]>
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 | ||
f745c6f5 SB |
1050 | /** |
1051 | * get_mem_cgroup_from_page: Obtain a reference on given page's memcg. | |
1052 | * @page: page from which memcg should be extracted. | |
1053 | * | |
1054 | * Obtain a reference on page->memcg and returns it if successful. Otherwise | |
1055 | * root_mem_cgroup is returned. | |
1056 | */ | |
1057 | struct mem_cgroup *get_mem_cgroup_from_page(struct page *page) | |
1058 | { | |
bcfe06bf | 1059 | struct mem_cgroup *memcg = page_memcg(page); |
f745c6f5 SB |
1060 | |
1061 | if (mem_cgroup_disabled()) | |
1062 | return NULL; | |
1063 | ||
1064 | rcu_read_lock(); | |
8965aa28 SB |
1065 | /* Page should not get uncharged and freed memcg under us. */ |
1066 | if (!memcg || WARN_ON_ONCE(!css_tryget(&memcg->css))) | |
f745c6f5 SB |
1067 | memcg = root_mem_cgroup; |
1068 | rcu_read_unlock(); | |
1069 | return memcg; | |
1070 | } | |
1071 | EXPORT_SYMBOL(get_mem_cgroup_from_page); | |
1072 | ||
37d5985c | 1073 | static __always_inline struct mem_cgroup *active_memcg(void) |
d46eb14b | 1074 | { |
37d5985c RG |
1075 | if (in_interrupt()) |
1076 | return this_cpu_read(int_active_memcg); | |
1077 | else | |
1078 | return current->active_memcg; | |
1079 | } | |
279c3393 | 1080 | |
37d5985c RG |
1081 | static __always_inline struct mem_cgroup *get_active_memcg(void) |
1082 | { | |
1083 | struct mem_cgroup *memcg; | |
d46eb14b | 1084 | |
37d5985c RG |
1085 | rcu_read_lock(); |
1086 | memcg = active_memcg(); | |
1087 | if (memcg) { | |
8965aa28 | 1088 | /* current->active_memcg must hold a ref. */ |
37d5985c | 1089 | if (WARN_ON_ONCE(!css_tryget(&memcg->css))) |
8965aa28 SB |
1090 | memcg = root_mem_cgroup; |
1091 | else | |
d46eb14b | 1092 | memcg = current->active_memcg; |
d46eb14b | 1093 | } |
37d5985c RG |
1094 | rcu_read_unlock(); |
1095 | ||
1096 | return memcg; | |
1097 | } | |
1098 | ||
4127c650 RG |
1099 | static __always_inline bool memcg_kmem_bypass(void) |
1100 | { | |
1101 | /* Allow remote memcg charging from any context. */ | |
1102 | if (unlikely(active_memcg())) | |
1103 | return false; | |
1104 | ||
1105 | /* Memcg to charge can't be determined. */ | |
1106 | if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD)) | |
1107 | return true; | |
1108 | ||
1109 | return false; | |
1110 | } | |
1111 | ||
37d5985c RG |
1112 | /** |
1113 | * If active memcg is set, do not fallback to current->mm->memcg. | |
1114 | */ | |
1115 | static __always_inline struct mem_cgroup *get_mem_cgroup_from_current(void) | |
1116 | { | |
1117 | if (memcg_kmem_bypass()) | |
1118 | return NULL; | |
1119 | ||
1120 | if (unlikely(active_memcg())) | |
1121 | return get_active_memcg(); | |
1122 | ||
d46eb14b SB |
1123 | return get_mem_cgroup_from_mm(current->mm); |
1124 | } | |
54595fe2 | 1125 | |
5660048c JW |
1126 | /** |
1127 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1128 | * @root: hierarchy root | |
1129 | * @prev: previously returned memcg, NULL on first invocation | |
1130 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1131 | * | |
1132 | * Returns references to children of the hierarchy below @root, or | |
1133 | * @root itself, or %NULL after a full round-trip. | |
1134 | * | |
1135 | * Caller must pass the return value in @prev on subsequent | |
1136 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1137 | * to cancel a hierarchy walk before the round-trip is complete. | |
1138 | * | |
05bdc520 ML |
1139 | * Reclaimers can specify a node in @reclaim to divide up the memcgs |
1140 | * in the hierarchy among all concurrent reclaimers operating on the | |
1141 | * same node. | |
5660048c | 1142 | */ |
694fbc0f | 1143 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1144 | struct mem_cgroup *prev, |
694fbc0f | 1145 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1146 | { |
3f649ab7 | 1147 | struct mem_cgroup_reclaim_iter *iter; |
5ac8fb31 | 1148 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 1149 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1150 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1151 | |
694fbc0f AM |
1152 | if (mem_cgroup_disabled()) |
1153 | return NULL; | |
5660048c | 1154 | |
9f3a0d09 JW |
1155 | if (!root) |
1156 | root = root_mem_cgroup; | |
7d74b06f | 1157 | |
9f3a0d09 | 1158 | if (prev && !reclaim) |
5ac8fb31 | 1159 | pos = prev; |
14067bb3 | 1160 | |
542f85f9 | 1161 | rcu_read_lock(); |
5f578161 | 1162 | |
5ac8fb31 | 1163 | if (reclaim) { |
ef8f2327 | 1164 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 1165 | |
ef8f2327 | 1166 | mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id); |
9da83f3f | 1167 | iter = &mz->iter; |
5ac8fb31 JW |
1168 | |
1169 | if (prev && reclaim->generation != iter->generation) | |
1170 | goto out_unlock; | |
1171 | ||
6df38689 | 1172 | while (1) { |
4db0c3c2 | 1173 | pos = READ_ONCE(iter->position); |
6df38689 VD |
1174 | if (!pos || css_tryget(&pos->css)) |
1175 | break; | |
5ac8fb31 | 1176 | /* |
6df38689 VD |
1177 | * css reference reached zero, so iter->position will |
1178 | * be cleared by ->css_released. However, we should not | |
1179 | * rely on this happening soon, because ->css_released | |
1180 | * is called from a work queue, and by busy-waiting we | |
1181 | * might block it. So we clear iter->position right | |
1182 | * away. | |
5ac8fb31 | 1183 | */ |
6df38689 VD |
1184 | (void)cmpxchg(&iter->position, pos, NULL); |
1185 | } | |
5ac8fb31 JW |
1186 | } |
1187 | ||
1188 | if (pos) | |
1189 | css = &pos->css; | |
1190 | ||
1191 | for (;;) { | |
1192 | css = css_next_descendant_pre(css, &root->css); | |
1193 | if (!css) { | |
1194 | /* | |
1195 | * Reclaimers share the hierarchy walk, and a | |
1196 | * new one might jump in right at the end of | |
1197 | * the hierarchy - make sure they see at least | |
1198 | * one group and restart from the beginning. | |
1199 | */ | |
1200 | if (!prev) | |
1201 | continue; | |
1202 | break; | |
527a5ec9 | 1203 | } |
7d74b06f | 1204 | |
5ac8fb31 JW |
1205 | /* |
1206 | * Verify the css and acquire a reference. The root | |
1207 | * is provided by the caller, so we know it's alive | |
1208 | * and kicking, and don't take an extra reference. | |
1209 | */ | |
1210 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 1211 | |
5ac8fb31 JW |
1212 | if (css == &root->css) |
1213 | break; | |
14067bb3 | 1214 | |
0b8f73e1 JW |
1215 | if (css_tryget(css)) |
1216 | break; | |
9f3a0d09 | 1217 | |
5ac8fb31 | 1218 | memcg = NULL; |
9f3a0d09 | 1219 | } |
5ac8fb31 JW |
1220 | |
1221 | if (reclaim) { | |
5ac8fb31 | 1222 | /* |
6df38689 VD |
1223 | * The position could have already been updated by a competing |
1224 | * thread, so check that the value hasn't changed since we read | |
1225 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 1226 | */ |
6df38689 VD |
1227 | (void)cmpxchg(&iter->position, pos, memcg); |
1228 | ||
5ac8fb31 JW |
1229 | if (pos) |
1230 | css_put(&pos->css); | |
1231 | ||
1232 | if (!memcg) | |
1233 | iter->generation++; | |
1234 | else if (!prev) | |
1235 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1236 | } |
5ac8fb31 | 1237 | |
542f85f9 MH |
1238 | out_unlock: |
1239 | rcu_read_unlock(); | |
c40046f3 MH |
1240 | if (prev && prev != root) |
1241 | css_put(&prev->css); | |
1242 | ||
9f3a0d09 | 1243 | return memcg; |
14067bb3 | 1244 | } |
7d74b06f | 1245 | |
5660048c JW |
1246 | /** |
1247 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1248 | * @root: hierarchy root | |
1249 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1250 | */ | |
1251 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1252 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1253 | { |
1254 | if (!root) | |
1255 | root = root_mem_cgroup; | |
1256 | if (prev && prev != root) | |
1257 | css_put(&prev->css); | |
1258 | } | |
7d74b06f | 1259 | |
54a83d6b MC |
1260 | static void __invalidate_reclaim_iterators(struct mem_cgroup *from, |
1261 | struct mem_cgroup *dead_memcg) | |
6df38689 | 1262 | { |
6df38689 | 1263 | struct mem_cgroup_reclaim_iter *iter; |
ef8f2327 MG |
1264 | struct mem_cgroup_per_node *mz; |
1265 | int nid; | |
6df38689 | 1266 | |
54a83d6b MC |
1267 | for_each_node(nid) { |
1268 | mz = mem_cgroup_nodeinfo(from, nid); | |
9da83f3f YS |
1269 | iter = &mz->iter; |
1270 | cmpxchg(&iter->position, dead_memcg, NULL); | |
6df38689 VD |
1271 | } |
1272 | } | |
1273 | ||
54a83d6b MC |
1274 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1275 | { | |
1276 | struct mem_cgroup *memcg = dead_memcg; | |
1277 | struct mem_cgroup *last; | |
1278 | ||
1279 | do { | |
1280 | __invalidate_reclaim_iterators(memcg, dead_memcg); | |
1281 | last = memcg; | |
1282 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1283 | ||
1284 | /* | |
1285 | * When cgruop1 non-hierarchy mode is used, | |
1286 | * parent_mem_cgroup() does not walk all the way up to the | |
1287 | * cgroup root (root_mem_cgroup). So we have to handle | |
1288 | * dead_memcg from cgroup root separately. | |
1289 | */ | |
1290 | if (last != root_mem_cgroup) | |
1291 | __invalidate_reclaim_iterators(root_mem_cgroup, | |
1292 | dead_memcg); | |
1293 | } | |
1294 | ||
7c5f64f8 VD |
1295 | /** |
1296 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1297 | * @memcg: hierarchy root | |
1298 | * @fn: function to call for each task | |
1299 | * @arg: argument passed to @fn | |
1300 | * | |
1301 | * This function iterates over tasks attached to @memcg or to any of its | |
1302 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
1303 | * value, the function breaks the iteration loop and returns the value. | |
1304 | * Otherwise, it will iterate over all tasks and return 0. | |
1305 | * | |
1306 | * This function must not be called for the root memory cgroup. | |
1307 | */ | |
1308 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
1309 | int (*fn)(struct task_struct *, void *), void *arg) | |
1310 | { | |
1311 | struct mem_cgroup *iter; | |
1312 | int ret = 0; | |
1313 | ||
1314 | BUG_ON(memcg == root_mem_cgroup); | |
1315 | ||
1316 | for_each_mem_cgroup_tree(iter, memcg) { | |
1317 | struct css_task_iter it; | |
1318 | struct task_struct *task; | |
1319 | ||
f168a9a5 | 1320 | css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it); |
7c5f64f8 VD |
1321 | while (!ret && (task = css_task_iter_next(&it))) |
1322 | ret = fn(task, arg); | |
1323 | css_task_iter_end(&it); | |
1324 | if (ret) { | |
1325 | mem_cgroup_iter_break(memcg, iter); | |
1326 | break; | |
1327 | } | |
1328 | } | |
1329 | return ret; | |
1330 | } | |
1331 | ||
6168d0da AS |
1332 | #ifdef CONFIG_DEBUG_VM |
1333 | void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page) | |
1334 | { | |
1335 | struct mem_cgroup *memcg; | |
1336 | ||
1337 | if (mem_cgroup_disabled()) | |
1338 | return; | |
1339 | ||
1340 | memcg = page_memcg(page); | |
1341 | ||
1342 | if (!memcg) | |
1343 | VM_BUG_ON_PAGE(lruvec_memcg(lruvec) != root_mem_cgroup, page); | |
1344 | else | |
1345 | VM_BUG_ON_PAGE(lruvec_memcg(lruvec) != memcg, page); | |
1346 | } | |
1347 | #endif | |
1348 | ||
6168d0da AS |
1349 | /** |
1350 | * lock_page_lruvec - lock and return lruvec for a given page. | |
1351 | * @page: the page | |
1352 | * | |
d7e3aba5 AS |
1353 | * These functions are safe to use under any of the following conditions: |
1354 | * - page locked | |
1355 | * - PageLRU cleared | |
1356 | * - lock_page_memcg() | |
1357 | * - page->_refcount is zero | |
6168d0da AS |
1358 | */ |
1359 | struct lruvec *lock_page_lruvec(struct page *page) | |
1360 | { | |
1361 | struct lruvec *lruvec; | |
1362 | struct pglist_data *pgdat = page_pgdat(page); | |
1363 | ||
6168d0da AS |
1364 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
1365 | spin_lock(&lruvec->lru_lock); | |
6168d0da AS |
1366 | |
1367 | lruvec_memcg_debug(lruvec, page); | |
1368 | ||
1369 | return lruvec; | |
1370 | } | |
1371 | ||
1372 | struct lruvec *lock_page_lruvec_irq(struct page *page) | |
1373 | { | |
1374 | struct lruvec *lruvec; | |
1375 | struct pglist_data *pgdat = page_pgdat(page); | |
1376 | ||
6168d0da AS |
1377 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
1378 | spin_lock_irq(&lruvec->lru_lock); | |
6168d0da AS |
1379 | |
1380 | lruvec_memcg_debug(lruvec, page); | |
1381 | ||
1382 | return lruvec; | |
1383 | } | |
1384 | ||
1385 | struct lruvec *lock_page_lruvec_irqsave(struct page *page, unsigned long *flags) | |
1386 | { | |
1387 | struct lruvec *lruvec; | |
1388 | struct pglist_data *pgdat = page_pgdat(page); | |
1389 | ||
6168d0da AS |
1390 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
1391 | spin_lock_irqsave(&lruvec->lru_lock, *flags); | |
6168d0da AS |
1392 | |
1393 | lruvec_memcg_debug(lruvec, page); | |
1394 | ||
1395 | return lruvec; | |
1396 | } | |
1397 | ||
925b7673 | 1398 | /** |
fa9add64 HD |
1399 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1400 | * @lruvec: mem_cgroup per zone lru vector | |
1401 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1402 | * @zid: zone id of the accounted pages |
fa9add64 | 1403 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1404 | * |
ca707239 HD |
1405 | * This function must be called under lru_lock, just before a page is added |
1406 | * to or just after a page is removed from an lru list (that ordering being | |
1407 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 1408 | */ |
fa9add64 | 1409 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1410 | int zid, int nr_pages) |
3f58a829 | 1411 | { |
ef8f2327 | 1412 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1413 | unsigned long *lru_size; |
ca707239 | 1414 | long size; |
3f58a829 MK |
1415 | |
1416 | if (mem_cgroup_disabled()) | |
1417 | return; | |
1418 | ||
ef8f2327 | 1419 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1420 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1421 | |
1422 | if (nr_pages < 0) | |
1423 | *lru_size += nr_pages; | |
1424 | ||
1425 | size = *lru_size; | |
b4536f0c MH |
1426 | if (WARN_ONCE(size < 0, |
1427 | "%s(%p, %d, %d): lru_size %ld\n", | |
1428 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1429 | VM_BUG_ON(1); |
1430 | *lru_size = 0; | |
1431 | } | |
1432 | ||
1433 | if (nr_pages > 0) | |
1434 | *lru_size += nr_pages; | |
08e552c6 | 1435 | } |
544122e5 | 1436 | |
19942822 | 1437 | /** |
9d11ea9f | 1438 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1439 | * @memcg: the memory cgroup |
19942822 | 1440 | * |
9d11ea9f | 1441 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1442 | * pages. |
19942822 | 1443 | */ |
c0ff4b85 | 1444 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1445 | { |
3e32cb2e JW |
1446 | unsigned long margin = 0; |
1447 | unsigned long count; | |
1448 | unsigned long limit; | |
9d11ea9f | 1449 | |
3e32cb2e | 1450 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1451 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1452 | if (count < limit) |
1453 | margin = limit - count; | |
1454 | ||
7941d214 | 1455 | if (do_memsw_account()) { |
3e32cb2e | 1456 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1457 | limit = READ_ONCE(memcg->memsw.max); |
1c4448ed | 1458 | if (count < limit) |
3e32cb2e | 1459 | margin = min(margin, limit - count); |
cbedbac3 LR |
1460 | else |
1461 | margin = 0; | |
3e32cb2e JW |
1462 | } |
1463 | ||
1464 | return margin; | |
19942822 JW |
1465 | } |
1466 | ||
32047e2a | 1467 | /* |
bdcbb659 | 1468 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1469 | * |
bdcbb659 QH |
1470 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1471 | * moving cgroups. This is for waiting at high-memory pressure | |
1472 | * caused by "move". | |
32047e2a | 1473 | */ |
c0ff4b85 | 1474 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1475 | { |
2bd9bb20 KH |
1476 | struct mem_cgroup *from; |
1477 | struct mem_cgroup *to; | |
4b534334 | 1478 | bool ret = false; |
2bd9bb20 KH |
1479 | /* |
1480 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1481 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1482 | */ | |
1483 | spin_lock(&mc.lock); | |
1484 | from = mc.from; | |
1485 | to = mc.to; | |
1486 | if (!from) | |
1487 | goto unlock; | |
3e92041d | 1488 | |
2314b42d JW |
1489 | ret = mem_cgroup_is_descendant(from, memcg) || |
1490 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1491 | unlock: |
1492 | spin_unlock(&mc.lock); | |
4b534334 KH |
1493 | return ret; |
1494 | } | |
1495 | ||
c0ff4b85 | 1496 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1497 | { |
1498 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1499 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1500 | DEFINE_WAIT(wait); |
1501 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1502 | /* moving charge context might have finished. */ | |
1503 | if (mc.moving_task) | |
1504 | schedule(); | |
1505 | finish_wait(&mc.waitq, &wait); | |
1506 | return true; | |
1507 | } | |
1508 | } | |
1509 | return false; | |
1510 | } | |
1511 | ||
5f9a4f4a MS |
1512 | struct memory_stat { |
1513 | const char *name; | |
5f9a4f4a MS |
1514 | unsigned int idx; |
1515 | }; | |
1516 | ||
57b2847d | 1517 | static const struct memory_stat memory_stats[] = { |
fff66b79 MS |
1518 | { "anon", NR_ANON_MAPPED }, |
1519 | { "file", NR_FILE_PAGES }, | |
1520 | { "kernel_stack", NR_KERNEL_STACK_KB }, | |
1521 | { "pagetables", NR_PAGETABLE }, | |
1522 | { "percpu", MEMCG_PERCPU_B }, | |
1523 | { "sock", MEMCG_SOCK }, | |
1524 | { "shmem", NR_SHMEM }, | |
1525 | { "file_mapped", NR_FILE_MAPPED }, | |
1526 | { "file_dirty", NR_FILE_DIRTY }, | |
1527 | { "file_writeback", NR_WRITEBACK }, | |
b6038942 SB |
1528 | #ifdef CONFIG_SWAP |
1529 | { "swapcached", NR_SWAPCACHE }, | |
1530 | #endif | |
5f9a4f4a | 1531 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
fff66b79 MS |
1532 | { "anon_thp", NR_ANON_THPS }, |
1533 | { "file_thp", NR_FILE_THPS }, | |
1534 | { "shmem_thp", NR_SHMEM_THPS }, | |
5f9a4f4a | 1535 | #endif |
fff66b79 MS |
1536 | { "inactive_anon", NR_INACTIVE_ANON }, |
1537 | { "active_anon", NR_ACTIVE_ANON }, | |
1538 | { "inactive_file", NR_INACTIVE_FILE }, | |
1539 | { "active_file", NR_ACTIVE_FILE }, | |
1540 | { "unevictable", NR_UNEVICTABLE }, | |
1541 | { "slab_reclaimable", NR_SLAB_RECLAIMABLE_B }, | |
1542 | { "slab_unreclaimable", NR_SLAB_UNRECLAIMABLE_B }, | |
5f9a4f4a MS |
1543 | |
1544 | /* The memory events */ | |
fff66b79 MS |
1545 | { "workingset_refault_anon", WORKINGSET_REFAULT_ANON }, |
1546 | { "workingset_refault_file", WORKINGSET_REFAULT_FILE }, | |
1547 | { "workingset_activate_anon", WORKINGSET_ACTIVATE_ANON }, | |
1548 | { "workingset_activate_file", WORKINGSET_ACTIVATE_FILE }, | |
1549 | { "workingset_restore_anon", WORKINGSET_RESTORE_ANON }, | |
1550 | { "workingset_restore_file", WORKINGSET_RESTORE_FILE }, | |
1551 | { "workingset_nodereclaim", WORKINGSET_NODERECLAIM }, | |
5f9a4f4a MS |
1552 | }; |
1553 | ||
fff66b79 MS |
1554 | /* Translate stat items to the correct unit for memory.stat output */ |
1555 | static int memcg_page_state_unit(int item) | |
1556 | { | |
1557 | switch (item) { | |
1558 | case MEMCG_PERCPU_B: | |
1559 | case NR_SLAB_RECLAIMABLE_B: | |
1560 | case NR_SLAB_UNRECLAIMABLE_B: | |
1561 | case WORKINGSET_REFAULT_ANON: | |
1562 | case WORKINGSET_REFAULT_FILE: | |
1563 | case WORKINGSET_ACTIVATE_ANON: | |
1564 | case WORKINGSET_ACTIVATE_FILE: | |
1565 | case WORKINGSET_RESTORE_ANON: | |
1566 | case WORKINGSET_RESTORE_FILE: | |
1567 | case WORKINGSET_NODERECLAIM: | |
1568 | return 1; | |
1569 | case NR_KERNEL_STACK_KB: | |
1570 | return SZ_1K; | |
1571 | default: | |
1572 | return PAGE_SIZE; | |
1573 | } | |
1574 | } | |
1575 | ||
1576 | static inline unsigned long memcg_page_state_output(struct mem_cgroup *memcg, | |
1577 | int item) | |
1578 | { | |
1579 | return memcg_page_state(memcg, item) * memcg_page_state_unit(item); | |
1580 | } | |
1581 | ||
c8713d0b JW |
1582 | static char *memory_stat_format(struct mem_cgroup *memcg) |
1583 | { | |
1584 | struct seq_buf s; | |
1585 | int i; | |
71cd3113 | 1586 | |
c8713d0b JW |
1587 | seq_buf_init(&s, kmalloc(PAGE_SIZE, GFP_KERNEL), PAGE_SIZE); |
1588 | if (!s.buffer) | |
1589 | return NULL; | |
1590 | ||
1591 | /* | |
1592 | * Provide statistics on the state of the memory subsystem as | |
1593 | * well as cumulative event counters that show past behavior. | |
1594 | * | |
1595 | * This list is ordered following a combination of these gradients: | |
1596 | * 1) generic big picture -> specifics and details | |
1597 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
1598 | * | |
1599 | * Current memory state: | |
1600 | */ | |
1601 | ||
5f9a4f4a MS |
1602 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { |
1603 | u64 size; | |
c8713d0b | 1604 | |
fff66b79 | 1605 | size = memcg_page_state_output(memcg, memory_stats[i].idx); |
5f9a4f4a | 1606 | seq_buf_printf(&s, "%s %llu\n", memory_stats[i].name, size); |
c8713d0b | 1607 | |
5f9a4f4a | 1608 | if (unlikely(memory_stats[i].idx == NR_SLAB_UNRECLAIMABLE_B)) { |
fff66b79 MS |
1609 | size += memcg_page_state_output(memcg, |
1610 | NR_SLAB_RECLAIMABLE_B); | |
5f9a4f4a MS |
1611 | seq_buf_printf(&s, "slab %llu\n", size); |
1612 | } | |
1613 | } | |
c8713d0b JW |
1614 | |
1615 | /* Accumulated memory events */ | |
1616 | ||
ebc5d83d KK |
1617 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGFAULT), |
1618 | memcg_events(memcg, PGFAULT)); | |
1619 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGMAJFAULT), | |
1620 | memcg_events(memcg, PGMAJFAULT)); | |
ebc5d83d KK |
1621 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGREFILL), |
1622 | memcg_events(memcg, PGREFILL)); | |
c8713d0b JW |
1623 | seq_buf_printf(&s, "pgscan %lu\n", |
1624 | memcg_events(memcg, PGSCAN_KSWAPD) + | |
1625 | memcg_events(memcg, PGSCAN_DIRECT)); | |
1626 | seq_buf_printf(&s, "pgsteal %lu\n", | |
1627 | memcg_events(memcg, PGSTEAL_KSWAPD) + | |
1628 | memcg_events(memcg, PGSTEAL_DIRECT)); | |
ebc5d83d KK |
1629 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGACTIVATE), |
1630 | memcg_events(memcg, PGACTIVATE)); | |
1631 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGDEACTIVATE), | |
1632 | memcg_events(memcg, PGDEACTIVATE)); | |
1633 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREE), | |
1634 | memcg_events(memcg, PGLAZYFREE)); | |
1635 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREED), | |
1636 | memcg_events(memcg, PGLAZYFREED)); | |
c8713d0b JW |
1637 | |
1638 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
ebc5d83d | 1639 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_FAULT_ALLOC), |
c8713d0b | 1640 | memcg_events(memcg, THP_FAULT_ALLOC)); |
ebc5d83d | 1641 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_COLLAPSE_ALLOC), |
c8713d0b JW |
1642 | memcg_events(memcg, THP_COLLAPSE_ALLOC)); |
1643 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ | |
1644 | ||
1645 | /* The above should easily fit into one page */ | |
1646 | WARN_ON_ONCE(seq_buf_has_overflowed(&s)); | |
1647 | ||
1648 | return s.buffer; | |
1649 | } | |
71cd3113 | 1650 | |
58cf188e | 1651 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1652 | /** |
f0c867d9 | 1653 | * mem_cgroup_print_oom_context: Print OOM information relevant to |
1654 | * memory controller. | |
e222432b BS |
1655 | * @memcg: The memory cgroup that went over limit |
1656 | * @p: Task that is going to be killed | |
1657 | * | |
1658 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1659 | * enabled | |
1660 | */ | |
f0c867d9 | 1661 | void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) |
e222432b | 1662 | { |
e222432b BS |
1663 | rcu_read_lock(); |
1664 | ||
f0c867d9 | 1665 | if (memcg) { |
1666 | pr_cont(",oom_memcg="); | |
1667 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1668 | } else | |
1669 | pr_cont(",global_oom"); | |
2415b9f5 | 1670 | if (p) { |
f0c867d9 | 1671 | pr_cont(",task_memcg="); |
2415b9f5 | 1672 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
2415b9f5 | 1673 | } |
e222432b | 1674 | rcu_read_unlock(); |
f0c867d9 | 1675 | } |
1676 | ||
1677 | /** | |
1678 | * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to | |
1679 | * memory controller. | |
1680 | * @memcg: The memory cgroup that went over limit | |
1681 | */ | |
1682 | void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) | |
1683 | { | |
c8713d0b | 1684 | char *buf; |
e222432b | 1685 | |
3e32cb2e JW |
1686 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1687 | K((u64)page_counter_read(&memcg->memory)), | |
15b42562 | 1688 | K((u64)READ_ONCE(memcg->memory.max)), memcg->memory.failcnt); |
c8713d0b JW |
1689 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
1690 | pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1691 | K((u64)page_counter_read(&memcg->swap)), | |
32d087cd | 1692 | K((u64)READ_ONCE(memcg->swap.max)), memcg->swap.failcnt); |
c8713d0b JW |
1693 | else { |
1694 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1695 | K((u64)page_counter_read(&memcg->memsw)), | |
1696 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); | |
1697 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1698 | K((u64)page_counter_read(&memcg->kmem)), | |
1699 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); | |
58cf188e | 1700 | } |
c8713d0b JW |
1701 | |
1702 | pr_info("Memory cgroup stats for "); | |
1703 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1704 | pr_cont(":"); | |
1705 | buf = memory_stat_format(memcg); | |
1706 | if (!buf) | |
1707 | return; | |
1708 | pr_info("%s", buf); | |
1709 | kfree(buf); | |
e222432b BS |
1710 | } |
1711 | ||
a63d83f4 DR |
1712 | /* |
1713 | * Return the memory (and swap, if configured) limit for a memcg. | |
1714 | */ | |
bbec2e15 | 1715 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1716 | { |
8d387a5f WL |
1717 | unsigned long max = READ_ONCE(memcg->memory.max); |
1718 | ||
1719 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) { | |
1720 | if (mem_cgroup_swappiness(memcg)) | |
1721 | max += min(READ_ONCE(memcg->swap.max), | |
1722 | (unsigned long)total_swap_pages); | |
1723 | } else { /* v1 */ | |
1724 | if (mem_cgroup_swappiness(memcg)) { | |
1725 | /* Calculate swap excess capacity from memsw limit */ | |
1726 | unsigned long swap = READ_ONCE(memcg->memsw.max) - max; | |
1727 | ||
1728 | max += min(swap, (unsigned long)total_swap_pages); | |
1729 | } | |
9a5a8f19 | 1730 | } |
bbec2e15 | 1731 | return max; |
a63d83f4 DR |
1732 | } |
1733 | ||
9783aa99 CD |
1734 | unsigned long mem_cgroup_size(struct mem_cgroup *memcg) |
1735 | { | |
1736 | return page_counter_read(&memcg->memory); | |
1737 | } | |
1738 | ||
b6e6edcf | 1739 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1740 | int order) |
9cbb78bb | 1741 | { |
6e0fc46d DR |
1742 | struct oom_control oc = { |
1743 | .zonelist = NULL, | |
1744 | .nodemask = NULL, | |
2a966b77 | 1745 | .memcg = memcg, |
6e0fc46d DR |
1746 | .gfp_mask = gfp_mask, |
1747 | .order = order, | |
6e0fc46d | 1748 | }; |
1378b37d | 1749 | bool ret = true; |
9cbb78bb | 1750 | |
7775face TH |
1751 | if (mutex_lock_killable(&oom_lock)) |
1752 | return true; | |
1378b37d YS |
1753 | |
1754 | if (mem_cgroup_margin(memcg) >= (1 << order)) | |
1755 | goto unlock; | |
1756 | ||
7775face TH |
1757 | /* |
1758 | * A few threads which were not waiting at mutex_lock_killable() can | |
1759 | * fail to bail out. Therefore, check again after holding oom_lock. | |
1760 | */ | |
1761 | ret = should_force_charge() || out_of_memory(&oc); | |
1378b37d YS |
1762 | |
1763 | unlock: | |
dc56401f | 1764 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1765 | return ret; |
9cbb78bb DR |
1766 | } |
1767 | ||
0608f43d | 1768 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1769 | pg_data_t *pgdat, |
0608f43d AM |
1770 | gfp_t gfp_mask, |
1771 | unsigned long *total_scanned) | |
1772 | { | |
1773 | struct mem_cgroup *victim = NULL; | |
1774 | int total = 0; | |
1775 | int loop = 0; | |
1776 | unsigned long excess; | |
1777 | unsigned long nr_scanned; | |
1778 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1779 | .pgdat = pgdat, |
0608f43d AM |
1780 | }; |
1781 | ||
3e32cb2e | 1782 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1783 | |
1784 | while (1) { | |
1785 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1786 | if (!victim) { | |
1787 | loop++; | |
1788 | if (loop >= 2) { | |
1789 | /* | |
1790 | * If we have not been able to reclaim | |
1791 | * anything, it might because there are | |
1792 | * no reclaimable pages under this hierarchy | |
1793 | */ | |
1794 | if (!total) | |
1795 | break; | |
1796 | /* | |
1797 | * We want to do more targeted reclaim. | |
1798 | * excess >> 2 is not to excessive so as to | |
1799 | * reclaim too much, nor too less that we keep | |
1800 | * coming back to reclaim from this cgroup | |
1801 | */ | |
1802 | if (total >= (excess >> 2) || | |
1803 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1804 | break; | |
1805 | } | |
1806 | continue; | |
1807 | } | |
a9dd0a83 | 1808 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1809 | pgdat, &nr_scanned); |
0608f43d | 1810 | *total_scanned += nr_scanned; |
3e32cb2e | 1811 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1812 | break; |
6d61ef40 | 1813 | } |
0608f43d AM |
1814 | mem_cgroup_iter_break(root_memcg, victim); |
1815 | return total; | |
6d61ef40 BS |
1816 | } |
1817 | ||
0056f4e6 JW |
1818 | #ifdef CONFIG_LOCKDEP |
1819 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1820 | .name = "memcg_oom_lock", | |
1821 | }; | |
1822 | #endif | |
1823 | ||
fb2a6fc5 JW |
1824 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1825 | ||
867578cb KH |
1826 | /* |
1827 | * Check OOM-Killer is already running under our hierarchy. | |
1828 | * If someone is running, return false. | |
1829 | */ | |
fb2a6fc5 | 1830 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1831 | { |
79dfdacc | 1832 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1833 | |
fb2a6fc5 JW |
1834 | spin_lock(&memcg_oom_lock); |
1835 | ||
9f3a0d09 | 1836 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1837 | if (iter->oom_lock) { |
79dfdacc MH |
1838 | /* |
1839 | * this subtree of our hierarchy is already locked | |
1840 | * so we cannot give a lock. | |
1841 | */ | |
79dfdacc | 1842 | failed = iter; |
9f3a0d09 JW |
1843 | mem_cgroup_iter_break(memcg, iter); |
1844 | break; | |
23751be0 JW |
1845 | } else |
1846 | iter->oom_lock = true; | |
7d74b06f | 1847 | } |
867578cb | 1848 | |
fb2a6fc5 JW |
1849 | if (failed) { |
1850 | /* | |
1851 | * OK, we failed to lock the whole subtree so we have | |
1852 | * to clean up what we set up to the failing subtree | |
1853 | */ | |
1854 | for_each_mem_cgroup_tree(iter, memcg) { | |
1855 | if (iter == failed) { | |
1856 | mem_cgroup_iter_break(memcg, iter); | |
1857 | break; | |
1858 | } | |
1859 | iter->oom_lock = false; | |
79dfdacc | 1860 | } |
0056f4e6 JW |
1861 | } else |
1862 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1863 | |
1864 | spin_unlock(&memcg_oom_lock); | |
1865 | ||
1866 | return !failed; | |
a636b327 | 1867 | } |
0b7f569e | 1868 | |
fb2a6fc5 | 1869 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1870 | { |
7d74b06f KH |
1871 | struct mem_cgroup *iter; |
1872 | ||
fb2a6fc5 | 1873 | spin_lock(&memcg_oom_lock); |
5facae4f | 1874 | mutex_release(&memcg_oom_lock_dep_map, _RET_IP_); |
c0ff4b85 | 1875 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1876 | iter->oom_lock = false; |
fb2a6fc5 | 1877 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1878 | } |
1879 | ||
c0ff4b85 | 1880 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1881 | { |
1882 | struct mem_cgroup *iter; | |
1883 | ||
c2b42d3c | 1884 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1885 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1886 | iter->under_oom++; |
1887 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1888 | } |
1889 | ||
c0ff4b85 | 1890 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1891 | { |
1892 | struct mem_cgroup *iter; | |
1893 | ||
867578cb | 1894 | /* |
7a52d4d8 ML |
1895 | * Be careful about under_oom underflows becase a child memcg |
1896 | * could have been added after mem_cgroup_mark_under_oom. | |
867578cb | 1897 | */ |
c2b42d3c | 1898 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1899 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1900 | if (iter->under_oom > 0) |
1901 | iter->under_oom--; | |
1902 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1903 | } |
1904 | ||
867578cb KH |
1905 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1906 | ||
dc98df5a | 1907 | struct oom_wait_info { |
d79154bb | 1908 | struct mem_cgroup *memcg; |
ac6424b9 | 1909 | wait_queue_entry_t wait; |
dc98df5a KH |
1910 | }; |
1911 | ||
ac6424b9 | 1912 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1913 | unsigned mode, int sync, void *arg) |
1914 | { | |
d79154bb HD |
1915 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1916 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1917 | struct oom_wait_info *oom_wait_info; |
1918 | ||
1919 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1920 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1921 | |
2314b42d JW |
1922 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1923 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1924 | return 0; |
dc98df5a KH |
1925 | return autoremove_wake_function(wait, mode, sync, arg); |
1926 | } | |
1927 | ||
c0ff4b85 | 1928 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1929 | { |
c2b42d3c TH |
1930 | /* |
1931 | * For the following lockless ->under_oom test, the only required | |
1932 | * guarantee is that it must see the state asserted by an OOM when | |
1933 | * this function is called as a result of userland actions | |
1934 | * triggered by the notification of the OOM. This is trivially | |
1935 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1936 | * triggering notification. | |
1937 | */ | |
1938 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1939 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1940 | } |
1941 | ||
29ef680a MH |
1942 | enum oom_status { |
1943 | OOM_SUCCESS, | |
1944 | OOM_FAILED, | |
1945 | OOM_ASYNC, | |
1946 | OOM_SKIPPED | |
1947 | }; | |
1948 | ||
1949 | static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 1950 | { |
7056d3a3 MH |
1951 | enum oom_status ret; |
1952 | bool locked; | |
1953 | ||
29ef680a MH |
1954 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
1955 | return OOM_SKIPPED; | |
1956 | ||
7a1adfdd RG |
1957 | memcg_memory_event(memcg, MEMCG_OOM); |
1958 | ||
867578cb | 1959 | /* |
49426420 JW |
1960 | * We are in the middle of the charge context here, so we |
1961 | * don't want to block when potentially sitting on a callstack | |
1962 | * that holds all kinds of filesystem and mm locks. | |
1963 | * | |
29ef680a MH |
1964 | * cgroup1 allows disabling the OOM killer and waiting for outside |
1965 | * handling until the charge can succeed; remember the context and put | |
1966 | * the task to sleep at the end of the page fault when all locks are | |
1967 | * released. | |
49426420 | 1968 | * |
29ef680a MH |
1969 | * On the other hand, in-kernel OOM killer allows for an async victim |
1970 | * memory reclaim (oom_reaper) and that means that we are not solely | |
1971 | * relying on the oom victim to make a forward progress and we can | |
1972 | * invoke the oom killer here. | |
1973 | * | |
1974 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
1975 | * victim and then we have to bail out from the charge path. | |
867578cb | 1976 | */ |
29ef680a MH |
1977 | if (memcg->oom_kill_disable) { |
1978 | if (!current->in_user_fault) | |
1979 | return OOM_SKIPPED; | |
1980 | css_get(&memcg->css); | |
1981 | current->memcg_in_oom = memcg; | |
1982 | current->memcg_oom_gfp_mask = mask; | |
1983 | current->memcg_oom_order = order; | |
1984 | ||
1985 | return OOM_ASYNC; | |
1986 | } | |
1987 | ||
7056d3a3 MH |
1988 | mem_cgroup_mark_under_oom(memcg); |
1989 | ||
1990 | locked = mem_cgroup_oom_trylock(memcg); | |
1991 | ||
1992 | if (locked) | |
1993 | mem_cgroup_oom_notify(memcg); | |
1994 | ||
1995 | mem_cgroup_unmark_under_oom(memcg); | |
29ef680a | 1996 | if (mem_cgroup_out_of_memory(memcg, mask, order)) |
7056d3a3 MH |
1997 | ret = OOM_SUCCESS; |
1998 | else | |
1999 | ret = OOM_FAILED; | |
2000 | ||
2001 | if (locked) | |
2002 | mem_cgroup_oom_unlock(memcg); | |
29ef680a | 2003 | |
7056d3a3 | 2004 | return ret; |
3812c8c8 JW |
2005 | } |
2006 | ||
2007 | /** | |
2008 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 2009 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 2010 | * |
49426420 JW |
2011 | * This has to be called at the end of a page fault if the memcg OOM |
2012 | * handler was enabled. | |
3812c8c8 | 2013 | * |
49426420 | 2014 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
2015 | * sleep on a waitqueue until the userspace task resolves the |
2016 | * situation. Sleeping directly in the charge context with all kinds | |
2017 | * of locks held is not a good idea, instead we remember an OOM state | |
2018 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 2019 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
2020 | * |
2021 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 2022 | * completed, %false otherwise. |
3812c8c8 | 2023 | */ |
49426420 | 2024 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 2025 | { |
626ebc41 | 2026 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 2027 | struct oom_wait_info owait; |
49426420 | 2028 | bool locked; |
3812c8c8 JW |
2029 | |
2030 | /* OOM is global, do not handle */ | |
3812c8c8 | 2031 | if (!memcg) |
49426420 | 2032 | return false; |
3812c8c8 | 2033 | |
7c5f64f8 | 2034 | if (!handle) |
49426420 | 2035 | goto cleanup; |
3812c8c8 JW |
2036 | |
2037 | owait.memcg = memcg; | |
2038 | owait.wait.flags = 0; | |
2039 | owait.wait.func = memcg_oom_wake_function; | |
2040 | owait.wait.private = current; | |
2055da97 | 2041 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 2042 | |
3812c8c8 | 2043 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
2044 | mem_cgroup_mark_under_oom(memcg); |
2045 | ||
2046 | locked = mem_cgroup_oom_trylock(memcg); | |
2047 | ||
2048 | if (locked) | |
2049 | mem_cgroup_oom_notify(memcg); | |
2050 | ||
2051 | if (locked && !memcg->oom_kill_disable) { | |
2052 | mem_cgroup_unmark_under_oom(memcg); | |
2053 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
2054 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
2055 | current->memcg_oom_order); | |
49426420 | 2056 | } else { |
3812c8c8 | 2057 | schedule(); |
49426420 JW |
2058 | mem_cgroup_unmark_under_oom(memcg); |
2059 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2060 | } | |
2061 | ||
2062 | if (locked) { | |
fb2a6fc5 JW |
2063 | mem_cgroup_oom_unlock(memcg); |
2064 | /* | |
2065 | * There is no guarantee that an OOM-lock contender | |
2066 | * sees the wakeups triggered by the OOM kill | |
2067 | * uncharges. Wake any sleepers explicitely. | |
2068 | */ | |
2069 | memcg_oom_recover(memcg); | |
2070 | } | |
49426420 | 2071 | cleanup: |
626ebc41 | 2072 | current->memcg_in_oom = NULL; |
3812c8c8 | 2073 | css_put(&memcg->css); |
867578cb | 2074 | return true; |
0b7f569e KH |
2075 | } |
2076 | ||
3d8b38eb RG |
2077 | /** |
2078 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
2079 | * @victim: task to be killed by the OOM killer | |
2080 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
2081 | * | |
2082 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
2083 | * by killing all belonging OOM-killable tasks. | |
2084 | * | |
2085 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
2086 | */ | |
2087 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
2088 | struct mem_cgroup *oom_domain) | |
2089 | { | |
2090 | struct mem_cgroup *oom_group = NULL; | |
2091 | struct mem_cgroup *memcg; | |
2092 | ||
2093 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
2094 | return NULL; | |
2095 | ||
2096 | if (!oom_domain) | |
2097 | oom_domain = root_mem_cgroup; | |
2098 | ||
2099 | rcu_read_lock(); | |
2100 | ||
2101 | memcg = mem_cgroup_from_task(victim); | |
2102 | if (memcg == root_mem_cgroup) | |
2103 | goto out; | |
2104 | ||
48fe267c RG |
2105 | /* |
2106 | * If the victim task has been asynchronously moved to a different | |
2107 | * memory cgroup, we might end up killing tasks outside oom_domain. | |
2108 | * In this case it's better to ignore memory.group.oom. | |
2109 | */ | |
2110 | if (unlikely(!mem_cgroup_is_descendant(memcg, oom_domain))) | |
2111 | goto out; | |
2112 | ||
3d8b38eb RG |
2113 | /* |
2114 | * Traverse the memory cgroup hierarchy from the victim task's | |
2115 | * cgroup up to the OOMing cgroup (or root) to find the | |
2116 | * highest-level memory cgroup with oom.group set. | |
2117 | */ | |
2118 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
2119 | if (memcg->oom_group) | |
2120 | oom_group = memcg; | |
2121 | ||
2122 | if (memcg == oom_domain) | |
2123 | break; | |
2124 | } | |
2125 | ||
2126 | if (oom_group) | |
2127 | css_get(&oom_group->css); | |
2128 | out: | |
2129 | rcu_read_unlock(); | |
2130 | ||
2131 | return oom_group; | |
2132 | } | |
2133 | ||
2134 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
2135 | { | |
2136 | pr_info("Tasks in "); | |
2137 | pr_cont_cgroup_path(memcg->css.cgroup); | |
2138 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
2139 | } | |
2140 | ||
d7365e78 | 2141 | /** |
bcfe06bf | 2142 | * lock_page_memcg - lock a page and memcg binding |
81f8c3a4 | 2143 | * @page: the page |
32047e2a | 2144 | * |
81f8c3a4 | 2145 | * This function protects unlocked LRU pages from being moved to |
739f79fc JW |
2146 | * another cgroup. |
2147 | * | |
2148 | * It ensures lifetime of the returned memcg. Caller is responsible | |
2149 | * for the lifetime of the page; __unlock_page_memcg() is available | |
2150 | * when @page might get freed inside the locked section. | |
d69b042f | 2151 | */ |
739f79fc | 2152 | struct mem_cgroup *lock_page_memcg(struct page *page) |
89c06bd5 | 2153 | { |
9da7b521 | 2154 | struct page *head = compound_head(page); /* rmap on tail pages */ |
89c06bd5 | 2155 | struct mem_cgroup *memcg; |
6de22619 | 2156 | unsigned long flags; |
89c06bd5 | 2157 | |
6de22619 JW |
2158 | /* |
2159 | * The RCU lock is held throughout the transaction. The fast | |
2160 | * path can get away without acquiring the memcg->move_lock | |
2161 | * because page moving starts with an RCU grace period. | |
739f79fc JW |
2162 | * |
2163 | * The RCU lock also protects the memcg from being freed when | |
2164 | * the page state that is going to change is the only thing | |
2165 | * preventing the page itself from being freed. E.g. writeback | |
2166 | * doesn't hold a page reference and relies on PG_writeback to | |
2167 | * keep off truncation, migration and so forth. | |
2168 | */ | |
d7365e78 JW |
2169 | rcu_read_lock(); |
2170 | ||
2171 | if (mem_cgroup_disabled()) | |
739f79fc | 2172 | return NULL; |
89c06bd5 | 2173 | again: |
bcfe06bf | 2174 | memcg = page_memcg(head); |
29833315 | 2175 | if (unlikely(!memcg)) |
739f79fc | 2176 | return NULL; |
d7365e78 | 2177 | |
20ad50d6 AS |
2178 | #ifdef CONFIG_PROVE_LOCKING |
2179 | local_irq_save(flags); | |
2180 | might_lock(&memcg->move_lock); | |
2181 | local_irq_restore(flags); | |
2182 | #endif | |
2183 | ||
bdcbb659 | 2184 | if (atomic_read(&memcg->moving_account) <= 0) |
739f79fc | 2185 | return memcg; |
89c06bd5 | 2186 | |
6de22619 | 2187 | spin_lock_irqsave(&memcg->move_lock, flags); |
bcfe06bf | 2188 | if (memcg != page_memcg(head)) { |
6de22619 | 2189 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
2190 | goto again; |
2191 | } | |
6de22619 JW |
2192 | |
2193 | /* | |
2194 | * When charge migration first begins, we can have locked and | |
2195 | * unlocked page stat updates happening concurrently. Track | |
81f8c3a4 | 2196 | * the task who has the lock for unlock_page_memcg(). |
6de22619 JW |
2197 | */ |
2198 | memcg->move_lock_task = current; | |
2199 | memcg->move_lock_flags = flags; | |
d7365e78 | 2200 | |
739f79fc | 2201 | return memcg; |
89c06bd5 | 2202 | } |
81f8c3a4 | 2203 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 2204 | |
d7365e78 | 2205 | /** |
739f79fc JW |
2206 | * __unlock_page_memcg - unlock and unpin a memcg |
2207 | * @memcg: the memcg | |
2208 | * | |
2209 | * Unlock and unpin a memcg returned by lock_page_memcg(). | |
d7365e78 | 2210 | */ |
739f79fc | 2211 | void __unlock_page_memcg(struct mem_cgroup *memcg) |
89c06bd5 | 2212 | { |
6de22619 JW |
2213 | if (memcg && memcg->move_lock_task == current) { |
2214 | unsigned long flags = memcg->move_lock_flags; | |
2215 | ||
2216 | memcg->move_lock_task = NULL; | |
2217 | memcg->move_lock_flags = 0; | |
2218 | ||
2219 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2220 | } | |
89c06bd5 | 2221 | |
d7365e78 | 2222 | rcu_read_unlock(); |
89c06bd5 | 2223 | } |
739f79fc JW |
2224 | |
2225 | /** | |
bcfe06bf | 2226 | * unlock_page_memcg - unlock a page and memcg binding |
739f79fc JW |
2227 | * @page: the page |
2228 | */ | |
2229 | void unlock_page_memcg(struct page *page) | |
2230 | { | |
9da7b521 JW |
2231 | struct page *head = compound_head(page); |
2232 | ||
bcfe06bf | 2233 | __unlock_page_memcg(page_memcg(head)); |
739f79fc | 2234 | } |
81f8c3a4 | 2235 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 2236 | |
cdec2e42 KH |
2237 | struct memcg_stock_pcp { |
2238 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2239 | unsigned int nr_pages; |
bf4f0599 RG |
2240 | |
2241 | #ifdef CONFIG_MEMCG_KMEM | |
2242 | struct obj_cgroup *cached_objcg; | |
2243 | unsigned int nr_bytes; | |
2244 | #endif | |
2245 | ||
cdec2e42 | 2246 | struct work_struct work; |
26fe6168 | 2247 | unsigned long flags; |
a0db00fc | 2248 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2249 | }; |
2250 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2251 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2252 | |
bf4f0599 RG |
2253 | #ifdef CONFIG_MEMCG_KMEM |
2254 | static void drain_obj_stock(struct memcg_stock_pcp *stock); | |
2255 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2256 | struct mem_cgroup *root_memcg); | |
2257 | ||
2258 | #else | |
2259 | static inline void drain_obj_stock(struct memcg_stock_pcp *stock) | |
2260 | { | |
2261 | } | |
2262 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2263 | struct mem_cgroup *root_memcg) | |
2264 | { | |
2265 | return false; | |
2266 | } | |
2267 | #endif | |
2268 | ||
a0956d54 SS |
2269 | /** |
2270 | * consume_stock: Try to consume stocked charge on this cpu. | |
2271 | * @memcg: memcg to consume from. | |
2272 | * @nr_pages: how many pages to charge. | |
2273 | * | |
2274 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2275 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2276 | * service an allocation will refill the stock. | |
2277 | * | |
2278 | * returns true if successful, false otherwise. | |
cdec2e42 | 2279 | */ |
a0956d54 | 2280 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2281 | { |
2282 | struct memcg_stock_pcp *stock; | |
db2ba40c | 2283 | unsigned long flags; |
3e32cb2e | 2284 | bool ret = false; |
cdec2e42 | 2285 | |
a983b5eb | 2286 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 2287 | return ret; |
a0956d54 | 2288 | |
db2ba40c JW |
2289 | local_irq_save(flags); |
2290 | ||
2291 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 2292 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2293 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2294 | ret = true; |
2295 | } | |
db2ba40c JW |
2296 | |
2297 | local_irq_restore(flags); | |
2298 | ||
cdec2e42 KH |
2299 | return ret; |
2300 | } | |
2301 | ||
2302 | /* | |
3e32cb2e | 2303 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2304 | */ |
2305 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2306 | { | |
2307 | struct mem_cgroup *old = stock->cached; | |
2308 | ||
1a3e1f40 JW |
2309 | if (!old) |
2310 | return; | |
2311 | ||
11c9ea4e | 2312 | if (stock->nr_pages) { |
3e32cb2e | 2313 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 2314 | if (do_memsw_account()) |
3e32cb2e | 2315 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
11c9ea4e | 2316 | stock->nr_pages = 0; |
cdec2e42 | 2317 | } |
1a3e1f40 JW |
2318 | |
2319 | css_put(&old->css); | |
cdec2e42 | 2320 | stock->cached = NULL; |
cdec2e42 KH |
2321 | } |
2322 | ||
cdec2e42 KH |
2323 | static void drain_local_stock(struct work_struct *dummy) |
2324 | { | |
db2ba40c JW |
2325 | struct memcg_stock_pcp *stock; |
2326 | unsigned long flags; | |
2327 | ||
72f0184c MH |
2328 | /* |
2329 | * The only protection from memory hotplug vs. drain_stock races is | |
2330 | * that we always operate on local CPU stock here with IRQ disabled | |
2331 | */ | |
db2ba40c JW |
2332 | local_irq_save(flags); |
2333 | ||
2334 | stock = this_cpu_ptr(&memcg_stock); | |
bf4f0599 | 2335 | drain_obj_stock(stock); |
cdec2e42 | 2336 | drain_stock(stock); |
26fe6168 | 2337 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c JW |
2338 | |
2339 | local_irq_restore(flags); | |
cdec2e42 KH |
2340 | } |
2341 | ||
2342 | /* | |
3e32cb2e | 2343 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2344 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2345 | */ |
c0ff4b85 | 2346 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2347 | { |
db2ba40c JW |
2348 | struct memcg_stock_pcp *stock; |
2349 | unsigned long flags; | |
2350 | ||
2351 | local_irq_save(flags); | |
cdec2e42 | 2352 | |
db2ba40c | 2353 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 2354 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2355 | drain_stock(stock); |
1a3e1f40 | 2356 | css_get(&memcg->css); |
c0ff4b85 | 2357 | stock->cached = memcg; |
cdec2e42 | 2358 | } |
11c9ea4e | 2359 | stock->nr_pages += nr_pages; |
db2ba40c | 2360 | |
a983b5eb | 2361 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 RG |
2362 | drain_stock(stock); |
2363 | ||
db2ba40c | 2364 | local_irq_restore(flags); |
cdec2e42 KH |
2365 | } |
2366 | ||
2367 | /* | |
c0ff4b85 | 2368 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2369 | * of the hierarchy under it. |
cdec2e42 | 2370 | */ |
6d3d6aa2 | 2371 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2372 | { |
26fe6168 | 2373 | int cpu, curcpu; |
d38144b7 | 2374 | |
6d3d6aa2 JW |
2375 | /* If someone's already draining, avoid adding running more workers. */ |
2376 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2377 | return; | |
72f0184c MH |
2378 | /* |
2379 | * Notify other cpus that system-wide "drain" is running | |
2380 | * We do not care about races with the cpu hotplug because cpu down | |
2381 | * as well as workers from this path always operate on the local | |
2382 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2383 | */ | |
5af12d0e | 2384 | curcpu = get_cpu(); |
cdec2e42 KH |
2385 | for_each_online_cpu(cpu) { |
2386 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2387 | struct mem_cgroup *memcg; |
e1a366be | 2388 | bool flush = false; |
26fe6168 | 2389 | |
e1a366be | 2390 | rcu_read_lock(); |
c0ff4b85 | 2391 | memcg = stock->cached; |
e1a366be RG |
2392 | if (memcg && stock->nr_pages && |
2393 | mem_cgroup_is_descendant(memcg, root_memcg)) | |
2394 | flush = true; | |
bf4f0599 RG |
2395 | if (obj_stock_flush_required(stock, root_memcg)) |
2396 | flush = true; | |
e1a366be RG |
2397 | rcu_read_unlock(); |
2398 | ||
2399 | if (flush && | |
2400 | !test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { | |
d1a05b69 MH |
2401 | if (cpu == curcpu) |
2402 | drain_local_stock(&stock->work); | |
2403 | else | |
2404 | schedule_work_on(cpu, &stock->work); | |
2405 | } | |
cdec2e42 | 2406 | } |
5af12d0e | 2407 | put_cpu(); |
9f50fad6 | 2408 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2409 | } |
2410 | ||
308167fc | 2411 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
cdec2e42 | 2412 | { |
cdec2e42 | 2413 | struct memcg_stock_pcp *stock; |
42a30035 | 2414 | struct mem_cgroup *memcg, *mi; |
cdec2e42 | 2415 | |
cdec2e42 KH |
2416 | stock = &per_cpu(memcg_stock, cpu); |
2417 | drain_stock(stock); | |
a983b5eb JW |
2418 | |
2419 | for_each_mem_cgroup(memcg) { | |
2420 | int i; | |
2421 | ||
2422 | for (i = 0; i < MEMCG_NR_STAT; i++) { | |
2423 | int nid; | |
2424 | long x; | |
2425 | ||
871789d4 | 2426 | x = this_cpu_xchg(memcg->vmstats_percpu->stat[i], 0); |
815744d7 | 2427 | if (x) |
42a30035 JW |
2428 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2429 | atomic_long_add(x, &memcg->vmstats[i]); | |
a983b5eb JW |
2430 | |
2431 | if (i >= NR_VM_NODE_STAT_ITEMS) | |
2432 | continue; | |
2433 | ||
2434 | for_each_node(nid) { | |
2435 | struct mem_cgroup_per_node *pn; | |
2436 | ||
2437 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
2438 | x = this_cpu_xchg(pn->lruvec_stat_cpu->count[i], 0); | |
815744d7 | 2439 | if (x) |
42a30035 JW |
2440 | do { |
2441 | atomic_long_add(x, &pn->lruvec_stat[i]); | |
2442 | } while ((pn = parent_nodeinfo(pn, nid))); | |
a983b5eb JW |
2443 | } |
2444 | } | |
2445 | ||
e27be240 | 2446 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { |
a983b5eb JW |
2447 | long x; |
2448 | ||
871789d4 | 2449 | x = this_cpu_xchg(memcg->vmstats_percpu->events[i], 0); |
815744d7 | 2450 | if (x) |
42a30035 JW |
2451 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2452 | atomic_long_add(x, &memcg->vmevents[i]); | |
a983b5eb JW |
2453 | } |
2454 | } | |
2455 | ||
308167fc | 2456 | return 0; |
cdec2e42 KH |
2457 | } |
2458 | ||
b3ff9291 CD |
2459 | static unsigned long reclaim_high(struct mem_cgroup *memcg, |
2460 | unsigned int nr_pages, | |
2461 | gfp_t gfp_mask) | |
f7e1cb6e | 2462 | { |
b3ff9291 CD |
2463 | unsigned long nr_reclaimed = 0; |
2464 | ||
f7e1cb6e | 2465 | do { |
e22c6ed9 JW |
2466 | unsigned long pflags; |
2467 | ||
d1663a90 JK |
2468 | if (page_counter_read(&memcg->memory) <= |
2469 | READ_ONCE(memcg->memory.high)) | |
f7e1cb6e | 2470 | continue; |
e22c6ed9 | 2471 | |
e27be240 | 2472 | memcg_memory_event(memcg, MEMCG_HIGH); |
e22c6ed9 JW |
2473 | |
2474 | psi_memstall_enter(&pflags); | |
b3ff9291 CD |
2475 | nr_reclaimed += try_to_free_mem_cgroup_pages(memcg, nr_pages, |
2476 | gfp_mask, true); | |
e22c6ed9 | 2477 | psi_memstall_leave(&pflags); |
4bf17307 CD |
2478 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2479 | !mem_cgroup_is_root(memcg)); | |
b3ff9291 CD |
2480 | |
2481 | return nr_reclaimed; | |
f7e1cb6e JW |
2482 | } |
2483 | ||
2484 | static void high_work_func(struct work_struct *work) | |
2485 | { | |
2486 | struct mem_cgroup *memcg; | |
2487 | ||
2488 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2489 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2490 | } |
2491 | ||
0e4b01df CD |
2492 | /* |
2493 | * Clamp the maximum sleep time per allocation batch to 2 seconds. This is | |
2494 | * enough to still cause a significant slowdown in most cases, while still | |
2495 | * allowing diagnostics and tracing to proceed without becoming stuck. | |
2496 | */ | |
2497 | #define MEMCG_MAX_HIGH_DELAY_JIFFIES (2UL*HZ) | |
2498 | ||
2499 | /* | |
2500 | * When calculating the delay, we use these either side of the exponentiation to | |
2501 | * maintain precision and scale to a reasonable number of jiffies (see the table | |
2502 | * below. | |
2503 | * | |
2504 | * - MEMCG_DELAY_PRECISION_SHIFT: Extra precision bits while translating the | |
2505 | * overage ratio to a delay. | |
ac5ddd0f | 2506 | * - MEMCG_DELAY_SCALING_SHIFT: The number of bits to scale down the |
0e4b01df CD |
2507 | * proposed penalty in order to reduce to a reasonable number of jiffies, and |
2508 | * to produce a reasonable delay curve. | |
2509 | * | |
2510 | * MEMCG_DELAY_SCALING_SHIFT just happens to be a number that produces a | |
2511 | * reasonable delay curve compared to precision-adjusted overage, not | |
2512 | * penalising heavily at first, but still making sure that growth beyond the | |
2513 | * limit penalises misbehaviour cgroups by slowing them down exponentially. For | |
2514 | * example, with a high of 100 megabytes: | |
2515 | * | |
2516 | * +-------+------------------------+ | |
2517 | * | usage | time to allocate in ms | | |
2518 | * +-------+------------------------+ | |
2519 | * | 100M | 0 | | |
2520 | * | 101M | 6 | | |
2521 | * | 102M | 25 | | |
2522 | * | 103M | 57 | | |
2523 | * | 104M | 102 | | |
2524 | * | 105M | 159 | | |
2525 | * | 106M | 230 | | |
2526 | * | 107M | 313 | | |
2527 | * | 108M | 409 | | |
2528 | * | 109M | 518 | | |
2529 | * | 110M | 639 | | |
2530 | * | 111M | 774 | | |
2531 | * | 112M | 921 | | |
2532 | * | 113M | 1081 | | |
2533 | * | 114M | 1254 | | |
2534 | * | 115M | 1439 | | |
2535 | * | 116M | 1638 | | |
2536 | * | 117M | 1849 | | |
2537 | * | 118M | 2000 | | |
2538 | * | 119M | 2000 | | |
2539 | * | 120M | 2000 | | |
2540 | * +-------+------------------------+ | |
2541 | */ | |
2542 | #define MEMCG_DELAY_PRECISION_SHIFT 20 | |
2543 | #define MEMCG_DELAY_SCALING_SHIFT 14 | |
2544 | ||
8a5dbc65 | 2545 | static u64 calculate_overage(unsigned long usage, unsigned long high) |
b23afb93 | 2546 | { |
8a5dbc65 | 2547 | u64 overage; |
b23afb93 | 2548 | |
8a5dbc65 JK |
2549 | if (usage <= high) |
2550 | return 0; | |
e26733e0 | 2551 | |
8a5dbc65 JK |
2552 | /* |
2553 | * Prevent division by 0 in overage calculation by acting as if | |
2554 | * it was a threshold of 1 page | |
2555 | */ | |
2556 | high = max(high, 1UL); | |
9b8b1754 | 2557 | |
8a5dbc65 JK |
2558 | overage = usage - high; |
2559 | overage <<= MEMCG_DELAY_PRECISION_SHIFT; | |
2560 | return div64_u64(overage, high); | |
2561 | } | |
e26733e0 | 2562 | |
8a5dbc65 JK |
2563 | static u64 mem_find_max_overage(struct mem_cgroup *memcg) |
2564 | { | |
2565 | u64 overage, max_overage = 0; | |
e26733e0 | 2566 | |
8a5dbc65 JK |
2567 | do { |
2568 | overage = calculate_overage(page_counter_read(&memcg->memory), | |
d1663a90 | 2569 | READ_ONCE(memcg->memory.high)); |
8a5dbc65 | 2570 | max_overage = max(overage, max_overage); |
e26733e0 CD |
2571 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2572 | !mem_cgroup_is_root(memcg)); | |
2573 | ||
8a5dbc65 JK |
2574 | return max_overage; |
2575 | } | |
2576 | ||
4b82ab4f JK |
2577 | static u64 swap_find_max_overage(struct mem_cgroup *memcg) |
2578 | { | |
2579 | u64 overage, max_overage = 0; | |
2580 | ||
2581 | do { | |
2582 | overage = calculate_overage(page_counter_read(&memcg->swap), | |
2583 | READ_ONCE(memcg->swap.high)); | |
2584 | if (overage) | |
2585 | memcg_memory_event(memcg, MEMCG_SWAP_HIGH); | |
2586 | max_overage = max(overage, max_overage); | |
2587 | } while ((memcg = parent_mem_cgroup(memcg)) && | |
2588 | !mem_cgroup_is_root(memcg)); | |
2589 | ||
2590 | return max_overage; | |
2591 | } | |
2592 | ||
8a5dbc65 JK |
2593 | /* |
2594 | * Get the number of jiffies that we should penalise a mischievous cgroup which | |
2595 | * is exceeding its memory.high by checking both it and its ancestors. | |
2596 | */ | |
2597 | static unsigned long calculate_high_delay(struct mem_cgroup *memcg, | |
2598 | unsigned int nr_pages, | |
2599 | u64 max_overage) | |
2600 | { | |
2601 | unsigned long penalty_jiffies; | |
2602 | ||
e26733e0 CD |
2603 | if (!max_overage) |
2604 | return 0; | |
0e4b01df CD |
2605 | |
2606 | /* | |
0e4b01df CD |
2607 | * We use overage compared to memory.high to calculate the number of |
2608 | * jiffies to sleep (penalty_jiffies). Ideally this value should be | |
2609 | * fairly lenient on small overages, and increasingly harsh when the | |
2610 | * memcg in question makes it clear that it has no intention of stopping | |
2611 | * its crazy behaviour, so we exponentially increase the delay based on | |
2612 | * overage amount. | |
2613 | */ | |
e26733e0 CD |
2614 | penalty_jiffies = max_overage * max_overage * HZ; |
2615 | penalty_jiffies >>= MEMCG_DELAY_PRECISION_SHIFT; | |
2616 | penalty_jiffies >>= MEMCG_DELAY_SCALING_SHIFT; | |
0e4b01df CD |
2617 | |
2618 | /* | |
2619 | * Factor in the task's own contribution to the overage, such that four | |
2620 | * N-sized allocations are throttled approximately the same as one | |
2621 | * 4N-sized allocation. | |
2622 | * | |
2623 | * MEMCG_CHARGE_BATCH pages is nominal, so work out how much smaller or | |
2624 | * larger the current charge patch is than that. | |
2625 | */ | |
ff144e69 | 2626 | return penalty_jiffies * nr_pages / MEMCG_CHARGE_BATCH; |
e26733e0 CD |
2627 | } |
2628 | ||
2629 | /* | |
2630 | * Scheduled by try_charge() to be executed from the userland return path | |
2631 | * and reclaims memory over the high limit. | |
2632 | */ | |
2633 | void mem_cgroup_handle_over_high(void) | |
2634 | { | |
2635 | unsigned long penalty_jiffies; | |
2636 | unsigned long pflags; | |
b3ff9291 | 2637 | unsigned long nr_reclaimed; |
e26733e0 | 2638 | unsigned int nr_pages = current->memcg_nr_pages_over_high; |
d977aa93 | 2639 | int nr_retries = MAX_RECLAIM_RETRIES; |
e26733e0 | 2640 | struct mem_cgroup *memcg; |
b3ff9291 | 2641 | bool in_retry = false; |
e26733e0 CD |
2642 | |
2643 | if (likely(!nr_pages)) | |
2644 | return; | |
2645 | ||
2646 | memcg = get_mem_cgroup_from_mm(current->mm); | |
e26733e0 CD |
2647 | current->memcg_nr_pages_over_high = 0; |
2648 | ||
b3ff9291 CD |
2649 | retry_reclaim: |
2650 | /* | |
2651 | * The allocating task should reclaim at least the batch size, but for | |
2652 | * subsequent retries we only want to do what's necessary to prevent oom | |
2653 | * or breaching resource isolation. | |
2654 | * | |
2655 | * This is distinct from memory.max or page allocator behaviour because | |
2656 | * memory.high is currently batched, whereas memory.max and the page | |
2657 | * allocator run every time an allocation is made. | |
2658 | */ | |
2659 | nr_reclaimed = reclaim_high(memcg, | |
2660 | in_retry ? SWAP_CLUSTER_MAX : nr_pages, | |
2661 | GFP_KERNEL); | |
2662 | ||
e26733e0 CD |
2663 | /* |
2664 | * memory.high is breached and reclaim is unable to keep up. Throttle | |
2665 | * allocators proactively to slow down excessive growth. | |
2666 | */ | |
8a5dbc65 JK |
2667 | penalty_jiffies = calculate_high_delay(memcg, nr_pages, |
2668 | mem_find_max_overage(memcg)); | |
0e4b01df | 2669 | |
4b82ab4f JK |
2670 | penalty_jiffies += calculate_high_delay(memcg, nr_pages, |
2671 | swap_find_max_overage(memcg)); | |
2672 | ||
ff144e69 JK |
2673 | /* |
2674 | * Clamp the max delay per usermode return so as to still keep the | |
2675 | * application moving forwards and also permit diagnostics, albeit | |
2676 | * extremely slowly. | |
2677 | */ | |
2678 | penalty_jiffies = min(penalty_jiffies, MEMCG_MAX_HIGH_DELAY_JIFFIES); | |
2679 | ||
0e4b01df CD |
2680 | /* |
2681 | * Don't sleep if the amount of jiffies this memcg owes us is so low | |
2682 | * that it's not even worth doing, in an attempt to be nice to those who | |
2683 | * go only a small amount over their memory.high value and maybe haven't | |
2684 | * been aggressively reclaimed enough yet. | |
2685 | */ | |
2686 | if (penalty_jiffies <= HZ / 100) | |
2687 | goto out; | |
2688 | ||
b3ff9291 CD |
2689 | /* |
2690 | * If reclaim is making forward progress but we're still over | |
2691 | * memory.high, we want to encourage that rather than doing allocator | |
2692 | * throttling. | |
2693 | */ | |
2694 | if (nr_reclaimed || nr_retries--) { | |
2695 | in_retry = true; | |
2696 | goto retry_reclaim; | |
2697 | } | |
2698 | ||
0e4b01df CD |
2699 | /* |
2700 | * If we exit early, we're guaranteed to die (since | |
2701 | * schedule_timeout_killable sets TASK_KILLABLE). This means we don't | |
2702 | * need to account for any ill-begotten jiffies to pay them off later. | |
2703 | */ | |
2704 | psi_memstall_enter(&pflags); | |
2705 | schedule_timeout_killable(penalty_jiffies); | |
2706 | psi_memstall_leave(&pflags); | |
2707 | ||
2708 | out: | |
2709 | css_put(&memcg->css); | |
b23afb93 TH |
2710 | } |
2711 | ||
00501b53 JW |
2712 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2713 | unsigned int nr_pages) | |
8a9f3ccd | 2714 | { |
a983b5eb | 2715 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
d977aa93 | 2716 | int nr_retries = MAX_RECLAIM_RETRIES; |
6539cc05 | 2717 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2718 | struct page_counter *counter; |
e22c6ed9 | 2719 | enum oom_status oom_status; |
6539cc05 | 2720 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2721 | bool may_swap = true; |
2722 | bool drained = false; | |
e22c6ed9 | 2723 | unsigned long pflags; |
a636b327 | 2724 | |
ce00a967 | 2725 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 2726 | return 0; |
6539cc05 | 2727 | retry: |
b6b6cc72 | 2728 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2729 | return 0; |
8a9f3ccd | 2730 | |
7941d214 | 2731 | if (!do_memsw_account() || |
6071ca52 JW |
2732 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2733 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2734 | goto done_restock; |
7941d214 | 2735 | if (do_memsw_account()) |
3e32cb2e JW |
2736 | page_counter_uncharge(&memcg->memsw, batch); |
2737 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2738 | } else { |
3e32cb2e | 2739 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2740 | may_swap = false; |
3fbe7244 | 2741 | } |
7a81b88c | 2742 | |
6539cc05 JW |
2743 | if (batch > nr_pages) { |
2744 | batch = nr_pages; | |
2745 | goto retry; | |
2746 | } | |
6d61ef40 | 2747 | |
869712fd JW |
2748 | /* |
2749 | * Memcg doesn't have a dedicated reserve for atomic | |
2750 | * allocations. But like the global atomic pool, we need to | |
2751 | * put the burden of reclaim on regular allocation requests | |
2752 | * and let these go through as privileged allocations. | |
2753 | */ | |
2754 | if (gfp_mask & __GFP_ATOMIC) | |
2755 | goto force; | |
2756 | ||
06b078fc JW |
2757 | /* |
2758 | * Unlike in global OOM situations, memcg is not in a physical | |
2759 | * memory shortage. Allow dying and OOM-killed tasks to | |
2760 | * bypass the last charges so that they can exit quickly and | |
2761 | * free their memory. | |
2762 | */ | |
7775face | 2763 | if (unlikely(should_force_charge())) |
10d53c74 | 2764 | goto force; |
06b078fc | 2765 | |
89a28483 JW |
2766 | /* |
2767 | * Prevent unbounded recursion when reclaim operations need to | |
2768 | * allocate memory. This might exceed the limits temporarily, | |
2769 | * but we prefer facilitating memory reclaim and getting back | |
2770 | * under the limit over triggering OOM kills in these cases. | |
2771 | */ | |
2772 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2773 | goto force; | |
2774 | ||
06b078fc JW |
2775 | if (unlikely(task_in_memcg_oom(current))) |
2776 | goto nomem; | |
2777 | ||
d0164adc | 2778 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2779 | goto nomem; |
4b534334 | 2780 | |
e27be240 | 2781 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
241994ed | 2782 | |
e22c6ed9 | 2783 | psi_memstall_enter(&pflags); |
b70a2a21 JW |
2784 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2785 | gfp_mask, may_swap); | |
e22c6ed9 | 2786 | psi_memstall_leave(&pflags); |
6539cc05 | 2787 | |
61e02c74 | 2788 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2789 | goto retry; |
28c34c29 | 2790 | |
b70a2a21 | 2791 | if (!drained) { |
6d3d6aa2 | 2792 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2793 | drained = true; |
2794 | goto retry; | |
2795 | } | |
2796 | ||
28c34c29 JW |
2797 | if (gfp_mask & __GFP_NORETRY) |
2798 | goto nomem; | |
6539cc05 JW |
2799 | /* |
2800 | * Even though the limit is exceeded at this point, reclaim | |
2801 | * may have been able to free some pages. Retry the charge | |
2802 | * before killing the task. | |
2803 | * | |
2804 | * Only for regular pages, though: huge pages are rather | |
2805 | * unlikely to succeed so close to the limit, and we fall back | |
2806 | * to regular pages anyway in case of failure. | |
2807 | */ | |
61e02c74 | 2808 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2809 | goto retry; |
2810 | /* | |
2811 | * At task move, charge accounts can be doubly counted. So, it's | |
2812 | * better to wait until the end of task_move if something is going on. | |
2813 | */ | |
2814 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2815 | goto retry; | |
2816 | ||
9b130619 JW |
2817 | if (nr_retries--) |
2818 | goto retry; | |
2819 | ||
38d38493 | 2820 | if (gfp_mask & __GFP_RETRY_MAYFAIL) |
29ef680a MH |
2821 | goto nomem; |
2822 | ||
06b078fc | 2823 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2824 | goto force; |
06b078fc | 2825 | |
6539cc05 | 2826 | if (fatal_signal_pending(current)) |
10d53c74 | 2827 | goto force; |
6539cc05 | 2828 | |
29ef680a MH |
2829 | /* |
2830 | * keep retrying as long as the memcg oom killer is able to make | |
2831 | * a forward progress or bypass the charge if the oom killer | |
2832 | * couldn't make any progress. | |
2833 | */ | |
2834 | oom_status = mem_cgroup_oom(mem_over_limit, gfp_mask, | |
3608de07 | 2835 | get_order(nr_pages * PAGE_SIZE)); |
29ef680a MH |
2836 | switch (oom_status) { |
2837 | case OOM_SUCCESS: | |
d977aa93 | 2838 | nr_retries = MAX_RECLAIM_RETRIES; |
29ef680a MH |
2839 | goto retry; |
2840 | case OOM_FAILED: | |
2841 | goto force; | |
2842 | default: | |
2843 | goto nomem; | |
2844 | } | |
7a81b88c | 2845 | nomem: |
6d1fdc48 | 2846 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2847 | return -ENOMEM; |
10d53c74 TH |
2848 | force: |
2849 | /* | |
2850 | * The allocation either can't fail or will lead to more memory | |
2851 | * being freed very soon. Allow memory usage go over the limit | |
2852 | * temporarily by force charging it. | |
2853 | */ | |
2854 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2855 | if (do_memsw_account()) |
10d53c74 | 2856 | page_counter_charge(&memcg->memsw, nr_pages); |
10d53c74 TH |
2857 | |
2858 | return 0; | |
6539cc05 JW |
2859 | |
2860 | done_restock: | |
2861 | if (batch > nr_pages) | |
2862 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2863 | |
241994ed | 2864 | /* |
b23afb93 TH |
2865 | * If the hierarchy is above the normal consumption range, schedule |
2866 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2867 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2868 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2869 | * not recorded as it most likely matches current's and won't | |
2870 | * change in the meantime. As high limit is checked again before | |
2871 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2872 | */ |
2873 | do { | |
4b82ab4f JK |
2874 | bool mem_high, swap_high; |
2875 | ||
2876 | mem_high = page_counter_read(&memcg->memory) > | |
2877 | READ_ONCE(memcg->memory.high); | |
2878 | swap_high = page_counter_read(&memcg->swap) > | |
2879 | READ_ONCE(memcg->swap.high); | |
2880 | ||
2881 | /* Don't bother a random interrupted task */ | |
2882 | if (in_interrupt()) { | |
2883 | if (mem_high) { | |
f7e1cb6e JW |
2884 | schedule_work(&memcg->high_work); |
2885 | break; | |
2886 | } | |
4b82ab4f JK |
2887 | continue; |
2888 | } | |
2889 | ||
2890 | if (mem_high || swap_high) { | |
2891 | /* | |
2892 | * The allocating tasks in this cgroup will need to do | |
2893 | * reclaim or be throttled to prevent further growth | |
2894 | * of the memory or swap footprints. | |
2895 | * | |
2896 | * Target some best-effort fairness between the tasks, | |
2897 | * and distribute reclaim work and delay penalties | |
2898 | * based on how much each task is actually allocating. | |
2899 | */ | |
9516a18a | 2900 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2901 | set_notify_resume(current); |
2902 | break; | |
2903 | } | |
241994ed | 2904 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2905 | |
2906 | return 0; | |
7a81b88c | 2907 | } |
8a9f3ccd | 2908 | |
f0e45fb4 | 2909 | #if defined(CONFIG_MEMCG_KMEM) || defined(CONFIG_MMU) |
00501b53 | 2910 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2911 | { |
ce00a967 JW |
2912 | if (mem_cgroup_is_root(memcg)) |
2913 | return; | |
2914 | ||
3e32cb2e | 2915 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2916 | if (do_memsw_account()) |
3e32cb2e | 2917 | page_counter_uncharge(&memcg->memsw, nr_pages); |
d01dd17f | 2918 | } |
f0e45fb4 | 2919 | #endif |
d01dd17f | 2920 | |
d9eb1ea2 | 2921 | static void commit_charge(struct page *page, struct mem_cgroup *memcg) |
0a31bc97 | 2922 | { |
bcfe06bf | 2923 | VM_BUG_ON_PAGE(page_memcg(page), page); |
0a31bc97 | 2924 | /* |
a5eb011a | 2925 | * Any of the following ensures page's memcg stability: |
0a31bc97 | 2926 | * |
a0b5b414 JW |
2927 | * - the page lock |
2928 | * - LRU isolation | |
2929 | * - lock_page_memcg() | |
2930 | * - exclusive reference | |
0a31bc97 | 2931 | */ |
bcfe06bf | 2932 | page->memcg_data = (unsigned long)memcg; |
7a81b88c | 2933 | } |
66e1707b | 2934 | |
84c07d11 | 2935 | #ifdef CONFIG_MEMCG_KMEM |
10befea9 | 2936 | int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s, |
2e9bd483 | 2937 | gfp_t gfp, bool new_page) |
10befea9 RG |
2938 | { |
2939 | unsigned int objects = objs_per_slab_page(s, page); | |
2e9bd483 | 2940 | unsigned long memcg_data; |
10befea9 RG |
2941 | void *vec; |
2942 | ||
2943 | vec = kcalloc_node(objects, sizeof(struct obj_cgroup *), gfp, | |
2944 | page_to_nid(page)); | |
2945 | if (!vec) | |
2946 | return -ENOMEM; | |
2947 | ||
2e9bd483 RG |
2948 | memcg_data = (unsigned long) vec | MEMCG_DATA_OBJCGS; |
2949 | if (new_page) { | |
2950 | /* | |
2951 | * If the slab page is brand new and nobody can yet access | |
2952 | * it's memcg_data, no synchronization is required and | |
2953 | * memcg_data can be simply assigned. | |
2954 | */ | |
2955 | page->memcg_data = memcg_data; | |
2956 | } else if (cmpxchg(&page->memcg_data, 0, memcg_data)) { | |
2957 | /* | |
2958 | * If the slab page is already in use, somebody can allocate | |
2959 | * and assign obj_cgroups in parallel. In this case the existing | |
2960 | * objcg vector should be reused. | |
2961 | */ | |
10befea9 | 2962 | kfree(vec); |
2e9bd483 RG |
2963 | return 0; |
2964 | } | |
10befea9 | 2965 | |
2e9bd483 | 2966 | kmemleak_not_leak(vec); |
10befea9 RG |
2967 | return 0; |
2968 | } | |
2969 | ||
8380ce47 RG |
2970 | /* |
2971 | * Returns a pointer to the memory cgroup to which the kernel object is charged. | |
2972 | * | |
bcfe06bf RG |
2973 | * A passed kernel object can be a slab object or a generic kernel page, so |
2974 | * different mechanisms for getting the memory cgroup pointer should be used. | |
2975 | * In certain cases (e.g. kernel stacks or large kmallocs with SLUB) the caller | |
2976 | * can not know for sure how the kernel object is implemented. | |
2977 | * mem_cgroup_from_obj() can be safely used in such cases. | |
2978 | * | |
8380ce47 RG |
2979 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), |
2980 | * cgroup_mutex, etc. | |
2981 | */ | |
2982 | struct mem_cgroup *mem_cgroup_from_obj(void *p) | |
2983 | { | |
2984 | struct page *page; | |
2985 | ||
2986 | if (mem_cgroup_disabled()) | |
2987 | return NULL; | |
2988 | ||
2989 | page = virt_to_head_page(p); | |
2990 | ||
2991 | /* | |
9855609b RG |
2992 | * Slab objects are accounted individually, not per-page. |
2993 | * Memcg membership data for each individual object is saved in | |
2994 | * the page->obj_cgroups. | |
8380ce47 | 2995 | */ |
270c6a71 | 2996 | if (page_objcgs_check(page)) { |
9855609b RG |
2997 | struct obj_cgroup *objcg; |
2998 | unsigned int off; | |
2999 | ||
3000 | off = obj_to_index(page->slab_cache, page, p); | |
270c6a71 | 3001 | objcg = page_objcgs(page)[off]; |
10befea9 RG |
3002 | if (objcg) |
3003 | return obj_cgroup_memcg(objcg); | |
3004 | ||
3005 | return NULL; | |
9855609b | 3006 | } |
8380ce47 | 3007 | |
bcfe06bf RG |
3008 | /* |
3009 | * page_memcg_check() is used here, because page_has_obj_cgroups() | |
3010 | * check above could fail because the object cgroups vector wasn't set | |
3011 | * at that moment, but it can be set concurrently. | |
3012 | * page_memcg_check(page) will guarantee that a proper memory | |
3013 | * cgroup pointer or NULL will be returned. | |
3014 | */ | |
3015 | return page_memcg_check(page); | |
8380ce47 RG |
3016 | } |
3017 | ||
bf4f0599 RG |
3018 | __always_inline struct obj_cgroup *get_obj_cgroup_from_current(void) |
3019 | { | |
3020 | struct obj_cgroup *objcg = NULL; | |
3021 | struct mem_cgroup *memcg; | |
3022 | ||
279c3393 RG |
3023 | if (memcg_kmem_bypass()) |
3024 | return NULL; | |
3025 | ||
bf4f0599 | 3026 | rcu_read_lock(); |
37d5985c RG |
3027 | if (unlikely(active_memcg())) |
3028 | memcg = active_memcg(); | |
bf4f0599 RG |
3029 | else |
3030 | memcg = mem_cgroup_from_task(current); | |
3031 | ||
3032 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { | |
3033 | objcg = rcu_dereference(memcg->objcg); | |
3034 | if (objcg && obj_cgroup_tryget(objcg)) | |
3035 | break; | |
2f7659a3 | 3036 | objcg = NULL; |
bf4f0599 RG |
3037 | } |
3038 | rcu_read_unlock(); | |
3039 | ||
3040 | return objcg; | |
3041 | } | |
3042 | ||
f3bb3043 | 3043 | static int memcg_alloc_cache_id(void) |
55007d84 | 3044 | { |
f3bb3043 VD |
3045 | int id, size; |
3046 | int err; | |
3047 | ||
dbcf73e2 | 3048 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
3049 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
3050 | if (id < 0) | |
3051 | return id; | |
55007d84 | 3052 | |
dbcf73e2 | 3053 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
3054 | return id; |
3055 | ||
3056 | /* | |
3057 | * There's no space for the new id in memcg_caches arrays, | |
3058 | * so we have to grow them. | |
3059 | */ | |
05257a1a | 3060 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
3061 | |
3062 | size = 2 * (id + 1); | |
55007d84 GC |
3063 | if (size < MEMCG_CACHES_MIN_SIZE) |
3064 | size = MEMCG_CACHES_MIN_SIZE; | |
3065 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
3066 | size = MEMCG_CACHES_MAX_SIZE; | |
3067 | ||
9855609b | 3068 | err = memcg_update_all_list_lrus(size); |
05257a1a VD |
3069 | if (!err) |
3070 | memcg_nr_cache_ids = size; | |
3071 | ||
3072 | up_write(&memcg_cache_ids_sem); | |
3073 | ||
f3bb3043 | 3074 | if (err) { |
dbcf73e2 | 3075 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
3076 | return err; |
3077 | } | |
3078 | return id; | |
3079 | } | |
3080 | ||
3081 | static void memcg_free_cache_id(int id) | |
3082 | { | |
dbcf73e2 | 3083 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
3084 | } |
3085 | ||
45264778 | 3086 | /** |
4b13f64d | 3087 | * __memcg_kmem_charge: charge a number of kernel pages to a memcg |
10eaec2f | 3088 | * @memcg: memory cgroup to charge |
45264778 | 3089 | * @gfp: reclaim mode |
92d0510c | 3090 | * @nr_pages: number of pages to charge |
45264778 VD |
3091 | * |
3092 | * Returns 0 on success, an error code on failure. | |
3093 | */ | |
c1a660de RG |
3094 | static int __memcg_kmem_charge(struct mem_cgroup *memcg, gfp_t gfp, |
3095 | unsigned int nr_pages) | |
7ae1e1d0 | 3096 | { |
f3ccb2c4 | 3097 | struct page_counter *counter; |
7ae1e1d0 GC |
3098 | int ret; |
3099 | ||
f3ccb2c4 | 3100 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 3101 | if (ret) |
f3ccb2c4 | 3102 | return ret; |
52c29b04 JW |
3103 | |
3104 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
3105 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
e55d9d9b MH |
3106 | |
3107 | /* | |
3108 | * Enforce __GFP_NOFAIL allocation because callers are not | |
3109 | * prepared to see failures and likely do not have any failure | |
3110 | * handling code. | |
3111 | */ | |
3112 | if (gfp & __GFP_NOFAIL) { | |
3113 | page_counter_charge(&memcg->kmem, nr_pages); | |
3114 | return 0; | |
3115 | } | |
52c29b04 JW |
3116 | cancel_charge(memcg, nr_pages); |
3117 | return -ENOMEM; | |
7ae1e1d0 | 3118 | } |
f3ccb2c4 | 3119 | return 0; |
7ae1e1d0 GC |
3120 | } |
3121 | ||
4b13f64d RG |
3122 | /** |
3123 | * __memcg_kmem_uncharge: uncharge a number of kernel pages from a memcg | |
3124 | * @memcg: memcg to uncharge | |
3125 | * @nr_pages: number of pages to uncharge | |
3126 | */ | |
c1a660de | 3127 | static void __memcg_kmem_uncharge(struct mem_cgroup *memcg, unsigned int nr_pages) |
4b13f64d RG |
3128 | { |
3129 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
3130 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
3131 | ||
3de7d4f2 | 3132 | refill_stock(memcg, nr_pages); |
4b13f64d RG |
3133 | } |
3134 | ||
45264778 | 3135 | /** |
f4b00eab | 3136 | * __memcg_kmem_charge_page: charge a kmem page to the current memory cgroup |
45264778 VD |
3137 | * @page: page to charge |
3138 | * @gfp: reclaim mode | |
3139 | * @order: allocation order | |
3140 | * | |
3141 | * Returns 0 on success, an error code on failure. | |
3142 | */ | |
f4b00eab | 3143 | int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 3144 | { |
f3ccb2c4 | 3145 | struct mem_cgroup *memcg; |
fcff7d7e | 3146 | int ret = 0; |
7ae1e1d0 | 3147 | |
d46eb14b | 3148 | memcg = get_mem_cgroup_from_current(); |
279c3393 | 3149 | if (memcg && !mem_cgroup_is_root(memcg)) { |
4b13f64d | 3150 | ret = __memcg_kmem_charge(memcg, gfp, 1 << order); |
4d96ba35 | 3151 | if (!ret) { |
18b2db3b RG |
3152 | page->memcg_data = (unsigned long)memcg | |
3153 | MEMCG_DATA_KMEM; | |
1a3e1f40 | 3154 | return 0; |
4d96ba35 | 3155 | } |
279c3393 | 3156 | css_put(&memcg->css); |
c4159a75 | 3157 | } |
d05e83a6 | 3158 | return ret; |
7ae1e1d0 | 3159 | } |
49a18eae | 3160 | |
45264778 | 3161 | /** |
f4b00eab | 3162 | * __memcg_kmem_uncharge_page: uncharge a kmem page |
45264778 VD |
3163 | * @page: page to uncharge |
3164 | * @order: allocation order | |
3165 | */ | |
f4b00eab | 3166 | void __memcg_kmem_uncharge_page(struct page *page, int order) |
7ae1e1d0 | 3167 | { |
bcfe06bf | 3168 | struct mem_cgroup *memcg = page_memcg(page); |
f3ccb2c4 | 3169 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 3170 | |
7ae1e1d0 GC |
3171 | if (!memcg) |
3172 | return; | |
3173 | ||
309381fe | 3174 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
4b13f64d | 3175 | __memcg_kmem_uncharge(memcg, nr_pages); |
bcfe06bf | 3176 | page->memcg_data = 0; |
1a3e1f40 | 3177 | css_put(&memcg->css); |
60d3fd32 | 3178 | } |
bf4f0599 RG |
3179 | |
3180 | static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) | |
3181 | { | |
3182 | struct memcg_stock_pcp *stock; | |
3183 | unsigned long flags; | |
3184 | bool ret = false; | |
3185 | ||
3186 | local_irq_save(flags); | |
3187 | ||
3188 | stock = this_cpu_ptr(&memcg_stock); | |
3189 | if (objcg == stock->cached_objcg && stock->nr_bytes >= nr_bytes) { | |
3190 | stock->nr_bytes -= nr_bytes; | |
3191 | ret = true; | |
3192 | } | |
3193 | ||
3194 | local_irq_restore(flags); | |
3195 | ||
3196 | return ret; | |
3197 | } | |
3198 | ||
3199 | static void drain_obj_stock(struct memcg_stock_pcp *stock) | |
3200 | { | |
3201 | struct obj_cgroup *old = stock->cached_objcg; | |
3202 | ||
3203 | if (!old) | |
3204 | return; | |
3205 | ||
3206 | if (stock->nr_bytes) { | |
3207 | unsigned int nr_pages = stock->nr_bytes >> PAGE_SHIFT; | |
3208 | unsigned int nr_bytes = stock->nr_bytes & (PAGE_SIZE - 1); | |
3209 | ||
3210 | if (nr_pages) { | |
3211 | rcu_read_lock(); | |
3212 | __memcg_kmem_uncharge(obj_cgroup_memcg(old), nr_pages); | |
3213 | rcu_read_unlock(); | |
3214 | } | |
3215 | ||
3216 | /* | |
3217 | * The leftover is flushed to the centralized per-memcg value. | |
3218 | * On the next attempt to refill obj stock it will be moved | |
3219 | * to a per-cpu stock (probably, on an other CPU), see | |
3220 | * refill_obj_stock(). | |
3221 | * | |
3222 | * How often it's flushed is a trade-off between the memory | |
3223 | * limit enforcement accuracy and potential CPU contention, | |
3224 | * so it might be changed in the future. | |
3225 | */ | |
3226 | atomic_add(nr_bytes, &old->nr_charged_bytes); | |
3227 | stock->nr_bytes = 0; | |
3228 | } | |
3229 | ||
3230 | obj_cgroup_put(old); | |
3231 | stock->cached_objcg = NULL; | |
3232 | } | |
3233 | ||
3234 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
3235 | struct mem_cgroup *root_memcg) | |
3236 | { | |
3237 | struct mem_cgroup *memcg; | |
3238 | ||
3239 | if (stock->cached_objcg) { | |
3240 | memcg = obj_cgroup_memcg(stock->cached_objcg); | |
3241 | if (memcg && mem_cgroup_is_descendant(memcg, root_memcg)) | |
3242 | return true; | |
3243 | } | |
3244 | ||
3245 | return false; | |
3246 | } | |
3247 | ||
3248 | static void refill_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) | |
3249 | { | |
3250 | struct memcg_stock_pcp *stock; | |
3251 | unsigned long flags; | |
3252 | ||
3253 | local_irq_save(flags); | |
3254 | ||
3255 | stock = this_cpu_ptr(&memcg_stock); | |
3256 | if (stock->cached_objcg != objcg) { /* reset if necessary */ | |
3257 | drain_obj_stock(stock); | |
3258 | obj_cgroup_get(objcg); | |
3259 | stock->cached_objcg = objcg; | |
3260 | stock->nr_bytes = atomic_xchg(&objcg->nr_charged_bytes, 0); | |
3261 | } | |
3262 | stock->nr_bytes += nr_bytes; | |
3263 | ||
3264 | if (stock->nr_bytes > PAGE_SIZE) | |
3265 | drain_obj_stock(stock); | |
3266 | ||
3267 | local_irq_restore(flags); | |
3268 | } | |
3269 | ||
3270 | int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size) | |
3271 | { | |
3272 | struct mem_cgroup *memcg; | |
3273 | unsigned int nr_pages, nr_bytes; | |
3274 | int ret; | |
3275 | ||
3276 | if (consume_obj_stock(objcg, size)) | |
3277 | return 0; | |
3278 | ||
3279 | /* | |
3280 | * In theory, memcg->nr_charged_bytes can have enough | |
3281 | * pre-charged bytes to satisfy the allocation. However, | |
3282 | * flushing memcg->nr_charged_bytes requires two atomic | |
3283 | * operations, and memcg->nr_charged_bytes can't be big, | |
3284 | * so it's better to ignore it and try grab some new pages. | |
3285 | * memcg->nr_charged_bytes will be flushed in | |
3286 | * refill_obj_stock(), called from this function or | |
3287 | * independently later. | |
3288 | */ | |
3289 | rcu_read_lock(); | |
eefbfa7f | 3290 | retry: |
bf4f0599 | 3291 | memcg = obj_cgroup_memcg(objcg); |
eefbfa7f MS |
3292 | if (unlikely(!css_tryget(&memcg->css))) |
3293 | goto retry; | |
bf4f0599 RG |
3294 | rcu_read_unlock(); |
3295 | ||
3296 | nr_pages = size >> PAGE_SHIFT; | |
3297 | nr_bytes = size & (PAGE_SIZE - 1); | |
3298 | ||
3299 | if (nr_bytes) | |
3300 | nr_pages += 1; | |
3301 | ||
3302 | ret = __memcg_kmem_charge(memcg, gfp, nr_pages); | |
3303 | if (!ret && nr_bytes) | |
3304 | refill_obj_stock(objcg, PAGE_SIZE - nr_bytes); | |
3305 | ||
3306 | css_put(&memcg->css); | |
3307 | return ret; | |
3308 | } | |
3309 | ||
3310 | void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size) | |
3311 | { | |
3312 | refill_obj_stock(objcg, size); | |
3313 | } | |
3314 | ||
84c07d11 | 3315 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 3316 | |
ca3e0214 | 3317 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
ca3e0214 | 3318 | /* |
6168d0da | 3319 | * Because page_memcg(head) is not set on compound tails, set it now. |
ca3e0214 | 3320 | */ |
e94c8a9c | 3321 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 3322 | { |
bcfe06bf | 3323 | struct mem_cgroup *memcg = page_memcg(head); |
e94c8a9c | 3324 | int i; |
ca3e0214 | 3325 | |
3d37c4a9 KH |
3326 | if (mem_cgroup_disabled()) |
3327 | return; | |
b070e65c | 3328 | |
1a3e1f40 JW |
3329 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
3330 | css_get(&memcg->css); | |
bcfe06bf | 3331 | head[i].memcg_data = (unsigned long)memcg; |
1a3e1f40 | 3332 | } |
ca3e0214 | 3333 | } |
12d27107 | 3334 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3335 | |
c255a458 | 3336 | #ifdef CONFIG_MEMCG_SWAP |
02491447 DN |
3337 | /** |
3338 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3339 | * @entry: swap entry to be moved | |
3340 | * @from: mem_cgroup which the entry is moved from | |
3341 | * @to: mem_cgroup which the entry is moved to | |
3342 | * | |
3343 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3344 | * as the mem_cgroup's id of @from. | |
3345 | * | |
3346 | * Returns 0 on success, -EINVAL on failure. | |
3347 | * | |
3e32cb2e | 3348 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
3349 | * both res and memsw, and called css_get(). |
3350 | */ | |
3351 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3352 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3353 | { |
3354 | unsigned short old_id, new_id; | |
3355 | ||
34c00c31 LZ |
3356 | old_id = mem_cgroup_id(from); |
3357 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3358 | |
3359 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
3360 | mod_memcg_state(from, MEMCG_SWAP, -1); |
3361 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
3362 | return 0; |
3363 | } | |
3364 | return -EINVAL; | |
3365 | } | |
3366 | #else | |
3367 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3368 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3369 | { |
3370 | return -EINVAL; | |
3371 | } | |
8c7c6e34 | 3372 | #endif |
d13d1443 | 3373 | |
bbec2e15 | 3374 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 3375 | |
bbec2e15 RG |
3376 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
3377 | unsigned long max, bool memsw) | |
628f4235 | 3378 | { |
3e32cb2e | 3379 | bool enlarge = false; |
bb4a7ea2 | 3380 | bool drained = false; |
3e32cb2e | 3381 | int ret; |
c054a78c YZ |
3382 | bool limits_invariant; |
3383 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 3384 | |
3e32cb2e | 3385 | do { |
628f4235 KH |
3386 | if (signal_pending(current)) { |
3387 | ret = -EINTR; | |
3388 | break; | |
3389 | } | |
3e32cb2e | 3390 | |
bbec2e15 | 3391 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
3392 | /* |
3393 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 3394 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 3395 | */ |
15b42562 | 3396 | limits_invariant = memsw ? max >= READ_ONCE(memcg->memory.max) : |
bbec2e15 | 3397 | max <= memcg->memsw.max; |
c054a78c | 3398 | if (!limits_invariant) { |
bbec2e15 | 3399 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 3400 | ret = -EINVAL; |
8c7c6e34 KH |
3401 | break; |
3402 | } | |
bbec2e15 | 3403 | if (max > counter->max) |
3e32cb2e | 3404 | enlarge = true; |
bbec2e15 RG |
3405 | ret = page_counter_set_max(counter, max); |
3406 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
3407 | |
3408 | if (!ret) | |
3409 | break; | |
3410 | ||
bb4a7ea2 SB |
3411 | if (!drained) { |
3412 | drain_all_stock(memcg); | |
3413 | drained = true; | |
3414 | continue; | |
3415 | } | |
3416 | ||
1ab5c056 AR |
3417 | if (!try_to_free_mem_cgroup_pages(memcg, 1, |
3418 | GFP_KERNEL, !memsw)) { | |
3419 | ret = -EBUSY; | |
3420 | break; | |
3421 | } | |
3422 | } while (true); | |
3e32cb2e | 3423 | |
3c11ecf4 KH |
3424 | if (!ret && enlarge) |
3425 | memcg_oom_recover(memcg); | |
3e32cb2e | 3426 | |
628f4235 KH |
3427 | return ret; |
3428 | } | |
3429 | ||
ef8f2327 | 3430 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
3431 | gfp_t gfp_mask, |
3432 | unsigned long *total_scanned) | |
3433 | { | |
3434 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 3435 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
3436 | unsigned long reclaimed; |
3437 | int loop = 0; | |
ef8f2327 | 3438 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 3439 | unsigned long excess; |
0608f43d AM |
3440 | unsigned long nr_scanned; |
3441 | ||
3442 | if (order > 0) | |
3443 | return 0; | |
3444 | ||
ef8f2327 | 3445 | mctz = soft_limit_tree_node(pgdat->node_id); |
d6507ff5 MH |
3446 | |
3447 | /* | |
3448 | * Do not even bother to check the largest node if the root | |
3449 | * is empty. Do it lockless to prevent lock bouncing. Races | |
3450 | * are acceptable as soft limit is best effort anyway. | |
3451 | */ | |
bfc7228b | 3452 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
3453 | return 0; |
3454 | ||
0608f43d AM |
3455 | /* |
3456 | * This loop can run a while, specially if mem_cgroup's continuously | |
3457 | * keep exceeding their soft limit and putting the system under | |
3458 | * pressure | |
3459 | */ | |
3460 | do { | |
3461 | if (next_mz) | |
3462 | mz = next_mz; | |
3463 | else | |
3464 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3465 | if (!mz) | |
3466 | break; | |
3467 | ||
3468 | nr_scanned = 0; | |
ef8f2327 | 3469 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
0608f43d AM |
3470 | gfp_mask, &nr_scanned); |
3471 | nr_reclaimed += reclaimed; | |
3472 | *total_scanned += nr_scanned; | |
0a31bc97 | 3473 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 3474 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
3475 | |
3476 | /* | |
3477 | * If we failed to reclaim anything from this memory cgroup | |
3478 | * it is time to move on to the next cgroup | |
3479 | */ | |
3480 | next_mz = NULL; | |
bc2f2e7f VD |
3481 | if (!reclaimed) |
3482 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3483 | ||
3e32cb2e | 3484 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3485 | /* |
3486 | * One school of thought says that we should not add | |
3487 | * back the node to the tree if reclaim returns 0. | |
3488 | * But our reclaim could return 0, simply because due | |
3489 | * to priority we are exposing a smaller subset of | |
3490 | * memory to reclaim from. Consider this as a longer | |
3491 | * term TODO. | |
3492 | */ | |
3493 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3494 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3495 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3496 | css_put(&mz->memcg->css); |
3497 | loop++; | |
3498 | /* | |
3499 | * Could not reclaim anything and there are no more | |
3500 | * mem cgroups to try or we seem to be looping without | |
3501 | * reclaiming anything. | |
3502 | */ | |
3503 | if (!nr_reclaimed && | |
3504 | (next_mz == NULL || | |
3505 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3506 | break; | |
3507 | } while (!nr_reclaimed); | |
3508 | if (next_mz) | |
3509 | css_put(&next_mz->memcg->css); | |
3510 | return nr_reclaimed; | |
3511 | } | |
3512 | ||
c26251f9 | 3513 | /* |
51038171 | 3514 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
3515 | * |
3516 | * Caller is responsible for holding css reference for memcg. | |
3517 | */ | |
3518 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3519 | { | |
d977aa93 | 3520 | int nr_retries = MAX_RECLAIM_RETRIES; |
c26251f9 | 3521 | |
c1e862c1 KH |
3522 | /* we call try-to-free pages for make this cgroup empty */ |
3523 | lru_add_drain_all(); | |
d12c60f6 JS |
3524 | |
3525 | drain_all_stock(memcg); | |
3526 | ||
f817ed48 | 3527 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3528 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 3529 | int progress; |
c1e862c1 | 3530 | |
c26251f9 MH |
3531 | if (signal_pending(current)) |
3532 | return -EINTR; | |
3533 | ||
b70a2a21 JW |
3534 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3535 | GFP_KERNEL, true); | |
c1e862c1 | 3536 | if (!progress) { |
f817ed48 | 3537 | nr_retries--; |
c1e862c1 | 3538 | /* maybe some writeback is necessary */ |
8aa7e847 | 3539 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3540 | } |
f817ed48 KH |
3541 | |
3542 | } | |
ab5196c2 MH |
3543 | |
3544 | return 0; | |
cc847582 KH |
3545 | } |
3546 | ||
6770c64e TH |
3547 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3548 | char *buf, size_t nbytes, | |
3549 | loff_t off) | |
c1e862c1 | 3550 | { |
6770c64e | 3551 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3552 | |
d8423011 MH |
3553 | if (mem_cgroup_is_root(memcg)) |
3554 | return -EINVAL; | |
6770c64e | 3555 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3556 | } |
3557 | ||
182446d0 TH |
3558 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3559 | struct cftype *cft) | |
18f59ea7 | 3560 | { |
bef8620c | 3561 | return 1; |
18f59ea7 BS |
3562 | } |
3563 | ||
182446d0 TH |
3564 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3565 | struct cftype *cft, u64 val) | |
18f59ea7 | 3566 | { |
bef8620c | 3567 | if (val == 1) |
0b8f73e1 | 3568 | return 0; |
567fb435 | 3569 | |
bef8620c RG |
3570 | pr_warn_once("Non-hierarchical mode is deprecated. " |
3571 | "Please report your usecase to linux-mm@kvack.org if you " | |
3572 | "depend on this functionality.\n"); | |
567fb435 | 3573 | |
bef8620c | 3574 | return -EINVAL; |
18f59ea7 BS |
3575 | } |
3576 | ||
6f646156 | 3577 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 3578 | { |
42a30035 | 3579 | unsigned long val; |
ce00a967 | 3580 | |
3e32cb2e | 3581 | if (mem_cgroup_is_root(memcg)) { |
0d1c2072 | 3582 | val = memcg_page_state(memcg, NR_FILE_PAGES) + |
be5d0a74 | 3583 | memcg_page_state(memcg, NR_ANON_MAPPED); |
42a30035 JW |
3584 | if (swap) |
3585 | val += memcg_page_state(memcg, MEMCG_SWAP); | |
3e32cb2e | 3586 | } else { |
ce00a967 | 3587 | if (!swap) |
3e32cb2e | 3588 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3589 | else |
3e32cb2e | 3590 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3591 | } |
c12176d3 | 3592 | return val; |
ce00a967 JW |
3593 | } |
3594 | ||
3e32cb2e JW |
3595 | enum { |
3596 | RES_USAGE, | |
3597 | RES_LIMIT, | |
3598 | RES_MAX_USAGE, | |
3599 | RES_FAILCNT, | |
3600 | RES_SOFT_LIMIT, | |
3601 | }; | |
ce00a967 | 3602 | |
791badbd | 3603 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3604 | struct cftype *cft) |
8cdea7c0 | 3605 | { |
182446d0 | 3606 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3607 | struct page_counter *counter; |
af36f906 | 3608 | |
3e32cb2e | 3609 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3610 | case _MEM: |
3e32cb2e JW |
3611 | counter = &memcg->memory; |
3612 | break; | |
8c7c6e34 | 3613 | case _MEMSWAP: |
3e32cb2e JW |
3614 | counter = &memcg->memsw; |
3615 | break; | |
510fc4e1 | 3616 | case _KMEM: |
3e32cb2e | 3617 | counter = &memcg->kmem; |
510fc4e1 | 3618 | break; |
d55f90bf | 3619 | case _TCP: |
0db15298 | 3620 | counter = &memcg->tcpmem; |
d55f90bf | 3621 | break; |
8c7c6e34 KH |
3622 | default: |
3623 | BUG(); | |
8c7c6e34 | 3624 | } |
3e32cb2e JW |
3625 | |
3626 | switch (MEMFILE_ATTR(cft->private)) { | |
3627 | case RES_USAGE: | |
3628 | if (counter == &memcg->memory) | |
c12176d3 | 3629 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 3630 | if (counter == &memcg->memsw) |
c12176d3 | 3631 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
3632 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
3633 | case RES_LIMIT: | |
bbec2e15 | 3634 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
3635 | case RES_MAX_USAGE: |
3636 | return (u64)counter->watermark * PAGE_SIZE; | |
3637 | case RES_FAILCNT: | |
3638 | return counter->failcnt; | |
3639 | case RES_SOFT_LIMIT: | |
3640 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3641 | default: | |
3642 | BUG(); | |
3643 | } | |
8cdea7c0 | 3644 | } |
510fc4e1 | 3645 | |
4a87e2a2 | 3646 | static void memcg_flush_percpu_vmstats(struct mem_cgroup *memcg) |
c350a99e | 3647 | { |
4a87e2a2 | 3648 | unsigned long stat[MEMCG_NR_STAT] = {0}; |
c350a99e RG |
3649 | struct mem_cgroup *mi; |
3650 | int node, cpu, i; | |
c350a99e RG |
3651 | |
3652 | for_each_online_cpu(cpu) | |
4a87e2a2 | 3653 | for (i = 0; i < MEMCG_NR_STAT; i++) |
6c1c2808 | 3654 | stat[i] += per_cpu(memcg->vmstats_percpu->stat[i], cpu); |
c350a99e RG |
3655 | |
3656 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
4a87e2a2 | 3657 | for (i = 0; i < MEMCG_NR_STAT; i++) |
c350a99e RG |
3658 | atomic_long_add(stat[i], &mi->vmstats[i]); |
3659 | ||
3660 | for_each_node(node) { | |
3661 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; | |
3662 | struct mem_cgroup_per_node *pi; | |
3663 | ||
4a87e2a2 | 3664 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
c350a99e RG |
3665 | stat[i] = 0; |
3666 | ||
3667 | for_each_online_cpu(cpu) | |
4a87e2a2 | 3668 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
6c1c2808 SB |
3669 | stat[i] += per_cpu( |
3670 | pn->lruvec_stat_cpu->count[i], cpu); | |
c350a99e RG |
3671 | |
3672 | for (pi = pn; pi; pi = parent_nodeinfo(pi, node)) | |
4a87e2a2 | 3673 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
c350a99e RG |
3674 | atomic_long_add(stat[i], &pi->lruvec_stat[i]); |
3675 | } | |
3676 | } | |
3677 | ||
bb65f89b RG |
3678 | static void memcg_flush_percpu_vmevents(struct mem_cgroup *memcg) |
3679 | { | |
3680 | unsigned long events[NR_VM_EVENT_ITEMS]; | |
3681 | struct mem_cgroup *mi; | |
3682 | int cpu, i; | |
3683 | ||
3684 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
3685 | events[i] = 0; | |
3686 | ||
3687 | for_each_online_cpu(cpu) | |
3688 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
6c1c2808 SB |
3689 | events[i] += per_cpu(memcg->vmstats_percpu->events[i], |
3690 | cpu); | |
bb65f89b RG |
3691 | |
3692 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
3693 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
3694 | atomic_long_add(events[i], &mi->vmevents[i]); | |
3695 | } | |
3696 | ||
84c07d11 | 3697 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 3698 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 3699 | { |
bf4f0599 | 3700 | struct obj_cgroup *objcg; |
d6441637 VD |
3701 | int memcg_id; |
3702 | ||
b313aeee VD |
3703 | if (cgroup_memory_nokmem) |
3704 | return 0; | |
3705 | ||
2a4db7eb | 3706 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 3707 | BUG_ON(memcg->kmem_state); |
d6441637 | 3708 | |
f3bb3043 | 3709 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
3710 | if (memcg_id < 0) |
3711 | return memcg_id; | |
d6441637 | 3712 | |
bf4f0599 RG |
3713 | objcg = obj_cgroup_alloc(); |
3714 | if (!objcg) { | |
3715 | memcg_free_cache_id(memcg_id); | |
3716 | return -ENOMEM; | |
3717 | } | |
3718 | objcg->memcg = memcg; | |
3719 | rcu_assign_pointer(memcg->objcg, objcg); | |
3720 | ||
d648bcc7 RG |
3721 | static_branch_enable(&memcg_kmem_enabled_key); |
3722 | ||
900a38f0 | 3723 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 3724 | memcg->kmem_state = KMEM_ONLINE; |
0b8f73e1 JW |
3725 | |
3726 | return 0; | |
d6441637 VD |
3727 | } |
3728 | ||
8e0a8912 JW |
3729 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
3730 | { | |
3731 | struct cgroup_subsys_state *css; | |
3732 | struct mem_cgroup *parent, *child; | |
3733 | int kmemcg_id; | |
3734 | ||
3735 | if (memcg->kmem_state != KMEM_ONLINE) | |
3736 | return; | |
9855609b | 3737 | |
8e0a8912 JW |
3738 | memcg->kmem_state = KMEM_ALLOCATED; |
3739 | ||
8e0a8912 JW |
3740 | parent = parent_mem_cgroup(memcg); |
3741 | if (!parent) | |
3742 | parent = root_mem_cgroup; | |
3743 | ||
bf4f0599 | 3744 | memcg_reparent_objcgs(memcg, parent); |
fb2f2b0a RG |
3745 | |
3746 | kmemcg_id = memcg->kmemcg_id; | |
3747 | BUG_ON(kmemcg_id < 0); | |
3748 | ||
8e0a8912 JW |
3749 | /* |
3750 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3751 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3752 | * to ones of the parent. After we have finished, all list_lrus | |
3753 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3754 | * ordering is imposed by list_lru_node->lock taken by | |
3755 | * memcg_drain_all_list_lrus(). | |
3756 | */ | |
3a06bb78 | 3757 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
8e0a8912 JW |
3758 | css_for_each_descendant_pre(css, &memcg->css) { |
3759 | child = mem_cgroup_from_css(css); | |
3760 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3761 | child->kmemcg_id = parent->kmemcg_id; | |
8e0a8912 | 3762 | } |
3a06bb78 TH |
3763 | rcu_read_unlock(); |
3764 | ||
9bec5c35 | 3765 | memcg_drain_all_list_lrus(kmemcg_id, parent); |
8e0a8912 JW |
3766 | |
3767 | memcg_free_cache_id(kmemcg_id); | |
3768 | } | |
3769 | ||
3770 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3771 | { | |
0b8f73e1 JW |
3772 | /* css_alloc() failed, offlining didn't happen */ |
3773 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
3774 | memcg_offline_kmem(memcg); | |
8e0a8912 | 3775 | } |
d6441637 | 3776 | #else |
0b8f73e1 | 3777 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
3778 | { |
3779 | return 0; | |
3780 | } | |
3781 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
3782 | { | |
3783 | } | |
3784 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3785 | { | |
3786 | } | |
84c07d11 | 3787 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 3788 | |
bbec2e15 RG |
3789 | static int memcg_update_kmem_max(struct mem_cgroup *memcg, |
3790 | unsigned long max) | |
d6441637 | 3791 | { |
b313aeee | 3792 | int ret; |
127424c8 | 3793 | |
bbec2e15 RG |
3794 | mutex_lock(&memcg_max_mutex); |
3795 | ret = page_counter_set_max(&memcg->kmem, max); | |
3796 | mutex_unlock(&memcg_max_mutex); | |
127424c8 | 3797 | return ret; |
d6441637 | 3798 | } |
510fc4e1 | 3799 | |
bbec2e15 | 3800 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
3801 | { |
3802 | int ret; | |
3803 | ||
bbec2e15 | 3804 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 3805 | |
bbec2e15 | 3806 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
3807 | if (ret) |
3808 | goto out; | |
3809 | ||
0db15298 | 3810 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
3811 | /* |
3812 | * The active flag needs to be written after the static_key | |
3813 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
3814 | * function is the last one to run. See mem_cgroup_sk_alloc() |
3815 | * for details, and note that we don't mark any socket as | |
3816 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
3817 | * |
3818 | * We need to do this, because static_keys will span multiple | |
3819 | * sites, but we can't control their order. If we mark a socket | |
3820 | * as accounted, but the accounting functions are not patched in | |
3821 | * yet, we'll lose accounting. | |
3822 | * | |
2d758073 | 3823 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
3824 | * because when this value change, the code to process it is not |
3825 | * patched in yet. | |
3826 | */ | |
3827 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 3828 | memcg->tcpmem_active = true; |
d55f90bf VD |
3829 | } |
3830 | out: | |
bbec2e15 | 3831 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
3832 | return ret; |
3833 | } | |
d55f90bf | 3834 | |
628f4235 KH |
3835 | /* |
3836 | * The user of this function is... | |
3837 | * RES_LIMIT. | |
3838 | */ | |
451af504 TH |
3839 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3840 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3841 | { |
451af504 | 3842 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3843 | unsigned long nr_pages; |
628f4235 KH |
3844 | int ret; |
3845 | ||
451af504 | 3846 | buf = strstrip(buf); |
650c5e56 | 3847 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3848 | if (ret) |
3849 | return ret; | |
af36f906 | 3850 | |
3e32cb2e | 3851 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3852 | case RES_LIMIT: |
4b3bde4c BS |
3853 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3854 | ret = -EINVAL; | |
3855 | break; | |
3856 | } | |
3e32cb2e JW |
3857 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3858 | case _MEM: | |
bbec2e15 | 3859 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 3860 | break; |
3e32cb2e | 3861 | case _MEMSWAP: |
bbec2e15 | 3862 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 3863 | break; |
3e32cb2e | 3864 | case _KMEM: |
0158115f MH |
3865 | pr_warn_once("kmem.limit_in_bytes is deprecated and will be removed. " |
3866 | "Please report your usecase to linux-mm@kvack.org if you " | |
3867 | "depend on this functionality.\n"); | |
bbec2e15 | 3868 | ret = memcg_update_kmem_max(memcg, nr_pages); |
3e32cb2e | 3869 | break; |
d55f90bf | 3870 | case _TCP: |
bbec2e15 | 3871 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 3872 | break; |
3e32cb2e | 3873 | } |
296c81d8 | 3874 | break; |
3e32cb2e JW |
3875 | case RES_SOFT_LIMIT: |
3876 | memcg->soft_limit = nr_pages; | |
3877 | ret = 0; | |
628f4235 KH |
3878 | break; |
3879 | } | |
451af504 | 3880 | return ret ?: nbytes; |
8cdea7c0 BS |
3881 | } |
3882 | ||
6770c64e TH |
3883 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3884 | size_t nbytes, loff_t off) | |
c84872e1 | 3885 | { |
6770c64e | 3886 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3887 | struct page_counter *counter; |
c84872e1 | 3888 | |
3e32cb2e JW |
3889 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3890 | case _MEM: | |
3891 | counter = &memcg->memory; | |
3892 | break; | |
3893 | case _MEMSWAP: | |
3894 | counter = &memcg->memsw; | |
3895 | break; | |
3896 | case _KMEM: | |
3897 | counter = &memcg->kmem; | |
3898 | break; | |
d55f90bf | 3899 | case _TCP: |
0db15298 | 3900 | counter = &memcg->tcpmem; |
d55f90bf | 3901 | break; |
3e32cb2e JW |
3902 | default: |
3903 | BUG(); | |
3904 | } | |
af36f906 | 3905 | |
3e32cb2e | 3906 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3907 | case RES_MAX_USAGE: |
3e32cb2e | 3908 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3909 | break; |
3910 | case RES_FAILCNT: | |
3e32cb2e | 3911 | counter->failcnt = 0; |
29f2a4da | 3912 | break; |
3e32cb2e JW |
3913 | default: |
3914 | BUG(); | |
29f2a4da | 3915 | } |
f64c3f54 | 3916 | |
6770c64e | 3917 | return nbytes; |
c84872e1 PE |
3918 | } |
3919 | ||
182446d0 | 3920 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3921 | struct cftype *cft) |
3922 | { | |
182446d0 | 3923 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3924 | } |
3925 | ||
02491447 | 3926 | #ifdef CONFIG_MMU |
182446d0 | 3927 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3928 | struct cftype *cft, u64 val) |
3929 | { | |
182446d0 | 3930 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3931 | |
1dfab5ab | 3932 | if (val & ~MOVE_MASK) |
7dc74be0 | 3933 | return -EINVAL; |
ee5e8472 | 3934 | |
7dc74be0 | 3935 | /* |
ee5e8472 GC |
3936 | * No kind of locking is needed in here, because ->can_attach() will |
3937 | * check this value once in the beginning of the process, and then carry | |
3938 | * on with stale data. This means that changes to this value will only | |
3939 | * affect task migrations starting after the change. | |
7dc74be0 | 3940 | */ |
c0ff4b85 | 3941 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3942 | return 0; |
3943 | } | |
02491447 | 3944 | #else |
182446d0 | 3945 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3946 | struct cftype *cft, u64 val) |
3947 | { | |
3948 | return -ENOSYS; | |
3949 | } | |
3950 | #endif | |
7dc74be0 | 3951 | |
406eb0c9 | 3952 | #ifdef CONFIG_NUMA |
113b7dfd JW |
3953 | |
3954 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
3955 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
3956 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) | |
3957 | ||
3958 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 | 3959 | int nid, unsigned int lru_mask, bool tree) |
113b7dfd | 3960 | { |
867e5e1d | 3961 | struct lruvec *lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); |
113b7dfd JW |
3962 | unsigned long nr = 0; |
3963 | enum lru_list lru; | |
3964 | ||
3965 | VM_BUG_ON((unsigned)nid >= nr_node_ids); | |
3966 | ||
3967 | for_each_lru(lru) { | |
3968 | if (!(BIT(lru) & lru_mask)) | |
3969 | continue; | |
dd8657b6 SB |
3970 | if (tree) |
3971 | nr += lruvec_page_state(lruvec, NR_LRU_BASE + lru); | |
3972 | else | |
3973 | nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru); | |
113b7dfd JW |
3974 | } |
3975 | return nr; | |
3976 | } | |
3977 | ||
3978 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 SB |
3979 | unsigned int lru_mask, |
3980 | bool tree) | |
113b7dfd JW |
3981 | { |
3982 | unsigned long nr = 0; | |
3983 | enum lru_list lru; | |
3984 | ||
3985 | for_each_lru(lru) { | |
3986 | if (!(BIT(lru) & lru_mask)) | |
3987 | continue; | |
dd8657b6 SB |
3988 | if (tree) |
3989 | nr += memcg_page_state(memcg, NR_LRU_BASE + lru); | |
3990 | else | |
3991 | nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru); | |
113b7dfd JW |
3992 | } |
3993 | return nr; | |
3994 | } | |
3995 | ||
2da8ca82 | 3996 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3997 | { |
25485de6 GT |
3998 | struct numa_stat { |
3999 | const char *name; | |
4000 | unsigned int lru_mask; | |
4001 | }; | |
4002 | ||
4003 | static const struct numa_stat stats[] = { | |
4004 | { "total", LRU_ALL }, | |
4005 | { "file", LRU_ALL_FILE }, | |
4006 | { "anon", LRU_ALL_ANON }, | |
4007 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
4008 | }; | |
4009 | const struct numa_stat *stat; | |
406eb0c9 | 4010 | int nid; |
aa9694bb | 4011 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
406eb0c9 | 4012 | |
25485de6 | 4013 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
4014 | seq_printf(m, "%s=%lu", stat->name, |
4015 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
4016 | false)); | |
4017 | for_each_node_state(nid, N_MEMORY) | |
4018 | seq_printf(m, " N%d=%lu", nid, | |
4019 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
4020 | stat->lru_mask, false)); | |
25485de6 | 4021 | seq_putc(m, '\n'); |
406eb0c9 | 4022 | } |
406eb0c9 | 4023 | |
071aee13 | 4024 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
4025 | |
4026 | seq_printf(m, "hierarchical_%s=%lu", stat->name, | |
4027 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
4028 | true)); | |
4029 | for_each_node_state(nid, N_MEMORY) | |
4030 | seq_printf(m, " N%d=%lu", nid, | |
4031 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
4032 | stat->lru_mask, true)); | |
071aee13 | 4033 | seq_putc(m, '\n'); |
406eb0c9 | 4034 | } |
406eb0c9 | 4035 | |
406eb0c9 YH |
4036 | return 0; |
4037 | } | |
4038 | #endif /* CONFIG_NUMA */ | |
4039 | ||
c8713d0b | 4040 | static const unsigned int memcg1_stats[] = { |
0d1c2072 | 4041 | NR_FILE_PAGES, |
be5d0a74 | 4042 | NR_ANON_MAPPED, |
468c3982 JW |
4043 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4044 | NR_ANON_THPS, | |
4045 | #endif | |
c8713d0b JW |
4046 | NR_SHMEM, |
4047 | NR_FILE_MAPPED, | |
4048 | NR_FILE_DIRTY, | |
4049 | NR_WRITEBACK, | |
4050 | MEMCG_SWAP, | |
4051 | }; | |
4052 | ||
4053 | static const char *const memcg1_stat_names[] = { | |
4054 | "cache", | |
4055 | "rss", | |
468c3982 | 4056 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
c8713d0b | 4057 | "rss_huge", |
468c3982 | 4058 | #endif |
c8713d0b JW |
4059 | "shmem", |
4060 | "mapped_file", | |
4061 | "dirty", | |
4062 | "writeback", | |
4063 | "swap", | |
4064 | }; | |
4065 | ||
df0e53d0 | 4066 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 4067 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
4068 | PGPGIN, |
4069 | PGPGOUT, | |
4070 | PGFAULT, | |
4071 | PGMAJFAULT, | |
4072 | }; | |
4073 | ||
2da8ca82 | 4074 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 4075 | { |
aa9694bb | 4076 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3e32cb2e | 4077 | unsigned long memory, memsw; |
af7c4b0e JW |
4078 | struct mem_cgroup *mi; |
4079 | unsigned int i; | |
406eb0c9 | 4080 | |
71cd3113 | 4081 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c | 4082 | |
71cd3113 | 4083 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
468c3982 JW |
4084 | unsigned long nr; |
4085 | ||
71cd3113 | 4086 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 4087 | continue; |
468c3982 | 4088 | nr = memcg_page_state_local(memcg, memcg1_stats[i]); |
468c3982 | 4089 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], nr * PAGE_SIZE); |
1dd3a273 | 4090 | } |
7b854121 | 4091 | |
df0e53d0 | 4092 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d | 4093 | seq_printf(m, "%s %lu\n", vm_event_name(memcg1_events[i]), |
205b20cc | 4094 | memcg_events_local(memcg, memcg1_events[i])); |
af7c4b0e JW |
4095 | |
4096 | for (i = 0; i < NR_LRU_LISTS; i++) | |
ebc5d83d | 4097 | seq_printf(m, "%s %lu\n", lru_list_name(i), |
205b20cc | 4098 | memcg_page_state_local(memcg, NR_LRU_BASE + i) * |
21d89d15 | 4099 | PAGE_SIZE); |
af7c4b0e | 4100 | |
14067bb3 | 4101 | /* Hierarchical information */ |
3e32cb2e JW |
4102 | memory = memsw = PAGE_COUNTER_MAX; |
4103 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
15b42562 CD |
4104 | memory = min(memory, READ_ONCE(mi->memory.max)); |
4105 | memsw = min(memsw, READ_ONCE(mi->memsw.max)); | |
fee7b548 | 4106 | } |
3e32cb2e JW |
4107 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
4108 | (u64)memory * PAGE_SIZE); | |
7941d214 | 4109 | if (do_memsw_account()) |
3e32cb2e JW |
4110 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
4111 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 4112 | |
8de7ecc6 | 4113 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
7de2e9f1 | 4114 | unsigned long nr; |
4115 | ||
71cd3113 | 4116 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 4117 | continue; |
7de2e9f1 | 4118 | nr = memcg_page_state(memcg, memcg1_stats[i]); |
8de7ecc6 | 4119 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], |
7de2e9f1 | 4120 | (u64)nr * PAGE_SIZE); |
af7c4b0e JW |
4121 | } |
4122 | ||
8de7ecc6 | 4123 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d KK |
4124 | seq_printf(m, "total_%s %llu\n", |
4125 | vm_event_name(memcg1_events[i]), | |
dd923990 | 4126 | (u64)memcg_events(memcg, memcg1_events[i])); |
af7c4b0e | 4127 | |
8de7ecc6 | 4128 | for (i = 0; i < NR_LRU_LISTS; i++) |
ebc5d83d | 4129 | seq_printf(m, "total_%s %llu\n", lru_list_name(i), |
42a30035 JW |
4130 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
4131 | PAGE_SIZE); | |
14067bb3 | 4132 | |
7f016ee8 | 4133 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 4134 | { |
ef8f2327 MG |
4135 | pg_data_t *pgdat; |
4136 | struct mem_cgroup_per_node *mz; | |
1431d4d1 JW |
4137 | unsigned long anon_cost = 0; |
4138 | unsigned long file_cost = 0; | |
7f016ee8 | 4139 | |
ef8f2327 MG |
4140 | for_each_online_pgdat(pgdat) { |
4141 | mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); | |
7f016ee8 | 4142 | |
1431d4d1 JW |
4143 | anon_cost += mz->lruvec.anon_cost; |
4144 | file_cost += mz->lruvec.file_cost; | |
ef8f2327 | 4145 | } |
1431d4d1 JW |
4146 | seq_printf(m, "anon_cost %lu\n", anon_cost); |
4147 | seq_printf(m, "file_cost %lu\n", file_cost); | |
7f016ee8 KM |
4148 | } |
4149 | #endif | |
4150 | ||
d2ceb9b7 KH |
4151 | return 0; |
4152 | } | |
4153 | ||
182446d0 TH |
4154 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
4155 | struct cftype *cft) | |
a7885eb8 | 4156 | { |
182446d0 | 4157 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4158 | |
1f4c025b | 4159 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4160 | } |
4161 | ||
182446d0 TH |
4162 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
4163 | struct cftype *cft, u64 val) | |
a7885eb8 | 4164 | { |
182446d0 | 4165 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4166 | |
3dae7fec | 4167 | if (val > 100) |
a7885eb8 KM |
4168 | return -EINVAL; |
4169 | ||
14208b0e | 4170 | if (css->parent) |
3dae7fec JW |
4171 | memcg->swappiness = val; |
4172 | else | |
4173 | vm_swappiness = val; | |
068b38c1 | 4174 | |
a7885eb8 KM |
4175 | return 0; |
4176 | } | |
4177 | ||
2e72b634 KS |
4178 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4179 | { | |
4180 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 4181 | unsigned long usage; |
2e72b634 KS |
4182 | int i; |
4183 | ||
4184 | rcu_read_lock(); | |
4185 | if (!swap) | |
2c488db2 | 4186 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4187 | else |
2c488db2 | 4188 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4189 | |
4190 | if (!t) | |
4191 | goto unlock; | |
4192 | ||
ce00a967 | 4193 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
4194 | |
4195 | /* | |
748dad36 | 4196 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4197 | * If it's not true, a threshold was crossed after last |
4198 | * call of __mem_cgroup_threshold(). | |
4199 | */ | |
5407a562 | 4200 | i = t->current_threshold; |
2e72b634 KS |
4201 | |
4202 | /* | |
4203 | * Iterate backward over array of thresholds starting from | |
4204 | * current_threshold and check if a threshold is crossed. | |
4205 | * If none of thresholds below usage is crossed, we read | |
4206 | * only one element of the array here. | |
4207 | */ | |
4208 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4209 | eventfd_signal(t->entries[i].eventfd, 1); | |
4210 | ||
4211 | /* i = current_threshold + 1 */ | |
4212 | i++; | |
4213 | ||
4214 | /* | |
4215 | * Iterate forward over array of thresholds starting from | |
4216 | * current_threshold+1 and check if a threshold is crossed. | |
4217 | * If none of thresholds above usage is crossed, we read | |
4218 | * only one element of the array here. | |
4219 | */ | |
4220 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4221 | eventfd_signal(t->entries[i].eventfd, 1); | |
4222 | ||
4223 | /* Update current_threshold */ | |
5407a562 | 4224 | t->current_threshold = i - 1; |
2e72b634 KS |
4225 | unlock: |
4226 | rcu_read_unlock(); | |
4227 | } | |
4228 | ||
4229 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4230 | { | |
ad4ca5f4 KS |
4231 | while (memcg) { |
4232 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 4233 | if (do_memsw_account()) |
ad4ca5f4 KS |
4234 | __mem_cgroup_threshold(memcg, true); |
4235 | ||
4236 | memcg = parent_mem_cgroup(memcg); | |
4237 | } | |
2e72b634 KS |
4238 | } |
4239 | ||
4240 | static int compare_thresholds(const void *a, const void *b) | |
4241 | { | |
4242 | const struct mem_cgroup_threshold *_a = a; | |
4243 | const struct mem_cgroup_threshold *_b = b; | |
4244 | ||
2bff24a3 GT |
4245 | if (_a->threshold > _b->threshold) |
4246 | return 1; | |
4247 | ||
4248 | if (_a->threshold < _b->threshold) | |
4249 | return -1; | |
4250 | ||
4251 | return 0; | |
2e72b634 KS |
4252 | } |
4253 | ||
c0ff4b85 | 4254 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4255 | { |
4256 | struct mem_cgroup_eventfd_list *ev; | |
4257 | ||
2bcf2e92 MH |
4258 | spin_lock(&memcg_oom_lock); |
4259 | ||
c0ff4b85 | 4260 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 4261 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
4262 | |
4263 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
4264 | return 0; |
4265 | } | |
4266 | ||
c0ff4b85 | 4267 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4268 | { |
7d74b06f KH |
4269 | struct mem_cgroup *iter; |
4270 | ||
c0ff4b85 | 4271 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4272 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4273 | } |
4274 | ||
59b6f873 | 4275 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4276 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4277 | { |
2c488db2 KS |
4278 | struct mem_cgroup_thresholds *thresholds; |
4279 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
4280 | unsigned long threshold; |
4281 | unsigned long usage; | |
2c488db2 | 4282 | int i, size, ret; |
2e72b634 | 4283 | |
650c5e56 | 4284 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
4285 | if (ret) |
4286 | return ret; | |
4287 | ||
4288 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4289 | |
05b84301 | 4290 | if (type == _MEM) { |
2c488db2 | 4291 | thresholds = &memcg->thresholds; |
ce00a967 | 4292 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4293 | } else if (type == _MEMSWAP) { |
2c488db2 | 4294 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4295 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4296 | } else |
2e72b634 KS |
4297 | BUG(); |
4298 | ||
2e72b634 | 4299 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4300 | if (thresholds->primary) |
2e72b634 KS |
4301 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4302 | ||
2c488db2 | 4303 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4304 | |
4305 | /* Allocate memory for new array of thresholds */ | |
67b8046f | 4306 | new = kmalloc(struct_size(new, entries, size), GFP_KERNEL); |
2c488db2 | 4307 | if (!new) { |
2e72b634 KS |
4308 | ret = -ENOMEM; |
4309 | goto unlock; | |
4310 | } | |
2c488db2 | 4311 | new->size = size; |
2e72b634 KS |
4312 | |
4313 | /* Copy thresholds (if any) to new array */ | |
e90342e6 GS |
4314 | if (thresholds->primary) |
4315 | memcpy(new->entries, thresholds->primary->entries, | |
4316 | flex_array_size(new, entries, size - 1)); | |
2c488db2 | 4317 | |
2e72b634 | 4318 | /* Add new threshold */ |
2c488db2 KS |
4319 | new->entries[size - 1].eventfd = eventfd; |
4320 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4321 | |
4322 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
61e604e6 | 4323 | sort(new->entries, size, sizeof(*new->entries), |
2e72b634 KS |
4324 | compare_thresholds, NULL); |
4325 | ||
4326 | /* Find current threshold */ | |
2c488db2 | 4327 | new->current_threshold = -1; |
2e72b634 | 4328 | for (i = 0; i < size; i++) { |
748dad36 | 4329 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4330 | /* |
2c488db2 KS |
4331 | * new->current_threshold will not be used until |
4332 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4333 | * it here. |
4334 | */ | |
2c488db2 | 4335 | ++new->current_threshold; |
748dad36 SZ |
4336 | } else |
4337 | break; | |
2e72b634 KS |
4338 | } |
4339 | ||
2c488db2 KS |
4340 | /* Free old spare buffer and save old primary buffer as spare */ |
4341 | kfree(thresholds->spare); | |
4342 | thresholds->spare = thresholds->primary; | |
4343 | ||
4344 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4345 | |
907860ed | 4346 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4347 | synchronize_rcu(); |
4348 | ||
2e72b634 KS |
4349 | unlock: |
4350 | mutex_unlock(&memcg->thresholds_lock); | |
4351 | ||
4352 | return ret; | |
4353 | } | |
4354 | ||
59b6f873 | 4355 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4356 | struct eventfd_ctx *eventfd, const char *args) |
4357 | { | |
59b6f873 | 4358 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4359 | } |
4360 | ||
59b6f873 | 4361 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4362 | struct eventfd_ctx *eventfd, const char *args) |
4363 | { | |
59b6f873 | 4364 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4365 | } |
4366 | ||
59b6f873 | 4367 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4368 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4369 | { |
2c488db2 KS |
4370 | struct mem_cgroup_thresholds *thresholds; |
4371 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 4372 | unsigned long usage; |
7d36665a | 4373 | int i, j, size, entries; |
2e72b634 KS |
4374 | |
4375 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4376 | |
4377 | if (type == _MEM) { | |
2c488db2 | 4378 | thresholds = &memcg->thresholds; |
ce00a967 | 4379 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4380 | } else if (type == _MEMSWAP) { |
2c488db2 | 4381 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4382 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4383 | } else |
2e72b634 KS |
4384 | BUG(); |
4385 | ||
371528ca AV |
4386 | if (!thresholds->primary) |
4387 | goto unlock; | |
4388 | ||
2e72b634 KS |
4389 | /* Check if a threshold crossed before removing */ |
4390 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4391 | ||
4392 | /* Calculate new number of threshold */ | |
7d36665a | 4393 | size = entries = 0; |
2c488db2 KS |
4394 | for (i = 0; i < thresholds->primary->size; i++) { |
4395 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 | 4396 | size++; |
7d36665a CX |
4397 | else |
4398 | entries++; | |
2e72b634 KS |
4399 | } |
4400 | ||
2c488db2 | 4401 | new = thresholds->spare; |
907860ed | 4402 | |
7d36665a CX |
4403 | /* If no items related to eventfd have been cleared, nothing to do */ |
4404 | if (!entries) | |
4405 | goto unlock; | |
4406 | ||
2e72b634 KS |
4407 | /* Set thresholds array to NULL if we don't have thresholds */ |
4408 | if (!size) { | |
2c488db2 KS |
4409 | kfree(new); |
4410 | new = NULL; | |
907860ed | 4411 | goto swap_buffers; |
2e72b634 KS |
4412 | } |
4413 | ||
2c488db2 | 4414 | new->size = size; |
2e72b634 KS |
4415 | |
4416 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4417 | new->current_threshold = -1; |
4418 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4419 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4420 | continue; |
4421 | ||
2c488db2 | 4422 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4423 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4424 | /* |
2c488db2 | 4425 | * new->current_threshold will not be used |
2e72b634 KS |
4426 | * until rcu_assign_pointer(), so it's safe to increment |
4427 | * it here. | |
4428 | */ | |
2c488db2 | 4429 | ++new->current_threshold; |
2e72b634 KS |
4430 | } |
4431 | j++; | |
4432 | } | |
4433 | ||
907860ed | 4434 | swap_buffers: |
2c488db2 KS |
4435 | /* Swap primary and spare array */ |
4436 | thresholds->spare = thresholds->primary; | |
8c757763 | 4437 | |
2c488db2 | 4438 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4439 | |
907860ed | 4440 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4441 | synchronize_rcu(); |
6611d8d7 MC |
4442 | |
4443 | /* If all events are unregistered, free the spare array */ | |
4444 | if (!new) { | |
4445 | kfree(thresholds->spare); | |
4446 | thresholds->spare = NULL; | |
4447 | } | |
371528ca | 4448 | unlock: |
2e72b634 | 4449 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4450 | } |
c1e862c1 | 4451 | |
59b6f873 | 4452 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4453 | struct eventfd_ctx *eventfd) |
4454 | { | |
59b6f873 | 4455 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4456 | } |
4457 | ||
59b6f873 | 4458 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4459 | struct eventfd_ctx *eventfd) |
4460 | { | |
59b6f873 | 4461 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4462 | } |
4463 | ||
59b6f873 | 4464 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4465 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4466 | { |
9490ff27 | 4467 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4468 | |
9490ff27 KH |
4469 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4470 | if (!event) | |
4471 | return -ENOMEM; | |
4472 | ||
1af8efe9 | 4473 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4474 | |
4475 | event->eventfd = eventfd; | |
4476 | list_add(&event->list, &memcg->oom_notify); | |
4477 | ||
4478 | /* already in OOM ? */ | |
c2b42d3c | 4479 | if (memcg->under_oom) |
9490ff27 | 4480 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4481 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4482 | |
4483 | return 0; | |
4484 | } | |
4485 | ||
59b6f873 | 4486 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4487 | struct eventfd_ctx *eventfd) |
9490ff27 | 4488 | { |
9490ff27 | 4489 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4490 | |
1af8efe9 | 4491 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4492 | |
c0ff4b85 | 4493 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4494 | if (ev->eventfd == eventfd) { |
4495 | list_del(&ev->list); | |
4496 | kfree(ev); | |
4497 | } | |
4498 | } | |
4499 | ||
1af8efe9 | 4500 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4501 | } |
4502 | ||
2da8ca82 | 4503 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4504 | { |
aa9694bb | 4505 | struct mem_cgroup *memcg = mem_cgroup_from_seq(sf); |
3c11ecf4 | 4506 | |
791badbd | 4507 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 4508 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
4509 | seq_printf(sf, "oom_kill %lu\n", |
4510 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
4511 | return 0; |
4512 | } | |
4513 | ||
182446d0 | 4514 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4515 | struct cftype *cft, u64 val) |
4516 | { | |
182446d0 | 4517 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4518 | |
4519 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 4520 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4521 | return -EINVAL; |
4522 | ||
c0ff4b85 | 4523 | memcg->oom_kill_disable = val; |
4d845ebf | 4524 | if (!val) |
c0ff4b85 | 4525 | memcg_oom_recover(memcg); |
3dae7fec | 4526 | |
3c11ecf4 KH |
4527 | return 0; |
4528 | } | |
4529 | ||
52ebea74 TH |
4530 | #ifdef CONFIG_CGROUP_WRITEBACK |
4531 | ||
3a8e9ac8 TH |
4532 | #include <trace/events/writeback.h> |
4533 | ||
841710aa TH |
4534 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
4535 | { | |
4536 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
4537 | } | |
4538 | ||
4539 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4540 | { | |
4541 | wb_domain_exit(&memcg->cgwb_domain); | |
4542 | } | |
4543 | ||
2529bb3a TH |
4544 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4545 | { | |
4546 | wb_domain_size_changed(&memcg->cgwb_domain); | |
4547 | } | |
4548 | ||
841710aa TH |
4549 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
4550 | { | |
4551 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4552 | ||
4553 | if (!memcg->css.parent) | |
4554 | return NULL; | |
4555 | ||
4556 | return &memcg->cgwb_domain; | |
4557 | } | |
4558 | ||
0b3d6e6f GT |
4559 | /* |
4560 | * idx can be of type enum memcg_stat_item or node_stat_item. | |
4561 | * Keep in sync with memcg_exact_page(). | |
4562 | */ | |
4563 | static unsigned long memcg_exact_page_state(struct mem_cgroup *memcg, int idx) | |
4564 | { | |
871789d4 | 4565 | long x = atomic_long_read(&memcg->vmstats[idx]); |
0b3d6e6f GT |
4566 | int cpu; |
4567 | ||
4568 | for_each_online_cpu(cpu) | |
871789d4 | 4569 | x += per_cpu_ptr(memcg->vmstats_percpu, cpu)->stat[idx]; |
0b3d6e6f GT |
4570 | if (x < 0) |
4571 | x = 0; | |
4572 | return x; | |
4573 | } | |
4574 | ||
c2aa723a TH |
4575 | /** |
4576 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
4577 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
4578 | * @pfilepages: out parameter for number of file pages |
4579 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
4580 | * @pdirty: out parameter for number of dirty pages |
4581 | * @pwriteback: out parameter for number of pages under writeback | |
4582 | * | |
c5edf9cd TH |
4583 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
4584 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
4585 | * is a bit more involved. | |
c2aa723a | 4586 | * |
c5edf9cd TH |
4587 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
4588 | * headroom is calculated as the lowest headroom of itself and the | |
4589 | * ancestors. Note that this doesn't consider the actual amount of | |
4590 | * available memory in the system. The caller should further cap | |
4591 | * *@pheadroom accordingly. | |
c2aa723a | 4592 | */ |
c5edf9cd TH |
4593 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
4594 | unsigned long *pheadroom, unsigned long *pdirty, | |
4595 | unsigned long *pwriteback) | |
c2aa723a TH |
4596 | { |
4597 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4598 | struct mem_cgroup *parent; | |
c2aa723a | 4599 | |
0b3d6e6f | 4600 | *pdirty = memcg_exact_page_state(memcg, NR_FILE_DIRTY); |
c2aa723a | 4601 | |
0b3d6e6f | 4602 | *pwriteback = memcg_exact_page_state(memcg, NR_WRITEBACK); |
21d89d15 JW |
4603 | *pfilepages = memcg_exact_page_state(memcg, NR_INACTIVE_FILE) + |
4604 | memcg_exact_page_state(memcg, NR_ACTIVE_FILE); | |
c5edf9cd | 4605 | *pheadroom = PAGE_COUNTER_MAX; |
c2aa723a | 4606 | |
c2aa723a | 4607 | while ((parent = parent_mem_cgroup(memcg))) { |
15b42562 | 4608 | unsigned long ceiling = min(READ_ONCE(memcg->memory.max), |
d1663a90 | 4609 | READ_ONCE(memcg->memory.high)); |
c2aa723a TH |
4610 | unsigned long used = page_counter_read(&memcg->memory); |
4611 | ||
c5edf9cd | 4612 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
4613 | memcg = parent; |
4614 | } | |
c2aa723a TH |
4615 | } |
4616 | ||
97b27821 TH |
4617 | /* |
4618 | * Foreign dirty flushing | |
4619 | * | |
4620 | * There's an inherent mismatch between memcg and writeback. The former | |
4621 | * trackes ownership per-page while the latter per-inode. This was a | |
4622 | * deliberate design decision because honoring per-page ownership in the | |
4623 | * writeback path is complicated, may lead to higher CPU and IO overheads | |
4624 | * and deemed unnecessary given that write-sharing an inode across | |
4625 | * different cgroups isn't a common use-case. | |
4626 | * | |
4627 | * Combined with inode majority-writer ownership switching, this works well | |
4628 | * enough in most cases but there are some pathological cases. For | |
4629 | * example, let's say there are two cgroups A and B which keep writing to | |
4630 | * different but confined parts of the same inode. B owns the inode and | |
4631 | * A's memory is limited far below B's. A's dirty ratio can rise enough to | |
4632 | * trigger balance_dirty_pages() sleeps but B's can be low enough to avoid | |
4633 | * triggering background writeback. A will be slowed down without a way to | |
4634 | * make writeback of the dirty pages happen. | |
4635 | * | |
4636 | * Conditions like the above can lead to a cgroup getting repatedly and | |
4637 | * severely throttled after making some progress after each | |
4638 | * dirty_expire_interval while the underyling IO device is almost | |
4639 | * completely idle. | |
4640 | * | |
4641 | * Solving this problem completely requires matching the ownership tracking | |
4642 | * granularities between memcg and writeback in either direction. However, | |
4643 | * the more egregious behaviors can be avoided by simply remembering the | |
4644 | * most recent foreign dirtying events and initiating remote flushes on | |
4645 | * them when local writeback isn't enough to keep the memory clean enough. | |
4646 | * | |
4647 | * The following two functions implement such mechanism. When a foreign | |
4648 | * page - a page whose memcg and writeback ownerships don't match - is | |
4649 | * dirtied, mem_cgroup_track_foreign_dirty() records the inode owning | |
4650 | * bdi_writeback on the page owning memcg. When balance_dirty_pages() | |
4651 | * decides that the memcg needs to sleep due to high dirty ratio, it calls | |
4652 | * mem_cgroup_flush_foreign() which queues writeback on the recorded | |
4653 | * foreign bdi_writebacks which haven't expired. Both the numbers of | |
4654 | * recorded bdi_writebacks and concurrent in-flight foreign writebacks are | |
4655 | * limited to MEMCG_CGWB_FRN_CNT. | |
4656 | * | |
4657 | * The mechanism only remembers IDs and doesn't hold any object references. | |
4658 | * As being wrong occasionally doesn't matter, updates and accesses to the | |
4659 | * records are lockless and racy. | |
4660 | */ | |
4661 | void mem_cgroup_track_foreign_dirty_slowpath(struct page *page, | |
4662 | struct bdi_writeback *wb) | |
4663 | { | |
bcfe06bf | 4664 | struct mem_cgroup *memcg = page_memcg(page); |
97b27821 TH |
4665 | struct memcg_cgwb_frn *frn; |
4666 | u64 now = get_jiffies_64(); | |
4667 | u64 oldest_at = now; | |
4668 | int oldest = -1; | |
4669 | int i; | |
4670 | ||
3a8e9ac8 TH |
4671 | trace_track_foreign_dirty(page, wb); |
4672 | ||
97b27821 TH |
4673 | /* |
4674 | * Pick the slot to use. If there is already a slot for @wb, keep | |
4675 | * using it. If not replace the oldest one which isn't being | |
4676 | * written out. | |
4677 | */ | |
4678 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4679 | frn = &memcg->cgwb_frn[i]; | |
4680 | if (frn->bdi_id == wb->bdi->id && | |
4681 | frn->memcg_id == wb->memcg_css->id) | |
4682 | break; | |
4683 | if (time_before64(frn->at, oldest_at) && | |
4684 | atomic_read(&frn->done.cnt) == 1) { | |
4685 | oldest = i; | |
4686 | oldest_at = frn->at; | |
4687 | } | |
4688 | } | |
4689 | ||
4690 | if (i < MEMCG_CGWB_FRN_CNT) { | |
4691 | /* | |
4692 | * Re-using an existing one. Update timestamp lazily to | |
4693 | * avoid making the cacheline hot. We want them to be | |
4694 | * reasonably up-to-date and significantly shorter than | |
4695 | * dirty_expire_interval as that's what expires the record. | |
4696 | * Use the shorter of 1s and dirty_expire_interval / 8. | |
4697 | */ | |
4698 | unsigned long update_intv = | |
4699 | min_t(unsigned long, HZ, | |
4700 | msecs_to_jiffies(dirty_expire_interval * 10) / 8); | |
4701 | ||
4702 | if (time_before64(frn->at, now - update_intv)) | |
4703 | frn->at = now; | |
4704 | } else if (oldest >= 0) { | |
4705 | /* replace the oldest free one */ | |
4706 | frn = &memcg->cgwb_frn[oldest]; | |
4707 | frn->bdi_id = wb->bdi->id; | |
4708 | frn->memcg_id = wb->memcg_css->id; | |
4709 | frn->at = now; | |
4710 | } | |
4711 | } | |
4712 | ||
4713 | /* issue foreign writeback flushes for recorded foreign dirtying events */ | |
4714 | void mem_cgroup_flush_foreign(struct bdi_writeback *wb) | |
4715 | { | |
4716 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4717 | unsigned long intv = msecs_to_jiffies(dirty_expire_interval * 10); | |
4718 | u64 now = jiffies_64; | |
4719 | int i; | |
4720 | ||
4721 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4722 | struct memcg_cgwb_frn *frn = &memcg->cgwb_frn[i]; | |
4723 | ||
4724 | /* | |
4725 | * If the record is older than dirty_expire_interval, | |
4726 | * writeback on it has already started. No need to kick it | |
4727 | * off again. Also, don't start a new one if there's | |
4728 | * already one in flight. | |
4729 | */ | |
4730 | if (time_after64(frn->at, now - intv) && | |
4731 | atomic_read(&frn->done.cnt) == 1) { | |
4732 | frn->at = 0; | |
3a8e9ac8 | 4733 | trace_flush_foreign(wb, frn->bdi_id, frn->memcg_id); |
97b27821 TH |
4734 | cgroup_writeback_by_id(frn->bdi_id, frn->memcg_id, 0, |
4735 | WB_REASON_FOREIGN_FLUSH, | |
4736 | &frn->done); | |
4737 | } | |
4738 | } | |
4739 | } | |
4740 | ||
841710aa TH |
4741 | #else /* CONFIG_CGROUP_WRITEBACK */ |
4742 | ||
4743 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
4744 | { | |
4745 | return 0; | |
4746 | } | |
4747 | ||
4748 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4749 | { | |
4750 | } | |
4751 | ||
2529bb3a TH |
4752 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4753 | { | |
4754 | } | |
4755 | ||
52ebea74 TH |
4756 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
4757 | ||
3bc942f3 TH |
4758 | /* |
4759 | * DO NOT USE IN NEW FILES. | |
4760 | * | |
4761 | * "cgroup.event_control" implementation. | |
4762 | * | |
4763 | * This is way over-engineered. It tries to support fully configurable | |
4764 | * events for each user. Such level of flexibility is completely | |
4765 | * unnecessary especially in the light of the planned unified hierarchy. | |
4766 | * | |
4767 | * Please deprecate this and replace with something simpler if at all | |
4768 | * possible. | |
4769 | */ | |
4770 | ||
79bd9814 TH |
4771 | /* |
4772 | * Unregister event and free resources. | |
4773 | * | |
4774 | * Gets called from workqueue. | |
4775 | */ | |
3bc942f3 | 4776 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4777 | { |
3bc942f3 TH |
4778 | struct mem_cgroup_event *event = |
4779 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4780 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4781 | |
4782 | remove_wait_queue(event->wqh, &event->wait); | |
4783 | ||
59b6f873 | 4784 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4785 | |
4786 | /* Notify userspace the event is going away. */ | |
4787 | eventfd_signal(event->eventfd, 1); | |
4788 | ||
4789 | eventfd_ctx_put(event->eventfd); | |
4790 | kfree(event); | |
59b6f873 | 4791 | css_put(&memcg->css); |
79bd9814 TH |
4792 | } |
4793 | ||
4794 | /* | |
a9a08845 | 4795 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
4796 | * |
4797 | * Called with wqh->lock held and interrupts disabled. | |
4798 | */ | |
ac6424b9 | 4799 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 4800 | int sync, void *key) |
79bd9814 | 4801 | { |
3bc942f3 TH |
4802 | struct mem_cgroup_event *event = |
4803 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4804 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 4805 | __poll_t flags = key_to_poll(key); |
79bd9814 | 4806 | |
a9a08845 | 4807 | if (flags & EPOLLHUP) { |
79bd9814 TH |
4808 | /* |
4809 | * If the event has been detached at cgroup removal, we | |
4810 | * can simply return knowing the other side will cleanup | |
4811 | * for us. | |
4812 | * | |
4813 | * We can't race against event freeing since the other | |
4814 | * side will require wqh->lock via remove_wait_queue(), | |
4815 | * which we hold. | |
4816 | */ | |
fba94807 | 4817 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4818 | if (!list_empty(&event->list)) { |
4819 | list_del_init(&event->list); | |
4820 | /* | |
4821 | * We are in atomic context, but cgroup_event_remove() | |
4822 | * may sleep, so we have to call it in workqueue. | |
4823 | */ | |
4824 | schedule_work(&event->remove); | |
4825 | } | |
fba94807 | 4826 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4827 | } |
4828 | ||
4829 | return 0; | |
4830 | } | |
4831 | ||
3bc942f3 | 4832 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4833 | wait_queue_head_t *wqh, poll_table *pt) |
4834 | { | |
3bc942f3 TH |
4835 | struct mem_cgroup_event *event = |
4836 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4837 | |
4838 | event->wqh = wqh; | |
4839 | add_wait_queue(wqh, &event->wait); | |
4840 | } | |
4841 | ||
4842 | /* | |
3bc942f3 TH |
4843 | * DO NOT USE IN NEW FILES. |
4844 | * | |
79bd9814 TH |
4845 | * Parse input and register new cgroup event handler. |
4846 | * | |
4847 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4848 | * Interpretation of args is defined by control file implementation. | |
4849 | */ | |
451af504 TH |
4850 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4851 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4852 | { |
451af504 | 4853 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4854 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4855 | struct mem_cgroup_event *event; |
79bd9814 TH |
4856 | struct cgroup_subsys_state *cfile_css; |
4857 | unsigned int efd, cfd; | |
4858 | struct fd efile; | |
4859 | struct fd cfile; | |
fba94807 | 4860 | const char *name; |
79bd9814 TH |
4861 | char *endp; |
4862 | int ret; | |
4863 | ||
451af504 TH |
4864 | buf = strstrip(buf); |
4865 | ||
4866 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4867 | if (*endp != ' ') |
4868 | return -EINVAL; | |
451af504 | 4869 | buf = endp + 1; |
79bd9814 | 4870 | |
451af504 | 4871 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4872 | if ((*endp != ' ') && (*endp != '\0')) |
4873 | return -EINVAL; | |
451af504 | 4874 | buf = endp + 1; |
79bd9814 TH |
4875 | |
4876 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4877 | if (!event) | |
4878 | return -ENOMEM; | |
4879 | ||
59b6f873 | 4880 | event->memcg = memcg; |
79bd9814 | 4881 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4882 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4883 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4884 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4885 | |
4886 | efile = fdget(efd); | |
4887 | if (!efile.file) { | |
4888 | ret = -EBADF; | |
4889 | goto out_kfree; | |
4890 | } | |
4891 | ||
4892 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4893 | if (IS_ERR(event->eventfd)) { | |
4894 | ret = PTR_ERR(event->eventfd); | |
4895 | goto out_put_efile; | |
4896 | } | |
4897 | ||
4898 | cfile = fdget(cfd); | |
4899 | if (!cfile.file) { | |
4900 | ret = -EBADF; | |
4901 | goto out_put_eventfd; | |
4902 | } | |
4903 | ||
4904 | /* the process need read permission on control file */ | |
4905 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
02f92b38 | 4906 | ret = file_permission(cfile.file, MAY_READ); |
79bd9814 TH |
4907 | if (ret < 0) |
4908 | goto out_put_cfile; | |
4909 | ||
fba94807 TH |
4910 | /* |
4911 | * Determine the event callbacks and set them in @event. This used | |
4912 | * to be done via struct cftype but cgroup core no longer knows | |
4913 | * about these events. The following is crude but the whole thing | |
4914 | * is for compatibility anyway. | |
3bc942f3 TH |
4915 | * |
4916 | * DO NOT ADD NEW FILES. | |
fba94807 | 4917 | */ |
b583043e | 4918 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4919 | |
4920 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4921 | event->register_event = mem_cgroup_usage_register_event; | |
4922 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4923 | } else if (!strcmp(name, "memory.oom_control")) { | |
4924 | event->register_event = mem_cgroup_oom_register_event; | |
4925 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4926 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4927 | event->register_event = vmpressure_register_event; | |
4928 | event->unregister_event = vmpressure_unregister_event; | |
4929 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4930 | event->register_event = memsw_cgroup_usage_register_event; |
4931 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4932 | } else { |
4933 | ret = -EINVAL; | |
4934 | goto out_put_cfile; | |
4935 | } | |
4936 | ||
79bd9814 | 4937 | /* |
b5557c4c TH |
4938 | * Verify @cfile should belong to @css. Also, remaining events are |
4939 | * automatically removed on cgroup destruction but the removal is | |
4940 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4941 | */ |
b583043e | 4942 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4943 | &memory_cgrp_subsys); |
79bd9814 | 4944 | ret = -EINVAL; |
5a17f543 | 4945 | if (IS_ERR(cfile_css)) |
79bd9814 | 4946 | goto out_put_cfile; |
5a17f543 TH |
4947 | if (cfile_css != css) { |
4948 | css_put(cfile_css); | |
79bd9814 | 4949 | goto out_put_cfile; |
5a17f543 | 4950 | } |
79bd9814 | 4951 | |
451af504 | 4952 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4953 | if (ret) |
4954 | goto out_put_css; | |
4955 | ||
9965ed17 | 4956 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 4957 | |
fba94807 TH |
4958 | spin_lock(&memcg->event_list_lock); |
4959 | list_add(&event->list, &memcg->event_list); | |
4960 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4961 | |
4962 | fdput(cfile); | |
4963 | fdput(efile); | |
4964 | ||
451af504 | 4965 | return nbytes; |
79bd9814 TH |
4966 | |
4967 | out_put_css: | |
b5557c4c | 4968 | css_put(css); |
79bd9814 TH |
4969 | out_put_cfile: |
4970 | fdput(cfile); | |
4971 | out_put_eventfd: | |
4972 | eventfd_ctx_put(event->eventfd); | |
4973 | out_put_efile: | |
4974 | fdput(efile); | |
4975 | out_kfree: | |
4976 | kfree(event); | |
4977 | ||
4978 | return ret; | |
4979 | } | |
4980 | ||
241994ed | 4981 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4982 | { |
0eea1030 | 4983 | .name = "usage_in_bytes", |
8c7c6e34 | 4984 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4985 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4986 | }, |
c84872e1 PE |
4987 | { |
4988 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4989 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4990 | .write = mem_cgroup_reset, |
791badbd | 4991 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4992 | }, |
8cdea7c0 | 4993 | { |
0eea1030 | 4994 | .name = "limit_in_bytes", |
8c7c6e34 | 4995 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4996 | .write = mem_cgroup_write, |
791badbd | 4997 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4998 | }, |
296c81d8 BS |
4999 | { |
5000 | .name = "soft_limit_in_bytes", | |
5001 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 5002 | .write = mem_cgroup_write, |
791badbd | 5003 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 5004 | }, |
8cdea7c0 BS |
5005 | { |
5006 | .name = "failcnt", | |
8c7c6e34 | 5007 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 5008 | .write = mem_cgroup_reset, |
791badbd | 5009 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5010 | }, |
d2ceb9b7 KH |
5011 | { |
5012 | .name = "stat", | |
2da8ca82 | 5013 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 5014 | }, |
c1e862c1 KH |
5015 | { |
5016 | .name = "force_empty", | |
6770c64e | 5017 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 5018 | }, |
18f59ea7 BS |
5019 | { |
5020 | .name = "use_hierarchy", | |
5021 | .write_u64 = mem_cgroup_hierarchy_write, | |
5022 | .read_u64 = mem_cgroup_hierarchy_read, | |
5023 | }, | |
79bd9814 | 5024 | { |
3bc942f3 | 5025 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 5026 | .write = memcg_write_event_control, |
7dbdb199 | 5027 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 5028 | }, |
a7885eb8 KM |
5029 | { |
5030 | .name = "swappiness", | |
5031 | .read_u64 = mem_cgroup_swappiness_read, | |
5032 | .write_u64 = mem_cgroup_swappiness_write, | |
5033 | }, | |
7dc74be0 DN |
5034 | { |
5035 | .name = "move_charge_at_immigrate", | |
5036 | .read_u64 = mem_cgroup_move_charge_read, | |
5037 | .write_u64 = mem_cgroup_move_charge_write, | |
5038 | }, | |
9490ff27 KH |
5039 | { |
5040 | .name = "oom_control", | |
2da8ca82 | 5041 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 5042 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
5043 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
5044 | }, | |
70ddf637 AV |
5045 | { |
5046 | .name = "pressure_level", | |
70ddf637 | 5047 | }, |
406eb0c9 YH |
5048 | #ifdef CONFIG_NUMA |
5049 | { | |
5050 | .name = "numa_stat", | |
2da8ca82 | 5051 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
5052 | }, |
5053 | #endif | |
510fc4e1 GC |
5054 | { |
5055 | .name = "kmem.limit_in_bytes", | |
5056 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 5057 | .write = mem_cgroup_write, |
791badbd | 5058 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5059 | }, |
5060 | { | |
5061 | .name = "kmem.usage_in_bytes", | |
5062 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 5063 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5064 | }, |
5065 | { | |
5066 | .name = "kmem.failcnt", | |
5067 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 5068 | .write = mem_cgroup_reset, |
791badbd | 5069 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5070 | }, |
5071 | { | |
5072 | .name = "kmem.max_usage_in_bytes", | |
5073 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 5074 | .write = mem_cgroup_reset, |
791badbd | 5075 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 5076 | }, |
a87425a3 YS |
5077 | #if defined(CONFIG_MEMCG_KMEM) && \ |
5078 | (defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)) | |
749c5415 GC |
5079 | { |
5080 | .name = "kmem.slabinfo", | |
b047501c | 5081 | .seq_show = memcg_slab_show, |
749c5415 GC |
5082 | }, |
5083 | #endif | |
d55f90bf VD |
5084 | { |
5085 | .name = "kmem.tcp.limit_in_bytes", | |
5086 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
5087 | .write = mem_cgroup_write, | |
5088 | .read_u64 = mem_cgroup_read_u64, | |
5089 | }, | |
5090 | { | |
5091 | .name = "kmem.tcp.usage_in_bytes", | |
5092 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
5093 | .read_u64 = mem_cgroup_read_u64, | |
5094 | }, | |
5095 | { | |
5096 | .name = "kmem.tcp.failcnt", | |
5097 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
5098 | .write = mem_cgroup_reset, | |
5099 | .read_u64 = mem_cgroup_read_u64, | |
5100 | }, | |
5101 | { | |
5102 | .name = "kmem.tcp.max_usage_in_bytes", | |
5103 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
5104 | .write = mem_cgroup_reset, | |
5105 | .read_u64 = mem_cgroup_read_u64, | |
5106 | }, | |
6bc10349 | 5107 | { }, /* terminate */ |
af36f906 | 5108 | }; |
8c7c6e34 | 5109 | |
73f576c0 JW |
5110 | /* |
5111 | * Private memory cgroup IDR | |
5112 | * | |
5113 | * Swap-out records and page cache shadow entries need to store memcg | |
5114 | * references in constrained space, so we maintain an ID space that is | |
5115 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
5116 | * memory-controlled cgroups to 64k. | |
5117 | * | |
b8f2935f | 5118 | * However, there usually are many references to the offline CSS after |
73f576c0 JW |
5119 | * the cgroup has been destroyed, such as page cache or reclaimable |
5120 | * slab objects, that don't need to hang on to the ID. We want to keep | |
5121 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
5122 | * relatively small ID space and prevent the creation of new cgroups | |
5123 | * even when there are much fewer than 64k cgroups - possibly none. | |
5124 | * | |
5125 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
5126 | * be freed and recycled when it's no longer needed, which is usually | |
5127 | * when the CSS is offlined. | |
5128 | * | |
5129 | * The only exception to that are records of swapped out tmpfs/shmem | |
5130 | * pages that need to be attributed to live ancestors on swapin. But | |
5131 | * those references are manageable from userspace. | |
5132 | */ | |
5133 | ||
5134 | static DEFINE_IDR(mem_cgroup_idr); | |
5135 | ||
7e97de0b KT |
5136 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
5137 | { | |
5138 | if (memcg->id.id > 0) { | |
5139 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
5140 | memcg->id.id = 0; | |
5141 | } | |
5142 | } | |
5143 | ||
c1514c0a VF |
5144 | static void __maybe_unused mem_cgroup_id_get_many(struct mem_cgroup *memcg, |
5145 | unsigned int n) | |
73f576c0 | 5146 | { |
1c2d479a | 5147 | refcount_add(n, &memcg->id.ref); |
73f576c0 JW |
5148 | } |
5149 | ||
615d66c3 | 5150 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 5151 | { |
1c2d479a | 5152 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 5153 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
5154 | |
5155 | /* Memcg ID pins CSS */ | |
5156 | css_put(&memcg->css); | |
5157 | } | |
5158 | } | |
5159 | ||
615d66c3 VD |
5160 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) |
5161 | { | |
5162 | mem_cgroup_id_put_many(memcg, 1); | |
5163 | } | |
5164 | ||
73f576c0 JW |
5165 | /** |
5166 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
5167 | * @id: the memcg id to look up | |
5168 | * | |
5169 | * Caller must hold rcu_read_lock(). | |
5170 | */ | |
5171 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
5172 | { | |
5173 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5174 | return idr_find(&mem_cgroup_idr, id); | |
5175 | } | |
5176 | ||
ef8f2327 | 5177 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
5178 | { |
5179 | struct mem_cgroup_per_node *pn; | |
ef8f2327 | 5180 | int tmp = node; |
1ecaab2b KH |
5181 | /* |
5182 | * This routine is called against possible nodes. | |
5183 | * But it's BUG to call kmalloc() against offline node. | |
5184 | * | |
5185 | * TODO: this routine can waste much memory for nodes which will | |
5186 | * never be onlined. It's better to use memory hotplug callback | |
5187 | * function. | |
5188 | */ | |
41e3355d KH |
5189 | if (!node_state(node, N_NORMAL_MEMORY)) |
5190 | tmp = -1; | |
17295c88 | 5191 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
5192 | if (!pn) |
5193 | return 1; | |
1ecaab2b | 5194 | |
3e38e0aa RG |
5195 | pn->lruvec_stat_local = alloc_percpu_gfp(struct lruvec_stat, |
5196 | GFP_KERNEL_ACCOUNT); | |
815744d7 JW |
5197 | if (!pn->lruvec_stat_local) { |
5198 | kfree(pn); | |
5199 | return 1; | |
5200 | } | |
5201 | ||
f3344adf | 5202 | pn->lruvec_stat_cpu = alloc_percpu_gfp(struct batched_lruvec_stat, |
3e38e0aa | 5203 | GFP_KERNEL_ACCOUNT); |
a983b5eb | 5204 | if (!pn->lruvec_stat_cpu) { |
815744d7 | 5205 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c JW |
5206 | kfree(pn); |
5207 | return 1; | |
5208 | } | |
5209 | ||
ef8f2327 MG |
5210 | lruvec_init(&pn->lruvec); |
5211 | pn->usage_in_excess = 0; | |
5212 | pn->on_tree = false; | |
5213 | pn->memcg = memcg; | |
5214 | ||
54f72fe0 | 5215 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
5216 | return 0; |
5217 | } | |
5218 | ||
ef8f2327 | 5219 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 5220 | { |
00f3ca2c JW |
5221 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
5222 | ||
4eaf431f MH |
5223 | if (!pn) |
5224 | return; | |
5225 | ||
a983b5eb | 5226 | free_percpu(pn->lruvec_stat_cpu); |
815744d7 | 5227 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c | 5228 | kfree(pn); |
1ecaab2b KH |
5229 | } |
5230 | ||
40e952f9 | 5231 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 5232 | { |
c8b2a36f | 5233 | int node; |
59927fb9 | 5234 | |
c8b2a36f | 5235 | for_each_node(node) |
ef8f2327 | 5236 | free_mem_cgroup_per_node_info(memcg, node); |
871789d4 | 5237 | free_percpu(memcg->vmstats_percpu); |
815744d7 | 5238 | free_percpu(memcg->vmstats_local); |
8ff69e2c | 5239 | kfree(memcg); |
59927fb9 | 5240 | } |
3afe36b1 | 5241 | |
40e952f9 TE |
5242 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
5243 | { | |
5244 | memcg_wb_domain_exit(memcg); | |
7961eee3 SB |
5245 | /* |
5246 | * Flush percpu vmstats and vmevents to guarantee the value correctness | |
5247 | * on parent's and all ancestor levels. | |
5248 | */ | |
4a87e2a2 | 5249 | memcg_flush_percpu_vmstats(memcg); |
7961eee3 | 5250 | memcg_flush_percpu_vmevents(memcg); |
40e952f9 TE |
5251 | __mem_cgroup_free(memcg); |
5252 | } | |
5253 | ||
0b8f73e1 | 5254 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 5255 | { |
d142e3e6 | 5256 | struct mem_cgroup *memcg; |
b9726c26 | 5257 | unsigned int size; |
6d12e2d8 | 5258 | int node; |
97b27821 | 5259 | int __maybe_unused i; |
11d67612 | 5260 | long error = -ENOMEM; |
8cdea7c0 | 5261 | |
0b8f73e1 JW |
5262 | size = sizeof(struct mem_cgroup); |
5263 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
5264 | ||
5265 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 5266 | if (!memcg) |
11d67612 | 5267 | return ERR_PTR(error); |
0b8f73e1 | 5268 | |
73f576c0 JW |
5269 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
5270 | 1, MEM_CGROUP_ID_MAX, | |
5271 | GFP_KERNEL); | |
11d67612 YS |
5272 | if (memcg->id.id < 0) { |
5273 | error = memcg->id.id; | |
73f576c0 | 5274 | goto fail; |
11d67612 | 5275 | } |
73f576c0 | 5276 | |
3e38e0aa RG |
5277 | memcg->vmstats_local = alloc_percpu_gfp(struct memcg_vmstats_percpu, |
5278 | GFP_KERNEL_ACCOUNT); | |
815744d7 JW |
5279 | if (!memcg->vmstats_local) |
5280 | goto fail; | |
5281 | ||
3e38e0aa RG |
5282 | memcg->vmstats_percpu = alloc_percpu_gfp(struct memcg_vmstats_percpu, |
5283 | GFP_KERNEL_ACCOUNT); | |
871789d4 | 5284 | if (!memcg->vmstats_percpu) |
0b8f73e1 | 5285 | goto fail; |
78fb7466 | 5286 | |
3ed28fa1 | 5287 | for_each_node(node) |
ef8f2327 | 5288 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 5289 | goto fail; |
f64c3f54 | 5290 | |
0b8f73e1 JW |
5291 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
5292 | goto fail; | |
28dbc4b6 | 5293 | |
f7e1cb6e | 5294 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 | 5295 | INIT_LIST_HEAD(&memcg->oom_notify); |
d142e3e6 GC |
5296 | mutex_init(&memcg->thresholds_lock); |
5297 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 5298 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
5299 | INIT_LIST_HEAD(&memcg->event_list); |
5300 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 5301 | memcg->socket_pressure = jiffies; |
84c07d11 | 5302 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 5303 | memcg->kmemcg_id = -1; |
bf4f0599 | 5304 | INIT_LIST_HEAD(&memcg->objcg_list); |
900a38f0 | 5305 | #endif |
52ebea74 TH |
5306 | #ifdef CONFIG_CGROUP_WRITEBACK |
5307 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
97b27821 TH |
5308 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) |
5309 | memcg->cgwb_frn[i].done = | |
5310 | __WB_COMPLETION_INIT(&memcg_cgwb_frn_waitq); | |
87eaceb3 YS |
5311 | #endif |
5312 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5313 | spin_lock_init(&memcg->deferred_split_queue.split_queue_lock); | |
5314 | INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue); | |
5315 | memcg->deferred_split_queue.split_queue_len = 0; | |
52ebea74 | 5316 | #endif |
73f576c0 | 5317 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
5318 | return memcg; |
5319 | fail: | |
7e97de0b | 5320 | mem_cgroup_id_remove(memcg); |
40e952f9 | 5321 | __mem_cgroup_free(memcg); |
11d67612 | 5322 | return ERR_PTR(error); |
d142e3e6 GC |
5323 | } |
5324 | ||
0b8f73e1 JW |
5325 | static struct cgroup_subsys_state * __ref |
5326 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 5327 | { |
0b8f73e1 | 5328 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
b87d8cef | 5329 | struct mem_cgroup *memcg, *old_memcg; |
0b8f73e1 | 5330 | long error = -ENOMEM; |
d142e3e6 | 5331 | |
b87d8cef | 5332 | old_memcg = set_active_memcg(parent); |
0b8f73e1 | 5333 | memcg = mem_cgroup_alloc(); |
b87d8cef | 5334 | set_active_memcg(old_memcg); |
11d67612 YS |
5335 | if (IS_ERR(memcg)) |
5336 | return ERR_CAST(memcg); | |
d142e3e6 | 5337 | |
d1663a90 | 5338 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
0b8f73e1 | 5339 | memcg->soft_limit = PAGE_COUNTER_MAX; |
4b82ab4f | 5340 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
0b8f73e1 JW |
5341 | if (parent) { |
5342 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
5343 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
bef8620c | 5344 | |
3e32cb2e | 5345 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 5346 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e | 5347 | page_counter_init(&memcg->kmem, &parent->kmem); |
0db15298 | 5348 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 5349 | } else { |
bef8620c RG |
5350 | page_counter_init(&memcg->memory, NULL); |
5351 | page_counter_init(&memcg->swap, NULL); | |
5352 | page_counter_init(&memcg->kmem, NULL); | |
5353 | page_counter_init(&memcg->tcpmem, NULL); | |
d6441637 | 5354 | |
0b8f73e1 JW |
5355 | root_mem_cgroup = memcg; |
5356 | return &memcg->css; | |
5357 | } | |
5358 | ||
bef8620c | 5359 | /* The following stuff does not apply to the root */ |
b313aeee | 5360 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
5361 | if (error) |
5362 | goto fail; | |
127424c8 | 5363 | |
f7e1cb6e | 5364 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5365 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 5366 | |
0b8f73e1 JW |
5367 | return &memcg->css; |
5368 | fail: | |
7e97de0b | 5369 | mem_cgroup_id_remove(memcg); |
0b8f73e1 | 5370 | mem_cgroup_free(memcg); |
11d67612 | 5371 | return ERR_PTR(error); |
0b8f73e1 JW |
5372 | } |
5373 | ||
73f576c0 | 5374 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 5375 | { |
58fa2a55 VD |
5376 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5377 | ||
0a4465d3 KT |
5378 | /* |
5379 | * A memcg must be visible for memcg_expand_shrinker_maps() | |
5380 | * by the time the maps are allocated. So, we allocate maps | |
5381 | * here, when for_each_mem_cgroup() can't skip it. | |
5382 | */ | |
5383 | if (memcg_alloc_shrinker_maps(memcg)) { | |
5384 | mem_cgroup_id_remove(memcg); | |
5385 | return -ENOMEM; | |
5386 | } | |
5387 | ||
73f576c0 | 5388 | /* Online state pins memcg ID, memcg ID pins CSS */ |
1c2d479a | 5389 | refcount_set(&memcg->id.ref, 1); |
73f576c0 | 5390 | css_get(css); |
2f7dd7a4 | 5391 | return 0; |
8cdea7c0 BS |
5392 | } |
5393 | ||
eb95419b | 5394 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 5395 | { |
eb95419b | 5396 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5397 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
5398 | |
5399 | /* | |
5400 | * Unregister events and notify userspace. | |
5401 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
5402 | * directory to avoid race between userspace and kernelspace. | |
5403 | */ | |
fba94807 TH |
5404 | spin_lock(&memcg->event_list_lock); |
5405 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
5406 | list_del_init(&event->list); |
5407 | schedule_work(&event->remove); | |
5408 | } | |
fba94807 | 5409 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 5410 | |
bf8d5d52 | 5411 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5412 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 5413 | |
567e9ab2 | 5414 | memcg_offline_kmem(memcg); |
52ebea74 | 5415 | wb_memcg_offline(memcg); |
73f576c0 | 5416 | |
591edfb1 RG |
5417 | drain_all_stock(memcg); |
5418 | ||
73f576c0 | 5419 | mem_cgroup_id_put(memcg); |
df878fb0 KH |
5420 | } |
5421 | ||
6df38689 VD |
5422 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
5423 | { | |
5424 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5425 | ||
5426 | invalidate_reclaim_iterators(memcg); | |
5427 | } | |
5428 | ||
eb95419b | 5429 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 5430 | { |
eb95419b | 5431 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
97b27821 | 5432 | int __maybe_unused i; |
c268e994 | 5433 | |
97b27821 TH |
5434 | #ifdef CONFIG_CGROUP_WRITEBACK |
5435 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) | |
5436 | wb_wait_for_completion(&memcg->cgwb_frn[i].done); | |
5437 | #endif | |
f7e1cb6e | 5438 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5439 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 5440 | |
0db15298 | 5441 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 5442 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 5443 | |
0b8f73e1 JW |
5444 | vmpressure_cleanup(&memcg->vmpressure); |
5445 | cancel_work_sync(&memcg->high_work); | |
5446 | mem_cgroup_remove_from_trees(memcg); | |
0a4465d3 | 5447 | memcg_free_shrinker_maps(memcg); |
d886f4e4 | 5448 | memcg_free_kmem(memcg); |
0b8f73e1 | 5449 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
5450 | } |
5451 | ||
1ced953b TH |
5452 | /** |
5453 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
5454 | * @css: the target css | |
5455 | * | |
5456 | * Reset the states of the mem_cgroup associated with @css. This is | |
5457 | * invoked when the userland requests disabling on the default hierarchy | |
5458 | * but the memcg is pinned through dependency. The memcg should stop | |
5459 | * applying policies and should revert to the vanilla state as it may be | |
5460 | * made visible again. | |
5461 | * | |
5462 | * The current implementation only resets the essential configurations. | |
5463 | * This needs to be expanded to cover all the visible parts. | |
5464 | */ | |
5465 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
5466 | { | |
5467 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5468 | ||
bbec2e15 RG |
5469 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
5470 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
bbec2e15 RG |
5471 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); |
5472 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 5473 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5474 | page_counter_set_low(&memcg->memory, 0); |
d1663a90 | 5475 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
24d404dc | 5476 | memcg->soft_limit = PAGE_COUNTER_MAX; |
4b82ab4f | 5477 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
2529bb3a | 5478 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
5479 | } |
5480 | ||
02491447 | 5481 | #ifdef CONFIG_MMU |
7dc74be0 | 5482 | /* Handlers for move charge at task migration. */ |
854ffa8d | 5483 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 5484 | { |
05b84301 | 5485 | int ret; |
9476db97 | 5486 | |
d0164adc MG |
5487 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
5488 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 5489 | if (!ret) { |
854ffa8d | 5490 | mc.precharge += count; |
854ffa8d DN |
5491 | return ret; |
5492 | } | |
9476db97 | 5493 | |
3674534b | 5494 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 5495 | while (count--) { |
3674534b | 5496 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 5497 | if (ret) |
38c5d72f | 5498 | return ret; |
854ffa8d | 5499 | mc.precharge++; |
9476db97 | 5500 | cond_resched(); |
854ffa8d | 5501 | } |
9476db97 | 5502 | return 0; |
4ffef5fe DN |
5503 | } |
5504 | ||
4ffef5fe DN |
5505 | union mc_target { |
5506 | struct page *page; | |
02491447 | 5507 | swp_entry_t ent; |
4ffef5fe DN |
5508 | }; |
5509 | ||
4ffef5fe | 5510 | enum mc_target_type { |
8d32ff84 | 5511 | MC_TARGET_NONE = 0, |
4ffef5fe | 5512 | MC_TARGET_PAGE, |
02491447 | 5513 | MC_TARGET_SWAP, |
c733a828 | 5514 | MC_TARGET_DEVICE, |
4ffef5fe DN |
5515 | }; |
5516 | ||
90254a65 DN |
5517 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5518 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5519 | { |
25b2995a | 5520 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5521 | |
90254a65 DN |
5522 | if (!page || !page_mapped(page)) |
5523 | return NULL; | |
5524 | if (PageAnon(page)) { | |
1dfab5ab | 5525 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5526 | return NULL; |
1dfab5ab JW |
5527 | } else { |
5528 | if (!(mc.flags & MOVE_FILE)) | |
5529 | return NULL; | |
5530 | } | |
90254a65 DN |
5531 | if (!get_page_unless_zero(page)) |
5532 | return NULL; | |
5533 | ||
5534 | return page; | |
5535 | } | |
5536 | ||
c733a828 | 5537 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 5538 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 5539 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 5540 | { |
90254a65 DN |
5541 | struct page *page = NULL; |
5542 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5543 | ||
9a137153 | 5544 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5545 | return NULL; |
c733a828 JG |
5546 | |
5547 | /* | |
5548 | * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to | |
5549 | * a device and because they are not accessible by CPU they are store | |
5550 | * as special swap entry in the CPU page table. | |
5551 | */ | |
5552 | if (is_device_private_entry(ent)) { | |
5553 | page = device_private_entry_to_page(ent); | |
5554 | /* | |
5555 | * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have | |
5556 | * a refcount of 1 when free (unlike normal page) | |
5557 | */ | |
5558 | if (!page_ref_add_unless(page, 1, 1)) | |
5559 | return NULL; | |
5560 | return page; | |
5561 | } | |
5562 | ||
9a137153 RC |
5563 | if (non_swap_entry(ent)) |
5564 | return NULL; | |
5565 | ||
4b91355e KH |
5566 | /* |
5567 | * Because lookup_swap_cache() updates some statistics counter, | |
5568 | * we call find_get_page() with swapper_space directly. | |
5569 | */ | |
f6ab1f7f | 5570 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
2d1c4980 | 5571 | entry->val = ent.val; |
90254a65 DN |
5572 | |
5573 | return page; | |
5574 | } | |
4b91355e KH |
5575 | #else |
5576 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 5577 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
5578 | { |
5579 | return NULL; | |
5580 | } | |
5581 | #endif | |
90254a65 | 5582 | |
87946a72 DN |
5583 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5584 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5585 | { | |
87946a72 DN |
5586 | if (!vma->vm_file) /* anonymous vma */ |
5587 | return NULL; | |
1dfab5ab | 5588 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
5589 | return NULL; |
5590 | ||
87946a72 | 5591 | /* page is moved even if it's not RSS of this task(page-faulted). */ |
aa3b1895 | 5592 | /* shmem/tmpfs may report page out on swap: account for that too. */ |
f5df8635 MWO |
5593 | return find_get_incore_page(vma->vm_file->f_mapping, |
5594 | linear_page_index(vma, addr)); | |
87946a72 DN |
5595 | } |
5596 | ||
b1b0deab CG |
5597 | /** |
5598 | * mem_cgroup_move_account - move account of the page | |
5599 | * @page: the page | |
25843c2b | 5600 | * @compound: charge the page as compound or small page |
b1b0deab CG |
5601 | * @from: mem_cgroup which the page is moved from. |
5602 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
5603 | * | |
3ac808fd | 5604 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
5605 | * |
5606 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
5607 | * from old cgroup. | |
5608 | */ | |
5609 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 5610 | bool compound, |
b1b0deab CG |
5611 | struct mem_cgroup *from, |
5612 | struct mem_cgroup *to) | |
5613 | { | |
ae8af438 KK |
5614 | struct lruvec *from_vec, *to_vec; |
5615 | struct pglist_data *pgdat; | |
6c357848 | 5616 | unsigned int nr_pages = compound ? thp_nr_pages(page) : 1; |
b1b0deab CG |
5617 | int ret; |
5618 | ||
5619 | VM_BUG_ON(from == to); | |
5620 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 5621 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
5622 | |
5623 | /* | |
6a93ca8f | 5624 | * Prevent mem_cgroup_migrate() from looking at |
bcfe06bf | 5625 | * page's memory cgroup of its source page while we change it. |
b1b0deab | 5626 | */ |
f627c2f5 | 5627 | ret = -EBUSY; |
b1b0deab CG |
5628 | if (!trylock_page(page)) |
5629 | goto out; | |
5630 | ||
5631 | ret = -EINVAL; | |
bcfe06bf | 5632 | if (page_memcg(page) != from) |
b1b0deab CG |
5633 | goto out_unlock; |
5634 | ||
ae8af438 | 5635 | pgdat = page_pgdat(page); |
867e5e1d JW |
5636 | from_vec = mem_cgroup_lruvec(from, pgdat); |
5637 | to_vec = mem_cgroup_lruvec(to, pgdat); | |
ae8af438 | 5638 | |
abb242f5 | 5639 | lock_page_memcg(page); |
b1b0deab | 5640 | |
be5d0a74 JW |
5641 | if (PageAnon(page)) { |
5642 | if (page_mapped(page)) { | |
5643 | __mod_lruvec_state(from_vec, NR_ANON_MAPPED, -nr_pages); | |
5644 | __mod_lruvec_state(to_vec, NR_ANON_MAPPED, nr_pages); | |
468c3982 | 5645 | if (PageTransHuge(page)) { |
69473e5d MS |
5646 | __mod_lruvec_state(from_vec, NR_ANON_THPS, |
5647 | -nr_pages); | |
5648 | __mod_lruvec_state(to_vec, NR_ANON_THPS, | |
5649 | nr_pages); | |
468c3982 | 5650 | } |
be5d0a74 JW |
5651 | } |
5652 | } else { | |
0d1c2072 JW |
5653 | __mod_lruvec_state(from_vec, NR_FILE_PAGES, -nr_pages); |
5654 | __mod_lruvec_state(to_vec, NR_FILE_PAGES, nr_pages); | |
5655 | ||
5656 | if (PageSwapBacked(page)) { | |
5657 | __mod_lruvec_state(from_vec, NR_SHMEM, -nr_pages); | |
5658 | __mod_lruvec_state(to_vec, NR_SHMEM, nr_pages); | |
5659 | } | |
5660 | ||
49e50d27 JW |
5661 | if (page_mapped(page)) { |
5662 | __mod_lruvec_state(from_vec, NR_FILE_MAPPED, -nr_pages); | |
5663 | __mod_lruvec_state(to_vec, NR_FILE_MAPPED, nr_pages); | |
5664 | } | |
b1b0deab | 5665 | |
49e50d27 JW |
5666 | if (PageDirty(page)) { |
5667 | struct address_space *mapping = page_mapping(page); | |
c4843a75 | 5668 | |
f56753ac | 5669 | if (mapping_can_writeback(mapping)) { |
49e50d27 JW |
5670 | __mod_lruvec_state(from_vec, NR_FILE_DIRTY, |
5671 | -nr_pages); | |
5672 | __mod_lruvec_state(to_vec, NR_FILE_DIRTY, | |
5673 | nr_pages); | |
5674 | } | |
c4843a75 GT |
5675 | } |
5676 | } | |
5677 | ||
b1b0deab | 5678 | if (PageWriteback(page)) { |
ae8af438 KK |
5679 | __mod_lruvec_state(from_vec, NR_WRITEBACK, -nr_pages); |
5680 | __mod_lruvec_state(to_vec, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
5681 | } |
5682 | ||
5683 | /* | |
abb242f5 JW |
5684 | * All state has been migrated, let's switch to the new memcg. |
5685 | * | |
bcfe06bf | 5686 | * It is safe to change page's memcg here because the page |
abb242f5 JW |
5687 | * is referenced, charged, isolated, and locked: we can't race |
5688 | * with (un)charging, migration, LRU putback, or anything else | |
bcfe06bf | 5689 | * that would rely on a stable page's memory cgroup. |
abb242f5 JW |
5690 | * |
5691 | * Note that lock_page_memcg is a memcg lock, not a page lock, | |
bcfe06bf | 5692 | * to save space. As soon as we switch page's memory cgroup to a |
abb242f5 JW |
5693 | * new memcg that isn't locked, the above state can change |
5694 | * concurrently again. Make sure we're truly done with it. | |
b1b0deab | 5695 | */ |
abb242f5 | 5696 | smp_mb(); |
b1b0deab | 5697 | |
1a3e1f40 JW |
5698 | css_get(&to->css); |
5699 | css_put(&from->css); | |
5700 | ||
bcfe06bf | 5701 | page->memcg_data = (unsigned long)to; |
87eaceb3 | 5702 | |
abb242f5 | 5703 | __unlock_page_memcg(from); |
b1b0deab CG |
5704 | |
5705 | ret = 0; | |
5706 | ||
5707 | local_irq_disable(); | |
3fba69a5 | 5708 | mem_cgroup_charge_statistics(to, page, nr_pages); |
b1b0deab | 5709 | memcg_check_events(to, page); |
3fba69a5 | 5710 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
b1b0deab CG |
5711 | memcg_check_events(from, page); |
5712 | local_irq_enable(); | |
5713 | out_unlock: | |
5714 | unlock_page(page); | |
5715 | out: | |
5716 | return ret; | |
5717 | } | |
5718 | ||
7cf7806c LR |
5719 | /** |
5720 | * get_mctgt_type - get target type of moving charge | |
5721 | * @vma: the vma the pte to be checked belongs | |
5722 | * @addr: the address corresponding to the pte to be checked | |
5723 | * @ptent: the pte to be checked | |
5724 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
5725 | * | |
5726 | * Returns | |
5727 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5728 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5729 | * move charge. if @target is not NULL, the page is stored in target->page | |
5730 | * with extra refcnt got(Callers should handle it). | |
5731 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
5732 | * target for charge migration. if @target is not NULL, the entry is stored | |
5733 | * in target->ent. | |
25b2995a CH |
5734 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PRIVATE |
5735 | * (so ZONE_DEVICE page and thus not on the lru). | |
df6ad698 JG |
5736 | * For now we such page is charge like a regular page would be as for all |
5737 | * intent and purposes it is just special memory taking the place of a | |
5738 | * regular page. | |
c733a828 JG |
5739 | * |
5740 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
5741 | * |
5742 | * Called with pte lock held. | |
5743 | */ | |
5744 | ||
8d32ff84 | 5745 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5746 | unsigned long addr, pte_t ptent, union mc_target *target) |
5747 | { | |
5748 | struct page *page = NULL; | |
8d32ff84 | 5749 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5750 | swp_entry_t ent = { .val = 0 }; |
5751 | ||
5752 | if (pte_present(ptent)) | |
5753 | page = mc_handle_present_pte(vma, addr, ptent); | |
5754 | else if (is_swap_pte(ptent)) | |
48406ef8 | 5755 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 5756 | else if (pte_none(ptent)) |
87946a72 | 5757 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
5758 | |
5759 | if (!page && !ent.val) | |
8d32ff84 | 5760 | return ret; |
02491447 | 5761 | if (page) { |
02491447 | 5762 | /* |
0a31bc97 | 5763 | * Do only loose check w/o serialization. |
1306a85a | 5764 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 5765 | * not under LRU exclusion. |
02491447 | 5766 | */ |
bcfe06bf | 5767 | if (page_memcg(page) == mc.from) { |
02491447 | 5768 | ret = MC_TARGET_PAGE; |
25b2995a | 5769 | if (is_device_private_page(page)) |
c733a828 | 5770 | ret = MC_TARGET_DEVICE; |
02491447 DN |
5771 | if (target) |
5772 | target->page = page; | |
5773 | } | |
5774 | if (!ret || !target) | |
5775 | put_page(page); | |
5776 | } | |
3e14a57b HY |
5777 | /* |
5778 | * There is a swap entry and a page doesn't exist or isn't charged. | |
5779 | * But we cannot move a tail-page in a THP. | |
5780 | */ | |
5781 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 5782 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5783 | ret = MC_TARGET_SWAP; |
5784 | if (target) | |
5785 | target->ent = ent; | |
4ffef5fe | 5786 | } |
4ffef5fe DN |
5787 | return ret; |
5788 | } | |
5789 | ||
12724850 NH |
5790 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5791 | /* | |
d6810d73 HY |
5792 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
5793 | * not support them for now. | |
12724850 NH |
5794 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
5795 | */ | |
5796 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5797 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5798 | { | |
5799 | struct page *page = NULL; | |
12724850 NH |
5800 | enum mc_target_type ret = MC_TARGET_NONE; |
5801 | ||
84c3fc4e ZY |
5802 | if (unlikely(is_swap_pmd(pmd))) { |
5803 | VM_BUG_ON(thp_migration_supported() && | |
5804 | !is_pmd_migration_entry(pmd)); | |
5805 | return ret; | |
5806 | } | |
12724850 | 5807 | page = pmd_page(pmd); |
309381fe | 5808 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 5809 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 5810 | return ret; |
bcfe06bf | 5811 | if (page_memcg(page) == mc.from) { |
12724850 NH |
5812 | ret = MC_TARGET_PAGE; |
5813 | if (target) { | |
5814 | get_page(page); | |
5815 | target->page = page; | |
5816 | } | |
5817 | } | |
5818 | return ret; | |
5819 | } | |
5820 | #else | |
5821 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5822 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5823 | { | |
5824 | return MC_TARGET_NONE; | |
5825 | } | |
5826 | #endif | |
5827 | ||
4ffef5fe DN |
5828 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5829 | unsigned long addr, unsigned long end, | |
5830 | struct mm_walk *walk) | |
5831 | { | |
26bcd64a | 5832 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5833 | pte_t *pte; |
5834 | spinlock_t *ptl; | |
5835 | ||
b6ec57f4 KS |
5836 | ptl = pmd_trans_huge_lock(pmd, vma); |
5837 | if (ptl) { | |
c733a828 JG |
5838 | /* |
5839 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
25b2995a CH |
5840 | * support transparent huge page with MEMORY_DEVICE_PRIVATE but |
5841 | * this might change. | |
c733a828 | 5842 | */ |
12724850 NH |
5843 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5844 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5845 | spin_unlock(ptl); |
1a5a9906 | 5846 | return 0; |
12724850 | 5847 | } |
03319327 | 5848 | |
45f83cef AA |
5849 | if (pmd_trans_unstable(pmd)) |
5850 | return 0; | |
4ffef5fe DN |
5851 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5852 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5853 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5854 | mc.precharge++; /* increment precharge temporarily */ |
5855 | pte_unmap_unlock(pte - 1, ptl); | |
5856 | cond_resched(); | |
5857 | ||
7dc74be0 DN |
5858 | return 0; |
5859 | } | |
5860 | ||
7b86ac33 CH |
5861 | static const struct mm_walk_ops precharge_walk_ops = { |
5862 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5863 | }; | |
5864 | ||
4ffef5fe DN |
5865 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5866 | { | |
5867 | unsigned long precharge; | |
4ffef5fe | 5868 | |
d8ed45c5 | 5869 | mmap_read_lock(mm); |
7b86ac33 | 5870 | walk_page_range(mm, 0, mm->highest_vm_end, &precharge_walk_ops, NULL); |
d8ed45c5 | 5871 | mmap_read_unlock(mm); |
4ffef5fe DN |
5872 | |
5873 | precharge = mc.precharge; | |
5874 | mc.precharge = 0; | |
5875 | ||
5876 | return precharge; | |
5877 | } | |
5878 | ||
4ffef5fe DN |
5879 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5880 | { | |
dfe076b0 DN |
5881 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5882 | ||
5883 | VM_BUG_ON(mc.moving_task); | |
5884 | mc.moving_task = current; | |
5885 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5886 | } |
5887 | ||
dfe076b0 DN |
5888 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5889 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5890 | { |
2bd9bb20 KH |
5891 | struct mem_cgroup *from = mc.from; |
5892 | struct mem_cgroup *to = mc.to; | |
5893 | ||
4ffef5fe | 5894 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5895 | if (mc.precharge) { |
00501b53 | 5896 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5897 | mc.precharge = 0; |
5898 | } | |
5899 | /* | |
5900 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5901 | * we must uncharge here. | |
5902 | */ | |
5903 | if (mc.moved_charge) { | |
00501b53 | 5904 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5905 | mc.moved_charge = 0; |
4ffef5fe | 5906 | } |
483c30b5 DN |
5907 | /* we must fixup refcnts and charges */ |
5908 | if (mc.moved_swap) { | |
483c30b5 | 5909 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5910 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5911 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5912 | |
615d66c3 VD |
5913 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
5914 | ||
05b84301 | 5915 | /* |
3e32cb2e JW |
5916 | * we charged both to->memory and to->memsw, so we |
5917 | * should uncharge to->memory. | |
05b84301 | 5918 | */ |
ce00a967 | 5919 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5920 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5921 | ||
483c30b5 DN |
5922 | mc.moved_swap = 0; |
5923 | } | |
dfe076b0 DN |
5924 | memcg_oom_recover(from); |
5925 | memcg_oom_recover(to); | |
5926 | wake_up_all(&mc.waitq); | |
5927 | } | |
5928 | ||
5929 | static void mem_cgroup_clear_mc(void) | |
5930 | { | |
264a0ae1 TH |
5931 | struct mm_struct *mm = mc.mm; |
5932 | ||
dfe076b0 DN |
5933 | /* |
5934 | * we must clear moving_task before waking up waiters at the end of | |
5935 | * task migration. | |
5936 | */ | |
5937 | mc.moving_task = NULL; | |
5938 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5939 | spin_lock(&mc.lock); |
4ffef5fe DN |
5940 | mc.from = NULL; |
5941 | mc.to = NULL; | |
264a0ae1 | 5942 | mc.mm = NULL; |
2bd9bb20 | 5943 | spin_unlock(&mc.lock); |
264a0ae1 TH |
5944 | |
5945 | mmput(mm); | |
4ffef5fe DN |
5946 | } |
5947 | ||
1f7dd3e5 | 5948 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5949 | { |
1f7dd3e5 | 5950 | struct cgroup_subsys_state *css; |
eed67d75 | 5951 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 5952 | struct mem_cgroup *from; |
4530eddb | 5953 | struct task_struct *leader, *p; |
9f2115f9 | 5954 | struct mm_struct *mm; |
1dfab5ab | 5955 | unsigned long move_flags; |
9f2115f9 | 5956 | int ret = 0; |
7dc74be0 | 5957 | |
1f7dd3e5 TH |
5958 | /* charge immigration isn't supported on the default hierarchy */ |
5959 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
5960 | return 0; |
5961 | ||
4530eddb TH |
5962 | /* |
5963 | * Multi-process migrations only happen on the default hierarchy | |
5964 | * where charge immigration is not used. Perform charge | |
5965 | * immigration if @tset contains a leader and whine if there are | |
5966 | * multiple. | |
5967 | */ | |
5968 | p = NULL; | |
1f7dd3e5 | 5969 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
5970 | WARN_ON_ONCE(p); |
5971 | p = leader; | |
1f7dd3e5 | 5972 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
5973 | } |
5974 | if (!p) | |
5975 | return 0; | |
5976 | ||
1f7dd3e5 TH |
5977 | /* |
5978 | * We are now commited to this value whatever it is. Changes in this | |
5979 | * tunable will only affect upcoming migrations, not the current one. | |
5980 | * So we need to save it, and keep it going. | |
5981 | */ | |
5982 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
5983 | if (!move_flags) | |
5984 | return 0; | |
5985 | ||
9f2115f9 TH |
5986 | from = mem_cgroup_from_task(p); |
5987 | ||
5988 | VM_BUG_ON(from == memcg); | |
5989 | ||
5990 | mm = get_task_mm(p); | |
5991 | if (!mm) | |
5992 | return 0; | |
5993 | /* We move charges only when we move a owner of the mm */ | |
5994 | if (mm->owner == p) { | |
5995 | VM_BUG_ON(mc.from); | |
5996 | VM_BUG_ON(mc.to); | |
5997 | VM_BUG_ON(mc.precharge); | |
5998 | VM_BUG_ON(mc.moved_charge); | |
5999 | VM_BUG_ON(mc.moved_swap); | |
6000 | ||
6001 | spin_lock(&mc.lock); | |
264a0ae1 | 6002 | mc.mm = mm; |
9f2115f9 TH |
6003 | mc.from = from; |
6004 | mc.to = memcg; | |
6005 | mc.flags = move_flags; | |
6006 | spin_unlock(&mc.lock); | |
6007 | /* We set mc.moving_task later */ | |
6008 | ||
6009 | ret = mem_cgroup_precharge_mc(mm); | |
6010 | if (ret) | |
6011 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
6012 | } else { |
6013 | mmput(mm); | |
7dc74be0 DN |
6014 | } |
6015 | return ret; | |
6016 | } | |
6017 | ||
1f7dd3e5 | 6018 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 6019 | { |
4e2f245d JW |
6020 | if (mc.to) |
6021 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
6022 | } |
6023 | ||
4ffef5fe DN |
6024 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
6025 | unsigned long addr, unsigned long end, | |
6026 | struct mm_walk *walk) | |
7dc74be0 | 6027 | { |
4ffef5fe | 6028 | int ret = 0; |
26bcd64a | 6029 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
6030 | pte_t *pte; |
6031 | spinlock_t *ptl; | |
12724850 NH |
6032 | enum mc_target_type target_type; |
6033 | union mc_target target; | |
6034 | struct page *page; | |
4ffef5fe | 6035 | |
b6ec57f4 KS |
6036 | ptl = pmd_trans_huge_lock(pmd, vma); |
6037 | if (ptl) { | |
62ade86a | 6038 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 6039 | spin_unlock(ptl); |
12724850 NH |
6040 | return 0; |
6041 | } | |
6042 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
6043 | if (target_type == MC_TARGET_PAGE) { | |
6044 | page = target.page; | |
6045 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 6046 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 6047 | mc.from, mc.to)) { |
12724850 NH |
6048 | mc.precharge -= HPAGE_PMD_NR; |
6049 | mc.moved_charge += HPAGE_PMD_NR; | |
6050 | } | |
6051 | putback_lru_page(page); | |
6052 | } | |
6053 | put_page(page); | |
c733a828 JG |
6054 | } else if (target_type == MC_TARGET_DEVICE) { |
6055 | page = target.page; | |
6056 | if (!mem_cgroup_move_account(page, true, | |
6057 | mc.from, mc.to)) { | |
6058 | mc.precharge -= HPAGE_PMD_NR; | |
6059 | mc.moved_charge += HPAGE_PMD_NR; | |
6060 | } | |
6061 | put_page(page); | |
12724850 | 6062 | } |
bf929152 | 6063 | spin_unlock(ptl); |
1a5a9906 | 6064 | return 0; |
12724850 NH |
6065 | } |
6066 | ||
45f83cef AA |
6067 | if (pmd_trans_unstable(pmd)) |
6068 | return 0; | |
4ffef5fe DN |
6069 | retry: |
6070 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
6071 | for (; addr != end; addr += PAGE_SIZE) { | |
6072 | pte_t ptent = *(pte++); | |
c733a828 | 6073 | bool device = false; |
02491447 | 6074 | swp_entry_t ent; |
4ffef5fe DN |
6075 | |
6076 | if (!mc.precharge) | |
6077 | break; | |
6078 | ||
8d32ff84 | 6079 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
6080 | case MC_TARGET_DEVICE: |
6081 | device = true; | |
e4a9bc58 | 6082 | fallthrough; |
4ffef5fe DN |
6083 | case MC_TARGET_PAGE: |
6084 | page = target.page; | |
53f9263b KS |
6085 | /* |
6086 | * We can have a part of the split pmd here. Moving it | |
6087 | * can be done but it would be too convoluted so simply | |
6088 | * ignore such a partial THP and keep it in original | |
6089 | * memcg. There should be somebody mapping the head. | |
6090 | */ | |
6091 | if (PageTransCompound(page)) | |
6092 | goto put; | |
c733a828 | 6093 | if (!device && isolate_lru_page(page)) |
4ffef5fe | 6094 | goto put; |
f627c2f5 KS |
6095 | if (!mem_cgroup_move_account(page, false, |
6096 | mc.from, mc.to)) { | |
4ffef5fe | 6097 | mc.precharge--; |
854ffa8d DN |
6098 | /* we uncharge from mc.from later. */ |
6099 | mc.moved_charge++; | |
4ffef5fe | 6100 | } |
c733a828 JG |
6101 | if (!device) |
6102 | putback_lru_page(page); | |
8d32ff84 | 6103 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
6104 | put_page(page); |
6105 | break; | |
02491447 DN |
6106 | case MC_TARGET_SWAP: |
6107 | ent = target.ent; | |
e91cbb42 | 6108 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6109 | mc.precharge--; |
8d22a935 HD |
6110 | mem_cgroup_id_get_many(mc.to, 1); |
6111 | /* we fixup other refcnts and charges later. */ | |
483c30b5 DN |
6112 | mc.moved_swap++; |
6113 | } | |
02491447 | 6114 | break; |
4ffef5fe DN |
6115 | default: |
6116 | break; | |
6117 | } | |
6118 | } | |
6119 | pte_unmap_unlock(pte - 1, ptl); | |
6120 | cond_resched(); | |
6121 | ||
6122 | if (addr != end) { | |
6123 | /* | |
6124 | * We have consumed all precharges we got in can_attach(). | |
6125 | * We try charge one by one, but don't do any additional | |
6126 | * charges to mc.to if we have failed in charge once in attach() | |
6127 | * phase. | |
6128 | */ | |
854ffa8d | 6129 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6130 | if (!ret) |
6131 | goto retry; | |
6132 | } | |
6133 | ||
6134 | return ret; | |
6135 | } | |
6136 | ||
7b86ac33 CH |
6137 | static const struct mm_walk_ops charge_walk_ops = { |
6138 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6139 | }; | |
6140 | ||
264a0ae1 | 6141 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 6142 | { |
4ffef5fe | 6143 | lru_add_drain_all(); |
312722cb | 6144 | /* |
81f8c3a4 JW |
6145 | * Signal lock_page_memcg() to take the memcg's move_lock |
6146 | * while we're moving its pages to another memcg. Then wait | |
6147 | * for already started RCU-only updates to finish. | |
312722cb JW |
6148 | */ |
6149 | atomic_inc(&mc.from->moving_account); | |
6150 | synchronize_rcu(); | |
dfe076b0 | 6151 | retry: |
d8ed45c5 | 6152 | if (unlikely(!mmap_read_trylock(mc.mm))) { |
dfe076b0 | 6153 | /* |
c1e8d7c6 | 6154 | * Someone who are holding the mmap_lock might be waiting in |
dfe076b0 DN |
6155 | * waitq. So we cancel all extra charges, wake up all waiters, |
6156 | * and retry. Because we cancel precharges, we might not be able | |
6157 | * to move enough charges, but moving charge is a best-effort | |
6158 | * feature anyway, so it wouldn't be a big problem. | |
6159 | */ | |
6160 | __mem_cgroup_clear_mc(); | |
6161 | cond_resched(); | |
6162 | goto retry; | |
6163 | } | |
26bcd64a NH |
6164 | /* |
6165 | * When we have consumed all precharges and failed in doing | |
6166 | * additional charge, the page walk just aborts. | |
6167 | */ | |
7b86ac33 CH |
6168 | walk_page_range(mc.mm, 0, mc.mm->highest_vm_end, &charge_walk_ops, |
6169 | NULL); | |
0247f3f4 | 6170 | |
d8ed45c5 | 6171 | mmap_read_unlock(mc.mm); |
312722cb | 6172 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
6173 | } |
6174 | ||
264a0ae1 | 6175 | static void mem_cgroup_move_task(void) |
67e465a7 | 6176 | { |
264a0ae1 TH |
6177 | if (mc.to) { |
6178 | mem_cgroup_move_charge(); | |
a433658c | 6179 | mem_cgroup_clear_mc(); |
264a0ae1 | 6180 | } |
67e465a7 | 6181 | } |
5cfb80a7 | 6182 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 6183 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6184 | { |
6185 | return 0; | |
6186 | } | |
1f7dd3e5 | 6187 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6188 | { |
6189 | } | |
264a0ae1 | 6190 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
6191 | { |
6192 | } | |
6193 | #endif | |
67e465a7 | 6194 | |
677dc973 CD |
6195 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
6196 | { | |
6197 | if (value == PAGE_COUNTER_MAX) | |
6198 | seq_puts(m, "max\n"); | |
6199 | else | |
6200 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); | |
6201 | ||
6202 | return 0; | |
6203 | } | |
6204 | ||
241994ed JW |
6205 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
6206 | struct cftype *cft) | |
6207 | { | |
f5fc3c5d JW |
6208 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
6209 | ||
6210 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
6211 | } |
6212 | ||
bf8d5d52 RG |
6213 | static int memory_min_show(struct seq_file *m, void *v) |
6214 | { | |
677dc973 CD |
6215 | return seq_puts_memcg_tunable(m, |
6216 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); | |
bf8d5d52 RG |
6217 | } |
6218 | ||
6219 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
6220 | char *buf, size_t nbytes, loff_t off) | |
6221 | { | |
6222 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6223 | unsigned long min; | |
6224 | int err; | |
6225 | ||
6226 | buf = strstrip(buf); | |
6227 | err = page_counter_memparse(buf, "max", &min); | |
6228 | if (err) | |
6229 | return err; | |
6230 | ||
6231 | page_counter_set_min(&memcg->memory, min); | |
6232 | ||
6233 | return nbytes; | |
6234 | } | |
6235 | ||
241994ed JW |
6236 | static int memory_low_show(struct seq_file *m, void *v) |
6237 | { | |
677dc973 CD |
6238 | return seq_puts_memcg_tunable(m, |
6239 | READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); | |
241994ed JW |
6240 | } |
6241 | ||
6242 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
6243 | char *buf, size_t nbytes, loff_t off) | |
6244 | { | |
6245 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6246 | unsigned long low; | |
6247 | int err; | |
6248 | ||
6249 | buf = strstrip(buf); | |
d2973697 | 6250 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
6251 | if (err) |
6252 | return err; | |
6253 | ||
23067153 | 6254 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
6255 | |
6256 | return nbytes; | |
6257 | } | |
6258 | ||
6259 | static int memory_high_show(struct seq_file *m, void *v) | |
6260 | { | |
d1663a90 JK |
6261 | return seq_puts_memcg_tunable(m, |
6262 | READ_ONCE(mem_cgroup_from_seq(m)->memory.high)); | |
241994ed JW |
6263 | } |
6264 | ||
6265 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
6266 | char *buf, size_t nbytes, loff_t off) | |
6267 | { | |
6268 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6269 | unsigned int nr_retries = MAX_RECLAIM_RETRIES; |
8c8c383c | 6270 | bool drained = false; |
241994ed JW |
6271 | unsigned long high; |
6272 | int err; | |
6273 | ||
6274 | buf = strstrip(buf); | |
d2973697 | 6275 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
6276 | if (err) |
6277 | return err; | |
6278 | ||
e82553c1 JW |
6279 | page_counter_set_high(&memcg->memory, high); |
6280 | ||
8c8c383c JW |
6281 | for (;;) { |
6282 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6283 | unsigned long reclaimed; | |
6284 | ||
6285 | if (nr_pages <= high) | |
6286 | break; | |
6287 | ||
6288 | if (signal_pending(current)) | |
6289 | break; | |
6290 | ||
6291 | if (!drained) { | |
6292 | drain_all_stock(memcg); | |
6293 | drained = true; | |
6294 | continue; | |
6295 | } | |
6296 | ||
6297 | reclaimed = try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
6298 | GFP_KERNEL, true); | |
6299 | ||
6300 | if (!reclaimed && !nr_retries--) | |
6301 | break; | |
6302 | } | |
588083bb | 6303 | |
19ce33ac | 6304 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
6305 | return nbytes; |
6306 | } | |
6307 | ||
6308 | static int memory_max_show(struct seq_file *m, void *v) | |
6309 | { | |
677dc973 CD |
6310 | return seq_puts_memcg_tunable(m, |
6311 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); | |
241994ed JW |
6312 | } |
6313 | ||
6314 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
6315 | char *buf, size_t nbytes, loff_t off) | |
6316 | { | |
6317 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6318 | unsigned int nr_reclaims = MAX_RECLAIM_RETRIES; |
b6e6edcf | 6319 | bool drained = false; |
241994ed JW |
6320 | unsigned long max; |
6321 | int err; | |
6322 | ||
6323 | buf = strstrip(buf); | |
d2973697 | 6324 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
6325 | if (err) |
6326 | return err; | |
6327 | ||
bbec2e15 | 6328 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
6329 | |
6330 | for (;;) { | |
6331 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6332 | ||
6333 | if (nr_pages <= max) | |
6334 | break; | |
6335 | ||
7249c9f0 | 6336 | if (signal_pending(current)) |
b6e6edcf | 6337 | break; |
b6e6edcf JW |
6338 | |
6339 | if (!drained) { | |
6340 | drain_all_stock(memcg); | |
6341 | drained = true; | |
6342 | continue; | |
6343 | } | |
6344 | ||
6345 | if (nr_reclaims) { | |
6346 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
6347 | GFP_KERNEL, true)) | |
6348 | nr_reclaims--; | |
6349 | continue; | |
6350 | } | |
6351 | ||
e27be240 | 6352 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
6353 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
6354 | break; | |
6355 | } | |
241994ed | 6356 | |
2529bb3a | 6357 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
6358 | return nbytes; |
6359 | } | |
6360 | ||
1e577f97 SB |
6361 | static void __memory_events_show(struct seq_file *m, atomic_long_t *events) |
6362 | { | |
6363 | seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW])); | |
6364 | seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH])); | |
6365 | seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX])); | |
6366 | seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM])); | |
6367 | seq_printf(m, "oom_kill %lu\n", | |
6368 | atomic_long_read(&events[MEMCG_OOM_KILL])); | |
6369 | } | |
6370 | ||
241994ed JW |
6371 | static int memory_events_show(struct seq_file *m, void *v) |
6372 | { | |
aa9694bb | 6373 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
241994ed | 6374 | |
1e577f97 SB |
6375 | __memory_events_show(m, memcg->memory_events); |
6376 | return 0; | |
6377 | } | |
6378 | ||
6379 | static int memory_events_local_show(struct seq_file *m, void *v) | |
6380 | { | |
6381 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
241994ed | 6382 | |
1e577f97 | 6383 | __memory_events_show(m, memcg->memory_events_local); |
241994ed JW |
6384 | return 0; |
6385 | } | |
6386 | ||
587d9f72 JW |
6387 | static int memory_stat_show(struct seq_file *m, void *v) |
6388 | { | |
aa9694bb | 6389 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
c8713d0b | 6390 | char *buf; |
1ff9e6e1 | 6391 | |
c8713d0b JW |
6392 | buf = memory_stat_format(memcg); |
6393 | if (!buf) | |
6394 | return -ENOMEM; | |
6395 | seq_puts(m, buf); | |
6396 | kfree(buf); | |
587d9f72 JW |
6397 | return 0; |
6398 | } | |
6399 | ||
5f9a4f4a | 6400 | #ifdef CONFIG_NUMA |
fff66b79 MS |
6401 | static inline unsigned long lruvec_page_state_output(struct lruvec *lruvec, |
6402 | int item) | |
6403 | { | |
6404 | return lruvec_page_state(lruvec, item) * memcg_page_state_unit(item); | |
6405 | } | |
6406 | ||
5f9a4f4a MS |
6407 | static int memory_numa_stat_show(struct seq_file *m, void *v) |
6408 | { | |
6409 | int i; | |
6410 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
6411 | ||
6412 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { | |
6413 | int nid; | |
6414 | ||
6415 | if (memory_stats[i].idx >= NR_VM_NODE_STAT_ITEMS) | |
6416 | continue; | |
6417 | ||
6418 | seq_printf(m, "%s", memory_stats[i].name); | |
6419 | for_each_node_state(nid, N_MEMORY) { | |
6420 | u64 size; | |
6421 | struct lruvec *lruvec; | |
6422 | ||
6423 | lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); | |
fff66b79 MS |
6424 | size = lruvec_page_state_output(lruvec, |
6425 | memory_stats[i].idx); | |
5f9a4f4a MS |
6426 | seq_printf(m, " N%d=%llu", nid, size); |
6427 | } | |
6428 | seq_putc(m, '\n'); | |
6429 | } | |
6430 | ||
6431 | return 0; | |
6432 | } | |
6433 | #endif | |
6434 | ||
3d8b38eb RG |
6435 | static int memory_oom_group_show(struct seq_file *m, void *v) |
6436 | { | |
aa9694bb | 6437 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3d8b38eb RG |
6438 | |
6439 | seq_printf(m, "%d\n", memcg->oom_group); | |
6440 | ||
6441 | return 0; | |
6442 | } | |
6443 | ||
6444 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
6445 | char *buf, size_t nbytes, loff_t off) | |
6446 | { | |
6447 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6448 | int ret, oom_group; | |
6449 | ||
6450 | buf = strstrip(buf); | |
6451 | if (!buf) | |
6452 | return -EINVAL; | |
6453 | ||
6454 | ret = kstrtoint(buf, 0, &oom_group); | |
6455 | if (ret) | |
6456 | return ret; | |
6457 | ||
6458 | if (oom_group != 0 && oom_group != 1) | |
6459 | return -EINVAL; | |
6460 | ||
6461 | memcg->oom_group = oom_group; | |
6462 | ||
6463 | return nbytes; | |
6464 | } | |
6465 | ||
241994ed JW |
6466 | static struct cftype memory_files[] = { |
6467 | { | |
6468 | .name = "current", | |
f5fc3c5d | 6469 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
6470 | .read_u64 = memory_current_read, |
6471 | }, | |
bf8d5d52 RG |
6472 | { |
6473 | .name = "min", | |
6474 | .flags = CFTYPE_NOT_ON_ROOT, | |
6475 | .seq_show = memory_min_show, | |
6476 | .write = memory_min_write, | |
6477 | }, | |
241994ed JW |
6478 | { |
6479 | .name = "low", | |
6480 | .flags = CFTYPE_NOT_ON_ROOT, | |
6481 | .seq_show = memory_low_show, | |
6482 | .write = memory_low_write, | |
6483 | }, | |
6484 | { | |
6485 | .name = "high", | |
6486 | .flags = CFTYPE_NOT_ON_ROOT, | |
6487 | .seq_show = memory_high_show, | |
6488 | .write = memory_high_write, | |
6489 | }, | |
6490 | { | |
6491 | .name = "max", | |
6492 | .flags = CFTYPE_NOT_ON_ROOT, | |
6493 | .seq_show = memory_max_show, | |
6494 | .write = memory_max_write, | |
6495 | }, | |
6496 | { | |
6497 | .name = "events", | |
6498 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 6499 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
6500 | .seq_show = memory_events_show, |
6501 | }, | |
1e577f97 SB |
6502 | { |
6503 | .name = "events.local", | |
6504 | .flags = CFTYPE_NOT_ON_ROOT, | |
6505 | .file_offset = offsetof(struct mem_cgroup, events_local_file), | |
6506 | .seq_show = memory_events_local_show, | |
6507 | }, | |
587d9f72 JW |
6508 | { |
6509 | .name = "stat", | |
587d9f72 JW |
6510 | .seq_show = memory_stat_show, |
6511 | }, | |
5f9a4f4a MS |
6512 | #ifdef CONFIG_NUMA |
6513 | { | |
6514 | .name = "numa_stat", | |
6515 | .seq_show = memory_numa_stat_show, | |
6516 | }, | |
6517 | #endif | |
3d8b38eb RG |
6518 | { |
6519 | .name = "oom.group", | |
6520 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
6521 | .seq_show = memory_oom_group_show, | |
6522 | .write = memory_oom_group_write, | |
6523 | }, | |
241994ed JW |
6524 | { } /* terminate */ |
6525 | }; | |
6526 | ||
073219e9 | 6527 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 6528 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6529 | .css_online = mem_cgroup_css_online, |
92fb9748 | 6530 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 6531 | .css_released = mem_cgroup_css_released, |
92fb9748 | 6532 | .css_free = mem_cgroup_css_free, |
1ced953b | 6533 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
6534 | .can_attach = mem_cgroup_can_attach, |
6535 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 6536 | .post_attach = mem_cgroup_move_task, |
241994ed JW |
6537 | .dfl_cftypes = memory_files, |
6538 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 6539 | .early_init = 0, |
8cdea7c0 | 6540 | }; |
c077719b | 6541 | |
bc50bcc6 JW |
6542 | /* |
6543 | * This function calculates an individual cgroup's effective | |
6544 | * protection which is derived from its own memory.min/low, its | |
6545 | * parent's and siblings' settings, as well as the actual memory | |
6546 | * distribution in the tree. | |
6547 | * | |
6548 | * The following rules apply to the effective protection values: | |
6549 | * | |
6550 | * 1. At the first level of reclaim, effective protection is equal to | |
6551 | * the declared protection in memory.min and memory.low. | |
6552 | * | |
6553 | * 2. To enable safe delegation of the protection configuration, at | |
6554 | * subsequent levels the effective protection is capped to the | |
6555 | * parent's effective protection. | |
6556 | * | |
6557 | * 3. To make complex and dynamic subtrees easier to configure, the | |
6558 | * user is allowed to overcommit the declared protection at a given | |
6559 | * level. If that is the case, the parent's effective protection is | |
6560 | * distributed to the children in proportion to how much protection | |
6561 | * they have declared and how much of it they are utilizing. | |
6562 | * | |
6563 | * This makes distribution proportional, but also work-conserving: | |
6564 | * if one cgroup claims much more protection than it uses memory, | |
6565 | * the unused remainder is available to its siblings. | |
6566 | * | |
6567 | * 4. Conversely, when the declared protection is undercommitted at a | |
6568 | * given level, the distribution of the larger parental protection | |
6569 | * budget is NOT proportional. A cgroup's protection from a sibling | |
6570 | * is capped to its own memory.min/low setting. | |
6571 | * | |
8a931f80 JW |
6572 | * 5. However, to allow protecting recursive subtrees from each other |
6573 | * without having to declare each individual cgroup's fixed share | |
6574 | * of the ancestor's claim to protection, any unutilized - | |
6575 | * "floating" - protection from up the tree is distributed in | |
6576 | * proportion to each cgroup's *usage*. This makes the protection | |
6577 | * neutral wrt sibling cgroups and lets them compete freely over | |
6578 | * the shared parental protection budget, but it protects the | |
6579 | * subtree as a whole from neighboring subtrees. | |
6580 | * | |
6581 | * Note that 4. and 5. are not in conflict: 4. is about protecting | |
6582 | * against immediate siblings whereas 5. is about protecting against | |
6583 | * neighboring subtrees. | |
bc50bcc6 JW |
6584 | */ |
6585 | static unsigned long effective_protection(unsigned long usage, | |
8a931f80 | 6586 | unsigned long parent_usage, |
bc50bcc6 JW |
6587 | unsigned long setting, |
6588 | unsigned long parent_effective, | |
6589 | unsigned long siblings_protected) | |
6590 | { | |
6591 | unsigned long protected; | |
8a931f80 | 6592 | unsigned long ep; |
bc50bcc6 JW |
6593 | |
6594 | protected = min(usage, setting); | |
6595 | /* | |
6596 | * If all cgroups at this level combined claim and use more | |
6597 | * protection then what the parent affords them, distribute | |
6598 | * shares in proportion to utilization. | |
6599 | * | |
6600 | * We are using actual utilization rather than the statically | |
6601 | * claimed protection in order to be work-conserving: claimed | |
6602 | * but unused protection is available to siblings that would | |
6603 | * otherwise get a smaller chunk than what they claimed. | |
6604 | */ | |
6605 | if (siblings_protected > parent_effective) | |
6606 | return protected * parent_effective / siblings_protected; | |
6607 | ||
6608 | /* | |
6609 | * Ok, utilized protection of all children is within what the | |
6610 | * parent affords them, so we know whatever this child claims | |
6611 | * and utilizes is effectively protected. | |
6612 | * | |
6613 | * If there is unprotected usage beyond this value, reclaim | |
6614 | * will apply pressure in proportion to that amount. | |
6615 | * | |
6616 | * If there is unutilized protection, the cgroup will be fully | |
6617 | * shielded from reclaim, but we do return a smaller value for | |
6618 | * protection than what the group could enjoy in theory. This | |
6619 | * is okay. With the overcommit distribution above, effective | |
6620 | * protection is always dependent on how memory is actually | |
6621 | * consumed among the siblings anyway. | |
6622 | */ | |
8a931f80 JW |
6623 | ep = protected; |
6624 | ||
6625 | /* | |
6626 | * If the children aren't claiming (all of) the protection | |
6627 | * afforded to them by the parent, distribute the remainder in | |
6628 | * proportion to the (unprotected) memory of each cgroup. That | |
6629 | * way, cgroups that aren't explicitly prioritized wrt each | |
6630 | * other compete freely over the allowance, but they are | |
6631 | * collectively protected from neighboring trees. | |
6632 | * | |
6633 | * We're using unprotected memory for the weight so that if | |
6634 | * some cgroups DO claim explicit protection, we don't protect | |
6635 | * the same bytes twice. | |
cd324edc JW |
6636 | * |
6637 | * Check both usage and parent_usage against the respective | |
6638 | * protected values. One should imply the other, but they | |
6639 | * aren't read atomically - make sure the division is sane. | |
8a931f80 JW |
6640 | */ |
6641 | if (!(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)) | |
6642 | return ep; | |
cd324edc JW |
6643 | if (parent_effective > siblings_protected && |
6644 | parent_usage > siblings_protected && | |
6645 | usage > protected) { | |
8a931f80 JW |
6646 | unsigned long unclaimed; |
6647 | ||
6648 | unclaimed = parent_effective - siblings_protected; | |
6649 | unclaimed *= usage - protected; | |
6650 | unclaimed /= parent_usage - siblings_protected; | |
6651 | ||
6652 | ep += unclaimed; | |
6653 | } | |
6654 | ||
6655 | return ep; | |
bc50bcc6 JW |
6656 | } |
6657 | ||
241994ed | 6658 | /** |
bf8d5d52 | 6659 | * mem_cgroup_protected - check if memory consumption is in the normal range |
34c81057 | 6660 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
6661 | * @memcg: the memory cgroup to check |
6662 | * | |
23067153 RG |
6663 | * WARNING: This function is not stateless! It can only be used as part |
6664 | * of a top-down tree iteration, not for isolated queries. | |
241994ed | 6665 | */ |
45c7f7e1 CD |
6666 | void mem_cgroup_calculate_protection(struct mem_cgroup *root, |
6667 | struct mem_cgroup *memcg) | |
241994ed | 6668 | { |
8a931f80 | 6669 | unsigned long usage, parent_usage; |
23067153 RG |
6670 | struct mem_cgroup *parent; |
6671 | ||
241994ed | 6672 | if (mem_cgroup_disabled()) |
45c7f7e1 | 6673 | return; |
241994ed | 6674 | |
34c81057 SC |
6675 | if (!root) |
6676 | root = root_mem_cgroup; | |
22f7496f YS |
6677 | |
6678 | /* | |
6679 | * Effective values of the reclaim targets are ignored so they | |
6680 | * can be stale. Have a look at mem_cgroup_protection for more | |
6681 | * details. | |
6682 | * TODO: calculation should be more robust so that we do not need | |
6683 | * that special casing. | |
6684 | */ | |
34c81057 | 6685 | if (memcg == root) |
45c7f7e1 | 6686 | return; |
241994ed | 6687 | |
23067153 | 6688 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 | 6689 | if (!usage) |
45c7f7e1 | 6690 | return; |
bf8d5d52 | 6691 | |
bf8d5d52 | 6692 | parent = parent_mem_cgroup(memcg); |
df2a4196 RG |
6693 | /* No parent means a non-hierarchical mode on v1 memcg */ |
6694 | if (!parent) | |
45c7f7e1 | 6695 | return; |
df2a4196 | 6696 | |
bc50bcc6 | 6697 | if (parent == root) { |
c3d53200 | 6698 | memcg->memory.emin = READ_ONCE(memcg->memory.min); |
03960e33 | 6699 | memcg->memory.elow = READ_ONCE(memcg->memory.low); |
45c7f7e1 | 6700 | return; |
bf8d5d52 RG |
6701 | } |
6702 | ||
8a931f80 JW |
6703 | parent_usage = page_counter_read(&parent->memory); |
6704 | ||
b3a7822e | 6705 | WRITE_ONCE(memcg->memory.emin, effective_protection(usage, parent_usage, |
c3d53200 CD |
6706 | READ_ONCE(memcg->memory.min), |
6707 | READ_ONCE(parent->memory.emin), | |
b3a7822e | 6708 | atomic_long_read(&parent->memory.children_min_usage))); |
23067153 | 6709 | |
b3a7822e | 6710 | WRITE_ONCE(memcg->memory.elow, effective_protection(usage, parent_usage, |
03960e33 CD |
6711 | READ_ONCE(memcg->memory.low), |
6712 | READ_ONCE(parent->memory.elow), | |
b3a7822e | 6713 | atomic_long_read(&parent->memory.children_low_usage))); |
241994ed JW |
6714 | } |
6715 | ||
00501b53 | 6716 | /** |
f0e45fb4 | 6717 | * mem_cgroup_charge - charge a newly allocated page to a cgroup |
00501b53 JW |
6718 | * @page: page to charge |
6719 | * @mm: mm context of the victim | |
6720 | * @gfp_mask: reclaim mode | |
00501b53 JW |
6721 | * |
6722 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
6723 | * pages according to @gfp_mask if necessary. | |
6724 | * | |
f0e45fb4 | 6725 | * Returns 0 on success. Otherwise, an error code is returned. |
00501b53 | 6726 | */ |
d9eb1ea2 | 6727 | int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask) |
00501b53 | 6728 | { |
6c357848 | 6729 | unsigned int nr_pages = thp_nr_pages(page); |
00501b53 | 6730 | struct mem_cgroup *memcg = NULL; |
00501b53 JW |
6731 | int ret = 0; |
6732 | ||
6733 | if (mem_cgroup_disabled()) | |
6734 | goto out; | |
6735 | ||
6736 | if (PageSwapCache(page)) { | |
2d1c4980 JW |
6737 | swp_entry_t ent = { .val = page_private(page), }; |
6738 | unsigned short id; | |
6739 | ||
00501b53 JW |
6740 | /* |
6741 | * Every swap fault against a single page tries to charge the | |
6742 | * page, bail as early as possible. shmem_unuse() encounters | |
bcfe06bf RG |
6743 | * already charged pages, too. page and memcg binding is |
6744 | * protected by the page lock, which serializes swap cache | |
6745 | * removal, which in turn serializes uncharging. | |
00501b53 | 6746 | */ |
e993d905 | 6747 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
bcfe06bf | 6748 | if (page_memcg(compound_head(page))) |
00501b53 | 6749 | goto out; |
e993d905 | 6750 | |
2d1c4980 JW |
6751 | id = lookup_swap_cgroup_id(ent); |
6752 | rcu_read_lock(); | |
6753 | memcg = mem_cgroup_from_id(id); | |
6754 | if (memcg && !css_tryget_online(&memcg->css)) | |
6755 | memcg = NULL; | |
6756 | rcu_read_unlock(); | |
00501b53 JW |
6757 | } |
6758 | ||
00501b53 JW |
6759 | if (!memcg) |
6760 | memcg = get_mem_cgroup_from_mm(mm); | |
6761 | ||
6762 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
f0e45fb4 JW |
6763 | if (ret) |
6764 | goto out_put; | |
00501b53 | 6765 | |
1a3e1f40 | 6766 | css_get(&memcg->css); |
d9eb1ea2 | 6767 | commit_charge(page, memcg); |
6abb5a86 | 6768 | |
6abb5a86 | 6769 | local_irq_disable(); |
3fba69a5 | 6770 | mem_cgroup_charge_statistics(memcg, page, nr_pages); |
6abb5a86 JW |
6771 | memcg_check_events(memcg, page); |
6772 | local_irq_enable(); | |
00501b53 | 6773 | |
2d1c4980 | 6774 | if (PageSwapCache(page)) { |
00501b53 JW |
6775 | swp_entry_t entry = { .val = page_private(page) }; |
6776 | /* | |
6777 | * The swap entry might not get freed for a long time, | |
6778 | * let's not wait for it. The page already received a | |
6779 | * memory+swap charge, drop the swap entry duplicate. | |
6780 | */ | |
38d8b4e6 | 6781 | mem_cgroup_uncharge_swap(entry, nr_pages); |
00501b53 | 6782 | } |
00501b53 | 6783 | |
f0e45fb4 JW |
6784 | out_put: |
6785 | css_put(&memcg->css); | |
6786 | out: | |
6787 | return ret; | |
3fea5a49 JW |
6788 | } |
6789 | ||
a9d5adee JG |
6790 | struct uncharge_gather { |
6791 | struct mem_cgroup *memcg; | |
9f762dbe | 6792 | unsigned long nr_pages; |
a9d5adee | 6793 | unsigned long pgpgout; |
a9d5adee | 6794 | unsigned long nr_kmem; |
a9d5adee JG |
6795 | struct page *dummy_page; |
6796 | }; | |
6797 | ||
6798 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 6799 | { |
a9d5adee JG |
6800 | memset(ug, 0, sizeof(*ug)); |
6801 | } | |
6802 | ||
6803 | static void uncharge_batch(const struct uncharge_gather *ug) | |
6804 | { | |
747db954 JW |
6805 | unsigned long flags; |
6806 | ||
a9d5adee | 6807 | if (!mem_cgroup_is_root(ug->memcg)) { |
9f762dbe | 6808 | page_counter_uncharge(&ug->memcg->memory, ug->nr_pages); |
7941d214 | 6809 | if (do_memsw_account()) |
9f762dbe | 6810 | page_counter_uncharge(&ug->memcg->memsw, ug->nr_pages); |
a9d5adee JG |
6811 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem) |
6812 | page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem); | |
6813 | memcg_oom_recover(ug->memcg); | |
ce00a967 | 6814 | } |
747db954 JW |
6815 | |
6816 | local_irq_save(flags); | |
c9019e9b | 6817 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); |
9f762dbe | 6818 | __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, ug->nr_pages); |
a9d5adee | 6819 | memcg_check_events(ug->memcg, ug->dummy_page); |
747db954 | 6820 | local_irq_restore(flags); |
f1796544 MH |
6821 | |
6822 | /* drop reference from uncharge_page */ | |
6823 | css_put(&ug->memcg->css); | |
a9d5adee JG |
6824 | } |
6825 | ||
6826 | static void uncharge_page(struct page *page, struct uncharge_gather *ug) | |
6827 | { | |
9f762dbe JW |
6828 | unsigned long nr_pages; |
6829 | ||
a9d5adee | 6830 | VM_BUG_ON_PAGE(PageLRU(page), page); |
a9d5adee | 6831 | |
bcfe06bf | 6832 | if (!page_memcg(page)) |
a9d5adee JG |
6833 | return; |
6834 | ||
6835 | /* | |
6836 | * Nobody should be changing or seriously looking at | |
bcfe06bf | 6837 | * page_memcg(page) at this point, we have fully |
a9d5adee JG |
6838 | * exclusive access to the page. |
6839 | */ | |
6840 | ||
bcfe06bf | 6841 | if (ug->memcg != page_memcg(page)) { |
a9d5adee JG |
6842 | if (ug->memcg) { |
6843 | uncharge_batch(ug); | |
6844 | uncharge_gather_clear(ug); | |
6845 | } | |
bcfe06bf | 6846 | ug->memcg = page_memcg(page); |
f1796544 MH |
6847 | |
6848 | /* pairs with css_put in uncharge_batch */ | |
6849 | css_get(&ug->memcg->css); | |
a9d5adee JG |
6850 | } |
6851 | ||
9f762dbe JW |
6852 | nr_pages = compound_nr(page); |
6853 | ug->nr_pages += nr_pages; | |
a9d5adee | 6854 | |
18b2db3b | 6855 | if (PageMemcgKmem(page)) |
9f762dbe | 6856 | ug->nr_kmem += nr_pages; |
18b2db3b RG |
6857 | else |
6858 | ug->pgpgout++; | |
a9d5adee JG |
6859 | |
6860 | ug->dummy_page = page; | |
bcfe06bf | 6861 | page->memcg_data = 0; |
1a3e1f40 | 6862 | css_put(&ug->memcg->css); |
747db954 JW |
6863 | } |
6864 | ||
6865 | static void uncharge_list(struct list_head *page_list) | |
6866 | { | |
a9d5adee | 6867 | struct uncharge_gather ug; |
747db954 | 6868 | struct list_head *next; |
a9d5adee JG |
6869 | |
6870 | uncharge_gather_clear(&ug); | |
747db954 | 6871 | |
8b592656 JW |
6872 | /* |
6873 | * Note that the list can be a single page->lru; hence the | |
6874 | * do-while loop instead of a simple list_for_each_entry(). | |
6875 | */ | |
747db954 JW |
6876 | next = page_list->next; |
6877 | do { | |
a9d5adee JG |
6878 | struct page *page; |
6879 | ||
747db954 JW |
6880 | page = list_entry(next, struct page, lru); |
6881 | next = page->lru.next; | |
6882 | ||
a9d5adee | 6883 | uncharge_page(page, &ug); |
747db954 JW |
6884 | } while (next != page_list); |
6885 | ||
a9d5adee JG |
6886 | if (ug.memcg) |
6887 | uncharge_batch(&ug); | |
747db954 JW |
6888 | } |
6889 | ||
0a31bc97 JW |
6890 | /** |
6891 | * mem_cgroup_uncharge - uncharge a page | |
6892 | * @page: page to uncharge | |
6893 | * | |
f0e45fb4 | 6894 | * Uncharge a page previously charged with mem_cgroup_charge(). |
0a31bc97 JW |
6895 | */ |
6896 | void mem_cgroup_uncharge(struct page *page) | |
6897 | { | |
a9d5adee JG |
6898 | struct uncharge_gather ug; |
6899 | ||
0a31bc97 JW |
6900 | if (mem_cgroup_disabled()) |
6901 | return; | |
6902 | ||
747db954 | 6903 | /* Don't touch page->lru of any random page, pre-check: */ |
bcfe06bf | 6904 | if (!page_memcg(page)) |
0a31bc97 JW |
6905 | return; |
6906 | ||
a9d5adee JG |
6907 | uncharge_gather_clear(&ug); |
6908 | uncharge_page(page, &ug); | |
6909 | uncharge_batch(&ug); | |
747db954 | 6910 | } |
0a31bc97 | 6911 | |
747db954 JW |
6912 | /** |
6913 | * mem_cgroup_uncharge_list - uncharge a list of page | |
6914 | * @page_list: list of pages to uncharge | |
6915 | * | |
6916 | * Uncharge a list of pages previously charged with | |
f0e45fb4 | 6917 | * mem_cgroup_charge(). |
747db954 JW |
6918 | */ |
6919 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
6920 | { | |
6921 | if (mem_cgroup_disabled()) | |
6922 | return; | |
0a31bc97 | 6923 | |
747db954 JW |
6924 | if (!list_empty(page_list)) |
6925 | uncharge_list(page_list); | |
0a31bc97 JW |
6926 | } |
6927 | ||
6928 | /** | |
6a93ca8f JW |
6929 | * mem_cgroup_migrate - charge a page's replacement |
6930 | * @oldpage: currently circulating page | |
6931 | * @newpage: replacement page | |
0a31bc97 | 6932 | * |
6a93ca8f JW |
6933 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
6934 | * be uncharged upon free. | |
0a31bc97 JW |
6935 | * |
6936 | * Both pages must be locked, @newpage->mapping must be set up. | |
6937 | */ | |
6a93ca8f | 6938 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 6939 | { |
29833315 | 6940 | struct mem_cgroup *memcg; |
44b7a8d3 | 6941 | unsigned int nr_pages; |
d93c4130 | 6942 | unsigned long flags; |
0a31bc97 JW |
6943 | |
6944 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6945 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 6946 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
6947 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6948 | newpage); | |
0a31bc97 JW |
6949 | |
6950 | if (mem_cgroup_disabled()) | |
6951 | return; | |
6952 | ||
6953 | /* Page cache replacement: new page already charged? */ | |
bcfe06bf | 6954 | if (page_memcg(newpage)) |
0a31bc97 JW |
6955 | return; |
6956 | ||
bcfe06bf | 6957 | memcg = page_memcg(oldpage); |
a4055888 | 6958 | VM_WARN_ON_ONCE_PAGE(!memcg, oldpage); |
29833315 | 6959 | if (!memcg) |
0a31bc97 JW |
6960 | return; |
6961 | ||
44b7a8d3 | 6962 | /* Force-charge the new page. The old one will be freed soon */ |
6c357848 | 6963 | nr_pages = thp_nr_pages(newpage); |
44b7a8d3 JW |
6964 | |
6965 | page_counter_charge(&memcg->memory, nr_pages); | |
6966 | if (do_memsw_account()) | |
6967 | page_counter_charge(&memcg->memsw, nr_pages); | |
0a31bc97 | 6968 | |
1a3e1f40 | 6969 | css_get(&memcg->css); |
d9eb1ea2 | 6970 | commit_charge(newpage, memcg); |
44b7a8d3 | 6971 | |
d93c4130 | 6972 | local_irq_save(flags); |
3fba69a5 | 6973 | mem_cgroup_charge_statistics(memcg, newpage, nr_pages); |
44b7a8d3 | 6974 | memcg_check_events(memcg, newpage); |
d93c4130 | 6975 | local_irq_restore(flags); |
0a31bc97 JW |
6976 | } |
6977 | ||
ef12947c | 6978 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
6979 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
6980 | ||
2d758073 | 6981 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
6982 | { |
6983 | struct mem_cgroup *memcg; | |
6984 | ||
2d758073 JW |
6985 | if (!mem_cgroup_sockets_enabled) |
6986 | return; | |
6987 | ||
e876ecc6 SB |
6988 | /* Do not associate the sock with unrelated interrupted task's memcg. */ |
6989 | if (in_interrupt()) | |
6990 | return; | |
6991 | ||
11092087 JW |
6992 | rcu_read_lock(); |
6993 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
6994 | if (memcg == root_mem_cgroup) |
6995 | goto out; | |
0db15298 | 6996 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 6997 | goto out; |
8965aa28 | 6998 | if (css_tryget(&memcg->css)) |
11092087 | 6999 | sk->sk_memcg = memcg; |
f7e1cb6e | 7000 | out: |
11092087 JW |
7001 | rcu_read_unlock(); |
7002 | } | |
11092087 | 7003 | |
2d758073 | 7004 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 7005 | { |
2d758073 JW |
7006 | if (sk->sk_memcg) |
7007 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
7008 | } |
7009 | ||
7010 | /** | |
7011 | * mem_cgroup_charge_skmem - charge socket memory | |
7012 | * @memcg: memcg to charge | |
7013 | * @nr_pages: number of pages to charge | |
7014 | * | |
7015 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
7016 | * @memcg's configured limit, %false if the charge had to be forced. | |
7017 | */ | |
7018 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
7019 | { | |
f7e1cb6e | 7020 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 7021 | |
f7e1cb6e | 7022 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 7023 | struct page_counter *fail; |
f7e1cb6e | 7024 | |
0db15298 JW |
7025 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
7026 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
7027 | return true; |
7028 | } | |
0db15298 JW |
7029 | page_counter_charge(&memcg->tcpmem, nr_pages); |
7030 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 7031 | return false; |
11092087 | 7032 | } |
d886f4e4 | 7033 | |
f7e1cb6e JW |
7034 | /* Don't block in the packet receive path */ |
7035 | if (in_softirq()) | |
7036 | gfp_mask = GFP_NOWAIT; | |
7037 | ||
c9019e9b | 7038 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); |
b2807f07 | 7039 | |
f7e1cb6e JW |
7040 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
7041 | return true; | |
7042 | ||
7043 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
7044 | return false; |
7045 | } | |
7046 | ||
7047 | /** | |
7048 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
7049 | * @memcg: memcg to uncharge |
7050 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
7051 | */ |
7052 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
7053 | { | |
f7e1cb6e | 7054 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 7055 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
7056 | return; |
7057 | } | |
d886f4e4 | 7058 | |
c9019e9b | 7059 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 7060 | |
475d0487 | 7061 | refill_stock(memcg, nr_pages); |
11092087 JW |
7062 | } |
7063 | ||
f7e1cb6e JW |
7064 | static int __init cgroup_memory(char *s) |
7065 | { | |
7066 | char *token; | |
7067 | ||
7068 | while ((token = strsep(&s, ",")) != NULL) { | |
7069 | if (!*token) | |
7070 | continue; | |
7071 | if (!strcmp(token, "nosocket")) | |
7072 | cgroup_memory_nosocket = true; | |
04823c83 VD |
7073 | if (!strcmp(token, "nokmem")) |
7074 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
7075 | } |
7076 | return 0; | |
7077 | } | |
7078 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 7079 | |
2d11085e | 7080 | /* |
1081312f MH |
7081 | * subsys_initcall() for memory controller. |
7082 | * | |
308167fc SAS |
7083 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
7084 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
7085 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
7086 | * should be initialized from here. | |
2d11085e MH |
7087 | */ |
7088 | static int __init mem_cgroup_init(void) | |
7089 | { | |
95a045f6 JW |
7090 | int cpu, node; |
7091 | ||
f3344adf MS |
7092 | /* |
7093 | * Currently s32 type (can refer to struct batched_lruvec_stat) is | |
7094 | * used for per-memcg-per-cpu caching of per-node statistics. In order | |
7095 | * to work fine, we should make sure that the overfill threshold can't | |
7096 | * exceed S32_MAX / PAGE_SIZE. | |
7097 | */ | |
7098 | BUILD_BUG_ON(MEMCG_CHARGE_BATCH > S32_MAX / PAGE_SIZE); | |
7099 | ||
308167fc SAS |
7100 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
7101 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
7102 | |
7103 | for_each_possible_cpu(cpu) | |
7104 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
7105 | drain_local_stock); | |
7106 | ||
7107 | for_each_node(node) { | |
7108 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
7109 | |
7110 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
7111 | node_online(node) ? node : NUMA_NO_NODE); | |
7112 | ||
ef8f2327 | 7113 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 7114 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 7115 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
7116 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
7117 | } | |
7118 | ||
2d11085e MH |
7119 | return 0; |
7120 | } | |
7121 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
7122 | |
7123 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
7124 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
7125 | { | |
1c2d479a | 7126 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
7127 | /* |
7128 | * The root cgroup cannot be destroyed, so it's refcount must | |
7129 | * always be >= 1. | |
7130 | */ | |
7131 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
7132 | VM_BUG_ON(1); | |
7133 | break; | |
7134 | } | |
7135 | memcg = parent_mem_cgroup(memcg); | |
7136 | if (!memcg) | |
7137 | memcg = root_mem_cgroup; | |
7138 | } | |
7139 | return memcg; | |
7140 | } | |
7141 | ||
21afa38e JW |
7142 | /** |
7143 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
7144 | * @page: page whose memsw charge to transfer | |
7145 | * @entry: swap entry to move the charge to | |
7146 | * | |
7147 | * Transfer the memsw charge of @page to @entry. | |
7148 | */ | |
7149 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
7150 | { | |
1f47b61f | 7151 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 7152 | unsigned int nr_entries; |
21afa38e JW |
7153 | unsigned short oldid; |
7154 | ||
7155 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
7156 | VM_BUG_ON_PAGE(page_count(page), page); | |
7157 | ||
76358ab5 AS |
7158 | if (mem_cgroup_disabled()) |
7159 | return; | |
7160 | ||
2d1c4980 | 7161 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
21afa38e JW |
7162 | return; |
7163 | ||
bcfe06bf | 7164 | memcg = page_memcg(page); |
21afa38e | 7165 | |
a4055888 | 7166 | VM_WARN_ON_ONCE_PAGE(!memcg, page); |
21afa38e JW |
7167 | if (!memcg) |
7168 | return; | |
7169 | ||
1f47b61f VD |
7170 | /* |
7171 | * In case the memcg owning these pages has been offlined and doesn't | |
7172 | * have an ID allocated to it anymore, charge the closest online | |
7173 | * ancestor for the swap instead and transfer the memory+swap charge. | |
7174 | */ | |
7175 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
6c357848 | 7176 | nr_entries = thp_nr_pages(page); |
d6810d73 HY |
7177 | /* Get references for the tail pages, too */ |
7178 | if (nr_entries > 1) | |
7179 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
7180 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
7181 | nr_entries); | |
21afa38e | 7182 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7183 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e | 7184 | |
bcfe06bf | 7185 | page->memcg_data = 0; |
21afa38e JW |
7186 | |
7187 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 7188 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 7189 | |
2d1c4980 | 7190 | if (!cgroup_memory_noswap && memcg != swap_memcg) { |
1f47b61f | 7191 | if (!mem_cgroup_is_root(swap_memcg)) |
d6810d73 HY |
7192 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
7193 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
7194 | } |
7195 | ||
ce9ce665 SAS |
7196 | /* |
7197 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 7198 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 7199 | * important here to have the interrupts disabled because it is the |
b93b0163 | 7200 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 SAS |
7201 | */ |
7202 | VM_BUG_ON(!irqs_disabled()); | |
3fba69a5 | 7203 | mem_cgroup_charge_statistics(memcg, page, -nr_entries); |
21afa38e | 7204 | memcg_check_events(memcg, page); |
73f576c0 | 7205 | |
1a3e1f40 | 7206 | css_put(&memcg->css); |
21afa38e JW |
7207 | } |
7208 | ||
38d8b4e6 HY |
7209 | /** |
7210 | * mem_cgroup_try_charge_swap - try charging swap space for a page | |
37e84351 VD |
7211 | * @page: page being added to swap |
7212 | * @entry: swap entry to charge | |
7213 | * | |
38d8b4e6 | 7214 | * Try to charge @page's memcg for the swap space at @entry. |
37e84351 VD |
7215 | * |
7216 | * Returns 0 on success, -ENOMEM on failure. | |
7217 | */ | |
7218 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
7219 | { | |
6c357848 | 7220 | unsigned int nr_pages = thp_nr_pages(page); |
37e84351 | 7221 | struct page_counter *counter; |
38d8b4e6 | 7222 | struct mem_cgroup *memcg; |
37e84351 VD |
7223 | unsigned short oldid; |
7224 | ||
76358ab5 AS |
7225 | if (mem_cgroup_disabled()) |
7226 | return 0; | |
7227 | ||
2d1c4980 | 7228 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
37e84351 VD |
7229 | return 0; |
7230 | ||
bcfe06bf | 7231 | memcg = page_memcg(page); |
37e84351 | 7232 | |
a4055888 | 7233 | VM_WARN_ON_ONCE_PAGE(!memcg, page); |
37e84351 VD |
7234 | if (!memcg) |
7235 | return 0; | |
7236 | ||
f3a53a3a TH |
7237 | if (!entry.val) { |
7238 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 7239 | return 0; |
f3a53a3a | 7240 | } |
bb98f2c5 | 7241 | |
1f47b61f VD |
7242 | memcg = mem_cgroup_id_get_online(memcg); |
7243 | ||
2d1c4980 | 7244 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg) && |
38d8b4e6 | 7245 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
7246 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
7247 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 7248 | mem_cgroup_id_put(memcg); |
37e84351 | 7249 | return -ENOMEM; |
1f47b61f | 7250 | } |
37e84351 | 7251 | |
38d8b4e6 HY |
7252 | /* Get references for the tail pages, too */ |
7253 | if (nr_pages > 1) | |
7254 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
7255 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
37e84351 | 7256 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7257 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 7258 | |
37e84351 VD |
7259 | return 0; |
7260 | } | |
7261 | ||
21afa38e | 7262 | /** |
38d8b4e6 | 7263 | * mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 7264 | * @entry: swap entry to uncharge |
38d8b4e6 | 7265 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 7266 | */ |
38d8b4e6 | 7267 | void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
7268 | { |
7269 | struct mem_cgroup *memcg; | |
7270 | unsigned short id; | |
7271 | ||
38d8b4e6 | 7272 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 7273 | rcu_read_lock(); |
adbe427b | 7274 | memcg = mem_cgroup_from_id(id); |
21afa38e | 7275 | if (memcg) { |
2d1c4980 | 7276 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg)) { |
37e84351 | 7277 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
38d8b4e6 | 7278 | page_counter_uncharge(&memcg->swap, nr_pages); |
37e84351 | 7279 | else |
38d8b4e6 | 7280 | page_counter_uncharge(&memcg->memsw, nr_pages); |
37e84351 | 7281 | } |
c9019e9b | 7282 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 7283 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
7284 | } |
7285 | rcu_read_unlock(); | |
7286 | } | |
7287 | ||
d8b38438 VD |
7288 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
7289 | { | |
7290 | long nr_swap_pages = get_nr_swap_pages(); | |
7291 | ||
eccb52e7 | 7292 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
d8b38438 VD |
7293 | return nr_swap_pages; |
7294 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
7295 | nr_swap_pages = min_t(long, nr_swap_pages, | |
bbec2e15 | 7296 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
7297 | page_counter_read(&memcg->swap)); |
7298 | return nr_swap_pages; | |
7299 | } | |
7300 | ||
5ccc5aba VD |
7301 | bool mem_cgroup_swap_full(struct page *page) |
7302 | { | |
7303 | struct mem_cgroup *memcg; | |
7304 | ||
7305 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
7306 | ||
7307 | if (vm_swap_full()) | |
7308 | return true; | |
eccb52e7 | 7309 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
5ccc5aba VD |
7310 | return false; |
7311 | ||
bcfe06bf | 7312 | memcg = page_memcg(page); |
5ccc5aba VD |
7313 | if (!memcg) |
7314 | return false; | |
7315 | ||
4b82ab4f JK |
7316 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { |
7317 | unsigned long usage = page_counter_read(&memcg->swap); | |
7318 | ||
7319 | if (usage * 2 >= READ_ONCE(memcg->swap.high) || | |
7320 | usage * 2 >= READ_ONCE(memcg->swap.max)) | |
5ccc5aba | 7321 | return true; |
4b82ab4f | 7322 | } |
5ccc5aba VD |
7323 | |
7324 | return false; | |
7325 | } | |
7326 | ||
eccb52e7 | 7327 | static int __init setup_swap_account(char *s) |
21afa38e JW |
7328 | { |
7329 | if (!strcmp(s, "1")) | |
5ab92901 | 7330 | cgroup_memory_noswap = false; |
21afa38e | 7331 | else if (!strcmp(s, "0")) |
5ab92901 | 7332 | cgroup_memory_noswap = true; |
21afa38e JW |
7333 | return 1; |
7334 | } | |
eccb52e7 | 7335 | __setup("swapaccount=", setup_swap_account); |
21afa38e | 7336 | |
37e84351 VD |
7337 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
7338 | struct cftype *cft) | |
7339 | { | |
7340 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
7341 | ||
7342 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
7343 | } | |
7344 | ||
4b82ab4f JK |
7345 | static int swap_high_show(struct seq_file *m, void *v) |
7346 | { | |
7347 | return seq_puts_memcg_tunable(m, | |
7348 | READ_ONCE(mem_cgroup_from_seq(m)->swap.high)); | |
7349 | } | |
7350 | ||
7351 | static ssize_t swap_high_write(struct kernfs_open_file *of, | |
7352 | char *buf, size_t nbytes, loff_t off) | |
7353 | { | |
7354 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7355 | unsigned long high; | |
7356 | int err; | |
7357 | ||
7358 | buf = strstrip(buf); | |
7359 | err = page_counter_memparse(buf, "max", &high); | |
7360 | if (err) | |
7361 | return err; | |
7362 | ||
7363 | page_counter_set_high(&memcg->swap, high); | |
7364 | ||
7365 | return nbytes; | |
7366 | } | |
7367 | ||
37e84351 VD |
7368 | static int swap_max_show(struct seq_file *m, void *v) |
7369 | { | |
677dc973 CD |
7370 | return seq_puts_memcg_tunable(m, |
7371 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); | |
37e84351 VD |
7372 | } |
7373 | ||
7374 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
7375 | char *buf, size_t nbytes, loff_t off) | |
7376 | { | |
7377 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7378 | unsigned long max; | |
7379 | int err; | |
7380 | ||
7381 | buf = strstrip(buf); | |
7382 | err = page_counter_memparse(buf, "max", &max); | |
7383 | if (err) | |
7384 | return err; | |
7385 | ||
be09102b | 7386 | xchg(&memcg->swap.max, max); |
37e84351 VD |
7387 | |
7388 | return nbytes; | |
7389 | } | |
7390 | ||
f3a53a3a TH |
7391 | static int swap_events_show(struct seq_file *m, void *v) |
7392 | { | |
aa9694bb | 7393 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
f3a53a3a | 7394 | |
4b82ab4f JK |
7395 | seq_printf(m, "high %lu\n", |
7396 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_HIGH])); | |
f3a53a3a TH |
7397 | seq_printf(m, "max %lu\n", |
7398 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
7399 | seq_printf(m, "fail %lu\n", | |
7400 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
7401 | ||
7402 | return 0; | |
7403 | } | |
7404 | ||
37e84351 VD |
7405 | static struct cftype swap_files[] = { |
7406 | { | |
7407 | .name = "swap.current", | |
7408 | .flags = CFTYPE_NOT_ON_ROOT, | |
7409 | .read_u64 = swap_current_read, | |
7410 | }, | |
4b82ab4f JK |
7411 | { |
7412 | .name = "swap.high", | |
7413 | .flags = CFTYPE_NOT_ON_ROOT, | |
7414 | .seq_show = swap_high_show, | |
7415 | .write = swap_high_write, | |
7416 | }, | |
37e84351 VD |
7417 | { |
7418 | .name = "swap.max", | |
7419 | .flags = CFTYPE_NOT_ON_ROOT, | |
7420 | .seq_show = swap_max_show, | |
7421 | .write = swap_max_write, | |
7422 | }, | |
f3a53a3a TH |
7423 | { |
7424 | .name = "swap.events", | |
7425 | .flags = CFTYPE_NOT_ON_ROOT, | |
7426 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
7427 | .seq_show = swap_events_show, | |
7428 | }, | |
37e84351 VD |
7429 | { } /* terminate */ |
7430 | }; | |
7431 | ||
eccb52e7 | 7432 | static struct cftype memsw_files[] = { |
21afa38e JW |
7433 | { |
7434 | .name = "memsw.usage_in_bytes", | |
7435 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
7436 | .read_u64 = mem_cgroup_read_u64, | |
7437 | }, | |
7438 | { | |
7439 | .name = "memsw.max_usage_in_bytes", | |
7440 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
7441 | .write = mem_cgroup_reset, | |
7442 | .read_u64 = mem_cgroup_read_u64, | |
7443 | }, | |
7444 | { | |
7445 | .name = "memsw.limit_in_bytes", | |
7446 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
7447 | .write = mem_cgroup_write, | |
7448 | .read_u64 = mem_cgroup_read_u64, | |
7449 | }, | |
7450 | { | |
7451 | .name = "memsw.failcnt", | |
7452 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
7453 | .write = mem_cgroup_reset, | |
7454 | .read_u64 = mem_cgroup_read_u64, | |
7455 | }, | |
7456 | { }, /* terminate */ | |
7457 | }; | |
7458 | ||
82ff165c BS |
7459 | /* |
7460 | * If mem_cgroup_swap_init() is implemented as a subsys_initcall() | |
7461 | * instead of a core_initcall(), this could mean cgroup_memory_noswap still | |
7462 | * remains set to false even when memcg is disabled via "cgroup_disable=memory" | |
7463 | * boot parameter. This may result in premature OOPS inside | |
7464 | * mem_cgroup_get_nr_swap_pages() function in corner cases. | |
7465 | */ | |
21afa38e JW |
7466 | static int __init mem_cgroup_swap_init(void) |
7467 | { | |
2d1c4980 JW |
7468 | /* No memory control -> no swap control */ |
7469 | if (mem_cgroup_disabled()) | |
7470 | cgroup_memory_noswap = true; | |
7471 | ||
7472 | if (cgroup_memory_noswap) | |
eccb52e7 JW |
7473 | return 0; |
7474 | ||
7475 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, swap_files)); | |
7476 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, memsw_files)); | |
7477 | ||
21afa38e JW |
7478 | return 0; |
7479 | } | |
82ff165c | 7480 | core_initcall(mem_cgroup_swap_init); |
21afa38e JW |
7481 | |
7482 | #endif /* CONFIG_MEMCG_SWAP */ |