]>
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> |
6e84f315 | 38 | #include <linux/sched/mm.h> |
3a4f8a0b | 39 | #include <linux/shmem_fs.h> |
4ffef5fe | 40 | #include <linux/hugetlb.h> |
d13d1443 | 41 | #include <linux/pagemap.h> |
d52aa412 | 42 | #include <linux/smp.h> |
8a9f3ccd | 43 | #include <linux/page-flags.h> |
66e1707b | 44 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
45 | #include <linux/bit_spinlock.h> |
46 | #include <linux/rcupdate.h> | |
e222432b | 47 | #include <linux/limits.h> |
b9e15baf | 48 | #include <linux/export.h> |
8c7c6e34 | 49 | #include <linux/mutex.h> |
bb4cc1a8 | 50 | #include <linux/rbtree.h> |
b6ac57d5 | 51 | #include <linux/slab.h> |
66e1707b | 52 | #include <linux/swap.h> |
02491447 | 53 | #include <linux/swapops.h> |
66e1707b | 54 | #include <linux/spinlock.h> |
2e72b634 | 55 | #include <linux/eventfd.h> |
79bd9814 | 56 | #include <linux/poll.h> |
2e72b634 | 57 | #include <linux/sort.h> |
66e1707b | 58 | #include <linux/fs.h> |
d2ceb9b7 | 59 | #include <linux/seq_file.h> |
70ddf637 | 60 | #include <linux/vmpressure.h> |
b69408e8 | 61 | #include <linux/mm_inline.h> |
5d1ea48b | 62 | #include <linux/swap_cgroup.h> |
cdec2e42 | 63 | #include <linux/cpu.h> |
158e0a2d | 64 | #include <linux/oom.h> |
0056f4e6 | 65 | #include <linux/lockdep.h> |
79bd9814 | 66 | #include <linux/file.h> |
b23afb93 | 67 | #include <linux/tracehook.h> |
08e552c6 | 68 | #include "internal.h" |
d1a4c0b3 | 69 | #include <net/sock.h> |
4bd2c1ee | 70 | #include <net/ip.h> |
f35c3a8e | 71 | #include "slab.h" |
8cdea7c0 | 72 | |
7c0f6ba6 | 73 | #include <linux/uaccess.h> |
8697d331 | 74 | |
cc8e970c KM |
75 | #include <trace/events/vmscan.h> |
76 | ||
073219e9 TH |
77 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
78 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 79 | |
7d828602 JW |
80 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
81 | ||
a181b0e8 | 82 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
8cdea7c0 | 83 | |
f7e1cb6e JW |
84 | /* Socket memory accounting disabled? */ |
85 | static bool cgroup_memory_nosocket; | |
86 | ||
04823c83 VD |
87 | /* Kernel memory accounting disabled? */ |
88 | static bool cgroup_memory_nokmem; | |
89 | ||
21afa38e | 90 | /* Whether the swap controller is active */ |
c255a458 | 91 | #ifdef CONFIG_MEMCG_SWAP |
c077719b | 92 | int do_swap_account __read_mostly; |
c077719b | 93 | #else |
a0db00fc | 94 | #define do_swap_account 0 |
c077719b KH |
95 | #endif |
96 | ||
7941d214 JW |
97 | /* Whether legacy memory+swap accounting is active */ |
98 | static bool do_memsw_account(void) | |
99 | { | |
100 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account; | |
101 | } | |
102 | ||
71cd3113 | 103 | static const char *const mem_cgroup_lru_names[] = { |
58cf188e SZ |
104 | "inactive_anon", |
105 | "active_anon", | |
106 | "inactive_file", | |
107 | "active_file", | |
108 | "unevictable", | |
109 | }; | |
110 | ||
a0db00fc KS |
111 | #define THRESHOLDS_EVENTS_TARGET 128 |
112 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
113 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 114 | |
bb4cc1a8 AM |
115 | /* |
116 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
117 | * their hierarchy representation | |
118 | */ | |
119 | ||
ef8f2327 | 120 | struct mem_cgroup_tree_per_node { |
bb4cc1a8 | 121 | struct rb_root rb_root; |
fa90b2fd | 122 | struct rb_node *rb_rightmost; |
bb4cc1a8 AM |
123 | spinlock_t lock; |
124 | }; | |
125 | ||
bb4cc1a8 AM |
126 | struct mem_cgroup_tree { |
127 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
128 | }; | |
129 | ||
130 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
131 | ||
9490ff27 KH |
132 | /* for OOM */ |
133 | struct mem_cgroup_eventfd_list { | |
134 | struct list_head list; | |
135 | struct eventfd_ctx *eventfd; | |
136 | }; | |
2e72b634 | 137 | |
79bd9814 TH |
138 | /* |
139 | * cgroup_event represents events which userspace want to receive. | |
140 | */ | |
3bc942f3 | 141 | struct mem_cgroup_event { |
79bd9814 | 142 | /* |
59b6f873 | 143 | * memcg which the event belongs to. |
79bd9814 | 144 | */ |
59b6f873 | 145 | struct mem_cgroup *memcg; |
79bd9814 TH |
146 | /* |
147 | * eventfd to signal userspace about the event. | |
148 | */ | |
149 | struct eventfd_ctx *eventfd; | |
150 | /* | |
151 | * Each of these stored in a list by the cgroup. | |
152 | */ | |
153 | struct list_head list; | |
fba94807 TH |
154 | /* |
155 | * register_event() callback will be used to add new userspace | |
156 | * waiter for changes related to this event. Use eventfd_signal() | |
157 | * on eventfd to send notification to userspace. | |
158 | */ | |
59b6f873 | 159 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 160 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
161 | /* |
162 | * unregister_event() callback will be called when userspace closes | |
163 | * the eventfd or on cgroup removing. This callback must be set, | |
164 | * if you want provide notification functionality. | |
165 | */ | |
59b6f873 | 166 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 167 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
168 | /* |
169 | * All fields below needed to unregister event when | |
170 | * userspace closes eventfd. | |
171 | */ | |
172 | poll_table pt; | |
173 | wait_queue_head_t *wqh; | |
ac6424b9 | 174 | wait_queue_entry_t wait; |
79bd9814 TH |
175 | struct work_struct remove; |
176 | }; | |
177 | ||
c0ff4b85 R |
178 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
179 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 180 | |
7dc74be0 DN |
181 | /* Stuffs for move charges at task migration. */ |
182 | /* | |
1dfab5ab | 183 | * Types of charges to be moved. |
7dc74be0 | 184 | */ |
1dfab5ab JW |
185 | #define MOVE_ANON 0x1U |
186 | #define MOVE_FILE 0x2U | |
187 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 188 | |
4ffef5fe DN |
189 | /* "mc" and its members are protected by cgroup_mutex */ |
190 | static struct move_charge_struct { | |
b1dd693e | 191 | spinlock_t lock; /* for from, to */ |
264a0ae1 | 192 | struct mm_struct *mm; |
4ffef5fe DN |
193 | struct mem_cgroup *from; |
194 | struct mem_cgroup *to; | |
1dfab5ab | 195 | unsigned long flags; |
4ffef5fe | 196 | unsigned long precharge; |
854ffa8d | 197 | unsigned long moved_charge; |
483c30b5 | 198 | unsigned long moved_swap; |
8033b97c DN |
199 | struct task_struct *moving_task; /* a task moving charges */ |
200 | wait_queue_head_t waitq; /* a waitq for other context */ | |
201 | } mc = { | |
2bd9bb20 | 202 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
203 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
204 | }; | |
4ffef5fe | 205 | |
4e416953 BS |
206 | /* |
207 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
208 | * limit reclaim to prevent infinite loops, if they ever occur. | |
209 | */ | |
a0db00fc | 210 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 211 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 212 | |
217bc319 KH |
213 | enum charge_type { |
214 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 215 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 216 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 217 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
218 | NR_CHARGE_TYPE, |
219 | }; | |
220 | ||
8c7c6e34 | 221 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
222 | enum res_type { |
223 | _MEM, | |
224 | _MEMSWAP, | |
225 | _OOM_TYPE, | |
510fc4e1 | 226 | _KMEM, |
d55f90bf | 227 | _TCP, |
86ae53e1 GC |
228 | }; |
229 | ||
a0db00fc KS |
230 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
231 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 232 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
233 | /* Used for OOM nofiier */ |
234 | #define OOM_CONTROL (0) | |
8c7c6e34 | 235 | |
b05706f1 KT |
236 | /* |
237 | * Iteration constructs for visiting all cgroups (under a tree). If | |
238 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
239 | * be used for reference counting. | |
240 | */ | |
241 | #define for_each_mem_cgroup_tree(iter, root) \ | |
242 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ | |
243 | iter != NULL; \ | |
244 | iter = mem_cgroup_iter(root, iter, NULL)) | |
245 | ||
246 | #define for_each_mem_cgroup(iter) \ | |
247 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ | |
248 | iter != NULL; \ | |
249 | iter = mem_cgroup_iter(NULL, iter, NULL)) | |
250 | ||
70ddf637 AV |
251 | /* Some nice accessors for the vmpressure. */ |
252 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
253 | { | |
254 | if (!memcg) | |
255 | memcg = root_mem_cgroup; | |
256 | return &memcg->vmpressure; | |
257 | } | |
258 | ||
259 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
260 | { | |
261 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
262 | } | |
263 | ||
7ffc0edc MH |
264 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
265 | { | |
266 | return (memcg == root_mem_cgroup); | |
267 | } | |
268 | ||
84c07d11 | 269 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 | 270 | /* |
f7ce3190 | 271 | * This will be the memcg's index in each cache's ->memcg_params.memcg_caches. |
b8627835 LZ |
272 | * The main reason for not using cgroup id for this: |
273 | * this works better in sparse environments, where we have a lot of memcgs, | |
274 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
275 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
276 | * 200 entry array for that. | |
55007d84 | 277 | * |
dbcf73e2 VD |
278 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
279 | * will double each time we have to increase it. | |
55007d84 | 280 | */ |
dbcf73e2 VD |
281 | static DEFINE_IDA(memcg_cache_ida); |
282 | int memcg_nr_cache_ids; | |
749c5415 | 283 | |
05257a1a VD |
284 | /* Protects memcg_nr_cache_ids */ |
285 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
286 | ||
287 | void memcg_get_cache_ids(void) | |
288 | { | |
289 | down_read(&memcg_cache_ids_sem); | |
290 | } | |
291 | ||
292 | void memcg_put_cache_ids(void) | |
293 | { | |
294 | up_read(&memcg_cache_ids_sem); | |
295 | } | |
296 | ||
55007d84 GC |
297 | /* |
298 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
299 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
300 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
301 | * tunable, but that is strictly not necessary. | |
302 | * | |
b8627835 | 303 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
304 | * this constant directly from cgroup, but it is understandable that this is |
305 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 306 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
307 | * increase ours as well if it increases. |
308 | */ | |
309 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 310 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 311 | |
d7f25f8a GC |
312 | /* |
313 | * A lot of the calls to the cache allocation functions are expected to be | |
314 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
315 | * conditional to this static branch, we'll have to allow modules that does | |
316 | * kmem_cache_alloc and the such to see this symbol as well | |
317 | */ | |
ef12947c | 318 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
d7f25f8a | 319 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 320 | |
17cc4dfe TH |
321 | struct workqueue_struct *memcg_kmem_cache_wq; |
322 | ||
0a4465d3 KT |
323 | static int memcg_shrinker_map_size; |
324 | static DEFINE_MUTEX(memcg_shrinker_map_mutex); | |
325 | ||
326 | static void memcg_free_shrinker_map_rcu(struct rcu_head *head) | |
327 | { | |
328 | kvfree(container_of(head, struct memcg_shrinker_map, rcu)); | |
329 | } | |
330 | ||
331 | static int memcg_expand_one_shrinker_map(struct mem_cgroup *memcg, | |
332 | int size, int old_size) | |
333 | { | |
334 | struct memcg_shrinker_map *new, *old; | |
335 | int nid; | |
336 | ||
337 | lockdep_assert_held(&memcg_shrinker_map_mutex); | |
338 | ||
339 | for_each_node(nid) { | |
340 | old = rcu_dereference_protected( | |
341 | mem_cgroup_nodeinfo(memcg, nid)->shrinker_map, true); | |
342 | /* Not yet online memcg */ | |
343 | if (!old) | |
344 | return 0; | |
345 | ||
346 | new = kvmalloc(sizeof(*new) + size, GFP_KERNEL); | |
347 | if (!new) | |
348 | return -ENOMEM; | |
349 | ||
350 | /* Set all old bits, clear all new bits */ | |
351 | memset(new->map, (int)0xff, old_size); | |
352 | memset((void *)new->map + old_size, 0, size - old_size); | |
353 | ||
354 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, new); | |
355 | call_rcu(&old->rcu, memcg_free_shrinker_map_rcu); | |
356 | } | |
357 | ||
358 | return 0; | |
359 | } | |
360 | ||
361 | static void memcg_free_shrinker_maps(struct mem_cgroup *memcg) | |
362 | { | |
363 | struct mem_cgroup_per_node *pn; | |
364 | struct memcg_shrinker_map *map; | |
365 | int nid; | |
366 | ||
367 | if (mem_cgroup_is_root(memcg)) | |
368 | return; | |
369 | ||
370 | for_each_node(nid) { | |
371 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
372 | map = rcu_dereference_protected(pn->shrinker_map, true); | |
373 | if (map) | |
374 | kvfree(map); | |
375 | rcu_assign_pointer(pn->shrinker_map, NULL); | |
376 | } | |
377 | } | |
378 | ||
379 | static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg) | |
380 | { | |
381 | struct memcg_shrinker_map *map; | |
382 | int nid, size, ret = 0; | |
383 | ||
384 | if (mem_cgroup_is_root(memcg)) | |
385 | return 0; | |
386 | ||
387 | mutex_lock(&memcg_shrinker_map_mutex); | |
388 | size = memcg_shrinker_map_size; | |
389 | for_each_node(nid) { | |
390 | map = kvzalloc(sizeof(*map) + size, GFP_KERNEL); | |
391 | if (!map) { | |
392 | memcg_free_shrinker_maps(memcg); | |
393 | ret = -ENOMEM; | |
394 | break; | |
395 | } | |
396 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, map); | |
397 | } | |
398 | mutex_unlock(&memcg_shrinker_map_mutex); | |
399 | ||
400 | return ret; | |
401 | } | |
402 | ||
403 | int memcg_expand_shrinker_maps(int new_id) | |
404 | { | |
405 | int size, old_size, ret = 0; | |
406 | struct mem_cgroup *memcg; | |
407 | ||
408 | size = DIV_ROUND_UP(new_id + 1, BITS_PER_LONG) * sizeof(unsigned long); | |
409 | old_size = memcg_shrinker_map_size; | |
410 | if (size <= old_size) | |
411 | return 0; | |
412 | ||
413 | mutex_lock(&memcg_shrinker_map_mutex); | |
414 | if (!root_mem_cgroup) | |
415 | goto unlock; | |
416 | ||
417 | for_each_mem_cgroup(memcg) { | |
418 | if (mem_cgroup_is_root(memcg)) | |
419 | continue; | |
420 | ret = memcg_expand_one_shrinker_map(memcg, size, old_size); | |
421 | if (ret) | |
422 | goto unlock; | |
423 | } | |
424 | unlock: | |
425 | if (!ret) | |
426 | memcg_shrinker_map_size = size; | |
427 | mutex_unlock(&memcg_shrinker_map_mutex); | |
428 | return ret; | |
429 | } | |
430 | #else /* CONFIG_MEMCG_KMEM */ | |
431 | static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg) | |
432 | { | |
433 | return 0; | |
434 | } | |
435 | static void memcg_free_shrinker_maps(struct mem_cgroup *memcg) { } | |
84c07d11 | 436 | #endif /* CONFIG_MEMCG_KMEM */ |
a8964b9b | 437 | |
ad7fa852 TH |
438 | /** |
439 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
440 | * @page: page of interest | |
441 | * | |
442 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
443 | * with @page is returned. The returned css remains associated with @page | |
444 | * until it is released. | |
445 | * | |
446 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
447 | * is returned. | |
ad7fa852 TH |
448 | */ |
449 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
450 | { | |
451 | struct mem_cgroup *memcg; | |
452 | ||
ad7fa852 TH |
453 | memcg = page->mem_cgroup; |
454 | ||
9e10a130 | 455 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
456 | memcg = root_mem_cgroup; |
457 | ||
ad7fa852 TH |
458 | return &memcg->css; |
459 | } | |
460 | ||
2fc04524 VD |
461 | /** |
462 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
463 | * @page: the page | |
464 | * | |
465 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
466 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
467 | * is safe to call this function without holding a reference to @page. | |
468 | * | |
469 | * Note, this function is inherently racy, because there is nothing to prevent | |
470 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
471 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
472 | * do not care (such as procfs interfaces). | |
473 | */ | |
474 | ino_t page_cgroup_ino(struct page *page) | |
475 | { | |
476 | struct mem_cgroup *memcg; | |
477 | unsigned long ino = 0; | |
478 | ||
479 | rcu_read_lock(); | |
480 | memcg = READ_ONCE(page->mem_cgroup); | |
481 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) | |
482 | memcg = parent_mem_cgroup(memcg); | |
483 | if (memcg) | |
484 | ino = cgroup_ino(memcg->css.cgroup); | |
485 | rcu_read_unlock(); | |
486 | return ino; | |
487 | } | |
488 | ||
ef8f2327 MG |
489 | static struct mem_cgroup_per_node * |
490 | mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page) | |
f64c3f54 | 491 | { |
97a6c37b | 492 | int nid = page_to_nid(page); |
f64c3f54 | 493 | |
ef8f2327 | 494 | return memcg->nodeinfo[nid]; |
f64c3f54 BS |
495 | } |
496 | ||
ef8f2327 MG |
497 | static struct mem_cgroup_tree_per_node * |
498 | soft_limit_tree_node(int nid) | |
bb4cc1a8 | 499 | { |
ef8f2327 | 500 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
501 | } |
502 | ||
ef8f2327 | 503 | static struct mem_cgroup_tree_per_node * |
bb4cc1a8 AM |
504 | soft_limit_tree_from_page(struct page *page) |
505 | { | |
506 | int nid = page_to_nid(page); | |
bb4cc1a8 | 507 | |
ef8f2327 | 508 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
509 | } |
510 | ||
ef8f2327 MG |
511 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
512 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 513 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
514 | { |
515 | struct rb_node **p = &mctz->rb_root.rb_node; | |
516 | struct rb_node *parent = NULL; | |
ef8f2327 | 517 | struct mem_cgroup_per_node *mz_node; |
fa90b2fd | 518 | bool rightmost = true; |
bb4cc1a8 AM |
519 | |
520 | if (mz->on_tree) | |
521 | return; | |
522 | ||
523 | mz->usage_in_excess = new_usage_in_excess; | |
524 | if (!mz->usage_in_excess) | |
525 | return; | |
526 | while (*p) { | |
527 | parent = *p; | |
ef8f2327 | 528 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 | 529 | tree_node); |
fa90b2fd | 530 | if (mz->usage_in_excess < mz_node->usage_in_excess) { |
bb4cc1a8 | 531 | p = &(*p)->rb_left; |
fa90b2fd DB |
532 | rightmost = false; |
533 | } | |
534 | ||
bb4cc1a8 AM |
535 | /* |
536 | * We can't avoid mem cgroups that are over their soft | |
537 | * limit by the same amount | |
538 | */ | |
539 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
540 | p = &(*p)->rb_right; | |
541 | } | |
fa90b2fd DB |
542 | |
543 | if (rightmost) | |
544 | mctz->rb_rightmost = &mz->tree_node; | |
545 | ||
bb4cc1a8 AM |
546 | rb_link_node(&mz->tree_node, parent, p); |
547 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
548 | mz->on_tree = true; | |
549 | } | |
550 | ||
ef8f2327 MG |
551 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
552 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
553 | { |
554 | if (!mz->on_tree) | |
555 | return; | |
fa90b2fd DB |
556 | |
557 | if (&mz->tree_node == mctz->rb_rightmost) | |
558 | mctz->rb_rightmost = rb_prev(&mz->tree_node); | |
559 | ||
bb4cc1a8 AM |
560 | rb_erase(&mz->tree_node, &mctz->rb_root); |
561 | mz->on_tree = false; | |
562 | } | |
563 | ||
ef8f2327 MG |
564 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
565 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 566 | { |
0a31bc97 JW |
567 | unsigned long flags; |
568 | ||
569 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 570 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 571 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
572 | } |
573 | ||
3e32cb2e JW |
574 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
575 | { | |
576 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 577 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
578 | unsigned long excess = 0; |
579 | ||
580 | if (nr_pages > soft_limit) | |
581 | excess = nr_pages - soft_limit; | |
582 | ||
583 | return excess; | |
584 | } | |
bb4cc1a8 AM |
585 | |
586 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
587 | { | |
3e32cb2e | 588 | unsigned long excess; |
ef8f2327 MG |
589 | struct mem_cgroup_per_node *mz; |
590 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 591 | |
e231875b | 592 | mctz = soft_limit_tree_from_page(page); |
bfc7228b LD |
593 | if (!mctz) |
594 | return; | |
bb4cc1a8 AM |
595 | /* |
596 | * Necessary to update all ancestors when hierarchy is used. | |
597 | * because their event counter is not touched. | |
598 | */ | |
599 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
ef8f2327 | 600 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
3e32cb2e | 601 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
602 | /* |
603 | * We have to update the tree if mz is on RB-tree or | |
604 | * mem is over its softlimit. | |
605 | */ | |
606 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
607 | unsigned long flags; |
608 | ||
609 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
610 | /* if on-tree, remove it */ |
611 | if (mz->on_tree) | |
cf2c8127 | 612 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
613 | /* |
614 | * Insert again. mz->usage_in_excess will be updated. | |
615 | * If excess is 0, no tree ops. | |
616 | */ | |
cf2c8127 | 617 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 618 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
619 | } |
620 | } | |
621 | } | |
622 | ||
623 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
624 | { | |
ef8f2327 MG |
625 | struct mem_cgroup_tree_per_node *mctz; |
626 | struct mem_cgroup_per_node *mz; | |
627 | int nid; | |
bb4cc1a8 | 628 | |
e231875b | 629 | for_each_node(nid) { |
ef8f2327 MG |
630 | mz = mem_cgroup_nodeinfo(memcg, nid); |
631 | mctz = soft_limit_tree_node(nid); | |
bfc7228b LD |
632 | if (mctz) |
633 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
634 | } |
635 | } | |
636 | ||
ef8f2327 MG |
637 | static struct mem_cgroup_per_node * |
638 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 639 | { |
ef8f2327 | 640 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
641 | |
642 | retry: | |
643 | mz = NULL; | |
fa90b2fd | 644 | if (!mctz->rb_rightmost) |
bb4cc1a8 AM |
645 | goto done; /* Nothing to reclaim from */ |
646 | ||
fa90b2fd DB |
647 | mz = rb_entry(mctz->rb_rightmost, |
648 | struct mem_cgroup_per_node, tree_node); | |
bb4cc1a8 AM |
649 | /* |
650 | * Remove the node now but someone else can add it back, | |
651 | * we will to add it back at the end of reclaim to its correct | |
652 | * position in the tree. | |
653 | */ | |
cf2c8127 | 654 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 655 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 656 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
657 | goto retry; |
658 | done: | |
659 | return mz; | |
660 | } | |
661 | ||
ef8f2327 MG |
662 | static struct mem_cgroup_per_node * |
663 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 664 | { |
ef8f2327 | 665 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 666 | |
0a31bc97 | 667 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 668 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 669 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
670 | return mz; |
671 | } | |
672 | ||
ccda7f43 | 673 | static unsigned long memcg_sum_events(struct mem_cgroup *memcg, |
04fecbf5 | 674 | int event) |
e9f8974f | 675 | { |
a983b5eb | 676 | return atomic_long_read(&memcg->events[event]); |
e9f8974f JW |
677 | } |
678 | ||
c0ff4b85 | 679 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 680 | struct page *page, |
f627c2f5 | 681 | bool compound, int nr_pages) |
d52aa412 | 682 | { |
b2402857 KH |
683 | /* |
684 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
685 | * counted as CACHE even if it's on ANON LRU. | |
686 | */ | |
0a31bc97 | 687 | if (PageAnon(page)) |
c9019e9b | 688 | __mod_memcg_state(memcg, MEMCG_RSS, nr_pages); |
9a4caf1e | 689 | else { |
c9019e9b | 690 | __mod_memcg_state(memcg, MEMCG_CACHE, nr_pages); |
9a4caf1e | 691 | if (PageSwapBacked(page)) |
c9019e9b | 692 | __mod_memcg_state(memcg, NR_SHMEM, nr_pages); |
9a4caf1e | 693 | } |
55e462b0 | 694 | |
f627c2f5 KS |
695 | if (compound) { |
696 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
c9019e9b | 697 | __mod_memcg_state(memcg, MEMCG_RSS_HUGE, nr_pages); |
f627c2f5 | 698 | } |
b070e65c | 699 | |
e401f176 KH |
700 | /* pagein of a big page is an event. So, ignore page size */ |
701 | if (nr_pages > 0) | |
c9019e9b | 702 | __count_memcg_events(memcg, PGPGIN, 1); |
3751d604 | 703 | else { |
c9019e9b | 704 | __count_memcg_events(memcg, PGPGOUT, 1); |
3751d604 KH |
705 | nr_pages = -nr_pages; /* for event */ |
706 | } | |
e401f176 | 707 | |
a983b5eb | 708 | __this_cpu_add(memcg->stat_cpu->nr_page_events, nr_pages); |
6d12e2d8 KH |
709 | } |
710 | ||
0a6b76dd VD |
711 | unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
712 | int nid, unsigned int lru_mask) | |
bb2a0de9 | 713 | { |
b4536f0c | 714 | struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg); |
e231875b | 715 | unsigned long nr = 0; |
ef8f2327 | 716 | enum lru_list lru; |
889976db | 717 | |
e231875b | 718 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 719 | |
ef8f2327 MG |
720 | for_each_lru(lru) { |
721 | if (!(BIT(lru) & lru_mask)) | |
722 | continue; | |
b4536f0c | 723 | nr += mem_cgroup_get_lru_size(lruvec, lru); |
e231875b JZ |
724 | } |
725 | return nr; | |
889976db | 726 | } |
bb2a0de9 | 727 | |
c0ff4b85 | 728 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 729 | unsigned int lru_mask) |
6d12e2d8 | 730 | { |
e231875b | 731 | unsigned long nr = 0; |
889976db | 732 | int nid; |
6d12e2d8 | 733 | |
31aaea4a | 734 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
735 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
736 | return nr; | |
d52aa412 KH |
737 | } |
738 | ||
f53d7ce3 JW |
739 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
740 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
741 | { |
742 | unsigned long val, next; | |
743 | ||
a983b5eb JW |
744 | val = __this_cpu_read(memcg->stat_cpu->nr_page_events); |
745 | next = __this_cpu_read(memcg->stat_cpu->targets[target]); | |
7a159cc9 | 746 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 747 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
748 | switch (target) { |
749 | case MEM_CGROUP_TARGET_THRESH: | |
750 | next = val + THRESHOLDS_EVENTS_TARGET; | |
751 | break; | |
bb4cc1a8 AM |
752 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
753 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
754 | break; | |
f53d7ce3 JW |
755 | case MEM_CGROUP_TARGET_NUMAINFO: |
756 | next = val + NUMAINFO_EVENTS_TARGET; | |
757 | break; | |
758 | default: | |
759 | break; | |
760 | } | |
a983b5eb | 761 | __this_cpu_write(memcg->stat_cpu->targets[target], next); |
f53d7ce3 | 762 | return true; |
7a159cc9 | 763 | } |
f53d7ce3 | 764 | return false; |
d2265e6f KH |
765 | } |
766 | ||
767 | /* | |
768 | * Check events in order. | |
769 | * | |
770 | */ | |
c0ff4b85 | 771 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
772 | { |
773 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
774 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
775 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 776 | bool do_softlimit; |
82b3f2a7 | 777 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 778 | |
bb4cc1a8 AM |
779 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
780 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
781 | #if MAX_NUMNODES > 1 |
782 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
783 | MEM_CGROUP_TARGET_NUMAINFO); | |
784 | #endif | |
c0ff4b85 | 785 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
786 | if (unlikely(do_softlimit)) |
787 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 788 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 789 | if (unlikely(do_numainfo)) |
c0ff4b85 | 790 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 791 | #endif |
0a31bc97 | 792 | } |
d2265e6f KH |
793 | } |
794 | ||
cf475ad2 | 795 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 796 | { |
31a78f23 BS |
797 | /* |
798 | * mm_update_next_owner() may clear mm->owner to NULL | |
799 | * if it races with swapoff, page migration, etc. | |
800 | * So this can be called with p == NULL. | |
801 | */ | |
802 | if (unlikely(!p)) | |
803 | return NULL; | |
804 | ||
073219e9 | 805 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 806 | } |
33398cf2 | 807 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 808 | |
d46eb14b SB |
809 | /** |
810 | * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg. | |
811 | * @mm: mm from which memcg should be extracted. It can be NULL. | |
812 | * | |
813 | * Obtain a reference on mm->memcg and returns it if successful. Otherwise | |
814 | * root_mem_cgroup is returned. However if mem_cgroup is disabled, NULL is | |
815 | * returned. | |
816 | */ | |
817 | struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) | |
54595fe2 | 818 | { |
d46eb14b SB |
819 | struct mem_cgroup *memcg; |
820 | ||
821 | if (mem_cgroup_disabled()) | |
822 | return NULL; | |
0b7f569e | 823 | |
54595fe2 KH |
824 | rcu_read_lock(); |
825 | do { | |
6f6acb00 MH |
826 | /* |
827 | * Page cache insertions can happen withou an | |
828 | * actual mm context, e.g. during disk probing | |
829 | * on boot, loopback IO, acct() writes etc. | |
830 | */ | |
831 | if (unlikely(!mm)) | |
df381975 | 832 | memcg = root_mem_cgroup; |
6f6acb00 MH |
833 | else { |
834 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
835 | if (unlikely(!memcg)) | |
836 | memcg = root_mem_cgroup; | |
837 | } | |
ec903c0c | 838 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 839 | rcu_read_unlock(); |
c0ff4b85 | 840 | return memcg; |
54595fe2 | 841 | } |
d46eb14b SB |
842 | EXPORT_SYMBOL(get_mem_cgroup_from_mm); |
843 | ||
f745c6f5 SB |
844 | /** |
845 | * get_mem_cgroup_from_page: Obtain a reference on given page's memcg. | |
846 | * @page: page from which memcg should be extracted. | |
847 | * | |
848 | * Obtain a reference on page->memcg and returns it if successful. Otherwise | |
849 | * root_mem_cgroup is returned. | |
850 | */ | |
851 | struct mem_cgroup *get_mem_cgroup_from_page(struct page *page) | |
852 | { | |
853 | struct mem_cgroup *memcg = page->mem_cgroup; | |
854 | ||
855 | if (mem_cgroup_disabled()) | |
856 | return NULL; | |
857 | ||
858 | rcu_read_lock(); | |
859 | if (!memcg || !css_tryget_online(&memcg->css)) | |
860 | memcg = root_mem_cgroup; | |
861 | rcu_read_unlock(); | |
862 | return memcg; | |
863 | } | |
864 | EXPORT_SYMBOL(get_mem_cgroup_from_page); | |
865 | ||
d46eb14b SB |
866 | /** |
867 | * If current->active_memcg is non-NULL, do not fallback to current->mm->memcg. | |
868 | */ | |
869 | static __always_inline struct mem_cgroup *get_mem_cgroup_from_current(void) | |
870 | { | |
871 | if (unlikely(current->active_memcg)) { | |
872 | struct mem_cgroup *memcg = root_mem_cgroup; | |
873 | ||
874 | rcu_read_lock(); | |
875 | if (css_tryget_online(¤t->active_memcg->css)) | |
876 | memcg = current->active_memcg; | |
877 | rcu_read_unlock(); | |
878 | return memcg; | |
879 | } | |
880 | return get_mem_cgroup_from_mm(current->mm); | |
881 | } | |
54595fe2 | 882 | |
5660048c JW |
883 | /** |
884 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
885 | * @root: hierarchy root | |
886 | * @prev: previously returned memcg, NULL on first invocation | |
887 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
888 | * | |
889 | * Returns references to children of the hierarchy below @root, or | |
890 | * @root itself, or %NULL after a full round-trip. | |
891 | * | |
892 | * Caller must pass the return value in @prev on subsequent | |
893 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
894 | * to cancel a hierarchy walk before the round-trip is complete. | |
895 | * | |
b213b54f | 896 | * Reclaimers can specify a node and a priority level in @reclaim to |
5660048c | 897 | * divide up the memcgs in the hierarchy among all concurrent |
b213b54f | 898 | * reclaimers operating on the same node and priority. |
5660048c | 899 | */ |
694fbc0f | 900 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 901 | struct mem_cgroup *prev, |
694fbc0f | 902 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 903 | { |
33398cf2 | 904 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
5ac8fb31 | 905 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 906 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 907 | struct mem_cgroup *pos = NULL; |
711d3d2c | 908 | |
694fbc0f AM |
909 | if (mem_cgroup_disabled()) |
910 | return NULL; | |
5660048c | 911 | |
9f3a0d09 JW |
912 | if (!root) |
913 | root = root_mem_cgroup; | |
7d74b06f | 914 | |
9f3a0d09 | 915 | if (prev && !reclaim) |
5ac8fb31 | 916 | pos = prev; |
14067bb3 | 917 | |
9f3a0d09 JW |
918 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
919 | if (prev) | |
5ac8fb31 | 920 | goto out; |
694fbc0f | 921 | return root; |
9f3a0d09 | 922 | } |
14067bb3 | 923 | |
542f85f9 | 924 | rcu_read_lock(); |
5f578161 | 925 | |
5ac8fb31 | 926 | if (reclaim) { |
ef8f2327 | 927 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 928 | |
ef8f2327 | 929 | mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id); |
5ac8fb31 JW |
930 | iter = &mz->iter[reclaim->priority]; |
931 | ||
932 | if (prev && reclaim->generation != iter->generation) | |
933 | goto out_unlock; | |
934 | ||
6df38689 | 935 | while (1) { |
4db0c3c2 | 936 | pos = READ_ONCE(iter->position); |
6df38689 VD |
937 | if (!pos || css_tryget(&pos->css)) |
938 | break; | |
5ac8fb31 | 939 | /* |
6df38689 VD |
940 | * css reference reached zero, so iter->position will |
941 | * be cleared by ->css_released. However, we should not | |
942 | * rely on this happening soon, because ->css_released | |
943 | * is called from a work queue, and by busy-waiting we | |
944 | * might block it. So we clear iter->position right | |
945 | * away. | |
5ac8fb31 | 946 | */ |
6df38689 VD |
947 | (void)cmpxchg(&iter->position, pos, NULL); |
948 | } | |
5ac8fb31 JW |
949 | } |
950 | ||
951 | if (pos) | |
952 | css = &pos->css; | |
953 | ||
954 | for (;;) { | |
955 | css = css_next_descendant_pre(css, &root->css); | |
956 | if (!css) { | |
957 | /* | |
958 | * Reclaimers share the hierarchy walk, and a | |
959 | * new one might jump in right at the end of | |
960 | * the hierarchy - make sure they see at least | |
961 | * one group and restart from the beginning. | |
962 | */ | |
963 | if (!prev) | |
964 | continue; | |
965 | break; | |
527a5ec9 | 966 | } |
7d74b06f | 967 | |
5ac8fb31 JW |
968 | /* |
969 | * Verify the css and acquire a reference. The root | |
970 | * is provided by the caller, so we know it's alive | |
971 | * and kicking, and don't take an extra reference. | |
972 | */ | |
973 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 974 | |
5ac8fb31 JW |
975 | if (css == &root->css) |
976 | break; | |
14067bb3 | 977 | |
0b8f73e1 JW |
978 | if (css_tryget(css)) |
979 | break; | |
9f3a0d09 | 980 | |
5ac8fb31 | 981 | memcg = NULL; |
9f3a0d09 | 982 | } |
5ac8fb31 JW |
983 | |
984 | if (reclaim) { | |
5ac8fb31 | 985 | /* |
6df38689 VD |
986 | * The position could have already been updated by a competing |
987 | * thread, so check that the value hasn't changed since we read | |
988 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 989 | */ |
6df38689 VD |
990 | (void)cmpxchg(&iter->position, pos, memcg); |
991 | ||
5ac8fb31 JW |
992 | if (pos) |
993 | css_put(&pos->css); | |
994 | ||
995 | if (!memcg) | |
996 | iter->generation++; | |
997 | else if (!prev) | |
998 | reclaim->generation = iter->generation; | |
9f3a0d09 | 999 | } |
5ac8fb31 | 1000 | |
542f85f9 MH |
1001 | out_unlock: |
1002 | rcu_read_unlock(); | |
5ac8fb31 | 1003 | out: |
c40046f3 MH |
1004 | if (prev && prev != root) |
1005 | css_put(&prev->css); | |
1006 | ||
9f3a0d09 | 1007 | return memcg; |
14067bb3 | 1008 | } |
7d74b06f | 1009 | |
5660048c JW |
1010 | /** |
1011 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1012 | * @root: hierarchy root | |
1013 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1014 | */ | |
1015 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1016 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1017 | { |
1018 | if (!root) | |
1019 | root = root_mem_cgroup; | |
1020 | if (prev && prev != root) | |
1021 | css_put(&prev->css); | |
1022 | } | |
7d74b06f | 1023 | |
6df38689 VD |
1024 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1025 | { | |
1026 | struct mem_cgroup *memcg = dead_memcg; | |
1027 | struct mem_cgroup_reclaim_iter *iter; | |
ef8f2327 MG |
1028 | struct mem_cgroup_per_node *mz; |
1029 | int nid; | |
6df38689 VD |
1030 | int i; |
1031 | ||
9f15bde6 | 1032 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
6df38689 | 1033 | for_each_node(nid) { |
ef8f2327 MG |
1034 | mz = mem_cgroup_nodeinfo(memcg, nid); |
1035 | for (i = 0; i <= DEF_PRIORITY; i++) { | |
1036 | iter = &mz->iter[i]; | |
1037 | cmpxchg(&iter->position, | |
1038 | dead_memcg, NULL); | |
6df38689 VD |
1039 | } |
1040 | } | |
1041 | } | |
1042 | } | |
1043 | ||
7c5f64f8 VD |
1044 | /** |
1045 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1046 | * @memcg: hierarchy root | |
1047 | * @fn: function to call for each task | |
1048 | * @arg: argument passed to @fn | |
1049 | * | |
1050 | * This function iterates over tasks attached to @memcg or to any of its | |
1051 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
1052 | * value, the function breaks the iteration loop and returns the value. | |
1053 | * Otherwise, it will iterate over all tasks and return 0. | |
1054 | * | |
1055 | * This function must not be called for the root memory cgroup. | |
1056 | */ | |
1057 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
1058 | int (*fn)(struct task_struct *, void *), void *arg) | |
1059 | { | |
1060 | struct mem_cgroup *iter; | |
1061 | int ret = 0; | |
1062 | ||
1063 | BUG_ON(memcg == root_mem_cgroup); | |
1064 | ||
1065 | for_each_mem_cgroup_tree(iter, memcg) { | |
1066 | struct css_task_iter it; | |
1067 | struct task_struct *task; | |
1068 | ||
bc2fb7ed | 1069 | css_task_iter_start(&iter->css, 0, &it); |
7c5f64f8 VD |
1070 | while (!ret && (task = css_task_iter_next(&it))) |
1071 | ret = fn(task, arg); | |
1072 | css_task_iter_end(&it); | |
1073 | if (ret) { | |
1074 | mem_cgroup_iter_break(memcg, iter); | |
1075 | break; | |
1076 | } | |
1077 | } | |
1078 | return ret; | |
1079 | } | |
1080 | ||
925b7673 | 1081 | /** |
dfe0e773 | 1082 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1083 | * @page: the page |
f144c390 | 1084 | * @pgdat: pgdat of the page |
dfe0e773 JW |
1085 | * |
1086 | * This function is only safe when following the LRU page isolation | |
1087 | * and putback protocol: the LRU lock must be held, and the page must | |
1088 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 1089 | */ |
599d0c95 | 1090 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat) |
08e552c6 | 1091 | { |
ef8f2327 | 1092 | struct mem_cgroup_per_node *mz; |
925b7673 | 1093 | struct mem_cgroup *memcg; |
bea8c150 | 1094 | struct lruvec *lruvec; |
6d12e2d8 | 1095 | |
bea8c150 | 1096 | if (mem_cgroup_disabled()) { |
599d0c95 | 1097 | lruvec = &pgdat->lruvec; |
bea8c150 HD |
1098 | goto out; |
1099 | } | |
925b7673 | 1100 | |
1306a85a | 1101 | memcg = page->mem_cgroup; |
7512102c | 1102 | /* |
dfe0e773 | 1103 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1104 | * possibly migrated - before they are charged. |
7512102c | 1105 | */ |
29833315 JW |
1106 | if (!memcg) |
1107 | memcg = root_mem_cgroup; | |
7512102c | 1108 | |
ef8f2327 | 1109 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
bea8c150 HD |
1110 | lruvec = &mz->lruvec; |
1111 | out: | |
1112 | /* | |
1113 | * Since a node can be onlined after the mem_cgroup was created, | |
1114 | * we have to be prepared to initialize lruvec->zone here; | |
1115 | * and if offlined then reonlined, we need to reinitialize it. | |
1116 | */ | |
599d0c95 MG |
1117 | if (unlikely(lruvec->pgdat != pgdat)) |
1118 | lruvec->pgdat = pgdat; | |
bea8c150 | 1119 | return lruvec; |
08e552c6 | 1120 | } |
b69408e8 | 1121 | |
925b7673 | 1122 | /** |
fa9add64 HD |
1123 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1124 | * @lruvec: mem_cgroup per zone lru vector | |
1125 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1126 | * @zid: zone id of the accounted pages |
fa9add64 | 1127 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1128 | * |
ca707239 HD |
1129 | * This function must be called under lru_lock, just before a page is added |
1130 | * to or just after a page is removed from an lru list (that ordering being | |
1131 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 1132 | */ |
fa9add64 | 1133 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1134 | int zid, int nr_pages) |
3f58a829 | 1135 | { |
ef8f2327 | 1136 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1137 | unsigned long *lru_size; |
ca707239 | 1138 | long size; |
3f58a829 MK |
1139 | |
1140 | if (mem_cgroup_disabled()) | |
1141 | return; | |
1142 | ||
ef8f2327 | 1143 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1144 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1145 | |
1146 | if (nr_pages < 0) | |
1147 | *lru_size += nr_pages; | |
1148 | ||
1149 | size = *lru_size; | |
b4536f0c MH |
1150 | if (WARN_ONCE(size < 0, |
1151 | "%s(%p, %d, %d): lru_size %ld\n", | |
1152 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1153 | VM_BUG_ON(1); |
1154 | *lru_size = 0; | |
1155 | } | |
1156 | ||
1157 | if (nr_pages > 0) | |
1158 | *lru_size += nr_pages; | |
08e552c6 | 1159 | } |
544122e5 | 1160 | |
2314b42d | 1161 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
c3ac9a8a | 1162 | { |
2314b42d | 1163 | struct mem_cgroup *task_memcg; |
158e0a2d | 1164 | struct task_struct *p; |
ffbdccf5 | 1165 | bool ret; |
4c4a2214 | 1166 | |
158e0a2d | 1167 | p = find_lock_task_mm(task); |
de077d22 | 1168 | if (p) { |
2314b42d | 1169 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1170 | task_unlock(p); |
1171 | } else { | |
1172 | /* | |
1173 | * All threads may have already detached their mm's, but the oom | |
1174 | * killer still needs to detect if they have already been oom | |
1175 | * killed to prevent needlessly killing additional tasks. | |
1176 | */ | |
ffbdccf5 | 1177 | rcu_read_lock(); |
2314b42d JW |
1178 | task_memcg = mem_cgroup_from_task(task); |
1179 | css_get(&task_memcg->css); | |
ffbdccf5 | 1180 | rcu_read_unlock(); |
de077d22 | 1181 | } |
2314b42d JW |
1182 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1183 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1184 | return ret; |
1185 | } | |
1186 | ||
19942822 | 1187 | /** |
9d11ea9f | 1188 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1189 | * @memcg: the memory cgroup |
19942822 | 1190 | * |
9d11ea9f | 1191 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1192 | * pages. |
19942822 | 1193 | */ |
c0ff4b85 | 1194 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1195 | { |
3e32cb2e JW |
1196 | unsigned long margin = 0; |
1197 | unsigned long count; | |
1198 | unsigned long limit; | |
9d11ea9f | 1199 | |
3e32cb2e | 1200 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1201 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1202 | if (count < limit) |
1203 | margin = limit - count; | |
1204 | ||
7941d214 | 1205 | if (do_memsw_account()) { |
3e32cb2e | 1206 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1207 | limit = READ_ONCE(memcg->memsw.max); |
3e32cb2e JW |
1208 | if (count <= limit) |
1209 | margin = min(margin, limit - count); | |
cbedbac3 LR |
1210 | else |
1211 | margin = 0; | |
3e32cb2e JW |
1212 | } |
1213 | ||
1214 | return margin; | |
19942822 JW |
1215 | } |
1216 | ||
32047e2a | 1217 | /* |
bdcbb659 | 1218 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1219 | * |
bdcbb659 QH |
1220 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1221 | * moving cgroups. This is for waiting at high-memory pressure | |
1222 | * caused by "move". | |
32047e2a | 1223 | */ |
c0ff4b85 | 1224 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1225 | { |
2bd9bb20 KH |
1226 | struct mem_cgroup *from; |
1227 | struct mem_cgroup *to; | |
4b534334 | 1228 | bool ret = false; |
2bd9bb20 KH |
1229 | /* |
1230 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1231 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1232 | */ | |
1233 | spin_lock(&mc.lock); | |
1234 | from = mc.from; | |
1235 | to = mc.to; | |
1236 | if (!from) | |
1237 | goto unlock; | |
3e92041d | 1238 | |
2314b42d JW |
1239 | ret = mem_cgroup_is_descendant(from, memcg) || |
1240 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1241 | unlock: |
1242 | spin_unlock(&mc.lock); | |
4b534334 KH |
1243 | return ret; |
1244 | } | |
1245 | ||
c0ff4b85 | 1246 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1247 | { |
1248 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1249 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1250 | DEFINE_WAIT(wait); |
1251 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1252 | /* moving charge context might have finished. */ | |
1253 | if (mc.moving_task) | |
1254 | schedule(); | |
1255 | finish_wait(&mc.waitq, &wait); | |
1256 | return true; | |
1257 | } | |
1258 | } | |
1259 | return false; | |
1260 | } | |
1261 | ||
8ad6e404 | 1262 | static const unsigned int memcg1_stats[] = { |
71cd3113 JW |
1263 | MEMCG_CACHE, |
1264 | MEMCG_RSS, | |
1265 | MEMCG_RSS_HUGE, | |
1266 | NR_SHMEM, | |
1267 | NR_FILE_MAPPED, | |
1268 | NR_FILE_DIRTY, | |
1269 | NR_WRITEBACK, | |
1270 | MEMCG_SWAP, | |
1271 | }; | |
1272 | ||
1273 | static const char *const memcg1_stat_names[] = { | |
1274 | "cache", | |
1275 | "rss", | |
1276 | "rss_huge", | |
1277 | "shmem", | |
1278 | "mapped_file", | |
1279 | "dirty", | |
1280 | "writeback", | |
1281 | "swap", | |
1282 | }; | |
1283 | ||
58cf188e | 1284 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1285 | /** |
58cf188e | 1286 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1287 | * @memcg: The memory cgroup that went over limit |
1288 | * @p: Task that is going to be killed | |
1289 | * | |
1290 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1291 | * enabled | |
1292 | */ | |
1293 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1294 | { | |
58cf188e SZ |
1295 | struct mem_cgroup *iter; |
1296 | unsigned int i; | |
e222432b | 1297 | |
e222432b BS |
1298 | rcu_read_lock(); |
1299 | ||
2415b9f5 BV |
1300 | if (p) { |
1301 | pr_info("Task in "); | |
1302 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1303 | pr_cont(" killed as a result of limit of "); | |
1304 | } else { | |
1305 | pr_info("Memory limit reached of cgroup "); | |
1306 | } | |
1307 | ||
e61734c5 | 1308 | pr_cont_cgroup_path(memcg->css.cgroup); |
0346dadb | 1309 | pr_cont("\n"); |
e222432b | 1310 | |
e222432b BS |
1311 | rcu_read_unlock(); |
1312 | ||
3e32cb2e JW |
1313 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1314 | K((u64)page_counter_read(&memcg->memory)), | |
bbec2e15 | 1315 | K((u64)memcg->memory.max), memcg->memory.failcnt); |
3e32cb2e JW |
1316 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", |
1317 | K((u64)page_counter_read(&memcg->memsw)), | |
bbec2e15 | 1318 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); |
3e32cb2e JW |
1319 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", |
1320 | K((u64)page_counter_read(&memcg->kmem)), | |
bbec2e15 | 1321 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); |
58cf188e SZ |
1322 | |
1323 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1324 | pr_info("Memory cgroup stats for "); |
1325 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1326 | pr_cont(":"); |
1327 | ||
71cd3113 JW |
1328 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
1329 | if (memcg1_stats[i] == MEMCG_SWAP && !do_swap_account) | |
58cf188e | 1330 | continue; |
71cd3113 | 1331 | pr_cont(" %s:%luKB", memcg1_stat_names[i], |
ccda7f43 | 1332 | K(memcg_page_state(iter, memcg1_stats[i]))); |
58cf188e SZ |
1333 | } |
1334 | ||
1335 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1336 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1337 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1338 | ||
1339 | pr_cont("\n"); | |
1340 | } | |
e222432b BS |
1341 | } |
1342 | ||
a63d83f4 DR |
1343 | /* |
1344 | * Return the memory (and swap, if configured) limit for a memcg. | |
1345 | */ | |
bbec2e15 | 1346 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1347 | { |
bbec2e15 | 1348 | unsigned long max; |
f3e8eb70 | 1349 | |
bbec2e15 | 1350 | max = memcg->memory.max; |
9a5a8f19 | 1351 | if (mem_cgroup_swappiness(memcg)) { |
bbec2e15 RG |
1352 | unsigned long memsw_max; |
1353 | unsigned long swap_max; | |
9a5a8f19 | 1354 | |
bbec2e15 RG |
1355 | memsw_max = memcg->memsw.max; |
1356 | swap_max = memcg->swap.max; | |
1357 | swap_max = min(swap_max, (unsigned long)total_swap_pages); | |
1358 | max = min(max + swap_max, memsw_max); | |
9a5a8f19 | 1359 | } |
bbec2e15 | 1360 | return max; |
a63d83f4 DR |
1361 | } |
1362 | ||
b6e6edcf | 1363 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1364 | int order) |
9cbb78bb | 1365 | { |
6e0fc46d DR |
1366 | struct oom_control oc = { |
1367 | .zonelist = NULL, | |
1368 | .nodemask = NULL, | |
2a966b77 | 1369 | .memcg = memcg, |
6e0fc46d DR |
1370 | .gfp_mask = gfp_mask, |
1371 | .order = order, | |
6e0fc46d | 1372 | }; |
7c5f64f8 | 1373 | bool ret; |
9cbb78bb | 1374 | |
dc56401f | 1375 | mutex_lock(&oom_lock); |
7c5f64f8 | 1376 | ret = out_of_memory(&oc); |
dc56401f | 1377 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1378 | return ret; |
9cbb78bb DR |
1379 | } |
1380 | ||
ae6e71d3 MC |
1381 | #if MAX_NUMNODES > 1 |
1382 | ||
4d0c066d KH |
1383 | /** |
1384 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1385 | * @memcg: the target memcg |
4d0c066d KH |
1386 | * @nid: the node ID to be checked. |
1387 | * @noswap : specify true here if the user wants flle only information. | |
1388 | * | |
1389 | * This function returns whether the specified memcg contains any | |
1390 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1391 | * pages in the node. | |
1392 | */ | |
c0ff4b85 | 1393 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1394 | int nid, bool noswap) |
1395 | { | |
c0ff4b85 | 1396 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1397 | return true; |
1398 | if (noswap || !total_swap_pages) | |
1399 | return false; | |
c0ff4b85 | 1400 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1401 | return true; |
1402 | return false; | |
1403 | ||
1404 | } | |
889976db YH |
1405 | |
1406 | /* | |
1407 | * Always updating the nodemask is not very good - even if we have an empty | |
1408 | * list or the wrong list here, we can start from some node and traverse all | |
1409 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1410 | * | |
1411 | */ | |
c0ff4b85 | 1412 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1413 | { |
1414 | int nid; | |
453a9bf3 KH |
1415 | /* |
1416 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1417 | * pagein/pageout changes since the last update. | |
1418 | */ | |
c0ff4b85 | 1419 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1420 | return; |
c0ff4b85 | 1421 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1422 | return; |
1423 | ||
889976db | 1424 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1425 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1426 | |
31aaea4a | 1427 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1428 | |
c0ff4b85 R |
1429 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1430 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1431 | } |
453a9bf3 | 1432 | |
c0ff4b85 R |
1433 | atomic_set(&memcg->numainfo_events, 0); |
1434 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1435 | } |
1436 | ||
1437 | /* | |
1438 | * Selecting a node where we start reclaim from. Because what we need is just | |
1439 | * reducing usage counter, start from anywhere is O,K. Considering | |
1440 | * memory reclaim from current node, there are pros. and cons. | |
1441 | * | |
1442 | * Freeing memory from current node means freeing memory from a node which | |
1443 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1444 | * hit limits, it will see a contention on a node. But freeing from remote | |
1445 | * node means more costs for memory reclaim because of memory latency. | |
1446 | * | |
1447 | * Now, we use round-robin. Better algorithm is welcomed. | |
1448 | */ | |
c0ff4b85 | 1449 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1450 | { |
1451 | int node; | |
1452 | ||
c0ff4b85 R |
1453 | mem_cgroup_may_update_nodemask(memcg); |
1454 | node = memcg->last_scanned_node; | |
889976db | 1455 | |
0edaf86c | 1456 | node = next_node_in(node, memcg->scan_nodes); |
889976db | 1457 | /* |
fda3d69b MH |
1458 | * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages |
1459 | * last time it really checked all the LRUs due to rate limiting. | |
1460 | * Fallback to the current node in that case for simplicity. | |
889976db YH |
1461 | */ |
1462 | if (unlikely(node == MAX_NUMNODES)) | |
1463 | node = numa_node_id(); | |
1464 | ||
c0ff4b85 | 1465 | memcg->last_scanned_node = node; |
889976db YH |
1466 | return node; |
1467 | } | |
889976db | 1468 | #else |
c0ff4b85 | 1469 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1470 | { |
1471 | return 0; | |
1472 | } | |
1473 | #endif | |
1474 | ||
0608f43d | 1475 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1476 | pg_data_t *pgdat, |
0608f43d AM |
1477 | gfp_t gfp_mask, |
1478 | unsigned long *total_scanned) | |
1479 | { | |
1480 | struct mem_cgroup *victim = NULL; | |
1481 | int total = 0; | |
1482 | int loop = 0; | |
1483 | unsigned long excess; | |
1484 | unsigned long nr_scanned; | |
1485 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1486 | .pgdat = pgdat, |
0608f43d AM |
1487 | .priority = 0, |
1488 | }; | |
1489 | ||
3e32cb2e | 1490 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1491 | |
1492 | while (1) { | |
1493 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1494 | if (!victim) { | |
1495 | loop++; | |
1496 | if (loop >= 2) { | |
1497 | /* | |
1498 | * If we have not been able to reclaim | |
1499 | * anything, it might because there are | |
1500 | * no reclaimable pages under this hierarchy | |
1501 | */ | |
1502 | if (!total) | |
1503 | break; | |
1504 | /* | |
1505 | * We want to do more targeted reclaim. | |
1506 | * excess >> 2 is not to excessive so as to | |
1507 | * reclaim too much, nor too less that we keep | |
1508 | * coming back to reclaim from this cgroup | |
1509 | */ | |
1510 | if (total >= (excess >> 2) || | |
1511 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1512 | break; | |
1513 | } | |
1514 | continue; | |
1515 | } | |
a9dd0a83 | 1516 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1517 | pgdat, &nr_scanned); |
0608f43d | 1518 | *total_scanned += nr_scanned; |
3e32cb2e | 1519 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1520 | break; |
6d61ef40 | 1521 | } |
0608f43d AM |
1522 | mem_cgroup_iter_break(root_memcg, victim); |
1523 | return total; | |
6d61ef40 BS |
1524 | } |
1525 | ||
0056f4e6 JW |
1526 | #ifdef CONFIG_LOCKDEP |
1527 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1528 | .name = "memcg_oom_lock", | |
1529 | }; | |
1530 | #endif | |
1531 | ||
fb2a6fc5 JW |
1532 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1533 | ||
867578cb KH |
1534 | /* |
1535 | * Check OOM-Killer is already running under our hierarchy. | |
1536 | * If someone is running, return false. | |
1537 | */ | |
fb2a6fc5 | 1538 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1539 | { |
79dfdacc | 1540 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1541 | |
fb2a6fc5 JW |
1542 | spin_lock(&memcg_oom_lock); |
1543 | ||
9f3a0d09 | 1544 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1545 | if (iter->oom_lock) { |
79dfdacc MH |
1546 | /* |
1547 | * this subtree of our hierarchy is already locked | |
1548 | * so we cannot give a lock. | |
1549 | */ | |
79dfdacc | 1550 | failed = iter; |
9f3a0d09 JW |
1551 | mem_cgroup_iter_break(memcg, iter); |
1552 | break; | |
23751be0 JW |
1553 | } else |
1554 | iter->oom_lock = true; | |
7d74b06f | 1555 | } |
867578cb | 1556 | |
fb2a6fc5 JW |
1557 | if (failed) { |
1558 | /* | |
1559 | * OK, we failed to lock the whole subtree so we have | |
1560 | * to clean up what we set up to the failing subtree | |
1561 | */ | |
1562 | for_each_mem_cgroup_tree(iter, memcg) { | |
1563 | if (iter == failed) { | |
1564 | mem_cgroup_iter_break(memcg, iter); | |
1565 | break; | |
1566 | } | |
1567 | iter->oom_lock = false; | |
79dfdacc | 1568 | } |
0056f4e6 JW |
1569 | } else |
1570 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1571 | |
1572 | spin_unlock(&memcg_oom_lock); | |
1573 | ||
1574 | return !failed; | |
a636b327 | 1575 | } |
0b7f569e | 1576 | |
fb2a6fc5 | 1577 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1578 | { |
7d74b06f KH |
1579 | struct mem_cgroup *iter; |
1580 | ||
fb2a6fc5 | 1581 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1582 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1583 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1584 | iter->oom_lock = false; |
fb2a6fc5 | 1585 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1586 | } |
1587 | ||
c0ff4b85 | 1588 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1589 | { |
1590 | struct mem_cgroup *iter; | |
1591 | ||
c2b42d3c | 1592 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1593 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1594 | iter->under_oom++; |
1595 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1596 | } |
1597 | ||
c0ff4b85 | 1598 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1599 | { |
1600 | struct mem_cgroup *iter; | |
1601 | ||
867578cb KH |
1602 | /* |
1603 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1604 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1605 | */ |
c2b42d3c | 1606 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1607 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1608 | if (iter->under_oom > 0) |
1609 | iter->under_oom--; | |
1610 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1611 | } |
1612 | ||
867578cb KH |
1613 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1614 | ||
dc98df5a | 1615 | struct oom_wait_info { |
d79154bb | 1616 | struct mem_cgroup *memcg; |
ac6424b9 | 1617 | wait_queue_entry_t wait; |
dc98df5a KH |
1618 | }; |
1619 | ||
ac6424b9 | 1620 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1621 | unsigned mode, int sync, void *arg) |
1622 | { | |
d79154bb HD |
1623 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1624 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1625 | struct oom_wait_info *oom_wait_info; |
1626 | ||
1627 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1628 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1629 | |
2314b42d JW |
1630 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1631 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1632 | return 0; |
dc98df5a KH |
1633 | return autoremove_wake_function(wait, mode, sync, arg); |
1634 | } | |
1635 | ||
c0ff4b85 | 1636 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1637 | { |
c2b42d3c TH |
1638 | /* |
1639 | * For the following lockless ->under_oom test, the only required | |
1640 | * guarantee is that it must see the state asserted by an OOM when | |
1641 | * this function is called as a result of userland actions | |
1642 | * triggered by the notification of the OOM. This is trivially | |
1643 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1644 | * triggering notification. | |
1645 | */ | |
1646 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1647 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1648 | } |
1649 | ||
29ef680a MH |
1650 | enum oom_status { |
1651 | OOM_SUCCESS, | |
1652 | OOM_FAILED, | |
1653 | OOM_ASYNC, | |
1654 | OOM_SKIPPED | |
1655 | }; | |
1656 | ||
1657 | static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 1658 | { |
29ef680a MH |
1659 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
1660 | return OOM_SKIPPED; | |
1661 | ||
867578cb | 1662 | /* |
49426420 JW |
1663 | * We are in the middle of the charge context here, so we |
1664 | * don't want to block when potentially sitting on a callstack | |
1665 | * that holds all kinds of filesystem and mm locks. | |
1666 | * | |
29ef680a MH |
1667 | * cgroup1 allows disabling the OOM killer and waiting for outside |
1668 | * handling until the charge can succeed; remember the context and put | |
1669 | * the task to sleep at the end of the page fault when all locks are | |
1670 | * released. | |
49426420 | 1671 | * |
29ef680a MH |
1672 | * On the other hand, in-kernel OOM killer allows for an async victim |
1673 | * memory reclaim (oom_reaper) and that means that we are not solely | |
1674 | * relying on the oom victim to make a forward progress and we can | |
1675 | * invoke the oom killer here. | |
1676 | * | |
1677 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
1678 | * victim and then we have to bail out from the charge path. | |
867578cb | 1679 | */ |
29ef680a MH |
1680 | if (memcg->oom_kill_disable) { |
1681 | if (!current->in_user_fault) | |
1682 | return OOM_SKIPPED; | |
1683 | css_get(&memcg->css); | |
1684 | current->memcg_in_oom = memcg; | |
1685 | current->memcg_oom_gfp_mask = mask; | |
1686 | current->memcg_oom_order = order; | |
1687 | ||
1688 | return OOM_ASYNC; | |
1689 | } | |
1690 | ||
1691 | if (mem_cgroup_out_of_memory(memcg, mask, order)) | |
1692 | return OOM_SUCCESS; | |
1693 | ||
1694 | WARN(1,"Memory cgroup charge failed because of no reclaimable memory! " | |
1695 | "This looks like a misconfiguration or a kernel bug."); | |
1696 | return OOM_FAILED; | |
3812c8c8 JW |
1697 | } |
1698 | ||
1699 | /** | |
1700 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1701 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1702 | * |
49426420 JW |
1703 | * This has to be called at the end of a page fault if the memcg OOM |
1704 | * handler was enabled. | |
3812c8c8 | 1705 | * |
49426420 | 1706 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1707 | * sleep on a waitqueue until the userspace task resolves the |
1708 | * situation. Sleeping directly in the charge context with all kinds | |
1709 | * of locks held is not a good idea, instead we remember an OOM state | |
1710 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1711 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1712 | * |
1713 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1714 | * completed, %false otherwise. |
3812c8c8 | 1715 | */ |
49426420 | 1716 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1717 | { |
626ebc41 | 1718 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1719 | struct oom_wait_info owait; |
49426420 | 1720 | bool locked; |
3812c8c8 JW |
1721 | |
1722 | /* OOM is global, do not handle */ | |
3812c8c8 | 1723 | if (!memcg) |
49426420 | 1724 | return false; |
3812c8c8 | 1725 | |
7c5f64f8 | 1726 | if (!handle) |
49426420 | 1727 | goto cleanup; |
3812c8c8 JW |
1728 | |
1729 | owait.memcg = memcg; | |
1730 | owait.wait.flags = 0; | |
1731 | owait.wait.func = memcg_oom_wake_function; | |
1732 | owait.wait.private = current; | |
2055da97 | 1733 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 1734 | |
3812c8c8 | 1735 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1736 | mem_cgroup_mark_under_oom(memcg); |
1737 | ||
1738 | locked = mem_cgroup_oom_trylock(memcg); | |
1739 | ||
1740 | if (locked) | |
1741 | mem_cgroup_oom_notify(memcg); | |
1742 | ||
1743 | if (locked && !memcg->oom_kill_disable) { | |
1744 | mem_cgroup_unmark_under_oom(memcg); | |
1745 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1746 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1747 | current->memcg_oom_order); | |
49426420 | 1748 | } else { |
3812c8c8 | 1749 | schedule(); |
49426420 JW |
1750 | mem_cgroup_unmark_under_oom(memcg); |
1751 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1752 | } | |
1753 | ||
1754 | if (locked) { | |
fb2a6fc5 JW |
1755 | mem_cgroup_oom_unlock(memcg); |
1756 | /* | |
1757 | * There is no guarantee that an OOM-lock contender | |
1758 | * sees the wakeups triggered by the OOM kill | |
1759 | * uncharges. Wake any sleepers explicitely. | |
1760 | */ | |
1761 | memcg_oom_recover(memcg); | |
1762 | } | |
49426420 | 1763 | cleanup: |
626ebc41 | 1764 | current->memcg_in_oom = NULL; |
3812c8c8 | 1765 | css_put(&memcg->css); |
867578cb | 1766 | return true; |
0b7f569e KH |
1767 | } |
1768 | ||
d7365e78 | 1769 | /** |
81f8c3a4 JW |
1770 | * lock_page_memcg - lock a page->mem_cgroup binding |
1771 | * @page: the page | |
32047e2a | 1772 | * |
81f8c3a4 | 1773 | * This function protects unlocked LRU pages from being moved to |
739f79fc JW |
1774 | * another cgroup. |
1775 | * | |
1776 | * It ensures lifetime of the returned memcg. Caller is responsible | |
1777 | * for the lifetime of the page; __unlock_page_memcg() is available | |
1778 | * when @page might get freed inside the locked section. | |
d69b042f | 1779 | */ |
739f79fc | 1780 | struct mem_cgroup *lock_page_memcg(struct page *page) |
89c06bd5 KH |
1781 | { |
1782 | struct mem_cgroup *memcg; | |
6de22619 | 1783 | unsigned long flags; |
89c06bd5 | 1784 | |
6de22619 JW |
1785 | /* |
1786 | * The RCU lock is held throughout the transaction. The fast | |
1787 | * path can get away without acquiring the memcg->move_lock | |
1788 | * because page moving starts with an RCU grace period. | |
739f79fc JW |
1789 | * |
1790 | * The RCU lock also protects the memcg from being freed when | |
1791 | * the page state that is going to change is the only thing | |
1792 | * preventing the page itself from being freed. E.g. writeback | |
1793 | * doesn't hold a page reference and relies on PG_writeback to | |
1794 | * keep off truncation, migration and so forth. | |
1795 | */ | |
d7365e78 JW |
1796 | rcu_read_lock(); |
1797 | ||
1798 | if (mem_cgroup_disabled()) | |
739f79fc | 1799 | return NULL; |
89c06bd5 | 1800 | again: |
1306a85a | 1801 | memcg = page->mem_cgroup; |
29833315 | 1802 | if (unlikely(!memcg)) |
739f79fc | 1803 | return NULL; |
d7365e78 | 1804 | |
bdcbb659 | 1805 | if (atomic_read(&memcg->moving_account) <= 0) |
739f79fc | 1806 | return memcg; |
89c06bd5 | 1807 | |
6de22619 | 1808 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 1809 | if (memcg != page->mem_cgroup) { |
6de22619 | 1810 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
1811 | goto again; |
1812 | } | |
6de22619 JW |
1813 | |
1814 | /* | |
1815 | * When charge migration first begins, we can have locked and | |
1816 | * unlocked page stat updates happening concurrently. Track | |
81f8c3a4 | 1817 | * the task who has the lock for unlock_page_memcg(). |
6de22619 JW |
1818 | */ |
1819 | memcg->move_lock_task = current; | |
1820 | memcg->move_lock_flags = flags; | |
d7365e78 | 1821 | |
739f79fc | 1822 | return memcg; |
89c06bd5 | 1823 | } |
81f8c3a4 | 1824 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 1825 | |
d7365e78 | 1826 | /** |
739f79fc JW |
1827 | * __unlock_page_memcg - unlock and unpin a memcg |
1828 | * @memcg: the memcg | |
1829 | * | |
1830 | * Unlock and unpin a memcg returned by lock_page_memcg(). | |
d7365e78 | 1831 | */ |
739f79fc | 1832 | void __unlock_page_memcg(struct mem_cgroup *memcg) |
89c06bd5 | 1833 | { |
6de22619 JW |
1834 | if (memcg && memcg->move_lock_task == current) { |
1835 | unsigned long flags = memcg->move_lock_flags; | |
1836 | ||
1837 | memcg->move_lock_task = NULL; | |
1838 | memcg->move_lock_flags = 0; | |
1839 | ||
1840 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
1841 | } | |
89c06bd5 | 1842 | |
d7365e78 | 1843 | rcu_read_unlock(); |
89c06bd5 | 1844 | } |
739f79fc JW |
1845 | |
1846 | /** | |
1847 | * unlock_page_memcg - unlock a page->mem_cgroup binding | |
1848 | * @page: the page | |
1849 | */ | |
1850 | void unlock_page_memcg(struct page *page) | |
1851 | { | |
1852 | __unlock_page_memcg(page->mem_cgroup); | |
1853 | } | |
81f8c3a4 | 1854 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 1855 | |
cdec2e42 KH |
1856 | struct memcg_stock_pcp { |
1857 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 1858 | unsigned int nr_pages; |
cdec2e42 | 1859 | struct work_struct work; |
26fe6168 | 1860 | unsigned long flags; |
a0db00fc | 1861 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
1862 | }; |
1863 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 1864 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 1865 | |
a0956d54 SS |
1866 | /** |
1867 | * consume_stock: Try to consume stocked charge on this cpu. | |
1868 | * @memcg: memcg to consume from. | |
1869 | * @nr_pages: how many pages to charge. | |
1870 | * | |
1871 | * The charges will only happen if @memcg matches the current cpu's memcg | |
1872 | * stock, and at least @nr_pages are available in that stock. Failure to | |
1873 | * service an allocation will refill the stock. | |
1874 | * | |
1875 | * returns true if successful, false otherwise. | |
cdec2e42 | 1876 | */ |
a0956d54 | 1877 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1878 | { |
1879 | struct memcg_stock_pcp *stock; | |
db2ba40c | 1880 | unsigned long flags; |
3e32cb2e | 1881 | bool ret = false; |
cdec2e42 | 1882 | |
a983b5eb | 1883 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 1884 | return ret; |
a0956d54 | 1885 | |
db2ba40c JW |
1886 | local_irq_save(flags); |
1887 | ||
1888 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 1889 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 1890 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
1891 | ret = true; |
1892 | } | |
db2ba40c JW |
1893 | |
1894 | local_irq_restore(flags); | |
1895 | ||
cdec2e42 KH |
1896 | return ret; |
1897 | } | |
1898 | ||
1899 | /* | |
3e32cb2e | 1900 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
1901 | */ |
1902 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1903 | { | |
1904 | struct mem_cgroup *old = stock->cached; | |
1905 | ||
11c9ea4e | 1906 | if (stock->nr_pages) { |
3e32cb2e | 1907 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 1908 | if (do_memsw_account()) |
3e32cb2e | 1909 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 1910 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 1911 | stock->nr_pages = 0; |
cdec2e42 KH |
1912 | } |
1913 | stock->cached = NULL; | |
cdec2e42 KH |
1914 | } |
1915 | ||
cdec2e42 KH |
1916 | static void drain_local_stock(struct work_struct *dummy) |
1917 | { | |
db2ba40c JW |
1918 | struct memcg_stock_pcp *stock; |
1919 | unsigned long flags; | |
1920 | ||
72f0184c MH |
1921 | /* |
1922 | * The only protection from memory hotplug vs. drain_stock races is | |
1923 | * that we always operate on local CPU stock here with IRQ disabled | |
1924 | */ | |
db2ba40c JW |
1925 | local_irq_save(flags); |
1926 | ||
1927 | stock = this_cpu_ptr(&memcg_stock); | |
cdec2e42 | 1928 | drain_stock(stock); |
26fe6168 | 1929 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c JW |
1930 | |
1931 | local_irq_restore(flags); | |
cdec2e42 KH |
1932 | } |
1933 | ||
1934 | /* | |
3e32cb2e | 1935 | * Cache charges(val) to local per_cpu area. |
320cc51d | 1936 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 1937 | */ |
c0ff4b85 | 1938 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 1939 | { |
db2ba40c JW |
1940 | struct memcg_stock_pcp *stock; |
1941 | unsigned long flags; | |
1942 | ||
1943 | local_irq_save(flags); | |
cdec2e42 | 1944 | |
db2ba40c | 1945 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 1946 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 1947 | drain_stock(stock); |
c0ff4b85 | 1948 | stock->cached = memcg; |
cdec2e42 | 1949 | } |
11c9ea4e | 1950 | stock->nr_pages += nr_pages; |
db2ba40c | 1951 | |
a983b5eb | 1952 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 RG |
1953 | drain_stock(stock); |
1954 | ||
db2ba40c | 1955 | local_irq_restore(flags); |
cdec2e42 KH |
1956 | } |
1957 | ||
1958 | /* | |
c0ff4b85 | 1959 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 1960 | * of the hierarchy under it. |
cdec2e42 | 1961 | */ |
6d3d6aa2 | 1962 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 1963 | { |
26fe6168 | 1964 | int cpu, curcpu; |
d38144b7 | 1965 | |
6d3d6aa2 JW |
1966 | /* If someone's already draining, avoid adding running more workers. */ |
1967 | if (!mutex_trylock(&percpu_charge_mutex)) | |
1968 | return; | |
72f0184c MH |
1969 | /* |
1970 | * Notify other cpus that system-wide "drain" is running | |
1971 | * We do not care about races with the cpu hotplug because cpu down | |
1972 | * as well as workers from this path always operate on the local | |
1973 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
1974 | */ | |
5af12d0e | 1975 | curcpu = get_cpu(); |
cdec2e42 KH |
1976 | for_each_online_cpu(cpu) { |
1977 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 1978 | struct mem_cgroup *memcg; |
26fe6168 | 1979 | |
c0ff4b85 | 1980 | memcg = stock->cached; |
72f0184c | 1981 | if (!memcg || !stock->nr_pages || !css_tryget(&memcg->css)) |
26fe6168 | 1982 | continue; |
72f0184c MH |
1983 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) { |
1984 | css_put(&memcg->css); | |
3e92041d | 1985 | continue; |
72f0184c | 1986 | } |
d1a05b69 MH |
1987 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
1988 | if (cpu == curcpu) | |
1989 | drain_local_stock(&stock->work); | |
1990 | else | |
1991 | schedule_work_on(cpu, &stock->work); | |
1992 | } | |
72f0184c | 1993 | css_put(&memcg->css); |
cdec2e42 | 1994 | } |
5af12d0e | 1995 | put_cpu(); |
9f50fad6 | 1996 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
1997 | } |
1998 | ||
308167fc | 1999 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
cdec2e42 | 2000 | { |
cdec2e42 | 2001 | struct memcg_stock_pcp *stock; |
a983b5eb | 2002 | struct mem_cgroup *memcg; |
cdec2e42 | 2003 | |
cdec2e42 KH |
2004 | stock = &per_cpu(memcg_stock, cpu); |
2005 | drain_stock(stock); | |
a983b5eb JW |
2006 | |
2007 | for_each_mem_cgroup(memcg) { | |
2008 | int i; | |
2009 | ||
2010 | for (i = 0; i < MEMCG_NR_STAT; i++) { | |
2011 | int nid; | |
2012 | long x; | |
2013 | ||
2014 | x = this_cpu_xchg(memcg->stat_cpu->count[i], 0); | |
2015 | if (x) | |
2016 | atomic_long_add(x, &memcg->stat[i]); | |
2017 | ||
2018 | if (i >= NR_VM_NODE_STAT_ITEMS) | |
2019 | continue; | |
2020 | ||
2021 | for_each_node(nid) { | |
2022 | struct mem_cgroup_per_node *pn; | |
2023 | ||
2024 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
2025 | x = this_cpu_xchg(pn->lruvec_stat_cpu->count[i], 0); | |
2026 | if (x) | |
2027 | atomic_long_add(x, &pn->lruvec_stat[i]); | |
2028 | } | |
2029 | } | |
2030 | ||
e27be240 | 2031 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { |
a983b5eb JW |
2032 | long x; |
2033 | ||
2034 | x = this_cpu_xchg(memcg->stat_cpu->events[i], 0); | |
2035 | if (x) | |
2036 | atomic_long_add(x, &memcg->events[i]); | |
2037 | } | |
2038 | } | |
2039 | ||
308167fc | 2040 | return 0; |
cdec2e42 KH |
2041 | } |
2042 | ||
f7e1cb6e JW |
2043 | static void reclaim_high(struct mem_cgroup *memcg, |
2044 | unsigned int nr_pages, | |
2045 | gfp_t gfp_mask) | |
2046 | { | |
2047 | do { | |
2048 | if (page_counter_read(&memcg->memory) <= memcg->high) | |
2049 | continue; | |
e27be240 | 2050 | memcg_memory_event(memcg, MEMCG_HIGH); |
f7e1cb6e JW |
2051 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); |
2052 | } while ((memcg = parent_mem_cgroup(memcg))); | |
2053 | } | |
2054 | ||
2055 | static void high_work_func(struct work_struct *work) | |
2056 | { | |
2057 | struct mem_cgroup *memcg; | |
2058 | ||
2059 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2060 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2061 | } |
2062 | ||
b23afb93 TH |
2063 | /* |
2064 | * Scheduled by try_charge() to be executed from the userland return path | |
2065 | * and reclaims memory over the high limit. | |
2066 | */ | |
2067 | void mem_cgroup_handle_over_high(void) | |
2068 | { | |
2069 | unsigned int nr_pages = current->memcg_nr_pages_over_high; | |
f7e1cb6e | 2070 | struct mem_cgroup *memcg; |
b23afb93 TH |
2071 | |
2072 | if (likely(!nr_pages)) | |
2073 | return; | |
2074 | ||
f7e1cb6e JW |
2075 | memcg = get_mem_cgroup_from_mm(current->mm); |
2076 | reclaim_high(memcg, nr_pages, GFP_KERNEL); | |
b23afb93 TH |
2077 | css_put(&memcg->css); |
2078 | current->memcg_nr_pages_over_high = 0; | |
2079 | } | |
2080 | ||
00501b53 JW |
2081 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2082 | unsigned int nr_pages) | |
8a9f3ccd | 2083 | { |
a983b5eb | 2084 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
9b130619 | 2085 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 2086 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2087 | struct page_counter *counter; |
6539cc05 | 2088 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2089 | bool may_swap = true; |
2090 | bool drained = false; | |
29ef680a MH |
2091 | bool oomed = false; |
2092 | enum oom_status oom_status; | |
a636b327 | 2093 | |
ce00a967 | 2094 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 2095 | return 0; |
6539cc05 | 2096 | retry: |
b6b6cc72 | 2097 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2098 | return 0; |
8a9f3ccd | 2099 | |
7941d214 | 2100 | if (!do_memsw_account() || |
6071ca52 JW |
2101 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2102 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2103 | goto done_restock; |
7941d214 | 2104 | if (do_memsw_account()) |
3e32cb2e JW |
2105 | page_counter_uncharge(&memcg->memsw, batch); |
2106 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2107 | } else { |
3e32cb2e | 2108 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2109 | may_swap = false; |
3fbe7244 | 2110 | } |
7a81b88c | 2111 | |
6539cc05 JW |
2112 | if (batch > nr_pages) { |
2113 | batch = nr_pages; | |
2114 | goto retry; | |
2115 | } | |
6d61ef40 | 2116 | |
06b078fc JW |
2117 | /* |
2118 | * Unlike in global OOM situations, memcg is not in a physical | |
2119 | * memory shortage. Allow dying and OOM-killed tasks to | |
2120 | * bypass the last charges so that they can exit quickly and | |
2121 | * free their memory. | |
2122 | */ | |
da99ecf1 | 2123 | if (unlikely(tsk_is_oom_victim(current) || |
06b078fc JW |
2124 | fatal_signal_pending(current) || |
2125 | current->flags & PF_EXITING)) | |
10d53c74 | 2126 | goto force; |
06b078fc | 2127 | |
89a28483 JW |
2128 | /* |
2129 | * Prevent unbounded recursion when reclaim operations need to | |
2130 | * allocate memory. This might exceed the limits temporarily, | |
2131 | * but we prefer facilitating memory reclaim and getting back | |
2132 | * under the limit over triggering OOM kills in these cases. | |
2133 | */ | |
2134 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2135 | goto force; | |
2136 | ||
06b078fc JW |
2137 | if (unlikely(task_in_memcg_oom(current))) |
2138 | goto nomem; | |
2139 | ||
d0164adc | 2140 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2141 | goto nomem; |
4b534334 | 2142 | |
e27be240 | 2143 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
241994ed | 2144 | |
b70a2a21 JW |
2145 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2146 | gfp_mask, may_swap); | |
6539cc05 | 2147 | |
61e02c74 | 2148 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2149 | goto retry; |
28c34c29 | 2150 | |
b70a2a21 | 2151 | if (!drained) { |
6d3d6aa2 | 2152 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2153 | drained = true; |
2154 | goto retry; | |
2155 | } | |
2156 | ||
28c34c29 JW |
2157 | if (gfp_mask & __GFP_NORETRY) |
2158 | goto nomem; | |
6539cc05 JW |
2159 | /* |
2160 | * Even though the limit is exceeded at this point, reclaim | |
2161 | * may have been able to free some pages. Retry the charge | |
2162 | * before killing the task. | |
2163 | * | |
2164 | * Only for regular pages, though: huge pages are rather | |
2165 | * unlikely to succeed so close to the limit, and we fall back | |
2166 | * to regular pages anyway in case of failure. | |
2167 | */ | |
61e02c74 | 2168 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2169 | goto retry; |
2170 | /* | |
2171 | * At task move, charge accounts can be doubly counted. So, it's | |
2172 | * better to wait until the end of task_move if something is going on. | |
2173 | */ | |
2174 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2175 | goto retry; | |
2176 | ||
9b130619 JW |
2177 | if (nr_retries--) |
2178 | goto retry; | |
2179 | ||
29ef680a MH |
2180 | if (gfp_mask & __GFP_RETRY_MAYFAIL && oomed) |
2181 | goto nomem; | |
2182 | ||
06b078fc | 2183 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2184 | goto force; |
06b078fc | 2185 | |
6539cc05 | 2186 | if (fatal_signal_pending(current)) |
10d53c74 | 2187 | goto force; |
6539cc05 | 2188 | |
e27be240 | 2189 | memcg_memory_event(mem_over_limit, MEMCG_OOM); |
241994ed | 2190 | |
29ef680a MH |
2191 | /* |
2192 | * keep retrying as long as the memcg oom killer is able to make | |
2193 | * a forward progress or bypass the charge if the oom killer | |
2194 | * couldn't make any progress. | |
2195 | */ | |
2196 | oom_status = mem_cgroup_oom(mem_over_limit, gfp_mask, | |
3608de07 | 2197 | get_order(nr_pages * PAGE_SIZE)); |
29ef680a MH |
2198 | switch (oom_status) { |
2199 | case OOM_SUCCESS: | |
2200 | nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
2201 | oomed = true; | |
2202 | goto retry; | |
2203 | case OOM_FAILED: | |
2204 | goto force; | |
2205 | default: | |
2206 | goto nomem; | |
2207 | } | |
7a81b88c | 2208 | nomem: |
6d1fdc48 | 2209 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2210 | return -ENOMEM; |
10d53c74 TH |
2211 | force: |
2212 | /* | |
2213 | * The allocation either can't fail or will lead to more memory | |
2214 | * being freed very soon. Allow memory usage go over the limit | |
2215 | * temporarily by force charging it. | |
2216 | */ | |
2217 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2218 | if (do_memsw_account()) |
10d53c74 TH |
2219 | page_counter_charge(&memcg->memsw, nr_pages); |
2220 | css_get_many(&memcg->css, nr_pages); | |
2221 | ||
2222 | return 0; | |
6539cc05 JW |
2223 | |
2224 | done_restock: | |
e8ea14cc | 2225 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2226 | if (batch > nr_pages) |
2227 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2228 | |
241994ed | 2229 | /* |
b23afb93 TH |
2230 | * If the hierarchy is above the normal consumption range, schedule |
2231 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2232 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2233 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2234 | * not recorded as it most likely matches current's and won't | |
2235 | * change in the meantime. As high limit is checked again before | |
2236 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2237 | */ |
2238 | do { | |
b23afb93 | 2239 | if (page_counter_read(&memcg->memory) > memcg->high) { |
f7e1cb6e JW |
2240 | /* Don't bother a random interrupted task */ |
2241 | if (in_interrupt()) { | |
2242 | schedule_work(&memcg->high_work); | |
2243 | break; | |
2244 | } | |
9516a18a | 2245 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2246 | set_notify_resume(current); |
2247 | break; | |
2248 | } | |
241994ed | 2249 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2250 | |
2251 | return 0; | |
7a81b88c | 2252 | } |
8a9f3ccd | 2253 | |
00501b53 | 2254 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2255 | { |
ce00a967 JW |
2256 | if (mem_cgroup_is_root(memcg)) |
2257 | return; | |
2258 | ||
3e32cb2e | 2259 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2260 | if (do_memsw_account()) |
3e32cb2e | 2261 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2262 | |
e8ea14cc | 2263 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2264 | } |
2265 | ||
0a31bc97 JW |
2266 | static void lock_page_lru(struct page *page, int *isolated) |
2267 | { | |
2268 | struct zone *zone = page_zone(page); | |
2269 | ||
a52633d8 | 2270 | spin_lock_irq(zone_lru_lock(zone)); |
0a31bc97 JW |
2271 | if (PageLRU(page)) { |
2272 | struct lruvec *lruvec; | |
2273 | ||
599d0c95 | 2274 | lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat); |
0a31bc97 JW |
2275 | ClearPageLRU(page); |
2276 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2277 | *isolated = 1; | |
2278 | } else | |
2279 | *isolated = 0; | |
2280 | } | |
2281 | ||
2282 | static void unlock_page_lru(struct page *page, int isolated) | |
2283 | { | |
2284 | struct zone *zone = page_zone(page); | |
2285 | ||
2286 | if (isolated) { | |
2287 | struct lruvec *lruvec; | |
2288 | ||
599d0c95 | 2289 | lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat); |
0a31bc97 JW |
2290 | VM_BUG_ON_PAGE(PageLRU(page), page); |
2291 | SetPageLRU(page); | |
2292 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2293 | } | |
a52633d8 | 2294 | spin_unlock_irq(zone_lru_lock(zone)); |
0a31bc97 JW |
2295 | } |
2296 | ||
00501b53 | 2297 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2298 | bool lrucare) |
7a81b88c | 2299 | { |
0a31bc97 | 2300 | int isolated; |
9ce70c02 | 2301 | |
1306a85a | 2302 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2303 | |
2304 | /* | |
2305 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2306 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2307 | */ | |
0a31bc97 JW |
2308 | if (lrucare) |
2309 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2310 | |
0a31bc97 JW |
2311 | /* |
2312 | * Nobody should be changing or seriously looking at | |
1306a85a | 2313 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2314 | * |
2315 | * - the page is uncharged | |
2316 | * | |
2317 | * - the page is off-LRU | |
2318 | * | |
2319 | * - an anonymous fault has exclusive page access, except for | |
2320 | * a locked page table | |
2321 | * | |
2322 | * - a page cache insertion, a swapin fault, or a migration | |
2323 | * have the page locked | |
2324 | */ | |
1306a85a | 2325 | page->mem_cgroup = memcg; |
9ce70c02 | 2326 | |
0a31bc97 JW |
2327 | if (lrucare) |
2328 | unlock_page_lru(page, isolated); | |
7a81b88c | 2329 | } |
66e1707b | 2330 | |
84c07d11 | 2331 | #ifdef CONFIG_MEMCG_KMEM |
f3bb3043 | 2332 | static int memcg_alloc_cache_id(void) |
55007d84 | 2333 | { |
f3bb3043 VD |
2334 | int id, size; |
2335 | int err; | |
2336 | ||
dbcf73e2 | 2337 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2338 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2339 | if (id < 0) | |
2340 | return id; | |
55007d84 | 2341 | |
dbcf73e2 | 2342 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2343 | return id; |
2344 | ||
2345 | /* | |
2346 | * There's no space for the new id in memcg_caches arrays, | |
2347 | * so we have to grow them. | |
2348 | */ | |
05257a1a | 2349 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2350 | |
2351 | size = 2 * (id + 1); | |
55007d84 GC |
2352 | if (size < MEMCG_CACHES_MIN_SIZE) |
2353 | size = MEMCG_CACHES_MIN_SIZE; | |
2354 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2355 | size = MEMCG_CACHES_MAX_SIZE; | |
2356 | ||
f3bb3043 | 2357 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2358 | if (!err) |
2359 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2360 | if (!err) |
2361 | memcg_nr_cache_ids = size; | |
2362 | ||
2363 | up_write(&memcg_cache_ids_sem); | |
2364 | ||
f3bb3043 | 2365 | if (err) { |
dbcf73e2 | 2366 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2367 | return err; |
2368 | } | |
2369 | return id; | |
2370 | } | |
2371 | ||
2372 | static void memcg_free_cache_id(int id) | |
2373 | { | |
dbcf73e2 | 2374 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2375 | } |
2376 | ||
d5b3cf71 | 2377 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2378 | struct mem_cgroup *memcg; |
2379 | struct kmem_cache *cachep; | |
2380 | struct work_struct work; | |
2381 | }; | |
2382 | ||
d5b3cf71 | 2383 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2384 | { |
d5b3cf71 VD |
2385 | struct memcg_kmem_cache_create_work *cw = |
2386 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2387 | struct mem_cgroup *memcg = cw->memcg; |
2388 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2389 | |
d5b3cf71 | 2390 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2391 | |
5722d094 | 2392 | css_put(&memcg->css); |
d7f25f8a GC |
2393 | kfree(cw); |
2394 | } | |
2395 | ||
2396 | /* | |
2397 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2398 | */ |
d5b3cf71 VD |
2399 | static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2400 | struct kmem_cache *cachep) | |
d7f25f8a | 2401 | { |
d5b3cf71 | 2402 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2403 | |
c892fd82 | 2404 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT | __GFP_NOWARN); |
8135be5a | 2405 | if (!cw) |
d7f25f8a | 2406 | return; |
8135be5a VD |
2407 | |
2408 | css_get(&memcg->css); | |
d7f25f8a GC |
2409 | |
2410 | cw->memcg = memcg; | |
2411 | cw->cachep = cachep; | |
d5b3cf71 | 2412 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2413 | |
17cc4dfe | 2414 | queue_work(memcg_kmem_cache_wq, &cw->work); |
d7f25f8a GC |
2415 | } |
2416 | ||
d5b3cf71 VD |
2417 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2418 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
2419 | { |
2420 | /* | |
2421 | * We need to stop accounting when we kmalloc, because if the | |
2422 | * corresponding kmalloc cache is not yet created, the first allocation | |
d5b3cf71 | 2423 | * in __memcg_schedule_kmem_cache_create will recurse. |
0e9d92f2 GC |
2424 | * |
2425 | * However, it is better to enclose the whole function. Depending on | |
2426 | * the debugging options enabled, INIT_WORK(), for instance, can | |
2427 | * trigger an allocation. This too, will make us recurse. Because at | |
2428 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
2429 | * the safest choice is to do it like this, wrapping the whole function. | |
2430 | */ | |
6f185c29 | 2431 | current->memcg_kmem_skip_account = 1; |
d5b3cf71 | 2432 | __memcg_schedule_kmem_cache_create(memcg, cachep); |
6f185c29 | 2433 | current->memcg_kmem_skip_account = 0; |
0e9d92f2 | 2434 | } |
c67a8a68 | 2435 | |
45264778 VD |
2436 | static inline bool memcg_kmem_bypass(void) |
2437 | { | |
2438 | if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD)) | |
2439 | return true; | |
2440 | return false; | |
2441 | } | |
2442 | ||
2443 | /** | |
2444 | * memcg_kmem_get_cache: select the correct per-memcg cache for allocation | |
2445 | * @cachep: the original global kmem cache | |
2446 | * | |
d7f25f8a GC |
2447 | * Return the kmem_cache we're supposed to use for a slab allocation. |
2448 | * We try to use the current memcg's version of the cache. | |
2449 | * | |
45264778 VD |
2450 | * If the cache does not exist yet, if we are the first user of it, we |
2451 | * create it asynchronously in a workqueue and let the current allocation | |
2452 | * go through with the original cache. | |
d7f25f8a | 2453 | * |
45264778 VD |
2454 | * This function takes a reference to the cache it returns to assure it |
2455 | * won't get destroyed while we are working with it. Once the caller is | |
2456 | * done with it, memcg_kmem_put_cache() must be called to release the | |
2457 | * reference. | |
d7f25f8a | 2458 | */ |
45264778 | 2459 | struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep) |
d7f25f8a GC |
2460 | { |
2461 | struct mem_cgroup *memcg; | |
959c8963 | 2462 | struct kmem_cache *memcg_cachep; |
2a4db7eb | 2463 | int kmemcg_id; |
d7f25f8a | 2464 | |
f7ce3190 | 2465 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2466 | |
45264778 | 2467 | if (memcg_kmem_bypass()) |
230e9fc2 VD |
2468 | return cachep; |
2469 | ||
9d100c5e | 2470 | if (current->memcg_kmem_skip_account) |
0e9d92f2 GC |
2471 | return cachep; |
2472 | ||
d46eb14b | 2473 | memcg = get_mem_cgroup_from_current(); |
4db0c3c2 | 2474 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
2a4db7eb | 2475 | if (kmemcg_id < 0) |
ca0dde97 | 2476 | goto out; |
d7f25f8a | 2477 | |
2a4db7eb | 2478 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
8135be5a VD |
2479 | if (likely(memcg_cachep)) |
2480 | return memcg_cachep; | |
ca0dde97 LZ |
2481 | |
2482 | /* | |
2483 | * If we are in a safe context (can wait, and not in interrupt | |
2484 | * context), we could be be predictable and return right away. | |
2485 | * This would guarantee that the allocation being performed | |
2486 | * already belongs in the new cache. | |
2487 | * | |
2488 | * However, there are some clashes that can arrive from locking. | |
2489 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2490 | * memcg_create_kmem_cache, this means no further allocation |
2491 | * could happen with the slab_mutex held. So it's better to | |
2492 | * defer everything. | |
ca0dde97 | 2493 | */ |
d5b3cf71 | 2494 | memcg_schedule_kmem_cache_create(memcg, cachep); |
ca0dde97 | 2495 | out: |
8135be5a | 2496 | css_put(&memcg->css); |
ca0dde97 | 2497 | return cachep; |
d7f25f8a | 2498 | } |
d7f25f8a | 2499 | |
45264778 VD |
2500 | /** |
2501 | * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache | |
2502 | * @cachep: the cache returned by memcg_kmem_get_cache | |
2503 | */ | |
2504 | void memcg_kmem_put_cache(struct kmem_cache *cachep) | |
8135be5a VD |
2505 | { |
2506 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2507 | css_put(&cachep->memcg_params.memcg->css); |
8135be5a VD |
2508 | } |
2509 | ||
45264778 | 2510 | /** |
b213b54f | 2511 | * memcg_kmem_charge_memcg: charge a kmem page |
45264778 VD |
2512 | * @page: page to charge |
2513 | * @gfp: reclaim mode | |
2514 | * @order: allocation order | |
2515 | * @memcg: memory cgroup to charge | |
2516 | * | |
2517 | * Returns 0 on success, an error code on failure. | |
2518 | */ | |
2519 | int memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, | |
2520 | struct mem_cgroup *memcg) | |
7ae1e1d0 | 2521 | { |
f3ccb2c4 VD |
2522 | unsigned int nr_pages = 1 << order; |
2523 | struct page_counter *counter; | |
7ae1e1d0 GC |
2524 | int ret; |
2525 | ||
f3ccb2c4 | 2526 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 2527 | if (ret) |
f3ccb2c4 | 2528 | return ret; |
52c29b04 JW |
2529 | |
2530 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
2531 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
2532 | cancel_charge(memcg, nr_pages); | |
2533 | return -ENOMEM; | |
7ae1e1d0 GC |
2534 | } |
2535 | ||
f3ccb2c4 | 2536 | page->mem_cgroup = memcg; |
7ae1e1d0 | 2537 | |
f3ccb2c4 | 2538 | return 0; |
7ae1e1d0 GC |
2539 | } |
2540 | ||
45264778 VD |
2541 | /** |
2542 | * memcg_kmem_charge: charge a kmem page to the current memory cgroup | |
2543 | * @page: page to charge | |
2544 | * @gfp: reclaim mode | |
2545 | * @order: allocation order | |
2546 | * | |
2547 | * Returns 0 on success, an error code on failure. | |
2548 | */ | |
2549 | int memcg_kmem_charge(struct page *page, gfp_t gfp, int order) | |
7ae1e1d0 | 2550 | { |
f3ccb2c4 | 2551 | struct mem_cgroup *memcg; |
fcff7d7e | 2552 | int ret = 0; |
7ae1e1d0 | 2553 | |
45264778 VD |
2554 | if (memcg_kmem_bypass()) |
2555 | return 0; | |
2556 | ||
d46eb14b | 2557 | memcg = get_mem_cgroup_from_current(); |
c4159a75 | 2558 | if (!mem_cgroup_is_root(memcg)) { |
45264778 | 2559 | ret = memcg_kmem_charge_memcg(page, gfp, order, memcg); |
c4159a75 VD |
2560 | if (!ret) |
2561 | __SetPageKmemcg(page); | |
2562 | } | |
7ae1e1d0 | 2563 | css_put(&memcg->css); |
d05e83a6 | 2564 | return ret; |
7ae1e1d0 | 2565 | } |
45264778 VD |
2566 | /** |
2567 | * memcg_kmem_uncharge: uncharge a kmem page | |
2568 | * @page: page to uncharge | |
2569 | * @order: allocation order | |
2570 | */ | |
2571 | void memcg_kmem_uncharge(struct page *page, int order) | |
7ae1e1d0 | 2572 | { |
1306a85a | 2573 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 2574 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 2575 | |
7ae1e1d0 GC |
2576 | if (!memcg) |
2577 | return; | |
2578 | ||
309381fe | 2579 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 2580 | |
52c29b04 JW |
2581 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
2582 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
2583 | ||
f3ccb2c4 | 2584 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2585 | if (do_memsw_account()) |
f3ccb2c4 | 2586 | page_counter_uncharge(&memcg->memsw, nr_pages); |
60d3fd32 | 2587 | |
1306a85a | 2588 | page->mem_cgroup = NULL; |
c4159a75 VD |
2589 | |
2590 | /* slab pages do not have PageKmemcg flag set */ | |
2591 | if (PageKmemcg(page)) | |
2592 | __ClearPageKmemcg(page); | |
2593 | ||
f3ccb2c4 | 2594 | css_put_many(&memcg->css, nr_pages); |
60d3fd32 | 2595 | } |
84c07d11 | 2596 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 2597 | |
ca3e0214 KH |
2598 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2599 | ||
ca3e0214 KH |
2600 | /* |
2601 | * Because tail pages are not marked as "used", set it. We're under | |
a52633d8 | 2602 | * zone_lru_lock and migration entries setup in all page mappings. |
ca3e0214 | 2603 | */ |
e94c8a9c | 2604 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2605 | { |
e94c8a9c | 2606 | int i; |
ca3e0214 | 2607 | |
3d37c4a9 KH |
2608 | if (mem_cgroup_disabled()) |
2609 | return; | |
b070e65c | 2610 | |
29833315 | 2611 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2612 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2613 | |
c9019e9b | 2614 | __mod_memcg_state(head->mem_cgroup, MEMCG_RSS_HUGE, -HPAGE_PMD_NR); |
ca3e0214 | 2615 | } |
12d27107 | 2616 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2617 | |
c255a458 | 2618 | #ifdef CONFIG_MEMCG_SWAP |
02491447 DN |
2619 | /** |
2620 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2621 | * @entry: swap entry to be moved | |
2622 | * @from: mem_cgroup which the entry is moved from | |
2623 | * @to: mem_cgroup which the entry is moved to | |
2624 | * | |
2625 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2626 | * as the mem_cgroup's id of @from. | |
2627 | * | |
2628 | * Returns 0 on success, -EINVAL on failure. | |
2629 | * | |
3e32cb2e | 2630 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2631 | * both res and memsw, and called css_get(). |
2632 | */ | |
2633 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2634 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2635 | { |
2636 | unsigned short old_id, new_id; | |
2637 | ||
34c00c31 LZ |
2638 | old_id = mem_cgroup_id(from); |
2639 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2640 | |
2641 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
2642 | mod_memcg_state(from, MEMCG_SWAP, -1); |
2643 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
2644 | return 0; |
2645 | } | |
2646 | return -EINVAL; | |
2647 | } | |
2648 | #else | |
2649 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2650 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2651 | { |
2652 | return -EINVAL; | |
2653 | } | |
8c7c6e34 | 2654 | #endif |
d13d1443 | 2655 | |
bbec2e15 | 2656 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 2657 | |
bbec2e15 RG |
2658 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
2659 | unsigned long max, bool memsw) | |
628f4235 | 2660 | { |
3e32cb2e | 2661 | bool enlarge = false; |
bb4a7ea2 | 2662 | bool drained = false; |
3e32cb2e | 2663 | int ret; |
c054a78c YZ |
2664 | bool limits_invariant; |
2665 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 2666 | |
3e32cb2e | 2667 | do { |
628f4235 KH |
2668 | if (signal_pending(current)) { |
2669 | ret = -EINTR; | |
2670 | break; | |
2671 | } | |
3e32cb2e | 2672 | |
bbec2e15 | 2673 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
2674 | /* |
2675 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 2676 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 2677 | */ |
bbec2e15 RG |
2678 | limits_invariant = memsw ? max >= memcg->memory.max : |
2679 | max <= memcg->memsw.max; | |
c054a78c | 2680 | if (!limits_invariant) { |
bbec2e15 | 2681 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 2682 | ret = -EINVAL; |
8c7c6e34 KH |
2683 | break; |
2684 | } | |
bbec2e15 | 2685 | if (max > counter->max) |
3e32cb2e | 2686 | enlarge = true; |
bbec2e15 RG |
2687 | ret = page_counter_set_max(counter, max); |
2688 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
2689 | |
2690 | if (!ret) | |
2691 | break; | |
2692 | ||
bb4a7ea2 SB |
2693 | if (!drained) { |
2694 | drain_all_stock(memcg); | |
2695 | drained = true; | |
2696 | continue; | |
2697 | } | |
2698 | ||
1ab5c056 AR |
2699 | if (!try_to_free_mem_cgroup_pages(memcg, 1, |
2700 | GFP_KERNEL, !memsw)) { | |
2701 | ret = -EBUSY; | |
2702 | break; | |
2703 | } | |
2704 | } while (true); | |
3e32cb2e | 2705 | |
3c11ecf4 KH |
2706 | if (!ret && enlarge) |
2707 | memcg_oom_recover(memcg); | |
3e32cb2e | 2708 | |
628f4235 KH |
2709 | return ret; |
2710 | } | |
2711 | ||
ef8f2327 | 2712 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
2713 | gfp_t gfp_mask, |
2714 | unsigned long *total_scanned) | |
2715 | { | |
2716 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 2717 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
2718 | unsigned long reclaimed; |
2719 | int loop = 0; | |
ef8f2327 | 2720 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 2721 | unsigned long excess; |
0608f43d AM |
2722 | unsigned long nr_scanned; |
2723 | ||
2724 | if (order > 0) | |
2725 | return 0; | |
2726 | ||
ef8f2327 | 2727 | mctz = soft_limit_tree_node(pgdat->node_id); |
d6507ff5 MH |
2728 | |
2729 | /* | |
2730 | * Do not even bother to check the largest node if the root | |
2731 | * is empty. Do it lockless to prevent lock bouncing. Races | |
2732 | * are acceptable as soft limit is best effort anyway. | |
2733 | */ | |
bfc7228b | 2734 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
2735 | return 0; |
2736 | ||
0608f43d AM |
2737 | /* |
2738 | * This loop can run a while, specially if mem_cgroup's continuously | |
2739 | * keep exceeding their soft limit and putting the system under | |
2740 | * pressure | |
2741 | */ | |
2742 | do { | |
2743 | if (next_mz) | |
2744 | mz = next_mz; | |
2745 | else | |
2746 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2747 | if (!mz) | |
2748 | break; | |
2749 | ||
2750 | nr_scanned = 0; | |
ef8f2327 | 2751 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
0608f43d AM |
2752 | gfp_mask, &nr_scanned); |
2753 | nr_reclaimed += reclaimed; | |
2754 | *total_scanned += nr_scanned; | |
0a31bc97 | 2755 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 2756 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
2757 | |
2758 | /* | |
2759 | * If we failed to reclaim anything from this memory cgroup | |
2760 | * it is time to move on to the next cgroup | |
2761 | */ | |
2762 | next_mz = NULL; | |
bc2f2e7f VD |
2763 | if (!reclaimed) |
2764 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
2765 | ||
3e32cb2e | 2766 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
2767 | /* |
2768 | * One school of thought says that we should not add | |
2769 | * back the node to the tree if reclaim returns 0. | |
2770 | * But our reclaim could return 0, simply because due | |
2771 | * to priority we are exposing a smaller subset of | |
2772 | * memory to reclaim from. Consider this as a longer | |
2773 | * term TODO. | |
2774 | */ | |
2775 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 2776 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 2777 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
2778 | css_put(&mz->memcg->css); |
2779 | loop++; | |
2780 | /* | |
2781 | * Could not reclaim anything and there are no more | |
2782 | * mem cgroups to try or we seem to be looping without | |
2783 | * reclaiming anything. | |
2784 | */ | |
2785 | if (!nr_reclaimed && | |
2786 | (next_mz == NULL || | |
2787 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2788 | break; | |
2789 | } while (!nr_reclaimed); | |
2790 | if (next_mz) | |
2791 | css_put(&next_mz->memcg->css); | |
2792 | return nr_reclaimed; | |
2793 | } | |
2794 | ||
ea280e7b TH |
2795 | /* |
2796 | * Test whether @memcg has children, dead or alive. Note that this | |
2797 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
2798 | * returns %true if there are child csses according to the cgroup | |
2799 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
2800 | */ | |
b5f99b53 GC |
2801 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
2802 | { | |
ea280e7b TH |
2803 | bool ret; |
2804 | ||
ea280e7b TH |
2805 | rcu_read_lock(); |
2806 | ret = css_next_child(NULL, &memcg->css); | |
2807 | rcu_read_unlock(); | |
2808 | return ret; | |
b5f99b53 GC |
2809 | } |
2810 | ||
c26251f9 | 2811 | /* |
51038171 | 2812 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
2813 | * |
2814 | * Caller is responsible for holding css reference for memcg. | |
2815 | */ | |
2816 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
2817 | { | |
2818 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 2819 | |
c1e862c1 KH |
2820 | /* we call try-to-free pages for make this cgroup empty */ |
2821 | lru_add_drain_all(); | |
d12c60f6 JS |
2822 | |
2823 | drain_all_stock(memcg); | |
2824 | ||
f817ed48 | 2825 | /* try to free all pages in this cgroup */ |
3e32cb2e | 2826 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 2827 | int progress; |
c1e862c1 | 2828 | |
c26251f9 MH |
2829 | if (signal_pending(current)) |
2830 | return -EINTR; | |
2831 | ||
b70a2a21 JW |
2832 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
2833 | GFP_KERNEL, true); | |
c1e862c1 | 2834 | if (!progress) { |
f817ed48 | 2835 | nr_retries--; |
c1e862c1 | 2836 | /* maybe some writeback is necessary */ |
8aa7e847 | 2837 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 2838 | } |
f817ed48 KH |
2839 | |
2840 | } | |
ab5196c2 MH |
2841 | |
2842 | return 0; | |
cc847582 KH |
2843 | } |
2844 | ||
6770c64e TH |
2845 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
2846 | char *buf, size_t nbytes, | |
2847 | loff_t off) | |
c1e862c1 | 2848 | { |
6770c64e | 2849 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 2850 | |
d8423011 MH |
2851 | if (mem_cgroup_is_root(memcg)) |
2852 | return -EINVAL; | |
6770c64e | 2853 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
2854 | } |
2855 | ||
182446d0 TH |
2856 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
2857 | struct cftype *cft) | |
18f59ea7 | 2858 | { |
182446d0 | 2859 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
2860 | } |
2861 | ||
182446d0 TH |
2862 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
2863 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
2864 | { |
2865 | int retval = 0; | |
182446d0 | 2866 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 2867 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 2868 | |
567fb435 | 2869 | if (memcg->use_hierarchy == val) |
0b8f73e1 | 2870 | return 0; |
567fb435 | 2871 | |
18f59ea7 | 2872 | /* |
af901ca1 | 2873 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
2874 | * in the child subtrees. If it is unset, then the change can |
2875 | * occur, provided the current cgroup has no children. | |
2876 | * | |
2877 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
2878 | * set if there are no children. | |
2879 | */ | |
c0ff4b85 | 2880 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 2881 | (val == 1 || val == 0)) { |
ea280e7b | 2882 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 2883 | memcg->use_hierarchy = val; |
18f59ea7 BS |
2884 | else |
2885 | retval = -EBUSY; | |
2886 | } else | |
2887 | retval = -EINVAL; | |
567fb435 | 2888 | |
18f59ea7 BS |
2889 | return retval; |
2890 | } | |
2891 | ||
72b54e73 | 2892 | static void tree_stat(struct mem_cgroup *memcg, unsigned long *stat) |
ce00a967 JW |
2893 | { |
2894 | struct mem_cgroup *iter; | |
72b54e73 | 2895 | int i; |
ce00a967 | 2896 | |
72b54e73 | 2897 | memset(stat, 0, sizeof(*stat) * MEMCG_NR_STAT); |
ce00a967 | 2898 | |
72b54e73 VD |
2899 | for_each_mem_cgroup_tree(iter, memcg) { |
2900 | for (i = 0; i < MEMCG_NR_STAT; i++) | |
ccda7f43 | 2901 | stat[i] += memcg_page_state(iter, i); |
72b54e73 | 2902 | } |
ce00a967 JW |
2903 | } |
2904 | ||
72b54e73 | 2905 | static void tree_events(struct mem_cgroup *memcg, unsigned long *events) |
587d9f72 JW |
2906 | { |
2907 | struct mem_cgroup *iter; | |
72b54e73 | 2908 | int i; |
587d9f72 | 2909 | |
e27be240 | 2910 | memset(events, 0, sizeof(*events) * NR_VM_EVENT_ITEMS); |
587d9f72 | 2911 | |
72b54e73 | 2912 | for_each_mem_cgroup_tree(iter, memcg) { |
e27be240 | 2913 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) |
ccda7f43 | 2914 | events[i] += memcg_sum_events(iter, i); |
72b54e73 | 2915 | } |
587d9f72 JW |
2916 | } |
2917 | ||
6f646156 | 2918 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 2919 | { |
72b54e73 | 2920 | unsigned long val = 0; |
ce00a967 | 2921 | |
3e32cb2e | 2922 | if (mem_cgroup_is_root(memcg)) { |
72b54e73 VD |
2923 | struct mem_cgroup *iter; |
2924 | ||
2925 | for_each_mem_cgroup_tree(iter, memcg) { | |
ccda7f43 JW |
2926 | val += memcg_page_state(iter, MEMCG_CACHE); |
2927 | val += memcg_page_state(iter, MEMCG_RSS); | |
72b54e73 | 2928 | if (swap) |
ccda7f43 | 2929 | val += memcg_page_state(iter, MEMCG_SWAP); |
72b54e73 | 2930 | } |
3e32cb2e | 2931 | } else { |
ce00a967 | 2932 | if (!swap) |
3e32cb2e | 2933 | val = page_counter_read(&memcg->memory); |
ce00a967 | 2934 | else |
3e32cb2e | 2935 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 2936 | } |
c12176d3 | 2937 | return val; |
ce00a967 JW |
2938 | } |
2939 | ||
3e32cb2e JW |
2940 | enum { |
2941 | RES_USAGE, | |
2942 | RES_LIMIT, | |
2943 | RES_MAX_USAGE, | |
2944 | RES_FAILCNT, | |
2945 | RES_SOFT_LIMIT, | |
2946 | }; | |
ce00a967 | 2947 | |
791badbd | 2948 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 2949 | struct cftype *cft) |
8cdea7c0 | 2950 | { |
182446d0 | 2951 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 2952 | struct page_counter *counter; |
af36f906 | 2953 | |
3e32cb2e | 2954 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 2955 | case _MEM: |
3e32cb2e JW |
2956 | counter = &memcg->memory; |
2957 | break; | |
8c7c6e34 | 2958 | case _MEMSWAP: |
3e32cb2e JW |
2959 | counter = &memcg->memsw; |
2960 | break; | |
510fc4e1 | 2961 | case _KMEM: |
3e32cb2e | 2962 | counter = &memcg->kmem; |
510fc4e1 | 2963 | break; |
d55f90bf | 2964 | case _TCP: |
0db15298 | 2965 | counter = &memcg->tcpmem; |
d55f90bf | 2966 | break; |
8c7c6e34 KH |
2967 | default: |
2968 | BUG(); | |
8c7c6e34 | 2969 | } |
3e32cb2e JW |
2970 | |
2971 | switch (MEMFILE_ATTR(cft->private)) { | |
2972 | case RES_USAGE: | |
2973 | if (counter == &memcg->memory) | |
c12176d3 | 2974 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 2975 | if (counter == &memcg->memsw) |
c12176d3 | 2976 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
2977 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
2978 | case RES_LIMIT: | |
bbec2e15 | 2979 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
2980 | case RES_MAX_USAGE: |
2981 | return (u64)counter->watermark * PAGE_SIZE; | |
2982 | case RES_FAILCNT: | |
2983 | return counter->failcnt; | |
2984 | case RES_SOFT_LIMIT: | |
2985 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
2986 | default: | |
2987 | BUG(); | |
2988 | } | |
8cdea7c0 | 2989 | } |
510fc4e1 | 2990 | |
84c07d11 | 2991 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 2992 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 2993 | { |
d6441637 VD |
2994 | int memcg_id; |
2995 | ||
b313aeee VD |
2996 | if (cgroup_memory_nokmem) |
2997 | return 0; | |
2998 | ||
2a4db7eb | 2999 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 3000 | BUG_ON(memcg->kmem_state); |
d6441637 | 3001 | |
f3bb3043 | 3002 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
3003 | if (memcg_id < 0) |
3004 | return memcg_id; | |
d6441637 | 3005 | |
ef12947c | 3006 | static_branch_inc(&memcg_kmem_enabled_key); |
d6441637 | 3007 | /* |
567e9ab2 | 3008 | * A memory cgroup is considered kmem-online as soon as it gets |
900a38f0 | 3009 | * kmemcg_id. Setting the id after enabling static branching will |
d6441637 VD |
3010 | * guarantee no one starts accounting before all call sites are |
3011 | * patched. | |
3012 | */ | |
900a38f0 | 3013 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 3014 | memcg->kmem_state = KMEM_ONLINE; |
bc2791f8 | 3015 | INIT_LIST_HEAD(&memcg->kmem_caches); |
0b8f73e1 JW |
3016 | |
3017 | return 0; | |
d6441637 VD |
3018 | } |
3019 | ||
8e0a8912 JW |
3020 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
3021 | { | |
3022 | struct cgroup_subsys_state *css; | |
3023 | struct mem_cgroup *parent, *child; | |
3024 | int kmemcg_id; | |
3025 | ||
3026 | if (memcg->kmem_state != KMEM_ONLINE) | |
3027 | return; | |
3028 | /* | |
3029 | * Clear the online state before clearing memcg_caches array | |
3030 | * entries. The slab_mutex in memcg_deactivate_kmem_caches() | |
3031 | * guarantees that no cache will be created for this cgroup | |
3032 | * after we are done (see memcg_create_kmem_cache()). | |
3033 | */ | |
3034 | memcg->kmem_state = KMEM_ALLOCATED; | |
3035 | ||
3036 | memcg_deactivate_kmem_caches(memcg); | |
3037 | ||
3038 | kmemcg_id = memcg->kmemcg_id; | |
3039 | BUG_ON(kmemcg_id < 0); | |
3040 | ||
3041 | parent = parent_mem_cgroup(memcg); | |
3042 | if (!parent) | |
3043 | parent = root_mem_cgroup; | |
3044 | ||
3045 | /* | |
3046 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3047 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3048 | * to ones of the parent. After we have finished, all list_lrus | |
3049 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3050 | * ordering is imposed by list_lru_node->lock taken by | |
3051 | * memcg_drain_all_list_lrus(). | |
3052 | */ | |
3a06bb78 | 3053 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
8e0a8912 JW |
3054 | css_for_each_descendant_pre(css, &memcg->css) { |
3055 | child = mem_cgroup_from_css(css); | |
3056 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3057 | child->kmemcg_id = parent->kmemcg_id; | |
3058 | if (!memcg->use_hierarchy) | |
3059 | break; | |
3060 | } | |
3a06bb78 TH |
3061 | rcu_read_unlock(); |
3062 | ||
8e0a8912 JW |
3063 | memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id); |
3064 | ||
3065 | memcg_free_cache_id(kmemcg_id); | |
3066 | } | |
3067 | ||
3068 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3069 | { | |
0b8f73e1 JW |
3070 | /* css_alloc() failed, offlining didn't happen */ |
3071 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
3072 | memcg_offline_kmem(memcg); | |
3073 | ||
8e0a8912 JW |
3074 | if (memcg->kmem_state == KMEM_ALLOCATED) { |
3075 | memcg_destroy_kmem_caches(memcg); | |
3076 | static_branch_dec(&memcg_kmem_enabled_key); | |
3077 | WARN_ON(page_counter_read(&memcg->kmem)); | |
3078 | } | |
8e0a8912 | 3079 | } |
d6441637 | 3080 | #else |
0b8f73e1 | 3081 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
3082 | { |
3083 | return 0; | |
3084 | } | |
3085 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
3086 | { | |
3087 | } | |
3088 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3089 | { | |
3090 | } | |
84c07d11 | 3091 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 3092 | |
bbec2e15 RG |
3093 | static int memcg_update_kmem_max(struct mem_cgroup *memcg, |
3094 | unsigned long max) | |
d6441637 | 3095 | { |
b313aeee | 3096 | int ret; |
127424c8 | 3097 | |
bbec2e15 RG |
3098 | mutex_lock(&memcg_max_mutex); |
3099 | ret = page_counter_set_max(&memcg->kmem, max); | |
3100 | mutex_unlock(&memcg_max_mutex); | |
127424c8 | 3101 | return ret; |
d6441637 | 3102 | } |
510fc4e1 | 3103 | |
bbec2e15 | 3104 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
3105 | { |
3106 | int ret; | |
3107 | ||
bbec2e15 | 3108 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 3109 | |
bbec2e15 | 3110 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
3111 | if (ret) |
3112 | goto out; | |
3113 | ||
0db15298 | 3114 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
3115 | /* |
3116 | * The active flag needs to be written after the static_key | |
3117 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
3118 | * function is the last one to run. See mem_cgroup_sk_alloc() |
3119 | * for details, and note that we don't mark any socket as | |
3120 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
3121 | * |
3122 | * We need to do this, because static_keys will span multiple | |
3123 | * sites, but we can't control their order. If we mark a socket | |
3124 | * as accounted, but the accounting functions are not patched in | |
3125 | * yet, we'll lose accounting. | |
3126 | * | |
2d758073 | 3127 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
3128 | * because when this value change, the code to process it is not |
3129 | * patched in yet. | |
3130 | */ | |
3131 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 3132 | memcg->tcpmem_active = true; |
d55f90bf VD |
3133 | } |
3134 | out: | |
bbec2e15 | 3135 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
3136 | return ret; |
3137 | } | |
d55f90bf | 3138 | |
628f4235 KH |
3139 | /* |
3140 | * The user of this function is... | |
3141 | * RES_LIMIT. | |
3142 | */ | |
451af504 TH |
3143 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3144 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3145 | { |
451af504 | 3146 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3147 | unsigned long nr_pages; |
628f4235 KH |
3148 | int ret; |
3149 | ||
451af504 | 3150 | buf = strstrip(buf); |
650c5e56 | 3151 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3152 | if (ret) |
3153 | return ret; | |
af36f906 | 3154 | |
3e32cb2e | 3155 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3156 | case RES_LIMIT: |
4b3bde4c BS |
3157 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3158 | ret = -EINVAL; | |
3159 | break; | |
3160 | } | |
3e32cb2e JW |
3161 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3162 | case _MEM: | |
bbec2e15 | 3163 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 3164 | break; |
3e32cb2e | 3165 | case _MEMSWAP: |
bbec2e15 | 3166 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 3167 | break; |
3e32cb2e | 3168 | case _KMEM: |
bbec2e15 | 3169 | ret = memcg_update_kmem_max(memcg, nr_pages); |
3e32cb2e | 3170 | break; |
d55f90bf | 3171 | case _TCP: |
bbec2e15 | 3172 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 3173 | break; |
3e32cb2e | 3174 | } |
296c81d8 | 3175 | break; |
3e32cb2e JW |
3176 | case RES_SOFT_LIMIT: |
3177 | memcg->soft_limit = nr_pages; | |
3178 | ret = 0; | |
628f4235 KH |
3179 | break; |
3180 | } | |
451af504 | 3181 | return ret ?: nbytes; |
8cdea7c0 BS |
3182 | } |
3183 | ||
6770c64e TH |
3184 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3185 | size_t nbytes, loff_t off) | |
c84872e1 | 3186 | { |
6770c64e | 3187 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3188 | struct page_counter *counter; |
c84872e1 | 3189 | |
3e32cb2e JW |
3190 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3191 | case _MEM: | |
3192 | counter = &memcg->memory; | |
3193 | break; | |
3194 | case _MEMSWAP: | |
3195 | counter = &memcg->memsw; | |
3196 | break; | |
3197 | case _KMEM: | |
3198 | counter = &memcg->kmem; | |
3199 | break; | |
d55f90bf | 3200 | case _TCP: |
0db15298 | 3201 | counter = &memcg->tcpmem; |
d55f90bf | 3202 | break; |
3e32cb2e JW |
3203 | default: |
3204 | BUG(); | |
3205 | } | |
af36f906 | 3206 | |
3e32cb2e | 3207 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3208 | case RES_MAX_USAGE: |
3e32cb2e | 3209 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3210 | break; |
3211 | case RES_FAILCNT: | |
3e32cb2e | 3212 | counter->failcnt = 0; |
29f2a4da | 3213 | break; |
3e32cb2e JW |
3214 | default: |
3215 | BUG(); | |
29f2a4da | 3216 | } |
f64c3f54 | 3217 | |
6770c64e | 3218 | return nbytes; |
c84872e1 PE |
3219 | } |
3220 | ||
182446d0 | 3221 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3222 | struct cftype *cft) |
3223 | { | |
182446d0 | 3224 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3225 | } |
3226 | ||
02491447 | 3227 | #ifdef CONFIG_MMU |
182446d0 | 3228 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3229 | struct cftype *cft, u64 val) |
3230 | { | |
182446d0 | 3231 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3232 | |
1dfab5ab | 3233 | if (val & ~MOVE_MASK) |
7dc74be0 | 3234 | return -EINVAL; |
ee5e8472 | 3235 | |
7dc74be0 | 3236 | /* |
ee5e8472 GC |
3237 | * No kind of locking is needed in here, because ->can_attach() will |
3238 | * check this value once in the beginning of the process, and then carry | |
3239 | * on with stale data. This means that changes to this value will only | |
3240 | * affect task migrations starting after the change. | |
7dc74be0 | 3241 | */ |
c0ff4b85 | 3242 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3243 | return 0; |
3244 | } | |
02491447 | 3245 | #else |
182446d0 | 3246 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3247 | struct cftype *cft, u64 val) |
3248 | { | |
3249 | return -ENOSYS; | |
3250 | } | |
3251 | #endif | |
7dc74be0 | 3252 | |
406eb0c9 | 3253 | #ifdef CONFIG_NUMA |
2da8ca82 | 3254 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3255 | { |
25485de6 GT |
3256 | struct numa_stat { |
3257 | const char *name; | |
3258 | unsigned int lru_mask; | |
3259 | }; | |
3260 | ||
3261 | static const struct numa_stat stats[] = { | |
3262 | { "total", LRU_ALL }, | |
3263 | { "file", LRU_ALL_FILE }, | |
3264 | { "anon", LRU_ALL_ANON }, | |
3265 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3266 | }; | |
3267 | const struct numa_stat *stat; | |
406eb0c9 | 3268 | int nid; |
25485de6 | 3269 | unsigned long nr; |
2da8ca82 | 3270 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 3271 | |
25485de6 GT |
3272 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3273 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3274 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3275 | for_each_node_state(nid, N_MEMORY) { | |
3276 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3277 | stat->lru_mask); | |
3278 | seq_printf(m, " N%d=%lu", nid, nr); | |
3279 | } | |
3280 | seq_putc(m, '\n'); | |
406eb0c9 | 3281 | } |
406eb0c9 | 3282 | |
071aee13 YH |
3283 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3284 | struct mem_cgroup *iter; | |
3285 | ||
3286 | nr = 0; | |
3287 | for_each_mem_cgroup_tree(iter, memcg) | |
3288 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3289 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3290 | for_each_node_state(nid, N_MEMORY) { | |
3291 | nr = 0; | |
3292 | for_each_mem_cgroup_tree(iter, memcg) | |
3293 | nr += mem_cgroup_node_nr_lru_pages( | |
3294 | iter, nid, stat->lru_mask); | |
3295 | seq_printf(m, " N%d=%lu", nid, nr); | |
3296 | } | |
3297 | seq_putc(m, '\n'); | |
406eb0c9 | 3298 | } |
406eb0c9 | 3299 | |
406eb0c9 YH |
3300 | return 0; |
3301 | } | |
3302 | #endif /* CONFIG_NUMA */ | |
3303 | ||
df0e53d0 | 3304 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 3305 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
3306 | PGPGIN, |
3307 | PGPGOUT, | |
3308 | PGFAULT, | |
3309 | PGMAJFAULT, | |
3310 | }; | |
3311 | ||
3312 | static const char *const memcg1_event_names[] = { | |
3313 | "pgpgin", | |
3314 | "pgpgout", | |
3315 | "pgfault", | |
3316 | "pgmajfault", | |
3317 | }; | |
3318 | ||
2da8ca82 | 3319 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3320 | { |
2da8ca82 | 3321 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 3322 | unsigned long memory, memsw; |
af7c4b0e JW |
3323 | struct mem_cgroup *mi; |
3324 | unsigned int i; | |
406eb0c9 | 3325 | |
71cd3113 | 3326 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c RS |
3327 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3328 | ||
71cd3113 JW |
3329 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
3330 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) | |
1dd3a273 | 3331 | continue; |
71cd3113 | 3332 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], |
ccda7f43 | 3333 | memcg_page_state(memcg, memcg1_stats[i]) * |
71cd3113 | 3334 | PAGE_SIZE); |
1dd3a273 | 3335 | } |
7b854121 | 3336 | |
df0e53d0 JW |
3337 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
3338 | seq_printf(m, "%s %lu\n", memcg1_event_names[i], | |
ccda7f43 | 3339 | memcg_sum_events(memcg, memcg1_events[i])); |
af7c4b0e JW |
3340 | |
3341 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3342 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
3343 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
3344 | ||
14067bb3 | 3345 | /* Hierarchical information */ |
3e32cb2e JW |
3346 | memory = memsw = PAGE_COUNTER_MAX; |
3347 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
bbec2e15 RG |
3348 | memory = min(memory, mi->memory.max); |
3349 | memsw = min(memsw, mi->memsw.max); | |
fee7b548 | 3350 | } |
3e32cb2e JW |
3351 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3352 | (u64)memory * PAGE_SIZE); | |
7941d214 | 3353 | if (do_memsw_account()) |
3e32cb2e JW |
3354 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
3355 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3356 | |
71cd3113 | 3357 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
484ebb3b | 3358 | unsigned long long val = 0; |
af7c4b0e | 3359 | |
71cd3113 | 3360 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 3361 | continue; |
af7c4b0e | 3362 | for_each_mem_cgroup_tree(mi, memcg) |
ccda7f43 | 3363 | val += memcg_page_state(mi, memcg1_stats[i]) * |
71cd3113 JW |
3364 | PAGE_SIZE; |
3365 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], val); | |
af7c4b0e JW |
3366 | } |
3367 | ||
df0e53d0 | 3368 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) { |
af7c4b0e JW |
3369 | unsigned long long val = 0; |
3370 | ||
3371 | for_each_mem_cgroup_tree(mi, memcg) | |
ccda7f43 | 3372 | val += memcg_sum_events(mi, memcg1_events[i]); |
df0e53d0 | 3373 | seq_printf(m, "total_%s %llu\n", memcg1_event_names[i], val); |
af7c4b0e JW |
3374 | } |
3375 | ||
3376 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
3377 | unsigned long long val = 0; | |
3378 | ||
3379 | for_each_mem_cgroup_tree(mi, memcg) | |
3380 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
3381 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 3382 | } |
14067bb3 | 3383 | |
7f016ee8 | 3384 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 3385 | { |
ef8f2327 MG |
3386 | pg_data_t *pgdat; |
3387 | struct mem_cgroup_per_node *mz; | |
89abfab1 | 3388 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3389 | unsigned long recent_rotated[2] = {0, 0}; |
3390 | unsigned long recent_scanned[2] = {0, 0}; | |
3391 | ||
ef8f2327 MG |
3392 | for_each_online_pgdat(pgdat) { |
3393 | mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); | |
3394 | rstat = &mz->lruvec.reclaim_stat; | |
7f016ee8 | 3395 | |
ef8f2327 MG |
3396 | recent_rotated[0] += rstat->recent_rotated[0]; |
3397 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3398 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3399 | recent_scanned[1] += rstat->recent_scanned[1]; | |
3400 | } | |
78ccf5b5 JW |
3401 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3402 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3403 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3404 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3405 | } |
3406 | #endif | |
3407 | ||
d2ceb9b7 KH |
3408 | return 0; |
3409 | } | |
3410 | ||
182446d0 TH |
3411 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3412 | struct cftype *cft) | |
a7885eb8 | 3413 | { |
182446d0 | 3414 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3415 | |
1f4c025b | 3416 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3417 | } |
3418 | ||
182446d0 TH |
3419 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3420 | struct cftype *cft, u64 val) | |
a7885eb8 | 3421 | { |
182446d0 | 3422 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3423 | |
3dae7fec | 3424 | if (val > 100) |
a7885eb8 KM |
3425 | return -EINVAL; |
3426 | ||
14208b0e | 3427 | if (css->parent) |
3dae7fec JW |
3428 | memcg->swappiness = val; |
3429 | else | |
3430 | vm_swappiness = val; | |
068b38c1 | 3431 | |
a7885eb8 KM |
3432 | return 0; |
3433 | } | |
3434 | ||
2e72b634 KS |
3435 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3436 | { | |
3437 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3438 | unsigned long usage; |
2e72b634 KS |
3439 | int i; |
3440 | ||
3441 | rcu_read_lock(); | |
3442 | if (!swap) | |
2c488db2 | 3443 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3444 | else |
2c488db2 | 3445 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3446 | |
3447 | if (!t) | |
3448 | goto unlock; | |
3449 | ||
ce00a967 | 3450 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3451 | |
3452 | /* | |
748dad36 | 3453 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3454 | * If it's not true, a threshold was crossed after last |
3455 | * call of __mem_cgroup_threshold(). | |
3456 | */ | |
5407a562 | 3457 | i = t->current_threshold; |
2e72b634 KS |
3458 | |
3459 | /* | |
3460 | * Iterate backward over array of thresholds starting from | |
3461 | * current_threshold and check if a threshold is crossed. | |
3462 | * If none of thresholds below usage is crossed, we read | |
3463 | * only one element of the array here. | |
3464 | */ | |
3465 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3466 | eventfd_signal(t->entries[i].eventfd, 1); | |
3467 | ||
3468 | /* i = current_threshold + 1 */ | |
3469 | i++; | |
3470 | ||
3471 | /* | |
3472 | * Iterate forward over array of thresholds starting from | |
3473 | * current_threshold+1 and check if a threshold is crossed. | |
3474 | * If none of thresholds above usage is crossed, we read | |
3475 | * only one element of the array here. | |
3476 | */ | |
3477 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3478 | eventfd_signal(t->entries[i].eventfd, 1); | |
3479 | ||
3480 | /* Update current_threshold */ | |
5407a562 | 3481 | t->current_threshold = i - 1; |
2e72b634 KS |
3482 | unlock: |
3483 | rcu_read_unlock(); | |
3484 | } | |
3485 | ||
3486 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3487 | { | |
ad4ca5f4 KS |
3488 | while (memcg) { |
3489 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 3490 | if (do_memsw_account()) |
ad4ca5f4 KS |
3491 | __mem_cgroup_threshold(memcg, true); |
3492 | ||
3493 | memcg = parent_mem_cgroup(memcg); | |
3494 | } | |
2e72b634 KS |
3495 | } |
3496 | ||
3497 | static int compare_thresholds(const void *a, const void *b) | |
3498 | { | |
3499 | const struct mem_cgroup_threshold *_a = a; | |
3500 | const struct mem_cgroup_threshold *_b = b; | |
3501 | ||
2bff24a3 GT |
3502 | if (_a->threshold > _b->threshold) |
3503 | return 1; | |
3504 | ||
3505 | if (_a->threshold < _b->threshold) | |
3506 | return -1; | |
3507 | ||
3508 | return 0; | |
2e72b634 KS |
3509 | } |
3510 | ||
c0ff4b85 | 3511 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3512 | { |
3513 | struct mem_cgroup_eventfd_list *ev; | |
3514 | ||
2bcf2e92 MH |
3515 | spin_lock(&memcg_oom_lock); |
3516 | ||
c0ff4b85 | 3517 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3518 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3519 | |
3520 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3521 | return 0; |
3522 | } | |
3523 | ||
c0ff4b85 | 3524 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3525 | { |
7d74b06f KH |
3526 | struct mem_cgroup *iter; |
3527 | ||
c0ff4b85 | 3528 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3529 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3530 | } |
3531 | ||
59b6f873 | 3532 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3533 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3534 | { |
2c488db2 KS |
3535 | struct mem_cgroup_thresholds *thresholds; |
3536 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3537 | unsigned long threshold; |
3538 | unsigned long usage; | |
2c488db2 | 3539 | int i, size, ret; |
2e72b634 | 3540 | |
650c5e56 | 3541 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
3542 | if (ret) |
3543 | return ret; | |
3544 | ||
3545 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3546 | |
05b84301 | 3547 | if (type == _MEM) { |
2c488db2 | 3548 | thresholds = &memcg->thresholds; |
ce00a967 | 3549 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3550 | } else if (type == _MEMSWAP) { |
2c488db2 | 3551 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3552 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3553 | } else |
2e72b634 KS |
3554 | BUG(); |
3555 | ||
2e72b634 | 3556 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3557 | if (thresholds->primary) |
2e72b634 KS |
3558 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3559 | ||
2c488db2 | 3560 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3561 | |
3562 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 3563 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 3564 | GFP_KERNEL); |
2c488db2 | 3565 | if (!new) { |
2e72b634 KS |
3566 | ret = -ENOMEM; |
3567 | goto unlock; | |
3568 | } | |
2c488db2 | 3569 | new->size = size; |
2e72b634 KS |
3570 | |
3571 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3572 | if (thresholds->primary) { |
3573 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3574 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3575 | } |
3576 | ||
2e72b634 | 3577 | /* Add new threshold */ |
2c488db2 KS |
3578 | new->entries[size - 1].eventfd = eventfd; |
3579 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3580 | |
3581 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3582 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3583 | compare_thresholds, NULL); |
3584 | ||
3585 | /* Find current threshold */ | |
2c488db2 | 3586 | new->current_threshold = -1; |
2e72b634 | 3587 | for (i = 0; i < size; i++) { |
748dad36 | 3588 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3589 | /* |
2c488db2 KS |
3590 | * new->current_threshold will not be used until |
3591 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3592 | * it here. |
3593 | */ | |
2c488db2 | 3594 | ++new->current_threshold; |
748dad36 SZ |
3595 | } else |
3596 | break; | |
2e72b634 KS |
3597 | } |
3598 | ||
2c488db2 KS |
3599 | /* Free old spare buffer and save old primary buffer as spare */ |
3600 | kfree(thresholds->spare); | |
3601 | thresholds->spare = thresholds->primary; | |
3602 | ||
3603 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3604 | |
907860ed | 3605 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3606 | synchronize_rcu(); |
3607 | ||
2e72b634 KS |
3608 | unlock: |
3609 | mutex_unlock(&memcg->thresholds_lock); | |
3610 | ||
3611 | return ret; | |
3612 | } | |
3613 | ||
59b6f873 | 3614 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3615 | struct eventfd_ctx *eventfd, const char *args) |
3616 | { | |
59b6f873 | 3617 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3618 | } |
3619 | ||
59b6f873 | 3620 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3621 | struct eventfd_ctx *eventfd, const char *args) |
3622 | { | |
59b6f873 | 3623 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3624 | } |
3625 | ||
59b6f873 | 3626 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3627 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3628 | { |
2c488db2 KS |
3629 | struct mem_cgroup_thresholds *thresholds; |
3630 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3631 | unsigned long usage; |
2c488db2 | 3632 | int i, j, size; |
2e72b634 KS |
3633 | |
3634 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3635 | |
3636 | if (type == _MEM) { | |
2c488db2 | 3637 | thresholds = &memcg->thresholds; |
ce00a967 | 3638 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3639 | } else if (type == _MEMSWAP) { |
2c488db2 | 3640 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3641 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3642 | } else |
2e72b634 KS |
3643 | BUG(); |
3644 | ||
371528ca AV |
3645 | if (!thresholds->primary) |
3646 | goto unlock; | |
3647 | ||
2e72b634 KS |
3648 | /* Check if a threshold crossed before removing */ |
3649 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3650 | ||
3651 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3652 | size = 0; |
3653 | for (i = 0; i < thresholds->primary->size; i++) { | |
3654 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3655 | size++; |
3656 | } | |
3657 | ||
2c488db2 | 3658 | new = thresholds->spare; |
907860ed | 3659 | |
2e72b634 KS |
3660 | /* Set thresholds array to NULL if we don't have thresholds */ |
3661 | if (!size) { | |
2c488db2 KS |
3662 | kfree(new); |
3663 | new = NULL; | |
907860ed | 3664 | goto swap_buffers; |
2e72b634 KS |
3665 | } |
3666 | ||
2c488db2 | 3667 | new->size = size; |
2e72b634 KS |
3668 | |
3669 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3670 | new->current_threshold = -1; |
3671 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3672 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3673 | continue; |
3674 | ||
2c488db2 | 3675 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3676 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3677 | /* |
2c488db2 | 3678 | * new->current_threshold will not be used |
2e72b634 KS |
3679 | * until rcu_assign_pointer(), so it's safe to increment |
3680 | * it here. | |
3681 | */ | |
2c488db2 | 3682 | ++new->current_threshold; |
2e72b634 KS |
3683 | } |
3684 | j++; | |
3685 | } | |
3686 | ||
907860ed | 3687 | swap_buffers: |
2c488db2 KS |
3688 | /* Swap primary and spare array */ |
3689 | thresholds->spare = thresholds->primary; | |
8c757763 | 3690 | |
2c488db2 | 3691 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3692 | |
907860ed | 3693 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3694 | synchronize_rcu(); |
6611d8d7 MC |
3695 | |
3696 | /* If all events are unregistered, free the spare array */ | |
3697 | if (!new) { | |
3698 | kfree(thresholds->spare); | |
3699 | thresholds->spare = NULL; | |
3700 | } | |
371528ca | 3701 | unlock: |
2e72b634 | 3702 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3703 | } |
c1e862c1 | 3704 | |
59b6f873 | 3705 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3706 | struct eventfd_ctx *eventfd) |
3707 | { | |
59b6f873 | 3708 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
3709 | } |
3710 | ||
59b6f873 | 3711 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3712 | struct eventfd_ctx *eventfd) |
3713 | { | |
59b6f873 | 3714 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
3715 | } |
3716 | ||
59b6f873 | 3717 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3718 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 3719 | { |
9490ff27 | 3720 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 3721 | |
9490ff27 KH |
3722 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
3723 | if (!event) | |
3724 | return -ENOMEM; | |
3725 | ||
1af8efe9 | 3726 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
3727 | |
3728 | event->eventfd = eventfd; | |
3729 | list_add(&event->list, &memcg->oom_notify); | |
3730 | ||
3731 | /* already in OOM ? */ | |
c2b42d3c | 3732 | if (memcg->under_oom) |
9490ff27 | 3733 | eventfd_signal(eventfd, 1); |
1af8efe9 | 3734 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3735 | |
3736 | return 0; | |
3737 | } | |
3738 | ||
59b6f873 | 3739 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3740 | struct eventfd_ctx *eventfd) |
9490ff27 | 3741 | { |
9490ff27 | 3742 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 3743 | |
1af8efe9 | 3744 | spin_lock(&memcg_oom_lock); |
9490ff27 | 3745 | |
c0ff4b85 | 3746 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
3747 | if (ev->eventfd == eventfd) { |
3748 | list_del(&ev->list); | |
3749 | kfree(ev); | |
3750 | } | |
3751 | } | |
3752 | ||
1af8efe9 | 3753 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3754 | } |
3755 | ||
2da8ca82 | 3756 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 3757 | { |
2da8ca82 | 3758 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 3759 | |
791badbd | 3760 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 3761 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
3762 | seq_printf(sf, "oom_kill %lu\n", |
3763 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
3764 | return 0; |
3765 | } | |
3766 | ||
182446d0 | 3767 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
3768 | struct cftype *cft, u64 val) |
3769 | { | |
182446d0 | 3770 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
3771 | |
3772 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 3773 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
3774 | return -EINVAL; |
3775 | ||
c0ff4b85 | 3776 | memcg->oom_kill_disable = val; |
4d845ebf | 3777 | if (!val) |
c0ff4b85 | 3778 | memcg_oom_recover(memcg); |
3dae7fec | 3779 | |
3c11ecf4 KH |
3780 | return 0; |
3781 | } | |
3782 | ||
52ebea74 TH |
3783 | #ifdef CONFIG_CGROUP_WRITEBACK |
3784 | ||
841710aa TH |
3785 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
3786 | { | |
3787 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
3788 | } | |
3789 | ||
3790 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3791 | { | |
3792 | wb_domain_exit(&memcg->cgwb_domain); | |
3793 | } | |
3794 | ||
2529bb3a TH |
3795 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3796 | { | |
3797 | wb_domain_size_changed(&memcg->cgwb_domain); | |
3798 | } | |
3799 | ||
841710aa TH |
3800 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
3801 | { | |
3802 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3803 | ||
3804 | if (!memcg->css.parent) | |
3805 | return NULL; | |
3806 | ||
3807 | return &memcg->cgwb_domain; | |
3808 | } | |
3809 | ||
c2aa723a TH |
3810 | /** |
3811 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
3812 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
3813 | * @pfilepages: out parameter for number of file pages |
3814 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
3815 | * @pdirty: out parameter for number of dirty pages |
3816 | * @pwriteback: out parameter for number of pages under writeback | |
3817 | * | |
c5edf9cd TH |
3818 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
3819 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
3820 | * is a bit more involved. | |
c2aa723a | 3821 | * |
c5edf9cd TH |
3822 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
3823 | * headroom is calculated as the lowest headroom of itself and the | |
3824 | * ancestors. Note that this doesn't consider the actual amount of | |
3825 | * available memory in the system. The caller should further cap | |
3826 | * *@pheadroom accordingly. | |
c2aa723a | 3827 | */ |
c5edf9cd TH |
3828 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
3829 | unsigned long *pheadroom, unsigned long *pdirty, | |
3830 | unsigned long *pwriteback) | |
c2aa723a TH |
3831 | { |
3832 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3833 | struct mem_cgroup *parent; | |
c2aa723a | 3834 | |
ccda7f43 | 3835 | *pdirty = memcg_page_state(memcg, NR_FILE_DIRTY); |
c2aa723a TH |
3836 | |
3837 | /* this should eventually include NR_UNSTABLE_NFS */ | |
ccda7f43 | 3838 | *pwriteback = memcg_page_state(memcg, NR_WRITEBACK); |
c5edf9cd TH |
3839 | *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) | |
3840 | (1 << LRU_ACTIVE_FILE)); | |
3841 | *pheadroom = PAGE_COUNTER_MAX; | |
c2aa723a | 3842 | |
c2aa723a | 3843 | while ((parent = parent_mem_cgroup(memcg))) { |
bbec2e15 | 3844 | unsigned long ceiling = min(memcg->memory.max, memcg->high); |
c2aa723a TH |
3845 | unsigned long used = page_counter_read(&memcg->memory); |
3846 | ||
c5edf9cd | 3847 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
3848 | memcg = parent; |
3849 | } | |
c2aa723a TH |
3850 | } |
3851 | ||
841710aa TH |
3852 | #else /* CONFIG_CGROUP_WRITEBACK */ |
3853 | ||
3854 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
3855 | { | |
3856 | return 0; | |
3857 | } | |
3858 | ||
3859 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3860 | { | |
3861 | } | |
3862 | ||
2529bb3a TH |
3863 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3864 | { | |
3865 | } | |
3866 | ||
52ebea74 TH |
3867 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
3868 | ||
3bc942f3 TH |
3869 | /* |
3870 | * DO NOT USE IN NEW FILES. | |
3871 | * | |
3872 | * "cgroup.event_control" implementation. | |
3873 | * | |
3874 | * This is way over-engineered. It tries to support fully configurable | |
3875 | * events for each user. Such level of flexibility is completely | |
3876 | * unnecessary especially in the light of the planned unified hierarchy. | |
3877 | * | |
3878 | * Please deprecate this and replace with something simpler if at all | |
3879 | * possible. | |
3880 | */ | |
3881 | ||
79bd9814 TH |
3882 | /* |
3883 | * Unregister event and free resources. | |
3884 | * | |
3885 | * Gets called from workqueue. | |
3886 | */ | |
3bc942f3 | 3887 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 3888 | { |
3bc942f3 TH |
3889 | struct mem_cgroup_event *event = |
3890 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 3891 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3892 | |
3893 | remove_wait_queue(event->wqh, &event->wait); | |
3894 | ||
59b6f873 | 3895 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
3896 | |
3897 | /* Notify userspace the event is going away. */ | |
3898 | eventfd_signal(event->eventfd, 1); | |
3899 | ||
3900 | eventfd_ctx_put(event->eventfd); | |
3901 | kfree(event); | |
59b6f873 | 3902 | css_put(&memcg->css); |
79bd9814 TH |
3903 | } |
3904 | ||
3905 | /* | |
a9a08845 | 3906 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
3907 | * |
3908 | * Called with wqh->lock held and interrupts disabled. | |
3909 | */ | |
ac6424b9 | 3910 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 3911 | int sync, void *key) |
79bd9814 | 3912 | { |
3bc942f3 TH |
3913 | struct mem_cgroup_event *event = |
3914 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 3915 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 3916 | __poll_t flags = key_to_poll(key); |
79bd9814 | 3917 | |
a9a08845 | 3918 | if (flags & EPOLLHUP) { |
79bd9814 TH |
3919 | /* |
3920 | * If the event has been detached at cgroup removal, we | |
3921 | * can simply return knowing the other side will cleanup | |
3922 | * for us. | |
3923 | * | |
3924 | * We can't race against event freeing since the other | |
3925 | * side will require wqh->lock via remove_wait_queue(), | |
3926 | * which we hold. | |
3927 | */ | |
fba94807 | 3928 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
3929 | if (!list_empty(&event->list)) { |
3930 | list_del_init(&event->list); | |
3931 | /* | |
3932 | * We are in atomic context, but cgroup_event_remove() | |
3933 | * may sleep, so we have to call it in workqueue. | |
3934 | */ | |
3935 | schedule_work(&event->remove); | |
3936 | } | |
fba94807 | 3937 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
3938 | } |
3939 | ||
3940 | return 0; | |
3941 | } | |
3942 | ||
3bc942f3 | 3943 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
3944 | wait_queue_head_t *wqh, poll_table *pt) |
3945 | { | |
3bc942f3 TH |
3946 | struct mem_cgroup_event *event = |
3947 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
3948 | |
3949 | event->wqh = wqh; | |
3950 | add_wait_queue(wqh, &event->wait); | |
3951 | } | |
3952 | ||
3953 | /* | |
3bc942f3 TH |
3954 | * DO NOT USE IN NEW FILES. |
3955 | * | |
79bd9814 TH |
3956 | * Parse input and register new cgroup event handler. |
3957 | * | |
3958 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
3959 | * Interpretation of args is defined by control file implementation. | |
3960 | */ | |
451af504 TH |
3961 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
3962 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 3963 | { |
451af504 | 3964 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 3965 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 3966 | struct mem_cgroup_event *event; |
79bd9814 TH |
3967 | struct cgroup_subsys_state *cfile_css; |
3968 | unsigned int efd, cfd; | |
3969 | struct fd efile; | |
3970 | struct fd cfile; | |
fba94807 | 3971 | const char *name; |
79bd9814 TH |
3972 | char *endp; |
3973 | int ret; | |
3974 | ||
451af504 TH |
3975 | buf = strstrip(buf); |
3976 | ||
3977 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
3978 | if (*endp != ' ') |
3979 | return -EINVAL; | |
451af504 | 3980 | buf = endp + 1; |
79bd9814 | 3981 | |
451af504 | 3982 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
3983 | if ((*endp != ' ') && (*endp != '\0')) |
3984 | return -EINVAL; | |
451af504 | 3985 | buf = endp + 1; |
79bd9814 TH |
3986 | |
3987 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
3988 | if (!event) | |
3989 | return -ENOMEM; | |
3990 | ||
59b6f873 | 3991 | event->memcg = memcg; |
79bd9814 | 3992 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
3993 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
3994 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
3995 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
3996 | |
3997 | efile = fdget(efd); | |
3998 | if (!efile.file) { | |
3999 | ret = -EBADF; | |
4000 | goto out_kfree; | |
4001 | } | |
4002 | ||
4003 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4004 | if (IS_ERR(event->eventfd)) { | |
4005 | ret = PTR_ERR(event->eventfd); | |
4006 | goto out_put_efile; | |
4007 | } | |
4008 | ||
4009 | cfile = fdget(cfd); | |
4010 | if (!cfile.file) { | |
4011 | ret = -EBADF; | |
4012 | goto out_put_eventfd; | |
4013 | } | |
4014 | ||
4015 | /* the process need read permission on control file */ | |
4016 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
4017 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
4018 | if (ret < 0) | |
4019 | goto out_put_cfile; | |
4020 | ||
fba94807 TH |
4021 | /* |
4022 | * Determine the event callbacks and set them in @event. This used | |
4023 | * to be done via struct cftype but cgroup core no longer knows | |
4024 | * about these events. The following is crude but the whole thing | |
4025 | * is for compatibility anyway. | |
3bc942f3 TH |
4026 | * |
4027 | * DO NOT ADD NEW FILES. | |
fba94807 | 4028 | */ |
b583043e | 4029 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4030 | |
4031 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4032 | event->register_event = mem_cgroup_usage_register_event; | |
4033 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4034 | } else if (!strcmp(name, "memory.oom_control")) { | |
4035 | event->register_event = mem_cgroup_oom_register_event; | |
4036 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4037 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4038 | event->register_event = vmpressure_register_event; | |
4039 | event->unregister_event = vmpressure_unregister_event; | |
4040 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4041 | event->register_event = memsw_cgroup_usage_register_event; |
4042 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4043 | } else { |
4044 | ret = -EINVAL; | |
4045 | goto out_put_cfile; | |
4046 | } | |
4047 | ||
79bd9814 | 4048 | /* |
b5557c4c TH |
4049 | * Verify @cfile should belong to @css. Also, remaining events are |
4050 | * automatically removed on cgroup destruction but the removal is | |
4051 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4052 | */ |
b583043e | 4053 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4054 | &memory_cgrp_subsys); |
79bd9814 | 4055 | ret = -EINVAL; |
5a17f543 | 4056 | if (IS_ERR(cfile_css)) |
79bd9814 | 4057 | goto out_put_cfile; |
5a17f543 TH |
4058 | if (cfile_css != css) { |
4059 | css_put(cfile_css); | |
79bd9814 | 4060 | goto out_put_cfile; |
5a17f543 | 4061 | } |
79bd9814 | 4062 | |
451af504 | 4063 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4064 | if (ret) |
4065 | goto out_put_css; | |
4066 | ||
9965ed17 | 4067 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 4068 | |
fba94807 TH |
4069 | spin_lock(&memcg->event_list_lock); |
4070 | list_add(&event->list, &memcg->event_list); | |
4071 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4072 | |
4073 | fdput(cfile); | |
4074 | fdput(efile); | |
4075 | ||
451af504 | 4076 | return nbytes; |
79bd9814 TH |
4077 | |
4078 | out_put_css: | |
b5557c4c | 4079 | css_put(css); |
79bd9814 TH |
4080 | out_put_cfile: |
4081 | fdput(cfile); | |
4082 | out_put_eventfd: | |
4083 | eventfd_ctx_put(event->eventfd); | |
4084 | out_put_efile: | |
4085 | fdput(efile); | |
4086 | out_kfree: | |
4087 | kfree(event); | |
4088 | ||
4089 | return ret; | |
4090 | } | |
4091 | ||
241994ed | 4092 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4093 | { |
0eea1030 | 4094 | .name = "usage_in_bytes", |
8c7c6e34 | 4095 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4096 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4097 | }, |
c84872e1 PE |
4098 | { |
4099 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4100 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4101 | .write = mem_cgroup_reset, |
791badbd | 4102 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4103 | }, |
8cdea7c0 | 4104 | { |
0eea1030 | 4105 | .name = "limit_in_bytes", |
8c7c6e34 | 4106 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4107 | .write = mem_cgroup_write, |
791badbd | 4108 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4109 | }, |
296c81d8 BS |
4110 | { |
4111 | .name = "soft_limit_in_bytes", | |
4112 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4113 | .write = mem_cgroup_write, |
791badbd | 4114 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4115 | }, |
8cdea7c0 BS |
4116 | { |
4117 | .name = "failcnt", | |
8c7c6e34 | 4118 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4119 | .write = mem_cgroup_reset, |
791badbd | 4120 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4121 | }, |
d2ceb9b7 KH |
4122 | { |
4123 | .name = "stat", | |
2da8ca82 | 4124 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4125 | }, |
c1e862c1 KH |
4126 | { |
4127 | .name = "force_empty", | |
6770c64e | 4128 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4129 | }, |
18f59ea7 BS |
4130 | { |
4131 | .name = "use_hierarchy", | |
4132 | .write_u64 = mem_cgroup_hierarchy_write, | |
4133 | .read_u64 = mem_cgroup_hierarchy_read, | |
4134 | }, | |
79bd9814 | 4135 | { |
3bc942f3 | 4136 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4137 | .write = memcg_write_event_control, |
7dbdb199 | 4138 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 4139 | }, |
a7885eb8 KM |
4140 | { |
4141 | .name = "swappiness", | |
4142 | .read_u64 = mem_cgroup_swappiness_read, | |
4143 | .write_u64 = mem_cgroup_swappiness_write, | |
4144 | }, | |
7dc74be0 DN |
4145 | { |
4146 | .name = "move_charge_at_immigrate", | |
4147 | .read_u64 = mem_cgroup_move_charge_read, | |
4148 | .write_u64 = mem_cgroup_move_charge_write, | |
4149 | }, | |
9490ff27 KH |
4150 | { |
4151 | .name = "oom_control", | |
2da8ca82 | 4152 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4153 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4154 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4155 | }, | |
70ddf637 AV |
4156 | { |
4157 | .name = "pressure_level", | |
70ddf637 | 4158 | }, |
406eb0c9 YH |
4159 | #ifdef CONFIG_NUMA |
4160 | { | |
4161 | .name = "numa_stat", | |
2da8ca82 | 4162 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4163 | }, |
4164 | #endif | |
510fc4e1 GC |
4165 | { |
4166 | .name = "kmem.limit_in_bytes", | |
4167 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4168 | .write = mem_cgroup_write, |
791badbd | 4169 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4170 | }, |
4171 | { | |
4172 | .name = "kmem.usage_in_bytes", | |
4173 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4174 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4175 | }, |
4176 | { | |
4177 | .name = "kmem.failcnt", | |
4178 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4179 | .write = mem_cgroup_reset, |
791badbd | 4180 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4181 | }, |
4182 | { | |
4183 | .name = "kmem.max_usage_in_bytes", | |
4184 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4185 | .write = mem_cgroup_reset, |
791badbd | 4186 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4187 | }, |
5b365771 | 4188 | #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) |
749c5415 GC |
4189 | { |
4190 | .name = "kmem.slabinfo", | |
bc2791f8 TH |
4191 | .seq_start = memcg_slab_start, |
4192 | .seq_next = memcg_slab_next, | |
4193 | .seq_stop = memcg_slab_stop, | |
b047501c | 4194 | .seq_show = memcg_slab_show, |
749c5415 GC |
4195 | }, |
4196 | #endif | |
d55f90bf VD |
4197 | { |
4198 | .name = "kmem.tcp.limit_in_bytes", | |
4199 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
4200 | .write = mem_cgroup_write, | |
4201 | .read_u64 = mem_cgroup_read_u64, | |
4202 | }, | |
4203 | { | |
4204 | .name = "kmem.tcp.usage_in_bytes", | |
4205 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
4206 | .read_u64 = mem_cgroup_read_u64, | |
4207 | }, | |
4208 | { | |
4209 | .name = "kmem.tcp.failcnt", | |
4210 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
4211 | .write = mem_cgroup_reset, | |
4212 | .read_u64 = mem_cgroup_read_u64, | |
4213 | }, | |
4214 | { | |
4215 | .name = "kmem.tcp.max_usage_in_bytes", | |
4216 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
4217 | .write = mem_cgroup_reset, | |
4218 | .read_u64 = mem_cgroup_read_u64, | |
4219 | }, | |
6bc10349 | 4220 | { }, /* terminate */ |
af36f906 | 4221 | }; |
8c7c6e34 | 4222 | |
73f576c0 JW |
4223 | /* |
4224 | * Private memory cgroup IDR | |
4225 | * | |
4226 | * Swap-out records and page cache shadow entries need to store memcg | |
4227 | * references in constrained space, so we maintain an ID space that is | |
4228 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
4229 | * memory-controlled cgroups to 64k. | |
4230 | * | |
4231 | * However, there usually are many references to the oflline CSS after | |
4232 | * the cgroup has been destroyed, such as page cache or reclaimable | |
4233 | * slab objects, that don't need to hang on to the ID. We want to keep | |
4234 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
4235 | * relatively small ID space and prevent the creation of new cgroups | |
4236 | * even when there are much fewer than 64k cgroups - possibly none. | |
4237 | * | |
4238 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
4239 | * be freed and recycled when it's no longer needed, which is usually | |
4240 | * when the CSS is offlined. | |
4241 | * | |
4242 | * The only exception to that are records of swapped out tmpfs/shmem | |
4243 | * pages that need to be attributed to live ancestors on swapin. But | |
4244 | * those references are manageable from userspace. | |
4245 | */ | |
4246 | ||
4247 | static DEFINE_IDR(mem_cgroup_idr); | |
4248 | ||
7e97de0b KT |
4249 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
4250 | { | |
4251 | if (memcg->id.id > 0) { | |
4252 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
4253 | memcg->id.id = 0; | |
4254 | } | |
4255 | } | |
4256 | ||
615d66c3 | 4257 | static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4258 | { |
58fa2a55 | 4259 | VM_BUG_ON(atomic_read(&memcg->id.ref) <= 0); |
615d66c3 | 4260 | atomic_add(n, &memcg->id.ref); |
73f576c0 JW |
4261 | } |
4262 | ||
615d66c3 | 4263 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4264 | { |
58fa2a55 | 4265 | VM_BUG_ON(atomic_read(&memcg->id.ref) < n); |
615d66c3 | 4266 | if (atomic_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 4267 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
4268 | |
4269 | /* Memcg ID pins CSS */ | |
4270 | css_put(&memcg->css); | |
4271 | } | |
4272 | } | |
4273 | ||
615d66c3 VD |
4274 | static inline void mem_cgroup_id_get(struct mem_cgroup *memcg) |
4275 | { | |
4276 | mem_cgroup_id_get_many(memcg, 1); | |
4277 | } | |
4278 | ||
4279 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) | |
4280 | { | |
4281 | mem_cgroup_id_put_many(memcg, 1); | |
4282 | } | |
4283 | ||
73f576c0 JW |
4284 | /** |
4285 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
4286 | * @id: the memcg id to look up | |
4287 | * | |
4288 | * Caller must hold rcu_read_lock(). | |
4289 | */ | |
4290 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
4291 | { | |
4292 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
4293 | return idr_find(&mem_cgroup_idr, id); | |
4294 | } | |
4295 | ||
ef8f2327 | 4296 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4297 | { |
4298 | struct mem_cgroup_per_node *pn; | |
ef8f2327 | 4299 | int tmp = node; |
1ecaab2b KH |
4300 | /* |
4301 | * This routine is called against possible nodes. | |
4302 | * But it's BUG to call kmalloc() against offline node. | |
4303 | * | |
4304 | * TODO: this routine can waste much memory for nodes which will | |
4305 | * never be onlined. It's better to use memory hotplug callback | |
4306 | * function. | |
4307 | */ | |
41e3355d KH |
4308 | if (!node_state(node, N_NORMAL_MEMORY)) |
4309 | tmp = -1; | |
17295c88 | 4310 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4311 | if (!pn) |
4312 | return 1; | |
1ecaab2b | 4313 | |
a983b5eb JW |
4314 | pn->lruvec_stat_cpu = alloc_percpu(struct lruvec_stat); |
4315 | if (!pn->lruvec_stat_cpu) { | |
00f3ca2c JW |
4316 | kfree(pn); |
4317 | return 1; | |
4318 | } | |
4319 | ||
ef8f2327 MG |
4320 | lruvec_init(&pn->lruvec); |
4321 | pn->usage_in_excess = 0; | |
4322 | pn->on_tree = false; | |
4323 | pn->memcg = memcg; | |
4324 | ||
54f72fe0 | 4325 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4326 | return 0; |
4327 | } | |
4328 | ||
ef8f2327 | 4329 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4330 | { |
00f3ca2c JW |
4331 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
4332 | ||
4eaf431f MH |
4333 | if (!pn) |
4334 | return; | |
4335 | ||
a983b5eb | 4336 | free_percpu(pn->lruvec_stat_cpu); |
00f3ca2c | 4337 | kfree(pn); |
1ecaab2b KH |
4338 | } |
4339 | ||
40e952f9 | 4340 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 4341 | { |
c8b2a36f | 4342 | int node; |
59927fb9 | 4343 | |
c8b2a36f | 4344 | for_each_node(node) |
ef8f2327 | 4345 | free_mem_cgroup_per_node_info(memcg, node); |
a983b5eb | 4346 | free_percpu(memcg->stat_cpu); |
8ff69e2c | 4347 | kfree(memcg); |
59927fb9 | 4348 | } |
3afe36b1 | 4349 | |
40e952f9 TE |
4350 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
4351 | { | |
4352 | memcg_wb_domain_exit(memcg); | |
4353 | __mem_cgroup_free(memcg); | |
4354 | } | |
4355 | ||
0b8f73e1 | 4356 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 4357 | { |
d142e3e6 | 4358 | struct mem_cgroup *memcg; |
0b8f73e1 | 4359 | size_t size; |
6d12e2d8 | 4360 | int node; |
8cdea7c0 | 4361 | |
0b8f73e1 JW |
4362 | size = sizeof(struct mem_cgroup); |
4363 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
4364 | ||
4365 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 4366 | if (!memcg) |
0b8f73e1 JW |
4367 | return NULL; |
4368 | ||
73f576c0 JW |
4369 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
4370 | 1, MEM_CGROUP_ID_MAX, | |
4371 | GFP_KERNEL); | |
4372 | if (memcg->id.id < 0) | |
4373 | goto fail; | |
4374 | ||
a983b5eb JW |
4375 | memcg->stat_cpu = alloc_percpu(struct mem_cgroup_stat_cpu); |
4376 | if (!memcg->stat_cpu) | |
0b8f73e1 | 4377 | goto fail; |
78fb7466 | 4378 | |
3ed28fa1 | 4379 | for_each_node(node) |
ef8f2327 | 4380 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 4381 | goto fail; |
f64c3f54 | 4382 | |
0b8f73e1 JW |
4383 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
4384 | goto fail; | |
28dbc4b6 | 4385 | |
f7e1cb6e | 4386 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 GC |
4387 | memcg->last_scanned_node = MAX_NUMNODES; |
4388 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4389 | mutex_init(&memcg->thresholds_lock); |
4390 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4391 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4392 | INIT_LIST_HEAD(&memcg->event_list); |
4393 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 4394 | memcg->socket_pressure = jiffies; |
84c07d11 | 4395 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 4396 | memcg->kmemcg_id = -1; |
900a38f0 | 4397 | #endif |
52ebea74 TH |
4398 | #ifdef CONFIG_CGROUP_WRITEBACK |
4399 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
4400 | #endif | |
73f576c0 | 4401 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
4402 | return memcg; |
4403 | fail: | |
7e97de0b | 4404 | mem_cgroup_id_remove(memcg); |
40e952f9 | 4405 | __mem_cgroup_free(memcg); |
0b8f73e1 | 4406 | return NULL; |
d142e3e6 GC |
4407 | } |
4408 | ||
0b8f73e1 JW |
4409 | static struct cgroup_subsys_state * __ref |
4410 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 4411 | { |
0b8f73e1 JW |
4412 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
4413 | struct mem_cgroup *memcg; | |
4414 | long error = -ENOMEM; | |
d142e3e6 | 4415 | |
0b8f73e1 JW |
4416 | memcg = mem_cgroup_alloc(); |
4417 | if (!memcg) | |
4418 | return ERR_PTR(error); | |
d142e3e6 | 4419 | |
0b8f73e1 JW |
4420 | memcg->high = PAGE_COUNTER_MAX; |
4421 | memcg->soft_limit = PAGE_COUNTER_MAX; | |
4422 | if (parent) { | |
4423 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4424 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4425 | } | |
4426 | if (parent && parent->use_hierarchy) { | |
4427 | memcg->use_hierarchy = true; | |
3e32cb2e | 4428 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 4429 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e JW |
4430 | page_counter_init(&memcg->memsw, &parent->memsw); |
4431 | page_counter_init(&memcg->kmem, &parent->kmem); | |
0db15298 | 4432 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 4433 | } else { |
3e32cb2e | 4434 | page_counter_init(&memcg->memory, NULL); |
37e84351 | 4435 | page_counter_init(&memcg->swap, NULL); |
3e32cb2e JW |
4436 | page_counter_init(&memcg->memsw, NULL); |
4437 | page_counter_init(&memcg->kmem, NULL); | |
0db15298 | 4438 | page_counter_init(&memcg->tcpmem, NULL); |
8c7f6edb TH |
4439 | /* |
4440 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4441 | * much sense so let cgroup subsystem know about this | |
4442 | * unfortunate state in our controller. | |
4443 | */ | |
d142e3e6 | 4444 | if (parent != root_mem_cgroup) |
073219e9 | 4445 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4446 | } |
d6441637 | 4447 | |
0b8f73e1 JW |
4448 | /* The following stuff does not apply to the root */ |
4449 | if (!parent) { | |
4450 | root_mem_cgroup = memcg; | |
4451 | return &memcg->css; | |
4452 | } | |
4453 | ||
b313aeee | 4454 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
4455 | if (error) |
4456 | goto fail; | |
127424c8 | 4457 | |
f7e1cb6e | 4458 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4459 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 4460 | |
0b8f73e1 JW |
4461 | return &memcg->css; |
4462 | fail: | |
7e97de0b | 4463 | mem_cgroup_id_remove(memcg); |
0b8f73e1 | 4464 | mem_cgroup_free(memcg); |
ea3a9645 | 4465 | return ERR_PTR(-ENOMEM); |
0b8f73e1 JW |
4466 | } |
4467 | ||
73f576c0 | 4468 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 4469 | { |
58fa2a55 VD |
4470 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
4471 | ||
0a4465d3 KT |
4472 | /* |
4473 | * A memcg must be visible for memcg_expand_shrinker_maps() | |
4474 | * by the time the maps are allocated. So, we allocate maps | |
4475 | * here, when for_each_mem_cgroup() can't skip it. | |
4476 | */ | |
4477 | if (memcg_alloc_shrinker_maps(memcg)) { | |
4478 | mem_cgroup_id_remove(memcg); | |
4479 | return -ENOMEM; | |
4480 | } | |
4481 | ||
73f576c0 | 4482 | /* Online state pins memcg ID, memcg ID pins CSS */ |
58fa2a55 | 4483 | atomic_set(&memcg->id.ref, 1); |
73f576c0 | 4484 | css_get(css); |
2f7dd7a4 | 4485 | return 0; |
8cdea7c0 BS |
4486 | } |
4487 | ||
eb95419b | 4488 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4489 | { |
eb95419b | 4490 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4491 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4492 | |
4493 | /* | |
4494 | * Unregister events and notify userspace. | |
4495 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4496 | * directory to avoid race between userspace and kernelspace. | |
4497 | */ | |
fba94807 TH |
4498 | spin_lock(&memcg->event_list_lock); |
4499 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4500 | list_del_init(&event->list); |
4501 | schedule_work(&event->remove); | |
4502 | } | |
fba94807 | 4503 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4504 | |
bf8d5d52 | 4505 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 4506 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 4507 | |
567e9ab2 | 4508 | memcg_offline_kmem(memcg); |
52ebea74 | 4509 | wb_memcg_offline(memcg); |
73f576c0 JW |
4510 | |
4511 | mem_cgroup_id_put(memcg); | |
df878fb0 KH |
4512 | } |
4513 | ||
6df38689 VD |
4514 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
4515 | { | |
4516 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4517 | ||
4518 | invalidate_reclaim_iterators(memcg); | |
4519 | } | |
4520 | ||
eb95419b | 4521 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4522 | { |
eb95419b | 4523 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4524 | |
f7e1cb6e | 4525 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4526 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 4527 | |
0db15298 | 4528 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 4529 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 4530 | |
0b8f73e1 JW |
4531 | vmpressure_cleanup(&memcg->vmpressure); |
4532 | cancel_work_sync(&memcg->high_work); | |
4533 | mem_cgroup_remove_from_trees(memcg); | |
0a4465d3 | 4534 | memcg_free_shrinker_maps(memcg); |
d886f4e4 | 4535 | memcg_free_kmem(memcg); |
0b8f73e1 | 4536 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
4537 | } |
4538 | ||
1ced953b TH |
4539 | /** |
4540 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4541 | * @css: the target css | |
4542 | * | |
4543 | * Reset the states of the mem_cgroup associated with @css. This is | |
4544 | * invoked when the userland requests disabling on the default hierarchy | |
4545 | * but the memcg is pinned through dependency. The memcg should stop | |
4546 | * applying policies and should revert to the vanilla state as it may be | |
4547 | * made visible again. | |
4548 | * | |
4549 | * The current implementation only resets the essential configurations. | |
4550 | * This needs to be expanded to cover all the visible parts. | |
4551 | */ | |
4552 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4553 | { | |
4554 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4555 | ||
bbec2e15 RG |
4556 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
4557 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
4558 | page_counter_set_max(&memcg->memsw, PAGE_COUNTER_MAX); | |
4559 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); | |
4560 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 4561 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 4562 | page_counter_set_low(&memcg->memory, 0); |
241994ed | 4563 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4564 | memcg->soft_limit = PAGE_COUNTER_MAX; |
2529bb3a | 4565 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
4566 | } |
4567 | ||
02491447 | 4568 | #ifdef CONFIG_MMU |
7dc74be0 | 4569 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4570 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4571 | { |
05b84301 | 4572 | int ret; |
9476db97 | 4573 | |
d0164adc MG |
4574 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
4575 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 4576 | if (!ret) { |
854ffa8d | 4577 | mc.precharge += count; |
854ffa8d DN |
4578 | return ret; |
4579 | } | |
9476db97 | 4580 | |
3674534b | 4581 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 4582 | while (count--) { |
3674534b | 4583 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 4584 | if (ret) |
38c5d72f | 4585 | return ret; |
854ffa8d | 4586 | mc.precharge++; |
9476db97 | 4587 | cond_resched(); |
854ffa8d | 4588 | } |
9476db97 | 4589 | return 0; |
4ffef5fe DN |
4590 | } |
4591 | ||
4ffef5fe DN |
4592 | union mc_target { |
4593 | struct page *page; | |
02491447 | 4594 | swp_entry_t ent; |
4ffef5fe DN |
4595 | }; |
4596 | ||
4ffef5fe | 4597 | enum mc_target_type { |
8d32ff84 | 4598 | MC_TARGET_NONE = 0, |
4ffef5fe | 4599 | MC_TARGET_PAGE, |
02491447 | 4600 | MC_TARGET_SWAP, |
c733a828 | 4601 | MC_TARGET_DEVICE, |
4ffef5fe DN |
4602 | }; |
4603 | ||
90254a65 DN |
4604 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4605 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4606 | { |
c733a828 | 4607 | struct page *page = _vm_normal_page(vma, addr, ptent, true); |
4ffef5fe | 4608 | |
90254a65 DN |
4609 | if (!page || !page_mapped(page)) |
4610 | return NULL; | |
4611 | if (PageAnon(page)) { | |
1dfab5ab | 4612 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 4613 | return NULL; |
1dfab5ab JW |
4614 | } else { |
4615 | if (!(mc.flags & MOVE_FILE)) | |
4616 | return NULL; | |
4617 | } | |
90254a65 DN |
4618 | if (!get_page_unless_zero(page)) |
4619 | return NULL; | |
4620 | ||
4621 | return page; | |
4622 | } | |
4623 | ||
c733a828 | 4624 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 4625 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 4626 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 4627 | { |
90254a65 DN |
4628 | struct page *page = NULL; |
4629 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4630 | ||
1dfab5ab | 4631 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 4632 | return NULL; |
c733a828 JG |
4633 | |
4634 | /* | |
4635 | * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to | |
4636 | * a device and because they are not accessible by CPU they are store | |
4637 | * as special swap entry in the CPU page table. | |
4638 | */ | |
4639 | if (is_device_private_entry(ent)) { | |
4640 | page = device_private_entry_to_page(ent); | |
4641 | /* | |
4642 | * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have | |
4643 | * a refcount of 1 when free (unlike normal page) | |
4644 | */ | |
4645 | if (!page_ref_add_unless(page, 1, 1)) | |
4646 | return NULL; | |
4647 | return page; | |
4648 | } | |
4649 | ||
4b91355e KH |
4650 | /* |
4651 | * Because lookup_swap_cache() updates some statistics counter, | |
4652 | * we call find_get_page() with swapper_space directly. | |
4653 | */ | |
f6ab1f7f | 4654 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
7941d214 | 4655 | if (do_memsw_account()) |
90254a65 DN |
4656 | entry->val = ent.val; |
4657 | ||
4658 | return page; | |
4659 | } | |
4b91355e KH |
4660 | #else |
4661 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 4662 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
4663 | { |
4664 | return NULL; | |
4665 | } | |
4666 | #endif | |
90254a65 | 4667 | |
87946a72 DN |
4668 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4669 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4670 | { | |
4671 | struct page *page = NULL; | |
87946a72 DN |
4672 | struct address_space *mapping; |
4673 | pgoff_t pgoff; | |
4674 | ||
4675 | if (!vma->vm_file) /* anonymous vma */ | |
4676 | return NULL; | |
1dfab5ab | 4677 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
4678 | return NULL; |
4679 | ||
87946a72 | 4680 | mapping = vma->vm_file->f_mapping; |
0661a336 | 4681 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
4682 | |
4683 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
4684 | #ifdef CONFIG_SWAP |
4685 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
4686 | if (shmem_mapping(mapping)) { |
4687 | page = find_get_entry(mapping, pgoff); | |
4688 | if (radix_tree_exceptional_entry(page)) { | |
4689 | swp_entry_t swp = radix_to_swp_entry(page); | |
7941d214 | 4690 | if (do_memsw_account()) |
139b6a6f | 4691 | *entry = swp; |
f6ab1f7f HY |
4692 | page = find_get_page(swap_address_space(swp), |
4693 | swp_offset(swp)); | |
139b6a6f JW |
4694 | } |
4695 | } else | |
4696 | page = find_get_page(mapping, pgoff); | |
4697 | #else | |
4698 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 4699 | #endif |
87946a72 DN |
4700 | return page; |
4701 | } | |
4702 | ||
b1b0deab CG |
4703 | /** |
4704 | * mem_cgroup_move_account - move account of the page | |
4705 | * @page: the page | |
25843c2b | 4706 | * @compound: charge the page as compound or small page |
b1b0deab CG |
4707 | * @from: mem_cgroup which the page is moved from. |
4708 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
4709 | * | |
3ac808fd | 4710 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
4711 | * |
4712 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
4713 | * from old cgroup. | |
4714 | */ | |
4715 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 4716 | bool compound, |
b1b0deab CG |
4717 | struct mem_cgroup *from, |
4718 | struct mem_cgroup *to) | |
4719 | { | |
4720 | unsigned long flags; | |
f627c2f5 | 4721 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
b1b0deab | 4722 | int ret; |
c4843a75 | 4723 | bool anon; |
b1b0deab CG |
4724 | |
4725 | VM_BUG_ON(from == to); | |
4726 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 4727 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
4728 | |
4729 | /* | |
6a93ca8f | 4730 | * Prevent mem_cgroup_migrate() from looking at |
45637bab | 4731 | * page->mem_cgroup of its source page while we change it. |
b1b0deab | 4732 | */ |
f627c2f5 | 4733 | ret = -EBUSY; |
b1b0deab CG |
4734 | if (!trylock_page(page)) |
4735 | goto out; | |
4736 | ||
4737 | ret = -EINVAL; | |
4738 | if (page->mem_cgroup != from) | |
4739 | goto out_unlock; | |
4740 | ||
c4843a75 GT |
4741 | anon = PageAnon(page); |
4742 | ||
b1b0deab CG |
4743 | spin_lock_irqsave(&from->move_lock, flags); |
4744 | ||
c4843a75 | 4745 | if (!anon && page_mapped(page)) { |
c9019e9b JW |
4746 | __mod_memcg_state(from, NR_FILE_MAPPED, -nr_pages); |
4747 | __mod_memcg_state(to, NR_FILE_MAPPED, nr_pages); | |
b1b0deab CG |
4748 | } |
4749 | ||
c4843a75 GT |
4750 | /* |
4751 | * move_lock grabbed above and caller set from->moving_account, so | |
ccda7f43 | 4752 | * mod_memcg_page_state will serialize updates to PageDirty. |
c4843a75 GT |
4753 | * So mapping should be stable for dirty pages. |
4754 | */ | |
4755 | if (!anon && PageDirty(page)) { | |
4756 | struct address_space *mapping = page_mapping(page); | |
4757 | ||
4758 | if (mapping_cap_account_dirty(mapping)) { | |
c9019e9b JW |
4759 | __mod_memcg_state(from, NR_FILE_DIRTY, -nr_pages); |
4760 | __mod_memcg_state(to, NR_FILE_DIRTY, nr_pages); | |
c4843a75 GT |
4761 | } |
4762 | } | |
4763 | ||
b1b0deab | 4764 | if (PageWriteback(page)) { |
c9019e9b JW |
4765 | __mod_memcg_state(from, NR_WRITEBACK, -nr_pages); |
4766 | __mod_memcg_state(to, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
4767 | } |
4768 | ||
4769 | /* | |
4770 | * It is safe to change page->mem_cgroup here because the page | |
4771 | * is referenced, charged, and isolated - we can't race with | |
4772 | * uncharging, charging, migration, or LRU putback. | |
4773 | */ | |
4774 | ||
4775 | /* caller should have done css_get */ | |
4776 | page->mem_cgroup = to; | |
4777 | spin_unlock_irqrestore(&from->move_lock, flags); | |
4778 | ||
4779 | ret = 0; | |
4780 | ||
4781 | local_irq_disable(); | |
f627c2f5 | 4782 | mem_cgroup_charge_statistics(to, page, compound, nr_pages); |
b1b0deab | 4783 | memcg_check_events(to, page); |
f627c2f5 | 4784 | mem_cgroup_charge_statistics(from, page, compound, -nr_pages); |
b1b0deab CG |
4785 | memcg_check_events(from, page); |
4786 | local_irq_enable(); | |
4787 | out_unlock: | |
4788 | unlock_page(page); | |
4789 | out: | |
4790 | return ret; | |
4791 | } | |
4792 | ||
7cf7806c LR |
4793 | /** |
4794 | * get_mctgt_type - get target type of moving charge | |
4795 | * @vma: the vma the pte to be checked belongs | |
4796 | * @addr: the address corresponding to the pte to be checked | |
4797 | * @ptent: the pte to be checked | |
4798 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
4799 | * | |
4800 | * Returns | |
4801 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4802 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4803 | * move charge. if @target is not NULL, the page is stored in target->page | |
4804 | * with extra refcnt got(Callers should handle it). | |
4805 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
4806 | * target for charge migration. if @target is not NULL, the entry is stored | |
4807 | * in target->ent. | |
df6ad698 JG |
4808 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PUBLIC |
4809 | * or MEMORY_DEVICE_PRIVATE (so ZONE_DEVICE page and thus not on the lru). | |
4810 | * For now we such page is charge like a regular page would be as for all | |
4811 | * intent and purposes it is just special memory taking the place of a | |
4812 | * regular page. | |
c733a828 JG |
4813 | * |
4814 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
4815 | * |
4816 | * Called with pte lock held. | |
4817 | */ | |
4818 | ||
8d32ff84 | 4819 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
4820 | unsigned long addr, pte_t ptent, union mc_target *target) |
4821 | { | |
4822 | struct page *page = NULL; | |
8d32ff84 | 4823 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
4824 | swp_entry_t ent = { .val = 0 }; |
4825 | ||
4826 | if (pte_present(ptent)) | |
4827 | page = mc_handle_present_pte(vma, addr, ptent); | |
4828 | else if (is_swap_pte(ptent)) | |
48406ef8 | 4829 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 4830 | else if (pte_none(ptent)) |
87946a72 | 4831 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
4832 | |
4833 | if (!page && !ent.val) | |
8d32ff84 | 4834 | return ret; |
02491447 | 4835 | if (page) { |
02491447 | 4836 | /* |
0a31bc97 | 4837 | * Do only loose check w/o serialization. |
1306a85a | 4838 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 4839 | * not under LRU exclusion. |
02491447 | 4840 | */ |
1306a85a | 4841 | if (page->mem_cgroup == mc.from) { |
02491447 | 4842 | ret = MC_TARGET_PAGE; |
df6ad698 JG |
4843 | if (is_device_private_page(page) || |
4844 | is_device_public_page(page)) | |
c733a828 | 4845 | ret = MC_TARGET_DEVICE; |
02491447 DN |
4846 | if (target) |
4847 | target->page = page; | |
4848 | } | |
4849 | if (!ret || !target) | |
4850 | put_page(page); | |
4851 | } | |
3e14a57b HY |
4852 | /* |
4853 | * There is a swap entry and a page doesn't exist or isn't charged. | |
4854 | * But we cannot move a tail-page in a THP. | |
4855 | */ | |
4856 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 4857 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
4858 | ret = MC_TARGET_SWAP; |
4859 | if (target) | |
4860 | target->ent = ent; | |
4ffef5fe | 4861 | } |
4ffef5fe DN |
4862 | return ret; |
4863 | } | |
4864 | ||
12724850 NH |
4865 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4866 | /* | |
d6810d73 HY |
4867 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
4868 | * not support them for now. | |
12724850 NH |
4869 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
4870 | */ | |
4871 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4872 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4873 | { | |
4874 | struct page *page = NULL; | |
12724850 NH |
4875 | enum mc_target_type ret = MC_TARGET_NONE; |
4876 | ||
84c3fc4e ZY |
4877 | if (unlikely(is_swap_pmd(pmd))) { |
4878 | VM_BUG_ON(thp_migration_supported() && | |
4879 | !is_pmd_migration_entry(pmd)); | |
4880 | return ret; | |
4881 | } | |
12724850 | 4882 | page = pmd_page(pmd); |
309381fe | 4883 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 4884 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 4885 | return ret; |
1306a85a | 4886 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
4887 | ret = MC_TARGET_PAGE; |
4888 | if (target) { | |
4889 | get_page(page); | |
4890 | target->page = page; | |
4891 | } | |
4892 | } | |
4893 | return ret; | |
4894 | } | |
4895 | #else | |
4896 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4897 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4898 | { | |
4899 | return MC_TARGET_NONE; | |
4900 | } | |
4901 | #endif | |
4902 | ||
4ffef5fe DN |
4903 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
4904 | unsigned long addr, unsigned long end, | |
4905 | struct mm_walk *walk) | |
4906 | { | |
26bcd64a | 4907 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4908 | pte_t *pte; |
4909 | spinlock_t *ptl; | |
4910 | ||
b6ec57f4 KS |
4911 | ptl = pmd_trans_huge_lock(pmd, vma); |
4912 | if (ptl) { | |
c733a828 JG |
4913 | /* |
4914 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
4915 | * support transparent huge page with MEMORY_DEVICE_PUBLIC or | |
4916 | * MEMORY_DEVICE_PRIVATE but this might change. | |
4917 | */ | |
12724850 NH |
4918 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
4919 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 4920 | spin_unlock(ptl); |
1a5a9906 | 4921 | return 0; |
12724850 | 4922 | } |
03319327 | 4923 | |
45f83cef AA |
4924 | if (pmd_trans_unstable(pmd)) |
4925 | return 0; | |
4ffef5fe DN |
4926 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
4927 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 4928 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
4929 | mc.precharge++; /* increment precharge temporarily */ |
4930 | pte_unmap_unlock(pte - 1, ptl); | |
4931 | cond_resched(); | |
4932 | ||
7dc74be0 DN |
4933 | return 0; |
4934 | } | |
4935 | ||
4ffef5fe DN |
4936 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
4937 | { | |
4938 | unsigned long precharge; | |
4ffef5fe | 4939 | |
26bcd64a NH |
4940 | struct mm_walk mem_cgroup_count_precharge_walk = { |
4941 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
4942 | .mm = mm, | |
4943 | }; | |
dfe076b0 | 4944 | down_read(&mm->mmap_sem); |
0247f3f4 JM |
4945 | walk_page_range(0, mm->highest_vm_end, |
4946 | &mem_cgroup_count_precharge_walk); | |
dfe076b0 | 4947 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
4948 | |
4949 | precharge = mc.precharge; | |
4950 | mc.precharge = 0; | |
4951 | ||
4952 | return precharge; | |
4953 | } | |
4954 | ||
4ffef5fe DN |
4955 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
4956 | { | |
dfe076b0 DN |
4957 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
4958 | ||
4959 | VM_BUG_ON(mc.moving_task); | |
4960 | mc.moving_task = current; | |
4961 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
4962 | } |
4963 | ||
dfe076b0 DN |
4964 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
4965 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 4966 | { |
2bd9bb20 KH |
4967 | struct mem_cgroup *from = mc.from; |
4968 | struct mem_cgroup *to = mc.to; | |
4969 | ||
4ffef5fe | 4970 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 4971 | if (mc.precharge) { |
00501b53 | 4972 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
4973 | mc.precharge = 0; |
4974 | } | |
4975 | /* | |
4976 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
4977 | * we must uncharge here. | |
4978 | */ | |
4979 | if (mc.moved_charge) { | |
00501b53 | 4980 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 4981 | mc.moved_charge = 0; |
4ffef5fe | 4982 | } |
483c30b5 DN |
4983 | /* we must fixup refcnts and charges */ |
4984 | if (mc.moved_swap) { | |
483c30b5 | 4985 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 4986 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 4987 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 4988 | |
615d66c3 VD |
4989 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
4990 | ||
05b84301 | 4991 | /* |
3e32cb2e JW |
4992 | * we charged both to->memory and to->memsw, so we |
4993 | * should uncharge to->memory. | |
05b84301 | 4994 | */ |
ce00a967 | 4995 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
4996 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
4997 | ||
615d66c3 VD |
4998 | mem_cgroup_id_get_many(mc.to, mc.moved_swap); |
4999 | css_put_many(&mc.to->css, mc.moved_swap); | |
3e32cb2e | 5000 | |
483c30b5 DN |
5001 | mc.moved_swap = 0; |
5002 | } | |
dfe076b0 DN |
5003 | memcg_oom_recover(from); |
5004 | memcg_oom_recover(to); | |
5005 | wake_up_all(&mc.waitq); | |
5006 | } | |
5007 | ||
5008 | static void mem_cgroup_clear_mc(void) | |
5009 | { | |
264a0ae1 TH |
5010 | struct mm_struct *mm = mc.mm; |
5011 | ||
dfe076b0 DN |
5012 | /* |
5013 | * we must clear moving_task before waking up waiters at the end of | |
5014 | * task migration. | |
5015 | */ | |
5016 | mc.moving_task = NULL; | |
5017 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5018 | spin_lock(&mc.lock); |
4ffef5fe DN |
5019 | mc.from = NULL; |
5020 | mc.to = NULL; | |
264a0ae1 | 5021 | mc.mm = NULL; |
2bd9bb20 | 5022 | spin_unlock(&mc.lock); |
264a0ae1 TH |
5023 | |
5024 | mmput(mm); | |
4ffef5fe DN |
5025 | } |
5026 | ||
1f7dd3e5 | 5027 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5028 | { |
1f7dd3e5 | 5029 | struct cgroup_subsys_state *css; |
eed67d75 | 5030 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 5031 | struct mem_cgroup *from; |
4530eddb | 5032 | struct task_struct *leader, *p; |
9f2115f9 | 5033 | struct mm_struct *mm; |
1dfab5ab | 5034 | unsigned long move_flags; |
9f2115f9 | 5035 | int ret = 0; |
7dc74be0 | 5036 | |
1f7dd3e5 TH |
5037 | /* charge immigration isn't supported on the default hierarchy */ |
5038 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
5039 | return 0; |
5040 | ||
4530eddb TH |
5041 | /* |
5042 | * Multi-process migrations only happen on the default hierarchy | |
5043 | * where charge immigration is not used. Perform charge | |
5044 | * immigration if @tset contains a leader and whine if there are | |
5045 | * multiple. | |
5046 | */ | |
5047 | p = NULL; | |
1f7dd3e5 | 5048 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
5049 | WARN_ON_ONCE(p); |
5050 | p = leader; | |
1f7dd3e5 | 5051 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
5052 | } |
5053 | if (!p) | |
5054 | return 0; | |
5055 | ||
1f7dd3e5 TH |
5056 | /* |
5057 | * We are now commited to this value whatever it is. Changes in this | |
5058 | * tunable will only affect upcoming migrations, not the current one. | |
5059 | * So we need to save it, and keep it going. | |
5060 | */ | |
5061 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
5062 | if (!move_flags) | |
5063 | return 0; | |
5064 | ||
9f2115f9 TH |
5065 | from = mem_cgroup_from_task(p); |
5066 | ||
5067 | VM_BUG_ON(from == memcg); | |
5068 | ||
5069 | mm = get_task_mm(p); | |
5070 | if (!mm) | |
5071 | return 0; | |
5072 | /* We move charges only when we move a owner of the mm */ | |
5073 | if (mm->owner == p) { | |
5074 | VM_BUG_ON(mc.from); | |
5075 | VM_BUG_ON(mc.to); | |
5076 | VM_BUG_ON(mc.precharge); | |
5077 | VM_BUG_ON(mc.moved_charge); | |
5078 | VM_BUG_ON(mc.moved_swap); | |
5079 | ||
5080 | spin_lock(&mc.lock); | |
264a0ae1 | 5081 | mc.mm = mm; |
9f2115f9 TH |
5082 | mc.from = from; |
5083 | mc.to = memcg; | |
5084 | mc.flags = move_flags; | |
5085 | spin_unlock(&mc.lock); | |
5086 | /* We set mc.moving_task later */ | |
5087 | ||
5088 | ret = mem_cgroup_precharge_mc(mm); | |
5089 | if (ret) | |
5090 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
5091 | } else { |
5092 | mmput(mm); | |
7dc74be0 DN |
5093 | } |
5094 | return ret; | |
5095 | } | |
5096 | ||
1f7dd3e5 | 5097 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5098 | { |
4e2f245d JW |
5099 | if (mc.to) |
5100 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5101 | } |
5102 | ||
4ffef5fe DN |
5103 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5104 | unsigned long addr, unsigned long end, | |
5105 | struct mm_walk *walk) | |
7dc74be0 | 5106 | { |
4ffef5fe | 5107 | int ret = 0; |
26bcd64a | 5108 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5109 | pte_t *pte; |
5110 | spinlock_t *ptl; | |
12724850 NH |
5111 | enum mc_target_type target_type; |
5112 | union mc_target target; | |
5113 | struct page *page; | |
4ffef5fe | 5114 | |
b6ec57f4 KS |
5115 | ptl = pmd_trans_huge_lock(pmd, vma); |
5116 | if (ptl) { | |
62ade86a | 5117 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5118 | spin_unlock(ptl); |
12724850 NH |
5119 | return 0; |
5120 | } | |
5121 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5122 | if (target_type == MC_TARGET_PAGE) { | |
5123 | page = target.page; | |
5124 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 5125 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 5126 | mc.from, mc.to)) { |
12724850 NH |
5127 | mc.precharge -= HPAGE_PMD_NR; |
5128 | mc.moved_charge += HPAGE_PMD_NR; | |
5129 | } | |
5130 | putback_lru_page(page); | |
5131 | } | |
5132 | put_page(page); | |
c733a828 JG |
5133 | } else if (target_type == MC_TARGET_DEVICE) { |
5134 | page = target.page; | |
5135 | if (!mem_cgroup_move_account(page, true, | |
5136 | mc.from, mc.to)) { | |
5137 | mc.precharge -= HPAGE_PMD_NR; | |
5138 | mc.moved_charge += HPAGE_PMD_NR; | |
5139 | } | |
5140 | put_page(page); | |
12724850 | 5141 | } |
bf929152 | 5142 | spin_unlock(ptl); |
1a5a9906 | 5143 | return 0; |
12724850 NH |
5144 | } |
5145 | ||
45f83cef AA |
5146 | if (pmd_trans_unstable(pmd)) |
5147 | return 0; | |
4ffef5fe DN |
5148 | retry: |
5149 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5150 | for (; addr != end; addr += PAGE_SIZE) { | |
5151 | pte_t ptent = *(pte++); | |
c733a828 | 5152 | bool device = false; |
02491447 | 5153 | swp_entry_t ent; |
4ffef5fe DN |
5154 | |
5155 | if (!mc.precharge) | |
5156 | break; | |
5157 | ||
8d32ff84 | 5158 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
5159 | case MC_TARGET_DEVICE: |
5160 | device = true; | |
5161 | /* fall through */ | |
4ffef5fe DN |
5162 | case MC_TARGET_PAGE: |
5163 | page = target.page; | |
53f9263b KS |
5164 | /* |
5165 | * We can have a part of the split pmd here. Moving it | |
5166 | * can be done but it would be too convoluted so simply | |
5167 | * ignore such a partial THP and keep it in original | |
5168 | * memcg. There should be somebody mapping the head. | |
5169 | */ | |
5170 | if (PageTransCompound(page)) | |
5171 | goto put; | |
c733a828 | 5172 | if (!device && isolate_lru_page(page)) |
4ffef5fe | 5173 | goto put; |
f627c2f5 KS |
5174 | if (!mem_cgroup_move_account(page, false, |
5175 | mc.from, mc.to)) { | |
4ffef5fe | 5176 | mc.precharge--; |
854ffa8d DN |
5177 | /* we uncharge from mc.from later. */ |
5178 | mc.moved_charge++; | |
4ffef5fe | 5179 | } |
c733a828 JG |
5180 | if (!device) |
5181 | putback_lru_page(page); | |
8d32ff84 | 5182 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5183 | put_page(page); |
5184 | break; | |
02491447 DN |
5185 | case MC_TARGET_SWAP: |
5186 | ent = target.ent; | |
e91cbb42 | 5187 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5188 | mc.precharge--; |
483c30b5 DN |
5189 | /* we fixup refcnts and charges later. */ |
5190 | mc.moved_swap++; | |
5191 | } | |
02491447 | 5192 | break; |
4ffef5fe DN |
5193 | default: |
5194 | break; | |
5195 | } | |
5196 | } | |
5197 | pte_unmap_unlock(pte - 1, ptl); | |
5198 | cond_resched(); | |
5199 | ||
5200 | if (addr != end) { | |
5201 | /* | |
5202 | * We have consumed all precharges we got in can_attach(). | |
5203 | * We try charge one by one, but don't do any additional | |
5204 | * charges to mc.to if we have failed in charge once in attach() | |
5205 | * phase. | |
5206 | */ | |
854ffa8d | 5207 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5208 | if (!ret) |
5209 | goto retry; | |
5210 | } | |
5211 | ||
5212 | return ret; | |
5213 | } | |
5214 | ||
264a0ae1 | 5215 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 5216 | { |
26bcd64a NH |
5217 | struct mm_walk mem_cgroup_move_charge_walk = { |
5218 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
264a0ae1 | 5219 | .mm = mc.mm, |
26bcd64a | 5220 | }; |
4ffef5fe DN |
5221 | |
5222 | lru_add_drain_all(); | |
312722cb | 5223 | /* |
81f8c3a4 JW |
5224 | * Signal lock_page_memcg() to take the memcg's move_lock |
5225 | * while we're moving its pages to another memcg. Then wait | |
5226 | * for already started RCU-only updates to finish. | |
312722cb JW |
5227 | */ |
5228 | atomic_inc(&mc.from->moving_account); | |
5229 | synchronize_rcu(); | |
dfe076b0 | 5230 | retry: |
264a0ae1 | 5231 | if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) { |
dfe076b0 DN |
5232 | /* |
5233 | * Someone who are holding the mmap_sem might be waiting in | |
5234 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5235 | * and retry. Because we cancel precharges, we might not be able | |
5236 | * to move enough charges, but moving charge is a best-effort | |
5237 | * feature anyway, so it wouldn't be a big problem. | |
5238 | */ | |
5239 | __mem_cgroup_clear_mc(); | |
5240 | cond_resched(); | |
5241 | goto retry; | |
5242 | } | |
26bcd64a NH |
5243 | /* |
5244 | * When we have consumed all precharges and failed in doing | |
5245 | * additional charge, the page walk just aborts. | |
5246 | */ | |
0247f3f4 JM |
5247 | walk_page_range(0, mc.mm->highest_vm_end, &mem_cgroup_move_charge_walk); |
5248 | ||
264a0ae1 | 5249 | up_read(&mc.mm->mmap_sem); |
312722cb | 5250 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
5251 | } |
5252 | ||
264a0ae1 | 5253 | static void mem_cgroup_move_task(void) |
67e465a7 | 5254 | { |
264a0ae1 TH |
5255 | if (mc.to) { |
5256 | mem_cgroup_move_charge(); | |
a433658c | 5257 | mem_cgroup_clear_mc(); |
264a0ae1 | 5258 | } |
67e465a7 | 5259 | } |
5cfb80a7 | 5260 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 5261 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5262 | { |
5263 | return 0; | |
5264 | } | |
1f7dd3e5 | 5265 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5266 | { |
5267 | } | |
264a0ae1 | 5268 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
5269 | { |
5270 | } | |
5271 | #endif | |
67e465a7 | 5272 | |
f00baae7 TH |
5273 | /* |
5274 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5275 | * to verify whether we're attached to the default hierarchy on each mount |
5276 | * attempt. | |
f00baae7 | 5277 | */ |
eb95419b | 5278 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5279 | { |
5280 | /* | |
aa6ec29b | 5281 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5282 | * guarantees that @root doesn't have any children, so turning it |
5283 | * on for the root memcg is enough. | |
5284 | */ | |
9e10a130 | 5285 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
5286 | root_mem_cgroup->use_hierarchy = true; |
5287 | else | |
5288 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
5289 | } |
5290 | ||
241994ed JW |
5291 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
5292 | struct cftype *cft) | |
5293 | { | |
f5fc3c5d JW |
5294 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5295 | ||
5296 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
5297 | } |
5298 | ||
bf8d5d52 RG |
5299 | static int memory_min_show(struct seq_file *m, void *v) |
5300 | { | |
5301 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5302 | unsigned long min = READ_ONCE(memcg->memory.min); | |
5303 | ||
5304 | if (min == PAGE_COUNTER_MAX) | |
5305 | seq_puts(m, "max\n"); | |
5306 | else | |
5307 | seq_printf(m, "%llu\n", (u64)min * PAGE_SIZE); | |
5308 | ||
5309 | return 0; | |
5310 | } | |
5311 | ||
5312 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
5313 | char *buf, size_t nbytes, loff_t off) | |
5314 | { | |
5315 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5316 | unsigned long min; | |
5317 | int err; | |
5318 | ||
5319 | buf = strstrip(buf); | |
5320 | err = page_counter_memparse(buf, "max", &min); | |
5321 | if (err) | |
5322 | return err; | |
5323 | ||
5324 | page_counter_set_min(&memcg->memory, min); | |
5325 | ||
5326 | return nbytes; | |
5327 | } | |
5328 | ||
241994ed JW |
5329 | static int memory_low_show(struct seq_file *m, void *v) |
5330 | { | |
5331 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
23067153 | 5332 | unsigned long low = READ_ONCE(memcg->memory.low); |
241994ed JW |
5333 | |
5334 | if (low == PAGE_COUNTER_MAX) | |
d2973697 | 5335 | seq_puts(m, "max\n"); |
241994ed JW |
5336 | else |
5337 | seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE); | |
5338 | ||
5339 | return 0; | |
5340 | } | |
5341 | ||
5342 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
5343 | char *buf, size_t nbytes, loff_t off) | |
5344 | { | |
5345 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5346 | unsigned long low; | |
5347 | int err; | |
5348 | ||
5349 | buf = strstrip(buf); | |
d2973697 | 5350 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
5351 | if (err) |
5352 | return err; | |
5353 | ||
23067153 | 5354 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
5355 | |
5356 | return nbytes; | |
5357 | } | |
5358 | ||
5359 | static int memory_high_show(struct seq_file *m, void *v) | |
5360 | { | |
5361 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5362 | unsigned long high = READ_ONCE(memcg->high); |
241994ed JW |
5363 | |
5364 | if (high == PAGE_COUNTER_MAX) | |
d2973697 | 5365 | seq_puts(m, "max\n"); |
241994ed JW |
5366 | else |
5367 | seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE); | |
5368 | ||
5369 | return 0; | |
5370 | } | |
5371 | ||
5372 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
5373 | char *buf, size_t nbytes, loff_t off) | |
5374 | { | |
5375 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
588083bb | 5376 | unsigned long nr_pages; |
241994ed JW |
5377 | unsigned long high; |
5378 | int err; | |
5379 | ||
5380 | buf = strstrip(buf); | |
d2973697 | 5381 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
5382 | if (err) |
5383 | return err; | |
5384 | ||
5385 | memcg->high = high; | |
5386 | ||
588083bb JW |
5387 | nr_pages = page_counter_read(&memcg->memory); |
5388 | if (nr_pages > high) | |
5389 | try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
5390 | GFP_KERNEL, true); | |
5391 | ||
2529bb3a | 5392 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5393 | return nbytes; |
5394 | } | |
5395 | ||
5396 | static int memory_max_show(struct seq_file *m, void *v) | |
5397 | { | |
5398 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
bbec2e15 | 5399 | unsigned long max = READ_ONCE(memcg->memory.max); |
241994ed JW |
5400 | |
5401 | if (max == PAGE_COUNTER_MAX) | |
d2973697 | 5402 | seq_puts(m, "max\n"); |
241994ed JW |
5403 | else |
5404 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
5405 | ||
5406 | return 0; | |
5407 | } | |
5408 | ||
5409 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
5410 | char *buf, size_t nbytes, loff_t off) | |
5411 | { | |
5412 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
b6e6edcf JW |
5413 | unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES; |
5414 | bool drained = false; | |
241994ed JW |
5415 | unsigned long max; |
5416 | int err; | |
5417 | ||
5418 | buf = strstrip(buf); | |
d2973697 | 5419 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
5420 | if (err) |
5421 | return err; | |
5422 | ||
bbec2e15 | 5423 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
5424 | |
5425 | for (;;) { | |
5426 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
5427 | ||
5428 | if (nr_pages <= max) | |
5429 | break; | |
5430 | ||
5431 | if (signal_pending(current)) { | |
5432 | err = -EINTR; | |
5433 | break; | |
5434 | } | |
5435 | ||
5436 | if (!drained) { | |
5437 | drain_all_stock(memcg); | |
5438 | drained = true; | |
5439 | continue; | |
5440 | } | |
5441 | ||
5442 | if (nr_reclaims) { | |
5443 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
5444 | GFP_KERNEL, true)) | |
5445 | nr_reclaims--; | |
5446 | continue; | |
5447 | } | |
5448 | ||
e27be240 | 5449 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
5450 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
5451 | break; | |
5452 | } | |
241994ed | 5453 | |
2529bb3a | 5454 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5455 | return nbytes; |
5456 | } | |
5457 | ||
5458 | static int memory_events_show(struct seq_file *m, void *v) | |
5459 | { | |
5460 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5461 | ||
e27be240 JW |
5462 | seq_printf(m, "low %lu\n", |
5463 | atomic_long_read(&memcg->memory_events[MEMCG_LOW])); | |
5464 | seq_printf(m, "high %lu\n", | |
5465 | atomic_long_read(&memcg->memory_events[MEMCG_HIGH])); | |
5466 | seq_printf(m, "max %lu\n", | |
5467 | atomic_long_read(&memcg->memory_events[MEMCG_MAX])); | |
5468 | seq_printf(m, "oom %lu\n", | |
5469 | atomic_long_read(&memcg->memory_events[MEMCG_OOM])); | |
fe6bdfc8 RG |
5470 | seq_printf(m, "oom_kill %lu\n", |
5471 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
241994ed JW |
5472 | |
5473 | return 0; | |
5474 | } | |
5475 | ||
587d9f72 JW |
5476 | static int memory_stat_show(struct seq_file *m, void *v) |
5477 | { | |
5478 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
72b54e73 | 5479 | unsigned long stat[MEMCG_NR_STAT]; |
e27be240 | 5480 | unsigned long events[NR_VM_EVENT_ITEMS]; |
587d9f72 JW |
5481 | int i; |
5482 | ||
5483 | /* | |
5484 | * Provide statistics on the state of the memory subsystem as | |
5485 | * well as cumulative event counters that show past behavior. | |
5486 | * | |
5487 | * This list is ordered following a combination of these gradients: | |
5488 | * 1) generic big picture -> specifics and details | |
5489 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
5490 | * | |
5491 | * Current memory state: | |
5492 | */ | |
5493 | ||
72b54e73 VD |
5494 | tree_stat(memcg, stat); |
5495 | tree_events(memcg, events); | |
5496 | ||
587d9f72 | 5497 | seq_printf(m, "anon %llu\n", |
71cd3113 | 5498 | (u64)stat[MEMCG_RSS] * PAGE_SIZE); |
587d9f72 | 5499 | seq_printf(m, "file %llu\n", |
71cd3113 | 5500 | (u64)stat[MEMCG_CACHE] * PAGE_SIZE); |
12580e4b | 5501 | seq_printf(m, "kernel_stack %llu\n", |
efdc9490 | 5502 | (u64)stat[MEMCG_KERNEL_STACK_KB] * 1024); |
27ee57c9 | 5503 | seq_printf(m, "slab %llu\n", |
32049296 JW |
5504 | (u64)(stat[NR_SLAB_RECLAIMABLE] + |
5505 | stat[NR_SLAB_UNRECLAIMABLE]) * PAGE_SIZE); | |
b2807f07 | 5506 | seq_printf(m, "sock %llu\n", |
72b54e73 | 5507 | (u64)stat[MEMCG_SOCK] * PAGE_SIZE); |
587d9f72 | 5508 | |
9a4caf1e | 5509 | seq_printf(m, "shmem %llu\n", |
71cd3113 | 5510 | (u64)stat[NR_SHMEM] * PAGE_SIZE); |
587d9f72 | 5511 | seq_printf(m, "file_mapped %llu\n", |
71cd3113 | 5512 | (u64)stat[NR_FILE_MAPPED] * PAGE_SIZE); |
587d9f72 | 5513 | seq_printf(m, "file_dirty %llu\n", |
71cd3113 | 5514 | (u64)stat[NR_FILE_DIRTY] * PAGE_SIZE); |
587d9f72 | 5515 | seq_printf(m, "file_writeback %llu\n", |
71cd3113 | 5516 | (u64)stat[NR_WRITEBACK] * PAGE_SIZE); |
587d9f72 JW |
5517 | |
5518 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
5519 | struct mem_cgroup *mi; | |
5520 | unsigned long val = 0; | |
5521 | ||
5522 | for_each_mem_cgroup_tree(mi, memcg) | |
5523 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)); | |
5524 | seq_printf(m, "%s %llu\n", | |
5525 | mem_cgroup_lru_names[i], (u64)val * PAGE_SIZE); | |
5526 | } | |
5527 | ||
27ee57c9 | 5528 | seq_printf(m, "slab_reclaimable %llu\n", |
32049296 | 5529 | (u64)stat[NR_SLAB_RECLAIMABLE] * PAGE_SIZE); |
27ee57c9 | 5530 | seq_printf(m, "slab_unreclaimable %llu\n", |
32049296 | 5531 | (u64)stat[NR_SLAB_UNRECLAIMABLE] * PAGE_SIZE); |
27ee57c9 | 5532 | |
587d9f72 JW |
5533 | /* Accumulated memory events */ |
5534 | ||
df0e53d0 JW |
5535 | seq_printf(m, "pgfault %lu\n", events[PGFAULT]); |
5536 | seq_printf(m, "pgmajfault %lu\n", events[PGMAJFAULT]); | |
587d9f72 | 5537 | |
2262185c RG |
5538 | seq_printf(m, "pgrefill %lu\n", events[PGREFILL]); |
5539 | seq_printf(m, "pgscan %lu\n", events[PGSCAN_KSWAPD] + | |
5540 | events[PGSCAN_DIRECT]); | |
5541 | seq_printf(m, "pgsteal %lu\n", events[PGSTEAL_KSWAPD] + | |
5542 | events[PGSTEAL_DIRECT]); | |
5543 | seq_printf(m, "pgactivate %lu\n", events[PGACTIVATE]); | |
5544 | seq_printf(m, "pgdeactivate %lu\n", events[PGDEACTIVATE]); | |
5545 | seq_printf(m, "pglazyfree %lu\n", events[PGLAZYFREE]); | |
5546 | seq_printf(m, "pglazyfreed %lu\n", events[PGLAZYFREED]); | |
5547 | ||
2a2e4885 | 5548 | seq_printf(m, "workingset_refault %lu\n", |
71cd3113 | 5549 | stat[WORKINGSET_REFAULT]); |
2a2e4885 | 5550 | seq_printf(m, "workingset_activate %lu\n", |
71cd3113 | 5551 | stat[WORKINGSET_ACTIVATE]); |
2a2e4885 | 5552 | seq_printf(m, "workingset_nodereclaim %lu\n", |
71cd3113 | 5553 | stat[WORKINGSET_NODERECLAIM]); |
2a2e4885 | 5554 | |
587d9f72 JW |
5555 | return 0; |
5556 | } | |
5557 | ||
241994ed JW |
5558 | static struct cftype memory_files[] = { |
5559 | { | |
5560 | .name = "current", | |
f5fc3c5d | 5561 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
5562 | .read_u64 = memory_current_read, |
5563 | }, | |
bf8d5d52 RG |
5564 | { |
5565 | .name = "min", | |
5566 | .flags = CFTYPE_NOT_ON_ROOT, | |
5567 | .seq_show = memory_min_show, | |
5568 | .write = memory_min_write, | |
5569 | }, | |
241994ed JW |
5570 | { |
5571 | .name = "low", | |
5572 | .flags = CFTYPE_NOT_ON_ROOT, | |
5573 | .seq_show = memory_low_show, | |
5574 | .write = memory_low_write, | |
5575 | }, | |
5576 | { | |
5577 | .name = "high", | |
5578 | .flags = CFTYPE_NOT_ON_ROOT, | |
5579 | .seq_show = memory_high_show, | |
5580 | .write = memory_high_write, | |
5581 | }, | |
5582 | { | |
5583 | .name = "max", | |
5584 | .flags = CFTYPE_NOT_ON_ROOT, | |
5585 | .seq_show = memory_max_show, | |
5586 | .write = memory_max_write, | |
5587 | }, | |
5588 | { | |
5589 | .name = "events", | |
5590 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 5591 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
5592 | .seq_show = memory_events_show, |
5593 | }, | |
587d9f72 JW |
5594 | { |
5595 | .name = "stat", | |
5596 | .flags = CFTYPE_NOT_ON_ROOT, | |
5597 | .seq_show = memory_stat_show, | |
5598 | }, | |
241994ed JW |
5599 | { } /* terminate */ |
5600 | }; | |
5601 | ||
073219e9 | 5602 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5603 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5604 | .css_online = mem_cgroup_css_online, |
92fb9748 | 5605 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 5606 | .css_released = mem_cgroup_css_released, |
92fb9748 | 5607 | .css_free = mem_cgroup_css_free, |
1ced953b | 5608 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5609 | .can_attach = mem_cgroup_can_attach, |
5610 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 5611 | .post_attach = mem_cgroup_move_task, |
f00baae7 | 5612 | .bind = mem_cgroup_bind, |
241994ed JW |
5613 | .dfl_cftypes = memory_files, |
5614 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 5615 | .early_init = 0, |
8cdea7c0 | 5616 | }; |
c077719b | 5617 | |
241994ed | 5618 | /** |
bf8d5d52 | 5619 | * mem_cgroup_protected - check if memory consumption is in the normal range |
34c81057 | 5620 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
5621 | * @memcg: the memory cgroup to check |
5622 | * | |
23067153 RG |
5623 | * WARNING: This function is not stateless! It can only be used as part |
5624 | * of a top-down tree iteration, not for isolated queries. | |
34c81057 | 5625 | * |
bf8d5d52 RG |
5626 | * Returns one of the following: |
5627 | * MEMCG_PROT_NONE: cgroup memory is not protected | |
5628 | * MEMCG_PROT_LOW: cgroup memory is protected as long there is | |
5629 | * an unprotected supply of reclaimable memory from other cgroups. | |
5630 | * MEMCG_PROT_MIN: cgroup memory is protected | |
34c81057 | 5631 | * |
bf8d5d52 | 5632 | * @root is exclusive; it is never protected when looked at directly |
34c81057 | 5633 | * |
bf8d5d52 RG |
5634 | * To provide a proper hierarchical behavior, effective memory.min/low values |
5635 | * are used. Below is the description of how effective memory.low is calculated. | |
5636 | * Effective memory.min values is calculated in the same way. | |
34c81057 | 5637 | * |
23067153 RG |
5638 | * Effective memory.low is always equal or less than the original memory.low. |
5639 | * If there is no memory.low overcommittment (which is always true for | |
5640 | * top-level memory cgroups), these two values are equal. | |
5641 | * Otherwise, it's a part of parent's effective memory.low, | |
5642 | * calculated as a cgroup's memory.low usage divided by sum of sibling's | |
5643 | * memory.low usages, where memory.low usage is the size of actually | |
5644 | * protected memory. | |
34c81057 | 5645 | * |
23067153 RG |
5646 | * low_usage |
5647 | * elow = min( memory.low, parent->elow * ------------------ ), | |
5648 | * siblings_low_usage | |
34c81057 | 5649 | * |
23067153 RG |
5650 | * | memory.current, if memory.current < memory.low |
5651 | * low_usage = | | |
5652 | | 0, otherwise. | |
34c81057 | 5653 | * |
23067153 RG |
5654 | * |
5655 | * Such definition of the effective memory.low provides the expected | |
5656 | * hierarchical behavior: parent's memory.low value is limiting | |
5657 | * children, unprotected memory is reclaimed first and cgroups, | |
5658 | * which are not using their guarantee do not affect actual memory | |
5659 | * distribution. | |
5660 | * | |
5661 | * For example, if there are memcgs A, A/B, A/C, A/D and A/E: | |
5662 | * | |
5663 | * A A/memory.low = 2G, A/memory.current = 6G | |
5664 | * //\\ | |
5665 | * BC DE B/memory.low = 3G B/memory.current = 2G | |
5666 | * C/memory.low = 1G C/memory.current = 2G | |
5667 | * D/memory.low = 0 D/memory.current = 2G | |
5668 | * E/memory.low = 10G E/memory.current = 0 | |
5669 | * | |
5670 | * and the memory pressure is applied, the following memory distribution | |
5671 | * is expected (approximately): | |
5672 | * | |
5673 | * A/memory.current = 2G | |
5674 | * | |
5675 | * B/memory.current = 1.3G | |
5676 | * C/memory.current = 0.6G | |
5677 | * D/memory.current = 0 | |
5678 | * E/memory.current = 0 | |
5679 | * | |
5680 | * These calculations require constant tracking of the actual low usages | |
bf8d5d52 RG |
5681 | * (see propagate_protected_usage()), as well as recursive calculation of |
5682 | * effective memory.low values. But as we do call mem_cgroup_protected() | |
23067153 RG |
5683 | * path for each memory cgroup top-down from the reclaim, |
5684 | * it's possible to optimize this part, and save calculated elow | |
5685 | * for next usage. This part is intentionally racy, but it's ok, | |
5686 | * as memory.low is a best-effort mechanism. | |
241994ed | 5687 | */ |
bf8d5d52 RG |
5688 | enum mem_cgroup_protection mem_cgroup_protected(struct mem_cgroup *root, |
5689 | struct mem_cgroup *memcg) | |
241994ed | 5690 | { |
23067153 | 5691 | struct mem_cgroup *parent; |
bf8d5d52 RG |
5692 | unsigned long emin, parent_emin; |
5693 | unsigned long elow, parent_elow; | |
5694 | unsigned long usage; | |
23067153 | 5695 | |
241994ed | 5696 | if (mem_cgroup_disabled()) |
bf8d5d52 | 5697 | return MEMCG_PROT_NONE; |
241994ed | 5698 | |
34c81057 SC |
5699 | if (!root) |
5700 | root = root_mem_cgroup; | |
5701 | if (memcg == root) | |
bf8d5d52 | 5702 | return MEMCG_PROT_NONE; |
241994ed | 5703 | |
23067153 | 5704 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 RG |
5705 | if (!usage) |
5706 | return MEMCG_PROT_NONE; | |
5707 | ||
5708 | emin = memcg->memory.min; | |
5709 | elow = memcg->memory.low; | |
34c81057 | 5710 | |
bf8d5d52 | 5711 | parent = parent_mem_cgroup(memcg); |
df2a4196 RG |
5712 | /* No parent means a non-hierarchical mode on v1 memcg */ |
5713 | if (!parent) | |
5714 | return MEMCG_PROT_NONE; | |
5715 | ||
23067153 RG |
5716 | if (parent == root) |
5717 | goto exit; | |
5718 | ||
bf8d5d52 RG |
5719 | parent_emin = READ_ONCE(parent->memory.emin); |
5720 | emin = min(emin, parent_emin); | |
5721 | if (emin && parent_emin) { | |
5722 | unsigned long min_usage, siblings_min_usage; | |
5723 | ||
5724 | min_usage = min(usage, memcg->memory.min); | |
5725 | siblings_min_usage = atomic_long_read( | |
5726 | &parent->memory.children_min_usage); | |
5727 | ||
5728 | if (min_usage && siblings_min_usage) | |
5729 | emin = min(emin, parent_emin * min_usage / | |
5730 | siblings_min_usage); | |
5731 | } | |
5732 | ||
23067153 RG |
5733 | parent_elow = READ_ONCE(parent->memory.elow); |
5734 | elow = min(elow, parent_elow); | |
bf8d5d52 RG |
5735 | if (elow && parent_elow) { |
5736 | unsigned long low_usage, siblings_low_usage; | |
23067153 | 5737 | |
bf8d5d52 RG |
5738 | low_usage = min(usage, memcg->memory.low); |
5739 | siblings_low_usage = atomic_long_read( | |
5740 | &parent->memory.children_low_usage); | |
23067153 | 5741 | |
bf8d5d52 RG |
5742 | if (low_usage && siblings_low_usage) |
5743 | elow = min(elow, parent_elow * low_usage / | |
5744 | siblings_low_usage); | |
5745 | } | |
23067153 | 5746 | |
23067153 | 5747 | exit: |
bf8d5d52 | 5748 | memcg->memory.emin = emin; |
23067153 | 5749 | memcg->memory.elow = elow; |
bf8d5d52 RG |
5750 | |
5751 | if (usage <= emin) | |
5752 | return MEMCG_PROT_MIN; | |
5753 | else if (usage <= elow) | |
5754 | return MEMCG_PROT_LOW; | |
5755 | else | |
5756 | return MEMCG_PROT_NONE; | |
241994ed JW |
5757 | } |
5758 | ||
00501b53 JW |
5759 | /** |
5760 | * mem_cgroup_try_charge - try charging a page | |
5761 | * @page: page to charge | |
5762 | * @mm: mm context of the victim | |
5763 | * @gfp_mask: reclaim mode | |
5764 | * @memcgp: charged memcg return | |
25843c2b | 5765 | * @compound: charge the page as compound or small page |
00501b53 JW |
5766 | * |
5767 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5768 | * pages according to @gfp_mask if necessary. | |
5769 | * | |
5770 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5771 | * Otherwise, an error code is returned. | |
5772 | * | |
5773 | * After page->mapping has been set up, the caller must finalize the | |
5774 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5775 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5776 | */ | |
5777 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
f627c2f5 KS |
5778 | gfp_t gfp_mask, struct mem_cgroup **memcgp, |
5779 | bool compound) | |
00501b53 JW |
5780 | { |
5781 | struct mem_cgroup *memcg = NULL; | |
f627c2f5 | 5782 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5783 | int ret = 0; |
5784 | ||
5785 | if (mem_cgroup_disabled()) | |
5786 | goto out; | |
5787 | ||
5788 | if (PageSwapCache(page)) { | |
00501b53 JW |
5789 | /* |
5790 | * Every swap fault against a single page tries to charge the | |
5791 | * page, bail as early as possible. shmem_unuse() encounters | |
5792 | * already charged pages, too. The USED bit is protected by | |
5793 | * the page lock, which serializes swap cache removal, which | |
5794 | * in turn serializes uncharging. | |
5795 | */ | |
e993d905 | 5796 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
abe2895b | 5797 | if (compound_head(page)->mem_cgroup) |
00501b53 | 5798 | goto out; |
e993d905 | 5799 | |
37e84351 | 5800 | if (do_swap_account) { |
e993d905 VD |
5801 | swp_entry_t ent = { .val = page_private(page), }; |
5802 | unsigned short id = lookup_swap_cgroup_id(ent); | |
5803 | ||
5804 | rcu_read_lock(); | |
5805 | memcg = mem_cgroup_from_id(id); | |
5806 | if (memcg && !css_tryget_online(&memcg->css)) | |
5807 | memcg = NULL; | |
5808 | rcu_read_unlock(); | |
5809 | } | |
00501b53 JW |
5810 | } |
5811 | ||
00501b53 JW |
5812 | if (!memcg) |
5813 | memcg = get_mem_cgroup_from_mm(mm); | |
5814 | ||
5815 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
5816 | ||
5817 | css_put(&memcg->css); | |
00501b53 JW |
5818 | out: |
5819 | *memcgp = memcg; | |
5820 | return ret; | |
5821 | } | |
5822 | ||
2cf85583 TH |
5823 | int mem_cgroup_try_charge_delay(struct page *page, struct mm_struct *mm, |
5824 | gfp_t gfp_mask, struct mem_cgroup **memcgp, | |
5825 | bool compound) | |
5826 | { | |
5827 | struct mem_cgroup *memcg; | |
5828 | int ret; | |
5829 | ||
5830 | ret = mem_cgroup_try_charge(page, mm, gfp_mask, memcgp, compound); | |
5831 | memcg = *memcgp; | |
5832 | mem_cgroup_throttle_swaprate(memcg, page_to_nid(page), gfp_mask); | |
5833 | return ret; | |
5834 | } | |
5835 | ||
00501b53 JW |
5836 | /** |
5837 | * mem_cgroup_commit_charge - commit a page charge | |
5838 | * @page: page to charge | |
5839 | * @memcg: memcg to charge the page to | |
5840 | * @lrucare: page might be on LRU already | |
25843c2b | 5841 | * @compound: charge the page as compound or small page |
00501b53 JW |
5842 | * |
5843 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
5844 | * after page->mapping has been set up. This must happen atomically | |
5845 | * as part of the page instantiation, i.e. under the page table lock | |
5846 | * for anonymous pages, under the page lock for page and swap cache. | |
5847 | * | |
5848 | * In addition, the page must not be on the LRU during the commit, to | |
5849 | * prevent racing with task migration. If it might be, use @lrucare. | |
5850 | * | |
5851 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
5852 | */ | |
5853 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
f627c2f5 | 5854 | bool lrucare, bool compound) |
00501b53 | 5855 | { |
f627c2f5 | 5856 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5857 | |
5858 | VM_BUG_ON_PAGE(!page->mapping, page); | |
5859 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
5860 | ||
5861 | if (mem_cgroup_disabled()) | |
5862 | return; | |
5863 | /* | |
5864 | * Swap faults will attempt to charge the same page multiple | |
5865 | * times. But reuse_swap_page() might have removed the page | |
5866 | * from swapcache already, so we can't check PageSwapCache(). | |
5867 | */ | |
5868 | if (!memcg) | |
5869 | return; | |
5870 | ||
6abb5a86 JW |
5871 | commit_charge(page, memcg, lrucare); |
5872 | ||
6abb5a86 | 5873 | local_irq_disable(); |
f627c2f5 | 5874 | mem_cgroup_charge_statistics(memcg, page, compound, nr_pages); |
6abb5a86 JW |
5875 | memcg_check_events(memcg, page); |
5876 | local_irq_enable(); | |
00501b53 | 5877 | |
7941d214 | 5878 | if (do_memsw_account() && PageSwapCache(page)) { |
00501b53 JW |
5879 | swp_entry_t entry = { .val = page_private(page) }; |
5880 | /* | |
5881 | * The swap entry might not get freed for a long time, | |
5882 | * let's not wait for it. The page already received a | |
5883 | * memory+swap charge, drop the swap entry duplicate. | |
5884 | */ | |
38d8b4e6 | 5885 | mem_cgroup_uncharge_swap(entry, nr_pages); |
00501b53 JW |
5886 | } |
5887 | } | |
5888 | ||
5889 | /** | |
5890 | * mem_cgroup_cancel_charge - cancel a page charge | |
5891 | * @page: page to charge | |
5892 | * @memcg: memcg to charge the page to | |
25843c2b | 5893 | * @compound: charge the page as compound or small page |
00501b53 JW |
5894 | * |
5895 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
5896 | */ | |
f627c2f5 KS |
5897 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg, |
5898 | bool compound) | |
00501b53 | 5899 | { |
f627c2f5 | 5900 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5901 | |
5902 | if (mem_cgroup_disabled()) | |
5903 | return; | |
5904 | /* | |
5905 | * Swap faults will attempt to charge the same page multiple | |
5906 | * times. But reuse_swap_page() might have removed the page | |
5907 | * from swapcache already, so we can't check PageSwapCache(). | |
5908 | */ | |
5909 | if (!memcg) | |
5910 | return; | |
5911 | ||
00501b53 JW |
5912 | cancel_charge(memcg, nr_pages); |
5913 | } | |
5914 | ||
a9d5adee JG |
5915 | struct uncharge_gather { |
5916 | struct mem_cgroup *memcg; | |
5917 | unsigned long pgpgout; | |
5918 | unsigned long nr_anon; | |
5919 | unsigned long nr_file; | |
5920 | unsigned long nr_kmem; | |
5921 | unsigned long nr_huge; | |
5922 | unsigned long nr_shmem; | |
5923 | struct page *dummy_page; | |
5924 | }; | |
5925 | ||
5926 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 5927 | { |
a9d5adee JG |
5928 | memset(ug, 0, sizeof(*ug)); |
5929 | } | |
5930 | ||
5931 | static void uncharge_batch(const struct uncharge_gather *ug) | |
5932 | { | |
5933 | unsigned long nr_pages = ug->nr_anon + ug->nr_file + ug->nr_kmem; | |
747db954 JW |
5934 | unsigned long flags; |
5935 | ||
a9d5adee JG |
5936 | if (!mem_cgroup_is_root(ug->memcg)) { |
5937 | page_counter_uncharge(&ug->memcg->memory, nr_pages); | |
7941d214 | 5938 | if (do_memsw_account()) |
a9d5adee JG |
5939 | page_counter_uncharge(&ug->memcg->memsw, nr_pages); |
5940 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem) | |
5941 | page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem); | |
5942 | memcg_oom_recover(ug->memcg); | |
ce00a967 | 5943 | } |
747db954 JW |
5944 | |
5945 | local_irq_save(flags); | |
c9019e9b JW |
5946 | __mod_memcg_state(ug->memcg, MEMCG_RSS, -ug->nr_anon); |
5947 | __mod_memcg_state(ug->memcg, MEMCG_CACHE, -ug->nr_file); | |
5948 | __mod_memcg_state(ug->memcg, MEMCG_RSS_HUGE, -ug->nr_huge); | |
5949 | __mod_memcg_state(ug->memcg, NR_SHMEM, -ug->nr_shmem); | |
5950 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); | |
a983b5eb | 5951 | __this_cpu_add(ug->memcg->stat_cpu->nr_page_events, nr_pages); |
a9d5adee | 5952 | memcg_check_events(ug->memcg, ug->dummy_page); |
747db954 | 5953 | local_irq_restore(flags); |
e8ea14cc | 5954 | |
a9d5adee JG |
5955 | if (!mem_cgroup_is_root(ug->memcg)) |
5956 | css_put_many(&ug->memcg->css, nr_pages); | |
5957 | } | |
5958 | ||
5959 | static void uncharge_page(struct page *page, struct uncharge_gather *ug) | |
5960 | { | |
5961 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
3f2eb028 JG |
5962 | VM_BUG_ON_PAGE(page_count(page) && !is_zone_device_page(page) && |
5963 | !PageHWPoison(page) , page); | |
a9d5adee JG |
5964 | |
5965 | if (!page->mem_cgroup) | |
5966 | return; | |
5967 | ||
5968 | /* | |
5969 | * Nobody should be changing or seriously looking at | |
5970 | * page->mem_cgroup at this point, we have fully | |
5971 | * exclusive access to the page. | |
5972 | */ | |
5973 | ||
5974 | if (ug->memcg != page->mem_cgroup) { | |
5975 | if (ug->memcg) { | |
5976 | uncharge_batch(ug); | |
5977 | uncharge_gather_clear(ug); | |
5978 | } | |
5979 | ug->memcg = page->mem_cgroup; | |
5980 | } | |
5981 | ||
5982 | if (!PageKmemcg(page)) { | |
5983 | unsigned int nr_pages = 1; | |
5984 | ||
5985 | if (PageTransHuge(page)) { | |
5986 | nr_pages <<= compound_order(page); | |
5987 | ug->nr_huge += nr_pages; | |
5988 | } | |
5989 | if (PageAnon(page)) | |
5990 | ug->nr_anon += nr_pages; | |
5991 | else { | |
5992 | ug->nr_file += nr_pages; | |
5993 | if (PageSwapBacked(page)) | |
5994 | ug->nr_shmem += nr_pages; | |
5995 | } | |
5996 | ug->pgpgout++; | |
5997 | } else { | |
5998 | ug->nr_kmem += 1 << compound_order(page); | |
5999 | __ClearPageKmemcg(page); | |
6000 | } | |
6001 | ||
6002 | ug->dummy_page = page; | |
6003 | page->mem_cgroup = NULL; | |
747db954 JW |
6004 | } |
6005 | ||
6006 | static void uncharge_list(struct list_head *page_list) | |
6007 | { | |
a9d5adee | 6008 | struct uncharge_gather ug; |
747db954 | 6009 | struct list_head *next; |
a9d5adee JG |
6010 | |
6011 | uncharge_gather_clear(&ug); | |
747db954 | 6012 | |
8b592656 JW |
6013 | /* |
6014 | * Note that the list can be a single page->lru; hence the | |
6015 | * do-while loop instead of a simple list_for_each_entry(). | |
6016 | */ | |
747db954 JW |
6017 | next = page_list->next; |
6018 | do { | |
a9d5adee JG |
6019 | struct page *page; |
6020 | ||
747db954 JW |
6021 | page = list_entry(next, struct page, lru); |
6022 | next = page->lru.next; | |
6023 | ||
a9d5adee | 6024 | uncharge_page(page, &ug); |
747db954 JW |
6025 | } while (next != page_list); |
6026 | ||
a9d5adee JG |
6027 | if (ug.memcg) |
6028 | uncharge_batch(&ug); | |
747db954 JW |
6029 | } |
6030 | ||
0a31bc97 JW |
6031 | /** |
6032 | * mem_cgroup_uncharge - uncharge a page | |
6033 | * @page: page to uncharge | |
6034 | * | |
6035 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
6036 | * mem_cgroup_commit_charge(). | |
6037 | */ | |
6038 | void mem_cgroup_uncharge(struct page *page) | |
6039 | { | |
a9d5adee JG |
6040 | struct uncharge_gather ug; |
6041 | ||
0a31bc97 JW |
6042 | if (mem_cgroup_disabled()) |
6043 | return; | |
6044 | ||
747db954 | 6045 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 6046 | if (!page->mem_cgroup) |
0a31bc97 JW |
6047 | return; |
6048 | ||
a9d5adee JG |
6049 | uncharge_gather_clear(&ug); |
6050 | uncharge_page(page, &ug); | |
6051 | uncharge_batch(&ug); | |
747db954 | 6052 | } |
0a31bc97 | 6053 | |
747db954 JW |
6054 | /** |
6055 | * mem_cgroup_uncharge_list - uncharge a list of page | |
6056 | * @page_list: list of pages to uncharge | |
6057 | * | |
6058 | * Uncharge a list of pages previously charged with | |
6059 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
6060 | */ | |
6061 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
6062 | { | |
6063 | if (mem_cgroup_disabled()) | |
6064 | return; | |
0a31bc97 | 6065 | |
747db954 JW |
6066 | if (!list_empty(page_list)) |
6067 | uncharge_list(page_list); | |
0a31bc97 JW |
6068 | } |
6069 | ||
6070 | /** | |
6a93ca8f JW |
6071 | * mem_cgroup_migrate - charge a page's replacement |
6072 | * @oldpage: currently circulating page | |
6073 | * @newpage: replacement page | |
0a31bc97 | 6074 | * |
6a93ca8f JW |
6075 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
6076 | * be uncharged upon free. | |
0a31bc97 JW |
6077 | * |
6078 | * Both pages must be locked, @newpage->mapping must be set up. | |
6079 | */ | |
6a93ca8f | 6080 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 6081 | { |
29833315 | 6082 | struct mem_cgroup *memcg; |
44b7a8d3 JW |
6083 | unsigned int nr_pages; |
6084 | bool compound; | |
d93c4130 | 6085 | unsigned long flags; |
0a31bc97 JW |
6086 | |
6087 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6088 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 6089 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
6090 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6091 | newpage); | |
0a31bc97 JW |
6092 | |
6093 | if (mem_cgroup_disabled()) | |
6094 | return; | |
6095 | ||
6096 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 6097 | if (newpage->mem_cgroup) |
0a31bc97 JW |
6098 | return; |
6099 | ||
45637bab | 6100 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 6101 | memcg = oldpage->mem_cgroup; |
29833315 | 6102 | if (!memcg) |
0a31bc97 JW |
6103 | return; |
6104 | ||
44b7a8d3 JW |
6105 | /* Force-charge the new page. The old one will be freed soon */ |
6106 | compound = PageTransHuge(newpage); | |
6107 | nr_pages = compound ? hpage_nr_pages(newpage) : 1; | |
6108 | ||
6109 | page_counter_charge(&memcg->memory, nr_pages); | |
6110 | if (do_memsw_account()) | |
6111 | page_counter_charge(&memcg->memsw, nr_pages); | |
6112 | css_get_many(&memcg->css, nr_pages); | |
0a31bc97 | 6113 | |
9cf7666a | 6114 | commit_charge(newpage, memcg, false); |
44b7a8d3 | 6115 | |
d93c4130 | 6116 | local_irq_save(flags); |
44b7a8d3 JW |
6117 | mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages); |
6118 | memcg_check_events(memcg, newpage); | |
d93c4130 | 6119 | local_irq_restore(flags); |
0a31bc97 JW |
6120 | } |
6121 | ||
ef12947c | 6122 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
6123 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
6124 | ||
2d758073 | 6125 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
6126 | { |
6127 | struct mem_cgroup *memcg; | |
6128 | ||
2d758073 JW |
6129 | if (!mem_cgroup_sockets_enabled) |
6130 | return; | |
6131 | ||
edbe69ef RG |
6132 | /* |
6133 | * Socket cloning can throw us here with sk_memcg already | |
6134 | * filled. It won't however, necessarily happen from | |
6135 | * process context. So the test for root memcg given | |
6136 | * the current task's memcg won't help us in this case. | |
6137 | * | |
6138 | * Respecting the original socket's memcg is a better | |
6139 | * decision in this case. | |
6140 | */ | |
6141 | if (sk->sk_memcg) { | |
6142 | css_get(&sk->sk_memcg->css); | |
6143 | return; | |
6144 | } | |
6145 | ||
11092087 JW |
6146 | rcu_read_lock(); |
6147 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
6148 | if (memcg == root_mem_cgroup) |
6149 | goto out; | |
0db15298 | 6150 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 6151 | goto out; |
f7e1cb6e | 6152 | if (css_tryget_online(&memcg->css)) |
11092087 | 6153 | sk->sk_memcg = memcg; |
f7e1cb6e | 6154 | out: |
11092087 JW |
6155 | rcu_read_unlock(); |
6156 | } | |
11092087 | 6157 | |
2d758073 | 6158 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 6159 | { |
2d758073 JW |
6160 | if (sk->sk_memcg) |
6161 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
6162 | } |
6163 | ||
6164 | /** | |
6165 | * mem_cgroup_charge_skmem - charge socket memory | |
6166 | * @memcg: memcg to charge | |
6167 | * @nr_pages: number of pages to charge | |
6168 | * | |
6169 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
6170 | * @memcg's configured limit, %false if the charge had to be forced. | |
6171 | */ | |
6172 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6173 | { | |
f7e1cb6e | 6174 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 6175 | |
f7e1cb6e | 6176 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6177 | struct page_counter *fail; |
f7e1cb6e | 6178 | |
0db15298 JW |
6179 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
6180 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
6181 | return true; |
6182 | } | |
0db15298 JW |
6183 | page_counter_charge(&memcg->tcpmem, nr_pages); |
6184 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 6185 | return false; |
11092087 | 6186 | } |
d886f4e4 | 6187 | |
f7e1cb6e JW |
6188 | /* Don't block in the packet receive path */ |
6189 | if (in_softirq()) | |
6190 | gfp_mask = GFP_NOWAIT; | |
6191 | ||
c9019e9b | 6192 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); |
b2807f07 | 6193 | |
f7e1cb6e JW |
6194 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
6195 | return true; | |
6196 | ||
6197 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
6198 | return false; |
6199 | } | |
6200 | ||
6201 | /** | |
6202 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
6203 | * @memcg: memcg to uncharge |
6204 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
6205 | */ |
6206 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6207 | { | |
f7e1cb6e | 6208 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6209 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
6210 | return; |
6211 | } | |
d886f4e4 | 6212 | |
c9019e9b | 6213 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 6214 | |
475d0487 | 6215 | refill_stock(memcg, nr_pages); |
11092087 JW |
6216 | } |
6217 | ||
f7e1cb6e JW |
6218 | static int __init cgroup_memory(char *s) |
6219 | { | |
6220 | char *token; | |
6221 | ||
6222 | while ((token = strsep(&s, ",")) != NULL) { | |
6223 | if (!*token) | |
6224 | continue; | |
6225 | if (!strcmp(token, "nosocket")) | |
6226 | cgroup_memory_nosocket = true; | |
04823c83 VD |
6227 | if (!strcmp(token, "nokmem")) |
6228 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
6229 | } |
6230 | return 0; | |
6231 | } | |
6232 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 6233 | |
2d11085e | 6234 | /* |
1081312f MH |
6235 | * subsys_initcall() for memory controller. |
6236 | * | |
308167fc SAS |
6237 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
6238 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
6239 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
6240 | * should be initialized from here. | |
2d11085e MH |
6241 | */ |
6242 | static int __init mem_cgroup_init(void) | |
6243 | { | |
95a045f6 JW |
6244 | int cpu, node; |
6245 | ||
84c07d11 | 6246 | #ifdef CONFIG_MEMCG_KMEM |
13583c3d VD |
6247 | /* |
6248 | * Kmem cache creation is mostly done with the slab_mutex held, | |
17cc4dfe TH |
6249 | * so use a workqueue with limited concurrency to avoid stalling |
6250 | * all worker threads in case lots of cgroups are created and | |
6251 | * destroyed simultaneously. | |
13583c3d | 6252 | */ |
17cc4dfe TH |
6253 | memcg_kmem_cache_wq = alloc_workqueue("memcg_kmem_cache", 0, 1); |
6254 | BUG_ON(!memcg_kmem_cache_wq); | |
13583c3d VD |
6255 | #endif |
6256 | ||
308167fc SAS |
6257 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
6258 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
6259 | |
6260 | for_each_possible_cpu(cpu) | |
6261 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
6262 | drain_local_stock); | |
6263 | ||
6264 | for_each_node(node) { | |
6265 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
6266 | |
6267 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
6268 | node_online(node) ? node : NUMA_NO_NODE); | |
6269 | ||
ef8f2327 | 6270 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 6271 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 6272 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
6273 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
6274 | } | |
6275 | ||
2d11085e MH |
6276 | return 0; |
6277 | } | |
6278 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
6279 | |
6280 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
6281 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
6282 | { | |
6283 | while (!atomic_inc_not_zero(&memcg->id.ref)) { | |
6284 | /* | |
6285 | * The root cgroup cannot be destroyed, so it's refcount must | |
6286 | * always be >= 1. | |
6287 | */ | |
6288 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
6289 | VM_BUG_ON(1); | |
6290 | break; | |
6291 | } | |
6292 | memcg = parent_mem_cgroup(memcg); | |
6293 | if (!memcg) | |
6294 | memcg = root_mem_cgroup; | |
6295 | } | |
6296 | return memcg; | |
6297 | } | |
6298 | ||
21afa38e JW |
6299 | /** |
6300 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
6301 | * @page: page whose memsw charge to transfer | |
6302 | * @entry: swap entry to move the charge to | |
6303 | * | |
6304 | * Transfer the memsw charge of @page to @entry. | |
6305 | */ | |
6306 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
6307 | { | |
1f47b61f | 6308 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 6309 | unsigned int nr_entries; |
21afa38e JW |
6310 | unsigned short oldid; |
6311 | ||
6312 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
6313 | VM_BUG_ON_PAGE(page_count(page), page); | |
6314 | ||
7941d214 | 6315 | if (!do_memsw_account()) |
21afa38e JW |
6316 | return; |
6317 | ||
6318 | memcg = page->mem_cgroup; | |
6319 | ||
6320 | /* Readahead page, never charged */ | |
6321 | if (!memcg) | |
6322 | return; | |
6323 | ||
1f47b61f VD |
6324 | /* |
6325 | * In case the memcg owning these pages has been offlined and doesn't | |
6326 | * have an ID allocated to it anymore, charge the closest online | |
6327 | * ancestor for the swap instead and transfer the memory+swap charge. | |
6328 | */ | |
6329 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
d6810d73 HY |
6330 | nr_entries = hpage_nr_pages(page); |
6331 | /* Get references for the tail pages, too */ | |
6332 | if (nr_entries > 1) | |
6333 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
6334 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
6335 | nr_entries); | |
21afa38e | 6336 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 6337 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e JW |
6338 | |
6339 | page->mem_cgroup = NULL; | |
6340 | ||
6341 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 6342 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 6343 | |
1f47b61f VD |
6344 | if (memcg != swap_memcg) { |
6345 | if (!mem_cgroup_is_root(swap_memcg)) | |
d6810d73 HY |
6346 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
6347 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
6348 | } |
6349 | ||
ce9ce665 SAS |
6350 | /* |
6351 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 6352 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 6353 | * important here to have the interrupts disabled because it is the |
b93b0163 | 6354 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 SAS |
6355 | */ |
6356 | VM_BUG_ON(!irqs_disabled()); | |
d6810d73 HY |
6357 | mem_cgroup_charge_statistics(memcg, page, PageTransHuge(page), |
6358 | -nr_entries); | |
21afa38e | 6359 | memcg_check_events(memcg, page); |
73f576c0 JW |
6360 | |
6361 | if (!mem_cgroup_is_root(memcg)) | |
d08afa14 | 6362 | css_put_many(&memcg->css, nr_entries); |
21afa38e JW |
6363 | } |
6364 | ||
38d8b4e6 HY |
6365 | /** |
6366 | * mem_cgroup_try_charge_swap - try charging swap space for a page | |
37e84351 VD |
6367 | * @page: page being added to swap |
6368 | * @entry: swap entry to charge | |
6369 | * | |
38d8b4e6 | 6370 | * Try to charge @page's memcg for the swap space at @entry. |
37e84351 VD |
6371 | * |
6372 | * Returns 0 on success, -ENOMEM on failure. | |
6373 | */ | |
6374 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
6375 | { | |
38d8b4e6 | 6376 | unsigned int nr_pages = hpage_nr_pages(page); |
37e84351 | 6377 | struct page_counter *counter; |
38d8b4e6 | 6378 | struct mem_cgroup *memcg; |
37e84351 VD |
6379 | unsigned short oldid; |
6380 | ||
6381 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account) | |
6382 | return 0; | |
6383 | ||
6384 | memcg = page->mem_cgroup; | |
6385 | ||
6386 | /* Readahead page, never charged */ | |
6387 | if (!memcg) | |
6388 | return 0; | |
6389 | ||
f3a53a3a TH |
6390 | if (!entry.val) { |
6391 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 6392 | return 0; |
f3a53a3a | 6393 | } |
bb98f2c5 | 6394 | |
1f47b61f VD |
6395 | memcg = mem_cgroup_id_get_online(memcg); |
6396 | ||
37e84351 | 6397 | if (!mem_cgroup_is_root(memcg) && |
38d8b4e6 | 6398 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
6399 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
6400 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 6401 | mem_cgroup_id_put(memcg); |
37e84351 | 6402 | return -ENOMEM; |
1f47b61f | 6403 | } |
37e84351 | 6404 | |
38d8b4e6 HY |
6405 | /* Get references for the tail pages, too */ |
6406 | if (nr_pages > 1) | |
6407 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
6408 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
37e84351 | 6409 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 6410 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 6411 | |
37e84351 VD |
6412 | return 0; |
6413 | } | |
6414 | ||
21afa38e | 6415 | /** |
38d8b4e6 | 6416 | * mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 6417 | * @entry: swap entry to uncharge |
38d8b4e6 | 6418 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 6419 | */ |
38d8b4e6 | 6420 | void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
6421 | { |
6422 | struct mem_cgroup *memcg; | |
6423 | unsigned short id; | |
6424 | ||
37e84351 | 6425 | if (!do_swap_account) |
21afa38e JW |
6426 | return; |
6427 | ||
38d8b4e6 | 6428 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 6429 | rcu_read_lock(); |
adbe427b | 6430 | memcg = mem_cgroup_from_id(id); |
21afa38e | 6431 | if (memcg) { |
37e84351 VD |
6432 | if (!mem_cgroup_is_root(memcg)) { |
6433 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
38d8b4e6 | 6434 | page_counter_uncharge(&memcg->swap, nr_pages); |
37e84351 | 6435 | else |
38d8b4e6 | 6436 | page_counter_uncharge(&memcg->memsw, nr_pages); |
37e84351 | 6437 | } |
c9019e9b | 6438 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 6439 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
6440 | } |
6441 | rcu_read_unlock(); | |
6442 | } | |
6443 | ||
d8b38438 VD |
6444 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
6445 | { | |
6446 | long nr_swap_pages = get_nr_swap_pages(); | |
6447 | ||
6448 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
6449 | return nr_swap_pages; | |
6450 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
6451 | nr_swap_pages = min_t(long, nr_swap_pages, | |
bbec2e15 | 6452 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
6453 | page_counter_read(&memcg->swap)); |
6454 | return nr_swap_pages; | |
6455 | } | |
6456 | ||
5ccc5aba VD |
6457 | bool mem_cgroup_swap_full(struct page *page) |
6458 | { | |
6459 | struct mem_cgroup *memcg; | |
6460 | ||
6461 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
6462 | ||
6463 | if (vm_swap_full()) | |
6464 | return true; | |
6465 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
6466 | return false; | |
6467 | ||
6468 | memcg = page->mem_cgroup; | |
6469 | if (!memcg) | |
6470 | return false; | |
6471 | ||
6472 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
bbec2e15 | 6473 | if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.max) |
5ccc5aba VD |
6474 | return true; |
6475 | ||
6476 | return false; | |
6477 | } | |
6478 | ||
21afa38e JW |
6479 | /* for remember boot option*/ |
6480 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
6481 | static int really_do_swap_account __initdata = 1; | |
6482 | #else | |
6483 | static int really_do_swap_account __initdata; | |
6484 | #endif | |
6485 | ||
6486 | static int __init enable_swap_account(char *s) | |
6487 | { | |
6488 | if (!strcmp(s, "1")) | |
6489 | really_do_swap_account = 1; | |
6490 | else if (!strcmp(s, "0")) | |
6491 | really_do_swap_account = 0; | |
6492 | return 1; | |
6493 | } | |
6494 | __setup("swapaccount=", enable_swap_account); | |
6495 | ||
37e84351 VD |
6496 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
6497 | struct cftype *cft) | |
6498 | { | |
6499 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
6500 | ||
6501 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
6502 | } | |
6503 | ||
6504 | static int swap_max_show(struct seq_file *m, void *v) | |
6505 | { | |
6506 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
bbec2e15 | 6507 | unsigned long max = READ_ONCE(memcg->swap.max); |
37e84351 VD |
6508 | |
6509 | if (max == PAGE_COUNTER_MAX) | |
6510 | seq_puts(m, "max\n"); | |
6511 | else | |
6512 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
6513 | ||
6514 | return 0; | |
6515 | } | |
6516 | ||
6517 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
6518 | char *buf, size_t nbytes, loff_t off) | |
6519 | { | |
6520 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6521 | unsigned long max; | |
6522 | int err; | |
6523 | ||
6524 | buf = strstrip(buf); | |
6525 | err = page_counter_memparse(buf, "max", &max); | |
6526 | if (err) | |
6527 | return err; | |
6528 | ||
be09102b | 6529 | xchg(&memcg->swap.max, max); |
37e84351 VD |
6530 | |
6531 | return nbytes; | |
6532 | } | |
6533 | ||
f3a53a3a TH |
6534 | static int swap_events_show(struct seq_file *m, void *v) |
6535 | { | |
6536 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
6537 | ||
6538 | seq_printf(m, "max %lu\n", | |
6539 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
6540 | seq_printf(m, "fail %lu\n", | |
6541 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
6542 | ||
6543 | return 0; | |
6544 | } | |
6545 | ||
37e84351 VD |
6546 | static struct cftype swap_files[] = { |
6547 | { | |
6548 | .name = "swap.current", | |
6549 | .flags = CFTYPE_NOT_ON_ROOT, | |
6550 | .read_u64 = swap_current_read, | |
6551 | }, | |
6552 | { | |
6553 | .name = "swap.max", | |
6554 | .flags = CFTYPE_NOT_ON_ROOT, | |
6555 | .seq_show = swap_max_show, | |
6556 | .write = swap_max_write, | |
6557 | }, | |
f3a53a3a TH |
6558 | { |
6559 | .name = "swap.events", | |
6560 | .flags = CFTYPE_NOT_ON_ROOT, | |
6561 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
6562 | .seq_show = swap_events_show, | |
6563 | }, | |
37e84351 VD |
6564 | { } /* terminate */ |
6565 | }; | |
6566 | ||
21afa38e JW |
6567 | static struct cftype memsw_cgroup_files[] = { |
6568 | { | |
6569 | .name = "memsw.usage_in_bytes", | |
6570 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
6571 | .read_u64 = mem_cgroup_read_u64, | |
6572 | }, | |
6573 | { | |
6574 | .name = "memsw.max_usage_in_bytes", | |
6575 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6576 | .write = mem_cgroup_reset, | |
6577 | .read_u64 = mem_cgroup_read_u64, | |
6578 | }, | |
6579 | { | |
6580 | .name = "memsw.limit_in_bytes", | |
6581 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
6582 | .write = mem_cgroup_write, | |
6583 | .read_u64 = mem_cgroup_read_u64, | |
6584 | }, | |
6585 | { | |
6586 | .name = "memsw.failcnt", | |
6587 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6588 | .write = mem_cgroup_reset, | |
6589 | .read_u64 = mem_cgroup_read_u64, | |
6590 | }, | |
6591 | { }, /* terminate */ | |
6592 | }; | |
6593 | ||
6594 | static int __init mem_cgroup_swap_init(void) | |
6595 | { | |
6596 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
6597 | do_swap_account = 1; | |
37e84351 VD |
6598 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, |
6599 | swap_files)); | |
21afa38e JW |
6600 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
6601 | memsw_cgroup_files)); | |
6602 | } | |
6603 | return 0; | |
6604 | } | |
6605 | subsys_initcall(mem_cgroup_swap_init); | |
6606 | ||
6607 | #endif /* CONFIG_MEMCG_SWAP */ |