Commit | Line | Data |
---|---|---|
8cdea7c0 BS |
1 | /* memcontrol.c - Memory Controller |
2 | * | |
3 | * Copyright IBM Corporation, 2007 | |
4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
5 | * | |
78fb7466 PE |
6 | * Copyright 2007 OpenVZ SWsoft Inc |
7 | * Author: Pavel Emelianov <xemul@openvz.org> | |
8 | * | |
2e72b634 KS |
9 | * Memory thresholds |
10 | * Copyright (C) 2009 Nokia Corporation | |
11 | * Author: Kirill A. Shutemov | |
12 | * | |
7ae1e1d0 GC |
13 | * Kernel Memory Controller |
14 | * Copyright (C) 2012 Parallels Inc. and Google Inc. | |
15 | * Authors: Glauber Costa and Suleiman Souhlal | |
16 | * | |
1575e68b JW |
17 | * Native page reclaim |
18 | * Charge lifetime sanitation | |
19 | * Lockless page tracking & accounting | |
20 | * Unified hierarchy configuration model | |
21 | * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner | |
22 | * | |
8cdea7c0 BS |
23 | * This program is free software; you can redistribute it and/or modify |
24 | * it under the terms of the GNU General Public License as published by | |
25 | * the Free Software Foundation; either version 2 of the License, or | |
26 | * (at your option) any later version. | |
27 | * | |
28 | * This program is distributed in the hope that it will be useful, | |
29 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
30 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
31 | * GNU General Public License for more details. | |
32 | */ | |
33 | ||
3e32cb2e | 34 | #include <linux/page_counter.h> |
8cdea7c0 BS |
35 | #include <linux/memcontrol.h> |
36 | #include <linux/cgroup.h> | |
78fb7466 | 37 | #include <linux/mm.h> |
4ffef5fe | 38 | #include <linux/hugetlb.h> |
d13d1443 | 39 | #include <linux/pagemap.h> |
d52aa412 | 40 | #include <linux/smp.h> |
8a9f3ccd | 41 | #include <linux/page-flags.h> |
66e1707b | 42 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
43 | #include <linux/bit_spinlock.h> |
44 | #include <linux/rcupdate.h> | |
e222432b | 45 | #include <linux/limits.h> |
b9e15baf | 46 | #include <linux/export.h> |
8c7c6e34 | 47 | #include <linux/mutex.h> |
bb4cc1a8 | 48 | #include <linux/rbtree.h> |
b6ac57d5 | 49 | #include <linux/slab.h> |
66e1707b | 50 | #include <linux/swap.h> |
02491447 | 51 | #include <linux/swapops.h> |
66e1707b | 52 | #include <linux/spinlock.h> |
2e72b634 | 53 | #include <linux/eventfd.h> |
79bd9814 | 54 | #include <linux/poll.h> |
2e72b634 | 55 | #include <linux/sort.h> |
66e1707b | 56 | #include <linux/fs.h> |
d2ceb9b7 | 57 | #include <linux/seq_file.h> |
70ddf637 | 58 | #include <linux/vmpressure.h> |
b69408e8 | 59 | #include <linux/mm_inline.h> |
5d1ea48b | 60 | #include <linux/swap_cgroup.h> |
cdec2e42 | 61 | #include <linux/cpu.h> |
158e0a2d | 62 | #include <linux/oom.h> |
0056f4e6 | 63 | #include <linux/lockdep.h> |
79bd9814 | 64 | #include <linux/file.h> |
b23afb93 | 65 | #include <linux/tracehook.h> |
08e552c6 | 66 | #include "internal.h" |
d1a4c0b3 | 67 | #include <net/sock.h> |
4bd2c1ee | 68 | #include <net/ip.h> |
d1a4c0b3 | 69 | #include <net/tcp_memcontrol.h> |
f35c3a8e | 70 | #include "slab.h" |
8cdea7c0 | 71 | |
8697d331 BS |
72 | #include <asm/uaccess.h> |
73 | ||
cc8e970c KM |
74 | #include <trace/events/vmscan.h> |
75 | ||
073219e9 TH |
76 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
77 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 78 | |
7d828602 JW |
79 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
80 | ||
a181b0e8 | 81 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
8cdea7c0 | 82 | |
21afa38e | 83 | /* Whether the swap controller is active */ |
c255a458 | 84 | #ifdef CONFIG_MEMCG_SWAP |
c077719b | 85 | int do_swap_account __read_mostly; |
c077719b | 86 | #else |
a0db00fc | 87 | #define do_swap_account 0 |
c077719b KH |
88 | #endif |
89 | ||
7941d214 JW |
90 | /* Whether legacy memory+swap accounting is active */ |
91 | static bool do_memsw_account(void) | |
92 | { | |
93 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account; | |
94 | } | |
95 | ||
af7c4b0e JW |
96 | static const char * const mem_cgroup_stat_names[] = { |
97 | "cache", | |
98 | "rss", | |
b070e65c | 99 | "rss_huge", |
af7c4b0e | 100 | "mapped_file", |
c4843a75 | 101 | "dirty", |
3ea67d06 | 102 | "writeback", |
af7c4b0e JW |
103 | "swap", |
104 | }; | |
105 | ||
af7c4b0e JW |
106 | static const char * const mem_cgroup_events_names[] = { |
107 | "pgpgin", | |
108 | "pgpgout", | |
109 | "pgfault", | |
110 | "pgmajfault", | |
111 | }; | |
112 | ||
58cf188e SZ |
113 | static const char * const mem_cgroup_lru_names[] = { |
114 | "inactive_anon", | |
115 | "active_anon", | |
116 | "inactive_file", | |
117 | "active_file", | |
118 | "unevictable", | |
119 | }; | |
120 | ||
a0db00fc KS |
121 | #define THRESHOLDS_EVENTS_TARGET 128 |
122 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
123 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 124 | |
bb4cc1a8 AM |
125 | /* |
126 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
127 | * their hierarchy representation | |
128 | */ | |
129 | ||
130 | struct mem_cgroup_tree_per_zone { | |
131 | struct rb_root rb_root; | |
132 | spinlock_t lock; | |
133 | }; | |
134 | ||
135 | struct mem_cgroup_tree_per_node { | |
136 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
137 | }; | |
138 | ||
139 | struct mem_cgroup_tree { | |
140 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
141 | }; | |
142 | ||
143 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
144 | ||
9490ff27 KH |
145 | /* for OOM */ |
146 | struct mem_cgroup_eventfd_list { | |
147 | struct list_head list; | |
148 | struct eventfd_ctx *eventfd; | |
149 | }; | |
2e72b634 | 150 | |
79bd9814 TH |
151 | /* |
152 | * cgroup_event represents events which userspace want to receive. | |
153 | */ | |
3bc942f3 | 154 | struct mem_cgroup_event { |
79bd9814 | 155 | /* |
59b6f873 | 156 | * memcg which the event belongs to. |
79bd9814 | 157 | */ |
59b6f873 | 158 | struct mem_cgroup *memcg; |
79bd9814 TH |
159 | /* |
160 | * eventfd to signal userspace about the event. | |
161 | */ | |
162 | struct eventfd_ctx *eventfd; | |
163 | /* | |
164 | * Each of these stored in a list by the cgroup. | |
165 | */ | |
166 | struct list_head list; | |
fba94807 TH |
167 | /* |
168 | * register_event() callback will be used to add new userspace | |
169 | * waiter for changes related to this event. Use eventfd_signal() | |
170 | * on eventfd to send notification to userspace. | |
171 | */ | |
59b6f873 | 172 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 173 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
174 | /* |
175 | * unregister_event() callback will be called when userspace closes | |
176 | * the eventfd or on cgroup removing. This callback must be set, | |
177 | * if you want provide notification functionality. | |
178 | */ | |
59b6f873 | 179 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 180 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
181 | /* |
182 | * All fields below needed to unregister event when | |
183 | * userspace closes eventfd. | |
184 | */ | |
185 | poll_table pt; | |
186 | wait_queue_head_t *wqh; | |
187 | wait_queue_t wait; | |
188 | struct work_struct remove; | |
189 | }; | |
190 | ||
c0ff4b85 R |
191 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
192 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 193 | |
7dc74be0 DN |
194 | /* Stuffs for move charges at task migration. */ |
195 | /* | |
1dfab5ab | 196 | * Types of charges to be moved. |
7dc74be0 | 197 | */ |
1dfab5ab JW |
198 | #define MOVE_ANON 0x1U |
199 | #define MOVE_FILE 0x2U | |
200 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 201 | |
4ffef5fe DN |
202 | /* "mc" and its members are protected by cgroup_mutex */ |
203 | static struct move_charge_struct { | |
b1dd693e | 204 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
205 | struct mem_cgroup *from; |
206 | struct mem_cgroup *to; | |
1dfab5ab | 207 | unsigned long flags; |
4ffef5fe | 208 | unsigned long precharge; |
854ffa8d | 209 | unsigned long moved_charge; |
483c30b5 | 210 | unsigned long moved_swap; |
8033b97c DN |
211 | struct task_struct *moving_task; /* a task moving charges */ |
212 | wait_queue_head_t waitq; /* a waitq for other context */ | |
213 | } mc = { | |
2bd9bb20 | 214 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
215 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
216 | }; | |
4ffef5fe | 217 | |
4e416953 BS |
218 | /* |
219 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
220 | * limit reclaim to prevent infinite loops, if they ever occur. | |
221 | */ | |
a0db00fc | 222 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 223 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 224 | |
217bc319 KH |
225 | enum charge_type { |
226 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 227 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 228 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 229 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
230 | NR_CHARGE_TYPE, |
231 | }; | |
232 | ||
8c7c6e34 | 233 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
234 | enum res_type { |
235 | _MEM, | |
236 | _MEMSWAP, | |
237 | _OOM_TYPE, | |
510fc4e1 | 238 | _KMEM, |
86ae53e1 GC |
239 | }; |
240 | ||
a0db00fc KS |
241 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
242 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 243 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
244 | /* Used for OOM nofiier */ |
245 | #define OOM_CONTROL (0) | |
8c7c6e34 | 246 | |
0999821b GC |
247 | /* |
248 | * The memcg_create_mutex will be held whenever a new cgroup is created. | |
249 | * As a consequence, any change that needs to protect against new child cgroups | |
250 | * appearing has to hold it as well. | |
251 | */ | |
252 | static DEFINE_MUTEX(memcg_create_mutex); | |
253 | ||
70ddf637 AV |
254 | /* Some nice accessors for the vmpressure. */ |
255 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
256 | { | |
257 | if (!memcg) | |
258 | memcg = root_mem_cgroup; | |
259 | return &memcg->vmpressure; | |
260 | } | |
261 | ||
262 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
263 | { | |
264 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
265 | } | |
266 | ||
7ffc0edc MH |
267 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
268 | { | |
269 | return (memcg == root_mem_cgroup); | |
270 | } | |
271 | ||
4219b2da LZ |
272 | /* |
273 | * We restrict the id in the range of [1, 65535], so it can fit into | |
274 | * an unsigned short. | |
275 | */ | |
276 | #define MEM_CGROUP_ID_MAX USHRT_MAX | |
277 | ||
34c00c31 LZ |
278 | static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) |
279 | { | |
15a4c835 | 280 | return memcg->css.id; |
34c00c31 LZ |
281 | } |
282 | ||
adbe427b VD |
283 | /* |
284 | * A helper function to get mem_cgroup from ID. must be called under | |
285 | * rcu_read_lock(). The caller is responsible for calling | |
286 | * css_tryget_online() if the mem_cgroup is used for charging. (dropping | |
287 | * refcnt from swap can be called against removed memcg.) | |
288 | */ | |
34c00c31 LZ |
289 | static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) |
290 | { | |
291 | struct cgroup_subsys_state *css; | |
292 | ||
7d699ddb | 293 | css = css_from_id(id, &memory_cgrp_subsys); |
34c00c31 LZ |
294 | return mem_cgroup_from_css(css); |
295 | } | |
296 | ||
a8964b9b | 297 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 | 298 | /* |
f7ce3190 | 299 | * This will be the memcg's index in each cache's ->memcg_params.memcg_caches. |
b8627835 LZ |
300 | * The main reason for not using cgroup id for this: |
301 | * this works better in sparse environments, where we have a lot of memcgs, | |
302 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
303 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
304 | * 200 entry array for that. | |
55007d84 | 305 | * |
dbcf73e2 VD |
306 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
307 | * will double each time we have to increase it. | |
55007d84 | 308 | */ |
dbcf73e2 VD |
309 | static DEFINE_IDA(memcg_cache_ida); |
310 | int memcg_nr_cache_ids; | |
749c5415 | 311 | |
05257a1a VD |
312 | /* Protects memcg_nr_cache_ids */ |
313 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
314 | ||
315 | void memcg_get_cache_ids(void) | |
316 | { | |
317 | down_read(&memcg_cache_ids_sem); | |
318 | } | |
319 | ||
320 | void memcg_put_cache_ids(void) | |
321 | { | |
322 | up_read(&memcg_cache_ids_sem); | |
323 | } | |
324 | ||
55007d84 GC |
325 | /* |
326 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
327 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
328 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
329 | * tunable, but that is strictly not necessary. | |
330 | * | |
b8627835 | 331 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
332 | * this constant directly from cgroup, but it is understandable that this is |
333 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 334 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
335 | * increase ours as well if it increases. |
336 | */ | |
337 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 338 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 339 | |
d7f25f8a GC |
340 | /* |
341 | * A lot of the calls to the cache allocation functions are expected to be | |
342 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
343 | * conditional to this static branch, we'll have to allow modules that does | |
344 | * kmem_cache_alloc and the such to see this symbol as well | |
345 | */ | |
a8964b9b | 346 | struct static_key memcg_kmem_enabled_key; |
d7f25f8a | 347 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 348 | |
a8964b9b GC |
349 | #endif /* CONFIG_MEMCG_KMEM */ |
350 | ||
f64c3f54 | 351 | static struct mem_cgroup_per_zone * |
e231875b | 352 | mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone) |
f64c3f54 | 353 | { |
e231875b JZ |
354 | int nid = zone_to_nid(zone); |
355 | int zid = zone_idx(zone); | |
356 | ||
54f72fe0 | 357 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
358 | } |
359 | ||
ad7fa852 TH |
360 | /** |
361 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
362 | * @page: page of interest | |
363 | * | |
364 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
365 | * with @page is returned. The returned css remains associated with @page | |
366 | * until it is released. | |
367 | * | |
368 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
369 | * is returned. | |
370 | * | |
371 | * XXX: The above description of behavior on the default hierarchy isn't | |
372 | * strictly true yet as replace_page_cache_page() can modify the | |
373 | * association before @page is released even on the default hierarchy; | |
374 | * however, the current and planned usages don't mix the the two functions | |
375 | * and replace_page_cache_page() will soon be updated to make the invariant | |
376 | * actually true. | |
377 | */ | |
378 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
379 | { | |
380 | struct mem_cgroup *memcg; | |
381 | ||
382 | rcu_read_lock(); | |
383 | ||
384 | memcg = page->mem_cgroup; | |
385 | ||
9e10a130 | 386 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
387 | memcg = root_mem_cgroup; |
388 | ||
389 | rcu_read_unlock(); | |
390 | return &memcg->css; | |
391 | } | |
392 | ||
2fc04524 VD |
393 | /** |
394 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
395 | * @page: the page | |
396 | * | |
397 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
398 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
399 | * is safe to call this function without holding a reference to @page. | |
400 | * | |
401 | * Note, this function is inherently racy, because there is nothing to prevent | |
402 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
403 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
404 | * do not care (such as procfs interfaces). | |
405 | */ | |
406 | ino_t page_cgroup_ino(struct page *page) | |
407 | { | |
408 | struct mem_cgroup *memcg; | |
409 | unsigned long ino = 0; | |
410 | ||
411 | rcu_read_lock(); | |
412 | memcg = READ_ONCE(page->mem_cgroup); | |
413 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) | |
414 | memcg = parent_mem_cgroup(memcg); | |
415 | if (memcg) | |
416 | ino = cgroup_ino(memcg->css.cgroup); | |
417 | rcu_read_unlock(); | |
418 | return ino; | |
419 | } | |
420 | ||
f64c3f54 | 421 | static struct mem_cgroup_per_zone * |
e231875b | 422 | mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 423 | { |
97a6c37b JW |
424 | int nid = page_to_nid(page); |
425 | int zid = page_zonenum(page); | |
f64c3f54 | 426 | |
e231875b | 427 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
428 | } |
429 | ||
bb4cc1a8 AM |
430 | static struct mem_cgroup_tree_per_zone * |
431 | soft_limit_tree_node_zone(int nid, int zid) | |
432 | { | |
433 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
434 | } | |
435 | ||
436 | static struct mem_cgroup_tree_per_zone * | |
437 | soft_limit_tree_from_page(struct page *page) | |
438 | { | |
439 | int nid = page_to_nid(page); | |
440 | int zid = page_zonenum(page); | |
441 | ||
442 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
443 | } | |
444 | ||
cf2c8127 JW |
445 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz, |
446 | struct mem_cgroup_tree_per_zone *mctz, | |
3e32cb2e | 447 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
448 | { |
449 | struct rb_node **p = &mctz->rb_root.rb_node; | |
450 | struct rb_node *parent = NULL; | |
451 | struct mem_cgroup_per_zone *mz_node; | |
452 | ||
453 | if (mz->on_tree) | |
454 | return; | |
455 | ||
456 | mz->usage_in_excess = new_usage_in_excess; | |
457 | if (!mz->usage_in_excess) | |
458 | return; | |
459 | while (*p) { | |
460 | parent = *p; | |
461 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
462 | tree_node); | |
463 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
464 | p = &(*p)->rb_left; | |
465 | /* | |
466 | * We can't avoid mem cgroups that are over their soft | |
467 | * limit by the same amount | |
468 | */ | |
469 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
470 | p = &(*p)->rb_right; | |
471 | } | |
472 | rb_link_node(&mz->tree_node, parent, p); | |
473 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
474 | mz->on_tree = true; | |
475 | } | |
476 | ||
cf2c8127 JW |
477 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
478 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 AM |
479 | { |
480 | if (!mz->on_tree) | |
481 | return; | |
482 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
483 | mz->on_tree = false; | |
484 | } | |
485 | ||
cf2c8127 JW |
486 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
487 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 | 488 | { |
0a31bc97 JW |
489 | unsigned long flags; |
490 | ||
491 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 492 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 493 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
494 | } |
495 | ||
3e32cb2e JW |
496 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
497 | { | |
498 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 499 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
500 | unsigned long excess = 0; |
501 | ||
502 | if (nr_pages > soft_limit) | |
503 | excess = nr_pages - soft_limit; | |
504 | ||
505 | return excess; | |
506 | } | |
bb4cc1a8 AM |
507 | |
508 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
509 | { | |
3e32cb2e | 510 | unsigned long excess; |
bb4cc1a8 AM |
511 | struct mem_cgroup_per_zone *mz; |
512 | struct mem_cgroup_tree_per_zone *mctz; | |
bb4cc1a8 | 513 | |
e231875b | 514 | mctz = soft_limit_tree_from_page(page); |
bb4cc1a8 AM |
515 | /* |
516 | * Necessary to update all ancestors when hierarchy is used. | |
517 | * because their event counter is not touched. | |
518 | */ | |
519 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
e231875b | 520 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
3e32cb2e | 521 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
522 | /* |
523 | * We have to update the tree if mz is on RB-tree or | |
524 | * mem is over its softlimit. | |
525 | */ | |
526 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
527 | unsigned long flags; |
528 | ||
529 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
530 | /* if on-tree, remove it */ |
531 | if (mz->on_tree) | |
cf2c8127 | 532 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
533 | /* |
534 | * Insert again. mz->usage_in_excess will be updated. | |
535 | * If excess is 0, no tree ops. | |
536 | */ | |
cf2c8127 | 537 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 538 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
539 | } |
540 | } | |
541 | } | |
542 | ||
543 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
544 | { | |
bb4cc1a8 | 545 | struct mem_cgroup_tree_per_zone *mctz; |
e231875b JZ |
546 | struct mem_cgroup_per_zone *mz; |
547 | int nid, zid; | |
bb4cc1a8 | 548 | |
e231875b JZ |
549 | for_each_node(nid) { |
550 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
551 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
552 | mctz = soft_limit_tree_node_zone(nid, zid); | |
cf2c8127 | 553 | mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
554 | } |
555 | } | |
556 | } | |
557 | ||
558 | static struct mem_cgroup_per_zone * | |
559 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
560 | { | |
561 | struct rb_node *rightmost = NULL; | |
562 | struct mem_cgroup_per_zone *mz; | |
563 | ||
564 | retry: | |
565 | mz = NULL; | |
566 | rightmost = rb_last(&mctz->rb_root); | |
567 | if (!rightmost) | |
568 | goto done; /* Nothing to reclaim from */ | |
569 | ||
570 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
571 | /* | |
572 | * Remove the node now but someone else can add it back, | |
573 | * we will to add it back at the end of reclaim to its correct | |
574 | * position in the tree. | |
575 | */ | |
cf2c8127 | 576 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 577 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 578 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
579 | goto retry; |
580 | done: | |
581 | return mz; | |
582 | } | |
583 | ||
584 | static struct mem_cgroup_per_zone * | |
585 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
586 | { | |
587 | struct mem_cgroup_per_zone *mz; | |
588 | ||
0a31bc97 | 589 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 590 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 591 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
592 | return mz; |
593 | } | |
594 | ||
711d3d2c | 595 | /* |
484ebb3b GT |
596 | * Return page count for single (non recursive) @memcg. |
597 | * | |
711d3d2c KH |
598 | * Implementation Note: reading percpu statistics for memcg. |
599 | * | |
600 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
601 | * synchronization to implement "quick" read. There are trade-off between | |
602 | * reading cost and precision of value. Then, we may have a chance to implement | |
484ebb3b | 603 | * a periodic synchronization of counter in memcg's counter. |
711d3d2c KH |
604 | * |
605 | * But this _read() function is used for user interface now. The user accounts | |
606 | * memory usage by memory cgroup and he _always_ requires exact value because | |
607 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
608 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
609 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
610 | * | |
611 | * If there are kernel internal actions which can make use of some not-exact | |
612 | * value, and reading all cpu value can be performance bottleneck in some | |
484ebb3b | 613 | * common workload, threshold and synchronization as vmstat[] should be |
711d3d2c KH |
614 | * implemented. |
615 | */ | |
484ebb3b GT |
616 | static unsigned long |
617 | mem_cgroup_read_stat(struct mem_cgroup *memcg, enum mem_cgroup_stat_index idx) | |
c62b1a3b | 618 | { |
7a159cc9 | 619 | long val = 0; |
c62b1a3b | 620 | int cpu; |
c62b1a3b | 621 | |
484ebb3b | 622 | /* Per-cpu values can be negative, use a signed accumulator */ |
733a572e | 623 | for_each_possible_cpu(cpu) |
c0ff4b85 | 624 | val += per_cpu(memcg->stat->count[idx], cpu); |
484ebb3b GT |
625 | /* |
626 | * Summing races with updates, so val may be negative. Avoid exposing | |
627 | * transient negative values. | |
628 | */ | |
629 | if (val < 0) | |
630 | val = 0; | |
c62b1a3b KH |
631 | return val; |
632 | } | |
633 | ||
c0ff4b85 | 634 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
635 | enum mem_cgroup_events_index idx) |
636 | { | |
637 | unsigned long val = 0; | |
638 | int cpu; | |
639 | ||
733a572e | 640 | for_each_possible_cpu(cpu) |
c0ff4b85 | 641 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f JW |
642 | return val; |
643 | } | |
644 | ||
c0ff4b85 | 645 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 646 | struct page *page, |
0a31bc97 | 647 | int nr_pages) |
d52aa412 | 648 | { |
b2402857 KH |
649 | /* |
650 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
651 | * counted as CACHE even if it's on ANON LRU. | |
652 | */ | |
0a31bc97 | 653 | if (PageAnon(page)) |
b2402857 | 654 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
c0ff4b85 | 655 | nr_pages); |
d52aa412 | 656 | else |
b2402857 | 657 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 658 | nr_pages); |
55e462b0 | 659 | |
b070e65c DR |
660 | if (PageTransHuge(page)) |
661 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], | |
662 | nr_pages); | |
663 | ||
e401f176 KH |
664 | /* pagein of a big page is an event. So, ignore page size */ |
665 | if (nr_pages > 0) | |
c0ff4b85 | 666 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 667 | else { |
c0ff4b85 | 668 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
669 | nr_pages = -nr_pages; /* for event */ |
670 | } | |
e401f176 | 671 | |
13114716 | 672 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
6d12e2d8 KH |
673 | } |
674 | ||
e231875b JZ |
675 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
676 | int nid, | |
677 | unsigned int lru_mask) | |
bb2a0de9 | 678 | { |
e231875b | 679 | unsigned long nr = 0; |
889976db YH |
680 | int zid; |
681 | ||
e231875b | 682 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 683 | |
e231875b JZ |
684 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
685 | struct mem_cgroup_per_zone *mz; | |
686 | enum lru_list lru; | |
687 | ||
688 | for_each_lru(lru) { | |
689 | if (!(BIT(lru) & lru_mask)) | |
690 | continue; | |
691 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
692 | nr += mz->lru_size[lru]; | |
693 | } | |
694 | } | |
695 | return nr; | |
889976db | 696 | } |
bb2a0de9 | 697 | |
c0ff4b85 | 698 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 699 | unsigned int lru_mask) |
6d12e2d8 | 700 | { |
e231875b | 701 | unsigned long nr = 0; |
889976db | 702 | int nid; |
6d12e2d8 | 703 | |
31aaea4a | 704 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
705 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
706 | return nr; | |
d52aa412 KH |
707 | } |
708 | ||
f53d7ce3 JW |
709 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
710 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
711 | { |
712 | unsigned long val, next; | |
713 | ||
13114716 | 714 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 715 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 716 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
717 | if ((long)next - (long)val < 0) { |
718 | switch (target) { | |
719 | case MEM_CGROUP_TARGET_THRESH: | |
720 | next = val + THRESHOLDS_EVENTS_TARGET; | |
721 | break; | |
bb4cc1a8 AM |
722 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
723 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
724 | break; | |
f53d7ce3 JW |
725 | case MEM_CGROUP_TARGET_NUMAINFO: |
726 | next = val + NUMAINFO_EVENTS_TARGET; | |
727 | break; | |
728 | default: | |
729 | break; | |
730 | } | |
731 | __this_cpu_write(memcg->stat->targets[target], next); | |
732 | return true; | |
7a159cc9 | 733 | } |
f53d7ce3 | 734 | return false; |
d2265e6f KH |
735 | } |
736 | ||
737 | /* | |
738 | * Check events in order. | |
739 | * | |
740 | */ | |
c0ff4b85 | 741 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
742 | { |
743 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
744 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
745 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 746 | bool do_softlimit; |
82b3f2a7 | 747 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 748 | |
bb4cc1a8 AM |
749 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
750 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
751 | #if MAX_NUMNODES > 1 |
752 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
753 | MEM_CGROUP_TARGET_NUMAINFO); | |
754 | #endif | |
c0ff4b85 | 755 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
756 | if (unlikely(do_softlimit)) |
757 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 758 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 759 | if (unlikely(do_numainfo)) |
c0ff4b85 | 760 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 761 | #endif |
0a31bc97 | 762 | } |
d2265e6f KH |
763 | } |
764 | ||
cf475ad2 | 765 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 766 | { |
31a78f23 BS |
767 | /* |
768 | * mm_update_next_owner() may clear mm->owner to NULL | |
769 | * if it races with swapoff, page migration, etc. | |
770 | * So this can be called with p == NULL. | |
771 | */ | |
772 | if (unlikely(!p)) | |
773 | return NULL; | |
774 | ||
073219e9 | 775 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 776 | } |
33398cf2 | 777 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 778 | |
df381975 | 779 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 780 | { |
c0ff4b85 | 781 | struct mem_cgroup *memcg = NULL; |
0b7f569e | 782 | |
54595fe2 KH |
783 | rcu_read_lock(); |
784 | do { | |
6f6acb00 MH |
785 | /* |
786 | * Page cache insertions can happen withou an | |
787 | * actual mm context, e.g. during disk probing | |
788 | * on boot, loopback IO, acct() writes etc. | |
789 | */ | |
790 | if (unlikely(!mm)) | |
df381975 | 791 | memcg = root_mem_cgroup; |
6f6acb00 MH |
792 | else { |
793 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
794 | if (unlikely(!memcg)) | |
795 | memcg = root_mem_cgroup; | |
796 | } | |
ec903c0c | 797 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 798 | rcu_read_unlock(); |
c0ff4b85 | 799 | return memcg; |
54595fe2 KH |
800 | } |
801 | ||
5660048c JW |
802 | /** |
803 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
804 | * @root: hierarchy root | |
805 | * @prev: previously returned memcg, NULL on first invocation | |
806 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
807 | * | |
808 | * Returns references to children of the hierarchy below @root, or | |
809 | * @root itself, or %NULL after a full round-trip. | |
810 | * | |
811 | * Caller must pass the return value in @prev on subsequent | |
812 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
813 | * to cancel a hierarchy walk before the round-trip is complete. | |
814 | * | |
815 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
816 | * divide up the memcgs in the hierarchy among all concurrent | |
817 | * reclaimers operating on the same zone and priority. | |
818 | */ | |
694fbc0f | 819 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 820 | struct mem_cgroup *prev, |
694fbc0f | 821 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 822 | { |
33398cf2 | 823 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
5ac8fb31 | 824 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 825 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 826 | struct mem_cgroup *pos = NULL; |
711d3d2c | 827 | |
694fbc0f AM |
828 | if (mem_cgroup_disabled()) |
829 | return NULL; | |
5660048c | 830 | |
9f3a0d09 JW |
831 | if (!root) |
832 | root = root_mem_cgroup; | |
7d74b06f | 833 | |
9f3a0d09 | 834 | if (prev && !reclaim) |
5ac8fb31 | 835 | pos = prev; |
14067bb3 | 836 | |
9f3a0d09 JW |
837 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
838 | if (prev) | |
5ac8fb31 | 839 | goto out; |
694fbc0f | 840 | return root; |
9f3a0d09 | 841 | } |
14067bb3 | 842 | |
542f85f9 | 843 | rcu_read_lock(); |
5f578161 | 844 | |
5ac8fb31 JW |
845 | if (reclaim) { |
846 | struct mem_cgroup_per_zone *mz; | |
847 | ||
848 | mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone); | |
849 | iter = &mz->iter[reclaim->priority]; | |
850 | ||
851 | if (prev && reclaim->generation != iter->generation) | |
852 | goto out_unlock; | |
853 | ||
6df38689 | 854 | while (1) { |
4db0c3c2 | 855 | pos = READ_ONCE(iter->position); |
6df38689 VD |
856 | if (!pos || css_tryget(&pos->css)) |
857 | break; | |
5ac8fb31 | 858 | /* |
6df38689 VD |
859 | * css reference reached zero, so iter->position will |
860 | * be cleared by ->css_released. However, we should not | |
861 | * rely on this happening soon, because ->css_released | |
862 | * is called from a work queue, and by busy-waiting we | |
863 | * might block it. So we clear iter->position right | |
864 | * away. | |
5ac8fb31 | 865 | */ |
6df38689 VD |
866 | (void)cmpxchg(&iter->position, pos, NULL); |
867 | } | |
5ac8fb31 JW |
868 | } |
869 | ||
870 | if (pos) | |
871 | css = &pos->css; | |
872 | ||
873 | for (;;) { | |
874 | css = css_next_descendant_pre(css, &root->css); | |
875 | if (!css) { | |
876 | /* | |
877 | * Reclaimers share the hierarchy walk, and a | |
878 | * new one might jump in right at the end of | |
879 | * the hierarchy - make sure they see at least | |
880 | * one group and restart from the beginning. | |
881 | */ | |
882 | if (!prev) | |
883 | continue; | |
884 | break; | |
527a5ec9 | 885 | } |
7d74b06f | 886 | |
5ac8fb31 JW |
887 | /* |
888 | * Verify the css and acquire a reference. The root | |
889 | * is provided by the caller, so we know it's alive | |
890 | * and kicking, and don't take an extra reference. | |
891 | */ | |
892 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 893 | |
5ac8fb31 JW |
894 | if (css == &root->css) |
895 | break; | |
14067bb3 | 896 | |
b2052564 | 897 | if (css_tryget(css)) { |
5ac8fb31 JW |
898 | /* |
899 | * Make sure the memcg is initialized: | |
900 | * mem_cgroup_css_online() orders the the | |
901 | * initialization against setting the flag. | |
902 | */ | |
903 | if (smp_load_acquire(&memcg->initialized)) | |
904 | break; | |
542f85f9 | 905 | |
5ac8fb31 | 906 | css_put(css); |
527a5ec9 | 907 | } |
9f3a0d09 | 908 | |
5ac8fb31 | 909 | memcg = NULL; |
9f3a0d09 | 910 | } |
5ac8fb31 JW |
911 | |
912 | if (reclaim) { | |
5ac8fb31 | 913 | /* |
6df38689 VD |
914 | * The position could have already been updated by a competing |
915 | * thread, so check that the value hasn't changed since we read | |
916 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 917 | */ |
6df38689 VD |
918 | (void)cmpxchg(&iter->position, pos, memcg); |
919 | ||
5ac8fb31 JW |
920 | if (pos) |
921 | css_put(&pos->css); | |
922 | ||
923 | if (!memcg) | |
924 | iter->generation++; | |
925 | else if (!prev) | |
926 | reclaim->generation = iter->generation; | |
9f3a0d09 | 927 | } |
5ac8fb31 | 928 | |
542f85f9 MH |
929 | out_unlock: |
930 | rcu_read_unlock(); | |
5ac8fb31 | 931 | out: |
c40046f3 MH |
932 | if (prev && prev != root) |
933 | css_put(&prev->css); | |
934 | ||
9f3a0d09 | 935 | return memcg; |
14067bb3 | 936 | } |
7d74b06f | 937 | |
5660048c JW |
938 | /** |
939 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
940 | * @root: hierarchy root | |
941 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
942 | */ | |
943 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
944 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
945 | { |
946 | if (!root) | |
947 | root = root_mem_cgroup; | |
948 | if (prev && prev != root) | |
949 | css_put(&prev->css); | |
950 | } | |
7d74b06f | 951 | |
6df38689 VD |
952 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
953 | { | |
954 | struct mem_cgroup *memcg = dead_memcg; | |
955 | struct mem_cgroup_reclaim_iter *iter; | |
956 | struct mem_cgroup_per_zone *mz; | |
957 | int nid, zid; | |
958 | int i; | |
959 | ||
960 | while ((memcg = parent_mem_cgroup(memcg))) { | |
961 | for_each_node(nid) { | |
962 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
963 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
964 | for (i = 0; i <= DEF_PRIORITY; i++) { | |
965 | iter = &mz->iter[i]; | |
966 | cmpxchg(&iter->position, | |
967 | dead_memcg, NULL); | |
968 | } | |
969 | } | |
970 | } | |
971 | } | |
972 | } | |
973 | ||
9f3a0d09 JW |
974 | /* |
975 | * Iteration constructs for visiting all cgroups (under a tree). If | |
976 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
977 | * be used for reference counting. | |
978 | */ | |
979 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 980 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 981 | iter != NULL; \ |
527a5ec9 | 982 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 983 | |
9f3a0d09 | 984 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 985 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 986 | iter != NULL; \ |
527a5ec9 | 987 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 988 | |
925b7673 JW |
989 | /** |
990 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
991 | * @zone: zone of the wanted lruvec | |
fa9add64 | 992 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
993 | * |
994 | * Returns the lru list vector holding pages for the given @zone and | |
995 | * @mem. This can be the global zone lruvec, if the memory controller | |
996 | * is disabled. | |
997 | */ | |
998 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
999 | struct mem_cgroup *memcg) | |
1000 | { | |
1001 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1002 | struct lruvec *lruvec; |
925b7673 | 1003 | |
bea8c150 HD |
1004 | if (mem_cgroup_disabled()) { |
1005 | lruvec = &zone->lruvec; | |
1006 | goto out; | |
1007 | } | |
925b7673 | 1008 | |
e231875b | 1009 | mz = mem_cgroup_zone_zoneinfo(memcg, zone); |
bea8c150 HD |
1010 | lruvec = &mz->lruvec; |
1011 | out: | |
1012 | /* | |
1013 | * Since a node can be onlined after the mem_cgroup was created, | |
1014 | * we have to be prepared to initialize lruvec->zone here; | |
1015 | * and if offlined then reonlined, we need to reinitialize it. | |
1016 | */ | |
1017 | if (unlikely(lruvec->zone != zone)) | |
1018 | lruvec->zone = zone; | |
1019 | return lruvec; | |
925b7673 JW |
1020 | } |
1021 | ||
925b7673 | 1022 | /** |
dfe0e773 | 1023 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1024 | * @page: the page |
fa9add64 | 1025 | * @zone: zone of the page |
dfe0e773 JW |
1026 | * |
1027 | * This function is only safe when following the LRU page isolation | |
1028 | * and putback protocol: the LRU lock must be held, and the page must | |
1029 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 1030 | */ |
fa9add64 | 1031 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1032 | { |
08e552c6 | 1033 | struct mem_cgroup_per_zone *mz; |
925b7673 | 1034 | struct mem_cgroup *memcg; |
bea8c150 | 1035 | struct lruvec *lruvec; |
6d12e2d8 | 1036 | |
bea8c150 HD |
1037 | if (mem_cgroup_disabled()) { |
1038 | lruvec = &zone->lruvec; | |
1039 | goto out; | |
1040 | } | |
925b7673 | 1041 | |
1306a85a | 1042 | memcg = page->mem_cgroup; |
7512102c | 1043 | /* |
dfe0e773 | 1044 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1045 | * possibly migrated - before they are charged. |
7512102c | 1046 | */ |
29833315 JW |
1047 | if (!memcg) |
1048 | memcg = root_mem_cgroup; | |
7512102c | 1049 | |
e231875b | 1050 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
bea8c150 HD |
1051 | lruvec = &mz->lruvec; |
1052 | out: | |
1053 | /* | |
1054 | * Since a node can be onlined after the mem_cgroup was created, | |
1055 | * we have to be prepared to initialize lruvec->zone here; | |
1056 | * and if offlined then reonlined, we need to reinitialize it. | |
1057 | */ | |
1058 | if (unlikely(lruvec->zone != zone)) | |
1059 | lruvec->zone = zone; | |
1060 | return lruvec; | |
08e552c6 | 1061 | } |
b69408e8 | 1062 | |
925b7673 | 1063 | /** |
fa9add64 HD |
1064 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1065 | * @lruvec: mem_cgroup per zone lru vector | |
1066 | * @lru: index of lru list the page is sitting on | |
1067 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1068 | * |
fa9add64 HD |
1069 | * This function must be called when a page is added to or removed from an |
1070 | * lru list. | |
3f58a829 | 1071 | */ |
fa9add64 HD |
1072 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1073 | int nr_pages) | |
3f58a829 MK |
1074 | { |
1075 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1076 | unsigned long *lru_size; |
3f58a829 MK |
1077 | |
1078 | if (mem_cgroup_disabled()) | |
1079 | return; | |
1080 | ||
fa9add64 HD |
1081 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1082 | lru_size = mz->lru_size + lru; | |
1083 | *lru_size += nr_pages; | |
1084 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1085 | } |
544122e5 | 1086 | |
2314b42d | 1087 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
c3ac9a8a | 1088 | { |
2314b42d | 1089 | struct mem_cgroup *task_memcg; |
158e0a2d | 1090 | struct task_struct *p; |
ffbdccf5 | 1091 | bool ret; |
4c4a2214 | 1092 | |
158e0a2d | 1093 | p = find_lock_task_mm(task); |
de077d22 | 1094 | if (p) { |
2314b42d | 1095 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1096 | task_unlock(p); |
1097 | } else { | |
1098 | /* | |
1099 | * All threads may have already detached their mm's, but the oom | |
1100 | * killer still needs to detect if they have already been oom | |
1101 | * killed to prevent needlessly killing additional tasks. | |
1102 | */ | |
ffbdccf5 | 1103 | rcu_read_lock(); |
2314b42d JW |
1104 | task_memcg = mem_cgroup_from_task(task); |
1105 | css_get(&task_memcg->css); | |
ffbdccf5 | 1106 | rcu_read_unlock(); |
de077d22 | 1107 | } |
2314b42d JW |
1108 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1109 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1110 | return ret; |
1111 | } | |
1112 | ||
3e32cb2e | 1113 | #define mem_cgroup_from_counter(counter, member) \ |
6d61ef40 BS |
1114 | container_of(counter, struct mem_cgroup, member) |
1115 | ||
19942822 | 1116 | /** |
9d11ea9f | 1117 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1118 | * @memcg: the memory cgroup |
19942822 | 1119 | * |
9d11ea9f | 1120 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1121 | * pages. |
19942822 | 1122 | */ |
c0ff4b85 | 1123 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1124 | { |
3e32cb2e JW |
1125 | unsigned long margin = 0; |
1126 | unsigned long count; | |
1127 | unsigned long limit; | |
9d11ea9f | 1128 | |
3e32cb2e | 1129 | count = page_counter_read(&memcg->memory); |
4db0c3c2 | 1130 | limit = READ_ONCE(memcg->memory.limit); |
3e32cb2e JW |
1131 | if (count < limit) |
1132 | margin = limit - count; | |
1133 | ||
7941d214 | 1134 | if (do_memsw_account()) { |
3e32cb2e | 1135 | count = page_counter_read(&memcg->memsw); |
4db0c3c2 | 1136 | limit = READ_ONCE(memcg->memsw.limit); |
3e32cb2e JW |
1137 | if (count <= limit) |
1138 | margin = min(margin, limit - count); | |
1139 | } | |
1140 | ||
1141 | return margin; | |
19942822 JW |
1142 | } |
1143 | ||
32047e2a | 1144 | /* |
bdcbb659 | 1145 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1146 | * |
bdcbb659 QH |
1147 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1148 | * moving cgroups. This is for waiting at high-memory pressure | |
1149 | * caused by "move". | |
32047e2a | 1150 | */ |
c0ff4b85 | 1151 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1152 | { |
2bd9bb20 KH |
1153 | struct mem_cgroup *from; |
1154 | struct mem_cgroup *to; | |
4b534334 | 1155 | bool ret = false; |
2bd9bb20 KH |
1156 | /* |
1157 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1158 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1159 | */ | |
1160 | spin_lock(&mc.lock); | |
1161 | from = mc.from; | |
1162 | to = mc.to; | |
1163 | if (!from) | |
1164 | goto unlock; | |
3e92041d | 1165 | |
2314b42d JW |
1166 | ret = mem_cgroup_is_descendant(from, memcg) || |
1167 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1168 | unlock: |
1169 | spin_unlock(&mc.lock); | |
4b534334 KH |
1170 | return ret; |
1171 | } | |
1172 | ||
c0ff4b85 | 1173 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1174 | { |
1175 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1176 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1177 | DEFINE_WAIT(wait); |
1178 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1179 | /* moving charge context might have finished. */ | |
1180 | if (mc.moving_task) | |
1181 | schedule(); | |
1182 | finish_wait(&mc.waitq, &wait); | |
1183 | return true; | |
1184 | } | |
1185 | } | |
1186 | return false; | |
1187 | } | |
1188 | ||
58cf188e | 1189 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1190 | /** |
58cf188e | 1191 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1192 | * @memcg: The memory cgroup that went over limit |
1193 | * @p: Task that is going to be killed | |
1194 | * | |
1195 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1196 | * enabled | |
1197 | */ | |
1198 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1199 | { | |
e61734c5 | 1200 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
08088cb9 | 1201 | static DEFINE_MUTEX(oom_info_lock); |
58cf188e SZ |
1202 | struct mem_cgroup *iter; |
1203 | unsigned int i; | |
e222432b | 1204 | |
08088cb9 | 1205 | mutex_lock(&oom_info_lock); |
e222432b BS |
1206 | rcu_read_lock(); |
1207 | ||
2415b9f5 BV |
1208 | if (p) { |
1209 | pr_info("Task in "); | |
1210 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1211 | pr_cont(" killed as a result of limit of "); | |
1212 | } else { | |
1213 | pr_info("Memory limit reached of cgroup "); | |
1214 | } | |
1215 | ||
e61734c5 | 1216 | pr_cont_cgroup_path(memcg->css.cgroup); |
0346dadb | 1217 | pr_cont("\n"); |
e222432b | 1218 | |
e222432b BS |
1219 | rcu_read_unlock(); |
1220 | ||
3e32cb2e JW |
1221 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1222 | K((u64)page_counter_read(&memcg->memory)), | |
1223 | K((u64)memcg->memory.limit), memcg->memory.failcnt); | |
1224 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1225 | K((u64)page_counter_read(&memcg->memsw)), | |
1226 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); | |
1227 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1228 | K((u64)page_counter_read(&memcg->kmem)), | |
1229 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); | |
58cf188e SZ |
1230 | |
1231 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1232 | pr_info("Memory cgroup stats for "); |
1233 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1234 | pr_cont(":"); |
1235 | ||
1236 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
7941d214 | 1237 | if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account()) |
58cf188e | 1238 | continue; |
484ebb3b | 1239 | pr_cont(" %s:%luKB", mem_cgroup_stat_names[i], |
58cf188e SZ |
1240 | K(mem_cgroup_read_stat(iter, i))); |
1241 | } | |
1242 | ||
1243 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1244 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1245 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1246 | ||
1247 | pr_cont("\n"); | |
1248 | } | |
08088cb9 | 1249 | mutex_unlock(&oom_info_lock); |
e222432b BS |
1250 | } |
1251 | ||
81d39c20 KH |
1252 | /* |
1253 | * This function returns the number of memcg under hierarchy tree. Returns | |
1254 | * 1(self count) if no children. | |
1255 | */ | |
c0ff4b85 | 1256 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1257 | { |
1258 | int num = 0; | |
7d74b06f KH |
1259 | struct mem_cgroup *iter; |
1260 | ||
c0ff4b85 | 1261 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1262 | num++; |
81d39c20 KH |
1263 | return num; |
1264 | } | |
1265 | ||
a63d83f4 DR |
1266 | /* |
1267 | * Return the memory (and swap, if configured) limit for a memcg. | |
1268 | */ | |
3e32cb2e | 1269 | static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 | 1270 | { |
3e32cb2e | 1271 | unsigned long limit; |
f3e8eb70 | 1272 | |
3e32cb2e | 1273 | limit = memcg->memory.limit; |
9a5a8f19 | 1274 | if (mem_cgroup_swappiness(memcg)) { |
3e32cb2e | 1275 | unsigned long memsw_limit; |
9a5a8f19 | 1276 | |
3e32cb2e JW |
1277 | memsw_limit = memcg->memsw.limit; |
1278 | limit = min(limit + total_swap_pages, memsw_limit); | |
9a5a8f19 | 1279 | } |
9a5a8f19 | 1280 | return limit; |
a63d83f4 DR |
1281 | } |
1282 | ||
19965460 DR |
1283 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1284 | int order) | |
9cbb78bb | 1285 | { |
6e0fc46d DR |
1286 | struct oom_control oc = { |
1287 | .zonelist = NULL, | |
1288 | .nodemask = NULL, | |
1289 | .gfp_mask = gfp_mask, | |
1290 | .order = order, | |
6e0fc46d | 1291 | }; |
9cbb78bb DR |
1292 | struct mem_cgroup *iter; |
1293 | unsigned long chosen_points = 0; | |
1294 | unsigned long totalpages; | |
1295 | unsigned int points = 0; | |
1296 | struct task_struct *chosen = NULL; | |
1297 | ||
dc56401f JW |
1298 | mutex_lock(&oom_lock); |
1299 | ||
876aafbf | 1300 | /* |
465adcf1 DR |
1301 | * If current has a pending SIGKILL or is exiting, then automatically |
1302 | * select it. The goal is to allow it to allocate so that it may | |
1303 | * quickly exit and free its memory. | |
876aafbf | 1304 | */ |
d003f371 | 1305 | if (fatal_signal_pending(current) || task_will_free_mem(current)) { |
16e95196 | 1306 | mark_oom_victim(current); |
dc56401f | 1307 | goto unlock; |
876aafbf DR |
1308 | } |
1309 | ||
6e0fc46d | 1310 | check_panic_on_oom(&oc, CONSTRAINT_MEMCG, memcg); |
3e32cb2e | 1311 | totalpages = mem_cgroup_get_limit(memcg) ? : 1; |
9cbb78bb | 1312 | for_each_mem_cgroup_tree(iter, memcg) { |
72ec7029 | 1313 | struct css_task_iter it; |
9cbb78bb DR |
1314 | struct task_struct *task; |
1315 | ||
72ec7029 TH |
1316 | css_task_iter_start(&iter->css, &it); |
1317 | while ((task = css_task_iter_next(&it))) { | |
6e0fc46d | 1318 | switch (oom_scan_process_thread(&oc, task, totalpages)) { |
9cbb78bb DR |
1319 | case OOM_SCAN_SELECT: |
1320 | if (chosen) | |
1321 | put_task_struct(chosen); | |
1322 | chosen = task; | |
1323 | chosen_points = ULONG_MAX; | |
1324 | get_task_struct(chosen); | |
1325 | /* fall through */ | |
1326 | case OOM_SCAN_CONTINUE: | |
1327 | continue; | |
1328 | case OOM_SCAN_ABORT: | |
72ec7029 | 1329 | css_task_iter_end(&it); |
9cbb78bb DR |
1330 | mem_cgroup_iter_break(memcg, iter); |
1331 | if (chosen) | |
1332 | put_task_struct(chosen); | |
dc56401f | 1333 | goto unlock; |
9cbb78bb DR |
1334 | case OOM_SCAN_OK: |
1335 | break; | |
1336 | }; | |
1337 | points = oom_badness(task, memcg, NULL, totalpages); | |
d49ad935 DR |
1338 | if (!points || points < chosen_points) |
1339 | continue; | |
1340 | /* Prefer thread group leaders for display purposes */ | |
1341 | if (points == chosen_points && | |
1342 | thread_group_leader(chosen)) | |
1343 | continue; | |
1344 | ||
1345 | if (chosen) | |
1346 | put_task_struct(chosen); | |
1347 | chosen = task; | |
1348 | chosen_points = points; | |
1349 | get_task_struct(chosen); | |
9cbb78bb | 1350 | } |
72ec7029 | 1351 | css_task_iter_end(&it); |
9cbb78bb DR |
1352 | } |
1353 | ||
dc56401f JW |
1354 | if (chosen) { |
1355 | points = chosen_points * 1000 / totalpages; | |
6e0fc46d DR |
1356 | oom_kill_process(&oc, chosen, points, totalpages, memcg, |
1357 | "Memory cgroup out of memory"); | |
dc56401f JW |
1358 | } |
1359 | unlock: | |
1360 | mutex_unlock(&oom_lock); | |
9cbb78bb DR |
1361 | } |
1362 | ||
ae6e71d3 MC |
1363 | #if MAX_NUMNODES > 1 |
1364 | ||
4d0c066d KH |
1365 | /** |
1366 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1367 | * @memcg: the target memcg |
4d0c066d KH |
1368 | * @nid: the node ID to be checked. |
1369 | * @noswap : specify true here if the user wants flle only information. | |
1370 | * | |
1371 | * This function returns whether the specified memcg contains any | |
1372 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1373 | * pages in the node. | |
1374 | */ | |
c0ff4b85 | 1375 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1376 | int nid, bool noswap) |
1377 | { | |
c0ff4b85 | 1378 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1379 | return true; |
1380 | if (noswap || !total_swap_pages) | |
1381 | return false; | |
c0ff4b85 | 1382 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1383 | return true; |
1384 | return false; | |
1385 | ||
1386 | } | |
889976db YH |
1387 | |
1388 | /* | |
1389 | * Always updating the nodemask is not very good - even if we have an empty | |
1390 | * list or the wrong list here, we can start from some node and traverse all | |
1391 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1392 | * | |
1393 | */ | |
c0ff4b85 | 1394 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1395 | { |
1396 | int nid; | |
453a9bf3 KH |
1397 | /* |
1398 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1399 | * pagein/pageout changes since the last update. | |
1400 | */ | |
c0ff4b85 | 1401 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1402 | return; |
c0ff4b85 | 1403 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1404 | return; |
1405 | ||
889976db | 1406 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1407 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1408 | |
31aaea4a | 1409 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1410 | |
c0ff4b85 R |
1411 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1412 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1413 | } |
453a9bf3 | 1414 | |
c0ff4b85 R |
1415 | atomic_set(&memcg->numainfo_events, 0); |
1416 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1417 | } |
1418 | ||
1419 | /* | |
1420 | * Selecting a node where we start reclaim from. Because what we need is just | |
1421 | * reducing usage counter, start from anywhere is O,K. Considering | |
1422 | * memory reclaim from current node, there are pros. and cons. | |
1423 | * | |
1424 | * Freeing memory from current node means freeing memory from a node which | |
1425 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1426 | * hit limits, it will see a contention on a node. But freeing from remote | |
1427 | * node means more costs for memory reclaim because of memory latency. | |
1428 | * | |
1429 | * Now, we use round-robin. Better algorithm is welcomed. | |
1430 | */ | |
c0ff4b85 | 1431 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1432 | { |
1433 | int node; | |
1434 | ||
c0ff4b85 R |
1435 | mem_cgroup_may_update_nodemask(memcg); |
1436 | node = memcg->last_scanned_node; | |
889976db | 1437 | |
c0ff4b85 | 1438 | node = next_node(node, memcg->scan_nodes); |
889976db | 1439 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1440 | node = first_node(memcg->scan_nodes); |
889976db YH |
1441 | /* |
1442 | * We call this when we hit limit, not when pages are added to LRU. | |
1443 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1444 | * memcg is too small and all pages are not on LRU. In that case, | |
1445 | * we use curret node. | |
1446 | */ | |
1447 | if (unlikely(node == MAX_NUMNODES)) | |
1448 | node = numa_node_id(); | |
1449 | ||
c0ff4b85 | 1450 | memcg->last_scanned_node = node; |
889976db YH |
1451 | return node; |
1452 | } | |
889976db | 1453 | #else |
c0ff4b85 | 1454 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1455 | { |
1456 | return 0; | |
1457 | } | |
1458 | #endif | |
1459 | ||
0608f43d AM |
1460 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1461 | struct zone *zone, | |
1462 | gfp_t gfp_mask, | |
1463 | unsigned long *total_scanned) | |
1464 | { | |
1465 | struct mem_cgroup *victim = NULL; | |
1466 | int total = 0; | |
1467 | int loop = 0; | |
1468 | unsigned long excess; | |
1469 | unsigned long nr_scanned; | |
1470 | struct mem_cgroup_reclaim_cookie reclaim = { | |
1471 | .zone = zone, | |
1472 | .priority = 0, | |
1473 | }; | |
1474 | ||
3e32cb2e | 1475 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1476 | |
1477 | while (1) { | |
1478 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1479 | if (!victim) { | |
1480 | loop++; | |
1481 | if (loop >= 2) { | |
1482 | /* | |
1483 | * If we have not been able to reclaim | |
1484 | * anything, it might because there are | |
1485 | * no reclaimable pages under this hierarchy | |
1486 | */ | |
1487 | if (!total) | |
1488 | break; | |
1489 | /* | |
1490 | * We want to do more targeted reclaim. | |
1491 | * excess >> 2 is not to excessive so as to | |
1492 | * reclaim too much, nor too less that we keep | |
1493 | * coming back to reclaim from this cgroup | |
1494 | */ | |
1495 | if (total >= (excess >> 2) || | |
1496 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1497 | break; | |
1498 | } | |
1499 | continue; | |
1500 | } | |
0608f43d AM |
1501 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1502 | zone, &nr_scanned); | |
1503 | *total_scanned += nr_scanned; | |
3e32cb2e | 1504 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1505 | break; |
6d61ef40 | 1506 | } |
0608f43d AM |
1507 | mem_cgroup_iter_break(root_memcg, victim); |
1508 | return total; | |
6d61ef40 BS |
1509 | } |
1510 | ||
0056f4e6 JW |
1511 | #ifdef CONFIG_LOCKDEP |
1512 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1513 | .name = "memcg_oom_lock", | |
1514 | }; | |
1515 | #endif | |
1516 | ||
fb2a6fc5 JW |
1517 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1518 | ||
867578cb KH |
1519 | /* |
1520 | * Check OOM-Killer is already running under our hierarchy. | |
1521 | * If someone is running, return false. | |
1522 | */ | |
fb2a6fc5 | 1523 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1524 | { |
79dfdacc | 1525 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1526 | |
fb2a6fc5 JW |
1527 | spin_lock(&memcg_oom_lock); |
1528 | ||
9f3a0d09 | 1529 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1530 | if (iter->oom_lock) { |
79dfdacc MH |
1531 | /* |
1532 | * this subtree of our hierarchy is already locked | |
1533 | * so we cannot give a lock. | |
1534 | */ | |
79dfdacc | 1535 | failed = iter; |
9f3a0d09 JW |
1536 | mem_cgroup_iter_break(memcg, iter); |
1537 | break; | |
23751be0 JW |
1538 | } else |
1539 | iter->oom_lock = true; | |
7d74b06f | 1540 | } |
867578cb | 1541 | |
fb2a6fc5 JW |
1542 | if (failed) { |
1543 | /* | |
1544 | * OK, we failed to lock the whole subtree so we have | |
1545 | * to clean up what we set up to the failing subtree | |
1546 | */ | |
1547 | for_each_mem_cgroup_tree(iter, memcg) { | |
1548 | if (iter == failed) { | |
1549 | mem_cgroup_iter_break(memcg, iter); | |
1550 | break; | |
1551 | } | |
1552 | iter->oom_lock = false; | |
79dfdacc | 1553 | } |
0056f4e6 JW |
1554 | } else |
1555 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1556 | |
1557 | spin_unlock(&memcg_oom_lock); | |
1558 | ||
1559 | return !failed; | |
a636b327 | 1560 | } |
0b7f569e | 1561 | |
fb2a6fc5 | 1562 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1563 | { |
7d74b06f KH |
1564 | struct mem_cgroup *iter; |
1565 | ||
fb2a6fc5 | 1566 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1567 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1568 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1569 | iter->oom_lock = false; |
fb2a6fc5 | 1570 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1571 | } |
1572 | ||
c0ff4b85 | 1573 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1574 | { |
1575 | struct mem_cgroup *iter; | |
1576 | ||
c2b42d3c | 1577 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1578 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1579 | iter->under_oom++; |
1580 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1581 | } |
1582 | ||
c0ff4b85 | 1583 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1584 | { |
1585 | struct mem_cgroup *iter; | |
1586 | ||
867578cb KH |
1587 | /* |
1588 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1589 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1590 | */ |
c2b42d3c | 1591 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1592 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1593 | if (iter->under_oom > 0) |
1594 | iter->under_oom--; | |
1595 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1596 | } |
1597 | ||
867578cb KH |
1598 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1599 | ||
dc98df5a | 1600 | struct oom_wait_info { |
d79154bb | 1601 | struct mem_cgroup *memcg; |
dc98df5a KH |
1602 | wait_queue_t wait; |
1603 | }; | |
1604 | ||
1605 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1606 | unsigned mode, int sync, void *arg) | |
1607 | { | |
d79154bb HD |
1608 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1609 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1610 | struct oom_wait_info *oom_wait_info; |
1611 | ||
1612 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1613 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1614 | |
2314b42d JW |
1615 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1616 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1617 | return 0; |
dc98df5a KH |
1618 | return autoremove_wake_function(wait, mode, sync, arg); |
1619 | } | |
1620 | ||
c0ff4b85 | 1621 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1622 | { |
c2b42d3c TH |
1623 | /* |
1624 | * For the following lockless ->under_oom test, the only required | |
1625 | * guarantee is that it must see the state asserted by an OOM when | |
1626 | * this function is called as a result of userland actions | |
1627 | * triggered by the notification of the OOM. This is trivially | |
1628 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1629 | * triggering notification. | |
1630 | */ | |
1631 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1632 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1633 | } |
1634 | ||
3812c8c8 | 1635 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 1636 | { |
626ebc41 | 1637 | if (!current->memcg_may_oom) |
3812c8c8 | 1638 | return; |
867578cb | 1639 | /* |
49426420 JW |
1640 | * We are in the middle of the charge context here, so we |
1641 | * don't want to block when potentially sitting on a callstack | |
1642 | * that holds all kinds of filesystem and mm locks. | |
1643 | * | |
1644 | * Also, the caller may handle a failed allocation gracefully | |
1645 | * (like optional page cache readahead) and so an OOM killer | |
1646 | * invocation might not even be necessary. | |
1647 | * | |
1648 | * That's why we don't do anything here except remember the | |
1649 | * OOM context and then deal with it at the end of the page | |
1650 | * fault when the stack is unwound, the locks are released, | |
1651 | * and when we know whether the fault was overall successful. | |
867578cb | 1652 | */ |
49426420 | 1653 | css_get(&memcg->css); |
626ebc41 TH |
1654 | current->memcg_in_oom = memcg; |
1655 | current->memcg_oom_gfp_mask = mask; | |
1656 | current->memcg_oom_order = order; | |
3812c8c8 JW |
1657 | } |
1658 | ||
1659 | /** | |
1660 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1661 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1662 | * |
49426420 JW |
1663 | * This has to be called at the end of a page fault if the memcg OOM |
1664 | * handler was enabled. | |
3812c8c8 | 1665 | * |
49426420 | 1666 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1667 | * sleep on a waitqueue until the userspace task resolves the |
1668 | * situation. Sleeping directly in the charge context with all kinds | |
1669 | * of locks held is not a good idea, instead we remember an OOM state | |
1670 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1671 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1672 | * |
1673 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1674 | * completed, %false otherwise. |
3812c8c8 | 1675 | */ |
49426420 | 1676 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1677 | { |
626ebc41 | 1678 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1679 | struct oom_wait_info owait; |
49426420 | 1680 | bool locked; |
3812c8c8 JW |
1681 | |
1682 | /* OOM is global, do not handle */ | |
3812c8c8 | 1683 | if (!memcg) |
49426420 | 1684 | return false; |
3812c8c8 | 1685 | |
c32b3cbe | 1686 | if (!handle || oom_killer_disabled) |
49426420 | 1687 | goto cleanup; |
3812c8c8 JW |
1688 | |
1689 | owait.memcg = memcg; | |
1690 | owait.wait.flags = 0; | |
1691 | owait.wait.func = memcg_oom_wake_function; | |
1692 | owait.wait.private = current; | |
1693 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 1694 | |
3812c8c8 | 1695 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1696 | mem_cgroup_mark_under_oom(memcg); |
1697 | ||
1698 | locked = mem_cgroup_oom_trylock(memcg); | |
1699 | ||
1700 | if (locked) | |
1701 | mem_cgroup_oom_notify(memcg); | |
1702 | ||
1703 | if (locked && !memcg->oom_kill_disable) { | |
1704 | mem_cgroup_unmark_under_oom(memcg); | |
1705 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1706 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1707 | current->memcg_oom_order); | |
49426420 | 1708 | } else { |
3812c8c8 | 1709 | schedule(); |
49426420 JW |
1710 | mem_cgroup_unmark_under_oom(memcg); |
1711 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1712 | } | |
1713 | ||
1714 | if (locked) { | |
fb2a6fc5 JW |
1715 | mem_cgroup_oom_unlock(memcg); |
1716 | /* | |
1717 | * There is no guarantee that an OOM-lock contender | |
1718 | * sees the wakeups triggered by the OOM kill | |
1719 | * uncharges. Wake any sleepers explicitely. | |
1720 | */ | |
1721 | memcg_oom_recover(memcg); | |
1722 | } | |
49426420 | 1723 | cleanup: |
626ebc41 | 1724 | current->memcg_in_oom = NULL; |
3812c8c8 | 1725 | css_put(&memcg->css); |
867578cb | 1726 | return true; |
0b7f569e KH |
1727 | } |
1728 | ||
d7365e78 JW |
1729 | /** |
1730 | * mem_cgroup_begin_page_stat - begin a page state statistics transaction | |
1731 | * @page: page that is going to change accounted state | |
32047e2a | 1732 | * |
d7365e78 JW |
1733 | * This function must mark the beginning of an accounted page state |
1734 | * change to prevent double accounting when the page is concurrently | |
1735 | * being moved to another memcg: | |
32047e2a | 1736 | * |
6de22619 | 1737 | * memcg = mem_cgroup_begin_page_stat(page); |
d7365e78 JW |
1738 | * if (TestClearPageState(page)) |
1739 | * mem_cgroup_update_page_stat(memcg, state, -1); | |
6de22619 | 1740 | * mem_cgroup_end_page_stat(memcg); |
d69b042f | 1741 | */ |
6de22619 | 1742 | struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page) |
89c06bd5 KH |
1743 | { |
1744 | struct mem_cgroup *memcg; | |
6de22619 | 1745 | unsigned long flags; |
89c06bd5 | 1746 | |
6de22619 JW |
1747 | /* |
1748 | * The RCU lock is held throughout the transaction. The fast | |
1749 | * path can get away without acquiring the memcg->move_lock | |
1750 | * because page moving starts with an RCU grace period. | |
1751 | * | |
1752 | * The RCU lock also protects the memcg from being freed when | |
1753 | * the page state that is going to change is the only thing | |
1754 | * preventing the page from being uncharged. | |
1755 | * E.g. end-writeback clearing PageWriteback(), which allows | |
1756 | * migration to go ahead and uncharge the page before the | |
1757 | * account transaction might be complete. | |
1758 | */ | |
d7365e78 JW |
1759 | rcu_read_lock(); |
1760 | ||
1761 | if (mem_cgroup_disabled()) | |
1762 | return NULL; | |
89c06bd5 | 1763 | again: |
1306a85a | 1764 | memcg = page->mem_cgroup; |
29833315 | 1765 | if (unlikely(!memcg)) |
d7365e78 JW |
1766 | return NULL; |
1767 | ||
bdcbb659 | 1768 | if (atomic_read(&memcg->moving_account) <= 0) |
d7365e78 | 1769 | return memcg; |
89c06bd5 | 1770 | |
6de22619 | 1771 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 1772 | if (memcg != page->mem_cgroup) { |
6de22619 | 1773 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
1774 | goto again; |
1775 | } | |
6de22619 JW |
1776 | |
1777 | /* | |
1778 | * When charge migration first begins, we can have locked and | |
1779 | * unlocked page stat updates happening concurrently. Track | |
1780 | * the task who has the lock for mem_cgroup_end_page_stat(). | |
1781 | */ | |
1782 | memcg->move_lock_task = current; | |
1783 | memcg->move_lock_flags = flags; | |
d7365e78 JW |
1784 | |
1785 | return memcg; | |
89c06bd5 | 1786 | } |
c4843a75 | 1787 | EXPORT_SYMBOL(mem_cgroup_begin_page_stat); |
89c06bd5 | 1788 | |
d7365e78 JW |
1789 | /** |
1790 | * mem_cgroup_end_page_stat - finish a page state statistics transaction | |
1791 | * @memcg: the memcg that was accounted against | |
d7365e78 | 1792 | */ |
6de22619 | 1793 | void mem_cgroup_end_page_stat(struct mem_cgroup *memcg) |
89c06bd5 | 1794 | { |
6de22619 JW |
1795 | if (memcg && memcg->move_lock_task == current) { |
1796 | unsigned long flags = memcg->move_lock_flags; | |
1797 | ||
1798 | memcg->move_lock_task = NULL; | |
1799 | memcg->move_lock_flags = 0; | |
1800 | ||
1801 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
1802 | } | |
89c06bd5 | 1803 | |
d7365e78 | 1804 | rcu_read_unlock(); |
89c06bd5 | 1805 | } |
c4843a75 | 1806 | EXPORT_SYMBOL(mem_cgroup_end_page_stat); |
89c06bd5 | 1807 | |
cdec2e42 KH |
1808 | /* |
1809 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1810 | * TODO: maybe necessary to use big numbers in big irons. | |
1811 | */ | |
7ec99d62 | 1812 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
1813 | struct memcg_stock_pcp { |
1814 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 1815 | unsigned int nr_pages; |
cdec2e42 | 1816 | struct work_struct work; |
26fe6168 | 1817 | unsigned long flags; |
a0db00fc | 1818 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
1819 | }; |
1820 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 1821 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 1822 | |
a0956d54 SS |
1823 | /** |
1824 | * consume_stock: Try to consume stocked charge on this cpu. | |
1825 | * @memcg: memcg to consume from. | |
1826 | * @nr_pages: how many pages to charge. | |
1827 | * | |
1828 | * The charges will only happen if @memcg matches the current cpu's memcg | |
1829 | * stock, and at least @nr_pages are available in that stock. Failure to | |
1830 | * service an allocation will refill the stock. | |
1831 | * | |
1832 | * returns true if successful, false otherwise. | |
cdec2e42 | 1833 | */ |
a0956d54 | 1834 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1835 | { |
1836 | struct memcg_stock_pcp *stock; | |
3e32cb2e | 1837 | bool ret = false; |
cdec2e42 | 1838 | |
a0956d54 | 1839 | if (nr_pages > CHARGE_BATCH) |
3e32cb2e | 1840 | return ret; |
a0956d54 | 1841 | |
cdec2e42 | 1842 | stock = &get_cpu_var(memcg_stock); |
3e32cb2e | 1843 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 1844 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
1845 | ret = true; |
1846 | } | |
cdec2e42 KH |
1847 | put_cpu_var(memcg_stock); |
1848 | return ret; | |
1849 | } | |
1850 | ||
1851 | /* | |
3e32cb2e | 1852 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
1853 | */ |
1854 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1855 | { | |
1856 | struct mem_cgroup *old = stock->cached; | |
1857 | ||
11c9ea4e | 1858 | if (stock->nr_pages) { |
3e32cb2e | 1859 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 1860 | if (do_memsw_account()) |
3e32cb2e | 1861 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 1862 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 1863 | stock->nr_pages = 0; |
cdec2e42 KH |
1864 | } |
1865 | stock->cached = NULL; | |
cdec2e42 KH |
1866 | } |
1867 | ||
1868 | /* | |
1869 | * This must be called under preempt disabled or must be called by | |
1870 | * a thread which is pinned to local cpu. | |
1871 | */ | |
1872 | static void drain_local_stock(struct work_struct *dummy) | |
1873 | { | |
7c8e0181 | 1874 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
cdec2e42 | 1875 | drain_stock(stock); |
26fe6168 | 1876 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
1877 | } |
1878 | ||
1879 | /* | |
3e32cb2e | 1880 | * Cache charges(val) to local per_cpu area. |
320cc51d | 1881 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 1882 | */ |
c0ff4b85 | 1883 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1884 | { |
1885 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
1886 | ||
c0ff4b85 | 1887 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 1888 | drain_stock(stock); |
c0ff4b85 | 1889 | stock->cached = memcg; |
cdec2e42 | 1890 | } |
11c9ea4e | 1891 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
1892 | put_cpu_var(memcg_stock); |
1893 | } | |
1894 | ||
1895 | /* | |
c0ff4b85 | 1896 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 1897 | * of the hierarchy under it. |
cdec2e42 | 1898 | */ |
6d3d6aa2 | 1899 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 1900 | { |
26fe6168 | 1901 | int cpu, curcpu; |
d38144b7 | 1902 | |
6d3d6aa2 JW |
1903 | /* If someone's already draining, avoid adding running more workers. */ |
1904 | if (!mutex_trylock(&percpu_charge_mutex)) | |
1905 | return; | |
cdec2e42 | 1906 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 1907 | get_online_cpus(); |
5af12d0e | 1908 | curcpu = get_cpu(); |
cdec2e42 KH |
1909 | for_each_online_cpu(cpu) { |
1910 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 1911 | struct mem_cgroup *memcg; |
26fe6168 | 1912 | |
c0ff4b85 R |
1913 | memcg = stock->cached; |
1914 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 1915 | continue; |
2314b42d | 1916 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) |
3e92041d | 1917 | continue; |
d1a05b69 MH |
1918 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
1919 | if (cpu == curcpu) | |
1920 | drain_local_stock(&stock->work); | |
1921 | else | |
1922 | schedule_work_on(cpu, &stock->work); | |
1923 | } | |
cdec2e42 | 1924 | } |
5af12d0e | 1925 | put_cpu(); |
f894ffa8 | 1926 | put_online_cpus(); |
9f50fad6 | 1927 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
1928 | } |
1929 | ||
0db0628d | 1930 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
1931 | unsigned long action, |
1932 | void *hcpu) | |
1933 | { | |
1934 | int cpu = (unsigned long)hcpu; | |
1935 | struct memcg_stock_pcp *stock; | |
1936 | ||
619d094b | 1937 | if (action == CPU_ONLINE) |
1489ebad | 1938 | return NOTIFY_OK; |
1489ebad | 1939 | |
d833049b | 1940 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 1941 | return NOTIFY_OK; |
711d3d2c | 1942 | |
cdec2e42 KH |
1943 | stock = &per_cpu(memcg_stock, cpu); |
1944 | drain_stock(stock); | |
1945 | return NOTIFY_OK; | |
1946 | } | |
1947 | ||
b23afb93 TH |
1948 | /* |
1949 | * Scheduled by try_charge() to be executed from the userland return path | |
1950 | * and reclaims memory over the high limit. | |
1951 | */ | |
1952 | void mem_cgroup_handle_over_high(void) | |
1953 | { | |
1954 | unsigned int nr_pages = current->memcg_nr_pages_over_high; | |
1955 | struct mem_cgroup *memcg, *pos; | |
1956 | ||
1957 | if (likely(!nr_pages)) | |
1958 | return; | |
1959 | ||
1960 | pos = memcg = get_mem_cgroup_from_mm(current->mm); | |
1961 | ||
1962 | do { | |
1963 | if (page_counter_read(&pos->memory) <= pos->high) | |
1964 | continue; | |
1965 | mem_cgroup_events(pos, MEMCG_HIGH, 1); | |
1966 | try_to_free_mem_cgroup_pages(pos, nr_pages, GFP_KERNEL, true); | |
1967 | } while ((pos = parent_mem_cgroup(pos))); | |
1968 | ||
1969 | css_put(&memcg->css); | |
1970 | current->memcg_nr_pages_over_high = 0; | |
1971 | } | |
1972 | ||
00501b53 JW |
1973 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1974 | unsigned int nr_pages) | |
8a9f3ccd | 1975 | { |
7ec99d62 | 1976 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 1977 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 1978 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 1979 | struct page_counter *counter; |
6539cc05 | 1980 | unsigned long nr_reclaimed; |
b70a2a21 JW |
1981 | bool may_swap = true; |
1982 | bool drained = false; | |
a636b327 | 1983 | |
ce00a967 | 1984 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 1985 | return 0; |
6539cc05 | 1986 | retry: |
b6b6cc72 | 1987 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 1988 | return 0; |
8a9f3ccd | 1989 | |
7941d214 | 1990 | if (!do_memsw_account() || |
6071ca52 JW |
1991 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
1992 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 1993 | goto done_restock; |
7941d214 | 1994 | if (do_memsw_account()) |
3e32cb2e JW |
1995 | page_counter_uncharge(&memcg->memsw, batch); |
1996 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 1997 | } else { |
3e32cb2e | 1998 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 1999 | may_swap = false; |
3fbe7244 | 2000 | } |
7a81b88c | 2001 | |
6539cc05 JW |
2002 | if (batch > nr_pages) { |
2003 | batch = nr_pages; | |
2004 | goto retry; | |
2005 | } | |
6d61ef40 | 2006 | |
06b078fc JW |
2007 | /* |
2008 | * Unlike in global OOM situations, memcg is not in a physical | |
2009 | * memory shortage. Allow dying and OOM-killed tasks to | |
2010 | * bypass the last charges so that they can exit quickly and | |
2011 | * free their memory. | |
2012 | */ | |
2013 | if (unlikely(test_thread_flag(TIF_MEMDIE) || | |
2014 | fatal_signal_pending(current) || | |
2015 | current->flags & PF_EXITING)) | |
10d53c74 | 2016 | goto force; |
06b078fc JW |
2017 | |
2018 | if (unlikely(task_in_memcg_oom(current))) | |
2019 | goto nomem; | |
2020 | ||
d0164adc | 2021 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2022 | goto nomem; |
4b534334 | 2023 | |
241994ed JW |
2024 | mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1); |
2025 | ||
b70a2a21 JW |
2026 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2027 | gfp_mask, may_swap); | |
6539cc05 | 2028 | |
61e02c74 | 2029 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2030 | goto retry; |
28c34c29 | 2031 | |
b70a2a21 | 2032 | if (!drained) { |
6d3d6aa2 | 2033 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2034 | drained = true; |
2035 | goto retry; | |
2036 | } | |
2037 | ||
28c34c29 JW |
2038 | if (gfp_mask & __GFP_NORETRY) |
2039 | goto nomem; | |
6539cc05 JW |
2040 | /* |
2041 | * Even though the limit is exceeded at this point, reclaim | |
2042 | * may have been able to free some pages. Retry the charge | |
2043 | * before killing the task. | |
2044 | * | |
2045 | * Only for regular pages, though: huge pages are rather | |
2046 | * unlikely to succeed so close to the limit, and we fall back | |
2047 | * to regular pages anyway in case of failure. | |
2048 | */ | |
61e02c74 | 2049 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2050 | goto retry; |
2051 | /* | |
2052 | * At task move, charge accounts can be doubly counted. So, it's | |
2053 | * better to wait until the end of task_move if something is going on. | |
2054 | */ | |
2055 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2056 | goto retry; | |
2057 | ||
9b130619 JW |
2058 | if (nr_retries--) |
2059 | goto retry; | |
2060 | ||
06b078fc | 2061 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2062 | goto force; |
06b078fc | 2063 | |
6539cc05 | 2064 | if (fatal_signal_pending(current)) |
10d53c74 | 2065 | goto force; |
6539cc05 | 2066 | |
241994ed JW |
2067 | mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1); |
2068 | ||
3608de07 JM |
2069 | mem_cgroup_oom(mem_over_limit, gfp_mask, |
2070 | get_order(nr_pages * PAGE_SIZE)); | |
7a81b88c | 2071 | nomem: |
6d1fdc48 | 2072 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2073 | return -ENOMEM; |
10d53c74 TH |
2074 | force: |
2075 | /* | |
2076 | * The allocation either can't fail or will lead to more memory | |
2077 | * being freed very soon. Allow memory usage go over the limit | |
2078 | * temporarily by force charging it. | |
2079 | */ | |
2080 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2081 | if (do_memsw_account()) |
10d53c74 TH |
2082 | page_counter_charge(&memcg->memsw, nr_pages); |
2083 | css_get_many(&memcg->css, nr_pages); | |
2084 | ||
2085 | return 0; | |
6539cc05 JW |
2086 | |
2087 | done_restock: | |
e8ea14cc | 2088 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2089 | if (batch > nr_pages) |
2090 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2091 | |
241994ed | 2092 | /* |
b23afb93 TH |
2093 | * If the hierarchy is above the normal consumption range, schedule |
2094 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2095 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2096 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2097 | * not recorded as it most likely matches current's and won't | |
2098 | * change in the meantime. As high limit is checked again before | |
2099 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2100 | */ |
2101 | do { | |
b23afb93 | 2102 | if (page_counter_read(&memcg->memory) > memcg->high) { |
9516a18a | 2103 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2104 | set_notify_resume(current); |
2105 | break; | |
2106 | } | |
241994ed | 2107 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2108 | |
2109 | return 0; | |
7a81b88c | 2110 | } |
8a9f3ccd | 2111 | |
00501b53 | 2112 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2113 | { |
ce00a967 JW |
2114 | if (mem_cgroup_is_root(memcg)) |
2115 | return; | |
2116 | ||
3e32cb2e | 2117 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2118 | if (do_memsw_account()) |
3e32cb2e | 2119 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2120 | |
e8ea14cc | 2121 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2122 | } |
2123 | ||
0a31bc97 JW |
2124 | static void lock_page_lru(struct page *page, int *isolated) |
2125 | { | |
2126 | struct zone *zone = page_zone(page); | |
2127 | ||
2128 | spin_lock_irq(&zone->lru_lock); | |
2129 | if (PageLRU(page)) { | |
2130 | struct lruvec *lruvec; | |
2131 | ||
2132 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2133 | ClearPageLRU(page); | |
2134 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2135 | *isolated = 1; | |
2136 | } else | |
2137 | *isolated = 0; | |
2138 | } | |
2139 | ||
2140 | static void unlock_page_lru(struct page *page, int isolated) | |
2141 | { | |
2142 | struct zone *zone = page_zone(page); | |
2143 | ||
2144 | if (isolated) { | |
2145 | struct lruvec *lruvec; | |
2146 | ||
2147 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2148 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
2149 | SetPageLRU(page); | |
2150 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2151 | } | |
2152 | spin_unlock_irq(&zone->lru_lock); | |
2153 | } | |
2154 | ||
00501b53 | 2155 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2156 | bool lrucare) |
7a81b88c | 2157 | { |
0a31bc97 | 2158 | int isolated; |
9ce70c02 | 2159 | |
1306a85a | 2160 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2161 | |
2162 | /* | |
2163 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2164 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2165 | */ | |
0a31bc97 JW |
2166 | if (lrucare) |
2167 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2168 | |
0a31bc97 JW |
2169 | /* |
2170 | * Nobody should be changing or seriously looking at | |
1306a85a | 2171 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2172 | * |
2173 | * - the page is uncharged | |
2174 | * | |
2175 | * - the page is off-LRU | |
2176 | * | |
2177 | * - an anonymous fault has exclusive page access, except for | |
2178 | * a locked page table | |
2179 | * | |
2180 | * - a page cache insertion, a swapin fault, or a migration | |
2181 | * have the page locked | |
2182 | */ | |
1306a85a | 2183 | page->mem_cgroup = memcg; |
9ce70c02 | 2184 | |
0a31bc97 JW |
2185 | if (lrucare) |
2186 | unlock_page_lru(page, isolated); | |
7a81b88c | 2187 | } |
66e1707b | 2188 | |
7ae1e1d0 | 2189 | #ifdef CONFIG_MEMCG_KMEM |
f3bb3043 | 2190 | static int memcg_alloc_cache_id(void) |
55007d84 | 2191 | { |
f3bb3043 VD |
2192 | int id, size; |
2193 | int err; | |
2194 | ||
dbcf73e2 | 2195 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2196 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2197 | if (id < 0) | |
2198 | return id; | |
55007d84 | 2199 | |
dbcf73e2 | 2200 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2201 | return id; |
2202 | ||
2203 | /* | |
2204 | * There's no space for the new id in memcg_caches arrays, | |
2205 | * so we have to grow them. | |
2206 | */ | |
05257a1a | 2207 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2208 | |
2209 | size = 2 * (id + 1); | |
55007d84 GC |
2210 | if (size < MEMCG_CACHES_MIN_SIZE) |
2211 | size = MEMCG_CACHES_MIN_SIZE; | |
2212 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2213 | size = MEMCG_CACHES_MAX_SIZE; | |
2214 | ||
f3bb3043 | 2215 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2216 | if (!err) |
2217 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2218 | if (!err) |
2219 | memcg_nr_cache_ids = size; | |
2220 | ||
2221 | up_write(&memcg_cache_ids_sem); | |
2222 | ||
f3bb3043 | 2223 | if (err) { |
dbcf73e2 | 2224 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2225 | return err; |
2226 | } | |
2227 | return id; | |
2228 | } | |
2229 | ||
2230 | static void memcg_free_cache_id(int id) | |
2231 | { | |
dbcf73e2 | 2232 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2233 | } |
2234 | ||
d5b3cf71 | 2235 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2236 | struct mem_cgroup *memcg; |
2237 | struct kmem_cache *cachep; | |
2238 | struct work_struct work; | |
2239 | }; | |
2240 | ||
d5b3cf71 | 2241 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2242 | { |
d5b3cf71 VD |
2243 | struct memcg_kmem_cache_create_work *cw = |
2244 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2245 | struct mem_cgroup *memcg = cw->memcg; |
2246 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2247 | |
d5b3cf71 | 2248 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2249 | |
5722d094 | 2250 | css_put(&memcg->css); |
d7f25f8a GC |
2251 | kfree(cw); |
2252 | } | |
2253 | ||
2254 | /* | |
2255 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2256 | */ |
d5b3cf71 VD |
2257 | static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2258 | struct kmem_cache *cachep) | |
d7f25f8a | 2259 | { |
d5b3cf71 | 2260 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2261 | |
776ed0f0 | 2262 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
8135be5a | 2263 | if (!cw) |
d7f25f8a | 2264 | return; |
8135be5a VD |
2265 | |
2266 | css_get(&memcg->css); | |
d7f25f8a GC |
2267 | |
2268 | cw->memcg = memcg; | |
2269 | cw->cachep = cachep; | |
d5b3cf71 | 2270 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2271 | |
d7f25f8a GC |
2272 | schedule_work(&cw->work); |
2273 | } | |
2274 | ||
d5b3cf71 VD |
2275 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2276 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
2277 | { |
2278 | /* | |
2279 | * We need to stop accounting when we kmalloc, because if the | |
2280 | * corresponding kmalloc cache is not yet created, the first allocation | |
d5b3cf71 | 2281 | * in __memcg_schedule_kmem_cache_create will recurse. |
0e9d92f2 GC |
2282 | * |
2283 | * However, it is better to enclose the whole function. Depending on | |
2284 | * the debugging options enabled, INIT_WORK(), for instance, can | |
2285 | * trigger an allocation. This too, will make us recurse. Because at | |
2286 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
2287 | * the safest choice is to do it like this, wrapping the whole function. | |
2288 | */ | |
6f185c29 | 2289 | current->memcg_kmem_skip_account = 1; |
d5b3cf71 | 2290 | __memcg_schedule_kmem_cache_create(memcg, cachep); |
6f185c29 | 2291 | current->memcg_kmem_skip_account = 0; |
0e9d92f2 | 2292 | } |
c67a8a68 | 2293 | |
d7f25f8a GC |
2294 | /* |
2295 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
2296 | * We try to use the current memcg's version of the cache. | |
2297 | * | |
2298 | * If the cache does not exist yet, if we are the first user of it, | |
2299 | * we either create it immediately, if possible, or create it asynchronously | |
2300 | * in a workqueue. | |
2301 | * In the latter case, we will let the current allocation go through with | |
2302 | * the original cache. | |
2303 | * | |
2304 | * Can't be called in interrupt context or from kernel threads. | |
2305 | * This function needs to be called with rcu_read_lock() held. | |
2306 | */ | |
230e9fc2 | 2307 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp) |
d7f25f8a GC |
2308 | { |
2309 | struct mem_cgroup *memcg; | |
959c8963 | 2310 | struct kmem_cache *memcg_cachep; |
2a4db7eb | 2311 | int kmemcg_id; |
d7f25f8a | 2312 | |
f7ce3190 | 2313 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2314 | |
230e9fc2 VD |
2315 | if (cachep->flags & SLAB_ACCOUNT) |
2316 | gfp |= __GFP_ACCOUNT; | |
2317 | ||
2318 | if (!(gfp & __GFP_ACCOUNT)) | |
2319 | return cachep; | |
2320 | ||
9d100c5e | 2321 | if (current->memcg_kmem_skip_account) |
0e9d92f2 GC |
2322 | return cachep; |
2323 | ||
8135be5a | 2324 | memcg = get_mem_cgroup_from_mm(current->mm); |
4db0c3c2 | 2325 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
2a4db7eb | 2326 | if (kmemcg_id < 0) |
ca0dde97 | 2327 | goto out; |
d7f25f8a | 2328 | |
2a4db7eb | 2329 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
8135be5a VD |
2330 | if (likely(memcg_cachep)) |
2331 | return memcg_cachep; | |
ca0dde97 LZ |
2332 | |
2333 | /* | |
2334 | * If we are in a safe context (can wait, and not in interrupt | |
2335 | * context), we could be be predictable and return right away. | |
2336 | * This would guarantee that the allocation being performed | |
2337 | * already belongs in the new cache. | |
2338 | * | |
2339 | * However, there are some clashes that can arrive from locking. | |
2340 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2341 | * memcg_create_kmem_cache, this means no further allocation |
2342 | * could happen with the slab_mutex held. So it's better to | |
2343 | * defer everything. | |
ca0dde97 | 2344 | */ |
d5b3cf71 | 2345 | memcg_schedule_kmem_cache_create(memcg, cachep); |
ca0dde97 | 2346 | out: |
8135be5a | 2347 | css_put(&memcg->css); |
ca0dde97 | 2348 | return cachep; |
d7f25f8a | 2349 | } |
d7f25f8a | 2350 | |
8135be5a VD |
2351 | void __memcg_kmem_put_cache(struct kmem_cache *cachep) |
2352 | { | |
2353 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2354 | css_put(&cachep->memcg_params.memcg->css); |
8135be5a VD |
2355 | } |
2356 | ||
f3ccb2c4 VD |
2357 | int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, |
2358 | struct mem_cgroup *memcg) | |
7ae1e1d0 | 2359 | { |
f3ccb2c4 VD |
2360 | unsigned int nr_pages = 1 << order; |
2361 | struct page_counter *counter; | |
7ae1e1d0 GC |
2362 | int ret; |
2363 | ||
f3ccb2c4 | 2364 | if (!memcg_kmem_is_active(memcg)) |
d05e83a6 | 2365 | return 0; |
6d42c232 | 2366 | |
6071ca52 JW |
2367 | if (!page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) |
2368 | return -ENOMEM; | |
7ae1e1d0 | 2369 | |
f3ccb2c4 VD |
2370 | ret = try_charge(memcg, gfp, nr_pages); |
2371 | if (ret) { | |
2372 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
2373 | return ret; | |
7ae1e1d0 GC |
2374 | } |
2375 | ||
f3ccb2c4 | 2376 | page->mem_cgroup = memcg; |
7ae1e1d0 | 2377 | |
f3ccb2c4 | 2378 | return 0; |
7ae1e1d0 GC |
2379 | } |
2380 | ||
f3ccb2c4 | 2381 | int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 2382 | { |
f3ccb2c4 VD |
2383 | struct mem_cgroup *memcg; |
2384 | int ret; | |
7ae1e1d0 | 2385 | |
f3ccb2c4 VD |
2386 | memcg = get_mem_cgroup_from_mm(current->mm); |
2387 | ret = __memcg_kmem_charge_memcg(page, gfp, order, memcg); | |
7ae1e1d0 | 2388 | css_put(&memcg->css); |
d05e83a6 | 2389 | return ret; |
7ae1e1d0 GC |
2390 | } |
2391 | ||
d05e83a6 | 2392 | void __memcg_kmem_uncharge(struct page *page, int order) |
7ae1e1d0 | 2393 | { |
1306a85a | 2394 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 2395 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 2396 | |
7ae1e1d0 GC |
2397 | if (!memcg) |
2398 | return; | |
2399 | ||
309381fe | 2400 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 2401 | |
f3ccb2c4 VD |
2402 | page_counter_uncharge(&memcg->kmem, nr_pages); |
2403 | page_counter_uncharge(&memcg->memory, nr_pages); | |
7941d214 | 2404 | if (do_memsw_account()) |
f3ccb2c4 | 2405 | page_counter_uncharge(&memcg->memsw, nr_pages); |
60d3fd32 | 2406 | |
1306a85a | 2407 | page->mem_cgroup = NULL; |
f3ccb2c4 | 2408 | css_put_many(&memcg->css, nr_pages); |
60d3fd32 | 2409 | } |
7ae1e1d0 GC |
2410 | #endif /* CONFIG_MEMCG_KMEM */ |
2411 | ||
ca3e0214 KH |
2412 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2413 | ||
ca3e0214 KH |
2414 | /* |
2415 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
2416 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
2417 | * charge/uncharge will be never happen and move_account() is done under | |
2418 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 2419 | */ |
e94c8a9c | 2420 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2421 | { |
e94c8a9c | 2422 | int i; |
ca3e0214 | 2423 | |
3d37c4a9 KH |
2424 | if (mem_cgroup_disabled()) |
2425 | return; | |
b070e65c | 2426 | |
29833315 | 2427 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2428 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2429 | |
1306a85a | 2430 | __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
b070e65c | 2431 | HPAGE_PMD_NR); |
ca3e0214 | 2432 | } |
12d27107 | 2433 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2434 | |
c255a458 | 2435 | #ifdef CONFIG_MEMCG_SWAP |
0a31bc97 JW |
2436 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
2437 | bool charge) | |
d13d1443 | 2438 | { |
0a31bc97 JW |
2439 | int val = (charge) ? 1 : -1; |
2440 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); | |
d13d1443 | 2441 | } |
02491447 DN |
2442 | |
2443 | /** | |
2444 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2445 | * @entry: swap entry to be moved | |
2446 | * @from: mem_cgroup which the entry is moved from | |
2447 | * @to: mem_cgroup which the entry is moved to | |
2448 | * | |
2449 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2450 | * as the mem_cgroup's id of @from. | |
2451 | * | |
2452 | * Returns 0 on success, -EINVAL on failure. | |
2453 | * | |
3e32cb2e | 2454 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2455 | * both res and memsw, and called css_get(). |
2456 | */ | |
2457 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2458 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2459 | { |
2460 | unsigned short old_id, new_id; | |
2461 | ||
34c00c31 LZ |
2462 | old_id = mem_cgroup_id(from); |
2463 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2464 | |
2465 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 2466 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 2467 | mem_cgroup_swap_statistics(to, true); |
02491447 DN |
2468 | return 0; |
2469 | } | |
2470 | return -EINVAL; | |
2471 | } | |
2472 | #else | |
2473 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2474 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2475 | { |
2476 | return -EINVAL; | |
2477 | } | |
8c7c6e34 | 2478 | #endif |
d13d1443 | 2479 | |
3e32cb2e | 2480 | static DEFINE_MUTEX(memcg_limit_mutex); |
f212ad7c | 2481 | |
d38d2a75 | 2482 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2483 | unsigned long limit) |
628f4235 | 2484 | { |
3e32cb2e JW |
2485 | unsigned long curusage; |
2486 | unsigned long oldusage; | |
2487 | bool enlarge = false; | |
81d39c20 | 2488 | int retry_count; |
3e32cb2e | 2489 | int ret; |
81d39c20 KH |
2490 | |
2491 | /* | |
2492 | * For keeping hierarchical_reclaim simple, how long we should retry | |
2493 | * is depends on callers. We set our retry-count to be function | |
2494 | * of # of children which we should visit in this loop. | |
2495 | */ | |
3e32cb2e JW |
2496 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2497 | mem_cgroup_count_children(memcg); | |
81d39c20 | 2498 | |
3e32cb2e | 2499 | oldusage = page_counter_read(&memcg->memory); |
628f4235 | 2500 | |
3e32cb2e | 2501 | do { |
628f4235 KH |
2502 | if (signal_pending(current)) { |
2503 | ret = -EINTR; | |
2504 | break; | |
2505 | } | |
3e32cb2e JW |
2506 | |
2507 | mutex_lock(&memcg_limit_mutex); | |
2508 | if (limit > memcg->memsw.limit) { | |
2509 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2510 | ret = -EINVAL; |
628f4235 KH |
2511 | break; |
2512 | } | |
3e32cb2e JW |
2513 | if (limit > memcg->memory.limit) |
2514 | enlarge = true; | |
2515 | ret = page_counter_limit(&memcg->memory, limit); | |
2516 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2517 | |
2518 | if (!ret) | |
2519 | break; | |
2520 | ||
b70a2a21 JW |
2521 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
2522 | ||
3e32cb2e | 2523 | curusage = page_counter_read(&memcg->memory); |
81d39c20 | 2524 | /* Usage is reduced ? */ |
f894ffa8 | 2525 | if (curusage >= oldusage) |
81d39c20 KH |
2526 | retry_count--; |
2527 | else | |
2528 | oldusage = curusage; | |
3e32cb2e JW |
2529 | } while (retry_count); |
2530 | ||
3c11ecf4 KH |
2531 | if (!ret && enlarge) |
2532 | memcg_oom_recover(memcg); | |
14797e23 | 2533 | |
8c7c6e34 KH |
2534 | return ret; |
2535 | } | |
2536 | ||
338c8431 | 2537 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2538 | unsigned long limit) |
8c7c6e34 | 2539 | { |
3e32cb2e JW |
2540 | unsigned long curusage; |
2541 | unsigned long oldusage; | |
2542 | bool enlarge = false; | |
81d39c20 | 2543 | int retry_count; |
3e32cb2e | 2544 | int ret; |
8c7c6e34 | 2545 | |
81d39c20 | 2546 | /* see mem_cgroup_resize_res_limit */ |
3e32cb2e JW |
2547 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2548 | mem_cgroup_count_children(memcg); | |
2549 | ||
2550 | oldusage = page_counter_read(&memcg->memsw); | |
2551 | ||
2552 | do { | |
8c7c6e34 KH |
2553 | if (signal_pending(current)) { |
2554 | ret = -EINTR; | |
2555 | break; | |
2556 | } | |
3e32cb2e JW |
2557 | |
2558 | mutex_lock(&memcg_limit_mutex); | |
2559 | if (limit < memcg->memory.limit) { | |
2560 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2561 | ret = -EINVAL; |
8c7c6e34 KH |
2562 | break; |
2563 | } | |
3e32cb2e JW |
2564 | if (limit > memcg->memsw.limit) |
2565 | enlarge = true; | |
2566 | ret = page_counter_limit(&memcg->memsw, limit); | |
2567 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2568 | |
2569 | if (!ret) | |
2570 | break; | |
2571 | ||
b70a2a21 JW |
2572 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
2573 | ||
3e32cb2e | 2574 | curusage = page_counter_read(&memcg->memsw); |
81d39c20 | 2575 | /* Usage is reduced ? */ |
8c7c6e34 | 2576 | if (curusage >= oldusage) |
628f4235 | 2577 | retry_count--; |
81d39c20 KH |
2578 | else |
2579 | oldusage = curusage; | |
3e32cb2e JW |
2580 | } while (retry_count); |
2581 | ||
3c11ecf4 KH |
2582 | if (!ret && enlarge) |
2583 | memcg_oom_recover(memcg); | |
3e32cb2e | 2584 | |
628f4235 KH |
2585 | return ret; |
2586 | } | |
2587 | ||
0608f43d AM |
2588 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
2589 | gfp_t gfp_mask, | |
2590 | unsigned long *total_scanned) | |
2591 | { | |
2592 | unsigned long nr_reclaimed = 0; | |
2593 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
2594 | unsigned long reclaimed; | |
2595 | int loop = 0; | |
2596 | struct mem_cgroup_tree_per_zone *mctz; | |
3e32cb2e | 2597 | unsigned long excess; |
0608f43d AM |
2598 | unsigned long nr_scanned; |
2599 | ||
2600 | if (order > 0) | |
2601 | return 0; | |
2602 | ||
2603 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
2604 | /* | |
2605 | * This loop can run a while, specially if mem_cgroup's continuously | |
2606 | * keep exceeding their soft limit and putting the system under | |
2607 | * pressure | |
2608 | */ | |
2609 | do { | |
2610 | if (next_mz) | |
2611 | mz = next_mz; | |
2612 | else | |
2613 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2614 | if (!mz) | |
2615 | break; | |
2616 | ||
2617 | nr_scanned = 0; | |
2618 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
2619 | gfp_mask, &nr_scanned); | |
2620 | nr_reclaimed += reclaimed; | |
2621 | *total_scanned += nr_scanned; | |
0a31bc97 | 2622 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 2623 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
2624 | |
2625 | /* | |
2626 | * If we failed to reclaim anything from this memory cgroup | |
2627 | * it is time to move on to the next cgroup | |
2628 | */ | |
2629 | next_mz = NULL; | |
bc2f2e7f VD |
2630 | if (!reclaimed) |
2631 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
2632 | ||
3e32cb2e | 2633 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
2634 | /* |
2635 | * One school of thought says that we should not add | |
2636 | * back the node to the tree if reclaim returns 0. | |
2637 | * But our reclaim could return 0, simply because due | |
2638 | * to priority we are exposing a smaller subset of | |
2639 | * memory to reclaim from. Consider this as a longer | |
2640 | * term TODO. | |
2641 | */ | |
2642 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 2643 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 2644 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
2645 | css_put(&mz->memcg->css); |
2646 | loop++; | |
2647 | /* | |
2648 | * Could not reclaim anything and there are no more | |
2649 | * mem cgroups to try or we seem to be looping without | |
2650 | * reclaiming anything. | |
2651 | */ | |
2652 | if (!nr_reclaimed && | |
2653 | (next_mz == NULL || | |
2654 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2655 | break; | |
2656 | } while (!nr_reclaimed); | |
2657 | if (next_mz) | |
2658 | css_put(&next_mz->memcg->css); | |
2659 | return nr_reclaimed; | |
2660 | } | |
2661 | ||
ea280e7b TH |
2662 | /* |
2663 | * Test whether @memcg has children, dead or alive. Note that this | |
2664 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
2665 | * returns %true if there are child csses according to the cgroup | |
2666 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
2667 | */ | |
b5f99b53 GC |
2668 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
2669 | { | |
ea280e7b TH |
2670 | bool ret; |
2671 | ||
696ac172 | 2672 | /* |
ea280e7b TH |
2673 | * The lock does not prevent addition or deletion of children, but |
2674 | * it prevents a new child from being initialized based on this | |
2675 | * parent in css_online(), so it's enough to decide whether | |
2676 | * hierarchically inherited attributes can still be changed or not. | |
696ac172 | 2677 | */ |
ea280e7b TH |
2678 | lockdep_assert_held(&memcg_create_mutex); |
2679 | ||
2680 | rcu_read_lock(); | |
2681 | ret = css_next_child(NULL, &memcg->css); | |
2682 | rcu_read_unlock(); | |
2683 | return ret; | |
b5f99b53 GC |
2684 | } |
2685 | ||
c26251f9 MH |
2686 | /* |
2687 | * Reclaims as many pages from the given memcg as possible and moves | |
2688 | * the rest to the parent. | |
2689 | * | |
2690 | * Caller is responsible for holding css reference for memcg. | |
2691 | */ | |
2692 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
2693 | { | |
2694 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 2695 | |
c1e862c1 KH |
2696 | /* we call try-to-free pages for make this cgroup empty */ |
2697 | lru_add_drain_all(); | |
f817ed48 | 2698 | /* try to free all pages in this cgroup */ |
3e32cb2e | 2699 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 2700 | int progress; |
c1e862c1 | 2701 | |
c26251f9 MH |
2702 | if (signal_pending(current)) |
2703 | return -EINTR; | |
2704 | ||
b70a2a21 JW |
2705 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
2706 | GFP_KERNEL, true); | |
c1e862c1 | 2707 | if (!progress) { |
f817ed48 | 2708 | nr_retries--; |
c1e862c1 | 2709 | /* maybe some writeback is necessary */ |
8aa7e847 | 2710 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 2711 | } |
f817ed48 KH |
2712 | |
2713 | } | |
ab5196c2 MH |
2714 | |
2715 | return 0; | |
cc847582 KH |
2716 | } |
2717 | ||
6770c64e TH |
2718 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
2719 | char *buf, size_t nbytes, | |
2720 | loff_t off) | |
c1e862c1 | 2721 | { |
6770c64e | 2722 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 2723 | |
d8423011 MH |
2724 | if (mem_cgroup_is_root(memcg)) |
2725 | return -EINVAL; | |
6770c64e | 2726 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
2727 | } |
2728 | ||
182446d0 TH |
2729 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
2730 | struct cftype *cft) | |
18f59ea7 | 2731 | { |
182446d0 | 2732 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
2733 | } |
2734 | ||
182446d0 TH |
2735 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
2736 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
2737 | { |
2738 | int retval = 0; | |
182446d0 | 2739 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 2740 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 2741 | |
0999821b | 2742 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
2743 | |
2744 | if (memcg->use_hierarchy == val) | |
2745 | goto out; | |
2746 | ||
18f59ea7 | 2747 | /* |
af901ca1 | 2748 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
2749 | * in the child subtrees. If it is unset, then the change can |
2750 | * occur, provided the current cgroup has no children. | |
2751 | * | |
2752 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
2753 | * set if there are no children. | |
2754 | */ | |
c0ff4b85 | 2755 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 2756 | (val == 1 || val == 0)) { |
ea280e7b | 2757 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 2758 | memcg->use_hierarchy = val; |
18f59ea7 BS |
2759 | else |
2760 | retval = -EBUSY; | |
2761 | } else | |
2762 | retval = -EINVAL; | |
567fb435 GC |
2763 | |
2764 | out: | |
0999821b | 2765 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
2766 | |
2767 | return retval; | |
2768 | } | |
2769 | ||
3e32cb2e JW |
2770 | static unsigned long tree_stat(struct mem_cgroup *memcg, |
2771 | enum mem_cgroup_stat_index idx) | |
ce00a967 JW |
2772 | { |
2773 | struct mem_cgroup *iter; | |
484ebb3b | 2774 | unsigned long val = 0; |
ce00a967 | 2775 | |
ce00a967 JW |
2776 | for_each_mem_cgroup_tree(iter, memcg) |
2777 | val += mem_cgroup_read_stat(iter, idx); | |
2778 | ||
ce00a967 JW |
2779 | return val; |
2780 | } | |
2781 | ||
6f646156 | 2782 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 2783 | { |
c12176d3 | 2784 | unsigned long val; |
ce00a967 | 2785 | |
3e32cb2e JW |
2786 | if (mem_cgroup_is_root(memcg)) { |
2787 | val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE); | |
2788 | val += tree_stat(memcg, MEM_CGROUP_STAT_RSS); | |
2789 | if (swap) | |
2790 | val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP); | |
2791 | } else { | |
ce00a967 | 2792 | if (!swap) |
3e32cb2e | 2793 | val = page_counter_read(&memcg->memory); |
ce00a967 | 2794 | else |
3e32cb2e | 2795 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 2796 | } |
c12176d3 | 2797 | return val; |
ce00a967 JW |
2798 | } |
2799 | ||
3e32cb2e JW |
2800 | enum { |
2801 | RES_USAGE, | |
2802 | RES_LIMIT, | |
2803 | RES_MAX_USAGE, | |
2804 | RES_FAILCNT, | |
2805 | RES_SOFT_LIMIT, | |
2806 | }; | |
ce00a967 | 2807 | |
791badbd | 2808 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 2809 | struct cftype *cft) |
8cdea7c0 | 2810 | { |
182446d0 | 2811 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 2812 | struct page_counter *counter; |
af36f906 | 2813 | |
3e32cb2e | 2814 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 2815 | case _MEM: |
3e32cb2e JW |
2816 | counter = &memcg->memory; |
2817 | break; | |
8c7c6e34 | 2818 | case _MEMSWAP: |
3e32cb2e JW |
2819 | counter = &memcg->memsw; |
2820 | break; | |
510fc4e1 | 2821 | case _KMEM: |
3e32cb2e | 2822 | counter = &memcg->kmem; |
510fc4e1 | 2823 | break; |
8c7c6e34 KH |
2824 | default: |
2825 | BUG(); | |
8c7c6e34 | 2826 | } |
3e32cb2e JW |
2827 | |
2828 | switch (MEMFILE_ATTR(cft->private)) { | |
2829 | case RES_USAGE: | |
2830 | if (counter == &memcg->memory) | |
c12176d3 | 2831 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 2832 | if (counter == &memcg->memsw) |
c12176d3 | 2833 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
2834 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
2835 | case RES_LIMIT: | |
2836 | return (u64)counter->limit * PAGE_SIZE; | |
2837 | case RES_MAX_USAGE: | |
2838 | return (u64)counter->watermark * PAGE_SIZE; | |
2839 | case RES_FAILCNT: | |
2840 | return counter->failcnt; | |
2841 | case RES_SOFT_LIMIT: | |
2842 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
2843 | default: | |
2844 | BUG(); | |
2845 | } | |
8cdea7c0 | 2846 | } |
510fc4e1 | 2847 | |
510fc4e1 | 2848 | #ifdef CONFIG_MEMCG_KMEM |
8c0145b6 VD |
2849 | static int memcg_activate_kmem(struct mem_cgroup *memcg, |
2850 | unsigned long nr_pages) | |
d6441637 VD |
2851 | { |
2852 | int err = 0; | |
2853 | int memcg_id; | |
2854 | ||
2a4db7eb | 2855 | BUG_ON(memcg->kmemcg_id >= 0); |
2788cf0c | 2856 | BUG_ON(memcg->kmem_acct_activated); |
2a4db7eb | 2857 | BUG_ON(memcg->kmem_acct_active); |
d6441637 | 2858 | |
510fc4e1 GC |
2859 | /* |
2860 | * For simplicity, we won't allow this to be disabled. It also can't | |
2861 | * be changed if the cgroup has children already, or if tasks had | |
2862 | * already joined. | |
2863 | * | |
2864 | * If tasks join before we set the limit, a person looking at | |
2865 | * kmem.usage_in_bytes will have no way to determine when it took | |
2866 | * place, which makes the value quite meaningless. | |
2867 | * | |
2868 | * After it first became limited, changes in the value of the limit are | |
2869 | * of course permitted. | |
510fc4e1 | 2870 | */ |
0999821b | 2871 | mutex_lock(&memcg_create_mutex); |
27bd4dbb | 2872 | if (cgroup_is_populated(memcg->css.cgroup) || |
ea280e7b | 2873 | (memcg->use_hierarchy && memcg_has_children(memcg))) |
d6441637 VD |
2874 | err = -EBUSY; |
2875 | mutex_unlock(&memcg_create_mutex); | |
2876 | if (err) | |
2877 | goto out; | |
510fc4e1 | 2878 | |
f3bb3043 | 2879 | memcg_id = memcg_alloc_cache_id(); |
d6441637 VD |
2880 | if (memcg_id < 0) { |
2881 | err = memcg_id; | |
2882 | goto out; | |
2883 | } | |
2884 | ||
d6441637 | 2885 | /* |
900a38f0 VD |
2886 | * We couldn't have accounted to this cgroup, because it hasn't got |
2887 | * activated yet, so this should succeed. | |
d6441637 | 2888 | */ |
3e32cb2e | 2889 | err = page_counter_limit(&memcg->kmem, nr_pages); |
d6441637 VD |
2890 | VM_BUG_ON(err); |
2891 | ||
2892 | static_key_slow_inc(&memcg_kmem_enabled_key); | |
2893 | /* | |
900a38f0 VD |
2894 | * A memory cgroup is considered kmem-active as soon as it gets |
2895 | * kmemcg_id. Setting the id after enabling static branching will | |
d6441637 VD |
2896 | * guarantee no one starts accounting before all call sites are |
2897 | * patched. | |
2898 | */ | |
900a38f0 | 2899 | memcg->kmemcg_id = memcg_id; |
2788cf0c | 2900 | memcg->kmem_acct_activated = true; |
2a4db7eb | 2901 | memcg->kmem_acct_active = true; |
510fc4e1 | 2902 | out: |
d6441637 | 2903 | return err; |
d6441637 VD |
2904 | } |
2905 | ||
d6441637 | 2906 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2907 | unsigned long limit) |
d6441637 VD |
2908 | { |
2909 | int ret; | |
2910 | ||
3e32cb2e | 2911 | mutex_lock(&memcg_limit_mutex); |
d6441637 | 2912 | if (!memcg_kmem_is_active(memcg)) |
3e32cb2e | 2913 | ret = memcg_activate_kmem(memcg, limit); |
d6441637 | 2914 | else |
3e32cb2e JW |
2915 | ret = page_counter_limit(&memcg->kmem, limit); |
2916 | mutex_unlock(&memcg_limit_mutex); | |
510fc4e1 GC |
2917 | return ret; |
2918 | } | |
2919 | ||
55007d84 | 2920 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 2921 | { |
55007d84 | 2922 | int ret = 0; |
510fc4e1 | 2923 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
55007d84 | 2924 | |
d6441637 VD |
2925 | if (!parent) |
2926 | return 0; | |
55007d84 | 2927 | |
8c0145b6 | 2928 | mutex_lock(&memcg_limit_mutex); |
55007d84 | 2929 | /* |
d6441637 VD |
2930 | * If the parent cgroup is not kmem-active now, it cannot be activated |
2931 | * after this point, because it has at least one child already. | |
55007d84 | 2932 | */ |
d6441637 | 2933 | if (memcg_kmem_is_active(parent)) |
8c0145b6 VD |
2934 | ret = memcg_activate_kmem(memcg, PAGE_COUNTER_MAX); |
2935 | mutex_unlock(&memcg_limit_mutex); | |
55007d84 | 2936 | return ret; |
510fc4e1 | 2937 | } |
d6441637 VD |
2938 | #else |
2939 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
3e32cb2e | 2940 | unsigned long limit) |
d6441637 VD |
2941 | { |
2942 | return -EINVAL; | |
2943 | } | |
6d043990 | 2944 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 2945 | |
628f4235 KH |
2946 | /* |
2947 | * The user of this function is... | |
2948 | * RES_LIMIT. | |
2949 | */ | |
451af504 TH |
2950 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
2951 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 2952 | { |
451af504 | 2953 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 2954 | unsigned long nr_pages; |
628f4235 KH |
2955 | int ret; |
2956 | ||
451af504 | 2957 | buf = strstrip(buf); |
650c5e56 | 2958 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
2959 | if (ret) |
2960 | return ret; | |
af36f906 | 2961 | |
3e32cb2e | 2962 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 2963 | case RES_LIMIT: |
4b3bde4c BS |
2964 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
2965 | ret = -EINVAL; | |
2966 | break; | |
2967 | } | |
3e32cb2e JW |
2968 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
2969 | case _MEM: | |
2970 | ret = mem_cgroup_resize_limit(memcg, nr_pages); | |
8c7c6e34 | 2971 | break; |
3e32cb2e JW |
2972 | case _MEMSWAP: |
2973 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); | |
296c81d8 | 2974 | break; |
3e32cb2e JW |
2975 | case _KMEM: |
2976 | ret = memcg_update_kmem_limit(memcg, nr_pages); | |
2977 | break; | |
2978 | } | |
296c81d8 | 2979 | break; |
3e32cb2e JW |
2980 | case RES_SOFT_LIMIT: |
2981 | memcg->soft_limit = nr_pages; | |
2982 | ret = 0; | |
628f4235 KH |
2983 | break; |
2984 | } | |
451af504 | 2985 | return ret ?: nbytes; |
8cdea7c0 BS |
2986 | } |
2987 | ||
6770c64e TH |
2988 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
2989 | size_t nbytes, loff_t off) | |
c84872e1 | 2990 | { |
6770c64e | 2991 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 2992 | struct page_counter *counter; |
c84872e1 | 2993 | |
3e32cb2e JW |
2994 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
2995 | case _MEM: | |
2996 | counter = &memcg->memory; | |
2997 | break; | |
2998 | case _MEMSWAP: | |
2999 | counter = &memcg->memsw; | |
3000 | break; | |
3001 | case _KMEM: | |
3002 | counter = &memcg->kmem; | |
3003 | break; | |
3004 | default: | |
3005 | BUG(); | |
3006 | } | |
af36f906 | 3007 | |
3e32cb2e | 3008 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3009 | case RES_MAX_USAGE: |
3e32cb2e | 3010 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3011 | break; |
3012 | case RES_FAILCNT: | |
3e32cb2e | 3013 | counter->failcnt = 0; |
29f2a4da | 3014 | break; |
3e32cb2e JW |
3015 | default: |
3016 | BUG(); | |
29f2a4da | 3017 | } |
f64c3f54 | 3018 | |
6770c64e | 3019 | return nbytes; |
c84872e1 PE |
3020 | } |
3021 | ||
182446d0 | 3022 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3023 | struct cftype *cft) |
3024 | { | |
182446d0 | 3025 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3026 | } |
3027 | ||
02491447 | 3028 | #ifdef CONFIG_MMU |
182446d0 | 3029 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3030 | struct cftype *cft, u64 val) |
3031 | { | |
182446d0 | 3032 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3033 | |
1dfab5ab | 3034 | if (val & ~MOVE_MASK) |
7dc74be0 | 3035 | return -EINVAL; |
ee5e8472 | 3036 | |
7dc74be0 | 3037 | /* |
ee5e8472 GC |
3038 | * No kind of locking is needed in here, because ->can_attach() will |
3039 | * check this value once in the beginning of the process, and then carry | |
3040 | * on with stale data. This means that changes to this value will only | |
3041 | * affect task migrations starting after the change. | |
7dc74be0 | 3042 | */ |
c0ff4b85 | 3043 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3044 | return 0; |
3045 | } | |
02491447 | 3046 | #else |
182446d0 | 3047 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3048 | struct cftype *cft, u64 val) |
3049 | { | |
3050 | return -ENOSYS; | |
3051 | } | |
3052 | #endif | |
7dc74be0 | 3053 | |
406eb0c9 | 3054 | #ifdef CONFIG_NUMA |
2da8ca82 | 3055 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3056 | { |
25485de6 GT |
3057 | struct numa_stat { |
3058 | const char *name; | |
3059 | unsigned int lru_mask; | |
3060 | }; | |
3061 | ||
3062 | static const struct numa_stat stats[] = { | |
3063 | { "total", LRU_ALL }, | |
3064 | { "file", LRU_ALL_FILE }, | |
3065 | { "anon", LRU_ALL_ANON }, | |
3066 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3067 | }; | |
3068 | const struct numa_stat *stat; | |
406eb0c9 | 3069 | int nid; |
25485de6 | 3070 | unsigned long nr; |
2da8ca82 | 3071 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 3072 | |
25485de6 GT |
3073 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3074 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3075 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3076 | for_each_node_state(nid, N_MEMORY) { | |
3077 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3078 | stat->lru_mask); | |
3079 | seq_printf(m, " N%d=%lu", nid, nr); | |
3080 | } | |
3081 | seq_putc(m, '\n'); | |
406eb0c9 | 3082 | } |
406eb0c9 | 3083 | |
071aee13 YH |
3084 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3085 | struct mem_cgroup *iter; | |
3086 | ||
3087 | nr = 0; | |
3088 | for_each_mem_cgroup_tree(iter, memcg) | |
3089 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3090 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3091 | for_each_node_state(nid, N_MEMORY) { | |
3092 | nr = 0; | |
3093 | for_each_mem_cgroup_tree(iter, memcg) | |
3094 | nr += mem_cgroup_node_nr_lru_pages( | |
3095 | iter, nid, stat->lru_mask); | |
3096 | seq_printf(m, " N%d=%lu", nid, nr); | |
3097 | } | |
3098 | seq_putc(m, '\n'); | |
406eb0c9 | 3099 | } |
406eb0c9 | 3100 | |
406eb0c9 YH |
3101 | return 0; |
3102 | } | |
3103 | #endif /* CONFIG_NUMA */ | |
3104 | ||
2da8ca82 | 3105 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3106 | { |
2da8ca82 | 3107 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 3108 | unsigned long memory, memsw; |
af7c4b0e JW |
3109 | struct mem_cgroup *mi; |
3110 | unsigned int i; | |
406eb0c9 | 3111 | |
0ca44b14 GT |
3112 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) != |
3113 | MEM_CGROUP_STAT_NSTATS); | |
3114 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) != | |
3115 | MEM_CGROUP_EVENTS_NSTATS); | |
70bc068c RS |
3116 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3117 | ||
af7c4b0e | 3118 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
7941d214 | 3119 | if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account()) |
1dd3a273 | 3120 | continue; |
484ebb3b | 3121 | seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i], |
af7c4b0e | 3122 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); |
1dd3a273 | 3123 | } |
7b854121 | 3124 | |
af7c4b0e JW |
3125 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
3126 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
3127 | mem_cgroup_read_events(memcg, i)); | |
3128 | ||
3129 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3130 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
3131 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
3132 | ||
14067bb3 | 3133 | /* Hierarchical information */ |
3e32cb2e JW |
3134 | memory = memsw = PAGE_COUNTER_MAX; |
3135 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
3136 | memory = min(memory, mi->memory.limit); | |
3137 | memsw = min(memsw, mi->memsw.limit); | |
fee7b548 | 3138 | } |
3e32cb2e JW |
3139 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3140 | (u64)memory * PAGE_SIZE); | |
7941d214 | 3141 | if (do_memsw_account()) |
3e32cb2e JW |
3142 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
3143 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3144 | |
af7c4b0e | 3145 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
484ebb3b | 3146 | unsigned long long val = 0; |
af7c4b0e | 3147 | |
7941d214 | 3148 | if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account()) |
1dd3a273 | 3149 | continue; |
af7c4b0e JW |
3150 | for_each_mem_cgroup_tree(mi, memcg) |
3151 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
484ebb3b | 3152 | seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val); |
af7c4b0e JW |
3153 | } |
3154 | ||
3155 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
3156 | unsigned long long val = 0; | |
3157 | ||
3158 | for_each_mem_cgroup_tree(mi, memcg) | |
3159 | val += mem_cgroup_read_events(mi, i); | |
3160 | seq_printf(m, "total_%s %llu\n", | |
3161 | mem_cgroup_events_names[i], val); | |
3162 | } | |
3163 | ||
3164 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
3165 | unsigned long long val = 0; | |
3166 | ||
3167 | for_each_mem_cgroup_tree(mi, memcg) | |
3168 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
3169 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 3170 | } |
14067bb3 | 3171 | |
7f016ee8 | 3172 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
3173 | { |
3174 | int nid, zid; | |
3175 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 3176 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3177 | unsigned long recent_rotated[2] = {0, 0}; |
3178 | unsigned long recent_scanned[2] = {0, 0}; | |
3179 | ||
3180 | for_each_online_node(nid) | |
3181 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
e231875b | 3182 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
89abfab1 | 3183 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 3184 | |
89abfab1 HD |
3185 | recent_rotated[0] += rstat->recent_rotated[0]; |
3186 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3187 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3188 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 3189 | } |
78ccf5b5 JW |
3190 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3191 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3192 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3193 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3194 | } |
3195 | #endif | |
3196 | ||
d2ceb9b7 KH |
3197 | return 0; |
3198 | } | |
3199 | ||
182446d0 TH |
3200 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3201 | struct cftype *cft) | |
a7885eb8 | 3202 | { |
182446d0 | 3203 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3204 | |
1f4c025b | 3205 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3206 | } |
3207 | ||
182446d0 TH |
3208 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3209 | struct cftype *cft, u64 val) | |
a7885eb8 | 3210 | { |
182446d0 | 3211 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3212 | |
3dae7fec | 3213 | if (val > 100) |
a7885eb8 KM |
3214 | return -EINVAL; |
3215 | ||
14208b0e | 3216 | if (css->parent) |
3dae7fec JW |
3217 | memcg->swappiness = val; |
3218 | else | |
3219 | vm_swappiness = val; | |
068b38c1 | 3220 | |
a7885eb8 KM |
3221 | return 0; |
3222 | } | |
3223 | ||
2e72b634 KS |
3224 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3225 | { | |
3226 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3227 | unsigned long usage; |
2e72b634 KS |
3228 | int i; |
3229 | ||
3230 | rcu_read_lock(); | |
3231 | if (!swap) | |
2c488db2 | 3232 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3233 | else |
2c488db2 | 3234 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3235 | |
3236 | if (!t) | |
3237 | goto unlock; | |
3238 | ||
ce00a967 | 3239 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3240 | |
3241 | /* | |
748dad36 | 3242 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3243 | * If it's not true, a threshold was crossed after last |
3244 | * call of __mem_cgroup_threshold(). | |
3245 | */ | |
5407a562 | 3246 | i = t->current_threshold; |
2e72b634 KS |
3247 | |
3248 | /* | |
3249 | * Iterate backward over array of thresholds starting from | |
3250 | * current_threshold and check if a threshold is crossed. | |
3251 | * If none of thresholds below usage is crossed, we read | |
3252 | * only one element of the array here. | |
3253 | */ | |
3254 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3255 | eventfd_signal(t->entries[i].eventfd, 1); | |
3256 | ||
3257 | /* i = current_threshold + 1 */ | |
3258 | i++; | |
3259 | ||
3260 | /* | |
3261 | * Iterate forward over array of thresholds starting from | |
3262 | * current_threshold+1 and check if a threshold is crossed. | |
3263 | * If none of thresholds above usage is crossed, we read | |
3264 | * only one element of the array here. | |
3265 | */ | |
3266 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3267 | eventfd_signal(t->entries[i].eventfd, 1); | |
3268 | ||
3269 | /* Update current_threshold */ | |
5407a562 | 3270 | t->current_threshold = i - 1; |
2e72b634 KS |
3271 | unlock: |
3272 | rcu_read_unlock(); | |
3273 | } | |
3274 | ||
3275 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3276 | { | |
ad4ca5f4 KS |
3277 | while (memcg) { |
3278 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 3279 | if (do_memsw_account()) |
ad4ca5f4 KS |
3280 | __mem_cgroup_threshold(memcg, true); |
3281 | ||
3282 | memcg = parent_mem_cgroup(memcg); | |
3283 | } | |
2e72b634 KS |
3284 | } |
3285 | ||
3286 | static int compare_thresholds(const void *a, const void *b) | |
3287 | { | |
3288 | const struct mem_cgroup_threshold *_a = a; | |
3289 | const struct mem_cgroup_threshold *_b = b; | |
3290 | ||
2bff24a3 GT |
3291 | if (_a->threshold > _b->threshold) |
3292 | return 1; | |
3293 | ||
3294 | if (_a->threshold < _b->threshold) | |
3295 | return -1; | |
3296 | ||
3297 | return 0; | |
2e72b634 KS |
3298 | } |
3299 | ||
c0ff4b85 | 3300 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3301 | { |
3302 | struct mem_cgroup_eventfd_list *ev; | |
3303 | ||
2bcf2e92 MH |
3304 | spin_lock(&memcg_oom_lock); |
3305 | ||
c0ff4b85 | 3306 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3307 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3308 | |
3309 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3310 | return 0; |
3311 | } | |
3312 | ||
c0ff4b85 | 3313 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3314 | { |
7d74b06f KH |
3315 | struct mem_cgroup *iter; |
3316 | ||
c0ff4b85 | 3317 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3318 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3319 | } |
3320 | ||
59b6f873 | 3321 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3322 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3323 | { |
2c488db2 KS |
3324 | struct mem_cgroup_thresholds *thresholds; |
3325 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3326 | unsigned long threshold; |
3327 | unsigned long usage; | |
2c488db2 | 3328 | int i, size, ret; |
2e72b634 | 3329 | |
650c5e56 | 3330 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
3331 | if (ret) |
3332 | return ret; | |
3333 | ||
3334 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3335 | |
05b84301 | 3336 | if (type == _MEM) { |
2c488db2 | 3337 | thresholds = &memcg->thresholds; |
ce00a967 | 3338 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3339 | } else if (type == _MEMSWAP) { |
2c488db2 | 3340 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3341 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3342 | } else |
2e72b634 KS |
3343 | BUG(); |
3344 | ||
2e72b634 | 3345 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3346 | if (thresholds->primary) |
2e72b634 KS |
3347 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3348 | ||
2c488db2 | 3349 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3350 | |
3351 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 3352 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 3353 | GFP_KERNEL); |
2c488db2 | 3354 | if (!new) { |
2e72b634 KS |
3355 | ret = -ENOMEM; |
3356 | goto unlock; | |
3357 | } | |
2c488db2 | 3358 | new->size = size; |
2e72b634 KS |
3359 | |
3360 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3361 | if (thresholds->primary) { |
3362 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3363 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3364 | } |
3365 | ||
2e72b634 | 3366 | /* Add new threshold */ |
2c488db2 KS |
3367 | new->entries[size - 1].eventfd = eventfd; |
3368 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3369 | |
3370 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3371 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3372 | compare_thresholds, NULL); |
3373 | ||
3374 | /* Find current threshold */ | |
2c488db2 | 3375 | new->current_threshold = -1; |
2e72b634 | 3376 | for (i = 0; i < size; i++) { |
748dad36 | 3377 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3378 | /* |
2c488db2 KS |
3379 | * new->current_threshold will not be used until |
3380 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3381 | * it here. |
3382 | */ | |
2c488db2 | 3383 | ++new->current_threshold; |
748dad36 SZ |
3384 | } else |
3385 | break; | |
2e72b634 KS |
3386 | } |
3387 | ||
2c488db2 KS |
3388 | /* Free old spare buffer and save old primary buffer as spare */ |
3389 | kfree(thresholds->spare); | |
3390 | thresholds->spare = thresholds->primary; | |
3391 | ||
3392 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3393 | |
907860ed | 3394 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3395 | synchronize_rcu(); |
3396 | ||
2e72b634 KS |
3397 | unlock: |
3398 | mutex_unlock(&memcg->thresholds_lock); | |
3399 | ||
3400 | return ret; | |
3401 | } | |
3402 | ||
59b6f873 | 3403 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3404 | struct eventfd_ctx *eventfd, const char *args) |
3405 | { | |
59b6f873 | 3406 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3407 | } |
3408 | ||
59b6f873 | 3409 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3410 | struct eventfd_ctx *eventfd, const char *args) |
3411 | { | |
59b6f873 | 3412 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3413 | } |
3414 | ||
59b6f873 | 3415 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3416 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3417 | { |
2c488db2 KS |
3418 | struct mem_cgroup_thresholds *thresholds; |
3419 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3420 | unsigned long usage; |
2c488db2 | 3421 | int i, j, size; |
2e72b634 KS |
3422 | |
3423 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3424 | |
3425 | if (type == _MEM) { | |
2c488db2 | 3426 | thresholds = &memcg->thresholds; |
ce00a967 | 3427 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3428 | } else if (type == _MEMSWAP) { |
2c488db2 | 3429 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3430 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3431 | } else |
2e72b634 KS |
3432 | BUG(); |
3433 | ||
371528ca AV |
3434 | if (!thresholds->primary) |
3435 | goto unlock; | |
3436 | ||
2e72b634 KS |
3437 | /* Check if a threshold crossed before removing */ |
3438 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3439 | ||
3440 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3441 | size = 0; |
3442 | for (i = 0; i < thresholds->primary->size; i++) { | |
3443 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3444 | size++; |
3445 | } | |
3446 | ||
2c488db2 | 3447 | new = thresholds->spare; |
907860ed | 3448 | |
2e72b634 KS |
3449 | /* Set thresholds array to NULL if we don't have thresholds */ |
3450 | if (!size) { | |
2c488db2 KS |
3451 | kfree(new); |
3452 | new = NULL; | |
907860ed | 3453 | goto swap_buffers; |
2e72b634 KS |
3454 | } |
3455 | ||
2c488db2 | 3456 | new->size = size; |
2e72b634 KS |
3457 | |
3458 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3459 | new->current_threshold = -1; |
3460 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3461 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3462 | continue; |
3463 | ||
2c488db2 | 3464 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3465 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3466 | /* |
2c488db2 | 3467 | * new->current_threshold will not be used |
2e72b634 KS |
3468 | * until rcu_assign_pointer(), so it's safe to increment |
3469 | * it here. | |
3470 | */ | |
2c488db2 | 3471 | ++new->current_threshold; |
2e72b634 KS |
3472 | } |
3473 | j++; | |
3474 | } | |
3475 | ||
907860ed | 3476 | swap_buffers: |
2c488db2 KS |
3477 | /* Swap primary and spare array */ |
3478 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
3479 | /* If all events are unregistered, free the spare array */ |
3480 | if (!new) { | |
3481 | kfree(thresholds->spare); | |
3482 | thresholds->spare = NULL; | |
3483 | } | |
3484 | ||
2c488db2 | 3485 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3486 | |
907860ed | 3487 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3488 | synchronize_rcu(); |
371528ca | 3489 | unlock: |
2e72b634 | 3490 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3491 | } |
c1e862c1 | 3492 | |
59b6f873 | 3493 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3494 | struct eventfd_ctx *eventfd) |
3495 | { | |
59b6f873 | 3496 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
3497 | } |
3498 | ||
59b6f873 | 3499 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3500 | struct eventfd_ctx *eventfd) |
3501 | { | |
59b6f873 | 3502 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
3503 | } |
3504 | ||
59b6f873 | 3505 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3506 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 3507 | { |
9490ff27 | 3508 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 3509 | |
9490ff27 KH |
3510 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
3511 | if (!event) | |
3512 | return -ENOMEM; | |
3513 | ||
1af8efe9 | 3514 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
3515 | |
3516 | event->eventfd = eventfd; | |
3517 | list_add(&event->list, &memcg->oom_notify); | |
3518 | ||
3519 | /* already in OOM ? */ | |
c2b42d3c | 3520 | if (memcg->under_oom) |
9490ff27 | 3521 | eventfd_signal(eventfd, 1); |
1af8efe9 | 3522 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3523 | |
3524 | return 0; | |
3525 | } | |
3526 | ||
59b6f873 | 3527 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3528 | struct eventfd_ctx *eventfd) |
9490ff27 | 3529 | { |
9490ff27 | 3530 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 3531 | |
1af8efe9 | 3532 | spin_lock(&memcg_oom_lock); |
9490ff27 | 3533 | |
c0ff4b85 | 3534 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
3535 | if (ev->eventfd == eventfd) { |
3536 | list_del(&ev->list); | |
3537 | kfree(ev); | |
3538 | } | |
3539 | } | |
3540 | ||
1af8efe9 | 3541 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3542 | } |
3543 | ||
2da8ca82 | 3544 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 3545 | { |
2da8ca82 | 3546 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 3547 | |
791badbd | 3548 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 3549 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
3c11ecf4 KH |
3550 | return 0; |
3551 | } | |
3552 | ||
182446d0 | 3553 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
3554 | struct cftype *cft, u64 val) |
3555 | { | |
182446d0 | 3556 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
3557 | |
3558 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 3559 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
3560 | return -EINVAL; |
3561 | ||
c0ff4b85 | 3562 | memcg->oom_kill_disable = val; |
4d845ebf | 3563 | if (!val) |
c0ff4b85 | 3564 | memcg_oom_recover(memcg); |
3dae7fec | 3565 | |
3c11ecf4 KH |
3566 | return 0; |
3567 | } | |
3568 | ||
c255a458 | 3569 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 3570 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 3571 | { |
55007d84 GC |
3572 | int ret; |
3573 | ||
55007d84 GC |
3574 | ret = memcg_propagate_kmem(memcg); |
3575 | if (ret) | |
3576 | return ret; | |
2633d7a0 | 3577 | |
baac50bb | 3578 | return tcp_init_cgroup(memcg, ss); |
573b400d | 3579 | } |
e5671dfa | 3580 | |
2a4db7eb VD |
3581 | static void memcg_deactivate_kmem(struct mem_cgroup *memcg) |
3582 | { | |
2788cf0c VD |
3583 | struct cgroup_subsys_state *css; |
3584 | struct mem_cgroup *parent, *child; | |
3585 | int kmemcg_id; | |
3586 | ||
2a4db7eb VD |
3587 | if (!memcg->kmem_acct_active) |
3588 | return; | |
3589 | ||
3590 | /* | |
3591 | * Clear the 'active' flag before clearing memcg_caches arrays entries. | |
3592 | * Since we take the slab_mutex in memcg_deactivate_kmem_caches(), it | |
3593 | * guarantees no cache will be created for this cgroup after we are | |
3594 | * done (see memcg_create_kmem_cache()). | |
3595 | */ | |
3596 | memcg->kmem_acct_active = false; | |
3597 | ||
3598 | memcg_deactivate_kmem_caches(memcg); | |
2788cf0c VD |
3599 | |
3600 | kmemcg_id = memcg->kmemcg_id; | |
3601 | BUG_ON(kmemcg_id < 0); | |
3602 | ||
3603 | parent = parent_mem_cgroup(memcg); | |
3604 | if (!parent) | |
3605 | parent = root_mem_cgroup; | |
3606 | ||
3607 | /* | |
3608 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3609 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3610 | * to ones of the parent. After we have finished, all list_lrus | |
3611 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3612 | * ordering is imposed by list_lru_node->lock taken by | |
3613 | * memcg_drain_all_list_lrus(). | |
3614 | */ | |
3615 | css_for_each_descendant_pre(css, &memcg->css) { | |
3616 | child = mem_cgroup_from_css(css); | |
3617 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3618 | child->kmemcg_id = parent->kmemcg_id; | |
3619 | if (!memcg->use_hierarchy) | |
3620 | break; | |
3621 | } | |
3622 | memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id); | |
3623 | ||
3624 | memcg_free_cache_id(kmemcg_id); | |
2a4db7eb VD |
3625 | } |
3626 | ||
10d5ebf4 | 3627 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 3628 | { |
f48b80a5 VD |
3629 | if (memcg->kmem_acct_activated) { |
3630 | memcg_destroy_kmem_caches(memcg); | |
3631 | static_key_slow_dec(&memcg_kmem_enabled_key); | |
3632 | WARN_ON(page_counter_read(&memcg->kmem)); | |
3633 | } | |
baac50bb | 3634 | tcp_destroy_cgroup(memcg); |
10d5ebf4 | 3635 | } |
e5671dfa | 3636 | #else |
cbe128e3 | 3637 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
3638 | { |
3639 | return 0; | |
3640 | } | |
d1a4c0b3 | 3641 | |
2a4db7eb VD |
3642 | static void memcg_deactivate_kmem(struct mem_cgroup *memcg) |
3643 | { | |
3644 | } | |
3645 | ||
10d5ebf4 LZ |
3646 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
3647 | { | |
3648 | } | |
e5671dfa GC |
3649 | #endif |
3650 | ||
52ebea74 TH |
3651 | #ifdef CONFIG_CGROUP_WRITEBACK |
3652 | ||
3653 | struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg) | |
3654 | { | |
3655 | return &memcg->cgwb_list; | |
3656 | } | |
3657 | ||
841710aa TH |
3658 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
3659 | { | |
3660 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
3661 | } | |
3662 | ||
3663 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3664 | { | |
3665 | wb_domain_exit(&memcg->cgwb_domain); | |
3666 | } | |
3667 | ||
2529bb3a TH |
3668 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3669 | { | |
3670 | wb_domain_size_changed(&memcg->cgwb_domain); | |
3671 | } | |
3672 | ||
841710aa TH |
3673 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
3674 | { | |
3675 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3676 | ||
3677 | if (!memcg->css.parent) | |
3678 | return NULL; | |
3679 | ||
3680 | return &memcg->cgwb_domain; | |
3681 | } | |
3682 | ||
c2aa723a TH |
3683 | /** |
3684 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
3685 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
3686 | * @pfilepages: out parameter for number of file pages |
3687 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
3688 | * @pdirty: out parameter for number of dirty pages |
3689 | * @pwriteback: out parameter for number of pages under writeback | |
3690 | * | |
c5edf9cd TH |
3691 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
3692 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
3693 | * is a bit more involved. | |
c2aa723a | 3694 | * |
c5edf9cd TH |
3695 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
3696 | * headroom is calculated as the lowest headroom of itself and the | |
3697 | * ancestors. Note that this doesn't consider the actual amount of | |
3698 | * available memory in the system. The caller should further cap | |
3699 | * *@pheadroom accordingly. | |
c2aa723a | 3700 | */ |
c5edf9cd TH |
3701 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
3702 | unsigned long *pheadroom, unsigned long *pdirty, | |
3703 | unsigned long *pwriteback) | |
c2aa723a TH |
3704 | { |
3705 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3706 | struct mem_cgroup *parent; | |
c2aa723a TH |
3707 | |
3708 | *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY); | |
3709 | ||
3710 | /* this should eventually include NR_UNSTABLE_NFS */ | |
3711 | *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK); | |
c5edf9cd TH |
3712 | *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) | |
3713 | (1 << LRU_ACTIVE_FILE)); | |
3714 | *pheadroom = PAGE_COUNTER_MAX; | |
c2aa723a | 3715 | |
c2aa723a TH |
3716 | while ((parent = parent_mem_cgroup(memcg))) { |
3717 | unsigned long ceiling = min(memcg->memory.limit, memcg->high); | |
3718 | unsigned long used = page_counter_read(&memcg->memory); | |
3719 | ||
c5edf9cd | 3720 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
3721 | memcg = parent; |
3722 | } | |
c2aa723a TH |
3723 | } |
3724 | ||
841710aa TH |
3725 | #else /* CONFIG_CGROUP_WRITEBACK */ |
3726 | ||
3727 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
3728 | { | |
3729 | return 0; | |
3730 | } | |
3731 | ||
3732 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3733 | { | |
3734 | } | |
3735 | ||
2529bb3a TH |
3736 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3737 | { | |
3738 | } | |
3739 | ||
52ebea74 TH |
3740 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
3741 | ||
3bc942f3 TH |
3742 | /* |
3743 | * DO NOT USE IN NEW FILES. | |
3744 | * | |
3745 | * "cgroup.event_control" implementation. | |
3746 | * | |
3747 | * This is way over-engineered. It tries to support fully configurable | |
3748 | * events for each user. Such level of flexibility is completely | |
3749 | * unnecessary especially in the light of the planned unified hierarchy. | |
3750 | * | |
3751 | * Please deprecate this and replace with something simpler if at all | |
3752 | * possible. | |
3753 | */ | |
3754 | ||
79bd9814 TH |
3755 | /* |
3756 | * Unregister event and free resources. | |
3757 | * | |
3758 | * Gets called from workqueue. | |
3759 | */ | |
3bc942f3 | 3760 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 3761 | { |
3bc942f3 TH |
3762 | struct mem_cgroup_event *event = |
3763 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 3764 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3765 | |
3766 | remove_wait_queue(event->wqh, &event->wait); | |
3767 | ||
59b6f873 | 3768 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
3769 | |
3770 | /* Notify userspace the event is going away. */ | |
3771 | eventfd_signal(event->eventfd, 1); | |
3772 | ||
3773 | eventfd_ctx_put(event->eventfd); | |
3774 | kfree(event); | |
59b6f873 | 3775 | css_put(&memcg->css); |
79bd9814 TH |
3776 | } |
3777 | ||
3778 | /* | |
3779 | * Gets called on POLLHUP on eventfd when user closes it. | |
3780 | * | |
3781 | * Called with wqh->lock held and interrupts disabled. | |
3782 | */ | |
3bc942f3 TH |
3783 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
3784 | int sync, void *key) | |
79bd9814 | 3785 | { |
3bc942f3 TH |
3786 | struct mem_cgroup_event *event = |
3787 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 3788 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3789 | unsigned long flags = (unsigned long)key; |
3790 | ||
3791 | if (flags & POLLHUP) { | |
3792 | /* | |
3793 | * If the event has been detached at cgroup removal, we | |
3794 | * can simply return knowing the other side will cleanup | |
3795 | * for us. | |
3796 | * | |
3797 | * We can't race against event freeing since the other | |
3798 | * side will require wqh->lock via remove_wait_queue(), | |
3799 | * which we hold. | |
3800 | */ | |
fba94807 | 3801 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
3802 | if (!list_empty(&event->list)) { |
3803 | list_del_init(&event->list); | |
3804 | /* | |
3805 | * We are in atomic context, but cgroup_event_remove() | |
3806 | * may sleep, so we have to call it in workqueue. | |
3807 | */ | |
3808 | schedule_work(&event->remove); | |
3809 | } | |
fba94807 | 3810 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
3811 | } |
3812 | ||
3813 | return 0; | |
3814 | } | |
3815 | ||
3bc942f3 | 3816 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
3817 | wait_queue_head_t *wqh, poll_table *pt) |
3818 | { | |
3bc942f3 TH |
3819 | struct mem_cgroup_event *event = |
3820 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
3821 | |
3822 | event->wqh = wqh; | |
3823 | add_wait_queue(wqh, &event->wait); | |
3824 | } | |
3825 | ||
3826 | /* | |
3bc942f3 TH |
3827 | * DO NOT USE IN NEW FILES. |
3828 | * | |
79bd9814 TH |
3829 | * Parse input and register new cgroup event handler. |
3830 | * | |
3831 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
3832 | * Interpretation of args is defined by control file implementation. | |
3833 | */ | |
451af504 TH |
3834 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
3835 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 3836 | { |
451af504 | 3837 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 3838 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 3839 | struct mem_cgroup_event *event; |
79bd9814 TH |
3840 | struct cgroup_subsys_state *cfile_css; |
3841 | unsigned int efd, cfd; | |
3842 | struct fd efile; | |
3843 | struct fd cfile; | |
fba94807 | 3844 | const char *name; |
79bd9814 TH |
3845 | char *endp; |
3846 | int ret; | |
3847 | ||
451af504 TH |
3848 | buf = strstrip(buf); |
3849 | ||
3850 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
3851 | if (*endp != ' ') |
3852 | return -EINVAL; | |
451af504 | 3853 | buf = endp + 1; |
79bd9814 | 3854 | |
451af504 | 3855 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
3856 | if ((*endp != ' ') && (*endp != '\0')) |
3857 | return -EINVAL; | |
451af504 | 3858 | buf = endp + 1; |
79bd9814 TH |
3859 | |
3860 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
3861 | if (!event) | |
3862 | return -ENOMEM; | |
3863 | ||
59b6f873 | 3864 | event->memcg = memcg; |
79bd9814 | 3865 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
3866 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
3867 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
3868 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
3869 | |
3870 | efile = fdget(efd); | |
3871 | if (!efile.file) { | |
3872 | ret = -EBADF; | |
3873 | goto out_kfree; | |
3874 | } | |
3875 | ||
3876 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
3877 | if (IS_ERR(event->eventfd)) { | |
3878 | ret = PTR_ERR(event->eventfd); | |
3879 | goto out_put_efile; | |
3880 | } | |
3881 | ||
3882 | cfile = fdget(cfd); | |
3883 | if (!cfile.file) { | |
3884 | ret = -EBADF; | |
3885 | goto out_put_eventfd; | |
3886 | } | |
3887 | ||
3888 | /* the process need read permission on control file */ | |
3889 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
3890 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
3891 | if (ret < 0) | |
3892 | goto out_put_cfile; | |
3893 | ||
fba94807 TH |
3894 | /* |
3895 | * Determine the event callbacks and set them in @event. This used | |
3896 | * to be done via struct cftype but cgroup core no longer knows | |
3897 | * about these events. The following is crude but the whole thing | |
3898 | * is for compatibility anyway. | |
3bc942f3 TH |
3899 | * |
3900 | * DO NOT ADD NEW FILES. | |
fba94807 | 3901 | */ |
b583043e | 3902 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
3903 | |
3904 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
3905 | event->register_event = mem_cgroup_usage_register_event; | |
3906 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
3907 | } else if (!strcmp(name, "memory.oom_control")) { | |
3908 | event->register_event = mem_cgroup_oom_register_event; | |
3909 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
3910 | } else if (!strcmp(name, "memory.pressure_level")) { | |
3911 | event->register_event = vmpressure_register_event; | |
3912 | event->unregister_event = vmpressure_unregister_event; | |
3913 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
3914 | event->register_event = memsw_cgroup_usage_register_event; |
3915 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
3916 | } else { |
3917 | ret = -EINVAL; | |
3918 | goto out_put_cfile; | |
3919 | } | |
3920 | ||
79bd9814 | 3921 | /* |
b5557c4c TH |
3922 | * Verify @cfile should belong to @css. Also, remaining events are |
3923 | * automatically removed on cgroup destruction but the removal is | |
3924 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 3925 | */ |
b583043e | 3926 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 3927 | &memory_cgrp_subsys); |
79bd9814 | 3928 | ret = -EINVAL; |
5a17f543 | 3929 | if (IS_ERR(cfile_css)) |
79bd9814 | 3930 | goto out_put_cfile; |
5a17f543 TH |
3931 | if (cfile_css != css) { |
3932 | css_put(cfile_css); | |
79bd9814 | 3933 | goto out_put_cfile; |
5a17f543 | 3934 | } |
79bd9814 | 3935 | |
451af504 | 3936 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
3937 | if (ret) |
3938 | goto out_put_css; | |
3939 | ||
3940 | efile.file->f_op->poll(efile.file, &event->pt); | |
3941 | ||
fba94807 TH |
3942 | spin_lock(&memcg->event_list_lock); |
3943 | list_add(&event->list, &memcg->event_list); | |
3944 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
3945 | |
3946 | fdput(cfile); | |
3947 | fdput(efile); | |
3948 | ||
451af504 | 3949 | return nbytes; |
79bd9814 TH |
3950 | |
3951 | out_put_css: | |
b5557c4c | 3952 | css_put(css); |
79bd9814 TH |
3953 | out_put_cfile: |
3954 | fdput(cfile); | |
3955 | out_put_eventfd: | |
3956 | eventfd_ctx_put(event->eventfd); | |
3957 | out_put_efile: | |
3958 | fdput(efile); | |
3959 | out_kfree: | |
3960 | kfree(event); | |
3961 | ||
3962 | return ret; | |
3963 | } | |
3964 | ||
241994ed | 3965 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 3966 | { |
0eea1030 | 3967 | .name = "usage_in_bytes", |
8c7c6e34 | 3968 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 3969 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 3970 | }, |
c84872e1 PE |
3971 | { |
3972 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 3973 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 3974 | .write = mem_cgroup_reset, |
791badbd | 3975 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 3976 | }, |
8cdea7c0 | 3977 | { |
0eea1030 | 3978 | .name = "limit_in_bytes", |
8c7c6e34 | 3979 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 3980 | .write = mem_cgroup_write, |
791badbd | 3981 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 3982 | }, |
296c81d8 BS |
3983 | { |
3984 | .name = "soft_limit_in_bytes", | |
3985 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 3986 | .write = mem_cgroup_write, |
791badbd | 3987 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 3988 | }, |
8cdea7c0 BS |
3989 | { |
3990 | .name = "failcnt", | |
8c7c6e34 | 3991 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 3992 | .write = mem_cgroup_reset, |
791badbd | 3993 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 3994 | }, |
d2ceb9b7 KH |
3995 | { |
3996 | .name = "stat", | |
2da8ca82 | 3997 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 3998 | }, |
c1e862c1 KH |
3999 | { |
4000 | .name = "force_empty", | |
6770c64e | 4001 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4002 | }, |
18f59ea7 BS |
4003 | { |
4004 | .name = "use_hierarchy", | |
4005 | .write_u64 = mem_cgroup_hierarchy_write, | |
4006 | .read_u64 = mem_cgroup_hierarchy_read, | |
4007 | }, | |
79bd9814 | 4008 | { |
3bc942f3 | 4009 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4010 | .write = memcg_write_event_control, |
7dbdb199 | 4011 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 4012 | }, |
a7885eb8 KM |
4013 | { |
4014 | .name = "swappiness", | |
4015 | .read_u64 = mem_cgroup_swappiness_read, | |
4016 | .write_u64 = mem_cgroup_swappiness_write, | |
4017 | }, | |
7dc74be0 DN |
4018 | { |
4019 | .name = "move_charge_at_immigrate", | |
4020 | .read_u64 = mem_cgroup_move_charge_read, | |
4021 | .write_u64 = mem_cgroup_move_charge_write, | |
4022 | }, | |
9490ff27 KH |
4023 | { |
4024 | .name = "oom_control", | |
2da8ca82 | 4025 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4026 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4027 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4028 | }, | |
70ddf637 AV |
4029 | { |
4030 | .name = "pressure_level", | |
70ddf637 | 4031 | }, |
406eb0c9 YH |
4032 | #ifdef CONFIG_NUMA |
4033 | { | |
4034 | .name = "numa_stat", | |
2da8ca82 | 4035 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4036 | }, |
4037 | #endif | |
510fc4e1 GC |
4038 | #ifdef CONFIG_MEMCG_KMEM |
4039 | { | |
4040 | .name = "kmem.limit_in_bytes", | |
4041 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4042 | .write = mem_cgroup_write, |
791badbd | 4043 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4044 | }, |
4045 | { | |
4046 | .name = "kmem.usage_in_bytes", | |
4047 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4048 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4049 | }, |
4050 | { | |
4051 | .name = "kmem.failcnt", | |
4052 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4053 | .write = mem_cgroup_reset, |
791badbd | 4054 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4055 | }, |
4056 | { | |
4057 | .name = "kmem.max_usage_in_bytes", | |
4058 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4059 | .write = mem_cgroup_reset, |
791badbd | 4060 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4061 | }, |
749c5415 GC |
4062 | #ifdef CONFIG_SLABINFO |
4063 | { | |
4064 | .name = "kmem.slabinfo", | |
b047501c VD |
4065 | .seq_start = slab_start, |
4066 | .seq_next = slab_next, | |
4067 | .seq_stop = slab_stop, | |
4068 | .seq_show = memcg_slab_show, | |
749c5415 GC |
4069 | }, |
4070 | #endif | |
8c7c6e34 | 4071 | #endif |
6bc10349 | 4072 | { }, /* terminate */ |
af36f906 | 4073 | }; |
8c7c6e34 | 4074 | |
c0ff4b85 | 4075 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4076 | { |
4077 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4078 | struct mem_cgroup_per_zone *mz; |
41e3355d | 4079 | int zone, tmp = node; |
1ecaab2b KH |
4080 | /* |
4081 | * This routine is called against possible nodes. | |
4082 | * But it's BUG to call kmalloc() against offline node. | |
4083 | * | |
4084 | * TODO: this routine can waste much memory for nodes which will | |
4085 | * never be onlined. It's better to use memory hotplug callback | |
4086 | * function. | |
4087 | */ | |
41e3355d KH |
4088 | if (!node_state(node, N_NORMAL_MEMORY)) |
4089 | tmp = -1; | |
17295c88 | 4090 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4091 | if (!pn) |
4092 | return 1; | |
1ecaab2b | 4093 | |
1ecaab2b KH |
4094 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4095 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 4096 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
4097 | mz->usage_in_excess = 0; |
4098 | mz->on_tree = false; | |
d79154bb | 4099 | mz->memcg = memcg; |
1ecaab2b | 4100 | } |
54f72fe0 | 4101 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4102 | return 0; |
4103 | } | |
4104 | ||
c0ff4b85 | 4105 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4106 | { |
54f72fe0 | 4107 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
4108 | } |
4109 | ||
33327948 KH |
4110 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4111 | { | |
d79154bb | 4112 | struct mem_cgroup *memcg; |
8ff69e2c | 4113 | size_t size; |
33327948 | 4114 | |
8ff69e2c VD |
4115 | size = sizeof(struct mem_cgroup); |
4116 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
33327948 | 4117 | |
8ff69e2c | 4118 | memcg = kzalloc(size, GFP_KERNEL); |
d79154bb | 4119 | if (!memcg) |
e7bbcdf3 DC |
4120 | return NULL; |
4121 | ||
d79154bb HD |
4122 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
4123 | if (!memcg->stat) | |
d2e61b8d | 4124 | goto out_free; |
841710aa TH |
4125 | |
4126 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) | |
4127 | goto out_free_stat; | |
4128 | ||
d79154bb | 4129 | return memcg; |
d2e61b8d | 4130 | |
841710aa TH |
4131 | out_free_stat: |
4132 | free_percpu(memcg->stat); | |
d2e61b8d | 4133 | out_free: |
8ff69e2c | 4134 | kfree(memcg); |
d2e61b8d | 4135 | return NULL; |
33327948 KH |
4136 | } |
4137 | ||
59927fb9 | 4138 | /* |
c8b2a36f GC |
4139 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
4140 | * (scanning all at force_empty is too costly...) | |
4141 | * | |
4142 | * Instead of clearing all references at force_empty, we remember | |
4143 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4144 | * it goes down to 0. | |
4145 | * | |
4146 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 4147 | */ |
c8b2a36f GC |
4148 | |
4149 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 4150 | { |
c8b2a36f | 4151 | int node; |
59927fb9 | 4152 | |
bb4cc1a8 | 4153 | mem_cgroup_remove_from_trees(memcg); |
c8b2a36f GC |
4154 | |
4155 | for_each_node(node) | |
4156 | free_mem_cgroup_per_zone_info(memcg, node); | |
4157 | ||
4158 | free_percpu(memcg->stat); | |
841710aa | 4159 | memcg_wb_domain_exit(memcg); |
8ff69e2c | 4160 | kfree(memcg); |
59927fb9 | 4161 | } |
3afe36b1 | 4162 | |
7bcc1bb1 DN |
4163 | /* |
4164 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4165 | */ | |
e1aab161 | 4166 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 4167 | { |
3e32cb2e | 4168 | if (!memcg->memory.parent) |
7bcc1bb1 | 4169 | return NULL; |
3e32cb2e | 4170 | return mem_cgroup_from_counter(memcg->memory.parent, memory); |
7bcc1bb1 | 4171 | } |
e1aab161 | 4172 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 4173 | |
0eb253e2 | 4174 | static struct cgroup_subsys_state * __ref |
eb95419b | 4175 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 4176 | { |
d142e3e6 | 4177 | struct mem_cgroup *memcg; |
04046e1a | 4178 | long error = -ENOMEM; |
6d12e2d8 | 4179 | int node; |
8cdea7c0 | 4180 | |
c0ff4b85 R |
4181 | memcg = mem_cgroup_alloc(); |
4182 | if (!memcg) | |
04046e1a | 4183 | return ERR_PTR(error); |
78fb7466 | 4184 | |
3ed28fa1 | 4185 | for_each_node(node) |
c0ff4b85 | 4186 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 4187 | goto free_out; |
f64c3f54 | 4188 | |
c077719b | 4189 | /* root ? */ |
eb95419b | 4190 | if (parent_css == NULL) { |
a41c58a6 | 4191 | root_mem_cgroup = memcg; |
3e32cb2e | 4192 | page_counter_init(&memcg->memory, NULL); |
241994ed | 4193 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4194 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4195 | page_counter_init(&memcg->memsw, NULL); |
4196 | page_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 4197 | } |
28dbc4b6 | 4198 | |
d142e3e6 GC |
4199 | memcg->last_scanned_node = MAX_NUMNODES; |
4200 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4201 | memcg->move_charge_at_immigrate = 0; |
4202 | mutex_init(&memcg->thresholds_lock); | |
4203 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4204 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4205 | INIT_LIST_HEAD(&memcg->event_list); |
4206 | spin_lock_init(&memcg->event_list_lock); | |
900a38f0 VD |
4207 | #ifdef CONFIG_MEMCG_KMEM |
4208 | memcg->kmemcg_id = -1; | |
900a38f0 | 4209 | #endif |
52ebea74 TH |
4210 | #ifdef CONFIG_CGROUP_WRITEBACK |
4211 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
4212 | #endif | |
d142e3e6 GC |
4213 | return &memcg->css; |
4214 | ||
4215 | free_out: | |
4216 | __mem_cgroup_free(memcg); | |
4217 | return ERR_PTR(error); | |
4218 | } | |
4219 | ||
4220 | static int | |
eb95419b | 4221 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 4222 | { |
eb95419b | 4223 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 4224 | struct mem_cgroup *parent = mem_cgroup_from_css(css->parent); |
2f7dd7a4 | 4225 | int ret; |
d142e3e6 | 4226 | |
15a4c835 | 4227 | if (css->id > MEM_CGROUP_ID_MAX) |
4219b2da LZ |
4228 | return -ENOSPC; |
4229 | ||
63876986 | 4230 | if (!parent) |
d142e3e6 GC |
4231 | return 0; |
4232 | ||
0999821b | 4233 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
4234 | |
4235 | memcg->use_hierarchy = parent->use_hierarchy; | |
4236 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4237 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4238 | ||
4239 | if (parent->use_hierarchy) { | |
3e32cb2e | 4240 | page_counter_init(&memcg->memory, &parent->memory); |
241994ed | 4241 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4242 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4243 | page_counter_init(&memcg->memsw, &parent->memsw); |
4244 | page_counter_init(&memcg->kmem, &parent->kmem); | |
55007d84 | 4245 | |
7bcc1bb1 | 4246 | /* |
8d76a979 LZ |
4247 | * No need to take a reference to the parent because cgroup |
4248 | * core guarantees its existence. | |
7bcc1bb1 | 4249 | */ |
18f59ea7 | 4250 | } else { |
3e32cb2e | 4251 | page_counter_init(&memcg->memory, NULL); |
241994ed | 4252 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4253 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4254 | page_counter_init(&memcg->memsw, NULL); |
4255 | page_counter_init(&memcg->kmem, NULL); | |
8c7f6edb TH |
4256 | /* |
4257 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4258 | * much sense so let cgroup subsystem know about this | |
4259 | * unfortunate state in our controller. | |
4260 | */ | |
d142e3e6 | 4261 | if (parent != root_mem_cgroup) |
073219e9 | 4262 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4263 | } |
0999821b | 4264 | mutex_unlock(&memcg_create_mutex); |
d6441637 | 4265 | |
2f7dd7a4 JW |
4266 | ret = memcg_init_kmem(memcg, &memory_cgrp_subsys); |
4267 | if (ret) | |
4268 | return ret; | |
4269 | ||
4270 | /* | |
4271 | * Make sure the memcg is initialized: mem_cgroup_iter() | |
4272 | * orders reading memcg->initialized against its callers | |
4273 | * reading the memcg members. | |
4274 | */ | |
4275 | smp_store_release(&memcg->initialized, 1); | |
4276 | ||
4277 | return 0; | |
8cdea7c0 BS |
4278 | } |
4279 | ||
eb95419b | 4280 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4281 | { |
eb95419b | 4282 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4283 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4284 | |
4285 | /* | |
4286 | * Unregister events and notify userspace. | |
4287 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4288 | * directory to avoid race between userspace and kernelspace. | |
4289 | */ | |
fba94807 TH |
4290 | spin_lock(&memcg->event_list_lock); |
4291 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4292 | list_del_init(&event->list); |
4293 | schedule_work(&event->remove); | |
4294 | } | |
fba94807 | 4295 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4296 | |
33cb876e | 4297 | vmpressure_cleanup(&memcg->vmpressure); |
2a4db7eb VD |
4298 | |
4299 | memcg_deactivate_kmem(memcg); | |
52ebea74 TH |
4300 | |
4301 | wb_memcg_offline(memcg); | |
df878fb0 KH |
4302 | } |
4303 | ||
6df38689 VD |
4304 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
4305 | { | |
4306 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4307 | ||
4308 | invalidate_reclaim_iterators(memcg); | |
4309 | } | |
4310 | ||
eb95419b | 4311 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4312 | { |
eb95419b | 4313 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4314 | |
10d5ebf4 | 4315 | memcg_destroy_kmem(memcg); |
465939a1 | 4316 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
4317 | } |
4318 | ||
1ced953b TH |
4319 | /** |
4320 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4321 | * @css: the target css | |
4322 | * | |
4323 | * Reset the states of the mem_cgroup associated with @css. This is | |
4324 | * invoked when the userland requests disabling on the default hierarchy | |
4325 | * but the memcg is pinned through dependency. The memcg should stop | |
4326 | * applying policies and should revert to the vanilla state as it may be | |
4327 | * made visible again. | |
4328 | * | |
4329 | * The current implementation only resets the essential configurations. | |
4330 | * This needs to be expanded to cover all the visible parts. | |
4331 | */ | |
4332 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4333 | { | |
4334 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4335 | ||
3e32cb2e JW |
4336 | mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX); |
4337 | mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX); | |
4338 | memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX); | |
241994ed JW |
4339 | memcg->low = 0; |
4340 | memcg->high = PAGE_COUNTER_MAX; | |
24d404dc | 4341 | memcg->soft_limit = PAGE_COUNTER_MAX; |
2529bb3a | 4342 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
4343 | } |
4344 | ||
02491447 | 4345 | #ifdef CONFIG_MMU |
7dc74be0 | 4346 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4347 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4348 | { |
05b84301 | 4349 | int ret; |
9476db97 | 4350 | |
d0164adc MG |
4351 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
4352 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 4353 | if (!ret) { |
854ffa8d | 4354 | mc.precharge += count; |
854ffa8d DN |
4355 | return ret; |
4356 | } | |
9476db97 JW |
4357 | |
4358 | /* Try charges one by one with reclaim */ | |
854ffa8d | 4359 | while (count--) { |
00501b53 | 4360 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1); |
38c5d72f | 4361 | if (ret) |
38c5d72f | 4362 | return ret; |
854ffa8d | 4363 | mc.precharge++; |
9476db97 | 4364 | cond_resched(); |
854ffa8d | 4365 | } |
9476db97 | 4366 | return 0; |
4ffef5fe DN |
4367 | } |
4368 | ||
4369 | /** | |
8d32ff84 | 4370 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
4371 | * @vma: the vma the pte to be checked belongs |
4372 | * @addr: the address corresponding to the pte to be checked | |
4373 | * @ptent: the pte to be checked | |
02491447 | 4374 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
4375 | * |
4376 | * Returns | |
4377 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4378 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4379 | * move charge. if @target is not NULL, the page is stored in target->page | |
4380 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
4381 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
4382 | * target for charge migration. if @target is not NULL, the entry is stored | |
4383 | * in target->ent. | |
4ffef5fe DN |
4384 | * |
4385 | * Called with pte lock held. | |
4386 | */ | |
4ffef5fe DN |
4387 | union mc_target { |
4388 | struct page *page; | |
02491447 | 4389 | swp_entry_t ent; |
4ffef5fe DN |
4390 | }; |
4391 | ||
4ffef5fe | 4392 | enum mc_target_type { |
8d32ff84 | 4393 | MC_TARGET_NONE = 0, |
4ffef5fe | 4394 | MC_TARGET_PAGE, |
02491447 | 4395 | MC_TARGET_SWAP, |
4ffef5fe DN |
4396 | }; |
4397 | ||
90254a65 DN |
4398 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4399 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4400 | { |
90254a65 | 4401 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 4402 | |
90254a65 DN |
4403 | if (!page || !page_mapped(page)) |
4404 | return NULL; | |
4405 | if (PageAnon(page)) { | |
1dfab5ab | 4406 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 4407 | return NULL; |
1dfab5ab JW |
4408 | } else { |
4409 | if (!(mc.flags & MOVE_FILE)) | |
4410 | return NULL; | |
4411 | } | |
90254a65 DN |
4412 | if (!get_page_unless_zero(page)) |
4413 | return NULL; | |
4414 | ||
4415 | return page; | |
4416 | } | |
4417 | ||
4b91355e | 4418 | #ifdef CONFIG_SWAP |
90254a65 DN |
4419 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
4420 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4421 | { | |
90254a65 DN |
4422 | struct page *page = NULL; |
4423 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4424 | ||
1dfab5ab | 4425 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 4426 | return NULL; |
4b91355e KH |
4427 | /* |
4428 | * Because lookup_swap_cache() updates some statistics counter, | |
4429 | * we call find_get_page() with swapper_space directly. | |
4430 | */ | |
33806f06 | 4431 | page = find_get_page(swap_address_space(ent), ent.val); |
7941d214 | 4432 | if (do_memsw_account()) |
90254a65 DN |
4433 | entry->val = ent.val; |
4434 | ||
4435 | return page; | |
4436 | } | |
4b91355e KH |
4437 | #else |
4438 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
4439 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4440 | { | |
4441 | return NULL; | |
4442 | } | |
4443 | #endif | |
90254a65 | 4444 | |
87946a72 DN |
4445 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4446 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4447 | { | |
4448 | struct page *page = NULL; | |
87946a72 DN |
4449 | struct address_space *mapping; |
4450 | pgoff_t pgoff; | |
4451 | ||
4452 | if (!vma->vm_file) /* anonymous vma */ | |
4453 | return NULL; | |
1dfab5ab | 4454 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
4455 | return NULL; |
4456 | ||
87946a72 | 4457 | mapping = vma->vm_file->f_mapping; |
0661a336 | 4458 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
4459 | |
4460 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
4461 | #ifdef CONFIG_SWAP |
4462 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
4463 | if (shmem_mapping(mapping)) { |
4464 | page = find_get_entry(mapping, pgoff); | |
4465 | if (radix_tree_exceptional_entry(page)) { | |
4466 | swp_entry_t swp = radix_to_swp_entry(page); | |
7941d214 | 4467 | if (do_memsw_account()) |
139b6a6f JW |
4468 | *entry = swp; |
4469 | page = find_get_page(swap_address_space(swp), swp.val); | |
4470 | } | |
4471 | } else | |
4472 | page = find_get_page(mapping, pgoff); | |
4473 | #else | |
4474 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 4475 | #endif |
87946a72 DN |
4476 | return page; |
4477 | } | |
4478 | ||
b1b0deab CG |
4479 | /** |
4480 | * mem_cgroup_move_account - move account of the page | |
4481 | * @page: the page | |
4482 | * @nr_pages: number of regular pages (>1 for huge pages) | |
4483 | * @from: mem_cgroup which the page is moved from. | |
4484 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
4485 | * | |
4486 | * The caller must confirm following. | |
4487 | * - page is not on LRU (isolate_page() is useful.) | |
4488 | * - compound_lock is held when nr_pages > 1 | |
4489 | * | |
4490 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
4491 | * from old cgroup. | |
4492 | */ | |
4493 | static int mem_cgroup_move_account(struct page *page, | |
4494 | unsigned int nr_pages, | |
4495 | struct mem_cgroup *from, | |
4496 | struct mem_cgroup *to) | |
4497 | { | |
4498 | unsigned long flags; | |
4499 | int ret; | |
c4843a75 | 4500 | bool anon; |
b1b0deab CG |
4501 | |
4502 | VM_BUG_ON(from == to); | |
4503 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
4504 | /* | |
4505 | * The page is isolated from LRU. So, collapse function | |
4506 | * will not handle this page. But page splitting can happen. | |
4507 | * Do this check under compound_page_lock(). The caller should | |
4508 | * hold it. | |
4509 | */ | |
4510 | ret = -EBUSY; | |
4511 | if (nr_pages > 1 && !PageTransHuge(page)) | |
4512 | goto out; | |
4513 | ||
4514 | /* | |
45637bab HD |
4515 | * Prevent mem_cgroup_replace_page() from looking at |
4516 | * page->mem_cgroup of its source page while we change it. | |
b1b0deab CG |
4517 | */ |
4518 | if (!trylock_page(page)) | |
4519 | goto out; | |
4520 | ||
4521 | ret = -EINVAL; | |
4522 | if (page->mem_cgroup != from) | |
4523 | goto out_unlock; | |
4524 | ||
c4843a75 GT |
4525 | anon = PageAnon(page); |
4526 | ||
b1b0deab CG |
4527 | spin_lock_irqsave(&from->move_lock, flags); |
4528 | ||
c4843a75 | 4529 | if (!anon && page_mapped(page)) { |
b1b0deab CG |
4530 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
4531 | nr_pages); | |
4532 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
4533 | nr_pages); | |
4534 | } | |
4535 | ||
c4843a75 GT |
4536 | /* |
4537 | * move_lock grabbed above and caller set from->moving_account, so | |
4538 | * mem_cgroup_update_page_stat() will serialize updates to PageDirty. | |
4539 | * So mapping should be stable for dirty pages. | |
4540 | */ | |
4541 | if (!anon && PageDirty(page)) { | |
4542 | struct address_space *mapping = page_mapping(page); | |
4543 | ||
4544 | if (mapping_cap_account_dirty(mapping)) { | |
4545 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY], | |
4546 | nr_pages); | |
4547 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY], | |
4548 | nr_pages); | |
4549 | } | |
4550 | } | |
4551 | ||
b1b0deab CG |
4552 | if (PageWriteback(page)) { |
4553 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
4554 | nr_pages); | |
4555 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
4556 | nr_pages); | |
4557 | } | |
4558 | ||
4559 | /* | |
4560 | * It is safe to change page->mem_cgroup here because the page | |
4561 | * is referenced, charged, and isolated - we can't race with | |
4562 | * uncharging, charging, migration, or LRU putback. | |
4563 | */ | |
4564 | ||
4565 | /* caller should have done css_get */ | |
4566 | page->mem_cgroup = to; | |
4567 | spin_unlock_irqrestore(&from->move_lock, flags); | |
4568 | ||
4569 | ret = 0; | |
4570 | ||
4571 | local_irq_disable(); | |
4572 | mem_cgroup_charge_statistics(to, page, nr_pages); | |
4573 | memcg_check_events(to, page); | |
4574 | mem_cgroup_charge_statistics(from, page, -nr_pages); | |
4575 | memcg_check_events(from, page); | |
4576 | local_irq_enable(); | |
4577 | out_unlock: | |
4578 | unlock_page(page); | |
4579 | out: | |
4580 | return ret; | |
4581 | } | |
4582 | ||
8d32ff84 | 4583 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
4584 | unsigned long addr, pte_t ptent, union mc_target *target) |
4585 | { | |
4586 | struct page *page = NULL; | |
8d32ff84 | 4587 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
4588 | swp_entry_t ent = { .val = 0 }; |
4589 | ||
4590 | if (pte_present(ptent)) | |
4591 | page = mc_handle_present_pte(vma, addr, ptent); | |
4592 | else if (is_swap_pte(ptent)) | |
4593 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
0661a336 | 4594 | else if (pte_none(ptent)) |
87946a72 | 4595 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
4596 | |
4597 | if (!page && !ent.val) | |
8d32ff84 | 4598 | return ret; |
02491447 | 4599 | if (page) { |
02491447 | 4600 | /* |
0a31bc97 | 4601 | * Do only loose check w/o serialization. |
1306a85a | 4602 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 4603 | * not under LRU exclusion. |
02491447 | 4604 | */ |
1306a85a | 4605 | if (page->mem_cgroup == mc.from) { |
02491447 DN |
4606 | ret = MC_TARGET_PAGE; |
4607 | if (target) | |
4608 | target->page = page; | |
4609 | } | |
4610 | if (!ret || !target) | |
4611 | put_page(page); | |
4612 | } | |
90254a65 DN |
4613 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
4614 | if (ent.val && !ret && | |
34c00c31 | 4615 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
4616 | ret = MC_TARGET_SWAP; |
4617 | if (target) | |
4618 | target->ent = ent; | |
4ffef5fe | 4619 | } |
4ffef5fe DN |
4620 | return ret; |
4621 | } | |
4622 | ||
12724850 NH |
4623 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4624 | /* | |
4625 | * We don't consider swapping or file mapped pages because THP does not | |
4626 | * support them for now. | |
4627 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
4628 | */ | |
4629 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4630 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4631 | { | |
4632 | struct page *page = NULL; | |
12724850 NH |
4633 | enum mc_target_type ret = MC_TARGET_NONE; |
4634 | ||
4635 | page = pmd_page(pmd); | |
309381fe | 4636 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 4637 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 4638 | return ret; |
1306a85a | 4639 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
4640 | ret = MC_TARGET_PAGE; |
4641 | if (target) { | |
4642 | get_page(page); | |
4643 | target->page = page; | |
4644 | } | |
4645 | } | |
4646 | return ret; | |
4647 | } | |
4648 | #else | |
4649 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4650 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4651 | { | |
4652 | return MC_TARGET_NONE; | |
4653 | } | |
4654 | #endif | |
4655 | ||
4ffef5fe DN |
4656 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
4657 | unsigned long addr, unsigned long end, | |
4658 | struct mm_walk *walk) | |
4659 | { | |
26bcd64a | 4660 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4661 | pte_t *pte; |
4662 | spinlock_t *ptl; | |
4663 | ||
bf929152 | 4664 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
12724850 NH |
4665 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
4666 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 4667 | spin_unlock(ptl); |
1a5a9906 | 4668 | return 0; |
12724850 | 4669 | } |
03319327 | 4670 | |
45f83cef AA |
4671 | if (pmd_trans_unstable(pmd)) |
4672 | return 0; | |
4ffef5fe DN |
4673 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
4674 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 4675 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
4676 | mc.precharge++; /* increment precharge temporarily */ |
4677 | pte_unmap_unlock(pte - 1, ptl); | |
4678 | cond_resched(); | |
4679 | ||
7dc74be0 DN |
4680 | return 0; |
4681 | } | |
4682 | ||
4ffef5fe DN |
4683 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
4684 | { | |
4685 | unsigned long precharge; | |
4ffef5fe | 4686 | |
26bcd64a NH |
4687 | struct mm_walk mem_cgroup_count_precharge_walk = { |
4688 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
4689 | .mm = mm, | |
4690 | }; | |
dfe076b0 | 4691 | down_read(&mm->mmap_sem); |
26bcd64a | 4692 | walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk); |
dfe076b0 | 4693 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
4694 | |
4695 | precharge = mc.precharge; | |
4696 | mc.precharge = 0; | |
4697 | ||
4698 | return precharge; | |
4699 | } | |
4700 | ||
4ffef5fe DN |
4701 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
4702 | { | |
dfe076b0 DN |
4703 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
4704 | ||
4705 | VM_BUG_ON(mc.moving_task); | |
4706 | mc.moving_task = current; | |
4707 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
4708 | } |
4709 | ||
dfe076b0 DN |
4710 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
4711 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 4712 | { |
2bd9bb20 KH |
4713 | struct mem_cgroup *from = mc.from; |
4714 | struct mem_cgroup *to = mc.to; | |
4715 | ||
4ffef5fe | 4716 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 4717 | if (mc.precharge) { |
00501b53 | 4718 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
4719 | mc.precharge = 0; |
4720 | } | |
4721 | /* | |
4722 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
4723 | * we must uncharge here. | |
4724 | */ | |
4725 | if (mc.moved_charge) { | |
00501b53 | 4726 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 4727 | mc.moved_charge = 0; |
4ffef5fe | 4728 | } |
483c30b5 DN |
4729 | /* we must fixup refcnts and charges */ |
4730 | if (mc.moved_swap) { | |
483c30b5 | 4731 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 4732 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 4733 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 4734 | |
05b84301 | 4735 | /* |
3e32cb2e JW |
4736 | * we charged both to->memory and to->memsw, so we |
4737 | * should uncharge to->memory. | |
05b84301 | 4738 | */ |
ce00a967 | 4739 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
4740 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
4741 | ||
e8ea14cc | 4742 | css_put_many(&mc.from->css, mc.moved_swap); |
3e32cb2e | 4743 | |
4050377b | 4744 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
4745 | mc.moved_swap = 0; |
4746 | } | |
dfe076b0 DN |
4747 | memcg_oom_recover(from); |
4748 | memcg_oom_recover(to); | |
4749 | wake_up_all(&mc.waitq); | |
4750 | } | |
4751 | ||
4752 | static void mem_cgroup_clear_mc(void) | |
4753 | { | |
dfe076b0 DN |
4754 | /* |
4755 | * we must clear moving_task before waking up waiters at the end of | |
4756 | * task migration. | |
4757 | */ | |
4758 | mc.moving_task = NULL; | |
4759 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 4760 | spin_lock(&mc.lock); |
4ffef5fe DN |
4761 | mc.from = NULL; |
4762 | mc.to = NULL; | |
2bd9bb20 | 4763 | spin_unlock(&mc.lock); |
4ffef5fe DN |
4764 | } |
4765 | ||
1f7dd3e5 | 4766 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 4767 | { |
1f7dd3e5 | 4768 | struct cgroup_subsys_state *css; |
eed67d75 | 4769 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 4770 | struct mem_cgroup *from; |
4530eddb | 4771 | struct task_struct *leader, *p; |
9f2115f9 | 4772 | struct mm_struct *mm; |
1dfab5ab | 4773 | unsigned long move_flags; |
9f2115f9 | 4774 | int ret = 0; |
7dc74be0 | 4775 | |
1f7dd3e5 TH |
4776 | /* charge immigration isn't supported on the default hierarchy */ |
4777 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
4778 | return 0; |
4779 | ||
4530eddb TH |
4780 | /* |
4781 | * Multi-process migrations only happen on the default hierarchy | |
4782 | * where charge immigration is not used. Perform charge | |
4783 | * immigration if @tset contains a leader and whine if there are | |
4784 | * multiple. | |
4785 | */ | |
4786 | p = NULL; | |
1f7dd3e5 | 4787 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
4788 | WARN_ON_ONCE(p); |
4789 | p = leader; | |
1f7dd3e5 | 4790 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
4791 | } |
4792 | if (!p) | |
4793 | return 0; | |
4794 | ||
1f7dd3e5 TH |
4795 | /* |
4796 | * We are now commited to this value whatever it is. Changes in this | |
4797 | * tunable will only affect upcoming migrations, not the current one. | |
4798 | * So we need to save it, and keep it going. | |
4799 | */ | |
4800 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
4801 | if (!move_flags) | |
4802 | return 0; | |
4803 | ||
9f2115f9 TH |
4804 | from = mem_cgroup_from_task(p); |
4805 | ||
4806 | VM_BUG_ON(from == memcg); | |
4807 | ||
4808 | mm = get_task_mm(p); | |
4809 | if (!mm) | |
4810 | return 0; | |
4811 | /* We move charges only when we move a owner of the mm */ | |
4812 | if (mm->owner == p) { | |
4813 | VM_BUG_ON(mc.from); | |
4814 | VM_BUG_ON(mc.to); | |
4815 | VM_BUG_ON(mc.precharge); | |
4816 | VM_BUG_ON(mc.moved_charge); | |
4817 | VM_BUG_ON(mc.moved_swap); | |
4818 | ||
4819 | spin_lock(&mc.lock); | |
4820 | mc.from = from; | |
4821 | mc.to = memcg; | |
4822 | mc.flags = move_flags; | |
4823 | spin_unlock(&mc.lock); | |
4824 | /* We set mc.moving_task later */ | |
4825 | ||
4826 | ret = mem_cgroup_precharge_mc(mm); | |
4827 | if (ret) | |
4828 | mem_cgroup_clear_mc(); | |
7dc74be0 | 4829 | } |
9f2115f9 | 4830 | mmput(mm); |
7dc74be0 DN |
4831 | return ret; |
4832 | } | |
4833 | ||
1f7dd3e5 | 4834 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 4835 | { |
4e2f245d JW |
4836 | if (mc.to) |
4837 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
4838 | } |
4839 | ||
4ffef5fe DN |
4840 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
4841 | unsigned long addr, unsigned long end, | |
4842 | struct mm_walk *walk) | |
7dc74be0 | 4843 | { |
4ffef5fe | 4844 | int ret = 0; |
26bcd64a | 4845 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4846 | pte_t *pte; |
4847 | spinlock_t *ptl; | |
12724850 NH |
4848 | enum mc_target_type target_type; |
4849 | union mc_target target; | |
4850 | struct page *page; | |
4ffef5fe | 4851 | |
12724850 NH |
4852 | /* |
4853 | * We don't take compound_lock() here but no race with splitting thp | |
4854 | * happens because: | |
4855 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
4856 | * under splitting, which means there's no concurrent thp split, | |
4857 | * - if another thread runs into split_huge_page() just after we | |
4858 | * entered this if-block, the thread must wait for page table lock | |
4859 | * to be unlocked in __split_huge_page_splitting(), where the main | |
4860 | * part of thp split is not executed yet. | |
4861 | */ | |
bf929152 | 4862 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
62ade86a | 4863 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 4864 | spin_unlock(ptl); |
12724850 NH |
4865 | return 0; |
4866 | } | |
4867 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
4868 | if (target_type == MC_TARGET_PAGE) { | |
4869 | page = target.page; | |
4870 | if (!isolate_lru_page(page)) { | |
12724850 | 4871 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, |
1306a85a | 4872 | mc.from, mc.to)) { |
12724850 NH |
4873 | mc.precharge -= HPAGE_PMD_NR; |
4874 | mc.moved_charge += HPAGE_PMD_NR; | |
4875 | } | |
4876 | putback_lru_page(page); | |
4877 | } | |
4878 | put_page(page); | |
4879 | } | |
bf929152 | 4880 | spin_unlock(ptl); |
1a5a9906 | 4881 | return 0; |
12724850 NH |
4882 | } |
4883 | ||
45f83cef AA |
4884 | if (pmd_trans_unstable(pmd)) |
4885 | return 0; | |
4ffef5fe DN |
4886 | retry: |
4887 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
4888 | for (; addr != end; addr += PAGE_SIZE) { | |
4889 | pte_t ptent = *(pte++); | |
02491447 | 4890 | swp_entry_t ent; |
4ffef5fe DN |
4891 | |
4892 | if (!mc.precharge) | |
4893 | break; | |
4894 | ||
8d32ff84 | 4895 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
4896 | case MC_TARGET_PAGE: |
4897 | page = target.page; | |
4898 | if (isolate_lru_page(page)) | |
4899 | goto put; | |
1306a85a | 4900 | if (!mem_cgroup_move_account(page, 1, mc.from, mc.to)) { |
4ffef5fe | 4901 | mc.precharge--; |
854ffa8d DN |
4902 | /* we uncharge from mc.from later. */ |
4903 | mc.moved_charge++; | |
4ffef5fe DN |
4904 | } |
4905 | putback_lru_page(page); | |
8d32ff84 | 4906 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
4907 | put_page(page); |
4908 | break; | |
02491447 DN |
4909 | case MC_TARGET_SWAP: |
4910 | ent = target.ent; | |
e91cbb42 | 4911 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 4912 | mc.precharge--; |
483c30b5 DN |
4913 | /* we fixup refcnts and charges later. */ |
4914 | mc.moved_swap++; | |
4915 | } | |
02491447 | 4916 | break; |
4ffef5fe DN |
4917 | default: |
4918 | break; | |
4919 | } | |
4920 | } | |
4921 | pte_unmap_unlock(pte - 1, ptl); | |
4922 | cond_resched(); | |
4923 | ||
4924 | if (addr != end) { | |
4925 | /* | |
4926 | * We have consumed all precharges we got in can_attach(). | |
4927 | * We try charge one by one, but don't do any additional | |
4928 | * charges to mc.to if we have failed in charge once in attach() | |
4929 | * phase. | |
4930 | */ | |
854ffa8d | 4931 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
4932 | if (!ret) |
4933 | goto retry; | |
4934 | } | |
4935 | ||
4936 | return ret; | |
4937 | } | |
4938 | ||
4939 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
4940 | { | |
26bcd64a NH |
4941 | struct mm_walk mem_cgroup_move_charge_walk = { |
4942 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
4943 | .mm = mm, | |
4944 | }; | |
4ffef5fe DN |
4945 | |
4946 | lru_add_drain_all(); | |
312722cb JW |
4947 | /* |
4948 | * Signal mem_cgroup_begin_page_stat() to take the memcg's | |
4949 | * move_lock while we're moving its pages to another memcg. | |
4950 | * Then wait for already started RCU-only updates to finish. | |
4951 | */ | |
4952 | atomic_inc(&mc.from->moving_account); | |
4953 | synchronize_rcu(); | |
dfe076b0 DN |
4954 | retry: |
4955 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
4956 | /* | |
4957 | * Someone who are holding the mmap_sem might be waiting in | |
4958 | * waitq. So we cancel all extra charges, wake up all waiters, | |
4959 | * and retry. Because we cancel precharges, we might not be able | |
4960 | * to move enough charges, but moving charge is a best-effort | |
4961 | * feature anyway, so it wouldn't be a big problem. | |
4962 | */ | |
4963 | __mem_cgroup_clear_mc(); | |
4964 | cond_resched(); | |
4965 | goto retry; | |
4966 | } | |
26bcd64a NH |
4967 | /* |
4968 | * When we have consumed all precharges and failed in doing | |
4969 | * additional charge, the page walk just aborts. | |
4970 | */ | |
4971 | walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk); | |
dfe076b0 | 4972 | up_read(&mm->mmap_sem); |
312722cb | 4973 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
4974 | } |
4975 | ||
1f7dd3e5 | 4976 | static void mem_cgroup_move_task(struct cgroup_taskset *tset) |
67e465a7 | 4977 | { |
1f7dd3e5 TH |
4978 | struct cgroup_subsys_state *css; |
4979 | struct task_struct *p = cgroup_taskset_first(tset, &css); | |
a433658c | 4980 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 4981 | |
dfe076b0 | 4982 | if (mm) { |
a433658c KM |
4983 | if (mc.to) |
4984 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
4985 | mmput(mm); |
4986 | } | |
a433658c KM |
4987 | if (mc.to) |
4988 | mem_cgroup_clear_mc(); | |
67e465a7 | 4989 | } |
5cfb80a7 | 4990 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 4991 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
4992 | { |
4993 | return 0; | |
4994 | } | |
1f7dd3e5 | 4995 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
4996 | { |
4997 | } | |
1f7dd3e5 | 4998 | static void mem_cgroup_move_task(struct cgroup_taskset *tset) |
5cfb80a7 DN |
4999 | { |
5000 | } | |
5001 | #endif | |
67e465a7 | 5002 | |
f00baae7 TH |
5003 | /* |
5004 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5005 | * to verify whether we're attached to the default hierarchy on each mount |
5006 | * attempt. | |
f00baae7 | 5007 | */ |
eb95419b | 5008 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5009 | { |
5010 | /* | |
aa6ec29b | 5011 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5012 | * guarantees that @root doesn't have any children, so turning it |
5013 | * on for the root memcg is enough. | |
5014 | */ | |
9e10a130 | 5015 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
5016 | root_mem_cgroup->use_hierarchy = true; |
5017 | else | |
5018 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
5019 | } |
5020 | ||
241994ed JW |
5021 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
5022 | struct cftype *cft) | |
5023 | { | |
f5fc3c5d JW |
5024 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5025 | ||
5026 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
5027 | } |
5028 | ||
5029 | static int memory_low_show(struct seq_file *m, void *v) | |
5030 | { | |
5031 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5032 | unsigned long low = READ_ONCE(memcg->low); |
241994ed JW |
5033 | |
5034 | if (low == PAGE_COUNTER_MAX) | |
d2973697 | 5035 | seq_puts(m, "max\n"); |
241994ed JW |
5036 | else |
5037 | seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE); | |
5038 | ||
5039 | return 0; | |
5040 | } | |
5041 | ||
5042 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
5043 | char *buf, size_t nbytes, loff_t off) | |
5044 | { | |
5045 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5046 | unsigned long low; | |
5047 | int err; | |
5048 | ||
5049 | buf = strstrip(buf); | |
d2973697 | 5050 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
5051 | if (err) |
5052 | return err; | |
5053 | ||
5054 | memcg->low = low; | |
5055 | ||
5056 | return nbytes; | |
5057 | } | |
5058 | ||
5059 | static int memory_high_show(struct seq_file *m, void *v) | |
5060 | { | |
5061 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5062 | unsigned long high = READ_ONCE(memcg->high); |
241994ed JW |
5063 | |
5064 | if (high == PAGE_COUNTER_MAX) | |
d2973697 | 5065 | seq_puts(m, "max\n"); |
241994ed JW |
5066 | else |
5067 | seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE); | |
5068 | ||
5069 | return 0; | |
5070 | } | |
5071 | ||
5072 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
5073 | char *buf, size_t nbytes, loff_t off) | |
5074 | { | |
5075 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5076 | unsigned long high; | |
5077 | int err; | |
5078 | ||
5079 | buf = strstrip(buf); | |
d2973697 | 5080 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
5081 | if (err) |
5082 | return err; | |
5083 | ||
5084 | memcg->high = high; | |
5085 | ||
2529bb3a | 5086 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5087 | return nbytes; |
5088 | } | |
5089 | ||
5090 | static int memory_max_show(struct seq_file *m, void *v) | |
5091 | { | |
5092 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5093 | unsigned long max = READ_ONCE(memcg->memory.limit); |
241994ed JW |
5094 | |
5095 | if (max == PAGE_COUNTER_MAX) | |
d2973697 | 5096 | seq_puts(m, "max\n"); |
241994ed JW |
5097 | else |
5098 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
5099 | ||
5100 | return 0; | |
5101 | } | |
5102 | ||
5103 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
5104 | char *buf, size_t nbytes, loff_t off) | |
5105 | { | |
5106 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5107 | unsigned long max; | |
5108 | int err; | |
5109 | ||
5110 | buf = strstrip(buf); | |
d2973697 | 5111 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
5112 | if (err) |
5113 | return err; | |
5114 | ||
5115 | err = mem_cgroup_resize_limit(memcg, max); | |
5116 | if (err) | |
5117 | return err; | |
5118 | ||
2529bb3a | 5119 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5120 | return nbytes; |
5121 | } | |
5122 | ||
5123 | static int memory_events_show(struct seq_file *m, void *v) | |
5124 | { | |
5125 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5126 | ||
5127 | seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW)); | |
5128 | seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH)); | |
5129 | seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX)); | |
5130 | seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM)); | |
5131 | ||
5132 | return 0; | |
5133 | } | |
5134 | ||
5135 | static struct cftype memory_files[] = { | |
5136 | { | |
5137 | .name = "current", | |
f5fc3c5d | 5138 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
5139 | .read_u64 = memory_current_read, |
5140 | }, | |
5141 | { | |
5142 | .name = "low", | |
5143 | .flags = CFTYPE_NOT_ON_ROOT, | |
5144 | .seq_show = memory_low_show, | |
5145 | .write = memory_low_write, | |
5146 | }, | |
5147 | { | |
5148 | .name = "high", | |
5149 | .flags = CFTYPE_NOT_ON_ROOT, | |
5150 | .seq_show = memory_high_show, | |
5151 | .write = memory_high_write, | |
5152 | }, | |
5153 | { | |
5154 | .name = "max", | |
5155 | .flags = CFTYPE_NOT_ON_ROOT, | |
5156 | .seq_show = memory_max_show, | |
5157 | .write = memory_max_write, | |
5158 | }, | |
5159 | { | |
5160 | .name = "events", | |
5161 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 5162 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
5163 | .seq_show = memory_events_show, |
5164 | }, | |
5165 | { } /* terminate */ | |
5166 | }; | |
5167 | ||
073219e9 | 5168 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5169 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5170 | .css_online = mem_cgroup_css_online, |
92fb9748 | 5171 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 5172 | .css_released = mem_cgroup_css_released, |
92fb9748 | 5173 | .css_free = mem_cgroup_css_free, |
1ced953b | 5174 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5175 | .can_attach = mem_cgroup_can_attach, |
5176 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5177 | .attach = mem_cgroup_move_task, |
f00baae7 | 5178 | .bind = mem_cgroup_bind, |
241994ed JW |
5179 | .dfl_cftypes = memory_files, |
5180 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 5181 | .early_init = 0, |
8cdea7c0 | 5182 | }; |
c077719b | 5183 | |
241994ed JW |
5184 | /** |
5185 | * mem_cgroup_low - check if memory consumption is below the normal range | |
5186 | * @root: the highest ancestor to consider | |
5187 | * @memcg: the memory cgroup to check | |
5188 | * | |
5189 | * Returns %true if memory consumption of @memcg, and that of all | |
5190 | * configurable ancestors up to @root, is below the normal range. | |
5191 | */ | |
5192 | bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg) | |
5193 | { | |
5194 | if (mem_cgroup_disabled()) | |
5195 | return false; | |
5196 | ||
5197 | /* | |
5198 | * The toplevel group doesn't have a configurable range, so | |
5199 | * it's never low when looked at directly, and it is not | |
5200 | * considered an ancestor when assessing the hierarchy. | |
5201 | */ | |
5202 | ||
5203 | if (memcg == root_mem_cgroup) | |
5204 | return false; | |
5205 | ||
4e54dede | 5206 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5207 | return false; |
5208 | ||
5209 | while (memcg != root) { | |
5210 | memcg = parent_mem_cgroup(memcg); | |
5211 | ||
5212 | if (memcg == root_mem_cgroup) | |
5213 | break; | |
5214 | ||
4e54dede | 5215 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5216 | return false; |
5217 | } | |
5218 | return true; | |
5219 | } | |
5220 | ||
00501b53 JW |
5221 | /** |
5222 | * mem_cgroup_try_charge - try charging a page | |
5223 | * @page: page to charge | |
5224 | * @mm: mm context of the victim | |
5225 | * @gfp_mask: reclaim mode | |
5226 | * @memcgp: charged memcg return | |
5227 | * | |
5228 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5229 | * pages according to @gfp_mask if necessary. | |
5230 | * | |
5231 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5232 | * Otherwise, an error code is returned. | |
5233 | * | |
5234 | * After page->mapping has been set up, the caller must finalize the | |
5235 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5236 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5237 | */ | |
5238 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
5239 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
5240 | { | |
5241 | struct mem_cgroup *memcg = NULL; | |
5242 | unsigned int nr_pages = 1; | |
5243 | int ret = 0; | |
5244 | ||
5245 | if (mem_cgroup_disabled()) | |
5246 | goto out; | |
5247 | ||
5248 | if (PageSwapCache(page)) { | |
00501b53 JW |
5249 | /* |
5250 | * Every swap fault against a single page tries to charge the | |
5251 | * page, bail as early as possible. shmem_unuse() encounters | |
5252 | * already charged pages, too. The USED bit is protected by | |
5253 | * the page lock, which serializes swap cache removal, which | |
5254 | * in turn serializes uncharging. | |
5255 | */ | |
e993d905 | 5256 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1306a85a | 5257 | if (page->mem_cgroup) |
00501b53 | 5258 | goto out; |
e993d905 | 5259 | |
7941d214 | 5260 | if (do_memsw_account()) { |
e993d905 VD |
5261 | swp_entry_t ent = { .val = page_private(page), }; |
5262 | unsigned short id = lookup_swap_cgroup_id(ent); | |
5263 | ||
5264 | rcu_read_lock(); | |
5265 | memcg = mem_cgroup_from_id(id); | |
5266 | if (memcg && !css_tryget_online(&memcg->css)) | |
5267 | memcg = NULL; | |
5268 | rcu_read_unlock(); | |
5269 | } | |
00501b53 JW |
5270 | } |
5271 | ||
5272 | if (PageTransHuge(page)) { | |
5273 | nr_pages <<= compound_order(page); | |
5274 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5275 | } | |
5276 | ||
00501b53 JW |
5277 | if (!memcg) |
5278 | memcg = get_mem_cgroup_from_mm(mm); | |
5279 | ||
5280 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
5281 | ||
5282 | css_put(&memcg->css); | |
00501b53 JW |
5283 | out: |
5284 | *memcgp = memcg; | |
5285 | return ret; | |
5286 | } | |
5287 | ||
5288 | /** | |
5289 | * mem_cgroup_commit_charge - commit a page charge | |
5290 | * @page: page to charge | |
5291 | * @memcg: memcg to charge the page to | |
5292 | * @lrucare: page might be on LRU already | |
5293 | * | |
5294 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
5295 | * after page->mapping has been set up. This must happen atomically | |
5296 | * as part of the page instantiation, i.e. under the page table lock | |
5297 | * for anonymous pages, under the page lock for page and swap cache. | |
5298 | * | |
5299 | * In addition, the page must not be on the LRU during the commit, to | |
5300 | * prevent racing with task migration. If it might be, use @lrucare. | |
5301 | * | |
5302 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
5303 | */ | |
5304 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
5305 | bool lrucare) | |
5306 | { | |
5307 | unsigned int nr_pages = 1; | |
5308 | ||
5309 | VM_BUG_ON_PAGE(!page->mapping, page); | |
5310 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
5311 | ||
5312 | if (mem_cgroup_disabled()) | |
5313 | return; | |
5314 | /* | |
5315 | * Swap faults will attempt to charge the same page multiple | |
5316 | * times. But reuse_swap_page() might have removed the page | |
5317 | * from swapcache already, so we can't check PageSwapCache(). | |
5318 | */ | |
5319 | if (!memcg) | |
5320 | return; | |
5321 | ||
6abb5a86 JW |
5322 | commit_charge(page, memcg, lrucare); |
5323 | ||
00501b53 JW |
5324 | if (PageTransHuge(page)) { |
5325 | nr_pages <<= compound_order(page); | |
5326 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5327 | } | |
5328 | ||
6abb5a86 JW |
5329 | local_irq_disable(); |
5330 | mem_cgroup_charge_statistics(memcg, page, nr_pages); | |
5331 | memcg_check_events(memcg, page); | |
5332 | local_irq_enable(); | |
00501b53 | 5333 | |
7941d214 | 5334 | if (do_memsw_account() && PageSwapCache(page)) { |
00501b53 JW |
5335 | swp_entry_t entry = { .val = page_private(page) }; |
5336 | /* | |
5337 | * The swap entry might not get freed for a long time, | |
5338 | * let's not wait for it. The page already received a | |
5339 | * memory+swap charge, drop the swap entry duplicate. | |
5340 | */ | |
5341 | mem_cgroup_uncharge_swap(entry); | |
5342 | } | |
5343 | } | |
5344 | ||
5345 | /** | |
5346 | * mem_cgroup_cancel_charge - cancel a page charge | |
5347 | * @page: page to charge | |
5348 | * @memcg: memcg to charge the page to | |
5349 | * | |
5350 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
5351 | */ | |
5352 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg) | |
5353 | { | |
5354 | unsigned int nr_pages = 1; | |
5355 | ||
5356 | if (mem_cgroup_disabled()) | |
5357 | return; | |
5358 | /* | |
5359 | * Swap faults will attempt to charge the same page multiple | |
5360 | * times. But reuse_swap_page() might have removed the page | |
5361 | * from swapcache already, so we can't check PageSwapCache(). | |
5362 | */ | |
5363 | if (!memcg) | |
5364 | return; | |
5365 | ||
5366 | if (PageTransHuge(page)) { | |
5367 | nr_pages <<= compound_order(page); | |
5368 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5369 | } | |
5370 | ||
5371 | cancel_charge(memcg, nr_pages); | |
5372 | } | |
5373 | ||
747db954 | 5374 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
747db954 JW |
5375 | unsigned long nr_anon, unsigned long nr_file, |
5376 | unsigned long nr_huge, struct page *dummy_page) | |
5377 | { | |
18eca2e6 | 5378 | unsigned long nr_pages = nr_anon + nr_file; |
747db954 JW |
5379 | unsigned long flags; |
5380 | ||
ce00a967 | 5381 | if (!mem_cgroup_is_root(memcg)) { |
18eca2e6 | 5382 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 5383 | if (do_memsw_account()) |
18eca2e6 | 5384 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 JW |
5385 | memcg_oom_recover(memcg); |
5386 | } | |
747db954 JW |
5387 | |
5388 | local_irq_save(flags); | |
5389 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); | |
5390 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); | |
5391 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); | |
5392 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); | |
18eca2e6 | 5393 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
747db954 JW |
5394 | memcg_check_events(memcg, dummy_page); |
5395 | local_irq_restore(flags); | |
e8ea14cc JW |
5396 | |
5397 | if (!mem_cgroup_is_root(memcg)) | |
18eca2e6 | 5398 | css_put_many(&memcg->css, nr_pages); |
747db954 JW |
5399 | } |
5400 | ||
5401 | static void uncharge_list(struct list_head *page_list) | |
5402 | { | |
5403 | struct mem_cgroup *memcg = NULL; | |
747db954 JW |
5404 | unsigned long nr_anon = 0; |
5405 | unsigned long nr_file = 0; | |
5406 | unsigned long nr_huge = 0; | |
5407 | unsigned long pgpgout = 0; | |
747db954 JW |
5408 | struct list_head *next; |
5409 | struct page *page; | |
5410 | ||
5411 | next = page_list->next; | |
5412 | do { | |
5413 | unsigned int nr_pages = 1; | |
747db954 JW |
5414 | |
5415 | page = list_entry(next, struct page, lru); | |
5416 | next = page->lru.next; | |
5417 | ||
5418 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5419 | VM_BUG_ON_PAGE(page_count(page), page); | |
5420 | ||
1306a85a | 5421 | if (!page->mem_cgroup) |
747db954 JW |
5422 | continue; |
5423 | ||
5424 | /* | |
5425 | * Nobody should be changing or seriously looking at | |
1306a85a | 5426 | * page->mem_cgroup at this point, we have fully |
29833315 | 5427 | * exclusive access to the page. |
747db954 JW |
5428 | */ |
5429 | ||
1306a85a | 5430 | if (memcg != page->mem_cgroup) { |
747db954 | 5431 | if (memcg) { |
18eca2e6 JW |
5432 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5433 | nr_huge, page); | |
5434 | pgpgout = nr_anon = nr_file = nr_huge = 0; | |
747db954 | 5435 | } |
1306a85a | 5436 | memcg = page->mem_cgroup; |
747db954 JW |
5437 | } |
5438 | ||
5439 | if (PageTransHuge(page)) { | |
5440 | nr_pages <<= compound_order(page); | |
5441 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5442 | nr_huge += nr_pages; | |
5443 | } | |
5444 | ||
5445 | if (PageAnon(page)) | |
5446 | nr_anon += nr_pages; | |
5447 | else | |
5448 | nr_file += nr_pages; | |
5449 | ||
1306a85a | 5450 | page->mem_cgroup = NULL; |
747db954 JW |
5451 | |
5452 | pgpgout++; | |
5453 | } while (next != page_list); | |
5454 | ||
5455 | if (memcg) | |
18eca2e6 JW |
5456 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5457 | nr_huge, page); | |
747db954 JW |
5458 | } |
5459 | ||
0a31bc97 JW |
5460 | /** |
5461 | * mem_cgroup_uncharge - uncharge a page | |
5462 | * @page: page to uncharge | |
5463 | * | |
5464 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
5465 | * mem_cgroup_commit_charge(). | |
5466 | */ | |
5467 | void mem_cgroup_uncharge(struct page *page) | |
5468 | { | |
0a31bc97 JW |
5469 | if (mem_cgroup_disabled()) |
5470 | return; | |
5471 | ||
747db954 | 5472 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 5473 | if (!page->mem_cgroup) |
0a31bc97 JW |
5474 | return; |
5475 | ||
747db954 JW |
5476 | INIT_LIST_HEAD(&page->lru); |
5477 | uncharge_list(&page->lru); | |
5478 | } | |
0a31bc97 | 5479 | |
747db954 JW |
5480 | /** |
5481 | * mem_cgroup_uncharge_list - uncharge a list of page | |
5482 | * @page_list: list of pages to uncharge | |
5483 | * | |
5484 | * Uncharge a list of pages previously charged with | |
5485 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
5486 | */ | |
5487 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
5488 | { | |
5489 | if (mem_cgroup_disabled()) | |
5490 | return; | |
0a31bc97 | 5491 | |
747db954 JW |
5492 | if (!list_empty(page_list)) |
5493 | uncharge_list(page_list); | |
0a31bc97 JW |
5494 | } |
5495 | ||
5496 | /** | |
45637bab | 5497 | * mem_cgroup_replace_page - migrate a charge to another page |
0a31bc97 JW |
5498 | * @oldpage: currently charged page |
5499 | * @newpage: page to transfer the charge to | |
0a31bc97 JW |
5500 | * |
5501 | * Migrate the charge from @oldpage to @newpage. | |
5502 | * | |
5503 | * Both pages must be locked, @newpage->mapping must be set up. | |
25be6a65 | 5504 | * Either or both pages might be on the LRU already. |
0a31bc97 | 5505 | */ |
45637bab | 5506 | void mem_cgroup_replace_page(struct page *oldpage, struct page *newpage) |
0a31bc97 | 5507 | { |
29833315 | 5508 | struct mem_cgroup *memcg; |
0a31bc97 JW |
5509 | int isolated; |
5510 | ||
5511 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
5512 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 5513 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
5514 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
5515 | newpage); | |
0a31bc97 JW |
5516 | |
5517 | if (mem_cgroup_disabled()) | |
5518 | return; | |
5519 | ||
5520 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 5521 | if (newpage->mem_cgroup) |
0a31bc97 JW |
5522 | return; |
5523 | ||
45637bab | 5524 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 5525 | memcg = oldpage->mem_cgroup; |
29833315 | 5526 | if (!memcg) |
0a31bc97 JW |
5527 | return; |
5528 | ||
45637bab | 5529 | lock_page_lru(oldpage, &isolated); |
1306a85a | 5530 | oldpage->mem_cgroup = NULL; |
45637bab | 5531 | unlock_page_lru(oldpage, isolated); |
0a31bc97 | 5532 | |
45637bab | 5533 | commit_charge(newpage, memcg, true); |
0a31bc97 JW |
5534 | } |
5535 | ||
11092087 JW |
5536 | /* Writing them here to avoid exposing memcg's inner layout */ |
5537 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) | |
5538 | ||
5539 | struct static_key memcg_sockets_enabled_key; | |
5540 | EXPORT_SYMBOL(memcg_sockets_enabled_key); | |
5541 | ||
5542 | void sock_update_memcg(struct sock *sk) | |
5543 | { | |
5544 | struct mem_cgroup *memcg; | |
5545 | ||
5546 | /* Socket cloning can throw us here with sk_cgrp already | |
5547 | * filled. It won't however, necessarily happen from | |
5548 | * process context. So the test for root memcg given | |
5549 | * the current task's memcg won't help us in this case. | |
5550 | * | |
5551 | * Respecting the original socket's memcg is a better | |
5552 | * decision in this case. | |
5553 | */ | |
5554 | if (sk->sk_memcg) { | |
5555 | BUG_ON(mem_cgroup_is_root(sk->sk_memcg)); | |
5556 | css_get(&sk->sk_memcg->css); | |
5557 | return; | |
5558 | } | |
5559 | ||
5560 | rcu_read_lock(); | |
5561 | memcg = mem_cgroup_from_task(current); | |
5562 | if (memcg != root_mem_cgroup && | |
5563 | memcg->tcp_mem.active && | |
5564 | css_tryget_online(&memcg->css)) | |
5565 | sk->sk_memcg = memcg; | |
5566 | rcu_read_unlock(); | |
5567 | } | |
5568 | EXPORT_SYMBOL(sock_update_memcg); | |
5569 | ||
5570 | void sock_release_memcg(struct sock *sk) | |
5571 | { | |
5572 | WARN_ON(!sk->sk_memcg); | |
5573 | css_put(&sk->sk_memcg->css); | |
5574 | } | |
5575 | ||
5576 | /** | |
5577 | * mem_cgroup_charge_skmem - charge socket memory | |
5578 | * @memcg: memcg to charge | |
5579 | * @nr_pages: number of pages to charge | |
5580 | * | |
5581 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
5582 | * @memcg's configured limit, %false if the charge had to be forced. | |
5583 | */ | |
5584 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
5585 | { | |
5586 | struct page_counter *counter; | |
5587 | ||
5588 | if (page_counter_try_charge(&memcg->tcp_mem.memory_allocated, | |
5589 | nr_pages, &counter)) { | |
5590 | memcg->tcp_mem.memory_pressure = 0; | |
5591 | return true; | |
5592 | } | |
5593 | page_counter_charge(&memcg->tcp_mem.memory_allocated, nr_pages); | |
5594 | memcg->tcp_mem.memory_pressure = 1; | |
5595 | return false; | |
5596 | } | |
5597 | ||
5598 | /** | |
5599 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
5600 | * @memcg - memcg to uncharge | |
5601 | * @nr_pages - number of pages to uncharge | |
5602 | */ | |
5603 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
5604 | { | |
5605 | page_counter_uncharge(&memcg->tcp_mem.memory_allocated, nr_pages); | |
5606 | } | |
5607 | ||
5608 | #endif | |
5609 | ||
2d11085e | 5610 | /* |
1081312f MH |
5611 | * subsys_initcall() for memory controller. |
5612 | * | |
5613 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
5614 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
5615 | * everything that doesn't depend on a specific mem_cgroup structure should | |
5616 | * be initialized from here. | |
2d11085e MH |
5617 | */ |
5618 | static int __init mem_cgroup_init(void) | |
5619 | { | |
95a045f6 JW |
5620 | int cpu, node; |
5621 | ||
2d11085e | 5622 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
95a045f6 JW |
5623 | |
5624 | for_each_possible_cpu(cpu) | |
5625 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
5626 | drain_local_stock); | |
5627 | ||
5628 | for_each_node(node) { | |
5629 | struct mem_cgroup_tree_per_node *rtpn; | |
5630 | int zone; | |
5631 | ||
5632 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
5633 | node_online(node) ? node : NUMA_NO_NODE); | |
5634 | ||
5635 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
5636 | struct mem_cgroup_tree_per_zone *rtpz; | |
5637 | ||
5638 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
5639 | rtpz->rb_root = RB_ROOT; | |
5640 | spin_lock_init(&rtpz->lock); | |
5641 | } | |
5642 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
5643 | } | |
5644 | ||
2d11085e MH |
5645 | return 0; |
5646 | } | |
5647 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
5648 | |
5649 | #ifdef CONFIG_MEMCG_SWAP | |
5650 | /** | |
5651 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
5652 | * @page: page whose memsw charge to transfer | |
5653 | * @entry: swap entry to move the charge to | |
5654 | * | |
5655 | * Transfer the memsw charge of @page to @entry. | |
5656 | */ | |
5657 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
5658 | { | |
5659 | struct mem_cgroup *memcg; | |
5660 | unsigned short oldid; | |
5661 | ||
5662 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5663 | VM_BUG_ON_PAGE(page_count(page), page); | |
5664 | ||
7941d214 | 5665 | if (!do_memsw_account()) |
21afa38e JW |
5666 | return; |
5667 | ||
5668 | memcg = page->mem_cgroup; | |
5669 | ||
5670 | /* Readahead page, never charged */ | |
5671 | if (!memcg) | |
5672 | return; | |
5673 | ||
5674 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); | |
5675 | VM_BUG_ON_PAGE(oldid, page); | |
5676 | mem_cgroup_swap_statistics(memcg, true); | |
5677 | ||
5678 | page->mem_cgroup = NULL; | |
5679 | ||
5680 | if (!mem_cgroup_is_root(memcg)) | |
5681 | page_counter_uncharge(&memcg->memory, 1); | |
5682 | ||
ce9ce665 SAS |
5683 | /* |
5684 | * Interrupts should be disabled here because the caller holds the | |
5685 | * mapping->tree_lock lock which is taken with interrupts-off. It is | |
5686 | * important here to have the interrupts disabled because it is the | |
5687 | * only synchronisation we have for udpating the per-CPU variables. | |
5688 | */ | |
5689 | VM_BUG_ON(!irqs_disabled()); | |
21afa38e JW |
5690 | mem_cgroup_charge_statistics(memcg, page, -1); |
5691 | memcg_check_events(memcg, page); | |
5692 | } | |
5693 | ||
5694 | /** | |
5695 | * mem_cgroup_uncharge_swap - uncharge a swap entry | |
5696 | * @entry: swap entry to uncharge | |
5697 | * | |
5698 | * Drop the memsw charge associated with @entry. | |
5699 | */ | |
5700 | void mem_cgroup_uncharge_swap(swp_entry_t entry) | |
5701 | { | |
5702 | struct mem_cgroup *memcg; | |
5703 | unsigned short id; | |
5704 | ||
7941d214 | 5705 | if (!do_memsw_account()) |
21afa38e JW |
5706 | return; |
5707 | ||
5708 | id = swap_cgroup_record(entry, 0); | |
5709 | rcu_read_lock(); | |
adbe427b | 5710 | memcg = mem_cgroup_from_id(id); |
21afa38e JW |
5711 | if (memcg) { |
5712 | if (!mem_cgroup_is_root(memcg)) | |
5713 | page_counter_uncharge(&memcg->memsw, 1); | |
5714 | mem_cgroup_swap_statistics(memcg, false); | |
5715 | css_put(&memcg->css); | |
5716 | } | |
5717 | rcu_read_unlock(); | |
5718 | } | |
5719 | ||
5720 | /* for remember boot option*/ | |
5721 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
5722 | static int really_do_swap_account __initdata = 1; | |
5723 | #else | |
5724 | static int really_do_swap_account __initdata; | |
5725 | #endif | |
5726 | ||
5727 | static int __init enable_swap_account(char *s) | |
5728 | { | |
5729 | if (!strcmp(s, "1")) | |
5730 | really_do_swap_account = 1; | |
5731 | else if (!strcmp(s, "0")) | |
5732 | really_do_swap_account = 0; | |
5733 | return 1; | |
5734 | } | |
5735 | __setup("swapaccount=", enable_swap_account); | |
5736 | ||
5737 | static struct cftype memsw_cgroup_files[] = { | |
5738 | { | |
5739 | .name = "memsw.usage_in_bytes", | |
5740 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
5741 | .read_u64 = mem_cgroup_read_u64, | |
5742 | }, | |
5743 | { | |
5744 | .name = "memsw.max_usage_in_bytes", | |
5745 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
5746 | .write = mem_cgroup_reset, | |
5747 | .read_u64 = mem_cgroup_read_u64, | |
5748 | }, | |
5749 | { | |
5750 | .name = "memsw.limit_in_bytes", | |
5751 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
5752 | .write = mem_cgroup_write, | |
5753 | .read_u64 = mem_cgroup_read_u64, | |
5754 | }, | |
5755 | { | |
5756 | .name = "memsw.failcnt", | |
5757 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
5758 | .write = mem_cgroup_reset, | |
5759 | .read_u64 = mem_cgroup_read_u64, | |
5760 | }, | |
5761 | { }, /* terminate */ | |
5762 | }; | |
5763 | ||
5764 | static int __init mem_cgroup_swap_init(void) | |
5765 | { | |
5766 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
5767 | do_swap_account = 1; | |
5768 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, | |
5769 | memsw_cgroup_files)); | |
5770 | } | |
5771 | return 0; | |
5772 | } | |
5773 | subsys_initcall(mem_cgroup_swap_init); | |
5774 | ||
5775 | #endif /* CONFIG_MEMCG_SWAP */ |