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 | |
8cdea7c0 BS |
23 | */ |
24 | ||
3e32cb2e | 25 | #include <linux/page_counter.h> |
8cdea7c0 BS |
26 | #include <linux/memcontrol.h> |
27 | #include <linux/cgroup.h> | |
a520110e | 28 | #include <linux/pagewalk.h> |
6e84f315 | 29 | #include <linux/sched/mm.h> |
3a4f8a0b | 30 | #include <linux/shmem_fs.h> |
4ffef5fe | 31 | #include <linux/hugetlb.h> |
d13d1443 | 32 | #include <linux/pagemap.h> |
1ff9e6e1 | 33 | #include <linux/vm_event_item.h> |
d52aa412 | 34 | #include <linux/smp.h> |
8a9f3ccd | 35 | #include <linux/page-flags.h> |
66e1707b | 36 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
37 | #include <linux/bit_spinlock.h> |
38 | #include <linux/rcupdate.h> | |
e222432b | 39 | #include <linux/limits.h> |
b9e15baf | 40 | #include <linux/export.h> |
8c7c6e34 | 41 | #include <linux/mutex.h> |
bb4cc1a8 | 42 | #include <linux/rbtree.h> |
b6ac57d5 | 43 | #include <linux/slab.h> |
66e1707b | 44 | #include <linux/swap.h> |
02491447 | 45 | #include <linux/swapops.h> |
66e1707b | 46 | #include <linux/spinlock.h> |
2e72b634 | 47 | #include <linux/eventfd.h> |
79bd9814 | 48 | #include <linux/poll.h> |
2e72b634 | 49 | #include <linux/sort.h> |
66e1707b | 50 | #include <linux/fs.h> |
d2ceb9b7 | 51 | #include <linux/seq_file.h> |
70ddf637 | 52 | #include <linux/vmpressure.h> |
b69408e8 | 53 | #include <linux/mm_inline.h> |
5d1ea48b | 54 | #include <linux/swap_cgroup.h> |
cdec2e42 | 55 | #include <linux/cpu.h> |
158e0a2d | 56 | #include <linux/oom.h> |
0056f4e6 | 57 | #include <linux/lockdep.h> |
79bd9814 | 58 | #include <linux/file.h> |
b23afb93 | 59 | #include <linux/tracehook.h> |
0e4b01df | 60 | #include <linux/psi.h> |
c8713d0b | 61 | #include <linux/seq_buf.h> |
08e552c6 | 62 | #include "internal.h" |
d1a4c0b3 | 63 | #include <net/sock.h> |
4bd2c1ee | 64 | #include <net/ip.h> |
f35c3a8e | 65 | #include "slab.h" |
8cdea7c0 | 66 | |
7c0f6ba6 | 67 | #include <linux/uaccess.h> |
8697d331 | 68 | |
cc8e970c KM |
69 | #include <trace/events/vmscan.h> |
70 | ||
073219e9 TH |
71 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
72 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 73 | |
7d828602 JW |
74 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
75 | ||
37d5985c RG |
76 | /* Active memory cgroup to use from an interrupt context */ |
77 | DEFINE_PER_CPU(struct mem_cgroup *, int_active_memcg); | |
78 | ||
f7e1cb6e JW |
79 | /* Socket memory accounting disabled? */ |
80 | static bool cgroup_memory_nosocket; | |
81 | ||
04823c83 VD |
82 | /* Kernel memory accounting disabled? */ |
83 | static bool cgroup_memory_nokmem; | |
84 | ||
21afa38e | 85 | /* Whether the swap controller is active */ |
c255a458 | 86 | #ifdef CONFIG_MEMCG_SWAP |
eccb52e7 | 87 | bool cgroup_memory_noswap __read_mostly; |
c077719b | 88 | #else |
eccb52e7 | 89 | #define cgroup_memory_noswap 1 |
2d1c4980 | 90 | #endif |
c077719b | 91 | |
97b27821 TH |
92 | #ifdef CONFIG_CGROUP_WRITEBACK |
93 | static DECLARE_WAIT_QUEUE_HEAD(memcg_cgwb_frn_waitq); | |
94 | #endif | |
95 | ||
7941d214 JW |
96 | /* Whether legacy memory+swap accounting is active */ |
97 | static bool do_memsw_account(void) | |
98 | { | |
eccb52e7 | 99 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_noswap; |
7941d214 JW |
100 | } |
101 | ||
a0db00fc KS |
102 | #define THRESHOLDS_EVENTS_TARGET 128 |
103 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
e9f8974f | 104 | |
bb4cc1a8 AM |
105 | /* |
106 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
107 | * their hierarchy representation | |
108 | */ | |
109 | ||
ef8f2327 | 110 | struct mem_cgroup_tree_per_node { |
bb4cc1a8 | 111 | struct rb_root rb_root; |
fa90b2fd | 112 | struct rb_node *rb_rightmost; |
bb4cc1a8 AM |
113 | spinlock_t lock; |
114 | }; | |
115 | ||
bb4cc1a8 AM |
116 | struct mem_cgroup_tree { |
117 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
118 | }; | |
119 | ||
120 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
121 | ||
9490ff27 KH |
122 | /* for OOM */ |
123 | struct mem_cgroup_eventfd_list { | |
124 | struct list_head list; | |
125 | struct eventfd_ctx *eventfd; | |
126 | }; | |
2e72b634 | 127 | |
79bd9814 TH |
128 | /* |
129 | * cgroup_event represents events which userspace want to receive. | |
130 | */ | |
3bc942f3 | 131 | struct mem_cgroup_event { |
79bd9814 | 132 | /* |
59b6f873 | 133 | * memcg which the event belongs to. |
79bd9814 | 134 | */ |
59b6f873 | 135 | struct mem_cgroup *memcg; |
79bd9814 TH |
136 | /* |
137 | * eventfd to signal userspace about the event. | |
138 | */ | |
139 | struct eventfd_ctx *eventfd; | |
140 | /* | |
141 | * Each of these stored in a list by the cgroup. | |
142 | */ | |
143 | struct list_head list; | |
fba94807 TH |
144 | /* |
145 | * register_event() callback will be used to add new userspace | |
146 | * waiter for changes related to this event. Use eventfd_signal() | |
147 | * on eventfd to send notification to userspace. | |
148 | */ | |
59b6f873 | 149 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 150 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
151 | /* |
152 | * unregister_event() callback will be called when userspace closes | |
153 | * the eventfd or on cgroup removing. This callback must be set, | |
154 | * if you want provide notification functionality. | |
155 | */ | |
59b6f873 | 156 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 157 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
158 | /* |
159 | * All fields below needed to unregister event when | |
160 | * userspace closes eventfd. | |
161 | */ | |
162 | poll_table pt; | |
163 | wait_queue_head_t *wqh; | |
ac6424b9 | 164 | wait_queue_entry_t wait; |
79bd9814 TH |
165 | struct work_struct remove; |
166 | }; | |
167 | ||
c0ff4b85 R |
168 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
169 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 170 | |
7dc74be0 DN |
171 | /* Stuffs for move charges at task migration. */ |
172 | /* | |
1dfab5ab | 173 | * Types of charges to be moved. |
7dc74be0 | 174 | */ |
1dfab5ab JW |
175 | #define MOVE_ANON 0x1U |
176 | #define MOVE_FILE 0x2U | |
177 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 178 | |
4ffef5fe DN |
179 | /* "mc" and its members are protected by cgroup_mutex */ |
180 | static struct move_charge_struct { | |
b1dd693e | 181 | spinlock_t lock; /* for from, to */ |
264a0ae1 | 182 | struct mm_struct *mm; |
4ffef5fe DN |
183 | struct mem_cgroup *from; |
184 | struct mem_cgroup *to; | |
1dfab5ab | 185 | unsigned long flags; |
4ffef5fe | 186 | unsigned long precharge; |
854ffa8d | 187 | unsigned long moved_charge; |
483c30b5 | 188 | unsigned long moved_swap; |
8033b97c DN |
189 | struct task_struct *moving_task; /* a task moving charges */ |
190 | wait_queue_head_t waitq; /* a waitq for other context */ | |
191 | } mc = { | |
2bd9bb20 | 192 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
193 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
194 | }; | |
4ffef5fe | 195 | |
4e416953 BS |
196 | /* |
197 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
198 | * limit reclaim to prevent infinite loops, if they ever occur. | |
199 | */ | |
a0db00fc | 200 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 201 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 202 | |
8c7c6e34 | 203 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
204 | enum res_type { |
205 | _MEM, | |
206 | _MEMSWAP, | |
207 | _OOM_TYPE, | |
510fc4e1 | 208 | _KMEM, |
d55f90bf | 209 | _TCP, |
86ae53e1 GC |
210 | }; |
211 | ||
a0db00fc KS |
212 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
213 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 214 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
215 | /* Used for OOM nofiier */ |
216 | #define OOM_CONTROL (0) | |
8c7c6e34 | 217 | |
b05706f1 KT |
218 | /* |
219 | * Iteration constructs for visiting all cgroups (under a tree). If | |
220 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
221 | * be used for reference counting. | |
222 | */ | |
223 | #define for_each_mem_cgroup_tree(iter, root) \ | |
224 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ | |
225 | iter != NULL; \ | |
226 | iter = mem_cgroup_iter(root, iter, NULL)) | |
227 | ||
228 | #define for_each_mem_cgroup(iter) \ | |
229 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ | |
230 | iter != NULL; \ | |
231 | iter = mem_cgroup_iter(NULL, iter, NULL)) | |
232 | ||
7775face TH |
233 | static inline bool should_force_charge(void) |
234 | { | |
235 | return tsk_is_oom_victim(current) || fatal_signal_pending(current) || | |
236 | (current->flags & PF_EXITING); | |
237 | } | |
238 | ||
70ddf637 AV |
239 | /* Some nice accessors for the vmpressure. */ |
240 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
241 | { | |
242 | if (!memcg) | |
243 | memcg = root_mem_cgroup; | |
244 | return &memcg->vmpressure; | |
245 | } | |
246 | ||
247 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
248 | { | |
249 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
250 | } | |
251 | ||
84c07d11 | 252 | #ifdef CONFIG_MEMCG_KMEM |
bf4f0599 RG |
253 | extern spinlock_t css_set_lock; |
254 | ||
255 | static void obj_cgroup_release(struct percpu_ref *ref) | |
256 | { | |
257 | struct obj_cgroup *objcg = container_of(ref, struct obj_cgroup, refcnt); | |
258 | struct mem_cgroup *memcg; | |
259 | unsigned int nr_bytes; | |
260 | unsigned int nr_pages; | |
261 | unsigned long flags; | |
262 | ||
263 | /* | |
264 | * At this point all allocated objects are freed, and | |
265 | * objcg->nr_charged_bytes can't have an arbitrary byte value. | |
266 | * However, it can be PAGE_SIZE or (x * PAGE_SIZE). | |
267 | * | |
268 | * The following sequence can lead to it: | |
269 | * 1) CPU0: objcg == stock->cached_objcg | |
270 | * 2) CPU1: we do a small allocation (e.g. 92 bytes), | |
271 | * PAGE_SIZE bytes are charged | |
272 | * 3) CPU1: a process from another memcg is allocating something, | |
273 | * the stock if flushed, | |
274 | * objcg->nr_charged_bytes = PAGE_SIZE - 92 | |
275 | * 5) CPU0: we do release this object, | |
276 | * 92 bytes are added to stock->nr_bytes | |
277 | * 6) CPU0: stock is flushed, | |
278 | * 92 bytes are added to objcg->nr_charged_bytes | |
279 | * | |
280 | * In the result, nr_charged_bytes == PAGE_SIZE. | |
281 | * This page will be uncharged in obj_cgroup_release(). | |
282 | */ | |
283 | nr_bytes = atomic_read(&objcg->nr_charged_bytes); | |
284 | WARN_ON_ONCE(nr_bytes & (PAGE_SIZE - 1)); | |
285 | nr_pages = nr_bytes >> PAGE_SHIFT; | |
286 | ||
287 | spin_lock_irqsave(&css_set_lock, flags); | |
288 | memcg = obj_cgroup_memcg(objcg); | |
289 | if (nr_pages) | |
290 | __memcg_kmem_uncharge(memcg, nr_pages); | |
291 | list_del(&objcg->list); | |
292 | mem_cgroup_put(memcg); | |
293 | spin_unlock_irqrestore(&css_set_lock, flags); | |
294 | ||
295 | percpu_ref_exit(ref); | |
296 | kfree_rcu(objcg, rcu); | |
297 | } | |
298 | ||
299 | static struct obj_cgroup *obj_cgroup_alloc(void) | |
300 | { | |
301 | struct obj_cgroup *objcg; | |
302 | int ret; | |
303 | ||
304 | objcg = kzalloc(sizeof(struct obj_cgroup), GFP_KERNEL); | |
305 | if (!objcg) | |
306 | return NULL; | |
307 | ||
308 | ret = percpu_ref_init(&objcg->refcnt, obj_cgroup_release, 0, | |
309 | GFP_KERNEL); | |
310 | if (ret) { | |
311 | kfree(objcg); | |
312 | return NULL; | |
313 | } | |
314 | INIT_LIST_HEAD(&objcg->list); | |
315 | return objcg; | |
316 | } | |
317 | ||
318 | static void memcg_reparent_objcgs(struct mem_cgroup *memcg, | |
319 | struct mem_cgroup *parent) | |
320 | { | |
321 | struct obj_cgroup *objcg, *iter; | |
322 | ||
323 | objcg = rcu_replace_pointer(memcg->objcg, NULL, true); | |
324 | ||
325 | spin_lock_irq(&css_set_lock); | |
326 | ||
327 | /* Move active objcg to the parent's list */ | |
328 | xchg(&objcg->memcg, parent); | |
329 | css_get(&parent->css); | |
330 | list_add(&objcg->list, &parent->objcg_list); | |
331 | ||
332 | /* Move already reparented objcgs to the parent's list */ | |
333 | list_for_each_entry(iter, &memcg->objcg_list, list) { | |
334 | css_get(&parent->css); | |
335 | xchg(&iter->memcg, parent); | |
336 | css_put(&memcg->css); | |
337 | } | |
338 | list_splice(&memcg->objcg_list, &parent->objcg_list); | |
339 | ||
340 | spin_unlock_irq(&css_set_lock); | |
341 | ||
342 | percpu_ref_kill(&objcg->refcnt); | |
343 | } | |
344 | ||
55007d84 | 345 | /* |
9855609b | 346 | * This will be used as a shrinker list's index. |
b8627835 LZ |
347 | * The main reason for not using cgroup id for this: |
348 | * this works better in sparse environments, where we have a lot of memcgs, | |
349 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
350 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
351 | * 200 entry array for that. | |
55007d84 | 352 | * |
dbcf73e2 VD |
353 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
354 | * will double each time we have to increase it. | |
55007d84 | 355 | */ |
dbcf73e2 VD |
356 | static DEFINE_IDA(memcg_cache_ida); |
357 | int memcg_nr_cache_ids; | |
749c5415 | 358 | |
05257a1a VD |
359 | /* Protects memcg_nr_cache_ids */ |
360 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
361 | ||
362 | void memcg_get_cache_ids(void) | |
363 | { | |
364 | down_read(&memcg_cache_ids_sem); | |
365 | } | |
366 | ||
367 | void memcg_put_cache_ids(void) | |
368 | { | |
369 | up_read(&memcg_cache_ids_sem); | |
370 | } | |
371 | ||
55007d84 GC |
372 | /* |
373 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
374 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
375 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
376 | * tunable, but that is strictly not necessary. | |
377 | * | |
b8627835 | 378 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
379 | * this constant directly from cgroup, but it is understandable that this is |
380 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 381 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
382 | * increase ours as well if it increases. |
383 | */ | |
384 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 385 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 386 | |
d7f25f8a GC |
387 | /* |
388 | * A lot of the calls to the cache allocation functions are expected to be | |
272911a4 | 389 | * inlined by the compiler. Since the calls to memcg_slab_pre_alloc_hook() are |
d7f25f8a GC |
390 | * conditional to this static branch, we'll have to allow modules that does |
391 | * kmem_cache_alloc and the such to see this symbol as well | |
392 | */ | |
ef12947c | 393 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
d7f25f8a | 394 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
0a432dcb | 395 | #endif |
17cc4dfe | 396 | |
0a4465d3 KT |
397 | static int memcg_shrinker_map_size; |
398 | static DEFINE_MUTEX(memcg_shrinker_map_mutex); | |
399 | ||
400 | static void memcg_free_shrinker_map_rcu(struct rcu_head *head) | |
401 | { | |
402 | kvfree(container_of(head, struct memcg_shrinker_map, rcu)); | |
403 | } | |
404 | ||
405 | static int memcg_expand_one_shrinker_map(struct mem_cgroup *memcg, | |
406 | int size, int old_size) | |
407 | { | |
408 | struct memcg_shrinker_map *new, *old; | |
409 | int nid; | |
410 | ||
411 | lockdep_assert_held(&memcg_shrinker_map_mutex); | |
412 | ||
413 | for_each_node(nid) { | |
414 | old = rcu_dereference_protected( | |
415 | mem_cgroup_nodeinfo(memcg, nid)->shrinker_map, true); | |
416 | /* Not yet online memcg */ | |
417 | if (!old) | |
418 | return 0; | |
419 | ||
86daf94e | 420 | new = kvmalloc_node(sizeof(*new) + size, GFP_KERNEL, nid); |
0a4465d3 KT |
421 | if (!new) |
422 | return -ENOMEM; | |
423 | ||
424 | /* Set all old bits, clear all new bits */ | |
425 | memset(new->map, (int)0xff, old_size); | |
426 | memset((void *)new->map + old_size, 0, size - old_size); | |
427 | ||
428 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, new); | |
429 | call_rcu(&old->rcu, memcg_free_shrinker_map_rcu); | |
430 | } | |
431 | ||
432 | return 0; | |
433 | } | |
434 | ||
435 | static void memcg_free_shrinker_maps(struct mem_cgroup *memcg) | |
436 | { | |
437 | struct mem_cgroup_per_node *pn; | |
438 | struct memcg_shrinker_map *map; | |
439 | int nid; | |
440 | ||
441 | if (mem_cgroup_is_root(memcg)) | |
442 | return; | |
443 | ||
444 | for_each_node(nid) { | |
445 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
446 | map = rcu_dereference_protected(pn->shrinker_map, true); | |
447 | if (map) | |
448 | kvfree(map); | |
449 | rcu_assign_pointer(pn->shrinker_map, NULL); | |
450 | } | |
451 | } | |
452 | ||
453 | static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg) | |
454 | { | |
455 | struct memcg_shrinker_map *map; | |
456 | int nid, size, ret = 0; | |
457 | ||
458 | if (mem_cgroup_is_root(memcg)) | |
459 | return 0; | |
460 | ||
461 | mutex_lock(&memcg_shrinker_map_mutex); | |
462 | size = memcg_shrinker_map_size; | |
463 | for_each_node(nid) { | |
86daf94e | 464 | map = kvzalloc_node(sizeof(*map) + size, GFP_KERNEL, nid); |
0a4465d3 KT |
465 | if (!map) { |
466 | memcg_free_shrinker_maps(memcg); | |
467 | ret = -ENOMEM; | |
468 | break; | |
469 | } | |
470 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, map); | |
471 | } | |
472 | mutex_unlock(&memcg_shrinker_map_mutex); | |
473 | ||
474 | return ret; | |
475 | } | |
476 | ||
477 | int memcg_expand_shrinker_maps(int new_id) | |
478 | { | |
479 | int size, old_size, ret = 0; | |
480 | struct mem_cgroup *memcg; | |
481 | ||
482 | size = DIV_ROUND_UP(new_id + 1, BITS_PER_LONG) * sizeof(unsigned long); | |
483 | old_size = memcg_shrinker_map_size; | |
484 | if (size <= old_size) | |
485 | return 0; | |
486 | ||
487 | mutex_lock(&memcg_shrinker_map_mutex); | |
488 | if (!root_mem_cgroup) | |
489 | goto unlock; | |
490 | ||
491 | for_each_mem_cgroup(memcg) { | |
492 | if (mem_cgroup_is_root(memcg)) | |
493 | continue; | |
494 | ret = memcg_expand_one_shrinker_map(memcg, size, old_size); | |
75866af6 VA |
495 | if (ret) { |
496 | mem_cgroup_iter_break(NULL, memcg); | |
0a4465d3 | 497 | goto unlock; |
75866af6 | 498 | } |
0a4465d3 KT |
499 | } |
500 | unlock: | |
501 | if (!ret) | |
502 | memcg_shrinker_map_size = size; | |
503 | mutex_unlock(&memcg_shrinker_map_mutex); | |
504 | return ret; | |
505 | } | |
fae91d6d KT |
506 | |
507 | void memcg_set_shrinker_bit(struct mem_cgroup *memcg, int nid, int shrinker_id) | |
508 | { | |
509 | if (shrinker_id >= 0 && memcg && !mem_cgroup_is_root(memcg)) { | |
510 | struct memcg_shrinker_map *map; | |
511 | ||
512 | rcu_read_lock(); | |
513 | map = rcu_dereference(memcg->nodeinfo[nid]->shrinker_map); | |
f90280d6 KT |
514 | /* Pairs with smp mb in shrink_slab() */ |
515 | smp_mb__before_atomic(); | |
fae91d6d KT |
516 | set_bit(shrinker_id, map->map); |
517 | rcu_read_unlock(); | |
518 | } | |
519 | } | |
520 | ||
ad7fa852 TH |
521 | /** |
522 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
523 | * @page: page of interest | |
524 | * | |
525 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
526 | * with @page is returned. The returned css remains associated with @page | |
527 | * until it is released. | |
528 | * | |
529 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
530 | * is returned. | |
ad7fa852 TH |
531 | */ |
532 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
533 | { | |
534 | struct mem_cgroup *memcg; | |
535 | ||
ad7fa852 TH |
536 | memcg = page->mem_cgroup; |
537 | ||
9e10a130 | 538 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
539 | memcg = root_mem_cgroup; |
540 | ||
ad7fa852 TH |
541 | return &memcg->css; |
542 | } | |
543 | ||
2fc04524 VD |
544 | /** |
545 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
546 | * @page: the page | |
547 | * | |
548 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
549 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
550 | * is safe to call this function without holding a reference to @page. | |
551 | * | |
552 | * Note, this function is inherently racy, because there is nothing to prevent | |
553 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
554 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
555 | * do not care (such as procfs interfaces). | |
556 | */ | |
557 | ino_t page_cgroup_ino(struct page *page) | |
558 | { | |
559 | struct mem_cgroup *memcg; | |
560 | unsigned long ino = 0; | |
561 | ||
562 | rcu_read_lock(); | |
9855609b | 563 | memcg = page->mem_cgroup; |
286e04b8 | 564 | |
9855609b RG |
565 | /* |
566 | * The lowest bit set means that memcg isn't a valid | |
567 | * memcg pointer, but a obj_cgroups pointer. | |
568 | * In this case the page is shared and doesn't belong | |
569 | * to any specific memory cgroup. | |
570 | */ | |
571 | if ((unsigned long) memcg & 0x1UL) | |
572 | memcg = NULL; | |
286e04b8 | 573 | |
2fc04524 VD |
574 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) |
575 | memcg = parent_mem_cgroup(memcg); | |
576 | if (memcg) | |
577 | ino = cgroup_ino(memcg->css.cgroup); | |
578 | rcu_read_unlock(); | |
579 | return ino; | |
580 | } | |
581 | ||
ef8f2327 MG |
582 | static struct mem_cgroup_per_node * |
583 | mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page) | |
f64c3f54 | 584 | { |
97a6c37b | 585 | int nid = page_to_nid(page); |
f64c3f54 | 586 | |
ef8f2327 | 587 | return memcg->nodeinfo[nid]; |
f64c3f54 BS |
588 | } |
589 | ||
ef8f2327 MG |
590 | static struct mem_cgroup_tree_per_node * |
591 | soft_limit_tree_node(int nid) | |
bb4cc1a8 | 592 | { |
ef8f2327 | 593 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
594 | } |
595 | ||
ef8f2327 | 596 | static struct mem_cgroup_tree_per_node * |
bb4cc1a8 AM |
597 | soft_limit_tree_from_page(struct page *page) |
598 | { | |
599 | int nid = page_to_nid(page); | |
bb4cc1a8 | 600 | |
ef8f2327 | 601 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
602 | } |
603 | ||
ef8f2327 MG |
604 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
605 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 606 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
607 | { |
608 | struct rb_node **p = &mctz->rb_root.rb_node; | |
609 | struct rb_node *parent = NULL; | |
ef8f2327 | 610 | struct mem_cgroup_per_node *mz_node; |
fa90b2fd | 611 | bool rightmost = true; |
bb4cc1a8 AM |
612 | |
613 | if (mz->on_tree) | |
614 | return; | |
615 | ||
616 | mz->usage_in_excess = new_usage_in_excess; | |
617 | if (!mz->usage_in_excess) | |
618 | return; | |
619 | while (*p) { | |
620 | parent = *p; | |
ef8f2327 | 621 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 | 622 | tree_node); |
fa90b2fd | 623 | if (mz->usage_in_excess < mz_node->usage_in_excess) { |
bb4cc1a8 | 624 | p = &(*p)->rb_left; |
fa90b2fd DB |
625 | rightmost = false; |
626 | } | |
627 | ||
bb4cc1a8 AM |
628 | /* |
629 | * We can't avoid mem cgroups that are over their soft | |
630 | * limit by the same amount | |
631 | */ | |
632 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
633 | p = &(*p)->rb_right; | |
634 | } | |
fa90b2fd DB |
635 | |
636 | if (rightmost) | |
637 | mctz->rb_rightmost = &mz->tree_node; | |
638 | ||
bb4cc1a8 AM |
639 | rb_link_node(&mz->tree_node, parent, p); |
640 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
641 | mz->on_tree = true; | |
642 | } | |
643 | ||
ef8f2327 MG |
644 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
645 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
646 | { |
647 | if (!mz->on_tree) | |
648 | return; | |
fa90b2fd DB |
649 | |
650 | if (&mz->tree_node == mctz->rb_rightmost) | |
651 | mctz->rb_rightmost = rb_prev(&mz->tree_node); | |
652 | ||
bb4cc1a8 AM |
653 | rb_erase(&mz->tree_node, &mctz->rb_root); |
654 | mz->on_tree = false; | |
655 | } | |
656 | ||
ef8f2327 MG |
657 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
658 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 659 | { |
0a31bc97 JW |
660 | unsigned long flags; |
661 | ||
662 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 663 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 664 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
665 | } |
666 | ||
3e32cb2e JW |
667 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
668 | { | |
669 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 670 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
671 | unsigned long excess = 0; |
672 | ||
673 | if (nr_pages > soft_limit) | |
674 | excess = nr_pages - soft_limit; | |
675 | ||
676 | return excess; | |
677 | } | |
bb4cc1a8 AM |
678 | |
679 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
680 | { | |
3e32cb2e | 681 | unsigned long excess; |
ef8f2327 MG |
682 | struct mem_cgroup_per_node *mz; |
683 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 684 | |
e231875b | 685 | mctz = soft_limit_tree_from_page(page); |
bfc7228b LD |
686 | if (!mctz) |
687 | return; | |
bb4cc1a8 AM |
688 | /* |
689 | * Necessary to update all ancestors when hierarchy is used. | |
690 | * because their event counter is not touched. | |
691 | */ | |
692 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
ef8f2327 | 693 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
3e32cb2e | 694 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
695 | /* |
696 | * We have to update the tree if mz is on RB-tree or | |
697 | * mem is over its softlimit. | |
698 | */ | |
699 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
700 | unsigned long flags; |
701 | ||
702 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
703 | /* if on-tree, remove it */ |
704 | if (mz->on_tree) | |
cf2c8127 | 705 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
706 | /* |
707 | * Insert again. mz->usage_in_excess will be updated. | |
708 | * If excess is 0, no tree ops. | |
709 | */ | |
cf2c8127 | 710 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 711 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
712 | } |
713 | } | |
714 | } | |
715 | ||
716 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
717 | { | |
ef8f2327 MG |
718 | struct mem_cgroup_tree_per_node *mctz; |
719 | struct mem_cgroup_per_node *mz; | |
720 | int nid; | |
bb4cc1a8 | 721 | |
e231875b | 722 | for_each_node(nid) { |
ef8f2327 MG |
723 | mz = mem_cgroup_nodeinfo(memcg, nid); |
724 | mctz = soft_limit_tree_node(nid); | |
bfc7228b LD |
725 | if (mctz) |
726 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
727 | } |
728 | } | |
729 | ||
ef8f2327 MG |
730 | static struct mem_cgroup_per_node * |
731 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 732 | { |
ef8f2327 | 733 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
734 | |
735 | retry: | |
736 | mz = NULL; | |
fa90b2fd | 737 | if (!mctz->rb_rightmost) |
bb4cc1a8 AM |
738 | goto done; /* Nothing to reclaim from */ |
739 | ||
fa90b2fd DB |
740 | mz = rb_entry(mctz->rb_rightmost, |
741 | struct mem_cgroup_per_node, tree_node); | |
bb4cc1a8 AM |
742 | /* |
743 | * Remove the node now but someone else can add it back, | |
744 | * we will to add it back at the end of reclaim to its correct | |
745 | * position in the tree. | |
746 | */ | |
cf2c8127 | 747 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 748 | if (!soft_limit_excess(mz->memcg) || |
8965aa28 | 749 | !css_tryget(&mz->memcg->css)) |
bb4cc1a8 AM |
750 | goto retry; |
751 | done: | |
752 | return mz; | |
753 | } | |
754 | ||
ef8f2327 MG |
755 | static struct mem_cgroup_per_node * |
756 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 757 | { |
ef8f2327 | 758 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 759 | |
0a31bc97 | 760 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 761 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 762 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
763 | return mz; |
764 | } | |
765 | ||
db9adbcb JW |
766 | /** |
767 | * __mod_memcg_state - update cgroup memory statistics | |
768 | * @memcg: the memory cgroup | |
769 | * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item | |
770 | * @val: delta to add to the counter, can be negative | |
771 | */ | |
772 | void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val) | |
773 | { | |
ea426c2a | 774 | long x, threshold = MEMCG_CHARGE_BATCH; |
db9adbcb JW |
775 | |
776 | if (mem_cgroup_disabled()) | |
777 | return; | |
778 | ||
772616b0 | 779 | if (memcg_stat_item_in_bytes(idx)) |
ea426c2a RG |
780 | threshold <<= PAGE_SHIFT; |
781 | ||
db9adbcb | 782 | x = val + __this_cpu_read(memcg->vmstats_percpu->stat[idx]); |
ea426c2a | 783 | if (unlikely(abs(x) > threshold)) { |
42a30035 JW |
784 | struct mem_cgroup *mi; |
785 | ||
766a4c19 YS |
786 | /* |
787 | * Batch local counters to keep them in sync with | |
788 | * the hierarchical ones. | |
789 | */ | |
790 | __this_cpu_add(memcg->vmstats_local->stat[idx], x); | |
42a30035 JW |
791 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
792 | atomic_long_add(x, &mi->vmstats[idx]); | |
db9adbcb JW |
793 | x = 0; |
794 | } | |
795 | __this_cpu_write(memcg->vmstats_percpu->stat[idx], x); | |
796 | } | |
797 | ||
42a30035 JW |
798 | static struct mem_cgroup_per_node * |
799 | parent_nodeinfo(struct mem_cgroup_per_node *pn, int nid) | |
800 | { | |
801 | struct mem_cgroup *parent; | |
802 | ||
803 | parent = parent_mem_cgroup(pn->memcg); | |
804 | if (!parent) | |
805 | return NULL; | |
806 | return mem_cgroup_nodeinfo(parent, nid); | |
807 | } | |
808 | ||
eedc4e5a RG |
809 | void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, |
810 | int val) | |
db9adbcb JW |
811 | { |
812 | struct mem_cgroup_per_node *pn; | |
42a30035 | 813 | struct mem_cgroup *memcg; |
ea426c2a | 814 | long x, threshold = MEMCG_CHARGE_BATCH; |
db9adbcb | 815 | |
db9adbcb | 816 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
42a30035 | 817 | memcg = pn->memcg; |
db9adbcb JW |
818 | |
819 | /* Update memcg */ | |
42a30035 | 820 | __mod_memcg_state(memcg, idx, val); |
db9adbcb | 821 | |
b4c46484 RG |
822 | /* Update lruvec */ |
823 | __this_cpu_add(pn->lruvec_stat_local->count[idx], val); | |
824 | ||
ea426c2a RG |
825 | if (vmstat_item_in_bytes(idx)) |
826 | threshold <<= PAGE_SHIFT; | |
827 | ||
db9adbcb | 828 | x = val + __this_cpu_read(pn->lruvec_stat_cpu->count[idx]); |
ea426c2a | 829 | if (unlikely(abs(x) > threshold)) { |
eedc4e5a | 830 | pg_data_t *pgdat = lruvec_pgdat(lruvec); |
42a30035 JW |
831 | struct mem_cgroup_per_node *pi; |
832 | ||
42a30035 JW |
833 | for (pi = pn; pi; pi = parent_nodeinfo(pi, pgdat->node_id)) |
834 | atomic_long_add(x, &pi->lruvec_stat[idx]); | |
db9adbcb JW |
835 | x = 0; |
836 | } | |
837 | __this_cpu_write(pn->lruvec_stat_cpu->count[idx], x); | |
838 | } | |
839 | ||
eedc4e5a RG |
840 | /** |
841 | * __mod_lruvec_state - update lruvec memory statistics | |
842 | * @lruvec: the lruvec | |
843 | * @idx: the stat item | |
844 | * @val: delta to add to the counter, can be negative | |
845 | * | |
846 | * The lruvec is the intersection of the NUMA node and a cgroup. This | |
847 | * function updates the all three counters that are affected by a | |
848 | * change of state at this level: per-node, per-cgroup, per-lruvec. | |
849 | */ | |
850 | void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, | |
851 | int val) | |
852 | { | |
853 | /* Update node */ | |
854 | __mod_node_page_state(lruvec_pgdat(lruvec), idx, val); | |
855 | ||
856 | /* Update memcg and lruvec */ | |
857 | if (!mem_cgroup_disabled()) | |
858 | __mod_memcg_lruvec_state(lruvec, idx, val); | |
859 | } | |
860 | ||
ec9f0238 RG |
861 | void __mod_lruvec_slab_state(void *p, enum node_stat_item idx, int val) |
862 | { | |
4f103c63 | 863 | pg_data_t *pgdat = page_pgdat(virt_to_page(p)); |
ec9f0238 RG |
864 | struct mem_cgroup *memcg; |
865 | struct lruvec *lruvec; | |
866 | ||
867 | rcu_read_lock(); | |
4f103c63 | 868 | memcg = mem_cgroup_from_obj(p); |
ec9f0238 RG |
869 | |
870 | /* Untracked pages have no memcg, no lruvec. Update only the node */ | |
871 | if (!memcg || memcg == root_mem_cgroup) { | |
872 | __mod_node_page_state(pgdat, idx, val); | |
873 | } else { | |
867e5e1d | 874 | lruvec = mem_cgroup_lruvec(memcg, pgdat); |
ec9f0238 RG |
875 | __mod_lruvec_state(lruvec, idx, val); |
876 | } | |
877 | rcu_read_unlock(); | |
878 | } | |
879 | ||
8380ce47 RG |
880 | void mod_memcg_obj_state(void *p, int idx, int val) |
881 | { | |
882 | struct mem_cgroup *memcg; | |
883 | ||
884 | rcu_read_lock(); | |
885 | memcg = mem_cgroup_from_obj(p); | |
886 | if (memcg) | |
887 | mod_memcg_state(memcg, idx, val); | |
888 | rcu_read_unlock(); | |
889 | } | |
890 | ||
db9adbcb JW |
891 | /** |
892 | * __count_memcg_events - account VM events in a cgroup | |
893 | * @memcg: the memory cgroup | |
894 | * @idx: the event item | |
895 | * @count: the number of events that occured | |
896 | */ | |
897 | void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, | |
898 | unsigned long count) | |
899 | { | |
900 | unsigned long x; | |
901 | ||
902 | if (mem_cgroup_disabled()) | |
903 | return; | |
904 | ||
905 | x = count + __this_cpu_read(memcg->vmstats_percpu->events[idx]); | |
906 | if (unlikely(x > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
907 | struct mem_cgroup *mi; |
908 | ||
766a4c19 YS |
909 | /* |
910 | * Batch local counters to keep them in sync with | |
911 | * the hierarchical ones. | |
912 | */ | |
913 | __this_cpu_add(memcg->vmstats_local->events[idx], x); | |
42a30035 JW |
914 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
915 | atomic_long_add(x, &mi->vmevents[idx]); | |
db9adbcb JW |
916 | x = 0; |
917 | } | |
918 | __this_cpu_write(memcg->vmstats_percpu->events[idx], x); | |
919 | } | |
920 | ||
42a30035 | 921 | static unsigned long memcg_events(struct mem_cgroup *memcg, int event) |
e9f8974f | 922 | { |
871789d4 | 923 | return atomic_long_read(&memcg->vmevents[event]); |
e9f8974f JW |
924 | } |
925 | ||
42a30035 JW |
926 | static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event) |
927 | { | |
815744d7 JW |
928 | long x = 0; |
929 | int cpu; | |
930 | ||
931 | for_each_possible_cpu(cpu) | |
932 | x += per_cpu(memcg->vmstats_local->events[event], cpu); | |
933 | return x; | |
42a30035 JW |
934 | } |
935 | ||
c0ff4b85 | 936 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 937 | struct page *page, |
3fba69a5 | 938 | int nr_pages) |
d52aa412 | 939 | { |
e401f176 KH |
940 | /* pagein of a big page is an event. So, ignore page size */ |
941 | if (nr_pages > 0) | |
c9019e9b | 942 | __count_memcg_events(memcg, PGPGIN, 1); |
3751d604 | 943 | else { |
c9019e9b | 944 | __count_memcg_events(memcg, PGPGOUT, 1); |
3751d604 KH |
945 | nr_pages = -nr_pages; /* for event */ |
946 | } | |
e401f176 | 947 | |
871789d4 | 948 | __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages); |
6d12e2d8 KH |
949 | } |
950 | ||
f53d7ce3 JW |
951 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
952 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
953 | { |
954 | unsigned long val, next; | |
955 | ||
871789d4 CD |
956 | val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events); |
957 | next = __this_cpu_read(memcg->vmstats_percpu->targets[target]); | |
7a159cc9 | 958 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 959 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
960 | switch (target) { |
961 | case MEM_CGROUP_TARGET_THRESH: | |
962 | next = val + THRESHOLDS_EVENTS_TARGET; | |
963 | break; | |
bb4cc1a8 AM |
964 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
965 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
966 | break; | |
f53d7ce3 JW |
967 | default: |
968 | break; | |
969 | } | |
871789d4 | 970 | __this_cpu_write(memcg->vmstats_percpu->targets[target], next); |
f53d7ce3 | 971 | return true; |
7a159cc9 | 972 | } |
f53d7ce3 | 973 | return false; |
d2265e6f KH |
974 | } |
975 | ||
976 | /* | |
977 | * Check events in order. | |
978 | * | |
979 | */ | |
c0ff4b85 | 980 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
981 | { |
982 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
983 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
984 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 985 | bool do_softlimit; |
f53d7ce3 | 986 | |
bb4cc1a8 AM |
987 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
988 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
c0ff4b85 | 989 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
990 | if (unlikely(do_softlimit)) |
991 | mem_cgroup_update_tree(memcg, page); | |
0a31bc97 | 992 | } |
d2265e6f KH |
993 | } |
994 | ||
cf475ad2 | 995 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 996 | { |
31a78f23 BS |
997 | /* |
998 | * mm_update_next_owner() may clear mm->owner to NULL | |
999 | * if it races with swapoff, page migration, etc. | |
1000 | * So this can be called with p == NULL. | |
1001 | */ | |
1002 | if (unlikely(!p)) | |
1003 | return NULL; | |
1004 | ||
073219e9 | 1005 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 1006 | } |
33398cf2 | 1007 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 1008 | |
d46eb14b SB |
1009 | /** |
1010 | * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg. | |
1011 | * @mm: mm from which memcg should be extracted. It can be NULL. | |
1012 | * | |
1013 | * Obtain a reference on mm->memcg and returns it if successful. Otherwise | |
1014 | * root_mem_cgroup is returned. However if mem_cgroup is disabled, NULL is | |
1015 | * returned. | |
1016 | */ | |
1017 | struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) | |
54595fe2 | 1018 | { |
d46eb14b SB |
1019 | struct mem_cgroup *memcg; |
1020 | ||
1021 | if (mem_cgroup_disabled()) | |
1022 | return NULL; | |
0b7f569e | 1023 | |
54595fe2 KH |
1024 | rcu_read_lock(); |
1025 | do { | |
6f6acb00 MH |
1026 | /* |
1027 | * Page cache insertions can happen withou an | |
1028 | * actual mm context, e.g. during disk probing | |
1029 | * on boot, loopback IO, acct() writes etc. | |
1030 | */ | |
1031 | if (unlikely(!mm)) | |
df381975 | 1032 | memcg = root_mem_cgroup; |
6f6acb00 MH |
1033 | else { |
1034 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
1035 | if (unlikely(!memcg)) | |
1036 | memcg = root_mem_cgroup; | |
1037 | } | |
00d484f3 | 1038 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 1039 | rcu_read_unlock(); |
c0ff4b85 | 1040 | return memcg; |
54595fe2 | 1041 | } |
d46eb14b SB |
1042 | EXPORT_SYMBOL(get_mem_cgroup_from_mm); |
1043 | ||
f745c6f5 SB |
1044 | /** |
1045 | * get_mem_cgroup_from_page: Obtain a reference on given page's memcg. | |
1046 | * @page: page from which memcg should be extracted. | |
1047 | * | |
1048 | * Obtain a reference on page->memcg and returns it if successful. Otherwise | |
1049 | * root_mem_cgroup is returned. | |
1050 | */ | |
1051 | struct mem_cgroup *get_mem_cgroup_from_page(struct page *page) | |
1052 | { | |
1053 | struct mem_cgroup *memcg = page->mem_cgroup; | |
1054 | ||
1055 | if (mem_cgroup_disabled()) | |
1056 | return NULL; | |
1057 | ||
1058 | rcu_read_lock(); | |
8965aa28 SB |
1059 | /* Page should not get uncharged and freed memcg under us. */ |
1060 | if (!memcg || WARN_ON_ONCE(!css_tryget(&memcg->css))) | |
f745c6f5 SB |
1061 | memcg = root_mem_cgroup; |
1062 | rcu_read_unlock(); | |
1063 | return memcg; | |
1064 | } | |
1065 | EXPORT_SYMBOL(get_mem_cgroup_from_page); | |
1066 | ||
37d5985c | 1067 | static __always_inline struct mem_cgroup *active_memcg(void) |
d46eb14b | 1068 | { |
37d5985c RG |
1069 | if (in_interrupt()) |
1070 | return this_cpu_read(int_active_memcg); | |
1071 | else | |
1072 | return current->active_memcg; | |
1073 | } | |
279c3393 | 1074 | |
37d5985c RG |
1075 | static __always_inline struct mem_cgroup *get_active_memcg(void) |
1076 | { | |
1077 | struct mem_cgroup *memcg; | |
d46eb14b | 1078 | |
37d5985c RG |
1079 | rcu_read_lock(); |
1080 | memcg = active_memcg(); | |
1081 | if (memcg) { | |
8965aa28 | 1082 | /* current->active_memcg must hold a ref. */ |
37d5985c | 1083 | if (WARN_ON_ONCE(!css_tryget(&memcg->css))) |
8965aa28 SB |
1084 | memcg = root_mem_cgroup; |
1085 | else | |
d46eb14b | 1086 | memcg = current->active_memcg; |
d46eb14b | 1087 | } |
37d5985c RG |
1088 | rcu_read_unlock(); |
1089 | ||
1090 | return memcg; | |
1091 | } | |
1092 | ||
4127c650 RG |
1093 | static __always_inline bool memcg_kmem_bypass(void) |
1094 | { | |
1095 | /* Allow remote memcg charging from any context. */ | |
1096 | if (unlikely(active_memcg())) | |
1097 | return false; | |
1098 | ||
1099 | /* Memcg to charge can't be determined. */ | |
1100 | if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD)) | |
1101 | return true; | |
1102 | ||
1103 | return false; | |
1104 | } | |
1105 | ||
37d5985c RG |
1106 | /** |
1107 | * If active memcg is set, do not fallback to current->mm->memcg. | |
1108 | */ | |
1109 | static __always_inline struct mem_cgroup *get_mem_cgroup_from_current(void) | |
1110 | { | |
1111 | if (memcg_kmem_bypass()) | |
1112 | return NULL; | |
1113 | ||
1114 | if (unlikely(active_memcg())) | |
1115 | return get_active_memcg(); | |
1116 | ||
d46eb14b SB |
1117 | return get_mem_cgroup_from_mm(current->mm); |
1118 | } | |
54595fe2 | 1119 | |
5660048c JW |
1120 | /** |
1121 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1122 | * @root: hierarchy root | |
1123 | * @prev: previously returned memcg, NULL on first invocation | |
1124 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1125 | * | |
1126 | * Returns references to children of the hierarchy below @root, or | |
1127 | * @root itself, or %NULL after a full round-trip. | |
1128 | * | |
1129 | * Caller must pass the return value in @prev on subsequent | |
1130 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1131 | * to cancel a hierarchy walk before the round-trip is complete. | |
1132 | * | |
05bdc520 ML |
1133 | * Reclaimers can specify a node in @reclaim to divide up the memcgs |
1134 | * in the hierarchy among all concurrent reclaimers operating on the | |
1135 | * same node. | |
5660048c | 1136 | */ |
694fbc0f | 1137 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1138 | struct mem_cgroup *prev, |
694fbc0f | 1139 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1140 | { |
3f649ab7 | 1141 | struct mem_cgroup_reclaim_iter *iter; |
5ac8fb31 | 1142 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 1143 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1144 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1145 | |
694fbc0f AM |
1146 | if (mem_cgroup_disabled()) |
1147 | return NULL; | |
5660048c | 1148 | |
9f3a0d09 JW |
1149 | if (!root) |
1150 | root = root_mem_cgroup; | |
7d74b06f | 1151 | |
9f3a0d09 | 1152 | if (prev && !reclaim) |
5ac8fb31 | 1153 | pos = prev; |
14067bb3 | 1154 | |
9f3a0d09 JW |
1155 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1156 | if (prev) | |
5ac8fb31 | 1157 | goto out; |
694fbc0f | 1158 | return root; |
9f3a0d09 | 1159 | } |
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(); | |
5ac8fb31 | 1240 | out: |
c40046f3 MH |
1241 | if (prev && prev != root) |
1242 | css_put(&prev->css); | |
1243 | ||
9f3a0d09 | 1244 | return memcg; |
14067bb3 | 1245 | } |
7d74b06f | 1246 | |
5660048c JW |
1247 | /** |
1248 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1249 | * @root: hierarchy root | |
1250 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1251 | */ | |
1252 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1253 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1254 | { |
1255 | if (!root) | |
1256 | root = root_mem_cgroup; | |
1257 | if (prev && prev != root) | |
1258 | css_put(&prev->css); | |
1259 | } | |
7d74b06f | 1260 | |
54a83d6b MC |
1261 | static void __invalidate_reclaim_iterators(struct mem_cgroup *from, |
1262 | struct mem_cgroup *dead_memcg) | |
6df38689 | 1263 | { |
6df38689 | 1264 | struct mem_cgroup_reclaim_iter *iter; |
ef8f2327 MG |
1265 | struct mem_cgroup_per_node *mz; |
1266 | int nid; | |
6df38689 | 1267 | |
54a83d6b MC |
1268 | for_each_node(nid) { |
1269 | mz = mem_cgroup_nodeinfo(from, nid); | |
9da83f3f YS |
1270 | iter = &mz->iter; |
1271 | cmpxchg(&iter->position, dead_memcg, NULL); | |
6df38689 VD |
1272 | } |
1273 | } | |
1274 | ||
54a83d6b MC |
1275 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1276 | { | |
1277 | struct mem_cgroup *memcg = dead_memcg; | |
1278 | struct mem_cgroup *last; | |
1279 | ||
1280 | do { | |
1281 | __invalidate_reclaim_iterators(memcg, dead_memcg); | |
1282 | last = memcg; | |
1283 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1284 | ||
1285 | /* | |
1286 | * When cgruop1 non-hierarchy mode is used, | |
1287 | * parent_mem_cgroup() does not walk all the way up to the | |
1288 | * cgroup root (root_mem_cgroup). So we have to handle | |
1289 | * dead_memcg from cgroup root separately. | |
1290 | */ | |
1291 | if (last != root_mem_cgroup) | |
1292 | __invalidate_reclaim_iterators(root_mem_cgroup, | |
1293 | dead_memcg); | |
1294 | } | |
1295 | ||
7c5f64f8 VD |
1296 | /** |
1297 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1298 | * @memcg: hierarchy root | |
1299 | * @fn: function to call for each task | |
1300 | * @arg: argument passed to @fn | |
1301 | * | |
1302 | * This function iterates over tasks attached to @memcg or to any of its | |
1303 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
1304 | * value, the function breaks the iteration loop and returns the value. | |
1305 | * Otherwise, it will iterate over all tasks and return 0. | |
1306 | * | |
1307 | * This function must not be called for the root memory cgroup. | |
1308 | */ | |
1309 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
1310 | int (*fn)(struct task_struct *, void *), void *arg) | |
1311 | { | |
1312 | struct mem_cgroup *iter; | |
1313 | int ret = 0; | |
1314 | ||
1315 | BUG_ON(memcg == root_mem_cgroup); | |
1316 | ||
1317 | for_each_mem_cgroup_tree(iter, memcg) { | |
1318 | struct css_task_iter it; | |
1319 | struct task_struct *task; | |
1320 | ||
f168a9a5 | 1321 | css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it); |
7c5f64f8 VD |
1322 | while (!ret && (task = css_task_iter_next(&it))) |
1323 | ret = fn(task, arg); | |
1324 | css_task_iter_end(&it); | |
1325 | if (ret) { | |
1326 | mem_cgroup_iter_break(memcg, iter); | |
1327 | break; | |
1328 | } | |
1329 | } | |
1330 | return ret; | |
1331 | } | |
1332 | ||
925b7673 | 1333 | /** |
dfe0e773 | 1334 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1335 | * @page: the page |
f144c390 | 1336 | * @pgdat: pgdat of the page |
dfe0e773 | 1337 | * |
a0b5b414 JW |
1338 | * This function relies on page->mem_cgroup being stable - see the |
1339 | * access rules in commit_charge(). | |
925b7673 | 1340 | */ |
599d0c95 | 1341 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat) |
08e552c6 | 1342 | { |
ef8f2327 | 1343 | struct mem_cgroup_per_node *mz; |
925b7673 | 1344 | struct mem_cgroup *memcg; |
bea8c150 | 1345 | struct lruvec *lruvec; |
6d12e2d8 | 1346 | |
bea8c150 | 1347 | if (mem_cgroup_disabled()) { |
867e5e1d | 1348 | lruvec = &pgdat->__lruvec; |
bea8c150 HD |
1349 | goto out; |
1350 | } | |
925b7673 | 1351 | |
1306a85a | 1352 | memcg = page->mem_cgroup; |
7512102c | 1353 | /* |
dfe0e773 | 1354 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1355 | * possibly migrated - before they are charged. |
7512102c | 1356 | */ |
29833315 JW |
1357 | if (!memcg) |
1358 | memcg = root_mem_cgroup; | |
7512102c | 1359 | |
ef8f2327 | 1360 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
bea8c150 HD |
1361 | lruvec = &mz->lruvec; |
1362 | out: | |
1363 | /* | |
1364 | * Since a node can be onlined after the mem_cgroup was created, | |
1365 | * we have to be prepared to initialize lruvec->zone here; | |
1366 | * and if offlined then reonlined, we need to reinitialize it. | |
1367 | */ | |
599d0c95 MG |
1368 | if (unlikely(lruvec->pgdat != pgdat)) |
1369 | lruvec->pgdat = pgdat; | |
bea8c150 | 1370 | return lruvec; |
08e552c6 | 1371 | } |
b69408e8 | 1372 | |
925b7673 | 1373 | /** |
fa9add64 HD |
1374 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1375 | * @lruvec: mem_cgroup per zone lru vector | |
1376 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1377 | * @zid: zone id of the accounted pages |
fa9add64 | 1378 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1379 | * |
ca707239 HD |
1380 | * This function must be called under lru_lock, just before a page is added |
1381 | * to or just after a page is removed from an lru list (that ordering being | |
1382 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 1383 | */ |
fa9add64 | 1384 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1385 | int zid, int nr_pages) |
3f58a829 | 1386 | { |
ef8f2327 | 1387 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1388 | unsigned long *lru_size; |
ca707239 | 1389 | long size; |
3f58a829 MK |
1390 | |
1391 | if (mem_cgroup_disabled()) | |
1392 | return; | |
1393 | ||
ef8f2327 | 1394 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1395 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1396 | |
1397 | if (nr_pages < 0) | |
1398 | *lru_size += nr_pages; | |
1399 | ||
1400 | size = *lru_size; | |
b4536f0c MH |
1401 | if (WARN_ONCE(size < 0, |
1402 | "%s(%p, %d, %d): lru_size %ld\n", | |
1403 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1404 | VM_BUG_ON(1); |
1405 | *lru_size = 0; | |
1406 | } | |
1407 | ||
1408 | if (nr_pages > 0) | |
1409 | *lru_size += nr_pages; | |
08e552c6 | 1410 | } |
544122e5 | 1411 | |
19942822 | 1412 | /** |
9d11ea9f | 1413 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1414 | * @memcg: the memory cgroup |
19942822 | 1415 | * |
9d11ea9f | 1416 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1417 | * pages. |
19942822 | 1418 | */ |
c0ff4b85 | 1419 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1420 | { |
3e32cb2e JW |
1421 | unsigned long margin = 0; |
1422 | unsigned long count; | |
1423 | unsigned long limit; | |
9d11ea9f | 1424 | |
3e32cb2e | 1425 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1426 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1427 | if (count < limit) |
1428 | margin = limit - count; | |
1429 | ||
7941d214 | 1430 | if (do_memsw_account()) { |
3e32cb2e | 1431 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1432 | limit = READ_ONCE(memcg->memsw.max); |
1c4448ed | 1433 | if (count < limit) |
3e32cb2e | 1434 | margin = min(margin, limit - count); |
cbedbac3 LR |
1435 | else |
1436 | margin = 0; | |
3e32cb2e JW |
1437 | } |
1438 | ||
1439 | return margin; | |
19942822 JW |
1440 | } |
1441 | ||
32047e2a | 1442 | /* |
bdcbb659 | 1443 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1444 | * |
bdcbb659 QH |
1445 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1446 | * moving cgroups. This is for waiting at high-memory pressure | |
1447 | * caused by "move". | |
32047e2a | 1448 | */ |
c0ff4b85 | 1449 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1450 | { |
2bd9bb20 KH |
1451 | struct mem_cgroup *from; |
1452 | struct mem_cgroup *to; | |
4b534334 | 1453 | bool ret = false; |
2bd9bb20 KH |
1454 | /* |
1455 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1456 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1457 | */ | |
1458 | spin_lock(&mc.lock); | |
1459 | from = mc.from; | |
1460 | to = mc.to; | |
1461 | if (!from) | |
1462 | goto unlock; | |
3e92041d | 1463 | |
2314b42d JW |
1464 | ret = mem_cgroup_is_descendant(from, memcg) || |
1465 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1466 | unlock: |
1467 | spin_unlock(&mc.lock); | |
4b534334 KH |
1468 | return ret; |
1469 | } | |
1470 | ||
c0ff4b85 | 1471 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1472 | { |
1473 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1474 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1475 | DEFINE_WAIT(wait); |
1476 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1477 | /* moving charge context might have finished. */ | |
1478 | if (mc.moving_task) | |
1479 | schedule(); | |
1480 | finish_wait(&mc.waitq, &wait); | |
1481 | return true; | |
1482 | } | |
1483 | } | |
1484 | return false; | |
1485 | } | |
1486 | ||
5f9a4f4a MS |
1487 | struct memory_stat { |
1488 | const char *name; | |
1489 | unsigned int ratio; | |
1490 | unsigned int idx; | |
1491 | }; | |
1492 | ||
1493 | static struct memory_stat memory_stats[] = { | |
1494 | { "anon", PAGE_SIZE, NR_ANON_MAPPED }, | |
1495 | { "file", PAGE_SIZE, NR_FILE_PAGES }, | |
1496 | { "kernel_stack", 1024, NR_KERNEL_STACK_KB }, | |
1497 | { "percpu", 1, MEMCG_PERCPU_B }, | |
1498 | { "sock", PAGE_SIZE, MEMCG_SOCK }, | |
1499 | { "shmem", PAGE_SIZE, NR_SHMEM }, | |
1500 | { "file_mapped", PAGE_SIZE, NR_FILE_MAPPED }, | |
1501 | { "file_dirty", PAGE_SIZE, NR_FILE_DIRTY }, | |
1502 | { "file_writeback", PAGE_SIZE, NR_WRITEBACK }, | |
1503 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
1504 | /* | |
1505 | * The ratio will be initialized in memory_stats_init(). Because | |
1506 | * on some architectures, the macro of HPAGE_PMD_SIZE is not | |
1507 | * constant(e.g. powerpc). | |
1508 | */ | |
1509 | { "anon_thp", 0, NR_ANON_THPS }, | |
1510 | #endif | |
1511 | { "inactive_anon", PAGE_SIZE, NR_INACTIVE_ANON }, | |
1512 | { "active_anon", PAGE_SIZE, NR_ACTIVE_ANON }, | |
1513 | { "inactive_file", PAGE_SIZE, NR_INACTIVE_FILE }, | |
1514 | { "active_file", PAGE_SIZE, NR_ACTIVE_FILE }, | |
1515 | { "unevictable", PAGE_SIZE, NR_UNEVICTABLE }, | |
1516 | ||
1517 | /* | |
1518 | * Note: The slab_reclaimable and slab_unreclaimable must be | |
1519 | * together and slab_reclaimable must be in front. | |
1520 | */ | |
1521 | { "slab_reclaimable", 1, NR_SLAB_RECLAIMABLE_B }, | |
1522 | { "slab_unreclaimable", 1, NR_SLAB_UNRECLAIMABLE_B }, | |
1523 | ||
1524 | /* The memory events */ | |
1525 | { "workingset_refault_anon", 1, WORKINGSET_REFAULT_ANON }, | |
1526 | { "workingset_refault_file", 1, WORKINGSET_REFAULT_FILE }, | |
1527 | { "workingset_activate_anon", 1, WORKINGSET_ACTIVATE_ANON }, | |
1528 | { "workingset_activate_file", 1, WORKINGSET_ACTIVATE_FILE }, | |
1529 | { "workingset_restore_anon", 1, WORKINGSET_RESTORE_ANON }, | |
1530 | { "workingset_restore_file", 1, WORKINGSET_RESTORE_FILE }, | |
1531 | { "workingset_nodereclaim", 1, WORKINGSET_NODERECLAIM }, | |
1532 | }; | |
1533 | ||
1534 | static int __init memory_stats_init(void) | |
1535 | { | |
1536 | int i; | |
1537 | ||
1538 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { | |
1539 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
1540 | if (memory_stats[i].idx == NR_ANON_THPS) | |
1541 | memory_stats[i].ratio = HPAGE_PMD_SIZE; | |
1542 | #endif | |
1543 | VM_BUG_ON(!memory_stats[i].ratio); | |
1544 | VM_BUG_ON(memory_stats[i].idx >= MEMCG_NR_STAT); | |
1545 | } | |
1546 | ||
1547 | return 0; | |
1548 | } | |
1549 | pure_initcall(memory_stats_init); | |
1550 | ||
c8713d0b JW |
1551 | static char *memory_stat_format(struct mem_cgroup *memcg) |
1552 | { | |
1553 | struct seq_buf s; | |
1554 | int i; | |
71cd3113 | 1555 | |
c8713d0b JW |
1556 | seq_buf_init(&s, kmalloc(PAGE_SIZE, GFP_KERNEL), PAGE_SIZE); |
1557 | if (!s.buffer) | |
1558 | return NULL; | |
1559 | ||
1560 | /* | |
1561 | * Provide statistics on the state of the memory subsystem as | |
1562 | * well as cumulative event counters that show past behavior. | |
1563 | * | |
1564 | * This list is ordered following a combination of these gradients: | |
1565 | * 1) generic big picture -> specifics and details | |
1566 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
1567 | * | |
1568 | * Current memory state: | |
1569 | */ | |
1570 | ||
5f9a4f4a MS |
1571 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { |
1572 | u64 size; | |
c8713d0b | 1573 | |
5f9a4f4a MS |
1574 | size = memcg_page_state(memcg, memory_stats[i].idx); |
1575 | size *= memory_stats[i].ratio; | |
1576 | seq_buf_printf(&s, "%s %llu\n", memory_stats[i].name, size); | |
c8713d0b | 1577 | |
5f9a4f4a MS |
1578 | if (unlikely(memory_stats[i].idx == NR_SLAB_UNRECLAIMABLE_B)) { |
1579 | size = memcg_page_state(memcg, NR_SLAB_RECLAIMABLE_B) + | |
1580 | memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE_B); | |
1581 | seq_buf_printf(&s, "slab %llu\n", size); | |
1582 | } | |
1583 | } | |
c8713d0b JW |
1584 | |
1585 | /* Accumulated memory events */ | |
1586 | ||
ebc5d83d KK |
1587 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGFAULT), |
1588 | memcg_events(memcg, PGFAULT)); | |
1589 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGMAJFAULT), | |
1590 | memcg_events(memcg, PGMAJFAULT)); | |
ebc5d83d KK |
1591 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGREFILL), |
1592 | memcg_events(memcg, PGREFILL)); | |
c8713d0b JW |
1593 | seq_buf_printf(&s, "pgscan %lu\n", |
1594 | memcg_events(memcg, PGSCAN_KSWAPD) + | |
1595 | memcg_events(memcg, PGSCAN_DIRECT)); | |
1596 | seq_buf_printf(&s, "pgsteal %lu\n", | |
1597 | memcg_events(memcg, PGSTEAL_KSWAPD) + | |
1598 | memcg_events(memcg, PGSTEAL_DIRECT)); | |
ebc5d83d KK |
1599 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGACTIVATE), |
1600 | memcg_events(memcg, PGACTIVATE)); | |
1601 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGDEACTIVATE), | |
1602 | memcg_events(memcg, PGDEACTIVATE)); | |
1603 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREE), | |
1604 | memcg_events(memcg, PGLAZYFREE)); | |
1605 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREED), | |
1606 | memcg_events(memcg, PGLAZYFREED)); | |
c8713d0b JW |
1607 | |
1608 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
ebc5d83d | 1609 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_FAULT_ALLOC), |
c8713d0b | 1610 | memcg_events(memcg, THP_FAULT_ALLOC)); |
ebc5d83d | 1611 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_COLLAPSE_ALLOC), |
c8713d0b JW |
1612 | memcg_events(memcg, THP_COLLAPSE_ALLOC)); |
1613 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ | |
1614 | ||
1615 | /* The above should easily fit into one page */ | |
1616 | WARN_ON_ONCE(seq_buf_has_overflowed(&s)); | |
1617 | ||
1618 | return s.buffer; | |
1619 | } | |
71cd3113 | 1620 | |
58cf188e | 1621 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1622 | /** |
f0c867d9 | 1623 | * mem_cgroup_print_oom_context: Print OOM information relevant to |
1624 | * memory controller. | |
e222432b BS |
1625 | * @memcg: The memory cgroup that went over limit |
1626 | * @p: Task that is going to be killed | |
1627 | * | |
1628 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1629 | * enabled | |
1630 | */ | |
f0c867d9 | 1631 | void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) |
e222432b | 1632 | { |
e222432b BS |
1633 | rcu_read_lock(); |
1634 | ||
f0c867d9 | 1635 | if (memcg) { |
1636 | pr_cont(",oom_memcg="); | |
1637 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1638 | } else | |
1639 | pr_cont(",global_oom"); | |
2415b9f5 | 1640 | if (p) { |
f0c867d9 | 1641 | pr_cont(",task_memcg="); |
2415b9f5 | 1642 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
2415b9f5 | 1643 | } |
e222432b | 1644 | rcu_read_unlock(); |
f0c867d9 | 1645 | } |
1646 | ||
1647 | /** | |
1648 | * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to | |
1649 | * memory controller. | |
1650 | * @memcg: The memory cgroup that went over limit | |
1651 | */ | |
1652 | void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) | |
1653 | { | |
c8713d0b | 1654 | char *buf; |
e222432b | 1655 | |
3e32cb2e JW |
1656 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1657 | K((u64)page_counter_read(&memcg->memory)), | |
15b42562 | 1658 | K((u64)READ_ONCE(memcg->memory.max)), memcg->memory.failcnt); |
c8713d0b JW |
1659 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
1660 | pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1661 | K((u64)page_counter_read(&memcg->swap)), | |
32d087cd | 1662 | K((u64)READ_ONCE(memcg->swap.max)), memcg->swap.failcnt); |
c8713d0b JW |
1663 | else { |
1664 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1665 | K((u64)page_counter_read(&memcg->memsw)), | |
1666 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); | |
1667 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1668 | K((u64)page_counter_read(&memcg->kmem)), | |
1669 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); | |
58cf188e | 1670 | } |
c8713d0b JW |
1671 | |
1672 | pr_info("Memory cgroup stats for "); | |
1673 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1674 | pr_cont(":"); | |
1675 | buf = memory_stat_format(memcg); | |
1676 | if (!buf) | |
1677 | return; | |
1678 | pr_info("%s", buf); | |
1679 | kfree(buf); | |
e222432b BS |
1680 | } |
1681 | ||
a63d83f4 DR |
1682 | /* |
1683 | * Return the memory (and swap, if configured) limit for a memcg. | |
1684 | */ | |
bbec2e15 | 1685 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1686 | { |
8d387a5f WL |
1687 | unsigned long max = READ_ONCE(memcg->memory.max); |
1688 | ||
1689 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) { | |
1690 | if (mem_cgroup_swappiness(memcg)) | |
1691 | max += min(READ_ONCE(memcg->swap.max), | |
1692 | (unsigned long)total_swap_pages); | |
1693 | } else { /* v1 */ | |
1694 | if (mem_cgroup_swappiness(memcg)) { | |
1695 | /* Calculate swap excess capacity from memsw limit */ | |
1696 | unsigned long swap = READ_ONCE(memcg->memsw.max) - max; | |
1697 | ||
1698 | max += min(swap, (unsigned long)total_swap_pages); | |
1699 | } | |
9a5a8f19 | 1700 | } |
bbec2e15 | 1701 | return max; |
a63d83f4 DR |
1702 | } |
1703 | ||
9783aa99 CD |
1704 | unsigned long mem_cgroup_size(struct mem_cgroup *memcg) |
1705 | { | |
1706 | return page_counter_read(&memcg->memory); | |
1707 | } | |
1708 | ||
b6e6edcf | 1709 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1710 | int order) |
9cbb78bb | 1711 | { |
6e0fc46d DR |
1712 | struct oom_control oc = { |
1713 | .zonelist = NULL, | |
1714 | .nodemask = NULL, | |
2a966b77 | 1715 | .memcg = memcg, |
6e0fc46d DR |
1716 | .gfp_mask = gfp_mask, |
1717 | .order = order, | |
6e0fc46d | 1718 | }; |
1378b37d | 1719 | bool ret = true; |
9cbb78bb | 1720 | |
7775face TH |
1721 | if (mutex_lock_killable(&oom_lock)) |
1722 | return true; | |
1378b37d YS |
1723 | |
1724 | if (mem_cgroup_margin(memcg) >= (1 << order)) | |
1725 | goto unlock; | |
1726 | ||
7775face TH |
1727 | /* |
1728 | * A few threads which were not waiting at mutex_lock_killable() can | |
1729 | * fail to bail out. Therefore, check again after holding oom_lock. | |
1730 | */ | |
1731 | ret = should_force_charge() || out_of_memory(&oc); | |
1378b37d YS |
1732 | |
1733 | unlock: | |
dc56401f | 1734 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1735 | return ret; |
9cbb78bb DR |
1736 | } |
1737 | ||
0608f43d | 1738 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1739 | pg_data_t *pgdat, |
0608f43d AM |
1740 | gfp_t gfp_mask, |
1741 | unsigned long *total_scanned) | |
1742 | { | |
1743 | struct mem_cgroup *victim = NULL; | |
1744 | int total = 0; | |
1745 | int loop = 0; | |
1746 | unsigned long excess; | |
1747 | unsigned long nr_scanned; | |
1748 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1749 | .pgdat = pgdat, |
0608f43d AM |
1750 | }; |
1751 | ||
3e32cb2e | 1752 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1753 | |
1754 | while (1) { | |
1755 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1756 | if (!victim) { | |
1757 | loop++; | |
1758 | if (loop >= 2) { | |
1759 | /* | |
1760 | * If we have not been able to reclaim | |
1761 | * anything, it might because there are | |
1762 | * no reclaimable pages under this hierarchy | |
1763 | */ | |
1764 | if (!total) | |
1765 | break; | |
1766 | /* | |
1767 | * We want to do more targeted reclaim. | |
1768 | * excess >> 2 is not to excessive so as to | |
1769 | * reclaim too much, nor too less that we keep | |
1770 | * coming back to reclaim from this cgroup | |
1771 | */ | |
1772 | if (total >= (excess >> 2) || | |
1773 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1774 | break; | |
1775 | } | |
1776 | continue; | |
1777 | } | |
a9dd0a83 | 1778 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1779 | pgdat, &nr_scanned); |
0608f43d | 1780 | *total_scanned += nr_scanned; |
3e32cb2e | 1781 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1782 | break; |
6d61ef40 | 1783 | } |
0608f43d AM |
1784 | mem_cgroup_iter_break(root_memcg, victim); |
1785 | return total; | |
6d61ef40 BS |
1786 | } |
1787 | ||
0056f4e6 JW |
1788 | #ifdef CONFIG_LOCKDEP |
1789 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1790 | .name = "memcg_oom_lock", | |
1791 | }; | |
1792 | #endif | |
1793 | ||
fb2a6fc5 JW |
1794 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1795 | ||
867578cb KH |
1796 | /* |
1797 | * Check OOM-Killer is already running under our hierarchy. | |
1798 | * If someone is running, return false. | |
1799 | */ | |
fb2a6fc5 | 1800 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1801 | { |
79dfdacc | 1802 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1803 | |
fb2a6fc5 JW |
1804 | spin_lock(&memcg_oom_lock); |
1805 | ||
9f3a0d09 | 1806 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1807 | if (iter->oom_lock) { |
79dfdacc MH |
1808 | /* |
1809 | * this subtree of our hierarchy is already locked | |
1810 | * so we cannot give a lock. | |
1811 | */ | |
79dfdacc | 1812 | failed = iter; |
9f3a0d09 JW |
1813 | mem_cgroup_iter_break(memcg, iter); |
1814 | break; | |
23751be0 JW |
1815 | } else |
1816 | iter->oom_lock = true; | |
7d74b06f | 1817 | } |
867578cb | 1818 | |
fb2a6fc5 JW |
1819 | if (failed) { |
1820 | /* | |
1821 | * OK, we failed to lock the whole subtree so we have | |
1822 | * to clean up what we set up to the failing subtree | |
1823 | */ | |
1824 | for_each_mem_cgroup_tree(iter, memcg) { | |
1825 | if (iter == failed) { | |
1826 | mem_cgroup_iter_break(memcg, iter); | |
1827 | break; | |
1828 | } | |
1829 | iter->oom_lock = false; | |
79dfdacc | 1830 | } |
0056f4e6 JW |
1831 | } else |
1832 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1833 | |
1834 | spin_unlock(&memcg_oom_lock); | |
1835 | ||
1836 | return !failed; | |
a636b327 | 1837 | } |
0b7f569e | 1838 | |
fb2a6fc5 | 1839 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1840 | { |
7d74b06f KH |
1841 | struct mem_cgroup *iter; |
1842 | ||
fb2a6fc5 | 1843 | spin_lock(&memcg_oom_lock); |
5facae4f | 1844 | mutex_release(&memcg_oom_lock_dep_map, _RET_IP_); |
c0ff4b85 | 1845 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1846 | iter->oom_lock = false; |
fb2a6fc5 | 1847 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1848 | } |
1849 | ||
c0ff4b85 | 1850 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1851 | { |
1852 | struct mem_cgroup *iter; | |
1853 | ||
c2b42d3c | 1854 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1855 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1856 | iter->under_oom++; |
1857 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1858 | } |
1859 | ||
c0ff4b85 | 1860 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1861 | { |
1862 | struct mem_cgroup *iter; | |
1863 | ||
867578cb | 1864 | /* |
7a52d4d8 ML |
1865 | * Be careful about under_oom underflows becase a child memcg |
1866 | * could have been added after mem_cgroup_mark_under_oom. | |
867578cb | 1867 | */ |
c2b42d3c | 1868 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1869 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1870 | if (iter->under_oom > 0) |
1871 | iter->under_oom--; | |
1872 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1873 | } |
1874 | ||
867578cb KH |
1875 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1876 | ||
dc98df5a | 1877 | struct oom_wait_info { |
d79154bb | 1878 | struct mem_cgroup *memcg; |
ac6424b9 | 1879 | wait_queue_entry_t wait; |
dc98df5a KH |
1880 | }; |
1881 | ||
ac6424b9 | 1882 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1883 | unsigned mode, int sync, void *arg) |
1884 | { | |
d79154bb HD |
1885 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1886 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1887 | struct oom_wait_info *oom_wait_info; |
1888 | ||
1889 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1890 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1891 | |
2314b42d JW |
1892 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1893 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1894 | return 0; |
dc98df5a KH |
1895 | return autoremove_wake_function(wait, mode, sync, arg); |
1896 | } | |
1897 | ||
c0ff4b85 | 1898 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1899 | { |
c2b42d3c TH |
1900 | /* |
1901 | * For the following lockless ->under_oom test, the only required | |
1902 | * guarantee is that it must see the state asserted by an OOM when | |
1903 | * this function is called as a result of userland actions | |
1904 | * triggered by the notification of the OOM. This is trivially | |
1905 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1906 | * triggering notification. | |
1907 | */ | |
1908 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1909 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1910 | } |
1911 | ||
29ef680a MH |
1912 | enum oom_status { |
1913 | OOM_SUCCESS, | |
1914 | OOM_FAILED, | |
1915 | OOM_ASYNC, | |
1916 | OOM_SKIPPED | |
1917 | }; | |
1918 | ||
1919 | static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 1920 | { |
7056d3a3 MH |
1921 | enum oom_status ret; |
1922 | bool locked; | |
1923 | ||
29ef680a MH |
1924 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
1925 | return OOM_SKIPPED; | |
1926 | ||
7a1adfdd RG |
1927 | memcg_memory_event(memcg, MEMCG_OOM); |
1928 | ||
867578cb | 1929 | /* |
49426420 JW |
1930 | * We are in the middle of the charge context here, so we |
1931 | * don't want to block when potentially sitting on a callstack | |
1932 | * that holds all kinds of filesystem and mm locks. | |
1933 | * | |
29ef680a MH |
1934 | * cgroup1 allows disabling the OOM killer and waiting for outside |
1935 | * handling until the charge can succeed; remember the context and put | |
1936 | * the task to sleep at the end of the page fault when all locks are | |
1937 | * released. | |
49426420 | 1938 | * |
29ef680a MH |
1939 | * On the other hand, in-kernel OOM killer allows for an async victim |
1940 | * memory reclaim (oom_reaper) and that means that we are not solely | |
1941 | * relying on the oom victim to make a forward progress and we can | |
1942 | * invoke the oom killer here. | |
1943 | * | |
1944 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
1945 | * victim and then we have to bail out from the charge path. | |
867578cb | 1946 | */ |
29ef680a MH |
1947 | if (memcg->oom_kill_disable) { |
1948 | if (!current->in_user_fault) | |
1949 | return OOM_SKIPPED; | |
1950 | css_get(&memcg->css); | |
1951 | current->memcg_in_oom = memcg; | |
1952 | current->memcg_oom_gfp_mask = mask; | |
1953 | current->memcg_oom_order = order; | |
1954 | ||
1955 | return OOM_ASYNC; | |
1956 | } | |
1957 | ||
7056d3a3 MH |
1958 | mem_cgroup_mark_under_oom(memcg); |
1959 | ||
1960 | locked = mem_cgroup_oom_trylock(memcg); | |
1961 | ||
1962 | if (locked) | |
1963 | mem_cgroup_oom_notify(memcg); | |
1964 | ||
1965 | mem_cgroup_unmark_under_oom(memcg); | |
29ef680a | 1966 | if (mem_cgroup_out_of_memory(memcg, mask, order)) |
7056d3a3 MH |
1967 | ret = OOM_SUCCESS; |
1968 | else | |
1969 | ret = OOM_FAILED; | |
1970 | ||
1971 | if (locked) | |
1972 | mem_cgroup_oom_unlock(memcg); | |
29ef680a | 1973 | |
7056d3a3 | 1974 | return ret; |
3812c8c8 JW |
1975 | } |
1976 | ||
1977 | /** | |
1978 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1979 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1980 | * |
49426420 JW |
1981 | * This has to be called at the end of a page fault if the memcg OOM |
1982 | * handler was enabled. | |
3812c8c8 | 1983 | * |
49426420 | 1984 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1985 | * sleep on a waitqueue until the userspace task resolves the |
1986 | * situation. Sleeping directly in the charge context with all kinds | |
1987 | * of locks held is not a good idea, instead we remember an OOM state | |
1988 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1989 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1990 | * |
1991 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1992 | * completed, %false otherwise. |
3812c8c8 | 1993 | */ |
49426420 | 1994 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1995 | { |
626ebc41 | 1996 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1997 | struct oom_wait_info owait; |
49426420 | 1998 | bool locked; |
3812c8c8 JW |
1999 | |
2000 | /* OOM is global, do not handle */ | |
3812c8c8 | 2001 | if (!memcg) |
49426420 | 2002 | return false; |
3812c8c8 | 2003 | |
7c5f64f8 | 2004 | if (!handle) |
49426420 | 2005 | goto cleanup; |
3812c8c8 JW |
2006 | |
2007 | owait.memcg = memcg; | |
2008 | owait.wait.flags = 0; | |
2009 | owait.wait.func = memcg_oom_wake_function; | |
2010 | owait.wait.private = current; | |
2055da97 | 2011 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 2012 | |
3812c8c8 | 2013 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
2014 | mem_cgroup_mark_under_oom(memcg); |
2015 | ||
2016 | locked = mem_cgroup_oom_trylock(memcg); | |
2017 | ||
2018 | if (locked) | |
2019 | mem_cgroup_oom_notify(memcg); | |
2020 | ||
2021 | if (locked && !memcg->oom_kill_disable) { | |
2022 | mem_cgroup_unmark_under_oom(memcg); | |
2023 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
2024 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
2025 | current->memcg_oom_order); | |
49426420 | 2026 | } else { |
3812c8c8 | 2027 | schedule(); |
49426420 JW |
2028 | mem_cgroup_unmark_under_oom(memcg); |
2029 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2030 | } | |
2031 | ||
2032 | if (locked) { | |
fb2a6fc5 JW |
2033 | mem_cgroup_oom_unlock(memcg); |
2034 | /* | |
2035 | * There is no guarantee that an OOM-lock contender | |
2036 | * sees the wakeups triggered by the OOM kill | |
2037 | * uncharges. Wake any sleepers explicitely. | |
2038 | */ | |
2039 | memcg_oom_recover(memcg); | |
2040 | } | |
49426420 | 2041 | cleanup: |
626ebc41 | 2042 | current->memcg_in_oom = NULL; |
3812c8c8 | 2043 | css_put(&memcg->css); |
867578cb | 2044 | return true; |
0b7f569e KH |
2045 | } |
2046 | ||
3d8b38eb RG |
2047 | /** |
2048 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
2049 | * @victim: task to be killed by the OOM killer | |
2050 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
2051 | * | |
2052 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
2053 | * by killing all belonging OOM-killable tasks. | |
2054 | * | |
2055 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
2056 | */ | |
2057 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
2058 | struct mem_cgroup *oom_domain) | |
2059 | { | |
2060 | struct mem_cgroup *oom_group = NULL; | |
2061 | struct mem_cgroup *memcg; | |
2062 | ||
2063 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
2064 | return NULL; | |
2065 | ||
2066 | if (!oom_domain) | |
2067 | oom_domain = root_mem_cgroup; | |
2068 | ||
2069 | rcu_read_lock(); | |
2070 | ||
2071 | memcg = mem_cgroup_from_task(victim); | |
2072 | if (memcg == root_mem_cgroup) | |
2073 | goto out; | |
2074 | ||
48fe267c RG |
2075 | /* |
2076 | * If the victim task has been asynchronously moved to a different | |
2077 | * memory cgroup, we might end up killing tasks outside oom_domain. | |
2078 | * In this case it's better to ignore memory.group.oom. | |
2079 | */ | |
2080 | if (unlikely(!mem_cgroup_is_descendant(memcg, oom_domain))) | |
2081 | goto out; | |
2082 | ||
3d8b38eb RG |
2083 | /* |
2084 | * Traverse the memory cgroup hierarchy from the victim task's | |
2085 | * cgroup up to the OOMing cgroup (or root) to find the | |
2086 | * highest-level memory cgroup with oom.group set. | |
2087 | */ | |
2088 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
2089 | if (memcg->oom_group) | |
2090 | oom_group = memcg; | |
2091 | ||
2092 | if (memcg == oom_domain) | |
2093 | break; | |
2094 | } | |
2095 | ||
2096 | if (oom_group) | |
2097 | css_get(&oom_group->css); | |
2098 | out: | |
2099 | rcu_read_unlock(); | |
2100 | ||
2101 | return oom_group; | |
2102 | } | |
2103 | ||
2104 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
2105 | { | |
2106 | pr_info("Tasks in "); | |
2107 | pr_cont_cgroup_path(memcg->css.cgroup); | |
2108 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
2109 | } | |
2110 | ||
d7365e78 | 2111 | /** |
81f8c3a4 JW |
2112 | * lock_page_memcg - lock a page->mem_cgroup binding |
2113 | * @page: the page | |
32047e2a | 2114 | * |
81f8c3a4 | 2115 | * This function protects unlocked LRU pages from being moved to |
739f79fc JW |
2116 | * another cgroup. |
2117 | * | |
2118 | * It ensures lifetime of the returned memcg. Caller is responsible | |
2119 | * for the lifetime of the page; __unlock_page_memcg() is available | |
2120 | * when @page might get freed inside the locked section. | |
d69b042f | 2121 | */ |
739f79fc | 2122 | struct mem_cgroup *lock_page_memcg(struct page *page) |
89c06bd5 | 2123 | { |
9da7b521 | 2124 | struct page *head = compound_head(page); /* rmap on tail pages */ |
89c06bd5 | 2125 | struct mem_cgroup *memcg; |
6de22619 | 2126 | unsigned long flags; |
89c06bd5 | 2127 | |
6de22619 JW |
2128 | /* |
2129 | * The RCU lock is held throughout the transaction. The fast | |
2130 | * path can get away without acquiring the memcg->move_lock | |
2131 | * because page moving starts with an RCU grace period. | |
739f79fc JW |
2132 | * |
2133 | * The RCU lock also protects the memcg from being freed when | |
2134 | * the page state that is going to change is the only thing | |
2135 | * preventing the page itself from being freed. E.g. writeback | |
2136 | * doesn't hold a page reference and relies on PG_writeback to | |
2137 | * keep off truncation, migration and so forth. | |
2138 | */ | |
d7365e78 JW |
2139 | rcu_read_lock(); |
2140 | ||
2141 | if (mem_cgroup_disabled()) | |
739f79fc | 2142 | return NULL; |
89c06bd5 | 2143 | again: |
9da7b521 | 2144 | memcg = head->mem_cgroup; |
29833315 | 2145 | if (unlikely(!memcg)) |
739f79fc | 2146 | return NULL; |
d7365e78 | 2147 | |
bdcbb659 | 2148 | if (atomic_read(&memcg->moving_account) <= 0) |
739f79fc | 2149 | return memcg; |
89c06bd5 | 2150 | |
6de22619 | 2151 | spin_lock_irqsave(&memcg->move_lock, flags); |
9da7b521 | 2152 | if (memcg != head->mem_cgroup) { |
6de22619 | 2153 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
2154 | goto again; |
2155 | } | |
6de22619 JW |
2156 | |
2157 | /* | |
2158 | * When charge migration first begins, we can have locked and | |
2159 | * unlocked page stat updates happening concurrently. Track | |
81f8c3a4 | 2160 | * the task who has the lock for unlock_page_memcg(). |
6de22619 JW |
2161 | */ |
2162 | memcg->move_lock_task = current; | |
2163 | memcg->move_lock_flags = flags; | |
d7365e78 | 2164 | |
739f79fc | 2165 | return memcg; |
89c06bd5 | 2166 | } |
81f8c3a4 | 2167 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 2168 | |
d7365e78 | 2169 | /** |
739f79fc JW |
2170 | * __unlock_page_memcg - unlock and unpin a memcg |
2171 | * @memcg: the memcg | |
2172 | * | |
2173 | * Unlock and unpin a memcg returned by lock_page_memcg(). | |
d7365e78 | 2174 | */ |
739f79fc | 2175 | void __unlock_page_memcg(struct mem_cgroup *memcg) |
89c06bd5 | 2176 | { |
6de22619 JW |
2177 | if (memcg && memcg->move_lock_task == current) { |
2178 | unsigned long flags = memcg->move_lock_flags; | |
2179 | ||
2180 | memcg->move_lock_task = NULL; | |
2181 | memcg->move_lock_flags = 0; | |
2182 | ||
2183 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2184 | } | |
89c06bd5 | 2185 | |
d7365e78 | 2186 | rcu_read_unlock(); |
89c06bd5 | 2187 | } |
739f79fc JW |
2188 | |
2189 | /** | |
2190 | * unlock_page_memcg - unlock a page->mem_cgroup binding | |
2191 | * @page: the page | |
2192 | */ | |
2193 | void unlock_page_memcg(struct page *page) | |
2194 | { | |
9da7b521 JW |
2195 | struct page *head = compound_head(page); |
2196 | ||
2197 | __unlock_page_memcg(head->mem_cgroup); | |
739f79fc | 2198 | } |
81f8c3a4 | 2199 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 2200 | |
cdec2e42 KH |
2201 | struct memcg_stock_pcp { |
2202 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2203 | unsigned int nr_pages; |
bf4f0599 RG |
2204 | |
2205 | #ifdef CONFIG_MEMCG_KMEM | |
2206 | struct obj_cgroup *cached_objcg; | |
2207 | unsigned int nr_bytes; | |
2208 | #endif | |
2209 | ||
cdec2e42 | 2210 | struct work_struct work; |
26fe6168 | 2211 | unsigned long flags; |
a0db00fc | 2212 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2213 | }; |
2214 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2215 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2216 | |
bf4f0599 RG |
2217 | #ifdef CONFIG_MEMCG_KMEM |
2218 | static void drain_obj_stock(struct memcg_stock_pcp *stock); | |
2219 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2220 | struct mem_cgroup *root_memcg); | |
2221 | ||
2222 | #else | |
2223 | static inline void drain_obj_stock(struct memcg_stock_pcp *stock) | |
2224 | { | |
2225 | } | |
2226 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2227 | struct mem_cgroup *root_memcg) | |
2228 | { | |
2229 | return false; | |
2230 | } | |
2231 | #endif | |
2232 | ||
a0956d54 SS |
2233 | /** |
2234 | * consume_stock: Try to consume stocked charge on this cpu. | |
2235 | * @memcg: memcg to consume from. | |
2236 | * @nr_pages: how many pages to charge. | |
2237 | * | |
2238 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2239 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2240 | * service an allocation will refill the stock. | |
2241 | * | |
2242 | * returns true if successful, false otherwise. | |
cdec2e42 | 2243 | */ |
a0956d54 | 2244 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2245 | { |
2246 | struct memcg_stock_pcp *stock; | |
db2ba40c | 2247 | unsigned long flags; |
3e32cb2e | 2248 | bool ret = false; |
cdec2e42 | 2249 | |
a983b5eb | 2250 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 2251 | return ret; |
a0956d54 | 2252 | |
db2ba40c JW |
2253 | local_irq_save(flags); |
2254 | ||
2255 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 2256 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2257 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2258 | ret = true; |
2259 | } | |
db2ba40c JW |
2260 | |
2261 | local_irq_restore(flags); | |
2262 | ||
cdec2e42 KH |
2263 | return ret; |
2264 | } | |
2265 | ||
2266 | /* | |
3e32cb2e | 2267 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2268 | */ |
2269 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2270 | { | |
2271 | struct mem_cgroup *old = stock->cached; | |
2272 | ||
1a3e1f40 JW |
2273 | if (!old) |
2274 | return; | |
2275 | ||
11c9ea4e | 2276 | if (stock->nr_pages) { |
3e32cb2e | 2277 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 2278 | if (do_memsw_account()) |
3e32cb2e | 2279 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
11c9ea4e | 2280 | stock->nr_pages = 0; |
cdec2e42 | 2281 | } |
1a3e1f40 JW |
2282 | |
2283 | css_put(&old->css); | |
cdec2e42 | 2284 | stock->cached = NULL; |
cdec2e42 KH |
2285 | } |
2286 | ||
cdec2e42 KH |
2287 | static void drain_local_stock(struct work_struct *dummy) |
2288 | { | |
db2ba40c JW |
2289 | struct memcg_stock_pcp *stock; |
2290 | unsigned long flags; | |
2291 | ||
72f0184c MH |
2292 | /* |
2293 | * The only protection from memory hotplug vs. drain_stock races is | |
2294 | * that we always operate on local CPU stock here with IRQ disabled | |
2295 | */ | |
db2ba40c JW |
2296 | local_irq_save(flags); |
2297 | ||
2298 | stock = this_cpu_ptr(&memcg_stock); | |
bf4f0599 | 2299 | drain_obj_stock(stock); |
cdec2e42 | 2300 | drain_stock(stock); |
26fe6168 | 2301 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c JW |
2302 | |
2303 | local_irq_restore(flags); | |
cdec2e42 KH |
2304 | } |
2305 | ||
2306 | /* | |
3e32cb2e | 2307 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2308 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2309 | */ |
c0ff4b85 | 2310 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2311 | { |
db2ba40c JW |
2312 | struct memcg_stock_pcp *stock; |
2313 | unsigned long flags; | |
2314 | ||
2315 | local_irq_save(flags); | |
cdec2e42 | 2316 | |
db2ba40c | 2317 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 2318 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2319 | drain_stock(stock); |
1a3e1f40 | 2320 | css_get(&memcg->css); |
c0ff4b85 | 2321 | stock->cached = memcg; |
cdec2e42 | 2322 | } |
11c9ea4e | 2323 | stock->nr_pages += nr_pages; |
db2ba40c | 2324 | |
a983b5eb | 2325 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 RG |
2326 | drain_stock(stock); |
2327 | ||
db2ba40c | 2328 | local_irq_restore(flags); |
cdec2e42 KH |
2329 | } |
2330 | ||
2331 | /* | |
c0ff4b85 | 2332 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2333 | * of the hierarchy under it. |
cdec2e42 | 2334 | */ |
6d3d6aa2 | 2335 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2336 | { |
26fe6168 | 2337 | int cpu, curcpu; |
d38144b7 | 2338 | |
6d3d6aa2 JW |
2339 | /* If someone's already draining, avoid adding running more workers. */ |
2340 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2341 | return; | |
72f0184c MH |
2342 | /* |
2343 | * Notify other cpus that system-wide "drain" is running | |
2344 | * We do not care about races with the cpu hotplug because cpu down | |
2345 | * as well as workers from this path always operate on the local | |
2346 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2347 | */ | |
5af12d0e | 2348 | curcpu = get_cpu(); |
cdec2e42 KH |
2349 | for_each_online_cpu(cpu) { |
2350 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2351 | struct mem_cgroup *memcg; |
e1a366be | 2352 | bool flush = false; |
26fe6168 | 2353 | |
e1a366be | 2354 | rcu_read_lock(); |
c0ff4b85 | 2355 | memcg = stock->cached; |
e1a366be RG |
2356 | if (memcg && stock->nr_pages && |
2357 | mem_cgroup_is_descendant(memcg, root_memcg)) | |
2358 | flush = true; | |
bf4f0599 RG |
2359 | if (obj_stock_flush_required(stock, root_memcg)) |
2360 | flush = true; | |
e1a366be RG |
2361 | rcu_read_unlock(); |
2362 | ||
2363 | if (flush && | |
2364 | !test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { | |
d1a05b69 MH |
2365 | if (cpu == curcpu) |
2366 | drain_local_stock(&stock->work); | |
2367 | else | |
2368 | schedule_work_on(cpu, &stock->work); | |
2369 | } | |
cdec2e42 | 2370 | } |
5af12d0e | 2371 | put_cpu(); |
9f50fad6 | 2372 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2373 | } |
2374 | ||
308167fc | 2375 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
cdec2e42 | 2376 | { |
cdec2e42 | 2377 | struct memcg_stock_pcp *stock; |
42a30035 | 2378 | struct mem_cgroup *memcg, *mi; |
cdec2e42 | 2379 | |
cdec2e42 KH |
2380 | stock = &per_cpu(memcg_stock, cpu); |
2381 | drain_stock(stock); | |
a983b5eb JW |
2382 | |
2383 | for_each_mem_cgroup(memcg) { | |
2384 | int i; | |
2385 | ||
2386 | for (i = 0; i < MEMCG_NR_STAT; i++) { | |
2387 | int nid; | |
2388 | long x; | |
2389 | ||
871789d4 | 2390 | x = this_cpu_xchg(memcg->vmstats_percpu->stat[i], 0); |
815744d7 | 2391 | if (x) |
42a30035 JW |
2392 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2393 | atomic_long_add(x, &memcg->vmstats[i]); | |
a983b5eb JW |
2394 | |
2395 | if (i >= NR_VM_NODE_STAT_ITEMS) | |
2396 | continue; | |
2397 | ||
2398 | for_each_node(nid) { | |
2399 | struct mem_cgroup_per_node *pn; | |
2400 | ||
2401 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
2402 | x = this_cpu_xchg(pn->lruvec_stat_cpu->count[i], 0); | |
815744d7 | 2403 | if (x) |
42a30035 JW |
2404 | do { |
2405 | atomic_long_add(x, &pn->lruvec_stat[i]); | |
2406 | } while ((pn = parent_nodeinfo(pn, nid))); | |
a983b5eb JW |
2407 | } |
2408 | } | |
2409 | ||
e27be240 | 2410 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { |
a983b5eb JW |
2411 | long x; |
2412 | ||
871789d4 | 2413 | x = this_cpu_xchg(memcg->vmstats_percpu->events[i], 0); |
815744d7 | 2414 | if (x) |
42a30035 JW |
2415 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2416 | atomic_long_add(x, &memcg->vmevents[i]); | |
a983b5eb JW |
2417 | } |
2418 | } | |
2419 | ||
308167fc | 2420 | return 0; |
cdec2e42 KH |
2421 | } |
2422 | ||
b3ff9291 CD |
2423 | static unsigned long reclaim_high(struct mem_cgroup *memcg, |
2424 | unsigned int nr_pages, | |
2425 | gfp_t gfp_mask) | |
f7e1cb6e | 2426 | { |
b3ff9291 CD |
2427 | unsigned long nr_reclaimed = 0; |
2428 | ||
f7e1cb6e | 2429 | do { |
e22c6ed9 JW |
2430 | unsigned long pflags; |
2431 | ||
d1663a90 JK |
2432 | if (page_counter_read(&memcg->memory) <= |
2433 | READ_ONCE(memcg->memory.high)) | |
f7e1cb6e | 2434 | continue; |
e22c6ed9 | 2435 | |
e27be240 | 2436 | memcg_memory_event(memcg, MEMCG_HIGH); |
e22c6ed9 JW |
2437 | |
2438 | psi_memstall_enter(&pflags); | |
b3ff9291 CD |
2439 | nr_reclaimed += try_to_free_mem_cgroup_pages(memcg, nr_pages, |
2440 | gfp_mask, true); | |
e22c6ed9 | 2441 | psi_memstall_leave(&pflags); |
4bf17307 CD |
2442 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2443 | !mem_cgroup_is_root(memcg)); | |
b3ff9291 CD |
2444 | |
2445 | return nr_reclaimed; | |
f7e1cb6e JW |
2446 | } |
2447 | ||
2448 | static void high_work_func(struct work_struct *work) | |
2449 | { | |
2450 | struct mem_cgroup *memcg; | |
2451 | ||
2452 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2453 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2454 | } |
2455 | ||
0e4b01df CD |
2456 | /* |
2457 | * Clamp the maximum sleep time per allocation batch to 2 seconds. This is | |
2458 | * enough to still cause a significant slowdown in most cases, while still | |
2459 | * allowing diagnostics and tracing to proceed without becoming stuck. | |
2460 | */ | |
2461 | #define MEMCG_MAX_HIGH_DELAY_JIFFIES (2UL*HZ) | |
2462 | ||
2463 | /* | |
2464 | * When calculating the delay, we use these either side of the exponentiation to | |
2465 | * maintain precision and scale to a reasonable number of jiffies (see the table | |
2466 | * below. | |
2467 | * | |
2468 | * - MEMCG_DELAY_PRECISION_SHIFT: Extra precision bits while translating the | |
2469 | * overage ratio to a delay. | |
ac5ddd0f | 2470 | * - MEMCG_DELAY_SCALING_SHIFT: The number of bits to scale down the |
0e4b01df CD |
2471 | * proposed penalty in order to reduce to a reasonable number of jiffies, and |
2472 | * to produce a reasonable delay curve. | |
2473 | * | |
2474 | * MEMCG_DELAY_SCALING_SHIFT just happens to be a number that produces a | |
2475 | * reasonable delay curve compared to precision-adjusted overage, not | |
2476 | * penalising heavily at first, but still making sure that growth beyond the | |
2477 | * limit penalises misbehaviour cgroups by slowing them down exponentially. For | |
2478 | * example, with a high of 100 megabytes: | |
2479 | * | |
2480 | * +-------+------------------------+ | |
2481 | * | usage | time to allocate in ms | | |
2482 | * +-------+------------------------+ | |
2483 | * | 100M | 0 | | |
2484 | * | 101M | 6 | | |
2485 | * | 102M | 25 | | |
2486 | * | 103M | 57 | | |
2487 | * | 104M | 102 | | |
2488 | * | 105M | 159 | | |
2489 | * | 106M | 230 | | |
2490 | * | 107M | 313 | | |
2491 | * | 108M | 409 | | |
2492 | * | 109M | 518 | | |
2493 | * | 110M | 639 | | |
2494 | * | 111M | 774 | | |
2495 | * | 112M | 921 | | |
2496 | * | 113M | 1081 | | |
2497 | * | 114M | 1254 | | |
2498 | * | 115M | 1439 | | |
2499 | * | 116M | 1638 | | |
2500 | * | 117M | 1849 | | |
2501 | * | 118M | 2000 | | |
2502 | * | 119M | 2000 | | |
2503 | * | 120M | 2000 | | |
2504 | * +-------+------------------------+ | |
2505 | */ | |
2506 | #define MEMCG_DELAY_PRECISION_SHIFT 20 | |
2507 | #define MEMCG_DELAY_SCALING_SHIFT 14 | |
2508 | ||
8a5dbc65 | 2509 | static u64 calculate_overage(unsigned long usage, unsigned long high) |
b23afb93 | 2510 | { |
8a5dbc65 | 2511 | u64 overage; |
b23afb93 | 2512 | |
8a5dbc65 JK |
2513 | if (usage <= high) |
2514 | return 0; | |
e26733e0 | 2515 | |
8a5dbc65 JK |
2516 | /* |
2517 | * Prevent division by 0 in overage calculation by acting as if | |
2518 | * it was a threshold of 1 page | |
2519 | */ | |
2520 | high = max(high, 1UL); | |
9b8b1754 | 2521 | |
8a5dbc65 JK |
2522 | overage = usage - high; |
2523 | overage <<= MEMCG_DELAY_PRECISION_SHIFT; | |
2524 | return div64_u64(overage, high); | |
2525 | } | |
e26733e0 | 2526 | |
8a5dbc65 JK |
2527 | static u64 mem_find_max_overage(struct mem_cgroup *memcg) |
2528 | { | |
2529 | u64 overage, max_overage = 0; | |
e26733e0 | 2530 | |
8a5dbc65 JK |
2531 | do { |
2532 | overage = calculate_overage(page_counter_read(&memcg->memory), | |
d1663a90 | 2533 | READ_ONCE(memcg->memory.high)); |
8a5dbc65 | 2534 | max_overage = max(overage, max_overage); |
e26733e0 CD |
2535 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2536 | !mem_cgroup_is_root(memcg)); | |
2537 | ||
8a5dbc65 JK |
2538 | return max_overage; |
2539 | } | |
2540 | ||
4b82ab4f JK |
2541 | static u64 swap_find_max_overage(struct mem_cgroup *memcg) |
2542 | { | |
2543 | u64 overage, max_overage = 0; | |
2544 | ||
2545 | do { | |
2546 | overage = calculate_overage(page_counter_read(&memcg->swap), | |
2547 | READ_ONCE(memcg->swap.high)); | |
2548 | if (overage) | |
2549 | memcg_memory_event(memcg, MEMCG_SWAP_HIGH); | |
2550 | max_overage = max(overage, max_overage); | |
2551 | } while ((memcg = parent_mem_cgroup(memcg)) && | |
2552 | !mem_cgroup_is_root(memcg)); | |
2553 | ||
2554 | return max_overage; | |
2555 | } | |
2556 | ||
8a5dbc65 JK |
2557 | /* |
2558 | * Get the number of jiffies that we should penalise a mischievous cgroup which | |
2559 | * is exceeding its memory.high by checking both it and its ancestors. | |
2560 | */ | |
2561 | static unsigned long calculate_high_delay(struct mem_cgroup *memcg, | |
2562 | unsigned int nr_pages, | |
2563 | u64 max_overage) | |
2564 | { | |
2565 | unsigned long penalty_jiffies; | |
2566 | ||
e26733e0 CD |
2567 | if (!max_overage) |
2568 | return 0; | |
0e4b01df CD |
2569 | |
2570 | /* | |
0e4b01df CD |
2571 | * We use overage compared to memory.high to calculate the number of |
2572 | * jiffies to sleep (penalty_jiffies). Ideally this value should be | |
2573 | * fairly lenient on small overages, and increasingly harsh when the | |
2574 | * memcg in question makes it clear that it has no intention of stopping | |
2575 | * its crazy behaviour, so we exponentially increase the delay based on | |
2576 | * overage amount. | |
2577 | */ | |
e26733e0 CD |
2578 | penalty_jiffies = max_overage * max_overage * HZ; |
2579 | penalty_jiffies >>= MEMCG_DELAY_PRECISION_SHIFT; | |
2580 | penalty_jiffies >>= MEMCG_DELAY_SCALING_SHIFT; | |
0e4b01df CD |
2581 | |
2582 | /* | |
2583 | * Factor in the task's own contribution to the overage, such that four | |
2584 | * N-sized allocations are throttled approximately the same as one | |
2585 | * 4N-sized allocation. | |
2586 | * | |
2587 | * MEMCG_CHARGE_BATCH pages is nominal, so work out how much smaller or | |
2588 | * larger the current charge patch is than that. | |
2589 | */ | |
ff144e69 | 2590 | return penalty_jiffies * nr_pages / MEMCG_CHARGE_BATCH; |
e26733e0 CD |
2591 | } |
2592 | ||
2593 | /* | |
2594 | * Scheduled by try_charge() to be executed from the userland return path | |
2595 | * and reclaims memory over the high limit. | |
2596 | */ | |
2597 | void mem_cgroup_handle_over_high(void) | |
2598 | { | |
2599 | unsigned long penalty_jiffies; | |
2600 | unsigned long pflags; | |
b3ff9291 | 2601 | unsigned long nr_reclaimed; |
e26733e0 | 2602 | unsigned int nr_pages = current->memcg_nr_pages_over_high; |
d977aa93 | 2603 | int nr_retries = MAX_RECLAIM_RETRIES; |
e26733e0 | 2604 | struct mem_cgroup *memcg; |
b3ff9291 | 2605 | bool in_retry = false; |
e26733e0 CD |
2606 | |
2607 | if (likely(!nr_pages)) | |
2608 | return; | |
2609 | ||
2610 | memcg = get_mem_cgroup_from_mm(current->mm); | |
e26733e0 CD |
2611 | current->memcg_nr_pages_over_high = 0; |
2612 | ||
b3ff9291 CD |
2613 | retry_reclaim: |
2614 | /* | |
2615 | * The allocating task should reclaim at least the batch size, but for | |
2616 | * subsequent retries we only want to do what's necessary to prevent oom | |
2617 | * or breaching resource isolation. | |
2618 | * | |
2619 | * This is distinct from memory.max or page allocator behaviour because | |
2620 | * memory.high is currently batched, whereas memory.max and the page | |
2621 | * allocator run every time an allocation is made. | |
2622 | */ | |
2623 | nr_reclaimed = reclaim_high(memcg, | |
2624 | in_retry ? SWAP_CLUSTER_MAX : nr_pages, | |
2625 | GFP_KERNEL); | |
2626 | ||
e26733e0 CD |
2627 | /* |
2628 | * memory.high is breached and reclaim is unable to keep up. Throttle | |
2629 | * allocators proactively to slow down excessive growth. | |
2630 | */ | |
8a5dbc65 JK |
2631 | penalty_jiffies = calculate_high_delay(memcg, nr_pages, |
2632 | mem_find_max_overage(memcg)); | |
0e4b01df | 2633 | |
4b82ab4f JK |
2634 | penalty_jiffies += calculate_high_delay(memcg, nr_pages, |
2635 | swap_find_max_overage(memcg)); | |
2636 | ||
ff144e69 JK |
2637 | /* |
2638 | * Clamp the max delay per usermode return so as to still keep the | |
2639 | * application moving forwards and also permit diagnostics, albeit | |
2640 | * extremely slowly. | |
2641 | */ | |
2642 | penalty_jiffies = min(penalty_jiffies, MEMCG_MAX_HIGH_DELAY_JIFFIES); | |
2643 | ||
0e4b01df CD |
2644 | /* |
2645 | * Don't sleep if the amount of jiffies this memcg owes us is so low | |
2646 | * that it's not even worth doing, in an attempt to be nice to those who | |
2647 | * go only a small amount over their memory.high value and maybe haven't | |
2648 | * been aggressively reclaimed enough yet. | |
2649 | */ | |
2650 | if (penalty_jiffies <= HZ / 100) | |
2651 | goto out; | |
2652 | ||
b3ff9291 CD |
2653 | /* |
2654 | * If reclaim is making forward progress but we're still over | |
2655 | * memory.high, we want to encourage that rather than doing allocator | |
2656 | * throttling. | |
2657 | */ | |
2658 | if (nr_reclaimed || nr_retries--) { | |
2659 | in_retry = true; | |
2660 | goto retry_reclaim; | |
2661 | } | |
2662 | ||
0e4b01df CD |
2663 | /* |
2664 | * If we exit early, we're guaranteed to die (since | |
2665 | * schedule_timeout_killable sets TASK_KILLABLE). This means we don't | |
2666 | * need to account for any ill-begotten jiffies to pay them off later. | |
2667 | */ | |
2668 | psi_memstall_enter(&pflags); | |
2669 | schedule_timeout_killable(penalty_jiffies); | |
2670 | psi_memstall_leave(&pflags); | |
2671 | ||
2672 | out: | |
2673 | css_put(&memcg->css); | |
b23afb93 TH |
2674 | } |
2675 | ||
00501b53 JW |
2676 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2677 | unsigned int nr_pages) | |
8a9f3ccd | 2678 | { |
a983b5eb | 2679 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
d977aa93 | 2680 | int nr_retries = MAX_RECLAIM_RETRIES; |
6539cc05 | 2681 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2682 | struct page_counter *counter; |
e22c6ed9 | 2683 | enum oom_status oom_status; |
6539cc05 | 2684 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2685 | bool may_swap = true; |
2686 | bool drained = false; | |
e22c6ed9 | 2687 | unsigned long pflags; |
a636b327 | 2688 | |
ce00a967 | 2689 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 2690 | return 0; |
6539cc05 | 2691 | retry: |
b6b6cc72 | 2692 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2693 | return 0; |
8a9f3ccd | 2694 | |
7941d214 | 2695 | if (!do_memsw_account() || |
6071ca52 JW |
2696 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2697 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2698 | goto done_restock; |
7941d214 | 2699 | if (do_memsw_account()) |
3e32cb2e JW |
2700 | page_counter_uncharge(&memcg->memsw, batch); |
2701 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2702 | } else { |
3e32cb2e | 2703 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2704 | may_swap = false; |
3fbe7244 | 2705 | } |
7a81b88c | 2706 | |
6539cc05 JW |
2707 | if (batch > nr_pages) { |
2708 | batch = nr_pages; | |
2709 | goto retry; | |
2710 | } | |
6d61ef40 | 2711 | |
869712fd JW |
2712 | /* |
2713 | * Memcg doesn't have a dedicated reserve for atomic | |
2714 | * allocations. But like the global atomic pool, we need to | |
2715 | * put the burden of reclaim on regular allocation requests | |
2716 | * and let these go through as privileged allocations. | |
2717 | */ | |
2718 | if (gfp_mask & __GFP_ATOMIC) | |
2719 | goto force; | |
2720 | ||
06b078fc JW |
2721 | /* |
2722 | * Unlike in global OOM situations, memcg is not in a physical | |
2723 | * memory shortage. Allow dying and OOM-killed tasks to | |
2724 | * bypass the last charges so that they can exit quickly and | |
2725 | * free their memory. | |
2726 | */ | |
7775face | 2727 | if (unlikely(should_force_charge())) |
10d53c74 | 2728 | goto force; |
06b078fc | 2729 | |
89a28483 JW |
2730 | /* |
2731 | * Prevent unbounded recursion when reclaim operations need to | |
2732 | * allocate memory. This might exceed the limits temporarily, | |
2733 | * but we prefer facilitating memory reclaim and getting back | |
2734 | * under the limit over triggering OOM kills in these cases. | |
2735 | */ | |
2736 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2737 | goto force; | |
2738 | ||
06b078fc JW |
2739 | if (unlikely(task_in_memcg_oom(current))) |
2740 | goto nomem; | |
2741 | ||
d0164adc | 2742 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2743 | goto nomem; |
4b534334 | 2744 | |
e27be240 | 2745 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
241994ed | 2746 | |
e22c6ed9 | 2747 | psi_memstall_enter(&pflags); |
b70a2a21 JW |
2748 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2749 | gfp_mask, may_swap); | |
e22c6ed9 | 2750 | psi_memstall_leave(&pflags); |
6539cc05 | 2751 | |
61e02c74 | 2752 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2753 | goto retry; |
28c34c29 | 2754 | |
b70a2a21 | 2755 | if (!drained) { |
6d3d6aa2 | 2756 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2757 | drained = true; |
2758 | goto retry; | |
2759 | } | |
2760 | ||
28c34c29 JW |
2761 | if (gfp_mask & __GFP_NORETRY) |
2762 | goto nomem; | |
6539cc05 JW |
2763 | /* |
2764 | * Even though the limit is exceeded at this point, reclaim | |
2765 | * may have been able to free some pages. Retry the charge | |
2766 | * before killing the task. | |
2767 | * | |
2768 | * Only for regular pages, though: huge pages are rather | |
2769 | * unlikely to succeed so close to the limit, and we fall back | |
2770 | * to regular pages anyway in case of failure. | |
2771 | */ | |
61e02c74 | 2772 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2773 | goto retry; |
2774 | /* | |
2775 | * At task move, charge accounts can be doubly counted. So, it's | |
2776 | * better to wait until the end of task_move if something is going on. | |
2777 | */ | |
2778 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2779 | goto retry; | |
2780 | ||
9b130619 JW |
2781 | if (nr_retries--) |
2782 | goto retry; | |
2783 | ||
38d38493 | 2784 | if (gfp_mask & __GFP_RETRY_MAYFAIL) |
29ef680a MH |
2785 | goto nomem; |
2786 | ||
06b078fc | 2787 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2788 | goto force; |
06b078fc | 2789 | |
6539cc05 | 2790 | if (fatal_signal_pending(current)) |
10d53c74 | 2791 | goto force; |
6539cc05 | 2792 | |
29ef680a MH |
2793 | /* |
2794 | * keep retrying as long as the memcg oom killer is able to make | |
2795 | * a forward progress or bypass the charge if the oom killer | |
2796 | * couldn't make any progress. | |
2797 | */ | |
2798 | oom_status = mem_cgroup_oom(mem_over_limit, gfp_mask, | |
3608de07 | 2799 | get_order(nr_pages * PAGE_SIZE)); |
29ef680a MH |
2800 | switch (oom_status) { |
2801 | case OOM_SUCCESS: | |
d977aa93 | 2802 | nr_retries = MAX_RECLAIM_RETRIES; |
29ef680a MH |
2803 | goto retry; |
2804 | case OOM_FAILED: | |
2805 | goto force; | |
2806 | default: | |
2807 | goto nomem; | |
2808 | } | |
7a81b88c | 2809 | nomem: |
6d1fdc48 | 2810 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2811 | return -ENOMEM; |
10d53c74 TH |
2812 | force: |
2813 | /* | |
2814 | * The allocation either can't fail or will lead to more memory | |
2815 | * being freed very soon. Allow memory usage go over the limit | |
2816 | * temporarily by force charging it. | |
2817 | */ | |
2818 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2819 | if (do_memsw_account()) |
10d53c74 | 2820 | page_counter_charge(&memcg->memsw, nr_pages); |
10d53c74 TH |
2821 | |
2822 | return 0; | |
6539cc05 JW |
2823 | |
2824 | done_restock: | |
2825 | if (batch > nr_pages) | |
2826 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2827 | |
241994ed | 2828 | /* |
b23afb93 TH |
2829 | * If the hierarchy is above the normal consumption range, schedule |
2830 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2831 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2832 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2833 | * not recorded as it most likely matches current's and won't | |
2834 | * change in the meantime. As high limit is checked again before | |
2835 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2836 | */ |
2837 | do { | |
4b82ab4f JK |
2838 | bool mem_high, swap_high; |
2839 | ||
2840 | mem_high = page_counter_read(&memcg->memory) > | |
2841 | READ_ONCE(memcg->memory.high); | |
2842 | swap_high = page_counter_read(&memcg->swap) > | |
2843 | READ_ONCE(memcg->swap.high); | |
2844 | ||
2845 | /* Don't bother a random interrupted task */ | |
2846 | if (in_interrupt()) { | |
2847 | if (mem_high) { | |
f7e1cb6e JW |
2848 | schedule_work(&memcg->high_work); |
2849 | break; | |
2850 | } | |
4b82ab4f JK |
2851 | continue; |
2852 | } | |
2853 | ||
2854 | if (mem_high || swap_high) { | |
2855 | /* | |
2856 | * The allocating tasks in this cgroup will need to do | |
2857 | * reclaim or be throttled to prevent further growth | |
2858 | * of the memory or swap footprints. | |
2859 | * | |
2860 | * Target some best-effort fairness between the tasks, | |
2861 | * and distribute reclaim work and delay penalties | |
2862 | * based on how much each task is actually allocating. | |
2863 | */ | |
9516a18a | 2864 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2865 | set_notify_resume(current); |
2866 | break; | |
2867 | } | |
241994ed | 2868 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2869 | |
2870 | return 0; | |
7a81b88c | 2871 | } |
8a9f3ccd | 2872 | |
f0e45fb4 | 2873 | #if defined(CONFIG_MEMCG_KMEM) || defined(CONFIG_MMU) |
00501b53 | 2874 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2875 | { |
ce00a967 JW |
2876 | if (mem_cgroup_is_root(memcg)) |
2877 | return; | |
2878 | ||
3e32cb2e | 2879 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2880 | if (do_memsw_account()) |
3e32cb2e | 2881 | page_counter_uncharge(&memcg->memsw, nr_pages); |
d01dd17f | 2882 | } |
f0e45fb4 | 2883 | #endif |
d01dd17f | 2884 | |
d9eb1ea2 | 2885 | static void commit_charge(struct page *page, struct mem_cgroup *memcg) |
0a31bc97 | 2886 | { |
1306a85a | 2887 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
0a31bc97 | 2888 | /* |
a0b5b414 | 2889 | * Any of the following ensures page->mem_cgroup stability: |
0a31bc97 | 2890 | * |
a0b5b414 JW |
2891 | * - the page lock |
2892 | * - LRU isolation | |
2893 | * - lock_page_memcg() | |
2894 | * - exclusive reference | |
0a31bc97 | 2895 | */ |
1306a85a | 2896 | page->mem_cgroup = memcg; |
7a81b88c | 2897 | } |
66e1707b | 2898 | |
84c07d11 | 2899 | #ifdef CONFIG_MEMCG_KMEM |
10befea9 RG |
2900 | int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s, |
2901 | gfp_t gfp) | |
2902 | { | |
2903 | unsigned int objects = objs_per_slab_page(s, page); | |
2904 | void *vec; | |
2905 | ||
2906 | vec = kcalloc_node(objects, sizeof(struct obj_cgroup *), gfp, | |
2907 | page_to_nid(page)); | |
2908 | if (!vec) | |
2909 | return -ENOMEM; | |
2910 | ||
2911 | if (cmpxchg(&page->obj_cgroups, NULL, | |
2912 | (struct obj_cgroup **) ((unsigned long)vec | 0x1UL))) | |
2913 | kfree(vec); | |
2914 | else | |
2915 | kmemleak_not_leak(vec); | |
2916 | ||
2917 | return 0; | |
2918 | } | |
2919 | ||
8380ce47 RG |
2920 | /* |
2921 | * Returns a pointer to the memory cgroup to which the kernel object is charged. | |
2922 | * | |
2923 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), | |
2924 | * cgroup_mutex, etc. | |
2925 | */ | |
2926 | struct mem_cgroup *mem_cgroup_from_obj(void *p) | |
2927 | { | |
2928 | struct page *page; | |
2929 | ||
2930 | if (mem_cgroup_disabled()) | |
2931 | return NULL; | |
2932 | ||
2933 | page = virt_to_head_page(p); | |
2934 | ||
19b629c9 RG |
2935 | /* |
2936 | * If page->mem_cgroup is set, it's either a simple mem_cgroup pointer | |
2937 | * or a pointer to obj_cgroup vector. In the latter case the lowest | |
2938 | * bit of the pointer is set. | |
2939 | * The page->mem_cgroup pointer can be asynchronously changed | |
2940 | * from NULL to (obj_cgroup_vec | 0x1UL), but can't be changed | |
2941 | * from a valid memcg pointer to objcg vector or back. | |
2942 | */ | |
2943 | if (!page->mem_cgroup) | |
2944 | return NULL; | |
2945 | ||
8380ce47 | 2946 | /* |
9855609b RG |
2947 | * Slab objects are accounted individually, not per-page. |
2948 | * Memcg membership data for each individual object is saved in | |
2949 | * the page->obj_cgroups. | |
8380ce47 | 2950 | */ |
9855609b RG |
2951 | if (page_has_obj_cgroups(page)) { |
2952 | struct obj_cgroup *objcg; | |
2953 | unsigned int off; | |
2954 | ||
2955 | off = obj_to_index(page->slab_cache, page, p); | |
2956 | objcg = page_obj_cgroups(page)[off]; | |
10befea9 RG |
2957 | if (objcg) |
2958 | return obj_cgroup_memcg(objcg); | |
2959 | ||
2960 | return NULL; | |
9855609b | 2961 | } |
8380ce47 RG |
2962 | |
2963 | /* All other pages use page->mem_cgroup */ | |
2964 | return page->mem_cgroup; | |
2965 | } | |
2966 | ||
bf4f0599 RG |
2967 | __always_inline struct obj_cgroup *get_obj_cgroup_from_current(void) |
2968 | { | |
2969 | struct obj_cgroup *objcg = NULL; | |
2970 | struct mem_cgroup *memcg; | |
2971 | ||
279c3393 RG |
2972 | if (memcg_kmem_bypass()) |
2973 | return NULL; | |
2974 | ||
bf4f0599 | 2975 | rcu_read_lock(); |
37d5985c RG |
2976 | if (unlikely(active_memcg())) |
2977 | memcg = active_memcg(); | |
bf4f0599 RG |
2978 | else |
2979 | memcg = mem_cgroup_from_task(current); | |
2980 | ||
2981 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { | |
2982 | objcg = rcu_dereference(memcg->objcg); | |
2983 | if (objcg && obj_cgroup_tryget(objcg)) | |
2984 | break; | |
2985 | } | |
2986 | rcu_read_unlock(); | |
2987 | ||
2988 | return objcg; | |
2989 | } | |
2990 | ||
f3bb3043 | 2991 | static int memcg_alloc_cache_id(void) |
55007d84 | 2992 | { |
f3bb3043 VD |
2993 | int id, size; |
2994 | int err; | |
2995 | ||
dbcf73e2 | 2996 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2997 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2998 | if (id < 0) | |
2999 | return id; | |
55007d84 | 3000 | |
dbcf73e2 | 3001 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
3002 | return id; |
3003 | ||
3004 | /* | |
3005 | * There's no space for the new id in memcg_caches arrays, | |
3006 | * so we have to grow them. | |
3007 | */ | |
05257a1a | 3008 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
3009 | |
3010 | size = 2 * (id + 1); | |
55007d84 GC |
3011 | if (size < MEMCG_CACHES_MIN_SIZE) |
3012 | size = MEMCG_CACHES_MIN_SIZE; | |
3013 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
3014 | size = MEMCG_CACHES_MAX_SIZE; | |
3015 | ||
9855609b | 3016 | err = memcg_update_all_list_lrus(size); |
05257a1a VD |
3017 | if (!err) |
3018 | memcg_nr_cache_ids = size; | |
3019 | ||
3020 | up_write(&memcg_cache_ids_sem); | |
3021 | ||
f3bb3043 | 3022 | if (err) { |
dbcf73e2 | 3023 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
3024 | return err; |
3025 | } | |
3026 | return id; | |
3027 | } | |
3028 | ||
3029 | static void memcg_free_cache_id(int id) | |
3030 | { | |
dbcf73e2 | 3031 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
3032 | } |
3033 | ||
45264778 | 3034 | /** |
4b13f64d | 3035 | * __memcg_kmem_charge: charge a number of kernel pages to a memcg |
10eaec2f | 3036 | * @memcg: memory cgroup to charge |
45264778 | 3037 | * @gfp: reclaim mode |
92d0510c | 3038 | * @nr_pages: number of pages to charge |
45264778 VD |
3039 | * |
3040 | * Returns 0 on success, an error code on failure. | |
3041 | */ | |
4b13f64d RG |
3042 | int __memcg_kmem_charge(struct mem_cgroup *memcg, gfp_t gfp, |
3043 | unsigned int nr_pages) | |
7ae1e1d0 | 3044 | { |
f3ccb2c4 | 3045 | struct page_counter *counter; |
7ae1e1d0 GC |
3046 | int ret; |
3047 | ||
f3ccb2c4 | 3048 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 3049 | if (ret) |
f3ccb2c4 | 3050 | return ret; |
52c29b04 JW |
3051 | |
3052 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
3053 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
e55d9d9b MH |
3054 | |
3055 | /* | |
3056 | * Enforce __GFP_NOFAIL allocation because callers are not | |
3057 | * prepared to see failures and likely do not have any failure | |
3058 | * handling code. | |
3059 | */ | |
3060 | if (gfp & __GFP_NOFAIL) { | |
3061 | page_counter_charge(&memcg->kmem, nr_pages); | |
3062 | return 0; | |
3063 | } | |
52c29b04 JW |
3064 | cancel_charge(memcg, nr_pages); |
3065 | return -ENOMEM; | |
7ae1e1d0 | 3066 | } |
f3ccb2c4 | 3067 | return 0; |
7ae1e1d0 GC |
3068 | } |
3069 | ||
4b13f64d RG |
3070 | /** |
3071 | * __memcg_kmem_uncharge: uncharge a number of kernel pages from a memcg | |
3072 | * @memcg: memcg to uncharge | |
3073 | * @nr_pages: number of pages to uncharge | |
3074 | */ | |
3075 | void __memcg_kmem_uncharge(struct mem_cgroup *memcg, unsigned int nr_pages) | |
3076 | { | |
3077 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
3078 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
3079 | ||
3080 | page_counter_uncharge(&memcg->memory, nr_pages); | |
3081 | if (do_memsw_account()) | |
3082 | page_counter_uncharge(&memcg->memsw, nr_pages); | |
3083 | } | |
3084 | ||
45264778 | 3085 | /** |
f4b00eab | 3086 | * __memcg_kmem_charge_page: charge a kmem page to the current memory cgroup |
45264778 VD |
3087 | * @page: page to charge |
3088 | * @gfp: reclaim mode | |
3089 | * @order: allocation order | |
3090 | * | |
3091 | * Returns 0 on success, an error code on failure. | |
3092 | */ | |
f4b00eab | 3093 | int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 3094 | { |
f3ccb2c4 | 3095 | struct mem_cgroup *memcg; |
fcff7d7e | 3096 | int ret = 0; |
7ae1e1d0 | 3097 | |
d46eb14b | 3098 | memcg = get_mem_cgroup_from_current(); |
279c3393 | 3099 | if (memcg && !mem_cgroup_is_root(memcg)) { |
4b13f64d | 3100 | ret = __memcg_kmem_charge(memcg, gfp, 1 << order); |
4d96ba35 RG |
3101 | if (!ret) { |
3102 | page->mem_cgroup = memcg; | |
c4159a75 | 3103 | __SetPageKmemcg(page); |
1a3e1f40 | 3104 | return 0; |
4d96ba35 | 3105 | } |
279c3393 | 3106 | css_put(&memcg->css); |
c4159a75 | 3107 | } |
d05e83a6 | 3108 | return ret; |
7ae1e1d0 | 3109 | } |
49a18eae | 3110 | |
45264778 | 3111 | /** |
f4b00eab | 3112 | * __memcg_kmem_uncharge_page: uncharge a kmem page |
45264778 VD |
3113 | * @page: page to uncharge |
3114 | * @order: allocation order | |
3115 | */ | |
f4b00eab | 3116 | void __memcg_kmem_uncharge_page(struct page *page, int order) |
7ae1e1d0 | 3117 | { |
1306a85a | 3118 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 3119 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 3120 | |
7ae1e1d0 GC |
3121 | if (!memcg) |
3122 | return; | |
3123 | ||
309381fe | 3124 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
4b13f64d | 3125 | __memcg_kmem_uncharge(memcg, nr_pages); |
1306a85a | 3126 | page->mem_cgroup = NULL; |
1a3e1f40 | 3127 | css_put(&memcg->css); |
c4159a75 VD |
3128 | |
3129 | /* slab pages do not have PageKmemcg flag set */ | |
3130 | if (PageKmemcg(page)) | |
3131 | __ClearPageKmemcg(page); | |
60d3fd32 | 3132 | } |
bf4f0599 RG |
3133 | |
3134 | static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) | |
3135 | { | |
3136 | struct memcg_stock_pcp *stock; | |
3137 | unsigned long flags; | |
3138 | bool ret = false; | |
3139 | ||
3140 | local_irq_save(flags); | |
3141 | ||
3142 | stock = this_cpu_ptr(&memcg_stock); | |
3143 | if (objcg == stock->cached_objcg && stock->nr_bytes >= nr_bytes) { | |
3144 | stock->nr_bytes -= nr_bytes; | |
3145 | ret = true; | |
3146 | } | |
3147 | ||
3148 | local_irq_restore(flags); | |
3149 | ||
3150 | return ret; | |
3151 | } | |
3152 | ||
3153 | static void drain_obj_stock(struct memcg_stock_pcp *stock) | |
3154 | { | |
3155 | struct obj_cgroup *old = stock->cached_objcg; | |
3156 | ||
3157 | if (!old) | |
3158 | return; | |
3159 | ||
3160 | if (stock->nr_bytes) { | |
3161 | unsigned int nr_pages = stock->nr_bytes >> PAGE_SHIFT; | |
3162 | unsigned int nr_bytes = stock->nr_bytes & (PAGE_SIZE - 1); | |
3163 | ||
3164 | if (nr_pages) { | |
3165 | rcu_read_lock(); | |
3166 | __memcg_kmem_uncharge(obj_cgroup_memcg(old), nr_pages); | |
3167 | rcu_read_unlock(); | |
3168 | } | |
3169 | ||
3170 | /* | |
3171 | * The leftover is flushed to the centralized per-memcg value. | |
3172 | * On the next attempt to refill obj stock it will be moved | |
3173 | * to a per-cpu stock (probably, on an other CPU), see | |
3174 | * refill_obj_stock(). | |
3175 | * | |
3176 | * How often it's flushed is a trade-off between the memory | |
3177 | * limit enforcement accuracy and potential CPU contention, | |
3178 | * so it might be changed in the future. | |
3179 | */ | |
3180 | atomic_add(nr_bytes, &old->nr_charged_bytes); | |
3181 | stock->nr_bytes = 0; | |
3182 | } | |
3183 | ||
3184 | obj_cgroup_put(old); | |
3185 | stock->cached_objcg = NULL; | |
3186 | } | |
3187 | ||
3188 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
3189 | struct mem_cgroup *root_memcg) | |
3190 | { | |
3191 | struct mem_cgroup *memcg; | |
3192 | ||
3193 | if (stock->cached_objcg) { | |
3194 | memcg = obj_cgroup_memcg(stock->cached_objcg); | |
3195 | if (memcg && mem_cgroup_is_descendant(memcg, root_memcg)) | |
3196 | return true; | |
3197 | } | |
3198 | ||
3199 | return false; | |
3200 | } | |
3201 | ||
3202 | static void refill_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) | |
3203 | { | |
3204 | struct memcg_stock_pcp *stock; | |
3205 | unsigned long flags; | |
3206 | ||
3207 | local_irq_save(flags); | |
3208 | ||
3209 | stock = this_cpu_ptr(&memcg_stock); | |
3210 | if (stock->cached_objcg != objcg) { /* reset if necessary */ | |
3211 | drain_obj_stock(stock); | |
3212 | obj_cgroup_get(objcg); | |
3213 | stock->cached_objcg = objcg; | |
3214 | stock->nr_bytes = atomic_xchg(&objcg->nr_charged_bytes, 0); | |
3215 | } | |
3216 | stock->nr_bytes += nr_bytes; | |
3217 | ||
3218 | if (stock->nr_bytes > PAGE_SIZE) | |
3219 | drain_obj_stock(stock); | |
3220 | ||
3221 | local_irq_restore(flags); | |
3222 | } | |
3223 | ||
3224 | int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size) | |
3225 | { | |
3226 | struct mem_cgroup *memcg; | |
3227 | unsigned int nr_pages, nr_bytes; | |
3228 | int ret; | |
3229 | ||
3230 | if (consume_obj_stock(objcg, size)) | |
3231 | return 0; | |
3232 | ||
3233 | /* | |
3234 | * In theory, memcg->nr_charged_bytes can have enough | |
3235 | * pre-charged bytes to satisfy the allocation. However, | |
3236 | * flushing memcg->nr_charged_bytes requires two atomic | |
3237 | * operations, and memcg->nr_charged_bytes can't be big, | |
3238 | * so it's better to ignore it and try grab some new pages. | |
3239 | * memcg->nr_charged_bytes will be flushed in | |
3240 | * refill_obj_stock(), called from this function or | |
3241 | * independently later. | |
3242 | */ | |
3243 | rcu_read_lock(); | |
3244 | memcg = obj_cgroup_memcg(objcg); | |
3245 | css_get(&memcg->css); | |
3246 | rcu_read_unlock(); | |
3247 | ||
3248 | nr_pages = size >> PAGE_SHIFT; | |
3249 | nr_bytes = size & (PAGE_SIZE - 1); | |
3250 | ||
3251 | if (nr_bytes) | |
3252 | nr_pages += 1; | |
3253 | ||
3254 | ret = __memcg_kmem_charge(memcg, gfp, nr_pages); | |
3255 | if (!ret && nr_bytes) | |
3256 | refill_obj_stock(objcg, PAGE_SIZE - nr_bytes); | |
3257 | ||
3258 | css_put(&memcg->css); | |
3259 | return ret; | |
3260 | } | |
3261 | ||
3262 | void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size) | |
3263 | { | |
3264 | refill_obj_stock(objcg, size); | |
3265 | } | |
3266 | ||
84c07d11 | 3267 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 3268 | |
ca3e0214 KH |
3269 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3270 | ||
ca3e0214 KH |
3271 | /* |
3272 | * Because tail pages are not marked as "used", set it. We're under | |
f4b7e272 | 3273 | * pgdat->lru_lock and migration entries setup in all page mappings. |
ca3e0214 | 3274 | */ |
e94c8a9c | 3275 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 3276 | { |
1a3e1f40 | 3277 | struct mem_cgroup *memcg = head->mem_cgroup; |
e94c8a9c | 3278 | int i; |
ca3e0214 | 3279 | |
3d37c4a9 KH |
3280 | if (mem_cgroup_disabled()) |
3281 | return; | |
b070e65c | 3282 | |
1a3e1f40 JW |
3283 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
3284 | css_get(&memcg->css); | |
3285 | head[i].mem_cgroup = memcg; | |
3286 | } | |
ca3e0214 | 3287 | } |
12d27107 | 3288 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3289 | |
c255a458 | 3290 | #ifdef CONFIG_MEMCG_SWAP |
02491447 DN |
3291 | /** |
3292 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3293 | * @entry: swap entry to be moved | |
3294 | * @from: mem_cgroup which the entry is moved from | |
3295 | * @to: mem_cgroup which the entry is moved to | |
3296 | * | |
3297 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3298 | * as the mem_cgroup's id of @from. | |
3299 | * | |
3300 | * Returns 0 on success, -EINVAL on failure. | |
3301 | * | |
3e32cb2e | 3302 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
3303 | * both res and memsw, and called css_get(). |
3304 | */ | |
3305 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3306 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3307 | { |
3308 | unsigned short old_id, new_id; | |
3309 | ||
34c00c31 LZ |
3310 | old_id = mem_cgroup_id(from); |
3311 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3312 | |
3313 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
3314 | mod_memcg_state(from, MEMCG_SWAP, -1); |
3315 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
3316 | return 0; |
3317 | } | |
3318 | return -EINVAL; | |
3319 | } | |
3320 | #else | |
3321 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3322 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3323 | { |
3324 | return -EINVAL; | |
3325 | } | |
8c7c6e34 | 3326 | #endif |
d13d1443 | 3327 | |
bbec2e15 | 3328 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 3329 | |
bbec2e15 RG |
3330 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
3331 | unsigned long max, bool memsw) | |
628f4235 | 3332 | { |
3e32cb2e | 3333 | bool enlarge = false; |
bb4a7ea2 | 3334 | bool drained = false; |
3e32cb2e | 3335 | int ret; |
c054a78c YZ |
3336 | bool limits_invariant; |
3337 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 3338 | |
3e32cb2e | 3339 | do { |
628f4235 KH |
3340 | if (signal_pending(current)) { |
3341 | ret = -EINTR; | |
3342 | break; | |
3343 | } | |
3e32cb2e | 3344 | |
bbec2e15 | 3345 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
3346 | /* |
3347 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 3348 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 3349 | */ |
15b42562 | 3350 | limits_invariant = memsw ? max >= READ_ONCE(memcg->memory.max) : |
bbec2e15 | 3351 | max <= memcg->memsw.max; |
c054a78c | 3352 | if (!limits_invariant) { |
bbec2e15 | 3353 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 3354 | ret = -EINVAL; |
8c7c6e34 KH |
3355 | break; |
3356 | } | |
bbec2e15 | 3357 | if (max > counter->max) |
3e32cb2e | 3358 | enlarge = true; |
bbec2e15 RG |
3359 | ret = page_counter_set_max(counter, max); |
3360 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
3361 | |
3362 | if (!ret) | |
3363 | break; | |
3364 | ||
bb4a7ea2 SB |
3365 | if (!drained) { |
3366 | drain_all_stock(memcg); | |
3367 | drained = true; | |
3368 | continue; | |
3369 | } | |
3370 | ||
1ab5c056 AR |
3371 | if (!try_to_free_mem_cgroup_pages(memcg, 1, |
3372 | GFP_KERNEL, !memsw)) { | |
3373 | ret = -EBUSY; | |
3374 | break; | |
3375 | } | |
3376 | } while (true); | |
3e32cb2e | 3377 | |
3c11ecf4 KH |
3378 | if (!ret && enlarge) |
3379 | memcg_oom_recover(memcg); | |
3e32cb2e | 3380 | |
628f4235 KH |
3381 | return ret; |
3382 | } | |
3383 | ||
ef8f2327 | 3384 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
3385 | gfp_t gfp_mask, |
3386 | unsigned long *total_scanned) | |
3387 | { | |
3388 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 3389 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
3390 | unsigned long reclaimed; |
3391 | int loop = 0; | |
ef8f2327 | 3392 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 3393 | unsigned long excess; |
0608f43d AM |
3394 | unsigned long nr_scanned; |
3395 | ||
3396 | if (order > 0) | |
3397 | return 0; | |
3398 | ||
ef8f2327 | 3399 | mctz = soft_limit_tree_node(pgdat->node_id); |
d6507ff5 MH |
3400 | |
3401 | /* | |
3402 | * Do not even bother to check the largest node if the root | |
3403 | * is empty. Do it lockless to prevent lock bouncing. Races | |
3404 | * are acceptable as soft limit is best effort anyway. | |
3405 | */ | |
bfc7228b | 3406 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
3407 | return 0; |
3408 | ||
0608f43d AM |
3409 | /* |
3410 | * This loop can run a while, specially if mem_cgroup's continuously | |
3411 | * keep exceeding their soft limit and putting the system under | |
3412 | * pressure | |
3413 | */ | |
3414 | do { | |
3415 | if (next_mz) | |
3416 | mz = next_mz; | |
3417 | else | |
3418 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3419 | if (!mz) | |
3420 | break; | |
3421 | ||
3422 | nr_scanned = 0; | |
ef8f2327 | 3423 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
0608f43d AM |
3424 | gfp_mask, &nr_scanned); |
3425 | nr_reclaimed += reclaimed; | |
3426 | *total_scanned += nr_scanned; | |
0a31bc97 | 3427 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 3428 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
3429 | |
3430 | /* | |
3431 | * If we failed to reclaim anything from this memory cgroup | |
3432 | * it is time to move on to the next cgroup | |
3433 | */ | |
3434 | next_mz = NULL; | |
bc2f2e7f VD |
3435 | if (!reclaimed) |
3436 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3437 | ||
3e32cb2e | 3438 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3439 | /* |
3440 | * One school of thought says that we should not add | |
3441 | * back the node to the tree if reclaim returns 0. | |
3442 | * But our reclaim could return 0, simply because due | |
3443 | * to priority we are exposing a smaller subset of | |
3444 | * memory to reclaim from. Consider this as a longer | |
3445 | * term TODO. | |
3446 | */ | |
3447 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3448 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3449 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3450 | css_put(&mz->memcg->css); |
3451 | loop++; | |
3452 | /* | |
3453 | * Could not reclaim anything and there are no more | |
3454 | * mem cgroups to try or we seem to be looping without | |
3455 | * reclaiming anything. | |
3456 | */ | |
3457 | if (!nr_reclaimed && | |
3458 | (next_mz == NULL || | |
3459 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3460 | break; | |
3461 | } while (!nr_reclaimed); | |
3462 | if (next_mz) | |
3463 | css_put(&next_mz->memcg->css); | |
3464 | return nr_reclaimed; | |
3465 | } | |
3466 | ||
ea280e7b TH |
3467 | /* |
3468 | * Test whether @memcg has children, dead or alive. Note that this | |
3469 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
3470 | * returns %true if there are child csses according to the cgroup | |
b8f2935f | 3471 | * hierarchy. Testing use_hierarchy is the caller's responsibility. |
ea280e7b | 3472 | */ |
b5f99b53 GC |
3473 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
3474 | { | |
ea280e7b TH |
3475 | bool ret; |
3476 | ||
ea280e7b TH |
3477 | rcu_read_lock(); |
3478 | ret = css_next_child(NULL, &memcg->css); | |
3479 | rcu_read_unlock(); | |
3480 | return ret; | |
b5f99b53 GC |
3481 | } |
3482 | ||
c26251f9 | 3483 | /* |
51038171 | 3484 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
3485 | * |
3486 | * Caller is responsible for holding css reference for memcg. | |
3487 | */ | |
3488 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3489 | { | |
d977aa93 | 3490 | int nr_retries = MAX_RECLAIM_RETRIES; |
c26251f9 | 3491 | |
c1e862c1 KH |
3492 | /* we call try-to-free pages for make this cgroup empty */ |
3493 | lru_add_drain_all(); | |
d12c60f6 JS |
3494 | |
3495 | drain_all_stock(memcg); | |
3496 | ||
f817ed48 | 3497 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3498 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 3499 | int progress; |
c1e862c1 | 3500 | |
c26251f9 MH |
3501 | if (signal_pending(current)) |
3502 | return -EINTR; | |
3503 | ||
b70a2a21 JW |
3504 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3505 | GFP_KERNEL, true); | |
c1e862c1 | 3506 | if (!progress) { |
f817ed48 | 3507 | nr_retries--; |
c1e862c1 | 3508 | /* maybe some writeback is necessary */ |
8aa7e847 | 3509 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3510 | } |
f817ed48 KH |
3511 | |
3512 | } | |
ab5196c2 MH |
3513 | |
3514 | return 0; | |
cc847582 KH |
3515 | } |
3516 | ||
6770c64e TH |
3517 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3518 | char *buf, size_t nbytes, | |
3519 | loff_t off) | |
c1e862c1 | 3520 | { |
6770c64e | 3521 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3522 | |
d8423011 MH |
3523 | if (mem_cgroup_is_root(memcg)) |
3524 | return -EINVAL; | |
6770c64e | 3525 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3526 | } |
3527 | ||
182446d0 TH |
3528 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3529 | struct cftype *cft) | |
18f59ea7 | 3530 | { |
182446d0 | 3531 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
3532 | } |
3533 | ||
182446d0 TH |
3534 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3535 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
3536 | { |
3537 | int retval = 0; | |
182446d0 | 3538 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 3539 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 3540 | |
567fb435 | 3541 | if (memcg->use_hierarchy == val) |
0b8f73e1 | 3542 | return 0; |
567fb435 | 3543 | |
18f59ea7 | 3544 | /* |
af901ca1 | 3545 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3546 | * in the child subtrees. If it is unset, then the change can |
3547 | * occur, provided the current cgroup has no children. | |
3548 | * | |
3549 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3550 | * set if there are no children. | |
3551 | */ | |
c0ff4b85 | 3552 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 3553 | (val == 1 || val == 0)) { |
ea280e7b | 3554 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 3555 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3556 | else |
3557 | retval = -EBUSY; | |
3558 | } else | |
3559 | retval = -EINVAL; | |
567fb435 | 3560 | |
18f59ea7 BS |
3561 | return retval; |
3562 | } | |
3563 | ||
6f646156 | 3564 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 3565 | { |
42a30035 | 3566 | unsigned long val; |
ce00a967 | 3567 | |
3e32cb2e | 3568 | if (mem_cgroup_is_root(memcg)) { |
0d1c2072 | 3569 | val = memcg_page_state(memcg, NR_FILE_PAGES) + |
be5d0a74 | 3570 | memcg_page_state(memcg, NR_ANON_MAPPED); |
42a30035 JW |
3571 | if (swap) |
3572 | val += memcg_page_state(memcg, MEMCG_SWAP); | |
3e32cb2e | 3573 | } else { |
ce00a967 | 3574 | if (!swap) |
3e32cb2e | 3575 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3576 | else |
3e32cb2e | 3577 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3578 | } |
c12176d3 | 3579 | return val; |
ce00a967 JW |
3580 | } |
3581 | ||
3e32cb2e JW |
3582 | enum { |
3583 | RES_USAGE, | |
3584 | RES_LIMIT, | |
3585 | RES_MAX_USAGE, | |
3586 | RES_FAILCNT, | |
3587 | RES_SOFT_LIMIT, | |
3588 | }; | |
ce00a967 | 3589 | |
791badbd | 3590 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3591 | struct cftype *cft) |
8cdea7c0 | 3592 | { |
182446d0 | 3593 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3594 | struct page_counter *counter; |
af36f906 | 3595 | |
3e32cb2e | 3596 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3597 | case _MEM: |
3e32cb2e JW |
3598 | counter = &memcg->memory; |
3599 | break; | |
8c7c6e34 | 3600 | case _MEMSWAP: |
3e32cb2e JW |
3601 | counter = &memcg->memsw; |
3602 | break; | |
510fc4e1 | 3603 | case _KMEM: |
3e32cb2e | 3604 | counter = &memcg->kmem; |
510fc4e1 | 3605 | break; |
d55f90bf | 3606 | case _TCP: |
0db15298 | 3607 | counter = &memcg->tcpmem; |
d55f90bf | 3608 | break; |
8c7c6e34 KH |
3609 | default: |
3610 | BUG(); | |
8c7c6e34 | 3611 | } |
3e32cb2e JW |
3612 | |
3613 | switch (MEMFILE_ATTR(cft->private)) { | |
3614 | case RES_USAGE: | |
3615 | if (counter == &memcg->memory) | |
c12176d3 | 3616 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 3617 | if (counter == &memcg->memsw) |
c12176d3 | 3618 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
3619 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
3620 | case RES_LIMIT: | |
bbec2e15 | 3621 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
3622 | case RES_MAX_USAGE: |
3623 | return (u64)counter->watermark * PAGE_SIZE; | |
3624 | case RES_FAILCNT: | |
3625 | return counter->failcnt; | |
3626 | case RES_SOFT_LIMIT: | |
3627 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3628 | default: | |
3629 | BUG(); | |
3630 | } | |
8cdea7c0 | 3631 | } |
510fc4e1 | 3632 | |
4a87e2a2 | 3633 | static void memcg_flush_percpu_vmstats(struct mem_cgroup *memcg) |
c350a99e | 3634 | { |
4a87e2a2 | 3635 | unsigned long stat[MEMCG_NR_STAT] = {0}; |
c350a99e RG |
3636 | struct mem_cgroup *mi; |
3637 | int node, cpu, i; | |
c350a99e RG |
3638 | |
3639 | for_each_online_cpu(cpu) | |
4a87e2a2 | 3640 | for (i = 0; i < MEMCG_NR_STAT; i++) |
6c1c2808 | 3641 | stat[i] += per_cpu(memcg->vmstats_percpu->stat[i], cpu); |
c350a99e RG |
3642 | |
3643 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
4a87e2a2 | 3644 | for (i = 0; i < MEMCG_NR_STAT; i++) |
c350a99e RG |
3645 | atomic_long_add(stat[i], &mi->vmstats[i]); |
3646 | ||
3647 | for_each_node(node) { | |
3648 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; | |
3649 | struct mem_cgroup_per_node *pi; | |
3650 | ||
4a87e2a2 | 3651 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
c350a99e RG |
3652 | stat[i] = 0; |
3653 | ||
3654 | for_each_online_cpu(cpu) | |
4a87e2a2 | 3655 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
6c1c2808 SB |
3656 | stat[i] += per_cpu( |
3657 | pn->lruvec_stat_cpu->count[i], cpu); | |
c350a99e RG |
3658 | |
3659 | for (pi = pn; pi; pi = parent_nodeinfo(pi, node)) | |
4a87e2a2 | 3660 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
c350a99e RG |
3661 | atomic_long_add(stat[i], &pi->lruvec_stat[i]); |
3662 | } | |
3663 | } | |
3664 | ||
bb65f89b RG |
3665 | static void memcg_flush_percpu_vmevents(struct mem_cgroup *memcg) |
3666 | { | |
3667 | unsigned long events[NR_VM_EVENT_ITEMS]; | |
3668 | struct mem_cgroup *mi; | |
3669 | int cpu, i; | |
3670 | ||
3671 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
3672 | events[i] = 0; | |
3673 | ||
3674 | for_each_online_cpu(cpu) | |
3675 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
6c1c2808 SB |
3676 | events[i] += per_cpu(memcg->vmstats_percpu->events[i], |
3677 | cpu); | |
bb65f89b RG |
3678 | |
3679 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
3680 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
3681 | atomic_long_add(events[i], &mi->vmevents[i]); | |
3682 | } | |
3683 | ||
84c07d11 | 3684 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 3685 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 3686 | { |
bf4f0599 | 3687 | struct obj_cgroup *objcg; |
d6441637 VD |
3688 | int memcg_id; |
3689 | ||
b313aeee VD |
3690 | if (cgroup_memory_nokmem) |
3691 | return 0; | |
3692 | ||
2a4db7eb | 3693 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 3694 | BUG_ON(memcg->kmem_state); |
d6441637 | 3695 | |
f3bb3043 | 3696 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
3697 | if (memcg_id < 0) |
3698 | return memcg_id; | |
d6441637 | 3699 | |
bf4f0599 RG |
3700 | objcg = obj_cgroup_alloc(); |
3701 | if (!objcg) { | |
3702 | memcg_free_cache_id(memcg_id); | |
3703 | return -ENOMEM; | |
3704 | } | |
3705 | objcg->memcg = memcg; | |
3706 | rcu_assign_pointer(memcg->objcg, objcg); | |
3707 | ||
d648bcc7 RG |
3708 | static_branch_enable(&memcg_kmem_enabled_key); |
3709 | ||
d6441637 | 3710 | /* |
567e9ab2 | 3711 | * A memory cgroup is considered kmem-online as soon as it gets |
900a38f0 | 3712 | * kmemcg_id. Setting the id after enabling static branching will |
d6441637 VD |
3713 | * guarantee no one starts accounting before all call sites are |
3714 | * patched. | |
3715 | */ | |
900a38f0 | 3716 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 3717 | memcg->kmem_state = KMEM_ONLINE; |
0b8f73e1 JW |
3718 | |
3719 | return 0; | |
d6441637 VD |
3720 | } |
3721 | ||
8e0a8912 JW |
3722 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
3723 | { | |
3724 | struct cgroup_subsys_state *css; | |
3725 | struct mem_cgroup *parent, *child; | |
3726 | int kmemcg_id; | |
3727 | ||
3728 | if (memcg->kmem_state != KMEM_ONLINE) | |
3729 | return; | |
9855609b | 3730 | |
8e0a8912 JW |
3731 | memcg->kmem_state = KMEM_ALLOCATED; |
3732 | ||
8e0a8912 JW |
3733 | parent = parent_mem_cgroup(memcg); |
3734 | if (!parent) | |
3735 | parent = root_mem_cgroup; | |
3736 | ||
bf4f0599 | 3737 | memcg_reparent_objcgs(memcg, parent); |
fb2f2b0a RG |
3738 | |
3739 | kmemcg_id = memcg->kmemcg_id; | |
3740 | BUG_ON(kmemcg_id < 0); | |
3741 | ||
8e0a8912 JW |
3742 | /* |
3743 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3744 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3745 | * to ones of the parent. After we have finished, all list_lrus | |
3746 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3747 | * ordering is imposed by list_lru_node->lock taken by | |
3748 | * memcg_drain_all_list_lrus(). | |
3749 | */ | |
3a06bb78 | 3750 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
8e0a8912 JW |
3751 | css_for_each_descendant_pre(css, &memcg->css) { |
3752 | child = mem_cgroup_from_css(css); | |
3753 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3754 | child->kmemcg_id = parent->kmemcg_id; | |
3755 | if (!memcg->use_hierarchy) | |
3756 | break; | |
3757 | } | |
3a06bb78 TH |
3758 | rcu_read_unlock(); |
3759 | ||
9bec5c35 | 3760 | memcg_drain_all_list_lrus(kmemcg_id, parent); |
8e0a8912 JW |
3761 | |
3762 | memcg_free_cache_id(kmemcg_id); | |
3763 | } | |
3764 | ||
3765 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3766 | { | |
0b8f73e1 JW |
3767 | /* css_alloc() failed, offlining didn't happen */ |
3768 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
3769 | memcg_offline_kmem(memcg); | |
8e0a8912 | 3770 | } |
d6441637 | 3771 | #else |
0b8f73e1 | 3772 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
3773 | { |
3774 | return 0; | |
3775 | } | |
3776 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
3777 | { | |
3778 | } | |
3779 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3780 | { | |
3781 | } | |
84c07d11 | 3782 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 3783 | |
bbec2e15 RG |
3784 | static int memcg_update_kmem_max(struct mem_cgroup *memcg, |
3785 | unsigned long max) | |
d6441637 | 3786 | { |
b313aeee | 3787 | int ret; |
127424c8 | 3788 | |
bbec2e15 RG |
3789 | mutex_lock(&memcg_max_mutex); |
3790 | ret = page_counter_set_max(&memcg->kmem, max); | |
3791 | mutex_unlock(&memcg_max_mutex); | |
127424c8 | 3792 | return ret; |
d6441637 | 3793 | } |
510fc4e1 | 3794 | |
bbec2e15 | 3795 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
3796 | { |
3797 | int ret; | |
3798 | ||
bbec2e15 | 3799 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 3800 | |
bbec2e15 | 3801 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
3802 | if (ret) |
3803 | goto out; | |
3804 | ||
0db15298 | 3805 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
3806 | /* |
3807 | * The active flag needs to be written after the static_key | |
3808 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
3809 | * function is the last one to run. See mem_cgroup_sk_alloc() |
3810 | * for details, and note that we don't mark any socket as | |
3811 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
3812 | * |
3813 | * We need to do this, because static_keys will span multiple | |
3814 | * sites, but we can't control their order. If we mark a socket | |
3815 | * as accounted, but the accounting functions are not patched in | |
3816 | * yet, we'll lose accounting. | |
3817 | * | |
2d758073 | 3818 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
3819 | * because when this value change, the code to process it is not |
3820 | * patched in yet. | |
3821 | */ | |
3822 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 3823 | memcg->tcpmem_active = true; |
d55f90bf VD |
3824 | } |
3825 | out: | |
bbec2e15 | 3826 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
3827 | return ret; |
3828 | } | |
d55f90bf | 3829 | |
628f4235 KH |
3830 | /* |
3831 | * The user of this function is... | |
3832 | * RES_LIMIT. | |
3833 | */ | |
451af504 TH |
3834 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3835 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3836 | { |
451af504 | 3837 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3838 | unsigned long nr_pages; |
628f4235 KH |
3839 | int ret; |
3840 | ||
451af504 | 3841 | buf = strstrip(buf); |
650c5e56 | 3842 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3843 | if (ret) |
3844 | return ret; | |
af36f906 | 3845 | |
3e32cb2e | 3846 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3847 | case RES_LIMIT: |
4b3bde4c BS |
3848 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3849 | ret = -EINVAL; | |
3850 | break; | |
3851 | } | |
3e32cb2e JW |
3852 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3853 | case _MEM: | |
bbec2e15 | 3854 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 3855 | break; |
3e32cb2e | 3856 | case _MEMSWAP: |
bbec2e15 | 3857 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 3858 | break; |
3e32cb2e | 3859 | case _KMEM: |
0158115f MH |
3860 | pr_warn_once("kmem.limit_in_bytes is deprecated and will be removed. " |
3861 | "Please report your usecase to linux-mm@kvack.org if you " | |
3862 | "depend on this functionality.\n"); | |
bbec2e15 | 3863 | ret = memcg_update_kmem_max(memcg, nr_pages); |
3e32cb2e | 3864 | break; |
d55f90bf | 3865 | case _TCP: |
bbec2e15 | 3866 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 3867 | break; |
3e32cb2e | 3868 | } |
296c81d8 | 3869 | break; |
3e32cb2e JW |
3870 | case RES_SOFT_LIMIT: |
3871 | memcg->soft_limit = nr_pages; | |
3872 | ret = 0; | |
628f4235 KH |
3873 | break; |
3874 | } | |
451af504 | 3875 | return ret ?: nbytes; |
8cdea7c0 BS |
3876 | } |
3877 | ||
6770c64e TH |
3878 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3879 | size_t nbytes, loff_t off) | |
c84872e1 | 3880 | { |
6770c64e | 3881 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3882 | struct page_counter *counter; |
c84872e1 | 3883 | |
3e32cb2e JW |
3884 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3885 | case _MEM: | |
3886 | counter = &memcg->memory; | |
3887 | break; | |
3888 | case _MEMSWAP: | |
3889 | counter = &memcg->memsw; | |
3890 | break; | |
3891 | case _KMEM: | |
3892 | counter = &memcg->kmem; | |
3893 | break; | |
d55f90bf | 3894 | case _TCP: |
0db15298 | 3895 | counter = &memcg->tcpmem; |
d55f90bf | 3896 | break; |
3e32cb2e JW |
3897 | default: |
3898 | BUG(); | |
3899 | } | |
af36f906 | 3900 | |
3e32cb2e | 3901 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3902 | case RES_MAX_USAGE: |
3e32cb2e | 3903 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3904 | break; |
3905 | case RES_FAILCNT: | |
3e32cb2e | 3906 | counter->failcnt = 0; |
29f2a4da | 3907 | break; |
3e32cb2e JW |
3908 | default: |
3909 | BUG(); | |
29f2a4da | 3910 | } |
f64c3f54 | 3911 | |
6770c64e | 3912 | return nbytes; |
c84872e1 PE |
3913 | } |
3914 | ||
182446d0 | 3915 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3916 | struct cftype *cft) |
3917 | { | |
182446d0 | 3918 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3919 | } |
3920 | ||
02491447 | 3921 | #ifdef CONFIG_MMU |
182446d0 | 3922 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3923 | struct cftype *cft, u64 val) |
3924 | { | |
182446d0 | 3925 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3926 | |
1dfab5ab | 3927 | if (val & ~MOVE_MASK) |
7dc74be0 | 3928 | return -EINVAL; |
ee5e8472 | 3929 | |
7dc74be0 | 3930 | /* |
ee5e8472 GC |
3931 | * No kind of locking is needed in here, because ->can_attach() will |
3932 | * check this value once in the beginning of the process, and then carry | |
3933 | * on with stale data. This means that changes to this value will only | |
3934 | * affect task migrations starting after the change. | |
7dc74be0 | 3935 | */ |
c0ff4b85 | 3936 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3937 | return 0; |
3938 | } | |
02491447 | 3939 | #else |
182446d0 | 3940 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3941 | struct cftype *cft, u64 val) |
3942 | { | |
3943 | return -ENOSYS; | |
3944 | } | |
3945 | #endif | |
7dc74be0 | 3946 | |
406eb0c9 | 3947 | #ifdef CONFIG_NUMA |
113b7dfd JW |
3948 | |
3949 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
3950 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
3951 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) | |
3952 | ||
3953 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 | 3954 | int nid, unsigned int lru_mask, bool tree) |
113b7dfd | 3955 | { |
867e5e1d | 3956 | struct lruvec *lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); |
113b7dfd JW |
3957 | unsigned long nr = 0; |
3958 | enum lru_list lru; | |
3959 | ||
3960 | VM_BUG_ON((unsigned)nid >= nr_node_ids); | |
3961 | ||
3962 | for_each_lru(lru) { | |
3963 | if (!(BIT(lru) & lru_mask)) | |
3964 | continue; | |
dd8657b6 SB |
3965 | if (tree) |
3966 | nr += lruvec_page_state(lruvec, NR_LRU_BASE + lru); | |
3967 | else | |
3968 | nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru); | |
113b7dfd JW |
3969 | } |
3970 | return nr; | |
3971 | } | |
3972 | ||
3973 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 SB |
3974 | unsigned int lru_mask, |
3975 | bool tree) | |
113b7dfd JW |
3976 | { |
3977 | unsigned long nr = 0; | |
3978 | enum lru_list lru; | |
3979 | ||
3980 | for_each_lru(lru) { | |
3981 | if (!(BIT(lru) & lru_mask)) | |
3982 | continue; | |
dd8657b6 SB |
3983 | if (tree) |
3984 | nr += memcg_page_state(memcg, NR_LRU_BASE + lru); | |
3985 | else | |
3986 | nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru); | |
113b7dfd JW |
3987 | } |
3988 | return nr; | |
3989 | } | |
3990 | ||
2da8ca82 | 3991 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3992 | { |
25485de6 GT |
3993 | struct numa_stat { |
3994 | const char *name; | |
3995 | unsigned int lru_mask; | |
3996 | }; | |
3997 | ||
3998 | static const struct numa_stat stats[] = { | |
3999 | { "total", LRU_ALL }, | |
4000 | { "file", LRU_ALL_FILE }, | |
4001 | { "anon", LRU_ALL_ANON }, | |
4002 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
4003 | }; | |
4004 | const struct numa_stat *stat; | |
406eb0c9 | 4005 | int nid; |
aa9694bb | 4006 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
406eb0c9 | 4007 | |
25485de6 | 4008 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
4009 | seq_printf(m, "%s=%lu", stat->name, |
4010 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
4011 | false)); | |
4012 | for_each_node_state(nid, N_MEMORY) | |
4013 | seq_printf(m, " N%d=%lu", nid, | |
4014 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
4015 | stat->lru_mask, false)); | |
25485de6 | 4016 | seq_putc(m, '\n'); |
406eb0c9 | 4017 | } |
406eb0c9 | 4018 | |
071aee13 | 4019 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
4020 | |
4021 | seq_printf(m, "hierarchical_%s=%lu", stat->name, | |
4022 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
4023 | true)); | |
4024 | for_each_node_state(nid, N_MEMORY) | |
4025 | seq_printf(m, " N%d=%lu", nid, | |
4026 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
4027 | stat->lru_mask, true)); | |
071aee13 | 4028 | seq_putc(m, '\n'); |
406eb0c9 | 4029 | } |
406eb0c9 | 4030 | |
406eb0c9 YH |
4031 | return 0; |
4032 | } | |
4033 | #endif /* CONFIG_NUMA */ | |
4034 | ||
c8713d0b | 4035 | static const unsigned int memcg1_stats[] = { |
0d1c2072 | 4036 | NR_FILE_PAGES, |
be5d0a74 | 4037 | NR_ANON_MAPPED, |
468c3982 JW |
4038 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4039 | NR_ANON_THPS, | |
4040 | #endif | |
c8713d0b JW |
4041 | NR_SHMEM, |
4042 | NR_FILE_MAPPED, | |
4043 | NR_FILE_DIRTY, | |
4044 | NR_WRITEBACK, | |
4045 | MEMCG_SWAP, | |
4046 | }; | |
4047 | ||
4048 | static const char *const memcg1_stat_names[] = { | |
4049 | "cache", | |
4050 | "rss", | |
468c3982 | 4051 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
c8713d0b | 4052 | "rss_huge", |
468c3982 | 4053 | #endif |
c8713d0b JW |
4054 | "shmem", |
4055 | "mapped_file", | |
4056 | "dirty", | |
4057 | "writeback", | |
4058 | "swap", | |
4059 | }; | |
4060 | ||
df0e53d0 | 4061 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 4062 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
4063 | PGPGIN, |
4064 | PGPGOUT, | |
4065 | PGFAULT, | |
4066 | PGMAJFAULT, | |
4067 | }; | |
4068 | ||
2da8ca82 | 4069 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 4070 | { |
aa9694bb | 4071 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3e32cb2e | 4072 | unsigned long memory, memsw; |
af7c4b0e JW |
4073 | struct mem_cgroup *mi; |
4074 | unsigned int i; | |
406eb0c9 | 4075 | |
71cd3113 | 4076 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c | 4077 | |
71cd3113 | 4078 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
468c3982 JW |
4079 | unsigned long nr; |
4080 | ||
71cd3113 | 4081 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 4082 | continue; |
468c3982 JW |
4083 | nr = memcg_page_state_local(memcg, memcg1_stats[i]); |
4084 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
4085 | if (memcg1_stats[i] == NR_ANON_THPS) | |
4086 | nr *= HPAGE_PMD_NR; | |
4087 | #endif | |
4088 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], nr * PAGE_SIZE); | |
1dd3a273 | 4089 | } |
7b854121 | 4090 | |
df0e53d0 | 4091 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d | 4092 | seq_printf(m, "%s %lu\n", vm_event_name(memcg1_events[i]), |
205b20cc | 4093 | memcg_events_local(memcg, memcg1_events[i])); |
af7c4b0e JW |
4094 | |
4095 | for (i = 0; i < NR_LRU_LISTS; i++) | |
ebc5d83d | 4096 | seq_printf(m, "%s %lu\n", lru_list_name(i), |
205b20cc | 4097 | memcg_page_state_local(memcg, NR_LRU_BASE + i) * |
21d89d15 | 4098 | PAGE_SIZE); |
af7c4b0e | 4099 | |
14067bb3 | 4100 | /* Hierarchical information */ |
3e32cb2e JW |
4101 | memory = memsw = PAGE_COUNTER_MAX; |
4102 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
15b42562 CD |
4103 | memory = min(memory, READ_ONCE(mi->memory.max)); |
4104 | memsw = min(memsw, READ_ONCE(mi->memsw.max)); | |
fee7b548 | 4105 | } |
3e32cb2e JW |
4106 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
4107 | (u64)memory * PAGE_SIZE); | |
7941d214 | 4108 | if (do_memsw_account()) |
3e32cb2e JW |
4109 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
4110 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 4111 | |
8de7ecc6 | 4112 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
71cd3113 | 4113 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 4114 | continue; |
8de7ecc6 | 4115 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], |
dd923990 YS |
4116 | (u64)memcg_page_state(memcg, memcg1_stats[i]) * |
4117 | PAGE_SIZE); | |
af7c4b0e JW |
4118 | } |
4119 | ||
8de7ecc6 | 4120 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d KK |
4121 | seq_printf(m, "total_%s %llu\n", |
4122 | vm_event_name(memcg1_events[i]), | |
dd923990 | 4123 | (u64)memcg_events(memcg, memcg1_events[i])); |
af7c4b0e | 4124 | |
8de7ecc6 | 4125 | for (i = 0; i < NR_LRU_LISTS; i++) |
ebc5d83d | 4126 | seq_printf(m, "total_%s %llu\n", lru_list_name(i), |
42a30035 JW |
4127 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
4128 | PAGE_SIZE); | |
14067bb3 | 4129 | |
7f016ee8 | 4130 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 4131 | { |
ef8f2327 MG |
4132 | pg_data_t *pgdat; |
4133 | struct mem_cgroup_per_node *mz; | |
1431d4d1 JW |
4134 | unsigned long anon_cost = 0; |
4135 | unsigned long file_cost = 0; | |
7f016ee8 | 4136 | |
ef8f2327 MG |
4137 | for_each_online_pgdat(pgdat) { |
4138 | mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); | |
7f016ee8 | 4139 | |
1431d4d1 JW |
4140 | anon_cost += mz->lruvec.anon_cost; |
4141 | file_cost += mz->lruvec.file_cost; | |
ef8f2327 | 4142 | } |
1431d4d1 JW |
4143 | seq_printf(m, "anon_cost %lu\n", anon_cost); |
4144 | seq_printf(m, "file_cost %lu\n", file_cost); | |
7f016ee8 KM |
4145 | } |
4146 | #endif | |
4147 | ||
d2ceb9b7 KH |
4148 | return 0; |
4149 | } | |
4150 | ||
182446d0 TH |
4151 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
4152 | struct cftype *cft) | |
a7885eb8 | 4153 | { |
182446d0 | 4154 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4155 | |
1f4c025b | 4156 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4157 | } |
4158 | ||
182446d0 TH |
4159 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
4160 | struct cftype *cft, u64 val) | |
a7885eb8 | 4161 | { |
182446d0 | 4162 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4163 | |
3dae7fec | 4164 | if (val > 100) |
a7885eb8 KM |
4165 | return -EINVAL; |
4166 | ||
14208b0e | 4167 | if (css->parent) |
3dae7fec JW |
4168 | memcg->swappiness = val; |
4169 | else | |
4170 | vm_swappiness = val; | |
068b38c1 | 4171 | |
a7885eb8 KM |
4172 | return 0; |
4173 | } | |
4174 | ||
2e72b634 KS |
4175 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4176 | { | |
4177 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 4178 | unsigned long usage; |
2e72b634 KS |
4179 | int i; |
4180 | ||
4181 | rcu_read_lock(); | |
4182 | if (!swap) | |
2c488db2 | 4183 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4184 | else |
2c488db2 | 4185 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4186 | |
4187 | if (!t) | |
4188 | goto unlock; | |
4189 | ||
ce00a967 | 4190 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
4191 | |
4192 | /* | |
748dad36 | 4193 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4194 | * If it's not true, a threshold was crossed after last |
4195 | * call of __mem_cgroup_threshold(). | |
4196 | */ | |
5407a562 | 4197 | i = t->current_threshold; |
2e72b634 KS |
4198 | |
4199 | /* | |
4200 | * Iterate backward over array of thresholds starting from | |
4201 | * current_threshold and check if a threshold is crossed. | |
4202 | * If none of thresholds below usage is crossed, we read | |
4203 | * only one element of the array here. | |
4204 | */ | |
4205 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4206 | eventfd_signal(t->entries[i].eventfd, 1); | |
4207 | ||
4208 | /* i = current_threshold + 1 */ | |
4209 | i++; | |
4210 | ||
4211 | /* | |
4212 | * Iterate forward over array of thresholds starting from | |
4213 | * current_threshold+1 and check if a threshold is crossed. | |
4214 | * If none of thresholds above usage is crossed, we read | |
4215 | * only one element of the array here. | |
4216 | */ | |
4217 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4218 | eventfd_signal(t->entries[i].eventfd, 1); | |
4219 | ||
4220 | /* Update current_threshold */ | |
5407a562 | 4221 | t->current_threshold = i - 1; |
2e72b634 KS |
4222 | unlock: |
4223 | rcu_read_unlock(); | |
4224 | } | |
4225 | ||
4226 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4227 | { | |
ad4ca5f4 KS |
4228 | while (memcg) { |
4229 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 4230 | if (do_memsw_account()) |
ad4ca5f4 KS |
4231 | __mem_cgroup_threshold(memcg, true); |
4232 | ||
4233 | memcg = parent_mem_cgroup(memcg); | |
4234 | } | |
2e72b634 KS |
4235 | } |
4236 | ||
4237 | static int compare_thresholds(const void *a, const void *b) | |
4238 | { | |
4239 | const struct mem_cgroup_threshold *_a = a; | |
4240 | const struct mem_cgroup_threshold *_b = b; | |
4241 | ||
2bff24a3 GT |
4242 | if (_a->threshold > _b->threshold) |
4243 | return 1; | |
4244 | ||
4245 | if (_a->threshold < _b->threshold) | |
4246 | return -1; | |
4247 | ||
4248 | return 0; | |
2e72b634 KS |
4249 | } |
4250 | ||
c0ff4b85 | 4251 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4252 | { |
4253 | struct mem_cgroup_eventfd_list *ev; | |
4254 | ||
2bcf2e92 MH |
4255 | spin_lock(&memcg_oom_lock); |
4256 | ||
c0ff4b85 | 4257 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 4258 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
4259 | |
4260 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
4261 | return 0; |
4262 | } | |
4263 | ||
c0ff4b85 | 4264 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4265 | { |
7d74b06f KH |
4266 | struct mem_cgroup *iter; |
4267 | ||
c0ff4b85 | 4268 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4269 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4270 | } |
4271 | ||
59b6f873 | 4272 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4273 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4274 | { |
2c488db2 KS |
4275 | struct mem_cgroup_thresholds *thresholds; |
4276 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
4277 | unsigned long threshold; |
4278 | unsigned long usage; | |
2c488db2 | 4279 | int i, size, ret; |
2e72b634 | 4280 | |
650c5e56 | 4281 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
4282 | if (ret) |
4283 | return ret; | |
4284 | ||
4285 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4286 | |
05b84301 | 4287 | if (type == _MEM) { |
2c488db2 | 4288 | thresholds = &memcg->thresholds; |
ce00a967 | 4289 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4290 | } else if (type == _MEMSWAP) { |
2c488db2 | 4291 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4292 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4293 | } else |
2e72b634 KS |
4294 | BUG(); |
4295 | ||
2e72b634 | 4296 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4297 | if (thresholds->primary) |
2e72b634 KS |
4298 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4299 | ||
2c488db2 | 4300 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4301 | |
4302 | /* Allocate memory for new array of thresholds */ | |
67b8046f | 4303 | new = kmalloc(struct_size(new, entries, size), GFP_KERNEL); |
2c488db2 | 4304 | if (!new) { |
2e72b634 KS |
4305 | ret = -ENOMEM; |
4306 | goto unlock; | |
4307 | } | |
2c488db2 | 4308 | new->size = size; |
2e72b634 KS |
4309 | |
4310 | /* Copy thresholds (if any) to new array */ | |
e90342e6 GS |
4311 | if (thresholds->primary) |
4312 | memcpy(new->entries, thresholds->primary->entries, | |
4313 | flex_array_size(new, entries, size - 1)); | |
2c488db2 | 4314 | |
2e72b634 | 4315 | /* Add new threshold */ |
2c488db2 KS |
4316 | new->entries[size - 1].eventfd = eventfd; |
4317 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4318 | |
4319 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
61e604e6 | 4320 | sort(new->entries, size, sizeof(*new->entries), |
2e72b634 KS |
4321 | compare_thresholds, NULL); |
4322 | ||
4323 | /* Find current threshold */ | |
2c488db2 | 4324 | new->current_threshold = -1; |
2e72b634 | 4325 | for (i = 0; i < size; i++) { |
748dad36 | 4326 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4327 | /* |
2c488db2 KS |
4328 | * new->current_threshold will not be used until |
4329 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4330 | * it here. |
4331 | */ | |
2c488db2 | 4332 | ++new->current_threshold; |
748dad36 SZ |
4333 | } else |
4334 | break; | |
2e72b634 KS |
4335 | } |
4336 | ||
2c488db2 KS |
4337 | /* Free old spare buffer and save old primary buffer as spare */ |
4338 | kfree(thresholds->spare); | |
4339 | thresholds->spare = thresholds->primary; | |
4340 | ||
4341 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4342 | |
907860ed | 4343 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4344 | synchronize_rcu(); |
4345 | ||
2e72b634 KS |
4346 | unlock: |
4347 | mutex_unlock(&memcg->thresholds_lock); | |
4348 | ||
4349 | return ret; | |
4350 | } | |
4351 | ||
59b6f873 | 4352 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4353 | struct eventfd_ctx *eventfd, const char *args) |
4354 | { | |
59b6f873 | 4355 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4356 | } |
4357 | ||
59b6f873 | 4358 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4359 | struct eventfd_ctx *eventfd, const char *args) |
4360 | { | |
59b6f873 | 4361 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4362 | } |
4363 | ||
59b6f873 | 4364 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4365 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4366 | { |
2c488db2 KS |
4367 | struct mem_cgroup_thresholds *thresholds; |
4368 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 4369 | unsigned long usage; |
7d36665a | 4370 | int i, j, size, entries; |
2e72b634 KS |
4371 | |
4372 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4373 | |
4374 | if (type == _MEM) { | |
2c488db2 | 4375 | thresholds = &memcg->thresholds; |
ce00a967 | 4376 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4377 | } else if (type == _MEMSWAP) { |
2c488db2 | 4378 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4379 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4380 | } else |
2e72b634 KS |
4381 | BUG(); |
4382 | ||
371528ca AV |
4383 | if (!thresholds->primary) |
4384 | goto unlock; | |
4385 | ||
2e72b634 KS |
4386 | /* Check if a threshold crossed before removing */ |
4387 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4388 | ||
4389 | /* Calculate new number of threshold */ | |
7d36665a | 4390 | size = entries = 0; |
2c488db2 KS |
4391 | for (i = 0; i < thresholds->primary->size; i++) { |
4392 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 | 4393 | size++; |
7d36665a CX |
4394 | else |
4395 | entries++; | |
2e72b634 KS |
4396 | } |
4397 | ||
2c488db2 | 4398 | new = thresholds->spare; |
907860ed | 4399 | |
7d36665a CX |
4400 | /* If no items related to eventfd have been cleared, nothing to do */ |
4401 | if (!entries) | |
4402 | goto unlock; | |
4403 | ||
2e72b634 KS |
4404 | /* Set thresholds array to NULL if we don't have thresholds */ |
4405 | if (!size) { | |
2c488db2 KS |
4406 | kfree(new); |
4407 | new = NULL; | |
907860ed | 4408 | goto swap_buffers; |
2e72b634 KS |
4409 | } |
4410 | ||
2c488db2 | 4411 | new->size = size; |
2e72b634 KS |
4412 | |
4413 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4414 | new->current_threshold = -1; |
4415 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4416 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4417 | continue; |
4418 | ||
2c488db2 | 4419 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4420 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4421 | /* |
2c488db2 | 4422 | * new->current_threshold will not be used |
2e72b634 KS |
4423 | * until rcu_assign_pointer(), so it's safe to increment |
4424 | * it here. | |
4425 | */ | |
2c488db2 | 4426 | ++new->current_threshold; |
2e72b634 KS |
4427 | } |
4428 | j++; | |
4429 | } | |
4430 | ||
907860ed | 4431 | swap_buffers: |
2c488db2 KS |
4432 | /* Swap primary and spare array */ |
4433 | thresholds->spare = thresholds->primary; | |
8c757763 | 4434 | |
2c488db2 | 4435 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4436 | |
907860ed | 4437 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4438 | synchronize_rcu(); |
6611d8d7 MC |
4439 | |
4440 | /* If all events are unregistered, free the spare array */ | |
4441 | if (!new) { | |
4442 | kfree(thresholds->spare); | |
4443 | thresholds->spare = NULL; | |
4444 | } | |
371528ca | 4445 | unlock: |
2e72b634 | 4446 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4447 | } |
c1e862c1 | 4448 | |
59b6f873 | 4449 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4450 | struct eventfd_ctx *eventfd) |
4451 | { | |
59b6f873 | 4452 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4453 | } |
4454 | ||
59b6f873 | 4455 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4456 | struct eventfd_ctx *eventfd) |
4457 | { | |
59b6f873 | 4458 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4459 | } |
4460 | ||
59b6f873 | 4461 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4462 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4463 | { |
9490ff27 | 4464 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4465 | |
9490ff27 KH |
4466 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4467 | if (!event) | |
4468 | return -ENOMEM; | |
4469 | ||
1af8efe9 | 4470 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4471 | |
4472 | event->eventfd = eventfd; | |
4473 | list_add(&event->list, &memcg->oom_notify); | |
4474 | ||
4475 | /* already in OOM ? */ | |
c2b42d3c | 4476 | if (memcg->under_oom) |
9490ff27 | 4477 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4478 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4479 | |
4480 | return 0; | |
4481 | } | |
4482 | ||
59b6f873 | 4483 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4484 | struct eventfd_ctx *eventfd) |
9490ff27 | 4485 | { |
9490ff27 | 4486 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4487 | |
1af8efe9 | 4488 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4489 | |
c0ff4b85 | 4490 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4491 | if (ev->eventfd == eventfd) { |
4492 | list_del(&ev->list); | |
4493 | kfree(ev); | |
4494 | } | |
4495 | } | |
4496 | ||
1af8efe9 | 4497 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4498 | } |
4499 | ||
2da8ca82 | 4500 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4501 | { |
aa9694bb | 4502 | struct mem_cgroup *memcg = mem_cgroup_from_seq(sf); |
3c11ecf4 | 4503 | |
791badbd | 4504 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 4505 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
4506 | seq_printf(sf, "oom_kill %lu\n", |
4507 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
4508 | return 0; |
4509 | } | |
4510 | ||
182446d0 | 4511 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4512 | struct cftype *cft, u64 val) |
4513 | { | |
182446d0 | 4514 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4515 | |
4516 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 4517 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4518 | return -EINVAL; |
4519 | ||
c0ff4b85 | 4520 | memcg->oom_kill_disable = val; |
4d845ebf | 4521 | if (!val) |
c0ff4b85 | 4522 | memcg_oom_recover(memcg); |
3dae7fec | 4523 | |
3c11ecf4 KH |
4524 | return 0; |
4525 | } | |
4526 | ||
52ebea74 TH |
4527 | #ifdef CONFIG_CGROUP_WRITEBACK |
4528 | ||
3a8e9ac8 TH |
4529 | #include <trace/events/writeback.h> |
4530 | ||
841710aa TH |
4531 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
4532 | { | |
4533 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
4534 | } | |
4535 | ||
4536 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4537 | { | |
4538 | wb_domain_exit(&memcg->cgwb_domain); | |
4539 | } | |
4540 | ||
2529bb3a TH |
4541 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4542 | { | |
4543 | wb_domain_size_changed(&memcg->cgwb_domain); | |
4544 | } | |
4545 | ||
841710aa TH |
4546 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
4547 | { | |
4548 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4549 | ||
4550 | if (!memcg->css.parent) | |
4551 | return NULL; | |
4552 | ||
4553 | return &memcg->cgwb_domain; | |
4554 | } | |
4555 | ||
0b3d6e6f GT |
4556 | /* |
4557 | * idx can be of type enum memcg_stat_item or node_stat_item. | |
4558 | * Keep in sync with memcg_exact_page(). | |
4559 | */ | |
4560 | static unsigned long memcg_exact_page_state(struct mem_cgroup *memcg, int idx) | |
4561 | { | |
871789d4 | 4562 | long x = atomic_long_read(&memcg->vmstats[idx]); |
0b3d6e6f GT |
4563 | int cpu; |
4564 | ||
4565 | for_each_online_cpu(cpu) | |
871789d4 | 4566 | x += per_cpu_ptr(memcg->vmstats_percpu, cpu)->stat[idx]; |
0b3d6e6f GT |
4567 | if (x < 0) |
4568 | x = 0; | |
4569 | return x; | |
4570 | } | |
4571 | ||
c2aa723a TH |
4572 | /** |
4573 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
4574 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
4575 | * @pfilepages: out parameter for number of file pages |
4576 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
4577 | * @pdirty: out parameter for number of dirty pages |
4578 | * @pwriteback: out parameter for number of pages under writeback | |
4579 | * | |
c5edf9cd TH |
4580 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
4581 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
4582 | * is a bit more involved. | |
c2aa723a | 4583 | * |
c5edf9cd TH |
4584 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
4585 | * headroom is calculated as the lowest headroom of itself and the | |
4586 | * ancestors. Note that this doesn't consider the actual amount of | |
4587 | * available memory in the system. The caller should further cap | |
4588 | * *@pheadroom accordingly. | |
c2aa723a | 4589 | */ |
c5edf9cd TH |
4590 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
4591 | unsigned long *pheadroom, unsigned long *pdirty, | |
4592 | unsigned long *pwriteback) | |
c2aa723a TH |
4593 | { |
4594 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4595 | struct mem_cgroup *parent; | |
c2aa723a | 4596 | |
0b3d6e6f | 4597 | *pdirty = memcg_exact_page_state(memcg, NR_FILE_DIRTY); |
c2aa723a | 4598 | |
0b3d6e6f | 4599 | *pwriteback = memcg_exact_page_state(memcg, NR_WRITEBACK); |
21d89d15 JW |
4600 | *pfilepages = memcg_exact_page_state(memcg, NR_INACTIVE_FILE) + |
4601 | memcg_exact_page_state(memcg, NR_ACTIVE_FILE); | |
c5edf9cd | 4602 | *pheadroom = PAGE_COUNTER_MAX; |
c2aa723a | 4603 | |
c2aa723a | 4604 | while ((parent = parent_mem_cgroup(memcg))) { |
15b42562 | 4605 | unsigned long ceiling = min(READ_ONCE(memcg->memory.max), |
d1663a90 | 4606 | READ_ONCE(memcg->memory.high)); |
c2aa723a TH |
4607 | unsigned long used = page_counter_read(&memcg->memory); |
4608 | ||
c5edf9cd | 4609 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
4610 | memcg = parent; |
4611 | } | |
c2aa723a TH |
4612 | } |
4613 | ||
97b27821 TH |
4614 | /* |
4615 | * Foreign dirty flushing | |
4616 | * | |
4617 | * There's an inherent mismatch between memcg and writeback. The former | |
4618 | * trackes ownership per-page while the latter per-inode. This was a | |
4619 | * deliberate design decision because honoring per-page ownership in the | |
4620 | * writeback path is complicated, may lead to higher CPU and IO overheads | |
4621 | * and deemed unnecessary given that write-sharing an inode across | |
4622 | * different cgroups isn't a common use-case. | |
4623 | * | |
4624 | * Combined with inode majority-writer ownership switching, this works well | |
4625 | * enough in most cases but there are some pathological cases. For | |
4626 | * example, let's say there are two cgroups A and B which keep writing to | |
4627 | * different but confined parts of the same inode. B owns the inode and | |
4628 | * A's memory is limited far below B's. A's dirty ratio can rise enough to | |
4629 | * trigger balance_dirty_pages() sleeps but B's can be low enough to avoid | |
4630 | * triggering background writeback. A will be slowed down without a way to | |
4631 | * make writeback of the dirty pages happen. | |
4632 | * | |
4633 | * Conditions like the above can lead to a cgroup getting repatedly and | |
4634 | * severely throttled after making some progress after each | |
4635 | * dirty_expire_interval while the underyling IO device is almost | |
4636 | * completely idle. | |
4637 | * | |
4638 | * Solving this problem completely requires matching the ownership tracking | |
4639 | * granularities between memcg and writeback in either direction. However, | |
4640 | * the more egregious behaviors can be avoided by simply remembering the | |
4641 | * most recent foreign dirtying events and initiating remote flushes on | |
4642 | * them when local writeback isn't enough to keep the memory clean enough. | |
4643 | * | |
4644 | * The following two functions implement such mechanism. When a foreign | |
4645 | * page - a page whose memcg and writeback ownerships don't match - is | |
4646 | * dirtied, mem_cgroup_track_foreign_dirty() records the inode owning | |
4647 | * bdi_writeback on the page owning memcg. When balance_dirty_pages() | |
4648 | * decides that the memcg needs to sleep due to high dirty ratio, it calls | |
4649 | * mem_cgroup_flush_foreign() which queues writeback on the recorded | |
4650 | * foreign bdi_writebacks which haven't expired. Both the numbers of | |
4651 | * recorded bdi_writebacks and concurrent in-flight foreign writebacks are | |
4652 | * limited to MEMCG_CGWB_FRN_CNT. | |
4653 | * | |
4654 | * The mechanism only remembers IDs and doesn't hold any object references. | |
4655 | * As being wrong occasionally doesn't matter, updates and accesses to the | |
4656 | * records are lockless and racy. | |
4657 | */ | |
4658 | void mem_cgroup_track_foreign_dirty_slowpath(struct page *page, | |
4659 | struct bdi_writeback *wb) | |
4660 | { | |
4661 | struct mem_cgroup *memcg = page->mem_cgroup; | |
4662 | struct memcg_cgwb_frn *frn; | |
4663 | u64 now = get_jiffies_64(); | |
4664 | u64 oldest_at = now; | |
4665 | int oldest = -1; | |
4666 | int i; | |
4667 | ||
3a8e9ac8 TH |
4668 | trace_track_foreign_dirty(page, wb); |
4669 | ||
97b27821 TH |
4670 | /* |
4671 | * Pick the slot to use. If there is already a slot for @wb, keep | |
4672 | * using it. If not replace the oldest one which isn't being | |
4673 | * written out. | |
4674 | */ | |
4675 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4676 | frn = &memcg->cgwb_frn[i]; | |
4677 | if (frn->bdi_id == wb->bdi->id && | |
4678 | frn->memcg_id == wb->memcg_css->id) | |
4679 | break; | |
4680 | if (time_before64(frn->at, oldest_at) && | |
4681 | atomic_read(&frn->done.cnt) == 1) { | |
4682 | oldest = i; | |
4683 | oldest_at = frn->at; | |
4684 | } | |
4685 | } | |
4686 | ||
4687 | if (i < MEMCG_CGWB_FRN_CNT) { | |
4688 | /* | |
4689 | * Re-using an existing one. Update timestamp lazily to | |
4690 | * avoid making the cacheline hot. We want them to be | |
4691 | * reasonably up-to-date and significantly shorter than | |
4692 | * dirty_expire_interval as that's what expires the record. | |
4693 | * Use the shorter of 1s and dirty_expire_interval / 8. | |
4694 | */ | |
4695 | unsigned long update_intv = | |
4696 | min_t(unsigned long, HZ, | |
4697 | msecs_to_jiffies(dirty_expire_interval * 10) / 8); | |
4698 | ||
4699 | if (time_before64(frn->at, now - update_intv)) | |
4700 | frn->at = now; | |
4701 | } else if (oldest >= 0) { | |
4702 | /* replace the oldest free one */ | |
4703 | frn = &memcg->cgwb_frn[oldest]; | |
4704 | frn->bdi_id = wb->bdi->id; | |
4705 | frn->memcg_id = wb->memcg_css->id; | |
4706 | frn->at = now; | |
4707 | } | |
4708 | } | |
4709 | ||
4710 | /* issue foreign writeback flushes for recorded foreign dirtying events */ | |
4711 | void mem_cgroup_flush_foreign(struct bdi_writeback *wb) | |
4712 | { | |
4713 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4714 | unsigned long intv = msecs_to_jiffies(dirty_expire_interval * 10); | |
4715 | u64 now = jiffies_64; | |
4716 | int i; | |
4717 | ||
4718 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4719 | struct memcg_cgwb_frn *frn = &memcg->cgwb_frn[i]; | |
4720 | ||
4721 | /* | |
4722 | * If the record is older than dirty_expire_interval, | |
4723 | * writeback on it has already started. No need to kick it | |
4724 | * off again. Also, don't start a new one if there's | |
4725 | * already one in flight. | |
4726 | */ | |
4727 | if (time_after64(frn->at, now - intv) && | |
4728 | atomic_read(&frn->done.cnt) == 1) { | |
4729 | frn->at = 0; | |
3a8e9ac8 | 4730 | trace_flush_foreign(wb, frn->bdi_id, frn->memcg_id); |
97b27821 TH |
4731 | cgroup_writeback_by_id(frn->bdi_id, frn->memcg_id, 0, |
4732 | WB_REASON_FOREIGN_FLUSH, | |
4733 | &frn->done); | |
4734 | } | |
4735 | } | |
4736 | } | |
4737 | ||
841710aa TH |
4738 | #else /* CONFIG_CGROUP_WRITEBACK */ |
4739 | ||
4740 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
4741 | { | |
4742 | return 0; | |
4743 | } | |
4744 | ||
4745 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4746 | { | |
4747 | } | |
4748 | ||
2529bb3a TH |
4749 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4750 | { | |
4751 | } | |
4752 | ||
52ebea74 TH |
4753 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
4754 | ||
3bc942f3 TH |
4755 | /* |
4756 | * DO NOT USE IN NEW FILES. | |
4757 | * | |
4758 | * "cgroup.event_control" implementation. | |
4759 | * | |
4760 | * This is way over-engineered. It tries to support fully configurable | |
4761 | * events for each user. Such level of flexibility is completely | |
4762 | * unnecessary especially in the light of the planned unified hierarchy. | |
4763 | * | |
4764 | * Please deprecate this and replace with something simpler if at all | |
4765 | * possible. | |
4766 | */ | |
4767 | ||
79bd9814 TH |
4768 | /* |
4769 | * Unregister event and free resources. | |
4770 | * | |
4771 | * Gets called from workqueue. | |
4772 | */ | |
3bc942f3 | 4773 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4774 | { |
3bc942f3 TH |
4775 | struct mem_cgroup_event *event = |
4776 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4777 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4778 | |
4779 | remove_wait_queue(event->wqh, &event->wait); | |
4780 | ||
59b6f873 | 4781 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4782 | |
4783 | /* Notify userspace the event is going away. */ | |
4784 | eventfd_signal(event->eventfd, 1); | |
4785 | ||
4786 | eventfd_ctx_put(event->eventfd); | |
4787 | kfree(event); | |
59b6f873 | 4788 | css_put(&memcg->css); |
79bd9814 TH |
4789 | } |
4790 | ||
4791 | /* | |
a9a08845 | 4792 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
4793 | * |
4794 | * Called with wqh->lock held and interrupts disabled. | |
4795 | */ | |
ac6424b9 | 4796 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 4797 | int sync, void *key) |
79bd9814 | 4798 | { |
3bc942f3 TH |
4799 | struct mem_cgroup_event *event = |
4800 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4801 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 4802 | __poll_t flags = key_to_poll(key); |
79bd9814 | 4803 | |
a9a08845 | 4804 | if (flags & EPOLLHUP) { |
79bd9814 TH |
4805 | /* |
4806 | * If the event has been detached at cgroup removal, we | |
4807 | * can simply return knowing the other side will cleanup | |
4808 | * for us. | |
4809 | * | |
4810 | * We can't race against event freeing since the other | |
4811 | * side will require wqh->lock via remove_wait_queue(), | |
4812 | * which we hold. | |
4813 | */ | |
fba94807 | 4814 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4815 | if (!list_empty(&event->list)) { |
4816 | list_del_init(&event->list); | |
4817 | /* | |
4818 | * We are in atomic context, but cgroup_event_remove() | |
4819 | * may sleep, so we have to call it in workqueue. | |
4820 | */ | |
4821 | schedule_work(&event->remove); | |
4822 | } | |
fba94807 | 4823 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4824 | } |
4825 | ||
4826 | return 0; | |
4827 | } | |
4828 | ||
3bc942f3 | 4829 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4830 | wait_queue_head_t *wqh, poll_table *pt) |
4831 | { | |
3bc942f3 TH |
4832 | struct mem_cgroup_event *event = |
4833 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4834 | |
4835 | event->wqh = wqh; | |
4836 | add_wait_queue(wqh, &event->wait); | |
4837 | } | |
4838 | ||
4839 | /* | |
3bc942f3 TH |
4840 | * DO NOT USE IN NEW FILES. |
4841 | * | |
79bd9814 TH |
4842 | * Parse input and register new cgroup event handler. |
4843 | * | |
4844 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4845 | * Interpretation of args is defined by control file implementation. | |
4846 | */ | |
451af504 TH |
4847 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4848 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4849 | { |
451af504 | 4850 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4851 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4852 | struct mem_cgroup_event *event; |
79bd9814 TH |
4853 | struct cgroup_subsys_state *cfile_css; |
4854 | unsigned int efd, cfd; | |
4855 | struct fd efile; | |
4856 | struct fd cfile; | |
fba94807 | 4857 | const char *name; |
79bd9814 TH |
4858 | char *endp; |
4859 | int ret; | |
4860 | ||
451af504 TH |
4861 | buf = strstrip(buf); |
4862 | ||
4863 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4864 | if (*endp != ' ') |
4865 | return -EINVAL; | |
451af504 | 4866 | buf = endp + 1; |
79bd9814 | 4867 | |
451af504 | 4868 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4869 | if ((*endp != ' ') && (*endp != '\0')) |
4870 | return -EINVAL; | |
451af504 | 4871 | buf = endp + 1; |
79bd9814 TH |
4872 | |
4873 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4874 | if (!event) | |
4875 | return -ENOMEM; | |
4876 | ||
59b6f873 | 4877 | event->memcg = memcg; |
79bd9814 | 4878 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4879 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4880 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4881 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4882 | |
4883 | efile = fdget(efd); | |
4884 | if (!efile.file) { | |
4885 | ret = -EBADF; | |
4886 | goto out_kfree; | |
4887 | } | |
4888 | ||
4889 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4890 | if (IS_ERR(event->eventfd)) { | |
4891 | ret = PTR_ERR(event->eventfd); | |
4892 | goto out_put_efile; | |
4893 | } | |
4894 | ||
4895 | cfile = fdget(cfd); | |
4896 | if (!cfile.file) { | |
4897 | ret = -EBADF; | |
4898 | goto out_put_eventfd; | |
4899 | } | |
4900 | ||
4901 | /* the process need read permission on control file */ | |
4902 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
4903 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
4904 | if (ret < 0) | |
4905 | goto out_put_cfile; | |
4906 | ||
fba94807 TH |
4907 | /* |
4908 | * Determine the event callbacks and set them in @event. This used | |
4909 | * to be done via struct cftype but cgroup core no longer knows | |
4910 | * about these events. The following is crude but the whole thing | |
4911 | * is for compatibility anyway. | |
3bc942f3 TH |
4912 | * |
4913 | * DO NOT ADD NEW FILES. | |
fba94807 | 4914 | */ |
b583043e | 4915 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4916 | |
4917 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4918 | event->register_event = mem_cgroup_usage_register_event; | |
4919 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4920 | } else if (!strcmp(name, "memory.oom_control")) { | |
4921 | event->register_event = mem_cgroup_oom_register_event; | |
4922 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4923 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4924 | event->register_event = vmpressure_register_event; | |
4925 | event->unregister_event = vmpressure_unregister_event; | |
4926 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4927 | event->register_event = memsw_cgroup_usage_register_event; |
4928 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4929 | } else { |
4930 | ret = -EINVAL; | |
4931 | goto out_put_cfile; | |
4932 | } | |
4933 | ||
79bd9814 | 4934 | /* |
b5557c4c TH |
4935 | * Verify @cfile should belong to @css. Also, remaining events are |
4936 | * automatically removed on cgroup destruction but the removal is | |
4937 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4938 | */ |
b583043e | 4939 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4940 | &memory_cgrp_subsys); |
79bd9814 | 4941 | ret = -EINVAL; |
5a17f543 | 4942 | if (IS_ERR(cfile_css)) |
79bd9814 | 4943 | goto out_put_cfile; |
5a17f543 TH |
4944 | if (cfile_css != css) { |
4945 | css_put(cfile_css); | |
79bd9814 | 4946 | goto out_put_cfile; |
5a17f543 | 4947 | } |
79bd9814 | 4948 | |
451af504 | 4949 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4950 | if (ret) |
4951 | goto out_put_css; | |
4952 | ||
9965ed17 | 4953 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 4954 | |
fba94807 TH |
4955 | spin_lock(&memcg->event_list_lock); |
4956 | list_add(&event->list, &memcg->event_list); | |
4957 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4958 | |
4959 | fdput(cfile); | |
4960 | fdput(efile); | |
4961 | ||
451af504 | 4962 | return nbytes; |
79bd9814 TH |
4963 | |
4964 | out_put_css: | |
b5557c4c | 4965 | css_put(css); |
79bd9814 TH |
4966 | out_put_cfile: |
4967 | fdput(cfile); | |
4968 | out_put_eventfd: | |
4969 | eventfd_ctx_put(event->eventfd); | |
4970 | out_put_efile: | |
4971 | fdput(efile); | |
4972 | out_kfree: | |
4973 | kfree(event); | |
4974 | ||
4975 | return ret; | |
4976 | } | |
4977 | ||
241994ed | 4978 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4979 | { |
0eea1030 | 4980 | .name = "usage_in_bytes", |
8c7c6e34 | 4981 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4982 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4983 | }, |
c84872e1 PE |
4984 | { |
4985 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4986 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4987 | .write = mem_cgroup_reset, |
791badbd | 4988 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4989 | }, |
8cdea7c0 | 4990 | { |
0eea1030 | 4991 | .name = "limit_in_bytes", |
8c7c6e34 | 4992 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4993 | .write = mem_cgroup_write, |
791badbd | 4994 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4995 | }, |
296c81d8 BS |
4996 | { |
4997 | .name = "soft_limit_in_bytes", | |
4998 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4999 | .write = mem_cgroup_write, |
791badbd | 5000 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 5001 | }, |
8cdea7c0 BS |
5002 | { |
5003 | .name = "failcnt", | |
8c7c6e34 | 5004 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 5005 | .write = mem_cgroup_reset, |
791badbd | 5006 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5007 | }, |
d2ceb9b7 KH |
5008 | { |
5009 | .name = "stat", | |
2da8ca82 | 5010 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 5011 | }, |
c1e862c1 KH |
5012 | { |
5013 | .name = "force_empty", | |
6770c64e | 5014 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 5015 | }, |
18f59ea7 BS |
5016 | { |
5017 | .name = "use_hierarchy", | |
5018 | .write_u64 = mem_cgroup_hierarchy_write, | |
5019 | .read_u64 = mem_cgroup_hierarchy_read, | |
5020 | }, | |
79bd9814 | 5021 | { |
3bc942f3 | 5022 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 5023 | .write = memcg_write_event_control, |
7dbdb199 | 5024 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 5025 | }, |
a7885eb8 KM |
5026 | { |
5027 | .name = "swappiness", | |
5028 | .read_u64 = mem_cgroup_swappiness_read, | |
5029 | .write_u64 = mem_cgroup_swappiness_write, | |
5030 | }, | |
7dc74be0 DN |
5031 | { |
5032 | .name = "move_charge_at_immigrate", | |
5033 | .read_u64 = mem_cgroup_move_charge_read, | |
5034 | .write_u64 = mem_cgroup_move_charge_write, | |
5035 | }, | |
9490ff27 KH |
5036 | { |
5037 | .name = "oom_control", | |
2da8ca82 | 5038 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 5039 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
5040 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
5041 | }, | |
70ddf637 AV |
5042 | { |
5043 | .name = "pressure_level", | |
70ddf637 | 5044 | }, |
406eb0c9 YH |
5045 | #ifdef CONFIG_NUMA |
5046 | { | |
5047 | .name = "numa_stat", | |
2da8ca82 | 5048 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
5049 | }, |
5050 | #endif | |
510fc4e1 GC |
5051 | { |
5052 | .name = "kmem.limit_in_bytes", | |
5053 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 5054 | .write = mem_cgroup_write, |
791badbd | 5055 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5056 | }, |
5057 | { | |
5058 | .name = "kmem.usage_in_bytes", | |
5059 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 5060 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5061 | }, |
5062 | { | |
5063 | .name = "kmem.failcnt", | |
5064 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 5065 | .write = mem_cgroup_reset, |
791badbd | 5066 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5067 | }, |
5068 | { | |
5069 | .name = "kmem.max_usage_in_bytes", | |
5070 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 5071 | .write = mem_cgroup_reset, |
791badbd | 5072 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 5073 | }, |
a87425a3 YS |
5074 | #if defined(CONFIG_MEMCG_KMEM) && \ |
5075 | (defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)) | |
749c5415 GC |
5076 | { |
5077 | .name = "kmem.slabinfo", | |
b047501c | 5078 | .seq_show = memcg_slab_show, |
749c5415 GC |
5079 | }, |
5080 | #endif | |
d55f90bf VD |
5081 | { |
5082 | .name = "kmem.tcp.limit_in_bytes", | |
5083 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
5084 | .write = mem_cgroup_write, | |
5085 | .read_u64 = mem_cgroup_read_u64, | |
5086 | }, | |
5087 | { | |
5088 | .name = "kmem.tcp.usage_in_bytes", | |
5089 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
5090 | .read_u64 = mem_cgroup_read_u64, | |
5091 | }, | |
5092 | { | |
5093 | .name = "kmem.tcp.failcnt", | |
5094 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
5095 | .write = mem_cgroup_reset, | |
5096 | .read_u64 = mem_cgroup_read_u64, | |
5097 | }, | |
5098 | { | |
5099 | .name = "kmem.tcp.max_usage_in_bytes", | |
5100 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
5101 | .write = mem_cgroup_reset, | |
5102 | .read_u64 = mem_cgroup_read_u64, | |
5103 | }, | |
6bc10349 | 5104 | { }, /* terminate */ |
af36f906 | 5105 | }; |
8c7c6e34 | 5106 | |
73f576c0 JW |
5107 | /* |
5108 | * Private memory cgroup IDR | |
5109 | * | |
5110 | * Swap-out records and page cache shadow entries need to store memcg | |
5111 | * references in constrained space, so we maintain an ID space that is | |
5112 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
5113 | * memory-controlled cgroups to 64k. | |
5114 | * | |
b8f2935f | 5115 | * However, there usually are many references to the offline CSS after |
73f576c0 JW |
5116 | * the cgroup has been destroyed, such as page cache or reclaimable |
5117 | * slab objects, that don't need to hang on to the ID. We want to keep | |
5118 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
5119 | * relatively small ID space and prevent the creation of new cgroups | |
5120 | * even when there are much fewer than 64k cgroups - possibly none. | |
5121 | * | |
5122 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
5123 | * be freed and recycled when it's no longer needed, which is usually | |
5124 | * when the CSS is offlined. | |
5125 | * | |
5126 | * The only exception to that are records of swapped out tmpfs/shmem | |
5127 | * pages that need to be attributed to live ancestors on swapin. But | |
5128 | * those references are manageable from userspace. | |
5129 | */ | |
5130 | ||
5131 | static DEFINE_IDR(mem_cgroup_idr); | |
5132 | ||
7e97de0b KT |
5133 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
5134 | { | |
5135 | if (memcg->id.id > 0) { | |
5136 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
5137 | memcg->id.id = 0; | |
5138 | } | |
5139 | } | |
5140 | ||
c1514c0a VF |
5141 | static void __maybe_unused mem_cgroup_id_get_many(struct mem_cgroup *memcg, |
5142 | unsigned int n) | |
73f576c0 | 5143 | { |
1c2d479a | 5144 | refcount_add(n, &memcg->id.ref); |
73f576c0 JW |
5145 | } |
5146 | ||
615d66c3 | 5147 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 5148 | { |
1c2d479a | 5149 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 5150 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
5151 | |
5152 | /* Memcg ID pins CSS */ | |
5153 | css_put(&memcg->css); | |
5154 | } | |
5155 | } | |
5156 | ||
615d66c3 VD |
5157 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) |
5158 | { | |
5159 | mem_cgroup_id_put_many(memcg, 1); | |
5160 | } | |
5161 | ||
73f576c0 JW |
5162 | /** |
5163 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
5164 | * @id: the memcg id to look up | |
5165 | * | |
5166 | * Caller must hold rcu_read_lock(). | |
5167 | */ | |
5168 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
5169 | { | |
5170 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5171 | return idr_find(&mem_cgroup_idr, id); | |
5172 | } | |
5173 | ||
ef8f2327 | 5174 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
5175 | { |
5176 | struct mem_cgroup_per_node *pn; | |
ef8f2327 | 5177 | int tmp = node; |
1ecaab2b KH |
5178 | /* |
5179 | * This routine is called against possible nodes. | |
5180 | * But it's BUG to call kmalloc() against offline node. | |
5181 | * | |
5182 | * TODO: this routine can waste much memory for nodes which will | |
5183 | * never be onlined. It's better to use memory hotplug callback | |
5184 | * function. | |
5185 | */ | |
41e3355d KH |
5186 | if (!node_state(node, N_NORMAL_MEMORY)) |
5187 | tmp = -1; | |
17295c88 | 5188 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
5189 | if (!pn) |
5190 | return 1; | |
1ecaab2b | 5191 | |
3e38e0aa RG |
5192 | pn->lruvec_stat_local = alloc_percpu_gfp(struct lruvec_stat, |
5193 | GFP_KERNEL_ACCOUNT); | |
815744d7 JW |
5194 | if (!pn->lruvec_stat_local) { |
5195 | kfree(pn); | |
5196 | return 1; | |
5197 | } | |
5198 | ||
3e38e0aa RG |
5199 | pn->lruvec_stat_cpu = alloc_percpu_gfp(struct lruvec_stat, |
5200 | GFP_KERNEL_ACCOUNT); | |
a983b5eb | 5201 | if (!pn->lruvec_stat_cpu) { |
815744d7 | 5202 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c JW |
5203 | kfree(pn); |
5204 | return 1; | |
5205 | } | |
5206 | ||
ef8f2327 MG |
5207 | lruvec_init(&pn->lruvec); |
5208 | pn->usage_in_excess = 0; | |
5209 | pn->on_tree = false; | |
5210 | pn->memcg = memcg; | |
5211 | ||
54f72fe0 | 5212 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
5213 | return 0; |
5214 | } | |
5215 | ||
ef8f2327 | 5216 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 5217 | { |
00f3ca2c JW |
5218 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
5219 | ||
4eaf431f MH |
5220 | if (!pn) |
5221 | return; | |
5222 | ||
a983b5eb | 5223 | free_percpu(pn->lruvec_stat_cpu); |
815744d7 | 5224 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c | 5225 | kfree(pn); |
1ecaab2b KH |
5226 | } |
5227 | ||
40e952f9 | 5228 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 5229 | { |
c8b2a36f | 5230 | int node; |
59927fb9 | 5231 | |
c8b2a36f | 5232 | for_each_node(node) |
ef8f2327 | 5233 | free_mem_cgroup_per_node_info(memcg, node); |
871789d4 | 5234 | free_percpu(memcg->vmstats_percpu); |
815744d7 | 5235 | free_percpu(memcg->vmstats_local); |
8ff69e2c | 5236 | kfree(memcg); |
59927fb9 | 5237 | } |
3afe36b1 | 5238 | |
40e952f9 TE |
5239 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
5240 | { | |
5241 | memcg_wb_domain_exit(memcg); | |
7961eee3 SB |
5242 | /* |
5243 | * Flush percpu vmstats and vmevents to guarantee the value correctness | |
5244 | * on parent's and all ancestor levels. | |
5245 | */ | |
4a87e2a2 | 5246 | memcg_flush_percpu_vmstats(memcg); |
7961eee3 | 5247 | memcg_flush_percpu_vmevents(memcg); |
40e952f9 TE |
5248 | __mem_cgroup_free(memcg); |
5249 | } | |
5250 | ||
0b8f73e1 | 5251 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 5252 | { |
d142e3e6 | 5253 | struct mem_cgroup *memcg; |
b9726c26 | 5254 | unsigned int size; |
6d12e2d8 | 5255 | int node; |
97b27821 | 5256 | int __maybe_unused i; |
11d67612 | 5257 | long error = -ENOMEM; |
8cdea7c0 | 5258 | |
0b8f73e1 JW |
5259 | size = sizeof(struct mem_cgroup); |
5260 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
5261 | ||
5262 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 5263 | if (!memcg) |
11d67612 | 5264 | return ERR_PTR(error); |
0b8f73e1 | 5265 | |
73f576c0 JW |
5266 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
5267 | 1, MEM_CGROUP_ID_MAX, | |
5268 | GFP_KERNEL); | |
11d67612 YS |
5269 | if (memcg->id.id < 0) { |
5270 | error = memcg->id.id; | |
73f576c0 | 5271 | goto fail; |
11d67612 | 5272 | } |
73f576c0 | 5273 | |
3e38e0aa RG |
5274 | memcg->vmstats_local = alloc_percpu_gfp(struct memcg_vmstats_percpu, |
5275 | GFP_KERNEL_ACCOUNT); | |
815744d7 JW |
5276 | if (!memcg->vmstats_local) |
5277 | goto fail; | |
5278 | ||
3e38e0aa RG |
5279 | memcg->vmstats_percpu = alloc_percpu_gfp(struct memcg_vmstats_percpu, |
5280 | GFP_KERNEL_ACCOUNT); | |
871789d4 | 5281 | if (!memcg->vmstats_percpu) |
0b8f73e1 | 5282 | goto fail; |
78fb7466 | 5283 | |
3ed28fa1 | 5284 | for_each_node(node) |
ef8f2327 | 5285 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 5286 | goto fail; |
f64c3f54 | 5287 | |
0b8f73e1 JW |
5288 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
5289 | goto fail; | |
28dbc4b6 | 5290 | |
f7e1cb6e | 5291 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 | 5292 | INIT_LIST_HEAD(&memcg->oom_notify); |
d142e3e6 GC |
5293 | mutex_init(&memcg->thresholds_lock); |
5294 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 5295 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
5296 | INIT_LIST_HEAD(&memcg->event_list); |
5297 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 5298 | memcg->socket_pressure = jiffies; |
84c07d11 | 5299 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 5300 | memcg->kmemcg_id = -1; |
bf4f0599 | 5301 | INIT_LIST_HEAD(&memcg->objcg_list); |
900a38f0 | 5302 | #endif |
52ebea74 TH |
5303 | #ifdef CONFIG_CGROUP_WRITEBACK |
5304 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
97b27821 TH |
5305 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) |
5306 | memcg->cgwb_frn[i].done = | |
5307 | __WB_COMPLETION_INIT(&memcg_cgwb_frn_waitq); | |
87eaceb3 YS |
5308 | #endif |
5309 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5310 | spin_lock_init(&memcg->deferred_split_queue.split_queue_lock); | |
5311 | INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue); | |
5312 | memcg->deferred_split_queue.split_queue_len = 0; | |
52ebea74 | 5313 | #endif |
73f576c0 | 5314 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
5315 | return memcg; |
5316 | fail: | |
7e97de0b | 5317 | mem_cgroup_id_remove(memcg); |
40e952f9 | 5318 | __mem_cgroup_free(memcg); |
11d67612 | 5319 | return ERR_PTR(error); |
d142e3e6 GC |
5320 | } |
5321 | ||
0b8f73e1 JW |
5322 | static struct cgroup_subsys_state * __ref |
5323 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 5324 | { |
0b8f73e1 | 5325 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
b87d8cef | 5326 | struct mem_cgroup *memcg, *old_memcg; |
0b8f73e1 | 5327 | long error = -ENOMEM; |
d142e3e6 | 5328 | |
b87d8cef | 5329 | old_memcg = set_active_memcg(parent); |
0b8f73e1 | 5330 | memcg = mem_cgroup_alloc(); |
b87d8cef | 5331 | set_active_memcg(old_memcg); |
11d67612 YS |
5332 | if (IS_ERR(memcg)) |
5333 | return ERR_CAST(memcg); | |
d142e3e6 | 5334 | |
d1663a90 | 5335 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
0b8f73e1 | 5336 | memcg->soft_limit = PAGE_COUNTER_MAX; |
4b82ab4f | 5337 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
0b8f73e1 JW |
5338 | if (parent) { |
5339 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
5340 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
5341 | } | |
5342 | if (parent && parent->use_hierarchy) { | |
5343 | memcg->use_hierarchy = true; | |
3e32cb2e | 5344 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 5345 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e | 5346 | page_counter_init(&memcg->kmem, &parent->kmem); |
0db15298 | 5347 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 5348 | } else { |
3e32cb2e | 5349 | page_counter_init(&memcg->memory, NULL); |
37e84351 | 5350 | page_counter_init(&memcg->swap, NULL); |
3e32cb2e | 5351 | page_counter_init(&memcg->kmem, NULL); |
0db15298 | 5352 | page_counter_init(&memcg->tcpmem, NULL); |
8c7f6edb TH |
5353 | /* |
5354 | * Deeper hierachy with use_hierarchy == false doesn't make | |
5355 | * much sense so let cgroup subsystem know about this | |
5356 | * unfortunate state in our controller. | |
5357 | */ | |
d142e3e6 | 5358 | if (parent != root_mem_cgroup) |
073219e9 | 5359 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 5360 | } |
d6441637 | 5361 | |
0b8f73e1 JW |
5362 | /* The following stuff does not apply to the root */ |
5363 | if (!parent) { | |
5364 | root_mem_cgroup = memcg; | |
5365 | return &memcg->css; | |
5366 | } | |
5367 | ||
b313aeee | 5368 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
5369 | if (error) |
5370 | goto fail; | |
127424c8 | 5371 | |
f7e1cb6e | 5372 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5373 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 5374 | |
0b8f73e1 JW |
5375 | return &memcg->css; |
5376 | fail: | |
7e97de0b | 5377 | mem_cgroup_id_remove(memcg); |
0b8f73e1 | 5378 | mem_cgroup_free(memcg); |
11d67612 | 5379 | return ERR_PTR(error); |
0b8f73e1 JW |
5380 | } |
5381 | ||
73f576c0 | 5382 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 5383 | { |
58fa2a55 VD |
5384 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5385 | ||
0a4465d3 KT |
5386 | /* |
5387 | * A memcg must be visible for memcg_expand_shrinker_maps() | |
5388 | * by the time the maps are allocated. So, we allocate maps | |
5389 | * here, when for_each_mem_cgroup() can't skip it. | |
5390 | */ | |
5391 | if (memcg_alloc_shrinker_maps(memcg)) { | |
5392 | mem_cgroup_id_remove(memcg); | |
5393 | return -ENOMEM; | |
5394 | } | |
5395 | ||
73f576c0 | 5396 | /* Online state pins memcg ID, memcg ID pins CSS */ |
1c2d479a | 5397 | refcount_set(&memcg->id.ref, 1); |
73f576c0 | 5398 | css_get(css); |
2f7dd7a4 | 5399 | return 0; |
8cdea7c0 BS |
5400 | } |
5401 | ||
eb95419b | 5402 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 5403 | { |
eb95419b | 5404 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5405 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
5406 | |
5407 | /* | |
5408 | * Unregister events and notify userspace. | |
5409 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
5410 | * directory to avoid race between userspace and kernelspace. | |
5411 | */ | |
fba94807 TH |
5412 | spin_lock(&memcg->event_list_lock); |
5413 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
5414 | list_del_init(&event->list); |
5415 | schedule_work(&event->remove); | |
5416 | } | |
fba94807 | 5417 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 5418 | |
bf8d5d52 | 5419 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5420 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 5421 | |
567e9ab2 | 5422 | memcg_offline_kmem(memcg); |
52ebea74 | 5423 | wb_memcg_offline(memcg); |
73f576c0 | 5424 | |
591edfb1 RG |
5425 | drain_all_stock(memcg); |
5426 | ||
73f576c0 | 5427 | mem_cgroup_id_put(memcg); |
df878fb0 KH |
5428 | } |
5429 | ||
6df38689 VD |
5430 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
5431 | { | |
5432 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5433 | ||
5434 | invalidate_reclaim_iterators(memcg); | |
5435 | } | |
5436 | ||
eb95419b | 5437 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 5438 | { |
eb95419b | 5439 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
97b27821 | 5440 | int __maybe_unused i; |
c268e994 | 5441 | |
97b27821 TH |
5442 | #ifdef CONFIG_CGROUP_WRITEBACK |
5443 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) | |
5444 | wb_wait_for_completion(&memcg->cgwb_frn[i].done); | |
5445 | #endif | |
f7e1cb6e | 5446 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5447 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 5448 | |
0db15298 | 5449 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 5450 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 5451 | |
0b8f73e1 JW |
5452 | vmpressure_cleanup(&memcg->vmpressure); |
5453 | cancel_work_sync(&memcg->high_work); | |
5454 | mem_cgroup_remove_from_trees(memcg); | |
0a4465d3 | 5455 | memcg_free_shrinker_maps(memcg); |
d886f4e4 | 5456 | memcg_free_kmem(memcg); |
0b8f73e1 | 5457 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
5458 | } |
5459 | ||
1ced953b TH |
5460 | /** |
5461 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
5462 | * @css: the target css | |
5463 | * | |
5464 | * Reset the states of the mem_cgroup associated with @css. This is | |
5465 | * invoked when the userland requests disabling on the default hierarchy | |
5466 | * but the memcg is pinned through dependency. The memcg should stop | |
5467 | * applying policies and should revert to the vanilla state as it may be | |
5468 | * made visible again. | |
5469 | * | |
5470 | * The current implementation only resets the essential configurations. | |
5471 | * This needs to be expanded to cover all the visible parts. | |
5472 | */ | |
5473 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
5474 | { | |
5475 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5476 | ||
bbec2e15 RG |
5477 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
5478 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
bbec2e15 RG |
5479 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); |
5480 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 5481 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5482 | page_counter_set_low(&memcg->memory, 0); |
d1663a90 | 5483 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
24d404dc | 5484 | memcg->soft_limit = PAGE_COUNTER_MAX; |
4b82ab4f | 5485 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
2529bb3a | 5486 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
5487 | } |
5488 | ||
02491447 | 5489 | #ifdef CONFIG_MMU |
7dc74be0 | 5490 | /* Handlers for move charge at task migration. */ |
854ffa8d | 5491 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 5492 | { |
05b84301 | 5493 | int ret; |
9476db97 | 5494 | |
d0164adc MG |
5495 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
5496 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 5497 | if (!ret) { |
854ffa8d | 5498 | mc.precharge += count; |
854ffa8d DN |
5499 | return ret; |
5500 | } | |
9476db97 | 5501 | |
3674534b | 5502 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 5503 | while (count--) { |
3674534b | 5504 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 5505 | if (ret) |
38c5d72f | 5506 | return ret; |
854ffa8d | 5507 | mc.precharge++; |
9476db97 | 5508 | cond_resched(); |
854ffa8d | 5509 | } |
9476db97 | 5510 | return 0; |
4ffef5fe DN |
5511 | } |
5512 | ||
4ffef5fe DN |
5513 | union mc_target { |
5514 | struct page *page; | |
02491447 | 5515 | swp_entry_t ent; |
4ffef5fe DN |
5516 | }; |
5517 | ||
4ffef5fe | 5518 | enum mc_target_type { |
8d32ff84 | 5519 | MC_TARGET_NONE = 0, |
4ffef5fe | 5520 | MC_TARGET_PAGE, |
02491447 | 5521 | MC_TARGET_SWAP, |
c733a828 | 5522 | MC_TARGET_DEVICE, |
4ffef5fe DN |
5523 | }; |
5524 | ||
90254a65 DN |
5525 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5526 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5527 | { |
25b2995a | 5528 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5529 | |
90254a65 DN |
5530 | if (!page || !page_mapped(page)) |
5531 | return NULL; | |
5532 | if (PageAnon(page)) { | |
1dfab5ab | 5533 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5534 | return NULL; |
1dfab5ab JW |
5535 | } else { |
5536 | if (!(mc.flags & MOVE_FILE)) | |
5537 | return NULL; | |
5538 | } | |
90254a65 DN |
5539 | if (!get_page_unless_zero(page)) |
5540 | return NULL; | |
5541 | ||
5542 | return page; | |
5543 | } | |
5544 | ||
c733a828 | 5545 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 5546 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 5547 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 5548 | { |
90254a65 DN |
5549 | struct page *page = NULL; |
5550 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5551 | ||
9a137153 | 5552 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5553 | return NULL; |
c733a828 JG |
5554 | |
5555 | /* | |
5556 | * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to | |
5557 | * a device and because they are not accessible by CPU they are store | |
5558 | * as special swap entry in the CPU page table. | |
5559 | */ | |
5560 | if (is_device_private_entry(ent)) { | |
5561 | page = device_private_entry_to_page(ent); | |
5562 | /* | |
5563 | * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have | |
5564 | * a refcount of 1 when free (unlike normal page) | |
5565 | */ | |
5566 | if (!page_ref_add_unless(page, 1, 1)) | |
5567 | return NULL; | |
5568 | return page; | |
5569 | } | |
5570 | ||
9a137153 RC |
5571 | if (non_swap_entry(ent)) |
5572 | return NULL; | |
5573 | ||
4b91355e KH |
5574 | /* |
5575 | * Because lookup_swap_cache() updates some statistics counter, | |
5576 | * we call find_get_page() with swapper_space directly. | |
5577 | */ | |
f6ab1f7f | 5578 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
2d1c4980 | 5579 | entry->val = ent.val; |
90254a65 DN |
5580 | |
5581 | return page; | |
5582 | } | |
4b91355e KH |
5583 | #else |
5584 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 5585 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
5586 | { |
5587 | return NULL; | |
5588 | } | |
5589 | #endif | |
90254a65 | 5590 | |
87946a72 DN |
5591 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5592 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5593 | { | |
87946a72 DN |
5594 | if (!vma->vm_file) /* anonymous vma */ |
5595 | return NULL; | |
1dfab5ab | 5596 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
5597 | return NULL; |
5598 | ||
87946a72 | 5599 | /* page is moved even if it's not RSS of this task(page-faulted). */ |
aa3b1895 | 5600 | /* shmem/tmpfs may report page out on swap: account for that too. */ |
f5df8635 MWO |
5601 | return find_get_incore_page(vma->vm_file->f_mapping, |
5602 | linear_page_index(vma, addr)); | |
87946a72 DN |
5603 | } |
5604 | ||
b1b0deab CG |
5605 | /** |
5606 | * mem_cgroup_move_account - move account of the page | |
5607 | * @page: the page | |
25843c2b | 5608 | * @compound: charge the page as compound or small page |
b1b0deab CG |
5609 | * @from: mem_cgroup which the page is moved from. |
5610 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
5611 | * | |
3ac808fd | 5612 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
5613 | * |
5614 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
5615 | * from old cgroup. | |
5616 | */ | |
5617 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 5618 | bool compound, |
b1b0deab CG |
5619 | struct mem_cgroup *from, |
5620 | struct mem_cgroup *to) | |
5621 | { | |
ae8af438 KK |
5622 | struct lruvec *from_vec, *to_vec; |
5623 | struct pglist_data *pgdat; | |
6c357848 | 5624 | unsigned int nr_pages = compound ? thp_nr_pages(page) : 1; |
b1b0deab CG |
5625 | int ret; |
5626 | ||
5627 | VM_BUG_ON(from == to); | |
5628 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 5629 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
5630 | |
5631 | /* | |
6a93ca8f | 5632 | * Prevent mem_cgroup_migrate() from looking at |
45637bab | 5633 | * page->mem_cgroup of its source page while we change it. |
b1b0deab | 5634 | */ |
f627c2f5 | 5635 | ret = -EBUSY; |
b1b0deab CG |
5636 | if (!trylock_page(page)) |
5637 | goto out; | |
5638 | ||
5639 | ret = -EINVAL; | |
5640 | if (page->mem_cgroup != from) | |
5641 | goto out_unlock; | |
5642 | ||
ae8af438 | 5643 | pgdat = page_pgdat(page); |
867e5e1d JW |
5644 | from_vec = mem_cgroup_lruvec(from, pgdat); |
5645 | to_vec = mem_cgroup_lruvec(to, pgdat); | |
ae8af438 | 5646 | |
abb242f5 | 5647 | lock_page_memcg(page); |
b1b0deab | 5648 | |
be5d0a74 JW |
5649 | if (PageAnon(page)) { |
5650 | if (page_mapped(page)) { | |
5651 | __mod_lruvec_state(from_vec, NR_ANON_MAPPED, -nr_pages); | |
5652 | __mod_lruvec_state(to_vec, NR_ANON_MAPPED, nr_pages); | |
468c3982 JW |
5653 | if (PageTransHuge(page)) { |
5654 | __mod_lruvec_state(from_vec, NR_ANON_THPS, | |
5655 | -nr_pages); | |
5656 | __mod_lruvec_state(to_vec, NR_ANON_THPS, | |
5657 | nr_pages); | |
5658 | } | |
5659 | ||
be5d0a74 JW |
5660 | } |
5661 | } else { | |
0d1c2072 JW |
5662 | __mod_lruvec_state(from_vec, NR_FILE_PAGES, -nr_pages); |
5663 | __mod_lruvec_state(to_vec, NR_FILE_PAGES, nr_pages); | |
5664 | ||
5665 | if (PageSwapBacked(page)) { | |
5666 | __mod_lruvec_state(from_vec, NR_SHMEM, -nr_pages); | |
5667 | __mod_lruvec_state(to_vec, NR_SHMEM, nr_pages); | |
5668 | } | |
5669 | ||
49e50d27 JW |
5670 | if (page_mapped(page)) { |
5671 | __mod_lruvec_state(from_vec, NR_FILE_MAPPED, -nr_pages); | |
5672 | __mod_lruvec_state(to_vec, NR_FILE_MAPPED, nr_pages); | |
5673 | } | |
b1b0deab | 5674 | |
49e50d27 JW |
5675 | if (PageDirty(page)) { |
5676 | struct address_space *mapping = page_mapping(page); | |
c4843a75 | 5677 | |
f56753ac | 5678 | if (mapping_can_writeback(mapping)) { |
49e50d27 JW |
5679 | __mod_lruvec_state(from_vec, NR_FILE_DIRTY, |
5680 | -nr_pages); | |
5681 | __mod_lruvec_state(to_vec, NR_FILE_DIRTY, | |
5682 | nr_pages); | |
5683 | } | |
c4843a75 GT |
5684 | } |
5685 | } | |
5686 | ||
b1b0deab | 5687 | if (PageWriteback(page)) { |
ae8af438 KK |
5688 | __mod_lruvec_state(from_vec, NR_WRITEBACK, -nr_pages); |
5689 | __mod_lruvec_state(to_vec, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
5690 | } |
5691 | ||
5692 | /* | |
abb242f5 JW |
5693 | * All state has been migrated, let's switch to the new memcg. |
5694 | * | |
b1b0deab | 5695 | * It is safe to change page->mem_cgroup here because the page |
abb242f5 JW |
5696 | * is referenced, charged, isolated, and locked: we can't race |
5697 | * with (un)charging, migration, LRU putback, or anything else | |
5698 | * that would rely on a stable page->mem_cgroup. | |
5699 | * | |
5700 | * Note that lock_page_memcg is a memcg lock, not a page lock, | |
5701 | * to save space. As soon as we switch page->mem_cgroup to a | |
5702 | * new memcg that isn't locked, the above state can change | |
5703 | * concurrently again. Make sure we're truly done with it. | |
b1b0deab | 5704 | */ |
abb242f5 | 5705 | smp_mb(); |
b1b0deab | 5706 | |
1a3e1f40 JW |
5707 | css_get(&to->css); |
5708 | css_put(&from->css); | |
5709 | ||
5710 | page->mem_cgroup = to; | |
87eaceb3 | 5711 | |
abb242f5 | 5712 | __unlock_page_memcg(from); |
b1b0deab CG |
5713 | |
5714 | ret = 0; | |
5715 | ||
5716 | local_irq_disable(); | |
3fba69a5 | 5717 | mem_cgroup_charge_statistics(to, page, nr_pages); |
b1b0deab | 5718 | memcg_check_events(to, page); |
3fba69a5 | 5719 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
b1b0deab CG |
5720 | memcg_check_events(from, page); |
5721 | local_irq_enable(); | |
5722 | out_unlock: | |
5723 | unlock_page(page); | |
5724 | out: | |
5725 | return ret; | |
5726 | } | |
5727 | ||
7cf7806c LR |
5728 | /** |
5729 | * get_mctgt_type - get target type of moving charge | |
5730 | * @vma: the vma the pte to be checked belongs | |
5731 | * @addr: the address corresponding to the pte to be checked | |
5732 | * @ptent: the pte to be checked | |
5733 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
5734 | * | |
5735 | * Returns | |
5736 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5737 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5738 | * move charge. if @target is not NULL, the page is stored in target->page | |
5739 | * with extra refcnt got(Callers should handle it). | |
5740 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
5741 | * target for charge migration. if @target is not NULL, the entry is stored | |
5742 | * in target->ent. | |
25b2995a CH |
5743 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PRIVATE |
5744 | * (so ZONE_DEVICE page and thus not on the lru). | |
df6ad698 JG |
5745 | * For now we such page is charge like a regular page would be as for all |
5746 | * intent and purposes it is just special memory taking the place of a | |
5747 | * regular page. | |
c733a828 JG |
5748 | * |
5749 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
5750 | * |
5751 | * Called with pte lock held. | |
5752 | */ | |
5753 | ||
8d32ff84 | 5754 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5755 | unsigned long addr, pte_t ptent, union mc_target *target) |
5756 | { | |
5757 | struct page *page = NULL; | |
8d32ff84 | 5758 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5759 | swp_entry_t ent = { .val = 0 }; |
5760 | ||
5761 | if (pte_present(ptent)) | |
5762 | page = mc_handle_present_pte(vma, addr, ptent); | |
5763 | else if (is_swap_pte(ptent)) | |
48406ef8 | 5764 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 5765 | else if (pte_none(ptent)) |
87946a72 | 5766 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
5767 | |
5768 | if (!page && !ent.val) | |
8d32ff84 | 5769 | return ret; |
02491447 | 5770 | if (page) { |
02491447 | 5771 | /* |
0a31bc97 | 5772 | * Do only loose check w/o serialization. |
1306a85a | 5773 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 5774 | * not under LRU exclusion. |
02491447 | 5775 | */ |
1306a85a | 5776 | if (page->mem_cgroup == mc.from) { |
02491447 | 5777 | ret = MC_TARGET_PAGE; |
25b2995a | 5778 | if (is_device_private_page(page)) |
c733a828 | 5779 | ret = MC_TARGET_DEVICE; |
02491447 DN |
5780 | if (target) |
5781 | target->page = page; | |
5782 | } | |
5783 | if (!ret || !target) | |
5784 | put_page(page); | |
5785 | } | |
3e14a57b HY |
5786 | /* |
5787 | * There is a swap entry and a page doesn't exist or isn't charged. | |
5788 | * But we cannot move a tail-page in a THP. | |
5789 | */ | |
5790 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 5791 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5792 | ret = MC_TARGET_SWAP; |
5793 | if (target) | |
5794 | target->ent = ent; | |
4ffef5fe | 5795 | } |
4ffef5fe DN |
5796 | return ret; |
5797 | } | |
5798 | ||
12724850 NH |
5799 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5800 | /* | |
d6810d73 HY |
5801 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
5802 | * not support them for now. | |
12724850 NH |
5803 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
5804 | */ | |
5805 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5806 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5807 | { | |
5808 | struct page *page = NULL; | |
12724850 NH |
5809 | enum mc_target_type ret = MC_TARGET_NONE; |
5810 | ||
84c3fc4e ZY |
5811 | if (unlikely(is_swap_pmd(pmd))) { |
5812 | VM_BUG_ON(thp_migration_supported() && | |
5813 | !is_pmd_migration_entry(pmd)); | |
5814 | return ret; | |
5815 | } | |
12724850 | 5816 | page = pmd_page(pmd); |
309381fe | 5817 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 5818 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 5819 | return ret; |
1306a85a | 5820 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
5821 | ret = MC_TARGET_PAGE; |
5822 | if (target) { | |
5823 | get_page(page); | |
5824 | target->page = page; | |
5825 | } | |
5826 | } | |
5827 | return ret; | |
5828 | } | |
5829 | #else | |
5830 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5831 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5832 | { | |
5833 | return MC_TARGET_NONE; | |
5834 | } | |
5835 | #endif | |
5836 | ||
4ffef5fe DN |
5837 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5838 | unsigned long addr, unsigned long end, | |
5839 | struct mm_walk *walk) | |
5840 | { | |
26bcd64a | 5841 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5842 | pte_t *pte; |
5843 | spinlock_t *ptl; | |
5844 | ||
b6ec57f4 KS |
5845 | ptl = pmd_trans_huge_lock(pmd, vma); |
5846 | if (ptl) { | |
c733a828 JG |
5847 | /* |
5848 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
25b2995a CH |
5849 | * support transparent huge page with MEMORY_DEVICE_PRIVATE but |
5850 | * this might change. | |
c733a828 | 5851 | */ |
12724850 NH |
5852 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5853 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5854 | spin_unlock(ptl); |
1a5a9906 | 5855 | return 0; |
12724850 | 5856 | } |
03319327 | 5857 | |
45f83cef AA |
5858 | if (pmd_trans_unstable(pmd)) |
5859 | return 0; | |
4ffef5fe DN |
5860 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5861 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5862 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5863 | mc.precharge++; /* increment precharge temporarily */ |
5864 | pte_unmap_unlock(pte - 1, ptl); | |
5865 | cond_resched(); | |
5866 | ||
7dc74be0 DN |
5867 | return 0; |
5868 | } | |
5869 | ||
7b86ac33 CH |
5870 | static const struct mm_walk_ops precharge_walk_ops = { |
5871 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5872 | }; | |
5873 | ||
4ffef5fe DN |
5874 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5875 | { | |
5876 | unsigned long precharge; | |
4ffef5fe | 5877 | |
d8ed45c5 | 5878 | mmap_read_lock(mm); |
7b86ac33 | 5879 | walk_page_range(mm, 0, mm->highest_vm_end, &precharge_walk_ops, NULL); |
d8ed45c5 | 5880 | mmap_read_unlock(mm); |
4ffef5fe DN |
5881 | |
5882 | precharge = mc.precharge; | |
5883 | mc.precharge = 0; | |
5884 | ||
5885 | return precharge; | |
5886 | } | |
5887 | ||
4ffef5fe DN |
5888 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5889 | { | |
dfe076b0 DN |
5890 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5891 | ||
5892 | VM_BUG_ON(mc.moving_task); | |
5893 | mc.moving_task = current; | |
5894 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5895 | } |
5896 | ||
dfe076b0 DN |
5897 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5898 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5899 | { |
2bd9bb20 KH |
5900 | struct mem_cgroup *from = mc.from; |
5901 | struct mem_cgroup *to = mc.to; | |
5902 | ||
4ffef5fe | 5903 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5904 | if (mc.precharge) { |
00501b53 | 5905 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5906 | mc.precharge = 0; |
5907 | } | |
5908 | /* | |
5909 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5910 | * we must uncharge here. | |
5911 | */ | |
5912 | if (mc.moved_charge) { | |
00501b53 | 5913 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5914 | mc.moved_charge = 0; |
4ffef5fe | 5915 | } |
483c30b5 DN |
5916 | /* we must fixup refcnts and charges */ |
5917 | if (mc.moved_swap) { | |
483c30b5 | 5918 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5919 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5920 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5921 | |
615d66c3 VD |
5922 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
5923 | ||
05b84301 | 5924 | /* |
3e32cb2e JW |
5925 | * we charged both to->memory and to->memsw, so we |
5926 | * should uncharge to->memory. | |
05b84301 | 5927 | */ |
ce00a967 | 5928 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5929 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5930 | ||
483c30b5 DN |
5931 | mc.moved_swap = 0; |
5932 | } | |
dfe076b0 DN |
5933 | memcg_oom_recover(from); |
5934 | memcg_oom_recover(to); | |
5935 | wake_up_all(&mc.waitq); | |
5936 | } | |
5937 | ||
5938 | static void mem_cgroup_clear_mc(void) | |
5939 | { | |
264a0ae1 TH |
5940 | struct mm_struct *mm = mc.mm; |
5941 | ||
dfe076b0 DN |
5942 | /* |
5943 | * we must clear moving_task before waking up waiters at the end of | |
5944 | * task migration. | |
5945 | */ | |
5946 | mc.moving_task = NULL; | |
5947 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5948 | spin_lock(&mc.lock); |
4ffef5fe DN |
5949 | mc.from = NULL; |
5950 | mc.to = NULL; | |
264a0ae1 | 5951 | mc.mm = NULL; |
2bd9bb20 | 5952 | spin_unlock(&mc.lock); |
264a0ae1 TH |
5953 | |
5954 | mmput(mm); | |
4ffef5fe DN |
5955 | } |
5956 | ||
1f7dd3e5 | 5957 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5958 | { |
1f7dd3e5 | 5959 | struct cgroup_subsys_state *css; |
eed67d75 | 5960 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 5961 | struct mem_cgroup *from; |
4530eddb | 5962 | struct task_struct *leader, *p; |
9f2115f9 | 5963 | struct mm_struct *mm; |
1dfab5ab | 5964 | unsigned long move_flags; |
9f2115f9 | 5965 | int ret = 0; |
7dc74be0 | 5966 | |
1f7dd3e5 TH |
5967 | /* charge immigration isn't supported on the default hierarchy */ |
5968 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
5969 | return 0; |
5970 | ||
4530eddb TH |
5971 | /* |
5972 | * Multi-process migrations only happen on the default hierarchy | |
5973 | * where charge immigration is not used. Perform charge | |
5974 | * immigration if @tset contains a leader and whine if there are | |
5975 | * multiple. | |
5976 | */ | |
5977 | p = NULL; | |
1f7dd3e5 | 5978 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
5979 | WARN_ON_ONCE(p); |
5980 | p = leader; | |
1f7dd3e5 | 5981 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
5982 | } |
5983 | if (!p) | |
5984 | return 0; | |
5985 | ||
1f7dd3e5 TH |
5986 | /* |
5987 | * We are now commited to this value whatever it is. Changes in this | |
5988 | * tunable will only affect upcoming migrations, not the current one. | |
5989 | * So we need to save it, and keep it going. | |
5990 | */ | |
5991 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
5992 | if (!move_flags) | |
5993 | return 0; | |
5994 | ||
9f2115f9 TH |
5995 | from = mem_cgroup_from_task(p); |
5996 | ||
5997 | VM_BUG_ON(from == memcg); | |
5998 | ||
5999 | mm = get_task_mm(p); | |
6000 | if (!mm) | |
6001 | return 0; | |
6002 | /* We move charges only when we move a owner of the mm */ | |
6003 | if (mm->owner == p) { | |
6004 | VM_BUG_ON(mc.from); | |
6005 | VM_BUG_ON(mc.to); | |
6006 | VM_BUG_ON(mc.precharge); | |
6007 | VM_BUG_ON(mc.moved_charge); | |
6008 | VM_BUG_ON(mc.moved_swap); | |
6009 | ||
6010 | spin_lock(&mc.lock); | |
264a0ae1 | 6011 | mc.mm = mm; |
9f2115f9 TH |
6012 | mc.from = from; |
6013 | mc.to = memcg; | |
6014 | mc.flags = move_flags; | |
6015 | spin_unlock(&mc.lock); | |
6016 | /* We set mc.moving_task later */ | |
6017 | ||
6018 | ret = mem_cgroup_precharge_mc(mm); | |
6019 | if (ret) | |
6020 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
6021 | } else { |
6022 | mmput(mm); | |
7dc74be0 DN |
6023 | } |
6024 | return ret; | |
6025 | } | |
6026 | ||
1f7dd3e5 | 6027 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 6028 | { |
4e2f245d JW |
6029 | if (mc.to) |
6030 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
6031 | } |
6032 | ||
4ffef5fe DN |
6033 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
6034 | unsigned long addr, unsigned long end, | |
6035 | struct mm_walk *walk) | |
7dc74be0 | 6036 | { |
4ffef5fe | 6037 | int ret = 0; |
26bcd64a | 6038 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
6039 | pte_t *pte; |
6040 | spinlock_t *ptl; | |
12724850 NH |
6041 | enum mc_target_type target_type; |
6042 | union mc_target target; | |
6043 | struct page *page; | |
4ffef5fe | 6044 | |
b6ec57f4 KS |
6045 | ptl = pmd_trans_huge_lock(pmd, vma); |
6046 | if (ptl) { | |
62ade86a | 6047 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 6048 | spin_unlock(ptl); |
12724850 NH |
6049 | return 0; |
6050 | } | |
6051 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
6052 | if (target_type == MC_TARGET_PAGE) { | |
6053 | page = target.page; | |
6054 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 6055 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 6056 | mc.from, mc.to)) { |
12724850 NH |
6057 | mc.precharge -= HPAGE_PMD_NR; |
6058 | mc.moved_charge += HPAGE_PMD_NR; | |
6059 | } | |
6060 | putback_lru_page(page); | |
6061 | } | |
6062 | put_page(page); | |
c733a828 JG |
6063 | } else if (target_type == MC_TARGET_DEVICE) { |
6064 | page = target.page; | |
6065 | if (!mem_cgroup_move_account(page, true, | |
6066 | mc.from, mc.to)) { | |
6067 | mc.precharge -= HPAGE_PMD_NR; | |
6068 | mc.moved_charge += HPAGE_PMD_NR; | |
6069 | } | |
6070 | put_page(page); | |
12724850 | 6071 | } |
bf929152 | 6072 | spin_unlock(ptl); |
1a5a9906 | 6073 | return 0; |
12724850 NH |
6074 | } |
6075 | ||
45f83cef AA |
6076 | if (pmd_trans_unstable(pmd)) |
6077 | return 0; | |
4ffef5fe DN |
6078 | retry: |
6079 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
6080 | for (; addr != end; addr += PAGE_SIZE) { | |
6081 | pte_t ptent = *(pte++); | |
c733a828 | 6082 | bool device = false; |
02491447 | 6083 | swp_entry_t ent; |
4ffef5fe DN |
6084 | |
6085 | if (!mc.precharge) | |
6086 | break; | |
6087 | ||
8d32ff84 | 6088 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
6089 | case MC_TARGET_DEVICE: |
6090 | device = true; | |
e4a9bc58 | 6091 | fallthrough; |
4ffef5fe DN |
6092 | case MC_TARGET_PAGE: |
6093 | page = target.page; | |
53f9263b KS |
6094 | /* |
6095 | * We can have a part of the split pmd here. Moving it | |
6096 | * can be done but it would be too convoluted so simply | |
6097 | * ignore such a partial THP and keep it in original | |
6098 | * memcg. There should be somebody mapping the head. | |
6099 | */ | |
6100 | if (PageTransCompound(page)) | |
6101 | goto put; | |
c733a828 | 6102 | if (!device && isolate_lru_page(page)) |
4ffef5fe | 6103 | goto put; |
f627c2f5 KS |
6104 | if (!mem_cgroup_move_account(page, false, |
6105 | mc.from, mc.to)) { | |
4ffef5fe | 6106 | mc.precharge--; |
854ffa8d DN |
6107 | /* we uncharge from mc.from later. */ |
6108 | mc.moved_charge++; | |
4ffef5fe | 6109 | } |
c733a828 JG |
6110 | if (!device) |
6111 | putback_lru_page(page); | |
8d32ff84 | 6112 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
6113 | put_page(page); |
6114 | break; | |
02491447 DN |
6115 | case MC_TARGET_SWAP: |
6116 | ent = target.ent; | |
e91cbb42 | 6117 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6118 | mc.precharge--; |
8d22a935 HD |
6119 | mem_cgroup_id_get_many(mc.to, 1); |
6120 | /* we fixup other refcnts and charges later. */ | |
483c30b5 DN |
6121 | mc.moved_swap++; |
6122 | } | |
02491447 | 6123 | break; |
4ffef5fe DN |
6124 | default: |
6125 | break; | |
6126 | } | |
6127 | } | |
6128 | pte_unmap_unlock(pte - 1, ptl); | |
6129 | cond_resched(); | |
6130 | ||
6131 | if (addr != end) { | |
6132 | /* | |
6133 | * We have consumed all precharges we got in can_attach(). | |
6134 | * We try charge one by one, but don't do any additional | |
6135 | * charges to mc.to if we have failed in charge once in attach() | |
6136 | * phase. | |
6137 | */ | |
854ffa8d | 6138 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6139 | if (!ret) |
6140 | goto retry; | |
6141 | } | |
6142 | ||
6143 | return ret; | |
6144 | } | |
6145 | ||
7b86ac33 CH |
6146 | static const struct mm_walk_ops charge_walk_ops = { |
6147 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6148 | }; | |
6149 | ||
264a0ae1 | 6150 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 6151 | { |
4ffef5fe | 6152 | lru_add_drain_all(); |
312722cb | 6153 | /* |
81f8c3a4 JW |
6154 | * Signal lock_page_memcg() to take the memcg's move_lock |
6155 | * while we're moving its pages to another memcg. Then wait | |
6156 | * for already started RCU-only updates to finish. | |
312722cb JW |
6157 | */ |
6158 | atomic_inc(&mc.from->moving_account); | |
6159 | synchronize_rcu(); | |
dfe076b0 | 6160 | retry: |
d8ed45c5 | 6161 | if (unlikely(!mmap_read_trylock(mc.mm))) { |
dfe076b0 | 6162 | /* |
c1e8d7c6 | 6163 | * Someone who are holding the mmap_lock might be waiting in |
dfe076b0 DN |
6164 | * waitq. So we cancel all extra charges, wake up all waiters, |
6165 | * and retry. Because we cancel precharges, we might not be able | |
6166 | * to move enough charges, but moving charge is a best-effort | |
6167 | * feature anyway, so it wouldn't be a big problem. | |
6168 | */ | |
6169 | __mem_cgroup_clear_mc(); | |
6170 | cond_resched(); | |
6171 | goto retry; | |
6172 | } | |
26bcd64a NH |
6173 | /* |
6174 | * When we have consumed all precharges and failed in doing | |
6175 | * additional charge, the page walk just aborts. | |
6176 | */ | |
7b86ac33 CH |
6177 | walk_page_range(mc.mm, 0, mc.mm->highest_vm_end, &charge_walk_ops, |
6178 | NULL); | |
0247f3f4 | 6179 | |
d8ed45c5 | 6180 | mmap_read_unlock(mc.mm); |
312722cb | 6181 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
6182 | } |
6183 | ||
264a0ae1 | 6184 | static void mem_cgroup_move_task(void) |
67e465a7 | 6185 | { |
264a0ae1 TH |
6186 | if (mc.to) { |
6187 | mem_cgroup_move_charge(); | |
a433658c | 6188 | mem_cgroup_clear_mc(); |
264a0ae1 | 6189 | } |
67e465a7 | 6190 | } |
5cfb80a7 | 6191 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 6192 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6193 | { |
6194 | return 0; | |
6195 | } | |
1f7dd3e5 | 6196 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6197 | { |
6198 | } | |
264a0ae1 | 6199 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
6200 | { |
6201 | } | |
6202 | #endif | |
67e465a7 | 6203 | |
f00baae7 TH |
6204 | /* |
6205 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
6206 | * to verify whether we're attached to the default hierarchy on each mount |
6207 | * attempt. | |
f00baae7 | 6208 | */ |
eb95419b | 6209 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
6210 | { |
6211 | /* | |
aa6ec29b | 6212 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
6213 | * guarantees that @root doesn't have any children, so turning it |
6214 | * on for the root memcg is enough. | |
6215 | */ | |
9e10a130 | 6216 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
6217 | root_mem_cgroup->use_hierarchy = true; |
6218 | else | |
6219 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
6220 | } |
6221 | ||
677dc973 CD |
6222 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
6223 | { | |
6224 | if (value == PAGE_COUNTER_MAX) | |
6225 | seq_puts(m, "max\n"); | |
6226 | else | |
6227 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); | |
6228 | ||
6229 | return 0; | |
6230 | } | |
6231 | ||
241994ed JW |
6232 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
6233 | struct cftype *cft) | |
6234 | { | |
f5fc3c5d JW |
6235 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
6236 | ||
6237 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
6238 | } |
6239 | ||
bf8d5d52 RG |
6240 | static int memory_min_show(struct seq_file *m, void *v) |
6241 | { | |
677dc973 CD |
6242 | return seq_puts_memcg_tunable(m, |
6243 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); | |
bf8d5d52 RG |
6244 | } |
6245 | ||
6246 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
6247 | char *buf, size_t nbytes, loff_t off) | |
6248 | { | |
6249 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6250 | unsigned long min; | |
6251 | int err; | |
6252 | ||
6253 | buf = strstrip(buf); | |
6254 | err = page_counter_memparse(buf, "max", &min); | |
6255 | if (err) | |
6256 | return err; | |
6257 | ||
6258 | page_counter_set_min(&memcg->memory, min); | |
6259 | ||
6260 | return nbytes; | |
6261 | } | |
6262 | ||
241994ed JW |
6263 | static int memory_low_show(struct seq_file *m, void *v) |
6264 | { | |
677dc973 CD |
6265 | return seq_puts_memcg_tunable(m, |
6266 | READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); | |
241994ed JW |
6267 | } |
6268 | ||
6269 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
6270 | char *buf, size_t nbytes, loff_t off) | |
6271 | { | |
6272 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6273 | unsigned long low; | |
6274 | int err; | |
6275 | ||
6276 | buf = strstrip(buf); | |
d2973697 | 6277 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
6278 | if (err) |
6279 | return err; | |
6280 | ||
23067153 | 6281 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
6282 | |
6283 | return nbytes; | |
6284 | } | |
6285 | ||
6286 | static int memory_high_show(struct seq_file *m, void *v) | |
6287 | { | |
d1663a90 JK |
6288 | return seq_puts_memcg_tunable(m, |
6289 | READ_ONCE(mem_cgroup_from_seq(m)->memory.high)); | |
241994ed JW |
6290 | } |
6291 | ||
6292 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
6293 | char *buf, size_t nbytes, loff_t off) | |
6294 | { | |
6295 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6296 | unsigned int nr_retries = MAX_RECLAIM_RETRIES; |
8c8c383c | 6297 | bool drained = false; |
241994ed JW |
6298 | unsigned long high; |
6299 | int err; | |
6300 | ||
6301 | buf = strstrip(buf); | |
d2973697 | 6302 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
6303 | if (err) |
6304 | return err; | |
6305 | ||
8c8c383c JW |
6306 | for (;;) { |
6307 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6308 | unsigned long reclaimed; | |
6309 | ||
6310 | if (nr_pages <= high) | |
6311 | break; | |
6312 | ||
6313 | if (signal_pending(current)) | |
6314 | break; | |
6315 | ||
6316 | if (!drained) { | |
6317 | drain_all_stock(memcg); | |
6318 | drained = true; | |
6319 | continue; | |
6320 | } | |
6321 | ||
6322 | reclaimed = try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
6323 | GFP_KERNEL, true); | |
6324 | ||
6325 | if (!reclaimed && !nr_retries--) | |
6326 | break; | |
6327 | } | |
588083bb | 6328 | |
536d3bf2 RG |
6329 | page_counter_set_high(&memcg->memory, high); |
6330 | ||
19ce33ac JW |
6331 | memcg_wb_domain_size_changed(memcg); |
6332 | ||
241994ed JW |
6333 | return nbytes; |
6334 | } | |
6335 | ||
6336 | static int memory_max_show(struct seq_file *m, void *v) | |
6337 | { | |
677dc973 CD |
6338 | return seq_puts_memcg_tunable(m, |
6339 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); | |
241994ed JW |
6340 | } |
6341 | ||
6342 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
6343 | char *buf, size_t nbytes, loff_t off) | |
6344 | { | |
6345 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6346 | unsigned int nr_reclaims = MAX_RECLAIM_RETRIES; |
b6e6edcf | 6347 | bool drained = false; |
241994ed JW |
6348 | unsigned long max; |
6349 | int err; | |
6350 | ||
6351 | buf = strstrip(buf); | |
d2973697 | 6352 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
6353 | if (err) |
6354 | return err; | |
6355 | ||
bbec2e15 | 6356 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
6357 | |
6358 | for (;;) { | |
6359 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6360 | ||
6361 | if (nr_pages <= max) | |
6362 | break; | |
6363 | ||
7249c9f0 | 6364 | if (signal_pending(current)) |
b6e6edcf | 6365 | break; |
b6e6edcf JW |
6366 | |
6367 | if (!drained) { | |
6368 | drain_all_stock(memcg); | |
6369 | drained = true; | |
6370 | continue; | |
6371 | } | |
6372 | ||
6373 | if (nr_reclaims) { | |
6374 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
6375 | GFP_KERNEL, true)) | |
6376 | nr_reclaims--; | |
6377 | continue; | |
6378 | } | |
6379 | ||
e27be240 | 6380 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
6381 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
6382 | break; | |
6383 | } | |
241994ed | 6384 | |
2529bb3a | 6385 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
6386 | return nbytes; |
6387 | } | |
6388 | ||
1e577f97 SB |
6389 | static void __memory_events_show(struct seq_file *m, atomic_long_t *events) |
6390 | { | |
6391 | seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW])); | |
6392 | seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH])); | |
6393 | seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX])); | |
6394 | seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM])); | |
6395 | seq_printf(m, "oom_kill %lu\n", | |
6396 | atomic_long_read(&events[MEMCG_OOM_KILL])); | |
6397 | } | |
6398 | ||
241994ed JW |
6399 | static int memory_events_show(struct seq_file *m, void *v) |
6400 | { | |
aa9694bb | 6401 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
241994ed | 6402 | |
1e577f97 SB |
6403 | __memory_events_show(m, memcg->memory_events); |
6404 | return 0; | |
6405 | } | |
6406 | ||
6407 | static int memory_events_local_show(struct seq_file *m, void *v) | |
6408 | { | |
6409 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
241994ed | 6410 | |
1e577f97 | 6411 | __memory_events_show(m, memcg->memory_events_local); |
241994ed JW |
6412 | return 0; |
6413 | } | |
6414 | ||
587d9f72 JW |
6415 | static int memory_stat_show(struct seq_file *m, void *v) |
6416 | { | |
aa9694bb | 6417 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
c8713d0b | 6418 | char *buf; |
1ff9e6e1 | 6419 | |
c8713d0b JW |
6420 | buf = memory_stat_format(memcg); |
6421 | if (!buf) | |
6422 | return -ENOMEM; | |
6423 | seq_puts(m, buf); | |
6424 | kfree(buf); | |
587d9f72 JW |
6425 | return 0; |
6426 | } | |
6427 | ||
5f9a4f4a MS |
6428 | #ifdef CONFIG_NUMA |
6429 | static int memory_numa_stat_show(struct seq_file *m, void *v) | |
6430 | { | |
6431 | int i; | |
6432 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
6433 | ||
6434 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { | |
6435 | int nid; | |
6436 | ||
6437 | if (memory_stats[i].idx >= NR_VM_NODE_STAT_ITEMS) | |
6438 | continue; | |
6439 | ||
6440 | seq_printf(m, "%s", memory_stats[i].name); | |
6441 | for_each_node_state(nid, N_MEMORY) { | |
6442 | u64 size; | |
6443 | struct lruvec *lruvec; | |
6444 | ||
6445 | lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); | |
6446 | size = lruvec_page_state(lruvec, memory_stats[i].idx); | |
6447 | size *= memory_stats[i].ratio; | |
6448 | seq_printf(m, " N%d=%llu", nid, size); | |
6449 | } | |
6450 | seq_putc(m, '\n'); | |
6451 | } | |
6452 | ||
6453 | return 0; | |
6454 | } | |
6455 | #endif | |
6456 | ||
3d8b38eb RG |
6457 | static int memory_oom_group_show(struct seq_file *m, void *v) |
6458 | { | |
aa9694bb | 6459 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3d8b38eb RG |
6460 | |
6461 | seq_printf(m, "%d\n", memcg->oom_group); | |
6462 | ||
6463 | return 0; | |
6464 | } | |
6465 | ||
6466 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
6467 | char *buf, size_t nbytes, loff_t off) | |
6468 | { | |
6469 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6470 | int ret, oom_group; | |
6471 | ||
6472 | buf = strstrip(buf); | |
6473 | if (!buf) | |
6474 | return -EINVAL; | |
6475 | ||
6476 | ret = kstrtoint(buf, 0, &oom_group); | |
6477 | if (ret) | |
6478 | return ret; | |
6479 | ||
6480 | if (oom_group != 0 && oom_group != 1) | |
6481 | return -EINVAL; | |
6482 | ||
6483 | memcg->oom_group = oom_group; | |
6484 | ||
6485 | return nbytes; | |
6486 | } | |
6487 | ||
241994ed JW |
6488 | static struct cftype memory_files[] = { |
6489 | { | |
6490 | .name = "current", | |
f5fc3c5d | 6491 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
6492 | .read_u64 = memory_current_read, |
6493 | }, | |
bf8d5d52 RG |
6494 | { |
6495 | .name = "min", | |
6496 | .flags = CFTYPE_NOT_ON_ROOT, | |
6497 | .seq_show = memory_min_show, | |
6498 | .write = memory_min_write, | |
6499 | }, | |
241994ed JW |
6500 | { |
6501 | .name = "low", | |
6502 | .flags = CFTYPE_NOT_ON_ROOT, | |
6503 | .seq_show = memory_low_show, | |
6504 | .write = memory_low_write, | |
6505 | }, | |
6506 | { | |
6507 | .name = "high", | |
6508 | .flags = CFTYPE_NOT_ON_ROOT, | |
6509 | .seq_show = memory_high_show, | |
6510 | .write = memory_high_write, | |
6511 | }, | |
6512 | { | |
6513 | .name = "max", | |
6514 | .flags = CFTYPE_NOT_ON_ROOT, | |
6515 | .seq_show = memory_max_show, | |
6516 | .write = memory_max_write, | |
6517 | }, | |
6518 | { | |
6519 | .name = "events", | |
6520 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 6521 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
6522 | .seq_show = memory_events_show, |
6523 | }, | |
1e577f97 SB |
6524 | { |
6525 | .name = "events.local", | |
6526 | .flags = CFTYPE_NOT_ON_ROOT, | |
6527 | .file_offset = offsetof(struct mem_cgroup, events_local_file), | |
6528 | .seq_show = memory_events_local_show, | |
6529 | }, | |
587d9f72 JW |
6530 | { |
6531 | .name = "stat", | |
587d9f72 JW |
6532 | .seq_show = memory_stat_show, |
6533 | }, | |
5f9a4f4a MS |
6534 | #ifdef CONFIG_NUMA |
6535 | { | |
6536 | .name = "numa_stat", | |
6537 | .seq_show = memory_numa_stat_show, | |
6538 | }, | |
6539 | #endif | |
3d8b38eb RG |
6540 | { |
6541 | .name = "oom.group", | |
6542 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
6543 | .seq_show = memory_oom_group_show, | |
6544 | .write = memory_oom_group_write, | |
6545 | }, | |
241994ed JW |
6546 | { } /* terminate */ |
6547 | }; | |
6548 | ||
073219e9 | 6549 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 6550 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6551 | .css_online = mem_cgroup_css_online, |
92fb9748 | 6552 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 6553 | .css_released = mem_cgroup_css_released, |
92fb9748 | 6554 | .css_free = mem_cgroup_css_free, |
1ced953b | 6555 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
6556 | .can_attach = mem_cgroup_can_attach, |
6557 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 6558 | .post_attach = mem_cgroup_move_task, |
f00baae7 | 6559 | .bind = mem_cgroup_bind, |
241994ed JW |
6560 | .dfl_cftypes = memory_files, |
6561 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 6562 | .early_init = 0, |
8cdea7c0 | 6563 | }; |
c077719b | 6564 | |
bc50bcc6 JW |
6565 | /* |
6566 | * This function calculates an individual cgroup's effective | |
6567 | * protection which is derived from its own memory.min/low, its | |
6568 | * parent's and siblings' settings, as well as the actual memory | |
6569 | * distribution in the tree. | |
6570 | * | |
6571 | * The following rules apply to the effective protection values: | |
6572 | * | |
6573 | * 1. At the first level of reclaim, effective protection is equal to | |
6574 | * the declared protection in memory.min and memory.low. | |
6575 | * | |
6576 | * 2. To enable safe delegation of the protection configuration, at | |
6577 | * subsequent levels the effective protection is capped to the | |
6578 | * parent's effective protection. | |
6579 | * | |
6580 | * 3. To make complex and dynamic subtrees easier to configure, the | |
6581 | * user is allowed to overcommit the declared protection at a given | |
6582 | * level. If that is the case, the parent's effective protection is | |
6583 | * distributed to the children in proportion to how much protection | |
6584 | * they have declared and how much of it they are utilizing. | |
6585 | * | |
6586 | * This makes distribution proportional, but also work-conserving: | |
6587 | * if one cgroup claims much more protection than it uses memory, | |
6588 | * the unused remainder is available to its siblings. | |
6589 | * | |
6590 | * 4. Conversely, when the declared protection is undercommitted at a | |
6591 | * given level, the distribution of the larger parental protection | |
6592 | * budget is NOT proportional. A cgroup's protection from a sibling | |
6593 | * is capped to its own memory.min/low setting. | |
6594 | * | |
8a931f80 JW |
6595 | * 5. However, to allow protecting recursive subtrees from each other |
6596 | * without having to declare each individual cgroup's fixed share | |
6597 | * of the ancestor's claim to protection, any unutilized - | |
6598 | * "floating" - protection from up the tree is distributed in | |
6599 | * proportion to each cgroup's *usage*. This makes the protection | |
6600 | * neutral wrt sibling cgroups and lets them compete freely over | |
6601 | * the shared parental protection budget, but it protects the | |
6602 | * subtree as a whole from neighboring subtrees. | |
6603 | * | |
6604 | * Note that 4. and 5. are not in conflict: 4. is about protecting | |
6605 | * against immediate siblings whereas 5. is about protecting against | |
6606 | * neighboring subtrees. | |
bc50bcc6 JW |
6607 | */ |
6608 | static unsigned long effective_protection(unsigned long usage, | |
8a931f80 | 6609 | unsigned long parent_usage, |
bc50bcc6 JW |
6610 | unsigned long setting, |
6611 | unsigned long parent_effective, | |
6612 | unsigned long siblings_protected) | |
6613 | { | |
6614 | unsigned long protected; | |
8a931f80 | 6615 | unsigned long ep; |
bc50bcc6 JW |
6616 | |
6617 | protected = min(usage, setting); | |
6618 | /* | |
6619 | * If all cgroups at this level combined claim and use more | |
6620 | * protection then what the parent affords them, distribute | |
6621 | * shares in proportion to utilization. | |
6622 | * | |
6623 | * We are using actual utilization rather than the statically | |
6624 | * claimed protection in order to be work-conserving: claimed | |
6625 | * but unused protection is available to siblings that would | |
6626 | * otherwise get a smaller chunk than what they claimed. | |
6627 | */ | |
6628 | if (siblings_protected > parent_effective) | |
6629 | return protected * parent_effective / siblings_protected; | |
6630 | ||
6631 | /* | |
6632 | * Ok, utilized protection of all children is within what the | |
6633 | * parent affords them, so we know whatever this child claims | |
6634 | * and utilizes is effectively protected. | |
6635 | * | |
6636 | * If there is unprotected usage beyond this value, reclaim | |
6637 | * will apply pressure in proportion to that amount. | |
6638 | * | |
6639 | * If there is unutilized protection, the cgroup will be fully | |
6640 | * shielded from reclaim, but we do return a smaller value for | |
6641 | * protection than what the group could enjoy in theory. This | |
6642 | * is okay. With the overcommit distribution above, effective | |
6643 | * protection is always dependent on how memory is actually | |
6644 | * consumed among the siblings anyway. | |
6645 | */ | |
8a931f80 JW |
6646 | ep = protected; |
6647 | ||
6648 | /* | |
6649 | * If the children aren't claiming (all of) the protection | |
6650 | * afforded to them by the parent, distribute the remainder in | |
6651 | * proportion to the (unprotected) memory of each cgroup. That | |
6652 | * way, cgroups that aren't explicitly prioritized wrt each | |
6653 | * other compete freely over the allowance, but they are | |
6654 | * collectively protected from neighboring trees. | |
6655 | * | |
6656 | * We're using unprotected memory for the weight so that if | |
6657 | * some cgroups DO claim explicit protection, we don't protect | |
6658 | * the same bytes twice. | |
cd324edc JW |
6659 | * |
6660 | * Check both usage and parent_usage against the respective | |
6661 | * protected values. One should imply the other, but they | |
6662 | * aren't read atomically - make sure the division is sane. | |
8a931f80 JW |
6663 | */ |
6664 | if (!(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)) | |
6665 | return ep; | |
cd324edc JW |
6666 | if (parent_effective > siblings_protected && |
6667 | parent_usage > siblings_protected && | |
6668 | usage > protected) { | |
8a931f80 JW |
6669 | unsigned long unclaimed; |
6670 | ||
6671 | unclaimed = parent_effective - siblings_protected; | |
6672 | unclaimed *= usage - protected; | |
6673 | unclaimed /= parent_usage - siblings_protected; | |
6674 | ||
6675 | ep += unclaimed; | |
6676 | } | |
6677 | ||
6678 | return ep; | |
bc50bcc6 JW |
6679 | } |
6680 | ||
241994ed | 6681 | /** |
bf8d5d52 | 6682 | * mem_cgroup_protected - check if memory consumption is in the normal range |
34c81057 | 6683 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
6684 | * @memcg: the memory cgroup to check |
6685 | * | |
23067153 RG |
6686 | * WARNING: This function is not stateless! It can only be used as part |
6687 | * of a top-down tree iteration, not for isolated queries. | |
241994ed | 6688 | */ |
45c7f7e1 CD |
6689 | void mem_cgroup_calculate_protection(struct mem_cgroup *root, |
6690 | struct mem_cgroup *memcg) | |
241994ed | 6691 | { |
8a931f80 | 6692 | unsigned long usage, parent_usage; |
23067153 RG |
6693 | struct mem_cgroup *parent; |
6694 | ||
241994ed | 6695 | if (mem_cgroup_disabled()) |
45c7f7e1 | 6696 | return; |
241994ed | 6697 | |
34c81057 SC |
6698 | if (!root) |
6699 | root = root_mem_cgroup; | |
22f7496f YS |
6700 | |
6701 | /* | |
6702 | * Effective values of the reclaim targets are ignored so they | |
6703 | * can be stale. Have a look at mem_cgroup_protection for more | |
6704 | * details. | |
6705 | * TODO: calculation should be more robust so that we do not need | |
6706 | * that special casing. | |
6707 | */ | |
34c81057 | 6708 | if (memcg == root) |
45c7f7e1 | 6709 | return; |
241994ed | 6710 | |
23067153 | 6711 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 | 6712 | if (!usage) |
45c7f7e1 | 6713 | return; |
bf8d5d52 | 6714 | |
bf8d5d52 | 6715 | parent = parent_mem_cgroup(memcg); |
df2a4196 RG |
6716 | /* No parent means a non-hierarchical mode on v1 memcg */ |
6717 | if (!parent) | |
45c7f7e1 | 6718 | return; |
df2a4196 | 6719 | |
bc50bcc6 | 6720 | if (parent == root) { |
c3d53200 | 6721 | memcg->memory.emin = READ_ONCE(memcg->memory.min); |
03960e33 | 6722 | memcg->memory.elow = READ_ONCE(memcg->memory.low); |
45c7f7e1 | 6723 | return; |
bf8d5d52 RG |
6724 | } |
6725 | ||
8a931f80 JW |
6726 | parent_usage = page_counter_read(&parent->memory); |
6727 | ||
b3a7822e | 6728 | WRITE_ONCE(memcg->memory.emin, effective_protection(usage, parent_usage, |
c3d53200 CD |
6729 | READ_ONCE(memcg->memory.min), |
6730 | READ_ONCE(parent->memory.emin), | |
b3a7822e | 6731 | atomic_long_read(&parent->memory.children_min_usage))); |
23067153 | 6732 | |
b3a7822e | 6733 | WRITE_ONCE(memcg->memory.elow, effective_protection(usage, parent_usage, |
03960e33 CD |
6734 | READ_ONCE(memcg->memory.low), |
6735 | READ_ONCE(parent->memory.elow), | |
b3a7822e | 6736 | atomic_long_read(&parent->memory.children_low_usage))); |
241994ed JW |
6737 | } |
6738 | ||
00501b53 | 6739 | /** |
f0e45fb4 | 6740 | * mem_cgroup_charge - charge a newly allocated page to a cgroup |
00501b53 JW |
6741 | * @page: page to charge |
6742 | * @mm: mm context of the victim | |
6743 | * @gfp_mask: reclaim mode | |
00501b53 JW |
6744 | * |
6745 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
6746 | * pages according to @gfp_mask if necessary. | |
6747 | * | |
f0e45fb4 | 6748 | * Returns 0 on success. Otherwise, an error code is returned. |
00501b53 | 6749 | */ |
d9eb1ea2 | 6750 | int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask) |
00501b53 | 6751 | { |
6c357848 | 6752 | unsigned int nr_pages = thp_nr_pages(page); |
00501b53 | 6753 | struct mem_cgroup *memcg = NULL; |
00501b53 JW |
6754 | int ret = 0; |
6755 | ||
6756 | if (mem_cgroup_disabled()) | |
6757 | goto out; | |
6758 | ||
6759 | if (PageSwapCache(page)) { | |
2d1c4980 JW |
6760 | swp_entry_t ent = { .val = page_private(page), }; |
6761 | unsigned short id; | |
6762 | ||
00501b53 JW |
6763 | /* |
6764 | * Every swap fault against a single page tries to charge the | |
6765 | * page, bail as early as possible. shmem_unuse() encounters | |
eccb52e7 JW |
6766 | * already charged pages, too. page->mem_cgroup is protected |
6767 | * by the page lock, which serializes swap cache removal, which | |
00501b53 JW |
6768 | * in turn serializes uncharging. |
6769 | */ | |
e993d905 | 6770 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
abe2895b | 6771 | if (compound_head(page)->mem_cgroup) |
00501b53 | 6772 | goto out; |
e993d905 | 6773 | |
2d1c4980 JW |
6774 | id = lookup_swap_cgroup_id(ent); |
6775 | rcu_read_lock(); | |
6776 | memcg = mem_cgroup_from_id(id); | |
6777 | if (memcg && !css_tryget_online(&memcg->css)) | |
6778 | memcg = NULL; | |
6779 | rcu_read_unlock(); | |
00501b53 JW |
6780 | } |
6781 | ||
00501b53 JW |
6782 | if (!memcg) |
6783 | memcg = get_mem_cgroup_from_mm(mm); | |
6784 | ||
6785 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
f0e45fb4 JW |
6786 | if (ret) |
6787 | goto out_put; | |
00501b53 | 6788 | |
1a3e1f40 | 6789 | css_get(&memcg->css); |
d9eb1ea2 | 6790 | commit_charge(page, memcg); |
6abb5a86 | 6791 | |
6abb5a86 | 6792 | local_irq_disable(); |
3fba69a5 | 6793 | mem_cgroup_charge_statistics(memcg, page, nr_pages); |
6abb5a86 JW |
6794 | memcg_check_events(memcg, page); |
6795 | local_irq_enable(); | |
00501b53 | 6796 | |
2d1c4980 | 6797 | if (PageSwapCache(page)) { |
00501b53 JW |
6798 | swp_entry_t entry = { .val = page_private(page) }; |
6799 | /* | |
6800 | * The swap entry might not get freed for a long time, | |
6801 | * let's not wait for it. The page already received a | |
6802 | * memory+swap charge, drop the swap entry duplicate. | |
6803 | */ | |
38d8b4e6 | 6804 | mem_cgroup_uncharge_swap(entry, nr_pages); |
00501b53 | 6805 | } |
00501b53 | 6806 | |
f0e45fb4 JW |
6807 | out_put: |
6808 | css_put(&memcg->css); | |
6809 | out: | |
6810 | return ret; | |
3fea5a49 JW |
6811 | } |
6812 | ||
a9d5adee JG |
6813 | struct uncharge_gather { |
6814 | struct mem_cgroup *memcg; | |
9f762dbe | 6815 | unsigned long nr_pages; |
a9d5adee | 6816 | unsigned long pgpgout; |
a9d5adee | 6817 | unsigned long nr_kmem; |
a9d5adee JG |
6818 | struct page *dummy_page; |
6819 | }; | |
6820 | ||
6821 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 6822 | { |
a9d5adee JG |
6823 | memset(ug, 0, sizeof(*ug)); |
6824 | } | |
6825 | ||
6826 | static void uncharge_batch(const struct uncharge_gather *ug) | |
6827 | { | |
747db954 JW |
6828 | unsigned long flags; |
6829 | ||
a9d5adee | 6830 | if (!mem_cgroup_is_root(ug->memcg)) { |
9f762dbe | 6831 | page_counter_uncharge(&ug->memcg->memory, ug->nr_pages); |
7941d214 | 6832 | if (do_memsw_account()) |
9f762dbe | 6833 | page_counter_uncharge(&ug->memcg->memsw, ug->nr_pages); |
a9d5adee JG |
6834 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem) |
6835 | page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem); | |
6836 | memcg_oom_recover(ug->memcg); | |
ce00a967 | 6837 | } |
747db954 JW |
6838 | |
6839 | local_irq_save(flags); | |
c9019e9b | 6840 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); |
9f762dbe | 6841 | __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, ug->nr_pages); |
a9d5adee | 6842 | memcg_check_events(ug->memcg, ug->dummy_page); |
747db954 | 6843 | local_irq_restore(flags); |
f1796544 MH |
6844 | |
6845 | /* drop reference from uncharge_page */ | |
6846 | css_put(&ug->memcg->css); | |
a9d5adee JG |
6847 | } |
6848 | ||
6849 | static void uncharge_page(struct page *page, struct uncharge_gather *ug) | |
6850 | { | |
9f762dbe JW |
6851 | unsigned long nr_pages; |
6852 | ||
a9d5adee | 6853 | VM_BUG_ON_PAGE(PageLRU(page), page); |
a9d5adee JG |
6854 | |
6855 | if (!page->mem_cgroup) | |
6856 | return; | |
6857 | ||
6858 | /* | |
6859 | * Nobody should be changing or seriously looking at | |
6860 | * page->mem_cgroup at this point, we have fully | |
6861 | * exclusive access to the page. | |
6862 | */ | |
6863 | ||
6864 | if (ug->memcg != page->mem_cgroup) { | |
6865 | if (ug->memcg) { | |
6866 | uncharge_batch(ug); | |
6867 | uncharge_gather_clear(ug); | |
6868 | } | |
6869 | ug->memcg = page->mem_cgroup; | |
f1796544 MH |
6870 | |
6871 | /* pairs with css_put in uncharge_batch */ | |
6872 | css_get(&ug->memcg->css); | |
a9d5adee JG |
6873 | } |
6874 | ||
9f762dbe JW |
6875 | nr_pages = compound_nr(page); |
6876 | ug->nr_pages += nr_pages; | |
a9d5adee | 6877 | |
9f762dbe | 6878 | if (!PageKmemcg(page)) { |
a9d5adee JG |
6879 | ug->pgpgout++; |
6880 | } else { | |
9f762dbe | 6881 | ug->nr_kmem += nr_pages; |
a9d5adee JG |
6882 | __ClearPageKmemcg(page); |
6883 | } | |
6884 | ||
6885 | ug->dummy_page = page; | |
6886 | page->mem_cgroup = NULL; | |
1a3e1f40 | 6887 | css_put(&ug->memcg->css); |
747db954 JW |
6888 | } |
6889 | ||
6890 | static void uncharge_list(struct list_head *page_list) | |
6891 | { | |
a9d5adee | 6892 | struct uncharge_gather ug; |
747db954 | 6893 | struct list_head *next; |
a9d5adee JG |
6894 | |
6895 | uncharge_gather_clear(&ug); | |
747db954 | 6896 | |
8b592656 JW |
6897 | /* |
6898 | * Note that the list can be a single page->lru; hence the | |
6899 | * do-while loop instead of a simple list_for_each_entry(). | |
6900 | */ | |
747db954 JW |
6901 | next = page_list->next; |
6902 | do { | |
a9d5adee JG |
6903 | struct page *page; |
6904 | ||
747db954 JW |
6905 | page = list_entry(next, struct page, lru); |
6906 | next = page->lru.next; | |
6907 | ||
a9d5adee | 6908 | uncharge_page(page, &ug); |
747db954 JW |
6909 | } while (next != page_list); |
6910 | ||
a9d5adee JG |
6911 | if (ug.memcg) |
6912 | uncharge_batch(&ug); | |
747db954 JW |
6913 | } |
6914 | ||
0a31bc97 JW |
6915 | /** |
6916 | * mem_cgroup_uncharge - uncharge a page | |
6917 | * @page: page to uncharge | |
6918 | * | |
f0e45fb4 | 6919 | * Uncharge a page previously charged with mem_cgroup_charge(). |
0a31bc97 JW |
6920 | */ |
6921 | void mem_cgroup_uncharge(struct page *page) | |
6922 | { | |
a9d5adee JG |
6923 | struct uncharge_gather ug; |
6924 | ||
0a31bc97 JW |
6925 | if (mem_cgroup_disabled()) |
6926 | return; | |
6927 | ||
747db954 | 6928 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 6929 | if (!page->mem_cgroup) |
0a31bc97 JW |
6930 | return; |
6931 | ||
a9d5adee JG |
6932 | uncharge_gather_clear(&ug); |
6933 | uncharge_page(page, &ug); | |
6934 | uncharge_batch(&ug); | |
747db954 | 6935 | } |
0a31bc97 | 6936 | |
747db954 JW |
6937 | /** |
6938 | * mem_cgroup_uncharge_list - uncharge a list of page | |
6939 | * @page_list: list of pages to uncharge | |
6940 | * | |
6941 | * Uncharge a list of pages previously charged with | |
f0e45fb4 | 6942 | * mem_cgroup_charge(). |
747db954 JW |
6943 | */ |
6944 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
6945 | { | |
6946 | if (mem_cgroup_disabled()) | |
6947 | return; | |
0a31bc97 | 6948 | |
747db954 JW |
6949 | if (!list_empty(page_list)) |
6950 | uncharge_list(page_list); | |
0a31bc97 JW |
6951 | } |
6952 | ||
6953 | /** | |
6a93ca8f JW |
6954 | * mem_cgroup_migrate - charge a page's replacement |
6955 | * @oldpage: currently circulating page | |
6956 | * @newpage: replacement page | |
0a31bc97 | 6957 | * |
6a93ca8f JW |
6958 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
6959 | * be uncharged upon free. | |
0a31bc97 JW |
6960 | * |
6961 | * Both pages must be locked, @newpage->mapping must be set up. | |
6962 | */ | |
6a93ca8f | 6963 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 6964 | { |
29833315 | 6965 | struct mem_cgroup *memcg; |
44b7a8d3 | 6966 | unsigned int nr_pages; |
d93c4130 | 6967 | unsigned long flags; |
0a31bc97 JW |
6968 | |
6969 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6970 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 6971 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
6972 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6973 | newpage); | |
0a31bc97 JW |
6974 | |
6975 | if (mem_cgroup_disabled()) | |
6976 | return; | |
6977 | ||
6978 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 6979 | if (newpage->mem_cgroup) |
0a31bc97 JW |
6980 | return; |
6981 | ||
45637bab | 6982 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 6983 | memcg = oldpage->mem_cgroup; |
29833315 | 6984 | if (!memcg) |
0a31bc97 JW |
6985 | return; |
6986 | ||
44b7a8d3 | 6987 | /* Force-charge the new page. The old one will be freed soon */ |
6c357848 | 6988 | nr_pages = thp_nr_pages(newpage); |
44b7a8d3 JW |
6989 | |
6990 | page_counter_charge(&memcg->memory, nr_pages); | |
6991 | if (do_memsw_account()) | |
6992 | page_counter_charge(&memcg->memsw, nr_pages); | |
0a31bc97 | 6993 | |
1a3e1f40 | 6994 | css_get(&memcg->css); |
d9eb1ea2 | 6995 | commit_charge(newpage, memcg); |
44b7a8d3 | 6996 | |
d93c4130 | 6997 | local_irq_save(flags); |
3fba69a5 | 6998 | mem_cgroup_charge_statistics(memcg, newpage, nr_pages); |
44b7a8d3 | 6999 | memcg_check_events(memcg, newpage); |
d93c4130 | 7000 | local_irq_restore(flags); |
0a31bc97 JW |
7001 | } |
7002 | ||
ef12947c | 7003 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
7004 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
7005 | ||
2d758073 | 7006 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
7007 | { |
7008 | struct mem_cgroup *memcg; | |
7009 | ||
2d758073 JW |
7010 | if (!mem_cgroup_sockets_enabled) |
7011 | return; | |
7012 | ||
e876ecc6 SB |
7013 | /* Do not associate the sock with unrelated interrupted task's memcg. */ |
7014 | if (in_interrupt()) | |
7015 | return; | |
7016 | ||
11092087 JW |
7017 | rcu_read_lock(); |
7018 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
7019 | if (memcg == root_mem_cgroup) |
7020 | goto out; | |
0db15298 | 7021 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 7022 | goto out; |
8965aa28 | 7023 | if (css_tryget(&memcg->css)) |
11092087 | 7024 | sk->sk_memcg = memcg; |
f7e1cb6e | 7025 | out: |
11092087 JW |
7026 | rcu_read_unlock(); |
7027 | } | |
11092087 | 7028 | |
2d758073 | 7029 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 7030 | { |
2d758073 JW |
7031 | if (sk->sk_memcg) |
7032 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
7033 | } |
7034 | ||
7035 | /** | |
7036 | * mem_cgroup_charge_skmem - charge socket memory | |
7037 | * @memcg: memcg to charge | |
7038 | * @nr_pages: number of pages to charge | |
7039 | * | |
7040 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
7041 | * @memcg's configured limit, %false if the charge had to be forced. | |
7042 | */ | |
7043 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
7044 | { | |
f7e1cb6e | 7045 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 7046 | |
f7e1cb6e | 7047 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 7048 | struct page_counter *fail; |
f7e1cb6e | 7049 | |
0db15298 JW |
7050 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
7051 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
7052 | return true; |
7053 | } | |
0db15298 JW |
7054 | page_counter_charge(&memcg->tcpmem, nr_pages); |
7055 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 7056 | return false; |
11092087 | 7057 | } |
d886f4e4 | 7058 | |
f7e1cb6e JW |
7059 | /* Don't block in the packet receive path */ |
7060 | if (in_softirq()) | |
7061 | gfp_mask = GFP_NOWAIT; | |
7062 | ||
c9019e9b | 7063 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); |
b2807f07 | 7064 | |
f7e1cb6e JW |
7065 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
7066 | return true; | |
7067 | ||
7068 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
7069 | return false; |
7070 | } | |
7071 | ||
7072 | /** | |
7073 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
7074 | * @memcg: memcg to uncharge |
7075 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
7076 | */ |
7077 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
7078 | { | |
f7e1cb6e | 7079 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 7080 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
7081 | return; |
7082 | } | |
d886f4e4 | 7083 | |
c9019e9b | 7084 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 7085 | |
475d0487 | 7086 | refill_stock(memcg, nr_pages); |
11092087 JW |
7087 | } |
7088 | ||
f7e1cb6e JW |
7089 | static int __init cgroup_memory(char *s) |
7090 | { | |
7091 | char *token; | |
7092 | ||
7093 | while ((token = strsep(&s, ",")) != NULL) { | |
7094 | if (!*token) | |
7095 | continue; | |
7096 | if (!strcmp(token, "nosocket")) | |
7097 | cgroup_memory_nosocket = true; | |
04823c83 VD |
7098 | if (!strcmp(token, "nokmem")) |
7099 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
7100 | } |
7101 | return 0; | |
7102 | } | |
7103 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 7104 | |
2d11085e | 7105 | /* |
1081312f MH |
7106 | * subsys_initcall() for memory controller. |
7107 | * | |
308167fc SAS |
7108 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
7109 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
7110 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
7111 | * should be initialized from here. | |
2d11085e MH |
7112 | */ |
7113 | static int __init mem_cgroup_init(void) | |
7114 | { | |
95a045f6 JW |
7115 | int cpu, node; |
7116 | ||
308167fc SAS |
7117 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
7118 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
7119 | |
7120 | for_each_possible_cpu(cpu) | |
7121 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
7122 | drain_local_stock); | |
7123 | ||
7124 | for_each_node(node) { | |
7125 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
7126 | |
7127 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
7128 | node_online(node) ? node : NUMA_NO_NODE); | |
7129 | ||
ef8f2327 | 7130 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 7131 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 7132 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
7133 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
7134 | } | |
7135 | ||
2d11085e MH |
7136 | return 0; |
7137 | } | |
7138 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
7139 | |
7140 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
7141 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
7142 | { | |
1c2d479a | 7143 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
7144 | /* |
7145 | * The root cgroup cannot be destroyed, so it's refcount must | |
7146 | * always be >= 1. | |
7147 | */ | |
7148 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
7149 | VM_BUG_ON(1); | |
7150 | break; | |
7151 | } | |
7152 | memcg = parent_mem_cgroup(memcg); | |
7153 | if (!memcg) | |
7154 | memcg = root_mem_cgroup; | |
7155 | } | |
7156 | return memcg; | |
7157 | } | |
7158 | ||
21afa38e JW |
7159 | /** |
7160 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
7161 | * @page: page whose memsw charge to transfer | |
7162 | * @entry: swap entry to move the charge to | |
7163 | * | |
7164 | * Transfer the memsw charge of @page to @entry. | |
7165 | */ | |
7166 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
7167 | { | |
1f47b61f | 7168 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 7169 | unsigned int nr_entries; |
21afa38e JW |
7170 | unsigned short oldid; |
7171 | ||
7172 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
7173 | VM_BUG_ON_PAGE(page_count(page), page); | |
7174 | ||
2d1c4980 | 7175 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
21afa38e JW |
7176 | return; |
7177 | ||
7178 | memcg = page->mem_cgroup; | |
7179 | ||
7180 | /* Readahead page, never charged */ | |
7181 | if (!memcg) | |
7182 | return; | |
7183 | ||
1f47b61f VD |
7184 | /* |
7185 | * In case the memcg owning these pages has been offlined and doesn't | |
7186 | * have an ID allocated to it anymore, charge the closest online | |
7187 | * ancestor for the swap instead and transfer the memory+swap charge. | |
7188 | */ | |
7189 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
6c357848 | 7190 | nr_entries = thp_nr_pages(page); |
d6810d73 HY |
7191 | /* Get references for the tail pages, too */ |
7192 | if (nr_entries > 1) | |
7193 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
7194 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
7195 | nr_entries); | |
21afa38e | 7196 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7197 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e JW |
7198 | |
7199 | page->mem_cgroup = NULL; | |
7200 | ||
7201 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 7202 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 7203 | |
2d1c4980 | 7204 | if (!cgroup_memory_noswap && memcg != swap_memcg) { |
1f47b61f | 7205 | if (!mem_cgroup_is_root(swap_memcg)) |
d6810d73 HY |
7206 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
7207 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
7208 | } |
7209 | ||
ce9ce665 SAS |
7210 | /* |
7211 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 7212 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 7213 | * important here to have the interrupts disabled because it is the |
b93b0163 | 7214 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 SAS |
7215 | */ |
7216 | VM_BUG_ON(!irqs_disabled()); | |
3fba69a5 | 7217 | mem_cgroup_charge_statistics(memcg, page, -nr_entries); |
21afa38e | 7218 | memcg_check_events(memcg, page); |
73f576c0 | 7219 | |
1a3e1f40 | 7220 | css_put(&memcg->css); |
21afa38e JW |
7221 | } |
7222 | ||
38d8b4e6 HY |
7223 | /** |
7224 | * mem_cgroup_try_charge_swap - try charging swap space for a page | |
37e84351 VD |
7225 | * @page: page being added to swap |
7226 | * @entry: swap entry to charge | |
7227 | * | |
38d8b4e6 | 7228 | * Try to charge @page's memcg for the swap space at @entry. |
37e84351 VD |
7229 | * |
7230 | * Returns 0 on success, -ENOMEM on failure. | |
7231 | */ | |
7232 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
7233 | { | |
6c357848 | 7234 | unsigned int nr_pages = thp_nr_pages(page); |
37e84351 | 7235 | struct page_counter *counter; |
38d8b4e6 | 7236 | struct mem_cgroup *memcg; |
37e84351 VD |
7237 | unsigned short oldid; |
7238 | ||
2d1c4980 | 7239 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
37e84351 VD |
7240 | return 0; |
7241 | ||
7242 | memcg = page->mem_cgroup; | |
7243 | ||
7244 | /* Readahead page, never charged */ | |
7245 | if (!memcg) | |
7246 | return 0; | |
7247 | ||
f3a53a3a TH |
7248 | if (!entry.val) { |
7249 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 7250 | return 0; |
f3a53a3a | 7251 | } |
bb98f2c5 | 7252 | |
1f47b61f VD |
7253 | memcg = mem_cgroup_id_get_online(memcg); |
7254 | ||
2d1c4980 | 7255 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg) && |
38d8b4e6 | 7256 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
7257 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
7258 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 7259 | mem_cgroup_id_put(memcg); |
37e84351 | 7260 | return -ENOMEM; |
1f47b61f | 7261 | } |
37e84351 | 7262 | |
38d8b4e6 HY |
7263 | /* Get references for the tail pages, too */ |
7264 | if (nr_pages > 1) | |
7265 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
7266 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
37e84351 | 7267 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7268 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 7269 | |
37e84351 VD |
7270 | return 0; |
7271 | } | |
7272 | ||
21afa38e | 7273 | /** |
38d8b4e6 | 7274 | * mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 7275 | * @entry: swap entry to uncharge |
38d8b4e6 | 7276 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 7277 | */ |
38d8b4e6 | 7278 | void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
7279 | { |
7280 | struct mem_cgroup *memcg; | |
7281 | unsigned short id; | |
7282 | ||
38d8b4e6 | 7283 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 7284 | rcu_read_lock(); |
adbe427b | 7285 | memcg = mem_cgroup_from_id(id); |
21afa38e | 7286 | if (memcg) { |
2d1c4980 | 7287 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg)) { |
37e84351 | 7288 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
38d8b4e6 | 7289 | page_counter_uncharge(&memcg->swap, nr_pages); |
37e84351 | 7290 | else |
38d8b4e6 | 7291 | page_counter_uncharge(&memcg->memsw, nr_pages); |
37e84351 | 7292 | } |
c9019e9b | 7293 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 7294 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
7295 | } |
7296 | rcu_read_unlock(); | |
7297 | } | |
7298 | ||
d8b38438 VD |
7299 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
7300 | { | |
7301 | long nr_swap_pages = get_nr_swap_pages(); | |
7302 | ||
eccb52e7 | 7303 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
d8b38438 VD |
7304 | return nr_swap_pages; |
7305 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
7306 | nr_swap_pages = min_t(long, nr_swap_pages, | |
bbec2e15 | 7307 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
7308 | page_counter_read(&memcg->swap)); |
7309 | return nr_swap_pages; | |
7310 | } | |
7311 | ||
5ccc5aba VD |
7312 | bool mem_cgroup_swap_full(struct page *page) |
7313 | { | |
7314 | struct mem_cgroup *memcg; | |
7315 | ||
7316 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
7317 | ||
7318 | if (vm_swap_full()) | |
7319 | return true; | |
eccb52e7 | 7320 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
5ccc5aba VD |
7321 | return false; |
7322 | ||
7323 | memcg = page->mem_cgroup; | |
7324 | if (!memcg) | |
7325 | return false; | |
7326 | ||
4b82ab4f JK |
7327 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { |
7328 | unsigned long usage = page_counter_read(&memcg->swap); | |
7329 | ||
7330 | if (usage * 2 >= READ_ONCE(memcg->swap.high) || | |
7331 | usage * 2 >= READ_ONCE(memcg->swap.max)) | |
5ccc5aba | 7332 | return true; |
4b82ab4f | 7333 | } |
5ccc5aba VD |
7334 | |
7335 | return false; | |
7336 | } | |
7337 | ||
eccb52e7 | 7338 | static int __init setup_swap_account(char *s) |
21afa38e JW |
7339 | { |
7340 | if (!strcmp(s, "1")) | |
eccb52e7 | 7341 | cgroup_memory_noswap = 0; |
21afa38e | 7342 | else if (!strcmp(s, "0")) |
eccb52e7 | 7343 | cgroup_memory_noswap = 1; |
21afa38e JW |
7344 | return 1; |
7345 | } | |
eccb52e7 | 7346 | __setup("swapaccount=", setup_swap_account); |
21afa38e | 7347 | |
37e84351 VD |
7348 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
7349 | struct cftype *cft) | |
7350 | { | |
7351 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
7352 | ||
7353 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
7354 | } | |
7355 | ||
4b82ab4f JK |
7356 | static int swap_high_show(struct seq_file *m, void *v) |
7357 | { | |
7358 | return seq_puts_memcg_tunable(m, | |
7359 | READ_ONCE(mem_cgroup_from_seq(m)->swap.high)); | |
7360 | } | |
7361 | ||
7362 | static ssize_t swap_high_write(struct kernfs_open_file *of, | |
7363 | char *buf, size_t nbytes, loff_t off) | |
7364 | { | |
7365 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7366 | unsigned long high; | |
7367 | int err; | |
7368 | ||
7369 | buf = strstrip(buf); | |
7370 | err = page_counter_memparse(buf, "max", &high); | |
7371 | if (err) | |
7372 | return err; | |
7373 | ||
7374 | page_counter_set_high(&memcg->swap, high); | |
7375 | ||
7376 | return nbytes; | |
7377 | } | |
7378 | ||
37e84351 VD |
7379 | static int swap_max_show(struct seq_file *m, void *v) |
7380 | { | |
677dc973 CD |
7381 | return seq_puts_memcg_tunable(m, |
7382 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); | |
37e84351 VD |
7383 | } |
7384 | ||
7385 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
7386 | char *buf, size_t nbytes, loff_t off) | |
7387 | { | |
7388 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7389 | unsigned long max; | |
7390 | int err; | |
7391 | ||
7392 | buf = strstrip(buf); | |
7393 | err = page_counter_memparse(buf, "max", &max); | |
7394 | if (err) | |
7395 | return err; | |
7396 | ||
be09102b | 7397 | xchg(&memcg->swap.max, max); |
37e84351 VD |
7398 | |
7399 | return nbytes; | |
7400 | } | |
7401 | ||
f3a53a3a TH |
7402 | static int swap_events_show(struct seq_file *m, void *v) |
7403 | { | |
aa9694bb | 7404 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
f3a53a3a | 7405 | |
4b82ab4f JK |
7406 | seq_printf(m, "high %lu\n", |
7407 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_HIGH])); | |
f3a53a3a TH |
7408 | seq_printf(m, "max %lu\n", |
7409 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
7410 | seq_printf(m, "fail %lu\n", | |
7411 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
7412 | ||
7413 | return 0; | |
7414 | } | |
7415 | ||
37e84351 VD |
7416 | static struct cftype swap_files[] = { |
7417 | { | |
7418 | .name = "swap.current", | |
7419 | .flags = CFTYPE_NOT_ON_ROOT, | |
7420 | .read_u64 = swap_current_read, | |
7421 | }, | |
4b82ab4f JK |
7422 | { |
7423 | .name = "swap.high", | |
7424 | .flags = CFTYPE_NOT_ON_ROOT, | |
7425 | .seq_show = swap_high_show, | |
7426 | .write = swap_high_write, | |
7427 | }, | |
37e84351 VD |
7428 | { |
7429 | .name = "swap.max", | |
7430 | .flags = CFTYPE_NOT_ON_ROOT, | |
7431 | .seq_show = swap_max_show, | |
7432 | .write = swap_max_write, | |
7433 | }, | |
f3a53a3a TH |
7434 | { |
7435 | .name = "swap.events", | |
7436 | .flags = CFTYPE_NOT_ON_ROOT, | |
7437 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
7438 | .seq_show = swap_events_show, | |
7439 | }, | |
37e84351 VD |
7440 | { } /* terminate */ |
7441 | }; | |
7442 | ||
eccb52e7 | 7443 | static struct cftype memsw_files[] = { |
21afa38e JW |
7444 | { |
7445 | .name = "memsw.usage_in_bytes", | |
7446 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
7447 | .read_u64 = mem_cgroup_read_u64, | |
7448 | }, | |
7449 | { | |
7450 | .name = "memsw.max_usage_in_bytes", | |
7451 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
7452 | .write = mem_cgroup_reset, | |
7453 | .read_u64 = mem_cgroup_read_u64, | |
7454 | }, | |
7455 | { | |
7456 | .name = "memsw.limit_in_bytes", | |
7457 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
7458 | .write = mem_cgroup_write, | |
7459 | .read_u64 = mem_cgroup_read_u64, | |
7460 | }, | |
7461 | { | |
7462 | .name = "memsw.failcnt", | |
7463 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
7464 | .write = mem_cgroup_reset, | |
7465 | .read_u64 = mem_cgroup_read_u64, | |
7466 | }, | |
7467 | { }, /* terminate */ | |
7468 | }; | |
7469 | ||
82ff165c BS |
7470 | /* |
7471 | * If mem_cgroup_swap_init() is implemented as a subsys_initcall() | |
7472 | * instead of a core_initcall(), this could mean cgroup_memory_noswap still | |
7473 | * remains set to false even when memcg is disabled via "cgroup_disable=memory" | |
7474 | * boot parameter. This may result in premature OOPS inside | |
7475 | * mem_cgroup_get_nr_swap_pages() function in corner cases. | |
7476 | */ | |
21afa38e JW |
7477 | static int __init mem_cgroup_swap_init(void) |
7478 | { | |
2d1c4980 JW |
7479 | /* No memory control -> no swap control */ |
7480 | if (mem_cgroup_disabled()) | |
7481 | cgroup_memory_noswap = true; | |
7482 | ||
7483 | if (cgroup_memory_noswap) | |
eccb52e7 JW |
7484 | return 0; |
7485 | ||
7486 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, swap_files)); | |
7487 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, memsw_files)); | |
7488 | ||
21afa38e JW |
7489 | return 0; |
7490 | } | |
82ff165c | 7491 | core_initcall(mem_cgroup_swap_init); |
21afa38e JW |
7492 | |
7493 | #endif /* CONFIG_MEMCG_SWAP */ |