]>
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 | ||
c1a660de RG |
258 | static int __memcg_kmem_charge(struct mem_cgroup *memcg, gfp_t gfp, |
259 | unsigned int nr_pages); | |
260 | static void __memcg_kmem_uncharge(struct mem_cgroup *memcg, | |
261 | unsigned int nr_pages); | |
262 | ||
bf4f0599 RG |
263 | static void obj_cgroup_release(struct percpu_ref *ref) |
264 | { | |
265 | struct obj_cgroup *objcg = container_of(ref, struct obj_cgroup, refcnt); | |
266 | struct mem_cgroup *memcg; | |
267 | unsigned int nr_bytes; | |
268 | unsigned int nr_pages; | |
269 | unsigned long flags; | |
270 | ||
271 | /* | |
272 | * At this point all allocated objects are freed, and | |
273 | * objcg->nr_charged_bytes can't have an arbitrary byte value. | |
274 | * However, it can be PAGE_SIZE or (x * PAGE_SIZE). | |
275 | * | |
276 | * The following sequence can lead to it: | |
277 | * 1) CPU0: objcg == stock->cached_objcg | |
278 | * 2) CPU1: we do a small allocation (e.g. 92 bytes), | |
279 | * PAGE_SIZE bytes are charged | |
280 | * 3) CPU1: a process from another memcg is allocating something, | |
281 | * the stock if flushed, | |
282 | * objcg->nr_charged_bytes = PAGE_SIZE - 92 | |
283 | * 5) CPU0: we do release this object, | |
284 | * 92 bytes are added to stock->nr_bytes | |
285 | * 6) CPU0: stock is flushed, | |
286 | * 92 bytes are added to objcg->nr_charged_bytes | |
287 | * | |
288 | * In the result, nr_charged_bytes == PAGE_SIZE. | |
289 | * This page will be uncharged in obj_cgroup_release(). | |
290 | */ | |
291 | nr_bytes = atomic_read(&objcg->nr_charged_bytes); | |
292 | WARN_ON_ONCE(nr_bytes & (PAGE_SIZE - 1)); | |
293 | nr_pages = nr_bytes >> PAGE_SHIFT; | |
294 | ||
295 | spin_lock_irqsave(&css_set_lock, flags); | |
296 | memcg = obj_cgroup_memcg(objcg); | |
297 | if (nr_pages) | |
298 | __memcg_kmem_uncharge(memcg, nr_pages); | |
299 | list_del(&objcg->list); | |
300 | mem_cgroup_put(memcg); | |
301 | spin_unlock_irqrestore(&css_set_lock, flags); | |
302 | ||
303 | percpu_ref_exit(ref); | |
304 | kfree_rcu(objcg, rcu); | |
305 | } | |
306 | ||
307 | static struct obj_cgroup *obj_cgroup_alloc(void) | |
308 | { | |
309 | struct obj_cgroup *objcg; | |
310 | int ret; | |
311 | ||
312 | objcg = kzalloc(sizeof(struct obj_cgroup), GFP_KERNEL); | |
313 | if (!objcg) | |
314 | return NULL; | |
315 | ||
316 | ret = percpu_ref_init(&objcg->refcnt, obj_cgroup_release, 0, | |
317 | GFP_KERNEL); | |
318 | if (ret) { | |
319 | kfree(objcg); | |
320 | return NULL; | |
321 | } | |
322 | INIT_LIST_HEAD(&objcg->list); | |
323 | return objcg; | |
324 | } | |
325 | ||
326 | static void memcg_reparent_objcgs(struct mem_cgroup *memcg, | |
327 | struct mem_cgroup *parent) | |
328 | { | |
329 | struct obj_cgroup *objcg, *iter; | |
330 | ||
331 | objcg = rcu_replace_pointer(memcg->objcg, NULL, true); | |
332 | ||
333 | spin_lock_irq(&css_set_lock); | |
334 | ||
335 | /* Move active objcg to the parent's list */ | |
336 | xchg(&objcg->memcg, parent); | |
337 | css_get(&parent->css); | |
338 | list_add(&objcg->list, &parent->objcg_list); | |
339 | ||
340 | /* Move already reparented objcgs to the parent's list */ | |
341 | list_for_each_entry(iter, &memcg->objcg_list, list) { | |
342 | css_get(&parent->css); | |
343 | xchg(&iter->memcg, parent); | |
344 | css_put(&memcg->css); | |
345 | } | |
346 | list_splice(&memcg->objcg_list, &parent->objcg_list); | |
347 | ||
348 | spin_unlock_irq(&css_set_lock); | |
349 | ||
350 | percpu_ref_kill(&objcg->refcnt); | |
351 | } | |
352 | ||
55007d84 | 353 | /* |
9855609b | 354 | * This will be used as a shrinker list's index. |
b8627835 LZ |
355 | * The main reason for not using cgroup id for this: |
356 | * this works better in sparse environments, where we have a lot of memcgs, | |
357 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
358 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
359 | * 200 entry array for that. | |
55007d84 | 360 | * |
dbcf73e2 VD |
361 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
362 | * will double each time we have to increase it. | |
55007d84 | 363 | */ |
dbcf73e2 VD |
364 | static DEFINE_IDA(memcg_cache_ida); |
365 | int memcg_nr_cache_ids; | |
749c5415 | 366 | |
05257a1a VD |
367 | /* Protects memcg_nr_cache_ids */ |
368 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
369 | ||
370 | void memcg_get_cache_ids(void) | |
371 | { | |
372 | down_read(&memcg_cache_ids_sem); | |
373 | } | |
374 | ||
375 | void memcg_put_cache_ids(void) | |
376 | { | |
377 | up_read(&memcg_cache_ids_sem); | |
378 | } | |
379 | ||
55007d84 GC |
380 | /* |
381 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
382 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
383 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
384 | * tunable, but that is strictly not necessary. | |
385 | * | |
b8627835 | 386 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
387 | * this constant directly from cgroup, but it is understandable that this is |
388 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 389 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
390 | * increase ours as well if it increases. |
391 | */ | |
392 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 393 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 394 | |
d7f25f8a GC |
395 | /* |
396 | * A lot of the calls to the cache allocation functions are expected to be | |
272911a4 | 397 | * inlined by the compiler. Since the calls to memcg_slab_pre_alloc_hook() are |
d7f25f8a GC |
398 | * conditional to this static branch, we'll have to allow modules that does |
399 | * kmem_cache_alloc and the such to see this symbol as well | |
400 | */ | |
ef12947c | 401 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
d7f25f8a | 402 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
0a432dcb | 403 | #endif |
17cc4dfe | 404 | |
0a4465d3 KT |
405 | static int memcg_shrinker_map_size; |
406 | static DEFINE_MUTEX(memcg_shrinker_map_mutex); | |
407 | ||
408 | static void memcg_free_shrinker_map_rcu(struct rcu_head *head) | |
409 | { | |
410 | kvfree(container_of(head, struct memcg_shrinker_map, rcu)); | |
411 | } | |
412 | ||
413 | static int memcg_expand_one_shrinker_map(struct mem_cgroup *memcg, | |
414 | int size, int old_size) | |
415 | { | |
416 | struct memcg_shrinker_map *new, *old; | |
417 | int nid; | |
418 | ||
419 | lockdep_assert_held(&memcg_shrinker_map_mutex); | |
420 | ||
421 | for_each_node(nid) { | |
422 | old = rcu_dereference_protected( | |
423 | mem_cgroup_nodeinfo(memcg, nid)->shrinker_map, true); | |
424 | /* Not yet online memcg */ | |
425 | if (!old) | |
426 | return 0; | |
427 | ||
86daf94e | 428 | new = kvmalloc_node(sizeof(*new) + size, GFP_KERNEL, nid); |
0a4465d3 KT |
429 | if (!new) |
430 | return -ENOMEM; | |
431 | ||
432 | /* Set all old bits, clear all new bits */ | |
433 | memset(new->map, (int)0xff, old_size); | |
434 | memset((void *)new->map + old_size, 0, size - old_size); | |
435 | ||
436 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, new); | |
437 | call_rcu(&old->rcu, memcg_free_shrinker_map_rcu); | |
438 | } | |
439 | ||
440 | return 0; | |
441 | } | |
442 | ||
443 | static void memcg_free_shrinker_maps(struct mem_cgroup *memcg) | |
444 | { | |
445 | struct mem_cgroup_per_node *pn; | |
446 | struct memcg_shrinker_map *map; | |
447 | int nid; | |
448 | ||
449 | if (mem_cgroup_is_root(memcg)) | |
450 | return; | |
451 | ||
452 | for_each_node(nid) { | |
453 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
454 | map = rcu_dereference_protected(pn->shrinker_map, true); | |
8a260162 | 455 | kvfree(map); |
0a4465d3 KT |
456 | rcu_assign_pointer(pn->shrinker_map, NULL); |
457 | } | |
458 | } | |
459 | ||
460 | static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg) | |
461 | { | |
462 | struct memcg_shrinker_map *map; | |
463 | int nid, size, ret = 0; | |
464 | ||
465 | if (mem_cgroup_is_root(memcg)) | |
466 | return 0; | |
467 | ||
468 | mutex_lock(&memcg_shrinker_map_mutex); | |
469 | size = memcg_shrinker_map_size; | |
470 | for_each_node(nid) { | |
86daf94e | 471 | map = kvzalloc_node(sizeof(*map) + size, GFP_KERNEL, nid); |
0a4465d3 KT |
472 | if (!map) { |
473 | memcg_free_shrinker_maps(memcg); | |
474 | ret = -ENOMEM; | |
475 | break; | |
476 | } | |
477 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, map); | |
478 | } | |
479 | mutex_unlock(&memcg_shrinker_map_mutex); | |
480 | ||
481 | return ret; | |
482 | } | |
483 | ||
484 | int memcg_expand_shrinker_maps(int new_id) | |
485 | { | |
486 | int size, old_size, ret = 0; | |
487 | struct mem_cgroup *memcg; | |
488 | ||
489 | size = DIV_ROUND_UP(new_id + 1, BITS_PER_LONG) * sizeof(unsigned long); | |
490 | old_size = memcg_shrinker_map_size; | |
491 | if (size <= old_size) | |
492 | return 0; | |
493 | ||
494 | mutex_lock(&memcg_shrinker_map_mutex); | |
495 | if (!root_mem_cgroup) | |
496 | goto unlock; | |
497 | ||
498 | for_each_mem_cgroup(memcg) { | |
499 | if (mem_cgroup_is_root(memcg)) | |
500 | continue; | |
501 | ret = memcg_expand_one_shrinker_map(memcg, size, old_size); | |
75866af6 VA |
502 | if (ret) { |
503 | mem_cgroup_iter_break(NULL, memcg); | |
0a4465d3 | 504 | goto unlock; |
75866af6 | 505 | } |
0a4465d3 KT |
506 | } |
507 | unlock: | |
508 | if (!ret) | |
509 | memcg_shrinker_map_size = size; | |
510 | mutex_unlock(&memcg_shrinker_map_mutex); | |
511 | return ret; | |
512 | } | |
fae91d6d KT |
513 | |
514 | void memcg_set_shrinker_bit(struct mem_cgroup *memcg, int nid, int shrinker_id) | |
515 | { | |
516 | if (shrinker_id >= 0 && memcg && !mem_cgroup_is_root(memcg)) { | |
517 | struct memcg_shrinker_map *map; | |
518 | ||
519 | rcu_read_lock(); | |
520 | map = rcu_dereference(memcg->nodeinfo[nid]->shrinker_map); | |
f90280d6 KT |
521 | /* Pairs with smp mb in shrink_slab() */ |
522 | smp_mb__before_atomic(); | |
fae91d6d KT |
523 | set_bit(shrinker_id, map->map); |
524 | rcu_read_unlock(); | |
525 | } | |
526 | } | |
527 | ||
ad7fa852 TH |
528 | /** |
529 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
530 | * @page: page of interest | |
531 | * | |
532 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
533 | * with @page is returned. The returned css remains associated with @page | |
534 | * until it is released. | |
535 | * | |
536 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
537 | * is returned. | |
ad7fa852 TH |
538 | */ |
539 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
540 | { | |
541 | struct mem_cgroup *memcg; | |
542 | ||
bcfe06bf | 543 | memcg = page_memcg(page); |
ad7fa852 | 544 | |
9e10a130 | 545 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
546 | memcg = root_mem_cgroup; |
547 | ||
ad7fa852 TH |
548 | return &memcg->css; |
549 | } | |
550 | ||
2fc04524 VD |
551 | /** |
552 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
553 | * @page: the page | |
554 | * | |
555 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
556 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
557 | * is safe to call this function without holding a reference to @page. | |
558 | * | |
559 | * Note, this function is inherently racy, because there is nothing to prevent | |
560 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
561 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
562 | * do not care (such as procfs interfaces). | |
563 | */ | |
564 | ino_t page_cgroup_ino(struct page *page) | |
565 | { | |
566 | struct mem_cgroup *memcg; | |
567 | unsigned long ino = 0; | |
568 | ||
569 | rcu_read_lock(); | |
bcfe06bf | 570 | memcg = page_memcg_check(page); |
286e04b8 | 571 | |
2fc04524 VD |
572 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) |
573 | memcg = parent_mem_cgroup(memcg); | |
574 | if (memcg) | |
575 | ino = cgroup_ino(memcg->css.cgroup); | |
576 | rcu_read_unlock(); | |
577 | return ino; | |
578 | } | |
579 | ||
ef8f2327 MG |
580 | static struct mem_cgroup_per_node * |
581 | mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page) | |
f64c3f54 | 582 | { |
97a6c37b | 583 | int nid = page_to_nid(page); |
f64c3f54 | 584 | |
ef8f2327 | 585 | return memcg->nodeinfo[nid]; |
f64c3f54 BS |
586 | } |
587 | ||
ef8f2327 MG |
588 | static struct mem_cgroup_tree_per_node * |
589 | soft_limit_tree_node(int nid) | |
bb4cc1a8 | 590 | { |
ef8f2327 | 591 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
592 | } |
593 | ||
ef8f2327 | 594 | static struct mem_cgroup_tree_per_node * |
bb4cc1a8 AM |
595 | soft_limit_tree_from_page(struct page *page) |
596 | { | |
597 | int nid = page_to_nid(page); | |
bb4cc1a8 | 598 | |
ef8f2327 | 599 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
600 | } |
601 | ||
ef8f2327 MG |
602 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
603 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 604 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
605 | { |
606 | struct rb_node **p = &mctz->rb_root.rb_node; | |
607 | struct rb_node *parent = NULL; | |
ef8f2327 | 608 | struct mem_cgroup_per_node *mz_node; |
fa90b2fd | 609 | bool rightmost = true; |
bb4cc1a8 AM |
610 | |
611 | if (mz->on_tree) | |
612 | return; | |
613 | ||
614 | mz->usage_in_excess = new_usage_in_excess; | |
615 | if (!mz->usage_in_excess) | |
616 | return; | |
617 | while (*p) { | |
618 | parent = *p; | |
ef8f2327 | 619 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 | 620 | tree_node); |
fa90b2fd | 621 | if (mz->usage_in_excess < mz_node->usage_in_excess) { |
bb4cc1a8 | 622 | p = &(*p)->rb_left; |
fa90b2fd | 623 | rightmost = false; |
378876b0 | 624 | } else { |
bb4cc1a8 | 625 | p = &(*p)->rb_right; |
378876b0 | 626 | } |
bb4cc1a8 | 627 | } |
fa90b2fd DB |
628 | |
629 | if (rightmost) | |
630 | mctz->rb_rightmost = &mz->tree_node; | |
631 | ||
bb4cc1a8 AM |
632 | rb_link_node(&mz->tree_node, parent, p); |
633 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
634 | mz->on_tree = true; | |
635 | } | |
636 | ||
ef8f2327 MG |
637 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
638 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
639 | { |
640 | if (!mz->on_tree) | |
641 | return; | |
fa90b2fd DB |
642 | |
643 | if (&mz->tree_node == mctz->rb_rightmost) | |
644 | mctz->rb_rightmost = rb_prev(&mz->tree_node); | |
645 | ||
bb4cc1a8 AM |
646 | rb_erase(&mz->tree_node, &mctz->rb_root); |
647 | mz->on_tree = false; | |
648 | } | |
649 | ||
ef8f2327 MG |
650 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
651 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 652 | { |
0a31bc97 JW |
653 | unsigned long flags; |
654 | ||
655 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 656 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 657 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
658 | } |
659 | ||
3e32cb2e JW |
660 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
661 | { | |
662 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 663 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
664 | unsigned long excess = 0; |
665 | ||
666 | if (nr_pages > soft_limit) | |
667 | excess = nr_pages - soft_limit; | |
668 | ||
669 | return excess; | |
670 | } | |
bb4cc1a8 AM |
671 | |
672 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
673 | { | |
3e32cb2e | 674 | unsigned long excess; |
ef8f2327 MG |
675 | struct mem_cgroup_per_node *mz; |
676 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 677 | |
e231875b | 678 | mctz = soft_limit_tree_from_page(page); |
bfc7228b LD |
679 | if (!mctz) |
680 | return; | |
bb4cc1a8 AM |
681 | /* |
682 | * Necessary to update all ancestors when hierarchy is used. | |
683 | * because their event counter is not touched. | |
684 | */ | |
685 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
ef8f2327 | 686 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
3e32cb2e | 687 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
688 | /* |
689 | * We have to update the tree if mz is on RB-tree or | |
690 | * mem is over its softlimit. | |
691 | */ | |
692 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
693 | unsigned long flags; |
694 | ||
695 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
696 | /* if on-tree, remove it */ |
697 | if (mz->on_tree) | |
cf2c8127 | 698 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
699 | /* |
700 | * Insert again. mz->usage_in_excess will be updated. | |
701 | * If excess is 0, no tree ops. | |
702 | */ | |
cf2c8127 | 703 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 704 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
705 | } |
706 | } | |
707 | } | |
708 | ||
709 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
710 | { | |
ef8f2327 MG |
711 | struct mem_cgroup_tree_per_node *mctz; |
712 | struct mem_cgroup_per_node *mz; | |
713 | int nid; | |
bb4cc1a8 | 714 | |
e231875b | 715 | for_each_node(nid) { |
ef8f2327 MG |
716 | mz = mem_cgroup_nodeinfo(memcg, nid); |
717 | mctz = soft_limit_tree_node(nid); | |
bfc7228b LD |
718 | if (mctz) |
719 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
720 | } |
721 | } | |
722 | ||
ef8f2327 MG |
723 | static struct mem_cgroup_per_node * |
724 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 725 | { |
ef8f2327 | 726 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
727 | |
728 | retry: | |
729 | mz = NULL; | |
fa90b2fd | 730 | if (!mctz->rb_rightmost) |
bb4cc1a8 AM |
731 | goto done; /* Nothing to reclaim from */ |
732 | ||
fa90b2fd DB |
733 | mz = rb_entry(mctz->rb_rightmost, |
734 | struct mem_cgroup_per_node, tree_node); | |
bb4cc1a8 AM |
735 | /* |
736 | * Remove the node now but someone else can add it back, | |
737 | * we will to add it back at the end of reclaim to its correct | |
738 | * position in the tree. | |
739 | */ | |
cf2c8127 | 740 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 741 | if (!soft_limit_excess(mz->memcg) || |
8965aa28 | 742 | !css_tryget(&mz->memcg->css)) |
bb4cc1a8 AM |
743 | goto retry; |
744 | done: | |
745 | return mz; | |
746 | } | |
747 | ||
ef8f2327 MG |
748 | static struct mem_cgroup_per_node * |
749 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 750 | { |
ef8f2327 | 751 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 752 | |
0a31bc97 | 753 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 754 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 755 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
756 | return mz; |
757 | } | |
758 | ||
db9adbcb JW |
759 | /** |
760 | * __mod_memcg_state - update cgroup memory statistics | |
761 | * @memcg: the memory cgroup | |
762 | * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item | |
763 | * @val: delta to add to the counter, can be negative | |
764 | */ | |
765 | void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val) | |
766 | { | |
ea426c2a | 767 | long x, threshold = MEMCG_CHARGE_BATCH; |
db9adbcb JW |
768 | |
769 | if (mem_cgroup_disabled()) | |
770 | return; | |
771 | ||
772616b0 | 772 | if (memcg_stat_item_in_bytes(idx)) |
ea426c2a RG |
773 | threshold <<= PAGE_SHIFT; |
774 | ||
db9adbcb | 775 | x = val + __this_cpu_read(memcg->vmstats_percpu->stat[idx]); |
ea426c2a | 776 | if (unlikely(abs(x) > threshold)) { |
42a30035 JW |
777 | struct mem_cgroup *mi; |
778 | ||
766a4c19 YS |
779 | /* |
780 | * Batch local counters to keep them in sync with | |
781 | * the hierarchical ones. | |
782 | */ | |
783 | __this_cpu_add(memcg->vmstats_local->stat[idx], x); | |
42a30035 JW |
784 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
785 | atomic_long_add(x, &mi->vmstats[idx]); | |
db9adbcb JW |
786 | x = 0; |
787 | } | |
788 | __this_cpu_write(memcg->vmstats_percpu->stat[idx], x); | |
789 | } | |
790 | ||
42a30035 JW |
791 | static struct mem_cgroup_per_node * |
792 | parent_nodeinfo(struct mem_cgroup_per_node *pn, int nid) | |
793 | { | |
794 | struct mem_cgroup *parent; | |
795 | ||
796 | parent = parent_mem_cgroup(pn->memcg); | |
797 | if (!parent) | |
798 | return NULL; | |
799 | return mem_cgroup_nodeinfo(parent, nid); | |
800 | } | |
801 | ||
eedc4e5a RG |
802 | void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, |
803 | int val) | |
db9adbcb JW |
804 | { |
805 | struct mem_cgroup_per_node *pn; | |
42a30035 | 806 | struct mem_cgroup *memcg; |
ea426c2a | 807 | long x, threshold = MEMCG_CHARGE_BATCH; |
db9adbcb | 808 | |
db9adbcb | 809 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
42a30035 | 810 | memcg = pn->memcg; |
db9adbcb JW |
811 | |
812 | /* Update memcg */ | |
42a30035 | 813 | __mod_memcg_state(memcg, idx, val); |
db9adbcb | 814 | |
b4c46484 RG |
815 | /* Update lruvec */ |
816 | __this_cpu_add(pn->lruvec_stat_local->count[idx], val); | |
817 | ||
ea426c2a RG |
818 | if (vmstat_item_in_bytes(idx)) |
819 | threshold <<= PAGE_SHIFT; | |
820 | ||
db9adbcb | 821 | x = val + __this_cpu_read(pn->lruvec_stat_cpu->count[idx]); |
ea426c2a | 822 | if (unlikely(abs(x) > threshold)) { |
eedc4e5a | 823 | pg_data_t *pgdat = lruvec_pgdat(lruvec); |
42a30035 JW |
824 | struct mem_cgroup_per_node *pi; |
825 | ||
42a30035 JW |
826 | for (pi = pn; pi; pi = parent_nodeinfo(pi, pgdat->node_id)) |
827 | atomic_long_add(x, &pi->lruvec_stat[idx]); | |
db9adbcb JW |
828 | x = 0; |
829 | } | |
830 | __this_cpu_write(pn->lruvec_stat_cpu->count[idx], x); | |
831 | } | |
832 | ||
eedc4e5a RG |
833 | /** |
834 | * __mod_lruvec_state - update lruvec memory statistics | |
835 | * @lruvec: the lruvec | |
836 | * @idx: the stat item | |
837 | * @val: delta to add to the counter, can be negative | |
838 | * | |
839 | * The lruvec is the intersection of the NUMA node and a cgroup. This | |
840 | * function updates the all three counters that are affected by a | |
841 | * change of state at this level: per-node, per-cgroup, per-lruvec. | |
842 | */ | |
843 | void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, | |
844 | int val) | |
845 | { | |
846 | /* Update node */ | |
847 | __mod_node_page_state(lruvec_pgdat(lruvec), idx, val); | |
848 | ||
849 | /* Update memcg and lruvec */ | |
850 | if (!mem_cgroup_disabled()) | |
851 | __mod_memcg_lruvec_state(lruvec, idx, val); | |
852 | } | |
853 | ||
c47d5032 SB |
854 | void __mod_lruvec_page_state(struct page *page, enum node_stat_item idx, |
855 | int val) | |
856 | { | |
857 | struct page *head = compound_head(page); /* rmap on tail pages */ | |
d635a69d | 858 | struct mem_cgroup *memcg = page_memcg(head); |
c47d5032 SB |
859 | pg_data_t *pgdat = page_pgdat(page); |
860 | struct lruvec *lruvec; | |
861 | ||
862 | /* Untracked pages have no memcg, no lruvec. Update only the node */ | |
d635a69d | 863 | if (!memcg) { |
c47d5032 SB |
864 | __mod_node_page_state(pgdat, idx, val); |
865 | return; | |
866 | } | |
867 | ||
d635a69d | 868 | lruvec = mem_cgroup_lruvec(memcg, pgdat); |
c47d5032 SB |
869 | __mod_lruvec_state(lruvec, idx, val); |
870 | } | |
f0c0c115 | 871 | EXPORT_SYMBOL(__mod_lruvec_page_state); |
c47d5032 | 872 | |
da3ceeff | 873 | void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, int val) |
ec9f0238 | 874 | { |
4f103c63 | 875 | pg_data_t *pgdat = page_pgdat(virt_to_page(p)); |
ec9f0238 RG |
876 | struct mem_cgroup *memcg; |
877 | struct lruvec *lruvec; | |
878 | ||
879 | rcu_read_lock(); | |
4f103c63 | 880 | memcg = mem_cgroup_from_obj(p); |
ec9f0238 | 881 | |
8faeb1ff MS |
882 | /* |
883 | * Untracked pages have no memcg, no lruvec. Update only the | |
884 | * node. If we reparent the slab objects to the root memcg, | |
885 | * when we free the slab object, we need to update the per-memcg | |
886 | * vmstats to keep it correct for the root memcg. | |
887 | */ | |
888 | if (!memcg) { | |
ec9f0238 RG |
889 | __mod_node_page_state(pgdat, idx, val); |
890 | } else { | |
867e5e1d | 891 | lruvec = mem_cgroup_lruvec(memcg, pgdat); |
ec9f0238 RG |
892 | __mod_lruvec_state(lruvec, idx, val); |
893 | } | |
894 | rcu_read_unlock(); | |
895 | } | |
896 | ||
db9adbcb JW |
897 | /** |
898 | * __count_memcg_events - account VM events in a cgroup | |
899 | * @memcg: the memory cgroup | |
900 | * @idx: the event item | |
901 | * @count: the number of events that occured | |
902 | */ | |
903 | void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, | |
904 | unsigned long count) | |
905 | { | |
906 | unsigned long x; | |
907 | ||
908 | if (mem_cgroup_disabled()) | |
909 | return; | |
910 | ||
911 | x = count + __this_cpu_read(memcg->vmstats_percpu->events[idx]); | |
912 | if (unlikely(x > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
913 | struct mem_cgroup *mi; |
914 | ||
766a4c19 YS |
915 | /* |
916 | * Batch local counters to keep them in sync with | |
917 | * the hierarchical ones. | |
918 | */ | |
919 | __this_cpu_add(memcg->vmstats_local->events[idx], x); | |
42a30035 JW |
920 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
921 | atomic_long_add(x, &mi->vmevents[idx]); | |
db9adbcb JW |
922 | x = 0; |
923 | } | |
924 | __this_cpu_write(memcg->vmstats_percpu->events[idx], x); | |
925 | } | |
926 | ||
42a30035 | 927 | static unsigned long memcg_events(struct mem_cgroup *memcg, int event) |
e9f8974f | 928 | { |
871789d4 | 929 | return atomic_long_read(&memcg->vmevents[event]); |
e9f8974f JW |
930 | } |
931 | ||
42a30035 JW |
932 | static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event) |
933 | { | |
815744d7 JW |
934 | long x = 0; |
935 | int cpu; | |
936 | ||
937 | for_each_possible_cpu(cpu) | |
938 | x += per_cpu(memcg->vmstats_local->events[event], cpu); | |
939 | return x; | |
42a30035 JW |
940 | } |
941 | ||
c0ff4b85 | 942 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 943 | struct page *page, |
3fba69a5 | 944 | int nr_pages) |
d52aa412 | 945 | { |
e401f176 KH |
946 | /* pagein of a big page is an event. So, ignore page size */ |
947 | if (nr_pages > 0) | |
c9019e9b | 948 | __count_memcg_events(memcg, PGPGIN, 1); |
3751d604 | 949 | else { |
c9019e9b | 950 | __count_memcg_events(memcg, PGPGOUT, 1); |
3751d604 KH |
951 | nr_pages = -nr_pages; /* for event */ |
952 | } | |
e401f176 | 953 | |
871789d4 | 954 | __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages); |
6d12e2d8 KH |
955 | } |
956 | ||
f53d7ce3 JW |
957 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
958 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
959 | { |
960 | unsigned long val, next; | |
961 | ||
871789d4 CD |
962 | val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events); |
963 | next = __this_cpu_read(memcg->vmstats_percpu->targets[target]); | |
7a159cc9 | 964 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 965 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
966 | switch (target) { |
967 | case MEM_CGROUP_TARGET_THRESH: | |
968 | next = val + THRESHOLDS_EVENTS_TARGET; | |
969 | break; | |
bb4cc1a8 AM |
970 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
971 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
972 | break; | |
f53d7ce3 JW |
973 | default: |
974 | break; | |
975 | } | |
871789d4 | 976 | __this_cpu_write(memcg->vmstats_percpu->targets[target], next); |
f53d7ce3 | 977 | return true; |
7a159cc9 | 978 | } |
f53d7ce3 | 979 | return false; |
d2265e6f KH |
980 | } |
981 | ||
982 | /* | |
983 | * Check events in order. | |
984 | * | |
985 | */ | |
c0ff4b85 | 986 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
987 | { |
988 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
989 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
990 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 991 | bool do_softlimit; |
f53d7ce3 | 992 | |
bb4cc1a8 AM |
993 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
994 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
c0ff4b85 | 995 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
996 | if (unlikely(do_softlimit)) |
997 | mem_cgroup_update_tree(memcg, page); | |
0a31bc97 | 998 | } |
d2265e6f KH |
999 | } |
1000 | ||
cf475ad2 | 1001 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 1002 | { |
31a78f23 BS |
1003 | /* |
1004 | * mm_update_next_owner() may clear mm->owner to NULL | |
1005 | * if it races with swapoff, page migration, etc. | |
1006 | * So this can be called with p == NULL. | |
1007 | */ | |
1008 | if (unlikely(!p)) | |
1009 | return NULL; | |
1010 | ||
073219e9 | 1011 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 1012 | } |
33398cf2 | 1013 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 1014 | |
d46eb14b SB |
1015 | /** |
1016 | * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg. | |
1017 | * @mm: mm from which memcg should be extracted. It can be NULL. | |
1018 | * | |
1019 | * Obtain a reference on mm->memcg and returns it if successful. Otherwise | |
1020 | * root_mem_cgroup is returned. However if mem_cgroup is disabled, NULL is | |
1021 | * returned. | |
1022 | */ | |
1023 | struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) | |
54595fe2 | 1024 | { |
d46eb14b SB |
1025 | struct mem_cgroup *memcg; |
1026 | ||
1027 | if (mem_cgroup_disabled()) | |
1028 | return NULL; | |
0b7f569e | 1029 | |
54595fe2 KH |
1030 | rcu_read_lock(); |
1031 | do { | |
6f6acb00 MH |
1032 | /* |
1033 | * Page cache insertions can happen withou an | |
1034 | * actual mm context, e.g. during disk probing | |
1035 | * on boot, loopback IO, acct() writes etc. | |
1036 | */ | |
1037 | if (unlikely(!mm)) | |
df381975 | 1038 | memcg = root_mem_cgroup; |
6f6acb00 MH |
1039 | else { |
1040 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
1041 | if (unlikely(!memcg)) | |
1042 | memcg = root_mem_cgroup; | |
1043 | } | |
00d484f3 | 1044 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 1045 | rcu_read_unlock(); |
c0ff4b85 | 1046 | return memcg; |
54595fe2 | 1047 | } |
d46eb14b SB |
1048 | EXPORT_SYMBOL(get_mem_cgroup_from_mm); |
1049 | ||
37d5985c | 1050 | static __always_inline struct mem_cgroup *active_memcg(void) |
d46eb14b | 1051 | { |
37d5985c RG |
1052 | if (in_interrupt()) |
1053 | return this_cpu_read(int_active_memcg); | |
1054 | else | |
1055 | return current->active_memcg; | |
1056 | } | |
279c3393 | 1057 | |
37d5985c RG |
1058 | static __always_inline struct mem_cgroup *get_active_memcg(void) |
1059 | { | |
1060 | struct mem_cgroup *memcg; | |
d46eb14b | 1061 | |
37d5985c RG |
1062 | rcu_read_lock(); |
1063 | memcg = active_memcg(); | |
1685bde6 MS |
1064 | /* remote memcg must hold a ref. */ |
1065 | if (memcg && WARN_ON_ONCE(!css_tryget(&memcg->css))) | |
1066 | memcg = root_mem_cgroup; | |
37d5985c RG |
1067 | rcu_read_unlock(); |
1068 | ||
1069 | return memcg; | |
1070 | } | |
1071 | ||
4127c650 RG |
1072 | static __always_inline bool memcg_kmem_bypass(void) |
1073 | { | |
1074 | /* Allow remote memcg charging from any context. */ | |
1075 | if (unlikely(active_memcg())) | |
1076 | return false; | |
1077 | ||
1078 | /* Memcg to charge can't be determined. */ | |
1079 | if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD)) | |
1080 | return true; | |
1081 | ||
1082 | return false; | |
1083 | } | |
1084 | ||
37d5985c RG |
1085 | /** |
1086 | * If active memcg is set, do not fallback to current->mm->memcg. | |
1087 | */ | |
1088 | static __always_inline struct mem_cgroup *get_mem_cgroup_from_current(void) | |
1089 | { | |
1090 | if (memcg_kmem_bypass()) | |
1091 | return NULL; | |
1092 | ||
1093 | if (unlikely(active_memcg())) | |
1094 | return get_active_memcg(); | |
1095 | ||
d46eb14b SB |
1096 | return get_mem_cgroup_from_mm(current->mm); |
1097 | } | |
54595fe2 | 1098 | |
5660048c JW |
1099 | /** |
1100 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1101 | * @root: hierarchy root | |
1102 | * @prev: previously returned memcg, NULL on first invocation | |
1103 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1104 | * | |
1105 | * Returns references to children of the hierarchy below @root, or | |
1106 | * @root itself, or %NULL after a full round-trip. | |
1107 | * | |
1108 | * Caller must pass the return value in @prev on subsequent | |
1109 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1110 | * to cancel a hierarchy walk before the round-trip is complete. | |
1111 | * | |
05bdc520 ML |
1112 | * Reclaimers can specify a node in @reclaim to divide up the memcgs |
1113 | * in the hierarchy among all concurrent reclaimers operating on the | |
1114 | * same node. | |
5660048c | 1115 | */ |
694fbc0f | 1116 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1117 | struct mem_cgroup *prev, |
694fbc0f | 1118 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1119 | { |
3f649ab7 | 1120 | struct mem_cgroup_reclaim_iter *iter; |
5ac8fb31 | 1121 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 1122 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1123 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1124 | |
694fbc0f AM |
1125 | if (mem_cgroup_disabled()) |
1126 | return NULL; | |
5660048c | 1127 | |
9f3a0d09 JW |
1128 | if (!root) |
1129 | root = root_mem_cgroup; | |
7d74b06f | 1130 | |
9f3a0d09 | 1131 | if (prev && !reclaim) |
5ac8fb31 | 1132 | pos = prev; |
14067bb3 | 1133 | |
542f85f9 | 1134 | rcu_read_lock(); |
5f578161 | 1135 | |
5ac8fb31 | 1136 | if (reclaim) { |
ef8f2327 | 1137 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 1138 | |
ef8f2327 | 1139 | mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id); |
9da83f3f | 1140 | iter = &mz->iter; |
5ac8fb31 JW |
1141 | |
1142 | if (prev && reclaim->generation != iter->generation) | |
1143 | goto out_unlock; | |
1144 | ||
6df38689 | 1145 | while (1) { |
4db0c3c2 | 1146 | pos = READ_ONCE(iter->position); |
6df38689 VD |
1147 | if (!pos || css_tryget(&pos->css)) |
1148 | break; | |
5ac8fb31 | 1149 | /* |
6df38689 VD |
1150 | * css reference reached zero, so iter->position will |
1151 | * be cleared by ->css_released. However, we should not | |
1152 | * rely on this happening soon, because ->css_released | |
1153 | * is called from a work queue, and by busy-waiting we | |
1154 | * might block it. So we clear iter->position right | |
1155 | * away. | |
5ac8fb31 | 1156 | */ |
6df38689 VD |
1157 | (void)cmpxchg(&iter->position, pos, NULL); |
1158 | } | |
5ac8fb31 JW |
1159 | } |
1160 | ||
1161 | if (pos) | |
1162 | css = &pos->css; | |
1163 | ||
1164 | for (;;) { | |
1165 | css = css_next_descendant_pre(css, &root->css); | |
1166 | if (!css) { | |
1167 | /* | |
1168 | * Reclaimers share the hierarchy walk, and a | |
1169 | * new one might jump in right at the end of | |
1170 | * the hierarchy - make sure they see at least | |
1171 | * one group and restart from the beginning. | |
1172 | */ | |
1173 | if (!prev) | |
1174 | continue; | |
1175 | break; | |
527a5ec9 | 1176 | } |
7d74b06f | 1177 | |
5ac8fb31 JW |
1178 | /* |
1179 | * Verify the css and acquire a reference. The root | |
1180 | * is provided by the caller, so we know it's alive | |
1181 | * and kicking, and don't take an extra reference. | |
1182 | */ | |
1183 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 1184 | |
5ac8fb31 JW |
1185 | if (css == &root->css) |
1186 | break; | |
14067bb3 | 1187 | |
0b8f73e1 JW |
1188 | if (css_tryget(css)) |
1189 | break; | |
9f3a0d09 | 1190 | |
5ac8fb31 | 1191 | memcg = NULL; |
9f3a0d09 | 1192 | } |
5ac8fb31 JW |
1193 | |
1194 | if (reclaim) { | |
5ac8fb31 | 1195 | /* |
6df38689 VD |
1196 | * The position could have already been updated by a competing |
1197 | * thread, so check that the value hasn't changed since we read | |
1198 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 1199 | */ |
6df38689 VD |
1200 | (void)cmpxchg(&iter->position, pos, memcg); |
1201 | ||
5ac8fb31 JW |
1202 | if (pos) |
1203 | css_put(&pos->css); | |
1204 | ||
1205 | if (!memcg) | |
1206 | iter->generation++; | |
1207 | else if (!prev) | |
1208 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1209 | } |
5ac8fb31 | 1210 | |
542f85f9 MH |
1211 | out_unlock: |
1212 | rcu_read_unlock(); | |
c40046f3 MH |
1213 | if (prev && prev != root) |
1214 | css_put(&prev->css); | |
1215 | ||
9f3a0d09 | 1216 | return memcg; |
14067bb3 | 1217 | } |
7d74b06f | 1218 | |
5660048c JW |
1219 | /** |
1220 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1221 | * @root: hierarchy root | |
1222 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1223 | */ | |
1224 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1225 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1226 | { |
1227 | if (!root) | |
1228 | root = root_mem_cgroup; | |
1229 | if (prev && prev != root) | |
1230 | css_put(&prev->css); | |
1231 | } | |
7d74b06f | 1232 | |
54a83d6b MC |
1233 | static void __invalidate_reclaim_iterators(struct mem_cgroup *from, |
1234 | struct mem_cgroup *dead_memcg) | |
6df38689 | 1235 | { |
6df38689 | 1236 | struct mem_cgroup_reclaim_iter *iter; |
ef8f2327 MG |
1237 | struct mem_cgroup_per_node *mz; |
1238 | int nid; | |
6df38689 | 1239 | |
54a83d6b MC |
1240 | for_each_node(nid) { |
1241 | mz = mem_cgroup_nodeinfo(from, nid); | |
9da83f3f YS |
1242 | iter = &mz->iter; |
1243 | cmpxchg(&iter->position, dead_memcg, NULL); | |
6df38689 VD |
1244 | } |
1245 | } | |
1246 | ||
54a83d6b MC |
1247 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1248 | { | |
1249 | struct mem_cgroup *memcg = dead_memcg; | |
1250 | struct mem_cgroup *last; | |
1251 | ||
1252 | do { | |
1253 | __invalidate_reclaim_iterators(memcg, dead_memcg); | |
1254 | last = memcg; | |
1255 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1256 | ||
1257 | /* | |
1258 | * When cgruop1 non-hierarchy mode is used, | |
1259 | * parent_mem_cgroup() does not walk all the way up to the | |
1260 | * cgroup root (root_mem_cgroup). So we have to handle | |
1261 | * dead_memcg from cgroup root separately. | |
1262 | */ | |
1263 | if (last != root_mem_cgroup) | |
1264 | __invalidate_reclaim_iterators(root_mem_cgroup, | |
1265 | dead_memcg); | |
1266 | } | |
1267 | ||
7c5f64f8 VD |
1268 | /** |
1269 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1270 | * @memcg: hierarchy root | |
1271 | * @fn: function to call for each task | |
1272 | * @arg: argument passed to @fn | |
1273 | * | |
1274 | * This function iterates over tasks attached to @memcg or to any of its | |
1275 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
1276 | * value, the function breaks the iteration loop and returns the value. | |
1277 | * Otherwise, it will iterate over all tasks and return 0. | |
1278 | * | |
1279 | * This function must not be called for the root memory cgroup. | |
1280 | */ | |
1281 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
1282 | int (*fn)(struct task_struct *, void *), void *arg) | |
1283 | { | |
1284 | struct mem_cgroup *iter; | |
1285 | int ret = 0; | |
1286 | ||
1287 | BUG_ON(memcg == root_mem_cgroup); | |
1288 | ||
1289 | for_each_mem_cgroup_tree(iter, memcg) { | |
1290 | struct css_task_iter it; | |
1291 | struct task_struct *task; | |
1292 | ||
f168a9a5 | 1293 | css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it); |
7c5f64f8 VD |
1294 | while (!ret && (task = css_task_iter_next(&it))) |
1295 | ret = fn(task, arg); | |
1296 | css_task_iter_end(&it); | |
1297 | if (ret) { | |
1298 | mem_cgroup_iter_break(memcg, iter); | |
1299 | break; | |
1300 | } | |
1301 | } | |
1302 | return ret; | |
1303 | } | |
1304 | ||
6168d0da AS |
1305 | #ifdef CONFIG_DEBUG_VM |
1306 | void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page) | |
1307 | { | |
1308 | struct mem_cgroup *memcg; | |
1309 | ||
1310 | if (mem_cgroup_disabled()) | |
1311 | return; | |
1312 | ||
1313 | memcg = page_memcg(page); | |
1314 | ||
1315 | if (!memcg) | |
1316 | VM_BUG_ON_PAGE(lruvec_memcg(lruvec) != root_mem_cgroup, page); | |
1317 | else | |
1318 | VM_BUG_ON_PAGE(lruvec_memcg(lruvec) != memcg, page); | |
1319 | } | |
1320 | #endif | |
1321 | ||
6168d0da AS |
1322 | /** |
1323 | * lock_page_lruvec - lock and return lruvec for a given page. | |
1324 | * @page: the page | |
1325 | * | |
d7e3aba5 AS |
1326 | * These functions are safe to use under any of the following conditions: |
1327 | * - page locked | |
1328 | * - PageLRU cleared | |
1329 | * - lock_page_memcg() | |
1330 | * - page->_refcount is zero | |
6168d0da AS |
1331 | */ |
1332 | struct lruvec *lock_page_lruvec(struct page *page) | |
1333 | { | |
1334 | struct lruvec *lruvec; | |
1335 | struct pglist_data *pgdat = page_pgdat(page); | |
1336 | ||
6168d0da AS |
1337 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
1338 | spin_lock(&lruvec->lru_lock); | |
6168d0da AS |
1339 | |
1340 | lruvec_memcg_debug(lruvec, page); | |
1341 | ||
1342 | return lruvec; | |
1343 | } | |
1344 | ||
1345 | struct lruvec *lock_page_lruvec_irq(struct page *page) | |
1346 | { | |
1347 | struct lruvec *lruvec; | |
1348 | struct pglist_data *pgdat = page_pgdat(page); | |
1349 | ||
6168d0da AS |
1350 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
1351 | spin_lock_irq(&lruvec->lru_lock); | |
6168d0da AS |
1352 | |
1353 | lruvec_memcg_debug(lruvec, page); | |
1354 | ||
1355 | return lruvec; | |
1356 | } | |
1357 | ||
1358 | struct lruvec *lock_page_lruvec_irqsave(struct page *page, unsigned long *flags) | |
1359 | { | |
1360 | struct lruvec *lruvec; | |
1361 | struct pglist_data *pgdat = page_pgdat(page); | |
1362 | ||
6168d0da AS |
1363 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
1364 | spin_lock_irqsave(&lruvec->lru_lock, *flags); | |
6168d0da AS |
1365 | |
1366 | lruvec_memcg_debug(lruvec, page); | |
1367 | ||
1368 | return lruvec; | |
1369 | } | |
1370 | ||
925b7673 | 1371 | /** |
fa9add64 HD |
1372 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1373 | * @lruvec: mem_cgroup per zone lru vector | |
1374 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1375 | * @zid: zone id of the accounted pages |
fa9add64 | 1376 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1377 | * |
ca707239 HD |
1378 | * This function must be called under lru_lock, just before a page is added |
1379 | * to or just after a page is removed from an lru list (that ordering being | |
1380 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 1381 | */ |
fa9add64 | 1382 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1383 | int zid, int nr_pages) |
3f58a829 | 1384 | { |
ef8f2327 | 1385 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1386 | unsigned long *lru_size; |
ca707239 | 1387 | long size; |
3f58a829 MK |
1388 | |
1389 | if (mem_cgroup_disabled()) | |
1390 | return; | |
1391 | ||
ef8f2327 | 1392 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1393 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1394 | |
1395 | if (nr_pages < 0) | |
1396 | *lru_size += nr_pages; | |
1397 | ||
1398 | size = *lru_size; | |
b4536f0c MH |
1399 | if (WARN_ONCE(size < 0, |
1400 | "%s(%p, %d, %d): lru_size %ld\n", | |
1401 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1402 | VM_BUG_ON(1); |
1403 | *lru_size = 0; | |
1404 | } | |
1405 | ||
1406 | if (nr_pages > 0) | |
1407 | *lru_size += nr_pages; | |
08e552c6 | 1408 | } |
544122e5 | 1409 | |
19942822 | 1410 | /** |
9d11ea9f | 1411 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1412 | * @memcg: the memory cgroup |
19942822 | 1413 | * |
9d11ea9f | 1414 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1415 | * pages. |
19942822 | 1416 | */ |
c0ff4b85 | 1417 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1418 | { |
3e32cb2e JW |
1419 | unsigned long margin = 0; |
1420 | unsigned long count; | |
1421 | unsigned long limit; | |
9d11ea9f | 1422 | |
3e32cb2e | 1423 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1424 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1425 | if (count < limit) |
1426 | margin = limit - count; | |
1427 | ||
7941d214 | 1428 | if (do_memsw_account()) { |
3e32cb2e | 1429 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1430 | limit = READ_ONCE(memcg->memsw.max); |
1c4448ed | 1431 | if (count < limit) |
3e32cb2e | 1432 | margin = min(margin, limit - count); |
cbedbac3 LR |
1433 | else |
1434 | margin = 0; | |
3e32cb2e JW |
1435 | } |
1436 | ||
1437 | return margin; | |
19942822 JW |
1438 | } |
1439 | ||
32047e2a | 1440 | /* |
bdcbb659 | 1441 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1442 | * |
bdcbb659 QH |
1443 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1444 | * moving cgroups. This is for waiting at high-memory pressure | |
1445 | * caused by "move". | |
32047e2a | 1446 | */ |
c0ff4b85 | 1447 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1448 | { |
2bd9bb20 KH |
1449 | struct mem_cgroup *from; |
1450 | struct mem_cgroup *to; | |
4b534334 | 1451 | bool ret = false; |
2bd9bb20 KH |
1452 | /* |
1453 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1454 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1455 | */ | |
1456 | spin_lock(&mc.lock); | |
1457 | from = mc.from; | |
1458 | to = mc.to; | |
1459 | if (!from) | |
1460 | goto unlock; | |
3e92041d | 1461 | |
2314b42d JW |
1462 | ret = mem_cgroup_is_descendant(from, memcg) || |
1463 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1464 | unlock: |
1465 | spin_unlock(&mc.lock); | |
4b534334 KH |
1466 | return ret; |
1467 | } | |
1468 | ||
c0ff4b85 | 1469 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1470 | { |
1471 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1472 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1473 | DEFINE_WAIT(wait); |
1474 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1475 | /* moving charge context might have finished. */ | |
1476 | if (mc.moving_task) | |
1477 | schedule(); | |
1478 | finish_wait(&mc.waitq, &wait); | |
1479 | return true; | |
1480 | } | |
1481 | } | |
1482 | return false; | |
1483 | } | |
1484 | ||
5f9a4f4a MS |
1485 | struct memory_stat { |
1486 | const char *name; | |
5f9a4f4a MS |
1487 | unsigned int idx; |
1488 | }; | |
1489 | ||
57b2847d | 1490 | static const struct memory_stat memory_stats[] = { |
fff66b79 MS |
1491 | { "anon", NR_ANON_MAPPED }, |
1492 | { "file", NR_FILE_PAGES }, | |
1493 | { "kernel_stack", NR_KERNEL_STACK_KB }, | |
1494 | { "pagetables", NR_PAGETABLE }, | |
1495 | { "percpu", MEMCG_PERCPU_B }, | |
1496 | { "sock", MEMCG_SOCK }, | |
1497 | { "shmem", NR_SHMEM }, | |
1498 | { "file_mapped", NR_FILE_MAPPED }, | |
1499 | { "file_dirty", NR_FILE_DIRTY }, | |
1500 | { "file_writeback", NR_WRITEBACK }, | |
b6038942 SB |
1501 | #ifdef CONFIG_SWAP |
1502 | { "swapcached", NR_SWAPCACHE }, | |
1503 | #endif | |
5f9a4f4a | 1504 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
fff66b79 MS |
1505 | { "anon_thp", NR_ANON_THPS }, |
1506 | { "file_thp", NR_FILE_THPS }, | |
1507 | { "shmem_thp", NR_SHMEM_THPS }, | |
5f9a4f4a | 1508 | #endif |
fff66b79 MS |
1509 | { "inactive_anon", NR_INACTIVE_ANON }, |
1510 | { "active_anon", NR_ACTIVE_ANON }, | |
1511 | { "inactive_file", NR_INACTIVE_FILE }, | |
1512 | { "active_file", NR_ACTIVE_FILE }, | |
1513 | { "unevictable", NR_UNEVICTABLE }, | |
1514 | { "slab_reclaimable", NR_SLAB_RECLAIMABLE_B }, | |
1515 | { "slab_unreclaimable", NR_SLAB_UNRECLAIMABLE_B }, | |
5f9a4f4a MS |
1516 | |
1517 | /* The memory events */ | |
fff66b79 MS |
1518 | { "workingset_refault_anon", WORKINGSET_REFAULT_ANON }, |
1519 | { "workingset_refault_file", WORKINGSET_REFAULT_FILE }, | |
1520 | { "workingset_activate_anon", WORKINGSET_ACTIVATE_ANON }, | |
1521 | { "workingset_activate_file", WORKINGSET_ACTIVATE_FILE }, | |
1522 | { "workingset_restore_anon", WORKINGSET_RESTORE_ANON }, | |
1523 | { "workingset_restore_file", WORKINGSET_RESTORE_FILE }, | |
1524 | { "workingset_nodereclaim", WORKINGSET_NODERECLAIM }, | |
5f9a4f4a MS |
1525 | }; |
1526 | ||
fff66b79 MS |
1527 | /* Translate stat items to the correct unit for memory.stat output */ |
1528 | static int memcg_page_state_unit(int item) | |
1529 | { | |
1530 | switch (item) { | |
1531 | case MEMCG_PERCPU_B: | |
1532 | case NR_SLAB_RECLAIMABLE_B: | |
1533 | case NR_SLAB_UNRECLAIMABLE_B: | |
1534 | case WORKINGSET_REFAULT_ANON: | |
1535 | case WORKINGSET_REFAULT_FILE: | |
1536 | case WORKINGSET_ACTIVATE_ANON: | |
1537 | case WORKINGSET_ACTIVATE_FILE: | |
1538 | case WORKINGSET_RESTORE_ANON: | |
1539 | case WORKINGSET_RESTORE_FILE: | |
1540 | case WORKINGSET_NODERECLAIM: | |
1541 | return 1; | |
1542 | case NR_KERNEL_STACK_KB: | |
1543 | return SZ_1K; | |
1544 | default: | |
1545 | return PAGE_SIZE; | |
1546 | } | |
1547 | } | |
1548 | ||
1549 | static inline unsigned long memcg_page_state_output(struct mem_cgroup *memcg, | |
1550 | int item) | |
1551 | { | |
1552 | return memcg_page_state(memcg, item) * memcg_page_state_unit(item); | |
1553 | } | |
1554 | ||
c8713d0b JW |
1555 | static char *memory_stat_format(struct mem_cgroup *memcg) |
1556 | { | |
1557 | struct seq_buf s; | |
1558 | int i; | |
71cd3113 | 1559 | |
c8713d0b JW |
1560 | seq_buf_init(&s, kmalloc(PAGE_SIZE, GFP_KERNEL), PAGE_SIZE); |
1561 | if (!s.buffer) | |
1562 | return NULL; | |
1563 | ||
1564 | /* | |
1565 | * Provide statistics on the state of the memory subsystem as | |
1566 | * well as cumulative event counters that show past behavior. | |
1567 | * | |
1568 | * This list is ordered following a combination of these gradients: | |
1569 | * 1) generic big picture -> specifics and details | |
1570 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
1571 | * | |
1572 | * Current memory state: | |
1573 | */ | |
1574 | ||
5f9a4f4a MS |
1575 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { |
1576 | u64 size; | |
c8713d0b | 1577 | |
fff66b79 | 1578 | size = memcg_page_state_output(memcg, memory_stats[i].idx); |
5f9a4f4a | 1579 | seq_buf_printf(&s, "%s %llu\n", memory_stats[i].name, size); |
c8713d0b | 1580 | |
5f9a4f4a | 1581 | if (unlikely(memory_stats[i].idx == NR_SLAB_UNRECLAIMABLE_B)) { |
fff66b79 MS |
1582 | size += memcg_page_state_output(memcg, |
1583 | NR_SLAB_RECLAIMABLE_B); | |
5f9a4f4a MS |
1584 | seq_buf_printf(&s, "slab %llu\n", size); |
1585 | } | |
1586 | } | |
c8713d0b JW |
1587 | |
1588 | /* Accumulated memory events */ | |
1589 | ||
ebc5d83d KK |
1590 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGFAULT), |
1591 | memcg_events(memcg, PGFAULT)); | |
1592 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGMAJFAULT), | |
1593 | memcg_events(memcg, PGMAJFAULT)); | |
ebc5d83d KK |
1594 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGREFILL), |
1595 | memcg_events(memcg, PGREFILL)); | |
c8713d0b JW |
1596 | seq_buf_printf(&s, "pgscan %lu\n", |
1597 | memcg_events(memcg, PGSCAN_KSWAPD) + | |
1598 | memcg_events(memcg, PGSCAN_DIRECT)); | |
1599 | seq_buf_printf(&s, "pgsteal %lu\n", | |
1600 | memcg_events(memcg, PGSTEAL_KSWAPD) + | |
1601 | memcg_events(memcg, PGSTEAL_DIRECT)); | |
ebc5d83d KK |
1602 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGACTIVATE), |
1603 | memcg_events(memcg, PGACTIVATE)); | |
1604 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGDEACTIVATE), | |
1605 | memcg_events(memcg, PGDEACTIVATE)); | |
1606 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREE), | |
1607 | memcg_events(memcg, PGLAZYFREE)); | |
1608 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREED), | |
1609 | memcg_events(memcg, PGLAZYFREED)); | |
c8713d0b JW |
1610 | |
1611 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
ebc5d83d | 1612 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_FAULT_ALLOC), |
c8713d0b | 1613 | memcg_events(memcg, THP_FAULT_ALLOC)); |
ebc5d83d | 1614 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_COLLAPSE_ALLOC), |
c8713d0b JW |
1615 | memcg_events(memcg, THP_COLLAPSE_ALLOC)); |
1616 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ | |
1617 | ||
1618 | /* The above should easily fit into one page */ | |
1619 | WARN_ON_ONCE(seq_buf_has_overflowed(&s)); | |
1620 | ||
1621 | return s.buffer; | |
1622 | } | |
71cd3113 | 1623 | |
58cf188e | 1624 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1625 | /** |
f0c867d9 | 1626 | * mem_cgroup_print_oom_context: Print OOM information relevant to |
1627 | * memory controller. | |
e222432b BS |
1628 | * @memcg: The memory cgroup that went over limit |
1629 | * @p: Task that is going to be killed | |
1630 | * | |
1631 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1632 | * enabled | |
1633 | */ | |
f0c867d9 | 1634 | void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) |
e222432b | 1635 | { |
e222432b BS |
1636 | rcu_read_lock(); |
1637 | ||
f0c867d9 | 1638 | if (memcg) { |
1639 | pr_cont(",oom_memcg="); | |
1640 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1641 | } else | |
1642 | pr_cont(",global_oom"); | |
2415b9f5 | 1643 | if (p) { |
f0c867d9 | 1644 | pr_cont(",task_memcg="); |
2415b9f5 | 1645 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
2415b9f5 | 1646 | } |
e222432b | 1647 | rcu_read_unlock(); |
f0c867d9 | 1648 | } |
1649 | ||
1650 | /** | |
1651 | * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to | |
1652 | * memory controller. | |
1653 | * @memcg: The memory cgroup that went over limit | |
1654 | */ | |
1655 | void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) | |
1656 | { | |
c8713d0b | 1657 | char *buf; |
e222432b | 1658 | |
3e32cb2e JW |
1659 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1660 | K((u64)page_counter_read(&memcg->memory)), | |
15b42562 | 1661 | K((u64)READ_ONCE(memcg->memory.max)), memcg->memory.failcnt); |
c8713d0b JW |
1662 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
1663 | pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1664 | K((u64)page_counter_read(&memcg->swap)), | |
32d087cd | 1665 | K((u64)READ_ONCE(memcg->swap.max)), memcg->swap.failcnt); |
c8713d0b JW |
1666 | else { |
1667 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1668 | K((u64)page_counter_read(&memcg->memsw)), | |
1669 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); | |
1670 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1671 | K((u64)page_counter_read(&memcg->kmem)), | |
1672 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); | |
58cf188e | 1673 | } |
c8713d0b JW |
1674 | |
1675 | pr_info("Memory cgroup stats for "); | |
1676 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1677 | pr_cont(":"); | |
1678 | buf = memory_stat_format(memcg); | |
1679 | if (!buf) | |
1680 | return; | |
1681 | pr_info("%s", buf); | |
1682 | kfree(buf); | |
e222432b BS |
1683 | } |
1684 | ||
a63d83f4 DR |
1685 | /* |
1686 | * Return the memory (and swap, if configured) limit for a memcg. | |
1687 | */ | |
bbec2e15 | 1688 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1689 | { |
8d387a5f WL |
1690 | unsigned long max = READ_ONCE(memcg->memory.max); |
1691 | ||
1692 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) { | |
1693 | if (mem_cgroup_swappiness(memcg)) | |
1694 | max += min(READ_ONCE(memcg->swap.max), | |
1695 | (unsigned long)total_swap_pages); | |
1696 | } else { /* v1 */ | |
1697 | if (mem_cgroup_swappiness(memcg)) { | |
1698 | /* Calculate swap excess capacity from memsw limit */ | |
1699 | unsigned long swap = READ_ONCE(memcg->memsw.max) - max; | |
1700 | ||
1701 | max += min(swap, (unsigned long)total_swap_pages); | |
1702 | } | |
9a5a8f19 | 1703 | } |
bbec2e15 | 1704 | return max; |
a63d83f4 DR |
1705 | } |
1706 | ||
9783aa99 CD |
1707 | unsigned long mem_cgroup_size(struct mem_cgroup *memcg) |
1708 | { | |
1709 | return page_counter_read(&memcg->memory); | |
1710 | } | |
1711 | ||
b6e6edcf | 1712 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1713 | int order) |
9cbb78bb | 1714 | { |
6e0fc46d DR |
1715 | struct oom_control oc = { |
1716 | .zonelist = NULL, | |
1717 | .nodemask = NULL, | |
2a966b77 | 1718 | .memcg = memcg, |
6e0fc46d DR |
1719 | .gfp_mask = gfp_mask, |
1720 | .order = order, | |
6e0fc46d | 1721 | }; |
1378b37d | 1722 | bool ret = true; |
9cbb78bb | 1723 | |
7775face TH |
1724 | if (mutex_lock_killable(&oom_lock)) |
1725 | return true; | |
1378b37d YS |
1726 | |
1727 | if (mem_cgroup_margin(memcg) >= (1 << order)) | |
1728 | goto unlock; | |
1729 | ||
7775face TH |
1730 | /* |
1731 | * A few threads which were not waiting at mutex_lock_killable() can | |
1732 | * fail to bail out. Therefore, check again after holding oom_lock. | |
1733 | */ | |
1734 | ret = should_force_charge() || out_of_memory(&oc); | |
1378b37d YS |
1735 | |
1736 | unlock: | |
dc56401f | 1737 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1738 | return ret; |
9cbb78bb DR |
1739 | } |
1740 | ||
0608f43d | 1741 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1742 | pg_data_t *pgdat, |
0608f43d AM |
1743 | gfp_t gfp_mask, |
1744 | unsigned long *total_scanned) | |
1745 | { | |
1746 | struct mem_cgroup *victim = NULL; | |
1747 | int total = 0; | |
1748 | int loop = 0; | |
1749 | unsigned long excess; | |
1750 | unsigned long nr_scanned; | |
1751 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1752 | .pgdat = pgdat, |
0608f43d AM |
1753 | }; |
1754 | ||
3e32cb2e | 1755 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1756 | |
1757 | while (1) { | |
1758 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1759 | if (!victim) { | |
1760 | loop++; | |
1761 | if (loop >= 2) { | |
1762 | /* | |
1763 | * If we have not been able to reclaim | |
1764 | * anything, it might because there are | |
1765 | * no reclaimable pages under this hierarchy | |
1766 | */ | |
1767 | if (!total) | |
1768 | break; | |
1769 | /* | |
1770 | * We want to do more targeted reclaim. | |
1771 | * excess >> 2 is not to excessive so as to | |
1772 | * reclaim too much, nor too less that we keep | |
1773 | * coming back to reclaim from this cgroup | |
1774 | */ | |
1775 | if (total >= (excess >> 2) || | |
1776 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1777 | break; | |
1778 | } | |
1779 | continue; | |
1780 | } | |
a9dd0a83 | 1781 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1782 | pgdat, &nr_scanned); |
0608f43d | 1783 | *total_scanned += nr_scanned; |
3e32cb2e | 1784 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1785 | break; |
6d61ef40 | 1786 | } |
0608f43d AM |
1787 | mem_cgroup_iter_break(root_memcg, victim); |
1788 | return total; | |
6d61ef40 BS |
1789 | } |
1790 | ||
0056f4e6 JW |
1791 | #ifdef CONFIG_LOCKDEP |
1792 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1793 | .name = "memcg_oom_lock", | |
1794 | }; | |
1795 | #endif | |
1796 | ||
fb2a6fc5 JW |
1797 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1798 | ||
867578cb KH |
1799 | /* |
1800 | * Check OOM-Killer is already running under our hierarchy. | |
1801 | * If someone is running, return false. | |
1802 | */ | |
fb2a6fc5 | 1803 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1804 | { |
79dfdacc | 1805 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1806 | |
fb2a6fc5 JW |
1807 | spin_lock(&memcg_oom_lock); |
1808 | ||
9f3a0d09 | 1809 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1810 | if (iter->oom_lock) { |
79dfdacc MH |
1811 | /* |
1812 | * this subtree of our hierarchy is already locked | |
1813 | * so we cannot give a lock. | |
1814 | */ | |
79dfdacc | 1815 | failed = iter; |
9f3a0d09 JW |
1816 | mem_cgroup_iter_break(memcg, iter); |
1817 | break; | |
23751be0 JW |
1818 | } else |
1819 | iter->oom_lock = true; | |
7d74b06f | 1820 | } |
867578cb | 1821 | |
fb2a6fc5 JW |
1822 | if (failed) { |
1823 | /* | |
1824 | * OK, we failed to lock the whole subtree so we have | |
1825 | * to clean up what we set up to the failing subtree | |
1826 | */ | |
1827 | for_each_mem_cgroup_tree(iter, memcg) { | |
1828 | if (iter == failed) { | |
1829 | mem_cgroup_iter_break(memcg, iter); | |
1830 | break; | |
1831 | } | |
1832 | iter->oom_lock = false; | |
79dfdacc | 1833 | } |
0056f4e6 JW |
1834 | } else |
1835 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1836 | |
1837 | spin_unlock(&memcg_oom_lock); | |
1838 | ||
1839 | return !failed; | |
a636b327 | 1840 | } |
0b7f569e | 1841 | |
fb2a6fc5 | 1842 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1843 | { |
7d74b06f KH |
1844 | struct mem_cgroup *iter; |
1845 | ||
fb2a6fc5 | 1846 | spin_lock(&memcg_oom_lock); |
5facae4f | 1847 | mutex_release(&memcg_oom_lock_dep_map, _RET_IP_); |
c0ff4b85 | 1848 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1849 | iter->oom_lock = false; |
fb2a6fc5 | 1850 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1851 | } |
1852 | ||
c0ff4b85 | 1853 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1854 | { |
1855 | struct mem_cgroup *iter; | |
1856 | ||
c2b42d3c | 1857 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1858 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1859 | iter->under_oom++; |
1860 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1861 | } |
1862 | ||
c0ff4b85 | 1863 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1864 | { |
1865 | struct mem_cgroup *iter; | |
1866 | ||
867578cb | 1867 | /* |
7a52d4d8 ML |
1868 | * Be careful about under_oom underflows becase a child memcg |
1869 | * could have been added after mem_cgroup_mark_under_oom. | |
867578cb | 1870 | */ |
c2b42d3c | 1871 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1872 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1873 | if (iter->under_oom > 0) |
1874 | iter->under_oom--; | |
1875 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1876 | } |
1877 | ||
867578cb KH |
1878 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1879 | ||
dc98df5a | 1880 | struct oom_wait_info { |
d79154bb | 1881 | struct mem_cgroup *memcg; |
ac6424b9 | 1882 | wait_queue_entry_t wait; |
dc98df5a KH |
1883 | }; |
1884 | ||
ac6424b9 | 1885 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1886 | unsigned mode, int sync, void *arg) |
1887 | { | |
d79154bb HD |
1888 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1889 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1890 | struct oom_wait_info *oom_wait_info; |
1891 | ||
1892 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1893 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1894 | |
2314b42d JW |
1895 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1896 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1897 | return 0; |
dc98df5a KH |
1898 | return autoremove_wake_function(wait, mode, sync, arg); |
1899 | } | |
1900 | ||
c0ff4b85 | 1901 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1902 | { |
c2b42d3c TH |
1903 | /* |
1904 | * For the following lockless ->under_oom test, the only required | |
1905 | * guarantee is that it must see the state asserted by an OOM when | |
1906 | * this function is called as a result of userland actions | |
1907 | * triggered by the notification of the OOM. This is trivially | |
1908 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1909 | * triggering notification. | |
1910 | */ | |
1911 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1912 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1913 | } |
1914 | ||
29ef680a MH |
1915 | enum oom_status { |
1916 | OOM_SUCCESS, | |
1917 | OOM_FAILED, | |
1918 | OOM_ASYNC, | |
1919 | OOM_SKIPPED | |
1920 | }; | |
1921 | ||
1922 | static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 1923 | { |
7056d3a3 MH |
1924 | enum oom_status ret; |
1925 | bool locked; | |
1926 | ||
29ef680a MH |
1927 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
1928 | return OOM_SKIPPED; | |
1929 | ||
7a1adfdd RG |
1930 | memcg_memory_event(memcg, MEMCG_OOM); |
1931 | ||
867578cb | 1932 | /* |
49426420 JW |
1933 | * We are in the middle of the charge context here, so we |
1934 | * don't want to block when potentially sitting on a callstack | |
1935 | * that holds all kinds of filesystem and mm locks. | |
1936 | * | |
29ef680a MH |
1937 | * cgroup1 allows disabling the OOM killer and waiting for outside |
1938 | * handling until the charge can succeed; remember the context and put | |
1939 | * the task to sleep at the end of the page fault when all locks are | |
1940 | * released. | |
49426420 | 1941 | * |
29ef680a MH |
1942 | * On the other hand, in-kernel OOM killer allows for an async victim |
1943 | * memory reclaim (oom_reaper) and that means that we are not solely | |
1944 | * relying on the oom victim to make a forward progress and we can | |
1945 | * invoke the oom killer here. | |
1946 | * | |
1947 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
1948 | * victim and then we have to bail out from the charge path. | |
867578cb | 1949 | */ |
29ef680a MH |
1950 | if (memcg->oom_kill_disable) { |
1951 | if (!current->in_user_fault) | |
1952 | return OOM_SKIPPED; | |
1953 | css_get(&memcg->css); | |
1954 | current->memcg_in_oom = memcg; | |
1955 | current->memcg_oom_gfp_mask = mask; | |
1956 | current->memcg_oom_order = order; | |
1957 | ||
1958 | return OOM_ASYNC; | |
1959 | } | |
1960 | ||
7056d3a3 MH |
1961 | mem_cgroup_mark_under_oom(memcg); |
1962 | ||
1963 | locked = mem_cgroup_oom_trylock(memcg); | |
1964 | ||
1965 | if (locked) | |
1966 | mem_cgroup_oom_notify(memcg); | |
1967 | ||
1968 | mem_cgroup_unmark_under_oom(memcg); | |
29ef680a | 1969 | if (mem_cgroup_out_of_memory(memcg, mask, order)) |
7056d3a3 MH |
1970 | ret = OOM_SUCCESS; |
1971 | else | |
1972 | ret = OOM_FAILED; | |
1973 | ||
1974 | if (locked) | |
1975 | mem_cgroup_oom_unlock(memcg); | |
29ef680a | 1976 | |
7056d3a3 | 1977 | return ret; |
3812c8c8 JW |
1978 | } |
1979 | ||
1980 | /** | |
1981 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1982 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1983 | * |
49426420 JW |
1984 | * This has to be called at the end of a page fault if the memcg OOM |
1985 | * handler was enabled. | |
3812c8c8 | 1986 | * |
49426420 | 1987 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1988 | * sleep on a waitqueue until the userspace task resolves the |
1989 | * situation. Sleeping directly in the charge context with all kinds | |
1990 | * of locks held is not a good idea, instead we remember an OOM state | |
1991 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1992 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1993 | * |
1994 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1995 | * completed, %false otherwise. |
3812c8c8 | 1996 | */ |
49426420 | 1997 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1998 | { |
626ebc41 | 1999 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 2000 | struct oom_wait_info owait; |
49426420 | 2001 | bool locked; |
3812c8c8 JW |
2002 | |
2003 | /* OOM is global, do not handle */ | |
3812c8c8 | 2004 | if (!memcg) |
49426420 | 2005 | return false; |
3812c8c8 | 2006 | |
7c5f64f8 | 2007 | if (!handle) |
49426420 | 2008 | goto cleanup; |
3812c8c8 JW |
2009 | |
2010 | owait.memcg = memcg; | |
2011 | owait.wait.flags = 0; | |
2012 | owait.wait.func = memcg_oom_wake_function; | |
2013 | owait.wait.private = current; | |
2055da97 | 2014 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 2015 | |
3812c8c8 | 2016 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
2017 | mem_cgroup_mark_under_oom(memcg); |
2018 | ||
2019 | locked = mem_cgroup_oom_trylock(memcg); | |
2020 | ||
2021 | if (locked) | |
2022 | mem_cgroup_oom_notify(memcg); | |
2023 | ||
2024 | if (locked && !memcg->oom_kill_disable) { | |
2025 | mem_cgroup_unmark_under_oom(memcg); | |
2026 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
2027 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
2028 | current->memcg_oom_order); | |
49426420 | 2029 | } else { |
3812c8c8 | 2030 | schedule(); |
49426420 JW |
2031 | mem_cgroup_unmark_under_oom(memcg); |
2032 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2033 | } | |
2034 | ||
2035 | if (locked) { | |
fb2a6fc5 JW |
2036 | mem_cgroup_oom_unlock(memcg); |
2037 | /* | |
2038 | * There is no guarantee that an OOM-lock contender | |
2039 | * sees the wakeups triggered by the OOM kill | |
2040 | * uncharges. Wake any sleepers explicitely. | |
2041 | */ | |
2042 | memcg_oom_recover(memcg); | |
2043 | } | |
49426420 | 2044 | cleanup: |
626ebc41 | 2045 | current->memcg_in_oom = NULL; |
3812c8c8 | 2046 | css_put(&memcg->css); |
867578cb | 2047 | return true; |
0b7f569e KH |
2048 | } |
2049 | ||
3d8b38eb RG |
2050 | /** |
2051 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
2052 | * @victim: task to be killed by the OOM killer | |
2053 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
2054 | * | |
2055 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
2056 | * by killing all belonging OOM-killable tasks. | |
2057 | * | |
2058 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
2059 | */ | |
2060 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
2061 | struct mem_cgroup *oom_domain) | |
2062 | { | |
2063 | struct mem_cgroup *oom_group = NULL; | |
2064 | struct mem_cgroup *memcg; | |
2065 | ||
2066 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
2067 | return NULL; | |
2068 | ||
2069 | if (!oom_domain) | |
2070 | oom_domain = root_mem_cgroup; | |
2071 | ||
2072 | rcu_read_lock(); | |
2073 | ||
2074 | memcg = mem_cgroup_from_task(victim); | |
2075 | if (memcg == root_mem_cgroup) | |
2076 | goto out; | |
2077 | ||
48fe267c RG |
2078 | /* |
2079 | * If the victim task has been asynchronously moved to a different | |
2080 | * memory cgroup, we might end up killing tasks outside oom_domain. | |
2081 | * In this case it's better to ignore memory.group.oom. | |
2082 | */ | |
2083 | if (unlikely(!mem_cgroup_is_descendant(memcg, oom_domain))) | |
2084 | goto out; | |
2085 | ||
3d8b38eb RG |
2086 | /* |
2087 | * Traverse the memory cgroup hierarchy from the victim task's | |
2088 | * cgroup up to the OOMing cgroup (or root) to find the | |
2089 | * highest-level memory cgroup with oom.group set. | |
2090 | */ | |
2091 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
2092 | if (memcg->oom_group) | |
2093 | oom_group = memcg; | |
2094 | ||
2095 | if (memcg == oom_domain) | |
2096 | break; | |
2097 | } | |
2098 | ||
2099 | if (oom_group) | |
2100 | css_get(&oom_group->css); | |
2101 | out: | |
2102 | rcu_read_unlock(); | |
2103 | ||
2104 | return oom_group; | |
2105 | } | |
2106 | ||
2107 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
2108 | { | |
2109 | pr_info("Tasks in "); | |
2110 | pr_cont_cgroup_path(memcg->css.cgroup); | |
2111 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
2112 | } | |
2113 | ||
d7365e78 | 2114 | /** |
bcfe06bf | 2115 | * lock_page_memcg - lock a page and memcg binding |
81f8c3a4 | 2116 | * @page: the page |
32047e2a | 2117 | * |
81f8c3a4 | 2118 | * This function protects unlocked LRU pages from being moved to |
739f79fc JW |
2119 | * another cgroup. |
2120 | * | |
1c824a68 JW |
2121 | * It ensures lifetime of the locked memcg. Caller is responsible |
2122 | * for the lifetime of the page. | |
d69b042f | 2123 | */ |
1c824a68 | 2124 | void lock_page_memcg(struct page *page) |
89c06bd5 | 2125 | { |
9da7b521 | 2126 | struct page *head = compound_head(page); /* rmap on tail pages */ |
89c06bd5 | 2127 | struct mem_cgroup *memcg; |
6de22619 | 2128 | unsigned long flags; |
89c06bd5 | 2129 | |
6de22619 JW |
2130 | /* |
2131 | * The RCU lock is held throughout the transaction. The fast | |
2132 | * path can get away without acquiring the memcg->move_lock | |
2133 | * because page moving starts with an RCU grace period. | |
739f79fc | 2134 | */ |
d7365e78 JW |
2135 | rcu_read_lock(); |
2136 | ||
2137 | if (mem_cgroup_disabled()) | |
1c824a68 | 2138 | return; |
89c06bd5 | 2139 | again: |
bcfe06bf | 2140 | memcg = page_memcg(head); |
29833315 | 2141 | if (unlikely(!memcg)) |
1c824a68 | 2142 | return; |
d7365e78 | 2143 | |
20ad50d6 AS |
2144 | #ifdef CONFIG_PROVE_LOCKING |
2145 | local_irq_save(flags); | |
2146 | might_lock(&memcg->move_lock); | |
2147 | local_irq_restore(flags); | |
2148 | #endif | |
2149 | ||
bdcbb659 | 2150 | if (atomic_read(&memcg->moving_account) <= 0) |
1c824a68 | 2151 | return; |
89c06bd5 | 2152 | |
6de22619 | 2153 | spin_lock_irqsave(&memcg->move_lock, flags); |
bcfe06bf | 2154 | if (memcg != page_memcg(head)) { |
6de22619 | 2155 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
2156 | goto again; |
2157 | } | |
6de22619 JW |
2158 | |
2159 | /* | |
1c824a68 JW |
2160 | * When charge migration first begins, we can have multiple |
2161 | * critical sections holding the fast-path RCU lock and one | |
2162 | * holding the slowpath move_lock. Track the task who has the | |
2163 | * move_lock for unlock_page_memcg(). | |
6de22619 JW |
2164 | */ |
2165 | memcg->move_lock_task = current; | |
2166 | memcg->move_lock_flags = flags; | |
89c06bd5 | 2167 | } |
81f8c3a4 | 2168 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 2169 | |
1c824a68 | 2170 | static void __unlock_page_memcg(struct mem_cgroup *memcg) |
89c06bd5 | 2171 | { |
6de22619 JW |
2172 | if (memcg && memcg->move_lock_task == current) { |
2173 | unsigned long flags = memcg->move_lock_flags; | |
2174 | ||
2175 | memcg->move_lock_task = NULL; | |
2176 | memcg->move_lock_flags = 0; | |
2177 | ||
2178 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2179 | } | |
89c06bd5 | 2180 | |
d7365e78 | 2181 | rcu_read_unlock(); |
89c06bd5 | 2182 | } |
739f79fc JW |
2183 | |
2184 | /** | |
bcfe06bf | 2185 | * unlock_page_memcg - unlock a page and memcg binding |
739f79fc JW |
2186 | * @page: the page |
2187 | */ | |
2188 | void unlock_page_memcg(struct page *page) | |
2189 | { | |
9da7b521 JW |
2190 | struct page *head = compound_head(page); |
2191 | ||
bcfe06bf | 2192 | __unlock_page_memcg(page_memcg(head)); |
739f79fc | 2193 | } |
81f8c3a4 | 2194 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 2195 | |
cdec2e42 KH |
2196 | struct memcg_stock_pcp { |
2197 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2198 | unsigned int nr_pages; |
bf4f0599 RG |
2199 | |
2200 | #ifdef CONFIG_MEMCG_KMEM | |
2201 | struct obj_cgroup *cached_objcg; | |
2202 | unsigned int nr_bytes; | |
2203 | #endif | |
2204 | ||
cdec2e42 | 2205 | struct work_struct work; |
26fe6168 | 2206 | unsigned long flags; |
a0db00fc | 2207 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2208 | }; |
2209 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2210 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2211 | |
bf4f0599 RG |
2212 | #ifdef CONFIG_MEMCG_KMEM |
2213 | static void drain_obj_stock(struct memcg_stock_pcp *stock); | |
2214 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2215 | struct mem_cgroup *root_memcg); | |
2216 | ||
2217 | #else | |
2218 | static inline void drain_obj_stock(struct memcg_stock_pcp *stock) | |
2219 | { | |
2220 | } | |
2221 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2222 | struct mem_cgroup *root_memcg) | |
2223 | { | |
2224 | return false; | |
2225 | } | |
2226 | #endif | |
2227 | ||
a0956d54 SS |
2228 | /** |
2229 | * consume_stock: Try to consume stocked charge on this cpu. | |
2230 | * @memcg: memcg to consume from. | |
2231 | * @nr_pages: how many pages to charge. | |
2232 | * | |
2233 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2234 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2235 | * service an allocation will refill the stock. | |
2236 | * | |
2237 | * returns true if successful, false otherwise. | |
cdec2e42 | 2238 | */ |
a0956d54 | 2239 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2240 | { |
2241 | struct memcg_stock_pcp *stock; | |
db2ba40c | 2242 | unsigned long flags; |
3e32cb2e | 2243 | bool ret = false; |
cdec2e42 | 2244 | |
a983b5eb | 2245 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 2246 | return ret; |
a0956d54 | 2247 | |
db2ba40c JW |
2248 | local_irq_save(flags); |
2249 | ||
2250 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 2251 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2252 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2253 | ret = true; |
2254 | } | |
db2ba40c JW |
2255 | |
2256 | local_irq_restore(flags); | |
2257 | ||
cdec2e42 KH |
2258 | return ret; |
2259 | } | |
2260 | ||
2261 | /* | |
3e32cb2e | 2262 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2263 | */ |
2264 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2265 | { | |
2266 | struct mem_cgroup *old = stock->cached; | |
2267 | ||
1a3e1f40 JW |
2268 | if (!old) |
2269 | return; | |
2270 | ||
11c9ea4e | 2271 | if (stock->nr_pages) { |
3e32cb2e | 2272 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 2273 | if (do_memsw_account()) |
3e32cb2e | 2274 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
11c9ea4e | 2275 | stock->nr_pages = 0; |
cdec2e42 | 2276 | } |
1a3e1f40 JW |
2277 | |
2278 | css_put(&old->css); | |
cdec2e42 | 2279 | stock->cached = NULL; |
cdec2e42 KH |
2280 | } |
2281 | ||
cdec2e42 KH |
2282 | static void drain_local_stock(struct work_struct *dummy) |
2283 | { | |
db2ba40c JW |
2284 | struct memcg_stock_pcp *stock; |
2285 | unsigned long flags; | |
2286 | ||
72f0184c MH |
2287 | /* |
2288 | * The only protection from memory hotplug vs. drain_stock races is | |
2289 | * that we always operate on local CPU stock here with IRQ disabled | |
2290 | */ | |
db2ba40c JW |
2291 | local_irq_save(flags); |
2292 | ||
2293 | stock = this_cpu_ptr(&memcg_stock); | |
bf4f0599 | 2294 | drain_obj_stock(stock); |
cdec2e42 | 2295 | drain_stock(stock); |
26fe6168 | 2296 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c JW |
2297 | |
2298 | local_irq_restore(flags); | |
cdec2e42 KH |
2299 | } |
2300 | ||
2301 | /* | |
3e32cb2e | 2302 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2303 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2304 | */ |
c0ff4b85 | 2305 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2306 | { |
db2ba40c JW |
2307 | struct memcg_stock_pcp *stock; |
2308 | unsigned long flags; | |
2309 | ||
2310 | local_irq_save(flags); | |
cdec2e42 | 2311 | |
db2ba40c | 2312 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 2313 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2314 | drain_stock(stock); |
1a3e1f40 | 2315 | css_get(&memcg->css); |
c0ff4b85 | 2316 | stock->cached = memcg; |
cdec2e42 | 2317 | } |
11c9ea4e | 2318 | stock->nr_pages += nr_pages; |
db2ba40c | 2319 | |
a983b5eb | 2320 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 RG |
2321 | drain_stock(stock); |
2322 | ||
db2ba40c | 2323 | local_irq_restore(flags); |
cdec2e42 KH |
2324 | } |
2325 | ||
2326 | /* | |
c0ff4b85 | 2327 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2328 | * of the hierarchy under it. |
cdec2e42 | 2329 | */ |
6d3d6aa2 | 2330 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2331 | { |
26fe6168 | 2332 | int cpu, curcpu; |
d38144b7 | 2333 | |
6d3d6aa2 JW |
2334 | /* If someone's already draining, avoid adding running more workers. */ |
2335 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2336 | return; | |
72f0184c MH |
2337 | /* |
2338 | * Notify other cpus that system-wide "drain" is running | |
2339 | * We do not care about races with the cpu hotplug because cpu down | |
2340 | * as well as workers from this path always operate on the local | |
2341 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2342 | */ | |
5af12d0e | 2343 | curcpu = get_cpu(); |
cdec2e42 KH |
2344 | for_each_online_cpu(cpu) { |
2345 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2346 | struct mem_cgroup *memcg; |
e1a366be | 2347 | bool flush = false; |
26fe6168 | 2348 | |
e1a366be | 2349 | rcu_read_lock(); |
c0ff4b85 | 2350 | memcg = stock->cached; |
e1a366be RG |
2351 | if (memcg && stock->nr_pages && |
2352 | mem_cgroup_is_descendant(memcg, root_memcg)) | |
2353 | flush = true; | |
bf4f0599 RG |
2354 | if (obj_stock_flush_required(stock, root_memcg)) |
2355 | flush = true; | |
e1a366be RG |
2356 | rcu_read_unlock(); |
2357 | ||
2358 | if (flush && | |
2359 | !test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { | |
d1a05b69 MH |
2360 | if (cpu == curcpu) |
2361 | drain_local_stock(&stock->work); | |
2362 | else | |
2363 | schedule_work_on(cpu, &stock->work); | |
2364 | } | |
cdec2e42 | 2365 | } |
5af12d0e | 2366 | put_cpu(); |
9f50fad6 | 2367 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2368 | } |
2369 | ||
308167fc | 2370 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
cdec2e42 | 2371 | { |
cdec2e42 | 2372 | struct memcg_stock_pcp *stock; |
a3d4c05a | 2373 | struct mem_cgroup *memcg; |
cdec2e42 | 2374 | |
cdec2e42 KH |
2375 | stock = &per_cpu(memcg_stock, cpu); |
2376 | drain_stock(stock); | |
a983b5eb JW |
2377 | |
2378 | for_each_mem_cgroup(memcg) { | |
a3d4c05a | 2379 | struct memcg_vmstats_percpu *statc; |
a983b5eb JW |
2380 | int i; |
2381 | ||
a3d4c05a JW |
2382 | statc = per_cpu_ptr(memcg->vmstats_percpu, cpu); |
2383 | ||
a983b5eb JW |
2384 | for (i = 0; i < MEMCG_NR_STAT; i++) { |
2385 | int nid; | |
a983b5eb | 2386 | |
a3d4c05a JW |
2387 | if (statc->stat[i]) { |
2388 | mod_memcg_state(memcg, i, statc->stat[i]); | |
2389 | statc->stat[i] = 0; | |
2390 | } | |
a983b5eb JW |
2391 | |
2392 | if (i >= NR_VM_NODE_STAT_ITEMS) | |
2393 | continue; | |
2394 | ||
2395 | for_each_node(nid) { | |
a3d4c05a | 2396 | struct batched_lruvec_stat *lstatc; |
a983b5eb | 2397 | struct mem_cgroup_per_node *pn; |
a3d4c05a | 2398 | long x; |
a983b5eb JW |
2399 | |
2400 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
a3d4c05a JW |
2401 | lstatc = per_cpu_ptr(pn->lruvec_stat_cpu, cpu); |
2402 | ||
2403 | x = lstatc->count[i]; | |
2404 | lstatc->count[i] = 0; | |
2405 | ||
2406 | if (x) { | |
42a30035 JW |
2407 | do { |
2408 | atomic_long_add(x, &pn->lruvec_stat[i]); | |
2409 | } while ((pn = parent_nodeinfo(pn, nid))); | |
a3d4c05a | 2410 | } |
a983b5eb JW |
2411 | } |
2412 | } | |
2413 | ||
e27be240 | 2414 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { |
a3d4c05a JW |
2415 | if (statc->events[i]) { |
2416 | count_memcg_events(memcg, i, statc->events[i]); | |
2417 | statc->events[i] = 0; | |
2418 | } | |
a983b5eb JW |
2419 | } |
2420 | } | |
2421 | ||
308167fc | 2422 | return 0; |
cdec2e42 KH |
2423 | } |
2424 | ||
b3ff9291 CD |
2425 | static unsigned long reclaim_high(struct mem_cgroup *memcg, |
2426 | unsigned int nr_pages, | |
2427 | gfp_t gfp_mask) | |
f7e1cb6e | 2428 | { |
b3ff9291 CD |
2429 | unsigned long nr_reclaimed = 0; |
2430 | ||
f7e1cb6e | 2431 | do { |
e22c6ed9 JW |
2432 | unsigned long pflags; |
2433 | ||
d1663a90 JK |
2434 | if (page_counter_read(&memcg->memory) <= |
2435 | READ_ONCE(memcg->memory.high)) | |
f7e1cb6e | 2436 | continue; |
e22c6ed9 | 2437 | |
e27be240 | 2438 | memcg_memory_event(memcg, MEMCG_HIGH); |
e22c6ed9 JW |
2439 | |
2440 | psi_memstall_enter(&pflags); | |
b3ff9291 CD |
2441 | nr_reclaimed += try_to_free_mem_cgroup_pages(memcg, nr_pages, |
2442 | gfp_mask, true); | |
e22c6ed9 | 2443 | psi_memstall_leave(&pflags); |
4bf17307 CD |
2444 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2445 | !mem_cgroup_is_root(memcg)); | |
b3ff9291 CD |
2446 | |
2447 | return nr_reclaimed; | |
f7e1cb6e JW |
2448 | } |
2449 | ||
2450 | static void high_work_func(struct work_struct *work) | |
2451 | { | |
2452 | struct mem_cgroup *memcg; | |
2453 | ||
2454 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2455 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2456 | } |
2457 | ||
0e4b01df CD |
2458 | /* |
2459 | * Clamp the maximum sleep time per allocation batch to 2 seconds. This is | |
2460 | * enough to still cause a significant slowdown in most cases, while still | |
2461 | * allowing diagnostics and tracing to proceed without becoming stuck. | |
2462 | */ | |
2463 | #define MEMCG_MAX_HIGH_DELAY_JIFFIES (2UL*HZ) | |
2464 | ||
2465 | /* | |
2466 | * When calculating the delay, we use these either side of the exponentiation to | |
2467 | * maintain precision and scale to a reasonable number of jiffies (see the table | |
2468 | * below. | |
2469 | * | |
2470 | * - MEMCG_DELAY_PRECISION_SHIFT: Extra precision bits while translating the | |
2471 | * overage ratio to a delay. | |
ac5ddd0f | 2472 | * - MEMCG_DELAY_SCALING_SHIFT: The number of bits to scale down the |
0e4b01df CD |
2473 | * proposed penalty in order to reduce to a reasonable number of jiffies, and |
2474 | * to produce a reasonable delay curve. | |
2475 | * | |
2476 | * MEMCG_DELAY_SCALING_SHIFT just happens to be a number that produces a | |
2477 | * reasonable delay curve compared to precision-adjusted overage, not | |
2478 | * penalising heavily at first, but still making sure that growth beyond the | |
2479 | * limit penalises misbehaviour cgroups by slowing them down exponentially. For | |
2480 | * example, with a high of 100 megabytes: | |
2481 | * | |
2482 | * +-------+------------------------+ | |
2483 | * | usage | time to allocate in ms | | |
2484 | * +-------+------------------------+ | |
2485 | * | 100M | 0 | | |
2486 | * | 101M | 6 | | |
2487 | * | 102M | 25 | | |
2488 | * | 103M | 57 | | |
2489 | * | 104M | 102 | | |
2490 | * | 105M | 159 | | |
2491 | * | 106M | 230 | | |
2492 | * | 107M | 313 | | |
2493 | * | 108M | 409 | | |
2494 | * | 109M | 518 | | |
2495 | * | 110M | 639 | | |
2496 | * | 111M | 774 | | |
2497 | * | 112M | 921 | | |
2498 | * | 113M | 1081 | | |
2499 | * | 114M | 1254 | | |
2500 | * | 115M | 1439 | | |
2501 | * | 116M | 1638 | | |
2502 | * | 117M | 1849 | | |
2503 | * | 118M | 2000 | | |
2504 | * | 119M | 2000 | | |
2505 | * | 120M | 2000 | | |
2506 | * +-------+------------------------+ | |
2507 | */ | |
2508 | #define MEMCG_DELAY_PRECISION_SHIFT 20 | |
2509 | #define MEMCG_DELAY_SCALING_SHIFT 14 | |
2510 | ||
8a5dbc65 | 2511 | static u64 calculate_overage(unsigned long usage, unsigned long high) |
b23afb93 | 2512 | { |
8a5dbc65 | 2513 | u64 overage; |
b23afb93 | 2514 | |
8a5dbc65 JK |
2515 | if (usage <= high) |
2516 | return 0; | |
e26733e0 | 2517 | |
8a5dbc65 JK |
2518 | /* |
2519 | * Prevent division by 0 in overage calculation by acting as if | |
2520 | * it was a threshold of 1 page | |
2521 | */ | |
2522 | high = max(high, 1UL); | |
9b8b1754 | 2523 | |
8a5dbc65 JK |
2524 | overage = usage - high; |
2525 | overage <<= MEMCG_DELAY_PRECISION_SHIFT; | |
2526 | return div64_u64(overage, high); | |
2527 | } | |
e26733e0 | 2528 | |
8a5dbc65 JK |
2529 | static u64 mem_find_max_overage(struct mem_cgroup *memcg) |
2530 | { | |
2531 | u64 overage, max_overage = 0; | |
e26733e0 | 2532 | |
8a5dbc65 JK |
2533 | do { |
2534 | overage = calculate_overage(page_counter_read(&memcg->memory), | |
d1663a90 | 2535 | READ_ONCE(memcg->memory.high)); |
8a5dbc65 | 2536 | max_overage = max(overage, max_overage); |
e26733e0 CD |
2537 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2538 | !mem_cgroup_is_root(memcg)); | |
2539 | ||
8a5dbc65 JK |
2540 | return max_overage; |
2541 | } | |
2542 | ||
4b82ab4f JK |
2543 | static u64 swap_find_max_overage(struct mem_cgroup *memcg) |
2544 | { | |
2545 | u64 overage, max_overage = 0; | |
2546 | ||
2547 | do { | |
2548 | overage = calculate_overage(page_counter_read(&memcg->swap), | |
2549 | READ_ONCE(memcg->swap.high)); | |
2550 | if (overage) | |
2551 | memcg_memory_event(memcg, MEMCG_SWAP_HIGH); | |
2552 | max_overage = max(overage, max_overage); | |
2553 | } while ((memcg = parent_mem_cgroup(memcg)) && | |
2554 | !mem_cgroup_is_root(memcg)); | |
2555 | ||
2556 | return max_overage; | |
2557 | } | |
2558 | ||
8a5dbc65 JK |
2559 | /* |
2560 | * Get the number of jiffies that we should penalise a mischievous cgroup which | |
2561 | * is exceeding its memory.high by checking both it and its ancestors. | |
2562 | */ | |
2563 | static unsigned long calculate_high_delay(struct mem_cgroup *memcg, | |
2564 | unsigned int nr_pages, | |
2565 | u64 max_overage) | |
2566 | { | |
2567 | unsigned long penalty_jiffies; | |
2568 | ||
e26733e0 CD |
2569 | if (!max_overage) |
2570 | return 0; | |
0e4b01df CD |
2571 | |
2572 | /* | |
0e4b01df CD |
2573 | * We use overage compared to memory.high to calculate the number of |
2574 | * jiffies to sleep (penalty_jiffies). Ideally this value should be | |
2575 | * fairly lenient on small overages, and increasingly harsh when the | |
2576 | * memcg in question makes it clear that it has no intention of stopping | |
2577 | * its crazy behaviour, so we exponentially increase the delay based on | |
2578 | * overage amount. | |
2579 | */ | |
e26733e0 CD |
2580 | penalty_jiffies = max_overage * max_overage * HZ; |
2581 | penalty_jiffies >>= MEMCG_DELAY_PRECISION_SHIFT; | |
2582 | penalty_jiffies >>= MEMCG_DELAY_SCALING_SHIFT; | |
0e4b01df CD |
2583 | |
2584 | /* | |
2585 | * Factor in the task's own contribution to the overage, such that four | |
2586 | * N-sized allocations are throttled approximately the same as one | |
2587 | * 4N-sized allocation. | |
2588 | * | |
2589 | * MEMCG_CHARGE_BATCH pages is nominal, so work out how much smaller or | |
2590 | * larger the current charge patch is than that. | |
2591 | */ | |
ff144e69 | 2592 | return penalty_jiffies * nr_pages / MEMCG_CHARGE_BATCH; |
e26733e0 CD |
2593 | } |
2594 | ||
2595 | /* | |
2596 | * Scheduled by try_charge() to be executed from the userland return path | |
2597 | * and reclaims memory over the high limit. | |
2598 | */ | |
2599 | void mem_cgroup_handle_over_high(void) | |
2600 | { | |
2601 | unsigned long penalty_jiffies; | |
2602 | unsigned long pflags; | |
b3ff9291 | 2603 | unsigned long nr_reclaimed; |
e26733e0 | 2604 | unsigned int nr_pages = current->memcg_nr_pages_over_high; |
d977aa93 | 2605 | int nr_retries = MAX_RECLAIM_RETRIES; |
e26733e0 | 2606 | struct mem_cgroup *memcg; |
b3ff9291 | 2607 | bool in_retry = false; |
e26733e0 CD |
2608 | |
2609 | if (likely(!nr_pages)) | |
2610 | return; | |
2611 | ||
2612 | memcg = get_mem_cgroup_from_mm(current->mm); | |
e26733e0 CD |
2613 | current->memcg_nr_pages_over_high = 0; |
2614 | ||
b3ff9291 CD |
2615 | retry_reclaim: |
2616 | /* | |
2617 | * The allocating task should reclaim at least the batch size, but for | |
2618 | * subsequent retries we only want to do what's necessary to prevent oom | |
2619 | * or breaching resource isolation. | |
2620 | * | |
2621 | * This is distinct from memory.max or page allocator behaviour because | |
2622 | * memory.high is currently batched, whereas memory.max and the page | |
2623 | * allocator run every time an allocation is made. | |
2624 | */ | |
2625 | nr_reclaimed = reclaim_high(memcg, | |
2626 | in_retry ? SWAP_CLUSTER_MAX : nr_pages, | |
2627 | GFP_KERNEL); | |
2628 | ||
e26733e0 CD |
2629 | /* |
2630 | * memory.high is breached and reclaim is unable to keep up. Throttle | |
2631 | * allocators proactively to slow down excessive growth. | |
2632 | */ | |
8a5dbc65 JK |
2633 | penalty_jiffies = calculate_high_delay(memcg, nr_pages, |
2634 | mem_find_max_overage(memcg)); | |
0e4b01df | 2635 | |
4b82ab4f JK |
2636 | penalty_jiffies += calculate_high_delay(memcg, nr_pages, |
2637 | swap_find_max_overage(memcg)); | |
2638 | ||
ff144e69 JK |
2639 | /* |
2640 | * Clamp the max delay per usermode return so as to still keep the | |
2641 | * application moving forwards and also permit diagnostics, albeit | |
2642 | * extremely slowly. | |
2643 | */ | |
2644 | penalty_jiffies = min(penalty_jiffies, MEMCG_MAX_HIGH_DELAY_JIFFIES); | |
2645 | ||
0e4b01df CD |
2646 | /* |
2647 | * Don't sleep if the amount of jiffies this memcg owes us is so low | |
2648 | * that it's not even worth doing, in an attempt to be nice to those who | |
2649 | * go only a small amount over their memory.high value and maybe haven't | |
2650 | * been aggressively reclaimed enough yet. | |
2651 | */ | |
2652 | if (penalty_jiffies <= HZ / 100) | |
2653 | goto out; | |
2654 | ||
b3ff9291 CD |
2655 | /* |
2656 | * If reclaim is making forward progress but we're still over | |
2657 | * memory.high, we want to encourage that rather than doing allocator | |
2658 | * throttling. | |
2659 | */ | |
2660 | if (nr_reclaimed || nr_retries--) { | |
2661 | in_retry = true; | |
2662 | goto retry_reclaim; | |
2663 | } | |
2664 | ||
0e4b01df CD |
2665 | /* |
2666 | * If we exit early, we're guaranteed to die (since | |
2667 | * schedule_timeout_killable sets TASK_KILLABLE). This means we don't | |
2668 | * need to account for any ill-begotten jiffies to pay them off later. | |
2669 | */ | |
2670 | psi_memstall_enter(&pflags); | |
2671 | schedule_timeout_killable(penalty_jiffies); | |
2672 | psi_memstall_leave(&pflags); | |
2673 | ||
2674 | out: | |
2675 | css_put(&memcg->css); | |
b23afb93 TH |
2676 | } |
2677 | ||
00501b53 JW |
2678 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2679 | unsigned int nr_pages) | |
8a9f3ccd | 2680 | { |
a983b5eb | 2681 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
d977aa93 | 2682 | int nr_retries = MAX_RECLAIM_RETRIES; |
6539cc05 | 2683 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2684 | struct page_counter *counter; |
e22c6ed9 | 2685 | enum oom_status oom_status; |
6539cc05 | 2686 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2687 | bool may_swap = true; |
2688 | bool drained = false; | |
e22c6ed9 | 2689 | unsigned long pflags; |
a636b327 | 2690 | |
ce00a967 | 2691 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 2692 | return 0; |
6539cc05 | 2693 | retry: |
b6b6cc72 | 2694 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2695 | return 0; |
8a9f3ccd | 2696 | |
7941d214 | 2697 | if (!do_memsw_account() || |
6071ca52 JW |
2698 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2699 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2700 | goto done_restock; |
7941d214 | 2701 | if (do_memsw_account()) |
3e32cb2e JW |
2702 | page_counter_uncharge(&memcg->memsw, batch); |
2703 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2704 | } else { |
3e32cb2e | 2705 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2706 | may_swap = false; |
3fbe7244 | 2707 | } |
7a81b88c | 2708 | |
6539cc05 JW |
2709 | if (batch > nr_pages) { |
2710 | batch = nr_pages; | |
2711 | goto retry; | |
2712 | } | |
6d61ef40 | 2713 | |
869712fd JW |
2714 | /* |
2715 | * Memcg doesn't have a dedicated reserve for atomic | |
2716 | * allocations. But like the global atomic pool, we need to | |
2717 | * put the burden of reclaim on regular allocation requests | |
2718 | * and let these go through as privileged allocations. | |
2719 | */ | |
2720 | if (gfp_mask & __GFP_ATOMIC) | |
2721 | goto force; | |
2722 | ||
06b078fc JW |
2723 | /* |
2724 | * Unlike in global OOM situations, memcg is not in a physical | |
2725 | * memory shortage. Allow dying and OOM-killed tasks to | |
2726 | * bypass the last charges so that they can exit quickly and | |
2727 | * free their memory. | |
2728 | */ | |
7775face | 2729 | if (unlikely(should_force_charge())) |
10d53c74 | 2730 | goto force; |
06b078fc | 2731 | |
89a28483 JW |
2732 | /* |
2733 | * Prevent unbounded recursion when reclaim operations need to | |
2734 | * allocate memory. This might exceed the limits temporarily, | |
2735 | * but we prefer facilitating memory reclaim and getting back | |
2736 | * under the limit over triggering OOM kills in these cases. | |
2737 | */ | |
2738 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2739 | goto force; | |
2740 | ||
06b078fc JW |
2741 | if (unlikely(task_in_memcg_oom(current))) |
2742 | goto nomem; | |
2743 | ||
d0164adc | 2744 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2745 | goto nomem; |
4b534334 | 2746 | |
e27be240 | 2747 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
241994ed | 2748 | |
e22c6ed9 | 2749 | psi_memstall_enter(&pflags); |
b70a2a21 JW |
2750 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2751 | gfp_mask, may_swap); | |
e22c6ed9 | 2752 | psi_memstall_leave(&pflags); |
6539cc05 | 2753 | |
61e02c74 | 2754 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2755 | goto retry; |
28c34c29 | 2756 | |
b70a2a21 | 2757 | if (!drained) { |
6d3d6aa2 | 2758 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2759 | drained = true; |
2760 | goto retry; | |
2761 | } | |
2762 | ||
28c34c29 JW |
2763 | if (gfp_mask & __GFP_NORETRY) |
2764 | goto nomem; | |
6539cc05 JW |
2765 | /* |
2766 | * Even though the limit is exceeded at this point, reclaim | |
2767 | * may have been able to free some pages. Retry the charge | |
2768 | * before killing the task. | |
2769 | * | |
2770 | * Only for regular pages, though: huge pages are rather | |
2771 | * unlikely to succeed so close to the limit, and we fall back | |
2772 | * to regular pages anyway in case of failure. | |
2773 | */ | |
61e02c74 | 2774 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2775 | goto retry; |
2776 | /* | |
2777 | * At task move, charge accounts can be doubly counted. So, it's | |
2778 | * better to wait until the end of task_move if something is going on. | |
2779 | */ | |
2780 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2781 | goto retry; | |
2782 | ||
9b130619 JW |
2783 | if (nr_retries--) |
2784 | goto retry; | |
2785 | ||
38d38493 | 2786 | if (gfp_mask & __GFP_RETRY_MAYFAIL) |
29ef680a MH |
2787 | goto nomem; |
2788 | ||
6539cc05 | 2789 | if (fatal_signal_pending(current)) |
10d53c74 | 2790 | goto force; |
6539cc05 | 2791 | |
29ef680a MH |
2792 | /* |
2793 | * keep retrying as long as the memcg oom killer is able to make | |
2794 | * a forward progress or bypass the charge if the oom killer | |
2795 | * couldn't make any progress. | |
2796 | */ | |
2797 | oom_status = mem_cgroup_oom(mem_over_limit, gfp_mask, | |
3608de07 | 2798 | get_order(nr_pages * PAGE_SIZE)); |
29ef680a MH |
2799 | switch (oom_status) { |
2800 | case OOM_SUCCESS: | |
d977aa93 | 2801 | nr_retries = MAX_RECLAIM_RETRIES; |
29ef680a MH |
2802 | goto retry; |
2803 | case OOM_FAILED: | |
2804 | goto force; | |
2805 | default: | |
2806 | goto nomem; | |
2807 | } | |
7a81b88c | 2808 | nomem: |
6d1fdc48 | 2809 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2810 | return -ENOMEM; |
10d53c74 TH |
2811 | force: |
2812 | /* | |
2813 | * The allocation either can't fail or will lead to more memory | |
2814 | * being freed very soon. Allow memory usage go over the limit | |
2815 | * temporarily by force charging it. | |
2816 | */ | |
2817 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2818 | if (do_memsw_account()) |
10d53c74 | 2819 | page_counter_charge(&memcg->memsw, nr_pages); |
10d53c74 TH |
2820 | |
2821 | return 0; | |
6539cc05 JW |
2822 | |
2823 | done_restock: | |
2824 | if (batch > nr_pages) | |
2825 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2826 | |
241994ed | 2827 | /* |
b23afb93 TH |
2828 | * If the hierarchy is above the normal consumption range, schedule |
2829 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2830 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2831 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2832 | * not recorded as it most likely matches current's and won't | |
2833 | * change in the meantime. As high limit is checked again before | |
2834 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2835 | */ |
2836 | do { | |
4b82ab4f JK |
2837 | bool mem_high, swap_high; |
2838 | ||
2839 | mem_high = page_counter_read(&memcg->memory) > | |
2840 | READ_ONCE(memcg->memory.high); | |
2841 | swap_high = page_counter_read(&memcg->swap) > | |
2842 | READ_ONCE(memcg->swap.high); | |
2843 | ||
2844 | /* Don't bother a random interrupted task */ | |
2845 | if (in_interrupt()) { | |
2846 | if (mem_high) { | |
f7e1cb6e JW |
2847 | schedule_work(&memcg->high_work); |
2848 | break; | |
2849 | } | |
4b82ab4f JK |
2850 | continue; |
2851 | } | |
2852 | ||
2853 | if (mem_high || swap_high) { | |
2854 | /* | |
2855 | * The allocating tasks in this cgroup will need to do | |
2856 | * reclaim or be throttled to prevent further growth | |
2857 | * of the memory or swap footprints. | |
2858 | * | |
2859 | * Target some best-effort fairness between the tasks, | |
2860 | * and distribute reclaim work and delay penalties | |
2861 | * based on how much each task is actually allocating. | |
2862 | */ | |
9516a18a | 2863 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2864 | set_notify_resume(current); |
2865 | break; | |
2866 | } | |
241994ed | 2867 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2868 | |
2869 | return 0; | |
7a81b88c | 2870 | } |
8a9f3ccd | 2871 | |
f0e45fb4 | 2872 | #if defined(CONFIG_MEMCG_KMEM) || defined(CONFIG_MMU) |
00501b53 | 2873 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2874 | { |
ce00a967 JW |
2875 | if (mem_cgroup_is_root(memcg)) |
2876 | return; | |
2877 | ||
3e32cb2e | 2878 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2879 | if (do_memsw_account()) |
3e32cb2e | 2880 | page_counter_uncharge(&memcg->memsw, nr_pages); |
d01dd17f | 2881 | } |
f0e45fb4 | 2882 | #endif |
d01dd17f | 2883 | |
d9eb1ea2 | 2884 | static void commit_charge(struct page *page, struct mem_cgroup *memcg) |
0a31bc97 | 2885 | { |
bcfe06bf | 2886 | VM_BUG_ON_PAGE(page_memcg(page), page); |
0a31bc97 | 2887 | /* |
a5eb011a | 2888 | * Any of the following ensures page's memcg stability: |
0a31bc97 | 2889 | * |
a0b5b414 JW |
2890 | * - the page lock |
2891 | * - LRU isolation | |
2892 | * - lock_page_memcg() | |
2893 | * - exclusive reference | |
0a31bc97 | 2894 | */ |
bcfe06bf | 2895 | page->memcg_data = (unsigned long)memcg; |
7a81b88c | 2896 | } |
66e1707b | 2897 | |
84c07d11 | 2898 | #ifdef CONFIG_MEMCG_KMEM |
10befea9 | 2899 | int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s, |
2e9bd483 | 2900 | gfp_t gfp, bool new_page) |
10befea9 RG |
2901 | { |
2902 | unsigned int objects = objs_per_slab_page(s, page); | |
2e9bd483 | 2903 | unsigned long memcg_data; |
10befea9 RG |
2904 | void *vec; |
2905 | ||
2906 | vec = kcalloc_node(objects, sizeof(struct obj_cgroup *), gfp, | |
2907 | page_to_nid(page)); | |
2908 | if (!vec) | |
2909 | return -ENOMEM; | |
2910 | ||
2e9bd483 RG |
2911 | memcg_data = (unsigned long) vec | MEMCG_DATA_OBJCGS; |
2912 | if (new_page) { | |
2913 | /* | |
2914 | * If the slab page is brand new and nobody can yet access | |
2915 | * it's memcg_data, no synchronization is required and | |
2916 | * memcg_data can be simply assigned. | |
2917 | */ | |
2918 | page->memcg_data = memcg_data; | |
2919 | } else if (cmpxchg(&page->memcg_data, 0, memcg_data)) { | |
2920 | /* | |
2921 | * If the slab page is already in use, somebody can allocate | |
2922 | * and assign obj_cgroups in parallel. In this case the existing | |
2923 | * objcg vector should be reused. | |
2924 | */ | |
10befea9 | 2925 | kfree(vec); |
2e9bd483 RG |
2926 | return 0; |
2927 | } | |
10befea9 | 2928 | |
2e9bd483 | 2929 | kmemleak_not_leak(vec); |
10befea9 RG |
2930 | return 0; |
2931 | } | |
2932 | ||
8380ce47 RG |
2933 | /* |
2934 | * Returns a pointer to the memory cgroup to which the kernel object is charged. | |
2935 | * | |
bcfe06bf RG |
2936 | * A passed kernel object can be a slab object or a generic kernel page, so |
2937 | * different mechanisms for getting the memory cgroup pointer should be used. | |
2938 | * In certain cases (e.g. kernel stacks or large kmallocs with SLUB) the caller | |
2939 | * can not know for sure how the kernel object is implemented. | |
2940 | * mem_cgroup_from_obj() can be safely used in such cases. | |
2941 | * | |
8380ce47 RG |
2942 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), |
2943 | * cgroup_mutex, etc. | |
2944 | */ | |
2945 | struct mem_cgroup *mem_cgroup_from_obj(void *p) | |
2946 | { | |
2947 | struct page *page; | |
2948 | ||
2949 | if (mem_cgroup_disabled()) | |
2950 | return NULL; | |
2951 | ||
2952 | page = virt_to_head_page(p); | |
2953 | ||
2954 | /* | |
9855609b RG |
2955 | * Slab objects are accounted individually, not per-page. |
2956 | * Memcg membership data for each individual object is saved in | |
2957 | * the page->obj_cgroups. | |
8380ce47 | 2958 | */ |
270c6a71 | 2959 | if (page_objcgs_check(page)) { |
9855609b RG |
2960 | struct obj_cgroup *objcg; |
2961 | unsigned int off; | |
2962 | ||
2963 | off = obj_to_index(page->slab_cache, page, p); | |
270c6a71 | 2964 | objcg = page_objcgs(page)[off]; |
10befea9 RG |
2965 | if (objcg) |
2966 | return obj_cgroup_memcg(objcg); | |
2967 | ||
2968 | return NULL; | |
9855609b | 2969 | } |
8380ce47 | 2970 | |
bcfe06bf RG |
2971 | /* |
2972 | * page_memcg_check() is used here, because page_has_obj_cgroups() | |
2973 | * check above could fail because the object cgroups vector wasn't set | |
2974 | * at that moment, but it can be set concurrently. | |
2975 | * page_memcg_check(page) will guarantee that a proper memory | |
2976 | * cgroup pointer or NULL will be returned. | |
2977 | */ | |
2978 | return page_memcg_check(page); | |
8380ce47 RG |
2979 | } |
2980 | ||
bf4f0599 RG |
2981 | __always_inline struct obj_cgroup *get_obj_cgroup_from_current(void) |
2982 | { | |
2983 | struct obj_cgroup *objcg = NULL; | |
2984 | struct mem_cgroup *memcg; | |
2985 | ||
279c3393 RG |
2986 | if (memcg_kmem_bypass()) |
2987 | return NULL; | |
2988 | ||
bf4f0599 | 2989 | rcu_read_lock(); |
37d5985c RG |
2990 | if (unlikely(active_memcg())) |
2991 | memcg = active_memcg(); | |
bf4f0599 RG |
2992 | else |
2993 | memcg = mem_cgroup_from_task(current); | |
2994 | ||
2995 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { | |
2996 | objcg = rcu_dereference(memcg->objcg); | |
2997 | if (objcg && obj_cgroup_tryget(objcg)) | |
2998 | break; | |
2f7659a3 | 2999 | objcg = NULL; |
bf4f0599 RG |
3000 | } |
3001 | rcu_read_unlock(); | |
3002 | ||
3003 | return objcg; | |
3004 | } | |
3005 | ||
f3bb3043 | 3006 | static int memcg_alloc_cache_id(void) |
55007d84 | 3007 | { |
f3bb3043 VD |
3008 | int id, size; |
3009 | int err; | |
3010 | ||
dbcf73e2 | 3011 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
3012 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
3013 | if (id < 0) | |
3014 | return id; | |
55007d84 | 3015 | |
dbcf73e2 | 3016 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
3017 | return id; |
3018 | ||
3019 | /* | |
3020 | * There's no space for the new id in memcg_caches arrays, | |
3021 | * so we have to grow them. | |
3022 | */ | |
05257a1a | 3023 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
3024 | |
3025 | size = 2 * (id + 1); | |
55007d84 GC |
3026 | if (size < MEMCG_CACHES_MIN_SIZE) |
3027 | size = MEMCG_CACHES_MIN_SIZE; | |
3028 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
3029 | size = MEMCG_CACHES_MAX_SIZE; | |
3030 | ||
9855609b | 3031 | err = memcg_update_all_list_lrus(size); |
05257a1a VD |
3032 | if (!err) |
3033 | memcg_nr_cache_ids = size; | |
3034 | ||
3035 | up_write(&memcg_cache_ids_sem); | |
3036 | ||
f3bb3043 | 3037 | if (err) { |
dbcf73e2 | 3038 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
3039 | return err; |
3040 | } | |
3041 | return id; | |
3042 | } | |
3043 | ||
3044 | static void memcg_free_cache_id(int id) | |
3045 | { | |
dbcf73e2 | 3046 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
3047 | } |
3048 | ||
45264778 | 3049 | /** |
4b13f64d | 3050 | * __memcg_kmem_charge: charge a number of kernel pages to a memcg |
10eaec2f | 3051 | * @memcg: memory cgroup to charge |
45264778 | 3052 | * @gfp: reclaim mode |
92d0510c | 3053 | * @nr_pages: number of pages to charge |
45264778 VD |
3054 | * |
3055 | * Returns 0 on success, an error code on failure. | |
3056 | */ | |
c1a660de RG |
3057 | static int __memcg_kmem_charge(struct mem_cgroup *memcg, gfp_t gfp, |
3058 | unsigned int nr_pages) | |
7ae1e1d0 | 3059 | { |
f3ccb2c4 | 3060 | struct page_counter *counter; |
7ae1e1d0 GC |
3061 | int ret; |
3062 | ||
f3ccb2c4 | 3063 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 3064 | if (ret) |
f3ccb2c4 | 3065 | return ret; |
52c29b04 JW |
3066 | |
3067 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
3068 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
e55d9d9b MH |
3069 | |
3070 | /* | |
3071 | * Enforce __GFP_NOFAIL allocation because callers are not | |
3072 | * prepared to see failures and likely do not have any failure | |
3073 | * handling code. | |
3074 | */ | |
3075 | if (gfp & __GFP_NOFAIL) { | |
3076 | page_counter_charge(&memcg->kmem, nr_pages); | |
3077 | return 0; | |
3078 | } | |
52c29b04 JW |
3079 | cancel_charge(memcg, nr_pages); |
3080 | return -ENOMEM; | |
7ae1e1d0 | 3081 | } |
f3ccb2c4 | 3082 | return 0; |
7ae1e1d0 GC |
3083 | } |
3084 | ||
4b13f64d RG |
3085 | /** |
3086 | * __memcg_kmem_uncharge: uncharge a number of kernel pages from a memcg | |
3087 | * @memcg: memcg to uncharge | |
3088 | * @nr_pages: number of pages to uncharge | |
3089 | */ | |
c1a660de | 3090 | static void __memcg_kmem_uncharge(struct mem_cgroup *memcg, unsigned int nr_pages) |
4b13f64d RG |
3091 | { |
3092 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
3093 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
3094 | ||
3de7d4f2 | 3095 | refill_stock(memcg, nr_pages); |
4b13f64d RG |
3096 | } |
3097 | ||
45264778 | 3098 | /** |
f4b00eab | 3099 | * __memcg_kmem_charge_page: charge a kmem page to the current memory cgroup |
45264778 VD |
3100 | * @page: page to charge |
3101 | * @gfp: reclaim mode | |
3102 | * @order: allocation order | |
3103 | * | |
3104 | * Returns 0 on success, an error code on failure. | |
3105 | */ | |
f4b00eab | 3106 | int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 3107 | { |
f3ccb2c4 | 3108 | struct mem_cgroup *memcg; |
fcff7d7e | 3109 | int ret = 0; |
7ae1e1d0 | 3110 | |
d46eb14b | 3111 | memcg = get_mem_cgroup_from_current(); |
279c3393 | 3112 | if (memcg && !mem_cgroup_is_root(memcg)) { |
4b13f64d | 3113 | ret = __memcg_kmem_charge(memcg, gfp, 1 << order); |
4d96ba35 | 3114 | if (!ret) { |
18b2db3b RG |
3115 | page->memcg_data = (unsigned long)memcg | |
3116 | MEMCG_DATA_KMEM; | |
1a3e1f40 | 3117 | return 0; |
4d96ba35 | 3118 | } |
279c3393 | 3119 | css_put(&memcg->css); |
c4159a75 | 3120 | } |
d05e83a6 | 3121 | return ret; |
7ae1e1d0 | 3122 | } |
49a18eae | 3123 | |
45264778 | 3124 | /** |
f4b00eab | 3125 | * __memcg_kmem_uncharge_page: uncharge a kmem page |
45264778 VD |
3126 | * @page: page to uncharge |
3127 | * @order: allocation order | |
3128 | */ | |
f4b00eab | 3129 | void __memcg_kmem_uncharge_page(struct page *page, int order) |
7ae1e1d0 | 3130 | { |
bcfe06bf | 3131 | struct mem_cgroup *memcg = page_memcg(page); |
f3ccb2c4 | 3132 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 3133 | |
7ae1e1d0 GC |
3134 | if (!memcg) |
3135 | return; | |
3136 | ||
309381fe | 3137 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
4b13f64d | 3138 | __memcg_kmem_uncharge(memcg, nr_pages); |
bcfe06bf | 3139 | page->memcg_data = 0; |
1a3e1f40 | 3140 | css_put(&memcg->css); |
60d3fd32 | 3141 | } |
bf4f0599 RG |
3142 | |
3143 | static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) | |
3144 | { | |
3145 | struct memcg_stock_pcp *stock; | |
3146 | unsigned long flags; | |
3147 | bool ret = false; | |
3148 | ||
3149 | local_irq_save(flags); | |
3150 | ||
3151 | stock = this_cpu_ptr(&memcg_stock); | |
3152 | if (objcg == stock->cached_objcg && stock->nr_bytes >= nr_bytes) { | |
3153 | stock->nr_bytes -= nr_bytes; | |
3154 | ret = true; | |
3155 | } | |
3156 | ||
3157 | local_irq_restore(flags); | |
3158 | ||
3159 | return ret; | |
3160 | } | |
3161 | ||
3162 | static void drain_obj_stock(struct memcg_stock_pcp *stock) | |
3163 | { | |
3164 | struct obj_cgroup *old = stock->cached_objcg; | |
3165 | ||
3166 | if (!old) | |
3167 | return; | |
3168 | ||
3169 | if (stock->nr_bytes) { | |
3170 | unsigned int nr_pages = stock->nr_bytes >> PAGE_SHIFT; | |
3171 | unsigned int nr_bytes = stock->nr_bytes & (PAGE_SIZE - 1); | |
3172 | ||
3173 | if (nr_pages) { | |
3174 | rcu_read_lock(); | |
3175 | __memcg_kmem_uncharge(obj_cgroup_memcg(old), nr_pages); | |
3176 | rcu_read_unlock(); | |
3177 | } | |
3178 | ||
3179 | /* | |
3180 | * The leftover is flushed to the centralized per-memcg value. | |
3181 | * On the next attempt to refill obj stock it will be moved | |
3182 | * to a per-cpu stock (probably, on an other CPU), see | |
3183 | * refill_obj_stock(). | |
3184 | * | |
3185 | * How often it's flushed is a trade-off between the memory | |
3186 | * limit enforcement accuracy and potential CPU contention, | |
3187 | * so it might be changed in the future. | |
3188 | */ | |
3189 | atomic_add(nr_bytes, &old->nr_charged_bytes); | |
3190 | stock->nr_bytes = 0; | |
3191 | } | |
3192 | ||
3193 | obj_cgroup_put(old); | |
3194 | stock->cached_objcg = NULL; | |
3195 | } | |
3196 | ||
3197 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
3198 | struct mem_cgroup *root_memcg) | |
3199 | { | |
3200 | struct mem_cgroup *memcg; | |
3201 | ||
3202 | if (stock->cached_objcg) { | |
3203 | memcg = obj_cgroup_memcg(stock->cached_objcg); | |
3204 | if (memcg && mem_cgroup_is_descendant(memcg, root_memcg)) | |
3205 | return true; | |
3206 | } | |
3207 | ||
3208 | return false; | |
3209 | } | |
3210 | ||
3211 | static void refill_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) | |
3212 | { | |
3213 | struct memcg_stock_pcp *stock; | |
3214 | unsigned long flags; | |
3215 | ||
3216 | local_irq_save(flags); | |
3217 | ||
3218 | stock = this_cpu_ptr(&memcg_stock); | |
3219 | if (stock->cached_objcg != objcg) { /* reset if necessary */ | |
3220 | drain_obj_stock(stock); | |
3221 | obj_cgroup_get(objcg); | |
3222 | stock->cached_objcg = objcg; | |
3223 | stock->nr_bytes = atomic_xchg(&objcg->nr_charged_bytes, 0); | |
3224 | } | |
3225 | stock->nr_bytes += nr_bytes; | |
3226 | ||
3227 | if (stock->nr_bytes > PAGE_SIZE) | |
3228 | drain_obj_stock(stock); | |
3229 | ||
3230 | local_irq_restore(flags); | |
3231 | } | |
3232 | ||
3233 | int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size) | |
3234 | { | |
3235 | struct mem_cgroup *memcg; | |
3236 | unsigned int nr_pages, nr_bytes; | |
3237 | int ret; | |
3238 | ||
3239 | if (consume_obj_stock(objcg, size)) | |
3240 | return 0; | |
3241 | ||
3242 | /* | |
3243 | * In theory, memcg->nr_charged_bytes can have enough | |
3244 | * pre-charged bytes to satisfy the allocation. However, | |
3245 | * flushing memcg->nr_charged_bytes requires two atomic | |
3246 | * operations, and memcg->nr_charged_bytes can't be big, | |
3247 | * so it's better to ignore it and try grab some new pages. | |
3248 | * memcg->nr_charged_bytes will be flushed in | |
3249 | * refill_obj_stock(), called from this function or | |
3250 | * independently later. | |
3251 | */ | |
3252 | rcu_read_lock(); | |
eefbfa7f | 3253 | retry: |
bf4f0599 | 3254 | memcg = obj_cgroup_memcg(objcg); |
eefbfa7f MS |
3255 | if (unlikely(!css_tryget(&memcg->css))) |
3256 | goto retry; | |
bf4f0599 RG |
3257 | rcu_read_unlock(); |
3258 | ||
3259 | nr_pages = size >> PAGE_SHIFT; | |
3260 | nr_bytes = size & (PAGE_SIZE - 1); | |
3261 | ||
3262 | if (nr_bytes) | |
3263 | nr_pages += 1; | |
3264 | ||
3265 | ret = __memcg_kmem_charge(memcg, gfp, nr_pages); | |
3266 | if (!ret && nr_bytes) | |
3267 | refill_obj_stock(objcg, PAGE_SIZE - nr_bytes); | |
3268 | ||
3269 | css_put(&memcg->css); | |
3270 | return ret; | |
3271 | } | |
3272 | ||
3273 | void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size) | |
3274 | { | |
3275 | refill_obj_stock(objcg, size); | |
3276 | } | |
3277 | ||
84c07d11 | 3278 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 3279 | |
ca3e0214 | 3280 | /* |
be6c8982 | 3281 | * Because page_memcg(head) is not set on tails, set it now. |
ca3e0214 | 3282 | */ |
be6c8982 | 3283 | void split_page_memcg(struct page *head, unsigned int nr) |
ca3e0214 | 3284 | { |
bcfe06bf | 3285 | struct mem_cgroup *memcg = page_memcg(head); |
e94c8a9c | 3286 | int i; |
ca3e0214 | 3287 | |
be6c8982 | 3288 | if (mem_cgroup_disabled() || !memcg) |
3d37c4a9 | 3289 | return; |
b070e65c | 3290 | |
be6c8982 ZG |
3291 | for (i = 1; i < nr; i++) |
3292 | head[i].memcg_data = head->memcg_data; | |
3293 | css_get_many(&memcg->css, nr - 1); | |
ca3e0214 | 3294 | } |
ca3e0214 | 3295 | |
c255a458 | 3296 | #ifdef CONFIG_MEMCG_SWAP |
02491447 DN |
3297 | /** |
3298 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3299 | * @entry: swap entry to be moved | |
3300 | * @from: mem_cgroup which the entry is moved from | |
3301 | * @to: mem_cgroup which the entry is moved to | |
3302 | * | |
3303 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3304 | * as the mem_cgroup's id of @from. | |
3305 | * | |
3306 | * Returns 0 on success, -EINVAL on failure. | |
3307 | * | |
3e32cb2e | 3308 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
3309 | * both res and memsw, and called css_get(). |
3310 | */ | |
3311 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3312 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3313 | { |
3314 | unsigned short old_id, new_id; | |
3315 | ||
34c00c31 LZ |
3316 | old_id = mem_cgroup_id(from); |
3317 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3318 | |
3319 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
3320 | mod_memcg_state(from, MEMCG_SWAP, -1); |
3321 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
3322 | return 0; |
3323 | } | |
3324 | return -EINVAL; | |
3325 | } | |
3326 | #else | |
3327 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3328 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3329 | { |
3330 | return -EINVAL; | |
3331 | } | |
8c7c6e34 | 3332 | #endif |
d13d1443 | 3333 | |
bbec2e15 | 3334 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 3335 | |
bbec2e15 RG |
3336 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
3337 | unsigned long max, bool memsw) | |
628f4235 | 3338 | { |
3e32cb2e | 3339 | bool enlarge = false; |
bb4a7ea2 | 3340 | bool drained = false; |
3e32cb2e | 3341 | int ret; |
c054a78c YZ |
3342 | bool limits_invariant; |
3343 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 3344 | |
3e32cb2e | 3345 | do { |
628f4235 KH |
3346 | if (signal_pending(current)) { |
3347 | ret = -EINTR; | |
3348 | break; | |
3349 | } | |
3e32cb2e | 3350 | |
bbec2e15 | 3351 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
3352 | /* |
3353 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 3354 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 3355 | */ |
15b42562 | 3356 | limits_invariant = memsw ? max >= READ_ONCE(memcg->memory.max) : |
bbec2e15 | 3357 | max <= memcg->memsw.max; |
c054a78c | 3358 | if (!limits_invariant) { |
bbec2e15 | 3359 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 3360 | ret = -EINVAL; |
8c7c6e34 KH |
3361 | break; |
3362 | } | |
bbec2e15 | 3363 | if (max > counter->max) |
3e32cb2e | 3364 | enlarge = true; |
bbec2e15 RG |
3365 | ret = page_counter_set_max(counter, max); |
3366 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
3367 | |
3368 | if (!ret) | |
3369 | break; | |
3370 | ||
bb4a7ea2 SB |
3371 | if (!drained) { |
3372 | drain_all_stock(memcg); | |
3373 | drained = true; | |
3374 | continue; | |
3375 | } | |
3376 | ||
1ab5c056 AR |
3377 | if (!try_to_free_mem_cgroup_pages(memcg, 1, |
3378 | GFP_KERNEL, !memsw)) { | |
3379 | ret = -EBUSY; | |
3380 | break; | |
3381 | } | |
3382 | } while (true); | |
3e32cb2e | 3383 | |
3c11ecf4 KH |
3384 | if (!ret && enlarge) |
3385 | memcg_oom_recover(memcg); | |
3e32cb2e | 3386 | |
628f4235 KH |
3387 | return ret; |
3388 | } | |
3389 | ||
ef8f2327 | 3390 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
3391 | gfp_t gfp_mask, |
3392 | unsigned long *total_scanned) | |
3393 | { | |
3394 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 3395 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
3396 | unsigned long reclaimed; |
3397 | int loop = 0; | |
ef8f2327 | 3398 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 3399 | unsigned long excess; |
0608f43d AM |
3400 | unsigned long nr_scanned; |
3401 | ||
3402 | if (order > 0) | |
3403 | return 0; | |
3404 | ||
ef8f2327 | 3405 | mctz = soft_limit_tree_node(pgdat->node_id); |
d6507ff5 MH |
3406 | |
3407 | /* | |
3408 | * Do not even bother to check the largest node if the root | |
3409 | * is empty. Do it lockless to prevent lock bouncing. Races | |
3410 | * are acceptable as soft limit is best effort anyway. | |
3411 | */ | |
bfc7228b | 3412 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
3413 | return 0; |
3414 | ||
0608f43d AM |
3415 | /* |
3416 | * This loop can run a while, specially if mem_cgroup's continuously | |
3417 | * keep exceeding their soft limit and putting the system under | |
3418 | * pressure | |
3419 | */ | |
3420 | do { | |
3421 | if (next_mz) | |
3422 | mz = next_mz; | |
3423 | else | |
3424 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3425 | if (!mz) | |
3426 | break; | |
3427 | ||
3428 | nr_scanned = 0; | |
ef8f2327 | 3429 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
0608f43d AM |
3430 | gfp_mask, &nr_scanned); |
3431 | nr_reclaimed += reclaimed; | |
3432 | *total_scanned += nr_scanned; | |
0a31bc97 | 3433 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 3434 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
3435 | |
3436 | /* | |
3437 | * If we failed to reclaim anything from this memory cgroup | |
3438 | * it is time to move on to the next cgroup | |
3439 | */ | |
3440 | next_mz = NULL; | |
bc2f2e7f VD |
3441 | if (!reclaimed) |
3442 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3443 | ||
3e32cb2e | 3444 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3445 | /* |
3446 | * One school of thought says that we should not add | |
3447 | * back the node to the tree if reclaim returns 0. | |
3448 | * But our reclaim could return 0, simply because due | |
3449 | * to priority we are exposing a smaller subset of | |
3450 | * memory to reclaim from. Consider this as a longer | |
3451 | * term TODO. | |
3452 | */ | |
3453 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3454 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3455 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3456 | css_put(&mz->memcg->css); |
3457 | loop++; | |
3458 | /* | |
3459 | * Could not reclaim anything and there are no more | |
3460 | * mem cgroups to try or we seem to be looping without | |
3461 | * reclaiming anything. | |
3462 | */ | |
3463 | if (!nr_reclaimed && | |
3464 | (next_mz == NULL || | |
3465 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3466 | break; | |
3467 | } while (!nr_reclaimed); | |
3468 | if (next_mz) | |
3469 | css_put(&next_mz->memcg->css); | |
3470 | return nr_reclaimed; | |
3471 | } | |
3472 | ||
c26251f9 | 3473 | /* |
51038171 | 3474 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
3475 | * |
3476 | * Caller is responsible for holding css reference for memcg. | |
3477 | */ | |
3478 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3479 | { | |
d977aa93 | 3480 | int nr_retries = MAX_RECLAIM_RETRIES; |
c26251f9 | 3481 | |
c1e862c1 KH |
3482 | /* we call try-to-free pages for make this cgroup empty */ |
3483 | lru_add_drain_all(); | |
d12c60f6 JS |
3484 | |
3485 | drain_all_stock(memcg); | |
3486 | ||
f817ed48 | 3487 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3488 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 3489 | int progress; |
c1e862c1 | 3490 | |
c26251f9 MH |
3491 | if (signal_pending(current)) |
3492 | return -EINTR; | |
3493 | ||
b70a2a21 JW |
3494 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3495 | GFP_KERNEL, true); | |
c1e862c1 | 3496 | if (!progress) { |
f817ed48 | 3497 | nr_retries--; |
c1e862c1 | 3498 | /* maybe some writeback is necessary */ |
8aa7e847 | 3499 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3500 | } |
f817ed48 KH |
3501 | |
3502 | } | |
ab5196c2 MH |
3503 | |
3504 | return 0; | |
cc847582 KH |
3505 | } |
3506 | ||
6770c64e TH |
3507 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3508 | char *buf, size_t nbytes, | |
3509 | loff_t off) | |
c1e862c1 | 3510 | { |
6770c64e | 3511 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3512 | |
d8423011 MH |
3513 | if (mem_cgroup_is_root(memcg)) |
3514 | return -EINVAL; | |
6770c64e | 3515 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3516 | } |
3517 | ||
182446d0 TH |
3518 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3519 | struct cftype *cft) | |
18f59ea7 | 3520 | { |
bef8620c | 3521 | return 1; |
18f59ea7 BS |
3522 | } |
3523 | ||
182446d0 TH |
3524 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3525 | struct cftype *cft, u64 val) | |
18f59ea7 | 3526 | { |
bef8620c | 3527 | if (val == 1) |
0b8f73e1 | 3528 | return 0; |
567fb435 | 3529 | |
bef8620c RG |
3530 | pr_warn_once("Non-hierarchical mode is deprecated. " |
3531 | "Please report your usecase to linux-mm@kvack.org if you " | |
3532 | "depend on this functionality.\n"); | |
567fb435 | 3533 | |
bef8620c | 3534 | return -EINVAL; |
18f59ea7 BS |
3535 | } |
3536 | ||
6f646156 | 3537 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 3538 | { |
42a30035 | 3539 | unsigned long val; |
ce00a967 | 3540 | |
3e32cb2e | 3541 | if (mem_cgroup_is_root(memcg)) { |
0d1c2072 | 3542 | val = memcg_page_state(memcg, NR_FILE_PAGES) + |
be5d0a74 | 3543 | memcg_page_state(memcg, NR_ANON_MAPPED); |
42a30035 JW |
3544 | if (swap) |
3545 | val += memcg_page_state(memcg, MEMCG_SWAP); | |
3e32cb2e | 3546 | } else { |
ce00a967 | 3547 | if (!swap) |
3e32cb2e | 3548 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3549 | else |
3e32cb2e | 3550 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3551 | } |
c12176d3 | 3552 | return val; |
ce00a967 JW |
3553 | } |
3554 | ||
3e32cb2e JW |
3555 | enum { |
3556 | RES_USAGE, | |
3557 | RES_LIMIT, | |
3558 | RES_MAX_USAGE, | |
3559 | RES_FAILCNT, | |
3560 | RES_SOFT_LIMIT, | |
3561 | }; | |
ce00a967 | 3562 | |
791badbd | 3563 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3564 | struct cftype *cft) |
8cdea7c0 | 3565 | { |
182446d0 | 3566 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3567 | struct page_counter *counter; |
af36f906 | 3568 | |
3e32cb2e | 3569 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3570 | case _MEM: |
3e32cb2e JW |
3571 | counter = &memcg->memory; |
3572 | break; | |
8c7c6e34 | 3573 | case _MEMSWAP: |
3e32cb2e JW |
3574 | counter = &memcg->memsw; |
3575 | break; | |
510fc4e1 | 3576 | case _KMEM: |
3e32cb2e | 3577 | counter = &memcg->kmem; |
510fc4e1 | 3578 | break; |
d55f90bf | 3579 | case _TCP: |
0db15298 | 3580 | counter = &memcg->tcpmem; |
d55f90bf | 3581 | break; |
8c7c6e34 KH |
3582 | default: |
3583 | BUG(); | |
8c7c6e34 | 3584 | } |
3e32cb2e JW |
3585 | |
3586 | switch (MEMFILE_ATTR(cft->private)) { | |
3587 | case RES_USAGE: | |
3588 | if (counter == &memcg->memory) | |
c12176d3 | 3589 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 3590 | if (counter == &memcg->memsw) |
c12176d3 | 3591 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
3592 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
3593 | case RES_LIMIT: | |
bbec2e15 | 3594 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
3595 | case RES_MAX_USAGE: |
3596 | return (u64)counter->watermark * PAGE_SIZE; | |
3597 | case RES_FAILCNT: | |
3598 | return counter->failcnt; | |
3599 | case RES_SOFT_LIMIT: | |
3600 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3601 | default: | |
3602 | BUG(); | |
3603 | } | |
8cdea7c0 | 3604 | } |
510fc4e1 | 3605 | |
4a87e2a2 | 3606 | static void memcg_flush_percpu_vmstats(struct mem_cgroup *memcg) |
c350a99e | 3607 | { |
4a87e2a2 | 3608 | unsigned long stat[MEMCG_NR_STAT] = {0}; |
c350a99e RG |
3609 | struct mem_cgroup *mi; |
3610 | int node, cpu, i; | |
c350a99e RG |
3611 | |
3612 | for_each_online_cpu(cpu) | |
4a87e2a2 | 3613 | for (i = 0; i < MEMCG_NR_STAT; i++) |
6c1c2808 | 3614 | stat[i] += per_cpu(memcg->vmstats_percpu->stat[i], cpu); |
c350a99e RG |
3615 | |
3616 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
4a87e2a2 | 3617 | for (i = 0; i < MEMCG_NR_STAT; i++) |
c350a99e RG |
3618 | atomic_long_add(stat[i], &mi->vmstats[i]); |
3619 | ||
3620 | for_each_node(node) { | |
3621 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; | |
3622 | struct mem_cgroup_per_node *pi; | |
3623 | ||
4a87e2a2 | 3624 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
c350a99e RG |
3625 | stat[i] = 0; |
3626 | ||
3627 | for_each_online_cpu(cpu) | |
4a87e2a2 | 3628 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
6c1c2808 SB |
3629 | stat[i] += per_cpu( |
3630 | pn->lruvec_stat_cpu->count[i], cpu); | |
c350a99e RG |
3631 | |
3632 | for (pi = pn; pi; pi = parent_nodeinfo(pi, node)) | |
4a87e2a2 | 3633 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
c350a99e RG |
3634 | atomic_long_add(stat[i], &pi->lruvec_stat[i]); |
3635 | } | |
3636 | } | |
3637 | ||
bb65f89b RG |
3638 | static void memcg_flush_percpu_vmevents(struct mem_cgroup *memcg) |
3639 | { | |
3640 | unsigned long events[NR_VM_EVENT_ITEMS]; | |
3641 | struct mem_cgroup *mi; | |
3642 | int cpu, i; | |
3643 | ||
3644 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
3645 | events[i] = 0; | |
3646 | ||
3647 | for_each_online_cpu(cpu) | |
3648 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
6c1c2808 SB |
3649 | events[i] += per_cpu(memcg->vmstats_percpu->events[i], |
3650 | cpu); | |
bb65f89b RG |
3651 | |
3652 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
3653 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
3654 | atomic_long_add(events[i], &mi->vmevents[i]); | |
3655 | } | |
3656 | ||
84c07d11 | 3657 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 3658 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 3659 | { |
bf4f0599 | 3660 | struct obj_cgroup *objcg; |
d6441637 VD |
3661 | int memcg_id; |
3662 | ||
b313aeee VD |
3663 | if (cgroup_memory_nokmem) |
3664 | return 0; | |
3665 | ||
2a4db7eb | 3666 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 3667 | BUG_ON(memcg->kmem_state); |
d6441637 | 3668 | |
f3bb3043 | 3669 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
3670 | if (memcg_id < 0) |
3671 | return memcg_id; | |
d6441637 | 3672 | |
bf4f0599 RG |
3673 | objcg = obj_cgroup_alloc(); |
3674 | if (!objcg) { | |
3675 | memcg_free_cache_id(memcg_id); | |
3676 | return -ENOMEM; | |
3677 | } | |
3678 | objcg->memcg = memcg; | |
3679 | rcu_assign_pointer(memcg->objcg, objcg); | |
3680 | ||
d648bcc7 RG |
3681 | static_branch_enable(&memcg_kmem_enabled_key); |
3682 | ||
900a38f0 | 3683 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 3684 | memcg->kmem_state = KMEM_ONLINE; |
0b8f73e1 JW |
3685 | |
3686 | return 0; | |
d6441637 VD |
3687 | } |
3688 | ||
8e0a8912 JW |
3689 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
3690 | { | |
3691 | struct cgroup_subsys_state *css; | |
3692 | struct mem_cgroup *parent, *child; | |
3693 | int kmemcg_id; | |
3694 | ||
3695 | if (memcg->kmem_state != KMEM_ONLINE) | |
3696 | return; | |
9855609b | 3697 | |
8e0a8912 JW |
3698 | memcg->kmem_state = KMEM_ALLOCATED; |
3699 | ||
8e0a8912 JW |
3700 | parent = parent_mem_cgroup(memcg); |
3701 | if (!parent) | |
3702 | parent = root_mem_cgroup; | |
3703 | ||
bf4f0599 | 3704 | memcg_reparent_objcgs(memcg, parent); |
fb2f2b0a RG |
3705 | |
3706 | kmemcg_id = memcg->kmemcg_id; | |
3707 | BUG_ON(kmemcg_id < 0); | |
3708 | ||
8e0a8912 JW |
3709 | /* |
3710 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3711 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3712 | * to ones of the parent. After we have finished, all list_lrus | |
3713 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3714 | * ordering is imposed by list_lru_node->lock taken by | |
3715 | * memcg_drain_all_list_lrus(). | |
3716 | */ | |
3a06bb78 | 3717 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
8e0a8912 JW |
3718 | css_for_each_descendant_pre(css, &memcg->css) { |
3719 | child = mem_cgroup_from_css(css); | |
3720 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3721 | child->kmemcg_id = parent->kmemcg_id; | |
8e0a8912 | 3722 | } |
3a06bb78 TH |
3723 | rcu_read_unlock(); |
3724 | ||
9bec5c35 | 3725 | memcg_drain_all_list_lrus(kmemcg_id, parent); |
8e0a8912 JW |
3726 | |
3727 | memcg_free_cache_id(kmemcg_id); | |
3728 | } | |
3729 | ||
3730 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3731 | { | |
0b8f73e1 JW |
3732 | /* css_alloc() failed, offlining didn't happen */ |
3733 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
3734 | memcg_offline_kmem(memcg); | |
8e0a8912 | 3735 | } |
d6441637 | 3736 | #else |
0b8f73e1 | 3737 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
3738 | { |
3739 | return 0; | |
3740 | } | |
3741 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
3742 | { | |
3743 | } | |
3744 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3745 | { | |
3746 | } | |
84c07d11 | 3747 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 3748 | |
bbec2e15 RG |
3749 | static int memcg_update_kmem_max(struct mem_cgroup *memcg, |
3750 | unsigned long max) | |
d6441637 | 3751 | { |
b313aeee | 3752 | int ret; |
127424c8 | 3753 | |
bbec2e15 RG |
3754 | mutex_lock(&memcg_max_mutex); |
3755 | ret = page_counter_set_max(&memcg->kmem, max); | |
3756 | mutex_unlock(&memcg_max_mutex); | |
127424c8 | 3757 | return ret; |
d6441637 | 3758 | } |
510fc4e1 | 3759 | |
bbec2e15 | 3760 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
3761 | { |
3762 | int ret; | |
3763 | ||
bbec2e15 | 3764 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 3765 | |
bbec2e15 | 3766 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
3767 | if (ret) |
3768 | goto out; | |
3769 | ||
0db15298 | 3770 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
3771 | /* |
3772 | * The active flag needs to be written after the static_key | |
3773 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
3774 | * function is the last one to run. See mem_cgroup_sk_alloc() |
3775 | * for details, and note that we don't mark any socket as | |
3776 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
3777 | * |
3778 | * We need to do this, because static_keys will span multiple | |
3779 | * sites, but we can't control their order. If we mark a socket | |
3780 | * as accounted, but the accounting functions are not patched in | |
3781 | * yet, we'll lose accounting. | |
3782 | * | |
2d758073 | 3783 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
3784 | * because when this value change, the code to process it is not |
3785 | * patched in yet. | |
3786 | */ | |
3787 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 3788 | memcg->tcpmem_active = true; |
d55f90bf VD |
3789 | } |
3790 | out: | |
bbec2e15 | 3791 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
3792 | return ret; |
3793 | } | |
d55f90bf | 3794 | |
628f4235 KH |
3795 | /* |
3796 | * The user of this function is... | |
3797 | * RES_LIMIT. | |
3798 | */ | |
451af504 TH |
3799 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3800 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3801 | { |
451af504 | 3802 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3803 | unsigned long nr_pages; |
628f4235 KH |
3804 | int ret; |
3805 | ||
451af504 | 3806 | buf = strstrip(buf); |
650c5e56 | 3807 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3808 | if (ret) |
3809 | return ret; | |
af36f906 | 3810 | |
3e32cb2e | 3811 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3812 | case RES_LIMIT: |
4b3bde4c BS |
3813 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3814 | ret = -EINVAL; | |
3815 | break; | |
3816 | } | |
3e32cb2e JW |
3817 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3818 | case _MEM: | |
bbec2e15 | 3819 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 3820 | break; |
3e32cb2e | 3821 | case _MEMSWAP: |
bbec2e15 | 3822 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 3823 | break; |
3e32cb2e | 3824 | case _KMEM: |
0158115f MH |
3825 | pr_warn_once("kmem.limit_in_bytes is deprecated and will be removed. " |
3826 | "Please report your usecase to linux-mm@kvack.org if you " | |
3827 | "depend on this functionality.\n"); | |
bbec2e15 | 3828 | ret = memcg_update_kmem_max(memcg, nr_pages); |
3e32cb2e | 3829 | break; |
d55f90bf | 3830 | case _TCP: |
bbec2e15 | 3831 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 3832 | break; |
3e32cb2e | 3833 | } |
296c81d8 | 3834 | break; |
3e32cb2e JW |
3835 | case RES_SOFT_LIMIT: |
3836 | memcg->soft_limit = nr_pages; | |
3837 | ret = 0; | |
628f4235 KH |
3838 | break; |
3839 | } | |
451af504 | 3840 | return ret ?: nbytes; |
8cdea7c0 BS |
3841 | } |
3842 | ||
6770c64e TH |
3843 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3844 | size_t nbytes, loff_t off) | |
c84872e1 | 3845 | { |
6770c64e | 3846 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3847 | struct page_counter *counter; |
c84872e1 | 3848 | |
3e32cb2e JW |
3849 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3850 | case _MEM: | |
3851 | counter = &memcg->memory; | |
3852 | break; | |
3853 | case _MEMSWAP: | |
3854 | counter = &memcg->memsw; | |
3855 | break; | |
3856 | case _KMEM: | |
3857 | counter = &memcg->kmem; | |
3858 | break; | |
d55f90bf | 3859 | case _TCP: |
0db15298 | 3860 | counter = &memcg->tcpmem; |
d55f90bf | 3861 | break; |
3e32cb2e JW |
3862 | default: |
3863 | BUG(); | |
3864 | } | |
af36f906 | 3865 | |
3e32cb2e | 3866 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3867 | case RES_MAX_USAGE: |
3e32cb2e | 3868 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3869 | break; |
3870 | case RES_FAILCNT: | |
3e32cb2e | 3871 | counter->failcnt = 0; |
29f2a4da | 3872 | break; |
3e32cb2e JW |
3873 | default: |
3874 | BUG(); | |
29f2a4da | 3875 | } |
f64c3f54 | 3876 | |
6770c64e | 3877 | return nbytes; |
c84872e1 PE |
3878 | } |
3879 | ||
182446d0 | 3880 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3881 | struct cftype *cft) |
3882 | { | |
182446d0 | 3883 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3884 | } |
3885 | ||
02491447 | 3886 | #ifdef CONFIG_MMU |
182446d0 | 3887 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3888 | struct cftype *cft, u64 val) |
3889 | { | |
182446d0 | 3890 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3891 | |
1dfab5ab | 3892 | if (val & ~MOVE_MASK) |
7dc74be0 | 3893 | return -EINVAL; |
ee5e8472 | 3894 | |
7dc74be0 | 3895 | /* |
ee5e8472 GC |
3896 | * No kind of locking is needed in here, because ->can_attach() will |
3897 | * check this value once in the beginning of the process, and then carry | |
3898 | * on with stale data. This means that changes to this value will only | |
3899 | * affect task migrations starting after the change. | |
7dc74be0 | 3900 | */ |
c0ff4b85 | 3901 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3902 | return 0; |
3903 | } | |
02491447 | 3904 | #else |
182446d0 | 3905 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3906 | struct cftype *cft, u64 val) |
3907 | { | |
3908 | return -ENOSYS; | |
3909 | } | |
3910 | #endif | |
7dc74be0 | 3911 | |
406eb0c9 | 3912 | #ifdef CONFIG_NUMA |
113b7dfd JW |
3913 | |
3914 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
3915 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
3916 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) | |
3917 | ||
3918 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 | 3919 | int nid, unsigned int lru_mask, bool tree) |
113b7dfd | 3920 | { |
867e5e1d | 3921 | struct lruvec *lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); |
113b7dfd JW |
3922 | unsigned long nr = 0; |
3923 | enum lru_list lru; | |
3924 | ||
3925 | VM_BUG_ON((unsigned)nid >= nr_node_ids); | |
3926 | ||
3927 | for_each_lru(lru) { | |
3928 | if (!(BIT(lru) & lru_mask)) | |
3929 | continue; | |
dd8657b6 SB |
3930 | if (tree) |
3931 | nr += lruvec_page_state(lruvec, NR_LRU_BASE + lru); | |
3932 | else | |
3933 | nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru); | |
113b7dfd JW |
3934 | } |
3935 | return nr; | |
3936 | } | |
3937 | ||
3938 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 SB |
3939 | unsigned int lru_mask, |
3940 | bool tree) | |
113b7dfd JW |
3941 | { |
3942 | unsigned long nr = 0; | |
3943 | enum lru_list lru; | |
3944 | ||
3945 | for_each_lru(lru) { | |
3946 | if (!(BIT(lru) & lru_mask)) | |
3947 | continue; | |
dd8657b6 SB |
3948 | if (tree) |
3949 | nr += memcg_page_state(memcg, NR_LRU_BASE + lru); | |
3950 | else | |
3951 | nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru); | |
113b7dfd JW |
3952 | } |
3953 | return nr; | |
3954 | } | |
3955 | ||
2da8ca82 | 3956 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3957 | { |
25485de6 GT |
3958 | struct numa_stat { |
3959 | const char *name; | |
3960 | unsigned int lru_mask; | |
3961 | }; | |
3962 | ||
3963 | static const struct numa_stat stats[] = { | |
3964 | { "total", LRU_ALL }, | |
3965 | { "file", LRU_ALL_FILE }, | |
3966 | { "anon", LRU_ALL_ANON }, | |
3967 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3968 | }; | |
3969 | const struct numa_stat *stat; | |
406eb0c9 | 3970 | int nid; |
aa9694bb | 3971 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
406eb0c9 | 3972 | |
25485de6 | 3973 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
3974 | seq_printf(m, "%s=%lu", stat->name, |
3975 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
3976 | false)); | |
3977 | for_each_node_state(nid, N_MEMORY) | |
3978 | seq_printf(m, " N%d=%lu", nid, | |
3979 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3980 | stat->lru_mask, false)); | |
25485de6 | 3981 | seq_putc(m, '\n'); |
406eb0c9 | 3982 | } |
406eb0c9 | 3983 | |
071aee13 | 3984 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
3985 | |
3986 | seq_printf(m, "hierarchical_%s=%lu", stat->name, | |
3987 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
3988 | true)); | |
3989 | for_each_node_state(nid, N_MEMORY) | |
3990 | seq_printf(m, " N%d=%lu", nid, | |
3991 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3992 | stat->lru_mask, true)); | |
071aee13 | 3993 | seq_putc(m, '\n'); |
406eb0c9 | 3994 | } |
406eb0c9 | 3995 | |
406eb0c9 YH |
3996 | return 0; |
3997 | } | |
3998 | #endif /* CONFIG_NUMA */ | |
3999 | ||
c8713d0b | 4000 | static const unsigned int memcg1_stats[] = { |
0d1c2072 | 4001 | NR_FILE_PAGES, |
be5d0a74 | 4002 | NR_ANON_MAPPED, |
468c3982 JW |
4003 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4004 | NR_ANON_THPS, | |
4005 | #endif | |
c8713d0b JW |
4006 | NR_SHMEM, |
4007 | NR_FILE_MAPPED, | |
4008 | NR_FILE_DIRTY, | |
4009 | NR_WRITEBACK, | |
4010 | MEMCG_SWAP, | |
4011 | }; | |
4012 | ||
4013 | static const char *const memcg1_stat_names[] = { | |
4014 | "cache", | |
4015 | "rss", | |
468c3982 | 4016 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
c8713d0b | 4017 | "rss_huge", |
468c3982 | 4018 | #endif |
c8713d0b JW |
4019 | "shmem", |
4020 | "mapped_file", | |
4021 | "dirty", | |
4022 | "writeback", | |
4023 | "swap", | |
4024 | }; | |
4025 | ||
df0e53d0 | 4026 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 4027 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
4028 | PGPGIN, |
4029 | PGPGOUT, | |
4030 | PGFAULT, | |
4031 | PGMAJFAULT, | |
4032 | }; | |
4033 | ||
2da8ca82 | 4034 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 4035 | { |
aa9694bb | 4036 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3e32cb2e | 4037 | unsigned long memory, memsw; |
af7c4b0e JW |
4038 | struct mem_cgroup *mi; |
4039 | unsigned int i; | |
406eb0c9 | 4040 | |
71cd3113 | 4041 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c | 4042 | |
71cd3113 | 4043 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
468c3982 JW |
4044 | unsigned long nr; |
4045 | ||
71cd3113 | 4046 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 4047 | continue; |
468c3982 | 4048 | nr = memcg_page_state_local(memcg, memcg1_stats[i]); |
468c3982 | 4049 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], nr * PAGE_SIZE); |
1dd3a273 | 4050 | } |
7b854121 | 4051 | |
df0e53d0 | 4052 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d | 4053 | seq_printf(m, "%s %lu\n", vm_event_name(memcg1_events[i]), |
205b20cc | 4054 | memcg_events_local(memcg, memcg1_events[i])); |
af7c4b0e JW |
4055 | |
4056 | for (i = 0; i < NR_LRU_LISTS; i++) | |
ebc5d83d | 4057 | seq_printf(m, "%s %lu\n", lru_list_name(i), |
205b20cc | 4058 | memcg_page_state_local(memcg, NR_LRU_BASE + i) * |
21d89d15 | 4059 | PAGE_SIZE); |
af7c4b0e | 4060 | |
14067bb3 | 4061 | /* Hierarchical information */ |
3e32cb2e JW |
4062 | memory = memsw = PAGE_COUNTER_MAX; |
4063 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
15b42562 CD |
4064 | memory = min(memory, READ_ONCE(mi->memory.max)); |
4065 | memsw = min(memsw, READ_ONCE(mi->memsw.max)); | |
fee7b548 | 4066 | } |
3e32cb2e JW |
4067 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
4068 | (u64)memory * PAGE_SIZE); | |
7941d214 | 4069 | if (do_memsw_account()) |
3e32cb2e JW |
4070 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
4071 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 4072 | |
8de7ecc6 | 4073 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
7de2e9f1 | 4074 | unsigned long nr; |
4075 | ||
71cd3113 | 4076 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 4077 | continue; |
7de2e9f1 | 4078 | nr = memcg_page_state(memcg, memcg1_stats[i]); |
8de7ecc6 | 4079 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], |
7de2e9f1 | 4080 | (u64)nr * PAGE_SIZE); |
af7c4b0e JW |
4081 | } |
4082 | ||
8de7ecc6 | 4083 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d KK |
4084 | seq_printf(m, "total_%s %llu\n", |
4085 | vm_event_name(memcg1_events[i]), | |
dd923990 | 4086 | (u64)memcg_events(memcg, memcg1_events[i])); |
af7c4b0e | 4087 | |
8de7ecc6 | 4088 | for (i = 0; i < NR_LRU_LISTS; i++) |
ebc5d83d | 4089 | seq_printf(m, "total_%s %llu\n", lru_list_name(i), |
42a30035 JW |
4090 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
4091 | PAGE_SIZE); | |
14067bb3 | 4092 | |
7f016ee8 | 4093 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 4094 | { |
ef8f2327 MG |
4095 | pg_data_t *pgdat; |
4096 | struct mem_cgroup_per_node *mz; | |
1431d4d1 JW |
4097 | unsigned long anon_cost = 0; |
4098 | unsigned long file_cost = 0; | |
7f016ee8 | 4099 | |
ef8f2327 MG |
4100 | for_each_online_pgdat(pgdat) { |
4101 | mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); | |
7f016ee8 | 4102 | |
1431d4d1 JW |
4103 | anon_cost += mz->lruvec.anon_cost; |
4104 | file_cost += mz->lruvec.file_cost; | |
ef8f2327 | 4105 | } |
1431d4d1 JW |
4106 | seq_printf(m, "anon_cost %lu\n", anon_cost); |
4107 | seq_printf(m, "file_cost %lu\n", file_cost); | |
7f016ee8 KM |
4108 | } |
4109 | #endif | |
4110 | ||
d2ceb9b7 KH |
4111 | return 0; |
4112 | } | |
4113 | ||
182446d0 TH |
4114 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
4115 | struct cftype *cft) | |
a7885eb8 | 4116 | { |
182446d0 | 4117 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4118 | |
1f4c025b | 4119 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4120 | } |
4121 | ||
182446d0 TH |
4122 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
4123 | struct cftype *cft, u64 val) | |
a7885eb8 | 4124 | { |
182446d0 | 4125 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4126 | |
3dae7fec | 4127 | if (val > 100) |
a7885eb8 KM |
4128 | return -EINVAL; |
4129 | ||
a4792030 | 4130 | if (!mem_cgroup_is_root(memcg)) |
3dae7fec JW |
4131 | memcg->swappiness = val; |
4132 | else | |
4133 | vm_swappiness = val; | |
068b38c1 | 4134 | |
a7885eb8 KM |
4135 | return 0; |
4136 | } | |
4137 | ||
2e72b634 KS |
4138 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4139 | { | |
4140 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 4141 | unsigned long usage; |
2e72b634 KS |
4142 | int i; |
4143 | ||
4144 | rcu_read_lock(); | |
4145 | if (!swap) | |
2c488db2 | 4146 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4147 | else |
2c488db2 | 4148 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4149 | |
4150 | if (!t) | |
4151 | goto unlock; | |
4152 | ||
ce00a967 | 4153 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
4154 | |
4155 | /* | |
748dad36 | 4156 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4157 | * If it's not true, a threshold was crossed after last |
4158 | * call of __mem_cgroup_threshold(). | |
4159 | */ | |
5407a562 | 4160 | i = t->current_threshold; |
2e72b634 KS |
4161 | |
4162 | /* | |
4163 | * Iterate backward over array of thresholds starting from | |
4164 | * current_threshold and check if a threshold is crossed. | |
4165 | * If none of thresholds below usage is crossed, we read | |
4166 | * only one element of the array here. | |
4167 | */ | |
4168 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4169 | eventfd_signal(t->entries[i].eventfd, 1); | |
4170 | ||
4171 | /* i = current_threshold + 1 */ | |
4172 | i++; | |
4173 | ||
4174 | /* | |
4175 | * Iterate forward over array of thresholds starting from | |
4176 | * current_threshold+1 and check if a threshold is crossed. | |
4177 | * If none of thresholds above usage is crossed, we read | |
4178 | * only one element of the array here. | |
4179 | */ | |
4180 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4181 | eventfd_signal(t->entries[i].eventfd, 1); | |
4182 | ||
4183 | /* Update current_threshold */ | |
5407a562 | 4184 | t->current_threshold = i - 1; |
2e72b634 KS |
4185 | unlock: |
4186 | rcu_read_unlock(); | |
4187 | } | |
4188 | ||
4189 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4190 | { | |
ad4ca5f4 KS |
4191 | while (memcg) { |
4192 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 4193 | if (do_memsw_account()) |
ad4ca5f4 KS |
4194 | __mem_cgroup_threshold(memcg, true); |
4195 | ||
4196 | memcg = parent_mem_cgroup(memcg); | |
4197 | } | |
2e72b634 KS |
4198 | } |
4199 | ||
4200 | static int compare_thresholds(const void *a, const void *b) | |
4201 | { | |
4202 | const struct mem_cgroup_threshold *_a = a; | |
4203 | const struct mem_cgroup_threshold *_b = b; | |
4204 | ||
2bff24a3 GT |
4205 | if (_a->threshold > _b->threshold) |
4206 | return 1; | |
4207 | ||
4208 | if (_a->threshold < _b->threshold) | |
4209 | return -1; | |
4210 | ||
4211 | return 0; | |
2e72b634 KS |
4212 | } |
4213 | ||
c0ff4b85 | 4214 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4215 | { |
4216 | struct mem_cgroup_eventfd_list *ev; | |
4217 | ||
2bcf2e92 MH |
4218 | spin_lock(&memcg_oom_lock); |
4219 | ||
c0ff4b85 | 4220 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 4221 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
4222 | |
4223 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
4224 | return 0; |
4225 | } | |
4226 | ||
c0ff4b85 | 4227 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4228 | { |
7d74b06f KH |
4229 | struct mem_cgroup *iter; |
4230 | ||
c0ff4b85 | 4231 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4232 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4233 | } |
4234 | ||
59b6f873 | 4235 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4236 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4237 | { |
2c488db2 KS |
4238 | struct mem_cgroup_thresholds *thresholds; |
4239 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
4240 | unsigned long threshold; |
4241 | unsigned long usage; | |
2c488db2 | 4242 | int i, size, ret; |
2e72b634 | 4243 | |
650c5e56 | 4244 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
4245 | if (ret) |
4246 | return ret; | |
4247 | ||
4248 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4249 | |
05b84301 | 4250 | if (type == _MEM) { |
2c488db2 | 4251 | thresholds = &memcg->thresholds; |
ce00a967 | 4252 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4253 | } else if (type == _MEMSWAP) { |
2c488db2 | 4254 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4255 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4256 | } else |
2e72b634 KS |
4257 | BUG(); |
4258 | ||
2e72b634 | 4259 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4260 | if (thresholds->primary) |
2e72b634 KS |
4261 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4262 | ||
2c488db2 | 4263 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4264 | |
4265 | /* Allocate memory for new array of thresholds */ | |
67b8046f | 4266 | new = kmalloc(struct_size(new, entries, size), GFP_KERNEL); |
2c488db2 | 4267 | if (!new) { |
2e72b634 KS |
4268 | ret = -ENOMEM; |
4269 | goto unlock; | |
4270 | } | |
2c488db2 | 4271 | new->size = size; |
2e72b634 KS |
4272 | |
4273 | /* Copy thresholds (if any) to new array */ | |
e90342e6 GS |
4274 | if (thresholds->primary) |
4275 | memcpy(new->entries, thresholds->primary->entries, | |
4276 | flex_array_size(new, entries, size - 1)); | |
2c488db2 | 4277 | |
2e72b634 | 4278 | /* Add new threshold */ |
2c488db2 KS |
4279 | new->entries[size - 1].eventfd = eventfd; |
4280 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4281 | |
4282 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
61e604e6 | 4283 | sort(new->entries, size, sizeof(*new->entries), |
2e72b634 KS |
4284 | compare_thresholds, NULL); |
4285 | ||
4286 | /* Find current threshold */ | |
2c488db2 | 4287 | new->current_threshold = -1; |
2e72b634 | 4288 | for (i = 0; i < size; i++) { |
748dad36 | 4289 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4290 | /* |
2c488db2 KS |
4291 | * new->current_threshold will not be used until |
4292 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4293 | * it here. |
4294 | */ | |
2c488db2 | 4295 | ++new->current_threshold; |
748dad36 SZ |
4296 | } else |
4297 | break; | |
2e72b634 KS |
4298 | } |
4299 | ||
2c488db2 KS |
4300 | /* Free old spare buffer and save old primary buffer as spare */ |
4301 | kfree(thresholds->spare); | |
4302 | thresholds->spare = thresholds->primary; | |
4303 | ||
4304 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4305 | |
907860ed | 4306 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4307 | synchronize_rcu(); |
4308 | ||
2e72b634 KS |
4309 | unlock: |
4310 | mutex_unlock(&memcg->thresholds_lock); | |
4311 | ||
4312 | return ret; | |
4313 | } | |
4314 | ||
59b6f873 | 4315 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4316 | struct eventfd_ctx *eventfd, const char *args) |
4317 | { | |
59b6f873 | 4318 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4319 | } |
4320 | ||
59b6f873 | 4321 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4322 | struct eventfd_ctx *eventfd, const char *args) |
4323 | { | |
59b6f873 | 4324 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4325 | } |
4326 | ||
59b6f873 | 4327 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4328 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4329 | { |
2c488db2 KS |
4330 | struct mem_cgroup_thresholds *thresholds; |
4331 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 4332 | unsigned long usage; |
7d36665a | 4333 | int i, j, size, entries; |
2e72b634 KS |
4334 | |
4335 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4336 | |
4337 | if (type == _MEM) { | |
2c488db2 | 4338 | thresholds = &memcg->thresholds; |
ce00a967 | 4339 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4340 | } else if (type == _MEMSWAP) { |
2c488db2 | 4341 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4342 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4343 | } else |
2e72b634 KS |
4344 | BUG(); |
4345 | ||
371528ca AV |
4346 | if (!thresholds->primary) |
4347 | goto unlock; | |
4348 | ||
2e72b634 KS |
4349 | /* Check if a threshold crossed before removing */ |
4350 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4351 | ||
4352 | /* Calculate new number of threshold */ | |
7d36665a | 4353 | size = entries = 0; |
2c488db2 KS |
4354 | for (i = 0; i < thresholds->primary->size; i++) { |
4355 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 | 4356 | size++; |
7d36665a CX |
4357 | else |
4358 | entries++; | |
2e72b634 KS |
4359 | } |
4360 | ||
2c488db2 | 4361 | new = thresholds->spare; |
907860ed | 4362 | |
7d36665a CX |
4363 | /* If no items related to eventfd have been cleared, nothing to do */ |
4364 | if (!entries) | |
4365 | goto unlock; | |
4366 | ||
2e72b634 KS |
4367 | /* Set thresholds array to NULL if we don't have thresholds */ |
4368 | if (!size) { | |
2c488db2 KS |
4369 | kfree(new); |
4370 | new = NULL; | |
907860ed | 4371 | goto swap_buffers; |
2e72b634 KS |
4372 | } |
4373 | ||
2c488db2 | 4374 | new->size = size; |
2e72b634 KS |
4375 | |
4376 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4377 | new->current_threshold = -1; |
4378 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4379 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4380 | continue; |
4381 | ||
2c488db2 | 4382 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4383 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4384 | /* |
2c488db2 | 4385 | * new->current_threshold will not be used |
2e72b634 KS |
4386 | * until rcu_assign_pointer(), so it's safe to increment |
4387 | * it here. | |
4388 | */ | |
2c488db2 | 4389 | ++new->current_threshold; |
2e72b634 KS |
4390 | } |
4391 | j++; | |
4392 | } | |
4393 | ||
907860ed | 4394 | swap_buffers: |
2c488db2 KS |
4395 | /* Swap primary and spare array */ |
4396 | thresholds->spare = thresholds->primary; | |
8c757763 | 4397 | |
2c488db2 | 4398 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4399 | |
907860ed | 4400 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4401 | synchronize_rcu(); |
6611d8d7 MC |
4402 | |
4403 | /* If all events are unregistered, free the spare array */ | |
4404 | if (!new) { | |
4405 | kfree(thresholds->spare); | |
4406 | thresholds->spare = NULL; | |
4407 | } | |
371528ca | 4408 | unlock: |
2e72b634 | 4409 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4410 | } |
c1e862c1 | 4411 | |
59b6f873 | 4412 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4413 | struct eventfd_ctx *eventfd) |
4414 | { | |
59b6f873 | 4415 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4416 | } |
4417 | ||
59b6f873 | 4418 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4419 | struct eventfd_ctx *eventfd) |
4420 | { | |
59b6f873 | 4421 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4422 | } |
4423 | ||
59b6f873 | 4424 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4425 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4426 | { |
9490ff27 | 4427 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4428 | |
9490ff27 KH |
4429 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4430 | if (!event) | |
4431 | return -ENOMEM; | |
4432 | ||
1af8efe9 | 4433 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4434 | |
4435 | event->eventfd = eventfd; | |
4436 | list_add(&event->list, &memcg->oom_notify); | |
4437 | ||
4438 | /* already in OOM ? */ | |
c2b42d3c | 4439 | if (memcg->under_oom) |
9490ff27 | 4440 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4441 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4442 | |
4443 | return 0; | |
4444 | } | |
4445 | ||
59b6f873 | 4446 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4447 | struct eventfd_ctx *eventfd) |
9490ff27 | 4448 | { |
9490ff27 | 4449 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4450 | |
1af8efe9 | 4451 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4452 | |
c0ff4b85 | 4453 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4454 | if (ev->eventfd == eventfd) { |
4455 | list_del(&ev->list); | |
4456 | kfree(ev); | |
4457 | } | |
4458 | } | |
4459 | ||
1af8efe9 | 4460 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4461 | } |
4462 | ||
2da8ca82 | 4463 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4464 | { |
aa9694bb | 4465 | struct mem_cgroup *memcg = mem_cgroup_from_seq(sf); |
3c11ecf4 | 4466 | |
791badbd | 4467 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 4468 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
4469 | seq_printf(sf, "oom_kill %lu\n", |
4470 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
4471 | return 0; |
4472 | } | |
4473 | ||
182446d0 | 4474 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4475 | struct cftype *cft, u64 val) |
4476 | { | |
182446d0 | 4477 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4478 | |
4479 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
a4792030 | 4480 | if (mem_cgroup_is_root(memcg) || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4481 | return -EINVAL; |
4482 | ||
c0ff4b85 | 4483 | memcg->oom_kill_disable = val; |
4d845ebf | 4484 | if (!val) |
c0ff4b85 | 4485 | memcg_oom_recover(memcg); |
3dae7fec | 4486 | |
3c11ecf4 KH |
4487 | return 0; |
4488 | } | |
4489 | ||
52ebea74 TH |
4490 | #ifdef CONFIG_CGROUP_WRITEBACK |
4491 | ||
3a8e9ac8 TH |
4492 | #include <trace/events/writeback.h> |
4493 | ||
841710aa TH |
4494 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
4495 | { | |
4496 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
4497 | } | |
4498 | ||
4499 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4500 | { | |
4501 | wb_domain_exit(&memcg->cgwb_domain); | |
4502 | } | |
4503 | ||
2529bb3a TH |
4504 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4505 | { | |
4506 | wb_domain_size_changed(&memcg->cgwb_domain); | |
4507 | } | |
4508 | ||
841710aa TH |
4509 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
4510 | { | |
4511 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4512 | ||
4513 | if (!memcg->css.parent) | |
4514 | return NULL; | |
4515 | ||
4516 | return &memcg->cgwb_domain; | |
4517 | } | |
4518 | ||
0b3d6e6f GT |
4519 | /* |
4520 | * idx can be of type enum memcg_stat_item or node_stat_item. | |
4521 | * Keep in sync with memcg_exact_page(). | |
4522 | */ | |
4523 | static unsigned long memcg_exact_page_state(struct mem_cgroup *memcg, int idx) | |
4524 | { | |
871789d4 | 4525 | long x = atomic_long_read(&memcg->vmstats[idx]); |
0b3d6e6f GT |
4526 | int cpu; |
4527 | ||
4528 | for_each_online_cpu(cpu) | |
871789d4 | 4529 | x += per_cpu_ptr(memcg->vmstats_percpu, cpu)->stat[idx]; |
0b3d6e6f GT |
4530 | if (x < 0) |
4531 | x = 0; | |
4532 | return x; | |
4533 | } | |
4534 | ||
c2aa723a TH |
4535 | /** |
4536 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
4537 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
4538 | * @pfilepages: out parameter for number of file pages |
4539 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
4540 | * @pdirty: out parameter for number of dirty pages |
4541 | * @pwriteback: out parameter for number of pages under writeback | |
4542 | * | |
c5edf9cd TH |
4543 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
4544 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
4545 | * is a bit more involved. | |
c2aa723a | 4546 | * |
c5edf9cd TH |
4547 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
4548 | * headroom is calculated as the lowest headroom of itself and the | |
4549 | * ancestors. Note that this doesn't consider the actual amount of | |
4550 | * available memory in the system. The caller should further cap | |
4551 | * *@pheadroom accordingly. | |
c2aa723a | 4552 | */ |
c5edf9cd TH |
4553 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
4554 | unsigned long *pheadroom, unsigned long *pdirty, | |
4555 | unsigned long *pwriteback) | |
c2aa723a TH |
4556 | { |
4557 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4558 | struct mem_cgroup *parent; | |
c2aa723a | 4559 | |
0b3d6e6f | 4560 | *pdirty = memcg_exact_page_state(memcg, NR_FILE_DIRTY); |
c2aa723a | 4561 | |
0b3d6e6f | 4562 | *pwriteback = memcg_exact_page_state(memcg, NR_WRITEBACK); |
21d89d15 JW |
4563 | *pfilepages = memcg_exact_page_state(memcg, NR_INACTIVE_FILE) + |
4564 | memcg_exact_page_state(memcg, NR_ACTIVE_FILE); | |
c5edf9cd | 4565 | *pheadroom = PAGE_COUNTER_MAX; |
c2aa723a | 4566 | |
c2aa723a | 4567 | while ((parent = parent_mem_cgroup(memcg))) { |
15b42562 | 4568 | unsigned long ceiling = min(READ_ONCE(memcg->memory.max), |
d1663a90 | 4569 | READ_ONCE(memcg->memory.high)); |
c2aa723a TH |
4570 | unsigned long used = page_counter_read(&memcg->memory); |
4571 | ||
c5edf9cd | 4572 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
4573 | memcg = parent; |
4574 | } | |
c2aa723a TH |
4575 | } |
4576 | ||
97b27821 TH |
4577 | /* |
4578 | * Foreign dirty flushing | |
4579 | * | |
4580 | * There's an inherent mismatch between memcg and writeback. The former | |
4581 | * trackes ownership per-page while the latter per-inode. This was a | |
4582 | * deliberate design decision because honoring per-page ownership in the | |
4583 | * writeback path is complicated, may lead to higher CPU and IO overheads | |
4584 | * and deemed unnecessary given that write-sharing an inode across | |
4585 | * different cgroups isn't a common use-case. | |
4586 | * | |
4587 | * Combined with inode majority-writer ownership switching, this works well | |
4588 | * enough in most cases but there are some pathological cases. For | |
4589 | * example, let's say there are two cgroups A and B which keep writing to | |
4590 | * different but confined parts of the same inode. B owns the inode and | |
4591 | * A's memory is limited far below B's. A's dirty ratio can rise enough to | |
4592 | * trigger balance_dirty_pages() sleeps but B's can be low enough to avoid | |
4593 | * triggering background writeback. A will be slowed down without a way to | |
4594 | * make writeback of the dirty pages happen. | |
4595 | * | |
4596 | * Conditions like the above can lead to a cgroup getting repatedly and | |
4597 | * severely throttled after making some progress after each | |
4598 | * dirty_expire_interval while the underyling IO device is almost | |
4599 | * completely idle. | |
4600 | * | |
4601 | * Solving this problem completely requires matching the ownership tracking | |
4602 | * granularities between memcg and writeback in either direction. However, | |
4603 | * the more egregious behaviors can be avoided by simply remembering the | |
4604 | * most recent foreign dirtying events and initiating remote flushes on | |
4605 | * them when local writeback isn't enough to keep the memory clean enough. | |
4606 | * | |
4607 | * The following two functions implement such mechanism. When a foreign | |
4608 | * page - a page whose memcg and writeback ownerships don't match - is | |
4609 | * dirtied, mem_cgroup_track_foreign_dirty() records the inode owning | |
4610 | * bdi_writeback on the page owning memcg. When balance_dirty_pages() | |
4611 | * decides that the memcg needs to sleep due to high dirty ratio, it calls | |
4612 | * mem_cgroup_flush_foreign() which queues writeback on the recorded | |
4613 | * foreign bdi_writebacks which haven't expired. Both the numbers of | |
4614 | * recorded bdi_writebacks and concurrent in-flight foreign writebacks are | |
4615 | * limited to MEMCG_CGWB_FRN_CNT. | |
4616 | * | |
4617 | * The mechanism only remembers IDs and doesn't hold any object references. | |
4618 | * As being wrong occasionally doesn't matter, updates and accesses to the | |
4619 | * records are lockless and racy. | |
4620 | */ | |
4621 | void mem_cgroup_track_foreign_dirty_slowpath(struct page *page, | |
4622 | struct bdi_writeback *wb) | |
4623 | { | |
bcfe06bf | 4624 | struct mem_cgroup *memcg = page_memcg(page); |
97b27821 TH |
4625 | struct memcg_cgwb_frn *frn; |
4626 | u64 now = get_jiffies_64(); | |
4627 | u64 oldest_at = now; | |
4628 | int oldest = -1; | |
4629 | int i; | |
4630 | ||
3a8e9ac8 TH |
4631 | trace_track_foreign_dirty(page, wb); |
4632 | ||
97b27821 TH |
4633 | /* |
4634 | * Pick the slot to use. If there is already a slot for @wb, keep | |
4635 | * using it. If not replace the oldest one which isn't being | |
4636 | * written out. | |
4637 | */ | |
4638 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4639 | frn = &memcg->cgwb_frn[i]; | |
4640 | if (frn->bdi_id == wb->bdi->id && | |
4641 | frn->memcg_id == wb->memcg_css->id) | |
4642 | break; | |
4643 | if (time_before64(frn->at, oldest_at) && | |
4644 | atomic_read(&frn->done.cnt) == 1) { | |
4645 | oldest = i; | |
4646 | oldest_at = frn->at; | |
4647 | } | |
4648 | } | |
4649 | ||
4650 | if (i < MEMCG_CGWB_FRN_CNT) { | |
4651 | /* | |
4652 | * Re-using an existing one. Update timestamp lazily to | |
4653 | * avoid making the cacheline hot. We want them to be | |
4654 | * reasonably up-to-date and significantly shorter than | |
4655 | * dirty_expire_interval as that's what expires the record. | |
4656 | * Use the shorter of 1s and dirty_expire_interval / 8. | |
4657 | */ | |
4658 | unsigned long update_intv = | |
4659 | min_t(unsigned long, HZ, | |
4660 | msecs_to_jiffies(dirty_expire_interval * 10) / 8); | |
4661 | ||
4662 | if (time_before64(frn->at, now - update_intv)) | |
4663 | frn->at = now; | |
4664 | } else if (oldest >= 0) { | |
4665 | /* replace the oldest free one */ | |
4666 | frn = &memcg->cgwb_frn[oldest]; | |
4667 | frn->bdi_id = wb->bdi->id; | |
4668 | frn->memcg_id = wb->memcg_css->id; | |
4669 | frn->at = now; | |
4670 | } | |
4671 | } | |
4672 | ||
4673 | /* issue foreign writeback flushes for recorded foreign dirtying events */ | |
4674 | void mem_cgroup_flush_foreign(struct bdi_writeback *wb) | |
4675 | { | |
4676 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4677 | unsigned long intv = msecs_to_jiffies(dirty_expire_interval * 10); | |
4678 | u64 now = jiffies_64; | |
4679 | int i; | |
4680 | ||
4681 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4682 | struct memcg_cgwb_frn *frn = &memcg->cgwb_frn[i]; | |
4683 | ||
4684 | /* | |
4685 | * If the record is older than dirty_expire_interval, | |
4686 | * writeback on it has already started. No need to kick it | |
4687 | * off again. Also, don't start a new one if there's | |
4688 | * already one in flight. | |
4689 | */ | |
4690 | if (time_after64(frn->at, now - intv) && | |
4691 | atomic_read(&frn->done.cnt) == 1) { | |
4692 | frn->at = 0; | |
3a8e9ac8 | 4693 | trace_flush_foreign(wb, frn->bdi_id, frn->memcg_id); |
97b27821 TH |
4694 | cgroup_writeback_by_id(frn->bdi_id, frn->memcg_id, 0, |
4695 | WB_REASON_FOREIGN_FLUSH, | |
4696 | &frn->done); | |
4697 | } | |
4698 | } | |
4699 | } | |
4700 | ||
841710aa TH |
4701 | #else /* CONFIG_CGROUP_WRITEBACK */ |
4702 | ||
4703 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
4704 | { | |
4705 | return 0; | |
4706 | } | |
4707 | ||
4708 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4709 | { | |
4710 | } | |
4711 | ||
2529bb3a TH |
4712 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4713 | { | |
4714 | } | |
4715 | ||
52ebea74 TH |
4716 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
4717 | ||
3bc942f3 TH |
4718 | /* |
4719 | * DO NOT USE IN NEW FILES. | |
4720 | * | |
4721 | * "cgroup.event_control" implementation. | |
4722 | * | |
4723 | * This is way over-engineered. It tries to support fully configurable | |
4724 | * events for each user. Such level of flexibility is completely | |
4725 | * unnecessary especially in the light of the planned unified hierarchy. | |
4726 | * | |
4727 | * Please deprecate this and replace with something simpler if at all | |
4728 | * possible. | |
4729 | */ | |
4730 | ||
79bd9814 TH |
4731 | /* |
4732 | * Unregister event and free resources. | |
4733 | * | |
4734 | * Gets called from workqueue. | |
4735 | */ | |
3bc942f3 | 4736 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4737 | { |
3bc942f3 TH |
4738 | struct mem_cgroup_event *event = |
4739 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4740 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4741 | |
4742 | remove_wait_queue(event->wqh, &event->wait); | |
4743 | ||
59b6f873 | 4744 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4745 | |
4746 | /* Notify userspace the event is going away. */ | |
4747 | eventfd_signal(event->eventfd, 1); | |
4748 | ||
4749 | eventfd_ctx_put(event->eventfd); | |
4750 | kfree(event); | |
59b6f873 | 4751 | css_put(&memcg->css); |
79bd9814 TH |
4752 | } |
4753 | ||
4754 | /* | |
a9a08845 | 4755 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
4756 | * |
4757 | * Called with wqh->lock held and interrupts disabled. | |
4758 | */ | |
ac6424b9 | 4759 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 4760 | int sync, void *key) |
79bd9814 | 4761 | { |
3bc942f3 TH |
4762 | struct mem_cgroup_event *event = |
4763 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4764 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 4765 | __poll_t flags = key_to_poll(key); |
79bd9814 | 4766 | |
a9a08845 | 4767 | if (flags & EPOLLHUP) { |
79bd9814 TH |
4768 | /* |
4769 | * If the event has been detached at cgroup removal, we | |
4770 | * can simply return knowing the other side will cleanup | |
4771 | * for us. | |
4772 | * | |
4773 | * We can't race against event freeing since the other | |
4774 | * side will require wqh->lock via remove_wait_queue(), | |
4775 | * which we hold. | |
4776 | */ | |
fba94807 | 4777 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4778 | if (!list_empty(&event->list)) { |
4779 | list_del_init(&event->list); | |
4780 | /* | |
4781 | * We are in atomic context, but cgroup_event_remove() | |
4782 | * may sleep, so we have to call it in workqueue. | |
4783 | */ | |
4784 | schedule_work(&event->remove); | |
4785 | } | |
fba94807 | 4786 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4787 | } |
4788 | ||
4789 | return 0; | |
4790 | } | |
4791 | ||
3bc942f3 | 4792 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4793 | wait_queue_head_t *wqh, poll_table *pt) |
4794 | { | |
3bc942f3 TH |
4795 | struct mem_cgroup_event *event = |
4796 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4797 | |
4798 | event->wqh = wqh; | |
4799 | add_wait_queue(wqh, &event->wait); | |
4800 | } | |
4801 | ||
4802 | /* | |
3bc942f3 TH |
4803 | * DO NOT USE IN NEW FILES. |
4804 | * | |
79bd9814 TH |
4805 | * Parse input and register new cgroup event handler. |
4806 | * | |
4807 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4808 | * Interpretation of args is defined by control file implementation. | |
4809 | */ | |
451af504 TH |
4810 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4811 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4812 | { |
451af504 | 4813 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4814 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4815 | struct mem_cgroup_event *event; |
79bd9814 TH |
4816 | struct cgroup_subsys_state *cfile_css; |
4817 | unsigned int efd, cfd; | |
4818 | struct fd efile; | |
4819 | struct fd cfile; | |
fba94807 | 4820 | const char *name; |
79bd9814 TH |
4821 | char *endp; |
4822 | int ret; | |
4823 | ||
451af504 TH |
4824 | buf = strstrip(buf); |
4825 | ||
4826 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4827 | if (*endp != ' ') |
4828 | return -EINVAL; | |
451af504 | 4829 | buf = endp + 1; |
79bd9814 | 4830 | |
451af504 | 4831 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4832 | if ((*endp != ' ') && (*endp != '\0')) |
4833 | return -EINVAL; | |
451af504 | 4834 | buf = endp + 1; |
79bd9814 TH |
4835 | |
4836 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4837 | if (!event) | |
4838 | return -ENOMEM; | |
4839 | ||
59b6f873 | 4840 | event->memcg = memcg; |
79bd9814 | 4841 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4842 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4843 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4844 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4845 | |
4846 | efile = fdget(efd); | |
4847 | if (!efile.file) { | |
4848 | ret = -EBADF; | |
4849 | goto out_kfree; | |
4850 | } | |
4851 | ||
4852 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4853 | if (IS_ERR(event->eventfd)) { | |
4854 | ret = PTR_ERR(event->eventfd); | |
4855 | goto out_put_efile; | |
4856 | } | |
4857 | ||
4858 | cfile = fdget(cfd); | |
4859 | if (!cfile.file) { | |
4860 | ret = -EBADF; | |
4861 | goto out_put_eventfd; | |
4862 | } | |
4863 | ||
4864 | /* the process need read permission on control file */ | |
4865 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
02f92b38 | 4866 | ret = file_permission(cfile.file, MAY_READ); |
79bd9814 TH |
4867 | if (ret < 0) |
4868 | goto out_put_cfile; | |
4869 | ||
fba94807 TH |
4870 | /* |
4871 | * Determine the event callbacks and set them in @event. This used | |
4872 | * to be done via struct cftype but cgroup core no longer knows | |
4873 | * about these events. The following is crude but the whole thing | |
4874 | * is for compatibility anyway. | |
3bc942f3 TH |
4875 | * |
4876 | * DO NOT ADD NEW FILES. | |
fba94807 | 4877 | */ |
b583043e | 4878 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4879 | |
4880 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4881 | event->register_event = mem_cgroup_usage_register_event; | |
4882 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4883 | } else if (!strcmp(name, "memory.oom_control")) { | |
4884 | event->register_event = mem_cgroup_oom_register_event; | |
4885 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4886 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4887 | event->register_event = vmpressure_register_event; | |
4888 | event->unregister_event = vmpressure_unregister_event; | |
4889 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4890 | event->register_event = memsw_cgroup_usage_register_event; |
4891 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4892 | } else { |
4893 | ret = -EINVAL; | |
4894 | goto out_put_cfile; | |
4895 | } | |
4896 | ||
79bd9814 | 4897 | /* |
b5557c4c TH |
4898 | * Verify @cfile should belong to @css. Also, remaining events are |
4899 | * automatically removed on cgroup destruction but the removal is | |
4900 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4901 | */ |
b583043e | 4902 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4903 | &memory_cgrp_subsys); |
79bd9814 | 4904 | ret = -EINVAL; |
5a17f543 | 4905 | if (IS_ERR(cfile_css)) |
79bd9814 | 4906 | goto out_put_cfile; |
5a17f543 TH |
4907 | if (cfile_css != css) { |
4908 | css_put(cfile_css); | |
79bd9814 | 4909 | goto out_put_cfile; |
5a17f543 | 4910 | } |
79bd9814 | 4911 | |
451af504 | 4912 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4913 | if (ret) |
4914 | goto out_put_css; | |
4915 | ||
9965ed17 | 4916 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 4917 | |
fba94807 TH |
4918 | spin_lock(&memcg->event_list_lock); |
4919 | list_add(&event->list, &memcg->event_list); | |
4920 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4921 | |
4922 | fdput(cfile); | |
4923 | fdput(efile); | |
4924 | ||
451af504 | 4925 | return nbytes; |
79bd9814 TH |
4926 | |
4927 | out_put_css: | |
b5557c4c | 4928 | css_put(css); |
79bd9814 TH |
4929 | out_put_cfile: |
4930 | fdput(cfile); | |
4931 | out_put_eventfd: | |
4932 | eventfd_ctx_put(event->eventfd); | |
4933 | out_put_efile: | |
4934 | fdput(efile); | |
4935 | out_kfree: | |
4936 | kfree(event); | |
4937 | ||
4938 | return ret; | |
4939 | } | |
4940 | ||
241994ed | 4941 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4942 | { |
0eea1030 | 4943 | .name = "usage_in_bytes", |
8c7c6e34 | 4944 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4945 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4946 | }, |
c84872e1 PE |
4947 | { |
4948 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4949 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4950 | .write = mem_cgroup_reset, |
791badbd | 4951 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4952 | }, |
8cdea7c0 | 4953 | { |
0eea1030 | 4954 | .name = "limit_in_bytes", |
8c7c6e34 | 4955 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4956 | .write = mem_cgroup_write, |
791badbd | 4957 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4958 | }, |
296c81d8 BS |
4959 | { |
4960 | .name = "soft_limit_in_bytes", | |
4961 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4962 | .write = mem_cgroup_write, |
791badbd | 4963 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4964 | }, |
8cdea7c0 BS |
4965 | { |
4966 | .name = "failcnt", | |
8c7c6e34 | 4967 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4968 | .write = mem_cgroup_reset, |
791badbd | 4969 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4970 | }, |
d2ceb9b7 KH |
4971 | { |
4972 | .name = "stat", | |
2da8ca82 | 4973 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4974 | }, |
c1e862c1 KH |
4975 | { |
4976 | .name = "force_empty", | |
6770c64e | 4977 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4978 | }, |
18f59ea7 BS |
4979 | { |
4980 | .name = "use_hierarchy", | |
4981 | .write_u64 = mem_cgroup_hierarchy_write, | |
4982 | .read_u64 = mem_cgroup_hierarchy_read, | |
4983 | }, | |
79bd9814 | 4984 | { |
3bc942f3 | 4985 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4986 | .write = memcg_write_event_control, |
7dbdb199 | 4987 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 4988 | }, |
a7885eb8 KM |
4989 | { |
4990 | .name = "swappiness", | |
4991 | .read_u64 = mem_cgroup_swappiness_read, | |
4992 | .write_u64 = mem_cgroup_swappiness_write, | |
4993 | }, | |
7dc74be0 DN |
4994 | { |
4995 | .name = "move_charge_at_immigrate", | |
4996 | .read_u64 = mem_cgroup_move_charge_read, | |
4997 | .write_u64 = mem_cgroup_move_charge_write, | |
4998 | }, | |
9490ff27 KH |
4999 | { |
5000 | .name = "oom_control", | |
2da8ca82 | 5001 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 5002 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
5003 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
5004 | }, | |
70ddf637 AV |
5005 | { |
5006 | .name = "pressure_level", | |
70ddf637 | 5007 | }, |
406eb0c9 YH |
5008 | #ifdef CONFIG_NUMA |
5009 | { | |
5010 | .name = "numa_stat", | |
2da8ca82 | 5011 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
5012 | }, |
5013 | #endif | |
510fc4e1 GC |
5014 | { |
5015 | .name = "kmem.limit_in_bytes", | |
5016 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 5017 | .write = mem_cgroup_write, |
791badbd | 5018 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5019 | }, |
5020 | { | |
5021 | .name = "kmem.usage_in_bytes", | |
5022 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 5023 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5024 | }, |
5025 | { | |
5026 | .name = "kmem.failcnt", | |
5027 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 5028 | .write = mem_cgroup_reset, |
791badbd | 5029 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5030 | }, |
5031 | { | |
5032 | .name = "kmem.max_usage_in_bytes", | |
5033 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 5034 | .write = mem_cgroup_reset, |
791badbd | 5035 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 5036 | }, |
a87425a3 YS |
5037 | #if defined(CONFIG_MEMCG_KMEM) && \ |
5038 | (defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)) | |
749c5415 GC |
5039 | { |
5040 | .name = "kmem.slabinfo", | |
b047501c | 5041 | .seq_show = memcg_slab_show, |
749c5415 GC |
5042 | }, |
5043 | #endif | |
d55f90bf VD |
5044 | { |
5045 | .name = "kmem.tcp.limit_in_bytes", | |
5046 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
5047 | .write = mem_cgroup_write, | |
5048 | .read_u64 = mem_cgroup_read_u64, | |
5049 | }, | |
5050 | { | |
5051 | .name = "kmem.tcp.usage_in_bytes", | |
5052 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
5053 | .read_u64 = mem_cgroup_read_u64, | |
5054 | }, | |
5055 | { | |
5056 | .name = "kmem.tcp.failcnt", | |
5057 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
5058 | .write = mem_cgroup_reset, | |
5059 | .read_u64 = mem_cgroup_read_u64, | |
5060 | }, | |
5061 | { | |
5062 | .name = "kmem.tcp.max_usage_in_bytes", | |
5063 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
5064 | .write = mem_cgroup_reset, | |
5065 | .read_u64 = mem_cgroup_read_u64, | |
5066 | }, | |
6bc10349 | 5067 | { }, /* terminate */ |
af36f906 | 5068 | }; |
8c7c6e34 | 5069 | |
73f576c0 JW |
5070 | /* |
5071 | * Private memory cgroup IDR | |
5072 | * | |
5073 | * Swap-out records and page cache shadow entries need to store memcg | |
5074 | * references in constrained space, so we maintain an ID space that is | |
5075 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
5076 | * memory-controlled cgroups to 64k. | |
5077 | * | |
b8f2935f | 5078 | * However, there usually are many references to the offline CSS after |
73f576c0 JW |
5079 | * the cgroup has been destroyed, such as page cache or reclaimable |
5080 | * slab objects, that don't need to hang on to the ID. We want to keep | |
5081 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
5082 | * relatively small ID space and prevent the creation of new cgroups | |
5083 | * even when there are much fewer than 64k cgroups - possibly none. | |
5084 | * | |
5085 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
5086 | * be freed and recycled when it's no longer needed, which is usually | |
5087 | * when the CSS is offlined. | |
5088 | * | |
5089 | * The only exception to that are records of swapped out tmpfs/shmem | |
5090 | * pages that need to be attributed to live ancestors on swapin. But | |
5091 | * those references are manageable from userspace. | |
5092 | */ | |
5093 | ||
5094 | static DEFINE_IDR(mem_cgroup_idr); | |
5095 | ||
7e97de0b KT |
5096 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
5097 | { | |
5098 | if (memcg->id.id > 0) { | |
5099 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
5100 | memcg->id.id = 0; | |
5101 | } | |
5102 | } | |
5103 | ||
c1514c0a VF |
5104 | static void __maybe_unused mem_cgroup_id_get_many(struct mem_cgroup *memcg, |
5105 | unsigned int n) | |
73f576c0 | 5106 | { |
1c2d479a | 5107 | refcount_add(n, &memcg->id.ref); |
73f576c0 JW |
5108 | } |
5109 | ||
615d66c3 | 5110 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 5111 | { |
1c2d479a | 5112 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 5113 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
5114 | |
5115 | /* Memcg ID pins CSS */ | |
5116 | css_put(&memcg->css); | |
5117 | } | |
5118 | } | |
5119 | ||
615d66c3 VD |
5120 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) |
5121 | { | |
5122 | mem_cgroup_id_put_many(memcg, 1); | |
5123 | } | |
5124 | ||
73f576c0 JW |
5125 | /** |
5126 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
5127 | * @id: the memcg id to look up | |
5128 | * | |
5129 | * Caller must hold rcu_read_lock(). | |
5130 | */ | |
5131 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
5132 | { | |
5133 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5134 | return idr_find(&mem_cgroup_idr, id); | |
5135 | } | |
5136 | ||
ef8f2327 | 5137 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
5138 | { |
5139 | struct mem_cgroup_per_node *pn; | |
ef8f2327 | 5140 | int tmp = node; |
1ecaab2b KH |
5141 | /* |
5142 | * This routine is called against possible nodes. | |
5143 | * But it's BUG to call kmalloc() against offline node. | |
5144 | * | |
5145 | * TODO: this routine can waste much memory for nodes which will | |
5146 | * never be onlined. It's better to use memory hotplug callback | |
5147 | * function. | |
5148 | */ | |
41e3355d KH |
5149 | if (!node_state(node, N_NORMAL_MEMORY)) |
5150 | tmp = -1; | |
17295c88 | 5151 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
5152 | if (!pn) |
5153 | return 1; | |
1ecaab2b | 5154 | |
3e38e0aa RG |
5155 | pn->lruvec_stat_local = alloc_percpu_gfp(struct lruvec_stat, |
5156 | GFP_KERNEL_ACCOUNT); | |
815744d7 JW |
5157 | if (!pn->lruvec_stat_local) { |
5158 | kfree(pn); | |
5159 | return 1; | |
5160 | } | |
5161 | ||
f3344adf | 5162 | pn->lruvec_stat_cpu = alloc_percpu_gfp(struct batched_lruvec_stat, |
3e38e0aa | 5163 | GFP_KERNEL_ACCOUNT); |
a983b5eb | 5164 | if (!pn->lruvec_stat_cpu) { |
815744d7 | 5165 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c JW |
5166 | kfree(pn); |
5167 | return 1; | |
5168 | } | |
5169 | ||
ef8f2327 MG |
5170 | lruvec_init(&pn->lruvec); |
5171 | pn->usage_in_excess = 0; | |
5172 | pn->on_tree = false; | |
5173 | pn->memcg = memcg; | |
5174 | ||
54f72fe0 | 5175 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
5176 | return 0; |
5177 | } | |
5178 | ||
ef8f2327 | 5179 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 5180 | { |
00f3ca2c JW |
5181 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
5182 | ||
4eaf431f MH |
5183 | if (!pn) |
5184 | return; | |
5185 | ||
a983b5eb | 5186 | free_percpu(pn->lruvec_stat_cpu); |
815744d7 | 5187 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c | 5188 | kfree(pn); |
1ecaab2b KH |
5189 | } |
5190 | ||
40e952f9 | 5191 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 5192 | { |
c8b2a36f | 5193 | int node; |
59927fb9 | 5194 | |
c8b2a36f | 5195 | for_each_node(node) |
ef8f2327 | 5196 | free_mem_cgroup_per_node_info(memcg, node); |
871789d4 | 5197 | free_percpu(memcg->vmstats_percpu); |
815744d7 | 5198 | free_percpu(memcg->vmstats_local); |
8ff69e2c | 5199 | kfree(memcg); |
59927fb9 | 5200 | } |
3afe36b1 | 5201 | |
40e952f9 TE |
5202 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
5203 | { | |
5204 | memcg_wb_domain_exit(memcg); | |
7961eee3 SB |
5205 | /* |
5206 | * Flush percpu vmstats and vmevents to guarantee the value correctness | |
5207 | * on parent's and all ancestor levels. | |
5208 | */ | |
4a87e2a2 | 5209 | memcg_flush_percpu_vmstats(memcg); |
7961eee3 | 5210 | memcg_flush_percpu_vmevents(memcg); |
40e952f9 TE |
5211 | __mem_cgroup_free(memcg); |
5212 | } | |
5213 | ||
0b8f73e1 | 5214 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 5215 | { |
d142e3e6 | 5216 | struct mem_cgroup *memcg; |
b9726c26 | 5217 | unsigned int size; |
6d12e2d8 | 5218 | int node; |
97b27821 | 5219 | int __maybe_unused i; |
11d67612 | 5220 | long error = -ENOMEM; |
8cdea7c0 | 5221 | |
0b8f73e1 JW |
5222 | size = sizeof(struct mem_cgroup); |
5223 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
5224 | ||
5225 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 5226 | if (!memcg) |
11d67612 | 5227 | return ERR_PTR(error); |
0b8f73e1 | 5228 | |
73f576c0 JW |
5229 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
5230 | 1, MEM_CGROUP_ID_MAX, | |
5231 | GFP_KERNEL); | |
11d67612 YS |
5232 | if (memcg->id.id < 0) { |
5233 | error = memcg->id.id; | |
73f576c0 | 5234 | goto fail; |
11d67612 | 5235 | } |
73f576c0 | 5236 | |
3e38e0aa RG |
5237 | memcg->vmstats_local = alloc_percpu_gfp(struct memcg_vmstats_percpu, |
5238 | GFP_KERNEL_ACCOUNT); | |
815744d7 JW |
5239 | if (!memcg->vmstats_local) |
5240 | goto fail; | |
5241 | ||
3e38e0aa RG |
5242 | memcg->vmstats_percpu = alloc_percpu_gfp(struct memcg_vmstats_percpu, |
5243 | GFP_KERNEL_ACCOUNT); | |
871789d4 | 5244 | if (!memcg->vmstats_percpu) |
0b8f73e1 | 5245 | goto fail; |
78fb7466 | 5246 | |
3ed28fa1 | 5247 | for_each_node(node) |
ef8f2327 | 5248 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 5249 | goto fail; |
f64c3f54 | 5250 | |
0b8f73e1 JW |
5251 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
5252 | goto fail; | |
28dbc4b6 | 5253 | |
f7e1cb6e | 5254 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 | 5255 | INIT_LIST_HEAD(&memcg->oom_notify); |
d142e3e6 GC |
5256 | mutex_init(&memcg->thresholds_lock); |
5257 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 5258 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
5259 | INIT_LIST_HEAD(&memcg->event_list); |
5260 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 5261 | memcg->socket_pressure = jiffies; |
84c07d11 | 5262 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 5263 | memcg->kmemcg_id = -1; |
bf4f0599 | 5264 | INIT_LIST_HEAD(&memcg->objcg_list); |
900a38f0 | 5265 | #endif |
52ebea74 TH |
5266 | #ifdef CONFIG_CGROUP_WRITEBACK |
5267 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
97b27821 TH |
5268 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) |
5269 | memcg->cgwb_frn[i].done = | |
5270 | __WB_COMPLETION_INIT(&memcg_cgwb_frn_waitq); | |
87eaceb3 YS |
5271 | #endif |
5272 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5273 | spin_lock_init(&memcg->deferred_split_queue.split_queue_lock); | |
5274 | INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue); | |
5275 | memcg->deferred_split_queue.split_queue_len = 0; | |
52ebea74 | 5276 | #endif |
73f576c0 | 5277 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
5278 | return memcg; |
5279 | fail: | |
7e97de0b | 5280 | mem_cgroup_id_remove(memcg); |
40e952f9 | 5281 | __mem_cgroup_free(memcg); |
11d67612 | 5282 | return ERR_PTR(error); |
d142e3e6 GC |
5283 | } |
5284 | ||
0b8f73e1 JW |
5285 | static struct cgroup_subsys_state * __ref |
5286 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 5287 | { |
0b8f73e1 | 5288 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
b87d8cef | 5289 | struct mem_cgroup *memcg, *old_memcg; |
0b8f73e1 | 5290 | long error = -ENOMEM; |
d142e3e6 | 5291 | |
b87d8cef | 5292 | old_memcg = set_active_memcg(parent); |
0b8f73e1 | 5293 | memcg = mem_cgroup_alloc(); |
b87d8cef | 5294 | set_active_memcg(old_memcg); |
11d67612 YS |
5295 | if (IS_ERR(memcg)) |
5296 | return ERR_CAST(memcg); | |
d142e3e6 | 5297 | |
d1663a90 | 5298 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
0b8f73e1 | 5299 | memcg->soft_limit = PAGE_COUNTER_MAX; |
4b82ab4f | 5300 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
0b8f73e1 JW |
5301 | if (parent) { |
5302 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
5303 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
bef8620c | 5304 | |
3e32cb2e | 5305 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 5306 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e | 5307 | page_counter_init(&memcg->kmem, &parent->kmem); |
0db15298 | 5308 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 5309 | } else { |
bef8620c RG |
5310 | page_counter_init(&memcg->memory, NULL); |
5311 | page_counter_init(&memcg->swap, NULL); | |
5312 | page_counter_init(&memcg->kmem, NULL); | |
5313 | page_counter_init(&memcg->tcpmem, NULL); | |
d6441637 | 5314 | |
0b8f73e1 JW |
5315 | root_mem_cgroup = memcg; |
5316 | return &memcg->css; | |
5317 | } | |
5318 | ||
bef8620c | 5319 | /* The following stuff does not apply to the root */ |
b313aeee | 5320 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
5321 | if (error) |
5322 | goto fail; | |
127424c8 | 5323 | |
f7e1cb6e | 5324 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5325 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 5326 | |
0b8f73e1 JW |
5327 | return &memcg->css; |
5328 | fail: | |
7e97de0b | 5329 | mem_cgroup_id_remove(memcg); |
0b8f73e1 | 5330 | mem_cgroup_free(memcg); |
11d67612 | 5331 | return ERR_PTR(error); |
0b8f73e1 JW |
5332 | } |
5333 | ||
73f576c0 | 5334 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 5335 | { |
58fa2a55 VD |
5336 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5337 | ||
0a4465d3 KT |
5338 | /* |
5339 | * A memcg must be visible for memcg_expand_shrinker_maps() | |
5340 | * by the time the maps are allocated. So, we allocate maps | |
5341 | * here, when for_each_mem_cgroup() can't skip it. | |
5342 | */ | |
5343 | if (memcg_alloc_shrinker_maps(memcg)) { | |
5344 | mem_cgroup_id_remove(memcg); | |
5345 | return -ENOMEM; | |
5346 | } | |
5347 | ||
73f576c0 | 5348 | /* Online state pins memcg ID, memcg ID pins CSS */ |
1c2d479a | 5349 | refcount_set(&memcg->id.ref, 1); |
73f576c0 | 5350 | css_get(css); |
2f7dd7a4 | 5351 | return 0; |
8cdea7c0 BS |
5352 | } |
5353 | ||
eb95419b | 5354 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 5355 | { |
eb95419b | 5356 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5357 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
5358 | |
5359 | /* | |
5360 | * Unregister events and notify userspace. | |
5361 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
5362 | * directory to avoid race between userspace and kernelspace. | |
5363 | */ | |
fba94807 TH |
5364 | spin_lock(&memcg->event_list_lock); |
5365 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
5366 | list_del_init(&event->list); |
5367 | schedule_work(&event->remove); | |
5368 | } | |
fba94807 | 5369 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 5370 | |
bf8d5d52 | 5371 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5372 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 5373 | |
567e9ab2 | 5374 | memcg_offline_kmem(memcg); |
52ebea74 | 5375 | wb_memcg_offline(memcg); |
73f576c0 | 5376 | |
591edfb1 RG |
5377 | drain_all_stock(memcg); |
5378 | ||
73f576c0 | 5379 | mem_cgroup_id_put(memcg); |
df878fb0 KH |
5380 | } |
5381 | ||
6df38689 VD |
5382 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
5383 | { | |
5384 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5385 | ||
5386 | invalidate_reclaim_iterators(memcg); | |
5387 | } | |
5388 | ||
eb95419b | 5389 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 5390 | { |
eb95419b | 5391 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
97b27821 | 5392 | int __maybe_unused i; |
c268e994 | 5393 | |
97b27821 TH |
5394 | #ifdef CONFIG_CGROUP_WRITEBACK |
5395 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) | |
5396 | wb_wait_for_completion(&memcg->cgwb_frn[i].done); | |
5397 | #endif | |
f7e1cb6e | 5398 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5399 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 5400 | |
0db15298 | 5401 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 5402 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 5403 | |
0b8f73e1 JW |
5404 | vmpressure_cleanup(&memcg->vmpressure); |
5405 | cancel_work_sync(&memcg->high_work); | |
5406 | mem_cgroup_remove_from_trees(memcg); | |
0a4465d3 | 5407 | memcg_free_shrinker_maps(memcg); |
d886f4e4 | 5408 | memcg_free_kmem(memcg); |
0b8f73e1 | 5409 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
5410 | } |
5411 | ||
1ced953b TH |
5412 | /** |
5413 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
5414 | * @css: the target css | |
5415 | * | |
5416 | * Reset the states of the mem_cgroup associated with @css. This is | |
5417 | * invoked when the userland requests disabling on the default hierarchy | |
5418 | * but the memcg is pinned through dependency. The memcg should stop | |
5419 | * applying policies and should revert to the vanilla state as it may be | |
5420 | * made visible again. | |
5421 | * | |
5422 | * The current implementation only resets the essential configurations. | |
5423 | * This needs to be expanded to cover all the visible parts. | |
5424 | */ | |
5425 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
5426 | { | |
5427 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5428 | ||
bbec2e15 RG |
5429 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
5430 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
bbec2e15 RG |
5431 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); |
5432 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 5433 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5434 | page_counter_set_low(&memcg->memory, 0); |
d1663a90 | 5435 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
24d404dc | 5436 | memcg->soft_limit = PAGE_COUNTER_MAX; |
4b82ab4f | 5437 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
2529bb3a | 5438 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
5439 | } |
5440 | ||
02491447 | 5441 | #ifdef CONFIG_MMU |
7dc74be0 | 5442 | /* Handlers for move charge at task migration. */ |
854ffa8d | 5443 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 5444 | { |
05b84301 | 5445 | int ret; |
9476db97 | 5446 | |
d0164adc MG |
5447 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
5448 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 5449 | if (!ret) { |
854ffa8d | 5450 | mc.precharge += count; |
854ffa8d DN |
5451 | return ret; |
5452 | } | |
9476db97 | 5453 | |
3674534b | 5454 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 5455 | while (count--) { |
3674534b | 5456 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 5457 | if (ret) |
38c5d72f | 5458 | return ret; |
854ffa8d | 5459 | mc.precharge++; |
9476db97 | 5460 | cond_resched(); |
854ffa8d | 5461 | } |
9476db97 | 5462 | return 0; |
4ffef5fe DN |
5463 | } |
5464 | ||
4ffef5fe DN |
5465 | union mc_target { |
5466 | struct page *page; | |
02491447 | 5467 | swp_entry_t ent; |
4ffef5fe DN |
5468 | }; |
5469 | ||
4ffef5fe | 5470 | enum mc_target_type { |
8d32ff84 | 5471 | MC_TARGET_NONE = 0, |
4ffef5fe | 5472 | MC_TARGET_PAGE, |
02491447 | 5473 | MC_TARGET_SWAP, |
c733a828 | 5474 | MC_TARGET_DEVICE, |
4ffef5fe DN |
5475 | }; |
5476 | ||
90254a65 DN |
5477 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5478 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5479 | { |
25b2995a | 5480 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5481 | |
90254a65 DN |
5482 | if (!page || !page_mapped(page)) |
5483 | return NULL; | |
5484 | if (PageAnon(page)) { | |
1dfab5ab | 5485 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5486 | return NULL; |
1dfab5ab JW |
5487 | } else { |
5488 | if (!(mc.flags & MOVE_FILE)) | |
5489 | return NULL; | |
5490 | } | |
90254a65 DN |
5491 | if (!get_page_unless_zero(page)) |
5492 | return NULL; | |
5493 | ||
5494 | return page; | |
5495 | } | |
5496 | ||
c733a828 | 5497 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 5498 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 5499 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 5500 | { |
90254a65 DN |
5501 | struct page *page = NULL; |
5502 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5503 | ||
9a137153 | 5504 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5505 | return NULL; |
c733a828 JG |
5506 | |
5507 | /* | |
5508 | * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to | |
5509 | * a device and because they are not accessible by CPU they are store | |
5510 | * as special swap entry in the CPU page table. | |
5511 | */ | |
5512 | if (is_device_private_entry(ent)) { | |
5513 | page = device_private_entry_to_page(ent); | |
5514 | /* | |
5515 | * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have | |
5516 | * a refcount of 1 when free (unlike normal page) | |
5517 | */ | |
5518 | if (!page_ref_add_unless(page, 1, 1)) | |
5519 | return NULL; | |
5520 | return page; | |
5521 | } | |
5522 | ||
9a137153 RC |
5523 | if (non_swap_entry(ent)) |
5524 | return NULL; | |
5525 | ||
4b91355e KH |
5526 | /* |
5527 | * Because lookup_swap_cache() updates some statistics counter, | |
5528 | * we call find_get_page() with swapper_space directly. | |
5529 | */ | |
f6ab1f7f | 5530 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
2d1c4980 | 5531 | entry->val = ent.val; |
90254a65 DN |
5532 | |
5533 | return page; | |
5534 | } | |
4b91355e KH |
5535 | #else |
5536 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 5537 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
5538 | { |
5539 | return NULL; | |
5540 | } | |
5541 | #endif | |
90254a65 | 5542 | |
87946a72 DN |
5543 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5544 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5545 | { | |
87946a72 DN |
5546 | if (!vma->vm_file) /* anonymous vma */ |
5547 | return NULL; | |
1dfab5ab | 5548 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
5549 | return NULL; |
5550 | ||
87946a72 | 5551 | /* page is moved even if it's not RSS of this task(page-faulted). */ |
aa3b1895 | 5552 | /* shmem/tmpfs may report page out on swap: account for that too. */ |
f5df8635 MWO |
5553 | return find_get_incore_page(vma->vm_file->f_mapping, |
5554 | linear_page_index(vma, addr)); | |
87946a72 DN |
5555 | } |
5556 | ||
b1b0deab CG |
5557 | /** |
5558 | * mem_cgroup_move_account - move account of the page | |
5559 | * @page: the page | |
25843c2b | 5560 | * @compound: charge the page as compound or small page |
b1b0deab CG |
5561 | * @from: mem_cgroup which the page is moved from. |
5562 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
5563 | * | |
3ac808fd | 5564 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
5565 | * |
5566 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
5567 | * from old cgroup. | |
5568 | */ | |
5569 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 5570 | bool compound, |
b1b0deab CG |
5571 | struct mem_cgroup *from, |
5572 | struct mem_cgroup *to) | |
5573 | { | |
ae8af438 KK |
5574 | struct lruvec *from_vec, *to_vec; |
5575 | struct pglist_data *pgdat; | |
6c357848 | 5576 | unsigned int nr_pages = compound ? thp_nr_pages(page) : 1; |
b1b0deab CG |
5577 | int ret; |
5578 | ||
5579 | VM_BUG_ON(from == to); | |
5580 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 5581 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
5582 | |
5583 | /* | |
6a93ca8f | 5584 | * Prevent mem_cgroup_migrate() from looking at |
bcfe06bf | 5585 | * page's memory cgroup of its source page while we change it. |
b1b0deab | 5586 | */ |
f627c2f5 | 5587 | ret = -EBUSY; |
b1b0deab CG |
5588 | if (!trylock_page(page)) |
5589 | goto out; | |
5590 | ||
5591 | ret = -EINVAL; | |
bcfe06bf | 5592 | if (page_memcg(page) != from) |
b1b0deab CG |
5593 | goto out_unlock; |
5594 | ||
ae8af438 | 5595 | pgdat = page_pgdat(page); |
867e5e1d JW |
5596 | from_vec = mem_cgroup_lruvec(from, pgdat); |
5597 | to_vec = mem_cgroup_lruvec(to, pgdat); | |
ae8af438 | 5598 | |
abb242f5 | 5599 | lock_page_memcg(page); |
b1b0deab | 5600 | |
be5d0a74 JW |
5601 | if (PageAnon(page)) { |
5602 | if (page_mapped(page)) { | |
5603 | __mod_lruvec_state(from_vec, NR_ANON_MAPPED, -nr_pages); | |
5604 | __mod_lruvec_state(to_vec, NR_ANON_MAPPED, nr_pages); | |
468c3982 | 5605 | if (PageTransHuge(page)) { |
69473e5d MS |
5606 | __mod_lruvec_state(from_vec, NR_ANON_THPS, |
5607 | -nr_pages); | |
5608 | __mod_lruvec_state(to_vec, NR_ANON_THPS, | |
5609 | nr_pages); | |
468c3982 | 5610 | } |
be5d0a74 JW |
5611 | } |
5612 | } else { | |
0d1c2072 JW |
5613 | __mod_lruvec_state(from_vec, NR_FILE_PAGES, -nr_pages); |
5614 | __mod_lruvec_state(to_vec, NR_FILE_PAGES, nr_pages); | |
5615 | ||
5616 | if (PageSwapBacked(page)) { | |
5617 | __mod_lruvec_state(from_vec, NR_SHMEM, -nr_pages); | |
5618 | __mod_lruvec_state(to_vec, NR_SHMEM, nr_pages); | |
5619 | } | |
5620 | ||
49e50d27 JW |
5621 | if (page_mapped(page)) { |
5622 | __mod_lruvec_state(from_vec, NR_FILE_MAPPED, -nr_pages); | |
5623 | __mod_lruvec_state(to_vec, NR_FILE_MAPPED, nr_pages); | |
5624 | } | |
b1b0deab | 5625 | |
49e50d27 JW |
5626 | if (PageDirty(page)) { |
5627 | struct address_space *mapping = page_mapping(page); | |
c4843a75 | 5628 | |
f56753ac | 5629 | if (mapping_can_writeback(mapping)) { |
49e50d27 JW |
5630 | __mod_lruvec_state(from_vec, NR_FILE_DIRTY, |
5631 | -nr_pages); | |
5632 | __mod_lruvec_state(to_vec, NR_FILE_DIRTY, | |
5633 | nr_pages); | |
5634 | } | |
c4843a75 GT |
5635 | } |
5636 | } | |
5637 | ||
b1b0deab | 5638 | if (PageWriteback(page)) { |
ae8af438 KK |
5639 | __mod_lruvec_state(from_vec, NR_WRITEBACK, -nr_pages); |
5640 | __mod_lruvec_state(to_vec, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
5641 | } |
5642 | ||
5643 | /* | |
abb242f5 JW |
5644 | * All state has been migrated, let's switch to the new memcg. |
5645 | * | |
bcfe06bf | 5646 | * It is safe to change page's memcg here because the page |
abb242f5 JW |
5647 | * is referenced, charged, isolated, and locked: we can't race |
5648 | * with (un)charging, migration, LRU putback, or anything else | |
bcfe06bf | 5649 | * that would rely on a stable page's memory cgroup. |
abb242f5 JW |
5650 | * |
5651 | * Note that lock_page_memcg is a memcg lock, not a page lock, | |
bcfe06bf | 5652 | * to save space. As soon as we switch page's memory cgroup to a |
abb242f5 JW |
5653 | * new memcg that isn't locked, the above state can change |
5654 | * concurrently again. Make sure we're truly done with it. | |
b1b0deab | 5655 | */ |
abb242f5 | 5656 | smp_mb(); |
b1b0deab | 5657 | |
1a3e1f40 JW |
5658 | css_get(&to->css); |
5659 | css_put(&from->css); | |
5660 | ||
bcfe06bf | 5661 | page->memcg_data = (unsigned long)to; |
87eaceb3 | 5662 | |
abb242f5 | 5663 | __unlock_page_memcg(from); |
b1b0deab CG |
5664 | |
5665 | ret = 0; | |
5666 | ||
5667 | local_irq_disable(); | |
3fba69a5 | 5668 | mem_cgroup_charge_statistics(to, page, nr_pages); |
b1b0deab | 5669 | memcg_check_events(to, page); |
3fba69a5 | 5670 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
b1b0deab CG |
5671 | memcg_check_events(from, page); |
5672 | local_irq_enable(); | |
5673 | out_unlock: | |
5674 | unlock_page(page); | |
5675 | out: | |
5676 | return ret; | |
5677 | } | |
5678 | ||
7cf7806c LR |
5679 | /** |
5680 | * get_mctgt_type - get target type of moving charge | |
5681 | * @vma: the vma the pte to be checked belongs | |
5682 | * @addr: the address corresponding to the pte to be checked | |
5683 | * @ptent: the pte to be checked | |
5684 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
5685 | * | |
5686 | * Returns | |
5687 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5688 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5689 | * move charge. if @target is not NULL, the page is stored in target->page | |
5690 | * with extra refcnt got(Callers should handle it). | |
5691 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
5692 | * target for charge migration. if @target is not NULL, the entry is stored | |
5693 | * in target->ent. | |
25b2995a CH |
5694 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PRIVATE |
5695 | * (so ZONE_DEVICE page and thus not on the lru). | |
df6ad698 JG |
5696 | * For now we such page is charge like a regular page would be as for all |
5697 | * intent and purposes it is just special memory taking the place of a | |
5698 | * regular page. | |
c733a828 JG |
5699 | * |
5700 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
5701 | * |
5702 | * Called with pte lock held. | |
5703 | */ | |
5704 | ||
8d32ff84 | 5705 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5706 | unsigned long addr, pte_t ptent, union mc_target *target) |
5707 | { | |
5708 | struct page *page = NULL; | |
8d32ff84 | 5709 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5710 | swp_entry_t ent = { .val = 0 }; |
5711 | ||
5712 | if (pte_present(ptent)) | |
5713 | page = mc_handle_present_pte(vma, addr, ptent); | |
5714 | else if (is_swap_pte(ptent)) | |
48406ef8 | 5715 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 5716 | else if (pte_none(ptent)) |
87946a72 | 5717 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
5718 | |
5719 | if (!page && !ent.val) | |
8d32ff84 | 5720 | return ret; |
02491447 | 5721 | if (page) { |
02491447 | 5722 | /* |
0a31bc97 | 5723 | * Do only loose check w/o serialization. |
1306a85a | 5724 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 5725 | * not under LRU exclusion. |
02491447 | 5726 | */ |
bcfe06bf | 5727 | if (page_memcg(page) == mc.from) { |
02491447 | 5728 | ret = MC_TARGET_PAGE; |
25b2995a | 5729 | if (is_device_private_page(page)) |
c733a828 | 5730 | ret = MC_TARGET_DEVICE; |
02491447 DN |
5731 | if (target) |
5732 | target->page = page; | |
5733 | } | |
5734 | if (!ret || !target) | |
5735 | put_page(page); | |
5736 | } | |
3e14a57b HY |
5737 | /* |
5738 | * There is a swap entry and a page doesn't exist or isn't charged. | |
5739 | * But we cannot move a tail-page in a THP. | |
5740 | */ | |
5741 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 5742 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5743 | ret = MC_TARGET_SWAP; |
5744 | if (target) | |
5745 | target->ent = ent; | |
4ffef5fe | 5746 | } |
4ffef5fe DN |
5747 | return ret; |
5748 | } | |
5749 | ||
12724850 NH |
5750 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5751 | /* | |
d6810d73 HY |
5752 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
5753 | * not support them for now. | |
12724850 NH |
5754 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
5755 | */ | |
5756 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5757 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5758 | { | |
5759 | struct page *page = NULL; | |
12724850 NH |
5760 | enum mc_target_type ret = MC_TARGET_NONE; |
5761 | ||
84c3fc4e ZY |
5762 | if (unlikely(is_swap_pmd(pmd))) { |
5763 | VM_BUG_ON(thp_migration_supported() && | |
5764 | !is_pmd_migration_entry(pmd)); | |
5765 | return ret; | |
5766 | } | |
12724850 | 5767 | page = pmd_page(pmd); |
309381fe | 5768 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 5769 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 5770 | return ret; |
bcfe06bf | 5771 | if (page_memcg(page) == mc.from) { |
12724850 NH |
5772 | ret = MC_TARGET_PAGE; |
5773 | if (target) { | |
5774 | get_page(page); | |
5775 | target->page = page; | |
5776 | } | |
5777 | } | |
5778 | return ret; | |
5779 | } | |
5780 | #else | |
5781 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5782 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5783 | { | |
5784 | return MC_TARGET_NONE; | |
5785 | } | |
5786 | #endif | |
5787 | ||
4ffef5fe DN |
5788 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5789 | unsigned long addr, unsigned long end, | |
5790 | struct mm_walk *walk) | |
5791 | { | |
26bcd64a | 5792 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5793 | pte_t *pte; |
5794 | spinlock_t *ptl; | |
5795 | ||
b6ec57f4 KS |
5796 | ptl = pmd_trans_huge_lock(pmd, vma); |
5797 | if (ptl) { | |
c733a828 JG |
5798 | /* |
5799 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
25b2995a CH |
5800 | * support transparent huge page with MEMORY_DEVICE_PRIVATE but |
5801 | * this might change. | |
c733a828 | 5802 | */ |
12724850 NH |
5803 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5804 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5805 | spin_unlock(ptl); |
1a5a9906 | 5806 | return 0; |
12724850 | 5807 | } |
03319327 | 5808 | |
45f83cef AA |
5809 | if (pmd_trans_unstable(pmd)) |
5810 | return 0; | |
4ffef5fe DN |
5811 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5812 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5813 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5814 | mc.precharge++; /* increment precharge temporarily */ |
5815 | pte_unmap_unlock(pte - 1, ptl); | |
5816 | cond_resched(); | |
5817 | ||
7dc74be0 DN |
5818 | return 0; |
5819 | } | |
5820 | ||
7b86ac33 CH |
5821 | static const struct mm_walk_ops precharge_walk_ops = { |
5822 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5823 | }; | |
5824 | ||
4ffef5fe DN |
5825 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5826 | { | |
5827 | unsigned long precharge; | |
4ffef5fe | 5828 | |
d8ed45c5 | 5829 | mmap_read_lock(mm); |
7b86ac33 | 5830 | walk_page_range(mm, 0, mm->highest_vm_end, &precharge_walk_ops, NULL); |
d8ed45c5 | 5831 | mmap_read_unlock(mm); |
4ffef5fe DN |
5832 | |
5833 | precharge = mc.precharge; | |
5834 | mc.precharge = 0; | |
5835 | ||
5836 | return precharge; | |
5837 | } | |
5838 | ||
4ffef5fe DN |
5839 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5840 | { | |
dfe076b0 DN |
5841 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5842 | ||
5843 | VM_BUG_ON(mc.moving_task); | |
5844 | mc.moving_task = current; | |
5845 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5846 | } |
5847 | ||
dfe076b0 DN |
5848 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5849 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5850 | { |
2bd9bb20 KH |
5851 | struct mem_cgroup *from = mc.from; |
5852 | struct mem_cgroup *to = mc.to; | |
5853 | ||
4ffef5fe | 5854 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5855 | if (mc.precharge) { |
00501b53 | 5856 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5857 | mc.precharge = 0; |
5858 | } | |
5859 | /* | |
5860 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5861 | * we must uncharge here. | |
5862 | */ | |
5863 | if (mc.moved_charge) { | |
00501b53 | 5864 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5865 | mc.moved_charge = 0; |
4ffef5fe | 5866 | } |
483c30b5 DN |
5867 | /* we must fixup refcnts and charges */ |
5868 | if (mc.moved_swap) { | |
483c30b5 | 5869 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5870 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5871 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5872 | |
615d66c3 VD |
5873 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
5874 | ||
05b84301 | 5875 | /* |
3e32cb2e JW |
5876 | * we charged both to->memory and to->memsw, so we |
5877 | * should uncharge to->memory. | |
05b84301 | 5878 | */ |
ce00a967 | 5879 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5880 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5881 | ||
483c30b5 DN |
5882 | mc.moved_swap = 0; |
5883 | } | |
dfe076b0 DN |
5884 | memcg_oom_recover(from); |
5885 | memcg_oom_recover(to); | |
5886 | wake_up_all(&mc.waitq); | |
5887 | } | |
5888 | ||
5889 | static void mem_cgroup_clear_mc(void) | |
5890 | { | |
264a0ae1 TH |
5891 | struct mm_struct *mm = mc.mm; |
5892 | ||
dfe076b0 DN |
5893 | /* |
5894 | * we must clear moving_task before waking up waiters at the end of | |
5895 | * task migration. | |
5896 | */ | |
5897 | mc.moving_task = NULL; | |
5898 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5899 | spin_lock(&mc.lock); |
4ffef5fe DN |
5900 | mc.from = NULL; |
5901 | mc.to = NULL; | |
264a0ae1 | 5902 | mc.mm = NULL; |
2bd9bb20 | 5903 | spin_unlock(&mc.lock); |
264a0ae1 TH |
5904 | |
5905 | mmput(mm); | |
4ffef5fe DN |
5906 | } |
5907 | ||
1f7dd3e5 | 5908 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5909 | { |
1f7dd3e5 | 5910 | struct cgroup_subsys_state *css; |
eed67d75 | 5911 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 5912 | struct mem_cgroup *from; |
4530eddb | 5913 | struct task_struct *leader, *p; |
9f2115f9 | 5914 | struct mm_struct *mm; |
1dfab5ab | 5915 | unsigned long move_flags; |
9f2115f9 | 5916 | int ret = 0; |
7dc74be0 | 5917 | |
1f7dd3e5 TH |
5918 | /* charge immigration isn't supported on the default hierarchy */ |
5919 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
5920 | return 0; |
5921 | ||
4530eddb TH |
5922 | /* |
5923 | * Multi-process migrations only happen on the default hierarchy | |
5924 | * where charge immigration is not used. Perform charge | |
5925 | * immigration if @tset contains a leader and whine if there are | |
5926 | * multiple. | |
5927 | */ | |
5928 | p = NULL; | |
1f7dd3e5 | 5929 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
5930 | WARN_ON_ONCE(p); |
5931 | p = leader; | |
1f7dd3e5 | 5932 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
5933 | } |
5934 | if (!p) | |
5935 | return 0; | |
5936 | ||
1f7dd3e5 TH |
5937 | /* |
5938 | * We are now commited to this value whatever it is. Changes in this | |
5939 | * tunable will only affect upcoming migrations, not the current one. | |
5940 | * So we need to save it, and keep it going. | |
5941 | */ | |
5942 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
5943 | if (!move_flags) | |
5944 | return 0; | |
5945 | ||
9f2115f9 TH |
5946 | from = mem_cgroup_from_task(p); |
5947 | ||
5948 | VM_BUG_ON(from == memcg); | |
5949 | ||
5950 | mm = get_task_mm(p); | |
5951 | if (!mm) | |
5952 | return 0; | |
5953 | /* We move charges only when we move a owner of the mm */ | |
5954 | if (mm->owner == p) { | |
5955 | VM_BUG_ON(mc.from); | |
5956 | VM_BUG_ON(mc.to); | |
5957 | VM_BUG_ON(mc.precharge); | |
5958 | VM_BUG_ON(mc.moved_charge); | |
5959 | VM_BUG_ON(mc.moved_swap); | |
5960 | ||
5961 | spin_lock(&mc.lock); | |
264a0ae1 | 5962 | mc.mm = mm; |
9f2115f9 TH |
5963 | mc.from = from; |
5964 | mc.to = memcg; | |
5965 | mc.flags = move_flags; | |
5966 | spin_unlock(&mc.lock); | |
5967 | /* We set mc.moving_task later */ | |
5968 | ||
5969 | ret = mem_cgroup_precharge_mc(mm); | |
5970 | if (ret) | |
5971 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
5972 | } else { |
5973 | mmput(mm); | |
7dc74be0 DN |
5974 | } |
5975 | return ret; | |
5976 | } | |
5977 | ||
1f7dd3e5 | 5978 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5979 | { |
4e2f245d JW |
5980 | if (mc.to) |
5981 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5982 | } |
5983 | ||
4ffef5fe DN |
5984 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5985 | unsigned long addr, unsigned long end, | |
5986 | struct mm_walk *walk) | |
7dc74be0 | 5987 | { |
4ffef5fe | 5988 | int ret = 0; |
26bcd64a | 5989 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5990 | pte_t *pte; |
5991 | spinlock_t *ptl; | |
12724850 NH |
5992 | enum mc_target_type target_type; |
5993 | union mc_target target; | |
5994 | struct page *page; | |
4ffef5fe | 5995 | |
b6ec57f4 KS |
5996 | ptl = pmd_trans_huge_lock(pmd, vma); |
5997 | if (ptl) { | |
62ade86a | 5998 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5999 | spin_unlock(ptl); |
12724850 NH |
6000 | return 0; |
6001 | } | |
6002 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
6003 | if (target_type == MC_TARGET_PAGE) { | |
6004 | page = target.page; | |
6005 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 6006 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 6007 | mc.from, mc.to)) { |
12724850 NH |
6008 | mc.precharge -= HPAGE_PMD_NR; |
6009 | mc.moved_charge += HPAGE_PMD_NR; | |
6010 | } | |
6011 | putback_lru_page(page); | |
6012 | } | |
6013 | put_page(page); | |
c733a828 JG |
6014 | } else if (target_type == MC_TARGET_DEVICE) { |
6015 | page = target.page; | |
6016 | if (!mem_cgroup_move_account(page, true, | |
6017 | mc.from, mc.to)) { | |
6018 | mc.precharge -= HPAGE_PMD_NR; | |
6019 | mc.moved_charge += HPAGE_PMD_NR; | |
6020 | } | |
6021 | put_page(page); | |
12724850 | 6022 | } |
bf929152 | 6023 | spin_unlock(ptl); |
1a5a9906 | 6024 | return 0; |
12724850 NH |
6025 | } |
6026 | ||
45f83cef AA |
6027 | if (pmd_trans_unstable(pmd)) |
6028 | return 0; | |
4ffef5fe DN |
6029 | retry: |
6030 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
6031 | for (; addr != end; addr += PAGE_SIZE) { | |
6032 | pte_t ptent = *(pte++); | |
c733a828 | 6033 | bool device = false; |
02491447 | 6034 | swp_entry_t ent; |
4ffef5fe DN |
6035 | |
6036 | if (!mc.precharge) | |
6037 | break; | |
6038 | ||
8d32ff84 | 6039 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
6040 | case MC_TARGET_DEVICE: |
6041 | device = true; | |
e4a9bc58 | 6042 | fallthrough; |
4ffef5fe DN |
6043 | case MC_TARGET_PAGE: |
6044 | page = target.page; | |
53f9263b KS |
6045 | /* |
6046 | * We can have a part of the split pmd here. Moving it | |
6047 | * can be done but it would be too convoluted so simply | |
6048 | * ignore such a partial THP and keep it in original | |
6049 | * memcg. There should be somebody mapping the head. | |
6050 | */ | |
6051 | if (PageTransCompound(page)) | |
6052 | goto put; | |
c733a828 | 6053 | if (!device && isolate_lru_page(page)) |
4ffef5fe | 6054 | goto put; |
f627c2f5 KS |
6055 | if (!mem_cgroup_move_account(page, false, |
6056 | mc.from, mc.to)) { | |
4ffef5fe | 6057 | mc.precharge--; |
854ffa8d DN |
6058 | /* we uncharge from mc.from later. */ |
6059 | mc.moved_charge++; | |
4ffef5fe | 6060 | } |
c733a828 JG |
6061 | if (!device) |
6062 | putback_lru_page(page); | |
8d32ff84 | 6063 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
6064 | put_page(page); |
6065 | break; | |
02491447 DN |
6066 | case MC_TARGET_SWAP: |
6067 | ent = target.ent; | |
e91cbb42 | 6068 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6069 | mc.precharge--; |
8d22a935 HD |
6070 | mem_cgroup_id_get_many(mc.to, 1); |
6071 | /* we fixup other refcnts and charges later. */ | |
483c30b5 DN |
6072 | mc.moved_swap++; |
6073 | } | |
02491447 | 6074 | break; |
4ffef5fe DN |
6075 | default: |
6076 | break; | |
6077 | } | |
6078 | } | |
6079 | pte_unmap_unlock(pte - 1, ptl); | |
6080 | cond_resched(); | |
6081 | ||
6082 | if (addr != end) { | |
6083 | /* | |
6084 | * We have consumed all precharges we got in can_attach(). | |
6085 | * We try charge one by one, but don't do any additional | |
6086 | * charges to mc.to if we have failed in charge once in attach() | |
6087 | * phase. | |
6088 | */ | |
854ffa8d | 6089 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6090 | if (!ret) |
6091 | goto retry; | |
6092 | } | |
6093 | ||
6094 | return ret; | |
6095 | } | |
6096 | ||
7b86ac33 CH |
6097 | static const struct mm_walk_ops charge_walk_ops = { |
6098 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6099 | }; | |
6100 | ||
264a0ae1 | 6101 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 6102 | { |
4ffef5fe | 6103 | lru_add_drain_all(); |
312722cb | 6104 | /* |
81f8c3a4 JW |
6105 | * Signal lock_page_memcg() to take the memcg's move_lock |
6106 | * while we're moving its pages to another memcg. Then wait | |
6107 | * for already started RCU-only updates to finish. | |
312722cb JW |
6108 | */ |
6109 | atomic_inc(&mc.from->moving_account); | |
6110 | synchronize_rcu(); | |
dfe076b0 | 6111 | retry: |
d8ed45c5 | 6112 | if (unlikely(!mmap_read_trylock(mc.mm))) { |
dfe076b0 | 6113 | /* |
c1e8d7c6 | 6114 | * Someone who are holding the mmap_lock might be waiting in |
dfe076b0 DN |
6115 | * waitq. So we cancel all extra charges, wake up all waiters, |
6116 | * and retry. Because we cancel precharges, we might not be able | |
6117 | * to move enough charges, but moving charge is a best-effort | |
6118 | * feature anyway, so it wouldn't be a big problem. | |
6119 | */ | |
6120 | __mem_cgroup_clear_mc(); | |
6121 | cond_resched(); | |
6122 | goto retry; | |
6123 | } | |
26bcd64a NH |
6124 | /* |
6125 | * When we have consumed all precharges and failed in doing | |
6126 | * additional charge, the page walk just aborts. | |
6127 | */ | |
7b86ac33 CH |
6128 | walk_page_range(mc.mm, 0, mc.mm->highest_vm_end, &charge_walk_ops, |
6129 | NULL); | |
0247f3f4 | 6130 | |
d8ed45c5 | 6131 | mmap_read_unlock(mc.mm); |
312722cb | 6132 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
6133 | } |
6134 | ||
264a0ae1 | 6135 | static void mem_cgroup_move_task(void) |
67e465a7 | 6136 | { |
264a0ae1 TH |
6137 | if (mc.to) { |
6138 | mem_cgroup_move_charge(); | |
a433658c | 6139 | mem_cgroup_clear_mc(); |
264a0ae1 | 6140 | } |
67e465a7 | 6141 | } |
5cfb80a7 | 6142 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 6143 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6144 | { |
6145 | return 0; | |
6146 | } | |
1f7dd3e5 | 6147 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6148 | { |
6149 | } | |
264a0ae1 | 6150 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
6151 | { |
6152 | } | |
6153 | #endif | |
67e465a7 | 6154 | |
677dc973 CD |
6155 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
6156 | { | |
6157 | if (value == PAGE_COUNTER_MAX) | |
6158 | seq_puts(m, "max\n"); | |
6159 | else | |
6160 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); | |
6161 | ||
6162 | return 0; | |
6163 | } | |
6164 | ||
241994ed JW |
6165 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
6166 | struct cftype *cft) | |
6167 | { | |
f5fc3c5d JW |
6168 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
6169 | ||
6170 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
6171 | } |
6172 | ||
bf8d5d52 RG |
6173 | static int memory_min_show(struct seq_file *m, void *v) |
6174 | { | |
677dc973 CD |
6175 | return seq_puts_memcg_tunable(m, |
6176 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); | |
bf8d5d52 RG |
6177 | } |
6178 | ||
6179 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
6180 | char *buf, size_t nbytes, loff_t off) | |
6181 | { | |
6182 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6183 | unsigned long min; | |
6184 | int err; | |
6185 | ||
6186 | buf = strstrip(buf); | |
6187 | err = page_counter_memparse(buf, "max", &min); | |
6188 | if (err) | |
6189 | return err; | |
6190 | ||
6191 | page_counter_set_min(&memcg->memory, min); | |
6192 | ||
6193 | return nbytes; | |
6194 | } | |
6195 | ||
241994ed JW |
6196 | static int memory_low_show(struct seq_file *m, void *v) |
6197 | { | |
677dc973 CD |
6198 | return seq_puts_memcg_tunable(m, |
6199 | READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); | |
241994ed JW |
6200 | } |
6201 | ||
6202 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
6203 | char *buf, size_t nbytes, loff_t off) | |
6204 | { | |
6205 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6206 | unsigned long low; | |
6207 | int err; | |
6208 | ||
6209 | buf = strstrip(buf); | |
d2973697 | 6210 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
6211 | if (err) |
6212 | return err; | |
6213 | ||
23067153 | 6214 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
6215 | |
6216 | return nbytes; | |
6217 | } | |
6218 | ||
6219 | static int memory_high_show(struct seq_file *m, void *v) | |
6220 | { | |
d1663a90 JK |
6221 | return seq_puts_memcg_tunable(m, |
6222 | READ_ONCE(mem_cgroup_from_seq(m)->memory.high)); | |
241994ed JW |
6223 | } |
6224 | ||
6225 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
6226 | char *buf, size_t nbytes, loff_t off) | |
6227 | { | |
6228 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6229 | unsigned int nr_retries = MAX_RECLAIM_RETRIES; |
8c8c383c | 6230 | bool drained = false; |
241994ed JW |
6231 | unsigned long high; |
6232 | int err; | |
6233 | ||
6234 | buf = strstrip(buf); | |
d2973697 | 6235 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
6236 | if (err) |
6237 | return err; | |
6238 | ||
e82553c1 JW |
6239 | page_counter_set_high(&memcg->memory, high); |
6240 | ||
8c8c383c JW |
6241 | for (;;) { |
6242 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6243 | unsigned long reclaimed; | |
6244 | ||
6245 | if (nr_pages <= high) | |
6246 | break; | |
6247 | ||
6248 | if (signal_pending(current)) | |
6249 | break; | |
6250 | ||
6251 | if (!drained) { | |
6252 | drain_all_stock(memcg); | |
6253 | drained = true; | |
6254 | continue; | |
6255 | } | |
6256 | ||
6257 | reclaimed = try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
6258 | GFP_KERNEL, true); | |
6259 | ||
6260 | if (!reclaimed && !nr_retries--) | |
6261 | break; | |
6262 | } | |
588083bb | 6263 | |
19ce33ac | 6264 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
6265 | return nbytes; |
6266 | } | |
6267 | ||
6268 | static int memory_max_show(struct seq_file *m, void *v) | |
6269 | { | |
677dc973 CD |
6270 | return seq_puts_memcg_tunable(m, |
6271 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); | |
241994ed JW |
6272 | } |
6273 | ||
6274 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
6275 | char *buf, size_t nbytes, loff_t off) | |
6276 | { | |
6277 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6278 | unsigned int nr_reclaims = MAX_RECLAIM_RETRIES; |
b6e6edcf | 6279 | bool drained = false; |
241994ed JW |
6280 | unsigned long max; |
6281 | int err; | |
6282 | ||
6283 | buf = strstrip(buf); | |
d2973697 | 6284 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
6285 | if (err) |
6286 | return err; | |
6287 | ||
bbec2e15 | 6288 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
6289 | |
6290 | for (;;) { | |
6291 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6292 | ||
6293 | if (nr_pages <= max) | |
6294 | break; | |
6295 | ||
7249c9f0 | 6296 | if (signal_pending(current)) |
b6e6edcf | 6297 | break; |
b6e6edcf JW |
6298 | |
6299 | if (!drained) { | |
6300 | drain_all_stock(memcg); | |
6301 | drained = true; | |
6302 | continue; | |
6303 | } | |
6304 | ||
6305 | if (nr_reclaims) { | |
6306 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
6307 | GFP_KERNEL, true)) | |
6308 | nr_reclaims--; | |
6309 | continue; | |
6310 | } | |
6311 | ||
e27be240 | 6312 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
6313 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
6314 | break; | |
6315 | } | |
241994ed | 6316 | |
2529bb3a | 6317 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
6318 | return nbytes; |
6319 | } | |
6320 | ||
1e577f97 SB |
6321 | static void __memory_events_show(struct seq_file *m, atomic_long_t *events) |
6322 | { | |
6323 | seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW])); | |
6324 | seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH])); | |
6325 | seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX])); | |
6326 | seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM])); | |
6327 | seq_printf(m, "oom_kill %lu\n", | |
6328 | atomic_long_read(&events[MEMCG_OOM_KILL])); | |
6329 | } | |
6330 | ||
241994ed JW |
6331 | static int memory_events_show(struct seq_file *m, void *v) |
6332 | { | |
aa9694bb | 6333 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
241994ed | 6334 | |
1e577f97 SB |
6335 | __memory_events_show(m, memcg->memory_events); |
6336 | return 0; | |
6337 | } | |
6338 | ||
6339 | static int memory_events_local_show(struct seq_file *m, void *v) | |
6340 | { | |
6341 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
241994ed | 6342 | |
1e577f97 | 6343 | __memory_events_show(m, memcg->memory_events_local); |
241994ed JW |
6344 | return 0; |
6345 | } | |
6346 | ||
587d9f72 JW |
6347 | static int memory_stat_show(struct seq_file *m, void *v) |
6348 | { | |
aa9694bb | 6349 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
c8713d0b | 6350 | char *buf; |
1ff9e6e1 | 6351 | |
c8713d0b JW |
6352 | buf = memory_stat_format(memcg); |
6353 | if (!buf) | |
6354 | return -ENOMEM; | |
6355 | seq_puts(m, buf); | |
6356 | kfree(buf); | |
587d9f72 JW |
6357 | return 0; |
6358 | } | |
6359 | ||
5f9a4f4a | 6360 | #ifdef CONFIG_NUMA |
fff66b79 MS |
6361 | static inline unsigned long lruvec_page_state_output(struct lruvec *lruvec, |
6362 | int item) | |
6363 | { | |
6364 | return lruvec_page_state(lruvec, item) * memcg_page_state_unit(item); | |
6365 | } | |
6366 | ||
5f9a4f4a MS |
6367 | static int memory_numa_stat_show(struct seq_file *m, void *v) |
6368 | { | |
6369 | int i; | |
6370 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
6371 | ||
6372 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { | |
6373 | int nid; | |
6374 | ||
6375 | if (memory_stats[i].idx >= NR_VM_NODE_STAT_ITEMS) | |
6376 | continue; | |
6377 | ||
6378 | seq_printf(m, "%s", memory_stats[i].name); | |
6379 | for_each_node_state(nid, N_MEMORY) { | |
6380 | u64 size; | |
6381 | struct lruvec *lruvec; | |
6382 | ||
6383 | lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); | |
fff66b79 MS |
6384 | size = lruvec_page_state_output(lruvec, |
6385 | memory_stats[i].idx); | |
5f9a4f4a MS |
6386 | seq_printf(m, " N%d=%llu", nid, size); |
6387 | } | |
6388 | seq_putc(m, '\n'); | |
6389 | } | |
6390 | ||
6391 | return 0; | |
6392 | } | |
6393 | #endif | |
6394 | ||
3d8b38eb RG |
6395 | static int memory_oom_group_show(struct seq_file *m, void *v) |
6396 | { | |
aa9694bb | 6397 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3d8b38eb RG |
6398 | |
6399 | seq_printf(m, "%d\n", memcg->oom_group); | |
6400 | ||
6401 | return 0; | |
6402 | } | |
6403 | ||
6404 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
6405 | char *buf, size_t nbytes, loff_t off) | |
6406 | { | |
6407 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6408 | int ret, oom_group; | |
6409 | ||
6410 | buf = strstrip(buf); | |
6411 | if (!buf) | |
6412 | return -EINVAL; | |
6413 | ||
6414 | ret = kstrtoint(buf, 0, &oom_group); | |
6415 | if (ret) | |
6416 | return ret; | |
6417 | ||
6418 | if (oom_group != 0 && oom_group != 1) | |
6419 | return -EINVAL; | |
6420 | ||
6421 | memcg->oom_group = oom_group; | |
6422 | ||
6423 | return nbytes; | |
6424 | } | |
6425 | ||
241994ed JW |
6426 | static struct cftype memory_files[] = { |
6427 | { | |
6428 | .name = "current", | |
f5fc3c5d | 6429 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
6430 | .read_u64 = memory_current_read, |
6431 | }, | |
bf8d5d52 RG |
6432 | { |
6433 | .name = "min", | |
6434 | .flags = CFTYPE_NOT_ON_ROOT, | |
6435 | .seq_show = memory_min_show, | |
6436 | .write = memory_min_write, | |
6437 | }, | |
241994ed JW |
6438 | { |
6439 | .name = "low", | |
6440 | .flags = CFTYPE_NOT_ON_ROOT, | |
6441 | .seq_show = memory_low_show, | |
6442 | .write = memory_low_write, | |
6443 | }, | |
6444 | { | |
6445 | .name = "high", | |
6446 | .flags = CFTYPE_NOT_ON_ROOT, | |
6447 | .seq_show = memory_high_show, | |
6448 | .write = memory_high_write, | |
6449 | }, | |
6450 | { | |
6451 | .name = "max", | |
6452 | .flags = CFTYPE_NOT_ON_ROOT, | |
6453 | .seq_show = memory_max_show, | |
6454 | .write = memory_max_write, | |
6455 | }, | |
6456 | { | |
6457 | .name = "events", | |
6458 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 6459 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
6460 | .seq_show = memory_events_show, |
6461 | }, | |
1e577f97 SB |
6462 | { |
6463 | .name = "events.local", | |
6464 | .flags = CFTYPE_NOT_ON_ROOT, | |
6465 | .file_offset = offsetof(struct mem_cgroup, events_local_file), | |
6466 | .seq_show = memory_events_local_show, | |
6467 | }, | |
587d9f72 JW |
6468 | { |
6469 | .name = "stat", | |
587d9f72 JW |
6470 | .seq_show = memory_stat_show, |
6471 | }, | |
5f9a4f4a MS |
6472 | #ifdef CONFIG_NUMA |
6473 | { | |
6474 | .name = "numa_stat", | |
6475 | .seq_show = memory_numa_stat_show, | |
6476 | }, | |
6477 | #endif | |
3d8b38eb RG |
6478 | { |
6479 | .name = "oom.group", | |
6480 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
6481 | .seq_show = memory_oom_group_show, | |
6482 | .write = memory_oom_group_write, | |
6483 | }, | |
241994ed JW |
6484 | { } /* terminate */ |
6485 | }; | |
6486 | ||
073219e9 | 6487 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 6488 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6489 | .css_online = mem_cgroup_css_online, |
92fb9748 | 6490 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 6491 | .css_released = mem_cgroup_css_released, |
92fb9748 | 6492 | .css_free = mem_cgroup_css_free, |
1ced953b | 6493 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
6494 | .can_attach = mem_cgroup_can_attach, |
6495 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 6496 | .post_attach = mem_cgroup_move_task, |
241994ed JW |
6497 | .dfl_cftypes = memory_files, |
6498 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 6499 | .early_init = 0, |
8cdea7c0 | 6500 | }; |
c077719b | 6501 | |
bc50bcc6 JW |
6502 | /* |
6503 | * This function calculates an individual cgroup's effective | |
6504 | * protection which is derived from its own memory.min/low, its | |
6505 | * parent's and siblings' settings, as well as the actual memory | |
6506 | * distribution in the tree. | |
6507 | * | |
6508 | * The following rules apply to the effective protection values: | |
6509 | * | |
6510 | * 1. At the first level of reclaim, effective protection is equal to | |
6511 | * the declared protection in memory.min and memory.low. | |
6512 | * | |
6513 | * 2. To enable safe delegation of the protection configuration, at | |
6514 | * subsequent levels the effective protection is capped to the | |
6515 | * parent's effective protection. | |
6516 | * | |
6517 | * 3. To make complex and dynamic subtrees easier to configure, the | |
6518 | * user is allowed to overcommit the declared protection at a given | |
6519 | * level. If that is the case, the parent's effective protection is | |
6520 | * distributed to the children in proportion to how much protection | |
6521 | * they have declared and how much of it they are utilizing. | |
6522 | * | |
6523 | * This makes distribution proportional, but also work-conserving: | |
6524 | * if one cgroup claims much more protection than it uses memory, | |
6525 | * the unused remainder is available to its siblings. | |
6526 | * | |
6527 | * 4. Conversely, when the declared protection is undercommitted at a | |
6528 | * given level, the distribution of the larger parental protection | |
6529 | * budget is NOT proportional. A cgroup's protection from a sibling | |
6530 | * is capped to its own memory.min/low setting. | |
6531 | * | |
8a931f80 JW |
6532 | * 5. However, to allow protecting recursive subtrees from each other |
6533 | * without having to declare each individual cgroup's fixed share | |
6534 | * of the ancestor's claim to protection, any unutilized - | |
6535 | * "floating" - protection from up the tree is distributed in | |
6536 | * proportion to each cgroup's *usage*. This makes the protection | |
6537 | * neutral wrt sibling cgroups and lets them compete freely over | |
6538 | * the shared parental protection budget, but it protects the | |
6539 | * subtree as a whole from neighboring subtrees. | |
6540 | * | |
6541 | * Note that 4. and 5. are not in conflict: 4. is about protecting | |
6542 | * against immediate siblings whereas 5. is about protecting against | |
6543 | * neighboring subtrees. | |
bc50bcc6 JW |
6544 | */ |
6545 | static unsigned long effective_protection(unsigned long usage, | |
8a931f80 | 6546 | unsigned long parent_usage, |
bc50bcc6 JW |
6547 | unsigned long setting, |
6548 | unsigned long parent_effective, | |
6549 | unsigned long siblings_protected) | |
6550 | { | |
6551 | unsigned long protected; | |
8a931f80 | 6552 | unsigned long ep; |
bc50bcc6 JW |
6553 | |
6554 | protected = min(usage, setting); | |
6555 | /* | |
6556 | * If all cgroups at this level combined claim and use more | |
6557 | * protection then what the parent affords them, distribute | |
6558 | * shares in proportion to utilization. | |
6559 | * | |
6560 | * We are using actual utilization rather than the statically | |
6561 | * claimed protection in order to be work-conserving: claimed | |
6562 | * but unused protection is available to siblings that would | |
6563 | * otherwise get a smaller chunk than what they claimed. | |
6564 | */ | |
6565 | if (siblings_protected > parent_effective) | |
6566 | return protected * parent_effective / siblings_protected; | |
6567 | ||
6568 | /* | |
6569 | * Ok, utilized protection of all children is within what the | |
6570 | * parent affords them, so we know whatever this child claims | |
6571 | * and utilizes is effectively protected. | |
6572 | * | |
6573 | * If there is unprotected usage beyond this value, reclaim | |
6574 | * will apply pressure in proportion to that amount. | |
6575 | * | |
6576 | * If there is unutilized protection, the cgroup will be fully | |
6577 | * shielded from reclaim, but we do return a smaller value for | |
6578 | * protection than what the group could enjoy in theory. This | |
6579 | * is okay. With the overcommit distribution above, effective | |
6580 | * protection is always dependent on how memory is actually | |
6581 | * consumed among the siblings anyway. | |
6582 | */ | |
8a931f80 JW |
6583 | ep = protected; |
6584 | ||
6585 | /* | |
6586 | * If the children aren't claiming (all of) the protection | |
6587 | * afforded to them by the parent, distribute the remainder in | |
6588 | * proportion to the (unprotected) memory of each cgroup. That | |
6589 | * way, cgroups that aren't explicitly prioritized wrt each | |
6590 | * other compete freely over the allowance, but they are | |
6591 | * collectively protected from neighboring trees. | |
6592 | * | |
6593 | * We're using unprotected memory for the weight so that if | |
6594 | * some cgroups DO claim explicit protection, we don't protect | |
6595 | * the same bytes twice. | |
cd324edc JW |
6596 | * |
6597 | * Check both usage and parent_usage against the respective | |
6598 | * protected values. One should imply the other, but they | |
6599 | * aren't read atomically - make sure the division is sane. | |
8a931f80 JW |
6600 | */ |
6601 | if (!(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)) | |
6602 | return ep; | |
cd324edc JW |
6603 | if (parent_effective > siblings_protected && |
6604 | parent_usage > siblings_protected && | |
6605 | usage > protected) { | |
8a931f80 JW |
6606 | unsigned long unclaimed; |
6607 | ||
6608 | unclaimed = parent_effective - siblings_protected; | |
6609 | unclaimed *= usage - protected; | |
6610 | unclaimed /= parent_usage - siblings_protected; | |
6611 | ||
6612 | ep += unclaimed; | |
6613 | } | |
6614 | ||
6615 | return ep; | |
bc50bcc6 JW |
6616 | } |
6617 | ||
241994ed | 6618 | /** |
bf8d5d52 | 6619 | * mem_cgroup_protected - check if memory consumption is in the normal range |
34c81057 | 6620 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
6621 | * @memcg: the memory cgroup to check |
6622 | * | |
23067153 RG |
6623 | * WARNING: This function is not stateless! It can only be used as part |
6624 | * of a top-down tree iteration, not for isolated queries. | |
241994ed | 6625 | */ |
45c7f7e1 CD |
6626 | void mem_cgroup_calculate_protection(struct mem_cgroup *root, |
6627 | struct mem_cgroup *memcg) | |
241994ed | 6628 | { |
8a931f80 | 6629 | unsigned long usage, parent_usage; |
23067153 RG |
6630 | struct mem_cgroup *parent; |
6631 | ||
241994ed | 6632 | if (mem_cgroup_disabled()) |
45c7f7e1 | 6633 | return; |
241994ed | 6634 | |
34c81057 SC |
6635 | if (!root) |
6636 | root = root_mem_cgroup; | |
22f7496f YS |
6637 | |
6638 | /* | |
6639 | * Effective values of the reclaim targets are ignored so they | |
6640 | * can be stale. Have a look at mem_cgroup_protection for more | |
6641 | * details. | |
6642 | * TODO: calculation should be more robust so that we do not need | |
6643 | * that special casing. | |
6644 | */ | |
34c81057 | 6645 | if (memcg == root) |
45c7f7e1 | 6646 | return; |
241994ed | 6647 | |
23067153 | 6648 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 | 6649 | if (!usage) |
45c7f7e1 | 6650 | return; |
bf8d5d52 | 6651 | |
bf8d5d52 | 6652 | parent = parent_mem_cgroup(memcg); |
df2a4196 RG |
6653 | /* No parent means a non-hierarchical mode on v1 memcg */ |
6654 | if (!parent) | |
45c7f7e1 | 6655 | return; |
df2a4196 | 6656 | |
bc50bcc6 | 6657 | if (parent == root) { |
c3d53200 | 6658 | memcg->memory.emin = READ_ONCE(memcg->memory.min); |
03960e33 | 6659 | memcg->memory.elow = READ_ONCE(memcg->memory.low); |
45c7f7e1 | 6660 | return; |
bf8d5d52 RG |
6661 | } |
6662 | ||
8a931f80 JW |
6663 | parent_usage = page_counter_read(&parent->memory); |
6664 | ||
b3a7822e | 6665 | WRITE_ONCE(memcg->memory.emin, effective_protection(usage, parent_usage, |
c3d53200 CD |
6666 | READ_ONCE(memcg->memory.min), |
6667 | READ_ONCE(parent->memory.emin), | |
b3a7822e | 6668 | atomic_long_read(&parent->memory.children_min_usage))); |
23067153 | 6669 | |
b3a7822e | 6670 | WRITE_ONCE(memcg->memory.elow, effective_protection(usage, parent_usage, |
03960e33 CD |
6671 | READ_ONCE(memcg->memory.low), |
6672 | READ_ONCE(parent->memory.elow), | |
b3a7822e | 6673 | atomic_long_read(&parent->memory.children_low_usage))); |
241994ed JW |
6674 | } |
6675 | ||
00501b53 | 6676 | /** |
f0e45fb4 | 6677 | * mem_cgroup_charge - charge a newly allocated page to a cgroup |
00501b53 JW |
6678 | * @page: page to charge |
6679 | * @mm: mm context of the victim | |
6680 | * @gfp_mask: reclaim mode | |
00501b53 JW |
6681 | * |
6682 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
6683 | * pages according to @gfp_mask if necessary. | |
6684 | * | |
f0e45fb4 | 6685 | * Returns 0 on success. Otherwise, an error code is returned. |
00501b53 | 6686 | */ |
d9eb1ea2 | 6687 | int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask) |
00501b53 | 6688 | { |
6c357848 | 6689 | unsigned int nr_pages = thp_nr_pages(page); |
00501b53 | 6690 | struct mem_cgroup *memcg = NULL; |
00501b53 JW |
6691 | int ret = 0; |
6692 | ||
6693 | if (mem_cgroup_disabled()) | |
6694 | goto out; | |
6695 | ||
6696 | if (PageSwapCache(page)) { | |
2d1c4980 JW |
6697 | swp_entry_t ent = { .val = page_private(page), }; |
6698 | unsigned short id; | |
6699 | ||
00501b53 JW |
6700 | /* |
6701 | * Every swap fault against a single page tries to charge the | |
6702 | * page, bail as early as possible. shmem_unuse() encounters | |
bcfe06bf RG |
6703 | * already charged pages, too. page and memcg binding is |
6704 | * protected by the page lock, which serializes swap cache | |
6705 | * removal, which in turn serializes uncharging. | |
00501b53 | 6706 | */ |
e993d905 | 6707 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
bcfe06bf | 6708 | if (page_memcg(compound_head(page))) |
00501b53 | 6709 | goto out; |
e993d905 | 6710 | |
2d1c4980 JW |
6711 | id = lookup_swap_cgroup_id(ent); |
6712 | rcu_read_lock(); | |
6713 | memcg = mem_cgroup_from_id(id); | |
6714 | if (memcg && !css_tryget_online(&memcg->css)) | |
6715 | memcg = NULL; | |
6716 | rcu_read_unlock(); | |
00501b53 JW |
6717 | } |
6718 | ||
00501b53 JW |
6719 | if (!memcg) |
6720 | memcg = get_mem_cgroup_from_mm(mm); | |
6721 | ||
6722 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
f0e45fb4 JW |
6723 | if (ret) |
6724 | goto out_put; | |
00501b53 | 6725 | |
1a3e1f40 | 6726 | css_get(&memcg->css); |
d9eb1ea2 | 6727 | commit_charge(page, memcg); |
6abb5a86 | 6728 | |
6abb5a86 | 6729 | local_irq_disable(); |
3fba69a5 | 6730 | mem_cgroup_charge_statistics(memcg, page, nr_pages); |
6abb5a86 JW |
6731 | memcg_check_events(memcg, page); |
6732 | local_irq_enable(); | |
00501b53 | 6733 | |
cae3af62 MS |
6734 | /* |
6735 | * Cgroup1's unified memory+swap counter has been charged with the | |
6736 | * new swapcache page, finish the transfer by uncharging the swap | |
6737 | * slot. The swap slot would also get uncharged when it dies, but | |
6738 | * it can stick around indefinitely and we'd count the page twice | |
6739 | * the entire time. | |
6740 | * | |
6741 | * Cgroup2 has separate resource counters for memory and swap, | |
6742 | * so this is a non-issue here. Memory and swap charge lifetimes | |
6743 | * correspond 1:1 to page and swap slot lifetimes: we charge the | |
6744 | * page to memory here, and uncharge swap when the slot is freed. | |
6745 | */ | |
6746 | if (do_memsw_account() && PageSwapCache(page)) { | |
00501b53 JW |
6747 | swp_entry_t entry = { .val = page_private(page) }; |
6748 | /* | |
6749 | * The swap entry might not get freed for a long time, | |
6750 | * let's not wait for it. The page already received a | |
6751 | * memory+swap charge, drop the swap entry duplicate. | |
6752 | */ | |
38d8b4e6 | 6753 | mem_cgroup_uncharge_swap(entry, nr_pages); |
00501b53 | 6754 | } |
00501b53 | 6755 | |
f0e45fb4 JW |
6756 | out_put: |
6757 | css_put(&memcg->css); | |
6758 | out: | |
6759 | return ret; | |
3fea5a49 JW |
6760 | } |
6761 | ||
a9d5adee JG |
6762 | struct uncharge_gather { |
6763 | struct mem_cgroup *memcg; | |
9f762dbe | 6764 | unsigned long nr_pages; |
a9d5adee | 6765 | unsigned long pgpgout; |
a9d5adee | 6766 | unsigned long nr_kmem; |
a9d5adee JG |
6767 | struct page *dummy_page; |
6768 | }; | |
6769 | ||
6770 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 6771 | { |
a9d5adee JG |
6772 | memset(ug, 0, sizeof(*ug)); |
6773 | } | |
6774 | ||
6775 | static void uncharge_batch(const struct uncharge_gather *ug) | |
6776 | { | |
747db954 JW |
6777 | unsigned long flags; |
6778 | ||
a9d5adee | 6779 | if (!mem_cgroup_is_root(ug->memcg)) { |
9f762dbe | 6780 | page_counter_uncharge(&ug->memcg->memory, ug->nr_pages); |
7941d214 | 6781 | if (do_memsw_account()) |
9f762dbe | 6782 | page_counter_uncharge(&ug->memcg->memsw, ug->nr_pages); |
a9d5adee JG |
6783 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem) |
6784 | page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem); | |
6785 | memcg_oom_recover(ug->memcg); | |
ce00a967 | 6786 | } |
747db954 JW |
6787 | |
6788 | local_irq_save(flags); | |
c9019e9b | 6789 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); |
9f762dbe | 6790 | __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, ug->nr_pages); |
a9d5adee | 6791 | memcg_check_events(ug->memcg, ug->dummy_page); |
747db954 | 6792 | local_irq_restore(flags); |
f1796544 MH |
6793 | |
6794 | /* drop reference from uncharge_page */ | |
6795 | css_put(&ug->memcg->css); | |
a9d5adee JG |
6796 | } |
6797 | ||
6798 | static void uncharge_page(struct page *page, struct uncharge_gather *ug) | |
6799 | { | |
9f762dbe JW |
6800 | unsigned long nr_pages; |
6801 | ||
a9d5adee | 6802 | VM_BUG_ON_PAGE(PageLRU(page), page); |
a9d5adee | 6803 | |
bcfe06bf | 6804 | if (!page_memcg(page)) |
a9d5adee JG |
6805 | return; |
6806 | ||
6807 | /* | |
6808 | * Nobody should be changing or seriously looking at | |
bcfe06bf | 6809 | * page_memcg(page) at this point, we have fully |
a9d5adee JG |
6810 | * exclusive access to the page. |
6811 | */ | |
6812 | ||
bcfe06bf | 6813 | if (ug->memcg != page_memcg(page)) { |
a9d5adee JG |
6814 | if (ug->memcg) { |
6815 | uncharge_batch(ug); | |
6816 | uncharge_gather_clear(ug); | |
6817 | } | |
bcfe06bf | 6818 | ug->memcg = page_memcg(page); |
f1796544 MH |
6819 | |
6820 | /* pairs with css_put in uncharge_batch */ | |
6821 | css_get(&ug->memcg->css); | |
a9d5adee JG |
6822 | } |
6823 | ||
9f762dbe JW |
6824 | nr_pages = compound_nr(page); |
6825 | ug->nr_pages += nr_pages; | |
a9d5adee | 6826 | |
18b2db3b | 6827 | if (PageMemcgKmem(page)) |
9f762dbe | 6828 | ug->nr_kmem += nr_pages; |
18b2db3b RG |
6829 | else |
6830 | ug->pgpgout++; | |
a9d5adee JG |
6831 | |
6832 | ug->dummy_page = page; | |
bcfe06bf | 6833 | page->memcg_data = 0; |
1a3e1f40 | 6834 | css_put(&ug->memcg->css); |
747db954 JW |
6835 | } |
6836 | ||
0a31bc97 JW |
6837 | /** |
6838 | * mem_cgroup_uncharge - uncharge a page | |
6839 | * @page: page to uncharge | |
6840 | * | |
f0e45fb4 | 6841 | * Uncharge a page previously charged with mem_cgroup_charge(). |
0a31bc97 JW |
6842 | */ |
6843 | void mem_cgroup_uncharge(struct page *page) | |
6844 | { | |
a9d5adee JG |
6845 | struct uncharge_gather ug; |
6846 | ||
0a31bc97 JW |
6847 | if (mem_cgroup_disabled()) |
6848 | return; | |
6849 | ||
747db954 | 6850 | /* Don't touch page->lru of any random page, pre-check: */ |
bcfe06bf | 6851 | if (!page_memcg(page)) |
0a31bc97 JW |
6852 | return; |
6853 | ||
a9d5adee JG |
6854 | uncharge_gather_clear(&ug); |
6855 | uncharge_page(page, &ug); | |
6856 | uncharge_batch(&ug); | |
747db954 | 6857 | } |
0a31bc97 | 6858 | |
747db954 JW |
6859 | /** |
6860 | * mem_cgroup_uncharge_list - uncharge a list of page | |
6861 | * @page_list: list of pages to uncharge | |
6862 | * | |
6863 | * Uncharge a list of pages previously charged with | |
f0e45fb4 | 6864 | * mem_cgroup_charge(). |
747db954 JW |
6865 | */ |
6866 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
6867 | { | |
c41a40b6 MS |
6868 | struct uncharge_gather ug; |
6869 | struct page *page; | |
6870 | ||
747db954 JW |
6871 | if (mem_cgroup_disabled()) |
6872 | return; | |
0a31bc97 | 6873 | |
c41a40b6 MS |
6874 | uncharge_gather_clear(&ug); |
6875 | list_for_each_entry(page, page_list, lru) | |
6876 | uncharge_page(page, &ug); | |
6877 | if (ug.memcg) | |
6878 | uncharge_batch(&ug); | |
0a31bc97 JW |
6879 | } |
6880 | ||
6881 | /** | |
6a93ca8f JW |
6882 | * mem_cgroup_migrate - charge a page's replacement |
6883 | * @oldpage: currently circulating page | |
6884 | * @newpage: replacement page | |
0a31bc97 | 6885 | * |
6a93ca8f JW |
6886 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
6887 | * be uncharged upon free. | |
0a31bc97 JW |
6888 | * |
6889 | * Both pages must be locked, @newpage->mapping must be set up. | |
6890 | */ | |
6a93ca8f | 6891 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 6892 | { |
29833315 | 6893 | struct mem_cgroup *memcg; |
44b7a8d3 | 6894 | unsigned int nr_pages; |
d93c4130 | 6895 | unsigned long flags; |
0a31bc97 JW |
6896 | |
6897 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6898 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 6899 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
6900 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6901 | newpage); | |
0a31bc97 JW |
6902 | |
6903 | if (mem_cgroup_disabled()) | |
6904 | return; | |
6905 | ||
6906 | /* Page cache replacement: new page already charged? */ | |
bcfe06bf | 6907 | if (page_memcg(newpage)) |
0a31bc97 JW |
6908 | return; |
6909 | ||
bcfe06bf | 6910 | memcg = page_memcg(oldpage); |
a4055888 | 6911 | VM_WARN_ON_ONCE_PAGE(!memcg, oldpage); |
29833315 | 6912 | if (!memcg) |
0a31bc97 JW |
6913 | return; |
6914 | ||
44b7a8d3 | 6915 | /* Force-charge the new page. The old one will be freed soon */ |
6c357848 | 6916 | nr_pages = thp_nr_pages(newpage); |
44b7a8d3 JW |
6917 | |
6918 | page_counter_charge(&memcg->memory, nr_pages); | |
6919 | if (do_memsw_account()) | |
6920 | page_counter_charge(&memcg->memsw, nr_pages); | |
0a31bc97 | 6921 | |
1a3e1f40 | 6922 | css_get(&memcg->css); |
d9eb1ea2 | 6923 | commit_charge(newpage, memcg); |
44b7a8d3 | 6924 | |
d93c4130 | 6925 | local_irq_save(flags); |
3fba69a5 | 6926 | mem_cgroup_charge_statistics(memcg, newpage, nr_pages); |
44b7a8d3 | 6927 | memcg_check_events(memcg, newpage); |
d93c4130 | 6928 | local_irq_restore(flags); |
0a31bc97 JW |
6929 | } |
6930 | ||
ef12947c | 6931 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
6932 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
6933 | ||
2d758073 | 6934 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
6935 | { |
6936 | struct mem_cgroup *memcg; | |
6937 | ||
2d758073 JW |
6938 | if (!mem_cgroup_sockets_enabled) |
6939 | return; | |
6940 | ||
e876ecc6 SB |
6941 | /* Do not associate the sock with unrelated interrupted task's memcg. */ |
6942 | if (in_interrupt()) | |
6943 | return; | |
6944 | ||
11092087 JW |
6945 | rcu_read_lock(); |
6946 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
6947 | if (memcg == root_mem_cgroup) |
6948 | goto out; | |
0db15298 | 6949 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 6950 | goto out; |
8965aa28 | 6951 | if (css_tryget(&memcg->css)) |
11092087 | 6952 | sk->sk_memcg = memcg; |
f7e1cb6e | 6953 | out: |
11092087 JW |
6954 | rcu_read_unlock(); |
6955 | } | |
11092087 | 6956 | |
2d758073 | 6957 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 6958 | { |
2d758073 JW |
6959 | if (sk->sk_memcg) |
6960 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
6961 | } |
6962 | ||
6963 | /** | |
6964 | * mem_cgroup_charge_skmem - charge socket memory | |
6965 | * @memcg: memcg to charge | |
6966 | * @nr_pages: number of pages to charge | |
6967 | * | |
6968 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
6969 | * @memcg's configured limit, %false if the charge had to be forced. | |
6970 | */ | |
6971 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6972 | { | |
f7e1cb6e | 6973 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 6974 | |
f7e1cb6e | 6975 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6976 | struct page_counter *fail; |
f7e1cb6e | 6977 | |
0db15298 JW |
6978 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
6979 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
6980 | return true; |
6981 | } | |
0db15298 JW |
6982 | page_counter_charge(&memcg->tcpmem, nr_pages); |
6983 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 6984 | return false; |
11092087 | 6985 | } |
d886f4e4 | 6986 | |
f7e1cb6e JW |
6987 | /* Don't block in the packet receive path */ |
6988 | if (in_softirq()) | |
6989 | gfp_mask = GFP_NOWAIT; | |
6990 | ||
c9019e9b | 6991 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); |
b2807f07 | 6992 | |
f7e1cb6e JW |
6993 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
6994 | return true; | |
6995 | ||
6996 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
6997 | return false; |
6998 | } | |
6999 | ||
7000 | /** | |
7001 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
7002 | * @memcg: memcg to uncharge |
7003 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
7004 | */ |
7005 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
7006 | { | |
f7e1cb6e | 7007 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 7008 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
7009 | return; |
7010 | } | |
d886f4e4 | 7011 | |
c9019e9b | 7012 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 7013 | |
475d0487 | 7014 | refill_stock(memcg, nr_pages); |
11092087 JW |
7015 | } |
7016 | ||
f7e1cb6e JW |
7017 | static int __init cgroup_memory(char *s) |
7018 | { | |
7019 | char *token; | |
7020 | ||
7021 | while ((token = strsep(&s, ",")) != NULL) { | |
7022 | if (!*token) | |
7023 | continue; | |
7024 | if (!strcmp(token, "nosocket")) | |
7025 | cgroup_memory_nosocket = true; | |
04823c83 VD |
7026 | if (!strcmp(token, "nokmem")) |
7027 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
7028 | } |
7029 | return 0; | |
7030 | } | |
7031 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 7032 | |
2d11085e | 7033 | /* |
1081312f MH |
7034 | * subsys_initcall() for memory controller. |
7035 | * | |
308167fc SAS |
7036 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
7037 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
7038 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
7039 | * should be initialized from here. | |
2d11085e MH |
7040 | */ |
7041 | static int __init mem_cgroup_init(void) | |
7042 | { | |
95a045f6 JW |
7043 | int cpu, node; |
7044 | ||
f3344adf MS |
7045 | /* |
7046 | * Currently s32 type (can refer to struct batched_lruvec_stat) is | |
7047 | * used for per-memcg-per-cpu caching of per-node statistics. In order | |
7048 | * to work fine, we should make sure that the overfill threshold can't | |
7049 | * exceed S32_MAX / PAGE_SIZE. | |
7050 | */ | |
7051 | BUILD_BUG_ON(MEMCG_CHARGE_BATCH > S32_MAX / PAGE_SIZE); | |
7052 | ||
308167fc SAS |
7053 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
7054 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
7055 | |
7056 | for_each_possible_cpu(cpu) | |
7057 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
7058 | drain_local_stock); | |
7059 | ||
7060 | for_each_node(node) { | |
7061 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
7062 | |
7063 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
7064 | node_online(node) ? node : NUMA_NO_NODE); | |
7065 | ||
ef8f2327 | 7066 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 7067 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 7068 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
7069 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
7070 | } | |
7071 | ||
2d11085e MH |
7072 | return 0; |
7073 | } | |
7074 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
7075 | |
7076 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
7077 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
7078 | { | |
1c2d479a | 7079 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
7080 | /* |
7081 | * The root cgroup cannot be destroyed, so it's refcount must | |
7082 | * always be >= 1. | |
7083 | */ | |
7084 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
7085 | VM_BUG_ON(1); | |
7086 | break; | |
7087 | } | |
7088 | memcg = parent_mem_cgroup(memcg); | |
7089 | if (!memcg) | |
7090 | memcg = root_mem_cgroup; | |
7091 | } | |
7092 | return memcg; | |
7093 | } | |
7094 | ||
21afa38e JW |
7095 | /** |
7096 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
7097 | * @page: page whose memsw charge to transfer | |
7098 | * @entry: swap entry to move the charge to | |
7099 | * | |
7100 | * Transfer the memsw charge of @page to @entry. | |
7101 | */ | |
7102 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
7103 | { | |
1f47b61f | 7104 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 7105 | unsigned int nr_entries; |
21afa38e JW |
7106 | unsigned short oldid; |
7107 | ||
7108 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
7109 | VM_BUG_ON_PAGE(page_count(page), page); | |
7110 | ||
76358ab5 AS |
7111 | if (mem_cgroup_disabled()) |
7112 | return; | |
7113 | ||
2d1c4980 | 7114 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
21afa38e JW |
7115 | return; |
7116 | ||
bcfe06bf | 7117 | memcg = page_memcg(page); |
21afa38e | 7118 | |
a4055888 | 7119 | VM_WARN_ON_ONCE_PAGE(!memcg, page); |
21afa38e JW |
7120 | if (!memcg) |
7121 | return; | |
7122 | ||
1f47b61f VD |
7123 | /* |
7124 | * In case the memcg owning these pages has been offlined and doesn't | |
7125 | * have an ID allocated to it anymore, charge the closest online | |
7126 | * ancestor for the swap instead and transfer the memory+swap charge. | |
7127 | */ | |
7128 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
6c357848 | 7129 | nr_entries = thp_nr_pages(page); |
d6810d73 HY |
7130 | /* Get references for the tail pages, too */ |
7131 | if (nr_entries > 1) | |
7132 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
7133 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
7134 | nr_entries); | |
21afa38e | 7135 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7136 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e | 7137 | |
bcfe06bf | 7138 | page->memcg_data = 0; |
21afa38e JW |
7139 | |
7140 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 7141 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 7142 | |
2d1c4980 | 7143 | if (!cgroup_memory_noswap && memcg != swap_memcg) { |
1f47b61f | 7144 | if (!mem_cgroup_is_root(swap_memcg)) |
d6810d73 HY |
7145 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
7146 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
7147 | } |
7148 | ||
ce9ce665 SAS |
7149 | /* |
7150 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 7151 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 7152 | * important here to have the interrupts disabled because it is the |
b93b0163 | 7153 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 SAS |
7154 | */ |
7155 | VM_BUG_ON(!irqs_disabled()); | |
3fba69a5 | 7156 | mem_cgroup_charge_statistics(memcg, page, -nr_entries); |
21afa38e | 7157 | memcg_check_events(memcg, page); |
73f576c0 | 7158 | |
1a3e1f40 | 7159 | css_put(&memcg->css); |
21afa38e JW |
7160 | } |
7161 | ||
38d8b4e6 HY |
7162 | /** |
7163 | * mem_cgroup_try_charge_swap - try charging swap space for a page | |
37e84351 VD |
7164 | * @page: page being added to swap |
7165 | * @entry: swap entry to charge | |
7166 | * | |
38d8b4e6 | 7167 | * Try to charge @page's memcg for the swap space at @entry. |
37e84351 VD |
7168 | * |
7169 | * Returns 0 on success, -ENOMEM on failure. | |
7170 | */ | |
7171 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
7172 | { | |
6c357848 | 7173 | unsigned int nr_pages = thp_nr_pages(page); |
37e84351 | 7174 | struct page_counter *counter; |
38d8b4e6 | 7175 | struct mem_cgroup *memcg; |
37e84351 VD |
7176 | unsigned short oldid; |
7177 | ||
76358ab5 AS |
7178 | if (mem_cgroup_disabled()) |
7179 | return 0; | |
7180 | ||
2d1c4980 | 7181 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
37e84351 VD |
7182 | return 0; |
7183 | ||
bcfe06bf | 7184 | memcg = page_memcg(page); |
37e84351 | 7185 | |
a4055888 | 7186 | VM_WARN_ON_ONCE_PAGE(!memcg, page); |
37e84351 VD |
7187 | if (!memcg) |
7188 | return 0; | |
7189 | ||
f3a53a3a TH |
7190 | if (!entry.val) { |
7191 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 7192 | return 0; |
f3a53a3a | 7193 | } |
bb98f2c5 | 7194 | |
1f47b61f VD |
7195 | memcg = mem_cgroup_id_get_online(memcg); |
7196 | ||
2d1c4980 | 7197 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg) && |
38d8b4e6 | 7198 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
7199 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
7200 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 7201 | mem_cgroup_id_put(memcg); |
37e84351 | 7202 | return -ENOMEM; |
1f47b61f | 7203 | } |
37e84351 | 7204 | |
38d8b4e6 HY |
7205 | /* Get references for the tail pages, too */ |
7206 | if (nr_pages > 1) | |
7207 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
7208 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
37e84351 | 7209 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7210 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 7211 | |
37e84351 VD |
7212 | return 0; |
7213 | } | |
7214 | ||
21afa38e | 7215 | /** |
38d8b4e6 | 7216 | * mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 7217 | * @entry: swap entry to uncharge |
38d8b4e6 | 7218 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 7219 | */ |
38d8b4e6 | 7220 | void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
7221 | { |
7222 | struct mem_cgroup *memcg; | |
7223 | unsigned short id; | |
7224 | ||
38d8b4e6 | 7225 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 7226 | rcu_read_lock(); |
adbe427b | 7227 | memcg = mem_cgroup_from_id(id); |
21afa38e | 7228 | if (memcg) { |
2d1c4980 | 7229 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg)) { |
37e84351 | 7230 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
38d8b4e6 | 7231 | page_counter_uncharge(&memcg->swap, nr_pages); |
37e84351 | 7232 | else |
38d8b4e6 | 7233 | page_counter_uncharge(&memcg->memsw, nr_pages); |
37e84351 | 7234 | } |
c9019e9b | 7235 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 7236 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
7237 | } |
7238 | rcu_read_unlock(); | |
7239 | } | |
7240 | ||
d8b38438 VD |
7241 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
7242 | { | |
7243 | long nr_swap_pages = get_nr_swap_pages(); | |
7244 | ||
eccb52e7 | 7245 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
d8b38438 VD |
7246 | return nr_swap_pages; |
7247 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
7248 | nr_swap_pages = min_t(long, nr_swap_pages, | |
bbec2e15 | 7249 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
7250 | page_counter_read(&memcg->swap)); |
7251 | return nr_swap_pages; | |
7252 | } | |
7253 | ||
5ccc5aba VD |
7254 | bool mem_cgroup_swap_full(struct page *page) |
7255 | { | |
7256 | struct mem_cgroup *memcg; | |
7257 | ||
7258 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
7259 | ||
7260 | if (vm_swap_full()) | |
7261 | return true; | |
eccb52e7 | 7262 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
5ccc5aba VD |
7263 | return false; |
7264 | ||
bcfe06bf | 7265 | memcg = page_memcg(page); |
5ccc5aba VD |
7266 | if (!memcg) |
7267 | return false; | |
7268 | ||
4b82ab4f JK |
7269 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { |
7270 | unsigned long usage = page_counter_read(&memcg->swap); | |
7271 | ||
7272 | if (usage * 2 >= READ_ONCE(memcg->swap.high) || | |
7273 | usage * 2 >= READ_ONCE(memcg->swap.max)) | |
5ccc5aba | 7274 | return true; |
4b82ab4f | 7275 | } |
5ccc5aba VD |
7276 | |
7277 | return false; | |
7278 | } | |
7279 | ||
eccb52e7 | 7280 | static int __init setup_swap_account(char *s) |
21afa38e JW |
7281 | { |
7282 | if (!strcmp(s, "1")) | |
5ab92901 | 7283 | cgroup_memory_noswap = false; |
21afa38e | 7284 | else if (!strcmp(s, "0")) |
5ab92901 | 7285 | cgroup_memory_noswap = true; |
21afa38e JW |
7286 | return 1; |
7287 | } | |
eccb52e7 | 7288 | __setup("swapaccount=", setup_swap_account); |
21afa38e | 7289 | |
37e84351 VD |
7290 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
7291 | struct cftype *cft) | |
7292 | { | |
7293 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
7294 | ||
7295 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
7296 | } | |
7297 | ||
4b82ab4f JK |
7298 | static int swap_high_show(struct seq_file *m, void *v) |
7299 | { | |
7300 | return seq_puts_memcg_tunable(m, | |
7301 | READ_ONCE(mem_cgroup_from_seq(m)->swap.high)); | |
7302 | } | |
7303 | ||
7304 | static ssize_t swap_high_write(struct kernfs_open_file *of, | |
7305 | char *buf, size_t nbytes, loff_t off) | |
7306 | { | |
7307 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7308 | unsigned long high; | |
7309 | int err; | |
7310 | ||
7311 | buf = strstrip(buf); | |
7312 | err = page_counter_memparse(buf, "max", &high); | |
7313 | if (err) | |
7314 | return err; | |
7315 | ||
7316 | page_counter_set_high(&memcg->swap, high); | |
7317 | ||
7318 | return nbytes; | |
7319 | } | |
7320 | ||
37e84351 VD |
7321 | static int swap_max_show(struct seq_file *m, void *v) |
7322 | { | |
677dc973 CD |
7323 | return seq_puts_memcg_tunable(m, |
7324 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); | |
37e84351 VD |
7325 | } |
7326 | ||
7327 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
7328 | char *buf, size_t nbytes, loff_t off) | |
7329 | { | |
7330 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7331 | unsigned long max; | |
7332 | int err; | |
7333 | ||
7334 | buf = strstrip(buf); | |
7335 | err = page_counter_memparse(buf, "max", &max); | |
7336 | if (err) | |
7337 | return err; | |
7338 | ||
be09102b | 7339 | xchg(&memcg->swap.max, max); |
37e84351 VD |
7340 | |
7341 | return nbytes; | |
7342 | } | |
7343 | ||
f3a53a3a TH |
7344 | static int swap_events_show(struct seq_file *m, void *v) |
7345 | { | |
aa9694bb | 7346 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
f3a53a3a | 7347 | |
4b82ab4f JK |
7348 | seq_printf(m, "high %lu\n", |
7349 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_HIGH])); | |
f3a53a3a TH |
7350 | seq_printf(m, "max %lu\n", |
7351 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
7352 | seq_printf(m, "fail %lu\n", | |
7353 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
7354 | ||
7355 | return 0; | |
7356 | } | |
7357 | ||
37e84351 VD |
7358 | static struct cftype swap_files[] = { |
7359 | { | |
7360 | .name = "swap.current", | |
7361 | .flags = CFTYPE_NOT_ON_ROOT, | |
7362 | .read_u64 = swap_current_read, | |
7363 | }, | |
4b82ab4f JK |
7364 | { |
7365 | .name = "swap.high", | |
7366 | .flags = CFTYPE_NOT_ON_ROOT, | |
7367 | .seq_show = swap_high_show, | |
7368 | .write = swap_high_write, | |
7369 | }, | |
37e84351 VD |
7370 | { |
7371 | .name = "swap.max", | |
7372 | .flags = CFTYPE_NOT_ON_ROOT, | |
7373 | .seq_show = swap_max_show, | |
7374 | .write = swap_max_write, | |
7375 | }, | |
f3a53a3a TH |
7376 | { |
7377 | .name = "swap.events", | |
7378 | .flags = CFTYPE_NOT_ON_ROOT, | |
7379 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
7380 | .seq_show = swap_events_show, | |
7381 | }, | |
37e84351 VD |
7382 | { } /* terminate */ |
7383 | }; | |
7384 | ||
eccb52e7 | 7385 | static struct cftype memsw_files[] = { |
21afa38e JW |
7386 | { |
7387 | .name = "memsw.usage_in_bytes", | |
7388 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
7389 | .read_u64 = mem_cgroup_read_u64, | |
7390 | }, | |
7391 | { | |
7392 | .name = "memsw.max_usage_in_bytes", | |
7393 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
7394 | .write = mem_cgroup_reset, | |
7395 | .read_u64 = mem_cgroup_read_u64, | |
7396 | }, | |
7397 | { | |
7398 | .name = "memsw.limit_in_bytes", | |
7399 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
7400 | .write = mem_cgroup_write, | |
7401 | .read_u64 = mem_cgroup_read_u64, | |
7402 | }, | |
7403 | { | |
7404 | .name = "memsw.failcnt", | |
7405 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
7406 | .write = mem_cgroup_reset, | |
7407 | .read_u64 = mem_cgroup_read_u64, | |
7408 | }, | |
7409 | { }, /* terminate */ | |
7410 | }; | |
7411 | ||
82ff165c BS |
7412 | /* |
7413 | * If mem_cgroup_swap_init() is implemented as a subsys_initcall() | |
7414 | * instead of a core_initcall(), this could mean cgroup_memory_noswap still | |
7415 | * remains set to false even when memcg is disabled via "cgroup_disable=memory" | |
7416 | * boot parameter. This may result in premature OOPS inside | |
7417 | * mem_cgroup_get_nr_swap_pages() function in corner cases. | |
7418 | */ | |
21afa38e JW |
7419 | static int __init mem_cgroup_swap_init(void) |
7420 | { | |
2d1c4980 JW |
7421 | /* No memory control -> no swap control */ |
7422 | if (mem_cgroup_disabled()) | |
7423 | cgroup_memory_noswap = true; | |
7424 | ||
7425 | if (cgroup_memory_noswap) | |
eccb52e7 JW |
7426 | return 0; |
7427 | ||
7428 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, swap_files)); | |
7429 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, memsw_files)); | |
7430 | ||
21afa38e JW |
7431 | return 0; |
7432 | } | |
82ff165c | 7433 | core_initcall(mem_cgroup_swap_init); |
21afa38e JW |
7434 | |
7435 | #endif /* CONFIG_MEMCG_SWAP */ |