]>
Commit | Line | Data |
---|---|---|
c942fddf | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
8cdea7c0 BS |
2 | /* memcontrol.c - Memory Controller |
3 | * | |
4 | * Copyright IBM Corporation, 2007 | |
5 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
6 | * | |
78fb7466 PE |
7 | * Copyright 2007 OpenVZ SWsoft Inc |
8 | * Author: Pavel Emelianov <xemul@openvz.org> | |
9 | * | |
2e72b634 KS |
10 | * Memory thresholds |
11 | * Copyright (C) 2009 Nokia Corporation | |
12 | * Author: Kirill A. Shutemov | |
13 | * | |
7ae1e1d0 GC |
14 | * Kernel Memory Controller |
15 | * Copyright (C) 2012 Parallels Inc. and Google Inc. | |
16 | * Authors: Glauber Costa and Suleiman Souhlal | |
17 | * | |
1575e68b JW |
18 | * Native page reclaim |
19 | * Charge lifetime sanitation | |
20 | * Lockless page tracking & accounting | |
21 | * Unified hierarchy configuration model | |
22 | * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner | |
6168d0da AS |
23 | * |
24 | * Per memcg lru locking | |
25 | * Copyright (C) 2020 Alibaba, Inc, Alex Shi | |
8cdea7c0 BS |
26 | */ |
27 | ||
3e32cb2e | 28 | #include <linux/page_counter.h> |
8cdea7c0 BS |
29 | #include <linux/memcontrol.h> |
30 | #include <linux/cgroup.h> | |
a520110e | 31 | #include <linux/pagewalk.h> |
6e84f315 | 32 | #include <linux/sched/mm.h> |
3a4f8a0b | 33 | #include <linux/shmem_fs.h> |
4ffef5fe | 34 | #include <linux/hugetlb.h> |
d13d1443 | 35 | #include <linux/pagemap.h> |
1ff9e6e1 | 36 | #include <linux/vm_event_item.h> |
d52aa412 | 37 | #include <linux/smp.h> |
8a9f3ccd | 38 | #include <linux/page-flags.h> |
66e1707b | 39 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
40 | #include <linux/bit_spinlock.h> |
41 | #include <linux/rcupdate.h> | |
e222432b | 42 | #include <linux/limits.h> |
b9e15baf | 43 | #include <linux/export.h> |
8c7c6e34 | 44 | #include <linux/mutex.h> |
bb4cc1a8 | 45 | #include <linux/rbtree.h> |
b6ac57d5 | 46 | #include <linux/slab.h> |
66e1707b | 47 | #include <linux/swap.h> |
02491447 | 48 | #include <linux/swapops.h> |
66e1707b | 49 | #include <linux/spinlock.h> |
2e72b634 | 50 | #include <linux/eventfd.h> |
79bd9814 | 51 | #include <linux/poll.h> |
2e72b634 | 52 | #include <linux/sort.h> |
66e1707b | 53 | #include <linux/fs.h> |
d2ceb9b7 | 54 | #include <linux/seq_file.h> |
70ddf637 | 55 | #include <linux/vmpressure.h> |
b69408e8 | 56 | #include <linux/mm_inline.h> |
5d1ea48b | 57 | #include <linux/swap_cgroup.h> |
cdec2e42 | 58 | #include <linux/cpu.h> |
158e0a2d | 59 | #include <linux/oom.h> |
0056f4e6 | 60 | #include <linux/lockdep.h> |
79bd9814 | 61 | #include <linux/file.h> |
b23afb93 | 62 | #include <linux/tracehook.h> |
0e4b01df | 63 | #include <linux/psi.h> |
c8713d0b | 64 | #include <linux/seq_buf.h> |
08e552c6 | 65 | #include "internal.h" |
d1a4c0b3 | 66 | #include <net/sock.h> |
4bd2c1ee | 67 | #include <net/ip.h> |
f35c3a8e | 68 | #include "slab.h" |
8cdea7c0 | 69 | |
7c0f6ba6 | 70 | #include <linux/uaccess.h> |
8697d331 | 71 | |
cc8e970c KM |
72 | #include <trace/events/vmscan.h> |
73 | ||
073219e9 TH |
74 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
75 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 76 | |
7d828602 JW |
77 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
78 | ||
37d5985c RG |
79 | /* Active memory cgroup to use from an interrupt context */ |
80 | DEFINE_PER_CPU(struct mem_cgroup *, int_active_memcg); | |
81 | ||
f7e1cb6e JW |
82 | /* Socket memory accounting disabled? */ |
83 | static bool cgroup_memory_nosocket; | |
84 | ||
04823c83 VD |
85 | /* Kernel memory accounting disabled? */ |
86 | static bool cgroup_memory_nokmem; | |
87 | ||
21afa38e | 88 | /* Whether the swap controller is active */ |
c255a458 | 89 | #ifdef CONFIG_MEMCG_SWAP |
eccb52e7 | 90 | bool cgroup_memory_noswap __read_mostly; |
c077719b | 91 | #else |
eccb52e7 | 92 | #define cgroup_memory_noswap 1 |
2d1c4980 | 93 | #endif |
c077719b | 94 | |
97b27821 TH |
95 | #ifdef CONFIG_CGROUP_WRITEBACK |
96 | static DECLARE_WAIT_QUEUE_HEAD(memcg_cgwb_frn_waitq); | |
97 | #endif | |
98 | ||
7941d214 JW |
99 | /* Whether legacy memory+swap accounting is active */ |
100 | static bool do_memsw_account(void) | |
101 | { | |
eccb52e7 | 102 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_noswap; |
7941d214 JW |
103 | } |
104 | ||
a0db00fc KS |
105 | #define THRESHOLDS_EVENTS_TARGET 128 |
106 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
e9f8974f | 107 | |
bb4cc1a8 AM |
108 | /* |
109 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
110 | * their hierarchy representation | |
111 | */ | |
112 | ||
ef8f2327 | 113 | struct mem_cgroup_tree_per_node { |
bb4cc1a8 | 114 | struct rb_root rb_root; |
fa90b2fd | 115 | struct rb_node *rb_rightmost; |
bb4cc1a8 AM |
116 | spinlock_t lock; |
117 | }; | |
118 | ||
bb4cc1a8 AM |
119 | struct mem_cgroup_tree { |
120 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
121 | }; | |
122 | ||
123 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
124 | ||
9490ff27 KH |
125 | /* for OOM */ |
126 | struct mem_cgroup_eventfd_list { | |
127 | struct list_head list; | |
128 | struct eventfd_ctx *eventfd; | |
129 | }; | |
2e72b634 | 130 | |
79bd9814 TH |
131 | /* |
132 | * cgroup_event represents events which userspace want to receive. | |
133 | */ | |
3bc942f3 | 134 | struct mem_cgroup_event { |
79bd9814 | 135 | /* |
59b6f873 | 136 | * memcg which the event belongs to. |
79bd9814 | 137 | */ |
59b6f873 | 138 | struct mem_cgroup *memcg; |
79bd9814 TH |
139 | /* |
140 | * eventfd to signal userspace about the event. | |
141 | */ | |
142 | struct eventfd_ctx *eventfd; | |
143 | /* | |
144 | * Each of these stored in a list by the cgroup. | |
145 | */ | |
146 | struct list_head list; | |
fba94807 TH |
147 | /* |
148 | * register_event() callback will be used to add new userspace | |
149 | * waiter for changes related to this event. Use eventfd_signal() | |
150 | * on eventfd to send notification to userspace. | |
151 | */ | |
59b6f873 | 152 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 153 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
154 | /* |
155 | * unregister_event() callback will be called when userspace closes | |
156 | * the eventfd or on cgroup removing. This callback must be set, | |
157 | * if you want provide notification functionality. | |
158 | */ | |
59b6f873 | 159 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 160 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
161 | /* |
162 | * All fields below needed to unregister event when | |
163 | * userspace closes eventfd. | |
164 | */ | |
165 | poll_table pt; | |
166 | wait_queue_head_t *wqh; | |
ac6424b9 | 167 | wait_queue_entry_t wait; |
79bd9814 TH |
168 | struct work_struct remove; |
169 | }; | |
170 | ||
c0ff4b85 R |
171 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
172 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 173 | |
7dc74be0 DN |
174 | /* Stuffs for move charges at task migration. */ |
175 | /* | |
1dfab5ab | 176 | * Types of charges to be moved. |
7dc74be0 | 177 | */ |
1dfab5ab JW |
178 | #define MOVE_ANON 0x1U |
179 | #define MOVE_FILE 0x2U | |
180 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 181 | |
4ffef5fe DN |
182 | /* "mc" and its members are protected by cgroup_mutex */ |
183 | static struct move_charge_struct { | |
b1dd693e | 184 | spinlock_t lock; /* for from, to */ |
264a0ae1 | 185 | struct mm_struct *mm; |
4ffef5fe DN |
186 | struct mem_cgroup *from; |
187 | struct mem_cgroup *to; | |
1dfab5ab | 188 | unsigned long flags; |
4ffef5fe | 189 | unsigned long precharge; |
854ffa8d | 190 | unsigned long moved_charge; |
483c30b5 | 191 | unsigned long moved_swap; |
8033b97c DN |
192 | struct task_struct *moving_task; /* a task moving charges */ |
193 | wait_queue_head_t waitq; /* a waitq for other context */ | |
194 | } mc = { | |
2bd9bb20 | 195 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
196 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
197 | }; | |
4ffef5fe | 198 | |
4e416953 BS |
199 | /* |
200 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
201 | * limit reclaim to prevent infinite loops, if they ever occur. | |
202 | */ | |
a0db00fc | 203 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 204 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 205 | |
8c7c6e34 | 206 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
207 | enum res_type { |
208 | _MEM, | |
209 | _MEMSWAP, | |
210 | _OOM_TYPE, | |
510fc4e1 | 211 | _KMEM, |
d55f90bf | 212 | _TCP, |
86ae53e1 GC |
213 | }; |
214 | ||
a0db00fc KS |
215 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
216 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 217 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
218 | /* Used for OOM nofiier */ |
219 | #define OOM_CONTROL (0) | |
8c7c6e34 | 220 | |
b05706f1 KT |
221 | /* |
222 | * Iteration constructs for visiting all cgroups (under a tree). If | |
223 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
224 | * be used for reference counting. | |
225 | */ | |
226 | #define for_each_mem_cgroup_tree(iter, root) \ | |
227 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ | |
228 | iter != NULL; \ | |
229 | iter = mem_cgroup_iter(root, iter, NULL)) | |
230 | ||
231 | #define for_each_mem_cgroup(iter) \ | |
232 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ | |
233 | iter != NULL; \ | |
234 | iter = mem_cgroup_iter(NULL, iter, NULL)) | |
235 | ||
7775face TH |
236 | static inline bool should_force_charge(void) |
237 | { | |
238 | return tsk_is_oom_victim(current) || fatal_signal_pending(current) || | |
239 | (current->flags & PF_EXITING); | |
240 | } | |
241 | ||
70ddf637 AV |
242 | /* Some nice accessors for the vmpressure. */ |
243 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
244 | { | |
245 | if (!memcg) | |
246 | memcg = root_mem_cgroup; | |
247 | return &memcg->vmpressure; | |
248 | } | |
249 | ||
250 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
251 | { | |
252 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
253 | } | |
254 | ||
84c07d11 | 255 | #ifdef CONFIG_MEMCG_KMEM |
bf4f0599 RG |
256 | extern spinlock_t css_set_lock; |
257 | ||
258 | static void obj_cgroup_release(struct percpu_ref *ref) | |
259 | { | |
260 | struct obj_cgroup *objcg = container_of(ref, struct obj_cgroup, refcnt); | |
261 | struct mem_cgroup *memcg; | |
262 | unsigned int nr_bytes; | |
263 | unsigned int nr_pages; | |
264 | unsigned long flags; | |
265 | ||
266 | /* | |
267 | * At this point all allocated objects are freed, and | |
268 | * objcg->nr_charged_bytes can't have an arbitrary byte value. | |
269 | * However, it can be PAGE_SIZE or (x * PAGE_SIZE). | |
270 | * | |
271 | * The following sequence can lead to it: | |
272 | * 1) CPU0: objcg == stock->cached_objcg | |
273 | * 2) CPU1: we do a small allocation (e.g. 92 bytes), | |
274 | * PAGE_SIZE bytes are charged | |
275 | * 3) CPU1: a process from another memcg is allocating something, | |
276 | * the stock if flushed, | |
277 | * objcg->nr_charged_bytes = PAGE_SIZE - 92 | |
278 | * 5) CPU0: we do release this object, | |
279 | * 92 bytes are added to stock->nr_bytes | |
280 | * 6) CPU0: stock is flushed, | |
281 | * 92 bytes are added to objcg->nr_charged_bytes | |
282 | * | |
283 | * In the result, nr_charged_bytes == PAGE_SIZE. | |
284 | * This page will be uncharged in obj_cgroup_release(). | |
285 | */ | |
286 | nr_bytes = atomic_read(&objcg->nr_charged_bytes); | |
287 | WARN_ON_ONCE(nr_bytes & (PAGE_SIZE - 1)); | |
288 | nr_pages = nr_bytes >> PAGE_SHIFT; | |
289 | ||
290 | spin_lock_irqsave(&css_set_lock, flags); | |
291 | memcg = obj_cgroup_memcg(objcg); | |
292 | if (nr_pages) | |
293 | __memcg_kmem_uncharge(memcg, nr_pages); | |
294 | list_del(&objcg->list); | |
295 | mem_cgroup_put(memcg); | |
296 | spin_unlock_irqrestore(&css_set_lock, flags); | |
297 | ||
298 | percpu_ref_exit(ref); | |
299 | kfree_rcu(objcg, rcu); | |
300 | } | |
301 | ||
302 | static struct obj_cgroup *obj_cgroup_alloc(void) | |
303 | { | |
304 | struct obj_cgroup *objcg; | |
305 | int ret; | |
306 | ||
307 | objcg = kzalloc(sizeof(struct obj_cgroup), GFP_KERNEL); | |
308 | if (!objcg) | |
309 | return NULL; | |
310 | ||
311 | ret = percpu_ref_init(&objcg->refcnt, obj_cgroup_release, 0, | |
312 | GFP_KERNEL); | |
313 | if (ret) { | |
314 | kfree(objcg); | |
315 | return NULL; | |
316 | } | |
317 | INIT_LIST_HEAD(&objcg->list); | |
318 | return objcg; | |
319 | } | |
320 | ||
321 | static void memcg_reparent_objcgs(struct mem_cgroup *memcg, | |
322 | struct mem_cgroup *parent) | |
323 | { | |
324 | struct obj_cgroup *objcg, *iter; | |
325 | ||
326 | objcg = rcu_replace_pointer(memcg->objcg, NULL, true); | |
327 | ||
328 | spin_lock_irq(&css_set_lock); | |
329 | ||
330 | /* Move active objcg to the parent's list */ | |
331 | xchg(&objcg->memcg, parent); | |
332 | css_get(&parent->css); | |
333 | list_add(&objcg->list, &parent->objcg_list); | |
334 | ||
335 | /* Move already reparented objcgs to the parent's list */ | |
336 | list_for_each_entry(iter, &memcg->objcg_list, list) { | |
337 | css_get(&parent->css); | |
338 | xchg(&iter->memcg, parent); | |
339 | css_put(&memcg->css); | |
340 | } | |
341 | list_splice(&memcg->objcg_list, &parent->objcg_list); | |
342 | ||
343 | spin_unlock_irq(&css_set_lock); | |
344 | ||
345 | percpu_ref_kill(&objcg->refcnt); | |
346 | } | |
347 | ||
55007d84 | 348 | /* |
9855609b | 349 | * This will be used as a shrinker list's index. |
b8627835 LZ |
350 | * The main reason for not using cgroup id for this: |
351 | * this works better in sparse environments, where we have a lot of memcgs, | |
352 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
353 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
354 | * 200 entry array for that. | |
55007d84 | 355 | * |
dbcf73e2 VD |
356 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
357 | * will double each time we have to increase it. | |
55007d84 | 358 | */ |
dbcf73e2 VD |
359 | static DEFINE_IDA(memcg_cache_ida); |
360 | int memcg_nr_cache_ids; | |
749c5415 | 361 | |
05257a1a VD |
362 | /* Protects memcg_nr_cache_ids */ |
363 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
364 | ||
365 | void memcg_get_cache_ids(void) | |
366 | { | |
367 | down_read(&memcg_cache_ids_sem); | |
368 | } | |
369 | ||
370 | void memcg_put_cache_ids(void) | |
371 | { | |
372 | up_read(&memcg_cache_ids_sem); | |
373 | } | |
374 | ||
55007d84 GC |
375 | /* |
376 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
377 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
378 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
379 | * tunable, but that is strictly not necessary. | |
380 | * | |
b8627835 | 381 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
382 | * this constant directly from cgroup, but it is understandable that this is |
383 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 384 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
385 | * increase ours as well if it increases. |
386 | */ | |
387 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 388 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 389 | |
d7f25f8a GC |
390 | /* |
391 | * A lot of the calls to the cache allocation functions are expected to be | |
272911a4 | 392 | * inlined by the compiler. Since the calls to memcg_slab_pre_alloc_hook() are |
d7f25f8a GC |
393 | * conditional to this static branch, we'll have to allow modules that does |
394 | * kmem_cache_alloc and the such to see this symbol as well | |
395 | */ | |
ef12947c | 396 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
d7f25f8a | 397 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
0a432dcb | 398 | #endif |
17cc4dfe | 399 | |
0a4465d3 KT |
400 | static int memcg_shrinker_map_size; |
401 | static DEFINE_MUTEX(memcg_shrinker_map_mutex); | |
402 | ||
403 | static void memcg_free_shrinker_map_rcu(struct rcu_head *head) | |
404 | { | |
405 | kvfree(container_of(head, struct memcg_shrinker_map, rcu)); | |
406 | } | |
407 | ||
408 | static int memcg_expand_one_shrinker_map(struct mem_cgroup *memcg, | |
409 | int size, int old_size) | |
410 | { | |
411 | struct memcg_shrinker_map *new, *old; | |
412 | int nid; | |
413 | ||
414 | lockdep_assert_held(&memcg_shrinker_map_mutex); | |
415 | ||
416 | for_each_node(nid) { | |
417 | old = rcu_dereference_protected( | |
418 | mem_cgroup_nodeinfo(memcg, nid)->shrinker_map, true); | |
419 | /* Not yet online memcg */ | |
420 | if (!old) | |
421 | return 0; | |
422 | ||
86daf94e | 423 | new = kvmalloc_node(sizeof(*new) + size, GFP_KERNEL, nid); |
0a4465d3 KT |
424 | if (!new) |
425 | return -ENOMEM; | |
426 | ||
427 | /* Set all old bits, clear all new bits */ | |
428 | memset(new->map, (int)0xff, old_size); | |
429 | memset((void *)new->map + old_size, 0, size - old_size); | |
430 | ||
431 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, new); | |
432 | call_rcu(&old->rcu, memcg_free_shrinker_map_rcu); | |
433 | } | |
434 | ||
435 | return 0; | |
436 | } | |
437 | ||
438 | static void memcg_free_shrinker_maps(struct mem_cgroup *memcg) | |
439 | { | |
440 | struct mem_cgroup_per_node *pn; | |
441 | struct memcg_shrinker_map *map; | |
442 | int nid; | |
443 | ||
444 | if (mem_cgroup_is_root(memcg)) | |
445 | return; | |
446 | ||
447 | for_each_node(nid) { | |
448 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
449 | map = rcu_dereference_protected(pn->shrinker_map, true); | |
450 | if (map) | |
451 | kvfree(map); | |
452 | rcu_assign_pointer(pn->shrinker_map, NULL); | |
453 | } | |
454 | } | |
455 | ||
456 | static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg) | |
457 | { | |
458 | struct memcg_shrinker_map *map; | |
459 | int nid, size, ret = 0; | |
460 | ||
461 | if (mem_cgroup_is_root(memcg)) | |
462 | return 0; | |
463 | ||
464 | mutex_lock(&memcg_shrinker_map_mutex); | |
465 | size = memcg_shrinker_map_size; | |
466 | for_each_node(nid) { | |
86daf94e | 467 | map = kvzalloc_node(sizeof(*map) + size, GFP_KERNEL, nid); |
0a4465d3 KT |
468 | if (!map) { |
469 | memcg_free_shrinker_maps(memcg); | |
470 | ret = -ENOMEM; | |
471 | break; | |
472 | } | |
473 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, map); | |
474 | } | |
475 | mutex_unlock(&memcg_shrinker_map_mutex); | |
476 | ||
477 | return ret; | |
478 | } | |
479 | ||
480 | int memcg_expand_shrinker_maps(int new_id) | |
481 | { | |
482 | int size, old_size, ret = 0; | |
483 | struct mem_cgroup *memcg; | |
484 | ||
485 | size = DIV_ROUND_UP(new_id + 1, BITS_PER_LONG) * sizeof(unsigned long); | |
486 | old_size = memcg_shrinker_map_size; | |
487 | if (size <= old_size) | |
488 | return 0; | |
489 | ||
490 | mutex_lock(&memcg_shrinker_map_mutex); | |
491 | if (!root_mem_cgroup) | |
492 | goto unlock; | |
493 | ||
494 | for_each_mem_cgroup(memcg) { | |
495 | if (mem_cgroup_is_root(memcg)) | |
496 | continue; | |
497 | ret = memcg_expand_one_shrinker_map(memcg, size, old_size); | |
75866af6 VA |
498 | if (ret) { |
499 | mem_cgroup_iter_break(NULL, memcg); | |
0a4465d3 | 500 | goto unlock; |
75866af6 | 501 | } |
0a4465d3 KT |
502 | } |
503 | unlock: | |
504 | if (!ret) | |
505 | memcg_shrinker_map_size = size; | |
506 | mutex_unlock(&memcg_shrinker_map_mutex); | |
507 | return ret; | |
508 | } | |
fae91d6d KT |
509 | |
510 | void memcg_set_shrinker_bit(struct mem_cgroup *memcg, int nid, int shrinker_id) | |
511 | { | |
512 | if (shrinker_id >= 0 && memcg && !mem_cgroup_is_root(memcg)) { | |
513 | struct memcg_shrinker_map *map; | |
514 | ||
515 | rcu_read_lock(); | |
516 | map = rcu_dereference(memcg->nodeinfo[nid]->shrinker_map); | |
f90280d6 KT |
517 | /* Pairs with smp mb in shrink_slab() */ |
518 | smp_mb__before_atomic(); | |
fae91d6d KT |
519 | set_bit(shrinker_id, map->map); |
520 | rcu_read_unlock(); | |
521 | } | |
522 | } | |
523 | ||
ad7fa852 TH |
524 | /** |
525 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
526 | * @page: page of interest | |
527 | * | |
528 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
529 | * with @page is returned. The returned css remains associated with @page | |
530 | * until it is released. | |
531 | * | |
532 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
533 | * is returned. | |
ad7fa852 TH |
534 | */ |
535 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
536 | { | |
537 | struct mem_cgroup *memcg; | |
538 | ||
bcfe06bf | 539 | memcg = page_memcg(page); |
ad7fa852 | 540 | |
9e10a130 | 541 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
542 | memcg = root_mem_cgroup; |
543 | ||
ad7fa852 TH |
544 | return &memcg->css; |
545 | } | |
546 | ||
2fc04524 VD |
547 | /** |
548 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
549 | * @page: the page | |
550 | * | |
551 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
552 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
553 | * is safe to call this function without holding a reference to @page. | |
554 | * | |
555 | * Note, this function is inherently racy, because there is nothing to prevent | |
556 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
557 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
558 | * do not care (such as procfs interfaces). | |
559 | */ | |
560 | ino_t page_cgroup_ino(struct page *page) | |
561 | { | |
562 | struct mem_cgroup *memcg; | |
563 | unsigned long ino = 0; | |
564 | ||
565 | rcu_read_lock(); | |
bcfe06bf | 566 | memcg = page_memcg_check(page); |
286e04b8 | 567 | |
2fc04524 VD |
568 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) |
569 | memcg = parent_mem_cgroup(memcg); | |
570 | if (memcg) | |
571 | ino = cgroup_ino(memcg->css.cgroup); | |
572 | rcu_read_unlock(); | |
573 | return ino; | |
574 | } | |
575 | ||
ef8f2327 MG |
576 | static struct mem_cgroup_per_node * |
577 | mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page) | |
f64c3f54 | 578 | { |
97a6c37b | 579 | int nid = page_to_nid(page); |
f64c3f54 | 580 | |
ef8f2327 | 581 | return memcg->nodeinfo[nid]; |
f64c3f54 BS |
582 | } |
583 | ||
ef8f2327 MG |
584 | static struct mem_cgroup_tree_per_node * |
585 | soft_limit_tree_node(int nid) | |
bb4cc1a8 | 586 | { |
ef8f2327 | 587 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
588 | } |
589 | ||
ef8f2327 | 590 | static struct mem_cgroup_tree_per_node * |
bb4cc1a8 AM |
591 | soft_limit_tree_from_page(struct page *page) |
592 | { | |
593 | int nid = page_to_nid(page); | |
bb4cc1a8 | 594 | |
ef8f2327 | 595 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
596 | } |
597 | ||
ef8f2327 MG |
598 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
599 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 600 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
601 | { |
602 | struct rb_node **p = &mctz->rb_root.rb_node; | |
603 | struct rb_node *parent = NULL; | |
ef8f2327 | 604 | struct mem_cgroup_per_node *mz_node; |
fa90b2fd | 605 | bool rightmost = true; |
bb4cc1a8 AM |
606 | |
607 | if (mz->on_tree) | |
608 | return; | |
609 | ||
610 | mz->usage_in_excess = new_usage_in_excess; | |
611 | if (!mz->usage_in_excess) | |
612 | return; | |
613 | while (*p) { | |
614 | parent = *p; | |
ef8f2327 | 615 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 | 616 | tree_node); |
fa90b2fd | 617 | if (mz->usage_in_excess < mz_node->usage_in_excess) { |
bb4cc1a8 | 618 | p = &(*p)->rb_left; |
fa90b2fd | 619 | rightmost = false; |
378876b0 | 620 | } else { |
bb4cc1a8 | 621 | p = &(*p)->rb_right; |
378876b0 | 622 | } |
bb4cc1a8 | 623 | } |
fa90b2fd DB |
624 | |
625 | if (rightmost) | |
626 | mctz->rb_rightmost = &mz->tree_node; | |
627 | ||
bb4cc1a8 AM |
628 | rb_link_node(&mz->tree_node, parent, p); |
629 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
630 | mz->on_tree = true; | |
631 | } | |
632 | ||
ef8f2327 MG |
633 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
634 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
635 | { |
636 | if (!mz->on_tree) | |
637 | return; | |
fa90b2fd DB |
638 | |
639 | if (&mz->tree_node == mctz->rb_rightmost) | |
640 | mctz->rb_rightmost = rb_prev(&mz->tree_node); | |
641 | ||
bb4cc1a8 AM |
642 | rb_erase(&mz->tree_node, &mctz->rb_root); |
643 | mz->on_tree = false; | |
644 | } | |
645 | ||
ef8f2327 MG |
646 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
647 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 648 | { |
0a31bc97 JW |
649 | unsigned long flags; |
650 | ||
651 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 652 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 653 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
654 | } |
655 | ||
3e32cb2e JW |
656 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
657 | { | |
658 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 659 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
660 | unsigned long excess = 0; |
661 | ||
662 | if (nr_pages > soft_limit) | |
663 | excess = nr_pages - soft_limit; | |
664 | ||
665 | return excess; | |
666 | } | |
bb4cc1a8 AM |
667 | |
668 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
669 | { | |
3e32cb2e | 670 | unsigned long excess; |
ef8f2327 MG |
671 | struct mem_cgroup_per_node *mz; |
672 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 673 | |
e231875b | 674 | mctz = soft_limit_tree_from_page(page); |
bfc7228b LD |
675 | if (!mctz) |
676 | return; | |
bb4cc1a8 AM |
677 | /* |
678 | * Necessary to update all ancestors when hierarchy is used. | |
679 | * because their event counter is not touched. | |
680 | */ | |
681 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
ef8f2327 | 682 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
3e32cb2e | 683 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
684 | /* |
685 | * We have to update the tree if mz is on RB-tree or | |
686 | * mem is over its softlimit. | |
687 | */ | |
688 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
689 | unsigned long flags; |
690 | ||
691 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
692 | /* if on-tree, remove it */ |
693 | if (mz->on_tree) | |
cf2c8127 | 694 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
695 | /* |
696 | * Insert again. mz->usage_in_excess will be updated. | |
697 | * If excess is 0, no tree ops. | |
698 | */ | |
cf2c8127 | 699 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 700 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
701 | } |
702 | } | |
703 | } | |
704 | ||
705 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
706 | { | |
ef8f2327 MG |
707 | struct mem_cgroup_tree_per_node *mctz; |
708 | struct mem_cgroup_per_node *mz; | |
709 | int nid; | |
bb4cc1a8 | 710 | |
e231875b | 711 | for_each_node(nid) { |
ef8f2327 MG |
712 | mz = mem_cgroup_nodeinfo(memcg, nid); |
713 | mctz = soft_limit_tree_node(nid); | |
bfc7228b LD |
714 | if (mctz) |
715 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
716 | } |
717 | } | |
718 | ||
ef8f2327 MG |
719 | static struct mem_cgroup_per_node * |
720 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 721 | { |
ef8f2327 | 722 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
723 | |
724 | retry: | |
725 | mz = NULL; | |
fa90b2fd | 726 | if (!mctz->rb_rightmost) |
bb4cc1a8 AM |
727 | goto done; /* Nothing to reclaim from */ |
728 | ||
fa90b2fd DB |
729 | mz = rb_entry(mctz->rb_rightmost, |
730 | struct mem_cgroup_per_node, tree_node); | |
bb4cc1a8 AM |
731 | /* |
732 | * Remove the node now but someone else can add it back, | |
733 | * we will to add it back at the end of reclaim to its correct | |
734 | * position in the tree. | |
735 | */ | |
cf2c8127 | 736 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 737 | if (!soft_limit_excess(mz->memcg) || |
8965aa28 | 738 | !css_tryget(&mz->memcg->css)) |
bb4cc1a8 AM |
739 | goto retry; |
740 | done: | |
741 | return mz; | |
742 | } | |
743 | ||
ef8f2327 MG |
744 | static struct mem_cgroup_per_node * |
745 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 746 | { |
ef8f2327 | 747 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 748 | |
0a31bc97 | 749 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 750 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 751 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
752 | return mz; |
753 | } | |
754 | ||
db9adbcb JW |
755 | /** |
756 | * __mod_memcg_state - update cgroup memory statistics | |
757 | * @memcg: the memory cgroup | |
758 | * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item | |
759 | * @val: delta to add to the counter, can be negative | |
760 | */ | |
761 | void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val) | |
762 | { | |
ea426c2a | 763 | long x, threshold = MEMCG_CHARGE_BATCH; |
db9adbcb JW |
764 | |
765 | if (mem_cgroup_disabled()) | |
766 | return; | |
767 | ||
772616b0 | 768 | if (memcg_stat_item_in_bytes(idx)) |
ea426c2a RG |
769 | threshold <<= PAGE_SHIFT; |
770 | ||
db9adbcb | 771 | x = val + __this_cpu_read(memcg->vmstats_percpu->stat[idx]); |
ea426c2a | 772 | if (unlikely(abs(x) > threshold)) { |
42a30035 JW |
773 | struct mem_cgroup *mi; |
774 | ||
766a4c19 YS |
775 | /* |
776 | * Batch local counters to keep them in sync with | |
777 | * the hierarchical ones. | |
778 | */ | |
779 | __this_cpu_add(memcg->vmstats_local->stat[idx], x); | |
42a30035 JW |
780 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
781 | atomic_long_add(x, &mi->vmstats[idx]); | |
db9adbcb JW |
782 | x = 0; |
783 | } | |
784 | __this_cpu_write(memcg->vmstats_percpu->stat[idx], x); | |
785 | } | |
786 | ||
42a30035 JW |
787 | static struct mem_cgroup_per_node * |
788 | parent_nodeinfo(struct mem_cgroup_per_node *pn, int nid) | |
789 | { | |
790 | struct mem_cgroup *parent; | |
791 | ||
792 | parent = parent_mem_cgroup(pn->memcg); | |
793 | if (!parent) | |
794 | return NULL; | |
795 | return mem_cgroup_nodeinfo(parent, nid); | |
796 | } | |
797 | ||
eedc4e5a RG |
798 | void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, |
799 | int val) | |
db9adbcb JW |
800 | { |
801 | struct mem_cgroup_per_node *pn; | |
42a30035 | 802 | struct mem_cgroup *memcg; |
ea426c2a | 803 | long x, threshold = MEMCG_CHARGE_BATCH; |
db9adbcb | 804 | |
db9adbcb | 805 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
42a30035 | 806 | memcg = pn->memcg; |
db9adbcb JW |
807 | |
808 | /* Update memcg */ | |
42a30035 | 809 | __mod_memcg_state(memcg, idx, val); |
db9adbcb | 810 | |
b4c46484 RG |
811 | /* Update lruvec */ |
812 | __this_cpu_add(pn->lruvec_stat_local->count[idx], val); | |
813 | ||
ea426c2a RG |
814 | if (vmstat_item_in_bytes(idx)) |
815 | threshold <<= PAGE_SHIFT; | |
816 | ||
db9adbcb | 817 | x = val + __this_cpu_read(pn->lruvec_stat_cpu->count[idx]); |
ea426c2a | 818 | if (unlikely(abs(x) > threshold)) { |
eedc4e5a | 819 | pg_data_t *pgdat = lruvec_pgdat(lruvec); |
42a30035 JW |
820 | struct mem_cgroup_per_node *pi; |
821 | ||
42a30035 JW |
822 | for (pi = pn; pi; pi = parent_nodeinfo(pi, pgdat->node_id)) |
823 | atomic_long_add(x, &pi->lruvec_stat[idx]); | |
db9adbcb JW |
824 | x = 0; |
825 | } | |
826 | __this_cpu_write(pn->lruvec_stat_cpu->count[idx], x); | |
827 | } | |
828 | ||
eedc4e5a RG |
829 | /** |
830 | * __mod_lruvec_state - update lruvec memory statistics | |
831 | * @lruvec: the lruvec | |
832 | * @idx: the stat item | |
833 | * @val: delta to add to the counter, can be negative | |
834 | * | |
835 | * The lruvec is the intersection of the NUMA node and a cgroup. This | |
836 | * function updates the all three counters that are affected by a | |
837 | * change of state at this level: per-node, per-cgroup, per-lruvec. | |
838 | */ | |
839 | void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, | |
840 | int val) | |
841 | { | |
842 | /* Update node */ | |
843 | __mod_node_page_state(lruvec_pgdat(lruvec), idx, val); | |
844 | ||
845 | /* Update memcg and lruvec */ | |
846 | if (!mem_cgroup_disabled()) | |
847 | __mod_memcg_lruvec_state(lruvec, idx, val); | |
848 | } | |
849 | ||
c47d5032 SB |
850 | void __mod_lruvec_page_state(struct page *page, enum node_stat_item idx, |
851 | int val) | |
852 | { | |
853 | struct page *head = compound_head(page); /* rmap on tail pages */ | |
d635a69d | 854 | struct mem_cgroup *memcg = page_memcg(head); |
c47d5032 SB |
855 | pg_data_t *pgdat = page_pgdat(page); |
856 | struct lruvec *lruvec; | |
857 | ||
858 | /* Untracked pages have no memcg, no lruvec. Update only the node */ | |
d635a69d | 859 | if (!memcg) { |
c47d5032 SB |
860 | __mod_node_page_state(pgdat, idx, val); |
861 | return; | |
862 | } | |
863 | ||
d635a69d | 864 | lruvec = mem_cgroup_lruvec(memcg, pgdat); |
c47d5032 SB |
865 | __mod_lruvec_state(lruvec, idx, val); |
866 | } | |
f0c0c115 | 867 | EXPORT_SYMBOL(__mod_lruvec_page_state); |
c47d5032 | 868 | |
da3ceeff | 869 | void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, int val) |
ec9f0238 | 870 | { |
4f103c63 | 871 | pg_data_t *pgdat = page_pgdat(virt_to_page(p)); |
ec9f0238 RG |
872 | struct mem_cgroup *memcg; |
873 | struct lruvec *lruvec; | |
874 | ||
875 | rcu_read_lock(); | |
4f103c63 | 876 | memcg = mem_cgroup_from_obj(p); |
ec9f0238 | 877 | |
8faeb1ff MS |
878 | /* |
879 | * Untracked pages have no memcg, no lruvec. Update only the | |
880 | * node. If we reparent the slab objects to the root memcg, | |
881 | * when we free the slab object, we need to update the per-memcg | |
882 | * vmstats to keep it correct for the root memcg. | |
883 | */ | |
884 | if (!memcg) { | |
ec9f0238 RG |
885 | __mod_node_page_state(pgdat, idx, val); |
886 | } else { | |
867e5e1d | 887 | lruvec = mem_cgroup_lruvec(memcg, pgdat); |
ec9f0238 RG |
888 | __mod_lruvec_state(lruvec, idx, val); |
889 | } | |
890 | rcu_read_unlock(); | |
891 | } | |
892 | ||
db9adbcb JW |
893 | /** |
894 | * __count_memcg_events - account VM events in a cgroup | |
895 | * @memcg: the memory cgroup | |
896 | * @idx: the event item | |
897 | * @count: the number of events that occured | |
898 | */ | |
899 | void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, | |
900 | unsigned long count) | |
901 | { | |
902 | unsigned long x; | |
903 | ||
904 | if (mem_cgroup_disabled()) | |
905 | return; | |
906 | ||
907 | x = count + __this_cpu_read(memcg->vmstats_percpu->events[idx]); | |
908 | if (unlikely(x > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
909 | struct mem_cgroup *mi; |
910 | ||
766a4c19 YS |
911 | /* |
912 | * Batch local counters to keep them in sync with | |
913 | * the hierarchical ones. | |
914 | */ | |
915 | __this_cpu_add(memcg->vmstats_local->events[idx], x); | |
42a30035 JW |
916 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
917 | atomic_long_add(x, &mi->vmevents[idx]); | |
db9adbcb JW |
918 | x = 0; |
919 | } | |
920 | __this_cpu_write(memcg->vmstats_percpu->events[idx], x); | |
921 | } | |
922 | ||
42a30035 | 923 | static unsigned long memcg_events(struct mem_cgroup *memcg, int event) |
e9f8974f | 924 | { |
871789d4 | 925 | return atomic_long_read(&memcg->vmevents[event]); |
e9f8974f JW |
926 | } |
927 | ||
42a30035 JW |
928 | static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event) |
929 | { | |
815744d7 JW |
930 | long x = 0; |
931 | int cpu; | |
932 | ||
933 | for_each_possible_cpu(cpu) | |
934 | x += per_cpu(memcg->vmstats_local->events[event], cpu); | |
935 | return x; | |
42a30035 JW |
936 | } |
937 | ||
c0ff4b85 | 938 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 939 | struct page *page, |
3fba69a5 | 940 | int nr_pages) |
d52aa412 | 941 | { |
e401f176 KH |
942 | /* pagein of a big page is an event. So, ignore page size */ |
943 | if (nr_pages > 0) | |
c9019e9b | 944 | __count_memcg_events(memcg, PGPGIN, 1); |
3751d604 | 945 | else { |
c9019e9b | 946 | __count_memcg_events(memcg, PGPGOUT, 1); |
3751d604 KH |
947 | nr_pages = -nr_pages; /* for event */ |
948 | } | |
e401f176 | 949 | |
871789d4 | 950 | __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages); |
6d12e2d8 KH |
951 | } |
952 | ||
f53d7ce3 JW |
953 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
954 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
955 | { |
956 | unsigned long val, next; | |
957 | ||
871789d4 CD |
958 | val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events); |
959 | next = __this_cpu_read(memcg->vmstats_percpu->targets[target]); | |
7a159cc9 | 960 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 961 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
962 | switch (target) { |
963 | case MEM_CGROUP_TARGET_THRESH: | |
964 | next = val + THRESHOLDS_EVENTS_TARGET; | |
965 | break; | |
bb4cc1a8 AM |
966 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
967 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
968 | break; | |
f53d7ce3 JW |
969 | default: |
970 | break; | |
971 | } | |
871789d4 | 972 | __this_cpu_write(memcg->vmstats_percpu->targets[target], next); |
f53d7ce3 | 973 | return true; |
7a159cc9 | 974 | } |
f53d7ce3 | 975 | return false; |
d2265e6f KH |
976 | } |
977 | ||
978 | /* | |
979 | * Check events in order. | |
980 | * | |
981 | */ | |
c0ff4b85 | 982 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
983 | { |
984 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
985 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
986 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 987 | bool do_softlimit; |
f53d7ce3 | 988 | |
bb4cc1a8 AM |
989 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
990 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
c0ff4b85 | 991 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
992 | if (unlikely(do_softlimit)) |
993 | mem_cgroup_update_tree(memcg, page); | |
0a31bc97 | 994 | } |
d2265e6f KH |
995 | } |
996 | ||
cf475ad2 | 997 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 998 | { |
31a78f23 BS |
999 | /* |
1000 | * mm_update_next_owner() may clear mm->owner to NULL | |
1001 | * if it races with swapoff, page migration, etc. | |
1002 | * So this can be called with p == NULL. | |
1003 | */ | |
1004 | if (unlikely(!p)) | |
1005 | return NULL; | |
1006 | ||
073219e9 | 1007 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 1008 | } |
33398cf2 | 1009 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 1010 | |
d46eb14b SB |
1011 | /** |
1012 | * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg. | |
1013 | * @mm: mm from which memcg should be extracted. It can be NULL. | |
1014 | * | |
1015 | * Obtain a reference on mm->memcg and returns it if successful. Otherwise | |
1016 | * root_mem_cgroup is returned. However if mem_cgroup is disabled, NULL is | |
1017 | * returned. | |
1018 | */ | |
1019 | struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) | |
54595fe2 | 1020 | { |
d46eb14b SB |
1021 | struct mem_cgroup *memcg; |
1022 | ||
1023 | if (mem_cgroup_disabled()) | |
1024 | return NULL; | |
0b7f569e | 1025 | |
54595fe2 KH |
1026 | rcu_read_lock(); |
1027 | do { | |
6f6acb00 MH |
1028 | /* |
1029 | * Page cache insertions can happen withou an | |
1030 | * actual mm context, e.g. during disk probing | |
1031 | * on boot, loopback IO, acct() writes etc. | |
1032 | */ | |
1033 | if (unlikely(!mm)) | |
df381975 | 1034 | memcg = root_mem_cgroup; |
6f6acb00 MH |
1035 | else { |
1036 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
1037 | if (unlikely(!memcg)) | |
1038 | memcg = root_mem_cgroup; | |
1039 | } | |
00d484f3 | 1040 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 1041 | rcu_read_unlock(); |
c0ff4b85 | 1042 | return memcg; |
54595fe2 | 1043 | } |
d46eb14b SB |
1044 | EXPORT_SYMBOL(get_mem_cgroup_from_mm); |
1045 | ||
f745c6f5 SB |
1046 | /** |
1047 | * get_mem_cgroup_from_page: Obtain a reference on given page's memcg. | |
1048 | * @page: page from which memcg should be extracted. | |
1049 | * | |
1050 | * Obtain a reference on page->memcg and returns it if successful. Otherwise | |
1051 | * root_mem_cgroup is returned. | |
1052 | */ | |
1053 | struct mem_cgroup *get_mem_cgroup_from_page(struct page *page) | |
1054 | { | |
bcfe06bf | 1055 | struct mem_cgroup *memcg = page_memcg(page); |
f745c6f5 SB |
1056 | |
1057 | if (mem_cgroup_disabled()) | |
1058 | return NULL; | |
1059 | ||
1060 | rcu_read_lock(); | |
8965aa28 SB |
1061 | /* Page should not get uncharged and freed memcg under us. */ |
1062 | if (!memcg || WARN_ON_ONCE(!css_tryget(&memcg->css))) | |
f745c6f5 SB |
1063 | memcg = root_mem_cgroup; |
1064 | rcu_read_unlock(); | |
1065 | return memcg; | |
1066 | } | |
1067 | EXPORT_SYMBOL(get_mem_cgroup_from_page); | |
1068 | ||
37d5985c | 1069 | static __always_inline struct mem_cgroup *active_memcg(void) |
d46eb14b | 1070 | { |
37d5985c RG |
1071 | if (in_interrupt()) |
1072 | return this_cpu_read(int_active_memcg); | |
1073 | else | |
1074 | return current->active_memcg; | |
1075 | } | |
279c3393 | 1076 | |
37d5985c RG |
1077 | static __always_inline struct mem_cgroup *get_active_memcg(void) |
1078 | { | |
1079 | struct mem_cgroup *memcg; | |
d46eb14b | 1080 | |
37d5985c RG |
1081 | rcu_read_lock(); |
1082 | memcg = active_memcg(); | |
1083 | if (memcg) { | |
8965aa28 | 1084 | /* current->active_memcg must hold a ref. */ |
37d5985c | 1085 | if (WARN_ON_ONCE(!css_tryget(&memcg->css))) |
8965aa28 SB |
1086 | memcg = root_mem_cgroup; |
1087 | else | |
d46eb14b | 1088 | memcg = current->active_memcg; |
d46eb14b | 1089 | } |
37d5985c RG |
1090 | rcu_read_unlock(); |
1091 | ||
1092 | return memcg; | |
1093 | } | |
1094 | ||
4127c650 RG |
1095 | static __always_inline bool memcg_kmem_bypass(void) |
1096 | { | |
1097 | /* Allow remote memcg charging from any context. */ | |
1098 | if (unlikely(active_memcg())) | |
1099 | return false; | |
1100 | ||
1101 | /* Memcg to charge can't be determined. */ | |
1102 | if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD)) | |
1103 | return true; | |
1104 | ||
1105 | return false; | |
1106 | } | |
1107 | ||
37d5985c RG |
1108 | /** |
1109 | * If active memcg is set, do not fallback to current->mm->memcg. | |
1110 | */ | |
1111 | static __always_inline struct mem_cgroup *get_mem_cgroup_from_current(void) | |
1112 | { | |
1113 | if (memcg_kmem_bypass()) | |
1114 | return NULL; | |
1115 | ||
1116 | if (unlikely(active_memcg())) | |
1117 | return get_active_memcg(); | |
1118 | ||
d46eb14b SB |
1119 | return get_mem_cgroup_from_mm(current->mm); |
1120 | } | |
54595fe2 | 1121 | |
5660048c JW |
1122 | /** |
1123 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1124 | * @root: hierarchy root | |
1125 | * @prev: previously returned memcg, NULL on first invocation | |
1126 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1127 | * | |
1128 | * Returns references to children of the hierarchy below @root, or | |
1129 | * @root itself, or %NULL after a full round-trip. | |
1130 | * | |
1131 | * Caller must pass the return value in @prev on subsequent | |
1132 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1133 | * to cancel a hierarchy walk before the round-trip is complete. | |
1134 | * | |
05bdc520 ML |
1135 | * Reclaimers can specify a node in @reclaim to divide up the memcgs |
1136 | * in the hierarchy among all concurrent reclaimers operating on the | |
1137 | * same node. | |
5660048c | 1138 | */ |
694fbc0f | 1139 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1140 | struct mem_cgroup *prev, |
694fbc0f | 1141 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1142 | { |
3f649ab7 | 1143 | struct mem_cgroup_reclaim_iter *iter; |
5ac8fb31 | 1144 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 1145 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1146 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1147 | |
694fbc0f AM |
1148 | if (mem_cgroup_disabled()) |
1149 | return NULL; | |
5660048c | 1150 | |
9f3a0d09 JW |
1151 | if (!root) |
1152 | root = root_mem_cgroup; | |
7d74b06f | 1153 | |
9f3a0d09 | 1154 | if (prev && !reclaim) |
5ac8fb31 | 1155 | pos = prev; |
14067bb3 | 1156 | |
542f85f9 | 1157 | rcu_read_lock(); |
5f578161 | 1158 | |
5ac8fb31 | 1159 | if (reclaim) { |
ef8f2327 | 1160 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 1161 | |
ef8f2327 | 1162 | mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id); |
9da83f3f | 1163 | iter = &mz->iter; |
5ac8fb31 JW |
1164 | |
1165 | if (prev && reclaim->generation != iter->generation) | |
1166 | goto out_unlock; | |
1167 | ||
6df38689 | 1168 | while (1) { |
4db0c3c2 | 1169 | pos = READ_ONCE(iter->position); |
6df38689 VD |
1170 | if (!pos || css_tryget(&pos->css)) |
1171 | break; | |
5ac8fb31 | 1172 | /* |
6df38689 VD |
1173 | * css reference reached zero, so iter->position will |
1174 | * be cleared by ->css_released. However, we should not | |
1175 | * rely on this happening soon, because ->css_released | |
1176 | * is called from a work queue, and by busy-waiting we | |
1177 | * might block it. So we clear iter->position right | |
1178 | * away. | |
5ac8fb31 | 1179 | */ |
6df38689 VD |
1180 | (void)cmpxchg(&iter->position, pos, NULL); |
1181 | } | |
5ac8fb31 JW |
1182 | } |
1183 | ||
1184 | if (pos) | |
1185 | css = &pos->css; | |
1186 | ||
1187 | for (;;) { | |
1188 | css = css_next_descendant_pre(css, &root->css); | |
1189 | if (!css) { | |
1190 | /* | |
1191 | * Reclaimers share the hierarchy walk, and a | |
1192 | * new one might jump in right at the end of | |
1193 | * the hierarchy - make sure they see at least | |
1194 | * one group and restart from the beginning. | |
1195 | */ | |
1196 | if (!prev) | |
1197 | continue; | |
1198 | break; | |
527a5ec9 | 1199 | } |
7d74b06f | 1200 | |
5ac8fb31 JW |
1201 | /* |
1202 | * Verify the css and acquire a reference. The root | |
1203 | * is provided by the caller, so we know it's alive | |
1204 | * and kicking, and don't take an extra reference. | |
1205 | */ | |
1206 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 1207 | |
5ac8fb31 JW |
1208 | if (css == &root->css) |
1209 | break; | |
14067bb3 | 1210 | |
0b8f73e1 JW |
1211 | if (css_tryget(css)) |
1212 | break; | |
9f3a0d09 | 1213 | |
5ac8fb31 | 1214 | memcg = NULL; |
9f3a0d09 | 1215 | } |
5ac8fb31 JW |
1216 | |
1217 | if (reclaim) { | |
5ac8fb31 | 1218 | /* |
6df38689 VD |
1219 | * The position could have already been updated by a competing |
1220 | * thread, so check that the value hasn't changed since we read | |
1221 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 1222 | */ |
6df38689 VD |
1223 | (void)cmpxchg(&iter->position, pos, memcg); |
1224 | ||
5ac8fb31 JW |
1225 | if (pos) |
1226 | css_put(&pos->css); | |
1227 | ||
1228 | if (!memcg) | |
1229 | iter->generation++; | |
1230 | else if (!prev) | |
1231 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1232 | } |
5ac8fb31 | 1233 | |
542f85f9 MH |
1234 | out_unlock: |
1235 | rcu_read_unlock(); | |
c40046f3 MH |
1236 | if (prev && prev != root) |
1237 | css_put(&prev->css); | |
1238 | ||
9f3a0d09 | 1239 | return memcg; |
14067bb3 | 1240 | } |
7d74b06f | 1241 | |
5660048c JW |
1242 | /** |
1243 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1244 | * @root: hierarchy root | |
1245 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1246 | */ | |
1247 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1248 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1249 | { |
1250 | if (!root) | |
1251 | root = root_mem_cgroup; | |
1252 | if (prev && prev != root) | |
1253 | css_put(&prev->css); | |
1254 | } | |
7d74b06f | 1255 | |
54a83d6b MC |
1256 | static void __invalidate_reclaim_iterators(struct mem_cgroup *from, |
1257 | struct mem_cgroup *dead_memcg) | |
6df38689 | 1258 | { |
6df38689 | 1259 | struct mem_cgroup_reclaim_iter *iter; |
ef8f2327 MG |
1260 | struct mem_cgroup_per_node *mz; |
1261 | int nid; | |
6df38689 | 1262 | |
54a83d6b MC |
1263 | for_each_node(nid) { |
1264 | mz = mem_cgroup_nodeinfo(from, nid); | |
9da83f3f YS |
1265 | iter = &mz->iter; |
1266 | cmpxchg(&iter->position, dead_memcg, NULL); | |
6df38689 VD |
1267 | } |
1268 | } | |
1269 | ||
54a83d6b MC |
1270 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1271 | { | |
1272 | struct mem_cgroup *memcg = dead_memcg; | |
1273 | struct mem_cgroup *last; | |
1274 | ||
1275 | do { | |
1276 | __invalidate_reclaim_iterators(memcg, dead_memcg); | |
1277 | last = memcg; | |
1278 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1279 | ||
1280 | /* | |
1281 | * When cgruop1 non-hierarchy mode is used, | |
1282 | * parent_mem_cgroup() does not walk all the way up to the | |
1283 | * cgroup root (root_mem_cgroup). So we have to handle | |
1284 | * dead_memcg from cgroup root separately. | |
1285 | */ | |
1286 | if (last != root_mem_cgroup) | |
1287 | __invalidate_reclaim_iterators(root_mem_cgroup, | |
1288 | dead_memcg); | |
1289 | } | |
1290 | ||
7c5f64f8 VD |
1291 | /** |
1292 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1293 | * @memcg: hierarchy root | |
1294 | * @fn: function to call for each task | |
1295 | * @arg: argument passed to @fn | |
1296 | * | |
1297 | * This function iterates over tasks attached to @memcg or to any of its | |
1298 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
1299 | * value, the function breaks the iteration loop and returns the value. | |
1300 | * Otherwise, it will iterate over all tasks and return 0. | |
1301 | * | |
1302 | * This function must not be called for the root memory cgroup. | |
1303 | */ | |
1304 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
1305 | int (*fn)(struct task_struct *, void *), void *arg) | |
1306 | { | |
1307 | struct mem_cgroup *iter; | |
1308 | int ret = 0; | |
1309 | ||
1310 | BUG_ON(memcg == root_mem_cgroup); | |
1311 | ||
1312 | for_each_mem_cgroup_tree(iter, memcg) { | |
1313 | struct css_task_iter it; | |
1314 | struct task_struct *task; | |
1315 | ||
f168a9a5 | 1316 | css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it); |
7c5f64f8 VD |
1317 | while (!ret && (task = css_task_iter_next(&it))) |
1318 | ret = fn(task, arg); | |
1319 | css_task_iter_end(&it); | |
1320 | if (ret) { | |
1321 | mem_cgroup_iter_break(memcg, iter); | |
1322 | break; | |
1323 | } | |
1324 | } | |
1325 | return ret; | |
1326 | } | |
1327 | ||
6168d0da AS |
1328 | #ifdef CONFIG_DEBUG_VM |
1329 | void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page) | |
1330 | { | |
1331 | struct mem_cgroup *memcg; | |
1332 | ||
1333 | if (mem_cgroup_disabled()) | |
1334 | return; | |
1335 | ||
1336 | memcg = page_memcg(page); | |
1337 | ||
1338 | if (!memcg) | |
1339 | VM_BUG_ON_PAGE(lruvec_memcg(lruvec) != root_mem_cgroup, page); | |
1340 | else | |
1341 | VM_BUG_ON_PAGE(lruvec_memcg(lruvec) != memcg, page); | |
1342 | } | |
1343 | #endif | |
1344 | ||
6168d0da AS |
1345 | /** |
1346 | * lock_page_lruvec - lock and return lruvec for a given page. | |
1347 | * @page: the page | |
1348 | * | |
1349 | * This series functions should be used in either conditions: | |
1350 | * PageLRU is cleared or unset | |
1351 | * or page->_refcount is zero | |
1352 | * or page is locked. | |
1353 | */ | |
1354 | struct lruvec *lock_page_lruvec(struct page *page) | |
1355 | { | |
1356 | struct lruvec *lruvec; | |
1357 | struct pglist_data *pgdat = page_pgdat(page); | |
1358 | ||
1359 | rcu_read_lock(); | |
1360 | lruvec = mem_cgroup_page_lruvec(page, pgdat); | |
1361 | spin_lock(&lruvec->lru_lock); | |
1362 | rcu_read_unlock(); | |
1363 | ||
1364 | lruvec_memcg_debug(lruvec, page); | |
1365 | ||
1366 | return lruvec; | |
1367 | } | |
1368 | ||
1369 | struct lruvec *lock_page_lruvec_irq(struct page *page) | |
1370 | { | |
1371 | struct lruvec *lruvec; | |
1372 | struct pglist_data *pgdat = page_pgdat(page); | |
1373 | ||
1374 | rcu_read_lock(); | |
1375 | lruvec = mem_cgroup_page_lruvec(page, pgdat); | |
1376 | spin_lock_irq(&lruvec->lru_lock); | |
1377 | rcu_read_unlock(); | |
1378 | ||
1379 | lruvec_memcg_debug(lruvec, page); | |
1380 | ||
1381 | return lruvec; | |
1382 | } | |
1383 | ||
1384 | struct lruvec *lock_page_lruvec_irqsave(struct page *page, unsigned long *flags) | |
1385 | { | |
1386 | struct lruvec *lruvec; | |
1387 | struct pglist_data *pgdat = page_pgdat(page); | |
1388 | ||
1389 | rcu_read_lock(); | |
1390 | lruvec = mem_cgroup_page_lruvec(page, pgdat); | |
1391 | spin_lock_irqsave(&lruvec->lru_lock, *flags); | |
1392 | rcu_read_unlock(); | |
1393 | ||
1394 | lruvec_memcg_debug(lruvec, page); | |
1395 | ||
1396 | return lruvec; | |
1397 | } | |
1398 | ||
925b7673 | 1399 | /** |
fa9add64 HD |
1400 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1401 | * @lruvec: mem_cgroup per zone lru vector | |
1402 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1403 | * @zid: zone id of the accounted pages |
fa9add64 | 1404 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1405 | * |
ca707239 HD |
1406 | * This function must be called under lru_lock, just before a page is added |
1407 | * to or just after a page is removed from an lru list (that ordering being | |
1408 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 1409 | */ |
fa9add64 | 1410 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1411 | int zid, int nr_pages) |
3f58a829 | 1412 | { |
ef8f2327 | 1413 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1414 | unsigned long *lru_size; |
ca707239 | 1415 | long size; |
3f58a829 MK |
1416 | |
1417 | if (mem_cgroup_disabled()) | |
1418 | return; | |
1419 | ||
ef8f2327 | 1420 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1421 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1422 | |
1423 | if (nr_pages < 0) | |
1424 | *lru_size += nr_pages; | |
1425 | ||
1426 | size = *lru_size; | |
b4536f0c MH |
1427 | if (WARN_ONCE(size < 0, |
1428 | "%s(%p, %d, %d): lru_size %ld\n", | |
1429 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1430 | VM_BUG_ON(1); |
1431 | *lru_size = 0; | |
1432 | } | |
1433 | ||
1434 | if (nr_pages > 0) | |
1435 | *lru_size += nr_pages; | |
08e552c6 | 1436 | } |
544122e5 | 1437 | |
19942822 | 1438 | /** |
9d11ea9f | 1439 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1440 | * @memcg: the memory cgroup |
19942822 | 1441 | * |
9d11ea9f | 1442 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1443 | * pages. |
19942822 | 1444 | */ |
c0ff4b85 | 1445 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1446 | { |
3e32cb2e JW |
1447 | unsigned long margin = 0; |
1448 | unsigned long count; | |
1449 | unsigned long limit; | |
9d11ea9f | 1450 | |
3e32cb2e | 1451 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1452 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1453 | if (count < limit) |
1454 | margin = limit - count; | |
1455 | ||
7941d214 | 1456 | if (do_memsw_account()) { |
3e32cb2e | 1457 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1458 | limit = READ_ONCE(memcg->memsw.max); |
1c4448ed | 1459 | if (count < limit) |
3e32cb2e | 1460 | margin = min(margin, limit - count); |
cbedbac3 LR |
1461 | else |
1462 | margin = 0; | |
3e32cb2e JW |
1463 | } |
1464 | ||
1465 | return margin; | |
19942822 JW |
1466 | } |
1467 | ||
32047e2a | 1468 | /* |
bdcbb659 | 1469 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1470 | * |
bdcbb659 QH |
1471 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1472 | * moving cgroups. This is for waiting at high-memory pressure | |
1473 | * caused by "move". | |
32047e2a | 1474 | */ |
c0ff4b85 | 1475 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1476 | { |
2bd9bb20 KH |
1477 | struct mem_cgroup *from; |
1478 | struct mem_cgroup *to; | |
4b534334 | 1479 | bool ret = false; |
2bd9bb20 KH |
1480 | /* |
1481 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1482 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1483 | */ | |
1484 | spin_lock(&mc.lock); | |
1485 | from = mc.from; | |
1486 | to = mc.to; | |
1487 | if (!from) | |
1488 | goto unlock; | |
3e92041d | 1489 | |
2314b42d JW |
1490 | ret = mem_cgroup_is_descendant(from, memcg) || |
1491 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1492 | unlock: |
1493 | spin_unlock(&mc.lock); | |
4b534334 KH |
1494 | return ret; |
1495 | } | |
1496 | ||
c0ff4b85 | 1497 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1498 | { |
1499 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1500 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1501 | DEFINE_WAIT(wait); |
1502 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1503 | /* moving charge context might have finished. */ | |
1504 | if (mc.moving_task) | |
1505 | schedule(); | |
1506 | finish_wait(&mc.waitq, &wait); | |
1507 | return true; | |
1508 | } | |
1509 | } | |
1510 | return false; | |
1511 | } | |
1512 | ||
5f9a4f4a MS |
1513 | struct memory_stat { |
1514 | const char *name; | |
1515 | unsigned int ratio; | |
1516 | unsigned int idx; | |
1517 | }; | |
1518 | ||
1519 | static struct memory_stat memory_stats[] = { | |
1520 | { "anon", PAGE_SIZE, NR_ANON_MAPPED }, | |
1521 | { "file", PAGE_SIZE, NR_FILE_PAGES }, | |
1522 | { "kernel_stack", 1024, NR_KERNEL_STACK_KB }, | |
f0c0c115 | 1523 | { "pagetables", PAGE_SIZE, NR_PAGETABLE }, |
5f9a4f4a MS |
1524 | { "percpu", 1, MEMCG_PERCPU_B }, |
1525 | { "sock", PAGE_SIZE, MEMCG_SOCK }, | |
1526 | { "shmem", PAGE_SIZE, NR_SHMEM }, | |
1527 | { "file_mapped", PAGE_SIZE, NR_FILE_MAPPED }, | |
1528 | { "file_dirty", PAGE_SIZE, NR_FILE_DIRTY }, | |
1529 | { "file_writeback", PAGE_SIZE, NR_WRITEBACK }, | |
1530 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
1531 | /* | |
1532 | * The ratio will be initialized in memory_stats_init(). Because | |
1533 | * on some architectures, the macro of HPAGE_PMD_SIZE is not | |
1534 | * constant(e.g. powerpc). | |
1535 | */ | |
69473e5d | 1536 | { "anon_thp", PAGE_SIZE, NR_ANON_THPS }, |
bf9ecead | 1537 | { "file_thp", PAGE_SIZE, NR_FILE_THPS }, |
b8eddff8 | 1538 | { "shmem_thp", 0, NR_SHMEM_THPS }, |
5f9a4f4a MS |
1539 | #endif |
1540 | { "inactive_anon", PAGE_SIZE, NR_INACTIVE_ANON }, | |
1541 | { "active_anon", PAGE_SIZE, NR_ACTIVE_ANON }, | |
1542 | { "inactive_file", PAGE_SIZE, NR_INACTIVE_FILE }, | |
1543 | { "active_file", PAGE_SIZE, NR_ACTIVE_FILE }, | |
1544 | { "unevictable", PAGE_SIZE, NR_UNEVICTABLE }, | |
1545 | ||
1546 | /* | |
1547 | * Note: The slab_reclaimable and slab_unreclaimable must be | |
1548 | * together and slab_reclaimable must be in front. | |
1549 | */ | |
1550 | { "slab_reclaimable", 1, NR_SLAB_RECLAIMABLE_B }, | |
1551 | { "slab_unreclaimable", 1, NR_SLAB_UNRECLAIMABLE_B }, | |
1552 | ||
1553 | /* The memory events */ | |
1554 | { "workingset_refault_anon", 1, WORKINGSET_REFAULT_ANON }, | |
1555 | { "workingset_refault_file", 1, WORKINGSET_REFAULT_FILE }, | |
1556 | { "workingset_activate_anon", 1, WORKINGSET_ACTIVATE_ANON }, | |
1557 | { "workingset_activate_file", 1, WORKINGSET_ACTIVATE_FILE }, | |
1558 | { "workingset_restore_anon", 1, WORKINGSET_RESTORE_ANON }, | |
1559 | { "workingset_restore_file", 1, WORKINGSET_RESTORE_FILE }, | |
1560 | { "workingset_nodereclaim", 1, WORKINGSET_NODERECLAIM }, | |
1561 | }; | |
1562 | ||
1563 | static int __init memory_stats_init(void) | |
1564 | { | |
1565 | int i; | |
1566 | ||
1567 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { | |
1568 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
bf9ecead | 1569 | if (memory_stats[i].idx == NR_SHMEM_THPS) |
5f9a4f4a MS |
1570 | memory_stats[i].ratio = HPAGE_PMD_SIZE; |
1571 | #endif | |
1572 | VM_BUG_ON(!memory_stats[i].ratio); | |
1573 | VM_BUG_ON(memory_stats[i].idx >= MEMCG_NR_STAT); | |
1574 | } | |
1575 | ||
1576 | return 0; | |
1577 | } | |
1578 | pure_initcall(memory_stats_init); | |
1579 | ||
c8713d0b JW |
1580 | static char *memory_stat_format(struct mem_cgroup *memcg) |
1581 | { | |
1582 | struct seq_buf s; | |
1583 | int i; | |
71cd3113 | 1584 | |
c8713d0b JW |
1585 | seq_buf_init(&s, kmalloc(PAGE_SIZE, GFP_KERNEL), PAGE_SIZE); |
1586 | if (!s.buffer) | |
1587 | return NULL; | |
1588 | ||
1589 | /* | |
1590 | * Provide statistics on the state of the memory subsystem as | |
1591 | * well as cumulative event counters that show past behavior. | |
1592 | * | |
1593 | * This list is ordered following a combination of these gradients: | |
1594 | * 1) generic big picture -> specifics and details | |
1595 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
1596 | * | |
1597 | * Current memory state: | |
1598 | */ | |
1599 | ||
5f9a4f4a MS |
1600 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { |
1601 | u64 size; | |
c8713d0b | 1602 | |
5f9a4f4a MS |
1603 | size = memcg_page_state(memcg, memory_stats[i].idx); |
1604 | size *= memory_stats[i].ratio; | |
1605 | seq_buf_printf(&s, "%s %llu\n", memory_stats[i].name, size); | |
c8713d0b | 1606 | |
5f9a4f4a MS |
1607 | if (unlikely(memory_stats[i].idx == NR_SLAB_UNRECLAIMABLE_B)) { |
1608 | size = memcg_page_state(memcg, NR_SLAB_RECLAIMABLE_B) + | |
1609 | memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE_B); | |
1610 | seq_buf_printf(&s, "slab %llu\n", size); | |
1611 | } | |
1612 | } | |
c8713d0b JW |
1613 | |
1614 | /* Accumulated memory events */ | |
1615 | ||
ebc5d83d KK |
1616 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGFAULT), |
1617 | memcg_events(memcg, PGFAULT)); | |
1618 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGMAJFAULT), | |
1619 | memcg_events(memcg, PGMAJFAULT)); | |
ebc5d83d KK |
1620 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGREFILL), |
1621 | memcg_events(memcg, PGREFILL)); | |
c8713d0b JW |
1622 | seq_buf_printf(&s, "pgscan %lu\n", |
1623 | memcg_events(memcg, PGSCAN_KSWAPD) + | |
1624 | memcg_events(memcg, PGSCAN_DIRECT)); | |
1625 | seq_buf_printf(&s, "pgsteal %lu\n", | |
1626 | memcg_events(memcg, PGSTEAL_KSWAPD) + | |
1627 | memcg_events(memcg, PGSTEAL_DIRECT)); | |
ebc5d83d KK |
1628 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGACTIVATE), |
1629 | memcg_events(memcg, PGACTIVATE)); | |
1630 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGDEACTIVATE), | |
1631 | memcg_events(memcg, PGDEACTIVATE)); | |
1632 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREE), | |
1633 | memcg_events(memcg, PGLAZYFREE)); | |
1634 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREED), | |
1635 | memcg_events(memcg, PGLAZYFREED)); | |
c8713d0b JW |
1636 | |
1637 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
ebc5d83d | 1638 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_FAULT_ALLOC), |
c8713d0b | 1639 | memcg_events(memcg, THP_FAULT_ALLOC)); |
ebc5d83d | 1640 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_COLLAPSE_ALLOC), |
c8713d0b JW |
1641 | memcg_events(memcg, THP_COLLAPSE_ALLOC)); |
1642 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ | |
1643 | ||
1644 | /* The above should easily fit into one page */ | |
1645 | WARN_ON_ONCE(seq_buf_has_overflowed(&s)); | |
1646 | ||
1647 | return s.buffer; | |
1648 | } | |
71cd3113 | 1649 | |
58cf188e | 1650 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1651 | /** |
f0c867d9 | 1652 | * mem_cgroup_print_oom_context: Print OOM information relevant to |
1653 | * memory controller. | |
e222432b BS |
1654 | * @memcg: The memory cgroup that went over limit |
1655 | * @p: Task that is going to be killed | |
1656 | * | |
1657 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1658 | * enabled | |
1659 | */ | |
f0c867d9 | 1660 | void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) |
e222432b | 1661 | { |
e222432b BS |
1662 | rcu_read_lock(); |
1663 | ||
f0c867d9 | 1664 | if (memcg) { |
1665 | pr_cont(",oom_memcg="); | |
1666 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1667 | } else | |
1668 | pr_cont(",global_oom"); | |
2415b9f5 | 1669 | if (p) { |
f0c867d9 | 1670 | pr_cont(",task_memcg="); |
2415b9f5 | 1671 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
2415b9f5 | 1672 | } |
e222432b | 1673 | rcu_read_unlock(); |
f0c867d9 | 1674 | } |
1675 | ||
1676 | /** | |
1677 | * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to | |
1678 | * memory controller. | |
1679 | * @memcg: The memory cgroup that went over limit | |
1680 | */ | |
1681 | void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) | |
1682 | { | |
c8713d0b | 1683 | char *buf; |
e222432b | 1684 | |
3e32cb2e JW |
1685 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1686 | K((u64)page_counter_read(&memcg->memory)), | |
15b42562 | 1687 | K((u64)READ_ONCE(memcg->memory.max)), memcg->memory.failcnt); |
c8713d0b JW |
1688 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
1689 | pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1690 | K((u64)page_counter_read(&memcg->swap)), | |
32d087cd | 1691 | K((u64)READ_ONCE(memcg->swap.max)), memcg->swap.failcnt); |
c8713d0b JW |
1692 | else { |
1693 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1694 | K((u64)page_counter_read(&memcg->memsw)), | |
1695 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); | |
1696 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1697 | K((u64)page_counter_read(&memcg->kmem)), | |
1698 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); | |
58cf188e | 1699 | } |
c8713d0b JW |
1700 | |
1701 | pr_info("Memory cgroup stats for "); | |
1702 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1703 | pr_cont(":"); | |
1704 | buf = memory_stat_format(memcg); | |
1705 | if (!buf) | |
1706 | return; | |
1707 | pr_info("%s", buf); | |
1708 | kfree(buf); | |
e222432b BS |
1709 | } |
1710 | ||
a63d83f4 DR |
1711 | /* |
1712 | * Return the memory (and swap, if configured) limit for a memcg. | |
1713 | */ | |
bbec2e15 | 1714 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1715 | { |
8d387a5f WL |
1716 | unsigned long max = READ_ONCE(memcg->memory.max); |
1717 | ||
1718 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) { | |
1719 | if (mem_cgroup_swappiness(memcg)) | |
1720 | max += min(READ_ONCE(memcg->swap.max), | |
1721 | (unsigned long)total_swap_pages); | |
1722 | } else { /* v1 */ | |
1723 | if (mem_cgroup_swappiness(memcg)) { | |
1724 | /* Calculate swap excess capacity from memsw limit */ | |
1725 | unsigned long swap = READ_ONCE(memcg->memsw.max) - max; | |
1726 | ||
1727 | max += min(swap, (unsigned long)total_swap_pages); | |
1728 | } | |
9a5a8f19 | 1729 | } |
bbec2e15 | 1730 | return max; |
a63d83f4 DR |
1731 | } |
1732 | ||
9783aa99 CD |
1733 | unsigned long mem_cgroup_size(struct mem_cgroup *memcg) |
1734 | { | |
1735 | return page_counter_read(&memcg->memory); | |
1736 | } | |
1737 | ||
b6e6edcf | 1738 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1739 | int order) |
9cbb78bb | 1740 | { |
6e0fc46d DR |
1741 | struct oom_control oc = { |
1742 | .zonelist = NULL, | |
1743 | .nodemask = NULL, | |
2a966b77 | 1744 | .memcg = memcg, |
6e0fc46d DR |
1745 | .gfp_mask = gfp_mask, |
1746 | .order = order, | |
6e0fc46d | 1747 | }; |
1378b37d | 1748 | bool ret = true; |
9cbb78bb | 1749 | |
7775face TH |
1750 | if (mutex_lock_killable(&oom_lock)) |
1751 | return true; | |
1378b37d YS |
1752 | |
1753 | if (mem_cgroup_margin(memcg) >= (1 << order)) | |
1754 | goto unlock; | |
1755 | ||
7775face TH |
1756 | /* |
1757 | * A few threads which were not waiting at mutex_lock_killable() can | |
1758 | * fail to bail out. Therefore, check again after holding oom_lock. | |
1759 | */ | |
1760 | ret = should_force_charge() || out_of_memory(&oc); | |
1378b37d YS |
1761 | |
1762 | unlock: | |
dc56401f | 1763 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1764 | return ret; |
9cbb78bb DR |
1765 | } |
1766 | ||
0608f43d | 1767 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1768 | pg_data_t *pgdat, |
0608f43d AM |
1769 | gfp_t gfp_mask, |
1770 | unsigned long *total_scanned) | |
1771 | { | |
1772 | struct mem_cgroup *victim = NULL; | |
1773 | int total = 0; | |
1774 | int loop = 0; | |
1775 | unsigned long excess; | |
1776 | unsigned long nr_scanned; | |
1777 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1778 | .pgdat = pgdat, |
0608f43d AM |
1779 | }; |
1780 | ||
3e32cb2e | 1781 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1782 | |
1783 | while (1) { | |
1784 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1785 | if (!victim) { | |
1786 | loop++; | |
1787 | if (loop >= 2) { | |
1788 | /* | |
1789 | * If we have not been able to reclaim | |
1790 | * anything, it might because there are | |
1791 | * no reclaimable pages under this hierarchy | |
1792 | */ | |
1793 | if (!total) | |
1794 | break; | |
1795 | /* | |
1796 | * We want to do more targeted reclaim. | |
1797 | * excess >> 2 is not to excessive so as to | |
1798 | * reclaim too much, nor too less that we keep | |
1799 | * coming back to reclaim from this cgroup | |
1800 | */ | |
1801 | if (total >= (excess >> 2) || | |
1802 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1803 | break; | |
1804 | } | |
1805 | continue; | |
1806 | } | |
a9dd0a83 | 1807 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1808 | pgdat, &nr_scanned); |
0608f43d | 1809 | *total_scanned += nr_scanned; |
3e32cb2e | 1810 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1811 | break; |
6d61ef40 | 1812 | } |
0608f43d AM |
1813 | mem_cgroup_iter_break(root_memcg, victim); |
1814 | return total; | |
6d61ef40 BS |
1815 | } |
1816 | ||
0056f4e6 JW |
1817 | #ifdef CONFIG_LOCKDEP |
1818 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1819 | .name = "memcg_oom_lock", | |
1820 | }; | |
1821 | #endif | |
1822 | ||
fb2a6fc5 JW |
1823 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1824 | ||
867578cb KH |
1825 | /* |
1826 | * Check OOM-Killer is already running under our hierarchy. | |
1827 | * If someone is running, return false. | |
1828 | */ | |
fb2a6fc5 | 1829 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1830 | { |
79dfdacc | 1831 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1832 | |
fb2a6fc5 JW |
1833 | spin_lock(&memcg_oom_lock); |
1834 | ||
9f3a0d09 | 1835 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1836 | if (iter->oom_lock) { |
79dfdacc MH |
1837 | /* |
1838 | * this subtree of our hierarchy is already locked | |
1839 | * so we cannot give a lock. | |
1840 | */ | |
79dfdacc | 1841 | failed = iter; |
9f3a0d09 JW |
1842 | mem_cgroup_iter_break(memcg, iter); |
1843 | break; | |
23751be0 JW |
1844 | } else |
1845 | iter->oom_lock = true; | |
7d74b06f | 1846 | } |
867578cb | 1847 | |
fb2a6fc5 JW |
1848 | if (failed) { |
1849 | /* | |
1850 | * OK, we failed to lock the whole subtree so we have | |
1851 | * to clean up what we set up to the failing subtree | |
1852 | */ | |
1853 | for_each_mem_cgroup_tree(iter, memcg) { | |
1854 | if (iter == failed) { | |
1855 | mem_cgroup_iter_break(memcg, iter); | |
1856 | break; | |
1857 | } | |
1858 | iter->oom_lock = false; | |
79dfdacc | 1859 | } |
0056f4e6 JW |
1860 | } else |
1861 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1862 | |
1863 | spin_unlock(&memcg_oom_lock); | |
1864 | ||
1865 | return !failed; | |
a636b327 | 1866 | } |
0b7f569e | 1867 | |
fb2a6fc5 | 1868 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1869 | { |
7d74b06f KH |
1870 | struct mem_cgroup *iter; |
1871 | ||
fb2a6fc5 | 1872 | spin_lock(&memcg_oom_lock); |
5facae4f | 1873 | mutex_release(&memcg_oom_lock_dep_map, _RET_IP_); |
c0ff4b85 | 1874 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1875 | iter->oom_lock = false; |
fb2a6fc5 | 1876 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1877 | } |
1878 | ||
c0ff4b85 | 1879 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1880 | { |
1881 | struct mem_cgroup *iter; | |
1882 | ||
c2b42d3c | 1883 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1884 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1885 | iter->under_oom++; |
1886 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1887 | } |
1888 | ||
c0ff4b85 | 1889 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1890 | { |
1891 | struct mem_cgroup *iter; | |
1892 | ||
867578cb | 1893 | /* |
7a52d4d8 ML |
1894 | * Be careful about under_oom underflows becase a child memcg |
1895 | * could have been added after mem_cgroup_mark_under_oom. | |
867578cb | 1896 | */ |
c2b42d3c | 1897 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1898 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1899 | if (iter->under_oom > 0) |
1900 | iter->under_oom--; | |
1901 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1902 | } |
1903 | ||
867578cb KH |
1904 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1905 | ||
dc98df5a | 1906 | struct oom_wait_info { |
d79154bb | 1907 | struct mem_cgroup *memcg; |
ac6424b9 | 1908 | wait_queue_entry_t wait; |
dc98df5a KH |
1909 | }; |
1910 | ||
ac6424b9 | 1911 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1912 | unsigned mode, int sync, void *arg) |
1913 | { | |
d79154bb HD |
1914 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1915 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1916 | struct oom_wait_info *oom_wait_info; |
1917 | ||
1918 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1919 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1920 | |
2314b42d JW |
1921 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1922 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1923 | return 0; |
dc98df5a KH |
1924 | return autoremove_wake_function(wait, mode, sync, arg); |
1925 | } | |
1926 | ||
c0ff4b85 | 1927 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1928 | { |
c2b42d3c TH |
1929 | /* |
1930 | * For the following lockless ->under_oom test, the only required | |
1931 | * guarantee is that it must see the state asserted by an OOM when | |
1932 | * this function is called as a result of userland actions | |
1933 | * triggered by the notification of the OOM. This is trivially | |
1934 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1935 | * triggering notification. | |
1936 | */ | |
1937 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1938 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1939 | } |
1940 | ||
29ef680a MH |
1941 | enum oom_status { |
1942 | OOM_SUCCESS, | |
1943 | OOM_FAILED, | |
1944 | OOM_ASYNC, | |
1945 | OOM_SKIPPED | |
1946 | }; | |
1947 | ||
1948 | static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 1949 | { |
7056d3a3 MH |
1950 | enum oom_status ret; |
1951 | bool locked; | |
1952 | ||
29ef680a MH |
1953 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
1954 | return OOM_SKIPPED; | |
1955 | ||
7a1adfdd RG |
1956 | memcg_memory_event(memcg, MEMCG_OOM); |
1957 | ||
867578cb | 1958 | /* |
49426420 JW |
1959 | * We are in the middle of the charge context here, so we |
1960 | * don't want to block when potentially sitting on a callstack | |
1961 | * that holds all kinds of filesystem and mm locks. | |
1962 | * | |
29ef680a MH |
1963 | * cgroup1 allows disabling the OOM killer and waiting for outside |
1964 | * handling until the charge can succeed; remember the context and put | |
1965 | * the task to sleep at the end of the page fault when all locks are | |
1966 | * released. | |
49426420 | 1967 | * |
29ef680a MH |
1968 | * On the other hand, in-kernel OOM killer allows for an async victim |
1969 | * memory reclaim (oom_reaper) and that means that we are not solely | |
1970 | * relying on the oom victim to make a forward progress and we can | |
1971 | * invoke the oom killer here. | |
1972 | * | |
1973 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
1974 | * victim and then we have to bail out from the charge path. | |
867578cb | 1975 | */ |
29ef680a MH |
1976 | if (memcg->oom_kill_disable) { |
1977 | if (!current->in_user_fault) | |
1978 | return OOM_SKIPPED; | |
1979 | css_get(&memcg->css); | |
1980 | current->memcg_in_oom = memcg; | |
1981 | current->memcg_oom_gfp_mask = mask; | |
1982 | current->memcg_oom_order = order; | |
1983 | ||
1984 | return OOM_ASYNC; | |
1985 | } | |
1986 | ||
7056d3a3 MH |
1987 | mem_cgroup_mark_under_oom(memcg); |
1988 | ||
1989 | locked = mem_cgroup_oom_trylock(memcg); | |
1990 | ||
1991 | if (locked) | |
1992 | mem_cgroup_oom_notify(memcg); | |
1993 | ||
1994 | mem_cgroup_unmark_under_oom(memcg); | |
29ef680a | 1995 | if (mem_cgroup_out_of_memory(memcg, mask, order)) |
7056d3a3 MH |
1996 | ret = OOM_SUCCESS; |
1997 | else | |
1998 | ret = OOM_FAILED; | |
1999 | ||
2000 | if (locked) | |
2001 | mem_cgroup_oom_unlock(memcg); | |
29ef680a | 2002 | |
7056d3a3 | 2003 | return ret; |
3812c8c8 JW |
2004 | } |
2005 | ||
2006 | /** | |
2007 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 2008 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 2009 | * |
49426420 JW |
2010 | * This has to be called at the end of a page fault if the memcg OOM |
2011 | * handler was enabled. | |
3812c8c8 | 2012 | * |
49426420 | 2013 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
2014 | * sleep on a waitqueue until the userspace task resolves the |
2015 | * situation. Sleeping directly in the charge context with all kinds | |
2016 | * of locks held is not a good idea, instead we remember an OOM state | |
2017 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 2018 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
2019 | * |
2020 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 2021 | * completed, %false otherwise. |
3812c8c8 | 2022 | */ |
49426420 | 2023 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 2024 | { |
626ebc41 | 2025 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 2026 | struct oom_wait_info owait; |
49426420 | 2027 | bool locked; |
3812c8c8 JW |
2028 | |
2029 | /* OOM is global, do not handle */ | |
3812c8c8 | 2030 | if (!memcg) |
49426420 | 2031 | return false; |
3812c8c8 | 2032 | |
7c5f64f8 | 2033 | if (!handle) |
49426420 | 2034 | goto cleanup; |
3812c8c8 JW |
2035 | |
2036 | owait.memcg = memcg; | |
2037 | owait.wait.flags = 0; | |
2038 | owait.wait.func = memcg_oom_wake_function; | |
2039 | owait.wait.private = current; | |
2055da97 | 2040 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 2041 | |
3812c8c8 | 2042 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
2043 | mem_cgroup_mark_under_oom(memcg); |
2044 | ||
2045 | locked = mem_cgroup_oom_trylock(memcg); | |
2046 | ||
2047 | if (locked) | |
2048 | mem_cgroup_oom_notify(memcg); | |
2049 | ||
2050 | if (locked && !memcg->oom_kill_disable) { | |
2051 | mem_cgroup_unmark_under_oom(memcg); | |
2052 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
2053 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
2054 | current->memcg_oom_order); | |
49426420 | 2055 | } else { |
3812c8c8 | 2056 | schedule(); |
49426420 JW |
2057 | mem_cgroup_unmark_under_oom(memcg); |
2058 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2059 | } | |
2060 | ||
2061 | if (locked) { | |
fb2a6fc5 JW |
2062 | mem_cgroup_oom_unlock(memcg); |
2063 | /* | |
2064 | * There is no guarantee that an OOM-lock contender | |
2065 | * sees the wakeups triggered by the OOM kill | |
2066 | * uncharges. Wake any sleepers explicitely. | |
2067 | */ | |
2068 | memcg_oom_recover(memcg); | |
2069 | } | |
49426420 | 2070 | cleanup: |
626ebc41 | 2071 | current->memcg_in_oom = NULL; |
3812c8c8 | 2072 | css_put(&memcg->css); |
867578cb | 2073 | return true; |
0b7f569e KH |
2074 | } |
2075 | ||
3d8b38eb RG |
2076 | /** |
2077 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
2078 | * @victim: task to be killed by the OOM killer | |
2079 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
2080 | * | |
2081 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
2082 | * by killing all belonging OOM-killable tasks. | |
2083 | * | |
2084 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
2085 | */ | |
2086 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
2087 | struct mem_cgroup *oom_domain) | |
2088 | { | |
2089 | struct mem_cgroup *oom_group = NULL; | |
2090 | struct mem_cgroup *memcg; | |
2091 | ||
2092 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
2093 | return NULL; | |
2094 | ||
2095 | if (!oom_domain) | |
2096 | oom_domain = root_mem_cgroup; | |
2097 | ||
2098 | rcu_read_lock(); | |
2099 | ||
2100 | memcg = mem_cgroup_from_task(victim); | |
2101 | if (memcg == root_mem_cgroup) | |
2102 | goto out; | |
2103 | ||
48fe267c RG |
2104 | /* |
2105 | * If the victim task has been asynchronously moved to a different | |
2106 | * memory cgroup, we might end up killing tasks outside oom_domain. | |
2107 | * In this case it's better to ignore memory.group.oom. | |
2108 | */ | |
2109 | if (unlikely(!mem_cgroup_is_descendant(memcg, oom_domain))) | |
2110 | goto out; | |
2111 | ||
3d8b38eb RG |
2112 | /* |
2113 | * Traverse the memory cgroup hierarchy from the victim task's | |
2114 | * cgroup up to the OOMing cgroup (or root) to find the | |
2115 | * highest-level memory cgroup with oom.group set. | |
2116 | */ | |
2117 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
2118 | if (memcg->oom_group) | |
2119 | oom_group = memcg; | |
2120 | ||
2121 | if (memcg == oom_domain) | |
2122 | break; | |
2123 | } | |
2124 | ||
2125 | if (oom_group) | |
2126 | css_get(&oom_group->css); | |
2127 | out: | |
2128 | rcu_read_unlock(); | |
2129 | ||
2130 | return oom_group; | |
2131 | } | |
2132 | ||
2133 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
2134 | { | |
2135 | pr_info("Tasks in "); | |
2136 | pr_cont_cgroup_path(memcg->css.cgroup); | |
2137 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
2138 | } | |
2139 | ||
d7365e78 | 2140 | /** |
bcfe06bf | 2141 | * lock_page_memcg - lock a page and memcg binding |
81f8c3a4 | 2142 | * @page: the page |
32047e2a | 2143 | * |
81f8c3a4 | 2144 | * This function protects unlocked LRU pages from being moved to |
739f79fc JW |
2145 | * another cgroup. |
2146 | * | |
2147 | * It ensures lifetime of the returned memcg. Caller is responsible | |
2148 | * for the lifetime of the page; __unlock_page_memcg() is available | |
2149 | * when @page might get freed inside the locked section. | |
d69b042f | 2150 | */ |
739f79fc | 2151 | struct mem_cgroup *lock_page_memcg(struct page *page) |
89c06bd5 | 2152 | { |
9da7b521 | 2153 | struct page *head = compound_head(page); /* rmap on tail pages */ |
89c06bd5 | 2154 | struct mem_cgroup *memcg; |
6de22619 | 2155 | unsigned long flags; |
89c06bd5 | 2156 | |
6de22619 JW |
2157 | /* |
2158 | * The RCU lock is held throughout the transaction. The fast | |
2159 | * path can get away without acquiring the memcg->move_lock | |
2160 | * because page moving starts with an RCU grace period. | |
739f79fc JW |
2161 | * |
2162 | * The RCU lock also protects the memcg from being freed when | |
2163 | * the page state that is going to change is the only thing | |
2164 | * preventing the page itself from being freed. E.g. writeback | |
2165 | * doesn't hold a page reference and relies on PG_writeback to | |
2166 | * keep off truncation, migration and so forth. | |
2167 | */ | |
d7365e78 JW |
2168 | rcu_read_lock(); |
2169 | ||
2170 | if (mem_cgroup_disabled()) | |
739f79fc | 2171 | return NULL; |
89c06bd5 | 2172 | again: |
bcfe06bf | 2173 | memcg = page_memcg(head); |
29833315 | 2174 | if (unlikely(!memcg)) |
739f79fc | 2175 | return NULL; |
d7365e78 | 2176 | |
20ad50d6 AS |
2177 | #ifdef CONFIG_PROVE_LOCKING |
2178 | local_irq_save(flags); | |
2179 | might_lock(&memcg->move_lock); | |
2180 | local_irq_restore(flags); | |
2181 | #endif | |
2182 | ||
bdcbb659 | 2183 | if (atomic_read(&memcg->moving_account) <= 0) |
739f79fc | 2184 | return memcg; |
89c06bd5 | 2185 | |
6de22619 | 2186 | spin_lock_irqsave(&memcg->move_lock, flags); |
bcfe06bf | 2187 | if (memcg != page_memcg(head)) { |
6de22619 | 2188 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
2189 | goto again; |
2190 | } | |
6de22619 JW |
2191 | |
2192 | /* | |
2193 | * When charge migration first begins, we can have locked and | |
2194 | * unlocked page stat updates happening concurrently. Track | |
81f8c3a4 | 2195 | * the task who has the lock for unlock_page_memcg(). |
6de22619 JW |
2196 | */ |
2197 | memcg->move_lock_task = current; | |
2198 | memcg->move_lock_flags = flags; | |
d7365e78 | 2199 | |
739f79fc | 2200 | return memcg; |
89c06bd5 | 2201 | } |
81f8c3a4 | 2202 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 2203 | |
d7365e78 | 2204 | /** |
739f79fc JW |
2205 | * __unlock_page_memcg - unlock and unpin a memcg |
2206 | * @memcg: the memcg | |
2207 | * | |
2208 | * Unlock and unpin a memcg returned by lock_page_memcg(). | |
d7365e78 | 2209 | */ |
739f79fc | 2210 | void __unlock_page_memcg(struct mem_cgroup *memcg) |
89c06bd5 | 2211 | { |
6de22619 JW |
2212 | if (memcg && memcg->move_lock_task == current) { |
2213 | unsigned long flags = memcg->move_lock_flags; | |
2214 | ||
2215 | memcg->move_lock_task = NULL; | |
2216 | memcg->move_lock_flags = 0; | |
2217 | ||
2218 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2219 | } | |
89c06bd5 | 2220 | |
d7365e78 | 2221 | rcu_read_unlock(); |
89c06bd5 | 2222 | } |
739f79fc JW |
2223 | |
2224 | /** | |
bcfe06bf | 2225 | * unlock_page_memcg - unlock a page and memcg binding |
739f79fc JW |
2226 | * @page: the page |
2227 | */ | |
2228 | void unlock_page_memcg(struct page *page) | |
2229 | { | |
9da7b521 JW |
2230 | struct page *head = compound_head(page); |
2231 | ||
bcfe06bf | 2232 | __unlock_page_memcg(page_memcg(head)); |
739f79fc | 2233 | } |
81f8c3a4 | 2234 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 2235 | |
cdec2e42 KH |
2236 | struct memcg_stock_pcp { |
2237 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2238 | unsigned int nr_pages; |
bf4f0599 RG |
2239 | |
2240 | #ifdef CONFIG_MEMCG_KMEM | |
2241 | struct obj_cgroup *cached_objcg; | |
2242 | unsigned int nr_bytes; | |
2243 | #endif | |
2244 | ||
cdec2e42 | 2245 | struct work_struct work; |
26fe6168 | 2246 | unsigned long flags; |
a0db00fc | 2247 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2248 | }; |
2249 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2250 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2251 | |
bf4f0599 RG |
2252 | #ifdef CONFIG_MEMCG_KMEM |
2253 | static void drain_obj_stock(struct memcg_stock_pcp *stock); | |
2254 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2255 | struct mem_cgroup *root_memcg); | |
2256 | ||
2257 | #else | |
2258 | static inline void drain_obj_stock(struct memcg_stock_pcp *stock) | |
2259 | { | |
2260 | } | |
2261 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2262 | struct mem_cgroup *root_memcg) | |
2263 | { | |
2264 | return false; | |
2265 | } | |
2266 | #endif | |
2267 | ||
a0956d54 SS |
2268 | /** |
2269 | * consume_stock: Try to consume stocked charge on this cpu. | |
2270 | * @memcg: memcg to consume from. | |
2271 | * @nr_pages: how many pages to charge. | |
2272 | * | |
2273 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2274 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2275 | * service an allocation will refill the stock. | |
2276 | * | |
2277 | * returns true if successful, false otherwise. | |
cdec2e42 | 2278 | */ |
a0956d54 | 2279 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2280 | { |
2281 | struct memcg_stock_pcp *stock; | |
db2ba40c | 2282 | unsigned long flags; |
3e32cb2e | 2283 | bool ret = false; |
cdec2e42 | 2284 | |
a983b5eb | 2285 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 2286 | return ret; |
a0956d54 | 2287 | |
db2ba40c JW |
2288 | local_irq_save(flags); |
2289 | ||
2290 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 2291 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2292 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2293 | ret = true; |
2294 | } | |
db2ba40c JW |
2295 | |
2296 | local_irq_restore(flags); | |
2297 | ||
cdec2e42 KH |
2298 | return ret; |
2299 | } | |
2300 | ||
2301 | /* | |
3e32cb2e | 2302 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2303 | */ |
2304 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2305 | { | |
2306 | struct mem_cgroup *old = stock->cached; | |
2307 | ||
1a3e1f40 JW |
2308 | if (!old) |
2309 | return; | |
2310 | ||
11c9ea4e | 2311 | if (stock->nr_pages) { |
3e32cb2e | 2312 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 2313 | if (do_memsw_account()) |
3e32cb2e | 2314 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
11c9ea4e | 2315 | stock->nr_pages = 0; |
cdec2e42 | 2316 | } |
1a3e1f40 JW |
2317 | |
2318 | css_put(&old->css); | |
cdec2e42 | 2319 | stock->cached = NULL; |
cdec2e42 KH |
2320 | } |
2321 | ||
cdec2e42 KH |
2322 | static void drain_local_stock(struct work_struct *dummy) |
2323 | { | |
db2ba40c JW |
2324 | struct memcg_stock_pcp *stock; |
2325 | unsigned long flags; | |
2326 | ||
72f0184c MH |
2327 | /* |
2328 | * The only protection from memory hotplug vs. drain_stock races is | |
2329 | * that we always operate on local CPU stock here with IRQ disabled | |
2330 | */ | |
db2ba40c JW |
2331 | local_irq_save(flags); |
2332 | ||
2333 | stock = this_cpu_ptr(&memcg_stock); | |
bf4f0599 | 2334 | drain_obj_stock(stock); |
cdec2e42 | 2335 | drain_stock(stock); |
26fe6168 | 2336 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c JW |
2337 | |
2338 | local_irq_restore(flags); | |
cdec2e42 KH |
2339 | } |
2340 | ||
2341 | /* | |
3e32cb2e | 2342 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2343 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2344 | */ |
c0ff4b85 | 2345 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2346 | { |
db2ba40c JW |
2347 | struct memcg_stock_pcp *stock; |
2348 | unsigned long flags; | |
2349 | ||
2350 | local_irq_save(flags); | |
cdec2e42 | 2351 | |
db2ba40c | 2352 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 2353 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2354 | drain_stock(stock); |
1a3e1f40 | 2355 | css_get(&memcg->css); |
c0ff4b85 | 2356 | stock->cached = memcg; |
cdec2e42 | 2357 | } |
11c9ea4e | 2358 | stock->nr_pages += nr_pages; |
db2ba40c | 2359 | |
a983b5eb | 2360 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 RG |
2361 | drain_stock(stock); |
2362 | ||
db2ba40c | 2363 | local_irq_restore(flags); |
cdec2e42 KH |
2364 | } |
2365 | ||
2366 | /* | |
c0ff4b85 | 2367 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2368 | * of the hierarchy under it. |
cdec2e42 | 2369 | */ |
6d3d6aa2 | 2370 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2371 | { |
26fe6168 | 2372 | int cpu, curcpu; |
d38144b7 | 2373 | |
6d3d6aa2 JW |
2374 | /* If someone's already draining, avoid adding running more workers. */ |
2375 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2376 | return; | |
72f0184c MH |
2377 | /* |
2378 | * Notify other cpus that system-wide "drain" is running | |
2379 | * We do not care about races with the cpu hotplug because cpu down | |
2380 | * as well as workers from this path always operate on the local | |
2381 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2382 | */ | |
5af12d0e | 2383 | curcpu = get_cpu(); |
cdec2e42 KH |
2384 | for_each_online_cpu(cpu) { |
2385 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2386 | struct mem_cgroup *memcg; |
e1a366be | 2387 | bool flush = false; |
26fe6168 | 2388 | |
e1a366be | 2389 | rcu_read_lock(); |
c0ff4b85 | 2390 | memcg = stock->cached; |
e1a366be RG |
2391 | if (memcg && stock->nr_pages && |
2392 | mem_cgroup_is_descendant(memcg, root_memcg)) | |
2393 | flush = true; | |
bf4f0599 RG |
2394 | if (obj_stock_flush_required(stock, root_memcg)) |
2395 | flush = true; | |
e1a366be RG |
2396 | rcu_read_unlock(); |
2397 | ||
2398 | if (flush && | |
2399 | !test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { | |
d1a05b69 MH |
2400 | if (cpu == curcpu) |
2401 | drain_local_stock(&stock->work); | |
2402 | else | |
2403 | schedule_work_on(cpu, &stock->work); | |
2404 | } | |
cdec2e42 | 2405 | } |
5af12d0e | 2406 | put_cpu(); |
9f50fad6 | 2407 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2408 | } |
2409 | ||
308167fc | 2410 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
cdec2e42 | 2411 | { |
cdec2e42 | 2412 | struct memcg_stock_pcp *stock; |
42a30035 | 2413 | struct mem_cgroup *memcg, *mi; |
cdec2e42 | 2414 | |
cdec2e42 KH |
2415 | stock = &per_cpu(memcg_stock, cpu); |
2416 | drain_stock(stock); | |
a983b5eb JW |
2417 | |
2418 | for_each_mem_cgroup(memcg) { | |
2419 | int i; | |
2420 | ||
2421 | for (i = 0; i < MEMCG_NR_STAT; i++) { | |
2422 | int nid; | |
2423 | long x; | |
2424 | ||
871789d4 | 2425 | x = this_cpu_xchg(memcg->vmstats_percpu->stat[i], 0); |
815744d7 | 2426 | if (x) |
42a30035 JW |
2427 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2428 | atomic_long_add(x, &memcg->vmstats[i]); | |
a983b5eb JW |
2429 | |
2430 | if (i >= NR_VM_NODE_STAT_ITEMS) | |
2431 | continue; | |
2432 | ||
2433 | for_each_node(nid) { | |
2434 | struct mem_cgroup_per_node *pn; | |
2435 | ||
2436 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
2437 | x = this_cpu_xchg(pn->lruvec_stat_cpu->count[i], 0); | |
815744d7 | 2438 | if (x) |
42a30035 JW |
2439 | do { |
2440 | atomic_long_add(x, &pn->lruvec_stat[i]); | |
2441 | } while ((pn = parent_nodeinfo(pn, nid))); | |
a983b5eb JW |
2442 | } |
2443 | } | |
2444 | ||
e27be240 | 2445 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { |
a983b5eb JW |
2446 | long x; |
2447 | ||
871789d4 | 2448 | x = this_cpu_xchg(memcg->vmstats_percpu->events[i], 0); |
815744d7 | 2449 | if (x) |
42a30035 JW |
2450 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2451 | atomic_long_add(x, &memcg->vmevents[i]); | |
a983b5eb JW |
2452 | } |
2453 | } | |
2454 | ||
308167fc | 2455 | return 0; |
cdec2e42 KH |
2456 | } |
2457 | ||
b3ff9291 CD |
2458 | static unsigned long reclaim_high(struct mem_cgroup *memcg, |
2459 | unsigned int nr_pages, | |
2460 | gfp_t gfp_mask) | |
f7e1cb6e | 2461 | { |
b3ff9291 CD |
2462 | unsigned long nr_reclaimed = 0; |
2463 | ||
f7e1cb6e | 2464 | do { |
e22c6ed9 JW |
2465 | unsigned long pflags; |
2466 | ||
d1663a90 JK |
2467 | if (page_counter_read(&memcg->memory) <= |
2468 | READ_ONCE(memcg->memory.high)) | |
f7e1cb6e | 2469 | continue; |
e22c6ed9 | 2470 | |
e27be240 | 2471 | memcg_memory_event(memcg, MEMCG_HIGH); |
e22c6ed9 JW |
2472 | |
2473 | psi_memstall_enter(&pflags); | |
b3ff9291 CD |
2474 | nr_reclaimed += try_to_free_mem_cgroup_pages(memcg, nr_pages, |
2475 | gfp_mask, true); | |
e22c6ed9 | 2476 | psi_memstall_leave(&pflags); |
4bf17307 CD |
2477 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2478 | !mem_cgroup_is_root(memcg)); | |
b3ff9291 CD |
2479 | |
2480 | return nr_reclaimed; | |
f7e1cb6e JW |
2481 | } |
2482 | ||
2483 | static void high_work_func(struct work_struct *work) | |
2484 | { | |
2485 | struct mem_cgroup *memcg; | |
2486 | ||
2487 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2488 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2489 | } |
2490 | ||
0e4b01df CD |
2491 | /* |
2492 | * Clamp the maximum sleep time per allocation batch to 2 seconds. This is | |
2493 | * enough to still cause a significant slowdown in most cases, while still | |
2494 | * allowing diagnostics and tracing to proceed without becoming stuck. | |
2495 | */ | |
2496 | #define MEMCG_MAX_HIGH_DELAY_JIFFIES (2UL*HZ) | |
2497 | ||
2498 | /* | |
2499 | * When calculating the delay, we use these either side of the exponentiation to | |
2500 | * maintain precision and scale to a reasonable number of jiffies (see the table | |
2501 | * below. | |
2502 | * | |
2503 | * - MEMCG_DELAY_PRECISION_SHIFT: Extra precision bits while translating the | |
2504 | * overage ratio to a delay. | |
ac5ddd0f | 2505 | * - MEMCG_DELAY_SCALING_SHIFT: The number of bits to scale down the |
0e4b01df CD |
2506 | * proposed penalty in order to reduce to a reasonable number of jiffies, and |
2507 | * to produce a reasonable delay curve. | |
2508 | * | |
2509 | * MEMCG_DELAY_SCALING_SHIFT just happens to be a number that produces a | |
2510 | * reasonable delay curve compared to precision-adjusted overage, not | |
2511 | * penalising heavily at first, but still making sure that growth beyond the | |
2512 | * limit penalises misbehaviour cgroups by slowing them down exponentially. For | |
2513 | * example, with a high of 100 megabytes: | |
2514 | * | |
2515 | * +-------+------------------------+ | |
2516 | * | usage | time to allocate in ms | | |
2517 | * +-------+------------------------+ | |
2518 | * | 100M | 0 | | |
2519 | * | 101M | 6 | | |
2520 | * | 102M | 25 | | |
2521 | * | 103M | 57 | | |
2522 | * | 104M | 102 | | |
2523 | * | 105M | 159 | | |
2524 | * | 106M | 230 | | |
2525 | * | 107M | 313 | | |
2526 | * | 108M | 409 | | |
2527 | * | 109M | 518 | | |
2528 | * | 110M | 639 | | |
2529 | * | 111M | 774 | | |
2530 | * | 112M | 921 | | |
2531 | * | 113M | 1081 | | |
2532 | * | 114M | 1254 | | |
2533 | * | 115M | 1439 | | |
2534 | * | 116M | 1638 | | |
2535 | * | 117M | 1849 | | |
2536 | * | 118M | 2000 | | |
2537 | * | 119M | 2000 | | |
2538 | * | 120M | 2000 | | |
2539 | * +-------+------------------------+ | |
2540 | */ | |
2541 | #define MEMCG_DELAY_PRECISION_SHIFT 20 | |
2542 | #define MEMCG_DELAY_SCALING_SHIFT 14 | |
2543 | ||
8a5dbc65 | 2544 | static u64 calculate_overage(unsigned long usage, unsigned long high) |
b23afb93 | 2545 | { |
8a5dbc65 | 2546 | u64 overage; |
b23afb93 | 2547 | |
8a5dbc65 JK |
2548 | if (usage <= high) |
2549 | return 0; | |
e26733e0 | 2550 | |
8a5dbc65 JK |
2551 | /* |
2552 | * Prevent division by 0 in overage calculation by acting as if | |
2553 | * it was a threshold of 1 page | |
2554 | */ | |
2555 | high = max(high, 1UL); | |
9b8b1754 | 2556 | |
8a5dbc65 JK |
2557 | overage = usage - high; |
2558 | overage <<= MEMCG_DELAY_PRECISION_SHIFT; | |
2559 | return div64_u64(overage, high); | |
2560 | } | |
e26733e0 | 2561 | |
8a5dbc65 JK |
2562 | static u64 mem_find_max_overage(struct mem_cgroup *memcg) |
2563 | { | |
2564 | u64 overage, max_overage = 0; | |
e26733e0 | 2565 | |
8a5dbc65 JK |
2566 | do { |
2567 | overage = calculate_overage(page_counter_read(&memcg->memory), | |
d1663a90 | 2568 | READ_ONCE(memcg->memory.high)); |
8a5dbc65 | 2569 | max_overage = max(overage, max_overage); |
e26733e0 CD |
2570 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2571 | !mem_cgroup_is_root(memcg)); | |
2572 | ||
8a5dbc65 JK |
2573 | return max_overage; |
2574 | } | |
2575 | ||
4b82ab4f JK |
2576 | static u64 swap_find_max_overage(struct mem_cgroup *memcg) |
2577 | { | |
2578 | u64 overage, max_overage = 0; | |
2579 | ||
2580 | do { | |
2581 | overage = calculate_overage(page_counter_read(&memcg->swap), | |
2582 | READ_ONCE(memcg->swap.high)); | |
2583 | if (overage) | |
2584 | memcg_memory_event(memcg, MEMCG_SWAP_HIGH); | |
2585 | max_overage = max(overage, max_overage); | |
2586 | } while ((memcg = parent_mem_cgroup(memcg)) && | |
2587 | !mem_cgroup_is_root(memcg)); | |
2588 | ||
2589 | return max_overage; | |
2590 | } | |
2591 | ||
8a5dbc65 JK |
2592 | /* |
2593 | * Get the number of jiffies that we should penalise a mischievous cgroup which | |
2594 | * is exceeding its memory.high by checking both it and its ancestors. | |
2595 | */ | |
2596 | static unsigned long calculate_high_delay(struct mem_cgroup *memcg, | |
2597 | unsigned int nr_pages, | |
2598 | u64 max_overage) | |
2599 | { | |
2600 | unsigned long penalty_jiffies; | |
2601 | ||
e26733e0 CD |
2602 | if (!max_overage) |
2603 | return 0; | |
0e4b01df CD |
2604 | |
2605 | /* | |
0e4b01df CD |
2606 | * We use overage compared to memory.high to calculate the number of |
2607 | * jiffies to sleep (penalty_jiffies). Ideally this value should be | |
2608 | * fairly lenient on small overages, and increasingly harsh when the | |
2609 | * memcg in question makes it clear that it has no intention of stopping | |
2610 | * its crazy behaviour, so we exponentially increase the delay based on | |
2611 | * overage amount. | |
2612 | */ | |
e26733e0 CD |
2613 | penalty_jiffies = max_overage * max_overage * HZ; |
2614 | penalty_jiffies >>= MEMCG_DELAY_PRECISION_SHIFT; | |
2615 | penalty_jiffies >>= MEMCG_DELAY_SCALING_SHIFT; | |
0e4b01df CD |
2616 | |
2617 | /* | |
2618 | * Factor in the task's own contribution to the overage, such that four | |
2619 | * N-sized allocations are throttled approximately the same as one | |
2620 | * 4N-sized allocation. | |
2621 | * | |
2622 | * MEMCG_CHARGE_BATCH pages is nominal, so work out how much smaller or | |
2623 | * larger the current charge patch is than that. | |
2624 | */ | |
ff144e69 | 2625 | return penalty_jiffies * nr_pages / MEMCG_CHARGE_BATCH; |
e26733e0 CD |
2626 | } |
2627 | ||
2628 | /* | |
2629 | * Scheduled by try_charge() to be executed from the userland return path | |
2630 | * and reclaims memory over the high limit. | |
2631 | */ | |
2632 | void mem_cgroup_handle_over_high(void) | |
2633 | { | |
2634 | unsigned long penalty_jiffies; | |
2635 | unsigned long pflags; | |
b3ff9291 | 2636 | unsigned long nr_reclaimed; |
e26733e0 | 2637 | unsigned int nr_pages = current->memcg_nr_pages_over_high; |
d977aa93 | 2638 | int nr_retries = MAX_RECLAIM_RETRIES; |
e26733e0 | 2639 | struct mem_cgroup *memcg; |
b3ff9291 | 2640 | bool in_retry = false; |
e26733e0 CD |
2641 | |
2642 | if (likely(!nr_pages)) | |
2643 | return; | |
2644 | ||
2645 | memcg = get_mem_cgroup_from_mm(current->mm); | |
e26733e0 CD |
2646 | current->memcg_nr_pages_over_high = 0; |
2647 | ||
b3ff9291 CD |
2648 | retry_reclaim: |
2649 | /* | |
2650 | * The allocating task should reclaim at least the batch size, but for | |
2651 | * subsequent retries we only want to do what's necessary to prevent oom | |
2652 | * or breaching resource isolation. | |
2653 | * | |
2654 | * This is distinct from memory.max or page allocator behaviour because | |
2655 | * memory.high is currently batched, whereas memory.max and the page | |
2656 | * allocator run every time an allocation is made. | |
2657 | */ | |
2658 | nr_reclaimed = reclaim_high(memcg, | |
2659 | in_retry ? SWAP_CLUSTER_MAX : nr_pages, | |
2660 | GFP_KERNEL); | |
2661 | ||
e26733e0 CD |
2662 | /* |
2663 | * memory.high is breached and reclaim is unable to keep up. Throttle | |
2664 | * allocators proactively to slow down excessive growth. | |
2665 | */ | |
8a5dbc65 JK |
2666 | penalty_jiffies = calculate_high_delay(memcg, nr_pages, |
2667 | mem_find_max_overage(memcg)); | |
0e4b01df | 2668 | |
4b82ab4f JK |
2669 | penalty_jiffies += calculate_high_delay(memcg, nr_pages, |
2670 | swap_find_max_overage(memcg)); | |
2671 | ||
ff144e69 JK |
2672 | /* |
2673 | * Clamp the max delay per usermode return so as to still keep the | |
2674 | * application moving forwards and also permit diagnostics, albeit | |
2675 | * extremely slowly. | |
2676 | */ | |
2677 | penalty_jiffies = min(penalty_jiffies, MEMCG_MAX_HIGH_DELAY_JIFFIES); | |
2678 | ||
0e4b01df CD |
2679 | /* |
2680 | * Don't sleep if the amount of jiffies this memcg owes us is so low | |
2681 | * that it's not even worth doing, in an attempt to be nice to those who | |
2682 | * go only a small amount over their memory.high value and maybe haven't | |
2683 | * been aggressively reclaimed enough yet. | |
2684 | */ | |
2685 | if (penalty_jiffies <= HZ / 100) | |
2686 | goto out; | |
2687 | ||
b3ff9291 CD |
2688 | /* |
2689 | * If reclaim is making forward progress but we're still over | |
2690 | * memory.high, we want to encourage that rather than doing allocator | |
2691 | * throttling. | |
2692 | */ | |
2693 | if (nr_reclaimed || nr_retries--) { | |
2694 | in_retry = true; | |
2695 | goto retry_reclaim; | |
2696 | } | |
2697 | ||
0e4b01df CD |
2698 | /* |
2699 | * If we exit early, we're guaranteed to die (since | |
2700 | * schedule_timeout_killable sets TASK_KILLABLE). This means we don't | |
2701 | * need to account for any ill-begotten jiffies to pay them off later. | |
2702 | */ | |
2703 | psi_memstall_enter(&pflags); | |
2704 | schedule_timeout_killable(penalty_jiffies); | |
2705 | psi_memstall_leave(&pflags); | |
2706 | ||
2707 | out: | |
2708 | css_put(&memcg->css); | |
b23afb93 TH |
2709 | } |
2710 | ||
00501b53 JW |
2711 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2712 | unsigned int nr_pages) | |
8a9f3ccd | 2713 | { |
a983b5eb | 2714 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
d977aa93 | 2715 | int nr_retries = MAX_RECLAIM_RETRIES; |
6539cc05 | 2716 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2717 | struct page_counter *counter; |
e22c6ed9 | 2718 | enum oom_status oom_status; |
6539cc05 | 2719 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2720 | bool may_swap = true; |
2721 | bool drained = false; | |
e22c6ed9 | 2722 | unsigned long pflags; |
a636b327 | 2723 | |
ce00a967 | 2724 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 2725 | return 0; |
6539cc05 | 2726 | retry: |
b6b6cc72 | 2727 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2728 | return 0; |
8a9f3ccd | 2729 | |
7941d214 | 2730 | if (!do_memsw_account() || |
6071ca52 JW |
2731 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2732 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2733 | goto done_restock; |
7941d214 | 2734 | if (do_memsw_account()) |
3e32cb2e JW |
2735 | page_counter_uncharge(&memcg->memsw, batch); |
2736 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2737 | } else { |
3e32cb2e | 2738 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2739 | may_swap = false; |
3fbe7244 | 2740 | } |
7a81b88c | 2741 | |
6539cc05 JW |
2742 | if (batch > nr_pages) { |
2743 | batch = nr_pages; | |
2744 | goto retry; | |
2745 | } | |
6d61ef40 | 2746 | |
869712fd JW |
2747 | /* |
2748 | * Memcg doesn't have a dedicated reserve for atomic | |
2749 | * allocations. But like the global atomic pool, we need to | |
2750 | * put the burden of reclaim on regular allocation requests | |
2751 | * and let these go through as privileged allocations. | |
2752 | */ | |
2753 | if (gfp_mask & __GFP_ATOMIC) | |
2754 | goto force; | |
2755 | ||
06b078fc JW |
2756 | /* |
2757 | * Unlike in global OOM situations, memcg is not in a physical | |
2758 | * memory shortage. Allow dying and OOM-killed tasks to | |
2759 | * bypass the last charges so that they can exit quickly and | |
2760 | * free their memory. | |
2761 | */ | |
7775face | 2762 | if (unlikely(should_force_charge())) |
10d53c74 | 2763 | goto force; |
06b078fc | 2764 | |
89a28483 JW |
2765 | /* |
2766 | * Prevent unbounded recursion when reclaim operations need to | |
2767 | * allocate memory. This might exceed the limits temporarily, | |
2768 | * but we prefer facilitating memory reclaim and getting back | |
2769 | * under the limit over triggering OOM kills in these cases. | |
2770 | */ | |
2771 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2772 | goto force; | |
2773 | ||
06b078fc JW |
2774 | if (unlikely(task_in_memcg_oom(current))) |
2775 | goto nomem; | |
2776 | ||
d0164adc | 2777 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2778 | goto nomem; |
4b534334 | 2779 | |
e27be240 | 2780 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
241994ed | 2781 | |
e22c6ed9 | 2782 | psi_memstall_enter(&pflags); |
b70a2a21 JW |
2783 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2784 | gfp_mask, may_swap); | |
e22c6ed9 | 2785 | psi_memstall_leave(&pflags); |
6539cc05 | 2786 | |
61e02c74 | 2787 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2788 | goto retry; |
28c34c29 | 2789 | |
b70a2a21 | 2790 | if (!drained) { |
6d3d6aa2 | 2791 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2792 | drained = true; |
2793 | goto retry; | |
2794 | } | |
2795 | ||
28c34c29 JW |
2796 | if (gfp_mask & __GFP_NORETRY) |
2797 | goto nomem; | |
6539cc05 JW |
2798 | /* |
2799 | * Even though the limit is exceeded at this point, reclaim | |
2800 | * may have been able to free some pages. Retry the charge | |
2801 | * before killing the task. | |
2802 | * | |
2803 | * Only for regular pages, though: huge pages are rather | |
2804 | * unlikely to succeed so close to the limit, and we fall back | |
2805 | * to regular pages anyway in case of failure. | |
2806 | */ | |
61e02c74 | 2807 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2808 | goto retry; |
2809 | /* | |
2810 | * At task move, charge accounts can be doubly counted. So, it's | |
2811 | * better to wait until the end of task_move if something is going on. | |
2812 | */ | |
2813 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2814 | goto retry; | |
2815 | ||
9b130619 JW |
2816 | if (nr_retries--) |
2817 | goto retry; | |
2818 | ||
38d38493 | 2819 | if (gfp_mask & __GFP_RETRY_MAYFAIL) |
29ef680a MH |
2820 | goto nomem; |
2821 | ||
06b078fc | 2822 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2823 | goto force; |
06b078fc | 2824 | |
6539cc05 | 2825 | if (fatal_signal_pending(current)) |
10d53c74 | 2826 | goto force; |
6539cc05 | 2827 | |
29ef680a MH |
2828 | /* |
2829 | * keep retrying as long as the memcg oom killer is able to make | |
2830 | * a forward progress or bypass the charge if the oom killer | |
2831 | * couldn't make any progress. | |
2832 | */ | |
2833 | oom_status = mem_cgroup_oom(mem_over_limit, gfp_mask, | |
3608de07 | 2834 | get_order(nr_pages * PAGE_SIZE)); |
29ef680a MH |
2835 | switch (oom_status) { |
2836 | case OOM_SUCCESS: | |
d977aa93 | 2837 | nr_retries = MAX_RECLAIM_RETRIES; |
29ef680a MH |
2838 | goto retry; |
2839 | case OOM_FAILED: | |
2840 | goto force; | |
2841 | default: | |
2842 | goto nomem; | |
2843 | } | |
7a81b88c | 2844 | nomem: |
6d1fdc48 | 2845 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2846 | return -ENOMEM; |
10d53c74 TH |
2847 | force: |
2848 | /* | |
2849 | * The allocation either can't fail or will lead to more memory | |
2850 | * being freed very soon. Allow memory usage go over the limit | |
2851 | * temporarily by force charging it. | |
2852 | */ | |
2853 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2854 | if (do_memsw_account()) |
10d53c74 | 2855 | page_counter_charge(&memcg->memsw, nr_pages); |
10d53c74 TH |
2856 | |
2857 | return 0; | |
6539cc05 JW |
2858 | |
2859 | done_restock: | |
2860 | if (batch > nr_pages) | |
2861 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2862 | |
241994ed | 2863 | /* |
b23afb93 TH |
2864 | * If the hierarchy is above the normal consumption range, schedule |
2865 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2866 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2867 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2868 | * not recorded as it most likely matches current's and won't | |
2869 | * change in the meantime. As high limit is checked again before | |
2870 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2871 | */ |
2872 | do { | |
4b82ab4f JK |
2873 | bool mem_high, swap_high; |
2874 | ||
2875 | mem_high = page_counter_read(&memcg->memory) > | |
2876 | READ_ONCE(memcg->memory.high); | |
2877 | swap_high = page_counter_read(&memcg->swap) > | |
2878 | READ_ONCE(memcg->swap.high); | |
2879 | ||
2880 | /* Don't bother a random interrupted task */ | |
2881 | if (in_interrupt()) { | |
2882 | if (mem_high) { | |
f7e1cb6e JW |
2883 | schedule_work(&memcg->high_work); |
2884 | break; | |
2885 | } | |
4b82ab4f JK |
2886 | continue; |
2887 | } | |
2888 | ||
2889 | if (mem_high || swap_high) { | |
2890 | /* | |
2891 | * The allocating tasks in this cgroup will need to do | |
2892 | * reclaim or be throttled to prevent further growth | |
2893 | * of the memory or swap footprints. | |
2894 | * | |
2895 | * Target some best-effort fairness between the tasks, | |
2896 | * and distribute reclaim work and delay penalties | |
2897 | * based on how much each task is actually allocating. | |
2898 | */ | |
9516a18a | 2899 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2900 | set_notify_resume(current); |
2901 | break; | |
2902 | } | |
241994ed | 2903 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2904 | |
2905 | return 0; | |
7a81b88c | 2906 | } |
8a9f3ccd | 2907 | |
f0e45fb4 | 2908 | #if defined(CONFIG_MEMCG_KMEM) || defined(CONFIG_MMU) |
00501b53 | 2909 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2910 | { |
ce00a967 JW |
2911 | if (mem_cgroup_is_root(memcg)) |
2912 | return; | |
2913 | ||
3e32cb2e | 2914 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2915 | if (do_memsw_account()) |
3e32cb2e | 2916 | page_counter_uncharge(&memcg->memsw, nr_pages); |
d01dd17f | 2917 | } |
f0e45fb4 | 2918 | #endif |
d01dd17f | 2919 | |
d9eb1ea2 | 2920 | static void commit_charge(struct page *page, struct mem_cgroup *memcg) |
0a31bc97 | 2921 | { |
bcfe06bf | 2922 | VM_BUG_ON_PAGE(page_memcg(page), page); |
0a31bc97 | 2923 | /* |
a5eb011a | 2924 | * Any of the following ensures page's memcg stability: |
0a31bc97 | 2925 | * |
a0b5b414 JW |
2926 | * - the page lock |
2927 | * - LRU isolation | |
2928 | * - lock_page_memcg() | |
2929 | * - exclusive reference | |
0a31bc97 | 2930 | */ |
bcfe06bf | 2931 | page->memcg_data = (unsigned long)memcg; |
7a81b88c | 2932 | } |
66e1707b | 2933 | |
84c07d11 | 2934 | #ifdef CONFIG_MEMCG_KMEM |
10befea9 | 2935 | int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s, |
2e9bd483 | 2936 | gfp_t gfp, bool new_page) |
10befea9 RG |
2937 | { |
2938 | unsigned int objects = objs_per_slab_page(s, page); | |
2e9bd483 | 2939 | unsigned long memcg_data; |
10befea9 RG |
2940 | void *vec; |
2941 | ||
2942 | vec = kcalloc_node(objects, sizeof(struct obj_cgroup *), gfp, | |
2943 | page_to_nid(page)); | |
2944 | if (!vec) | |
2945 | return -ENOMEM; | |
2946 | ||
2e9bd483 RG |
2947 | memcg_data = (unsigned long) vec | MEMCG_DATA_OBJCGS; |
2948 | if (new_page) { | |
2949 | /* | |
2950 | * If the slab page is brand new and nobody can yet access | |
2951 | * it's memcg_data, no synchronization is required and | |
2952 | * memcg_data can be simply assigned. | |
2953 | */ | |
2954 | page->memcg_data = memcg_data; | |
2955 | } else if (cmpxchg(&page->memcg_data, 0, memcg_data)) { | |
2956 | /* | |
2957 | * If the slab page is already in use, somebody can allocate | |
2958 | * and assign obj_cgroups in parallel. In this case the existing | |
2959 | * objcg vector should be reused. | |
2960 | */ | |
10befea9 | 2961 | kfree(vec); |
2e9bd483 RG |
2962 | return 0; |
2963 | } | |
10befea9 | 2964 | |
2e9bd483 | 2965 | kmemleak_not_leak(vec); |
10befea9 RG |
2966 | return 0; |
2967 | } | |
2968 | ||
8380ce47 RG |
2969 | /* |
2970 | * Returns a pointer to the memory cgroup to which the kernel object is charged. | |
2971 | * | |
bcfe06bf RG |
2972 | * A passed kernel object can be a slab object or a generic kernel page, so |
2973 | * different mechanisms for getting the memory cgroup pointer should be used. | |
2974 | * In certain cases (e.g. kernel stacks or large kmallocs with SLUB) the caller | |
2975 | * can not know for sure how the kernel object is implemented. | |
2976 | * mem_cgroup_from_obj() can be safely used in such cases. | |
2977 | * | |
8380ce47 RG |
2978 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), |
2979 | * cgroup_mutex, etc. | |
2980 | */ | |
2981 | struct mem_cgroup *mem_cgroup_from_obj(void *p) | |
2982 | { | |
2983 | struct page *page; | |
2984 | ||
2985 | if (mem_cgroup_disabled()) | |
2986 | return NULL; | |
2987 | ||
2988 | page = virt_to_head_page(p); | |
2989 | ||
2990 | /* | |
9855609b RG |
2991 | * Slab objects are accounted individually, not per-page. |
2992 | * Memcg membership data for each individual object is saved in | |
2993 | * the page->obj_cgroups. | |
8380ce47 | 2994 | */ |
270c6a71 | 2995 | if (page_objcgs_check(page)) { |
9855609b RG |
2996 | struct obj_cgroup *objcg; |
2997 | unsigned int off; | |
2998 | ||
2999 | off = obj_to_index(page->slab_cache, page, p); | |
270c6a71 | 3000 | objcg = page_objcgs(page)[off]; |
10befea9 RG |
3001 | if (objcg) |
3002 | return obj_cgroup_memcg(objcg); | |
3003 | ||
3004 | return NULL; | |
9855609b | 3005 | } |
8380ce47 | 3006 | |
bcfe06bf RG |
3007 | /* |
3008 | * page_memcg_check() is used here, because page_has_obj_cgroups() | |
3009 | * check above could fail because the object cgroups vector wasn't set | |
3010 | * at that moment, but it can be set concurrently. | |
3011 | * page_memcg_check(page) will guarantee that a proper memory | |
3012 | * cgroup pointer or NULL will be returned. | |
3013 | */ | |
3014 | return page_memcg_check(page); | |
8380ce47 RG |
3015 | } |
3016 | ||
bf4f0599 RG |
3017 | __always_inline struct obj_cgroup *get_obj_cgroup_from_current(void) |
3018 | { | |
3019 | struct obj_cgroup *objcg = NULL; | |
3020 | struct mem_cgroup *memcg; | |
3021 | ||
279c3393 RG |
3022 | if (memcg_kmem_bypass()) |
3023 | return NULL; | |
3024 | ||
bf4f0599 | 3025 | rcu_read_lock(); |
37d5985c RG |
3026 | if (unlikely(active_memcg())) |
3027 | memcg = active_memcg(); | |
bf4f0599 RG |
3028 | else |
3029 | memcg = mem_cgroup_from_task(current); | |
3030 | ||
3031 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { | |
3032 | objcg = rcu_dereference(memcg->objcg); | |
3033 | if (objcg && obj_cgroup_tryget(objcg)) | |
3034 | break; | |
2f7659a3 | 3035 | objcg = NULL; |
bf4f0599 RG |
3036 | } |
3037 | rcu_read_unlock(); | |
3038 | ||
3039 | return objcg; | |
3040 | } | |
3041 | ||
f3bb3043 | 3042 | static int memcg_alloc_cache_id(void) |
55007d84 | 3043 | { |
f3bb3043 VD |
3044 | int id, size; |
3045 | int err; | |
3046 | ||
dbcf73e2 | 3047 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
3048 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
3049 | if (id < 0) | |
3050 | return id; | |
55007d84 | 3051 | |
dbcf73e2 | 3052 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
3053 | return id; |
3054 | ||
3055 | /* | |
3056 | * There's no space for the new id in memcg_caches arrays, | |
3057 | * so we have to grow them. | |
3058 | */ | |
05257a1a | 3059 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
3060 | |
3061 | size = 2 * (id + 1); | |
55007d84 GC |
3062 | if (size < MEMCG_CACHES_MIN_SIZE) |
3063 | size = MEMCG_CACHES_MIN_SIZE; | |
3064 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
3065 | size = MEMCG_CACHES_MAX_SIZE; | |
3066 | ||
9855609b | 3067 | err = memcg_update_all_list_lrus(size); |
05257a1a VD |
3068 | if (!err) |
3069 | memcg_nr_cache_ids = size; | |
3070 | ||
3071 | up_write(&memcg_cache_ids_sem); | |
3072 | ||
f3bb3043 | 3073 | if (err) { |
dbcf73e2 | 3074 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
3075 | return err; |
3076 | } | |
3077 | return id; | |
3078 | } | |
3079 | ||
3080 | static void memcg_free_cache_id(int id) | |
3081 | { | |
dbcf73e2 | 3082 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
3083 | } |
3084 | ||
45264778 | 3085 | /** |
4b13f64d | 3086 | * __memcg_kmem_charge: charge a number of kernel pages to a memcg |
10eaec2f | 3087 | * @memcg: memory cgroup to charge |
45264778 | 3088 | * @gfp: reclaim mode |
92d0510c | 3089 | * @nr_pages: number of pages to charge |
45264778 VD |
3090 | * |
3091 | * Returns 0 on success, an error code on failure. | |
3092 | */ | |
4b13f64d RG |
3093 | int __memcg_kmem_charge(struct mem_cgroup *memcg, gfp_t gfp, |
3094 | unsigned int nr_pages) | |
7ae1e1d0 | 3095 | { |
f3ccb2c4 | 3096 | struct page_counter *counter; |
7ae1e1d0 GC |
3097 | int ret; |
3098 | ||
f3ccb2c4 | 3099 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 3100 | if (ret) |
f3ccb2c4 | 3101 | return ret; |
52c29b04 JW |
3102 | |
3103 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
3104 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
e55d9d9b MH |
3105 | |
3106 | /* | |
3107 | * Enforce __GFP_NOFAIL allocation because callers are not | |
3108 | * prepared to see failures and likely do not have any failure | |
3109 | * handling code. | |
3110 | */ | |
3111 | if (gfp & __GFP_NOFAIL) { | |
3112 | page_counter_charge(&memcg->kmem, nr_pages); | |
3113 | return 0; | |
3114 | } | |
52c29b04 JW |
3115 | cancel_charge(memcg, nr_pages); |
3116 | return -ENOMEM; | |
7ae1e1d0 | 3117 | } |
f3ccb2c4 | 3118 | return 0; |
7ae1e1d0 GC |
3119 | } |
3120 | ||
4b13f64d RG |
3121 | /** |
3122 | * __memcg_kmem_uncharge: uncharge a number of kernel pages from a memcg | |
3123 | * @memcg: memcg to uncharge | |
3124 | * @nr_pages: number of pages to uncharge | |
3125 | */ | |
3126 | void __memcg_kmem_uncharge(struct mem_cgroup *memcg, unsigned int nr_pages) | |
3127 | { | |
3128 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
3129 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
3130 | ||
3de7d4f2 | 3131 | refill_stock(memcg, nr_pages); |
4b13f64d RG |
3132 | } |
3133 | ||
45264778 | 3134 | /** |
f4b00eab | 3135 | * __memcg_kmem_charge_page: charge a kmem page to the current memory cgroup |
45264778 VD |
3136 | * @page: page to charge |
3137 | * @gfp: reclaim mode | |
3138 | * @order: allocation order | |
3139 | * | |
3140 | * Returns 0 on success, an error code on failure. | |
3141 | */ | |
f4b00eab | 3142 | int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 3143 | { |
f3ccb2c4 | 3144 | struct mem_cgroup *memcg; |
fcff7d7e | 3145 | int ret = 0; |
7ae1e1d0 | 3146 | |
d46eb14b | 3147 | memcg = get_mem_cgroup_from_current(); |
279c3393 | 3148 | if (memcg && !mem_cgroup_is_root(memcg)) { |
4b13f64d | 3149 | ret = __memcg_kmem_charge(memcg, gfp, 1 << order); |
4d96ba35 | 3150 | if (!ret) { |
18b2db3b RG |
3151 | page->memcg_data = (unsigned long)memcg | |
3152 | MEMCG_DATA_KMEM; | |
1a3e1f40 | 3153 | return 0; |
4d96ba35 | 3154 | } |
279c3393 | 3155 | css_put(&memcg->css); |
c4159a75 | 3156 | } |
d05e83a6 | 3157 | return ret; |
7ae1e1d0 | 3158 | } |
49a18eae | 3159 | |
45264778 | 3160 | /** |
f4b00eab | 3161 | * __memcg_kmem_uncharge_page: uncharge a kmem page |
45264778 VD |
3162 | * @page: page to uncharge |
3163 | * @order: allocation order | |
3164 | */ | |
f4b00eab | 3165 | void __memcg_kmem_uncharge_page(struct page *page, int order) |
7ae1e1d0 | 3166 | { |
bcfe06bf | 3167 | struct mem_cgroup *memcg = page_memcg(page); |
f3ccb2c4 | 3168 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 3169 | |
7ae1e1d0 GC |
3170 | if (!memcg) |
3171 | return; | |
3172 | ||
309381fe | 3173 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
4b13f64d | 3174 | __memcg_kmem_uncharge(memcg, nr_pages); |
bcfe06bf | 3175 | page->memcg_data = 0; |
1a3e1f40 | 3176 | css_put(&memcg->css); |
60d3fd32 | 3177 | } |
bf4f0599 RG |
3178 | |
3179 | static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) | |
3180 | { | |
3181 | struct memcg_stock_pcp *stock; | |
3182 | unsigned long flags; | |
3183 | bool ret = false; | |
3184 | ||
3185 | local_irq_save(flags); | |
3186 | ||
3187 | stock = this_cpu_ptr(&memcg_stock); | |
3188 | if (objcg == stock->cached_objcg && stock->nr_bytes >= nr_bytes) { | |
3189 | stock->nr_bytes -= nr_bytes; | |
3190 | ret = true; | |
3191 | } | |
3192 | ||
3193 | local_irq_restore(flags); | |
3194 | ||
3195 | return ret; | |
3196 | } | |
3197 | ||
3198 | static void drain_obj_stock(struct memcg_stock_pcp *stock) | |
3199 | { | |
3200 | struct obj_cgroup *old = stock->cached_objcg; | |
3201 | ||
3202 | if (!old) | |
3203 | return; | |
3204 | ||
3205 | if (stock->nr_bytes) { | |
3206 | unsigned int nr_pages = stock->nr_bytes >> PAGE_SHIFT; | |
3207 | unsigned int nr_bytes = stock->nr_bytes & (PAGE_SIZE - 1); | |
3208 | ||
3209 | if (nr_pages) { | |
3210 | rcu_read_lock(); | |
3211 | __memcg_kmem_uncharge(obj_cgroup_memcg(old), nr_pages); | |
3212 | rcu_read_unlock(); | |
3213 | } | |
3214 | ||
3215 | /* | |
3216 | * The leftover is flushed to the centralized per-memcg value. | |
3217 | * On the next attempt to refill obj stock it will be moved | |
3218 | * to a per-cpu stock (probably, on an other CPU), see | |
3219 | * refill_obj_stock(). | |
3220 | * | |
3221 | * How often it's flushed is a trade-off between the memory | |
3222 | * limit enforcement accuracy and potential CPU contention, | |
3223 | * so it might be changed in the future. | |
3224 | */ | |
3225 | atomic_add(nr_bytes, &old->nr_charged_bytes); | |
3226 | stock->nr_bytes = 0; | |
3227 | } | |
3228 | ||
3229 | obj_cgroup_put(old); | |
3230 | stock->cached_objcg = NULL; | |
3231 | } | |
3232 | ||
3233 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
3234 | struct mem_cgroup *root_memcg) | |
3235 | { | |
3236 | struct mem_cgroup *memcg; | |
3237 | ||
3238 | if (stock->cached_objcg) { | |
3239 | memcg = obj_cgroup_memcg(stock->cached_objcg); | |
3240 | if (memcg && mem_cgroup_is_descendant(memcg, root_memcg)) | |
3241 | return true; | |
3242 | } | |
3243 | ||
3244 | return false; | |
3245 | } | |
3246 | ||
3247 | static void refill_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) | |
3248 | { | |
3249 | struct memcg_stock_pcp *stock; | |
3250 | unsigned long flags; | |
3251 | ||
3252 | local_irq_save(flags); | |
3253 | ||
3254 | stock = this_cpu_ptr(&memcg_stock); | |
3255 | if (stock->cached_objcg != objcg) { /* reset if necessary */ | |
3256 | drain_obj_stock(stock); | |
3257 | obj_cgroup_get(objcg); | |
3258 | stock->cached_objcg = objcg; | |
3259 | stock->nr_bytes = atomic_xchg(&objcg->nr_charged_bytes, 0); | |
3260 | } | |
3261 | stock->nr_bytes += nr_bytes; | |
3262 | ||
3263 | if (stock->nr_bytes > PAGE_SIZE) | |
3264 | drain_obj_stock(stock); | |
3265 | ||
3266 | local_irq_restore(flags); | |
3267 | } | |
3268 | ||
3269 | int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size) | |
3270 | { | |
3271 | struct mem_cgroup *memcg; | |
3272 | unsigned int nr_pages, nr_bytes; | |
3273 | int ret; | |
3274 | ||
3275 | if (consume_obj_stock(objcg, size)) | |
3276 | return 0; | |
3277 | ||
3278 | /* | |
3279 | * In theory, memcg->nr_charged_bytes can have enough | |
3280 | * pre-charged bytes to satisfy the allocation. However, | |
3281 | * flushing memcg->nr_charged_bytes requires two atomic | |
3282 | * operations, and memcg->nr_charged_bytes can't be big, | |
3283 | * so it's better to ignore it and try grab some new pages. | |
3284 | * memcg->nr_charged_bytes will be flushed in | |
3285 | * refill_obj_stock(), called from this function or | |
3286 | * independently later. | |
3287 | */ | |
3288 | rcu_read_lock(); | |
eefbfa7f | 3289 | retry: |
bf4f0599 | 3290 | memcg = obj_cgroup_memcg(objcg); |
eefbfa7f MS |
3291 | if (unlikely(!css_tryget(&memcg->css))) |
3292 | goto retry; | |
bf4f0599 RG |
3293 | rcu_read_unlock(); |
3294 | ||
3295 | nr_pages = size >> PAGE_SHIFT; | |
3296 | nr_bytes = size & (PAGE_SIZE - 1); | |
3297 | ||
3298 | if (nr_bytes) | |
3299 | nr_pages += 1; | |
3300 | ||
3301 | ret = __memcg_kmem_charge(memcg, gfp, nr_pages); | |
3302 | if (!ret && nr_bytes) | |
3303 | refill_obj_stock(objcg, PAGE_SIZE - nr_bytes); | |
3304 | ||
3305 | css_put(&memcg->css); | |
3306 | return ret; | |
3307 | } | |
3308 | ||
3309 | void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size) | |
3310 | { | |
3311 | refill_obj_stock(objcg, size); | |
3312 | } | |
3313 | ||
84c07d11 | 3314 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 3315 | |
ca3e0214 | 3316 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
ca3e0214 | 3317 | /* |
6168d0da | 3318 | * Because page_memcg(head) is not set on compound tails, set it now. |
ca3e0214 | 3319 | */ |
e94c8a9c | 3320 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 3321 | { |
bcfe06bf | 3322 | struct mem_cgroup *memcg = page_memcg(head); |
e94c8a9c | 3323 | int i; |
ca3e0214 | 3324 | |
3d37c4a9 KH |
3325 | if (mem_cgroup_disabled()) |
3326 | return; | |
b070e65c | 3327 | |
1a3e1f40 JW |
3328 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
3329 | css_get(&memcg->css); | |
bcfe06bf | 3330 | head[i].memcg_data = (unsigned long)memcg; |
1a3e1f40 | 3331 | } |
ca3e0214 | 3332 | } |
12d27107 | 3333 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3334 | |
c255a458 | 3335 | #ifdef CONFIG_MEMCG_SWAP |
02491447 DN |
3336 | /** |
3337 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3338 | * @entry: swap entry to be moved | |
3339 | * @from: mem_cgroup which the entry is moved from | |
3340 | * @to: mem_cgroup which the entry is moved to | |
3341 | * | |
3342 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3343 | * as the mem_cgroup's id of @from. | |
3344 | * | |
3345 | * Returns 0 on success, -EINVAL on failure. | |
3346 | * | |
3e32cb2e | 3347 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
3348 | * both res and memsw, and called css_get(). |
3349 | */ | |
3350 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3351 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3352 | { |
3353 | unsigned short old_id, new_id; | |
3354 | ||
34c00c31 LZ |
3355 | old_id = mem_cgroup_id(from); |
3356 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3357 | |
3358 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
3359 | mod_memcg_state(from, MEMCG_SWAP, -1); |
3360 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
3361 | return 0; |
3362 | } | |
3363 | return -EINVAL; | |
3364 | } | |
3365 | #else | |
3366 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3367 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3368 | { |
3369 | return -EINVAL; | |
3370 | } | |
8c7c6e34 | 3371 | #endif |
d13d1443 | 3372 | |
bbec2e15 | 3373 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 3374 | |
bbec2e15 RG |
3375 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
3376 | unsigned long max, bool memsw) | |
628f4235 | 3377 | { |
3e32cb2e | 3378 | bool enlarge = false; |
bb4a7ea2 | 3379 | bool drained = false; |
3e32cb2e | 3380 | int ret; |
c054a78c YZ |
3381 | bool limits_invariant; |
3382 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 3383 | |
3e32cb2e | 3384 | do { |
628f4235 KH |
3385 | if (signal_pending(current)) { |
3386 | ret = -EINTR; | |
3387 | break; | |
3388 | } | |
3e32cb2e | 3389 | |
bbec2e15 | 3390 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
3391 | /* |
3392 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 3393 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 3394 | */ |
15b42562 | 3395 | limits_invariant = memsw ? max >= READ_ONCE(memcg->memory.max) : |
bbec2e15 | 3396 | max <= memcg->memsw.max; |
c054a78c | 3397 | if (!limits_invariant) { |
bbec2e15 | 3398 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 3399 | ret = -EINVAL; |
8c7c6e34 KH |
3400 | break; |
3401 | } | |
bbec2e15 | 3402 | if (max > counter->max) |
3e32cb2e | 3403 | enlarge = true; |
bbec2e15 RG |
3404 | ret = page_counter_set_max(counter, max); |
3405 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
3406 | |
3407 | if (!ret) | |
3408 | break; | |
3409 | ||
bb4a7ea2 SB |
3410 | if (!drained) { |
3411 | drain_all_stock(memcg); | |
3412 | drained = true; | |
3413 | continue; | |
3414 | } | |
3415 | ||
1ab5c056 AR |
3416 | if (!try_to_free_mem_cgroup_pages(memcg, 1, |
3417 | GFP_KERNEL, !memsw)) { | |
3418 | ret = -EBUSY; | |
3419 | break; | |
3420 | } | |
3421 | } while (true); | |
3e32cb2e | 3422 | |
3c11ecf4 KH |
3423 | if (!ret && enlarge) |
3424 | memcg_oom_recover(memcg); | |
3e32cb2e | 3425 | |
628f4235 KH |
3426 | return ret; |
3427 | } | |
3428 | ||
ef8f2327 | 3429 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
3430 | gfp_t gfp_mask, |
3431 | unsigned long *total_scanned) | |
3432 | { | |
3433 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 3434 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
3435 | unsigned long reclaimed; |
3436 | int loop = 0; | |
ef8f2327 | 3437 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 3438 | unsigned long excess; |
0608f43d AM |
3439 | unsigned long nr_scanned; |
3440 | ||
3441 | if (order > 0) | |
3442 | return 0; | |
3443 | ||
ef8f2327 | 3444 | mctz = soft_limit_tree_node(pgdat->node_id); |
d6507ff5 MH |
3445 | |
3446 | /* | |
3447 | * Do not even bother to check the largest node if the root | |
3448 | * is empty. Do it lockless to prevent lock bouncing. Races | |
3449 | * are acceptable as soft limit is best effort anyway. | |
3450 | */ | |
bfc7228b | 3451 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
3452 | return 0; |
3453 | ||
0608f43d AM |
3454 | /* |
3455 | * This loop can run a while, specially if mem_cgroup's continuously | |
3456 | * keep exceeding their soft limit and putting the system under | |
3457 | * pressure | |
3458 | */ | |
3459 | do { | |
3460 | if (next_mz) | |
3461 | mz = next_mz; | |
3462 | else | |
3463 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3464 | if (!mz) | |
3465 | break; | |
3466 | ||
3467 | nr_scanned = 0; | |
ef8f2327 | 3468 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
0608f43d AM |
3469 | gfp_mask, &nr_scanned); |
3470 | nr_reclaimed += reclaimed; | |
3471 | *total_scanned += nr_scanned; | |
0a31bc97 | 3472 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 3473 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
3474 | |
3475 | /* | |
3476 | * If we failed to reclaim anything from this memory cgroup | |
3477 | * it is time to move on to the next cgroup | |
3478 | */ | |
3479 | next_mz = NULL; | |
bc2f2e7f VD |
3480 | if (!reclaimed) |
3481 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3482 | ||
3e32cb2e | 3483 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3484 | /* |
3485 | * One school of thought says that we should not add | |
3486 | * back the node to the tree if reclaim returns 0. | |
3487 | * But our reclaim could return 0, simply because due | |
3488 | * to priority we are exposing a smaller subset of | |
3489 | * memory to reclaim from. Consider this as a longer | |
3490 | * term TODO. | |
3491 | */ | |
3492 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3493 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3494 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3495 | css_put(&mz->memcg->css); |
3496 | loop++; | |
3497 | /* | |
3498 | * Could not reclaim anything and there are no more | |
3499 | * mem cgroups to try or we seem to be looping without | |
3500 | * reclaiming anything. | |
3501 | */ | |
3502 | if (!nr_reclaimed && | |
3503 | (next_mz == NULL || | |
3504 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3505 | break; | |
3506 | } while (!nr_reclaimed); | |
3507 | if (next_mz) | |
3508 | css_put(&next_mz->memcg->css); | |
3509 | return nr_reclaimed; | |
3510 | } | |
3511 | ||
c26251f9 | 3512 | /* |
51038171 | 3513 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
3514 | * |
3515 | * Caller is responsible for holding css reference for memcg. | |
3516 | */ | |
3517 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3518 | { | |
d977aa93 | 3519 | int nr_retries = MAX_RECLAIM_RETRIES; |
c26251f9 | 3520 | |
c1e862c1 KH |
3521 | /* we call try-to-free pages for make this cgroup empty */ |
3522 | lru_add_drain_all(); | |
d12c60f6 JS |
3523 | |
3524 | drain_all_stock(memcg); | |
3525 | ||
f817ed48 | 3526 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3527 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 3528 | int progress; |
c1e862c1 | 3529 | |
c26251f9 MH |
3530 | if (signal_pending(current)) |
3531 | return -EINTR; | |
3532 | ||
b70a2a21 JW |
3533 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3534 | GFP_KERNEL, true); | |
c1e862c1 | 3535 | if (!progress) { |
f817ed48 | 3536 | nr_retries--; |
c1e862c1 | 3537 | /* maybe some writeback is necessary */ |
8aa7e847 | 3538 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3539 | } |
f817ed48 KH |
3540 | |
3541 | } | |
ab5196c2 MH |
3542 | |
3543 | return 0; | |
cc847582 KH |
3544 | } |
3545 | ||
6770c64e TH |
3546 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3547 | char *buf, size_t nbytes, | |
3548 | loff_t off) | |
c1e862c1 | 3549 | { |
6770c64e | 3550 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3551 | |
d8423011 MH |
3552 | if (mem_cgroup_is_root(memcg)) |
3553 | return -EINVAL; | |
6770c64e | 3554 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3555 | } |
3556 | ||
182446d0 TH |
3557 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3558 | struct cftype *cft) | |
18f59ea7 | 3559 | { |
bef8620c | 3560 | return 1; |
18f59ea7 BS |
3561 | } |
3562 | ||
182446d0 TH |
3563 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3564 | struct cftype *cft, u64 val) | |
18f59ea7 | 3565 | { |
bef8620c | 3566 | if (val == 1) |
0b8f73e1 | 3567 | return 0; |
567fb435 | 3568 | |
bef8620c RG |
3569 | pr_warn_once("Non-hierarchical mode is deprecated. " |
3570 | "Please report your usecase to linux-mm@kvack.org if you " | |
3571 | "depend on this functionality.\n"); | |
567fb435 | 3572 | |
bef8620c | 3573 | return -EINVAL; |
18f59ea7 BS |
3574 | } |
3575 | ||
6f646156 | 3576 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 3577 | { |
42a30035 | 3578 | unsigned long val; |
ce00a967 | 3579 | |
3e32cb2e | 3580 | if (mem_cgroup_is_root(memcg)) { |
0d1c2072 | 3581 | val = memcg_page_state(memcg, NR_FILE_PAGES) + |
be5d0a74 | 3582 | memcg_page_state(memcg, NR_ANON_MAPPED); |
42a30035 JW |
3583 | if (swap) |
3584 | val += memcg_page_state(memcg, MEMCG_SWAP); | |
3e32cb2e | 3585 | } else { |
ce00a967 | 3586 | if (!swap) |
3e32cb2e | 3587 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3588 | else |
3e32cb2e | 3589 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3590 | } |
c12176d3 | 3591 | return val; |
ce00a967 JW |
3592 | } |
3593 | ||
3e32cb2e JW |
3594 | enum { |
3595 | RES_USAGE, | |
3596 | RES_LIMIT, | |
3597 | RES_MAX_USAGE, | |
3598 | RES_FAILCNT, | |
3599 | RES_SOFT_LIMIT, | |
3600 | }; | |
ce00a967 | 3601 | |
791badbd | 3602 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3603 | struct cftype *cft) |
8cdea7c0 | 3604 | { |
182446d0 | 3605 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3606 | struct page_counter *counter; |
af36f906 | 3607 | |
3e32cb2e | 3608 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3609 | case _MEM: |
3e32cb2e JW |
3610 | counter = &memcg->memory; |
3611 | break; | |
8c7c6e34 | 3612 | case _MEMSWAP: |
3e32cb2e JW |
3613 | counter = &memcg->memsw; |
3614 | break; | |
510fc4e1 | 3615 | case _KMEM: |
3e32cb2e | 3616 | counter = &memcg->kmem; |
510fc4e1 | 3617 | break; |
d55f90bf | 3618 | case _TCP: |
0db15298 | 3619 | counter = &memcg->tcpmem; |
d55f90bf | 3620 | break; |
8c7c6e34 KH |
3621 | default: |
3622 | BUG(); | |
8c7c6e34 | 3623 | } |
3e32cb2e JW |
3624 | |
3625 | switch (MEMFILE_ATTR(cft->private)) { | |
3626 | case RES_USAGE: | |
3627 | if (counter == &memcg->memory) | |
c12176d3 | 3628 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 3629 | if (counter == &memcg->memsw) |
c12176d3 | 3630 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
3631 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
3632 | case RES_LIMIT: | |
bbec2e15 | 3633 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
3634 | case RES_MAX_USAGE: |
3635 | return (u64)counter->watermark * PAGE_SIZE; | |
3636 | case RES_FAILCNT: | |
3637 | return counter->failcnt; | |
3638 | case RES_SOFT_LIMIT: | |
3639 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3640 | default: | |
3641 | BUG(); | |
3642 | } | |
8cdea7c0 | 3643 | } |
510fc4e1 | 3644 | |
4a87e2a2 | 3645 | static void memcg_flush_percpu_vmstats(struct mem_cgroup *memcg) |
c350a99e | 3646 | { |
4a87e2a2 | 3647 | unsigned long stat[MEMCG_NR_STAT] = {0}; |
c350a99e RG |
3648 | struct mem_cgroup *mi; |
3649 | int node, cpu, i; | |
c350a99e RG |
3650 | |
3651 | for_each_online_cpu(cpu) | |
4a87e2a2 | 3652 | for (i = 0; i < MEMCG_NR_STAT; i++) |
6c1c2808 | 3653 | stat[i] += per_cpu(memcg->vmstats_percpu->stat[i], cpu); |
c350a99e RG |
3654 | |
3655 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
4a87e2a2 | 3656 | for (i = 0; i < MEMCG_NR_STAT; i++) |
c350a99e RG |
3657 | atomic_long_add(stat[i], &mi->vmstats[i]); |
3658 | ||
3659 | for_each_node(node) { | |
3660 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; | |
3661 | struct mem_cgroup_per_node *pi; | |
3662 | ||
4a87e2a2 | 3663 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
c350a99e RG |
3664 | stat[i] = 0; |
3665 | ||
3666 | for_each_online_cpu(cpu) | |
4a87e2a2 | 3667 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
6c1c2808 SB |
3668 | stat[i] += per_cpu( |
3669 | pn->lruvec_stat_cpu->count[i], cpu); | |
c350a99e RG |
3670 | |
3671 | for (pi = pn; pi; pi = parent_nodeinfo(pi, node)) | |
4a87e2a2 | 3672 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
c350a99e RG |
3673 | atomic_long_add(stat[i], &pi->lruvec_stat[i]); |
3674 | } | |
3675 | } | |
3676 | ||
bb65f89b RG |
3677 | static void memcg_flush_percpu_vmevents(struct mem_cgroup *memcg) |
3678 | { | |
3679 | unsigned long events[NR_VM_EVENT_ITEMS]; | |
3680 | struct mem_cgroup *mi; | |
3681 | int cpu, i; | |
3682 | ||
3683 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
3684 | events[i] = 0; | |
3685 | ||
3686 | for_each_online_cpu(cpu) | |
3687 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
6c1c2808 SB |
3688 | events[i] += per_cpu(memcg->vmstats_percpu->events[i], |
3689 | cpu); | |
bb65f89b RG |
3690 | |
3691 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
3692 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
3693 | atomic_long_add(events[i], &mi->vmevents[i]); | |
3694 | } | |
3695 | ||
84c07d11 | 3696 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 3697 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 3698 | { |
bf4f0599 | 3699 | struct obj_cgroup *objcg; |
d6441637 VD |
3700 | int memcg_id; |
3701 | ||
b313aeee VD |
3702 | if (cgroup_memory_nokmem) |
3703 | return 0; | |
3704 | ||
2a4db7eb | 3705 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 3706 | BUG_ON(memcg->kmem_state); |
d6441637 | 3707 | |
f3bb3043 | 3708 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
3709 | if (memcg_id < 0) |
3710 | return memcg_id; | |
d6441637 | 3711 | |
bf4f0599 RG |
3712 | objcg = obj_cgroup_alloc(); |
3713 | if (!objcg) { | |
3714 | memcg_free_cache_id(memcg_id); | |
3715 | return -ENOMEM; | |
3716 | } | |
3717 | objcg->memcg = memcg; | |
3718 | rcu_assign_pointer(memcg->objcg, objcg); | |
3719 | ||
d648bcc7 RG |
3720 | static_branch_enable(&memcg_kmem_enabled_key); |
3721 | ||
900a38f0 | 3722 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 3723 | memcg->kmem_state = KMEM_ONLINE; |
0b8f73e1 JW |
3724 | |
3725 | return 0; | |
d6441637 VD |
3726 | } |
3727 | ||
8e0a8912 JW |
3728 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
3729 | { | |
3730 | struct cgroup_subsys_state *css; | |
3731 | struct mem_cgroup *parent, *child; | |
3732 | int kmemcg_id; | |
3733 | ||
3734 | if (memcg->kmem_state != KMEM_ONLINE) | |
3735 | return; | |
9855609b | 3736 | |
8e0a8912 JW |
3737 | memcg->kmem_state = KMEM_ALLOCATED; |
3738 | ||
8e0a8912 JW |
3739 | parent = parent_mem_cgroup(memcg); |
3740 | if (!parent) | |
3741 | parent = root_mem_cgroup; | |
3742 | ||
bf4f0599 | 3743 | memcg_reparent_objcgs(memcg, parent); |
fb2f2b0a RG |
3744 | |
3745 | kmemcg_id = memcg->kmemcg_id; | |
3746 | BUG_ON(kmemcg_id < 0); | |
3747 | ||
8e0a8912 JW |
3748 | /* |
3749 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3750 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3751 | * to ones of the parent. After we have finished, all list_lrus | |
3752 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3753 | * ordering is imposed by list_lru_node->lock taken by | |
3754 | * memcg_drain_all_list_lrus(). | |
3755 | */ | |
3a06bb78 | 3756 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
8e0a8912 JW |
3757 | css_for_each_descendant_pre(css, &memcg->css) { |
3758 | child = mem_cgroup_from_css(css); | |
3759 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3760 | child->kmemcg_id = parent->kmemcg_id; | |
8e0a8912 | 3761 | } |
3a06bb78 TH |
3762 | rcu_read_unlock(); |
3763 | ||
9bec5c35 | 3764 | memcg_drain_all_list_lrus(kmemcg_id, parent); |
8e0a8912 JW |
3765 | |
3766 | memcg_free_cache_id(kmemcg_id); | |
3767 | } | |
3768 | ||
3769 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3770 | { | |
0b8f73e1 JW |
3771 | /* css_alloc() failed, offlining didn't happen */ |
3772 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
3773 | memcg_offline_kmem(memcg); | |
8e0a8912 | 3774 | } |
d6441637 | 3775 | #else |
0b8f73e1 | 3776 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
3777 | { |
3778 | return 0; | |
3779 | } | |
3780 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
3781 | { | |
3782 | } | |
3783 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3784 | { | |
3785 | } | |
84c07d11 | 3786 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 3787 | |
bbec2e15 RG |
3788 | static int memcg_update_kmem_max(struct mem_cgroup *memcg, |
3789 | unsigned long max) | |
d6441637 | 3790 | { |
b313aeee | 3791 | int ret; |
127424c8 | 3792 | |
bbec2e15 RG |
3793 | mutex_lock(&memcg_max_mutex); |
3794 | ret = page_counter_set_max(&memcg->kmem, max); | |
3795 | mutex_unlock(&memcg_max_mutex); | |
127424c8 | 3796 | return ret; |
d6441637 | 3797 | } |
510fc4e1 | 3798 | |
bbec2e15 | 3799 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
3800 | { |
3801 | int ret; | |
3802 | ||
bbec2e15 | 3803 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 3804 | |
bbec2e15 | 3805 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
3806 | if (ret) |
3807 | goto out; | |
3808 | ||
0db15298 | 3809 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
3810 | /* |
3811 | * The active flag needs to be written after the static_key | |
3812 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
3813 | * function is the last one to run. See mem_cgroup_sk_alloc() |
3814 | * for details, and note that we don't mark any socket as | |
3815 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
3816 | * |
3817 | * We need to do this, because static_keys will span multiple | |
3818 | * sites, but we can't control their order. If we mark a socket | |
3819 | * as accounted, but the accounting functions are not patched in | |
3820 | * yet, we'll lose accounting. | |
3821 | * | |
2d758073 | 3822 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
3823 | * because when this value change, the code to process it is not |
3824 | * patched in yet. | |
3825 | */ | |
3826 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 3827 | memcg->tcpmem_active = true; |
d55f90bf VD |
3828 | } |
3829 | out: | |
bbec2e15 | 3830 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
3831 | return ret; |
3832 | } | |
d55f90bf | 3833 | |
628f4235 KH |
3834 | /* |
3835 | * The user of this function is... | |
3836 | * RES_LIMIT. | |
3837 | */ | |
451af504 TH |
3838 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3839 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3840 | { |
451af504 | 3841 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3842 | unsigned long nr_pages; |
628f4235 KH |
3843 | int ret; |
3844 | ||
451af504 | 3845 | buf = strstrip(buf); |
650c5e56 | 3846 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3847 | if (ret) |
3848 | return ret; | |
af36f906 | 3849 | |
3e32cb2e | 3850 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3851 | case RES_LIMIT: |
4b3bde4c BS |
3852 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3853 | ret = -EINVAL; | |
3854 | break; | |
3855 | } | |
3e32cb2e JW |
3856 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3857 | case _MEM: | |
bbec2e15 | 3858 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 3859 | break; |
3e32cb2e | 3860 | case _MEMSWAP: |
bbec2e15 | 3861 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 3862 | break; |
3e32cb2e | 3863 | case _KMEM: |
0158115f MH |
3864 | pr_warn_once("kmem.limit_in_bytes is deprecated and will be removed. " |
3865 | "Please report your usecase to linux-mm@kvack.org if you " | |
3866 | "depend on this functionality.\n"); | |
bbec2e15 | 3867 | ret = memcg_update_kmem_max(memcg, nr_pages); |
3e32cb2e | 3868 | break; |
d55f90bf | 3869 | case _TCP: |
bbec2e15 | 3870 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 3871 | break; |
3e32cb2e | 3872 | } |
296c81d8 | 3873 | break; |
3e32cb2e JW |
3874 | case RES_SOFT_LIMIT: |
3875 | memcg->soft_limit = nr_pages; | |
3876 | ret = 0; | |
628f4235 KH |
3877 | break; |
3878 | } | |
451af504 | 3879 | return ret ?: nbytes; |
8cdea7c0 BS |
3880 | } |
3881 | ||
6770c64e TH |
3882 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3883 | size_t nbytes, loff_t off) | |
c84872e1 | 3884 | { |
6770c64e | 3885 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3886 | struct page_counter *counter; |
c84872e1 | 3887 | |
3e32cb2e JW |
3888 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3889 | case _MEM: | |
3890 | counter = &memcg->memory; | |
3891 | break; | |
3892 | case _MEMSWAP: | |
3893 | counter = &memcg->memsw; | |
3894 | break; | |
3895 | case _KMEM: | |
3896 | counter = &memcg->kmem; | |
3897 | break; | |
d55f90bf | 3898 | case _TCP: |
0db15298 | 3899 | counter = &memcg->tcpmem; |
d55f90bf | 3900 | break; |
3e32cb2e JW |
3901 | default: |
3902 | BUG(); | |
3903 | } | |
af36f906 | 3904 | |
3e32cb2e | 3905 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3906 | case RES_MAX_USAGE: |
3e32cb2e | 3907 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3908 | break; |
3909 | case RES_FAILCNT: | |
3e32cb2e | 3910 | counter->failcnt = 0; |
29f2a4da | 3911 | break; |
3e32cb2e JW |
3912 | default: |
3913 | BUG(); | |
29f2a4da | 3914 | } |
f64c3f54 | 3915 | |
6770c64e | 3916 | return nbytes; |
c84872e1 PE |
3917 | } |
3918 | ||
182446d0 | 3919 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3920 | struct cftype *cft) |
3921 | { | |
182446d0 | 3922 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3923 | } |
3924 | ||
02491447 | 3925 | #ifdef CONFIG_MMU |
182446d0 | 3926 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3927 | struct cftype *cft, u64 val) |
3928 | { | |
182446d0 | 3929 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3930 | |
1dfab5ab | 3931 | if (val & ~MOVE_MASK) |
7dc74be0 | 3932 | return -EINVAL; |
ee5e8472 | 3933 | |
7dc74be0 | 3934 | /* |
ee5e8472 GC |
3935 | * No kind of locking is needed in here, because ->can_attach() will |
3936 | * check this value once in the beginning of the process, and then carry | |
3937 | * on with stale data. This means that changes to this value will only | |
3938 | * affect task migrations starting after the change. | |
7dc74be0 | 3939 | */ |
c0ff4b85 | 3940 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3941 | return 0; |
3942 | } | |
02491447 | 3943 | #else |
182446d0 | 3944 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3945 | struct cftype *cft, u64 val) |
3946 | { | |
3947 | return -ENOSYS; | |
3948 | } | |
3949 | #endif | |
7dc74be0 | 3950 | |
406eb0c9 | 3951 | #ifdef CONFIG_NUMA |
113b7dfd JW |
3952 | |
3953 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
3954 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
3955 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) | |
3956 | ||
3957 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 | 3958 | int nid, unsigned int lru_mask, bool tree) |
113b7dfd | 3959 | { |
867e5e1d | 3960 | struct lruvec *lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); |
113b7dfd JW |
3961 | unsigned long nr = 0; |
3962 | enum lru_list lru; | |
3963 | ||
3964 | VM_BUG_ON((unsigned)nid >= nr_node_ids); | |
3965 | ||
3966 | for_each_lru(lru) { | |
3967 | if (!(BIT(lru) & lru_mask)) | |
3968 | continue; | |
dd8657b6 SB |
3969 | if (tree) |
3970 | nr += lruvec_page_state(lruvec, NR_LRU_BASE + lru); | |
3971 | else | |
3972 | nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru); | |
113b7dfd JW |
3973 | } |
3974 | return nr; | |
3975 | } | |
3976 | ||
3977 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 SB |
3978 | unsigned int lru_mask, |
3979 | bool tree) | |
113b7dfd JW |
3980 | { |
3981 | unsigned long nr = 0; | |
3982 | enum lru_list lru; | |
3983 | ||
3984 | for_each_lru(lru) { | |
3985 | if (!(BIT(lru) & lru_mask)) | |
3986 | continue; | |
dd8657b6 SB |
3987 | if (tree) |
3988 | nr += memcg_page_state(memcg, NR_LRU_BASE + lru); | |
3989 | else | |
3990 | nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru); | |
113b7dfd JW |
3991 | } |
3992 | return nr; | |
3993 | } | |
3994 | ||
2da8ca82 | 3995 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3996 | { |
25485de6 GT |
3997 | struct numa_stat { |
3998 | const char *name; | |
3999 | unsigned int lru_mask; | |
4000 | }; | |
4001 | ||
4002 | static const struct numa_stat stats[] = { | |
4003 | { "total", LRU_ALL }, | |
4004 | { "file", LRU_ALL_FILE }, | |
4005 | { "anon", LRU_ALL_ANON }, | |
4006 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
4007 | }; | |
4008 | const struct numa_stat *stat; | |
406eb0c9 | 4009 | int nid; |
aa9694bb | 4010 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
406eb0c9 | 4011 | |
25485de6 | 4012 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
4013 | seq_printf(m, "%s=%lu", stat->name, |
4014 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
4015 | false)); | |
4016 | for_each_node_state(nid, N_MEMORY) | |
4017 | seq_printf(m, " N%d=%lu", nid, | |
4018 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
4019 | stat->lru_mask, false)); | |
25485de6 | 4020 | seq_putc(m, '\n'); |
406eb0c9 | 4021 | } |
406eb0c9 | 4022 | |
071aee13 | 4023 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
4024 | |
4025 | seq_printf(m, "hierarchical_%s=%lu", stat->name, | |
4026 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
4027 | true)); | |
4028 | for_each_node_state(nid, N_MEMORY) | |
4029 | seq_printf(m, " N%d=%lu", nid, | |
4030 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
4031 | stat->lru_mask, true)); | |
071aee13 | 4032 | seq_putc(m, '\n'); |
406eb0c9 | 4033 | } |
406eb0c9 | 4034 | |
406eb0c9 YH |
4035 | return 0; |
4036 | } | |
4037 | #endif /* CONFIG_NUMA */ | |
4038 | ||
c8713d0b | 4039 | static const unsigned int memcg1_stats[] = { |
0d1c2072 | 4040 | NR_FILE_PAGES, |
be5d0a74 | 4041 | NR_ANON_MAPPED, |
468c3982 JW |
4042 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4043 | NR_ANON_THPS, | |
4044 | #endif | |
c8713d0b JW |
4045 | NR_SHMEM, |
4046 | NR_FILE_MAPPED, | |
4047 | NR_FILE_DIRTY, | |
4048 | NR_WRITEBACK, | |
4049 | MEMCG_SWAP, | |
4050 | }; | |
4051 | ||
4052 | static const char *const memcg1_stat_names[] = { | |
4053 | "cache", | |
4054 | "rss", | |
468c3982 | 4055 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
c8713d0b | 4056 | "rss_huge", |
468c3982 | 4057 | #endif |
c8713d0b JW |
4058 | "shmem", |
4059 | "mapped_file", | |
4060 | "dirty", | |
4061 | "writeback", | |
4062 | "swap", | |
4063 | }; | |
4064 | ||
df0e53d0 | 4065 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 4066 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
4067 | PGPGIN, |
4068 | PGPGOUT, | |
4069 | PGFAULT, | |
4070 | PGMAJFAULT, | |
4071 | }; | |
4072 | ||
2da8ca82 | 4073 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 4074 | { |
aa9694bb | 4075 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3e32cb2e | 4076 | unsigned long memory, memsw; |
af7c4b0e JW |
4077 | struct mem_cgroup *mi; |
4078 | unsigned int i; | |
406eb0c9 | 4079 | |
71cd3113 | 4080 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c | 4081 | |
71cd3113 | 4082 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
468c3982 JW |
4083 | unsigned long nr; |
4084 | ||
71cd3113 | 4085 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 4086 | continue; |
468c3982 | 4087 | nr = memcg_page_state_local(memcg, memcg1_stats[i]); |
468c3982 | 4088 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], nr * PAGE_SIZE); |
1dd3a273 | 4089 | } |
7b854121 | 4090 | |
df0e53d0 | 4091 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d | 4092 | seq_printf(m, "%s %lu\n", vm_event_name(memcg1_events[i]), |
205b20cc | 4093 | memcg_events_local(memcg, memcg1_events[i])); |
af7c4b0e JW |
4094 | |
4095 | for (i = 0; i < NR_LRU_LISTS; i++) | |
ebc5d83d | 4096 | seq_printf(m, "%s %lu\n", lru_list_name(i), |
205b20cc | 4097 | memcg_page_state_local(memcg, NR_LRU_BASE + i) * |
21d89d15 | 4098 | PAGE_SIZE); |
af7c4b0e | 4099 | |
14067bb3 | 4100 | /* Hierarchical information */ |
3e32cb2e JW |
4101 | memory = memsw = PAGE_COUNTER_MAX; |
4102 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
15b42562 CD |
4103 | memory = min(memory, READ_ONCE(mi->memory.max)); |
4104 | memsw = min(memsw, READ_ONCE(mi->memsw.max)); | |
fee7b548 | 4105 | } |
3e32cb2e JW |
4106 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
4107 | (u64)memory * PAGE_SIZE); | |
7941d214 | 4108 | if (do_memsw_account()) |
3e32cb2e JW |
4109 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
4110 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 4111 | |
8de7ecc6 | 4112 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
7de2e9f1 | 4113 | unsigned long nr; |
4114 | ||
71cd3113 | 4115 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 4116 | continue; |
7de2e9f1 | 4117 | nr = memcg_page_state(memcg, memcg1_stats[i]); |
8de7ecc6 | 4118 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], |
7de2e9f1 | 4119 | (u64)nr * PAGE_SIZE); |
af7c4b0e JW |
4120 | } |
4121 | ||
8de7ecc6 | 4122 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d KK |
4123 | seq_printf(m, "total_%s %llu\n", |
4124 | vm_event_name(memcg1_events[i]), | |
dd923990 | 4125 | (u64)memcg_events(memcg, memcg1_events[i])); |
af7c4b0e | 4126 | |
8de7ecc6 | 4127 | for (i = 0; i < NR_LRU_LISTS; i++) |
ebc5d83d | 4128 | seq_printf(m, "total_%s %llu\n", lru_list_name(i), |
42a30035 JW |
4129 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
4130 | PAGE_SIZE); | |
14067bb3 | 4131 | |
7f016ee8 | 4132 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 4133 | { |
ef8f2327 MG |
4134 | pg_data_t *pgdat; |
4135 | struct mem_cgroup_per_node *mz; | |
1431d4d1 JW |
4136 | unsigned long anon_cost = 0; |
4137 | unsigned long file_cost = 0; | |
7f016ee8 | 4138 | |
ef8f2327 MG |
4139 | for_each_online_pgdat(pgdat) { |
4140 | mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); | |
7f016ee8 | 4141 | |
1431d4d1 JW |
4142 | anon_cost += mz->lruvec.anon_cost; |
4143 | file_cost += mz->lruvec.file_cost; | |
ef8f2327 | 4144 | } |
1431d4d1 JW |
4145 | seq_printf(m, "anon_cost %lu\n", anon_cost); |
4146 | seq_printf(m, "file_cost %lu\n", file_cost); | |
7f016ee8 KM |
4147 | } |
4148 | #endif | |
4149 | ||
d2ceb9b7 KH |
4150 | return 0; |
4151 | } | |
4152 | ||
182446d0 TH |
4153 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
4154 | struct cftype *cft) | |
a7885eb8 | 4155 | { |
182446d0 | 4156 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4157 | |
1f4c025b | 4158 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4159 | } |
4160 | ||
182446d0 TH |
4161 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
4162 | struct cftype *cft, u64 val) | |
a7885eb8 | 4163 | { |
182446d0 | 4164 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4165 | |
3dae7fec | 4166 | if (val > 100) |
a7885eb8 KM |
4167 | return -EINVAL; |
4168 | ||
14208b0e | 4169 | if (css->parent) |
3dae7fec JW |
4170 | memcg->swappiness = val; |
4171 | else | |
4172 | vm_swappiness = val; | |
068b38c1 | 4173 | |
a7885eb8 KM |
4174 | return 0; |
4175 | } | |
4176 | ||
2e72b634 KS |
4177 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4178 | { | |
4179 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 4180 | unsigned long usage; |
2e72b634 KS |
4181 | int i; |
4182 | ||
4183 | rcu_read_lock(); | |
4184 | if (!swap) | |
2c488db2 | 4185 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4186 | else |
2c488db2 | 4187 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4188 | |
4189 | if (!t) | |
4190 | goto unlock; | |
4191 | ||
ce00a967 | 4192 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
4193 | |
4194 | /* | |
748dad36 | 4195 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4196 | * If it's not true, a threshold was crossed after last |
4197 | * call of __mem_cgroup_threshold(). | |
4198 | */ | |
5407a562 | 4199 | i = t->current_threshold; |
2e72b634 KS |
4200 | |
4201 | /* | |
4202 | * Iterate backward over array of thresholds starting from | |
4203 | * current_threshold and check if a threshold is crossed. | |
4204 | * If none of thresholds below usage is crossed, we read | |
4205 | * only one element of the array here. | |
4206 | */ | |
4207 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4208 | eventfd_signal(t->entries[i].eventfd, 1); | |
4209 | ||
4210 | /* i = current_threshold + 1 */ | |
4211 | i++; | |
4212 | ||
4213 | /* | |
4214 | * Iterate forward over array of thresholds starting from | |
4215 | * current_threshold+1 and check if a threshold is crossed. | |
4216 | * If none of thresholds above usage is crossed, we read | |
4217 | * only one element of the array here. | |
4218 | */ | |
4219 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4220 | eventfd_signal(t->entries[i].eventfd, 1); | |
4221 | ||
4222 | /* Update current_threshold */ | |
5407a562 | 4223 | t->current_threshold = i - 1; |
2e72b634 KS |
4224 | unlock: |
4225 | rcu_read_unlock(); | |
4226 | } | |
4227 | ||
4228 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4229 | { | |
ad4ca5f4 KS |
4230 | while (memcg) { |
4231 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 4232 | if (do_memsw_account()) |
ad4ca5f4 KS |
4233 | __mem_cgroup_threshold(memcg, true); |
4234 | ||
4235 | memcg = parent_mem_cgroup(memcg); | |
4236 | } | |
2e72b634 KS |
4237 | } |
4238 | ||
4239 | static int compare_thresholds(const void *a, const void *b) | |
4240 | { | |
4241 | const struct mem_cgroup_threshold *_a = a; | |
4242 | const struct mem_cgroup_threshold *_b = b; | |
4243 | ||
2bff24a3 GT |
4244 | if (_a->threshold > _b->threshold) |
4245 | return 1; | |
4246 | ||
4247 | if (_a->threshold < _b->threshold) | |
4248 | return -1; | |
4249 | ||
4250 | return 0; | |
2e72b634 KS |
4251 | } |
4252 | ||
c0ff4b85 | 4253 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4254 | { |
4255 | struct mem_cgroup_eventfd_list *ev; | |
4256 | ||
2bcf2e92 MH |
4257 | spin_lock(&memcg_oom_lock); |
4258 | ||
c0ff4b85 | 4259 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 4260 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
4261 | |
4262 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
4263 | return 0; |
4264 | } | |
4265 | ||
c0ff4b85 | 4266 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4267 | { |
7d74b06f KH |
4268 | struct mem_cgroup *iter; |
4269 | ||
c0ff4b85 | 4270 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4271 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4272 | } |
4273 | ||
59b6f873 | 4274 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4275 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4276 | { |
2c488db2 KS |
4277 | struct mem_cgroup_thresholds *thresholds; |
4278 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
4279 | unsigned long threshold; |
4280 | unsigned long usage; | |
2c488db2 | 4281 | int i, size, ret; |
2e72b634 | 4282 | |
650c5e56 | 4283 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
4284 | if (ret) |
4285 | return ret; | |
4286 | ||
4287 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4288 | |
05b84301 | 4289 | if (type == _MEM) { |
2c488db2 | 4290 | thresholds = &memcg->thresholds; |
ce00a967 | 4291 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4292 | } else if (type == _MEMSWAP) { |
2c488db2 | 4293 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4294 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4295 | } else |
2e72b634 KS |
4296 | BUG(); |
4297 | ||
2e72b634 | 4298 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4299 | if (thresholds->primary) |
2e72b634 KS |
4300 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4301 | ||
2c488db2 | 4302 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4303 | |
4304 | /* Allocate memory for new array of thresholds */ | |
67b8046f | 4305 | new = kmalloc(struct_size(new, entries, size), GFP_KERNEL); |
2c488db2 | 4306 | if (!new) { |
2e72b634 KS |
4307 | ret = -ENOMEM; |
4308 | goto unlock; | |
4309 | } | |
2c488db2 | 4310 | new->size = size; |
2e72b634 KS |
4311 | |
4312 | /* Copy thresholds (if any) to new array */ | |
e90342e6 GS |
4313 | if (thresholds->primary) |
4314 | memcpy(new->entries, thresholds->primary->entries, | |
4315 | flex_array_size(new, entries, size - 1)); | |
2c488db2 | 4316 | |
2e72b634 | 4317 | /* Add new threshold */ |
2c488db2 KS |
4318 | new->entries[size - 1].eventfd = eventfd; |
4319 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4320 | |
4321 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
61e604e6 | 4322 | sort(new->entries, size, sizeof(*new->entries), |
2e72b634 KS |
4323 | compare_thresholds, NULL); |
4324 | ||
4325 | /* Find current threshold */ | |
2c488db2 | 4326 | new->current_threshold = -1; |
2e72b634 | 4327 | for (i = 0; i < size; i++) { |
748dad36 | 4328 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4329 | /* |
2c488db2 KS |
4330 | * new->current_threshold will not be used until |
4331 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4332 | * it here. |
4333 | */ | |
2c488db2 | 4334 | ++new->current_threshold; |
748dad36 SZ |
4335 | } else |
4336 | break; | |
2e72b634 KS |
4337 | } |
4338 | ||
2c488db2 KS |
4339 | /* Free old spare buffer and save old primary buffer as spare */ |
4340 | kfree(thresholds->spare); | |
4341 | thresholds->spare = thresholds->primary; | |
4342 | ||
4343 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4344 | |
907860ed | 4345 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4346 | synchronize_rcu(); |
4347 | ||
2e72b634 KS |
4348 | unlock: |
4349 | mutex_unlock(&memcg->thresholds_lock); | |
4350 | ||
4351 | return ret; | |
4352 | } | |
4353 | ||
59b6f873 | 4354 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4355 | struct eventfd_ctx *eventfd, const char *args) |
4356 | { | |
59b6f873 | 4357 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4358 | } |
4359 | ||
59b6f873 | 4360 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4361 | struct eventfd_ctx *eventfd, const char *args) |
4362 | { | |
59b6f873 | 4363 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4364 | } |
4365 | ||
59b6f873 | 4366 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4367 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4368 | { |
2c488db2 KS |
4369 | struct mem_cgroup_thresholds *thresholds; |
4370 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 4371 | unsigned long usage; |
7d36665a | 4372 | int i, j, size, entries; |
2e72b634 KS |
4373 | |
4374 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4375 | |
4376 | if (type == _MEM) { | |
2c488db2 | 4377 | thresholds = &memcg->thresholds; |
ce00a967 | 4378 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4379 | } else if (type == _MEMSWAP) { |
2c488db2 | 4380 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4381 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4382 | } else |
2e72b634 KS |
4383 | BUG(); |
4384 | ||
371528ca AV |
4385 | if (!thresholds->primary) |
4386 | goto unlock; | |
4387 | ||
2e72b634 KS |
4388 | /* Check if a threshold crossed before removing */ |
4389 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4390 | ||
4391 | /* Calculate new number of threshold */ | |
7d36665a | 4392 | size = entries = 0; |
2c488db2 KS |
4393 | for (i = 0; i < thresholds->primary->size; i++) { |
4394 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 | 4395 | size++; |
7d36665a CX |
4396 | else |
4397 | entries++; | |
2e72b634 KS |
4398 | } |
4399 | ||
2c488db2 | 4400 | new = thresholds->spare; |
907860ed | 4401 | |
7d36665a CX |
4402 | /* If no items related to eventfd have been cleared, nothing to do */ |
4403 | if (!entries) | |
4404 | goto unlock; | |
4405 | ||
2e72b634 KS |
4406 | /* Set thresholds array to NULL if we don't have thresholds */ |
4407 | if (!size) { | |
2c488db2 KS |
4408 | kfree(new); |
4409 | new = NULL; | |
907860ed | 4410 | goto swap_buffers; |
2e72b634 KS |
4411 | } |
4412 | ||
2c488db2 | 4413 | new->size = size; |
2e72b634 KS |
4414 | |
4415 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4416 | new->current_threshold = -1; |
4417 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4418 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4419 | continue; |
4420 | ||
2c488db2 | 4421 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4422 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4423 | /* |
2c488db2 | 4424 | * new->current_threshold will not be used |
2e72b634 KS |
4425 | * until rcu_assign_pointer(), so it's safe to increment |
4426 | * it here. | |
4427 | */ | |
2c488db2 | 4428 | ++new->current_threshold; |
2e72b634 KS |
4429 | } |
4430 | j++; | |
4431 | } | |
4432 | ||
907860ed | 4433 | swap_buffers: |
2c488db2 KS |
4434 | /* Swap primary and spare array */ |
4435 | thresholds->spare = thresholds->primary; | |
8c757763 | 4436 | |
2c488db2 | 4437 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4438 | |
907860ed | 4439 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4440 | synchronize_rcu(); |
6611d8d7 MC |
4441 | |
4442 | /* If all events are unregistered, free the spare array */ | |
4443 | if (!new) { | |
4444 | kfree(thresholds->spare); | |
4445 | thresholds->spare = NULL; | |
4446 | } | |
371528ca | 4447 | unlock: |
2e72b634 | 4448 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4449 | } |
c1e862c1 | 4450 | |
59b6f873 | 4451 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4452 | struct eventfd_ctx *eventfd) |
4453 | { | |
59b6f873 | 4454 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4455 | } |
4456 | ||
59b6f873 | 4457 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4458 | struct eventfd_ctx *eventfd) |
4459 | { | |
59b6f873 | 4460 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4461 | } |
4462 | ||
59b6f873 | 4463 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4464 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4465 | { |
9490ff27 | 4466 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4467 | |
9490ff27 KH |
4468 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4469 | if (!event) | |
4470 | return -ENOMEM; | |
4471 | ||
1af8efe9 | 4472 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4473 | |
4474 | event->eventfd = eventfd; | |
4475 | list_add(&event->list, &memcg->oom_notify); | |
4476 | ||
4477 | /* already in OOM ? */ | |
c2b42d3c | 4478 | if (memcg->under_oom) |
9490ff27 | 4479 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4480 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4481 | |
4482 | return 0; | |
4483 | } | |
4484 | ||
59b6f873 | 4485 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4486 | struct eventfd_ctx *eventfd) |
9490ff27 | 4487 | { |
9490ff27 | 4488 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4489 | |
1af8efe9 | 4490 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4491 | |
c0ff4b85 | 4492 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4493 | if (ev->eventfd == eventfd) { |
4494 | list_del(&ev->list); | |
4495 | kfree(ev); | |
4496 | } | |
4497 | } | |
4498 | ||
1af8efe9 | 4499 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4500 | } |
4501 | ||
2da8ca82 | 4502 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4503 | { |
aa9694bb | 4504 | struct mem_cgroup *memcg = mem_cgroup_from_seq(sf); |
3c11ecf4 | 4505 | |
791badbd | 4506 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 4507 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
4508 | seq_printf(sf, "oom_kill %lu\n", |
4509 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
4510 | return 0; |
4511 | } | |
4512 | ||
182446d0 | 4513 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4514 | struct cftype *cft, u64 val) |
4515 | { | |
182446d0 | 4516 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4517 | |
4518 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 4519 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4520 | return -EINVAL; |
4521 | ||
c0ff4b85 | 4522 | memcg->oom_kill_disable = val; |
4d845ebf | 4523 | if (!val) |
c0ff4b85 | 4524 | memcg_oom_recover(memcg); |
3dae7fec | 4525 | |
3c11ecf4 KH |
4526 | return 0; |
4527 | } | |
4528 | ||
52ebea74 TH |
4529 | #ifdef CONFIG_CGROUP_WRITEBACK |
4530 | ||
3a8e9ac8 TH |
4531 | #include <trace/events/writeback.h> |
4532 | ||
841710aa TH |
4533 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
4534 | { | |
4535 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
4536 | } | |
4537 | ||
4538 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4539 | { | |
4540 | wb_domain_exit(&memcg->cgwb_domain); | |
4541 | } | |
4542 | ||
2529bb3a TH |
4543 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4544 | { | |
4545 | wb_domain_size_changed(&memcg->cgwb_domain); | |
4546 | } | |
4547 | ||
841710aa TH |
4548 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
4549 | { | |
4550 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4551 | ||
4552 | if (!memcg->css.parent) | |
4553 | return NULL; | |
4554 | ||
4555 | return &memcg->cgwb_domain; | |
4556 | } | |
4557 | ||
0b3d6e6f GT |
4558 | /* |
4559 | * idx can be of type enum memcg_stat_item or node_stat_item. | |
4560 | * Keep in sync with memcg_exact_page(). | |
4561 | */ | |
4562 | static unsigned long memcg_exact_page_state(struct mem_cgroup *memcg, int idx) | |
4563 | { | |
871789d4 | 4564 | long x = atomic_long_read(&memcg->vmstats[idx]); |
0b3d6e6f GT |
4565 | int cpu; |
4566 | ||
4567 | for_each_online_cpu(cpu) | |
871789d4 | 4568 | x += per_cpu_ptr(memcg->vmstats_percpu, cpu)->stat[idx]; |
0b3d6e6f GT |
4569 | if (x < 0) |
4570 | x = 0; | |
4571 | return x; | |
4572 | } | |
4573 | ||
c2aa723a TH |
4574 | /** |
4575 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
4576 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
4577 | * @pfilepages: out parameter for number of file pages |
4578 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
4579 | * @pdirty: out parameter for number of dirty pages |
4580 | * @pwriteback: out parameter for number of pages under writeback | |
4581 | * | |
c5edf9cd TH |
4582 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
4583 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
4584 | * is a bit more involved. | |
c2aa723a | 4585 | * |
c5edf9cd TH |
4586 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
4587 | * headroom is calculated as the lowest headroom of itself and the | |
4588 | * ancestors. Note that this doesn't consider the actual amount of | |
4589 | * available memory in the system. The caller should further cap | |
4590 | * *@pheadroom accordingly. | |
c2aa723a | 4591 | */ |
c5edf9cd TH |
4592 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
4593 | unsigned long *pheadroom, unsigned long *pdirty, | |
4594 | unsigned long *pwriteback) | |
c2aa723a TH |
4595 | { |
4596 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4597 | struct mem_cgroup *parent; | |
c2aa723a | 4598 | |
0b3d6e6f | 4599 | *pdirty = memcg_exact_page_state(memcg, NR_FILE_DIRTY); |
c2aa723a | 4600 | |
0b3d6e6f | 4601 | *pwriteback = memcg_exact_page_state(memcg, NR_WRITEBACK); |
21d89d15 JW |
4602 | *pfilepages = memcg_exact_page_state(memcg, NR_INACTIVE_FILE) + |
4603 | memcg_exact_page_state(memcg, NR_ACTIVE_FILE); | |
c5edf9cd | 4604 | *pheadroom = PAGE_COUNTER_MAX; |
c2aa723a | 4605 | |
c2aa723a | 4606 | while ((parent = parent_mem_cgroup(memcg))) { |
15b42562 | 4607 | unsigned long ceiling = min(READ_ONCE(memcg->memory.max), |
d1663a90 | 4608 | READ_ONCE(memcg->memory.high)); |
c2aa723a TH |
4609 | unsigned long used = page_counter_read(&memcg->memory); |
4610 | ||
c5edf9cd | 4611 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
4612 | memcg = parent; |
4613 | } | |
c2aa723a TH |
4614 | } |
4615 | ||
97b27821 TH |
4616 | /* |
4617 | * Foreign dirty flushing | |
4618 | * | |
4619 | * There's an inherent mismatch between memcg and writeback. The former | |
4620 | * trackes ownership per-page while the latter per-inode. This was a | |
4621 | * deliberate design decision because honoring per-page ownership in the | |
4622 | * writeback path is complicated, may lead to higher CPU and IO overheads | |
4623 | * and deemed unnecessary given that write-sharing an inode across | |
4624 | * different cgroups isn't a common use-case. | |
4625 | * | |
4626 | * Combined with inode majority-writer ownership switching, this works well | |
4627 | * enough in most cases but there are some pathological cases. For | |
4628 | * example, let's say there are two cgroups A and B which keep writing to | |
4629 | * different but confined parts of the same inode. B owns the inode and | |
4630 | * A's memory is limited far below B's. A's dirty ratio can rise enough to | |
4631 | * trigger balance_dirty_pages() sleeps but B's can be low enough to avoid | |
4632 | * triggering background writeback. A will be slowed down without a way to | |
4633 | * make writeback of the dirty pages happen. | |
4634 | * | |
4635 | * Conditions like the above can lead to a cgroup getting repatedly and | |
4636 | * severely throttled after making some progress after each | |
4637 | * dirty_expire_interval while the underyling IO device is almost | |
4638 | * completely idle. | |
4639 | * | |
4640 | * Solving this problem completely requires matching the ownership tracking | |
4641 | * granularities between memcg and writeback in either direction. However, | |
4642 | * the more egregious behaviors can be avoided by simply remembering the | |
4643 | * most recent foreign dirtying events and initiating remote flushes on | |
4644 | * them when local writeback isn't enough to keep the memory clean enough. | |
4645 | * | |
4646 | * The following two functions implement such mechanism. When a foreign | |
4647 | * page - a page whose memcg and writeback ownerships don't match - is | |
4648 | * dirtied, mem_cgroup_track_foreign_dirty() records the inode owning | |
4649 | * bdi_writeback on the page owning memcg. When balance_dirty_pages() | |
4650 | * decides that the memcg needs to sleep due to high dirty ratio, it calls | |
4651 | * mem_cgroup_flush_foreign() which queues writeback on the recorded | |
4652 | * foreign bdi_writebacks which haven't expired. Both the numbers of | |
4653 | * recorded bdi_writebacks and concurrent in-flight foreign writebacks are | |
4654 | * limited to MEMCG_CGWB_FRN_CNT. | |
4655 | * | |
4656 | * The mechanism only remembers IDs and doesn't hold any object references. | |
4657 | * As being wrong occasionally doesn't matter, updates and accesses to the | |
4658 | * records are lockless and racy. | |
4659 | */ | |
4660 | void mem_cgroup_track_foreign_dirty_slowpath(struct page *page, | |
4661 | struct bdi_writeback *wb) | |
4662 | { | |
bcfe06bf | 4663 | struct mem_cgroup *memcg = page_memcg(page); |
97b27821 TH |
4664 | struct memcg_cgwb_frn *frn; |
4665 | u64 now = get_jiffies_64(); | |
4666 | u64 oldest_at = now; | |
4667 | int oldest = -1; | |
4668 | int i; | |
4669 | ||
3a8e9ac8 TH |
4670 | trace_track_foreign_dirty(page, wb); |
4671 | ||
97b27821 TH |
4672 | /* |
4673 | * Pick the slot to use. If there is already a slot for @wb, keep | |
4674 | * using it. If not replace the oldest one which isn't being | |
4675 | * written out. | |
4676 | */ | |
4677 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4678 | frn = &memcg->cgwb_frn[i]; | |
4679 | if (frn->bdi_id == wb->bdi->id && | |
4680 | frn->memcg_id == wb->memcg_css->id) | |
4681 | break; | |
4682 | if (time_before64(frn->at, oldest_at) && | |
4683 | atomic_read(&frn->done.cnt) == 1) { | |
4684 | oldest = i; | |
4685 | oldest_at = frn->at; | |
4686 | } | |
4687 | } | |
4688 | ||
4689 | if (i < MEMCG_CGWB_FRN_CNT) { | |
4690 | /* | |
4691 | * Re-using an existing one. Update timestamp lazily to | |
4692 | * avoid making the cacheline hot. We want them to be | |
4693 | * reasonably up-to-date and significantly shorter than | |
4694 | * dirty_expire_interval as that's what expires the record. | |
4695 | * Use the shorter of 1s and dirty_expire_interval / 8. | |
4696 | */ | |
4697 | unsigned long update_intv = | |
4698 | min_t(unsigned long, HZ, | |
4699 | msecs_to_jiffies(dirty_expire_interval * 10) / 8); | |
4700 | ||
4701 | if (time_before64(frn->at, now - update_intv)) | |
4702 | frn->at = now; | |
4703 | } else if (oldest >= 0) { | |
4704 | /* replace the oldest free one */ | |
4705 | frn = &memcg->cgwb_frn[oldest]; | |
4706 | frn->bdi_id = wb->bdi->id; | |
4707 | frn->memcg_id = wb->memcg_css->id; | |
4708 | frn->at = now; | |
4709 | } | |
4710 | } | |
4711 | ||
4712 | /* issue foreign writeback flushes for recorded foreign dirtying events */ | |
4713 | void mem_cgroup_flush_foreign(struct bdi_writeback *wb) | |
4714 | { | |
4715 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4716 | unsigned long intv = msecs_to_jiffies(dirty_expire_interval * 10); | |
4717 | u64 now = jiffies_64; | |
4718 | int i; | |
4719 | ||
4720 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4721 | struct memcg_cgwb_frn *frn = &memcg->cgwb_frn[i]; | |
4722 | ||
4723 | /* | |
4724 | * If the record is older than dirty_expire_interval, | |
4725 | * writeback on it has already started. No need to kick it | |
4726 | * off again. Also, don't start a new one if there's | |
4727 | * already one in flight. | |
4728 | */ | |
4729 | if (time_after64(frn->at, now - intv) && | |
4730 | atomic_read(&frn->done.cnt) == 1) { | |
4731 | frn->at = 0; | |
3a8e9ac8 | 4732 | trace_flush_foreign(wb, frn->bdi_id, frn->memcg_id); |
97b27821 TH |
4733 | cgroup_writeback_by_id(frn->bdi_id, frn->memcg_id, 0, |
4734 | WB_REASON_FOREIGN_FLUSH, | |
4735 | &frn->done); | |
4736 | } | |
4737 | } | |
4738 | } | |
4739 | ||
841710aa TH |
4740 | #else /* CONFIG_CGROUP_WRITEBACK */ |
4741 | ||
4742 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
4743 | { | |
4744 | return 0; | |
4745 | } | |
4746 | ||
4747 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4748 | { | |
4749 | } | |
4750 | ||
2529bb3a TH |
4751 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4752 | { | |
4753 | } | |
4754 | ||
52ebea74 TH |
4755 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
4756 | ||
3bc942f3 TH |
4757 | /* |
4758 | * DO NOT USE IN NEW FILES. | |
4759 | * | |
4760 | * "cgroup.event_control" implementation. | |
4761 | * | |
4762 | * This is way over-engineered. It tries to support fully configurable | |
4763 | * events for each user. Such level of flexibility is completely | |
4764 | * unnecessary especially in the light of the planned unified hierarchy. | |
4765 | * | |
4766 | * Please deprecate this and replace with something simpler if at all | |
4767 | * possible. | |
4768 | */ | |
4769 | ||
79bd9814 TH |
4770 | /* |
4771 | * Unregister event and free resources. | |
4772 | * | |
4773 | * Gets called from workqueue. | |
4774 | */ | |
3bc942f3 | 4775 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4776 | { |
3bc942f3 TH |
4777 | struct mem_cgroup_event *event = |
4778 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4779 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4780 | |
4781 | remove_wait_queue(event->wqh, &event->wait); | |
4782 | ||
59b6f873 | 4783 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4784 | |
4785 | /* Notify userspace the event is going away. */ | |
4786 | eventfd_signal(event->eventfd, 1); | |
4787 | ||
4788 | eventfd_ctx_put(event->eventfd); | |
4789 | kfree(event); | |
59b6f873 | 4790 | css_put(&memcg->css); |
79bd9814 TH |
4791 | } |
4792 | ||
4793 | /* | |
a9a08845 | 4794 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
4795 | * |
4796 | * Called with wqh->lock held and interrupts disabled. | |
4797 | */ | |
ac6424b9 | 4798 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 4799 | int sync, void *key) |
79bd9814 | 4800 | { |
3bc942f3 TH |
4801 | struct mem_cgroup_event *event = |
4802 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4803 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 4804 | __poll_t flags = key_to_poll(key); |
79bd9814 | 4805 | |
a9a08845 | 4806 | if (flags & EPOLLHUP) { |
79bd9814 TH |
4807 | /* |
4808 | * If the event has been detached at cgroup removal, we | |
4809 | * can simply return knowing the other side will cleanup | |
4810 | * for us. | |
4811 | * | |
4812 | * We can't race against event freeing since the other | |
4813 | * side will require wqh->lock via remove_wait_queue(), | |
4814 | * which we hold. | |
4815 | */ | |
fba94807 | 4816 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4817 | if (!list_empty(&event->list)) { |
4818 | list_del_init(&event->list); | |
4819 | /* | |
4820 | * We are in atomic context, but cgroup_event_remove() | |
4821 | * may sleep, so we have to call it in workqueue. | |
4822 | */ | |
4823 | schedule_work(&event->remove); | |
4824 | } | |
fba94807 | 4825 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4826 | } |
4827 | ||
4828 | return 0; | |
4829 | } | |
4830 | ||
3bc942f3 | 4831 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4832 | wait_queue_head_t *wqh, poll_table *pt) |
4833 | { | |
3bc942f3 TH |
4834 | struct mem_cgroup_event *event = |
4835 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4836 | |
4837 | event->wqh = wqh; | |
4838 | add_wait_queue(wqh, &event->wait); | |
4839 | } | |
4840 | ||
4841 | /* | |
3bc942f3 TH |
4842 | * DO NOT USE IN NEW FILES. |
4843 | * | |
79bd9814 TH |
4844 | * Parse input and register new cgroup event handler. |
4845 | * | |
4846 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4847 | * Interpretation of args is defined by control file implementation. | |
4848 | */ | |
451af504 TH |
4849 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4850 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4851 | { |
451af504 | 4852 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4853 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4854 | struct mem_cgroup_event *event; |
79bd9814 TH |
4855 | struct cgroup_subsys_state *cfile_css; |
4856 | unsigned int efd, cfd; | |
4857 | struct fd efile; | |
4858 | struct fd cfile; | |
fba94807 | 4859 | const char *name; |
79bd9814 TH |
4860 | char *endp; |
4861 | int ret; | |
4862 | ||
451af504 TH |
4863 | buf = strstrip(buf); |
4864 | ||
4865 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4866 | if (*endp != ' ') |
4867 | return -EINVAL; | |
451af504 | 4868 | buf = endp + 1; |
79bd9814 | 4869 | |
451af504 | 4870 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4871 | if ((*endp != ' ') && (*endp != '\0')) |
4872 | return -EINVAL; | |
451af504 | 4873 | buf = endp + 1; |
79bd9814 TH |
4874 | |
4875 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4876 | if (!event) | |
4877 | return -ENOMEM; | |
4878 | ||
59b6f873 | 4879 | event->memcg = memcg; |
79bd9814 | 4880 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4881 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4882 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4883 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4884 | |
4885 | efile = fdget(efd); | |
4886 | if (!efile.file) { | |
4887 | ret = -EBADF; | |
4888 | goto out_kfree; | |
4889 | } | |
4890 | ||
4891 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4892 | if (IS_ERR(event->eventfd)) { | |
4893 | ret = PTR_ERR(event->eventfd); | |
4894 | goto out_put_efile; | |
4895 | } | |
4896 | ||
4897 | cfile = fdget(cfd); | |
4898 | if (!cfile.file) { | |
4899 | ret = -EBADF; | |
4900 | goto out_put_eventfd; | |
4901 | } | |
4902 | ||
4903 | /* the process need read permission on control file */ | |
4904 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
02f92b38 | 4905 | ret = file_permission(cfile.file, MAY_READ); |
79bd9814 TH |
4906 | if (ret < 0) |
4907 | goto out_put_cfile; | |
4908 | ||
fba94807 TH |
4909 | /* |
4910 | * Determine the event callbacks and set them in @event. This used | |
4911 | * to be done via struct cftype but cgroup core no longer knows | |
4912 | * about these events. The following is crude but the whole thing | |
4913 | * is for compatibility anyway. | |
3bc942f3 TH |
4914 | * |
4915 | * DO NOT ADD NEW FILES. | |
fba94807 | 4916 | */ |
b583043e | 4917 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4918 | |
4919 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4920 | event->register_event = mem_cgroup_usage_register_event; | |
4921 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4922 | } else if (!strcmp(name, "memory.oom_control")) { | |
4923 | event->register_event = mem_cgroup_oom_register_event; | |
4924 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4925 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4926 | event->register_event = vmpressure_register_event; | |
4927 | event->unregister_event = vmpressure_unregister_event; | |
4928 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4929 | event->register_event = memsw_cgroup_usage_register_event; |
4930 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4931 | } else { |
4932 | ret = -EINVAL; | |
4933 | goto out_put_cfile; | |
4934 | } | |
4935 | ||
79bd9814 | 4936 | /* |
b5557c4c TH |
4937 | * Verify @cfile should belong to @css. Also, remaining events are |
4938 | * automatically removed on cgroup destruction but the removal is | |
4939 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4940 | */ |
b583043e | 4941 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4942 | &memory_cgrp_subsys); |
79bd9814 | 4943 | ret = -EINVAL; |
5a17f543 | 4944 | if (IS_ERR(cfile_css)) |
79bd9814 | 4945 | goto out_put_cfile; |
5a17f543 TH |
4946 | if (cfile_css != css) { |
4947 | css_put(cfile_css); | |
79bd9814 | 4948 | goto out_put_cfile; |
5a17f543 | 4949 | } |
79bd9814 | 4950 | |
451af504 | 4951 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4952 | if (ret) |
4953 | goto out_put_css; | |
4954 | ||
9965ed17 | 4955 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 4956 | |
fba94807 TH |
4957 | spin_lock(&memcg->event_list_lock); |
4958 | list_add(&event->list, &memcg->event_list); | |
4959 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4960 | |
4961 | fdput(cfile); | |
4962 | fdput(efile); | |
4963 | ||
451af504 | 4964 | return nbytes; |
79bd9814 TH |
4965 | |
4966 | out_put_css: | |
b5557c4c | 4967 | css_put(css); |
79bd9814 TH |
4968 | out_put_cfile: |
4969 | fdput(cfile); | |
4970 | out_put_eventfd: | |
4971 | eventfd_ctx_put(event->eventfd); | |
4972 | out_put_efile: | |
4973 | fdput(efile); | |
4974 | out_kfree: | |
4975 | kfree(event); | |
4976 | ||
4977 | return ret; | |
4978 | } | |
4979 | ||
241994ed | 4980 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4981 | { |
0eea1030 | 4982 | .name = "usage_in_bytes", |
8c7c6e34 | 4983 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4984 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4985 | }, |
c84872e1 PE |
4986 | { |
4987 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4988 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4989 | .write = mem_cgroup_reset, |
791badbd | 4990 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4991 | }, |
8cdea7c0 | 4992 | { |
0eea1030 | 4993 | .name = "limit_in_bytes", |
8c7c6e34 | 4994 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4995 | .write = mem_cgroup_write, |
791badbd | 4996 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4997 | }, |
296c81d8 BS |
4998 | { |
4999 | .name = "soft_limit_in_bytes", | |
5000 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 5001 | .write = mem_cgroup_write, |
791badbd | 5002 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 5003 | }, |
8cdea7c0 BS |
5004 | { |
5005 | .name = "failcnt", | |
8c7c6e34 | 5006 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 5007 | .write = mem_cgroup_reset, |
791badbd | 5008 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5009 | }, |
d2ceb9b7 KH |
5010 | { |
5011 | .name = "stat", | |
2da8ca82 | 5012 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 5013 | }, |
c1e862c1 KH |
5014 | { |
5015 | .name = "force_empty", | |
6770c64e | 5016 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 5017 | }, |
18f59ea7 BS |
5018 | { |
5019 | .name = "use_hierarchy", | |
5020 | .write_u64 = mem_cgroup_hierarchy_write, | |
5021 | .read_u64 = mem_cgroup_hierarchy_read, | |
5022 | }, | |
79bd9814 | 5023 | { |
3bc942f3 | 5024 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 5025 | .write = memcg_write_event_control, |
7dbdb199 | 5026 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 5027 | }, |
a7885eb8 KM |
5028 | { |
5029 | .name = "swappiness", | |
5030 | .read_u64 = mem_cgroup_swappiness_read, | |
5031 | .write_u64 = mem_cgroup_swappiness_write, | |
5032 | }, | |
7dc74be0 DN |
5033 | { |
5034 | .name = "move_charge_at_immigrate", | |
5035 | .read_u64 = mem_cgroup_move_charge_read, | |
5036 | .write_u64 = mem_cgroup_move_charge_write, | |
5037 | }, | |
9490ff27 KH |
5038 | { |
5039 | .name = "oom_control", | |
2da8ca82 | 5040 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 5041 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
5042 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
5043 | }, | |
70ddf637 AV |
5044 | { |
5045 | .name = "pressure_level", | |
70ddf637 | 5046 | }, |
406eb0c9 YH |
5047 | #ifdef CONFIG_NUMA |
5048 | { | |
5049 | .name = "numa_stat", | |
2da8ca82 | 5050 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
5051 | }, |
5052 | #endif | |
510fc4e1 GC |
5053 | { |
5054 | .name = "kmem.limit_in_bytes", | |
5055 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 5056 | .write = mem_cgroup_write, |
791badbd | 5057 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5058 | }, |
5059 | { | |
5060 | .name = "kmem.usage_in_bytes", | |
5061 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 5062 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5063 | }, |
5064 | { | |
5065 | .name = "kmem.failcnt", | |
5066 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 5067 | .write = mem_cgroup_reset, |
791badbd | 5068 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5069 | }, |
5070 | { | |
5071 | .name = "kmem.max_usage_in_bytes", | |
5072 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 5073 | .write = mem_cgroup_reset, |
791badbd | 5074 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 5075 | }, |
a87425a3 YS |
5076 | #if defined(CONFIG_MEMCG_KMEM) && \ |
5077 | (defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)) | |
749c5415 GC |
5078 | { |
5079 | .name = "kmem.slabinfo", | |
b047501c | 5080 | .seq_show = memcg_slab_show, |
749c5415 GC |
5081 | }, |
5082 | #endif | |
d55f90bf VD |
5083 | { |
5084 | .name = "kmem.tcp.limit_in_bytes", | |
5085 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
5086 | .write = mem_cgroup_write, | |
5087 | .read_u64 = mem_cgroup_read_u64, | |
5088 | }, | |
5089 | { | |
5090 | .name = "kmem.tcp.usage_in_bytes", | |
5091 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
5092 | .read_u64 = mem_cgroup_read_u64, | |
5093 | }, | |
5094 | { | |
5095 | .name = "kmem.tcp.failcnt", | |
5096 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
5097 | .write = mem_cgroup_reset, | |
5098 | .read_u64 = mem_cgroup_read_u64, | |
5099 | }, | |
5100 | { | |
5101 | .name = "kmem.tcp.max_usage_in_bytes", | |
5102 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
5103 | .write = mem_cgroup_reset, | |
5104 | .read_u64 = mem_cgroup_read_u64, | |
5105 | }, | |
6bc10349 | 5106 | { }, /* terminate */ |
af36f906 | 5107 | }; |
8c7c6e34 | 5108 | |
73f576c0 JW |
5109 | /* |
5110 | * Private memory cgroup IDR | |
5111 | * | |
5112 | * Swap-out records and page cache shadow entries need to store memcg | |
5113 | * references in constrained space, so we maintain an ID space that is | |
5114 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
5115 | * memory-controlled cgroups to 64k. | |
5116 | * | |
b8f2935f | 5117 | * However, there usually are many references to the offline CSS after |
73f576c0 JW |
5118 | * the cgroup has been destroyed, such as page cache or reclaimable |
5119 | * slab objects, that don't need to hang on to the ID. We want to keep | |
5120 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
5121 | * relatively small ID space and prevent the creation of new cgroups | |
5122 | * even when there are much fewer than 64k cgroups - possibly none. | |
5123 | * | |
5124 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
5125 | * be freed and recycled when it's no longer needed, which is usually | |
5126 | * when the CSS is offlined. | |
5127 | * | |
5128 | * The only exception to that are records of swapped out tmpfs/shmem | |
5129 | * pages that need to be attributed to live ancestors on swapin. But | |
5130 | * those references are manageable from userspace. | |
5131 | */ | |
5132 | ||
5133 | static DEFINE_IDR(mem_cgroup_idr); | |
5134 | ||
7e97de0b KT |
5135 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
5136 | { | |
5137 | if (memcg->id.id > 0) { | |
5138 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
5139 | memcg->id.id = 0; | |
5140 | } | |
5141 | } | |
5142 | ||
c1514c0a VF |
5143 | static void __maybe_unused mem_cgroup_id_get_many(struct mem_cgroup *memcg, |
5144 | unsigned int n) | |
73f576c0 | 5145 | { |
1c2d479a | 5146 | refcount_add(n, &memcg->id.ref); |
73f576c0 JW |
5147 | } |
5148 | ||
615d66c3 | 5149 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 5150 | { |
1c2d479a | 5151 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 5152 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
5153 | |
5154 | /* Memcg ID pins CSS */ | |
5155 | css_put(&memcg->css); | |
5156 | } | |
5157 | } | |
5158 | ||
615d66c3 VD |
5159 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) |
5160 | { | |
5161 | mem_cgroup_id_put_many(memcg, 1); | |
5162 | } | |
5163 | ||
73f576c0 JW |
5164 | /** |
5165 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
5166 | * @id: the memcg id to look up | |
5167 | * | |
5168 | * Caller must hold rcu_read_lock(). | |
5169 | */ | |
5170 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
5171 | { | |
5172 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5173 | return idr_find(&mem_cgroup_idr, id); | |
5174 | } | |
5175 | ||
ef8f2327 | 5176 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
5177 | { |
5178 | struct mem_cgroup_per_node *pn; | |
ef8f2327 | 5179 | int tmp = node; |
1ecaab2b KH |
5180 | /* |
5181 | * This routine is called against possible nodes. | |
5182 | * But it's BUG to call kmalloc() against offline node. | |
5183 | * | |
5184 | * TODO: this routine can waste much memory for nodes which will | |
5185 | * never be onlined. It's better to use memory hotplug callback | |
5186 | * function. | |
5187 | */ | |
41e3355d KH |
5188 | if (!node_state(node, N_NORMAL_MEMORY)) |
5189 | tmp = -1; | |
17295c88 | 5190 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
5191 | if (!pn) |
5192 | return 1; | |
1ecaab2b | 5193 | |
3e38e0aa RG |
5194 | pn->lruvec_stat_local = alloc_percpu_gfp(struct lruvec_stat, |
5195 | GFP_KERNEL_ACCOUNT); | |
815744d7 JW |
5196 | if (!pn->lruvec_stat_local) { |
5197 | kfree(pn); | |
5198 | return 1; | |
5199 | } | |
5200 | ||
f3344adf | 5201 | pn->lruvec_stat_cpu = alloc_percpu_gfp(struct batched_lruvec_stat, |
3e38e0aa | 5202 | GFP_KERNEL_ACCOUNT); |
a983b5eb | 5203 | if (!pn->lruvec_stat_cpu) { |
815744d7 | 5204 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c JW |
5205 | kfree(pn); |
5206 | return 1; | |
5207 | } | |
5208 | ||
ef8f2327 MG |
5209 | lruvec_init(&pn->lruvec); |
5210 | pn->usage_in_excess = 0; | |
5211 | pn->on_tree = false; | |
5212 | pn->memcg = memcg; | |
5213 | ||
54f72fe0 | 5214 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
5215 | return 0; |
5216 | } | |
5217 | ||
ef8f2327 | 5218 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 5219 | { |
00f3ca2c JW |
5220 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
5221 | ||
4eaf431f MH |
5222 | if (!pn) |
5223 | return; | |
5224 | ||
a983b5eb | 5225 | free_percpu(pn->lruvec_stat_cpu); |
815744d7 | 5226 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c | 5227 | kfree(pn); |
1ecaab2b KH |
5228 | } |
5229 | ||
40e952f9 | 5230 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 5231 | { |
c8b2a36f | 5232 | int node; |
59927fb9 | 5233 | |
c8b2a36f | 5234 | for_each_node(node) |
ef8f2327 | 5235 | free_mem_cgroup_per_node_info(memcg, node); |
871789d4 | 5236 | free_percpu(memcg->vmstats_percpu); |
815744d7 | 5237 | free_percpu(memcg->vmstats_local); |
8ff69e2c | 5238 | kfree(memcg); |
59927fb9 | 5239 | } |
3afe36b1 | 5240 | |
40e952f9 TE |
5241 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
5242 | { | |
5243 | memcg_wb_domain_exit(memcg); | |
7961eee3 SB |
5244 | /* |
5245 | * Flush percpu vmstats and vmevents to guarantee the value correctness | |
5246 | * on parent's and all ancestor levels. | |
5247 | */ | |
4a87e2a2 | 5248 | memcg_flush_percpu_vmstats(memcg); |
7961eee3 | 5249 | memcg_flush_percpu_vmevents(memcg); |
40e952f9 TE |
5250 | __mem_cgroup_free(memcg); |
5251 | } | |
5252 | ||
0b8f73e1 | 5253 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 5254 | { |
d142e3e6 | 5255 | struct mem_cgroup *memcg; |
b9726c26 | 5256 | unsigned int size; |
6d12e2d8 | 5257 | int node; |
97b27821 | 5258 | int __maybe_unused i; |
11d67612 | 5259 | long error = -ENOMEM; |
8cdea7c0 | 5260 | |
0b8f73e1 JW |
5261 | size = sizeof(struct mem_cgroup); |
5262 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
5263 | ||
5264 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 5265 | if (!memcg) |
11d67612 | 5266 | return ERR_PTR(error); |
0b8f73e1 | 5267 | |
73f576c0 JW |
5268 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
5269 | 1, MEM_CGROUP_ID_MAX, | |
5270 | GFP_KERNEL); | |
11d67612 YS |
5271 | if (memcg->id.id < 0) { |
5272 | error = memcg->id.id; | |
73f576c0 | 5273 | goto fail; |
11d67612 | 5274 | } |
73f576c0 | 5275 | |
3e38e0aa RG |
5276 | memcg->vmstats_local = alloc_percpu_gfp(struct memcg_vmstats_percpu, |
5277 | GFP_KERNEL_ACCOUNT); | |
815744d7 JW |
5278 | if (!memcg->vmstats_local) |
5279 | goto fail; | |
5280 | ||
3e38e0aa RG |
5281 | memcg->vmstats_percpu = alloc_percpu_gfp(struct memcg_vmstats_percpu, |
5282 | GFP_KERNEL_ACCOUNT); | |
871789d4 | 5283 | if (!memcg->vmstats_percpu) |
0b8f73e1 | 5284 | goto fail; |
78fb7466 | 5285 | |
3ed28fa1 | 5286 | for_each_node(node) |
ef8f2327 | 5287 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 5288 | goto fail; |
f64c3f54 | 5289 | |
0b8f73e1 JW |
5290 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
5291 | goto fail; | |
28dbc4b6 | 5292 | |
f7e1cb6e | 5293 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 | 5294 | INIT_LIST_HEAD(&memcg->oom_notify); |
d142e3e6 GC |
5295 | mutex_init(&memcg->thresholds_lock); |
5296 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 5297 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
5298 | INIT_LIST_HEAD(&memcg->event_list); |
5299 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 5300 | memcg->socket_pressure = jiffies; |
84c07d11 | 5301 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 5302 | memcg->kmemcg_id = -1; |
bf4f0599 | 5303 | INIT_LIST_HEAD(&memcg->objcg_list); |
900a38f0 | 5304 | #endif |
52ebea74 TH |
5305 | #ifdef CONFIG_CGROUP_WRITEBACK |
5306 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
97b27821 TH |
5307 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) |
5308 | memcg->cgwb_frn[i].done = | |
5309 | __WB_COMPLETION_INIT(&memcg_cgwb_frn_waitq); | |
87eaceb3 YS |
5310 | #endif |
5311 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5312 | spin_lock_init(&memcg->deferred_split_queue.split_queue_lock); | |
5313 | INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue); | |
5314 | memcg->deferred_split_queue.split_queue_len = 0; | |
52ebea74 | 5315 | #endif |
73f576c0 | 5316 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
5317 | return memcg; |
5318 | fail: | |
7e97de0b | 5319 | mem_cgroup_id_remove(memcg); |
40e952f9 | 5320 | __mem_cgroup_free(memcg); |
11d67612 | 5321 | return ERR_PTR(error); |
d142e3e6 GC |
5322 | } |
5323 | ||
0b8f73e1 JW |
5324 | static struct cgroup_subsys_state * __ref |
5325 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 5326 | { |
0b8f73e1 | 5327 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
b87d8cef | 5328 | struct mem_cgroup *memcg, *old_memcg; |
0b8f73e1 | 5329 | long error = -ENOMEM; |
d142e3e6 | 5330 | |
b87d8cef | 5331 | old_memcg = set_active_memcg(parent); |
0b8f73e1 | 5332 | memcg = mem_cgroup_alloc(); |
b87d8cef | 5333 | set_active_memcg(old_memcg); |
11d67612 YS |
5334 | if (IS_ERR(memcg)) |
5335 | return ERR_CAST(memcg); | |
d142e3e6 | 5336 | |
d1663a90 | 5337 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
0b8f73e1 | 5338 | memcg->soft_limit = PAGE_COUNTER_MAX; |
4b82ab4f | 5339 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
0b8f73e1 JW |
5340 | if (parent) { |
5341 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
5342 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
bef8620c | 5343 | |
3e32cb2e | 5344 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 5345 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e | 5346 | page_counter_init(&memcg->kmem, &parent->kmem); |
0db15298 | 5347 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 5348 | } else { |
bef8620c RG |
5349 | page_counter_init(&memcg->memory, NULL); |
5350 | page_counter_init(&memcg->swap, NULL); | |
5351 | page_counter_init(&memcg->kmem, NULL); | |
5352 | page_counter_init(&memcg->tcpmem, NULL); | |
d6441637 | 5353 | |
0b8f73e1 JW |
5354 | root_mem_cgroup = memcg; |
5355 | return &memcg->css; | |
5356 | } | |
5357 | ||
bef8620c | 5358 | /* The following stuff does not apply to the root */ |
b313aeee | 5359 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
5360 | if (error) |
5361 | goto fail; | |
127424c8 | 5362 | |
f7e1cb6e | 5363 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5364 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 5365 | |
0b8f73e1 JW |
5366 | return &memcg->css; |
5367 | fail: | |
7e97de0b | 5368 | mem_cgroup_id_remove(memcg); |
0b8f73e1 | 5369 | mem_cgroup_free(memcg); |
11d67612 | 5370 | return ERR_PTR(error); |
0b8f73e1 JW |
5371 | } |
5372 | ||
73f576c0 | 5373 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 5374 | { |
58fa2a55 VD |
5375 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5376 | ||
0a4465d3 KT |
5377 | /* |
5378 | * A memcg must be visible for memcg_expand_shrinker_maps() | |
5379 | * by the time the maps are allocated. So, we allocate maps | |
5380 | * here, when for_each_mem_cgroup() can't skip it. | |
5381 | */ | |
5382 | if (memcg_alloc_shrinker_maps(memcg)) { | |
5383 | mem_cgroup_id_remove(memcg); | |
5384 | return -ENOMEM; | |
5385 | } | |
5386 | ||
73f576c0 | 5387 | /* Online state pins memcg ID, memcg ID pins CSS */ |
1c2d479a | 5388 | refcount_set(&memcg->id.ref, 1); |
73f576c0 | 5389 | css_get(css); |
2f7dd7a4 | 5390 | return 0; |
8cdea7c0 BS |
5391 | } |
5392 | ||
eb95419b | 5393 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 5394 | { |
eb95419b | 5395 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5396 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
5397 | |
5398 | /* | |
5399 | * Unregister events and notify userspace. | |
5400 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
5401 | * directory to avoid race between userspace and kernelspace. | |
5402 | */ | |
fba94807 TH |
5403 | spin_lock(&memcg->event_list_lock); |
5404 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
5405 | list_del_init(&event->list); |
5406 | schedule_work(&event->remove); | |
5407 | } | |
fba94807 | 5408 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 5409 | |
bf8d5d52 | 5410 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5411 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 5412 | |
567e9ab2 | 5413 | memcg_offline_kmem(memcg); |
52ebea74 | 5414 | wb_memcg_offline(memcg); |
73f576c0 | 5415 | |
591edfb1 RG |
5416 | drain_all_stock(memcg); |
5417 | ||
73f576c0 | 5418 | mem_cgroup_id_put(memcg); |
df878fb0 KH |
5419 | } |
5420 | ||
6df38689 VD |
5421 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
5422 | { | |
5423 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5424 | ||
5425 | invalidate_reclaim_iterators(memcg); | |
5426 | } | |
5427 | ||
eb95419b | 5428 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 5429 | { |
eb95419b | 5430 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
97b27821 | 5431 | int __maybe_unused i; |
c268e994 | 5432 | |
97b27821 TH |
5433 | #ifdef CONFIG_CGROUP_WRITEBACK |
5434 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) | |
5435 | wb_wait_for_completion(&memcg->cgwb_frn[i].done); | |
5436 | #endif | |
f7e1cb6e | 5437 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5438 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 5439 | |
0db15298 | 5440 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 5441 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 5442 | |
0b8f73e1 JW |
5443 | vmpressure_cleanup(&memcg->vmpressure); |
5444 | cancel_work_sync(&memcg->high_work); | |
5445 | mem_cgroup_remove_from_trees(memcg); | |
0a4465d3 | 5446 | memcg_free_shrinker_maps(memcg); |
d886f4e4 | 5447 | memcg_free_kmem(memcg); |
0b8f73e1 | 5448 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
5449 | } |
5450 | ||
1ced953b TH |
5451 | /** |
5452 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
5453 | * @css: the target css | |
5454 | * | |
5455 | * Reset the states of the mem_cgroup associated with @css. This is | |
5456 | * invoked when the userland requests disabling on the default hierarchy | |
5457 | * but the memcg is pinned through dependency. The memcg should stop | |
5458 | * applying policies and should revert to the vanilla state as it may be | |
5459 | * made visible again. | |
5460 | * | |
5461 | * The current implementation only resets the essential configurations. | |
5462 | * This needs to be expanded to cover all the visible parts. | |
5463 | */ | |
5464 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
5465 | { | |
5466 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5467 | ||
bbec2e15 RG |
5468 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
5469 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
bbec2e15 RG |
5470 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); |
5471 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 5472 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5473 | page_counter_set_low(&memcg->memory, 0); |
d1663a90 | 5474 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
24d404dc | 5475 | memcg->soft_limit = PAGE_COUNTER_MAX; |
4b82ab4f | 5476 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
2529bb3a | 5477 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
5478 | } |
5479 | ||
02491447 | 5480 | #ifdef CONFIG_MMU |
7dc74be0 | 5481 | /* Handlers for move charge at task migration. */ |
854ffa8d | 5482 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 5483 | { |
05b84301 | 5484 | int ret; |
9476db97 | 5485 | |
d0164adc MG |
5486 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
5487 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 5488 | if (!ret) { |
854ffa8d | 5489 | mc.precharge += count; |
854ffa8d DN |
5490 | return ret; |
5491 | } | |
9476db97 | 5492 | |
3674534b | 5493 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 5494 | while (count--) { |
3674534b | 5495 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 5496 | if (ret) |
38c5d72f | 5497 | return ret; |
854ffa8d | 5498 | mc.precharge++; |
9476db97 | 5499 | cond_resched(); |
854ffa8d | 5500 | } |
9476db97 | 5501 | return 0; |
4ffef5fe DN |
5502 | } |
5503 | ||
4ffef5fe DN |
5504 | union mc_target { |
5505 | struct page *page; | |
02491447 | 5506 | swp_entry_t ent; |
4ffef5fe DN |
5507 | }; |
5508 | ||
4ffef5fe | 5509 | enum mc_target_type { |
8d32ff84 | 5510 | MC_TARGET_NONE = 0, |
4ffef5fe | 5511 | MC_TARGET_PAGE, |
02491447 | 5512 | MC_TARGET_SWAP, |
c733a828 | 5513 | MC_TARGET_DEVICE, |
4ffef5fe DN |
5514 | }; |
5515 | ||
90254a65 DN |
5516 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5517 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5518 | { |
25b2995a | 5519 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5520 | |
90254a65 DN |
5521 | if (!page || !page_mapped(page)) |
5522 | return NULL; | |
5523 | if (PageAnon(page)) { | |
1dfab5ab | 5524 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5525 | return NULL; |
1dfab5ab JW |
5526 | } else { |
5527 | if (!(mc.flags & MOVE_FILE)) | |
5528 | return NULL; | |
5529 | } | |
90254a65 DN |
5530 | if (!get_page_unless_zero(page)) |
5531 | return NULL; | |
5532 | ||
5533 | return page; | |
5534 | } | |
5535 | ||
c733a828 | 5536 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 5537 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 5538 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 5539 | { |
90254a65 DN |
5540 | struct page *page = NULL; |
5541 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5542 | ||
9a137153 | 5543 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5544 | return NULL; |
c733a828 JG |
5545 | |
5546 | /* | |
5547 | * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to | |
5548 | * a device and because they are not accessible by CPU they are store | |
5549 | * as special swap entry in the CPU page table. | |
5550 | */ | |
5551 | if (is_device_private_entry(ent)) { | |
5552 | page = device_private_entry_to_page(ent); | |
5553 | /* | |
5554 | * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have | |
5555 | * a refcount of 1 when free (unlike normal page) | |
5556 | */ | |
5557 | if (!page_ref_add_unless(page, 1, 1)) | |
5558 | return NULL; | |
5559 | return page; | |
5560 | } | |
5561 | ||
9a137153 RC |
5562 | if (non_swap_entry(ent)) |
5563 | return NULL; | |
5564 | ||
4b91355e KH |
5565 | /* |
5566 | * Because lookup_swap_cache() updates some statistics counter, | |
5567 | * we call find_get_page() with swapper_space directly. | |
5568 | */ | |
f6ab1f7f | 5569 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
2d1c4980 | 5570 | entry->val = ent.val; |
90254a65 DN |
5571 | |
5572 | return page; | |
5573 | } | |
4b91355e KH |
5574 | #else |
5575 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 5576 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
5577 | { |
5578 | return NULL; | |
5579 | } | |
5580 | #endif | |
90254a65 | 5581 | |
87946a72 DN |
5582 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5583 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5584 | { | |
87946a72 DN |
5585 | if (!vma->vm_file) /* anonymous vma */ |
5586 | return NULL; | |
1dfab5ab | 5587 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
5588 | return NULL; |
5589 | ||
87946a72 | 5590 | /* page is moved even if it's not RSS of this task(page-faulted). */ |
aa3b1895 | 5591 | /* shmem/tmpfs may report page out on swap: account for that too. */ |
f5df8635 MWO |
5592 | return find_get_incore_page(vma->vm_file->f_mapping, |
5593 | linear_page_index(vma, addr)); | |
87946a72 DN |
5594 | } |
5595 | ||
b1b0deab CG |
5596 | /** |
5597 | * mem_cgroup_move_account - move account of the page | |
5598 | * @page: the page | |
25843c2b | 5599 | * @compound: charge the page as compound or small page |
b1b0deab CG |
5600 | * @from: mem_cgroup which the page is moved from. |
5601 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
5602 | * | |
3ac808fd | 5603 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
5604 | * |
5605 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
5606 | * from old cgroup. | |
5607 | */ | |
5608 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 5609 | bool compound, |
b1b0deab CG |
5610 | struct mem_cgroup *from, |
5611 | struct mem_cgroup *to) | |
5612 | { | |
ae8af438 KK |
5613 | struct lruvec *from_vec, *to_vec; |
5614 | struct pglist_data *pgdat; | |
6c357848 | 5615 | unsigned int nr_pages = compound ? thp_nr_pages(page) : 1; |
b1b0deab CG |
5616 | int ret; |
5617 | ||
5618 | VM_BUG_ON(from == to); | |
5619 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 5620 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
5621 | |
5622 | /* | |
6a93ca8f | 5623 | * Prevent mem_cgroup_migrate() from looking at |
bcfe06bf | 5624 | * page's memory cgroup of its source page while we change it. |
b1b0deab | 5625 | */ |
f627c2f5 | 5626 | ret = -EBUSY; |
b1b0deab CG |
5627 | if (!trylock_page(page)) |
5628 | goto out; | |
5629 | ||
5630 | ret = -EINVAL; | |
bcfe06bf | 5631 | if (page_memcg(page) != from) |
b1b0deab CG |
5632 | goto out_unlock; |
5633 | ||
ae8af438 | 5634 | pgdat = page_pgdat(page); |
867e5e1d JW |
5635 | from_vec = mem_cgroup_lruvec(from, pgdat); |
5636 | to_vec = mem_cgroup_lruvec(to, pgdat); | |
ae8af438 | 5637 | |
abb242f5 | 5638 | lock_page_memcg(page); |
b1b0deab | 5639 | |
be5d0a74 JW |
5640 | if (PageAnon(page)) { |
5641 | if (page_mapped(page)) { | |
5642 | __mod_lruvec_state(from_vec, NR_ANON_MAPPED, -nr_pages); | |
5643 | __mod_lruvec_state(to_vec, NR_ANON_MAPPED, nr_pages); | |
468c3982 | 5644 | if (PageTransHuge(page)) { |
69473e5d MS |
5645 | __mod_lruvec_state(from_vec, NR_ANON_THPS, |
5646 | -nr_pages); | |
5647 | __mod_lruvec_state(to_vec, NR_ANON_THPS, | |
5648 | nr_pages); | |
468c3982 | 5649 | } |
be5d0a74 JW |
5650 | } |
5651 | } else { | |
0d1c2072 JW |
5652 | __mod_lruvec_state(from_vec, NR_FILE_PAGES, -nr_pages); |
5653 | __mod_lruvec_state(to_vec, NR_FILE_PAGES, nr_pages); | |
5654 | ||
5655 | if (PageSwapBacked(page)) { | |
5656 | __mod_lruvec_state(from_vec, NR_SHMEM, -nr_pages); | |
5657 | __mod_lruvec_state(to_vec, NR_SHMEM, nr_pages); | |
5658 | } | |
5659 | ||
49e50d27 JW |
5660 | if (page_mapped(page)) { |
5661 | __mod_lruvec_state(from_vec, NR_FILE_MAPPED, -nr_pages); | |
5662 | __mod_lruvec_state(to_vec, NR_FILE_MAPPED, nr_pages); | |
5663 | } | |
b1b0deab | 5664 | |
49e50d27 JW |
5665 | if (PageDirty(page)) { |
5666 | struct address_space *mapping = page_mapping(page); | |
c4843a75 | 5667 | |
f56753ac | 5668 | if (mapping_can_writeback(mapping)) { |
49e50d27 JW |
5669 | __mod_lruvec_state(from_vec, NR_FILE_DIRTY, |
5670 | -nr_pages); | |
5671 | __mod_lruvec_state(to_vec, NR_FILE_DIRTY, | |
5672 | nr_pages); | |
5673 | } | |
c4843a75 GT |
5674 | } |
5675 | } | |
5676 | ||
b1b0deab | 5677 | if (PageWriteback(page)) { |
ae8af438 KK |
5678 | __mod_lruvec_state(from_vec, NR_WRITEBACK, -nr_pages); |
5679 | __mod_lruvec_state(to_vec, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
5680 | } |
5681 | ||
5682 | /* | |
abb242f5 JW |
5683 | * All state has been migrated, let's switch to the new memcg. |
5684 | * | |
bcfe06bf | 5685 | * It is safe to change page's memcg here because the page |
abb242f5 JW |
5686 | * is referenced, charged, isolated, and locked: we can't race |
5687 | * with (un)charging, migration, LRU putback, or anything else | |
bcfe06bf | 5688 | * that would rely on a stable page's memory cgroup. |
abb242f5 JW |
5689 | * |
5690 | * Note that lock_page_memcg is a memcg lock, not a page lock, | |
bcfe06bf | 5691 | * to save space. As soon as we switch page's memory cgroup to a |
abb242f5 JW |
5692 | * new memcg that isn't locked, the above state can change |
5693 | * concurrently again. Make sure we're truly done with it. | |
b1b0deab | 5694 | */ |
abb242f5 | 5695 | smp_mb(); |
b1b0deab | 5696 | |
1a3e1f40 JW |
5697 | css_get(&to->css); |
5698 | css_put(&from->css); | |
5699 | ||
bcfe06bf | 5700 | page->memcg_data = (unsigned long)to; |
87eaceb3 | 5701 | |
abb242f5 | 5702 | __unlock_page_memcg(from); |
b1b0deab CG |
5703 | |
5704 | ret = 0; | |
5705 | ||
5706 | local_irq_disable(); | |
3fba69a5 | 5707 | mem_cgroup_charge_statistics(to, page, nr_pages); |
b1b0deab | 5708 | memcg_check_events(to, page); |
3fba69a5 | 5709 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
b1b0deab CG |
5710 | memcg_check_events(from, page); |
5711 | local_irq_enable(); | |
5712 | out_unlock: | |
5713 | unlock_page(page); | |
5714 | out: | |
5715 | return ret; | |
5716 | } | |
5717 | ||
7cf7806c LR |
5718 | /** |
5719 | * get_mctgt_type - get target type of moving charge | |
5720 | * @vma: the vma the pte to be checked belongs | |
5721 | * @addr: the address corresponding to the pte to be checked | |
5722 | * @ptent: the pte to be checked | |
5723 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
5724 | * | |
5725 | * Returns | |
5726 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5727 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5728 | * move charge. if @target is not NULL, the page is stored in target->page | |
5729 | * with extra refcnt got(Callers should handle it). | |
5730 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
5731 | * target for charge migration. if @target is not NULL, the entry is stored | |
5732 | * in target->ent. | |
25b2995a CH |
5733 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PRIVATE |
5734 | * (so ZONE_DEVICE page and thus not on the lru). | |
df6ad698 JG |
5735 | * For now we such page is charge like a regular page would be as for all |
5736 | * intent and purposes it is just special memory taking the place of a | |
5737 | * regular page. | |
c733a828 JG |
5738 | * |
5739 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
5740 | * |
5741 | * Called with pte lock held. | |
5742 | */ | |
5743 | ||
8d32ff84 | 5744 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5745 | unsigned long addr, pte_t ptent, union mc_target *target) |
5746 | { | |
5747 | struct page *page = NULL; | |
8d32ff84 | 5748 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5749 | swp_entry_t ent = { .val = 0 }; |
5750 | ||
5751 | if (pte_present(ptent)) | |
5752 | page = mc_handle_present_pte(vma, addr, ptent); | |
5753 | else if (is_swap_pte(ptent)) | |
48406ef8 | 5754 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 5755 | else if (pte_none(ptent)) |
87946a72 | 5756 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
5757 | |
5758 | if (!page && !ent.val) | |
8d32ff84 | 5759 | return ret; |
02491447 | 5760 | if (page) { |
02491447 | 5761 | /* |
0a31bc97 | 5762 | * Do only loose check w/o serialization. |
1306a85a | 5763 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 5764 | * not under LRU exclusion. |
02491447 | 5765 | */ |
bcfe06bf | 5766 | if (page_memcg(page) == mc.from) { |
02491447 | 5767 | ret = MC_TARGET_PAGE; |
25b2995a | 5768 | if (is_device_private_page(page)) |
c733a828 | 5769 | ret = MC_TARGET_DEVICE; |
02491447 DN |
5770 | if (target) |
5771 | target->page = page; | |
5772 | } | |
5773 | if (!ret || !target) | |
5774 | put_page(page); | |
5775 | } | |
3e14a57b HY |
5776 | /* |
5777 | * There is a swap entry and a page doesn't exist or isn't charged. | |
5778 | * But we cannot move a tail-page in a THP. | |
5779 | */ | |
5780 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 5781 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5782 | ret = MC_TARGET_SWAP; |
5783 | if (target) | |
5784 | target->ent = ent; | |
4ffef5fe | 5785 | } |
4ffef5fe DN |
5786 | return ret; |
5787 | } | |
5788 | ||
12724850 NH |
5789 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5790 | /* | |
d6810d73 HY |
5791 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
5792 | * not support them for now. | |
12724850 NH |
5793 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
5794 | */ | |
5795 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5796 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5797 | { | |
5798 | struct page *page = NULL; | |
12724850 NH |
5799 | enum mc_target_type ret = MC_TARGET_NONE; |
5800 | ||
84c3fc4e ZY |
5801 | if (unlikely(is_swap_pmd(pmd))) { |
5802 | VM_BUG_ON(thp_migration_supported() && | |
5803 | !is_pmd_migration_entry(pmd)); | |
5804 | return ret; | |
5805 | } | |
12724850 | 5806 | page = pmd_page(pmd); |
309381fe | 5807 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 5808 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 5809 | return ret; |
bcfe06bf | 5810 | if (page_memcg(page) == mc.from) { |
12724850 NH |
5811 | ret = MC_TARGET_PAGE; |
5812 | if (target) { | |
5813 | get_page(page); | |
5814 | target->page = page; | |
5815 | } | |
5816 | } | |
5817 | return ret; | |
5818 | } | |
5819 | #else | |
5820 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5821 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5822 | { | |
5823 | return MC_TARGET_NONE; | |
5824 | } | |
5825 | #endif | |
5826 | ||
4ffef5fe DN |
5827 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5828 | unsigned long addr, unsigned long end, | |
5829 | struct mm_walk *walk) | |
5830 | { | |
26bcd64a | 5831 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5832 | pte_t *pte; |
5833 | spinlock_t *ptl; | |
5834 | ||
b6ec57f4 KS |
5835 | ptl = pmd_trans_huge_lock(pmd, vma); |
5836 | if (ptl) { | |
c733a828 JG |
5837 | /* |
5838 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
25b2995a CH |
5839 | * support transparent huge page with MEMORY_DEVICE_PRIVATE but |
5840 | * this might change. | |
c733a828 | 5841 | */ |
12724850 NH |
5842 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5843 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5844 | spin_unlock(ptl); |
1a5a9906 | 5845 | return 0; |
12724850 | 5846 | } |
03319327 | 5847 | |
45f83cef AA |
5848 | if (pmd_trans_unstable(pmd)) |
5849 | return 0; | |
4ffef5fe DN |
5850 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5851 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5852 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5853 | mc.precharge++; /* increment precharge temporarily */ |
5854 | pte_unmap_unlock(pte - 1, ptl); | |
5855 | cond_resched(); | |
5856 | ||
7dc74be0 DN |
5857 | return 0; |
5858 | } | |
5859 | ||
7b86ac33 CH |
5860 | static const struct mm_walk_ops precharge_walk_ops = { |
5861 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5862 | }; | |
5863 | ||
4ffef5fe DN |
5864 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5865 | { | |
5866 | unsigned long precharge; | |
4ffef5fe | 5867 | |
d8ed45c5 | 5868 | mmap_read_lock(mm); |
7b86ac33 | 5869 | walk_page_range(mm, 0, mm->highest_vm_end, &precharge_walk_ops, NULL); |
d8ed45c5 | 5870 | mmap_read_unlock(mm); |
4ffef5fe DN |
5871 | |
5872 | precharge = mc.precharge; | |
5873 | mc.precharge = 0; | |
5874 | ||
5875 | return precharge; | |
5876 | } | |
5877 | ||
4ffef5fe DN |
5878 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5879 | { | |
dfe076b0 DN |
5880 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5881 | ||
5882 | VM_BUG_ON(mc.moving_task); | |
5883 | mc.moving_task = current; | |
5884 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5885 | } |
5886 | ||
dfe076b0 DN |
5887 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5888 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5889 | { |
2bd9bb20 KH |
5890 | struct mem_cgroup *from = mc.from; |
5891 | struct mem_cgroup *to = mc.to; | |
5892 | ||
4ffef5fe | 5893 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5894 | if (mc.precharge) { |
00501b53 | 5895 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5896 | mc.precharge = 0; |
5897 | } | |
5898 | /* | |
5899 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5900 | * we must uncharge here. | |
5901 | */ | |
5902 | if (mc.moved_charge) { | |
00501b53 | 5903 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5904 | mc.moved_charge = 0; |
4ffef5fe | 5905 | } |
483c30b5 DN |
5906 | /* we must fixup refcnts and charges */ |
5907 | if (mc.moved_swap) { | |
483c30b5 | 5908 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5909 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5910 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5911 | |
615d66c3 VD |
5912 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
5913 | ||
05b84301 | 5914 | /* |
3e32cb2e JW |
5915 | * we charged both to->memory and to->memsw, so we |
5916 | * should uncharge to->memory. | |
05b84301 | 5917 | */ |
ce00a967 | 5918 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5919 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5920 | ||
483c30b5 DN |
5921 | mc.moved_swap = 0; |
5922 | } | |
dfe076b0 DN |
5923 | memcg_oom_recover(from); |
5924 | memcg_oom_recover(to); | |
5925 | wake_up_all(&mc.waitq); | |
5926 | } | |
5927 | ||
5928 | static void mem_cgroup_clear_mc(void) | |
5929 | { | |
264a0ae1 TH |
5930 | struct mm_struct *mm = mc.mm; |
5931 | ||
dfe076b0 DN |
5932 | /* |
5933 | * we must clear moving_task before waking up waiters at the end of | |
5934 | * task migration. | |
5935 | */ | |
5936 | mc.moving_task = NULL; | |
5937 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5938 | spin_lock(&mc.lock); |
4ffef5fe DN |
5939 | mc.from = NULL; |
5940 | mc.to = NULL; | |
264a0ae1 | 5941 | mc.mm = NULL; |
2bd9bb20 | 5942 | spin_unlock(&mc.lock); |
264a0ae1 TH |
5943 | |
5944 | mmput(mm); | |
4ffef5fe DN |
5945 | } |
5946 | ||
1f7dd3e5 | 5947 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5948 | { |
1f7dd3e5 | 5949 | struct cgroup_subsys_state *css; |
eed67d75 | 5950 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 5951 | struct mem_cgroup *from; |
4530eddb | 5952 | struct task_struct *leader, *p; |
9f2115f9 | 5953 | struct mm_struct *mm; |
1dfab5ab | 5954 | unsigned long move_flags; |
9f2115f9 | 5955 | int ret = 0; |
7dc74be0 | 5956 | |
1f7dd3e5 TH |
5957 | /* charge immigration isn't supported on the default hierarchy */ |
5958 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
5959 | return 0; |
5960 | ||
4530eddb TH |
5961 | /* |
5962 | * Multi-process migrations only happen on the default hierarchy | |
5963 | * where charge immigration is not used. Perform charge | |
5964 | * immigration if @tset contains a leader and whine if there are | |
5965 | * multiple. | |
5966 | */ | |
5967 | p = NULL; | |
1f7dd3e5 | 5968 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
5969 | WARN_ON_ONCE(p); |
5970 | p = leader; | |
1f7dd3e5 | 5971 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
5972 | } |
5973 | if (!p) | |
5974 | return 0; | |
5975 | ||
1f7dd3e5 TH |
5976 | /* |
5977 | * We are now commited to this value whatever it is. Changes in this | |
5978 | * tunable will only affect upcoming migrations, not the current one. | |
5979 | * So we need to save it, and keep it going. | |
5980 | */ | |
5981 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
5982 | if (!move_flags) | |
5983 | return 0; | |
5984 | ||
9f2115f9 TH |
5985 | from = mem_cgroup_from_task(p); |
5986 | ||
5987 | VM_BUG_ON(from == memcg); | |
5988 | ||
5989 | mm = get_task_mm(p); | |
5990 | if (!mm) | |
5991 | return 0; | |
5992 | /* We move charges only when we move a owner of the mm */ | |
5993 | if (mm->owner == p) { | |
5994 | VM_BUG_ON(mc.from); | |
5995 | VM_BUG_ON(mc.to); | |
5996 | VM_BUG_ON(mc.precharge); | |
5997 | VM_BUG_ON(mc.moved_charge); | |
5998 | VM_BUG_ON(mc.moved_swap); | |
5999 | ||
6000 | spin_lock(&mc.lock); | |
264a0ae1 | 6001 | mc.mm = mm; |
9f2115f9 TH |
6002 | mc.from = from; |
6003 | mc.to = memcg; | |
6004 | mc.flags = move_flags; | |
6005 | spin_unlock(&mc.lock); | |
6006 | /* We set mc.moving_task later */ | |
6007 | ||
6008 | ret = mem_cgroup_precharge_mc(mm); | |
6009 | if (ret) | |
6010 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
6011 | } else { |
6012 | mmput(mm); | |
7dc74be0 DN |
6013 | } |
6014 | return ret; | |
6015 | } | |
6016 | ||
1f7dd3e5 | 6017 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 6018 | { |
4e2f245d JW |
6019 | if (mc.to) |
6020 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
6021 | } |
6022 | ||
4ffef5fe DN |
6023 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
6024 | unsigned long addr, unsigned long end, | |
6025 | struct mm_walk *walk) | |
7dc74be0 | 6026 | { |
4ffef5fe | 6027 | int ret = 0; |
26bcd64a | 6028 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
6029 | pte_t *pte; |
6030 | spinlock_t *ptl; | |
12724850 NH |
6031 | enum mc_target_type target_type; |
6032 | union mc_target target; | |
6033 | struct page *page; | |
4ffef5fe | 6034 | |
b6ec57f4 KS |
6035 | ptl = pmd_trans_huge_lock(pmd, vma); |
6036 | if (ptl) { | |
62ade86a | 6037 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 6038 | spin_unlock(ptl); |
12724850 NH |
6039 | return 0; |
6040 | } | |
6041 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
6042 | if (target_type == MC_TARGET_PAGE) { | |
6043 | page = target.page; | |
6044 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 6045 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 6046 | mc.from, mc.to)) { |
12724850 NH |
6047 | mc.precharge -= HPAGE_PMD_NR; |
6048 | mc.moved_charge += HPAGE_PMD_NR; | |
6049 | } | |
6050 | putback_lru_page(page); | |
6051 | } | |
6052 | put_page(page); | |
c733a828 JG |
6053 | } else if (target_type == MC_TARGET_DEVICE) { |
6054 | page = target.page; | |
6055 | if (!mem_cgroup_move_account(page, true, | |
6056 | mc.from, mc.to)) { | |
6057 | mc.precharge -= HPAGE_PMD_NR; | |
6058 | mc.moved_charge += HPAGE_PMD_NR; | |
6059 | } | |
6060 | put_page(page); | |
12724850 | 6061 | } |
bf929152 | 6062 | spin_unlock(ptl); |
1a5a9906 | 6063 | return 0; |
12724850 NH |
6064 | } |
6065 | ||
45f83cef AA |
6066 | if (pmd_trans_unstable(pmd)) |
6067 | return 0; | |
4ffef5fe DN |
6068 | retry: |
6069 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
6070 | for (; addr != end; addr += PAGE_SIZE) { | |
6071 | pte_t ptent = *(pte++); | |
c733a828 | 6072 | bool device = false; |
02491447 | 6073 | swp_entry_t ent; |
4ffef5fe DN |
6074 | |
6075 | if (!mc.precharge) | |
6076 | break; | |
6077 | ||
8d32ff84 | 6078 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
6079 | case MC_TARGET_DEVICE: |
6080 | device = true; | |
e4a9bc58 | 6081 | fallthrough; |
4ffef5fe DN |
6082 | case MC_TARGET_PAGE: |
6083 | page = target.page; | |
53f9263b KS |
6084 | /* |
6085 | * We can have a part of the split pmd here. Moving it | |
6086 | * can be done but it would be too convoluted so simply | |
6087 | * ignore such a partial THP and keep it in original | |
6088 | * memcg. There should be somebody mapping the head. | |
6089 | */ | |
6090 | if (PageTransCompound(page)) | |
6091 | goto put; | |
c733a828 | 6092 | if (!device && isolate_lru_page(page)) |
4ffef5fe | 6093 | goto put; |
f627c2f5 KS |
6094 | if (!mem_cgroup_move_account(page, false, |
6095 | mc.from, mc.to)) { | |
4ffef5fe | 6096 | mc.precharge--; |
854ffa8d DN |
6097 | /* we uncharge from mc.from later. */ |
6098 | mc.moved_charge++; | |
4ffef5fe | 6099 | } |
c733a828 JG |
6100 | if (!device) |
6101 | putback_lru_page(page); | |
8d32ff84 | 6102 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
6103 | put_page(page); |
6104 | break; | |
02491447 DN |
6105 | case MC_TARGET_SWAP: |
6106 | ent = target.ent; | |
e91cbb42 | 6107 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6108 | mc.precharge--; |
8d22a935 HD |
6109 | mem_cgroup_id_get_many(mc.to, 1); |
6110 | /* we fixup other refcnts and charges later. */ | |
483c30b5 DN |
6111 | mc.moved_swap++; |
6112 | } | |
02491447 | 6113 | break; |
4ffef5fe DN |
6114 | default: |
6115 | break; | |
6116 | } | |
6117 | } | |
6118 | pte_unmap_unlock(pte - 1, ptl); | |
6119 | cond_resched(); | |
6120 | ||
6121 | if (addr != end) { | |
6122 | /* | |
6123 | * We have consumed all precharges we got in can_attach(). | |
6124 | * We try charge one by one, but don't do any additional | |
6125 | * charges to mc.to if we have failed in charge once in attach() | |
6126 | * phase. | |
6127 | */ | |
854ffa8d | 6128 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6129 | if (!ret) |
6130 | goto retry; | |
6131 | } | |
6132 | ||
6133 | return ret; | |
6134 | } | |
6135 | ||
7b86ac33 CH |
6136 | static const struct mm_walk_ops charge_walk_ops = { |
6137 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6138 | }; | |
6139 | ||
264a0ae1 | 6140 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 6141 | { |
4ffef5fe | 6142 | lru_add_drain_all(); |
312722cb | 6143 | /* |
81f8c3a4 JW |
6144 | * Signal lock_page_memcg() to take the memcg's move_lock |
6145 | * while we're moving its pages to another memcg. Then wait | |
6146 | * for already started RCU-only updates to finish. | |
312722cb JW |
6147 | */ |
6148 | atomic_inc(&mc.from->moving_account); | |
6149 | synchronize_rcu(); | |
dfe076b0 | 6150 | retry: |
d8ed45c5 | 6151 | if (unlikely(!mmap_read_trylock(mc.mm))) { |
dfe076b0 | 6152 | /* |
c1e8d7c6 | 6153 | * Someone who are holding the mmap_lock might be waiting in |
dfe076b0 DN |
6154 | * waitq. So we cancel all extra charges, wake up all waiters, |
6155 | * and retry. Because we cancel precharges, we might not be able | |
6156 | * to move enough charges, but moving charge is a best-effort | |
6157 | * feature anyway, so it wouldn't be a big problem. | |
6158 | */ | |
6159 | __mem_cgroup_clear_mc(); | |
6160 | cond_resched(); | |
6161 | goto retry; | |
6162 | } | |
26bcd64a NH |
6163 | /* |
6164 | * When we have consumed all precharges and failed in doing | |
6165 | * additional charge, the page walk just aborts. | |
6166 | */ | |
7b86ac33 CH |
6167 | walk_page_range(mc.mm, 0, mc.mm->highest_vm_end, &charge_walk_ops, |
6168 | NULL); | |
0247f3f4 | 6169 | |
d8ed45c5 | 6170 | mmap_read_unlock(mc.mm); |
312722cb | 6171 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
6172 | } |
6173 | ||
264a0ae1 | 6174 | static void mem_cgroup_move_task(void) |
67e465a7 | 6175 | { |
264a0ae1 TH |
6176 | if (mc.to) { |
6177 | mem_cgroup_move_charge(); | |
a433658c | 6178 | mem_cgroup_clear_mc(); |
264a0ae1 | 6179 | } |
67e465a7 | 6180 | } |
5cfb80a7 | 6181 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 6182 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6183 | { |
6184 | return 0; | |
6185 | } | |
1f7dd3e5 | 6186 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6187 | { |
6188 | } | |
264a0ae1 | 6189 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
6190 | { |
6191 | } | |
6192 | #endif | |
67e465a7 | 6193 | |
677dc973 CD |
6194 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
6195 | { | |
6196 | if (value == PAGE_COUNTER_MAX) | |
6197 | seq_puts(m, "max\n"); | |
6198 | else | |
6199 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); | |
6200 | ||
6201 | return 0; | |
6202 | } | |
6203 | ||
241994ed JW |
6204 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
6205 | struct cftype *cft) | |
6206 | { | |
f5fc3c5d JW |
6207 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
6208 | ||
6209 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
6210 | } |
6211 | ||
bf8d5d52 RG |
6212 | static int memory_min_show(struct seq_file *m, void *v) |
6213 | { | |
677dc973 CD |
6214 | return seq_puts_memcg_tunable(m, |
6215 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); | |
bf8d5d52 RG |
6216 | } |
6217 | ||
6218 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
6219 | char *buf, size_t nbytes, loff_t off) | |
6220 | { | |
6221 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6222 | unsigned long min; | |
6223 | int err; | |
6224 | ||
6225 | buf = strstrip(buf); | |
6226 | err = page_counter_memparse(buf, "max", &min); | |
6227 | if (err) | |
6228 | return err; | |
6229 | ||
6230 | page_counter_set_min(&memcg->memory, min); | |
6231 | ||
6232 | return nbytes; | |
6233 | } | |
6234 | ||
241994ed JW |
6235 | static int memory_low_show(struct seq_file *m, void *v) |
6236 | { | |
677dc973 CD |
6237 | return seq_puts_memcg_tunable(m, |
6238 | READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); | |
241994ed JW |
6239 | } |
6240 | ||
6241 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
6242 | char *buf, size_t nbytes, loff_t off) | |
6243 | { | |
6244 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6245 | unsigned long low; | |
6246 | int err; | |
6247 | ||
6248 | buf = strstrip(buf); | |
d2973697 | 6249 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
6250 | if (err) |
6251 | return err; | |
6252 | ||
23067153 | 6253 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
6254 | |
6255 | return nbytes; | |
6256 | } | |
6257 | ||
6258 | static int memory_high_show(struct seq_file *m, void *v) | |
6259 | { | |
d1663a90 JK |
6260 | return seq_puts_memcg_tunable(m, |
6261 | READ_ONCE(mem_cgroup_from_seq(m)->memory.high)); | |
241994ed JW |
6262 | } |
6263 | ||
6264 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
6265 | char *buf, size_t nbytes, loff_t off) | |
6266 | { | |
6267 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6268 | unsigned int nr_retries = MAX_RECLAIM_RETRIES; |
8c8c383c | 6269 | bool drained = false; |
241994ed JW |
6270 | unsigned long high; |
6271 | int err; | |
6272 | ||
6273 | buf = strstrip(buf); | |
d2973697 | 6274 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
6275 | if (err) |
6276 | return err; | |
6277 | ||
e82553c1 JW |
6278 | page_counter_set_high(&memcg->memory, high); |
6279 | ||
8c8c383c JW |
6280 | for (;;) { |
6281 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6282 | unsigned long reclaimed; | |
6283 | ||
6284 | if (nr_pages <= high) | |
6285 | break; | |
6286 | ||
6287 | if (signal_pending(current)) | |
6288 | break; | |
6289 | ||
6290 | if (!drained) { | |
6291 | drain_all_stock(memcg); | |
6292 | drained = true; | |
6293 | continue; | |
6294 | } | |
6295 | ||
6296 | reclaimed = try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
6297 | GFP_KERNEL, true); | |
6298 | ||
6299 | if (!reclaimed && !nr_retries--) | |
6300 | break; | |
6301 | } | |
588083bb | 6302 | |
19ce33ac | 6303 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
6304 | return nbytes; |
6305 | } | |
6306 | ||
6307 | static int memory_max_show(struct seq_file *m, void *v) | |
6308 | { | |
677dc973 CD |
6309 | return seq_puts_memcg_tunable(m, |
6310 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); | |
241994ed JW |
6311 | } |
6312 | ||
6313 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
6314 | char *buf, size_t nbytes, loff_t off) | |
6315 | { | |
6316 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6317 | unsigned int nr_reclaims = MAX_RECLAIM_RETRIES; |
b6e6edcf | 6318 | bool drained = false; |
241994ed JW |
6319 | unsigned long max; |
6320 | int err; | |
6321 | ||
6322 | buf = strstrip(buf); | |
d2973697 | 6323 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
6324 | if (err) |
6325 | return err; | |
6326 | ||
bbec2e15 | 6327 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
6328 | |
6329 | for (;;) { | |
6330 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6331 | ||
6332 | if (nr_pages <= max) | |
6333 | break; | |
6334 | ||
7249c9f0 | 6335 | if (signal_pending(current)) |
b6e6edcf | 6336 | break; |
b6e6edcf JW |
6337 | |
6338 | if (!drained) { | |
6339 | drain_all_stock(memcg); | |
6340 | drained = true; | |
6341 | continue; | |
6342 | } | |
6343 | ||
6344 | if (nr_reclaims) { | |
6345 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
6346 | GFP_KERNEL, true)) | |
6347 | nr_reclaims--; | |
6348 | continue; | |
6349 | } | |
6350 | ||
e27be240 | 6351 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
6352 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
6353 | break; | |
6354 | } | |
241994ed | 6355 | |
2529bb3a | 6356 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
6357 | return nbytes; |
6358 | } | |
6359 | ||
1e577f97 SB |
6360 | static void __memory_events_show(struct seq_file *m, atomic_long_t *events) |
6361 | { | |
6362 | seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW])); | |
6363 | seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH])); | |
6364 | seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX])); | |
6365 | seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM])); | |
6366 | seq_printf(m, "oom_kill %lu\n", | |
6367 | atomic_long_read(&events[MEMCG_OOM_KILL])); | |
6368 | } | |
6369 | ||
241994ed JW |
6370 | static int memory_events_show(struct seq_file *m, void *v) |
6371 | { | |
aa9694bb | 6372 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
241994ed | 6373 | |
1e577f97 SB |
6374 | __memory_events_show(m, memcg->memory_events); |
6375 | return 0; | |
6376 | } | |
6377 | ||
6378 | static int memory_events_local_show(struct seq_file *m, void *v) | |
6379 | { | |
6380 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
241994ed | 6381 | |
1e577f97 | 6382 | __memory_events_show(m, memcg->memory_events_local); |
241994ed JW |
6383 | return 0; |
6384 | } | |
6385 | ||
587d9f72 JW |
6386 | static int memory_stat_show(struct seq_file *m, void *v) |
6387 | { | |
aa9694bb | 6388 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
c8713d0b | 6389 | char *buf; |
1ff9e6e1 | 6390 | |
c8713d0b JW |
6391 | buf = memory_stat_format(memcg); |
6392 | if (!buf) | |
6393 | return -ENOMEM; | |
6394 | seq_puts(m, buf); | |
6395 | kfree(buf); | |
587d9f72 JW |
6396 | return 0; |
6397 | } | |
6398 | ||
5f9a4f4a MS |
6399 | #ifdef CONFIG_NUMA |
6400 | static int memory_numa_stat_show(struct seq_file *m, void *v) | |
6401 | { | |
6402 | int i; | |
6403 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
6404 | ||
6405 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { | |
6406 | int nid; | |
6407 | ||
6408 | if (memory_stats[i].idx >= NR_VM_NODE_STAT_ITEMS) | |
6409 | continue; | |
6410 | ||
6411 | seq_printf(m, "%s", memory_stats[i].name); | |
6412 | for_each_node_state(nid, N_MEMORY) { | |
6413 | u64 size; | |
6414 | struct lruvec *lruvec; | |
6415 | ||
6416 | lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); | |
6417 | size = lruvec_page_state(lruvec, memory_stats[i].idx); | |
6418 | size *= memory_stats[i].ratio; | |
6419 | seq_printf(m, " N%d=%llu", nid, size); | |
6420 | } | |
6421 | seq_putc(m, '\n'); | |
6422 | } | |
6423 | ||
6424 | return 0; | |
6425 | } | |
6426 | #endif | |
6427 | ||
3d8b38eb RG |
6428 | static int memory_oom_group_show(struct seq_file *m, void *v) |
6429 | { | |
aa9694bb | 6430 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3d8b38eb RG |
6431 | |
6432 | seq_printf(m, "%d\n", memcg->oom_group); | |
6433 | ||
6434 | return 0; | |
6435 | } | |
6436 | ||
6437 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
6438 | char *buf, size_t nbytes, loff_t off) | |
6439 | { | |
6440 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6441 | int ret, oom_group; | |
6442 | ||
6443 | buf = strstrip(buf); | |
6444 | if (!buf) | |
6445 | return -EINVAL; | |
6446 | ||
6447 | ret = kstrtoint(buf, 0, &oom_group); | |
6448 | if (ret) | |
6449 | return ret; | |
6450 | ||
6451 | if (oom_group != 0 && oom_group != 1) | |
6452 | return -EINVAL; | |
6453 | ||
6454 | memcg->oom_group = oom_group; | |
6455 | ||
6456 | return nbytes; | |
6457 | } | |
6458 | ||
241994ed JW |
6459 | static struct cftype memory_files[] = { |
6460 | { | |
6461 | .name = "current", | |
f5fc3c5d | 6462 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
6463 | .read_u64 = memory_current_read, |
6464 | }, | |
bf8d5d52 RG |
6465 | { |
6466 | .name = "min", | |
6467 | .flags = CFTYPE_NOT_ON_ROOT, | |
6468 | .seq_show = memory_min_show, | |
6469 | .write = memory_min_write, | |
6470 | }, | |
241994ed JW |
6471 | { |
6472 | .name = "low", | |
6473 | .flags = CFTYPE_NOT_ON_ROOT, | |
6474 | .seq_show = memory_low_show, | |
6475 | .write = memory_low_write, | |
6476 | }, | |
6477 | { | |
6478 | .name = "high", | |
6479 | .flags = CFTYPE_NOT_ON_ROOT, | |
6480 | .seq_show = memory_high_show, | |
6481 | .write = memory_high_write, | |
6482 | }, | |
6483 | { | |
6484 | .name = "max", | |
6485 | .flags = CFTYPE_NOT_ON_ROOT, | |
6486 | .seq_show = memory_max_show, | |
6487 | .write = memory_max_write, | |
6488 | }, | |
6489 | { | |
6490 | .name = "events", | |
6491 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 6492 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
6493 | .seq_show = memory_events_show, |
6494 | }, | |
1e577f97 SB |
6495 | { |
6496 | .name = "events.local", | |
6497 | .flags = CFTYPE_NOT_ON_ROOT, | |
6498 | .file_offset = offsetof(struct mem_cgroup, events_local_file), | |
6499 | .seq_show = memory_events_local_show, | |
6500 | }, | |
587d9f72 JW |
6501 | { |
6502 | .name = "stat", | |
587d9f72 JW |
6503 | .seq_show = memory_stat_show, |
6504 | }, | |
5f9a4f4a MS |
6505 | #ifdef CONFIG_NUMA |
6506 | { | |
6507 | .name = "numa_stat", | |
6508 | .seq_show = memory_numa_stat_show, | |
6509 | }, | |
6510 | #endif | |
3d8b38eb RG |
6511 | { |
6512 | .name = "oom.group", | |
6513 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
6514 | .seq_show = memory_oom_group_show, | |
6515 | .write = memory_oom_group_write, | |
6516 | }, | |
241994ed JW |
6517 | { } /* terminate */ |
6518 | }; | |
6519 | ||
073219e9 | 6520 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 6521 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6522 | .css_online = mem_cgroup_css_online, |
92fb9748 | 6523 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 6524 | .css_released = mem_cgroup_css_released, |
92fb9748 | 6525 | .css_free = mem_cgroup_css_free, |
1ced953b | 6526 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
6527 | .can_attach = mem_cgroup_can_attach, |
6528 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 6529 | .post_attach = mem_cgroup_move_task, |
241994ed JW |
6530 | .dfl_cftypes = memory_files, |
6531 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 6532 | .early_init = 0, |
8cdea7c0 | 6533 | }; |
c077719b | 6534 | |
bc50bcc6 JW |
6535 | /* |
6536 | * This function calculates an individual cgroup's effective | |
6537 | * protection which is derived from its own memory.min/low, its | |
6538 | * parent's and siblings' settings, as well as the actual memory | |
6539 | * distribution in the tree. | |
6540 | * | |
6541 | * The following rules apply to the effective protection values: | |
6542 | * | |
6543 | * 1. At the first level of reclaim, effective protection is equal to | |
6544 | * the declared protection in memory.min and memory.low. | |
6545 | * | |
6546 | * 2. To enable safe delegation of the protection configuration, at | |
6547 | * subsequent levels the effective protection is capped to the | |
6548 | * parent's effective protection. | |
6549 | * | |
6550 | * 3. To make complex and dynamic subtrees easier to configure, the | |
6551 | * user is allowed to overcommit the declared protection at a given | |
6552 | * level. If that is the case, the parent's effective protection is | |
6553 | * distributed to the children in proportion to how much protection | |
6554 | * they have declared and how much of it they are utilizing. | |
6555 | * | |
6556 | * This makes distribution proportional, but also work-conserving: | |
6557 | * if one cgroup claims much more protection than it uses memory, | |
6558 | * the unused remainder is available to its siblings. | |
6559 | * | |
6560 | * 4. Conversely, when the declared protection is undercommitted at a | |
6561 | * given level, the distribution of the larger parental protection | |
6562 | * budget is NOT proportional. A cgroup's protection from a sibling | |
6563 | * is capped to its own memory.min/low setting. | |
6564 | * | |
8a931f80 JW |
6565 | * 5. However, to allow protecting recursive subtrees from each other |
6566 | * without having to declare each individual cgroup's fixed share | |
6567 | * of the ancestor's claim to protection, any unutilized - | |
6568 | * "floating" - protection from up the tree is distributed in | |
6569 | * proportion to each cgroup's *usage*. This makes the protection | |
6570 | * neutral wrt sibling cgroups and lets them compete freely over | |
6571 | * the shared parental protection budget, but it protects the | |
6572 | * subtree as a whole from neighboring subtrees. | |
6573 | * | |
6574 | * Note that 4. and 5. are not in conflict: 4. is about protecting | |
6575 | * against immediate siblings whereas 5. is about protecting against | |
6576 | * neighboring subtrees. | |
bc50bcc6 JW |
6577 | */ |
6578 | static unsigned long effective_protection(unsigned long usage, | |
8a931f80 | 6579 | unsigned long parent_usage, |
bc50bcc6 JW |
6580 | unsigned long setting, |
6581 | unsigned long parent_effective, | |
6582 | unsigned long siblings_protected) | |
6583 | { | |
6584 | unsigned long protected; | |
8a931f80 | 6585 | unsigned long ep; |
bc50bcc6 JW |
6586 | |
6587 | protected = min(usage, setting); | |
6588 | /* | |
6589 | * If all cgroups at this level combined claim and use more | |
6590 | * protection then what the parent affords them, distribute | |
6591 | * shares in proportion to utilization. | |
6592 | * | |
6593 | * We are using actual utilization rather than the statically | |
6594 | * claimed protection in order to be work-conserving: claimed | |
6595 | * but unused protection is available to siblings that would | |
6596 | * otherwise get a smaller chunk than what they claimed. | |
6597 | */ | |
6598 | if (siblings_protected > parent_effective) | |
6599 | return protected * parent_effective / siblings_protected; | |
6600 | ||
6601 | /* | |
6602 | * Ok, utilized protection of all children is within what the | |
6603 | * parent affords them, so we know whatever this child claims | |
6604 | * and utilizes is effectively protected. | |
6605 | * | |
6606 | * If there is unprotected usage beyond this value, reclaim | |
6607 | * will apply pressure in proportion to that amount. | |
6608 | * | |
6609 | * If there is unutilized protection, the cgroup will be fully | |
6610 | * shielded from reclaim, but we do return a smaller value for | |
6611 | * protection than what the group could enjoy in theory. This | |
6612 | * is okay. With the overcommit distribution above, effective | |
6613 | * protection is always dependent on how memory is actually | |
6614 | * consumed among the siblings anyway. | |
6615 | */ | |
8a931f80 JW |
6616 | ep = protected; |
6617 | ||
6618 | /* | |
6619 | * If the children aren't claiming (all of) the protection | |
6620 | * afforded to them by the parent, distribute the remainder in | |
6621 | * proportion to the (unprotected) memory of each cgroup. That | |
6622 | * way, cgroups that aren't explicitly prioritized wrt each | |
6623 | * other compete freely over the allowance, but they are | |
6624 | * collectively protected from neighboring trees. | |
6625 | * | |
6626 | * We're using unprotected memory for the weight so that if | |
6627 | * some cgroups DO claim explicit protection, we don't protect | |
6628 | * the same bytes twice. | |
cd324edc JW |
6629 | * |
6630 | * Check both usage and parent_usage against the respective | |
6631 | * protected values. One should imply the other, but they | |
6632 | * aren't read atomically - make sure the division is sane. | |
8a931f80 JW |
6633 | */ |
6634 | if (!(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)) | |
6635 | return ep; | |
cd324edc JW |
6636 | if (parent_effective > siblings_protected && |
6637 | parent_usage > siblings_protected && | |
6638 | usage > protected) { | |
8a931f80 JW |
6639 | unsigned long unclaimed; |
6640 | ||
6641 | unclaimed = parent_effective - siblings_protected; | |
6642 | unclaimed *= usage - protected; | |
6643 | unclaimed /= parent_usage - siblings_protected; | |
6644 | ||
6645 | ep += unclaimed; | |
6646 | } | |
6647 | ||
6648 | return ep; | |
bc50bcc6 JW |
6649 | } |
6650 | ||
241994ed | 6651 | /** |
bf8d5d52 | 6652 | * mem_cgroup_protected - check if memory consumption is in the normal range |
34c81057 | 6653 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
6654 | * @memcg: the memory cgroup to check |
6655 | * | |
23067153 RG |
6656 | * WARNING: This function is not stateless! It can only be used as part |
6657 | * of a top-down tree iteration, not for isolated queries. | |
241994ed | 6658 | */ |
45c7f7e1 CD |
6659 | void mem_cgroup_calculate_protection(struct mem_cgroup *root, |
6660 | struct mem_cgroup *memcg) | |
241994ed | 6661 | { |
8a931f80 | 6662 | unsigned long usage, parent_usage; |
23067153 RG |
6663 | struct mem_cgroup *parent; |
6664 | ||
241994ed | 6665 | if (mem_cgroup_disabled()) |
45c7f7e1 | 6666 | return; |
241994ed | 6667 | |
34c81057 SC |
6668 | if (!root) |
6669 | root = root_mem_cgroup; | |
22f7496f YS |
6670 | |
6671 | /* | |
6672 | * Effective values of the reclaim targets are ignored so they | |
6673 | * can be stale. Have a look at mem_cgroup_protection for more | |
6674 | * details. | |
6675 | * TODO: calculation should be more robust so that we do not need | |
6676 | * that special casing. | |
6677 | */ | |
34c81057 | 6678 | if (memcg == root) |
45c7f7e1 | 6679 | return; |
241994ed | 6680 | |
23067153 | 6681 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 | 6682 | if (!usage) |
45c7f7e1 | 6683 | return; |
bf8d5d52 | 6684 | |
bf8d5d52 | 6685 | parent = parent_mem_cgroup(memcg); |
df2a4196 RG |
6686 | /* No parent means a non-hierarchical mode on v1 memcg */ |
6687 | if (!parent) | |
45c7f7e1 | 6688 | return; |
df2a4196 | 6689 | |
bc50bcc6 | 6690 | if (parent == root) { |
c3d53200 | 6691 | memcg->memory.emin = READ_ONCE(memcg->memory.min); |
03960e33 | 6692 | memcg->memory.elow = READ_ONCE(memcg->memory.low); |
45c7f7e1 | 6693 | return; |
bf8d5d52 RG |
6694 | } |
6695 | ||
8a931f80 JW |
6696 | parent_usage = page_counter_read(&parent->memory); |
6697 | ||
b3a7822e | 6698 | WRITE_ONCE(memcg->memory.emin, effective_protection(usage, parent_usage, |
c3d53200 CD |
6699 | READ_ONCE(memcg->memory.min), |
6700 | READ_ONCE(parent->memory.emin), | |
b3a7822e | 6701 | atomic_long_read(&parent->memory.children_min_usage))); |
23067153 | 6702 | |
b3a7822e | 6703 | WRITE_ONCE(memcg->memory.elow, effective_protection(usage, parent_usage, |
03960e33 CD |
6704 | READ_ONCE(memcg->memory.low), |
6705 | READ_ONCE(parent->memory.elow), | |
b3a7822e | 6706 | atomic_long_read(&parent->memory.children_low_usage))); |
241994ed JW |
6707 | } |
6708 | ||
00501b53 | 6709 | /** |
f0e45fb4 | 6710 | * mem_cgroup_charge - charge a newly allocated page to a cgroup |
00501b53 JW |
6711 | * @page: page to charge |
6712 | * @mm: mm context of the victim | |
6713 | * @gfp_mask: reclaim mode | |
00501b53 JW |
6714 | * |
6715 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
6716 | * pages according to @gfp_mask if necessary. | |
6717 | * | |
f0e45fb4 | 6718 | * Returns 0 on success. Otherwise, an error code is returned. |
00501b53 | 6719 | */ |
d9eb1ea2 | 6720 | int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask) |
00501b53 | 6721 | { |
6c357848 | 6722 | unsigned int nr_pages = thp_nr_pages(page); |
00501b53 | 6723 | struct mem_cgroup *memcg = NULL; |
00501b53 JW |
6724 | int ret = 0; |
6725 | ||
6726 | if (mem_cgroup_disabled()) | |
6727 | goto out; | |
6728 | ||
6729 | if (PageSwapCache(page)) { | |
2d1c4980 JW |
6730 | swp_entry_t ent = { .val = page_private(page), }; |
6731 | unsigned short id; | |
6732 | ||
00501b53 JW |
6733 | /* |
6734 | * Every swap fault against a single page tries to charge the | |
6735 | * page, bail as early as possible. shmem_unuse() encounters | |
bcfe06bf RG |
6736 | * already charged pages, too. page and memcg binding is |
6737 | * protected by the page lock, which serializes swap cache | |
6738 | * removal, which in turn serializes uncharging. | |
00501b53 | 6739 | */ |
e993d905 | 6740 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
bcfe06bf | 6741 | if (page_memcg(compound_head(page))) |
00501b53 | 6742 | goto out; |
e993d905 | 6743 | |
2d1c4980 JW |
6744 | id = lookup_swap_cgroup_id(ent); |
6745 | rcu_read_lock(); | |
6746 | memcg = mem_cgroup_from_id(id); | |
6747 | if (memcg && !css_tryget_online(&memcg->css)) | |
6748 | memcg = NULL; | |
6749 | rcu_read_unlock(); | |
00501b53 JW |
6750 | } |
6751 | ||
00501b53 JW |
6752 | if (!memcg) |
6753 | memcg = get_mem_cgroup_from_mm(mm); | |
6754 | ||
6755 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
f0e45fb4 JW |
6756 | if (ret) |
6757 | goto out_put; | |
00501b53 | 6758 | |
1a3e1f40 | 6759 | css_get(&memcg->css); |
d9eb1ea2 | 6760 | commit_charge(page, memcg); |
6abb5a86 | 6761 | |
6abb5a86 | 6762 | local_irq_disable(); |
3fba69a5 | 6763 | mem_cgroup_charge_statistics(memcg, page, nr_pages); |
6abb5a86 JW |
6764 | memcg_check_events(memcg, page); |
6765 | local_irq_enable(); | |
00501b53 | 6766 | |
2d1c4980 | 6767 | if (PageSwapCache(page)) { |
00501b53 JW |
6768 | swp_entry_t entry = { .val = page_private(page) }; |
6769 | /* | |
6770 | * The swap entry might not get freed for a long time, | |
6771 | * let's not wait for it. The page already received a | |
6772 | * memory+swap charge, drop the swap entry duplicate. | |
6773 | */ | |
38d8b4e6 | 6774 | mem_cgroup_uncharge_swap(entry, nr_pages); |
00501b53 | 6775 | } |
00501b53 | 6776 | |
f0e45fb4 JW |
6777 | out_put: |
6778 | css_put(&memcg->css); | |
6779 | out: | |
6780 | return ret; | |
3fea5a49 JW |
6781 | } |
6782 | ||
a9d5adee JG |
6783 | struct uncharge_gather { |
6784 | struct mem_cgroup *memcg; | |
9f762dbe | 6785 | unsigned long nr_pages; |
a9d5adee | 6786 | unsigned long pgpgout; |
a9d5adee | 6787 | unsigned long nr_kmem; |
a9d5adee JG |
6788 | struct page *dummy_page; |
6789 | }; | |
6790 | ||
6791 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 6792 | { |
a9d5adee JG |
6793 | memset(ug, 0, sizeof(*ug)); |
6794 | } | |
6795 | ||
6796 | static void uncharge_batch(const struct uncharge_gather *ug) | |
6797 | { | |
747db954 JW |
6798 | unsigned long flags; |
6799 | ||
a9d5adee | 6800 | if (!mem_cgroup_is_root(ug->memcg)) { |
9f762dbe | 6801 | page_counter_uncharge(&ug->memcg->memory, ug->nr_pages); |
7941d214 | 6802 | if (do_memsw_account()) |
9f762dbe | 6803 | page_counter_uncharge(&ug->memcg->memsw, ug->nr_pages); |
a9d5adee JG |
6804 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem) |
6805 | page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem); | |
6806 | memcg_oom_recover(ug->memcg); | |
ce00a967 | 6807 | } |
747db954 JW |
6808 | |
6809 | local_irq_save(flags); | |
c9019e9b | 6810 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); |
9f762dbe | 6811 | __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, ug->nr_pages); |
a9d5adee | 6812 | memcg_check_events(ug->memcg, ug->dummy_page); |
747db954 | 6813 | local_irq_restore(flags); |
f1796544 MH |
6814 | |
6815 | /* drop reference from uncharge_page */ | |
6816 | css_put(&ug->memcg->css); | |
a9d5adee JG |
6817 | } |
6818 | ||
6819 | static void uncharge_page(struct page *page, struct uncharge_gather *ug) | |
6820 | { | |
9f762dbe JW |
6821 | unsigned long nr_pages; |
6822 | ||
a9d5adee | 6823 | VM_BUG_ON_PAGE(PageLRU(page), page); |
a9d5adee | 6824 | |
bcfe06bf | 6825 | if (!page_memcg(page)) |
a9d5adee JG |
6826 | return; |
6827 | ||
6828 | /* | |
6829 | * Nobody should be changing or seriously looking at | |
bcfe06bf | 6830 | * page_memcg(page) at this point, we have fully |
a9d5adee JG |
6831 | * exclusive access to the page. |
6832 | */ | |
6833 | ||
bcfe06bf | 6834 | if (ug->memcg != page_memcg(page)) { |
a9d5adee JG |
6835 | if (ug->memcg) { |
6836 | uncharge_batch(ug); | |
6837 | uncharge_gather_clear(ug); | |
6838 | } | |
bcfe06bf | 6839 | ug->memcg = page_memcg(page); |
f1796544 MH |
6840 | |
6841 | /* pairs with css_put in uncharge_batch */ | |
6842 | css_get(&ug->memcg->css); | |
a9d5adee JG |
6843 | } |
6844 | ||
9f762dbe JW |
6845 | nr_pages = compound_nr(page); |
6846 | ug->nr_pages += nr_pages; | |
a9d5adee | 6847 | |
18b2db3b | 6848 | if (PageMemcgKmem(page)) |
9f762dbe | 6849 | ug->nr_kmem += nr_pages; |
18b2db3b RG |
6850 | else |
6851 | ug->pgpgout++; | |
a9d5adee JG |
6852 | |
6853 | ug->dummy_page = page; | |
bcfe06bf | 6854 | page->memcg_data = 0; |
1a3e1f40 | 6855 | css_put(&ug->memcg->css); |
747db954 JW |
6856 | } |
6857 | ||
6858 | static void uncharge_list(struct list_head *page_list) | |
6859 | { | |
a9d5adee | 6860 | struct uncharge_gather ug; |
747db954 | 6861 | struct list_head *next; |
a9d5adee JG |
6862 | |
6863 | uncharge_gather_clear(&ug); | |
747db954 | 6864 | |
8b592656 JW |
6865 | /* |
6866 | * Note that the list can be a single page->lru; hence the | |
6867 | * do-while loop instead of a simple list_for_each_entry(). | |
6868 | */ | |
747db954 JW |
6869 | next = page_list->next; |
6870 | do { | |
a9d5adee JG |
6871 | struct page *page; |
6872 | ||
747db954 JW |
6873 | page = list_entry(next, struct page, lru); |
6874 | next = page->lru.next; | |
6875 | ||
a9d5adee | 6876 | uncharge_page(page, &ug); |
747db954 JW |
6877 | } while (next != page_list); |
6878 | ||
a9d5adee JG |
6879 | if (ug.memcg) |
6880 | uncharge_batch(&ug); | |
747db954 JW |
6881 | } |
6882 | ||
0a31bc97 JW |
6883 | /** |
6884 | * mem_cgroup_uncharge - uncharge a page | |
6885 | * @page: page to uncharge | |
6886 | * | |
f0e45fb4 | 6887 | * Uncharge a page previously charged with mem_cgroup_charge(). |
0a31bc97 JW |
6888 | */ |
6889 | void mem_cgroup_uncharge(struct page *page) | |
6890 | { | |
a9d5adee JG |
6891 | struct uncharge_gather ug; |
6892 | ||
0a31bc97 JW |
6893 | if (mem_cgroup_disabled()) |
6894 | return; | |
6895 | ||
747db954 | 6896 | /* Don't touch page->lru of any random page, pre-check: */ |
bcfe06bf | 6897 | if (!page_memcg(page)) |
0a31bc97 JW |
6898 | return; |
6899 | ||
a9d5adee JG |
6900 | uncharge_gather_clear(&ug); |
6901 | uncharge_page(page, &ug); | |
6902 | uncharge_batch(&ug); | |
747db954 | 6903 | } |
0a31bc97 | 6904 | |
747db954 JW |
6905 | /** |
6906 | * mem_cgroup_uncharge_list - uncharge a list of page | |
6907 | * @page_list: list of pages to uncharge | |
6908 | * | |
6909 | * Uncharge a list of pages previously charged with | |
f0e45fb4 | 6910 | * mem_cgroup_charge(). |
747db954 JW |
6911 | */ |
6912 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
6913 | { | |
6914 | if (mem_cgroup_disabled()) | |
6915 | return; | |
0a31bc97 | 6916 | |
747db954 JW |
6917 | if (!list_empty(page_list)) |
6918 | uncharge_list(page_list); | |
0a31bc97 JW |
6919 | } |
6920 | ||
6921 | /** | |
6a93ca8f JW |
6922 | * mem_cgroup_migrate - charge a page's replacement |
6923 | * @oldpage: currently circulating page | |
6924 | * @newpage: replacement page | |
0a31bc97 | 6925 | * |
6a93ca8f JW |
6926 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
6927 | * be uncharged upon free. | |
0a31bc97 JW |
6928 | * |
6929 | * Both pages must be locked, @newpage->mapping must be set up. | |
6930 | */ | |
6a93ca8f | 6931 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 6932 | { |
29833315 | 6933 | struct mem_cgroup *memcg; |
44b7a8d3 | 6934 | unsigned int nr_pages; |
d93c4130 | 6935 | unsigned long flags; |
0a31bc97 JW |
6936 | |
6937 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6938 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 6939 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
6940 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6941 | newpage); | |
0a31bc97 JW |
6942 | |
6943 | if (mem_cgroup_disabled()) | |
6944 | return; | |
6945 | ||
6946 | /* Page cache replacement: new page already charged? */ | |
bcfe06bf | 6947 | if (page_memcg(newpage)) |
0a31bc97 JW |
6948 | return; |
6949 | ||
bcfe06bf | 6950 | memcg = page_memcg(oldpage); |
a4055888 | 6951 | VM_WARN_ON_ONCE_PAGE(!memcg, oldpage); |
29833315 | 6952 | if (!memcg) |
0a31bc97 JW |
6953 | return; |
6954 | ||
44b7a8d3 | 6955 | /* Force-charge the new page. The old one will be freed soon */ |
6c357848 | 6956 | nr_pages = thp_nr_pages(newpage); |
44b7a8d3 JW |
6957 | |
6958 | page_counter_charge(&memcg->memory, nr_pages); | |
6959 | if (do_memsw_account()) | |
6960 | page_counter_charge(&memcg->memsw, nr_pages); | |
0a31bc97 | 6961 | |
1a3e1f40 | 6962 | css_get(&memcg->css); |
d9eb1ea2 | 6963 | commit_charge(newpage, memcg); |
44b7a8d3 | 6964 | |
d93c4130 | 6965 | local_irq_save(flags); |
3fba69a5 | 6966 | mem_cgroup_charge_statistics(memcg, newpage, nr_pages); |
44b7a8d3 | 6967 | memcg_check_events(memcg, newpage); |
d93c4130 | 6968 | local_irq_restore(flags); |
0a31bc97 JW |
6969 | } |
6970 | ||
ef12947c | 6971 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
6972 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
6973 | ||
2d758073 | 6974 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
6975 | { |
6976 | struct mem_cgroup *memcg; | |
6977 | ||
2d758073 JW |
6978 | if (!mem_cgroup_sockets_enabled) |
6979 | return; | |
6980 | ||
e876ecc6 SB |
6981 | /* Do not associate the sock with unrelated interrupted task's memcg. */ |
6982 | if (in_interrupt()) | |
6983 | return; | |
6984 | ||
11092087 JW |
6985 | rcu_read_lock(); |
6986 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
6987 | if (memcg == root_mem_cgroup) |
6988 | goto out; | |
0db15298 | 6989 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 6990 | goto out; |
8965aa28 | 6991 | if (css_tryget(&memcg->css)) |
11092087 | 6992 | sk->sk_memcg = memcg; |
f7e1cb6e | 6993 | out: |
11092087 JW |
6994 | rcu_read_unlock(); |
6995 | } | |
11092087 | 6996 | |
2d758073 | 6997 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 6998 | { |
2d758073 JW |
6999 | if (sk->sk_memcg) |
7000 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
7001 | } |
7002 | ||
7003 | /** | |
7004 | * mem_cgroup_charge_skmem - charge socket memory | |
7005 | * @memcg: memcg to charge | |
7006 | * @nr_pages: number of pages to charge | |
7007 | * | |
7008 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
7009 | * @memcg's configured limit, %false if the charge had to be forced. | |
7010 | */ | |
7011 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
7012 | { | |
f7e1cb6e | 7013 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 7014 | |
f7e1cb6e | 7015 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 7016 | struct page_counter *fail; |
f7e1cb6e | 7017 | |
0db15298 JW |
7018 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
7019 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
7020 | return true; |
7021 | } | |
0db15298 JW |
7022 | page_counter_charge(&memcg->tcpmem, nr_pages); |
7023 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 7024 | return false; |
11092087 | 7025 | } |
d886f4e4 | 7026 | |
f7e1cb6e JW |
7027 | /* Don't block in the packet receive path */ |
7028 | if (in_softirq()) | |
7029 | gfp_mask = GFP_NOWAIT; | |
7030 | ||
c9019e9b | 7031 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); |
b2807f07 | 7032 | |
f7e1cb6e JW |
7033 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
7034 | return true; | |
7035 | ||
7036 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
7037 | return false; |
7038 | } | |
7039 | ||
7040 | /** | |
7041 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
7042 | * @memcg: memcg to uncharge |
7043 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
7044 | */ |
7045 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
7046 | { | |
f7e1cb6e | 7047 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 7048 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
7049 | return; |
7050 | } | |
d886f4e4 | 7051 | |
c9019e9b | 7052 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 7053 | |
475d0487 | 7054 | refill_stock(memcg, nr_pages); |
11092087 JW |
7055 | } |
7056 | ||
f7e1cb6e JW |
7057 | static int __init cgroup_memory(char *s) |
7058 | { | |
7059 | char *token; | |
7060 | ||
7061 | while ((token = strsep(&s, ",")) != NULL) { | |
7062 | if (!*token) | |
7063 | continue; | |
7064 | if (!strcmp(token, "nosocket")) | |
7065 | cgroup_memory_nosocket = true; | |
04823c83 VD |
7066 | if (!strcmp(token, "nokmem")) |
7067 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
7068 | } |
7069 | return 0; | |
7070 | } | |
7071 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 7072 | |
2d11085e | 7073 | /* |
1081312f MH |
7074 | * subsys_initcall() for memory controller. |
7075 | * | |
308167fc SAS |
7076 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
7077 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
7078 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
7079 | * should be initialized from here. | |
2d11085e MH |
7080 | */ |
7081 | static int __init mem_cgroup_init(void) | |
7082 | { | |
95a045f6 JW |
7083 | int cpu, node; |
7084 | ||
f3344adf MS |
7085 | /* |
7086 | * Currently s32 type (can refer to struct batched_lruvec_stat) is | |
7087 | * used for per-memcg-per-cpu caching of per-node statistics. In order | |
7088 | * to work fine, we should make sure that the overfill threshold can't | |
7089 | * exceed S32_MAX / PAGE_SIZE. | |
7090 | */ | |
7091 | BUILD_BUG_ON(MEMCG_CHARGE_BATCH > S32_MAX / PAGE_SIZE); | |
7092 | ||
308167fc SAS |
7093 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
7094 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
7095 | |
7096 | for_each_possible_cpu(cpu) | |
7097 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
7098 | drain_local_stock); | |
7099 | ||
7100 | for_each_node(node) { | |
7101 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
7102 | |
7103 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
7104 | node_online(node) ? node : NUMA_NO_NODE); | |
7105 | ||
ef8f2327 | 7106 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 7107 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 7108 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
7109 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
7110 | } | |
7111 | ||
2d11085e MH |
7112 | return 0; |
7113 | } | |
7114 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
7115 | |
7116 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
7117 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
7118 | { | |
1c2d479a | 7119 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
7120 | /* |
7121 | * The root cgroup cannot be destroyed, so it's refcount must | |
7122 | * always be >= 1. | |
7123 | */ | |
7124 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
7125 | VM_BUG_ON(1); | |
7126 | break; | |
7127 | } | |
7128 | memcg = parent_mem_cgroup(memcg); | |
7129 | if (!memcg) | |
7130 | memcg = root_mem_cgroup; | |
7131 | } | |
7132 | return memcg; | |
7133 | } | |
7134 | ||
21afa38e JW |
7135 | /** |
7136 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
7137 | * @page: page whose memsw charge to transfer | |
7138 | * @entry: swap entry to move the charge to | |
7139 | * | |
7140 | * Transfer the memsw charge of @page to @entry. | |
7141 | */ | |
7142 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
7143 | { | |
1f47b61f | 7144 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 7145 | unsigned int nr_entries; |
21afa38e JW |
7146 | unsigned short oldid; |
7147 | ||
7148 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
7149 | VM_BUG_ON_PAGE(page_count(page), page); | |
7150 | ||
76358ab5 AS |
7151 | if (mem_cgroup_disabled()) |
7152 | return; | |
7153 | ||
2d1c4980 | 7154 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
21afa38e JW |
7155 | return; |
7156 | ||
bcfe06bf | 7157 | memcg = page_memcg(page); |
21afa38e | 7158 | |
a4055888 | 7159 | VM_WARN_ON_ONCE_PAGE(!memcg, page); |
21afa38e JW |
7160 | if (!memcg) |
7161 | return; | |
7162 | ||
1f47b61f VD |
7163 | /* |
7164 | * In case the memcg owning these pages has been offlined and doesn't | |
7165 | * have an ID allocated to it anymore, charge the closest online | |
7166 | * ancestor for the swap instead and transfer the memory+swap charge. | |
7167 | */ | |
7168 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
6c357848 | 7169 | nr_entries = thp_nr_pages(page); |
d6810d73 HY |
7170 | /* Get references for the tail pages, too */ |
7171 | if (nr_entries > 1) | |
7172 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
7173 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
7174 | nr_entries); | |
21afa38e | 7175 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7176 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e | 7177 | |
bcfe06bf | 7178 | page->memcg_data = 0; |
21afa38e JW |
7179 | |
7180 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 7181 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 7182 | |
2d1c4980 | 7183 | if (!cgroup_memory_noswap && memcg != swap_memcg) { |
1f47b61f | 7184 | if (!mem_cgroup_is_root(swap_memcg)) |
d6810d73 HY |
7185 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
7186 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
7187 | } |
7188 | ||
ce9ce665 SAS |
7189 | /* |
7190 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 7191 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 7192 | * important here to have the interrupts disabled because it is the |
b93b0163 | 7193 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 SAS |
7194 | */ |
7195 | VM_BUG_ON(!irqs_disabled()); | |
3fba69a5 | 7196 | mem_cgroup_charge_statistics(memcg, page, -nr_entries); |
21afa38e | 7197 | memcg_check_events(memcg, page); |
73f576c0 | 7198 | |
1a3e1f40 | 7199 | css_put(&memcg->css); |
21afa38e JW |
7200 | } |
7201 | ||
38d8b4e6 HY |
7202 | /** |
7203 | * mem_cgroup_try_charge_swap - try charging swap space for a page | |
37e84351 VD |
7204 | * @page: page being added to swap |
7205 | * @entry: swap entry to charge | |
7206 | * | |
38d8b4e6 | 7207 | * Try to charge @page's memcg for the swap space at @entry. |
37e84351 VD |
7208 | * |
7209 | * Returns 0 on success, -ENOMEM on failure. | |
7210 | */ | |
7211 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
7212 | { | |
6c357848 | 7213 | unsigned int nr_pages = thp_nr_pages(page); |
37e84351 | 7214 | struct page_counter *counter; |
38d8b4e6 | 7215 | struct mem_cgroup *memcg; |
37e84351 VD |
7216 | unsigned short oldid; |
7217 | ||
76358ab5 AS |
7218 | if (mem_cgroup_disabled()) |
7219 | return 0; | |
7220 | ||
2d1c4980 | 7221 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
37e84351 VD |
7222 | return 0; |
7223 | ||
bcfe06bf | 7224 | memcg = page_memcg(page); |
37e84351 | 7225 | |
a4055888 | 7226 | VM_WARN_ON_ONCE_PAGE(!memcg, page); |
37e84351 VD |
7227 | if (!memcg) |
7228 | return 0; | |
7229 | ||
f3a53a3a TH |
7230 | if (!entry.val) { |
7231 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 7232 | return 0; |
f3a53a3a | 7233 | } |
bb98f2c5 | 7234 | |
1f47b61f VD |
7235 | memcg = mem_cgroup_id_get_online(memcg); |
7236 | ||
2d1c4980 | 7237 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg) && |
38d8b4e6 | 7238 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
7239 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
7240 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 7241 | mem_cgroup_id_put(memcg); |
37e84351 | 7242 | return -ENOMEM; |
1f47b61f | 7243 | } |
37e84351 | 7244 | |
38d8b4e6 HY |
7245 | /* Get references for the tail pages, too */ |
7246 | if (nr_pages > 1) | |
7247 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
7248 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
37e84351 | 7249 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7250 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 7251 | |
37e84351 VD |
7252 | return 0; |
7253 | } | |
7254 | ||
21afa38e | 7255 | /** |
38d8b4e6 | 7256 | * mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 7257 | * @entry: swap entry to uncharge |
38d8b4e6 | 7258 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 7259 | */ |
38d8b4e6 | 7260 | void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
7261 | { |
7262 | struct mem_cgroup *memcg; | |
7263 | unsigned short id; | |
7264 | ||
38d8b4e6 | 7265 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 7266 | rcu_read_lock(); |
adbe427b | 7267 | memcg = mem_cgroup_from_id(id); |
21afa38e | 7268 | if (memcg) { |
2d1c4980 | 7269 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg)) { |
37e84351 | 7270 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
38d8b4e6 | 7271 | page_counter_uncharge(&memcg->swap, nr_pages); |
37e84351 | 7272 | else |
38d8b4e6 | 7273 | page_counter_uncharge(&memcg->memsw, nr_pages); |
37e84351 | 7274 | } |
c9019e9b | 7275 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 7276 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
7277 | } |
7278 | rcu_read_unlock(); | |
7279 | } | |
7280 | ||
d8b38438 VD |
7281 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
7282 | { | |
7283 | long nr_swap_pages = get_nr_swap_pages(); | |
7284 | ||
eccb52e7 | 7285 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
d8b38438 VD |
7286 | return nr_swap_pages; |
7287 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
7288 | nr_swap_pages = min_t(long, nr_swap_pages, | |
bbec2e15 | 7289 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
7290 | page_counter_read(&memcg->swap)); |
7291 | return nr_swap_pages; | |
7292 | } | |
7293 | ||
5ccc5aba VD |
7294 | bool mem_cgroup_swap_full(struct page *page) |
7295 | { | |
7296 | struct mem_cgroup *memcg; | |
7297 | ||
7298 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
7299 | ||
7300 | if (vm_swap_full()) | |
7301 | return true; | |
eccb52e7 | 7302 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
5ccc5aba VD |
7303 | return false; |
7304 | ||
bcfe06bf | 7305 | memcg = page_memcg(page); |
5ccc5aba VD |
7306 | if (!memcg) |
7307 | return false; | |
7308 | ||
4b82ab4f JK |
7309 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { |
7310 | unsigned long usage = page_counter_read(&memcg->swap); | |
7311 | ||
7312 | if (usage * 2 >= READ_ONCE(memcg->swap.high) || | |
7313 | usage * 2 >= READ_ONCE(memcg->swap.max)) | |
5ccc5aba | 7314 | return true; |
4b82ab4f | 7315 | } |
5ccc5aba VD |
7316 | |
7317 | return false; | |
7318 | } | |
7319 | ||
eccb52e7 | 7320 | static int __init setup_swap_account(char *s) |
21afa38e JW |
7321 | { |
7322 | if (!strcmp(s, "1")) | |
5ab92901 | 7323 | cgroup_memory_noswap = false; |
21afa38e | 7324 | else if (!strcmp(s, "0")) |
5ab92901 | 7325 | cgroup_memory_noswap = true; |
21afa38e JW |
7326 | return 1; |
7327 | } | |
eccb52e7 | 7328 | __setup("swapaccount=", setup_swap_account); |
21afa38e | 7329 | |
37e84351 VD |
7330 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
7331 | struct cftype *cft) | |
7332 | { | |
7333 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
7334 | ||
7335 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
7336 | } | |
7337 | ||
4b82ab4f JK |
7338 | static int swap_high_show(struct seq_file *m, void *v) |
7339 | { | |
7340 | return seq_puts_memcg_tunable(m, | |
7341 | READ_ONCE(mem_cgroup_from_seq(m)->swap.high)); | |
7342 | } | |
7343 | ||
7344 | static ssize_t swap_high_write(struct kernfs_open_file *of, | |
7345 | char *buf, size_t nbytes, loff_t off) | |
7346 | { | |
7347 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7348 | unsigned long high; | |
7349 | int err; | |
7350 | ||
7351 | buf = strstrip(buf); | |
7352 | err = page_counter_memparse(buf, "max", &high); | |
7353 | if (err) | |
7354 | return err; | |
7355 | ||
7356 | page_counter_set_high(&memcg->swap, high); | |
7357 | ||
7358 | return nbytes; | |
7359 | } | |
7360 | ||
37e84351 VD |
7361 | static int swap_max_show(struct seq_file *m, void *v) |
7362 | { | |
677dc973 CD |
7363 | return seq_puts_memcg_tunable(m, |
7364 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); | |
37e84351 VD |
7365 | } |
7366 | ||
7367 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
7368 | char *buf, size_t nbytes, loff_t off) | |
7369 | { | |
7370 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7371 | unsigned long max; | |
7372 | int err; | |
7373 | ||
7374 | buf = strstrip(buf); | |
7375 | err = page_counter_memparse(buf, "max", &max); | |
7376 | if (err) | |
7377 | return err; | |
7378 | ||
be09102b | 7379 | xchg(&memcg->swap.max, max); |
37e84351 VD |
7380 | |
7381 | return nbytes; | |
7382 | } | |
7383 | ||
f3a53a3a TH |
7384 | static int swap_events_show(struct seq_file *m, void *v) |
7385 | { | |
aa9694bb | 7386 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
f3a53a3a | 7387 | |
4b82ab4f JK |
7388 | seq_printf(m, "high %lu\n", |
7389 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_HIGH])); | |
f3a53a3a TH |
7390 | seq_printf(m, "max %lu\n", |
7391 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
7392 | seq_printf(m, "fail %lu\n", | |
7393 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
7394 | ||
7395 | return 0; | |
7396 | } | |
7397 | ||
37e84351 VD |
7398 | static struct cftype swap_files[] = { |
7399 | { | |
7400 | .name = "swap.current", | |
7401 | .flags = CFTYPE_NOT_ON_ROOT, | |
7402 | .read_u64 = swap_current_read, | |
7403 | }, | |
4b82ab4f JK |
7404 | { |
7405 | .name = "swap.high", | |
7406 | .flags = CFTYPE_NOT_ON_ROOT, | |
7407 | .seq_show = swap_high_show, | |
7408 | .write = swap_high_write, | |
7409 | }, | |
37e84351 VD |
7410 | { |
7411 | .name = "swap.max", | |
7412 | .flags = CFTYPE_NOT_ON_ROOT, | |
7413 | .seq_show = swap_max_show, | |
7414 | .write = swap_max_write, | |
7415 | }, | |
f3a53a3a TH |
7416 | { |
7417 | .name = "swap.events", | |
7418 | .flags = CFTYPE_NOT_ON_ROOT, | |
7419 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
7420 | .seq_show = swap_events_show, | |
7421 | }, | |
37e84351 VD |
7422 | { } /* terminate */ |
7423 | }; | |
7424 | ||
eccb52e7 | 7425 | static struct cftype memsw_files[] = { |
21afa38e JW |
7426 | { |
7427 | .name = "memsw.usage_in_bytes", | |
7428 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
7429 | .read_u64 = mem_cgroup_read_u64, | |
7430 | }, | |
7431 | { | |
7432 | .name = "memsw.max_usage_in_bytes", | |
7433 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
7434 | .write = mem_cgroup_reset, | |
7435 | .read_u64 = mem_cgroup_read_u64, | |
7436 | }, | |
7437 | { | |
7438 | .name = "memsw.limit_in_bytes", | |
7439 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
7440 | .write = mem_cgroup_write, | |
7441 | .read_u64 = mem_cgroup_read_u64, | |
7442 | }, | |
7443 | { | |
7444 | .name = "memsw.failcnt", | |
7445 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
7446 | .write = mem_cgroup_reset, | |
7447 | .read_u64 = mem_cgroup_read_u64, | |
7448 | }, | |
7449 | { }, /* terminate */ | |
7450 | }; | |
7451 | ||
82ff165c BS |
7452 | /* |
7453 | * If mem_cgroup_swap_init() is implemented as a subsys_initcall() | |
7454 | * instead of a core_initcall(), this could mean cgroup_memory_noswap still | |
7455 | * remains set to false even when memcg is disabled via "cgroup_disable=memory" | |
7456 | * boot parameter. This may result in premature OOPS inside | |
7457 | * mem_cgroup_get_nr_swap_pages() function in corner cases. | |
7458 | */ | |
21afa38e JW |
7459 | static int __init mem_cgroup_swap_init(void) |
7460 | { | |
2d1c4980 JW |
7461 | /* No memory control -> no swap control */ |
7462 | if (mem_cgroup_disabled()) | |
7463 | cgroup_memory_noswap = true; | |
7464 | ||
7465 | if (cgroup_memory_noswap) | |
eccb52e7 JW |
7466 | return 0; |
7467 | ||
7468 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, swap_files)); | |
7469 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, memsw_files)); | |
7470 | ||
21afa38e JW |
7471 | return 0; |
7472 | } | |
82ff165c | 7473 | core_initcall(mem_cgroup_swap_init); |
21afa38e JW |
7474 | |
7475 | #endif /* CONFIG_MEMCG_SWAP */ |