176 KH |
716 | /* pagein of a big page is an event. So, ignore page size */ |
717 | if (nr_pages > 0) | |
c0ff4b85 | 718 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 719 | else { |
c0ff4b85 | 720 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
721 | nr_pages = -nr_pages; /* for event */ |
722 | } | |
e401f176 | 723 | |
13114716 | 724 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
6d12e2d8 KH |
725 | } |
726 | ||
e231875b JZ |
727 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
728 | int nid, | |
729 | unsigned int lru_mask) | |
bb2a0de9 | 730 | { |
e231875b | 731 | unsigned long nr = 0; |
889976db YH |
732 | int zid; |
733 | ||
e231875b | 734 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 735 | |
e231875b JZ |
736 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
737 | struct mem_cgroup_per_zone *mz; | |
738 | enum lru_list lru; | |
739 | ||
740 | for_each_lru(lru) { | |
741 | if (!(BIT(lru) & lru_mask)) | |
742 | continue; | |
743 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
744 | nr += mz->lru_size[lru]; | |
745 | } | |
746 | } | |
747 | return nr; | |
889976db | 748 | } |
bb2a0de9 | 749 | |
c0ff4b85 | 750 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 751 | unsigned int lru_mask) |
6d12e2d8 | 752 | { |
e231875b | 753 | unsigned long nr = 0; |
889976db | 754 | int nid; |
6d12e2d8 | 755 | |
31aaea4a | 756 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
757 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
758 | return nr; | |
d52aa412 KH |
759 | } |
760 | ||
f53d7ce3 JW |
761 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
762 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
763 | { |
764 | unsigned long val, next; | |
765 | ||
13114716 | 766 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 767 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 768 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
769 | if ((long)next - (long)val < 0) { |
770 | switch (target) { | |
771 | case MEM_CGROUP_TARGET_THRESH: | |
772 | next = val + THRESHOLDS_EVENTS_TARGET; | |
773 | break; | |
bb4cc1a8 AM |
774 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
775 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
776 | break; | |
f53d7ce3 JW |
777 | case MEM_CGROUP_TARGET_NUMAINFO: |
778 | next = val + NUMAINFO_EVENTS_TARGET; | |
779 | break; | |
780 | default: | |
781 | break; | |
782 | } | |
783 | __this_cpu_write(memcg->stat->targets[target], next); | |
784 | return true; | |
7a159cc9 | 785 | } |
f53d7ce3 | 786 | return false; |
d2265e6f KH |
787 | } |
788 | ||
789 | /* | |
790 | * Check events in order. | |
791 | * | |
792 | */ | |
c0ff4b85 | 793 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
794 | { |
795 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
796 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
797 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 798 | bool do_softlimit; |
82b3f2a7 | 799 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 800 | |
bb4cc1a8 AM |
801 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
802 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
803 | #if MAX_NUMNODES > 1 |
804 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
805 | MEM_CGROUP_TARGET_NUMAINFO); | |
806 | #endif | |
c0ff4b85 | 807 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
808 | if (unlikely(do_softlimit)) |
809 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 810 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 811 | if (unlikely(do_numainfo)) |
c0ff4b85 | 812 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 813 | #endif |
0a31bc97 | 814 | } |
d2265e6f KH |
815 | } |
816 | ||
cf475ad2 | 817 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 818 | { |
31a78f23 BS |
819 | /* |
820 | * mm_update_next_owner() may clear mm->owner to NULL | |
821 | * if it races with swapoff, page migration, etc. | |
822 | * So this can be called with p == NULL. | |
823 | */ | |
824 | if (unlikely(!p)) | |
825 | return NULL; | |
826 | ||
073219e9 | 827 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 828 | } |
33398cf2 | 829 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 830 | |
df381975 | 831 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 832 | { |
c0ff4b85 | 833 | struct mem_cgroup *memcg = NULL; |
0b7f569e | 834 | |
54595fe2 KH |
835 | rcu_read_lock(); |
836 | do { | |
6f6acb00 MH |
837 | /* |
838 | * Page cache insertions can happen withou an | |
839 | * actual mm context, e.g. during disk probing | |
840 | * on boot, loopback IO, acct() writes etc. | |
841 | */ | |
842 | if (unlikely(!mm)) | |
df381975 | 843 | memcg = root_mem_cgroup; |
6f6acb00 MH |
844 | else { |
845 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
846 | if (unlikely(!memcg)) | |
847 | memcg = root_mem_cgroup; | |
848 | } | |
ec903c0c | 849 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 850 | rcu_read_unlock(); |
c0ff4b85 | 851 | return memcg; |
54595fe2 KH |
852 | } |
853 | ||
5660048c JW |
854 | /** |
855 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
856 | * @root: hierarchy root | |
857 | * @prev: previously returned memcg, NULL on first invocation | |
858 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
859 | * | |
860 | * Returns references to children of the hierarchy below @root, or | |
861 | * @root itself, or %NULL after a full round-trip. | |
862 | * | |
863 | * Caller must pass the return value in @prev on subsequent | |
864 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
865 | * to cancel a hierarchy walk before the round-trip is complete. | |
866 | * | |
867 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
868 | * divide up the memcgs in the hierarchy among all concurrent | |
869 | * reclaimers operating on the same zone and priority. | |
870 | */ | |
694fbc0f | 871 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 872 | struct mem_cgroup *prev, |
694fbc0f | 873 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 874 | { |
33398cf2 | 875 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
5ac8fb31 | 876 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 877 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 878 | struct mem_cgroup *pos = NULL; |
711d3d2c | 879 | |
694fbc0f AM |
880 | if (mem_cgroup_disabled()) |
881 | return NULL; | |
5660048c | 882 | |
9f3a0d09 JW |
883 | if (!root) |
884 | root = root_mem_cgroup; | |
7d74b06f | 885 | |
9f3a0d09 | 886 | if (prev && !reclaim) |
5ac8fb31 | 887 | pos = prev; |
14067bb3 | 888 | |
9f3a0d09 JW |
889 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
890 | if (prev) | |
5ac8fb31 | 891 | goto out; |
694fbc0f | 892 | return root; |
9f3a0d09 | 893 | } |
14067bb3 | 894 | |
542f85f9 | 895 | rcu_read_lock(); |
5f578161 | 896 | |
5ac8fb31 JW |
897 | if (reclaim) { |
898 | struct mem_cgroup_per_zone *mz; | |
899 | ||
900 | mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone); | |
901 | iter = &mz->iter[reclaim->priority]; | |
902 | ||
903 | if (prev && reclaim->generation != iter->generation) | |
904 | goto out_unlock; | |
905 | ||
6df38689 | 906 | while (1) { |
4db0c3c2 | 907 | pos = READ_ONCE(iter->position); |
6df38689 VD |
908 | if (!pos || css_tryget(&pos->css)) |
909 | break; | |
5ac8fb31 | 910 | /* |
6df38689 VD |
911 | * css reference reached zero, so iter->position will |
912 | * be cleared by ->css_released. However, we should not | |
913 | * rely on this happening soon, because ->css_released | |
914 | * is called from a work queue, and by busy-waiting we | |
915 | * might block it. So we clear iter->position right | |
916 | * away. | |
5ac8fb31 | 917 | */ |
6df38689 VD |
918 | (void)cmpxchg(&iter->position, pos, NULL); |
919 | } | |
5ac8fb31 JW |
920 | } |
921 | ||
922 | if (pos) | |
923 | css = &pos->css; | |
924 | ||
925 | for (;;) { | |
926 | css = css_next_descendant_pre(css, &root->css); | |
927 | if (!css) { | |
928 | /* | |
929 | * Reclaimers share the hierarchy walk, and a | |
930 | * new one might jump in right at the end of | |
931 | * the hierarchy - make sure they see at least | |
932 | * one group and restart from the beginning. | |
933 | */ | |
934 | if (!prev) | |
935 | continue; | |
936 | break; | |
527a5ec9 | 937 | } |
7d74b06f | 938 | |
5ac8fb31 JW |
939 | /* |
940 | * Verify the css and acquire a reference. The root | |
941 | * is provided by the caller, so we know it's alive | |
942 | * and kicking, and don't take an extra reference. | |
943 | */ | |
944 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 945 | |
5ac8fb31 JW |
946 | if (css == &root->css) |
947 | break; | |
14067bb3 | 948 | |
b2052564 | 949 | if (css_tryget(css)) { |
5ac8fb31 JW |
950 | /* |
951 | * Make sure the memcg is initialized: | |
952 | * mem_cgroup_css_online() orders the the | |
953 | * initialization against setting the flag. | |
954 | */ | |
955 | if (smp_load_acquire(&memcg->initialized)) | |
956 | break; | |
542f85f9 | 957 | |
5ac8fb31 | 958 | css_put(css); |
527a5ec9 | 959 | } |
9f3a0d09 | 960 | |
5ac8fb31 | 961 | memcg = NULL; |
9f3a0d09 | 962 | } |
5ac8fb31 JW |
963 | |
964 | if (reclaim) { | |
5ac8fb31 | 965 | /* |
6df38689 VD |
966 | * The position could have already been updated by a competing |
967 | * thread, so check that the value hasn't changed since we read | |
968 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 969 | */ |
6df38689 VD |
970 | (void)cmpxchg(&iter->position, pos, memcg); |
971 | ||
5ac8fb31 JW |
972 | if (pos) |
973 | css_put(&pos->css); | |
974 | ||
975 | if (!memcg) | |
976 | iter->generation++; | |
977 | else if (!prev) | |
978 | reclaim->generation = iter->generation; | |
9f3a0d09 | 979 | } |
5ac8fb31 | 980 | |
542f85f9 MH |
981 | out_unlock: |
982 | rcu_read_unlock(); | |
5ac8fb31 | 983 | out: |
c40046f3 MH |
984 | if (prev && prev != root) |
985 | css_put(&prev->css); | |
986 | ||
9f3a0d09 | 987 | return memcg; |
14067bb3 | 988 | } |
7d74b06f | 989 | |
5660048c JW |
990 | /** |
991 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
992 | * @root: hierarchy root | |
993 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
994 | */ | |
995 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
996 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
997 | { |
998 | if (!root) | |
999 | root = root_mem_cgroup; | |
1000 | if (prev && prev != root) | |
1001 | css_put(&prev->css); | |
1002 | } | |
7d74b06f | 1003 | |
6df38689 VD |
1004 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1005 | { | |
1006 | struct mem_cgroup *memcg = dead_memcg; | |
1007 | struct mem_cgroup_reclaim_iter *iter; | |
1008 | struct mem_cgroup_per_zone *mz; | |
1009 | int nid, zid; | |
1010 | int i; | |
1011 | ||
1012 | while ((memcg = parent_mem_cgroup(memcg))) { | |
1013 | for_each_node(nid) { | |
1014 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1015 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
1016 | for (i = 0; i <= DEF_PRIORITY; i++) { | |
1017 | iter = &mz->iter[i]; | |
1018 | cmpxchg(&iter->position, | |
1019 | dead_memcg, NULL); | |
1020 | } | |
1021 | } | |
1022 | } | |
1023 | } | |
1024 | } | |
1025 | ||
9f3a0d09 JW |
1026 | /* |
1027 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1028 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1029 | * be used for reference counting. | |
1030 | */ | |
1031 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1032 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1033 | iter != NULL; \ |
527a5ec9 | 1034 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1035 | |
9f3a0d09 | 1036 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1037 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1038 | iter != NULL; \ |
527a5ec9 | 1039 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1040 | |
925b7673 JW |
1041 | /** |
1042 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1043 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1044 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1045 | * |
1046 | * Returns the lru list vector holding pages for the given @zone and | |
1047 | * @mem. This can be the global zone lruvec, if the memory controller | |
1048 | * is disabled. | |
1049 | */ | |
1050 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1051 | struct mem_cgroup *memcg) | |
1052 | { | |
1053 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1054 | struct lruvec *lruvec; |
925b7673 | 1055 | |
bea8c150 HD |
1056 | if (mem_cgroup_disabled()) { |
1057 | lruvec = &zone->lruvec; | |
1058 | goto out; | |
1059 | } | |
925b7673 | 1060 | |
e231875b | 1061 | mz = mem_cgroup_zone_zoneinfo(memcg, zone); |
bea8c150 HD |
1062 | lruvec = &mz->lruvec; |
1063 | out: | |
1064 | /* | |
1065 | * Since a node can be onlined after the mem_cgroup was created, | |
1066 | * we have to be prepared to initialize lruvec->zone here; | |
1067 | * and if offlined then reonlined, we need to reinitialize it. | |
1068 | */ | |
1069 | if (unlikely(lruvec->zone != zone)) | |
1070 | lruvec->zone = zone; | |
1071 | return lruvec; | |
925b7673 JW |
1072 | } |
1073 | ||
925b7673 | 1074 | /** |
dfe0e773 | 1075 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1076 | * @page: the page |
fa9add64 | 1077 | * @zone: zone of the page |
dfe0e773 JW |
1078 | * |
1079 | * This function is only safe when following the LRU page isolation | |
1080 | * and putback protocol: the LRU lock must be held, and the page must | |
1081 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 1082 | */ |
fa9add64 | 1083 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1084 | { |
08e552c6 | 1085 | struct mem_cgroup_per_zone *mz; |
925b7673 | 1086 | struct mem_cgroup *memcg; |
bea8c150 | 1087 | struct lruvec *lruvec; |
6d12e2d8 | 1088 | |
bea8c150 HD |
1089 | if (mem_cgroup_disabled()) { |
1090 | lruvec = &zone->lruvec; | |
1091 | goto out; | |
1092 | } | |
925b7673 | 1093 | |
1306a85a | 1094 | memcg = page->mem_cgroup; |
7512102c | 1095 | /* |
dfe0e773 | 1096 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1097 | * possibly migrated - before they are charged. |
7512102c | 1098 | */ |
29833315 JW |
1099 | if (!memcg) |
1100 | memcg = root_mem_cgroup; | |
7512102c | 1101 | |
e231875b | 1102 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
bea8c150 HD |
1103 | lruvec = &mz->lruvec; |
1104 | out: | |
1105 | /* | |
1106 | * Since a node can be onlined after the mem_cgroup was created, | |
1107 | * we have to be prepared to initialize lruvec->zone here; | |
1108 | * and if offlined then reonlined, we need to reinitialize it. | |
1109 | */ | |
1110 | if (unlikely(lruvec->zone != zone)) | |
1111 | lruvec->zone = zone; | |
1112 | return lruvec; | |
08e552c6 | 1113 | } |
b69408e8 | 1114 | |
925b7673 | 1115 | /** |
fa9add64 HD |
1116 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1117 | * @lruvec: mem_cgroup per zone lru vector | |
1118 | * @lru: index of lru list the page is sitting on | |
1119 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1120 | * |
fa9add64 HD |
1121 | * This function must be called when a page is added to or removed from an |
1122 | * lru list. | |
3f58a829 | 1123 | */ |
fa9add64 HD |
1124 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1125 | int nr_pages) | |
3f58a829 MK |
1126 | { |
1127 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1128 | unsigned long *lru_size; |
3f58a829 MK |
1129 | |
1130 | if (mem_cgroup_disabled()) | |
1131 | return; | |
1132 | ||
fa9add64 HD |
1133 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1134 | lru_size = mz->lru_size + lru; | |
1135 | *lru_size += nr_pages; | |
1136 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1137 | } |
544122e5 | 1138 | |
2314b42d | 1139 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
c3ac9a8a | 1140 | { |
2314b42d | 1141 | struct mem_cgroup *task_memcg; |
158e0a2d | 1142 | struct task_struct *p; |
ffbdccf5 | 1143 | bool ret; |
4c4a2214 | 1144 | |
158e0a2d | 1145 | p = find_lock_task_mm(task); |
de077d22 | 1146 | if (p) { |
2314b42d | 1147 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1148 | task_unlock(p); |
1149 | } else { | |
1150 | /* | |
1151 | * All threads may have already detached their mm's, but the oom | |
1152 | * killer still needs to detect if they have already been oom | |
1153 | * killed to prevent needlessly killing additional tasks. | |
1154 | */ | |
ffbdccf5 | 1155 | rcu_read_lock(); |
2314b42d JW |
1156 | task_memcg = mem_cgroup_from_task(task); |
1157 | css_get(&task_memcg->css); | |
ffbdccf5 | 1158 | rcu_read_unlock(); |
de077d22 | 1159 | } |
2314b42d JW |
1160 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1161 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1162 | return ret; |
1163 | } | |
1164 | ||
3e32cb2e | 1165 | #define mem_cgroup_from_counter(counter, member) \ |
6d61ef40 BS |
1166 | container_of(counter, struct mem_cgroup, member) |
1167 | ||
19942822 | 1168 | /** |
9d11ea9f | 1169 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1170 | * @memcg: the memory cgroup |
19942822 | 1171 | * |
9d11ea9f | 1172 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1173 | * pages. |
19942822 | 1174 | */ |
c0ff4b85 | 1175 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1176 | { |
3e32cb2e JW |
1177 | unsigned long margin = 0; |
1178 | unsigned long count; | |
1179 | unsigned long limit; | |
9d11ea9f | 1180 | |
3e32cb2e | 1181 | count = page_counter_read(&memcg->memory); |
4db0c3c2 | 1182 | limit = READ_ONCE(memcg->memory.limit); |
3e32cb2e JW |
1183 | if (count < limit) |
1184 | margin = limit - count; | |
1185 | ||
1186 | if (do_swap_account) { | |
1187 | count = page_counter_read(&memcg->memsw); | |
4db0c3c2 | 1188 | limit = READ_ONCE(memcg->memsw.limit); |
3e32cb2e JW |
1189 | if (count <= limit) |
1190 | margin = min(margin, limit - count); | |
1191 | } | |
1192 | ||
1193 | return margin; | |
19942822 JW |
1194 | } |
1195 | ||
32047e2a | 1196 | /* |
bdcbb659 | 1197 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1198 | * |
bdcbb659 QH |
1199 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1200 | * moving cgroups. This is for waiting at high-memory pressure | |
1201 | * caused by "move". | |
32047e2a | 1202 | */ |
c0ff4b85 | 1203 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1204 | { |
2bd9bb20 KH |
1205 | struct mem_cgroup *from; |
1206 | struct mem_cgroup *to; | |
4b534334 | 1207 | bool ret = false; |
2bd9bb20 KH |
1208 | /* |
1209 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1210 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1211 | */ | |
1212 | spin_lock(&mc.lock); | |
1213 | from = mc.from; | |
1214 | to = mc.to; | |
1215 | if (!from) | |
1216 | goto unlock; | |
3e92041d | 1217 | |
2314b42d JW |
1218 | ret = mem_cgroup_is_descendant(from, memcg) || |
1219 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1220 | unlock: |
1221 | spin_unlock(&mc.lock); | |
4b534334 KH |
1222 | return ret; |
1223 | } | |
1224 | ||
c0ff4b85 | 1225 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1226 | { |
1227 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1228 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1229 | DEFINE_WAIT(wait); |
1230 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1231 | /* moving charge context might have finished. */ | |
1232 | if (mc.moving_task) | |
1233 | schedule(); | |
1234 | finish_wait(&mc.waitq, &wait); | |
1235 | return true; | |
1236 | } | |
1237 | } | |
1238 | return false; | |
1239 | } | |
1240 | ||
58cf188e | 1241 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1242 | /** |
58cf188e | 1243 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1244 | * @memcg: The memory cgroup that went over limit |
1245 | * @p: Task that is going to be killed | |
1246 | * | |
1247 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1248 | * enabled | |
1249 | */ | |
1250 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1251 | { | |
e61734c5 | 1252 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
08088cb9 | 1253 | static DEFINE_MUTEX(oom_info_lock); |
58cf188e SZ |
1254 | struct mem_cgroup *iter; |
1255 | unsigned int i; | |
e222432b | 1256 | |
08088cb9 | 1257 | mutex_lock(&oom_info_lock); |
e222432b BS |
1258 | rcu_read_lock(); |
1259 | ||
2415b9f5 BV |
1260 | if (p) { |
1261 | pr_info("Task in "); | |
1262 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1263 | pr_cont(" killed as a result of limit of "); | |
1264 | } else { | |
1265 | pr_info("Memory limit reached of cgroup "); | |
1266 | } | |
1267 | ||
e61734c5 | 1268 | pr_cont_cgroup_path(memcg->css.cgroup); |
0346dadb | 1269 | pr_cont("\n"); |
e222432b | 1270 | |
e222432b BS |
1271 | rcu_read_unlock(); |
1272 | ||
3e32cb2e JW |
1273 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1274 | K((u64)page_counter_read(&memcg->memory)), | |
1275 | K((u64)memcg->memory.limit), memcg->memory.failcnt); | |
1276 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1277 | K((u64)page_counter_read(&memcg->memsw)), | |
1278 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); | |
1279 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1280 | K((u64)page_counter_read(&memcg->kmem)), | |
1281 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); | |
58cf188e SZ |
1282 | |
1283 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1284 | pr_info("Memory cgroup stats for "); |
1285 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1286 | pr_cont(":"); |
1287 | ||
1288 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1289 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1290 | continue; | |
484ebb3b | 1291 | pr_cont(" %s:%luKB", mem_cgroup_stat_names[i], |
58cf188e SZ |
1292 | K(mem_cgroup_read_stat(iter, i))); |
1293 | } | |
1294 | ||
1295 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1296 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1297 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1298 | ||
1299 | pr_cont("\n"); | |
1300 | } | |
08088cb9 | 1301 | mutex_unlock(&oom_info_lock); |
e222432b BS |
1302 | } |
1303 | ||
81d39c20 KH |
1304 | /* |
1305 | * This function returns the number of memcg under hierarchy tree. Returns | |
1306 | * 1(self count) if no children. | |
1307 | */ | |
c0ff4b85 | 1308 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1309 | { |
1310 | int num = 0; | |
7d74b06f KH |
1311 | struct mem_cgroup *iter; |
1312 | ||
c0ff4b85 | 1313 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1314 | num++; |
81d39c20 KH |
1315 | return num; |
1316 | } | |
1317 | ||
a63d83f4 DR |
1318 | /* |
1319 | * Return the memory (and swap, if configured) limit for a memcg. | |
1320 | */ | |
3e32cb2e | 1321 | static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 | 1322 | { |
3e32cb2e | 1323 | unsigned long limit; |
f3e8eb70 | 1324 | |
3e32cb2e | 1325 | limit = memcg->memory.limit; |
9a5a8f19 | 1326 | if (mem_cgroup_swappiness(memcg)) { |
3e32cb2e | 1327 | unsigned long memsw_limit; |
9a5a8f19 | 1328 | |
3e32cb2e JW |
1329 | memsw_limit = memcg->memsw.limit; |
1330 | limit = min(limit + total_swap_pages, memsw_limit); | |
9a5a8f19 | 1331 | } |
9a5a8f19 | 1332 | return limit; |
a63d83f4 DR |
1333 | } |
1334 | ||
0ccab5b1 | 1335 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1336 | int order) |
9cbb78bb | 1337 | { |
6e0fc46d DR |
1338 | struct oom_control oc = { |
1339 | .zonelist = NULL, | |
1340 | .nodemask = NULL, | |
1341 | .gfp_mask = gfp_mask, | |
1342 | .order = order, | |
6e0fc46d | 1343 | }; |
9cbb78bb DR |
1344 | struct mem_cgroup *iter; |
1345 | unsigned long chosen_points = 0; | |
1346 | unsigned long totalpages; | |
1347 | unsigned int points = 0; | |
1348 | struct task_struct *chosen = NULL; | |
1349 | ||
dc56401f JW |
1350 | mutex_lock(&oom_lock); |
1351 | ||
876aafbf | 1352 | /* |
465adcf1 DR |
1353 | * If current has a pending SIGKILL or is exiting, then automatically |
1354 | * select it. The goal is to allow it to allocate so that it may | |
1355 | * quickly exit and free its memory. | |
876aafbf | 1356 | */ |
d003f371 | 1357 | if (fatal_signal_pending(current) || task_will_free_mem(current)) { |
16e95196 | 1358 | mark_oom_victim(current); |
dc56401f | 1359 | goto unlock; |
876aafbf DR |
1360 | } |
1361 | ||
6e0fc46d | 1362 | check_panic_on_oom(&oc, CONSTRAINT_MEMCG, memcg); |
3e32cb2e | 1363 | totalpages = mem_cgroup_get_limit(memcg) ? : 1; |
9cbb78bb | 1364 | for_each_mem_cgroup_tree(iter, memcg) { |
72ec7029 | 1365 | struct css_task_iter it; |
9cbb78bb DR |
1366 | struct task_struct *task; |
1367 | ||
72ec7029 TH |
1368 | css_task_iter_start(&iter->css, &it); |
1369 | while ((task = css_task_iter_next(&it))) { | |
6e0fc46d | 1370 | switch (oom_scan_process_thread(&oc, task, totalpages)) { |
9cbb78bb DR |
1371 | case OOM_SCAN_SELECT: |
1372 | if (chosen) | |
1373 | put_task_struct(chosen); | |
1374 | chosen = task; | |
1375 | chosen_points = ULONG_MAX; | |
1376 | get_task_struct(chosen); | |
1377 | /* fall through */ | |
1378 | case OOM_SCAN_CONTINUE: | |
1379 | continue; | |
1380 | case OOM_SCAN_ABORT: | |
72ec7029 | 1381 | css_task_iter_end(&it); |
9cbb78bb DR |
1382 | mem_cgroup_iter_break(memcg, iter); |
1383 | if (chosen) | |
1384 | put_task_struct(chosen); | |
dc56401f | 1385 | goto unlock; |
9cbb78bb DR |
1386 | case OOM_SCAN_OK: |
1387 | break; | |
1388 | }; | |
1389 | points = oom_badness(task, memcg, NULL, totalpages); | |
d49ad935 DR |
1390 | if (!points || points < chosen_points) |
1391 | continue; | |
1392 | /* Prefer thread group leaders for display purposes */ | |
1393 | if (points == chosen_points && | |
1394 | thread_group_leader(chosen)) | |
1395 | continue; | |
1396 | ||
1397 | if (chosen) | |
1398 | put_task_struct(chosen); | |
1399 | chosen = task; | |
1400 | chosen_points = points; | |
1401 | get_task_struct(chosen); | |
9cbb78bb | 1402 | } |
72ec7029 | 1403 | css_task_iter_end(&it); |
9cbb78bb DR |
1404 | } |
1405 | ||
dc56401f JW |
1406 | if (chosen) { |
1407 | points = chosen_points * 1000 / totalpages; | |
6e0fc46d DR |
1408 | oom_kill_process(&oc, chosen, points, totalpages, memcg, |
1409 | "Memory cgroup out of memory"); | |
dc56401f JW |
1410 | } |
1411 | unlock: | |
1412 | mutex_unlock(&oom_lock); | |
0ccab5b1 | 1413 | return chosen; |
9cbb78bb DR |
1414 | } |
1415 | ||
ae6e71d3 MC |
1416 | #if MAX_NUMNODES > 1 |
1417 | ||
4d0c066d KH |
1418 | /** |
1419 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1420 | * @memcg: the target memcg |
4d0c066d KH |
1421 | * @nid: the node ID to be checked. |
1422 | * @noswap : specify true here if the user wants flle only information. | |
1423 | * | |
1424 | * This function returns whether the specified memcg contains any | |
1425 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1426 | * pages in the node. | |
1427 | */ | |
c0ff4b85 | 1428 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1429 | int nid, bool noswap) |
1430 | { | |
c0ff4b85 | 1431 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1432 | return true; |
1433 | if (noswap || !total_swap_pages) | |
1434 | return false; | |
c0ff4b85 | 1435 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1436 | return true; |
1437 | return false; | |
1438 | ||
1439 | } | |
889976db YH |
1440 | |
1441 | /* | |
1442 | * Always updating the nodemask is not very good - even if we have an empty | |
1443 | * list or the wrong list here, we can start from some node and traverse all | |
1444 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1445 | * | |
1446 | */ | |
c0ff4b85 | 1447 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1448 | { |
1449 | int nid; | |
453a9bf3 KH |
1450 | /* |
1451 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1452 | * pagein/pageout changes since the last update. | |
1453 | */ | |
c0ff4b85 | 1454 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1455 | return; |
c0ff4b85 | 1456 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1457 | return; |
1458 | ||
889976db | 1459 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1460 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1461 | |
31aaea4a | 1462 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1463 | |
c0ff4b85 R |
1464 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1465 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1466 | } |
453a9bf3 | 1467 | |
c0ff4b85 R |
1468 | atomic_set(&memcg->numainfo_events, 0); |
1469 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1470 | } |
1471 | ||
1472 | /* | |
1473 | * Selecting a node where we start reclaim from. Because what we need is just | |
1474 | * reducing usage counter, start from anywhere is O,K. Considering | |
1475 | * memory reclaim from current node, there are pros. and cons. | |
1476 | * | |
1477 | * Freeing memory from current node means freeing memory from a node which | |
1478 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1479 | * hit limits, it will see a contention on a node. But freeing from remote | |
1480 | * node means more costs for memory reclaim because of memory latency. | |
1481 | * | |
1482 | * Now, we use round-robin. Better algorithm is welcomed. | |
1483 | */ | |
c0ff4b85 | 1484 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1485 | { |
1486 | int node; | |
1487 | ||
c0ff4b85 R |
1488 | mem_cgroup_may_update_nodemask(memcg); |
1489 | node = memcg->last_scanned_node; | |
889976db | 1490 | |
c0ff4b85 | 1491 | node = next_node(node, memcg->scan_nodes); |
889976db | 1492 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1493 | node = first_node(memcg->scan_nodes); |
889976db YH |
1494 | /* |
1495 | * We call this when we hit limit, not when pages are added to LRU. | |
1496 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1497 | * memcg is too small and all pages are not on LRU. In that case, | |
1498 | * we use curret node. | |
1499 | */ | |
1500 | if (unlikely(node == MAX_NUMNODES)) | |
1501 | node = numa_node_id(); | |
1502 | ||
c0ff4b85 | 1503 | memcg->last_scanned_node = node; |
889976db YH |
1504 | return node; |
1505 | } | |
889976db | 1506 | #else |
c0ff4b85 | 1507 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1508 | { |
1509 | return 0; | |
1510 | } | |
1511 | #endif | |
1512 | ||
0608f43d AM |
1513 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1514 | struct zone *zone, | |
1515 | gfp_t gfp_mask, | |
1516 | unsigned long *total_scanned) | |
1517 | { | |
1518 | struct mem_cgroup *victim = NULL; | |
1519 | int total = 0; | |
1520 | int loop = 0; | |
1521 | unsigned long excess; | |
1522 | unsigned long nr_scanned; | |
1523 | struct mem_cgroup_reclaim_cookie reclaim = { | |
1524 | .zone = zone, | |
1525 | .priority = 0, | |
1526 | }; | |
1527 | ||
3e32cb2e | 1528 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1529 | |
1530 | while (1) { | |
1531 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1532 | if (!victim) { | |
1533 | loop++; | |
1534 | if (loop >= 2) { | |
1535 | /* | |
1536 | * If we have not been able to reclaim | |
1537 | * anything, it might because there are | |
1538 | * no reclaimable pages under this hierarchy | |
1539 | */ | |
1540 | if (!total) | |
1541 | break; | |
1542 | /* | |
1543 | * We want to do more targeted reclaim. | |
1544 | * excess >> 2 is not to excessive so as to | |
1545 | * reclaim too much, nor too less that we keep | |
1546 | * coming back to reclaim from this cgroup | |
1547 | */ | |
1548 | if (total >= (excess >> 2) || | |
1549 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1550 | break; | |
1551 | } | |
1552 | continue; | |
1553 | } | |
0608f43d AM |
1554 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1555 | zone, &nr_scanned); | |
1556 | *total_scanned += nr_scanned; | |
3e32cb2e | 1557 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1558 | break; |
6d61ef40 | 1559 | } |
0608f43d AM |
1560 | mem_cgroup_iter_break(root_memcg, victim); |
1561 | return total; | |
6d61ef40 BS |
1562 | } |
1563 | ||
0056f4e6 JW |
1564 | #ifdef CONFIG_LOCKDEP |
1565 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1566 | .name = "memcg_oom_lock", | |
1567 | }; | |
1568 | #endif | |
1569 | ||
fb2a6fc5 JW |
1570 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1571 | ||
867578cb KH |
1572 | /* |
1573 | * Check OOM-Killer is already running under our hierarchy. | |
1574 | * If someone is running, return false. | |
1575 | */ | |
fb2a6fc5 | 1576 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1577 | { |
79dfdacc | 1578 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1579 | |
fb2a6fc5 JW |
1580 | spin_lock(&memcg_oom_lock); |
1581 | ||
9f3a0d09 | 1582 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1583 | if (iter->oom_lock) { |
79dfdacc MH |
1584 | /* |
1585 | * this subtree of our hierarchy is already locked | |
1586 | * so we cannot give a lock. | |
1587 | */ | |
79dfdacc | 1588 | failed = iter; |
9f3a0d09 JW |
1589 | mem_cgroup_iter_break(memcg, iter); |
1590 | break; | |
23751be0 JW |
1591 | } else |
1592 | iter->oom_lock = true; | |
7d74b06f | 1593 | } |
867578cb | 1594 | |
fb2a6fc5 JW |
1595 | if (failed) { |
1596 | /* | |
1597 | * OK, we failed to lock the whole subtree so we have | |
1598 | * to clean up what we set up to the failing subtree | |
1599 | */ | |
1600 | for_each_mem_cgroup_tree(iter, memcg) { | |
1601 | if (iter == failed) { | |
1602 | mem_cgroup_iter_break(memcg, iter); | |
1603 | break; | |
1604 | } | |
1605 | iter->oom_lock = false; | |
79dfdacc | 1606 | } |
0056f4e6 JW |
1607 | } else |
1608 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1609 | |
1610 | spin_unlock(&memcg_oom_lock); | |
1611 | ||
1612 | return !failed; | |
a636b327 | 1613 | } |
0b7f569e | 1614 | |
fb2a6fc5 | 1615 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1616 | { |
7d74b06f KH |
1617 | struct mem_cgroup *iter; |
1618 | ||
fb2a6fc5 | 1619 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1620 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1621 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1622 | iter->oom_lock = false; |
fb2a6fc5 | 1623 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1624 | } |
1625 | ||
c0ff4b85 | 1626 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1627 | { |
1628 | struct mem_cgroup *iter; | |
1629 | ||
c2b42d3c | 1630 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1631 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1632 | iter->under_oom++; |
1633 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1634 | } |
1635 | ||
c0ff4b85 | 1636 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1637 | { |
1638 | struct mem_cgroup *iter; | |
1639 | ||
867578cb KH |
1640 | /* |
1641 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1642 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1643 | */ |
c2b42d3c | 1644 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1645 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1646 | if (iter->under_oom > 0) |
1647 | iter->under_oom--; | |
1648 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1649 | } |
1650 | ||
867578cb KH |
1651 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1652 | ||
dc98df5a | 1653 | struct oom_wait_info { |
d79154bb | 1654 | struct mem_cgroup *memcg; |
dc98df5a KH |
1655 | wait_queue_t wait; |
1656 | }; | |
1657 | ||
1658 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1659 | unsigned mode, int sync, void *arg) | |
1660 | { | |
d79154bb HD |
1661 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1662 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1663 | struct oom_wait_info *oom_wait_info; |
1664 | ||
1665 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1666 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1667 | |
2314b42d JW |
1668 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1669 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1670 | return 0; |
dc98df5a KH |
1671 | return autoremove_wake_function(wait, mode, sync, arg); |
1672 | } | |
1673 | ||
c0ff4b85 | 1674 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1675 | { |
c2b42d3c TH |
1676 | /* |
1677 | * For the following lockless ->under_oom test, the only required | |
1678 | * guarantee is that it must see the state asserted by an OOM when | |
1679 | * this function is called as a result of userland actions | |
1680 | * triggered by the notification of the OOM. This is trivially | |
1681 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1682 | * triggering notification. | |
1683 | */ | |
1684 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1685 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1686 | } |
1687 | ||
3812c8c8 | 1688 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 1689 | { |
626ebc41 | 1690 | if (!current->memcg_may_oom) |
3812c8c8 | 1691 | return; |
867578cb | 1692 | /* |
49426420 JW |
1693 | * We are in the middle of the charge context here, so we |
1694 | * don't want to block when potentially sitting on a callstack | |
1695 | * that holds all kinds of filesystem and mm locks. | |
1696 | * | |
1697 | * Also, the caller may handle a failed allocation gracefully | |
1698 | * (like optional page cache readahead) and so an OOM killer | |
1699 | * invocation might not even be necessary. | |
1700 | * | |
1701 | * That's why we don't do anything here except remember the | |
1702 | * OOM context and then deal with it at the end of the page | |
1703 | * fault when the stack is unwound, the locks are released, | |
1704 | * and when we know whether the fault was overall successful. | |
867578cb | 1705 | */ |
49426420 | 1706 | css_get(&memcg->css); |
626ebc41 TH |
1707 | current->memcg_in_oom = memcg; |
1708 | current->memcg_oom_gfp_mask = mask; | |
1709 | current->memcg_oom_order = order; | |
3812c8c8 JW |
1710 | } |
1711 | ||
1712 | /** | |
1713 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1714 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1715 | * |
49426420 JW |
1716 | * This has to be called at the end of a page fault if the memcg OOM |
1717 | * handler was enabled. | |
3812c8c8 | 1718 | * |
49426420 | 1719 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1720 | * sleep on a waitqueue until the userspace task resolves the |
1721 | * situation. Sleeping directly in the charge context with all kinds | |
1722 | * of locks held is not a good idea, instead we remember an OOM state | |
1723 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1724 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1725 | * |
1726 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1727 | * completed, %false otherwise. |
3812c8c8 | 1728 | */ |
49426420 | 1729 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1730 | { |
626ebc41 | 1731 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1732 | struct oom_wait_info owait; |
49426420 | 1733 | bool locked; |
3812c8c8 JW |
1734 | |
1735 | /* OOM is global, do not handle */ | |
3812c8c8 | 1736 | if (!memcg) |
49426420 | 1737 | return false; |
3812c8c8 | 1738 | |
c32b3cbe | 1739 | if (!handle || oom_killer_disabled) |
49426420 | 1740 | goto cleanup; |
3812c8c8 JW |
1741 | |
1742 | owait.memcg = memcg; | |
1743 | owait.wait.flags = 0; | |
1744 | owait.wait.func = memcg_oom_wake_function; | |
1745 | owait.wait.private = current; | |
1746 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 1747 | |
3812c8c8 | 1748 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1749 | mem_cgroup_mark_under_oom(memcg); |
1750 | ||
1751 | locked = mem_cgroup_oom_trylock(memcg); | |
1752 | ||
1753 | if (locked) | |
1754 | mem_cgroup_oom_notify(memcg); | |
1755 | ||
1756 | if (locked && !memcg->oom_kill_disable) { | |
1757 | mem_cgroup_unmark_under_oom(memcg); | |
1758 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1759 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1760 | current->memcg_oom_order); | |
49426420 | 1761 | } else { |
3812c8c8 | 1762 | schedule(); |
49426420 JW |
1763 | mem_cgroup_unmark_under_oom(memcg); |
1764 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1765 | } | |
1766 | ||
1767 | if (locked) { | |
fb2a6fc5 JW |
1768 | mem_cgroup_oom_unlock(memcg); |
1769 | /* | |
1770 | * There is no guarantee that an OOM-lock contender | |
1771 | * sees the wakeups triggered by the OOM kill | |
1772 | * uncharges. Wake any sleepers explicitely. | |
1773 | */ | |
1774 | memcg_oom_recover(memcg); | |
1775 | } | |
49426420 | 1776 | cleanup: |
626ebc41 | 1777 | current->memcg_in_oom = NULL; |
3812c8c8 | 1778 | css_put(&memcg->css); |
867578cb | 1779 | return true; |
0b7f569e KH |
1780 | } |
1781 | ||
d7365e78 JW |
1782 | /** |
1783 | * mem_cgroup_begin_page_stat - begin a page state statistics transaction | |
1784 | * @page: page that is going to change accounted state | |
32047e2a | 1785 | * |
d7365e78 JW |
1786 | * This function must mark the beginning of an accounted page state |
1787 | * change to prevent double accounting when the page is concurrently | |
1788 | * being moved to another memcg: | |
32047e2a | 1789 | * |
6de22619 | 1790 | * memcg = mem_cgroup_begin_page_stat(page); |
d7365e78 JW |
1791 | * if (TestClearPageState(page)) |
1792 | * mem_cgroup_update_page_stat(memcg, state, -1); | |
6de22619 | 1793 | * mem_cgroup_end_page_stat(memcg); |
d69b042f | 1794 | */ |
6de22619 | 1795 | struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page) |
89c06bd5 KH |
1796 | { |
1797 | struct mem_cgroup *memcg; | |
6de22619 | 1798 | unsigned long flags; |
89c06bd5 | 1799 | |
6de22619 JW |
1800 | /* |
1801 | * The RCU lock is held throughout the transaction. The fast | |
1802 | * path can get away without acquiring the memcg->move_lock | |
1803 | * because page moving starts with an RCU grace period. | |
1804 | * | |
1805 | * The RCU lock also protects the memcg from being freed when | |
1806 | * the page state that is going to change is the only thing | |
1807 | * preventing the page from being uncharged. | |
1808 | * E.g. end-writeback clearing PageWriteback(), which allows | |
1809 | * migration to go ahead and uncharge the page before the | |
1810 | * account transaction might be complete. | |
1811 | */ | |
d7365e78 JW |
1812 | rcu_read_lock(); |
1813 | ||
1814 | if (mem_cgroup_disabled()) | |
1815 | return NULL; | |
89c06bd5 | 1816 | again: |
1306a85a | 1817 | memcg = page->mem_cgroup; |
29833315 | 1818 | if (unlikely(!memcg)) |
d7365e78 JW |
1819 | return NULL; |
1820 | ||
bdcbb659 | 1821 | if (atomic_read(&memcg->moving_account) <= 0) |
d7365e78 | 1822 | return memcg; |
89c06bd5 | 1823 | |
6de22619 | 1824 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 1825 | if (memcg != page->mem_cgroup) { |
6de22619 | 1826 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
1827 | goto again; |
1828 | } | |
6de22619 JW |
1829 | |
1830 | /* | |
1831 | * When charge migration first begins, we can have locked and | |
1832 | * unlocked page stat updates happening concurrently. Track | |
1833 | * the task who has the lock for mem_cgroup_end_page_stat(). | |
1834 | */ | |
1835 | memcg->move_lock_task = current; | |
1836 | memcg->move_lock_flags = flags; | |
d7365e78 JW |
1837 | |
1838 | return memcg; | |
89c06bd5 | 1839 | } |
c4843a75 | 1840 | EXPORT_SYMBOL(mem_cgroup_begin_page_stat); |
89c06bd5 | 1841 | |
d7365e78 JW |
1842 | /** |
1843 | * mem_cgroup_end_page_stat - finish a page state statistics transaction | |
1844 | * @memcg: the memcg that was accounted against | |
d7365e78 | 1845 | */ |
6de22619 | 1846 | void mem_cgroup_end_page_stat(struct mem_cgroup *memcg) |
89c06bd5 | 1847 | { |
6de22619 JW |
1848 | if (memcg && memcg->move_lock_task == current) { |
1849 | unsigned long flags = memcg->move_lock_flags; | |
1850 | ||
1851 | memcg->move_lock_task = NULL; | |
1852 | memcg->move_lock_flags = 0; | |
1853 | ||
1854 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
1855 | } | |
89c06bd5 | 1856 | |
d7365e78 | 1857 | rcu_read_unlock(); |
89c06bd5 | 1858 | } |
c4843a75 | 1859 | EXPORT_SYMBOL(mem_cgroup_end_page_stat); |
89c06bd5 | 1860 | |
cdec2e42 KH |
1861 | /* |
1862 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1863 | * TODO: maybe necessary to use big numbers in big irons. | |
1864 | */ | |
7ec99d62 | 1865 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
1866 | struct memcg_stock_pcp { |
1867 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 1868 | unsigned int nr_pages; |
cdec2e42 | 1869 | struct work_struct work; |
26fe6168 | 1870 | unsigned long flags; |
a0db00fc | 1871 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
1872 | }; |
1873 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 1874 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 1875 | |
a0956d54 SS |
1876 | /** |
1877 | * consume_stock: Try to consume stocked charge on this cpu. | |
1878 | * @memcg: memcg to consume from. | |
1879 | * @nr_pages: how many pages to charge. | |
1880 | * | |
1881 | * The charges will only happen if @memcg matches the current cpu's memcg | |
1882 | * stock, and at least @nr_pages are available in that stock. Failure to | |
1883 | * service an allocation will refill the stock. | |
1884 | * | |
1885 | * returns true if successful, false otherwise. | |
cdec2e42 | 1886 | */ |
a0956d54 | 1887 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1888 | { |
1889 | struct memcg_stock_pcp *stock; | |
3e32cb2e | 1890 | bool ret = false; |
cdec2e42 | 1891 | |
a0956d54 | 1892 | if (nr_pages > CHARGE_BATCH) |
3e32cb2e | 1893 | return ret; |
a0956d54 | 1894 | |
cdec2e42 | 1895 | stock = &get_cpu_var(memcg_stock); |
3e32cb2e | 1896 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 1897 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
1898 | ret = true; |
1899 | } | |
cdec2e42 KH |
1900 | put_cpu_var(memcg_stock); |
1901 | return ret; | |
1902 | } | |
1903 | ||
1904 | /* | |
3e32cb2e | 1905 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
1906 | */ |
1907 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1908 | { | |
1909 | struct mem_cgroup *old = stock->cached; | |
1910 | ||
11c9ea4e | 1911 | if (stock->nr_pages) { |
3e32cb2e | 1912 | page_counter_uncharge(&old->memory, stock->nr_pages); |
cdec2e42 | 1913 | if (do_swap_account) |
3e32cb2e | 1914 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 1915 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 1916 | stock->nr_pages = 0; |
cdec2e42 KH |
1917 | } |
1918 | stock->cached = NULL; | |
cdec2e42 KH |
1919 | } |
1920 | ||
1921 | /* | |
1922 | * This must be called under preempt disabled or must be called by | |
1923 | * a thread which is pinned to local cpu. | |
1924 | */ | |
1925 | static void drain_local_stock(struct work_struct *dummy) | |
1926 | { | |
7c8e0181 | 1927 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
cdec2e42 | 1928 | drain_stock(stock); |
26fe6168 | 1929 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
1930 | } |
1931 | ||
1932 | /* | |
3e32cb2e | 1933 | * Cache charges(val) to local per_cpu area. |
320cc51d | 1934 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 1935 | */ |
c0ff4b85 | 1936 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1937 | { |
1938 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
1939 | ||
c0ff4b85 | 1940 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 1941 | drain_stock(stock); |
c0ff4b85 | 1942 | stock->cached = memcg; |
cdec2e42 | 1943 | } |
11c9ea4e | 1944 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
1945 | put_cpu_var(memcg_stock); |
1946 | } | |
1947 | ||
1948 | /* | |
c0ff4b85 | 1949 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 1950 | * of the hierarchy under it. |
cdec2e42 | 1951 | */ |
6d3d6aa2 | 1952 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 1953 | { |
26fe6168 | 1954 | int cpu, curcpu; |
d38144b7 | 1955 | |
6d3d6aa2 JW |
1956 | /* If someone's already draining, avoid adding running more workers. */ |
1957 | if (!mutex_trylock(&percpu_charge_mutex)) | |
1958 | return; | |
cdec2e42 | 1959 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 1960 | get_online_cpus(); |
5af12d0e | 1961 | curcpu = get_cpu(); |
cdec2e42 KH |
1962 | for_each_online_cpu(cpu) { |
1963 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 1964 | struct mem_cgroup *memcg; |
26fe6168 | 1965 | |
c0ff4b85 R |
1966 | memcg = stock->cached; |
1967 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 1968 | continue; |
2314b42d | 1969 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) |
3e92041d | 1970 | continue; |
d1a05b69 MH |
1971 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
1972 | if (cpu == curcpu) | |
1973 | drain_local_stock(&stock->work); | |
1974 | else | |
1975 | schedule_work_on(cpu, &stock->work); | |
1976 | } | |
cdec2e42 | 1977 | } |
5af12d0e | 1978 | put_cpu(); |
f894ffa8 | 1979 | put_online_cpus(); |
9f50fad6 | 1980 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
1981 | } |
1982 | ||
0db0628d | 1983 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
1984 | unsigned long action, |
1985 | void *hcpu) | |
1986 | { | |
1987 | int cpu = (unsigned long)hcpu; | |
1988 | struct memcg_stock_pcp *stock; | |
1989 | ||
619d094b | 1990 | if (action == CPU_ONLINE) |
1489ebad | 1991 | return NOTIFY_OK; |
1489ebad | 1992 | |
d833049b | 1993 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 1994 | return NOTIFY_OK; |
711d3d2c | 1995 | |
cdec2e42 KH |
1996 | stock = &per_cpu(memcg_stock, cpu); |
1997 | drain_stock(stock); | |
1998 | return NOTIFY_OK; | |
1999 | } | |
2000 | ||
b23afb93 TH |
2001 | /* |
2002 | * Scheduled by try_charge() to be executed from the userland return path | |
2003 | * and reclaims memory over the high limit. | |
2004 | */ | |
2005 | void mem_cgroup_handle_over_high(void) | |
2006 | { | |
2007 | unsigned int nr_pages = current->memcg_nr_pages_over_high; | |
2008 | struct mem_cgroup *memcg, *pos; | |
2009 | ||
2010 | if (likely(!nr_pages)) | |
2011 | return; | |
2012 | ||
2013 | pos = memcg = get_mem_cgroup_from_mm(current->mm); | |
2014 | ||
2015 | do { | |
2016 | if (page_counter_read(&pos->memory) <= pos->high) | |
2017 | continue; | |
2018 | mem_cgroup_events(pos, MEMCG_HIGH, 1); | |
2019 | try_to_free_mem_cgroup_pages(pos, nr_pages, GFP_KERNEL, true); | |
2020 | } while ((pos = parent_mem_cgroup(pos))); | |
2021 | ||
2022 | css_put(&memcg->css); | |
2023 | current->memcg_nr_pages_over_high = 0; | |
2024 | } | |
2025 | ||
00501b53 JW |
2026 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2027 | unsigned int nr_pages) | |
8a9f3ccd | 2028 | { |
7ec99d62 | 2029 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 2030 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 2031 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2032 | struct page_counter *counter; |
6539cc05 | 2033 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2034 | bool may_swap = true; |
2035 | bool drained = false; | |
a636b327 | 2036 | |
ce00a967 | 2037 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 2038 | return 0; |
6539cc05 | 2039 | retry: |
b6b6cc72 | 2040 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2041 | return 0; |
8a9f3ccd | 2042 | |
3fbe7244 | 2043 | if (!do_swap_account || |
6071ca52 JW |
2044 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2045 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2046 | goto done_restock; |
3fbe7244 | 2047 | if (do_swap_account) |
3e32cb2e JW |
2048 | page_counter_uncharge(&memcg->memsw, batch); |
2049 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2050 | } else { |
3e32cb2e | 2051 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2052 | may_swap = false; |
3fbe7244 | 2053 | } |
7a81b88c | 2054 | |
6539cc05 JW |
2055 | if (batch > nr_pages) { |
2056 | batch = nr_pages; | |
2057 | goto retry; | |
2058 | } | |
6d61ef40 | 2059 | |
06b078fc JW |
2060 | /* |
2061 | * Unlike in global OOM situations, memcg is not in a physical | |
2062 | * memory shortage. Allow dying and OOM-killed tasks to | |
2063 | * bypass the last charges so that they can exit quickly and | |
2064 | * free their memory. | |
2065 | */ | |
2066 | if (unlikely(test_thread_flag(TIF_MEMDIE) || | |
2067 | fatal_signal_pending(current) || | |
2068 | current->flags & PF_EXITING)) | |
10d53c74 | 2069 | goto force; |
06b078fc JW |
2070 | |
2071 | if (unlikely(task_in_memcg_oom(current))) | |
2072 | goto nomem; | |
2073 | ||
d0164adc | 2074 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2075 | goto nomem; |
4b534334 | 2076 | |
241994ed JW |
2077 | mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1); |
2078 | ||
b70a2a21 JW |
2079 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2080 | gfp_mask, may_swap); | |
6539cc05 | 2081 | |
61e02c74 | 2082 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2083 | goto retry; |
28c34c29 | 2084 | |
b70a2a21 | 2085 | if (!drained) { |
6d3d6aa2 | 2086 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2087 | drained = true; |
2088 | goto retry; | |
2089 | } | |
2090 | ||
28c34c29 JW |
2091 | if (gfp_mask & __GFP_NORETRY) |
2092 | goto nomem; | |
6539cc05 JW |
2093 | /* |
2094 | * Even though the limit is exceeded at this point, reclaim | |
2095 | * may have been able to free some pages. Retry the charge | |
2096 | * before killing the task. | |
2097 | * | |
2098 | * Only for regular pages, though: huge pages are rather | |
2099 | * unlikely to succeed so close to the limit, and we fall back | |
2100 | * to regular pages anyway in case of failure. | |
2101 | */ | |
61e02c74 | 2102 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2103 | goto retry; |
2104 | /* | |
2105 | * At task move, charge accounts can be doubly counted. So, it's | |
2106 | * better to wait until the end of task_move if something is going on. | |
2107 | */ | |
2108 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2109 | goto retry; | |
2110 | ||
9b130619 JW |
2111 | if (nr_retries--) |
2112 | goto retry; | |
2113 | ||
06b078fc | 2114 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2115 | goto force; |
06b078fc | 2116 | |
6539cc05 | 2117 | if (fatal_signal_pending(current)) |
10d53c74 | 2118 | goto force; |
6539cc05 | 2119 | |
241994ed JW |
2120 | mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1); |
2121 | ||
3608de07 JM |
2122 | mem_cgroup_oom(mem_over_limit, gfp_mask, |
2123 | get_order(nr_pages * PAGE_SIZE)); | |
7a81b88c | 2124 | nomem: |
6d1fdc48 | 2125 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2126 | return -ENOMEM; |
10d53c74 TH |
2127 | force: |
2128 | /* | |
2129 | * The allocation either can't fail or will lead to more memory | |
2130 | * being freed very soon. Allow memory usage go over the limit | |
2131 | * temporarily by force charging it. | |
2132 | */ | |
2133 | page_counter_charge(&memcg->memory, nr_pages); | |
2134 | if (do_swap_account) | |
2135 | page_counter_charge(&memcg->memsw, nr_pages); | |
2136 | css_get_many(&memcg->css, nr_pages); | |
2137 | ||
2138 | return 0; | |
6539cc05 JW |
2139 | |
2140 | done_restock: | |
e8ea14cc | 2141 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2142 | if (batch > nr_pages) |
2143 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2144 | |
241994ed | 2145 | /* |
b23afb93 TH |
2146 | * If the hierarchy is above the normal consumption range, schedule |
2147 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2148 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2149 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2150 | * not recorded as it most likely matches current's and won't | |
2151 | * change in the meantime. As high limit is checked again before | |
2152 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2153 | */ |
2154 | do { | |
b23afb93 | 2155 | if (page_counter_read(&memcg->memory) > memcg->high) { |
9516a18a | 2156 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2157 | set_notify_resume(current); |
2158 | break; | |
2159 | } | |
241994ed | 2160 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2161 | |
2162 | return 0; | |
7a81b88c | 2163 | } |
8a9f3ccd | 2164 | |
00501b53 | 2165 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2166 | { |
ce00a967 JW |
2167 | if (mem_cgroup_is_root(memcg)) |
2168 | return; | |
2169 | ||
3e32cb2e | 2170 | page_counter_uncharge(&memcg->memory, nr_pages); |
05b84301 | 2171 | if (do_swap_account) |
3e32cb2e | 2172 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2173 | |
e8ea14cc | 2174 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2175 | } |
2176 | ||
0a31bc97 JW |
2177 | static void lock_page_lru(struct page *page, int *isolated) |
2178 | { | |
2179 | struct zone *zone = page_zone(page); | |
2180 | ||
2181 | spin_lock_irq(&zone->lru_lock); | |
2182 | if (PageLRU(page)) { | |
2183 | struct lruvec *lruvec; | |
2184 | ||
2185 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2186 | ClearPageLRU(page); | |
2187 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2188 | *isolated = 1; | |
2189 | } else | |
2190 | *isolated = 0; | |
2191 | } | |
2192 | ||
2193 | static void unlock_page_lru(struct page *page, int isolated) | |
2194 | { | |
2195 | struct zone *zone = page_zone(page); | |
2196 | ||
2197 | if (isolated) { | |
2198 | struct lruvec *lruvec; | |
2199 | ||
2200 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2201 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
2202 | SetPageLRU(page); | |
2203 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2204 | } | |
2205 | spin_unlock_irq(&zone->lru_lock); | |
2206 | } | |
2207 | ||
00501b53 | 2208 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2209 | bool lrucare) |
7a81b88c | 2210 | { |
0a31bc97 | 2211 | int isolated; |
9ce70c02 | 2212 | |
1306a85a | 2213 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2214 | |
2215 | /* | |
2216 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2217 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2218 | */ | |
0a31bc97 JW |
2219 | if (lrucare) |
2220 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2221 | |
0a31bc97 JW |
2222 | /* |
2223 | * Nobody should be changing or seriously looking at | |
1306a85a | 2224 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2225 | * |
2226 | * - the page is uncharged | |
2227 | * | |
2228 | * - the page is off-LRU | |
2229 | * | |
2230 | * - an anonymous fault has exclusive page access, except for | |
2231 | * a locked page table | |
2232 | * | |
2233 | * - a page cache insertion, a swapin fault, or a migration | |
2234 | * have the page locked | |
2235 | */ | |
1306a85a | 2236 | page->mem_cgroup = memcg; |
9ce70c02 | 2237 | |
0a31bc97 JW |
2238 | if (lrucare) |
2239 | unlock_page_lru(page, isolated); | |
7a81b88c | 2240 | } |
66e1707b | 2241 | |
7ae1e1d0 | 2242 | #ifdef CONFIG_MEMCG_KMEM |
f3bb3043 | 2243 | static int memcg_alloc_cache_id(void) |
55007d84 | 2244 | { |
f3bb3043 VD |
2245 | int id, size; |
2246 | int err; | |
2247 | ||
dbcf73e2 | 2248 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2249 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2250 | if (id < 0) | |
2251 | return id; | |
55007d84 | 2252 | |
dbcf73e2 | 2253 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2254 | return id; |
2255 | ||
2256 | /* | |
2257 | * There's no space for the new id in memcg_caches arrays, | |
2258 | * so we have to grow them. | |
2259 | */ | |
05257a1a | 2260 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2261 | |
2262 | size = 2 * (id + 1); | |
55007d84 GC |
2263 | if (size < MEMCG_CACHES_MIN_SIZE) |
2264 | size = MEMCG_CACHES_MIN_SIZE; | |
2265 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2266 | size = MEMCG_CACHES_MAX_SIZE; | |
2267 | ||
f3bb3043 | 2268 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2269 | if (!err) |
2270 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2271 | if (!err) |
2272 | memcg_nr_cache_ids = size; | |
2273 | ||
2274 | up_write(&memcg_cache_ids_sem); | |
2275 | ||
f3bb3043 | 2276 | if (err) { |
dbcf73e2 | 2277 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2278 | return err; |
2279 | } | |
2280 | return id; | |
2281 | } | |
2282 | ||
2283 | static void memcg_free_cache_id(int id) | |
2284 | { | |
dbcf73e2 | 2285 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2286 | } |
2287 | ||
d5b3cf71 | 2288 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2289 | struct mem_cgroup *memcg; |
2290 | struct kmem_cache *cachep; | |
2291 | struct work_struct work; | |
2292 | }; | |
2293 | ||
d5b3cf71 | 2294 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2295 | { |
d5b3cf71 VD |
2296 | struct memcg_kmem_cache_create_work *cw = |
2297 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2298 | struct mem_cgroup *memcg = cw->memcg; |
2299 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2300 | |
d5b3cf71 | 2301 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2302 | |
5722d094 | 2303 | css_put(&memcg->css); |
d7f25f8a GC |
2304 | kfree(cw); |
2305 | } | |
2306 | ||
2307 | /* | |
2308 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2309 | */ |
d5b3cf71 VD |
2310 | static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2311 | struct kmem_cache *cachep) | |
d7f25f8a | 2312 | { |
d5b3cf71 | 2313 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2314 | |
776ed0f0 | 2315 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
8135be5a | 2316 | if (!cw) |
d7f25f8a | 2317 | return; |
8135be5a VD |
2318 | |
2319 | css_get(&memcg->css); | |
d7f25f8a GC |
2320 | |
2321 | cw->memcg = memcg; | |
2322 | cw->cachep = cachep; | |
d5b3cf71 | 2323 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2324 | |
d7f25f8a GC |
2325 | schedule_work(&cw->work); |
2326 | } | |
2327 | ||
d5b3cf71 VD |
2328 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2329 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
2330 | { |
2331 | /* | |
2332 | * We need to stop accounting when we kmalloc, because if the | |
2333 | * corresponding kmalloc cache is not yet created, the first allocation | |
d5b3cf71 | 2334 | * in __memcg_schedule_kmem_cache_create will recurse. |
0e9d92f2 GC |
2335 | * |
2336 | * However, it is better to enclose the whole function. Depending on | |
2337 | * the debugging options enabled, INIT_WORK(), for instance, can | |
2338 | * trigger an allocation. This too, will make us recurse. Because at | |
2339 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
2340 | * the safest choice is to do it like this, wrapping the whole function. | |
2341 | */ | |
6f185c29 | 2342 | current->memcg_kmem_skip_account = 1; |
d5b3cf71 | 2343 | __memcg_schedule_kmem_cache_create(memcg, cachep); |
6f185c29 | 2344 | current->memcg_kmem_skip_account = 0; |
0e9d92f2 | 2345 | } |
c67a8a68 | 2346 | |
d7f25f8a GC |
2347 | /* |
2348 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
2349 | * We try to use the current memcg's version of the cache. | |
2350 | * | |
2351 | * If the cache does not exist yet, if we are the first user of it, | |
2352 | * we either create it immediately, if possible, or create it asynchronously | |
2353 | * in a workqueue. | |
2354 | * In the latter case, we will let the current allocation go through with | |
2355 | * the original cache. | |
2356 | * | |
2357 | * Can't be called in interrupt context or from kernel threads. | |
2358 | * This function needs to be called with rcu_read_lock() held. | |
2359 | */ | |
056b7cce | 2360 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep) |
d7f25f8a GC |
2361 | { |
2362 | struct mem_cgroup *memcg; | |
959c8963 | 2363 | struct kmem_cache *memcg_cachep; |
2a4db7eb | 2364 | int kmemcg_id; |
d7f25f8a | 2365 | |
f7ce3190 | 2366 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2367 | |
9d100c5e | 2368 | if (current->memcg_kmem_skip_account) |
0e9d92f2 GC |
2369 | return cachep; |
2370 | ||
8135be5a | 2371 | memcg = get_mem_cgroup_from_mm(current->mm); |
4db0c3c2 | 2372 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
2a4db7eb | 2373 | if (kmemcg_id < 0) |
ca0dde97 | 2374 | goto out; |
d7f25f8a | 2375 | |
2a4db7eb | 2376 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
8135be5a VD |
2377 | if (likely(memcg_cachep)) |
2378 | return memcg_cachep; | |
ca0dde97 LZ |
2379 | |
2380 | /* | |
2381 | * If we are in a safe context (can wait, and not in interrupt | |
2382 | * context), we could be be predictable and return right away. | |
2383 | * This would guarantee that the allocation being performed | |
2384 | * already belongs in the new cache. | |
2385 | * | |
2386 | * However, there are some clashes that can arrive from locking. | |
2387 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2388 | * memcg_create_kmem_cache, this means no further allocation |
2389 | * could happen with the slab_mutex held. So it's better to | |
2390 | * defer everything. | |
ca0dde97 | 2391 | */ |
d5b3cf71 | 2392 | memcg_schedule_kmem_cache_create(memcg, cachep); |
ca0dde97 | 2393 | out: |
8135be5a | 2394 | css_put(&memcg->css); |
ca0dde97 | 2395 | return cachep; |
d7f25f8a | 2396 | } |
d7f25f8a | 2397 | |
8135be5a VD |
2398 | void __memcg_kmem_put_cache(struct kmem_cache *cachep) |
2399 | { | |
2400 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2401 | css_put(&cachep->memcg_params.memcg->css); |
8135be5a VD |
2402 | } |
2403 | ||
f3ccb2c4 VD |
2404 | int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, |
2405 | struct mem_cgroup *memcg) | |
7ae1e1d0 | 2406 | { |
f3ccb2c4 VD |
2407 | unsigned int nr_pages = 1 << order; |
2408 | struct page_counter *counter; | |
7ae1e1d0 GC |
2409 | int ret; |
2410 | ||
f3ccb2c4 | 2411 | if (!memcg_kmem_is_active(memcg)) |
d05e83a6 | 2412 | return 0; |
6d42c232 | 2413 | |
6071ca52 JW |
2414 | if (!page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) |
2415 | return -ENOMEM; | |
7ae1e1d0 | 2416 | |
f3ccb2c4 VD |
2417 | ret = try_charge(memcg, gfp, nr_pages); |
2418 | if (ret) { | |
2419 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
2420 | return ret; | |
7ae1e1d0 GC |
2421 | } |
2422 | ||
f3ccb2c4 | 2423 | page->mem_cgroup = memcg; |
7ae1e1d0 | 2424 | |
f3ccb2c4 | 2425 | return 0; |
7ae1e1d0 GC |
2426 | } |
2427 | ||
f3ccb2c4 | 2428 | int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 2429 | { |
f3ccb2c4 VD |
2430 | struct mem_cgroup *memcg; |
2431 | int ret; | |
7ae1e1d0 | 2432 | |
f3ccb2c4 VD |
2433 | memcg = get_mem_cgroup_from_mm(current->mm); |
2434 | ret = __memcg_kmem_charge_memcg(page, gfp, order, memcg); | |
7ae1e1d0 | 2435 | css_put(&memcg->css); |
d05e83a6 | 2436 | return ret; |
7ae1e1d0 GC |
2437 | } |
2438 | ||
d05e83a6 | 2439 | void __memcg_kmem_uncharge(struct page *page, int order) |
7ae1e1d0 | 2440 | { |
1306a85a | 2441 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 2442 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 2443 | |
7ae1e1d0 GC |
2444 | if (!memcg) |
2445 | return; | |
2446 | ||
309381fe | 2447 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 2448 | |
f3ccb2c4 VD |
2449 | page_counter_uncharge(&memcg->kmem, nr_pages); |
2450 | page_counter_uncharge(&memcg->memory, nr_pages); | |
2451 | if (do_swap_account) | |
2452 | page_counter_uncharge(&memcg->memsw, nr_pages); | |
60d3fd32 | 2453 | |
1306a85a | 2454 | page->mem_cgroup = NULL; |
f3ccb2c4 | 2455 | css_put_many(&memcg->css, nr_pages); |
60d3fd32 | 2456 | } |
7ae1e1d0 GC |
2457 | #endif /* CONFIG_MEMCG_KMEM */ |
2458 | ||
ca3e0214 KH |
2459 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2460 | ||
ca3e0214 KH |
2461 | /* |
2462 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
2463 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
2464 | * charge/uncharge will be never happen and move_account() is done under | |
2465 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 2466 | */ |
e94c8a9c | 2467 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2468 | { |
e94c8a9c | 2469 | int i; |
ca3e0214 | 2470 | |
3d37c4a9 KH |
2471 | if (mem_cgroup_disabled()) |
2472 | return; | |
b070e65c | 2473 | |
29833315 | 2474 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2475 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2476 | |
1306a85a | 2477 | __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
b070e65c | 2478 | HPAGE_PMD_NR); |
ca3e0214 | 2479 | } |
12d27107 | 2480 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2481 | |
c255a458 | 2482 | #ifdef CONFIG_MEMCG_SWAP |
0a31bc97 JW |
2483 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
2484 | bool charge) | |
d13d1443 | 2485 | { |
0a31bc97 JW |
2486 | int val = (charge) ? 1 : -1; |
2487 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); | |
d13d1443 | 2488 | } |
02491447 DN |
2489 | |
2490 | /** | |
2491 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2492 | * @entry: swap entry to be moved | |
2493 | * @from: mem_cgroup which the entry is moved from | |
2494 | * @to: mem_cgroup which the entry is moved to | |
2495 | * | |
2496 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2497 | * as the mem_cgroup's id of @from. | |
2498 | * | |
2499 | * Returns 0 on success, -EINVAL on failure. | |
2500 | * | |
3e32cb2e | 2501 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2502 | * both res and memsw, and called css_get(). |
2503 | */ | |
2504 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2505 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2506 | { |
2507 | unsigned short old_id, new_id; | |
2508 | ||
34c00c31 LZ |
2509 | old_id = mem_cgroup_id(from); |
2510 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2511 | |
2512 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 2513 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 2514 | mem_cgroup_swap_statistics(to, true); |
02491447 DN |
2515 | return 0; |
2516 | } | |
2517 | return -EINVAL; | |
2518 | } | |
2519 | #else | |
2520 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2521 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2522 | { |
2523 | return -EINVAL; | |
2524 | } | |
8c7c6e34 | 2525 | #endif |
d13d1443 | 2526 | |
3e32cb2e | 2527 | static DEFINE_MUTEX(memcg_limit_mutex); |
f212ad7c | 2528 | |
d38d2a75 | 2529 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2530 | unsigned long limit) |
628f4235 | 2531 | { |
3e32cb2e JW |
2532 | unsigned long curusage; |
2533 | unsigned long oldusage; | |
2534 | bool enlarge = false; | |
81d39c20 | 2535 | int retry_count; |
3e32cb2e | 2536 | int ret; |
81d39c20 KH |
2537 | |
2538 | /* | |
2539 | * For keeping hierarchical_reclaim simple, how long we should retry | |
2540 | * is depends on callers. We set our retry-count to be function | |
2541 | * of # of children which we should visit in this loop. | |
2542 | */ | |
3e32cb2e JW |
2543 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2544 | mem_cgroup_count_children(memcg); | |
81d39c20 | 2545 | |
3e32cb2e | 2546 | oldusage = page_counter_read(&memcg->memory); |
628f4235 | 2547 | |
3e32cb2e | 2548 | do { |
628f4235 KH |
2549 | if (signal_pending(current)) { |
2550 | ret = -EINTR; | |
2551 | break; | |
2552 | } | |
3e32cb2e JW |
2553 | |
2554 | mutex_lock(&memcg_limit_mutex); | |
2555 | if (limit > memcg->memsw.limit) { | |
2556 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2557 | ret = -EINVAL; |
628f4235 KH |
2558 | break; |
2559 | } | |
3e32cb2e JW |
2560 | if (limit > memcg->memory.limit) |
2561 | enlarge = true; | |
2562 | ret = page_counter_limit(&memcg->memory, limit); | |
2563 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2564 | |
2565 | if (!ret) | |
2566 | break; | |
2567 | ||
b70a2a21 JW |
2568 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
2569 | ||
3e32cb2e | 2570 | curusage = page_counter_read(&memcg->memory); |
81d39c20 | 2571 | /* Usage is reduced ? */ |
f894ffa8 | 2572 | if (curusage >= oldusage) |
81d39c20 KH |
2573 | retry_count--; |
2574 | else | |
2575 | oldusage = curusage; | |
3e32cb2e JW |
2576 | } while (retry_count); |
2577 | ||
3c11ecf4 KH |
2578 | if (!ret && enlarge) |
2579 | memcg_oom_recover(memcg); | |
14797e23 | 2580 | |
8c7c6e34 KH |
2581 | return ret; |
2582 | } | |
2583 | ||
338c8431 | 2584 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2585 | unsigned long limit) |
8c7c6e34 | 2586 | { |
3e32cb2e JW |
2587 | unsigned long curusage; |
2588 | unsigned long oldusage; | |
2589 | bool enlarge = false; | |
81d39c20 | 2590 | int retry_count; |
3e32cb2e | 2591 | int ret; |
8c7c6e34 | 2592 | |
81d39c20 | 2593 | /* see mem_cgroup_resize_res_limit */ |
3e32cb2e JW |
2594 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2595 | mem_cgroup_count_children(memcg); | |
2596 | ||
2597 | oldusage = page_counter_read(&memcg->memsw); | |
2598 | ||
2599 | do { | |
8c7c6e34 KH |
2600 | if (signal_pending(current)) { |
2601 | ret = -EINTR; | |
2602 | break; | |
2603 | } | |
3e32cb2e JW |
2604 | |
2605 | mutex_lock(&memcg_limit_mutex); | |
2606 | if (limit < memcg->memory.limit) { | |
2607 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2608 | ret = -EINVAL; |
8c7c6e34 KH |
2609 | break; |
2610 | } | |
3e32cb2e JW |
2611 | if (limit > memcg->memsw.limit) |
2612 | enlarge = true; | |
2613 | ret = page_counter_limit(&memcg->memsw, limit); | |
2614 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2615 | |
2616 | if (!ret) | |
2617 | break; | |
2618 | ||
b70a2a21 JW |
2619 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
2620 | ||
3e32cb2e | 2621 | curusage = page_counter_read(&memcg->memsw); |
81d39c20 | 2622 | /* Usage is reduced ? */ |
8c7c6e34 | 2623 | if (curusage >= oldusage) |
628f4235 | 2624 | retry_count--; |
81d39c20 KH |
2625 | else |
2626 | oldusage = curusage; | |
3e32cb2e JW |
2627 | } while (retry_count); |
2628 | ||
3c11ecf4 KH |
2629 | if (!ret && enlarge) |
2630 | memcg_oom_recover(memcg); | |
3e32cb2e | 2631 | |
628f4235 KH |
2632 | return ret; |
2633 | } | |
2634 | ||
0608f43d AM |
2635 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
2636 | gfp_t gfp_mask, | |
2637 | unsigned long *total_scanned) | |
2638 | { | |
2639 | unsigned long nr_reclaimed = 0; | |
2640 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
2641 | unsigned long reclaimed; | |
2642 | int loop = 0; | |
2643 | struct mem_cgroup_tree_per_zone *mctz; | |
3e32cb2e | 2644 | unsigned long excess; |
0608f43d AM |
2645 | unsigned long nr_scanned; |
2646 | ||
2647 | if (order > 0) | |
2648 | return 0; | |
2649 | ||
2650 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
2651 | /* | |
2652 | * This loop can run a while, specially if mem_cgroup's continuously | |
2653 | * keep exceeding their soft limit and putting the system under | |
2654 | * pressure | |
2655 | */ | |
2656 | do { | |
2657 | if (next_mz) | |
2658 | mz = next_mz; | |
2659 | else | |
2660 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2661 | if (!mz) | |
2662 | break; | |
2663 | ||
2664 | nr_scanned = 0; | |
2665 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
2666 | gfp_mask, &nr_scanned); | |
2667 | nr_reclaimed += reclaimed; | |
2668 | *total_scanned += nr_scanned; | |
0a31bc97 | 2669 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 2670 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
2671 | |
2672 | /* | |
2673 | * If we failed to reclaim anything from this memory cgroup | |
2674 | * it is time to move on to the next cgroup | |
2675 | */ | |
2676 | next_mz = NULL; | |
bc2f2e7f VD |
2677 | if (!reclaimed) |
2678 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
2679 | ||
3e32cb2e | 2680 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
2681 | /* |
2682 | * One school of thought says that we should not add | |
2683 | * back the node to the tree if reclaim returns 0. | |
2684 | * But our reclaim could return 0, simply because due | |
2685 | * to priority we are exposing a smaller subset of | |
2686 | * memory to reclaim from. Consider this as a longer | |
2687 | * term TODO. | |
2688 | */ | |
2689 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 2690 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 2691 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
2692 | css_put(&mz->memcg->css); |
2693 | loop++; | |
2694 | /* | |
2695 | * Could not reclaim anything and there are no more | |
2696 | * mem cgroups to try or we seem to be looping without | |
2697 | * reclaiming anything. | |
2698 | */ | |
2699 | if (!nr_reclaimed && | |
2700 | (next_mz == NULL || | |
2701 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2702 | break; | |
2703 | } while (!nr_reclaimed); | |
2704 | if (next_mz) | |
2705 | css_put(&next_mz->memcg->css); | |
2706 | return nr_reclaimed; | |
2707 | } | |
2708 | ||
ea280e7b TH |
2709 | /* |
2710 | * Test whether @memcg has children, dead or alive. Note that this | |
2711 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
2712 | * returns %true if there are child csses according to the cgroup | |
2713 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
2714 | */ | |
b5f99b53 GC |
2715 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
2716 | { | |
ea280e7b TH |
2717 | bool ret; |
2718 | ||
696ac172 | 2719 | /* |
ea280e7b TH |
2720 | * The lock does not prevent addition or deletion of children, but |
2721 | * it prevents a new child from being initialized based on this | |
2722 | * parent in css_online(), so it's enough to decide whether | |
2723 | * hierarchically inherited attributes can still be changed or not. | |
696ac172 | 2724 | */ |
ea280e7b TH |
2725 | lockdep_assert_held(&memcg_create_mutex); |
2726 | ||
2727 | rcu_read_lock(); | |
2728 | ret = css_next_child(NULL, &memcg->css); | |
2729 | rcu_read_unlock(); | |
2730 | return ret; | |
b5f99b53 GC |
2731 | } |
2732 | ||
c26251f9 MH |
2733 | /* |
2734 | * Reclaims as many pages from the given memcg as possible and moves | |
2735 | * the rest to the parent. | |
2736 | * | |
2737 | * Caller is responsible for holding css reference for memcg. | |
2738 | */ | |
2739 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
2740 | { | |
2741 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 2742 | |
c1e862c1 KH |
2743 | /* we call try-to-free pages for make this cgroup empty */ |
2744 | lru_add_drain_all(); | |
f817ed48 | 2745 | /* try to free all pages in this cgroup */ |
3e32cb2e | 2746 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 2747 | int progress; |
c1e862c1 | 2748 | |
c26251f9 MH |
2749 | if (signal_pending(current)) |
2750 | return -EINTR; | |
2751 | ||
b70a2a21 JW |
2752 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
2753 | GFP_KERNEL, true); | |
c1e862c1 | 2754 | if (!progress) { |
f817ed48 | 2755 | nr_retries--; |
c1e862c1 | 2756 | /* maybe some writeback is necessary */ |
8aa7e847 | 2757 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 2758 | } |
f817ed48 KH |
2759 | |
2760 | } | |
ab5196c2 MH |
2761 | |
2762 | return 0; | |
cc847582 KH |
2763 | } |
2764 | ||
6770c64e TH |
2765 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
2766 | char *buf, size_t nbytes, | |
2767 | loff_t off) | |
c1e862c1 | 2768 | { |
6770c64e | 2769 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 2770 | |
d8423011 MH |
2771 | if (mem_cgroup_is_root(memcg)) |
2772 | return -EINVAL; | |
6770c64e | 2773 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
2774 | } |
2775 | ||
182446d0 TH |
2776 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
2777 | struct cftype *cft) | |
18f59ea7 | 2778 | { |
182446d0 | 2779 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
2780 | } |
2781 | ||
182446d0 TH |
2782 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
2783 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
2784 | { |
2785 | int retval = 0; | |
182446d0 | 2786 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 2787 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 2788 | |
0999821b | 2789 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
2790 | |
2791 | if (memcg->use_hierarchy == val) | |
2792 | goto out; | |
2793 | ||
18f59ea7 | 2794 | /* |
af901ca1 | 2795 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
2796 | * in the child subtrees. If it is unset, then the change can |
2797 | * occur, provided the current cgroup has no children. | |
2798 | * | |
2799 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
2800 | * set if there are no children. | |
2801 | */ | |
c0ff4b85 | 2802 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 2803 | (val == 1 || val == 0)) { |
ea280e7b | 2804 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 2805 | memcg->use_hierarchy = val; |
18f59ea7 BS |
2806 | else |
2807 | retval = -EBUSY; | |
2808 | } else | |
2809 | retval = -EINVAL; | |
567fb435 GC |
2810 | |
2811 | out: | |
0999821b | 2812 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
2813 | |
2814 | return retval; | |
2815 | } | |
2816 | ||
3e32cb2e JW |
2817 | static unsigned long tree_stat(struct mem_cgroup *memcg, |
2818 | enum mem_cgroup_stat_index idx) | |
ce00a967 JW |
2819 | { |
2820 | struct mem_cgroup *iter; | |
484ebb3b | 2821 | unsigned long val = 0; |
ce00a967 | 2822 | |
ce00a967 JW |
2823 | for_each_mem_cgroup_tree(iter, memcg) |
2824 | val += mem_cgroup_read_stat(iter, idx); | |
2825 | ||
ce00a967 JW |
2826 | return val; |
2827 | } | |
2828 | ||
6f646156 | 2829 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 2830 | { |
c12176d3 | 2831 | unsigned long val; |
ce00a967 | 2832 | |
3e32cb2e JW |
2833 | if (mem_cgroup_is_root(memcg)) { |
2834 | val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE); | |
2835 | val += tree_stat(memcg, MEM_CGROUP_STAT_RSS); | |
2836 | if (swap) | |
2837 | val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP); | |
2838 | } else { | |
ce00a967 | 2839 | if (!swap) |
3e32cb2e | 2840 | val = page_counter_read(&memcg->memory); |
ce00a967 | 2841 | else |
3e32cb2e | 2842 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 2843 | } |
c12176d3 | 2844 | return val; |
ce00a967 JW |
2845 | } |
2846 | ||
3e32cb2e JW |
2847 | enum { |
2848 | RES_USAGE, | |
2849 | RES_LIMIT, | |
2850 | RES_MAX_USAGE, | |
2851 | RES_FAILCNT, | |
2852 | RES_SOFT_LIMIT, | |
2853 | }; | |
ce00a967 | 2854 | |
791badbd | 2855 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 2856 | struct cftype *cft) |
8cdea7c0 | 2857 | { |
182446d0 | 2858 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 2859 | struct page_counter *counter; |
af36f906 | 2860 | |
3e32cb2e | 2861 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 2862 | case _MEM: |
3e32cb2e JW |
2863 | counter = &memcg->memory; |
2864 | break; | |
8c7c6e34 | 2865 | case _MEMSWAP: |
3e32cb2e JW |
2866 | counter = &memcg->memsw; |
2867 | break; | |
510fc4e1 | 2868 | case _KMEM: |
3e32cb2e | 2869 | counter = &memcg->kmem; |
510fc4e1 | 2870 | break; |
8c7c6e34 KH |
2871 | default: |
2872 | BUG(); | |
8c7c6e34 | 2873 | } |
3e32cb2e JW |
2874 | |
2875 | switch (MEMFILE_ATTR(cft->private)) { | |
2876 | case RES_USAGE: | |
2877 | if (counter == &memcg->memory) | |
c12176d3 | 2878 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 2879 | if (counter == &memcg->memsw) |
c12176d3 | 2880 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
2881 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
2882 | case RES_LIMIT: | |
2883 | return (u64)counter->limit * PAGE_SIZE; | |
2884 | case RES_MAX_USAGE: | |
2885 | return (u64)counter->watermark * PAGE_SIZE; | |
2886 | case RES_FAILCNT: | |
2887 | return counter->failcnt; | |
2888 | case RES_SOFT_LIMIT: | |
2889 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
2890 | default: | |
2891 | BUG(); | |
2892 | } | |
8cdea7c0 | 2893 | } |
510fc4e1 | 2894 | |
510fc4e1 | 2895 | #ifdef CONFIG_MEMCG_KMEM |
8c0145b6 VD |
2896 | static int memcg_activate_kmem(struct mem_cgroup *memcg, |
2897 | unsigned long nr_pages) | |
d6441637 VD |
2898 | { |
2899 | int err = 0; | |
2900 | int memcg_id; | |
2901 | ||
2a4db7eb | 2902 | BUG_ON(memcg->kmemcg_id >= 0); |
2788cf0c | 2903 | BUG_ON(memcg->kmem_acct_activated); |
2a4db7eb | 2904 | BUG_ON(memcg->kmem_acct_active); |
d6441637 | 2905 | |
510fc4e1 GC |
2906 | /* |
2907 | * For simplicity, we won't allow this to be disabled. It also can't | |
2908 | * be changed if the cgroup has children already, or if tasks had | |
2909 | * already joined. | |
2910 | * | |
2911 | * If tasks join before we set the limit, a person looking at | |
2912 | * kmem.usage_in_bytes will have no way to determine when it took | |
2913 | * place, which makes the value quite meaningless. | |
2914 | * | |
2915 | * After it first became limited, changes in the value of the limit are | |
2916 | * of course permitted. | |
510fc4e1 | 2917 | */ |
0999821b | 2918 | mutex_lock(&memcg_create_mutex); |
27bd4dbb | 2919 | if (cgroup_is_populated(memcg->css.cgroup) || |
ea280e7b | 2920 | (memcg->use_hierarchy && memcg_has_children(memcg))) |
d6441637 VD |
2921 | err = -EBUSY; |
2922 | mutex_unlock(&memcg_create_mutex); | |
2923 | if (err) | |
2924 | goto out; | |
510fc4e1 | 2925 | |
f3bb3043 | 2926 | memcg_id = memcg_alloc_cache_id(); |
d6441637 VD |
2927 | if (memcg_id < 0) { |
2928 | err = memcg_id; | |
2929 | goto out; | |
2930 | } | |
2931 | ||
d6441637 | 2932 | /* |
900a38f0 VD |
2933 | * We couldn't have accounted to this cgroup, because it hasn't got |
2934 | * activated yet, so this should succeed. | |
d6441637 | 2935 | */ |
3e32cb2e | 2936 | err = page_counter_limit(&memcg->kmem, nr_pages); |
d6441637 VD |
2937 | VM_BUG_ON(err); |
2938 | ||
2939 | static_key_slow_inc(&memcg_kmem_enabled_key); | |
2940 | /* | |
900a38f0 VD |
2941 | * A memory cgroup is considered kmem-active as soon as it gets |
2942 | * kmemcg_id. Setting the id after enabling static branching will | |
d6441637 VD |
2943 | * guarantee no one starts accounting before all call sites are |
2944 | * patched. | |
2945 | */ | |
900a38f0 | 2946 | memcg->kmemcg_id = memcg_id; |
2788cf0c | 2947 | memcg->kmem_acct_activated = true; |
2a4db7eb | 2948 | memcg->kmem_acct_active = true; |
510fc4e1 | 2949 | out: |
d6441637 | 2950 | return err; |
d6441637 VD |
2951 | } |
2952 | ||
d6441637 | 2953 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2954 | unsigned long limit) |
d6441637 VD |
2955 | { |
2956 | int ret; | |
2957 | ||
3e32cb2e | 2958 | mutex_lock(&memcg_limit_mutex); |
d6441637 | 2959 | if (!memcg_kmem_is_active(memcg)) |
3e32cb2e | 2960 | ret = memcg_activate_kmem(memcg, limit); |
d6441637 | 2961 | else |
3e32cb2e JW |
2962 | ret = page_counter_limit(&memcg->kmem, limit); |
2963 | mutex_unlock(&memcg_limit_mutex); | |
510fc4e1 GC |
2964 | return ret; |
2965 | } | |
2966 | ||
55007d84 | 2967 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 2968 | { |
55007d84 | 2969 | int ret = 0; |
510fc4e1 | 2970 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
55007d84 | 2971 | |
d6441637 VD |
2972 | if (!parent) |
2973 | return 0; | |
55007d84 | 2974 | |
8c0145b6 | 2975 | mutex_lock(&memcg_limit_mutex); |
55007d84 | 2976 | /* |
d6441637 VD |
2977 | * If the parent cgroup is not kmem-active now, it cannot be activated |
2978 | * after this point, because it has at least one child already. | |
55007d84 | 2979 | */ |
d6441637 | 2980 | if (memcg_kmem_is_active(parent)) |
8c0145b6 VD |
2981 | ret = memcg_activate_kmem(memcg, PAGE_COUNTER_MAX); |
2982 | mutex_unlock(&memcg_limit_mutex); | |
55007d84 | 2983 | return ret; |
510fc4e1 | 2984 | } |
d6441637 VD |
2985 | #else |
2986 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
3e32cb2e | 2987 | unsigned long limit) |
d6441637 VD |
2988 | { |
2989 | return -EINVAL; | |
2990 | } | |
6d043990 | 2991 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 2992 | |
628f4235 KH |
2993 | /* |
2994 | * The user of this function is... | |
2995 | * RES_LIMIT. | |
2996 | */ | |
451af504 TH |
2997 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
2998 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 2999 | { |
451af504 | 3000 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3001 | unsigned long nr_pages; |
628f4235 KH |
3002 | int ret; |
3003 | ||
451af504 | 3004 | buf = strstrip(buf); |
650c5e56 | 3005 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3006 | if (ret) |
3007 | return ret; | |
af36f906 | 3008 | |
3e32cb2e | 3009 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3010 | case RES_LIMIT: |
4b3bde4c BS |
3011 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3012 | ret = -EINVAL; | |
3013 | break; | |
3014 | } | |
3e32cb2e JW |
3015 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3016 | case _MEM: | |
3017 | ret = mem_cgroup_resize_limit(memcg, nr_pages); | |
8c7c6e34 | 3018 | break; |
3e32cb2e JW |
3019 | case _MEMSWAP: |
3020 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); | |
296c81d8 | 3021 | break; |
3e32cb2e JW |
3022 | case _KMEM: |
3023 | ret = memcg_update_kmem_limit(memcg, nr_pages); | |
3024 | break; | |
3025 | } | |
296c81d8 | 3026 | break; |
3e32cb2e JW |
3027 | case RES_SOFT_LIMIT: |
3028 | memcg->soft_limit = nr_pages; | |
3029 | ret = 0; | |
628f4235 KH |
3030 | break; |
3031 | } | |
451af504 | 3032 | return ret ?: nbytes; |
8cdea7c0 BS |
3033 | } |
3034 | ||
6770c64e TH |
3035 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3036 | size_t nbytes, loff_t off) | |
c84872e1 | 3037 | { |
6770c64e | 3038 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3039 | struct page_counter *counter; |
c84872e1 | 3040 | |
3e32cb2e JW |
3041 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3042 | case _MEM: | |
3043 | counter = &memcg->memory; | |
3044 | break; | |
3045 | case _MEMSWAP: | |
3046 | counter = &memcg->memsw; | |
3047 | break; | |
3048 | case _KMEM: | |
3049 | counter = &memcg->kmem; | |
3050 | break; | |
3051 | default: | |
3052 | BUG(); | |
3053 | } | |
af36f906 | 3054 | |
3e32cb2e | 3055 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3056 | case RES_MAX_USAGE: |
3e32cb2e | 3057 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3058 | break; |
3059 | case RES_FAILCNT: | |
3e32cb2e | 3060 | counter->failcnt = 0; |
29f2a4da | 3061 | break; |
3e32cb2e JW |
3062 | default: |
3063 | BUG(); | |
29f2a4da | 3064 | } |
f64c3f54 | 3065 | |
6770c64e | 3066 | return nbytes; |
c84872e1 PE |
3067 | } |
3068 | ||
182446d0 | 3069 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3070 | struct cftype *cft) |
3071 | { | |
182446d0 | 3072 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3073 | } |
3074 | ||
02491447 | 3075 | #ifdef CONFIG_MMU |
182446d0 | 3076 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3077 | struct cftype *cft, u64 val) |
3078 | { | |
182446d0 | 3079 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3080 | |
1dfab5ab | 3081 | if (val & ~MOVE_MASK) |
7dc74be0 | 3082 | return -EINVAL; |
ee5e8472 | 3083 | |
7dc74be0 | 3084 | /* |
ee5e8472 GC |
3085 | * No kind of locking is needed in here, because ->can_attach() will |
3086 | * check this value once in the beginning of the process, and then carry | |
3087 | * on with stale data. This means that changes to this value will only | |
3088 | * affect task migrations starting after the change. | |
7dc74be0 | 3089 | */ |
c0ff4b85 | 3090 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3091 | return 0; |
3092 | } | |
02491447 | 3093 | #else |
182446d0 | 3094 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3095 | struct cftype *cft, u64 val) |
3096 | { | |
3097 | return -ENOSYS; | |
3098 | } | |
3099 | #endif | |
7dc74be0 | 3100 | |
406eb0c9 | 3101 | #ifdef CONFIG_NUMA |
2da8ca82 | 3102 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3103 | { |
25485de6 GT |
3104 | struct numa_stat { |
3105 | const char *name; | |
3106 | unsigned int lru_mask; | |
3107 | }; | |
3108 | ||
3109 | static const struct numa_stat stats[] = { | |
3110 | { "total", LRU_ALL }, | |
3111 | { "file", LRU_ALL_FILE }, | |
3112 | { "anon", LRU_ALL_ANON }, | |
3113 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3114 | }; | |
3115 | const struct numa_stat *stat; | |
406eb0c9 | 3116 | int nid; |
25485de6 | 3117 | unsigned long nr; |
2da8ca82 | 3118 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 3119 | |
25485de6 GT |
3120 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3121 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3122 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3123 | for_each_node_state(nid, N_MEMORY) { | |
3124 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3125 | stat->lru_mask); | |
3126 | seq_printf(m, " N%d=%lu", nid, nr); | |
3127 | } | |
3128 | seq_putc(m, '\n'); | |
406eb0c9 | 3129 | } |
406eb0c9 | 3130 | |
071aee13 YH |
3131 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3132 | struct mem_cgroup *iter; | |
3133 | ||
3134 | nr = 0; | |
3135 | for_each_mem_cgroup_tree(iter, memcg) | |
3136 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3137 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3138 | for_each_node_state(nid, N_MEMORY) { | |
3139 | nr = 0; | |
3140 | for_each_mem_cgroup_tree(iter, memcg) | |
3141 | nr += mem_cgroup_node_nr_lru_pages( | |
3142 | iter, nid, stat->lru_mask); | |
3143 | seq_printf(m, " N%d=%lu", nid, nr); | |
3144 | } | |
3145 | seq_putc(m, '\n'); | |
406eb0c9 | 3146 | } |
406eb0c9 | 3147 | |
406eb0c9 YH |
3148 | return 0; |
3149 | } | |
3150 | #endif /* CONFIG_NUMA */ | |
3151 | ||
2da8ca82 | 3152 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3153 | { |
2da8ca82 | 3154 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 3155 | unsigned long memory, memsw; |
af7c4b0e JW |
3156 | struct mem_cgroup *mi; |
3157 | unsigned int i; | |
406eb0c9 | 3158 | |
0ca44b14 GT |
3159 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) != |
3160 | MEM_CGROUP_STAT_NSTATS); | |
3161 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) != | |
3162 | MEM_CGROUP_EVENTS_NSTATS); | |
70bc068c RS |
3163 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3164 | ||
af7c4b0e | 3165 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 3166 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 3167 | continue; |
484ebb3b | 3168 | seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i], |
af7c4b0e | 3169 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); |
1dd3a273 | 3170 | } |
7b854121 | 3171 | |
af7c4b0e JW |
3172 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
3173 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
3174 | mem_cgroup_read_events(memcg, i)); | |
3175 | ||
3176 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3177 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
3178 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
3179 | ||
14067bb3 | 3180 | /* Hierarchical information */ |
3e32cb2e JW |
3181 | memory = memsw = PAGE_COUNTER_MAX; |
3182 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
3183 | memory = min(memory, mi->memory.limit); | |
3184 | memsw = min(memsw, mi->memsw.limit); | |
fee7b548 | 3185 | } |
3e32cb2e JW |
3186 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3187 | (u64)memory * PAGE_SIZE); | |
3188 | if (do_swap_account) | |
3189 | seq_printf(m, "hierarchical_memsw_limit %llu\n", | |
3190 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3191 | |
af7c4b0e | 3192 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
484ebb3b | 3193 | unsigned long long val = 0; |
af7c4b0e | 3194 | |
bff6bb83 | 3195 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 3196 | continue; |
af7c4b0e JW |
3197 | for_each_mem_cgroup_tree(mi, memcg) |
3198 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
484ebb3b | 3199 | seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val); |
af7c4b0e JW |
3200 | } |
3201 | ||
3202 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
3203 | unsigned long long val = 0; | |
3204 | ||
3205 | for_each_mem_cgroup_tree(mi, memcg) | |
3206 | val += mem_cgroup_read_events(mi, i); | |
3207 | seq_printf(m, "total_%s %llu\n", | |
3208 | mem_cgroup_events_names[i], val); | |
3209 | } | |
3210 | ||
3211 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
3212 | unsigned long long val = 0; | |
3213 | ||
3214 | for_each_mem_cgroup_tree(mi, memcg) | |
3215 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
3216 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 3217 | } |
14067bb3 | 3218 | |
7f016ee8 | 3219 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
3220 | { |
3221 | int nid, zid; | |
3222 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 3223 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3224 | unsigned long recent_rotated[2] = {0, 0}; |
3225 | unsigned long recent_scanned[2] = {0, 0}; | |
3226 | ||
3227 | for_each_online_node(nid) | |
3228 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
e231875b | 3229 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
89abfab1 | 3230 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 3231 | |
89abfab1 HD |
3232 | recent_rotated[0] += rstat->recent_rotated[0]; |
3233 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3234 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3235 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 3236 | } |
78ccf5b5 JW |
3237 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3238 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3239 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3240 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3241 | } |
3242 | #endif | |
3243 | ||
d2ceb9b7 KH |
3244 | return 0; |
3245 | } | |
3246 | ||
182446d0 TH |
3247 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3248 | struct cftype *cft) | |
a7885eb8 | 3249 | { |
182446d0 | 3250 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3251 | |
1f4c025b | 3252 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3253 | } |
3254 | ||
182446d0 TH |
3255 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3256 | struct cftype *cft, u64 val) | |
a7885eb8 | 3257 | { |
182446d0 | 3258 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3259 | |
3dae7fec | 3260 | if (val > 100) |
a7885eb8 KM |
3261 | return -EINVAL; |
3262 | ||
14208b0e | 3263 | if (css->parent) |
3dae7fec JW |
3264 | memcg->swappiness = val; |
3265 | else | |
3266 | vm_swappiness = val; | |
068b38c1 | 3267 | |
a7885eb8 KM |
3268 | return 0; |
3269 | } | |
3270 | ||
2e72b634 KS |
3271 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3272 | { | |
3273 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3274 | unsigned long usage; |
2e72b634 KS |
3275 | int i; |
3276 | ||
3277 | rcu_read_lock(); | |
3278 | if (!swap) | |
2c488db2 | 3279 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3280 | else |
2c488db2 | 3281 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3282 | |
3283 | if (!t) | |
3284 | goto unlock; | |
3285 | ||
ce00a967 | 3286 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3287 | |
3288 | /* | |
748dad36 | 3289 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3290 | * If it's not true, a threshold was crossed after last |
3291 | * call of __mem_cgroup_threshold(). | |
3292 | */ | |
5407a562 | 3293 | i = t->current_threshold; |
2e72b634 KS |
3294 | |
3295 | /* | |
3296 | * Iterate backward over array of thresholds starting from | |
3297 | * current_threshold and check if a threshold is crossed. | |
3298 | * If none of thresholds below usage is crossed, we read | |
3299 | * only one element of the array here. | |
3300 | */ | |
3301 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3302 | eventfd_signal(t->entries[i].eventfd, 1); | |
3303 | ||
3304 | /* i = current_threshold + 1 */ | |
3305 | i++; | |
3306 | ||
3307 | /* | |
3308 | * Iterate forward over array of thresholds starting from | |
3309 | * current_threshold+1 and check if a threshold is crossed. | |
3310 | * If none of thresholds above usage is crossed, we read | |
3311 | * only one element of the array here. | |
3312 | */ | |
3313 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3314 | eventfd_signal(t->entries[i].eventfd, 1); | |
3315 | ||
3316 | /* Update current_threshold */ | |
5407a562 | 3317 | t->current_threshold = i - 1; |
2e72b634 KS |
3318 | unlock: |
3319 | rcu_read_unlock(); | |
3320 | } | |
3321 | ||
3322 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3323 | { | |
ad4ca5f4 KS |
3324 | while (memcg) { |
3325 | __mem_cgroup_threshold(memcg, false); | |
3326 | if (do_swap_account) | |
3327 | __mem_cgroup_threshold(memcg, true); | |
3328 | ||
3329 | memcg = parent_mem_cgroup(memcg); | |
3330 | } | |
2e72b634 KS |
3331 | } |
3332 | ||
3333 | static int compare_thresholds(const void *a, const void *b) | |
3334 | { | |
3335 | const struct mem_cgroup_threshold *_a = a; | |
3336 | const struct mem_cgroup_threshold *_b = b; | |
3337 | ||
2bff24a3 GT |
3338 | if (_a->threshold > _b->threshold) |
3339 | return 1; | |
3340 | ||
3341 | if (_a->threshold < _b->threshold) | |
3342 | return -1; | |
3343 | ||
3344 | return 0; | |
2e72b634 KS |
3345 | } |
3346 | ||
c0ff4b85 | 3347 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3348 | { |
3349 | struct mem_cgroup_eventfd_list *ev; | |
3350 | ||
2bcf2e92 MH |
3351 | spin_lock(&memcg_oom_lock); |
3352 | ||
c0ff4b85 | 3353 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3354 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3355 | |
3356 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3357 | return 0; |
3358 | } | |
3359 | ||
c0ff4b85 | 3360 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3361 | { |
7d74b06f KH |
3362 | struct mem_cgroup *iter; |
3363 | ||
c0ff4b85 | 3364 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3365 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3366 | } |
3367 | ||
59b6f873 | 3368 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3369 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3370 | { |
2c488db2 KS |
3371 | struct mem_cgroup_thresholds *thresholds; |
3372 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3373 | unsigned long threshold; |
3374 | unsigned long usage; | |
2c488db2 | 3375 | int i, size, ret; |
2e72b634 | 3376 | |
650c5e56 | 3377 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
3378 | if (ret) |
3379 | return ret; | |
3380 | ||
3381 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3382 | |
05b84301 | 3383 | if (type == _MEM) { |
2c488db2 | 3384 | thresholds = &memcg->thresholds; |
ce00a967 | 3385 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3386 | } else if (type == _MEMSWAP) { |
2c488db2 | 3387 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3388 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3389 | } else |
2e72b634 KS |
3390 | BUG(); |
3391 | ||
2e72b634 | 3392 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3393 | if (thresholds->primary) |
2e72b634 KS |
3394 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3395 | ||
2c488db2 | 3396 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3397 | |
3398 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 3399 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 3400 | GFP_KERNEL); |
2c488db2 | 3401 | if (!new) { |
2e72b634 KS |
3402 | ret = -ENOMEM; |
3403 | goto unlock; | |
3404 | } | |
2c488db2 | 3405 | new->size = size; |
2e72b634 KS |
3406 | |
3407 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3408 | if (thresholds->primary) { |
3409 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3410 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3411 | } |
3412 | ||
2e72b634 | 3413 | /* Add new threshold */ |
2c488db2 KS |
3414 | new->entries[size - 1].eventfd = eventfd; |
3415 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3416 | |
3417 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3418 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3419 | compare_thresholds, NULL); |
3420 | ||
3421 | /* Find current threshold */ | |
2c488db2 | 3422 | new->current_threshold = -1; |
2e72b634 | 3423 | for (i = 0; i < size; i++) { |
748dad36 | 3424 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3425 | /* |
2c488db2 KS |
3426 | * new->current_threshold will not be used until |
3427 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3428 | * it here. |
3429 | */ | |
2c488db2 | 3430 | ++new->current_threshold; |
748dad36 SZ |
3431 | } else |
3432 | break; | |
2e72b634 KS |
3433 | } |
3434 | ||
2c488db2 KS |
3435 | /* Free old spare buffer and save old primary buffer as spare */ |
3436 | kfree(thresholds->spare); | |
3437 | thresholds->spare = thresholds->primary; | |
3438 | ||
3439 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3440 | |
907860ed | 3441 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3442 | synchronize_rcu(); |
3443 | ||
2e72b634 KS |
3444 | unlock: |
3445 | mutex_unlock(&memcg->thresholds_lock); | |
3446 | ||
3447 | return ret; | |
3448 | } | |
3449 | ||
59b6f873 | 3450 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3451 | struct eventfd_ctx *eventfd, const char *args) |
3452 | { | |
59b6f873 | 3453 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3454 | } |
3455 | ||
59b6f873 | 3456 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3457 | struct eventfd_ctx *eventfd, const char *args) |
3458 | { | |
59b6f873 | 3459 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3460 | } |
3461 | ||
59b6f873 | 3462 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3463 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3464 | { |
2c488db2 KS |
3465 | struct mem_cgroup_thresholds *thresholds; |
3466 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3467 | unsigned long usage; |
2c488db2 | 3468 | int i, j, size; |
2e72b634 KS |
3469 | |
3470 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3471 | |
3472 | if (type == _MEM) { | |
2c488db2 | 3473 | thresholds = &memcg->thresholds; |
ce00a967 | 3474 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3475 | } else if (type == _MEMSWAP) { |
2c488db2 | 3476 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3477 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3478 | } else |
2e72b634 KS |
3479 | BUG(); |
3480 | ||
371528ca AV |
3481 | if (!thresholds->primary) |
3482 | goto unlock; | |
3483 | ||
2e72b634 KS |
3484 | /* Check if a threshold crossed before removing */ |
3485 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3486 | ||
3487 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3488 | size = 0; |
3489 | for (i = 0; i < thresholds->primary->size; i++) { | |
3490 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3491 | size++; |
3492 | } | |
3493 | ||
2c488db2 | 3494 | new = thresholds->spare; |
907860ed | 3495 | |
2e72b634 KS |
3496 | /* Set thresholds array to NULL if we don't have thresholds */ |
3497 | if (!size) { | |
2c488db2 KS |
3498 | kfree(new); |
3499 | new = NULL; | |
907860ed | 3500 | goto swap_buffers; |
2e72b634 KS |
3501 | } |
3502 | ||
2c488db2 | 3503 | new->size = size; |
2e72b634 KS |
3504 | |
3505 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3506 | new->current_threshold = -1; |
3507 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3508 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3509 | continue; |
3510 | ||
2c488db2 | 3511 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3512 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3513 | /* |
2c488db2 | 3514 | * new->current_threshold will not be used |
2e72b634 KS |
3515 | * until rcu_assign_pointer(), so it's safe to increment |
3516 | * it here. | |
3517 | */ | |
2c488db2 | 3518 | ++new->current_threshold; |
2e72b634 KS |
3519 | } |
3520 | j++; | |
3521 | } | |
3522 | ||
907860ed | 3523 | swap_buffers: |
2c488db2 KS |
3524 | /* Swap primary and spare array */ |
3525 | thresholds->spare = thresholds->primary; | |
8c757763 | 3526 | |
2c488db2 | 3527 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3528 | |
907860ed | 3529 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3530 | synchronize_rcu(); |
ececa3eb MC |
3531 | |
3532 | /* If all events are unregistered, free the spare array */ | |
3533 | if (!new) { | |
3534 | kfree(thresholds->spare); | |
3535 | thresholds->spare = NULL; | |
3536 | } | |
371528ca | 3537 | unlock: |
2e72b634 | 3538 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3539 | } |
c1e862c1 | 3540 | |
59b6f873 | 3541 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3542 | struct eventfd_ctx *eventfd) |
3543 | { | |
59b6f873 | 3544 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
3545 | } |
3546 | ||
59b6f873 | 3547 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3548 | struct eventfd_ctx *eventfd) |
3549 | { | |
59b6f873 | 3550 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
3551 | } |
3552 | ||
59b6f873 | 3553 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3554 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 3555 | { |
9490ff27 | 3556 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 3557 | |
9490ff27 KH |
3558 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
3559 | if (!event) | |
3560 | return -ENOMEM; | |
3561 | ||
1af8efe9 | 3562 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
3563 | |
3564 | event->eventfd = eventfd; | |
3565 | list_add(&event->list, &memcg->oom_notify); | |
3566 | ||
3567 | /* already in OOM ? */ | |
c2b42d3c | 3568 | if (memcg->under_oom) |
9490ff27 | 3569 | eventfd_signal(eventfd, 1); |
1af8efe9 | 3570 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3571 | |
3572 | return 0; | |
3573 | } | |
3574 | ||
59b6f873 | 3575 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3576 | struct eventfd_ctx *eventfd) |
9490ff27 | 3577 | { |
9490ff27 | 3578 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 3579 | |
1af8efe9 | 3580 | spin_lock(&memcg_oom_lock); |
9490ff27 | 3581 | |
c0ff4b85 | 3582 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
3583 | if (ev->eventfd == eventfd) { |
3584 | list_del(&ev->list); | |
3585 | kfree(ev); | |
3586 | } | |
3587 | } | |
3588 | ||
1af8efe9 | 3589 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3590 | } |
3591 | ||
2da8ca82 | 3592 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 3593 | { |
2da8ca82 | 3594 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 3595 | |
791badbd | 3596 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 3597 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
3c11ecf4 KH |
3598 | return 0; |
3599 | } | |
3600 | ||
182446d0 | 3601 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
3602 | struct cftype *cft, u64 val) |
3603 | { | |
182446d0 | 3604 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
3605 | |
3606 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 3607 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
3608 | return -EINVAL; |
3609 | ||
c0ff4b85 | 3610 | memcg->oom_kill_disable = val; |
4d845ebf | 3611 | if (!val) |
c0ff4b85 | 3612 | memcg_oom_recover(memcg); |
3dae7fec | 3613 | |
3c11ecf4 KH |
3614 | return 0; |
3615 | } | |
3616 | ||
c255a458 | 3617 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 3618 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 3619 | { |
55007d84 GC |
3620 | int ret; |
3621 | ||
55007d84 GC |
3622 | ret = memcg_propagate_kmem(memcg); |
3623 | if (ret) | |
3624 | return ret; | |
2633d7a0 | 3625 | |
1d62e436 | 3626 | return mem_cgroup_sockets_init(memcg, ss); |
573b400d | 3627 | } |
e5671dfa | 3628 | |
2a4db7eb VD |
3629 | static void memcg_deactivate_kmem(struct mem_cgroup *memcg) |
3630 | { | |
2788cf0c VD |
3631 | struct cgroup_subsys_state *css; |
3632 | struct mem_cgroup *parent, *child; | |
3633 | int kmemcg_id; | |
3634 | ||
2a4db7eb VD |
3635 | if (!memcg->kmem_acct_active) |
3636 | return; | |
3637 | ||
3638 | /* | |
3639 | * Clear the 'active' flag before clearing memcg_caches arrays entries. | |
3640 | * Since we take the slab_mutex in memcg_deactivate_kmem_caches(), it | |
3641 | * guarantees no cache will be created for this cgroup after we are | |
3642 | * done (see memcg_create_kmem_cache()). | |
3643 | */ | |
3644 | memcg->kmem_acct_active = false; | |
3645 | ||
3646 | memcg_deactivate_kmem_caches(memcg); | |
2788cf0c VD |
3647 | |
3648 | kmemcg_id = memcg->kmemcg_id; | |
3649 | BUG_ON(kmemcg_id < 0); | |
3650 | ||
3651 | parent = parent_mem_cgroup(memcg); | |
3652 | if (!parent) | |
3653 | parent = root_mem_cgroup; | |
3654 | ||
3655 | /* | |
3656 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3657 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3658 | * to ones of the parent. After we have finished, all list_lrus | |
3659 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3660 | * ordering is imposed by list_lru_node->lock taken by | |
3661 | * memcg_drain_all_list_lrus(). | |
3662 | */ | |
3663 | css_for_each_descendant_pre(css, &memcg->css) { | |
3664 | child = mem_cgroup_from_css(css); | |
3665 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3666 | child->kmemcg_id = parent->kmemcg_id; | |
3667 | if (!memcg->use_hierarchy) | |
3668 | break; | |
3669 | } | |
3670 | memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id); | |
3671 | ||
3672 | memcg_free_cache_id(kmemcg_id); | |
2a4db7eb VD |
3673 | } |
3674 | ||
10d5ebf4 | 3675 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 3676 | { |
f48b80a5 VD |
3677 | if (memcg->kmem_acct_activated) { |
3678 | memcg_destroy_kmem_caches(memcg); | |
3679 | static_key_slow_dec(&memcg_kmem_enabled_key); | |
3680 | WARN_ON(page_counter_read(&memcg->kmem)); | |
3681 | } | |
1d62e436 | 3682 | mem_cgroup_sockets_destroy(memcg); |
10d5ebf4 | 3683 | } |
e5671dfa | 3684 | #else |
cbe128e3 | 3685 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
3686 | { |
3687 | return 0; | |
3688 | } | |
d1a4c0b3 | 3689 | |
2a4db7eb VD |
3690 | static void memcg_deactivate_kmem(struct mem_cgroup *memcg) |
3691 | { | |
3692 | } | |
3693 | ||
10d5ebf4 LZ |
3694 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
3695 | { | |
3696 | } | |
e5671dfa GC |
3697 | #endif |
3698 | ||
52ebea74 TH |
3699 | #ifdef CONFIG_CGROUP_WRITEBACK |
3700 | ||
3701 | struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg) | |
3702 | { | |
3703 | return &memcg->cgwb_list; | |
3704 | } | |
3705 | ||
841710aa TH |
3706 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
3707 | { | |
3708 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
3709 | } | |
3710 | ||
3711 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3712 | { | |
3713 | wb_domain_exit(&memcg->cgwb_domain); | |
3714 | } | |
3715 | ||
2529bb3a TH |
3716 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3717 | { | |
3718 | wb_domain_size_changed(&memcg->cgwb_domain); | |
3719 | } | |
3720 | ||
841710aa TH |
3721 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
3722 | { | |
3723 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3724 | ||
3725 | if (!memcg->css.parent) | |
3726 | return NULL; | |
3727 | ||
3728 | return &memcg->cgwb_domain; | |
3729 | } | |
3730 | ||
c2aa723a TH |
3731 | /** |
3732 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
3733 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
3734 | * @pfilepages: out parameter for number of file pages |
3735 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
3736 | * @pdirty: out parameter for number of dirty pages |
3737 | * @pwriteback: out parameter for number of pages under writeback | |
3738 | * | |
c5edf9cd TH |
3739 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
3740 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
3741 | * is a bit more involved. | |
c2aa723a | 3742 | * |
c5edf9cd TH |
3743 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
3744 | * headroom is calculated as the lowest headroom of itself and the | |
3745 | * ancestors. Note that this doesn't consider the actual amount of | |
3746 | * available memory in the system. The caller should further cap | |
3747 | * *@pheadroom accordingly. | |
c2aa723a | 3748 | */ |
c5edf9cd TH |
3749 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
3750 | unsigned long *pheadroom, unsigned long *pdirty, | |
3751 | unsigned long *pwriteback) | |
c2aa723a TH |
3752 | { |
3753 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3754 | struct mem_cgroup *parent; | |
c2aa723a TH |
3755 | |
3756 | *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY); | |
3757 | ||
3758 | /* this should eventually include NR_UNSTABLE_NFS */ | |
3759 | *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK); | |
c5edf9cd TH |
3760 | *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) | |
3761 | (1 << LRU_ACTIVE_FILE)); | |
3762 | *pheadroom = PAGE_COUNTER_MAX; | |
c2aa723a | 3763 | |
c2aa723a TH |
3764 | while ((parent = parent_mem_cgroup(memcg))) { |
3765 | unsigned long ceiling = min(memcg->memory.limit, memcg->high); | |
3766 | unsigned long used = page_counter_read(&memcg->memory); | |
3767 | ||
c5edf9cd | 3768 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
3769 | memcg = parent; |
3770 | } | |
c2aa723a TH |
3771 | } |
3772 | ||
841710aa TH |
3773 | #else /* CONFIG_CGROUP_WRITEBACK */ |
3774 | ||
3775 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
3776 | { | |
3777 | return 0; | |
3778 | } | |
3779 | ||
3780 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3781 | { | |
3782 | } | |
3783 | ||
2529bb3a TH |
3784 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3785 | { | |
3786 | } | |
3787 | ||
52ebea74 TH |
3788 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
3789 | ||
3bc942f3 TH |
3790 | /* |
3791 | * DO NOT USE IN NEW FILES. | |
3792 | * | |
3793 | * "cgroup.event_control" implementation. | |
3794 | * | |
3795 | * This is way over-engineered. It tries to support fully configurable | |
3796 | * events for each user. Such level of flexibility is completely | |
3797 | * unnecessary especially in the light of the planned unified hierarchy. | |
3798 | * | |
3799 | * Please deprecate this and replace with something simpler if at all | |
3800 | * possible. | |
3801 | */ | |
3802 | ||
79bd9814 TH |
3803 | /* |
3804 | * Unregister event and free resources. | |
3805 | * | |
3806 | * Gets called from workqueue. | |
3807 | */ | |
3bc942f3 | 3808 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 3809 | { |
3bc942f3 TH |
3810 | struct mem_cgroup_event *event = |
3811 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 3812 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3813 | |
3814 | remove_wait_queue(event->wqh, &event->wait); | |
3815 | ||
59b6f873 | 3816 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
3817 | |
3818 | /* Notify userspace the event is going away. */ | |
3819 | eventfd_signal(event->eventfd, 1); | |
3820 | ||
3821 | eventfd_ctx_put(event->eventfd); | |
3822 | kfree(event); | |
59b6f873 | 3823 | css_put(&memcg->css); |
79bd9814 TH |
3824 | } |
3825 | ||
3826 | /* | |
3827 | * Gets called on POLLHUP on eventfd when user closes it. | |
3828 | * | |
3829 | * Called with wqh->lock held and interrupts disabled. | |
3830 | */ | |
3bc942f3 TH |
3831 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
3832 | int sync, void *key) | |
79bd9814 | 3833 | { |
3bc942f3 TH |
3834 | struct mem_cgroup_event *event = |
3835 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 3836 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3837 | unsigned long flags = (unsigned long)key; |
3838 | ||
3839 | if (flags & POLLHUP) { | |
3840 | /* | |
3841 | * If the event has been detached at cgroup removal, we | |
3842 | * can simply return knowing the other side will cleanup | |
3843 | * for us. | |
3844 | * | |
3845 | * We can't race against event freeing since the other | |
3846 | * side will require wqh->lock via remove_wait_queue(), | |
3847 | * which we hold. | |
3848 | */ | |
fba94807 | 3849 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
3850 | if (!list_empty(&event->list)) { |
3851 | list_del_init(&event->list); | |
3852 | /* | |
3853 | * We are in atomic context, but cgroup_event_remove() | |
3854 | * may sleep, so we have to call it in workqueue. | |
3855 | */ | |
3856 | schedule_work(&event->remove); | |
3857 | } | |
fba94807 | 3858 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
3859 | } |
3860 | ||
3861 | return 0; | |
3862 | } | |
3863 | ||
3bc942f3 | 3864 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
3865 | wait_queue_head_t *wqh, poll_table *pt) |
3866 | { | |
3bc942f3 TH |
3867 | struct mem_cgroup_event *event = |
3868 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
3869 | |
3870 | event->wqh = wqh; | |
3871 | add_wait_queue(wqh, &event->wait); | |
3872 | } | |
3873 | ||
3874 | /* | |
3bc942f3 TH |
3875 | * DO NOT USE IN NEW FILES. |
3876 | * | |
79bd9814 TH |
3877 | * Parse input and register new cgroup event handler. |
3878 | * | |
3879 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
3880 | * Interpretation of args is defined by control file implementation. | |
3881 | */ | |
451af504 TH |
3882 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
3883 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 3884 | { |
451af504 | 3885 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 3886 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 3887 | struct mem_cgroup_event *event; |
79bd9814 TH |
3888 | struct cgroup_subsys_state *cfile_css; |
3889 | unsigned int efd, cfd; | |
3890 | struct fd efile; | |
3891 | struct fd cfile; | |
fba94807 | 3892 | const char *name; |
79bd9814 TH |
3893 | char *endp; |
3894 | int ret; | |
3895 | ||
451af504 TH |
3896 | buf = strstrip(buf); |
3897 | ||
3898 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
3899 | if (*endp != ' ') |
3900 | return -EINVAL; | |
451af504 | 3901 | buf = endp + 1; |
79bd9814 | 3902 | |
451af504 | 3903 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
3904 | if ((*endp != ' ') && (*endp != '\0')) |
3905 | return -EINVAL; | |
451af504 | 3906 | buf = endp + 1; |
79bd9814 TH |
3907 | |
3908 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
3909 | if (!event) | |
3910 | return -ENOMEM; | |
3911 | ||
59b6f873 | 3912 | event->memcg = memcg; |
79bd9814 | 3913 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
3914 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
3915 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
3916 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
3917 | |
3918 | efile = fdget(efd); | |
3919 | if (!efile.file) { | |
3920 | ret = -EBADF; | |
3921 | goto out_kfree; | |
3922 | } | |
3923 | ||
3924 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
3925 | if (IS_ERR(event->eventfd)) { | |
3926 | ret = PTR_ERR(event->eventfd); | |
3927 | goto out_put_efile; | |
3928 | } | |
3929 | ||
3930 | cfile = fdget(cfd); | |
3931 | if (!cfile.file) { | |
3932 | ret = -EBADF; | |
3933 | goto out_put_eventfd; | |
3934 | } | |
3935 | ||
3936 | /* the process need read permission on control file */ | |
3937 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
3938 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
3939 | if (ret < 0) | |
3940 | goto out_put_cfile; | |
3941 | ||
fba94807 TH |
3942 | /* |
3943 | * Determine the event callbacks and set them in @event. This used | |
3944 | * to be done via struct cftype but cgroup core no longer knows | |
3945 | * about these events. The following is crude but the whole thing | |
3946 | * is for compatibility anyway. | |
3bc942f3 TH |
3947 | * |
3948 | * DO NOT ADD NEW FILES. | |
fba94807 | 3949 | */ |
b583043e | 3950 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
3951 | |
3952 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
3953 | event->register_event = mem_cgroup_usage_register_event; | |
3954 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
3955 | } else if (!strcmp(name, "memory.oom_control")) { | |
3956 | event->register_event = mem_cgroup_oom_register_event; | |
3957 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
3958 | } else if (!strcmp(name, "memory.pressure_level")) { | |
3959 | event->register_event = vmpressure_register_event; | |
3960 | event->unregister_event = vmpressure_unregister_event; | |
3961 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
3962 | event->register_event = memsw_cgroup_usage_register_event; |
3963 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
3964 | } else { |
3965 | ret = -EINVAL; | |
3966 | goto out_put_cfile; | |
3967 | } | |
3968 | ||
79bd9814 | 3969 | /* |
b5557c4c TH |
3970 | * Verify @cfile should belong to @css. Also, remaining events are |
3971 | * automatically removed on cgroup destruction but the removal is | |
3972 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 3973 | */ |
b583043e | 3974 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 3975 | &memory_cgrp_subsys); |
79bd9814 | 3976 | ret = -EINVAL; |
5a17f543 | 3977 | if (IS_ERR(cfile_css)) |
79bd9814 | 3978 | goto out_put_cfile; |
5a17f543 TH |
3979 | if (cfile_css != css) { |
3980 | css_put(cfile_css); | |
79bd9814 | 3981 | goto out_put_cfile; |
5a17f543 | 3982 | } |
79bd9814 | 3983 | |
451af504 | 3984 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
3985 | if (ret) |
3986 | goto out_put_css; | |
3987 | ||
3988 | efile.file->f_op->poll(efile.file, &event->pt); | |
3989 | ||
fba94807 TH |
3990 | spin_lock(&memcg->event_list_lock); |
3991 | list_add(&event->list, &memcg->event_list); | |
3992 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
3993 | |
3994 | fdput(cfile); | |
3995 | fdput(efile); | |
3996 | ||
451af504 | 3997 | return nbytes; |
79bd9814 TH |
3998 | |
3999 | out_put_css: | |
b5557c4c | 4000 | css_put(css); |
79bd9814 TH |
4001 | out_put_cfile: |
4002 | fdput(cfile); | |
4003 | out_put_eventfd: | |
4004 | eventfd_ctx_put(event->eventfd); | |
4005 | out_put_efile: | |
4006 | fdput(efile); | |
4007 | out_kfree: | |
4008 | kfree(event); | |
4009 | ||
4010 | return ret; | |
4011 | } | |
4012 | ||
241994ed | 4013 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4014 | { |
0eea1030 | 4015 | .name = "usage_in_bytes", |
8c7c6e34 | 4016 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4017 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4018 | }, |
c84872e1 PE |
4019 | { |
4020 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4021 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4022 | .write = mem_cgroup_reset, |
791badbd | 4023 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4024 | }, |
8cdea7c0 | 4025 | { |
0eea1030 | 4026 | .name = "limit_in_bytes", |
8c7c6e34 | 4027 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4028 | .write = mem_cgroup_write, |
791badbd | 4029 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4030 | }, |
296c81d8 BS |
4031 | { |
4032 | .name = "soft_limit_in_bytes", | |
4033 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4034 | .write = mem_cgroup_write, |
791badbd | 4035 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4036 | }, |
8cdea7c0 BS |
4037 | { |
4038 | .name = "failcnt", | |
8c7c6e34 | 4039 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4040 | .write = mem_cgroup_reset, |
791badbd | 4041 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4042 | }, |
d2ceb9b7 KH |
4043 | { |
4044 | .name = "stat", | |
2da8ca82 | 4045 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4046 | }, |
c1e862c1 KH |
4047 | { |
4048 | .name = "force_empty", | |
6770c64e | 4049 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4050 | }, |
18f59ea7 BS |
4051 | { |
4052 | .name = "use_hierarchy", | |
4053 | .write_u64 = mem_cgroup_hierarchy_write, | |
4054 | .read_u64 = mem_cgroup_hierarchy_read, | |
4055 | }, | |
79bd9814 | 4056 | { |
3bc942f3 | 4057 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4058 | .write = memcg_write_event_control, |
7dbdb199 | 4059 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 4060 | }, |
a7885eb8 KM |
4061 | { |
4062 | .name = "swappiness", | |
4063 | .read_u64 = mem_cgroup_swappiness_read, | |
4064 | .write_u64 = mem_cgroup_swappiness_write, | |
4065 | }, | |
7dc74be0 DN |
4066 | { |
4067 | .name = "move_charge_at_immigrate", | |
4068 | .read_u64 = mem_cgroup_move_charge_read, | |
4069 | .write_u64 = mem_cgroup_move_charge_write, | |
4070 | }, | |
9490ff27 KH |
4071 | { |
4072 | .name = "oom_control", | |
2da8ca82 | 4073 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4074 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4075 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4076 | }, | |
70ddf637 AV |
4077 | { |
4078 | .name = "pressure_level", | |
70ddf637 | 4079 | }, |
406eb0c9 YH |
4080 | #ifdef CONFIG_NUMA |
4081 | { | |
4082 | .name = "numa_stat", | |
2da8ca82 | 4083 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4084 | }, |
4085 | #endif | |
510fc4e1 GC |
4086 | #ifdef CONFIG_MEMCG_KMEM |
4087 | { | |
4088 | .name = "kmem.limit_in_bytes", | |
4089 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4090 | .write = mem_cgroup_write, |
791badbd | 4091 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4092 | }, |
4093 | { | |
4094 | .name = "kmem.usage_in_bytes", | |
4095 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4096 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4097 | }, |
4098 | { | |
4099 | .name = "kmem.failcnt", | |
4100 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4101 | .write = mem_cgroup_reset, |
791badbd | 4102 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4103 | }, |
4104 | { | |
4105 | .name = "kmem.max_usage_in_bytes", | |
4106 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4107 | .write = mem_cgroup_reset, |
791badbd | 4108 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4109 | }, |
749c5415 GC |
4110 | #ifdef CONFIG_SLABINFO |
4111 | { | |
4112 | .name = "kmem.slabinfo", | |
b047501c VD |
4113 | .seq_start = slab_start, |
4114 | .seq_next = slab_next, | |
4115 | .seq_stop = slab_stop, | |
4116 | .seq_show = memcg_slab_show, | |
749c5415 GC |
4117 | }, |
4118 | #endif | |
8c7c6e34 | 4119 | #endif |
6bc10349 | 4120 | { }, /* terminate */ |
af36f906 | 4121 | }; |
8c7c6e34 | 4122 | |
c0ff4b85 | 4123 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4124 | { |
4125 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4126 | struct mem_cgroup_per_zone *mz; |
41e3355d | 4127 | int zone, tmp = node; |
1ecaab2b KH |
4128 | /* |
4129 | * This routine is called against possible nodes. | |
4130 | * But it's BUG to call kmalloc() against offline node. | |
4131 | * | |
4132 | * TODO: this routine can waste much memory for nodes which will | |
4133 | * never be onlined. It's better to use memory hotplug callback | |
4134 | * function. | |
4135 | */ | |
41e3355d KH |
4136 | if (!node_state(node, N_NORMAL_MEMORY)) |
4137 | tmp = -1; | |
17295c88 | 4138 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4139 | if (!pn) |
4140 | return 1; | |
1ecaab2b | 4141 | |
1ecaab2b KH |
4142 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4143 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 4144 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
4145 | mz->usage_in_excess = 0; |
4146 | mz->on_tree = false; | |
d79154bb | 4147 | mz->memcg = memcg; |
1ecaab2b | 4148 | } |
54f72fe0 | 4149 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4150 | return 0; |
4151 | } | |
4152 | ||
c0ff4b85 | 4153 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4154 | { |
54f72fe0 | 4155 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
4156 | } |
4157 | ||
33327948 KH |
4158 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4159 | { | |
d79154bb | 4160 | struct mem_cgroup *memcg; |
8ff69e2c | 4161 | size_t size; |
33327948 | 4162 | |
8ff69e2c VD |
4163 | size = sizeof(struct mem_cgroup); |
4164 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
33327948 | 4165 | |
8ff69e2c | 4166 | memcg = kzalloc(size, GFP_KERNEL); |
d79154bb | 4167 | if (!memcg) |
e7bbcdf3 DC |
4168 | return NULL; |
4169 | ||
d79154bb HD |
4170 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
4171 | if (!memcg->stat) | |
d2e61b8d | 4172 | goto out_free; |
841710aa TH |
4173 | |
4174 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) | |
4175 | goto out_free_stat; | |
4176 | ||
d79154bb | 4177 | return memcg; |
d2e61b8d | 4178 | |
841710aa TH |
4179 | out_free_stat: |
4180 | free_percpu(memcg->stat); | |
d2e61b8d | 4181 | out_free: |
8ff69e2c | 4182 | kfree(memcg); |
d2e61b8d | 4183 | return NULL; |
33327948 KH |
4184 | } |
4185 | ||
59927fb9 | 4186 | /* |
c8b2a36f GC |
4187 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
4188 | * (scanning all at force_empty is too costly...) | |
4189 | * | |
4190 | * Instead of clearing all references at force_empty, we remember | |
4191 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4192 | * it goes down to 0. | |
4193 | * | |
4194 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 4195 | */ |
c8b2a36f GC |
4196 | |
4197 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 4198 | { |
c8b2a36f | 4199 | int node; |
59927fb9 | 4200 | |
bb4cc1a8 | 4201 | mem_cgroup_remove_from_trees(memcg); |
c8b2a36f GC |
4202 | |
4203 | for_each_node(node) | |
4204 | free_mem_cgroup_per_zone_info(memcg, node); | |
4205 | ||
4206 | free_percpu(memcg->stat); | |
841710aa | 4207 | memcg_wb_domain_exit(memcg); |
8ff69e2c | 4208 | kfree(memcg); |
59927fb9 | 4209 | } |
3afe36b1 | 4210 | |
7bcc1bb1 DN |
4211 | /* |
4212 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4213 | */ | |
e1aab161 | 4214 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 4215 | { |
3e32cb2e | 4216 | if (!memcg->memory.parent) |
7bcc1bb1 | 4217 | return NULL; |
3e32cb2e | 4218 | return mem_cgroup_from_counter(memcg->memory.parent, memory); |
7bcc1bb1 | 4219 | } |
e1aab161 | 4220 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 4221 | |
0eb253e2 | 4222 | static struct cgroup_subsys_state * __ref |
eb95419b | 4223 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 4224 | { |
d142e3e6 | 4225 | struct mem_cgroup *memcg; |
04046e1a | 4226 | long error = -ENOMEM; |
6d12e2d8 | 4227 | int node; |
8cdea7c0 | 4228 | |
c0ff4b85 R |
4229 | memcg = mem_cgroup_alloc(); |
4230 | if (!memcg) | |
04046e1a | 4231 | return ERR_PTR(error); |
78fb7466 | 4232 | |
3ed28fa1 | 4233 | for_each_node(node) |
c0ff4b85 | 4234 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 4235 | goto free_out; |
f64c3f54 | 4236 | |
c077719b | 4237 | /* root ? */ |
eb95419b | 4238 | if (parent_css == NULL) { |
a41c58a6 | 4239 | root_mem_cgroup = memcg; |
56161634 | 4240 | mem_cgroup_root_css = &memcg->css; |
3e32cb2e | 4241 | page_counter_init(&memcg->memory, NULL); |
241994ed | 4242 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4243 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4244 | page_counter_init(&memcg->memsw, NULL); |
4245 | page_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 4246 | } |
28dbc4b6 | 4247 | |
d142e3e6 GC |
4248 | memcg->last_scanned_node = MAX_NUMNODES; |
4249 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4250 | memcg->move_charge_at_immigrate = 0; |
4251 | mutex_init(&memcg->thresholds_lock); | |
4252 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4253 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4254 | INIT_LIST_HEAD(&memcg->event_list); |
4255 | spin_lock_init(&memcg->event_list_lock); | |
900a38f0 VD |
4256 | #ifdef CONFIG_MEMCG_KMEM |
4257 | memcg->kmemcg_id = -1; | |
900a38f0 | 4258 | #endif |
52ebea74 TH |
4259 | #ifdef CONFIG_CGROUP_WRITEBACK |
4260 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
4261 | #endif | |
d142e3e6 GC |
4262 | return &memcg->css; |
4263 | ||
4264 | free_out: | |
4265 | __mem_cgroup_free(memcg); | |
4266 | return ERR_PTR(error); | |
4267 | } | |
4268 | ||
4269 | static int | |
eb95419b | 4270 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 4271 | { |
eb95419b | 4272 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 4273 | struct mem_cgroup *parent = mem_cgroup_from_css(css->parent); |
2f7dd7a4 | 4274 | int ret; |
d142e3e6 | 4275 | |
15a4c835 | 4276 | if (css->id > MEM_CGROUP_ID_MAX) |
4219b2da LZ |
4277 | return -ENOSPC; |
4278 | ||
63876986 | 4279 | if (!parent) |
d142e3e6 GC |
4280 | return 0; |
4281 | ||
0999821b | 4282 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
4283 | |
4284 | memcg->use_hierarchy = parent->use_hierarchy; | |
4285 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4286 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4287 | ||
4288 | if (parent->use_hierarchy) { | |
3e32cb2e | 4289 | page_counter_init(&memcg->memory, &parent->memory); |
241994ed | 4290 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4291 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4292 | page_counter_init(&memcg->memsw, &parent->memsw); |
4293 | page_counter_init(&memcg->kmem, &parent->kmem); | |
55007d84 | 4294 | |
7bcc1bb1 | 4295 | /* |
8d76a979 LZ |
4296 | * No need to take a reference to the parent because cgroup |
4297 | * core guarantees its existence. | |
7bcc1bb1 | 4298 | */ |
18f59ea7 | 4299 | } else { |
3e32cb2e | 4300 | page_counter_init(&memcg->memory, NULL); |
241994ed | 4301 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4302 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4303 | page_counter_init(&memcg->memsw, NULL); |
4304 | page_counter_init(&memcg->kmem, NULL); | |
8c7f6edb TH |
4305 | /* |
4306 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4307 | * much sense so let cgroup subsystem know about this | |
4308 | * unfortunate state in our controller. | |
4309 | */ | |
d142e3e6 | 4310 | if (parent != root_mem_cgroup) |
073219e9 | 4311 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4312 | } |
0999821b | 4313 | mutex_unlock(&memcg_create_mutex); |
d6441637 | 4314 | |
2f7dd7a4 JW |
4315 | ret = memcg_init_kmem(memcg, &memory_cgrp_subsys); |
4316 | if (ret) | |
4317 | return ret; | |
4318 | ||
4319 | /* | |
4320 | * Make sure the memcg is initialized: mem_cgroup_iter() | |
4321 | * orders reading memcg->initialized against its callers | |
4322 | * reading the memcg members. | |
4323 | */ | |
4324 | smp_store_release(&memcg->initialized, 1); | |
4325 | ||
4326 | return 0; | |
8cdea7c0 BS |
4327 | } |
4328 | ||
eb95419b | 4329 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4330 | { |
eb95419b | 4331 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4332 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4333 | |
4334 | /* | |
4335 | * Unregister events and notify userspace. | |
4336 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4337 | * directory to avoid race between userspace and kernelspace. | |
4338 | */ | |
fba94807 TH |
4339 | spin_lock(&memcg->event_list_lock); |
4340 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4341 | list_del_init(&event->list); |
4342 | schedule_work(&event->remove); | |
4343 | } | |
fba94807 | 4344 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4345 | |
33cb876e | 4346 | vmpressure_cleanup(&memcg->vmpressure); |
2a4db7eb VD |
4347 | |
4348 | memcg_deactivate_kmem(memcg); | |
52ebea74 TH |
4349 | |
4350 | wb_memcg_offline(memcg); | |
df878fb0 KH |
4351 | } |
4352 | ||
6df38689 VD |
4353 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
4354 | { | |
4355 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4356 | ||
4357 | invalidate_reclaim_iterators(memcg); | |
4358 | } | |
4359 | ||
eb95419b | 4360 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4361 | { |
eb95419b | 4362 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4363 | |
10d5ebf4 | 4364 | memcg_destroy_kmem(memcg); |
465939a1 | 4365 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
4366 | } |
4367 | ||
1ced953b TH |
4368 | /** |
4369 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4370 | * @css: the target css | |
4371 | * | |
4372 | * Reset the states of the mem_cgroup associated with @css. This is | |
4373 | * invoked when the userland requests disabling on the default hierarchy | |
4374 | * but the memcg is pinned through dependency. The memcg should stop | |
4375 | * applying policies and should revert to the vanilla state as it may be | |
4376 | * made visible again. | |
4377 | * | |
4378 | * The current implementation only resets the essential configurations. | |
4379 | * This needs to be expanded to cover all the visible parts. | |
4380 | */ | |
4381 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4382 | { | |
4383 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4384 | ||
3e32cb2e JW |
4385 | mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX); |
4386 | mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX); | |
4387 | memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX); | |
241994ed JW |
4388 | memcg->low = 0; |
4389 | memcg->high = PAGE_COUNTER_MAX; | |
24d404dc | 4390 | memcg->soft_limit = PAGE_COUNTER_MAX; |
2529bb3a | 4391 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
4392 | } |
4393 | ||
02491447 | 4394 | #ifdef CONFIG_MMU |
7dc74be0 | 4395 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4396 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4397 | { |
05b84301 | 4398 | int ret; |
9476db97 | 4399 | |
d0164adc MG |
4400 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
4401 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 4402 | if (!ret) { |
854ffa8d | 4403 | mc.precharge += count; |
854ffa8d DN |
4404 | return ret; |
4405 | } | |
9476db97 JW |
4406 | |
4407 | /* Try charges one by one with reclaim */ | |
854ffa8d | 4408 | while (count--) { |
00501b53 | 4409 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1); |
38c5d72f | 4410 | if (ret) |
38c5d72f | 4411 | return ret; |
854ffa8d | 4412 | mc.precharge++; |
9476db97 | 4413 | cond_resched(); |
854ffa8d | 4414 | } |
9476db97 | 4415 | return 0; |
4ffef5fe DN |
4416 | } |
4417 | ||
4418 | /** | |
8d32ff84 | 4419 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
4420 | * @vma: the vma the pte to be checked belongs |
4421 | * @addr: the address corresponding to the pte to be checked | |
4422 | * @ptent: the pte to be checked | |
02491447 | 4423 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
4424 | * |
4425 | * Returns | |
4426 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4427 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4428 | * move charge. if @target is not NULL, the page is stored in target->page | |
4429 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
4430 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
4431 | * target for charge migration. if @target is not NULL, the entry is stored | |
4432 | * in target->ent. | |
4ffef5fe DN |
4433 | * |
4434 | * Called with pte lock held. | |
4435 | */ | |
4ffef5fe DN |
4436 | union mc_target { |
4437 | struct page *page; | |
02491447 | 4438 | swp_entry_t ent; |
4ffef5fe DN |
4439 | }; |
4440 | ||
4ffef5fe | 4441 | enum mc_target_type { |
8d32ff84 | 4442 | MC_TARGET_NONE = 0, |
4ffef5fe | 4443 | MC_TARGET_PAGE, |
02491447 | 4444 | MC_TARGET_SWAP, |
4ffef5fe DN |
4445 | }; |
4446 | ||
90254a65 DN |
4447 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4448 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4449 | { |
90254a65 | 4450 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 4451 | |
90254a65 DN |
4452 | if (!page || !page_mapped(page)) |
4453 | return NULL; | |
4454 | if (PageAnon(page)) { | |
1dfab5ab | 4455 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 4456 | return NULL; |
1dfab5ab JW |
4457 | } else { |
4458 | if (!(mc.flags & MOVE_FILE)) | |
4459 | return NULL; | |
4460 | } | |
90254a65 DN |
4461 | if (!get_page_unless_zero(page)) |
4462 | return NULL; | |
4463 | ||
4464 | return page; | |
4465 | } | |
4466 | ||
4b91355e | 4467 | #ifdef CONFIG_SWAP |
90254a65 DN |
4468 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
4469 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4470 | { | |
90254a65 DN |
4471 | struct page *page = NULL; |
4472 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4473 | ||
1dfab5ab | 4474 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 4475 | return NULL; |
4b91355e KH |
4476 | /* |
4477 | * Because lookup_swap_cache() updates some statistics counter, | |
4478 | * we call find_get_page() with swapper_space directly. | |
4479 | */ | |
33806f06 | 4480 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
4481 | if (do_swap_account) |
4482 | entry->val = ent.val; | |
4483 | ||
4484 | return page; | |
4485 | } | |
4b91355e KH |
4486 | #else |
4487 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
4488 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4489 | { | |
4490 | return NULL; | |
4491 | } | |
4492 | #endif | |
90254a65 | 4493 | |
87946a72 DN |
4494 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4495 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4496 | { | |
4497 | struct page *page = NULL; | |
87946a72 DN |
4498 | struct address_space *mapping; |
4499 | pgoff_t pgoff; | |
4500 | ||
4501 | if (!vma->vm_file) /* anonymous vma */ | |
4502 | return NULL; | |
1dfab5ab | 4503 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
4504 | return NULL; |
4505 | ||
87946a72 | 4506 | mapping = vma->vm_file->f_mapping; |
0661a336 | 4507 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
4508 | |
4509 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
4510 | #ifdef CONFIG_SWAP |
4511 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
4512 | if (shmem_mapping(mapping)) { |
4513 | page = find_get_entry(mapping, pgoff); | |
4514 | if (radix_tree_exceptional_entry(page)) { | |
4515 | swp_entry_t swp = radix_to_swp_entry(page); | |
4516 | if (do_swap_account) | |
4517 | *entry = swp; | |
4518 | page = find_get_page(swap_address_space(swp), swp.val); | |
4519 | } | |
4520 | } else | |
4521 | page = find_get_page(mapping, pgoff); | |
4522 | #else | |
4523 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 4524 | #endif |
87946a72 DN |
4525 | return page; |
4526 | } | |
4527 | ||
b1b0deab CG |
4528 | /** |
4529 | * mem_cgroup_move_account - move account of the page | |
4530 | * @page: the page | |
4531 | * @nr_pages: number of regular pages (>1 for huge pages) | |
4532 | * @from: mem_cgroup which the page is moved from. | |
4533 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
4534 | * | |
4535 | * The caller must confirm following. | |
4536 | * - page is not on LRU (isolate_page() is useful.) | |
4537 | * - compound_lock is held when nr_pages > 1 | |
4538 | * | |
4539 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
4540 | * from old cgroup. | |
4541 | */ | |
4542 | static int mem_cgroup_move_account(struct page *page, | |
4543 | unsigned int nr_pages, | |
4544 | struct mem_cgroup *from, | |
4545 | struct mem_cgroup *to) | |
4546 | { | |
4547 | unsigned long flags; | |
4548 | int ret; | |
c4843a75 | 4549 | bool anon; |
b1b0deab CG |
4550 | |
4551 | VM_BUG_ON(from == to); | |
4552 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
4553 | /* | |
4554 | * The page is isolated from LRU. So, collapse function | |
4555 | * will not handle this page. But page splitting can happen. | |
4556 | * Do this check under compound_page_lock(). The caller should | |
4557 | * hold it. | |
4558 | */ | |
4559 | ret = -EBUSY; | |
4560 | if (nr_pages > 1 && !PageTransHuge(page)) | |
4561 | goto out; | |
4562 | ||
4563 | /* | |
45637bab HD |
4564 | * Prevent mem_cgroup_replace_page() from looking at |
4565 | * page->mem_cgroup of its source page while we change it. | |
b1b0deab CG |
4566 | */ |
4567 | if (!trylock_page(page)) | |
4568 | goto out; | |
4569 | ||
4570 | ret = -EINVAL; | |
4571 | if (page->mem_cgroup != from) | |
4572 | goto out_unlock; | |
4573 | ||
c4843a75 GT |
4574 | anon = PageAnon(page); |
4575 | ||
b1b0deab CG |
4576 | spin_lock_irqsave(&from->move_lock, flags); |
4577 | ||
c4843a75 | 4578 | if (!anon && page_mapped(page)) { |
b1b0deab CG |
4579 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
4580 | nr_pages); | |
4581 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
4582 | nr_pages); | |
4583 | } | |
4584 | ||
c4843a75 GT |
4585 | /* |
4586 | * move_lock grabbed above and caller set from->moving_account, so | |
4587 | * mem_cgroup_update_page_stat() will serialize updates to PageDirty. | |
4588 | * So mapping should be stable for dirty pages. | |
4589 | */ | |
4590 | if (!anon && PageDirty(page)) { | |
4591 | struct address_space *mapping = page_mapping(page); | |
4592 | ||
4593 | if (mapping_cap_account_dirty(mapping)) { | |
4594 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY], | |
4595 | nr_pages); | |
4596 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY], | |
4597 | nr_pages); | |
4598 | } | |
4599 | } | |
4600 | ||
b1b0deab CG |
4601 | if (PageWriteback(page)) { |
4602 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
4603 | nr_pages); | |
4604 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
4605 | nr_pages); | |
4606 | } | |
4607 | ||
4608 | /* | |
4609 | * It is safe to change page->mem_cgroup here because the page | |
4610 | * is referenced, charged, and isolated - we can't race with | |
4611 | * uncharging, charging, migration, or LRU putback. | |
4612 | */ | |
4613 | ||
4614 | /* caller should have done css_get */ | |
4615 | page->mem_cgroup = to; | |
4616 | spin_unlock_irqrestore(&from->move_lock, flags); | |
4617 | ||
4618 | ret = 0; | |
4619 | ||
4620 | local_irq_disable(); | |
4621 | mem_cgroup_charge_statistics(to, page, nr_pages); | |
4622 | memcg_check_events(to, page); | |
4623 | mem_cgroup_charge_statistics(from, page, -nr_pages); | |
4624 | memcg_check_events(from, page); | |
4625 | local_irq_enable(); | |
4626 | out_unlock: | |
4627 | unlock_page(page); | |
4628 | out: | |
4629 | return ret; | |
4630 | } | |
4631 | ||
8d32ff84 | 4632 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
4633 | unsigned long addr, pte_t ptent, union mc_target *target) |
4634 | { | |
4635 | struct page *page = NULL; | |
8d32ff84 | 4636 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
4637 | swp_entry_t ent = { .val = 0 }; |
4638 | ||
4639 | if (pte_present(ptent)) | |
4640 | page = mc_handle_present_pte(vma, addr, ptent); | |
4641 | else if (is_swap_pte(ptent)) | |
4642 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
0661a336 | 4643 | else if (pte_none(ptent)) |
87946a72 | 4644 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
4645 | |
4646 | if (!page && !ent.val) | |
8d32ff84 | 4647 | return ret; |
02491447 | 4648 | if (page) { |
02491447 | 4649 | /* |
0a31bc97 | 4650 | * Do only loose check w/o serialization. |
1306a85a | 4651 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 4652 | * not under LRU exclusion. |
02491447 | 4653 | */ |
1306a85a | 4654 | if (page->mem_cgroup == mc.from) { |
02491447 DN |
4655 | ret = MC_TARGET_PAGE; |
4656 | if (target) | |
4657 | target->page = page; | |
4658 | } | |
4659 | if (!ret || !target) | |
4660 | put_page(page); | |
4661 | } | |
90254a65 DN |
4662 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
4663 | if (ent.val && !ret && | |
34c00c31 | 4664 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
4665 | ret = MC_TARGET_SWAP; |
4666 | if (target) | |
4667 | target->ent = ent; | |
4ffef5fe | 4668 | } |
4ffef5fe DN |
4669 | return ret; |
4670 | } | |
4671 | ||
12724850 NH |
4672 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4673 | /* | |
4674 | * We don't consider swapping or file mapped pages because THP does not | |
4675 | * support them for now. | |
4676 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
4677 | */ | |
4678 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4679 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4680 | { | |
4681 | struct page *page = NULL; | |
12724850 NH |
4682 | enum mc_target_type ret = MC_TARGET_NONE; |
4683 | ||
4684 | page = pmd_page(pmd); | |
309381fe | 4685 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 4686 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 4687 | return ret; |
1306a85a | 4688 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
4689 | ret = MC_TARGET_PAGE; |
4690 | if (target) { | |
4691 | get_page(page); | |
4692 | target->page = page; | |
4693 | } | |
4694 | } | |
4695 | return ret; | |
4696 | } | |
4697 | #else | |
4698 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4699 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4700 | { | |
4701 | return MC_TARGET_NONE; | |
4702 | } | |
4703 | #endif | |
4704 | ||
4ffef5fe DN |
4705 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
4706 | unsigned long addr, unsigned long end, | |
4707 | struct mm_walk *walk) | |
4708 | { | |
26bcd64a | 4709 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4710 | pte_t *pte; |
4711 | spinlock_t *ptl; | |
4712 | ||
bf929152 | 4713 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
12724850 NH |
4714 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
4715 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 4716 | spin_unlock(ptl); |
1a5a9906 | 4717 | return 0; |
12724850 | 4718 | } |
03319327 | 4719 | |
45f83cef AA |
4720 | if (pmd_trans_unstable(pmd)) |
4721 | return 0; | |
4ffef5fe DN |
4722 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
4723 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 4724 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
4725 | mc.precharge++; /* increment precharge temporarily */ |
4726 | pte_unmap_unlock(pte - 1, ptl); | |
4727 | cond_resched(); | |
4728 | ||
7dc74be0 DN |
4729 | return 0; |
4730 | } | |
4731 | ||
4ffef5fe DN |
4732 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
4733 | { | |
4734 | unsigned long precharge; | |
4ffef5fe | 4735 | |
26bcd64a NH |
4736 | struct mm_walk mem_cgroup_count_precharge_walk = { |
4737 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
4738 | .mm = mm, | |
4739 | }; | |
dfe076b0 | 4740 | down_read(&mm->mmap_sem); |
26bcd64a | 4741 | walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk); |
dfe076b0 | 4742 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
4743 | |
4744 | precharge = mc.precharge; | |
4745 | mc.precharge = 0; | |
4746 | ||
4747 | return precharge; | |
4748 | } | |
4749 | ||
4ffef5fe DN |
4750 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
4751 | { | |
dfe076b0 DN |
4752 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
4753 | ||
4754 | VM_BUG_ON(mc.moving_task); | |
4755 | mc.moving_task = current; | |
4756 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
4757 | } |
4758 | ||
dfe076b0 DN |
4759 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
4760 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 4761 | { |
2bd9bb20 KH |
4762 | struct mem_cgroup *from = mc.from; |
4763 | struct mem_cgroup *to = mc.to; | |
4764 | ||
4ffef5fe | 4765 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 4766 | if (mc.precharge) { |
00501b53 | 4767 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
4768 | mc.precharge = 0; |
4769 | } | |
4770 | /* | |
4771 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
4772 | * we must uncharge here. | |
4773 | */ | |
4774 | if (mc.moved_charge) { | |
00501b53 | 4775 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 4776 | mc.moved_charge = 0; |
4ffef5fe | 4777 | } |
483c30b5 DN |
4778 | /* we must fixup refcnts and charges */ |
4779 | if (mc.moved_swap) { | |
483c30b5 | 4780 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 4781 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 4782 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 4783 | |
05b84301 | 4784 | /* |
3e32cb2e JW |
4785 | * we charged both to->memory and to->memsw, so we |
4786 | * should uncharge to->memory. | |
05b84301 | 4787 | */ |
ce00a967 | 4788 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
4789 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
4790 | ||
e8ea14cc | 4791 | css_put_many(&mc.from->css, mc.moved_swap); |
3e32cb2e | 4792 | |
4050377b | 4793 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
4794 | mc.moved_swap = 0; |
4795 | } | |
dfe076b0 DN |
4796 | memcg_oom_recover(from); |
4797 | memcg_oom_recover(to); | |
4798 | wake_up_all(&mc.waitq); | |
4799 | } | |
4800 | ||
4801 | static void mem_cgroup_clear_mc(void) | |
4802 | { | |
dfe076b0 DN |
4803 | /* |
4804 | * we must clear moving_task before waking up waiters at the end of | |
4805 | * task migration. | |
4806 | */ | |
4807 | mc.moving_task = NULL; | |
4808 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 4809 | spin_lock(&mc.lock); |
4ffef5fe DN |
4810 | mc.from = NULL; |
4811 | mc.to = NULL; | |
2bd9bb20 | 4812 | spin_unlock(&mc.lock); |
4ffef5fe DN |
4813 | } |
4814 | ||
1f7dd3e5 | 4815 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 4816 | { |
1f7dd3e5 TH |
4817 | struct cgroup_subsys_state *css; |
4818 | struct mem_cgroup *memcg; | |
9f2115f9 | 4819 | struct mem_cgroup *from; |
4530eddb | 4820 | struct task_struct *leader, *p; |
9f2115f9 | 4821 | struct mm_struct *mm; |
1dfab5ab | 4822 | unsigned long move_flags; |
9f2115f9 | 4823 | int ret = 0; |
7dc74be0 | 4824 | |
1f7dd3e5 TH |
4825 | /* charge immigration isn't supported on the default hierarchy */ |
4826 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
4827 | return 0; |
4828 | ||
4530eddb TH |
4829 | /* |
4830 | * Multi-process migrations only happen on the default hierarchy | |
4831 | * where charge immigration is not used. Perform charge | |
4832 | * immigration if @tset contains a leader and whine if there are | |
4833 | * multiple. | |
4834 | */ | |
4835 | p = NULL; | |
1f7dd3e5 | 4836 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
4837 | WARN_ON_ONCE(p); |
4838 | p = leader; | |
1f7dd3e5 | 4839 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
4840 | } |
4841 | if (!p) | |
4842 | return 0; | |
4843 | ||
1f7dd3e5 TH |
4844 | /* |
4845 | * We are now commited to this value whatever it is. Changes in this | |
4846 | * tunable will only affect upcoming migrations, not the current one. | |
4847 | * So we need to save it, and keep it going. | |
4848 | */ | |
4849 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
4850 | if (!move_flags) | |
4851 | return 0; | |
4852 | ||
9f2115f9 TH |
4853 | from = mem_cgroup_from_task(p); |
4854 | ||
4855 | VM_BUG_ON(from == memcg); | |
4856 | ||
4857 | mm = get_task_mm(p); | |
4858 | if (!mm) | |
4859 | return 0; | |
4860 | /* We move charges only when we move a owner of the mm */ | |
4861 | if (mm->owner == p) { | |
4862 | VM_BUG_ON(mc.from); | |
4863 | VM_BUG_ON(mc.to); | |
4864 | VM_BUG_ON(mc.precharge); | |
4865 | VM_BUG_ON(mc.moved_charge); | |
4866 | VM_BUG_ON(mc.moved_swap); | |
4867 | ||
4868 | spin_lock(&mc.lock); | |
4869 | mc.from = from; | |
4870 | mc.to = memcg; | |
4871 | mc.flags = move_flags; | |
4872 | spin_unlock(&mc.lock); | |
4873 | /* We set mc.moving_task later */ | |
4874 | ||
4875 | ret = mem_cgroup_precharge_mc(mm); | |
4876 | if (ret) | |
4877 | mem_cgroup_clear_mc(); | |
7dc74be0 | 4878 | } |
9f2115f9 | 4879 | mmput(mm); |
7dc74be0 DN |
4880 | return ret; |
4881 | } | |
4882 | ||
1f7dd3e5 | 4883 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 4884 | { |
4e2f245d JW |
4885 | if (mc.to) |
4886 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
4887 | } |
4888 | ||
4ffef5fe DN |
4889 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
4890 | unsigned long addr, unsigned long end, | |
4891 | struct mm_walk *walk) | |
7dc74be0 | 4892 | { |
4ffef5fe | 4893 | int ret = 0; |
26bcd64a | 4894 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4895 | pte_t *pte; |
4896 | spinlock_t *ptl; | |
12724850 NH |
4897 | enum mc_target_type target_type; |
4898 | union mc_target target; | |
4899 | struct page *page; | |
4ffef5fe | 4900 | |
12724850 NH |
4901 | /* |
4902 | * We don't take compound_lock() here but no race with splitting thp | |
4903 | * happens because: | |
4904 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
4905 | * under splitting, which means there's no concurrent thp split, | |
4906 | * - if another thread runs into split_huge_page() just after we | |
4907 | * entered this if-block, the thread must wait for page table lock | |
4908 | * to be unlocked in __split_huge_page_splitting(), where the main | |
4909 | * part of thp split is not executed yet. | |
4910 | */ | |
bf929152 | 4911 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
62ade86a | 4912 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 4913 | spin_unlock(ptl); |
12724850 NH |
4914 | return 0; |
4915 | } | |
4916 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
4917 | if (target_type == MC_TARGET_PAGE) { | |
4918 | page = target.page; | |
4919 | if (!isolate_lru_page(page)) { | |
12724850 | 4920 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, |
1306a85a | 4921 | mc.from, mc.to)) { |
12724850 NH |
4922 | mc.precharge -= HPAGE_PMD_NR; |
4923 | mc.moved_charge += HPAGE_PMD_NR; | |
4924 | } | |
4925 | putback_lru_page(page); | |
4926 | } | |
4927 | put_page(page); | |
4928 | } | |
bf929152 | 4929 | spin_unlock(ptl); |
1a5a9906 | 4930 | return 0; |
12724850 NH |
4931 | } |
4932 | ||
45f83cef AA |
4933 | if (pmd_trans_unstable(pmd)) |
4934 | return 0; | |
4ffef5fe DN |
4935 | retry: |
4936 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
4937 | for (; addr != end; addr += PAGE_SIZE) { | |
4938 | pte_t ptent = *(pte++); | |
02491447 | 4939 | swp_entry_t ent; |
4ffef5fe DN |
4940 | |
4941 | if (!mc.precharge) | |
4942 | break; | |
4943 | ||
8d32ff84 | 4944 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
4945 | case MC_TARGET_PAGE: |
4946 | page = target.page; | |
4947 | if (isolate_lru_page(page)) | |
4948 | goto put; | |
1306a85a | 4949 | if (!mem_cgroup_move_account(page, 1, mc.from, mc.to)) { |
4ffef5fe | 4950 | mc.precharge--; |
854ffa8d DN |
4951 | /* we uncharge from mc.from later. */ |
4952 | mc.moved_charge++; | |
4ffef5fe DN |
4953 | } |
4954 | putback_lru_page(page); | |
8d32ff84 | 4955 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
4956 | put_page(page); |
4957 | break; | |
02491447 DN |
4958 | case MC_TARGET_SWAP: |
4959 | ent = target.ent; | |
e91cbb42 | 4960 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 4961 | mc.precharge--; |
483c30b5 DN |
4962 | /* we fixup refcnts and charges later. */ |
4963 | mc.moved_swap++; | |
4964 | } | |
02491447 | 4965 | break; |
4ffef5fe DN |
4966 | default: |
4967 | break; | |
4968 | } | |
4969 | } | |
4970 | pte_unmap_unlock(pte - 1, ptl); | |
4971 | cond_resched(); | |
4972 | ||
4973 | if (addr != end) { | |
4974 | /* | |
4975 | * We have consumed all precharges we got in can_attach(). | |
4976 | * We try charge one by one, but don't do any additional | |
4977 | * charges to mc.to if we have failed in charge once in attach() | |
4978 | * phase. | |
4979 | */ | |
854ffa8d | 4980 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
4981 | if (!ret) |
4982 | goto retry; | |
4983 | } | |
4984 | ||
4985 | return ret; | |
4986 | } | |
4987 | ||
4988 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
4989 | { | |
26bcd64a NH |
4990 | struct mm_walk mem_cgroup_move_charge_walk = { |
4991 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
4992 | .mm = mm, | |
4993 | }; | |
4ffef5fe DN |
4994 | |
4995 | lru_add_drain_all(); | |
312722cb JW |
4996 | /* |
4997 | * Signal mem_cgroup_begin_page_stat() to take the memcg's | |
4998 | * move_lock while we're moving its pages to another memcg. | |
4999 | * Then wait for already started RCU-only updates to finish. | |
5000 | */ | |
5001 | atomic_inc(&mc.from->moving_account); | |
5002 | synchronize_rcu(); | |
dfe076b0 DN |
5003 | retry: |
5004 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5005 | /* | |
5006 | * Someone who are holding the mmap_sem might be waiting in | |
5007 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5008 | * and retry. Because we cancel precharges, we might not be able | |
5009 | * to move enough charges, but moving charge is a best-effort | |
5010 | * feature anyway, so it wouldn't be a big problem. | |
5011 | */ | |
5012 | __mem_cgroup_clear_mc(); | |
5013 | cond_resched(); | |
5014 | goto retry; | |
5015 | } | |
26bcd64a NH |
5016 | /* |
5017 | * When we have consumed all precharges and failed in doing | |
5018 | * additional charge, the page walk just aborts. | |
5019 | */ | |
5020 | walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk); | |
dfe076b0 | 5021 | up_read(&mm->mmap_sem); |
312722cb | 5022 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
5023 | } |
5024 | ||
1f7dd3e5 | 5025 | static void mem_cgroup_move_task(struct cgroup_taskset *tset) |
67e465a7 | 5026 | { |
1f7dd3e5 TH |
5027 | struct cgroup_subsys_state *css; |
5028 | struct task_struct *p = cgroup_taskset_first(tset, &css); | |
a433658c | 5029 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 5030 | |
dfe076b0 | 5031 | if (mm) { |
a433658c KM |
5032 | if (mc.to) |
5033 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
5034 | mmput(mm); |
5035 | } | |
a433658c KM |
5036 | if (mc.to) |
5037 | mem_cgroup_clear_mc(); | |
67e465a7 | 5038 | } |
5cfb80a7 | 5039 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 5040 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5041 | { |
5042 | return 0; | |
5043 | } | |
1f7dd3e5 | 5044 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5045 | { |
5046 | } | |
1f7dd3e5 | 5047 | static void mem_cgroup_move_task(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5048 | { |
5049 | } | |
5050 | #endif | |
67e465a7 | 5051 | |
f00baae7 TH |
5052 | /* |
5053 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5054 | * to verify whether we're attached to the default hierarchy on each mount |
5055 | * attempt. | |
f00baae7 | 5056 | */ |
eb95419b | 5057 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5058 | { |
5059 | /* | |
aa6ec29b | 5060 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5061 | * guarantees that @root doesn't have any children, so turning it |
5062 | * on for the root memcg is enough. | |
5063 | */ | |
9e10a130 | 5064 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
5065 | root_mem_cgroup->use_hierarchy = true; |
5066 | else | |
5067 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
5068 | } |
5069 | ||
241994ed JW |
5070 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
5071 | struct cftype *cft) | |
5072 | { | |
f5fc3c5d JW |
5073 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5074 | ||
5075 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
5076 | } |
5077 | ||
5078 | static int memory_low_show(struct seq_file *m, void *v) | |
5079 | { | |
5080 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5081 | unsigned long low = READ_ONCE(memcg->low); |
241994ed JW |
5082 | |
5083 | if (low == PAGE_COUNTER_MAX) | |
d2973697 | 5084 | seq_puts(m, "max\n"); |
241994ed JW |
5085 | else |
5086 | seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE); | |
5087 | ||
5088 | return 0; | |
5089 | } | |
5090 | ||
5091 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
5092 | char *buf, size_t nbytes, loff_t off) | |
5093 | { | |
5094 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5095 | unsigned long low; | |
5096 | int err; | |
5097 | ||
5098 | buf = strstrip(buf); | |
d2973697 | 5099 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
5100 | if (err) |
5101 | return err; | |
5102 | ||
5103 | memcg->low = low; | |
5104 | ||
5105 | return nbytes; | |
5106 | } | |
5107 | ||
5108 | static int memory_high_show(struct seq_file *m, void *v) | |
5109 | { | |
5110 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5111 | unsigned long high = READ_ONCE(memcg->high); |
241994ed JW |
5112 | |
5113 | if (high == PAGE_COUNTER_MAX) | |
d2973697 | 5114 | seq_puts(m, "max\n"); |
241994ed JW |
5115 | else |
5116 | seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE); | |
5117 | ||
5118 | return 0; | |
5119 | } | |
5120 | ||
5121 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
5122 | char *buf, size_t nbytes, loff_t off) | |
5123 | { | |
5124 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
8b42fc47 | 5125 | unsigned long nr_pages; |
241994ed JW |
5126 | unsigned long high; |
5127 | int err; | |
5128 | ||
5129 | buf = strstrip(buf); | |
d2973697 | 5130 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
5131 | if (err) |
5132 | return err; | |
5133 | ||
5134 | memcg->high = high; | |
5135 | ||
8b42fc47 JW |
5136 | nr_pages = page_counter_read(&memcg->memory); |
5137 | if (nr_pages > high) | |
5138 | try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
5139 | GFP_KERNEL, true); | |
5140 | ||
2529bb3a | 5141 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5142 | return nbytes; |
5143 | } | |
5144 | ||
5145 | static int memory_max_show(struct seq_file *m, void *v) | |
5146 | { | |
5147 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5148 | unsigned long max = READ_ONCE(memcg->memory.limit); |
241994ed JW |
5149 | |
5150 | if (max == PAGE_COUNTER_MAX) | |
d2973697 | 5151 | seq_puts(m, "max\n"); |
241994ed JW |
5152 | else |
5153 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
5154 | ||
5155 | return 0; | |
5156 | } | |
5157 | ||
5158 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
5159 | char *buf, size_t nbytes, loff_t off) | |
5160 | { | |
5161 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
0ccab5b1 JW |
5162 | unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES; |
5163 | bool drained = false; | |
241994ed JW |
5164 | unsigned long max; |
5165 | int err; | |
5166 | ||
5167 | buf = strstrip(buf); | |
d2973697 | 5168 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
5169 | if (err) |
5170 | return err; | |
5171 | ||
0ccab5b1 JW |
5172 | xchg(&memcg->memory.limit, max); |
5173 | ||
5174 | for (;;) { | |
5175 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
5176 | ||
5177 | if (nr_pages <= max) | |
5178 | break; | |
5179 | ||
5180 | if (signal_pending(current)) { | |
5181 | err = -EINTR; | |
5182 | break; | |
5183 | } | |
5184 | ||
5185 | if (!drained) { | |
5186 | drain_all_stock(memcg); | |
5187 | drained = true; | |
5188 | continue; | |
5189 | } | |
5190 | ||
5191 | if (nr_reclaims) { | |
5192 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
5193 | GFP_KERNEL, true)) | |
5194 | nr_reclaims--; | |
5195 | continue; | |
5196 | } | |
5197 | ||
5198 | mem_cgroup_events(memcg, MEMCG_OOM, 1); | |
5199 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) | |
5200 | break; | |
5201 | } | |
241994ed | 5202 | |
2529bb3a | 5203 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5204 | return nbytes; |
5205 | } | |
5206 | ||
5207 | static int memory_events_show(struct seq_file *m, void *v) | |
5208 | { | |
5209 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5210 | ||
5211 | seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW)); | |
5212 | seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH)); | |
5213 | seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX)); | |
5214 | seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM)); | |
5215 | ||
5216 | return 0; | |
5217 | } | |
5218 | ||
5219 | static struct cftype memory_files[] = { | |
5220 | { | |
5221 | .name = "current", | |
f5fc3c5d | 5222 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
5223 | .read_u64 = memory_current_read, |
5224 | }, | |
5225 | { | |
5226 | .name = "low", | |
5227 | .flags = CFTYPE_NOT_ON_ROOT, | |
5228 | .seq_show = memory_low_show, | |
5229 | .write = memory_low_write, | |
5230 | }, | |
5231 | { | |
5232 | .name = "high", | |
5233 | .flags = CFTYPE_NOT_ON_ROOT, | |
5234 | .seq_show = memory_high_show, | |
5235 | .write = memory_high_write, | |
5236 | }, | |
5237 | { | |
5238 | .name = "max", | |
5239 | .flags = CFTYPE_NOT_ON_ROOT, | |
5240 | .seq_show = memory_max_show, | |
5241 | .write = memory_max_write, | |
5242 | }, | |
5243 | { | |
5244 | .name = "events", | |
5245 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 5246 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
5247 | .seq_show = memory_events_show, |
5248 | }, | |
5249 | { } /* terminate */ | |
5250 | }; | |
5251 | ||
073219e9 | 5252 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5253 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5254 | .css_online = mem_cgroup_css_online, |
92fb9748 | 5255 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 5256 | .css_released = mem_cgroup_css_released, |
92fb9748 | 5257 | .css_free = mem_cgroup_css_free, |
1ced953b | 5258 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5259 | .can_attach = mem_cgroup_can_attach, |
5260 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5261 | .attach = mem_cgroup_move_task, |
f00baae7 | 5262 | .bind = mem_cgroup_bind, |
241994ed JW |
5263 | .dfl_cftypes = memory_files, |
5264 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 5265 | .early_init = 0, |
8cdea7c0 | 5266 | }; |
c077719b | 5267 | |
241994ed JW |
5268 | /** |
5269 | * mem_cgroup_low - check if memory consumption is below the normal range | |
5270 | * @root: the highest ancestor to consider | |
5271 | * @memcg: the memory cgroup to check | |
5272 | * | |
5273 | * Returns %true if memory consumption of @memcg, and that of all | |
5274 | * configurable ancestors up to @root, is below the normal range. | |
5275 | */ | |
5276 | bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg) | |
5277 | { | |
5278 | if (mem_cgroup_disabled()) | |
5279 | return false; | |
5280 | ||
5281 | /* | |
5282 | * The toplevel group doesn't have a configurable range, so | |
5283 | * it's never low when looked at directly, and it is not | |
5284 | * considered an ancestor when assessing the hierarchy. | |
5285 | */ | |
5286 | ||
5287 | if (memcg == root_mem_cgroup) | |
5288 | return false; | |
5289 | ||
4e54dede | 5290 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5291 | return false; |
5292 | ||
5293 | while (memcg != root) { | |
5294 | memcg = parent_mem_cgroup(memcg); | |
5295 | ||
5296 | if (memcg == root_mem_cgroup) | |
5297 | break; | |
5298 | ||
4e54dede | 5299 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5300 | return false; |
5301 | } | |
5302 | return true; | |
5303 | } | |
5304 | ||
00501b53 JW |
5305 | /** |
5306 | * mem_cgroup_try_charge - try charging a page | |
5307 | * @page: page to charge | |
5308 | * @mm: mm context of the victim | |
5309 | * @gfp_mask: reclaim mode | |
5310 | * @memcgp: charged memcg return | |
5311 | * | |
5312 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5313 | * pages according to @gfp_mask if necessary. | |
5314 | * | |
5315 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5316 | * Otherwise, an error code is returned. | |
5317 | * | |
5318 | * After page->mapping has been set up, the caller must finalize the | |
5319 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5320 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5321 | */ | |
5322 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
5323 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
5324 | { | |
5325 | struct mem_cgroup *memcg = NULL; | |
5326 | unsigned int nr_pages = 1; | |
5327 | int ret = 0; | |
5328 | ||
5329 | if (mem_cgroup_disabled()) | |
5330 | goto out; | |
5331 | ||
5332 | if (PageSwapCache(page)) { | |
00501b53 JW |
5333 | /* |
5334 | * Every swap fault against a single page tries to charge the | |
5335 | * page, bail as early as possible. shmem_unuse() encounters | |
5336 | * already charged pages, too. The USED bit is protected by | |
5337 | * the page lock, which serializes swap cache removal, which | |
5338 | * in turn serializes uncharging. | |
5339 | */ | |
e993d905 | 5340 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1306a85a | 5341 | if (page->mem_cgroup) |
00501b53 | 5342 | goto out; |
e993d905 VD |
5343 | |
5344 | if (do_swap_account) { | |
5345 | swp_entry_t ent = { .val = page_private(page), }; | |
5346 | unsigned short id = lookup_swap_cgroup_id(ent); | |
5347 | ||
5348 | rcu_read_lock(); | |
5349 | memcg = mem_cgroup_from_id(id); | |
5350 | if (memcg && !css_tryget_online(&memcg->css)) | |
5351 | memcg = NULL; | |
5352 | rcu_read_unlock(); | |
5353 | } | |
00501b53 JW |
5354 | } |
5355 | ||
5356 | if (PageTransHuge(page)) { | |
5357 | nr_pages <<= compound_order(page); | |
5358 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5359 | } | |
5360 | ||
00501b53 JW |
5361 | if (!memcg) |
5362 | memcg = get_mem_cgroup_from_mm(mm); | |
5363 | ||
5364 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
5365 | ||
5366 | css_put(&memcg->css); | |
00501b53 JW |
5367 | out: |
5368 | *memcgp = memcg; | |
5369 | return ret; | |
5370 | } | |
5371 | ||
5372 | /** | |
5373 | * mem_cgroup_commit_charge - commit a page charge | |
5374 | * @page: page to charge | |
5375 | * @memcg: memcg to charge the page to | |
5376 | * @lrucare: page might be on LRU already | |
5377 | * | |
5378 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
5379 | * after page->mapping has been set up. This must happen atomically | |
5380 | * as part of the page instantiation, i.e. under the page table lock | |
5381 | * for anonymous pages, under the page lock for page and swap cache. | |
5382 | * | |
5383 | * In addition, the page must not be on the LRU during the commit, to | |
5384 | * prevent racing with task migration. If it might be, use @lrucare. | |
5385 | * | |
5386 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
5387 | */ | |
5388 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
5389 | bool lrucare) | |
5390 | { | |
5391 | unsigned int nr_pages = 1; | |
5392 | ||
5393 | VM_BUG_ON_PAGE(!page->mapping, page); | |
5394 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
5395 | ||
5396 | if (mem_cgroup_disabled()) | |
5397 | return; | |
5398 | /* | |
5399 | * Swap faults will attempt to charge the same page multiple | |
5400 | * times. But reuse_swap_page() might have removed the page | |
5401 | * from swapcache already, so we can't check PageSwapCache(). | |
5402 | */ | |
5403 | if (!memcg) | |
5404 | return; | |
5405 | ||
6abb5a86 JW |
5406 | commit_charge(page, memcg, lrucare); |
5407 | ||
00501b53 JW |
5408 | if (PageTransHuge(page)) { |
5409 | nr_pages <<= compound_order(page); | |
5410 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5411 | } | |
5412 | ||
6abb5a86 JW |
5413 | local_irq_disable(); |
5414 | mem_cgroup_charge_statistics(memcg, page, nr_pages); | |
5415 | memcg_check_events(memcg, page); | |
5416 | local_irq_enable(); | |
00501b53 JW |
5417 | |
5418 | if (do_swap_account && PageSwapCache(page)) { | |
5419 | swp_entry_t entry = { .val = page_private(page) }; | |
5420 | /* | |
5421 | * The swap entry might not get freed for a long time, | |
5422 | * let's not wait for it. The page already received a | |
5423 | * memory+swap charge, drop the swap entry duplicate. | |
5424 | */ | |
5425 | mem_cgroup_uncharge_swap(entry); | |
5426 | } | |
5427 | } | |
5428 | ||
5429 | /** | |
5430 | * mem_cgroup_cancel_charge - cancel a page charge | |
5431 | * @page: page to charge | |
5432 | * @memcg: memcg to charge the page to | |
5433 | * | |
5434 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
5435 | */ | |
5436 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg) | |
5437 | { | |
5438 | unsigned int nr_pages = 1; | |
5439 | ||
5440 | if (mem_cgroup_disabled()) | |
5441 | return; | |
5442 | /* | |
5443 | * Swap faults will attempt to charge the same page multiple | |
5444 | * times. But reuse_swap_page() might have removed the page | |
5445 | * from swapcache already, so we can't check PageSwapCache(). | |
5446 | */ | |
5447 | if (!memcg) | |
5448 | return; | |
5449 | ||
5450 | if (PageTransHuge(page)) { | |
5451 | nr_pages <<= compound_order(page); | |
5452 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5453 | } | |
5454 | ||
5455 | cancel_charge(memcg, nr_pages); | |
5456 | } | |
5457 | ||
747db954 | 5458 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
747db954 JW |
5459 | unsigned long nr_anon, unsigned long nr_file, |
5460 | unsigned long nr_huge, struct page *dummy_page) | |
5461 | { | |
18eca2e6 | 5462 | unsigned long nr_pages = nr_anon + nr_file; |
747db954 JW |
5463 | unsigned long flags; |
5464 | ||
ce00a967 | 5465 | if (!mem_cgroup_is_root(memcg)) { |
18eca2e6 JW |
5466 | page_counter_uncharge(&memcg->memory, nr_pages); |
5467 | if (do_swap_account) | |
5468 | page_counter_uncharge(&memcg->memsw, nr_pages); | |
ce00a967 JW |
5469 | memcg_oom_recover(memcg); |
5470 | } | |
747db954 JW |
5471 | |
5472 | local_irq_save(flags); | |
5473 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); | |
5474 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); | |
5475 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); | |
5476 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); | |
18eca2e6 | 5477 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
747db954 JW |
5478 | memcg_check_events(memcg, dummy_page); |
5479 | local_irq_restore(flags); | |
e8ea14cc JW |
5480 | |
5481 | if (!mem_cgroup_is_root(memcg)) | |
18eca2e6 | 5482 | css_put_many(&memcg->css, nr_pages); |
747db954 JW |
5483 | } |
5484 | ||
5485 | static void uncharge_list(struct list_head *page_list) | |
5486 | { | |
5487 | struct mem_cgroup *memcg = NULL; | |
747db954 JW |
5488 | unsigned long nr_anon = 0; |
5489 | unsigned long nr_file = 0; | |
5490 | unsigned long nr_huge = 0; | |
5491 | unsigned long pgpgout = 0; | |
747db954 JW |
5492 | struct list_head *next; |
5493 | struct page *page; | |
5494 | ||
5495 | next = page_list->next; | |
5496 | do { | |
5497 | unsigned int nr_pages = 1; | |
747db954 JW |
5498 | |
5499 | page = list_entry(next, struct page, lru); | |
5500 | next = page->lru.next; | |
5501 | ||
5502 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5503 | VM_BUG_ON_PAGE(page_count(page), page); | |
5504 | ||
1306a85a | 5505 | if (!page->mem_cgroup) |
747db954 JW |
5506 | continue; |
5507 | ||
5508 | /* | |
5509 | * Nobody should be changing or seriously looking at | |
1306a85a | 5510 | * page->mem_cgroup at this point, we have fully |
29833315 | 5511 | * exclusive access to the page. |
747db954 JW |
5512 | */ |
5513 | ||
1306a85a | 5514 | if (memcg != page->mem_cgroup) { |
747db954 | 5515 | if (memcg) { |
18eca2e6 JW |
5516 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5517 | nr_huge, page); | |
5518 | pgpgout = nr_anon = nr_file = nr_huge = 0; | |
747db954 | 5519 | } |
1306a85a | 5520 | memcg = page->mem_cgroup; |
747db954 JW |
5521 | } |
5522 | ||
5523 | if (PageTransHuge(page)) { | |
5524 | nr_pages <<= compound_order(page); | |
5525 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5526 | nr_huge += nr_pages; | |
5527 | } | |
5528 | ||
5529 | if (PageAnon(page)) | |
5530 | nr_anon += nr_pages; | |
5531 | else | |
5532 | nr_file += nr_pages; | |
5533 | ||
1306a85a | 5534 | page->mem_cgroup = NULL; |
747db954 JW |
5535 | |
5536 | pgpgout++; | |
5537 | } while (next != page_list); | |
5538 | ||
5539 | if (memcg) | |
18eca2e6 JW |
5540 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5541 | nr_huge, page); | |
747db954 JW |
5542 | } |
5543 | ||
0a31bc97 JW |
5544 | /** |
5545 | * mem_cgroup_uncharge - uncharge a page | |
5546 | * @page: page to uncharge | |
5547 | * | |
5548 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
5549 | * mem_cgroup_commit_charge(). | |
5550 | */ | |
5551 | void mem_cgroup_uncharge(struct page *page) | |
5552 | { | |
0a31bc97 JW |
5553 | if (mem_cgroup_disabled()) |
5554 | return; | |
5555 | ||
747db954 | 5556 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 5557 | if (!page->mem_cgroup) |
0a31bc97 JW |
5558 | return; |
5559 | ||
747db954 JW |
5560 | INIT_LIST_HEAD(&page->lru); |
5561 | uncharge_list(&page->lru); | |
5562 | } | |
0a31bc97 | 5563 | |
747db954 JW |
5564 | /** |
5565 | * mem_cgroup_uncharge_list - uncharge a list of page | |
5566 | * @page_list: list of pages to uncharge | |
5567 | * | |
5568 | * Uncharge a list of pages previously charged with | |
5569 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
5570 | */ | |
5571 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
5572 | { | |
5573 | if (mem_cgroup_disabled()) | |
5574 | return; | |
0a31bc97 | 5575 | |
747db954 JW |
5576 | if (!list_empty(page_list)) |
5577 | uncharge_list(page_list); | |
0a31bc97 JW |
5578 | } |
5579 | ||
5580 | /** | |
45637bab | 5581 | * mem_cgroup_replace_page - migrate a charge to another page |
0a31bc97 JW |
5582 | * @oldpage: currently charged page |
5583 | * @newpage: page to transfer the charge to | |
0a31bc97 JW |
5584 | * |
5585 | * Migrate the charge from @oldpage to @newpage. | |
5586 | * | |
5587 | * Both pages must be locked, @newpage->mapping must be set up. | |
25be6a65 | 5588 | * Either or both pages might be on the LRU already. |
0a31bc97 | 5589 | */ |
45637bab | 5590 | void mem_cgroup_replace_page(struct page *oldpage, struct page *newpage) |
0a31bc97 | 5591 | { |
29833315 | 5592 | struct mem_cgroup *memcg; |
0a31bc97 JW |
5593 | int isolated; |
5594 | ||
5595 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
5596 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 5597 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
5598 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
5599 | newpage); | |
0a31bc97 JW |
5600 | |
5601 | if (mem_cgroup_disabled()) | |
5602 | return; | |
5603 | ||
5604 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 5605 | if (newpage->mem_cgroup) |
0a31bc97 JW |
5606 | return; |
5607 | ||
45637bab | 5608 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 5609 | memcg = oldpage->mem_cgroup; |
29833315 | 5610 | if (!memcg) |
0a31bc97 JW |
5611 | return; |
5612 | ||
45637bab | 5613 | lock_page_lru(oldpage, &isolated); |
1306a85a | 5614 | oldpage->mem_cgroup = NULL; |
45637bab | 5615 | unlock_page_lru(oldpage, isolated); |
0a31bc97 | 5616 | |
45637bab | 5617 | commit_charge(newpage, memcg, true); |
0a31bc97 JW |
5618 | } |
5619 | ||
2d11085e | 5620 | /* |
1081312f MH |
5621 | * subsys_initcall() for memory controller. |
5622 | * | |
5623 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
5624 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
5625 | * everything that doesn't depend on a specific mem_cgroup structure should | |
5626 | * be initialized from here. | |
2d11085e MH |
5627 | */ |
5628 | static int __init mem_cgroup_init(void) | |
5629 | { | |
95a045f6 JW |
5630 | int cpu, node; |
5631 | ||
2d11085e | 5632 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
95a045f6 JW |
5633 | |
5634 | for_each_possible_cpu(cpu) | |
5635 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
5636 | drain_local_stock); | |
5637 | ||
5638 | for_each_node(node) { | |
5639 | struct mem_cgroup_tree_per_node *rtpn; | |
5640 | int zone; | |
5641 | ||
5642 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
5643 | node_online(node) ? node : NUMA_NO_NODE); | |
5644 | ||
5645 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
5646 | struct mem_cgroup_tree_per_zone *rtpz; | |
5647 | ||
5648 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
5649 | rtpz->rb_root = RB_ROOT; | |
5650 | spin_lock_init(&rtpz->lock); | |
5651 | } | |
5652 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
5653 | } | |
5654 | ||
2d11085e MH |
5655 | return 0; |
5656 | } | |
5657 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
5658 | |
5659 | #ifdef CONFIG_MEMCG_SWAP | |
5660 | /** | |
5661 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
5662 | * @page: page whose memsw charge to transfer | |
5663 | * @entry: swap entry to move the charge to | |
5664 | * | |
5665 | * Transfer the memsw charge of @page to @entry. | |
5666 | */ | |
5667 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
5668 | { | |
5669 | struct mem_cgroup *memcg; | |
5670 | unsigned short oldid; | |
5671 | ||
5672 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5673 | VM_BUG_ON_PAGE(page_count(page), page); | |
5674 | ||
5675 | if (!do_swap_account) | |
5676 | return; | |
5677 | ||
5678 | memcg = page->mem_cgroup; | |
5679 | ||
5680 | /* Readahead page, never charged */ | |
5681 | if (!memcg) | |
5682 | return; | |
5683 | ||
5684 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); | |
5685 | VM_BUG_ON_PAGE(oldid, page); | |
5686 | mem_cgroup_swap_statistics(memcg, true); | |
5687 | ||
5688 | page->mem_cgroup = NULL; | |
5689 | ||
5690 | if (!mem_cgroup_is_root(memcg)) | |
5691 | page_counter_uncharge(&memcg->memory, 1); | |
5692 | ||
ce9ce665 SAS |
5693 | /* |
5694 | * Interrupts should be disabled here because the caller holds the | |
5695 | * mapping->tree_lock lock which is taken with interrupts-off. It is | |
5696 | * important here to have the interrupts disabled because it is the | |
5697 | * only synchronisation we have for udpating the per-CPU variables. | |
5698 | */ | |
5699 | VM_BUG_ON(!irqs_disabled()); | |
21afa38e JW |
5700 | mem_cgroup_charge_statistics(memcg, page, -1); |
5701 | memcg_check_events(memcg, page); | |
5702 | } | |
5703 | ||
5704 | /** | |
5705 | * mem_cgroup_uncharge_swap - uncharge a swap entry | |
5706 | * @entry: swap entry to uncharge | |
5707 | * | |
5708 | * Drop the memsw charge associated with @entry. | |
5709 | */ | |
5710 | void mem_cgroup_uncharge_swap(swp_entry_t entry) | |
5711 | { | |
5712 | struct mem_cgroup *memcg; | |
5713 | unsigned short id; | |
5714 | ||
5715 | if (!do_swap_account) | |
5716 | return; | |
5717 | ||
5718 | id = swap_cgroup_record(entry, 0); | |
5719 | rcu_read_lock(); | |
adbe427b | 5720 | memcg = mem_cgroup_from_id(id); |
21afa38e JW |
5721 | if (memcg) { |
5722 | if (!mem_cgroup_is_root(memcg)) | |
5723 | page_counter_uncharge(&memcg->memsw, 1); | |
5724 | mem_cgroup_swap_statistics(memcg, false); | |
5725 | css_put(&memcg->css); | |
5726 | } | |
5727 | rcu_read_unlock(); | |
5728 | } | |
5729 | ||
5730 | /* for remember boot option*/ | |
5731 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
5732 | static int really_do_swap_account __initdata = 1; | |
5733 | #else | |
5734 | static int really_do_swap_account __initdata; | |
5735 | #endif | |
5736 | ||
5737 | static int __init enable_swap_account(char *s) | |
5738 | { | |
5739 | if (!strcmp(s, "1")) | |
5740 | really_do_swap_account = 1; | |
5741 | else if (!strcmp(s, "0")) | |
5742 | really_do_swap_account = 0; | |
5743 | return 1; | |
5744 | } | |
5745 | __setup("swapaccount=", enable_swap_account); | |
5746 | ||
5747 | static struct cftype memsw_cgroup_files[] = { | |
5748 | { | |
5749 | .name = "memsw.usage_in_bytes", | |
5750 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
5751 | .read_u64 = mem_cgroup_read_u64, | |
5752 | }, | |
5753 | { | |
5754 | .name = "memsw.max_usage_in_bytes", | |
5755 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
5756 | .write = mem_cgroup_reset, | |
5757 | .read_u64 = mem_cgroup_read_u64, | |
5758 | }, | |
5759 | { | |
5760 | .name = "memsw.limit_in_bytes", | |
5761 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
5762 | .write = mem_cgroup_write, | |
5763 | .read_u64 = mem_cgroup_read_u64, | |
5764 | }, | |
5765 | { | |
5766 | .name = "memsw.failcnt", | |
5767 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
5768 | .write = mem_cgroup_reset, | |
5769 | .read_u64 = mem_cgroup_read_u64, | |
5770 | }, | |
5771 | { }, /* terminate */ | |
5772 | }; | |
5773 | ||
5774 | static int __init mem_cgroup_swap_init(void) | |
5775 | { | |
5776 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
5777 | do_swap_account = 1; | |
5778 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, | |
5779 | memsw_cgroup_files)); | |
5780 | } | |
5781 | return 0; | |
5782 | } | |
5783 | subsys_initcall(mem_cgroup_swap_init); | |
5784 | ||
5785 | #endif /* CONFIG_MEMCG_SWAP */ |