]>
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 | 625 | rightmost = false; |
378876b0 | 626 | } else { |
bb4cc1a8 | 627 | p = &(*p)->rb_right; |
378876b0 | 628 | } |
bb4cc1a8 | 629 | } |
fa90b2fd DB |
630 | |
631 | if (rightmost) | |
632 | mctz->rb_rightmost = &mz->tree_node; | |
633 | ||
bb4cc1a8 AM |
634 | rb_link_node(&mz->tree_node, parent, p); |
635 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
636 | mz->on_tree = true; | |
637 | } | |
638 | ||
ef8f2327 MG |
639 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
640 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
641 | { |
642 | if (!mz->on_tree) | |
643 | return; | |
fa90b2fd DB |
644 | |
645 | if (&mz->tree_node == mctz->rb_rightmost) | |
646 | mctz->rb_rightmost = rb_prev(&mz->tree_node); | |
647 | ||
bb4cc1a8 AM |
648 | rb_erase(&mz->tree_node, &mctz->rb_root); |
649 | mz->on_tree = false; | |
650 | } | |
651 | ||
ef8f2327 MG |
652 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
653 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 654 | { |
0a31bc97 JW |
655 | unsigned long flags; |
656 | ||
657 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 658 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 659 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
660 | } |
661 | ||
3e32cb2e JW |
662 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
663 | { | |
664 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 665 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
666 | unsigned long excess = 0; |
667 | ||
668 | if (nr_pages > soft_limit) | |
669 | excess = nr_pages - soft_limit; | |
670 | ||
671 | return excess; | |
672 | } | |
bb4cc1a8 AM |
673 | |
674 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
675 | { | |
3e32cb2e | 676 | unsigned long excess; |
ef8f2327 MG |
677 | struct mem_cgroup_per_node *mz; |
678 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 679 | |
e231875b | 680 | mctz = soft_limit_tree_from_page(page); |
bfc7228b LD |
681 | if (!mctz) |
682 | return; | |
bb4cc1a8 AM |
683 | /* |
684 | * Necessary to update all ancestors when hierarchy is used. | |
685 | * because their event counter is not touched. | |
686 | */ | |
687 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
ef8f2327 | 688 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
3e32cb2e | 689 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
690 | /* |
691 | * We have to update the tree if mz is on RB-tree or | |
692 | * mem is over its softlimit. | |
693 | */ | |
694 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
695 | unsigned long flags; |
696 | ||
697 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
698 | /* if on-tree, remove it */ |
699 | if (mz->on_tree) | |
cf2c8127 | 700 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
701 | /* |
702 | * Insert again. mz->usage_in_excess will be updated. | |
703 | * If excess is 0, no tree ops. | |
704 | */ | |
cf2c8127 | 705 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 706 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
707 | } |
708 | } | |
709 | } | |
710 | ||
711 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
712 | { | |
ef8f2327 MG |
713 | struct mem_cgroup_tree_per_node *mctz; |
714 | struct mem_cgroup_per_node *mz; | |
715 | int nid; | |
bb4cc1a8 | 716 | |
e231875b | 717 | for_each_node(nid) { |
ef8f2327 MG |
718 | mz = mem_cgroup_nodeinfo(memcg, nid); |
719 | mctz = soft_limit_tree_node(nid); | |
bfc7228b LD |
720 | if (mctz) |
721 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
722 | } |
723 | } | |
724 | ||
ef8f2327 MG |
725 | static struct mem_cgroup_per_node * |
726 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 727 | { |
ef8f2327 | 728 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
729 | |
730 | retry: | |
731 | mz = NULL; | |
fa90b2fd | 732 | if (!mctz->rb_rightmost) |
bb4cc1a8 AM |
733 | goto done; /* Nothing to reclaim from */ |
734 | ||
fa90b2fd DB |
735 | mz = rb_entry(mctz->rb_rightmost, |
736 | struct mem_cgroup_per_node, tree_node); | |
bb4cc1a8 AM |
737 | /* |
738 | * Remove the node now but someone else can add it back, | |
739 | * we will to add it back at the end of reclaim to its correct | |
740 | * position in the tree. | |
741 | */ | |
cf2c8127 | 742 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 743 | if (!soft_limit_excess(mz->memcg) || |
8965aa28 | 744 | !css_tryget(&mz->memcg->css)) |
bb4cc1a8 AM |
745 | goto retry; |
746 | done: | |
747 | return mz; | |
748 | } | |
749 | ||
ef8f2327 MG |
750 | static struct mem_cgroup_per_node * |
751 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 752 | { |
ef8f2327 | 753 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 754 | |
0a31bc97 | 755 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 756 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 757 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
758 | return mz; |
759 | } | |
760 | ||
db9adbcb JW |
761 | /** |
762 | * __mod_memcg_state - update cgroup memory statistics | |
763 | * @memcg: the memory cgroup | |
764 | * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item | |
765 | * @val: delta to add to the counter, can be negative | |
766 | */ | |
767 | void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val) | |
768 | { | |
ea426c2a | 769 | long x, threshold = MEMCG_CHARGE_BATCH; |
db9adbcb JW |
770 | |
771 | if (mem_cgroup_disabled()) | |
772 | return; | |
773 | ||
772616b0 | 774 | if (memcg_stat_item_in_bytes(idx)) |
ea426c2a RG |
775 | threshold <<= PAGE_SHIFT; |
776 | ||
db9adbcb | 777 | x = val + __this_cpu_read(memcg->vmstats_percpu->stat[idx]); |
ea426c2a | 778 | if (unlikely(abs(x) > threshold)) { |
42a30035 JW |
779 | struct mem_cgroup *mi; |
780 | ||
766a4c19 YS |
781 | /* |
782 | * Batch local counters to keep them in sync with | |
783 | * the hierarchical ones. | |
784 | */ | |
785 | __this_cpu_add(memcg->vmstats_local->stat[idx], x); | |
42a30035 JW |
786 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
787 | atomic_long_add(x, &mi->vmstats[idx]); | |
db9adbcb JW |
788 | x = 0; |
789 | } | |
790 | __this_cpu_write(memcg->vmstats_percpu->stat[idx], x); | |
791 | } | |
792 | ||
42a30035 JW |
793 | static struct mem_cgroup_per_node * |
794 | parent_nodeinfo(struct mem_cgroup_per_node *pn, int nid) | |
795 | { | |
796 | struct mem_cgroup *parent; | |
797 | ||
798 | parent = parent_mem_cgroup(pn->memcg); | |
799 | if (!parent) | |
800 | return NULL; | |
801 | return mem_cgroup_nodeinfo(parent, nid); | |
802 | } | |
803 | ||
eedc4e5a RG |
804 | void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, |
805 | int val) | |
db9adbcb JW |
806 | { |
807 | struct mem_cgroup_per_node *pn; | |
42a30035 | 808 | struct mem_cgroup *memcg; |
ea426c2a | 809 | long x, threshold = MEMCG_CHARGE_BATCH; |
db9adbcb | 810 | |
db9adbcb | 811 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
42a30035 | 812 | memcg = pn->memcg; |
db9adbcb JW |
813 | |
814 | /* Update memcg */ | |
42a30035 | 815 | __mod_memcg_state(memcg, idx, val); |
db9adbcb | 816 | |
b4c46484 RG |
817 | /* Update lruvec */ |
818 | __this_cpu_add(pn->lruvec_stat_local->count[idx], val); | |
819 | ||
ea426c2a RG |
820 | if (vmstat_item_in_bytes(idx)) |
821 | threshold <<= PAGE_SHIFT; | |
822 | ||
db9adbcb | 823 | x = val + __this_cpu_read(pn->lruvec_stat_cpu->count[idx]); |
ea426c2a | 824 | if (unlikely(abs(x) > threshold)) { |
eedc4e5a | 825 | pg_data_t *pgdat = lruvec_pgdat(lruvec); |
42a30035 JW |
826 | struct mem_cgroup_per_node *pi; |
827 | ||
42a30035 JW |
828 | for (pi = pn; pi; pi = parent_nodeinfo(pi, pgdat->node_id)) |
829 | atomic_long_add(x, &pi->lruvec_stat[idx]); | |
db9adbcb JW |
830 | x = 0; |
831 | } | |
832 | __this_cpu_write(pn->lruvec_stat_cpu->count[idx], x); | |
833 | } | |
834 | ||
eedc4e5a RG |
835 | /** |
836 | * __mod_lruvec_state - update lruvec memory statistics | |
837 | * @lruvec: the lruvec | |
838 | * @idx: the stat item | |
839 | * @val: delta to add to the counter, can be negative | |
840 | * | |
841 | * The lruvec is the intersection of the NUMA node and a cgroup. This | |
842 | * function updates the all three counters that are affected by a | |
843 | * change of state at this level: per-node, per-cgroup, per-lruvec. | |
844 | */ | |
845 | void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, | |
846 | int val) | |
847 | { | |
848 | /* Update node */ | |
849 | __mod_node_page_state(lruvec_pgdat(lruvec), idx, val); | |
850 | ||
851 | /* Update memcg and lruvec */ | |
852 | if (!mem_cgroup_disabled()) | |
853 | __mod_memcg_lruvec_state(lruvec, idx, val); | |
854 | } | |
855 | ||
da3ceeff | 856 | void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, int val) |
ec9f0238 | 857 | { |
4f103c63 | 858 | pg_data_t *pgdat = page_pgdat(virt_to_page(p)); |
ec9f0238 RG |
859 | struct mem_cgroup *memcg; |
860 | struct lruvec *lruvec; | |
861 | ||
862 | rcu_read_lock(); | |
4f103c63 | 863 | memcg = mem_cgroup_from_obj(p); |
ec9f0238 | 864 | |
8faeb1ff MS |
865 | /* |
866 | * Untracked pages have no memcg, no lruvec. Update only the | |
867 | * node. If we reparent the slab objects to the root memcg, | |
868 | * when we free the slab object, we need to update the per-memcg | |
869 | * vmstats to keep it correct for the root memcg. | |
870 | */ | |
871 | if (!memcg) { | |
ec9f0238 RG |
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 | ||
db9adbcb JW |
880 | /** |
881 | * __count_memcg_events - account VM events in a cgroup | |
882 | * @memcg: the memory cgroup | |
883 | * @idx: the event item | |
884 | * @count: the number of events that occured | |
885 | */ | |
886 | void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, | |
887 | unsigned long count) | |
888 | { | |
889 | unsigned long x; | |
890 | ||
891 | if (mem_cgroup_disabled()) | |
892 | return; | |
893 | ||
894 | x = count + __this_cpu_read(memcg->vmstats_percpu->events[idx]); | |
895 | if (unlikely(x > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
896 | struct mem_cgroup *mi; |
897 | ||
766a4c19 YS |
898 | /* |
899 | * Batch local counters to keep them in sync with | |
900 | * the hierarchical ones. | |
901 | */ | |
902 | __this_cpu_add(memcg->vmstats_local->events[idx], x); | |
42a30035 JW |
903 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
904 | atomic_long_add(x, &mi->vmevents[idx]); | |
db9adbcb JW |
905 | x = 0; |
906 | } | |
907 | __this_cpu_write(memcg->vmstats_percpu->events[idx], x); | |
908 | } | |
909 | ||
42a30035 | 910 | static unsigned long memcg_events(struct mem_cgroup *memcg, int event) |
e9f8974f | 911 | { |
871789d4 | 912 | return atomic_long_read(&memcg->vmevents[event]); |
e9f8974f JW |
913 | } |
914 | ||
42a30035 JW |
915 | static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event) |
916 | { | |
815744d7 JW |
917 | long x = 0; |
918 | int cpu; | |
919 | ||
920 | for_each_possible_cpu(cpu) | |
921 | x += per_cpu(memcg->vmstats_local->events[event], cpu); | |
922 | return x; | |
42a30035 JW |
923 | } |
924 | ||
c0ff4b85 | 925 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 926 | struct page *page, |
3fba69a5 | 927 | int nr_pages) |
d52aa412 | 928 | { |
e401f176 KH |
929 | /* pagein of a big page is an event. So, ignore page size */ |
930 | if (nr_pages > 0) | |
c9019e9b | 931 | __count_memcg_events(memcg, PGPGIN, 1); |
3751d604 | 932 | else { |
c9019e9b | 933 | __count_memcg_events(memcg, PGPGOUT, 1); |
3751d604 KH |
934 | nr_pages = -nr_pages; /* for event */ |
935 | } | |
e401f176 | 936 | |
871789d4 | 937 | __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages); |
6d12e2d8 KH |
938 | } |
939 | ||
f53d7ce3 JW |
940 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
941 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
942 | { |
943 | unsigned long val, next; | |
944 | ||
871789d4 CD |
945 | val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events); |
946 | next = __this_cpu_read(memcg->vmstats_percpu->targets[target]); | |
7a159cc9 | 947 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 948 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
949 | switch (target) { |
950 | case MEM_CGROUP_TARGET_THRESH: | |
951 | next = val + THRESHOLDS_EVENTS_TARGET; | |
952 | break; | |
bb4cc1a8 AM |
953 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
954 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
955 | break; | |
f53d7ce3 JW |
956 | default: |
957 | break; | |
958 | } | |
871789d4 | 959 | __this_cpu_write(memcg->vmstats_percpu->targets[target], next); |
f53d7ce3 | 960 | return true; |
7a159cc9 | 961 | } |
f53d7ce3 | 962 | return false; |
d2265e6f KH |
963 | } |
964 | ||
965 | /* | |
966 | * Check events in order. | |
967 | * | |
968 | */ | |
c0ff4b85 | 969 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
970 | { |
971 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
972 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
973 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 974 | bool do_softlimit; |
f53d7ce3 | 975 | |
bb4cc1a8 AM |
976 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
977 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
c0ff4b85 | 978 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
979 | if (unlikely(do_softlimit)) |
980 | mem_cgroup_update_tree(memcg, page); | |
0a31bc97 | 981 | } |
d2265e6f KH |
982 | } |
983 | ||
cf475ad2 | 984 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 985 | { |
31a78f23 BS |
986 | /* |
987 | * mm_update_next_owner() may clear mm->owner to NULL | |
988 | * if it races with swapoff, page migration, etc. | |
989 | * So this can be called with p == NULL. | |
990 | */ | |
991 | if (unlikely(!p)) | |
992 | return NULL; | |
993 | ||
073219e9 | 994 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 995 | } |
33398cf2 | 996 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 997 | |
d46eb14b SB |
998 | /** |
999 | * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg. | |
1000 | * @mm: mm from which memcg should be extracted. It can be NULL. | |
1001 | * | |
1002 | * Obtain a reference on mm->memcg and returns it if successful. Otherwise | |
1003 | * root_mem_cgroup is returned. However if mem_cgroup is disabled, NULL is | |
1004 | * returned. | |
1005 | */ | |
1006 | struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) | |
54595fe2 | 1007 | { |
d46eb14b SB |
1008 | struct mem_cgroup *memcg; |
1009 | ||
1010 | if (mem_cgroup_disabled()) | |
1011 | return NULL; | |
0b7f569e | 1012 | |
54595fe2 KH |
1013 | rcu_read_lock(); |
1014 | do { | |
6f6acb00 MH |
1015 | /* |
1016 | * Page cache insertions can happen withou an | |
1017 | * actual mm context, e.g. during disk probing | |
1018 | * on boot, loopback IO, acct() writes etc. | |
1019 | */ | |
1020 | if (unlikely(!mm)) | |
df381975 | 1021 | memcg = root_mem_cgroup; |
6f6acb00 MH |
1022 | else { |
1023 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
1024 | if (unlikely(!memcg)) | |
1025 | memcg = root_mem_cgroup; | |
1026 | } | |
00d484f3 | 1027 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 1028 | rcu_read_unlock(); |
c0ff4b85 | 1029 | return memcg; |
54595fe2 | 1030 | } |
d46eb14b SB |
1031 | EXPORT_SYMBOL(get_mem_cgroup_from_mm); |
1032 | ||
f745c6f5 SB |
1033 | /** |
1034 | * get_mem_cgroup_from_page: Obtain a reference on given page's memcg. | |
1035 | * @page: page from which memcg should be extracted. | |
1036 | * | |
1037 | * Obtain a reference on page->memcg and returns it if successful. Otherwise | |
1038 | * root_mem_cgroup is returned. | |
1039 | */ | |
1040 | struct mem_cgroup *get_mem_cgroup_from_page(struct page *page) | |
1041 | { | |
1042 | struct mem_cgroup *memcg = page->mem_cgroup; | |
1043 | ||
1044 | if (mem_cgroup_disabled()) | |
1045 | return NULL; | |
1046 | ||
1047 | rcu_read_lock(); | |
8965aa28 SB |
1048 | /* Page should not get uncharged and freed memcg under us. */ |
1049 | if (!memcg || WARN_ON_ONCE(!css_tryget(&memcg->css))) | |
f745c6f5 SB |
1050 | memcg = root_mem_cgroup; |
1051 | rcu_read_unlock(); | |
1052 | return memcg; | |
1053 | } | |
1054 | EXPORT_SYMBOL(get_mem_cgroup_from_page); | |
1055 | ||
37d5985c | 1056 | static __always_inline struct mem_cgroup *active_memcg(void) |
d46eb14b | 1057 | { |
37d5985c RG |
1058 | if (in_interrupt()) |
1059 | return this_cpu_read(int_active_memcg); | |
1060 | else | |
1061 | return current->active_memcg; | |
1062 | } | |
279c3393 | 1063 | |
37d5985c RG |
1064 | static __always_inline struct mem_cgroup *get_active_memcg(void) |
1065 | { | |
1066 | struct mem_cgroup *memcg; | |
d46eb14b | 1067 | |
37d5985c RG |
1068 | rcu_read_lock(); |
1069 | memcg = active_memcg(); | |
1070 | if (memcg) { | |
8965aa28 | 1071 | /* current->active_memcg must hold a ref. */ |
37d5985c | 1072 | if (WARN_ON_ONCE(!css_tryget(&memcg->css))) |
8965aa28 SB |
1073 | memcg = root_mem_cgroup; |
1074 | else | |
d46eb14b | 1075 | memcg = current->active_memcg; |
d46eb14b | 1076 | } |
37d5985c RG |
1077 | rcu_read_unlock(); |
1078 | ||
1079 | return memcg; | |
1080 | } | |
1081 | ||
4127c650 RG |
1082 | static __always_inline bool memcg_kmem_bypass(void) |
1083 | { | |
1084 | /* Allow remote memcg charging from any context. */ | |
1085 | if (unlikely(active_memcg())) | |
1086 | return false; | |
1087 | ||
1088 | /* Memcg to charge can't be determined. */ | |
1089 | if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD)) | |
1090 | return true; | |
1091 | ||
1092 | return false; | |
1093 | } | |
1094 | ||
37d5985c RG |
1095 | /** |
1096 | * If active memcg is set, do not fallback to current->mm->memcg. | |
1097 | */ | |
1098 | static __always_inline struct mem_cgroup *get_mem_cgroup_from_current(void) | |
1099 | { | |
1100 | if (memcg_kmem_bypass()) | |
1101 | return NULL; | |
1102 | ||
1103 | if (unlikely(active_memcg())) | |
1104 | return get_active_memcg(); | |
1105 | ||
d46eb14b SB |
1106 | return get_mem_cgroup_from_mm(current->mm); |
1107 | } | |
54595fe2 | 1108 | |
5660048c JW |
1109 | /** |
1110 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1111 | * @root: hierarchy root | |
1112 | * @prev: previously returned memcg, NULL on first invocation | |
1113 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1114 | * | |
1115 | * Returns references to children of the hierarchy below @root, or | |
1116 | * @root itself, or %NULL after a full round-trip. | |
1117 | * | |
1118 | * Caller must pass the return value in @prev on subsequent | |
1119 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1120 | * to cancel a hierarchy walk before the round-trip is complete. | |
1121 | * | |
05bdc520 ML |
1122 | * Reclaimers can specify a node in @reclaim to divide up the memcgs |
1123 | * in the hierarchy among all concurrent reclaimers operating on the | |
1124 | * same node. | |
5660048c | 1125 | */ |
694fbc0f | 1126 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1127 | struct mem_cgroup *prev, |
694fbc0f | 1128 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1129 | { |
3f649ab7 | 1130 | struct mem_cgroup_reclaim_iter *iter; |
5ac8fb31 | 1131 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 1132 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1133 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1134 | |
694fbc0f AM |
1135 | if (mem_cgroup_disabled()) |
1136 | return NULL; | |
5660048c | 1137 | |
9f3a0d09 JW |
1138 | if (!root) |
1139 | root = root_mem_cgroup; | |
7d74b06f | 1140 | |
9f3a0d09 | 1141 | if (prev && !reclaim) |
5ac8fb31 | 1142 | pos = prev; |
14067bb3 | 1143 | |
542f85f9 | 1144 | rcu_read_lock(); |
5f578161 | 1145 | |
5ac8fb31 | 1146 | if (reclaim) { |
ef8f2327 | 1147 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 1148 | |
ef8f2327 | 1149 | mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id); |
9da83f3f | 1150 | iter = &mz->iter; |
5ac8fb31 JW |
1151 | |
1152 | if (prev && reclaim->generation != iter->generation) | |
1153 | goto out_unlock; | |
1154 | ||
6df38689 | 1155 | while (1) { |
4db0c3c2 | 1156 | pos = READ_ONCE(iter->position); |
6df38689 VD |
1157 | if (!pos || css_tryget(&pos->css)) |
1158 | break; | |
5ac8fb31 | 1159 | /* |
6df38689 VD |
1160 | * css reference reached zero, so iter->position will |
1161 | * be cleared by ->css_released. However, we should not | |
1162 | * rely on this happening soon, because ->css_released | |
1163 | * is called from a work queue, and by busy-waiting we | |
1164 | * might block it. So we clear iter->position right | |
1165 | * away. | |
5ac8fb31 | 1166 | */ |
6df38689 VD |
1167 | (void)cmpxchg(&iter->position, pos, NULL); |
1168 | } | |
5ac8fb31 JW |
1169 | } |
1170 | ||
1171 | if (pos) | |
1172 | css = &pos->css; | |
1173 | ||
1174 | for (;;) { | |
1175 | css = css_next_descendant_pre(css, &root->css); | |
1176 | if (!css) { | |
1177 | /* | |
1178 | * Reclaimers share the hierarchy walk, and a | |
1179 | * new one might jump in right at the end of | |
1180 | * the hierarchy - make sure they see at least | |
1181 | * one group and restart from the beginning. | |
1182 | */ | |
1183 | if (!prev) | |
1184 | continue; | |
1185 | break; | |
527a5ec9 | 1186 | } |
7d74b06f | 1187 | |
5ac8fb31 JW |
1188 | /* |
1189 | * Verify the css and acquire a reference. The root | |
1190 | * is provided by the caller, so we know it's alive | |
1191 | * and kicking, and don't take an extra reference. | |
1192 | */ | |
1193 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 1194 | |
5ac8fb31 JW |
1195 | if (css == &root->css) |
1196 | break; | |
14067bb3 | 1197 | |
0b8f73e1 JW |
1198 | if (css_tryget(css)) |
1199 | break; | |
9f3a0d09 | 1200 | |
5ac8fb31 | 1201 | memcg = NULL; |
9f3a0d09 | 1202 | } |
5ac8fb31 JW |
1203 | |
1204 | if (reclaim) { | |
5ac8fb31 | 1205 | /* |
6df38689 VD |
1206 | * The position could have already been updated by a competing |
1207 | * thread, so check that the value hasn't changed since we read | |
1208 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 1209 | */ |
6df38689 VD |
1210 | (void)cmpxchg(&iter->position, pos, memcg); |
1211 | ||
5ac8fb31 JW |
1212 | if (pos) |
1213 | css_put(&pos->css); | |
1214 | ||
1215 | if (!memcg) | |
1216 | iter->generation++; | |
1217 | else if (!prev) | |
1218 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1219 | } |
5ac8fb31 | 1220 | |
542f85f9 MH |
1221 | out_unlock: |
1222 | rcu_read_unlock(); | |
c40046f3 MH |
1223 | if (prev && prev != root) |
1224 | css_put(&prev->css); | |
1225 | ||
9f3a0d09 | 1226 | return memcg; |
14067bb3 | 1227 | } |
7d74b06f | 1228 | |
5660048c JW |
1229 | /** |
1230 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1231 | * @root: hierarchy root | |
1232 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1233 | */ | |
1234 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1235 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1236 | { |
1237 | if (!root) | |
1238 | root = root_mem_cgroup; | |
1239 | if (prev && prev != root) | |
1240 | css_put(&prev->css); | |
1241 | } | |
7d74b06f | 1242 | |
54a83d6b MC |
1243 | static void __invalidate_reclaim_iterators(struct mem_cgroup *from, |
1244 | struct mem_cgroup *dead_memcg) | |
6df38689 | 1245 | { |
6df38689 | 1246 | struct mem_cgroup_reclaim_iter *iter; |
ef8f2327 MG |
1247 | struct mem_cgroup_per_node *mz; |
1248 | int nid; | |
6df38689 | 1249 | |
54a83d6b MC |
1250 | for_each_node(nid) { |
1251 | mz = mem_cgroup_nodeinfo(from, nid); | |
9da83f3f YS |
1252 | iter = &mz->iter; |
1253 | cmpxchg(&iter->position, dead_memcg, NULL); | |
6df38689 VD |
1254 | } |
1255 | } | |
1256 | ||
54a83d6b MC |
1257 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1258 | { | |
1259 | struct mem_cgroup *memcg = dead_memcg; | |
1260 | struct mem_cgroup *last; | |
1261 | ||
1262 | do { | |
1263 | __invalidate_reclaim_iterators(memcg, dead_memcg); | |
1264 | last = memcg; | |
1265 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1266 | ||
1267 | /* | |
1268 | * When cgruop1 non-hierarchy mode is used, | |
1269 | * parent_mem_cgroup() does not walk all the way up to the | |
1270 | * cgroup root (root_mem_cgroup). So we have to handle | |
1271 | * dead_memcg from cgroup root separately. | |
1272 | */ | |
1273 | if (last != root_mem_cgroup) | |
1274 | __invalidate_reclaim_iterators(root_mem_cgroup, | |
1275 | dead_memcg); | |
1276 | } | |
1277 | ||
7c5f64f8 VD |
1278 | /** |
1279 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1280 | * @memcg: hierarchy root | |
1281 | * @fn: function to call for each task | |
1282 | * @arg: argument passed to @fn | |
1283 | * | |
1284 | * This function iterates over tasks attached to @memcg or to any of its | |
1285 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
1286 | * value, the function breaks the iteration loop and returns the value. | |
1287 | * Otherwise, it will iterate over all tasks and return 0. | |
1288 | * | |
1289 | * This function must not be called for the root memory cgroup. | |
1290 | */ | |
1291 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
1292 | int (*fn)(struct task_struct *, void *), void *arg) | |
1293 | { | |
1294 | struct mem_cgroup *iter; | |
1295 | int ret = 0; | |
1296 | ||
1297 | BUG_ON(memcg == root_mem_cgroup); | |
1298 | ||
1299 | for_each_mem_cgroup_tree(iter, memcg) { | |
1300 | struct css_task_iter it; | |
1301 | struct task_struct *task; | |
1302 | ||
f168a9a5 | 1303 | css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it); |
7c5f64f8 VD |
1304 | while (!ret && (task = css_task_iter_next(&it))) |
1305 | ret = fn(task, arg); | |
1306 | css_task_iter_end(&it); | |
1307 | if (ret) { | |
1308 | mem_cgroup_iter_break(memcg, iter); | |
1309 | break; | |
1310 | } | |
1311 | } | |
1312 | return ret; | |
1313 | } | |
1314 | ||
925b7673 | 1315 | /** |
dfe0e773 | 1316 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1317 | * @page: the page |
f144c390 | 1318 | * @pgdat: pgdat of the page |
dfe0e773 | 1319 | * |
a5eb011a | 1320 | * This function relies on page's memcg being stable - see the |
a0b5b414 | 1321 | * access rules in commit_charge(). |
925b7673 | 1322 | */ |
599d0c95 | 1323 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat) |
08e552c6 | 1324 | { |
ef8f2327 | 1325 | struct mem_cgroup_per_node *mz; |
925b7673 | 1326 | struct mem_cgroup *memcg; |
bea8c150 | 1327 | struct lruvec *lruvec; |
6d12e2d8 | 1328 | |
bea8c150 | 1329 | if (mem_cgroup_disabled()) { |
867e5e1d | 1330 | lruvec = &pgdat->__lruvec; |
bea8c150 HD |
1331 | goto out; |
1332 | } | |
925b7673 | 1333 | |
1306a85a | 1334 | memcg = page->mem_cgroup; |
7512102c | 1335 | /* |
dfe0e773 | 1336 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1337 | * possibly migrated - before they are charged. |
7512102c | 1338 | */ |
29833315 JW |
1339 | if (!memcg) |
1340 | memcg = root_mem_cgroup; | |
7512102c | 1341 | |
ef8f2327 | 1342 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
bea8c150 HD |
1343 | lruvec = &mz->lruvec; |
1344 | out: | |
1345 | /* | |
1346 | * Since a node can be onlined after the mem_cgroup was created, | |
1347 | * we have to be prepared to initialize lruvec->zone here; | |
1348 | * and if offlined then reonlined, we need to reinitialize it. | |
1349 | */ | |
599d0c95 MG |
1350 | if (unlikely(lruvec->pgdat != pgdat)) |
1351 | lruvec->pgdat = pgdat; | |
bea8c150 | 1352 | return lruvec; |
08e552c6 | 1353 | } |
b69408e8 | 1354 | |
925b7673 | 1355 | /** |
fa9add64 HD |
1356 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1357 | * @lruvec: mem_cgroup per zone lru vector | |
1358 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1359 | * @zid: zone id of the accounted pages |
fa9add64 | 1360 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1361 | * |
ca707239 HD |
1362 | * This function must be called under lru_lock, just before a page is added |
1363 | * to or just after a page is removed from an lru list (that ordering being | |
1364 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 1365 | */ |
fa9add64 | 1366 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1367 | int zid, int nr_pages) |
3f58a829 | 1368 | { |
ef8f2327 | 1369 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1370 | unsigned long *lru_size; |
ca707239 | 1371 | long size; |
3f58a829 MK |
1372 | |
1373 | if (mem_cgroup_disabled()) | |
1374 | return; | |
1375 | ||
ef8f2327 | 1376 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1377 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1378 | |
1379 | if (nr_pages < 0) | |
1380 | *lru_size += nr_pages; | |
1381 | ||
1382 | size = *lru_size; | |
b4536f0c MH |
1383 | if (WARN_ONCE(size < 0, |
1384 | "%s(%p, %d, %d): lru_size %ld\n", | |
1385 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1386 | VM_BUG_ON(1); |
1387 | *lru_size = 0; | |
1388 | } | |
1389 | ||
1390 | if (nr_pages > 0) | |
1391 | *lru_size += nr_pages; | |
08e552c6 | 1392 | } |
544122e5 | 1393 | |
19942822 | 1394 | /** |
9d11ea9f | 1395 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1396 | * @memcg: the memory cgroup |
19942822 | 1397 | * |
9d11ea9f | 1398 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1399 | * pages. |
19942822 | 1400 | */ |
c0ff4b85 | 1401 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1402 | { |
3e32cb2e JW |
1403 | unsigned long margin = 0; |
1404 | unsigned long count; | |
1405 | unsigned long limit; | |
9d11ea9f | 1406 | |
3e32cb2e | 1407 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1408 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1409 | if (count < limit) |
1410 | margin = limit - count; | |
1411 | ||
7941d214 | 1412 | if (do_memsw_account()) { |
3e32cb2e | 1413 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1414 | limit = READ_ONCE(memcg->memsw.max); |
1c4448ed | 1415 | if (count < limit) |
3e32cb2e | 1416 | margin = min(margin, limit - count); |
cbedbac3 LR |
1417 | else |
1418 | margin = 0; | |
3e32cb2e JW |
1419 | } |
1420 | ||
1421 | return margin; | |
19942822 JW |
1422 | } |
1423 | ||
32047e2a | 1424 | /* |
bdcbb659 | 1425 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1426 | * |
bdcbb659 QH |
1427 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1428 | * moving cgroups. This is for waiting at high-memory pressure | |
1429 | * caused by "move". | |
32047e2a | 1430 | */ |
c0ff4b85 | 1431 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1432 | { |
2bd9bb20 KH |
1433 | struct mem_cgroup *from; |
1434 | struct mem_cgroup *to; | |
4b534334 | 1435 | bool ret = false; |
2bd9bb20 KH |
1436 | /* |
1437 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1438 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1439 | */ | |
1440 | spin_lock(&mc.lock); | |
1441 | from = mc.from; | |
1442 | to = mc.to; | |
1443 | if (!from) | |
1444 | goto unlock; | |
3e92041d | 1445 | |
2314b42d JW |
1446 | ret = mem_cgroup_is_descendant(from, memcg) || |
1447 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1448 | unlock: |
1449 | spin_unlock(&mc.lock); | |
4b534334 KH |
1450 | return ret; |
1451 | } | |
1452 | ||
c0ff4b85 | 1453 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1454 | { |
1455 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1456 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1457 | DEFINE_WAIT(wait); |
1458 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1459 | /* moving charge context might have finished. */ | |
1460 | if (mc.moving_task) | |
1461 | schedule(); | |
1462 | finish_wait(&mc.waitq, &wait); | |
1463 | return true; | |
1464 | } | |
1465 | } | |
1466 | return false; | |
1467 | } | |
1468 | ||
5f9a4f4a MS |
1469 | struct memory_stat { |
1470 | const char *name; | |
1471 | unsigned int ratio; | |
1472 | unsigned int idx; | |
1473 | }; | |
1474 | ||
1475 | static struct memory_stat memory_stats[] = { | |
1476 | { "anon", PAGE_SIZE, NR_ANON_MAPPED }, | |
1477 | { "file", PAGE_SIZE, NR_FILE_PAGES }, | |
1478 | { "kernel_stack", 1024, NR_KERNEL_STACK_KB }, | |
1479 | { "percpu", 1, MEMCG_PERCPU_B }, | |
1480 | { "sock", PAGE_SIZE, MEMCG_SOCK }, | |
1481 | { "shmem", PAGE_SIZE, NR_SHMEM }, | |
1482 | { "file_mapped", PAGE_SIZE, NR_FILE_MAPPED }, | |
1483 | { "file_dirty", PAGE_SIZE, NR_FILE_DIRTY }, | |
1484 | { "file_writeback", PAGE_SIZE, NR_WRITEBACK }, | |
1485 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
1486 | /* | |
1487 | * The ratio will be initialized in memory_stats_init(). Because | |
1488 | * on some architectures, the macro of HPAGE_PMD_SIZE is not | |
1489 | * constant(e.g. powerpc). | |
1490 | */ | |
1491 | { "anon_thp", 0, NR_ANON_THPS }, | |
b8eddff8 JW |
1492 | { "file_thp", 0, NR_FILE_THPS }, |
1493 | { "shmem_thp", 0, NR_SHMEM_THPS }, | |
5f9a4f4a MS |
1494 | #endif |
1495 | { "inactive_anon", PAGE_SIZE, NR_INACTIVE_ANON }, | |
1496 | { "active_anon", PAGE_SIZE, NR_ACTIVE_ANON }, | |
1497 | { "inactive_file", PAGE_SIZE, NR_INACTIVE_FILE }, | |
1498 | { "active_file", PAGE_SIZE, NR_ACTIVE_FILE }, | |
1499 | { "unevictable", PAGE_SIZE, NR_UNEVICTABLE }, | |
1500 | ||
1501 | /* | |
1502 | * Note: The slab_reclaimable and slab_unreclaimable must be | |
1503 | * together and slab_reclaimable must be in front. | |
1504 | */ | |
1505 | { "slab_reclaimable", 1, NR_SLAB_RECLAIMABLE_B }, | |
1506 | { "slab_unreclaimable", 1, NR_SLAB_UNRECLAIMABLE_B }, | |
1507 | ||
1508 | /* The memory events */ | |
1509 | { "workingset_refault_anon", 1, WORKINGSET_REFAULT_ANON }, | |
1510 | { "workingset_refault_file", 1, WORKINGSET_REFAULT_FILE }, | |
1511 | { "workingset_activate_anon", 1, WORKINGSET_ACTIVATE_ANON }, | |
1512 | { "workingset_activate_file", 1, WORKINGSET_ACTIVATE_FILE }, | |
1513 | { "workingset_restore_anon", 1, WORKINGSET_RESTORE_ANON }, | |
1514 | { "workingset_restore_file", 1, WORKINGSET_RESTORE_FILE }, | |
1515 | { "workingset_nodereclaim", 1, WORKINGSET_NODERECLAIM }, | |
1516 | }; | |
1517 | ||
1518 | static int __init memory_stats_init(void) | |
1519 | { | |
1520 | int i; | |
1521 | ||
1522 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { | |
1523 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
b8eddff8 JW |
1524 | if (memory_stats[i].idx == NR_ANON_THPS || |
1525 | memory_stats[i].idx == NR_FILE_THPS || | |
1526 | memory_stats[i].idx == NR_SHMEM_THPS) | |
5f9a4f4a MS |
1527 | memory_stats[i].ratio = HPAGE_PMD_SIZE; |
1528 | #endif | |
1529 | VM_BUG_ON(!memory_stats[i].ratio); | |
1530 | VM_BUG_ON(memory_stats[i].idx >= MEMCG_NR_STAT); | |
1531 | } | |
1532 | ||
1533 | return 0; | |
1534 | } | |
1535 | pure_initcall(memory_stats_init); | |
1536 | ||
c8713d0b JW |
1537 | static char *memory_stat_format(struct mem_cgroup *memcg) |
1538 | { | |
1539 | struct seq_buf s; | |
1540 | int i; | |
71cd3113 | 1541 | |
c8713d0b JW |
1542 | seq_buf_init(&s, kmalloc(PAGE_SIZE, GFP_KERNEL), PAGE_SIZE); |
1543 | if (!s.buffer) | |
1544 | return NULL; | |
1545 | ||
1546 | /* | |
1547 | * Provide statistics on the state of the memory subsystem as | |
1548 | * well as cumulative event counters that show past behavior. | |
1549 | * | |
1550 | * This list is ordered following a combination of these gradients: | |
1551 | * 1) generic big picture -> specifics and details | |
1552 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
1553 | * | |
1554 | * Current memory state: | |
1555 | */ | |
1556 | ||
5f9a4f4a MS |
1557 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { |
1558 | u64 size; | |
c8713d0b | 1559 | |
5f9a4f4a MS |
1560 | size = memcg_page_state(memcg, memory_stats[i].idx); |
1561 | size *= memory_stats[i].ratio; | |
1562 | seq_buf_printf(&s, "%s %llu\n", memory_stats[i].name, size); | |
c8713d0b | 1563 | |
5f9a4f4a MS |
1564 | if (unlikely(memory_stats[i].idx == NR_SLAB_UNRECLAIMABLE_B)) { |
1565 | size = memcg_page_state(memcg, NR_SLAB_RECLAIMABLE_B) + | |
1566 | memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE_B); | |
1567 | seq_buf_printf(&s, "slab %llu\n", size); | |
1568 | } | |
1569 | } | |
c8713d0b JW |
1570 | |
1571 | /* Accumulated memory events */ | |
1572 | ||
ebc5d83d KK |
1573 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGFAULT), |
1574 | memcg_events(memcg, PGFAULT)); | |
1575 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGMAJFAULT), | |
1576 | memcg_events(memcg, PGMAJFAULT)); | |
ebc5d83d KK |
1577 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGREFILL), |
1578 | memcg_events(memcg, PGREFILL)); | |
c8713d0b JW |
1579 | seq_buf_printf(&s, "pgscan %lu\n", |
1580 | memcg_events(memcg, PGSCAN_KSWAPD) + | |
1581 | memcg_events(memcg, PGSCAN_DIRECT)); | |
1582 | seq_buf_printf(&s, "pgsteal %lu\n", | |
1583 | memcg_events(memcg, PGSTEAL_KSWAPD) + | |
1584 | memcg_events(memcg, PGSTEAL_DIRECT)); | |
ebc5d83d KK |
1585 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGACTIVATE), |
1586 | memcg_events(memcg, PGACTIVATE)); | |
1587 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGDEACTIVATE), | |
1588 | memcg_events(memcg, PGDEACTIVATE)); | |
1589 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREE), | |
1590 | memcg_events(memcg, PGLAZYFREE)); | |
1591 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREED), | |
1592 | memcg_events(memcg, PGLAZYFREED)); | |
c8713d0b JW |
1593 | |
1594 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
ebc5d83d | 1595 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_FAULT_ALLOC), |
c8713d0b | 1596 | memcg_events(memcg, THP_FAULT_ALLOC)); |
ebc5d83d | 1597 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_COLLAPSE_ALLOC), |
c8713d0b JW |
1598 | memcg_events(memcg, THP_COLLAPSE_ALLOC)); |
1599 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ | |
1600 | ||
1601 | /* The above should easily fit into one page */ | |
1602 | WARN_ON_ONCE(seq_buf_has_overflowed(&s)); | |
1603 | ||
1604 | return s.buffer; | |
1605 | } | |
71cd3113 | 1606 | |
58cf188e | 1607 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1608 | /** |
f0c867d9 | 1609 | * mem_cgroup_print_oom_context: Print OOM information relevant to |
1610 | * memory controller. | |
e222432b BS |
1611 | * @memcg: The memory cgroup that went over limit |
1612 | * @p: Task that is going to be killed | |
1613 | * | |
1614 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1615 | * enabled | |
1616 | */ | |
f0c867d9 | 1617 | void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) |
e222432b | 1618 | { |
e222432b BS |
1619 | rcu_read_lock(); |
1620 | ||
f0c867d9 | 1621 | if (memcg) { |
1622 | pr_cont(",oom_memcg="); | |
1623 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1624 | } else | |
1625 | pr_cont(",global_oom"); | |
2415b9f5 | 1626 | if (p) { |
f0c867d9 | 1627 | pr_cont(",task_memcg="); |
2415b9f5 | 1628 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
2415b9f5 | 1629 | } |
e222432b | 1630 | rcu_read_unlock(); |
f0c867d9 | 1631 | } |
1632 | ||
1633 | /** | |
1634 | * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to | |
1635 | * memory controller. | |
1636 | * @memcg: The memory cgroup that went over limit | |
1637 | */ | |
1638 | void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) | |
1639 | { | |
c8713d0b | 1640 | char *buf; |
e222432b | 1641 | |
3e32cb2e JW |
1642 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1643 | K((u64)page_counter_read(&memcg->memory)), | |
15b42562 | 1644 | K((u64)READ_ONCE(memcg->memory.max)), memcg->memory.failcnt); |
c8713d0b JW |
1645 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
1646 | pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1647 | K((u64)page_counter_read(&memcg->swap)), | |
32d087cd | 1648 | K((u64)READ_ONCE(memcg->swap.max)), memcg->swap.failcnt); |
c8713d0b JW |
1649 | else { |
1650 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1651 | K((u64)page_counter_read(&memcg->memsw)), | |
1652 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); | |
1653 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1654 | K((u64)page_counter_read(&memcg->kmem)), | |
1655 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); | |
58cf188e | 1656 | } |
c8713d0b JW |
1657 | |
1658 | pr_info("Memory cgroup stats for "); | |
1659 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1660 | pr_cont(":"); | |
1661 | buf = memory_stat_format(memcg); | |
1662 | if (!buf) | |
1663 | return; | |
1664 | pr_info("%s", buf); | |
1665 | kfree(buf); | |
e222432b BS |
1666 | } |
1667 | ||
a63d83f4 DR |
1668 | /* |
1669 | * Return the memory (and swap, if configured) limit for a memcg. | |
1670 | */ | |
bbec2e15 | 1671 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1672 | { |
8d387a5f WL |
1673 | unsigned long max = READ_ONCE(memcg->memory.max); |
1674 | ||
1675 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) { | |
1676 | if (mem_cgroup_swappiness(memcg)) | |
1677 | max += min(READ_ONCE(memcg->swap.max), | |
1678 | (unsigned long)total_swap_pages); | |
1679 | } else { /* v1 */ | |
1680 | if (mem_cgroup_swappiness(memcg)) { | |
1681 | /* Calculate swap excess capacity from memsw limit */ | |
1682 | unsigned long swap = READ_ONCE(memcg->memsw.max) - max; | |
1683 | ||
1684 | max += min(swap, (unsigned long)total_swap_pages); | |
1685 | } | |
9a5a8f19 | 1686 | } |
bbec2e15 | 1687 | return max; |
a63d83f4 DR |
1688 | } |
1689 | ||
9783aa99 CD |
1690 | unsigned long mem_cgroup_size(struct mem_cgroup *memcg) |
1691 | { | |
1692 | return page_counter_read(&memcg->memory); | |
1693 | } | |
1694 | ||
b6e6edcf | 1695 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1696 | int order) |
9cbb78bb | 1697 | { |
6e0fc46d DR |
1698 | struct oom_control oc = { |
1699 | .zonelist = NULL, | |
1700 | .nodemask = NULL, | |
2a966b77 | 1701 | .memcg = memcg, |
6e0fc46d DR |
1702 | .gfp_mask = gfp_mask, |
1703 | .order = order, | |
6e0fc46d | 1704 | }; |
1378b37d | 1705 | bool ret = true; |
9cbb78bb | 1706 | |
7775face TH |
1707 | if (mutex_lock_killable(&oom_lock)) |
1708 | return true; | |
1378b37d YS |
1709 | |
1710 | if (mem_cgroup_margin(memcg) >= (1 << order)) | |
1711 | goto unlock; | |
1712 | ||
7775face TH |
1713 | /* |
1714 | * A few threads which were not waiting at mutex_lock_killable() can | |
1715 | * fail to bail out. Therefore, check again after holding oom_lock. | |
1716 | */ | |
1717 | ret = should_force_charge() || out_of_memory(&oc); | |
1378b37d YS |
1718 | |
1719 | unlock: | |
dc56401f | 1720 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1721 | return ret; |
9cbb78bb DR |
1722 | } |
1723 | ||
0608f43d | 1724 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1725 | pg_data_t *pgdat, |
0608f43d AM |
1726 | gfp_t gfp_mask, |
1727 | unsigned long *total_scanned) | |
1728 | { | |
1729 | struct mem_cgroup *victim = NULL; | |
1730 | int total = 0; | |
1731 | int loop = 0; | |
1732 | unsigned long excess; | |
1733 | unsigned long nr_scanned; | |
1734 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1735 | .pgdat = pgdat, |
0608f43d AM |
1736 | }; |
1737 | ||
3e32cb2e | 1738 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1739 | |
1740 | while (1) { | |
1741 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1742 | if (!victim) { | |
1743 | loop++; | |
1744 | if (loop >= 2) { | |
1745 | /* | |
1746 | * If we have not been able to reclaim | |
1747 | * anything, it might because there are | |
1748 | * no reclaimable pages under this hierarchy | |
1749 | */ | |
1750 | if (!total) | |
1751 | break; | |
1752 | /* | |
1753 | * We want to do more targeted reclaim. | |
1754 | * excess >> 2 is not to excessive so as to | |
1755 | * reclaim too much, nor too less that we keep | |
1756 | * coming back to reclaim from this cgroup | |
1757 | */ | |
1758 | if (total >= (excess >> 2) || | |
1759 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1760 | break; | |
1761 | } | |
1762 | continue; | |
1763 | } | |
a9dd0a83 | 1764 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1765 | pgdat, &nr_scanned); |
0608f43d | 1766 | *total_scanned += nr_scanned; |
3e32cb2e | 1767 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1768 | break; |
6d61ef40 | 1769 | } |
0608f43d AM |
1770 | mem_cgroup_iter_break(root_memcg, victim); |
1771 | return total; | |
6d61ef40 BS |
1772 | } |
1773 | ||
0056f4e6 JW |
1774 | #ifdef CONFIG_LOCKDEP |
1775 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1776 | .name = "memcg_oom_lock", | |
1777 | }; | |
1778 | #endif | |
1779 | ||
fb2a6fc5 JW |
1780 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1781 | ||
867578cb KH |
1782 | /* |
1783 | * Check OOM-Killer is already running under our hierarchy. | |
1784 | * If someone is running, return false. | |
1785 | */ | |
fb2a6fc5 | 1786 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1787 | { |
79dfdacc | 1788 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1789 | |
fb2a6fc5 JW |
1790 | spin_lock(&memcg_oom_lock); |
1791 | ||
9f3a0d09 | 1792 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1793 | if (iter->oom_lock) { |
79dfdacc MH |
1794 | /* |
1795 | * this subtree of our hierarchy is already locked | |
1796 | * so we cannot give a lock. | |
1797 | */ | |
79dfdacc | 1798 | failed = iter; |
9f3a0d09 JW |
1799 | mem_cgroup_iter_break(memcg, iter); |
1800 | break; | |
23751be0 JW |
1801 | } else |
1802 | iter->oom_lock = true; | |
7d74b06f | 1803 | } |
867578cb | 1804 | |
fb2a6fc5 JW |
1805 | if (failed) { |
1806 | /* | |
1807 | * OK, we failed to lock the whole subtree so we have | |
1808 | * to clean up what we set up to the failing subtree | |
1809 | */ | |
1810 | for_each_mem_cgroup_tree(iter, memcg) { | |
1811 | if (iter == failed) { | |
1812 | mem_cgroup_iter_break(memcg, iter); | |
1813 | break; | |
1814 | } | |
1815 | iter->oom_lock = false; | |
79dfdacc | 1816 | } |
0056f4e6 JW |
1817 | } else |
1818 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1819 | |
1820 | spin_unlock(&memcg_oom_lock); | |
1821 | ||
1822 | return !failed; | |
a636b327 | 1823 | } |
0b7f569e | 1824 | |
fb2a6fc5 | 1825 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1826 | { |
7d74b06f KH |
1827 | struct mem_cgroup *iter; |
1828 | ||
fb2a6fc5 | 1829 | spin_lock(&memcg_oom_lock); |
5facae4f | 1830 | mutex_release(&memcg_oom_lock_dep_map, _RET_IP_); |
c0ff4b85 | 1831 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1832 | iter->oom_lock = false; |
fb2a6fc5 | 1833 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1834 | } |
1835 | ||
c0ff4b85 | 1836 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1837 | { |
1838 | struct mem_cgroup *iter; | |
1839 | ||
c2b42d3c | 1840 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1841 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1842 | iter->under_oom++; |
1843 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1844 | } |
1845 | ||
c0ff4b85 | 1846 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1847 | { |
1848 | struct mem_cgroup *iter; | |
1849 | ||
867578cb | 1850 | /* |
7a52d4d8 ML |
1851 | * Be careful about under_oom underflows becase a child memcg |
1852 | * could have been added after mem_cgroup_mark_under_oom. | |
867578cb | 1853 | */ |
c2b42d3c | 1854 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1855 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1856 | if (iter->under_oom > 0) |
1857 | iter->under_oom--; | |
1858 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1859 | } |
1860 | ||
867578cb KH |
1861 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1862 | ||
dc98df5a | 1863 | struct oom_wait_info { |
d79154bb | 1864 | struct mem_cgroup *memcg; |
ac6424b9 | 1865 | wait_queue_entry_t wait; |
dc98df5a KH |
1866 | }; |
1867 | ||
ac6424b9 | 1868 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1869 | unsigned mode, int sync, void *arg) |
1870 | { | |
d79154bb HD |
1871 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1872 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1873 | struct oom_wait_info *oom_wait_info; |
1874 | ||
1875 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1876 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1877 | |
2314b42d JW |
1878 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1879 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1880 | return 0; |
dc98df5a KH |
1881 | return autoremove_wake_function(wait, mode, sync, arg); |
1882 | } | |
1883 | ||
c0ff4b85 | 1884 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1885 | { |
c2b42d3c TH |
1886 | /* |
1887 | * For the following lockless ->under_oom test, the only required | |
1888 | * guarantee is that it must see the state asserted by an OOM when | |
1889 | * this function is called as a result of userland actions | |
1890 | * triggered by the notification of the OOM. This is trivially | |
1891 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1892 | * triggering notification. | |
1893 | */ | |
1894 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1895 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1896 | } |
1897 | ||
29ef680a MH |
1898 | enum oom_status { |
1899 | OOM_SUCCESS, | |
1900 | OOM_FAILED, | |
1901 | OOM_ASYNC, | |
1902 | OOM_SKIPPED | |
1903 | }; | |
1904 | ||
1905 | static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 1906 | { |
7056d3a3 MH |
1907 | enum oom_status ret; |
1908 | bool locked; | |
1909 | ||
29ef680a MH |
1910 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
1911 | return OOM_SKIPPED; | |
1912 | ||
7a1adfdd RG |
1913 | memcg_memory_event(memcg, MEMCG_OOM); |
1914 | ||
867578cb | 1915 | /* |
49426420 JW |
1916 | * We are in the middle of the charge context here, so we |
1917 | * don't want to block when potentially sitting on a callstack | |
1918 | * that holds all kinds of filesystem and mm locks. | |
1919 | * | |
29ef680a MH |
1920 | * cgroup1 allows disabling the OOM killer and waiting for outside |
1921 | * handling until the charge can succeed; remember the context and put | |
1922 | * the task to sleep at the end of the page fault when all locks are | |
1923 | * released. | |
49426420 | 1924 | * |
29ef680a MH |
1925 | * On the other hand, in-kernel OOM killer allows for an async victim |
1926 | * memory reclaim (oom_reaper) and that means that we are not solely | |
1927 | * relying on the oom victim to make a forward progress and we can | |
1928 | * invoke the oom killer here. | |
1929 | * | |
1930 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
1931 | * victim and then we have to bail out from the charge path. | |
867578cb | 1932 | */ |
29ef680a MH |
1933 | if (memcg->oom_kill_disable) { |
1934 | if (!current->in_user_fault) | |
1935 | return OOM_SKIPPED; | |
1936 | css_get(&memcg->css); | |
1937 | current->memcg_in_oom = memcg; | |
1938 | current->memcg_oom_gfp_mask = mask; | |
1939 | current->memcg_oom_order = order; | |
1940 | ||
1941 | return OOM_ASYNC; | |
1942 | } | |
1943 | ||
7056d3a3 MH |
1944 | mem_cgroup_mark_under_oom(memcg); |
1945 | ||
1946 | locked = mem_cgroup_oom_trylock(memcg); | |
1947 | ||
1948 | if (locked) | |
1949 | mem_cgroup_oom_notify(memcg); | |
1950 | ||
1951 | mem_cgroup_unmark_under_oom(memcg); | |
29ef680a | 1952 | if (mem_cgroup_out_of_memory(memcg, mask, order)) |
7056d3a3 MH |
1953 | ret = OOM_SUCCESS; |
1954 | else | |
1955 | ret = OOM_FAILED; | |
1956 | ||
1957 | if (locked) | |
1958 | mem_cgroup_oom_unlock(memcg); | |
29ef680a | 1959 | |
7056d3a3 | 1960 | return ret; |
3812c8c8 JW |
1961 | } |
1962 | ||
1963 | /** | |
1964 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1965 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1966 | * |
49426420 JW |
1967 | * This has to be called at the end of a page fault if the memcg OOM |
1968 | * handler was enabled. | |
3812c8c8 | 1969 | * |
49426420 | 1970 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1971 | * sleep on a waitqueue until the userspace task resolves the |
1972 | * situation. Sleeping directly in the charge context with all kinds | |
1973 | * of locks held is not a good idea, instead we remember an OOM state | |
1974 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1975 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1976 | * |
1977 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1978 | * completed, %false otherwise. |
3812c8c8 | 1979 | */ |
49426420 | 1980 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1981 | { |
626ebc41 | 1982 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1983 | struct oom_wait_info owait; |
49426420 | 1984 | bool locked; |
3812c8c8 JW |
1985 | |
1986 | /* OOM is global, do not handle */ | |
3812c8c8 | 1987 | if (!memcg) |
49426420 | 1988 | return false; |
3812c8c8 | 1989 | |
7c5f64f8 | 1990 | if (!handle) |
49426420 | 1991 | goto cleanup; |
3812c8c8 JW |
1992 | |
1993 | owait.memcg = memcg; | |
1994 | owait.wait.flags = 0; | |
1995 | owait.wait.func = memcg_oom_wake_function; | |
1996 | owait.wait.private = current; | |
2055da97 | 1997 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 1998 | |
3812c8c8 | 1999 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
2000 | mem_cgroup_mark_under_oom(memcg); |
2001 | ||
2002 | locked = mem_cgroup_oom_trylock(memcg); | |
2003 | ||
2004 | if (locked) | |
2005 | mem_cgroup_oom_notify(memcg); | |
2006 | ||
2007 | if (locked && !memcg->oom_kill_disable) { | |
2008 | mem_cgroup_unmark_under_oom(memcg); | |
2009 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
2010 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
2011 | current->memcg_oom_order); | |
49426420 | 2012 | } else { |
3812c8c8 | 2013 | schedule(); |
49426420 JW |
2014 | mem_cgroup_unmark_under_oom(memcg); |
2015 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2016 | } | |
2017 | ||
2018 | if (locked) { | |
fb2a6fc5 JW |
2019 | mem_cgroup_oom_unlock(memcg); |
2020 | /* | |
2021 | * There is no guarantee that an OOM-lock contender | |
2022 | * sees the wakeups triggered by the OOM kill | |
2023 | * uncharges. Wake any sleepers explicitely. | |
2024 | */ | |
2025 | memcg_oom_recover(memcg); | |
2026 | } | |
49426420 | 2027 | cleanup: |
626ebc41 | 2028 | current->memcg_in_oom = NULL; |
3812c8c8 | 2029 | css_put(&memcg->css); |
867578cb | 2030 | return true; |
0b7f569e KH |
2031 | } |
2032 | ||
3d8b38eb RG |
2033 | /** |
2034 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
2035 | * @victim: task to be killed by the OOM killer | |
2036 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
2037 | * | |
2038 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
2039 | * by killing all belonging OOM-killable tasks. | |
2040 | * | |
2041 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
2042 | */ | |
2043 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
2044 | struct mem_cgroup *oom_domain) | |
2045 | { | |
2046 | struct mem_cgroup *oom_group = NULL; | |
2047 | struct mem_cgroup *memcg; | |
2048 | ||
2049 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
2050 | return NULL; | |
2051 | ||
2052 | if (!oom_domain) | |
2053 | oom_domain = root_mem_cgroup; | |
2054 | ||
2055 | rcu_read_lock(); | |
2056 | ||
2057 | memcg = mem_cgroup_from_task(victim); | |
2058 | if (memcg == root_mem_cgroup) | |
2059 | goto out; | |
2060 | ||
48fe267c RG |
2061 | /* |
2062 | * If the victim task has been asynchronously moved to a different | |
2063 | * memory cgroup, we might end up killing tasks outside oom_domain. | |
2064 | * In this case it's better to ignore memory.group.oom. | |
2065 | */ | |
2066 | if (unlikely(!mem_cgroup_is_descendant(memcg, oom_domain))) | |
2067 | goto out; | |
2068 | ||
3d8b38eb RG |
2069 | /* |
2070 | * Traverse the memory cgroup hierarchy from the victim task's | |
2071 | * cgroup up to the OOMing cgroup (or root) to find the | |
2072 | * highest-level memory cgroup with oom.group set. | |
2073 | */ | |
2074 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
2075 | if (memcg->oom_group) | |
2076 | oom_group = memcg; | |
2077 | ||
2078 | if (memcg == oom_domain) | |
2079 | break; | |
2080 | } | |
2081 | ||
2082 | if (oom_group) | |
2083 | css_get(&oom_group->css); | |
2084 | out: | |
2085 | rcu_read_unlock(); | |
2086 | ||
2087 | return oom_group; | |
2088 | } | |
2089 | ||
2090 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
2091 | { | |
2092 | pr_info("Tasks in "); | |
2093 | pr_cont_cgroup_path(memcg->css.cgroup); | |
2094 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
2095 | } | |
2096 | ||
d7365e78 | 2097 | /** |
81f8c3a4 JW |
2098 | * lock_page_memcg - lock a page->mem_cgroup binding |
2099 | * @page: the page | |
32047e2a | 2100 | * |
81f8c3a4 | 2101 | * This function protects unlocked LRU pages from being moved to |
739f79fc JW |
2102 | * another cgroup. |
2103 | * | |
2104 | * It ensures lifetime of the returned memcg. Caller is responsible | |
2105 | * for the lifetime of the page; __unlock_page_memcg() is available | |
2106 | * when @page might get freed inside the locked section. | |
d69b042f | 2107 | */ |
739f79fc | 2108 | struct mem_cgroup *lock_page_memcg(struct page *page) |
89c06bd5 | 2109 | { |
9da7b521 | 2110 | struct page *head = compound_head(page); /* rmap on tail pages */ |
89c06bd5 | 2111 | struct mem_cgroup *memcg; |
6de22619 | 2112 | unsigned long flags; |
89c06bd5 | 2113 | |
6de22619 JW |
2114 | /* |
2115 | * The RCU lock is held throughout the transaction. The fast | |
2116 | * path can get away without acquiring the memcg->move_lock | |
2117 | * because page moving starts with an RCU grace period. | |
739f79fc JW |
2118 | * |
2119 | * The RCU lock also protects the memcg from being freed when | |
2120 | * the page state that is going to change is the only thing | |
2121 | * preventing the page itself from being freed. E.g. writeback | |
2122 | * doesn't hold a page reference and relies on PG_writeback to | |
2123 | * keep off truncation, migration and so forth. | |
2124 | */ | |
d7365e78 JW |
2125 | rcu_read_lock(); |
2126 | ||
2127 | if (mem_cgroup_disabled()) | |
739f79fc | 2128 | return NULL; |
89c06bd5 | 2129 | again: |
9da7b521 | 2130 | memcg = head->mem_cgroup; |
29833315 | 2131 | if (unlikely(!memcg)) |
739f79fc | 2132 | return NULL; |
d7365e78 | 2133 | |
bdcbb659 | 2134 | if (atomic_read(&memcg->moving_account) <= 0) |
739f79fc | 2135 | return memcg; |
89c06bd5 | 2136 | |
6de22619 | 2137 | spin_lock_irqsave(&memcg->move_lock, flags); |
9da7b521 | 2138 | if (memcg != head->mem_cgroup) { |
6de22619 | 2139 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
2140 | goto again; |
2141 | } | |
6de22619 JW |
2142 | |
2143 | /* | |
2144 | * When charge migration first begins, we can have locked and | |
2145 | * unlocked page stat updates happening concurrently. Track | |
81f8c3a4 | 2146 | * the task who has the lock for unlock_page_memcg(). |
6de22619 JW |
2147 | */ |
2148 | memcg->move_lock_task = current; | |
2149 | memcg->move_lock_flags = flags; | |
d7365e78 | 2150 | |
739f79fc | 2151 | return memcg; |
89c06bd5 | 2152 | } |
81f8c3a4 | 2153 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 2154 | |
d7365e78 | 2155 | /** |
739f79fc JW |
2156 | * __unlock_page_memcg - unlock and unpin a memcg |
2157 | * @memcg: the memcg | |
2158 | * | |
2159 | * Unlock and unpin a memcg returned by lock_page_memcg(). | |
d7365e78 | 2160 | */ |
739f79fc | 2161 | void __unlock_page_memcg(struct mem_cgroup *memcg) |
89c06bd5 | 2162 | { |
6de22619 JW |
2163 | if (memcg && memcg->move_lock_task == current) { |
2164 | unsigned long flags = memcg->move_lock_flags; | |
2165 | ||
2166 | memcg->move_lock_task = NULL; | |
2167 | memcg->move_lock_flags = 0; | |
2168 | ||
2169 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2170 | } | |
89c06bd5 | 2171 | |
d7365e78 | 2172 | rcu_read_unlock(); |
89c06bd5 | 2173 | } |
739f79fc JW |
2174 | |
2175 | /** | |
2176 | * unlock_page_memcg - unlock a page->mem_cgroup binding | |
2177 | * @page: the page | |
2178 | */ | |
2179 | void unlock_page_memcg(struct page *page) | |
2180 | { | |
9da7b521 JW |
2181 | struct page *head = compound_head(page); |
2182 | ||
2183 | __unlock_page_memcg(head->mem_cgroup); | |
739f79fc | 2184 | } |
81f8c3a4 | 2185 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 2186 | |
cdec2e42 KH |
2187 | struct memcg_stock_pcp { |
2188 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2189 | unsigned int nr_pages; |
bf4f0599 RG |
2190 | |
2191 | #ifdef CONFIG_MEMCG_KMEM | |
2192 | struct obj_cgroup *cached_objcg; | |
2193 | unsigned int nr_bytes; | |
2194 | #endif | |
2195 | ||
cdec2e42 | 2196 | struct work_struct work; |
26fe6168 | 2197 | unsigned long flags; |
a0db00fc | 2198 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2199 | }; |
2200 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2201 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2202 | |
bf4f0599 RG |
2203 | #ifdef CONFIG_MEMCG_KMEM |
2204 | static void drain_obj_stock(struct memcg_stock_pcp *stock); | |
2205 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2206 | struct mem_cgroup *root_memcg); | |
2207 | ||
2208 | #else | |
2209 | static inline void drain_obj_stock(struct memcg_stock_pcp *stock) | |
2210 | { | |
2211 | } | |
2212 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2213 | struct mem_cgroup *root_memcg) | |
2214 | { | |
2215 | return false; | |
2216 | } | |
2217 | #endif | |
2218 | ||
a0956d54 SS |
2219 | /** |
2220 | * consume_stock: Try to consume stocked charge on this cpu. | |
2221 | * @memcg: memcg to consume from. | |
2222 | * @nr_pages: how many pages to charge. | |
2223 | * | |
2224 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2225 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2226 | * service an allocation will refill the stock. | |
2227 | * | |
2228 | * returns true if successful, false otherwise. | |
cdec2e42 | 2229 | */ |
a0956d54 | 2230 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2231 | { |
2232 | struct memcg_stock_pcp *stock; | |
db2ba40c | 2233 | unsigned long flags; |
3e32cb2e | 2234 | bool ret = false; |
cdec2e42 | 2235 | |
a983b5eb | 2236 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 2237 | return ret; |
a0956d54 | 2238 | |
db2ba40c JW |
2239 | local_irq_save(flags); |
2240 | ||
2241 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 2242 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2243 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2244 | ret = true; |
2245 | } | |
db2ba40c JW |
2246 | |
2247 | local_irq_restore(flags); | |
2248 | ||
cdec2e42 KH |
2249 | return ret; |
2250 | } | |
2251 | ||
2252 | /* | |
3e32cb2e | 2253 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2254 | */ |
2255 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2256 | { | |
2257 | struct mem_cgroup *old = stock->cached; | |
2258 | ||
1a3e1f40 JW |
2259 | if (!old) |
2260 | return; | |
2261 | ||
11c9ea4e | 2262 | if (stock->nr_pages) { |
3e32cb2e | 2263 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 2264 | if (do_memsw_account()) |
3e32cb2e | 2265 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
11c9ea4e | 2266 | stock->nr_pages = 0; |
cdec2e42 | 2267 | } |
1a3e1f40 JW |
2268 | |
2269 | css_put(&old->css); | |
cdec2e42 | 2270 | stock->cached = NULL; |
cdec2e42 KH |
2271 | } |
2272 | ||
cdec2e42 KH |
2273 | static void drain_local_stock(struct work_struct *dummy) |
2274 | { | |
db2ba40c JW |
2275 | struct memcg_stock_pcp *stock; |
2276 | unsigned long flags; | |
2277 | ||
72f0184c MH |
2278 | /* |
2279 | * The only protection from memory hotplug vs. drain_stock races is | |
2280 | * that we always operate on local CPU stock here with IRQ disabled | |
2281 | */ | |
db2ba40c JW |
2282 | local_irq_save(flags); |
2283 | ||
2284 | stock = this_cpu_ptr(&memcg_stock); | |
bf4f0599 | 2285 | drain_obj_stock(stock); |
cdec2e42 | 2286 | drain_stock(stock); |
26fe6168 | 2287 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c JW |
2288 | |
2289 | local_irq_restore(flags); | |
cdec2e42 KH |
2290 | } |
2291 | ||
2292 | /* | |
3e32cb2e | 2293 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2294 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2295 | */ |
c0ff4b85 | 2296 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2297 | { |
db2ba40c JW |
2298 | struct memcg_stock_pcp *stock; |
2299 | unsigned long flags; | |
2300 | ||
2301 | local_irq_save(flags); | |
cdec2e42 | 2302 | |
db2ba40c | 2303 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 2304 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2305 | drain_stock(stock); |
1a3e1f40 | 2306 | css_get(&memcg->css); |
c0ff4b85 | 2307 | stock->cached = memcg; |
cdec2e42 | 2308 | } |
11c9ea4e | 2309 | stock->nr_pages += nr_pages; |
db2ba40c | 2310 | |
a983b5eb | 2311 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 RG |
2312 | drain_stock(stock); |
2313 | ||
db2ba40c | 2314 | local_irq_restore(flags); |
cdec2e42 KH |
2315 | } |
2316 | ||
2317 | /* | |
c0ff4b85 | 2318 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2319 | * of the hierarchy under it. |
cdec2e42 | 2320 | */ |
6d3d6aa2 | 2321 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2322 | { |
26fe6168 | 2323 | int cpu, curcpu; |
d38144b7 | 2324 | |
6d3d6aa2 JW |
2325 | /* If someone's already draining, avoid adding running more workers. */ |
2326 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2327 | return; | |
72f0184c MH |
2328 | /* |
2329 | * Notify other cpus that system-wide "drain" is running | |
2330 | * We do not care about races with the cpu hotplug because cpu down | |
2331 | * as well as workers from this path always operate on the local | |
2332 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2333 | */ | |
5af12d0e | 2334 | curcpu = get_cpu(); |
cdec2e42 KH |
2335 | for_each_online_cpu(cpu) { |
2336 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2337 | struct mem_cgroup *memcg; |
e1a366be | 2338 | bool flush = false; |
26fe6168 | 2339 | |
e1a366be | 2340 | rcu_read_lock(); |
c0ff4b85 | 2341 | memcg = stock->cached; |
e1a366be RG |
2342 | if (memcg && stock->nr_pages && |
2343 | mem_cgroup_is_descendant(memcg, root_memcg)) | |
2344 | flush = true; | |
bf4f0599 RG |
2345 | if (obj_stock_flush_required(stock, root_memcg)) |
2346 | flush = true; | |
e1a366be RG |
2347 | rcu_read_unlock(); |
2348 | ||
2349 | if (flush && | |
2350 | !test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { | |
d1a05b69 MH |
2351 | if (cpu == curcpu) |
2352 | drain_local_stock(&stock->work); | |
2353 | else | |
2354 | schedule_work_on(cpu, &stock->work); | |
2355 | } | |
cdec2e42 | 2356 | } |
5af12d0e | 2357 | put_cpu(); |
9f50fad6 | 2358 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2359 | } |
2360 | ||
308167fc | 2361 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
cdec2e42 | 2362 | { |
cdec2e42 | 2363 | struct memcg_stock_pcp *stock; |
42a30035 | 2364 | struct mem_cgroup *memcg, *mi; |
cdec2e42 | 2365 | |
cdec2e42 KH |
2366 | stock = &per_cpu(memcg_stock, cpu); |
2367 | drain_stock(stock); | |
a983b5eb JW |
2368 | |
2369 | for_each_mem_cgroup(memcg) { | |
2370 | int i; | |
2371 | ||
2372 | for (i = 0; i < MEMCG_NR_STAT; i++) { | |
2373 | int nid; | |
2374 | long x; | |
2375 | ||
871789d4 | 2376 | x = this_cpu_xchg(memcg->vmstats_percpu->stat[i], 0); |
815744d7 | 2377 | if (x) |
42a30035 JW |
2378 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2379 | atomic_long_add(x, &memcg->vmstats[i]); | |
a983b5eb JW |
2380 | |
2381 | if (i >= NR_VM_NODE_STAT_ITEMS) | |
2382 | continue; | |
2383 | ||
2384 | for_each_node(nid) { | |
2385 | struct mem_cgroup_per_node *pn; | |
2386 | ||
2387 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
2388 | x = this_cpu_xchg(pn->lruvec_stat_cpu->count[i], 0); | |
815744d7 | 2389 | if (x) |
42a30035 JW |
2390 | do { |
2391 | atomic_long_add(x, &pn->lruvec_stat[i]); | |
2392 | } while ((pn = parent_nodeinfo(pn, nid))); | |
a983b5eb JW |
2393 | } |
2394 | } | |
2395 | ||
e27be240 | 2396 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { |
a983b5eb JW |
2397 | long x; |
2398 | ||
871789d4 | 2399 | x = this_cpu_xchg(memcg->vmstats_percpu->events[i], 0); |
815744d7 | 2400 | if (x) |
42a30035 JW |
2401 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2402 | atomic_long_add(x, &memcg->vmevents[i]); | |
a983b5eb JW |
2403 | } |
2404 | } | |
2405 | ||
308167fc | 2406 | return 0; |
cdec2e42 KH |
2407 | } |
2408 | ||
b3ff9291 CD |
2409 | static unsigned long reclaim_high(struct mem_cgroup *memcg, |
2410 | unsigned int nr_pages, | |
2411 | gfp_t gfp_mask) | |
f7e1cb6e | 2412 | { |
b3ff9291 CD |
2413 | unsigned long nr_reclaimed = 0; |
2414 | ||
f7e1cb6e | 2415 | do { |
e22c6ed9 JW |
2416 | unsigned long pflags; |
2417 | ||
d1663a90 JK |
2418 | if (page_counter_read(&memcg->memory) <= |
2419 | READ_ONCE(memcg->memory.high)) | |
f7e1cb6e | 2420 | continue; |
e22c6ed9 | 2421 | |
e27be240 | 2422 | memcg_memory_event(memcg, MEMCG_HIGH); |
e22c6ed9 JW |
2423 | |
2424 | psi_memstall_enter(&pflags); | |
b3ff9291 CD |
2425 | nr_reclaimed += try_to_free_mem_cgroup_pages(memcg, nr_pages, |
2426 | gfp_mask, true); | |
e22c6ed9 | 2427 | psi_memstall_leave(&pflags); |
4bf17307 CD |
2428 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2429 | !mem_cgroup_is_root(memcg)); | |
b3ff9291 CD |
2430 | |
2431 | return nr_reclaimed; | |
f7e1cb6e JW |
2432 | } |
2433 | ||
2434 | static void high_work_func(struct work_struct *work) | |
2435 | { | |
2436 | struct mem_cgroup *memcg; | |
2437 | ||
2438 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2439 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2440 | } |
2441 | ||
0e4b01df CD |
2442 | /* |
2443 | * Clamp the maximum sleep time per allocation batch to 2 seconds. This is | |
2444 | * enough to still cause a significant slowdown in most cases, while still | |
2445 | * allowing diagnostics and tracing to proceed without becoming stuck. | |
2446 | */ | |
2447 | #define MEMCG_MAX_HIGH_DELAY_JIFFIES (2UL*HZ) | |
2448 | ||
2449 | /* | |
2450 | * When calculating the delay, we use these either side of the exponentiation to | |
2451 | * maintain precision and scale to a reasonable number of jiffies (see the table | |
2452 | * below. | |
2453 | * | |
2454 | * - MEMCG_DELAY_PRECISION_SHIFT: Extra precision bits while translating the | |
2455 | * overage ratio to a delay. | |
ac5ddd0f | 2456 | * - MEMCG_DELAY_SCALING_SHIFT: The number of bits to scale down the |
0e4b01df CD |
2457 | * proposed penalty in order to reduce to a reasonable number of jiffies, and |
2458 | * to produce a reasonable delay curve. | |
2459 | * | |
2460 | * MEMCG_DELAY_SCALING_SHIFT just happens to be a number that produces a | |
2461 | * reasonable delay curve compared to precision-adjusted overage, not | |
2462 | * penalising heavily at first, but still making sure that growth beyond the | |
2463 | * limit penalises misbehaviour cgroups by slowing them down exponentially. For | |
2464 | * example, with a high of 100 megabytes: | |
2465 | * | |
2466 | * +-------+------------------------+ | |
2467 | * | usage | time to allocate in ms | | |
2468 | * +-------+------------------------+ | |
2469 | * | 100M | 0 | | |
2470 | * | 101M | 6 | | |
2471 | * | 102M | 25 | | |
2472 | * | 103M | 57 | | |
2473 | * | 104M | 102 | | |
2474 | * | 105M | 159 | | |
2475 | * | 106M | 230 | | |
2476 | * | 107M | 313 | | |
2477 | * | 108M | 409 | | |
2478 | * | 109M | 518 | | |
2479 | * | 110M | 639 | | |
2480 | * | 111M | 774 | | |
2481 | * | 112M | 921 | | |
2482 | * | 113M | 1081 | | |
2483 | * | 114M | 1254 | | |
2484 | * | 115M | 1439 | | |
2485 | * | 116M | 1638 | | |
2486 | * | 117M | 1849 | | |
2487 | * | 118M | 2000 | | |
2488 | * | 119M | 2000 | | |
2489 | * | 120M | 2000 | | |
2490 | * +-------+------------------------+ | |
2491 | */ | |
2492 | #define MEMCG_DELAY_PRECISION_SHIFT 20 | |
2493 | #define MEMCG_DELAY_SCALING_SHIFT 14 | |
2494 | ||
8a5dbc65 | 2495 | static u64 calculate_overage(unsigned long usage, unsigned long high) |
b23afb93 | 2496 | { |
8a5dbc65 | 2497 | u64 overage; |
b23afb93 | 2498 | |
8a5dbc65 JK |
2499 | if (usage <= high) |
2500 | return 0; | |
e26733e0 | 2501 | |
8a5dbc65 JK |
2502 | /* |
2503 | * Prevent division by 0 in overage calculation by acting as if | |
2504 | * it was a threshold of 1 page | |
2505 | */ | |
2506 | high = max(high, 1UL); | |
9b8b1754 | 2507 | |
8a5dbc65 JK |
2508 | overage = usage - high; |
2509 | overage <<= MEMCG_DELAY_PRECISION_SHIFT; | |
2510 | return div64_u64(overage, high); | |
2511 | } | |
e26733e0 | 2512 | |
8a5dbc65 JK |
2513 | static u64 mem_find_max_overage(struct mem_cgroup *memcg) |
2514 | { | |
2515 | u64 overage, max_overage = 0; | |
e26733e0 | 2516 | |
8a5dbc65 JK |
2517 | do { |
2518 | overage = calculate_overage(page_counter_read(&memcg->memory), | |
d1663a90 | 2519 | READ_ONCE(memcg->memory.high)); |
8a5dbc65 | 2520 | max_overage = max(overage, max_overage); |
e26733e0 CD |
2521 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2522 | !mem_cgroup_is_root(memcg)); | |
2523 | ||
8a5dbc65 JK |
2524 | return max_overage; |
2525 | } | |
2526 | ||
4b82ab4f JK |
2527 | static u64 swap_find_max_overage(struct mem_cgroup *memcg) |
2528 | { | |
2529 | u64 overage, max_overage = 0; | |
2530 | ||
2531 | do { | |
2532 | overage = calculate_overage(page_counter_read(&memcg->swap), | |
2533 | READ_ONCE(memcg->swap.high)); | |
2534 | if (overage) | |
2535 | memcg_memory_event(memcg, MEMCG_SWAP_HIGH); | |
2536 | max_overage = max(overage, max_overage); | |
2537 | } while ((memcg = parent_mem_cgroup(memcg)) && | |
2538 | !mem_cgroup_is_root(memcg)); | |
2539 | ||
2540 | return max_overage; | |
2541 | } | |
2542 | ||
8a5dbc65 JK |
2543 | /* |
2544 | * Get the number of jiffies that we should penalise a mischievous cgroup which | |
2545 | * is exceeding its memory.high by checking both it and its ancestors. | |
2546 | */ | |
2547 | static unsigned long calculate_high_delay(struct mem_cgroup *memcg, | |
2548 | unsigned int nr_pages, | |
2549 | u64 max_overage) | |
2550 | { | |
2551 | unsigned long penalty_jiffies; | |
2552 | ||
e26733e0 CD |
2553 | if (!max_overage) |
2554 | return 0; | |
0e4b01df CD |
2555 | |
2556 | /* | |
0e4b01df CD |
2557 | * We use overage compared to memory.high to calculate the number of |
2558 | * jiffies to sleep (penalty_jiffies). Ideally this value should be | |
2559 | * fairly lenient on small overages, and increasingly harsh when the | |
2560 | * memcg in question makes it clear that it has no intention of stopping | |
2561 | * its crazy behaviour, so we exponentially increase the delay based on | |
2562 | * overage amount. | |
2563 | */ | |
e26733e0 CD |
2564 | penalty_jiffies = max_overage * max_overage * HZ; |
2565 | penalty_jiffies >>= MEMCG_DELAY_PRECISION_SHIFT; | |
2566 | penalty_jiffies >>= MEMCG_DELAY_SCALING_SHIFT; | |
0e4b01df CD |
2567 | |
2568 | /* | |
2569 | * Factor in the task's own contribution to the overage, such that four | |
2570 | * N-sized allocations are throttled approximately the same as one | |
2571 | * 4N-sized allocation. | |
2572 | * | |
2573 | * MEMCG_CHARGE_BATCH pages is nominal, so work out how much smaller or | |
2574 | * larger the current charge patch is than that. | |
2575 | */ | |
ff144e69 | 2576 | return penalty_jiffies * nr_pages / MEMCG_CHARGE_BATCH; |
e26733e0 CD |
2577 | } |
2578 | ||
2579 | /* | |
2580 | * Scheduled by try_charge() to be executed from the userland return path | |
2581 | * and reclaims memory over the high limit. | |
2582 | */ | |
2583 | void mem_cgroup_handle_over_high(void) | |
2584 | { | |
2585 | unsigned long penalty_jiffies; | |
2586 | unsigned long pflags; | |
b3ff9291 | 2587 | unsigned long nr_reclaimed; |
e26733e0 | 2588 | unsigned int nr_pages = current->memcg_nr_pages_over_high; |
d977aa93 | 2589 | int nr_retries = MAX_RECLAIM_RETRIES; |
e26733e0 | 2590 | struct mem_cgroup *memcg; |
b3ff9291 | 2591 | bool in_retry = false; |
e26733e0 CD |
2592 | |
2593 | if (likely(!nr_pages)) | |
2594 | return; | |
2595 | ||
2596 | memcg = get_mem_cgroup_from_mm(current->mm); | |
e26733e0 CD |
2597 | current->memcg_nr_pages_over_high = 0; |
2598 | ||
b3ff9291 CD |
2599 | retry_reclaim: |
2600 | /* | |
2601 | * The allocating task should reclaim at least the batch size, but for | |
2602 | * subsequent retries we only want to do what's necessary to prevent oom | |
2603 | * or breaching resource isolation. | |
2604 | * | |
2605 | * This is distinct from memory.max or page allocator behaviour because | |
2606 | * memory.high is currently batched, whereas memory.max and the page | |
2607 | * allocator run every time an allocation is made. | |
2608 | */ | |
2609 | nr_reclaimed = reclaim_high(memcg, | |
2610 | in_retry ? SWAP_CLUSTER_MAX : nr_pages, | |
2611 | GFP_KERNEL); | |
2612 | ||
e26733e0 CD |
2613 | /* |
2614 | * memory.high is breached and reclaim is unable to keep up. Throttle | |
2615 | * allocators proactively to slow down excessive growth. | |
2616 | */ | |
8a5dbc65 JK |
2617 | penalty_jiffies = calculate_high_delay(memcg, nr_pages, |
2618 | mem_find_max_overage(memcg)); | |
0e4b01df | 2619 | |
4b82ab4f JK |
2620 | penalty_jiffies += calculate_high_delay(memcg, nr_pages, |
2621 | swap_find_max_overage(memcg)); | |
2622 | ||
ff144e69 JK |
2623 | /* |
2624 | * Clamp the max delay per usermode return so as to still keep the | |
2625 | * application moving forwards and also permit diagnostics, albeit | |
2626 | * extremely slowly. | |
2627 | */ | |
2628 | penalty_jiffies = min(penalty_jiffies, MEMCG_MAX_HIGH_DELAY_JIFFIES); | |
2629 | ||
0e4b01df CD |
2630 | /* |
2631 | * Don't sleep if the amount of jiffies this memcg owes us is so low | |
2632 | * that it's not even worth doing, in an attempt to be nice to those who | |
2633 | * go only a small amount over their memory.high value and maybe haven't | |
2634 | * been aggressively reclaimed enough yet. | |
2635 | */ | |
2636 | if (penalty_jiffies <= HZ / 100) | |
2637 | goto out; | |
2638 | ||
b3ff9291 CD |
2639 | /* |
2640 | * If reclaim is making forward progress but we're still over | |
2641 | * memory.high, we want to encourage that rather than doing allocator | |
2642 | * throttling. | |
2643 | */ | |
2644 | if (nr_reclaimed || nr_retries--) { | |
2645 | in_retry = true; | |
2646 | goto retry_reclaim; | |
2647 | } | |
2648 | ||
0e4b01df CD |
2649 | /* |
2650 | * If we exit early, we're guaranteed to die (since | |
2651 | * schedule_timeout_killable sets TASK_KILLABLE). This means we don't | |
2652 | * need to account for any ill-begotten jiffies to pay them off later. | |
2653 | */ | |
2654 | psi_memstall_enter(&pflags); | |
2655 | schedule_timeout_killable(penalty_jiffies); | |
2656 | psi_memstall_leave(&pflags); | |
2657 | ||
2658 | out: | |
2659 | css_put(&memcg->css); | |
b23afb93 TH |
2660 | } |
2661 | ||
00501b53 JW |
2662 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2663 | unsigned int nr_pages) | |
8a9f3ccd | 2664 | { |
a983b5eb | 2665 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
d977aa93 | 2666 | int nr_retries = MAX_RECLAIM_RETRIES; |
6539cc05 | 2667 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2668 | struct page_counter *counter; |
e22c6ed9 | 2669 | enum oom_status oom_status; |
6539cc05 | 2670 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2671 | bool may_swap = true; |
2672 | bool drained = false; | |
e22c6ed9 | 2673 | unsigned long pflags; |
a636b327 | 2674 | |
ce00a967 | 2675 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 2676 | return 0; |
6539cc05 | 2677 | retry: |
b6b6cc72 | 2678 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2679 | return 0; |
8a9f3ccd | 2680 | |
7941d214 | 2681 | if (!do_memsw_account() || |
6071ca52 JW |
2682 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2683 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2684 | goto done_restock; |
7941d214 | 2685 | if (do_memsw_account()) |
3e32cb2e JW |
2686 | page_counter_uncharge(&memcg->memsw, batch); |
2687 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2688 | } else { |
3e32cb2e | 2689 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2690 | may_swap = false; |
3fbe7244 | 2691 | } |
7a81b88c | 2692 | |
6539cc05 JW |
2693 | if (batch > nr_pages) { |
2694 | batch = nr_pages; | |
2695 | goto retry; | |
2696 | } | |
6d61ef40 | 2697 | |
869712fd JW |
2698 | /* |
2699 | * Memcg doesn't have a dedicated reserve for atomic | |
2700 | * allocations. But like the global atomic pool, we need to | |
2701 | * put the burden of reclaim on regular allocation requests | |
2702 | * and let these go through as privileged allocations. | |
2703 | */ | |
2704 | if (gfp_mask & __GFP_ATOMIC) | |
2705 | goto force; | |
2706 | ||
06b078fc JW |
2707 | /* |
2708 | * Unlike in global OOM situations, memcg is not in a physical | |
2709 | * memory shortage. Allow dying and OOM-killed tasks to | |
2710 | * bypass the last charges so that they can exit quickly and | |
2711 | * free their memory. | |
2712 | */ | |
7775face | 2713 | if (unlikely(should_force_charge())) |
10d53c74 | 2714 | goto force; |
06b078fc | 2715 | |
89a28483 JW |
2716 | /* |
2717 | * Prevent unbounded recursion when reclaim operations need to | |
2718 | * allocate memory. This might exceed the limits temporarily, | |
2719 | * but we prefer facilitating memory reclaim and getting back | |
2720 | * under the limit over triggering OOM kills in these cases. | |
2721 | */ | |
2722 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2723 | goto force; | |
2724 | ||
06b078fc JW |
2725 | if (unlikely(task_in_memcg_oom(current))) |
2726 | goto nomem; | |
2727 | ||
d0164adc | 2728 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2729 | goto nomem; |
4b534334 | 2730 | |
e27be240 | 2731 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
241994ed | 2732 | |
e22c6ed9 | 2733 | psi_memstall_enter(&pflags); |
b70a2a21 JW |
2734 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2735 | gfp_mask, may_swap); | |
e22c6ed9 | 2736 | psi_memstall_leave(&pflags); |
6539cc05 | 2737 | |
61e02c74 | 2738 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2739 | goto retry; |
28c34c29 | 2740 | |
b70a2a21 | 2741 | if (!drained) { |
6d3d6aa2 | 2742 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2743 | drained = true; |
2744 | goto retry; | |
2745 | } | |
2746 | ||
28c34c29 JW |
2747 | if (gfp_mask & __GFP_NORETRY) |
2748 | goto nomem; | |
6539cc05 JW |
2749 | /* |
2750 | * Even though the limit is exceeded at this point, reclaim | |
2751 | * may have been able to free some pages. Retry the charge | |
2752 | * before killing the task. | |
2753 | * | |
2754 | * Only for regular pages, though: huge pages are rather | |
2755 | * unlikely to succeed so close to the limit, and we fall back | |
2756 | * to regular pages anyway in case of failure. | |
2757 | */ | |
61e02c74 | 2758 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2759 | goto retry; |
2760 | /* | |
2761 | * At task move, charge accounts can be doubly counted. So, it's | |
2762 | * better to wait until the end of task_move if something is going on. | |
2763 | */ | |
2764 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2765 | goto retry; | |
2766 | ||
9b130619 JW |
2767 | if (nr_retries--) |
2768 | goto retry; | |
2769 | ||
38d38493 | 2770 | if (gfp_mask & __GFP_RETRY_MAYFAIL) |
29ef680a MH |
2771 | goto nomem; |
2772 | ||
06b078fc | 2773 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2774 | goto force; |
06b078fc | 2775 | |
6539cc05 | 2776 | if (fatal_signal_pending(current)) |
10d53c74 | 2777 | goto force; |
6539cc05 | 2778 | |
29ef680a MH |
2779 | /* |
2780 | * keep retrying as long as the memcg oom killer is able to make | |
2781 | * a forward progress or bypass the charge if the oom killer | |
2782 | * couldn't make any progress. | |
2783 | */ | |
2784 | oom_status = mem_cgroup_oom(mem_over_limit, gfp_mask, | |
3608de07 | 2785 | get_order(nr_pages * PAGE_SIZE)); |
29ef680a MH |
2786 | switch (oom_status) { |
2787 | case OOM_SUCCESS: | |
d977aa93 | 2788 | nr_retries = MAX_RECLAIM_RETRIES; |
29ef680a MH |
2789 | goto retry; |
2790 | case OOM_FAILED: | |
2791 | goto force; | |
2792 | default: | |
2793 | goto nomem; | |
2794 | } | |
7a81b88c | 2795 | nomem: |
6d1fdc48 | 2796 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2797 | return -ENOMEM; |
10d53c74 TH |
2798 | force: |
2799 | /* | |
2800 | * The allocation either can't fail or will lead to more memory | |
2801 | * being freed very soon. Allow memory usage go over the limit | |
2802 | * temporarily by force charging it. | |
2803 | */ | |
2804 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2805 | if (do_memsw_account()) |
10d53c74 | 2806 | page_counter_charge(&memcg->memsw, nr_pages); |
10d53c74 TH |
2807 | |
2808 | return 0; | |
6539cc05 JW |
2809 | |
2810 | done_restock: | |
2811 | if (batch > nr_pages) | |
2812 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2813 | |
241994ed | 2814 | /* |
b23afb93 TH |
2815 | * If the hierarchy is above the normal consumption range, schedule |
2816 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2817 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2818 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2819 | * not recorded as it most likely matches current's and won't | |
2820 | * change in the meantime. As high limit is checked again before | |
2821 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2822 | */ |
2823 | do { | |
4b82ab4f JK |
2824 | bool mem_high, swap_high; |
2825 | ||
2826 | mem_high = page_counter_read(&memcg->memory) > | |
2827 | READ_ONCE(memcg->memory.high); | |
2828 | swap_high = page_counter_read(&memcg->swap) > | |
2829 | READ_ONCE(memcg->swap.high); | |
2830 | ||
2831 | /* Don't bother a random interrupted task */ | |
2832 | if (in_interrupt()) { | |
2833 | if (mem_high) { | |
f7e1cb6e JW |
2834 | schedule_work(&memcg->high_work); |
2835 | break; | |
2836 | } | |
4b82ab4f JK |
2837 | continue; |
2838 | } | |
2839 | ||
2840 | if (mem_high || swap_high) { | |
2841 | /* | |
2842 | * The allocating tasks in this cgroup will need to do | |
2843 | * reclaim or be throttled to prevent further growth | |
2844 | * of the memory or swap footprints. | |
2845 | * | |
2846 | * Target some best-effort fairness between the tasks, | |
2847 | * and distribute reclaim work and delay penalties | |
2848 | * based on how much each task is actually allocating. | |
2849 | */ | |
9516a18a | 2850 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2851 | set_notify_resume(current); |
2852 | break; | |
2853 | } | |
241994ed | 2854 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2855 | |
2856 | return 0; | |
7a81b88c | 2857 | } |
8a9f3ccd | 2858 | |
f0e45fb4 | 2859 | #if defined(CONFIG_MEMCG_KMEM) || defined(CONFIG_MMU) |
00501b53 | 2860 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2861 | { |
ce00a967 JW |
2862 | if (mem_cgroup_is_root(memcg)) |
2863 | return; | |
2864 | ||
3e32cb2e | 2865 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2866 | if (do_memsw_account()) |
3e32cb2e | 2867 | page_counter_uncharge(&memcg->memsw, nr_pages); |
d01dd17f | 2868 | } |
f0e45fb4 | 2869 | #endif |
d01dd17f | 2870 | |
d9eb1ea2 | 2871 | static void commit_charge(struct page *page, struct mem_cgroup *memcg) |
0a31bc97 | 2872 | { |
1306a85a | 2873 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
0a31bc97 | 2874 | /* |
a5eb011a | 2875 | * Any of the following ensures page's memcg stability: |
0a31bc97 | 2876 | * |
a0b5b414 JW |
2877 | * - the page lock |
2878 | * - LRU isolation | |
2879 | * - lock_page_memcg() | |
2880 | * - exclusive reference | |
0a31bc97 | 2881 | */ |
1306a85a | 2882 | page->mem_cgroup = memcg; |
7a81b88c | 2883 | } |
66e1707b | 2884 | |
84c07d11 | 2885 | #ifdef CONFIG_MEMCG_KMEM |
10befea9 RG |
2886 | int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s, |
2887 | gfp_t gfp) | |
2888 | { | |
2889 | unsigned int objects = objs_per_slab_page(s, page); | |
2890 | void *vec; | |
2891 | ||
2892 | vec = kcalloc_node(objects, sizeof(struct obj_cgroup *), gfp, | |
2893 | page_to_nid(page)); | |
2894 | if (!vec) | |
2895 | return -ENOMEM; | |
2896 | ||
2897 | if (cmpxchg(&page->obj_cgroups, NULL, | |
2898 | (struct obj_cgroup **) ((unsigned long)vec | 0x1UL))) | |
2899 | kfree(vec); | |
2900 | else | |
2901 | kmemleak_not_leak(vec); | |
2902 | ||
2903 | return 0; | |
2904 | } | |
2905 | ||
8380ce47 RG |
2906 | /* |
2907 | * Returns a pointer to the memory cgroup to which the kernel object is charged. | |
2908 | * | |
2909 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), | |
2910 | * cgroup_mutex, etc. | |
2911 | */ | |
2912 | struct mem_cgroup *mem_cgroup_from_obj(void *p) | |
2913 | { | |
2914 | struct page *page; | |
2915 | ||
2916 | if (mem_cgroup_disabled()) | |
2917 | return NULL; | |
2918 | ||
2919 | page = virt_to_head_page(p); | |
2920 | ||
19b629c9 RG |
2921 | /* |
2922 | * If page->mem_cgroup is set, it's either a simple mem_cgroup pointer | |
2923 | * or a pointer to obj_cgroup vector. In the latter case the lowest | |
2924 | * bit of the pointer is set. | |
2925 | * The page->mem_cgroup pointer can be asynchronously changed | |
2926 | * from NULL to (obj_cgroup_vec | 0x1UL), but can't be changed | |
2927 | * from a valid memcg pointer to objcg vector or back. | |
2928 | */ | |
2929 | if (!page->mem_cgroup) | |
2930 | return NULL; | |
2931 | ||
8380ce47 | 2932 | /* |
9855609b RG |
2933 | * Slab objects are accounted individually, not per-page. |
2934 | * Memcg membership data for each individual object is saved in | |
2935 | * the page->obj_cgroups. | |
8380ce47 | 2936 | */ |
9855609b RG |
2937 | if (page_has_obj_cgroups(page)) { |
2938 | struct obj_cgroup *objcg; | |
2939 | unsigned int off; | |
2940 | ||
2941 | off = obj_to_index(page->slab_cache, page, p); | |
2942 | objcg = page_obj_cgroups(page)[off]; | |
10befea9 RG |
2943 | if (objcg) |
2944 | return obj_cgroup_memcg(objcg); | |
2945 | ||
2946 | return NULL; | |
9855609b | 2947 | } |
8380ce47 RG |
2948 | |
2949 | /* All other pages use page->mem_cgroup */ | |
2950 | return page->mem_cgroup; | |
2951 | } | |
2952 | ||
bf4f0599 RG |
2953 | __always_inline struct obj_cgroup *get_obj_cgroup_from_current(void) |
2954 | { | |
2955 | struct obj_cgroup *objcg = NULL; | |
2956 | struct mem_cgroup *memcg; | |
2957 | ||
279c3393 RG |
2958 | if (memcg_kmem_bypass()) |
2959 | return NULL; | |
2960 | ||
bf4f0599 | 2961 | rcu_read_lock(); |
37d5985c RG |
2962 | if (unlikely(active_memcg())) |
2963 | memcg = active_memcg(); | |
bf4f0599 RG |
2964 | else |
2965 | memcg = mem_cgroup_from_task(current); | |
2966 | ||
2967 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { | |
2968 | objcg = rcu_dereference(memcg->objcg); | |
2969 | if (objcg && obj_cgroup_tryget(objcg)) | |
2970 | break; | |
2f7659a3 | 2971 | objcg = NULL; |
bf4f0599 RG |
2972 | } |
2973 | rcu_read_unlock(); | |
2974 | ||
2975 | return objcg; | |
2976 | } | |
2977 | ||
f3bb3043 | 2978 | static int memcg_alloc_cache_id(void) |
55007d84 | 2979 | { |
f3bb3043 VD |
2980 | int id, size; |
2981 | int err; | |
2982 | ||
dbcf73e2 | 2983 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2984 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2985 | if (id < 0) | |
2986 | return id; | |
55007d84 | 2987 | |
dbcf73e2 | 2988 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2989 | return id; |
2990 | ||
2991 | /* | |
2992 | * There's no space for the new id in memcg_caches arrays, | |
2993 | * so we have to grow them. | |
2994 | */ | |
05257a1a | 2995 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2996 | |
2997 | size = 2 * (id + 1); | |
55007d84 GC |
2998 | if (size < MEMCG_CACHES_MIN_SIZE) |
2999 | size = MEMCG_CACHES_MIN_SIZE; | |
3000 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
3001 | size = MEMCG_CACHES_MAX_SIZE; | |
3002 | ||
9855609b | 3003 | err = memcg_update_all_list_lrus(size); |
05257a1a VD |
3004 | if (!err) |
3005 | memcg_nr_cache_ids = size; | |
3006 | ||
3007 | up_write(&memcg_cache_ids_sem); | |
3008 | ||
f3bb3043 | 3009 | if (err) { |
dbcf73e2 | 3010 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
3011 | return err; |
3012 | } | |
3013 | return id; | |
3014 | } | |
3015 | ||
3016 | static void memcg_free_cache_id(int id) | |
3017 | { | |
dbcf73e2 | 3018 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
3019 | } |
3020 | ||
45264778 | 3021 | /** |
4b13f64d | 3022 | * __memcg_kmem_charge: charge a number of kernel pages to a memcg |
10eaec2f | 3023 | * @memcg: memory cgroup to charge |
45264778 | 3024 | * @gfp: reclaim mode |
92d0510c | 3025 | * @nr_pages: number of pages to charge |
45264778 VD |
3026 | * |
3027 | * Returns 0 on success, an error code on failure. | |
3028 | */ | |
4b13f64d RG |
3029 | int __memcg_kmem_charge(struct mem_cgroup *memcg, gfp_t gfp, |
3030 | unsigned int nr_pages) | |
7ae1e1d0 | 3031 | { |
f3ccb2c4 | 3032 | struct page_counter *counter; |
7ae1e1d0 GC |
3033 | int ret; |
3034 | ||
f3ccb2c4 | 3035 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 3036 | if (ret) |
f3ccb2c4 | 3037 | return ret; |
52c29b04 JW |
3038 | |
3039 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
3040 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
e55d9d9b MH |
3041 | |
3042 | /* | |
3043 | * Enforce __GFP_NOFAIL allocation because callers are not | |
3044 | * prepared to see failures and likely do not have any failure | |
3045 | * handling code. | |
3046 | */ | |
3047 | if (gfp & __GFP_NOFAIL) { | |
3048 | page_counter_charge(&memcg->kmem, nr_pages); | |
3049 | return 0; | |
3050 | } | |
52c29b04 JW |
3051 | cancel_charge(memcg, nr_pages); |
3052 | return -ENOMEM; | |
7ae1e1d0 | 3053 | } |
f3ccb2c4 | 3054 | return 0; |
7ae1e1d0 GC |
3055 | } |
3056 | ||
4b13f64d RG |
3057 | /** |
3058 | * __memcg_kmem_uncharge: uncharge a number of kernel pages from a memcg | |
3059 | * @memcg: memcg to uncharge | |
3060 | * @nr_pages: number of pages to uncharge | |
3061 | */ | |
3062 | void __memcg_kmem_uncharge(struct mem_cgroup *memcg, unsigned int nr_pages) | |
3063 | { | |
3064 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
3065 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
3066 | ||
3067 | page_counter_uncharge(&memcg->memory, nr_pages); | |
3068 | if (do_memsw_account()) | |
3069 | page_counter_uncharge(&memcg->memsw, nr_pages); | |
3070 | } | |
3071 | ||
45264778 | 3072 | /** |
f4b00eab | 3073 | * __memcg_kmem_charge_page: charge a kmem page to the current memory cgroup |
45264778 VD |
3074 | * @page: page to charge |
3075 | * @gfp: reclaim mode | |
3076 | * @order: allocation order | |
3077 | * | |
3078 | * Returns 0 on success, an error code on failure. | |
3079 | */ | |
f4b00eab | 3080 | int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 3081 | { |
f3ccb2c4 | 3082 | struct mem_cgroup *memcg; |
fcff7d7e | 3083 | int ret = 0; |
7ae1e1d0 | 3084 | |
d46eb14b | 3085 | memcg = get_mem_cgroup_from_current(); |
279c3393 | 3086 | if (memcg && !mem_cgroup_is_root(memcg)) { |
4b13f64d | 3087 | ret = __memcg_kmem_charge(memcg, gfp, 1 << order); |
4d96ba35 RG |
3088 | if (!ret) { |
3089 | page->mem_cgroup = memcg; | |
c4159a75 | 3090 | __SetPageKmemcg(page); |
1a3e1f40 | 3091 | return 0; |
4d96ba35 | 3092 | } |
279c3393 | 3093 | css_put(&memcg->css); |
c4159a75 | 3094 | } |
d05e83a6 | 3095 | return ret; |
7ae1e1d0 | 3096 | } |
49a18eae | 3097 | |
45264778 | 3098 | /** |
f4b00eab | 3099 | * __memcg_kmem_uncharge_page: uncharge a kmem page |
45264778 VD |
3100 | * @page: page to uncharge |
3101 | * @order: allocation order | |
3102 | */ | |
f4b00eab | 3103 | void __memcg_kmem_uncharge_page(struct page *page, int order) |
7ae1e1d0 | 3104 | { |
1306a85a | 3105 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 3106 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 3107 | |
7ae1e1d0 GC |
3108 | if (!memcg) |
3109 | return; | |
3110 | ||
309381fe | 3111 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
4b13f64d | 3112 | __memcg_kmem_uncharge(memcg, nr_pages); |
1306a85a | 3113 | page->mem_cgroup = NULL; |
1a3e1f40 | 3114 | css_put(&memcg->css); |
c4159a75 VD |
3115 | |
3116 | /* slab pages do not have PageKmemcg flag set */ | |
3117 | if (PageKmemcg(page)) | |
3118 | __ClearPageKmemcg(page); | |
60d3fd32 | 3119 | } |
bf4f0599 RG |
3120 | |
3121 | static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) | |
3122 | { | |
3123 | struct memcg_stock_pcp *stock; | |
3124 | unsigned long flags; | |
3125 | bool ret = false; | |
3126 | ||
3127 | local_irq_save(flags); | |
3128 | ||
3129 | stock = this_cpu_ptr(&memcg_stock); | |
3130 | if (objcg == stock->cached_objcg && stock->nr_bytes >= nr_bytes) { | |
3131 | stock->nr_bytes -= nr_bytes; | |
3132 | ret = true; | |
3133 | } | |
3134 | ||
3135 | local_irq_restore(flags); | |
3136 | ||
3137 | return ret; | |
3138 | } | |
3139 | ||
3140 | static void drain_obj_stock(struct memcg_stock_pcp *stock) | |
3141 | { | |
3142 | struct obj_cgroup *old = stock->cached_objcg; | |
3143 | ||
3144 | if (!old) | |
3145 | return; | |
3146 | ||
3147 | if (stock->nr_bytes) { | |
3148 | unsigned int nr_pages = stock->nr_bytes >> PAGE_SHIFT; | |
3149 | unsigned int nr_bytes = stock->nr_bytes & (PAGE_SIZE - 1); | |
3150 | ||
3151 | if (nr_pages) { | |
3152 | rcu_read_lock(); | |
3153 | __memcg_kmem_uncharge(obj_cgroup_memcg(old), nr_pages); | |
3154 | rcu_read_unlock(); | |
3155 | } | |
3156 | ||
3157 | /* | |
3158 | * The leftover is flushed to the centralized per-memcg value. | |
3159 | * On the next attempt to refill obj stock it will be moved | |
3160 | * to a per-cpu stock (probably, on an other CPU), see | |
3161 | * refill_obj_stock(). | |
3162 | * | |
3163 | * How often it's flushed is a trade-off between the memory | |
3164 | * limit enforcement accuracy and potential CPU contention, | |
3165 | * so it might be changed in the future. | |
3166 | */ | |
3167 | atomic_add(nr_bytes, &old->nr_charged_bytes); | |
3168 | stock->nr_bytes = 0; | |
3169 | } | |
3170 | ||
3171 | obj_cgroup_put(old); | |
3172 | stock->cached_objcg = NULL; | |
3173 | } | |
3174 | ||
3175 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
3176 | struct mem_cgroup *root_memcg) | |
3177 | { | |
3178 | struct mem_cgroup *memcg; | |
3179 | ||
3180 | if (stock->cached_objcg) { | |
3181 | memcg = obj_cgroup_memcg(stock->cached_objcg); | |
3182 | if (memcg && mem_cgroup_is_descendant(memcg, root_memcg)) | |
3183 | return true; | |
3184 | } | |
3185 | ||
3186 | return false; | |
3187 | } | |
3188 | ||
3189 | static void refill_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) | |
3190 | { | |
3191 | struct memcg_stock_pcp *stock; | |
3192 | unsigned long flags; | |
3193 | ||
3194 | local_irq_save(flags); | |
3195 | ||
3196 | stock = this_cpu_ptr(&memcg_stock); | |
3197 | if (stock->cached_objcg != objcg) { /* reset if necessary */ | |
3198 | drain_obj_stock(stock); | |
3199 | obj_cgroup_get(objcg); | |
3200 | stock->cached_objcg = objcg; | |
3201 | stock->nr_bytes = atomic_xchg(&objcg->nr_charged_bytes, 0); | |
3202 | } | |
3203 | stock->nr_bytes += nr_bytes; | |
3204 | ||
3205 | if (stock->nr_bytes > PAGE_SIZE) | |
3206 | drain_obj_stock(stock); | |
3207 | ||
3208 | local_irq_restore(flags); | |
3209 | } | |
3210 | ||
3211 | int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size) | |
3212 | { | |
3213 | struct mem_cgroup *memcg; | |
3214 | unsigned int nr_pages, nr_bytes; | |
3215 | int ret; | |
3216 | ||
3217 | if (consume_obj_stock(objcg, size)) | |
3218 | return 0; | |
3219 | ||
3220 | /* | |
3221 | * In theory, memcg->nr_charged_bytes can have enough | |
3222 | * pre-charged bytes to satisfy the allocation. However, | |
3223 | * flushing memcg->nr_charged_bytes requires two atomic | |
3224 | * operations, and memcg->nr_charged_bytes can't be big, | |
3225 | * so it's better to ignore it and try grab some new pages. | |
3226 | * memcg->nr_charged_bytes will be flushed in | |
3227 | * refill_obj_stock(), called from this function or | |
3228 | * independently later. | |
3229 | */ | |
3230 | rcu_read_lock(); | |
eefbfa7f | 3231 | retry: |
bf4f0599 | 3232 | memcg = obj_cgroup_memcg(objcg); |
eefbfa7f MS |
3233 | if (unlikely(!css_tryget(&memcg->css))) |
3234 | goto retry; | |
bf4f0599 RG |
3235 | rcu_read_unlock(); |
3236 | ||
3237 | nr_pages = size >> PAGE_SHIFT; | |
3238 | nr_bytes = size & (PAGE_SIZE - 1); | |
3239 | ||
3240 | if (nr_bytes) | |
3241 | nr_pages += 1; | |
3242 | ||
3243 | ret = __memcg_kmem_charge(memcg, gfp, nr_pages); | |
3244 | if (!ret && nr_bytes) | |
3245 | refill_obj_stock(objcg, PAGE_SIZE - nr_bytes); | |
3246 | ||
3247 | css_put(&memcg->css); | |
3248 | return ret; | |
3249 | } | |
3250 | ||
3251 | void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size) | |
3252 | { | |
3253 | refill_obj_stock(objcg, size); | |
3254 | } | |
3255 | ||
84c07d11 | 3256 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 3257 | |
ca3e0214 KH |
3258 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3259 | ||
ca3e0214 KH |
3260 | /* |
3261 | * Because tail pages are not marked as "used", set it. We're under | |
f4b7e272 | 3262 | * pgdat->lru_lock and migration entries setup in all page mappings. |
ca3e0214 | 3263 | */ |
e94c8a9c | 3264 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 3265 | { |
1a3e1f40 | 3266 | struct mem_cgroup *memcg = head->mem_cgroup; |
e94c8a9c | 3267 | int i; |
ca3e0214 | 3268 | |
3d37c4a9 KH |
3269 | if (mem_cgroup_disabled()) |
3270 | return; | |
b070e65c | 3271 | |
1a3e1f40 JW |
3272 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
3273 | css_get(&memcg->css); | |
3274 | head[i].mem_cgroup = memcg; | |
3275 | } | |
ca3e0214 | 3276 | } |
12d27107 | 3277 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3278 | |
c255a458 | 3279 | #ifdef CONFIG_MEMCG_SWAP |
02491447 DN |
3280 | /** |
3281 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3282 | * @entry: swap entry to be moved | |
3283 | * @from: mem_cgroup which the entry is moved from | |
3284 | * @to: mem_cgroup which the entry is moved to | |
3285 | * | |
3286 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3287 | * as the mem_cgroup's id of @from. | |
3288 | * | |
3289 | * Returns 0 on success, -EINVAL on failure. | |
3290 | * | |
3e32cb2e | 3291 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
3292 | * both res and memsw, and called css_get(). |
3293 | */ | |
3294 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3295 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3296 | { |
3297 | unsigned short old_id, new_id; | |
3298 | ||
34c00c31 LZ |
3299 | old_id = mem_cgroup_id(from); |
3300 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3301 | |
3302 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
3303 | mod_memcg_state(from, MEMCG_SWAP, -1); |
3304 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
3305 | return 0; |
3306 | } | |
3307 | return -EINVAL; | |
3308 | } | |
3309 | #else | |
3310 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3311 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3312 | { |
3313 | return -EINVAL; | |
3314 | } | |
8c7c6e34 | 3315 | #endif |
d13d1443 | 3316 | |
bbec2e15 | 3317 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 3318 | |
bbec2e15 RG |
3319 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
3320 | unsigned long max, bool memsw) | |
628f4235 | 3321 | { |
3e32cb2e | 3322 | bool enlarge = false; |
bb4a7ea2 | 3323 | bool drained = false; |
3e32cb2e | 3324 | int ret; |
c054a78c YZ |
3325 | bool limits_invariant; |
3326 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 3327 | |
3e32cb2e | 3328 | do { |
628f4235 KH |
3329 | if (signal_pending(current)) { |
3330 | ret = -EINTR; | |
3331 | break; | |
3332 | } | |
3e32cb2e | 3333 | |
bbec2e15 | 3334 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
3335 | /* |
3336 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 3337 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 3338 | */ |
15b42562 | 3339 | limits_invariant = memsw ? max >= READ_ONCE(memcg->memory.max) : |
bbec2e15 | 3340 | max <= memcg->memsw.max; |
c054a78c | 3341 | if (!limits_invariant) { |
bbec2e15 | 3342 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 3343 | ret = -EINVAL; |
8c7c6e34 KH |
3344 | break; |
3345 | } | |
bbec2e15 | 3346 | if (max > counter->max) |
3e32cb2e | 3347 | enlarge = true; |
bbec2e15 RG |
3348 | ret = page_counter_set_max(counter, max); |
3349 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
3350 | |
3351 | if (!ret) | |
3352 | break; | |
3353 | ||
bb4a7ea2 SB |
3354 | if (!drained) { |
3355 | drain_all_stock(memcg); | |
3356 | drained = true; | |
3357 | continue; | |
3358 | } | |
3359 | ||
1ab5c056 AR |
3360 | if (!try_to_free_mem_cgroup_pages(memcg, 1, |
3361 | GFP_KERNEL, !memsw)) { | |
3362 | ret = -EBUSY; | |
3363 | break; | |
3364 | } | |
3365 | } while (true); | |
3e32cb2e | 3366 | |
3c11ecf4 KH |
3367 | if (!ret && enlarge) |
3368 | memcg_oom_recover(memcg); | |
3e32cb2e | 3369 | |
628f4235 KH |
3370 | return ret; |
3371 | } | |
3372 | ||
ef8f2327 | 3373 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
3374 | gfp_t gfp_mask, |
3375 | unsigned long *total_scanned) | |
3376 | { | |
3377 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 3378 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
3379 | unsigned long reclaimed; |
3380 | int loop = 0; | |
ef8f2327 | 3381 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 3382 | unsigned long excess; |
0608f43d AM |
3383 | unsigned long nr_scanned; |
3384 | ||
3385 | if (order > 0) | |
3386 | return 0; | |
3387 | ||
ef8f2327 | 3388 | mctz = soft_limit_tree_node(pgdat->node_id); |
d6507ff5 MH |
3389 | |
3390 | /* | |
3391 | * Do not even bother to check the largest node if the root | |
3392 | * is empty. Do it lockless to prevent lock bouncing. Races | |
3393 | * are acceptable as soft limit is best effort anyway. | |
3394 | */ | |
bfc7228b | 3395 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
3396 | return 0; |
3397 | ||
0608f43d AM |
3398 | /* |
3399 | * This loop can run a while, specially if mem_cgroup's continuously | |
3400 | * keep exceeding their soft limit and putting the system under | |
3401 | * pressure | |
3402 | */ | |
3403 | do { | |
3404 | if (next_mz) | |
3405 | mz = next_mz; | |
3406 | else | |
3407 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3408 | if (!mz) | |
3409 | break; | |
3410 | ||
3411 | nr_scanned = 0; | |
ef8f2327 | 3412 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
0608f43d AM |
3413 | gfp_mask, &nr_scanned); |
3414 | nr_reclaimed += reclaimed; | |
3415 | *total_scanned += nr_scanned; | |
0a31bc97 | 3416 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 3417 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
3418 | |
3419 | /* | |
3420 | * If we failed to reclaim anything from this memory cgroup | |
3421 | * it is time to move on to the next cgroup | |
3422 | */ | |
3423 | next_mz = NULL; | |
bc2f2e7f VD |
3424 | if (!reclaimed) |
3425 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3426 | ||
3e32cb2e | 3427 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3428 | /* |
3429 | * One school of thought says that we should not add | |
3430 | * back the node to the tree if reclaim returns 0. | |
3431 | * But our reclaim could return 0, simply because due | |
3432 | * to priority we are exposing a smaller subset of | |
3433 | * memory to reclaim from. Consider this as a longer | |
3434 | * term TODO. | |
3435 | */ | |
3436 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3437 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3438 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3439 | css_put(&mz->memcg->css); |
3440 | loop++; | |
3441 | /* | |
3442 | * Could not reclaim anything and there are no more | |
3443 | * mem cgroups to try or we seem to be looping without | |
3444 | * reclaiming anything. | |
3445 | */ | |
3446 | if (!nr_reclaimed && | |
3447 | (next_mz == NULL || | |
3448 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3449 | break; | |
3450 | } while (!nr_reclaimed); | |
3451 | if (next_mz) | |
3452 | css_put(&next_mz->memcg->css); | |
3453 | return nr_reclaimed; | |
3454 | } | |
3455 | ||
c26251f9 | 3456 | /* |
51038171 | 3457 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
3458 | * |
3459 | * Caller is responsible for holding css reference for memcg. | |
3460 | */ | |
3461 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3462 | { | |
d977aa93 | 3463 | int nr_retries = MAX_RECLAIM_RETRIES; |
c26251f9 | 3464 | |
c1e862c1 KH |
3465 | /* we call try-to-free pages for make this cgroup empty */ |
3466 | lru_add_drain_all(); | |
d12c60f6 JS |
3467 | |
3468 | drain_all_stock(memcg); | |
3469 | ||
f817ed48 | 3470 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3471 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 3472 | int progress; |
c1e862c1 | 3473 | |
c26251f9 MH |
3474 | if (signal_pending(current)) |
3475 | return -EINTR; | |
3476 | ||
b70a2a21 JW |
3477 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3478 | GFP_KERNEL, true); | |
c1e862c1 | 3479 | if (!progress) { |
f817ed48 | 3480 | nr_retries--; |
c1e862c1 | 3481 | /* maybe some writeback is necessary */ |
8aa7e847 | 3482 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3483 | } |
f817ed48 KH |
3484 | |
3485 | } | |
ab5196c2 MH |
3486 | |
3487 | return 0; | |
cc847582 KH |
3488 | } |
3489 | ||
6770c64e TH |
3490 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3491 | char *buf, size_t nbytes, | |
3492 | loff_t off) | |
c1e862c1 | 3493 | { |
6770c64e | 3494 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3495 | |
d8423011 MH |
3496 | if (mem_cgroup_is_root(memcg)) |
3497 | return -EINVAL; | |
6770c64e | 3498 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3499 | } |
3500 | ||
182446d0 TH |
3501 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3502 | struct cftype *cft) | |
18f59ea7 | 3503 | { |
bef8620c | 3504 | return 1; |
18f59ea7 BS |
3505 | } |
3506 | ||
182446d0 TH |
3507 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3508 | struct cftype *cft, u64 val) | |
18f59ea7 | 3509 | { |
bef8620c | 3510 | if (val == 1) |
0b8f73e1 | 3511 | return 0; |
567fb435 | 3512 | |
bef8620c RG |
3513 | pr_warn_once("Non-hierarchical mode is deprecated. " |
3514 | "Please report your usecase to linux-mm@kvack.org if you " | |
3515 | "depend on this functionality.\n"); | |
567fb435 | 3516 | |
bef8620c | 3517 | return -EINVAL; |
18f59ea7 BS |
3518 | } |
3519 | ||
6f646156 | 3520 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 3521 | { |
42a30035 | 3522 | unsigned long val; |
ce00a967 | 3523 | |
3e32cb2e | 3524 | if (mem_cgroup_is_root(memcg)) { |
0d1c2072 | 3525 | val = memcg_page_state(memcg, NR_FILE_PAGES) + |
be5d0a74 | 3526 | memcg_page_state(memcg, NR_ANON_MAPPED); |
42a30035 JW |
3527 | if (swap) |
3528 | val += memcg_page_state(memcg, MEMCG_SWAP); | |
3e32cb2e | 3529 | } else { |
ce00a967 | 3530 | if (!swap) |
3e32cb2e | 3531 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3532 | else |
3e32cb2e | 3533 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3534 | } |
c12176d3 | 3535 | return val; |
ce00a967 JW |
3536 | } |
3537 | ||
3e32cb2e JW |
3538 | enum { |
3539 | RES_USAGE, | |
3540 | RES_LIMIT, | |
3541 | RES_MAX_USAGE, | |
3542 | RES_FAILCNT, | |
3543 | RES_SOFT_LIMIT, | |
3544 | }; | |
ce00a967 | 3545 | |
791badbd | 3546 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3547 | struct cftype *cft) |
8cdea7c0 | 3548 | { |
182446d0 | 3549 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3550 | struct page_counter *counter; |
af36f906 | 3551 | |
3e32cb2e | 3552 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3553 | case _MEM: |
3e32cb2e JW |
3554 | counter = &memcg->memory; |
3555 | break; | |
8c7c6e34 | 3556 | case _MEMSWAP: |
3e32cb2e JW |
3557 | counter = &memcg->memsw; |
3558 | break; | |
510fc4e1 | 3559 | case _KMEM: |
3e32cb2e | 3560 | counter = &memcg->kmem; |
510fc4e1 | 3561 | break; |
d55f90bf | 3562 | case _TCP: |
0db15298 | 3563 | counter = &memcg->tcpmem; |
d55f90bf | 3564 | break; |
8c7c6e34 KH |
3565 | default: |
3566 | BUG(); | |
8c7c6e34 | 3567 | } |
3e32cb2e JW |
3568 | |
3569 | switch (MEMFILE_ATTR(cft->private)) { | |
3570 | case RES_USAGE: | |
3571 | if (counter == &memcg->memory) | |
c12176d3 | 3572 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 3573 | if (counter == &memcg->memsw) |
c12176d3 | 3574 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
3575 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
3576 | case RES_LIMIT: | |
bbec2e15 | 3577 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
3578 | case RES_MAX_USAGE: |
3579 | return (u64)counter->watermark * PAGE_SIZE; | |
3580 | case RES_FAILCNT: | |
3581 | return counter->failcnt; | |
3582 | case RES_SOFT_LIMIT: | |
3583 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3584 | default: | |
3585 | BUG(); | |
3586 | } | |
8cdea7c0 | 3587 | } |
510fc4e1 | 3588 | |
4a87e2a2 | 3589 | static void memcg_flush_percpu_vmstats(struct mem_cgroup *memcg) |
c350a99e | 3590 | { |
4a87e2a2 | 3591 | unsigned long stat[MEMCG_NR_STAT] = {0}; |
c350a99e RG |
3592 | struct mem_cgroup *mi; |
3593 | int node, cpu, i; | |
c350a99e RG |
3594 | |
3595 | for_each_online_cpu(cpu) | |
4a87e2a2 | 3596 | for (i = 0; i < MEMCG_NR_STAT; i++) |
6c1c2808 | 3597 | stat[i] += per_cpu(memcg->vmstats_percpu->stat[i], cpu); |
c350a99e RG |
3598 | |
3599 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
4a87e2a2 | 3600 | for (i = 0; i < MEMCG_NR_STAT; i++) |
c350a99e RG |
3601 | atomic_long_add(stat[i], &mi->vmstats[i]); |
3602 | ||
3603 | for_each_node(node) { | |
3604 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; | |
3605 | struct mem_cgroup_per_node *pi; | |
3606 | ||
4a87e2a2 | 3607 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
c350a99e RG |
3608 | stat[i] = 0; |
3609 | ||
3610 | for_each_online_cpu(cpu) | |
4a87e2a2 | 3611 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
6c1c2808 SB |
3612 | stat[i] += per_cpu( |
3613 | pn->lruvec_stat_cpu->count[i], cpu); | |
c350a99e RG |
3614 | |
3615 | for (pi = pn; pi; pi = parent_nodeinfo(pi, node)) | |
4a87e2a2 | 3616 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
c350a99e RG |
3617 | atomic_long_add(stat[i], &pi->lruvec_stat[i]); |
3618 | } | |
3619 | } | |
3620 | ||
bb65f89b RG |
3621 | static void memcg_flush_percpu_vmevents(struct mem_cgroup *memcg) |
3622 | { | |
3623 | unsigned long events[NR_VM_EVENT_ITEMS]; | |
3624 | struct mem_cgroup *mi; | |
3625 | int cpu, i; | |
3626 | ||
3627 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
3628 | events[i] = 0; | |
3629 | ||
3630 | for_each_online_cpu(cpu) | |
3631 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
6c1c2808 SB |
3632 | events[i] += per_cpu(memcg->vmstats_percpu->events[i], |
3633 | cpu); | |
bb65f89b RG |
3634 | |
3635 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
3636 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
3637 | atomic_long_add(events[i], &mi->vmevents[i]); | |
3638 | } | |
3639 | ||
84c07d11 | 3640 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 3641 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 3642 | { |
bf4f0599 | 3643 | struct obj_cgroup *objcg; |
d6441637 VD |
3644 | int memcg_id; |
3645 | ||
b313aeee VD |
3646 | if (cgroup_memory_nokmem) |
3647 | return 0; | |
3648 | ||
2a4db7eb | 3649 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 3650 | BUG_ON(memcg->kmem_state); |
d6441637 | 3651 | |
f3bb3043 | 3652 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
3653 | if (memcg_id < 0) |
3654 | return memcg_id; | |
d6441637 | 3655 | |
bf4f0599 RG |
3656 | objcg = obj_cgroup_alloc(); |
3657 | if (!objcg) { | |
3658 | memcg_free_cache_id(memcg_id); | |
3659 | return -ENOMEM; | |
3660 | } | |
3661 | objcg->memcg = memcg; | |
3662 | rcu_assign_pointer(memcg->objcg, objcg); | |
3663 | ||
d648bcc7 RG |
3664 | static_branch_enable(&memcg_kmem_enabled_key); |
3665 | ||
900a38f0 | 3666 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 3667 | memcg->kmem_state = KMEM_ONLINE; |
0b8f73e1 JW |
3668 | |
3669 | return 0; | |
d6441637 VD |
3670 | } |
3671 | ||
8e0a8912 JW |
3672 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
3673 | { | |
3674 | struct cgroup_subsys_state *css; | |
3675 | struct mem_cgroup *parent, *child; | |
3676 | int kmemcg_id; | |
3677 | ||
3678 | if (memcg->kmem_state != KMEM_ONLINE) | |
3679 | return; | |
9855609b | 3680 | |
8e0a8912 JW |
3681 | memcg->kmem_state = KMEM_ALLOCATED; |
3682 | ||
8e0a8912 JW |
3683 | parent = parent_mem_cgroup(memcg); |
3684 | if (!parent) | |
3685 | parent = root_mem_cgroup; | |
3686 | ||
bf4f0599 | 3687 | memcg_reparent_objcgs(memcg, parent); |
fb2f2b0a RG |
3688 | |
3689 | kmemcg_id = memcg->kmemcg_id; | |
3690 | BUG_ON(kmemcg_id < 0); | |
3691 | ||
8e0a8912 JW |
3692 | /* |
3693 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3694 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3695 | * to ones of the parent. After we have finished, all list_lrus | |
3696 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3697 | * ordering is imposed by list_lru_node->lock taken by | |
3698 | * memcg_drain_all_list_lrus(). | |
3699 | */ | |
3a06bb78 | 3700 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
8e0a8912 JW |
3701 | css_for_each_descendant_pre(css, &memcg->css) { |
3702 | child = mem_cgroup_from_css(css); | |
3703 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3704 | child->kmemcg_id = parent->kmemcg_id; | |
8e0a8912 | 3705 | } |
3a06bb78 TH |
3706 | rcu_read_unlock(); |
3707 | ||
9bec5c35 | 3708 | memcg_drain_all_list_lrus(kmemcg_id, parent); |
8e0a8912 JW |
3709 | |
3710 | memcg_free_cache_id(kmemcg_id); | |
3711 | } | |
3712 | ||
3713 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3714 | { | |
0b8f73e1 JW |
3715 | /* css_alloc() failed, offlining didn't happen */ |
3716 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
3717 | memcg_offline_kmem(memcg); | |
8e0a8912 | 3718 | } |
d6441637 | 3719 | #else |
0b8f73e1 | 3720 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
3721 | { |
3722 | return 0; | |
3723 | } | |
3724 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
3725 | { | |
3726 | } | |
3727 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3728 | { | |
3729 | } | |
84c07d11 | 3730 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 3731 | |
bbec2e15 RG |
3732 | static int memcg_update_kmem_max(struct mem_cgroup *memcg, |
3733 | unsigned long max) | |
d6441637 | 3734 | { |
b313aeee | 3735 | int ret; |
127424c8 | 3736 | |
bbec2e15 RG |
3737 | mutex_lock(&memcg_max_mutex); |
3738 | ret = page_counter_set_max(&memcg->kmem, max); | |
3739 | mutex_unlock(&memcg_max_mutex); | |
127424c8 | 3740 | return ret; |
d6441637 | 3741 | } |
510fc4e1 | 3742 | |
bbec2e15 | 3743 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
3744 | { |
3745 | int ret; | |
3746 | ||
bbec2e15 | 3747 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 3748 | |
bbec2e15 | 3749 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
3750 | if (ret) |
3751 | goto out; | |
3752 | ||
0db15298 | 3753 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
3754 | /* |
3755 | * The active flag needs to be written after the static_key | |
3756 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
3757 | * function is the last one to run. See mem_cgroup_sk_alloc() |
3758 | * for details, and note that we don't mark any socket as | |
3759 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
3760 | * |
3761 | * We need to do this, because static_keys will span multiple | |
3762 | * sites, but we can't control their order. If we mark a socket | |
3763 | * as accounted, but the accounting functions are not patched in | |
3764 | * yet, we'll lose accounting. | |
3765 | * | |
2d758073 | 3766 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
3767 | * because when this value change, the code to process it is not |
3768 | * patched in yet. | |
3769 | */ | |
3770 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 3771 | memcg->tcpmem_active = true; |
d55f90bf VD |
3772 | } |
3773 | out: | |
bbec2e15 | 3774 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
3775 | return ret; |
3776 | } | |
d55f90bf | 3777 | |
628f4235 KH |
3778 | /* |
3779 | * The user of this function is... | |
3780 | * RES_LIMIT. | |
3781 | */ | |
451af504 TH |
3782 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3783 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3784 | { |
451af504 | 3785 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3786 | unsigned long nr_pages; |
628f4235 KH |
3787 | int ret; |
3788 | ||
451af504 | 3789 | buf = strstrip(buf); |
650c5e56 | 3790 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3791 | if (ret) |
3792 | return ret; | |
af36f906 | 3793 | |
3e32cb2e | 3794 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3795 | case RES_LIMIT: |
4b3bde4c BS |
3796 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3797 | ret = -EINVAL; | |
3798 | break; | |
3799 | } | |
3e32cb2e JW |
3800 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3801 | case _MEM: | |
bbec2e15 | 3802 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 3803 | break; |
3e32cb2e | 3804 | case _MEMSWAP: |
bbec2e15 | 3805 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 3806 | break; |
3e32cb2e | 3807 | case _KMEM: |
0158115f MH |
3808 | pr_warn_once("kmem.limit_in_bytes is deprecated and will be removed. " |
3809 | "Please report your usecase to linux-mm@kvack.org if you " | |
3810 | "depend on this functionality.\n"); | |
bbec2e15 | 3811 | ret = memcg_update_kmem_max(memcg, nr_pages); |
3e32cb2e | 3812 | break; |
d55f90bf | 3813 | case _TCP: |
bbec2e15 | 3814 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 3815 | break; |
3e32cb2e | 3816 | } |
296c81d8 | 3817 | break; |
3e32cb2e JW |
3818 | case RES_SOFT_LIMIT: |
3819 | memcg->soft_limit = nr_pages; | |
3820 | ret = 0; | |
628f4235 KH |
3821 | break; |
3822 | } | |
451af504 | 3823 | return ret ?: nbytes; |
8cdea7c0 BS |
3824 | } |
3825 | ||
6770c64e TH |
3826 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3827 | size_t nbytes, loff_t off) | |
c84872e1 | 3828 | { |
6770c64e | 3829 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3830 | struct page_counter *counter; |
c84872e1 | 3831 | |
3e32cb2e JW |
3832 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3833 | case _MEM: | |
3834 | counter = &memcg->memory; | |
3835 | break; | |
3836 | case _MEMSWAP: | |
3837 | counter = &memcg->memsw; | |
3838 | break; | |
3839 | case _KMEM: | |
3840 | counter = &memcg->kmem; | |
3841 | break; | |
d55f90bf | 3842 | case _TCP: |
0db15298 | 3843 | counter = &memcg->tcpmem; |
d55f90bf | 3844 | break; |
3e32cb2e JW |
3845 | default: |
3846 | BUG(); | |
3847 | } | |
af36f906 | 3848 | |
3e32cb2e | 3849 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3850 | case RES_MAX_USAGE: |
3e32cb2e | 3851 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3852 | break; |
3853 | case RES_FAILCNT: | |
3e32cb2e | 3854 | counter->failcnt = 0; |
29f2a4da | 3855 | break; |
3e32cb2e JW |
3856 | default: |
3857 | BUG(); | |
29f2a4da | 3858 | } |
f64c3f54 | 3859 | |
6770c64e | 3860 | return nbytes; |
c84872e1 PE |
3861 | } |
3862 | ||
182446d0 | 3863 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3864 | struct cftype *cft) |
3865 | { | |
182446d0 | 3866 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3867 | } |
3868 | ||
02491447 | 3869 | #ifdef CONFIG_MMU |
182446d0 | 3870 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3871 | struct cftype *cft, u64 val) |
3872 | { | |
182446d0 | 3873 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3874 | |
1dfab5ab | 3875 | if (val & ~MOVE_MASK) |
7dc74be0 | 3876 | return -EINVAL; |
ee5e8472 | 3877 | |
7dc74be0 | 3878 | /* |
ee5e8472 GC |
3879 | * No kind of locking is needed in here, because ->can_attach() will |
3880 | * check this value once in the beginning of the process, and then carry | |
3881 | * on with stale data. This means that changes to this value will only | |
3882 | * affect task migrations starting after the change. | |
7dc74be0 | 3883 | */ |
c0ff4b85 | 3884 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3885 | return 0; |
3886 | } | |
02491447 | 3887 | #else |
182446d0 | 3888 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3889 | struct cftype *cft, u64 val) |
3890 | { | |
3891 | return -ENOSYS; | |
3892 | } | |
3893 | #endif | |
7dc74be0 | 3894 | |
406eb0c9 | 3895 | #ifdef CONFIG_NUMA |
113b7dfd JW |
3896 | |
3897 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
3898 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
3899 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) | |
3900 | ||
3901 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 | 3902 | int nid, unsigned int lru_mask, bool tree) |
113b7dfd | 3903 | { |
867e5e1d | 3904 | struct lruvec *lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); |
113b7dfd JW |
3905 | unsigned long nr = 0; |
3906 | enum lru_list lru; | |
3907 | ||
3908 | VM_BUG_ON((unsigned)nid >= nr_node_ids); | |
3909 | ||
3910 | for_each_lru(lru) { | |
3911 | if (!(BIT(lru) & lru_mask)) | |
3912 | continue; | |
dd8657b6 SB |
3913 | if (tree) |
3914 | nr += lruvec_page_state(lruvec, NR_LRU_BASE + lru); | |
3915 | else | |
3916 | nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru); | |
113b7dfd JW |
3917 | } |
3918 | return nr; | |
3919 | } | |
3920 | ||
3921 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 SB |
3922 | unsigned int lru_mask, |
3923 | bool tree) | |
113b7dfd JW |
3924 | { |
3925 | unsigned long nr = 0; | |
3926 | enum lru_list lru; | |
3927 | ||
3928 | for_each_lru(lru) { | |
3929 | if (!(BIT(lru) & lru_mask)) | |
3930 | continue; | |
dd8657b6 SB |
3931 | if (tree) |
3932 | nr += memcg_page_state(memcg, NR_LRU_BASE + lru); | |
3933 | else | |
3934 | nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru); | |
113b7dfd JW |
3935 | } |
3936 | return nr; | |
3937 | } | |
3938 | ||
2da8ca82 | 3939 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3940 | { |
25485de6 GT |
3941 | struct numa_stat { |
3942 | const char *name; | |
3943 | unsigned int lru_mask; | |
3944 | }; | |
3945 | ||
3946 | static const struct numa_stat stats[] = { | |
3947 | { "total", LRU_ALL }, | |
3948 | { "file", LRU_ALL_FILE }, | |
3949 | { "anon", LRU_ALL_ANON }, | |
3950 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3951 | }; | |
3952 | const struct numa_stat *stat; | |
406eb0c9 | 3953 | int nid; |
aa9694bb | 3954 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
406eb0c9 | 3955 | |
25485de6 | 3956 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
3957 | seq_printf(m, "%s=%lu", stat->name, |
3958 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
3959 | false)); | |
3960 | for_each_node_state(nid, N_MEMORY) | |
3961 | seq_printf(m, " N%d=%lu", nid, | |
3962 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3963 | stat->lru_mask, false)); | |
25485de6 | 3964 | seq_putc(m, '\n'); |
406eb0c9 | 3965 | } |
406eb0c9 | 3966 | |
071aee13 | 3967 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
3968 | |
3969 | seq_printf(m, "hierarchical_%s=%lu", stat->name, | |
3970 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
3971 | true)); | |
3972 | for_each_node_state(nid, N_MEMORY) | |
3973 | seq_printf(m, " N%d=%lu", nid, | |
3974 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3975 | stat->lru_mask, true)); | |
071aee13 | 3976 | seq_putc(m, '\n'); |
406eb0c9 | 3977 | } |
406eb0c9 | 3978 | |
406eb0c9 YH |
3979 | return 0; |
3980 | } | |
3981 | #endif /* CONFIG_NUMA */ | |
3982 | ||
c8713d0b | 3983 | static const unsigned int memcg1_stats[] = { |
0d1c2072 | 3984 | NR_FILE_PAGES, |
be5d0a74 | 3985 | NR_ANON_MAPPED, |
468c3982 JW |
3986 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3987 | NR_ANON_THPS, | |
3988 | #endif | |
c8713d0b JW |
3989 | NR_SHMEM, |
3990 | NR_FILE_MAPPED, | |
3991 | NR_FILE_DIRTY, | |
3992 | NR_WRITEBACK, | |
3993 | MEMCG_SWAP, | |
3994 | }; | |
3995 | ||
3996 | static const char *const memcg1_stat_names[] = { | |
3997 | "cache", | |
3998 | "rss", | |
468c3982 | 3999 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
c8713d0b | 4000 | "rss_huge", |
468c3982 | 4001 | #endif |
c8713d0b JW |
4002 | "shmem", |
4003 | "mapped_file", | |
4004 | "dirty", | |
4005 | "writeback", | |
4006 | "swap", | |
4007 | }; | |
4008 | ||
df0e53d0 | 4009 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 4010 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
4011 | PGPGIN, |
4012 | PGPGOUT, | |
4013 | PGFAULT, | |
4014 | PGMAJFAULT, | |
4015 | }; | |
4016 | ||
2da8ca82 | 4017 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 4018 | { |
aa9694bb | 4019 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3e32cb2e | 4020 | unsigned long memory, memsw; |
af7c4b0e JW |
4021 | struct mem_cgroup *mi; |
4022 | unsigned int i; | |
406eb0c9 | 4023 | |
71cd3113 | 4024 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c | 4025 | |
71cd3113 | 4026 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
468c3982 JW |
4027 | unsigned long nr; |
4028 | ||
71cd3113 | 4029 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 4030 | continue; |
468c3982 JW |
4031 | nr = memcg_page_state_local(memcg, memcg1_stats[i]); |
4032 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
4033 | if (memcg1_stats[i] == NR_ANON_THPS) | |
4034 | nr *= HPAGE_PMD_NR; | |
4035 | #endif | |
4036 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], nr * PAGE_SIZE); | |
1dd3a273 | 4037 | } |
7b854121 | 4038 | |
df0e53d0 | 4039 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d | 4040 | seq_printf(m, "%s %lu\n", vm_event_name(memcg1_events[i]), |
205b20cc | 4041 | memcg_events_local(memcg, memcg1_events[i])); |
af7c4b0e JW |
4042 | |
4043 | for (i = 0; i < NR_LRU_LISTS; i++) | |
ebc5d83d | 4044 | seq_printf(m, "%s %lu\n", lru_list_name(i), |
205b20cc | 4045 | memcg_page_state_local(memcg, NR_LRU_BASE + i) * |
21d89d15 | 4046 | PAGE_SIZE); |
af7c4b0e | 4047 | |
14067bb3 | 4048 | /* Hierarchical information */ |
3e32cb2e JW |
4049 | memory = memsw = PAGE_COUNTER_MAX; |
4050 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
15b42562 CD |
4051 | memory = min(memory, READ_ONCE(mi->memory.max)); |
4052 | memsw = min(memsw, READ_ONCE(mi->memsw.max)); | |
fee7b548 | 4053 | } |
3e32cb2e JW |
4054 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
4055 | (u64)memory * PAGE_SIZE); | |
7941d214 | 4056 | if (do_memsw_account()) |
3e32cb2e JW |
4057 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
4058 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 4059 | |
8de7ecc6 | 4060 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
7de2e9f1 | 4061 | unsigned long nr; |
4062 | ||
71cd3113 | 4063 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 4064 | continue; |
7de2e9f1 | 4065 | nr = memcg_page_state(memcg, memcg1_stats[i]); |
4066 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
4067 | if (memcg1_stats[i] == NR_ANON_THPS) | |
4068 | nr *= HPAGE_PMD_NR; | |
4069 | #endif | |
8de7ecc6 | 4070 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], |
7de2e9f1 | 4071 | (u64)nr * PAGE_SIZE); |
af7c4b0e JW |
4072 | } |
4073 | ||
8de7ecc6 | 4074 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d KK |
4075 | seq_printf(m, "total_%s %llu\n", |
4076 | vm_event_name(memcg1_events[i]), | |
dd923990 | 4077 | (u64)memcg_events(memcg, memcg1_events[i])); |
af7c4b0e | 4078 | |
8de7ecc6 | 4079 | for (i = 0; i < NR_LRU_LISTS; i++) |
ebc5d83d | 4080 | seq_printf(m, "total_%s %llu\n", lru_list_name(i), |
42a30035 JW |
4081 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
4082 | PAGE_SIZE); | |
14067bb3 | 4083 | |
7f016ee8 | 4084 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 4085 | { |
ef8f2327 MG |
4086 | pg_data_t *pgdat; |
4087 | struct mem_cgroup_per_node *mz; | |
1431d4d1 JW |
4088 | unsigned long anon_cost = 0; |
4089 | unsigned long file_cost = 0; | |
7f016ee8 | 4090 | |
ef8f2327 MG |
4091 | for_each_online_pgdat(pgdat) { |
4092 | mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); | |
7f016ee8 | 4093 | |
1431d4d1 JW |
4094 | anon_cost += mz->lruvec.anon_cost; |
4095 | file_cost += mz->lruvec.file_cost; | |
ef8f2327 | 4096 | } |
1431d4d1 JW |
4097 | seq_printf(m, "anon_cost %lu\n", anon_cost); |
4098 | seq_printf(m, "file_cost %lu\n", file_cost); | |
7f016ee8 KM |
4099 | } |
4100 | #endif | |
4101 | ||
d2ceb9b7 KH |
4102 | return 0; |
4103 | } | |
4104 | ||
182446d0 TH |
4105 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
4106 | struct cftype *cft) | |
a7885eb8 | 4107 | { |
182446d0 | 4108 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4109 | |
1f4c025b | 4110 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4111 | } |
4112 | ||
182446d0 TH |
4113 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
4114 | struct cftype *cft, u64 val) | |
a7885eb8 | 4115 | { |
182446d0 | 4116 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4117 | |
3dae7fec | 4118 | if (val > 100) |
a7885eb8 KM |
4119 | return -EINVAL; |
4120 | ||
14208b0e | 4121 | if (css->parent) |
3dae7fec JW |
4122 | memcg->swappiness = val; |
4123 | else | |
4124 | vm_swappiness = val; | |
068b38c1 | 4125 | |
a7885eb8 KM |
4126 | return 0; |
4127 | } | |
4128 | ||
2e72b634 KS |
4129 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4130 | { | |
4131 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 4132 | unsigned long usage; |
2e72b634 KS |
4133 | int i; |
4134 | ||
4135 | rcu_read_lock(); | |
4136 | if (!swap) | |
2c488db2 | 4137 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4138 | else |
2c488db2 | 4139 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4140 | |
4141 | if (!t) | |
4142 | goto unlock; | |
4143 | ||
ce00a967 | 4144 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
4145 | |
4146 | /* | |
748dad36 | 4147 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4148 | * If it's not true, a threshold was crossed after last |
4149 | * call of __mem_cgroup_threshold(). | |
4150 | */ | |
5407a562 | 4151 | i = t->current_threshold; |
2e72b634 KS |
4152 | |
4153 | /* | |
4154 | * Iterate backward over array of thresholds starting from | |
4155 | * current_threshold and check if a threshold is crossed. | |
4156 | * If none of thresholds below usage is crossed, we read | |
4157 | * only one element of the array here. | |
4158 | */ | |
4159 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4160 | eventfd_signal(t->entries[i].eventfd, 1); | |
4161 | ||
4162 | /* i = current_threshold + 1 */ | |
4163 | i++; | |
4164 | ||
4165 | /* | |
4166 | * Iterate forward over array of thresholds starting from | |
4167 | * current_threshold+1 and check if a threshold is crossed. | |
4168 | * If none of thresholds above usage is crossed, we read | |
4169 | * only one element of the array here. | |
4170 | */ | |
4171 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4172 | eventfd_signal(t->entries[i].eventfd, 1); | |
4173 | ||
4174 | /* Update current_threshold */ | |
5407a562 | 4175 | t->current_threshold = i - 1; |
2e72b634 KS |
4176 | unlock: |
4177 | rcu_read_unlock(); | |
4178 | } | |
4179 | ||
4180 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4181 | { | |
ad4ca5f4 KS |
4182 | while (memcg) { |
4183 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 4184 | if (do_memsw_account()) |
ad4ca5f4 KS |
4185 | __mem_cgroup_threshold(memcg, true); |
4186 | ||
4187 | memcg = parent_mem_cgroup(memcg); | |
4188 | } | |
2e72b634 KS |
4189 | } |
4190 | ||
4191 | static int compare_thresholds(const void *a, const void *b) | |
4192 | { | |
4193 | const struct mem_cgroup_threshold *_a = a; | |
4194 | const struct mem_cgroup_threshold *_b = b; | |
4195 | ||
2bff24a3 GT |
4196 | if (_a->threshold > _b->threshold) |
4197 | return 1; | |
4198 | ||
4199 | if (_a->threshold < _b->threshold) | |
4200 | return -1; | |
4201 | ||
4202 | return 0; | |
2e72b634 KS |
4203 | } |
4204 | ||
c0ff4b85 | 4205 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4206 | { |
4207 | struct mem_cgroup_eventfd_list *ev; | |
4208 | ||
2bcf2e92 MH |
4209 | spin_lock(&memcg_oom_lock); |
4210 | ||
c0ff4b85 | 4211 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 4212 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
4213 | |
4214 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
4215 | return 0; |
4216 | } | |
4217 | ||
c0ff4b85 | 4218 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4219 | { |
7d74b06f KH |
4220 | struct mem_cgroup *iter; |
4221 | ||
c0ff4b85 | 4222 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4223 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4224 | } |
4225 | ||
59b6f873 | 4226 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4227 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4228 | { |
2c488db2 KS |
4229 | struct mem_cgroup_thresholds *thresholds; |
4230 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
4231 | unsigned long threshold; |
4232 | unsigned long usage; | |
2c488db2 | 4233 | int i, size, ret; |
2e72b634 | 4234 | |
650c5e56 | 4235 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
4236 | if (ret) |
4237 | return ret; | |
4238 | ||
4239 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4240 | |
05b84301 | 4241 | if (type == _MEM) { |
2c488db2 | 4242 | thresholds = &memcg->thresholds; |
ce00a967 | 4243 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4244 | } else if (type == _MEMSWAP) { |
2c488db2 | 4245 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4246 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4247 | } else |
2e72b634 KS |
4248 | BUG(); |
4249 | ||
2e72b634 | 4250 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4251 | if (thresholds->primary) |
2e72b634 KS |
4252 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4253 | ||
2c488db2 | 4254 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4255 | |
4256 | /* Allocate memory for new array of thresholds */ | |
67b8046f | 4257 | new = kmalloc(struct_size(new, entries, size), GFP_KERNEL); |
2c488db2 | 4258 | if (!new) { |
2e72b634 KS |
4259 | ret = -ENOMEM; |
4260 | goto unlock; | |
4261 | } | |
2c488db2 | 4262 | new->size = size; |
2e72b634 KS |
4263 | |
4264 | /* Copy thresholds (if any) to new array */ | |
e90342e6 GS |
4265 | if (thresholds->primary) |
4266 | memcpy(new->entries, thresholds->primary->entries, | |
4267 | flex_array_size(new, entries, size - 1)); | |
2c488db2 | 4268 | |
2e72b634 | 4269 | /* Add new threshold */ |
2c488db2 KS |
4270 | new->entries[size - 1].eventfd = eventfd; |
4271 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4272 | |
4273 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
61e604e6 | 4274 | sort(new->entries, size, sizeof(*new->entries), |
2e72b634 KS |
4275 | compare_thresholds, NULL); |
4276 | ||
4277 | /* Find current threshold */ | |
2c488db2 | 4278 | new->current_threshold = -1; |
2e72b634 | 4279 | for (i = 0; i < size; i++) { |
748dad36 | 4280 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4281 | /* |
2c488db2 KS |
4282 | * new->current_threshold will not be used until |
4283 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4284 | * it here. |
4285 | */ | |
2c488db2 | 4286 | ++new->current_threshold; |
748dad36 SZ |
4287 | } else |
4288 | break; | |
2e72b634 KS |
4289 | } |
4290 | ||
2c488db2 KS |
4291 | /* Free old spare buffer and save old primary buffer as spare */ |
4292 | kfree(thresholds->spare); | |
4293 | thresholds->spare = thresholds->primary; | |
4294 | ||
4295 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4296 | |
907860ed | 4297 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4298 | synchronize_rcu(); |
4299 | ||
2e72b634 KS |
4300 | unlock: |
4301 | mutex_unlock(&memcg->thresholds_lock); | |
4302 | ||
4303 | return ret; | |
4304 | } | |
4305 | ||
59b6f873 | 4306 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4307 | struct eventfd_ctx *eventfd, const char *args) |
4308 | { | |
59b6f873 | 4309 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4310 | } |
4311 | ||
59b6f873 | 4312 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4313 | struct eventfd_ctx *eventfd, const char *args) |
4314 | { | |
59b6f873 | 4315 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4316 | } |
4317 | ||
59b6f873 | 4318 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4319 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4320 | { |
2c488db2 KS |
4321 | struct mem_cgroup_thresholds *thresholds; |
4322 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 4323 | unsigned long usage; |
7d36665a | 4324 | int i, j, size, entries; |
2e72b634 KS |
4325 | |
4326 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4327 | |
4328 | if (type == _MEM) { | |
2c488db2 | 4329 | thresholds = &memcg->thresholds; |
ce00a967 | 4330 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4331 | } else if (type == _MEMSWAP) { |
2c488db2 | 4332 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4333 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4334 | } else |
2e72b634 KS |
4335 | BUG(); |
4336 | ||
371528ca AV |
4337 | if (!thresholds->primary) |
4338 | goto unlock; | |
4339 | ||
2e72b634 KS |
4340 | /* Check if a threshold crossed before removing */ |
4341 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4342 | ||
4343 | /* Calculate new number of threshold */ | |
7d36665a | 4344 | size = entries = 0; |
2c488db2 KS |
4345 | for (i = 0; i < thresholds->primary->size; i++) { |
4346 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 | 4347 | size++; |
7d36665a CX |
4348 | else |
4349 | entries++; | |
2e72b634 KS |
4350 | } |
4351 | ||
2c488db2 | 4352 | new = thresholds->spare; |
907860ed | 4353 | |
7d36665a CX |
4354 | /* If no items related to eventfd have been cleared, nothing to do */ |
4355 | if (!entries) | |
4356 | goto unlock; | |
4357 | ||
2e72b634 KS |
4358 | /* Set thresholds array to NULL if we don't have thresholds */ |
4359 | if (!size) { | |
2c488db2 KS |
4360 | kfree(new); |
4361 | new = NULL; | |
907860ed | 4362 | goto swap_buffers; |
2e72b634 KS |
4363 | } |
4364 | ||
2c488db2 | 4365 | new->size = size; |
2e72b634 KS |
4366 | |
4367 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4368 | new->current_threshold = -1; |
4369 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4370 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4371 | continue; |
4372 | ||
2c488db2 | 4373 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4374 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4375 | /* |
2c488db2 | 4376 | * new->current_threshold will not be used |
2e72b634 KS |
4377 | * until rcu_assign_pointer(), so it's safe to increment |
4378 | * it here. | |
4379 | */ | |
2c488db2 | 4380 | ++new->current_threshold; |
2e72b634 KS |
4381 | } |
4382 | j++; | |
4383 | } | |
4384 | ||
907860ed | 4385 | swap_buffers: |
2c488db2 KS |
4386 | /* Swap primary and spare array */ |
4387 | thresholds->spare = thresholds->primary; | |
8c757763 | 4388 | |
2c488db2 | 4389 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4390 | |
907860ed | 4391 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4392 | synchronize_rcu(); |
6611d8d7 MC |
4393 | |
4394 | /* If all events are unregistered, free the spare array */ | |
4395 | if (!new) { | |
4396 | kfree(thresholds->spare); | |
4397 | thresholds->spare = NULL; | |
4398 | } | |
371528ca | 4399 | unlock: |
2e72b634 | 4400 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4401 | } |
c1e862c1 | 4402 | |
59b6f873 | 4403 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4404 | struct eventfd_ctx *eventfd) |
4405 | { | |
59b6f873 | 4406 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4407 | } |
4408 | ||
59b6f873 | 4409 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4410 | struct eventfd_ctx *eventfd) |
4411 | { | |
59b6f873 | 4412 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4413 | } |
4414 | ||
59b6f873 | 4415 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4416 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4417 | { |
9490ff27 | 4418 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4419 | |
9490ff27 KH |
4420 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4421 | if (!event) | |
4422 | return -ENOMEM; | |
4423 | ||
1af8efe9 | 4424 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4425 | |
4426 | event->eventfd = eventfd; | |
4427 | list_add(&event->list, &memcg->oom_notify); | |
4428 | ||
4429 | /* already in OOM ? */ | |
c2b42d3c | 4430 | if (memcg->under_oom) |
9490ff27 | 4431 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4432 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4433 | |
4434 | return 0; | |
4435 | } | |
4436 | ||
59b6f873 | 4437 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4438 | struct eventfd_ctx *eventfd) |
9490ff27 | 4439 | { |
9490ff27 | 4440 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4441 | |
1af8efe9 | 4442 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4443 | |
c0ff4b85 | 4444 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4445 | if (ev->eventfd == eventfd) { |
4446 | list_del(&ev->list); | |
4447 | kfree(ev); | |
4448 | } | |
4449 | } | |
4450 | ||
1af8efe9 | 4451 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4452 | } |
4453 | ||
2da8ca82 | 4454 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4455 | { |
aa9694bb | 4456 | struct mem_cgroup *memcg = mem_cgroup_from_seq(sf); |
3c11ecf4 | 4457 | |
791badbd | 4458 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 4459 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
4460 | seq_printf(sf, "oom_kill %lu\n", |
4461 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
4462 | return 0; |
4463 | } | |
4464 | ||
182446d0 | 4465 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4466 | struct cftype *cft, u64 val) |
4467 | { | |
182446d0 | 4468 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4469 | |
4470 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 4471 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4472 | return -EINVAL; |
4473 | ||
c0ff4b85 | 4474 | memcg->oom_kill_disable = val; |
4d845ebf | 4475 | if (!val) |
c0ff4b85 | 4476 | memcg_oom_recover(memcg); |
3dae7fec | 4477 | |
3c11ecf4 KH |
4478 | return 0; |
4479 | } | |
4480 | ||
52ebea74 TH |
4481 | #ifdef CONFIG_CGROUP_WRITEBACK |
4482 | ||
3a8e9ac8 TH |
4483 | #include <trace/events/writeback.h> |
4484 | ||
841710aa TH |
4485 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
4486 | { | |
4487 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
4488 | } | |
4489 | ||
4490 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4491 | { | |
4492 | wb_domain_exit(&memcg->cgwb_domain); | |
4493 | } | |
4494 | ||
2529bb3a TH |
4495 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4496 | { | |
4497 | wb_domain_size_changed(&memcg->cgwb_domain); | |
4498 | } | |
4499 | ||
841710aa TH |
4500 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
4501 | { | |
4502 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4503 | ||
4504 | if (!memcg->css.parent) | |
4505 | return NULL; | |
4506 | ||
4507 | return &memcg->cgwb_domain; | |
4508 | } | |
4509 | ||
0b3d6e6f GT |
4510 | /* |
4511 | * idx can be of type enum memcg_stat_item or node_stat_item. | |
4512 | * Keep in sync with memcg_exact_page(). | |
4513 | */ | |
4514 | static unsigned long memcg_exact_page_state(struct mem_cgroup *memcg, int idx) | |
4515 | { | |
871789d4 | 4516 | long x = atomic_long_read(&memcg->vmstats[idx]); |
0b3d6e6f GT |
4517 | int cpu; |
4518 | ||
4519 | for_each_online_cpu(cpu) | |
871789d4 | 4520 | x += per_cpu_ptr(memcg->vmstats_percpu, cpu)->stat[idx]; |
0b3d6e6f GT |
4521 | if (x < 0) |
4522 | x = 0; | |
4523 | return x; | |
4524 | } | |
4525 | ||
c2aa723a TH |
4526 | /** |
4527 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
4528 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
4529 | * @pfilepages: out parameter for number of file pages |
4530 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
4531 | * @pdirty: out parameter for number of dirty pages |
4532 | * @pwriteback: out parameter for number of pages under writeback | |
4533 | * | |
c5edf9cd TH |
4534 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
4535 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
4536 | * is a bit more involved. | |
c2aa723a | 4537 | * |
c5edf9cd TH |
4538 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
4539 | * headroom is calculated as the lowest headroom of itself and the | |
4540 | * ancestors. Note that this doesn't consider the actual amount of | |
4541 | * available memory in the system. The caller should further cap | |
4542 | * *@pheadroom accordingly. | |
c2aa723a | 4543 | */ |
c5edf9cd TH |
4544 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
4545 | unsigned long *pheadroom, unsigned long *pdirty, | |
4546 | unsigned long *pwriteback) | |
c2aa723a TH |
4547 | { |
4548 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4549 | struct mem_cgroup *parent; | |
c2aa723a | 4550 | |
0b3d6e6f | 4551 | *pdirty = memcg_exact_page_state(memcg, NR_FILE_DIRTY); |
c2aa723a | 4552 | |
0b3d6e6f | 4553 | *pwriteback = memcg_exact_page_state(memcg, NR_WRITEBACK); |
21d89d15 JW |
4554 | *pfilepages = memcg_exact_page_state(memcg, NR_INACTIVE_FILE) + |
4555 | memcg_exact_page_state(memcg, NR_ACTIVE_FILE); | |
c5edf9cd | 4556 | *pheadroom = PAGE_COUNTER_MAX; |
c2aa723a | 4557 | |
c2aa723a | 4558 | while ((parent = parent_mem_cgroup(memcg))) { |
15b42562 | 4559 | unsigned long ceiling = min(READ_ONCE(memcg->memory.max), |
d1663a90 | 4560 | READ_ONCE(memcg->memory.high)); |
c2aa723a TH |
4561 | unsigned long used = page_counter_read(&memcg->memory); |
4562 | ||
c5edf9cd | 4563 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
4564 | memcg = parent; |
4565 | } | |
c2aa723a TH |
4566 | } |
4567 | ||
97b27821 TH |
4568 | /* |
4569 | * Foreign dirty flushing | |
4570 | * | |
4571 | * There's an inherent mismatch between memcg and writeback. The former | |
4572 | * trackes ownership per-page while the latter per-inode. This was a | |
4573 | * deliberate design decision because honoring per-page ownership in the | |
4574 | * writeback path is complicated, may lead to higher CPU and IO overheads | |
4575 | * and deemed unnecessary given that write-sharing an inode across | |
4576 | * different cgroups isn't a common use-case. | |
4577 | * | |
4578 | * Combined with inode majority-writer ownership switching, this works well | |
4579 | * enough in most cases but there are some pathological cases. For | |
4580 | * example, let's say there are two cgroups A and B which keep writing to | |
4581 | * different but confined parts of the same inode. B owns the inode and | |
4582 | * A's memory is limited far below B's. A's dirty ratio can rise enough to | |
4583 | * trigger balance_dirty_pages() sleeps but B's can be low enough to avoid | |
4584 | * triggering background writeback. A will be slowed down without a way to | |
4585 | * make writeback of the dirty pages happen. | |
4586 | * | |
4587 | * Conditions like the above can lead to a cgroup getting repatedly and | |
4588 | * severely throttled after making some progress after each | |
4589 | * dirty_expire_interval while the underyling IO device is almost | |
4590 | * completely idle. | |
4591 | * | |
4592 | * Solving this problem completely requires matching the ownership tracking | |
4593 | * granularities between memcg and writeback in either direction. However, | |
4594 | * the more egregious behaviors can be avoided by simply remembering the | |
4595 | * most recent foreign dirtying events and initiating remote flushes on | |
4596 | * them when local writeback isn't enough to keep the memory clean enough. | |
4597 | * | |
4598 | * The following two functions implement such mechanism. When a foreign | |
4599 | * page - a page whose memcg and writeback ownerships don't match - is | |
4600 | * dirtied, mem_cgroup_track_foreign_dirty() records the inode owning | |
4601 | * bdi_writeback on the page owning memcg. When balance_dirty_pages() | |
4602 | * decides that the memcg needs to sleep due to high dirty ratio, it calls | |
4603 | * mem_cgroup_flush_foreign() which queues writeback on the recorded | |
4604 | * foreign bdi_writebacks which haven't expired. Both the numbers of | |
4605 | * recorded bdi_writebacks and concurrent in-flight foreign writebacks are | |
4606 | * limited to MEMCG_CGWB_FRN_CNT. | |
4607 | * | |
4608 | * The mechanism only remembers IDs and doesn't hold any object references. | |
4609 | * As being wrong occasionally doesn't matter, updates and accesses to the | |
4610 | * records are lockless and racy. | |
4611 | */ | |
4612 | void mem_cgroup_track_foreign_dirty_slowpath(struct page *page, | |
4613 | struct bdi_writeback *wb) | |
4614 | { | |
4615 | struct mem_cgroup *memcg = page->mem_cgroup; | |
4616 | struct memcg_cgwb_frn *frn; | |
4617 | u64 now = get_jiffies_64(); | |
4618 | u64 oldest_at = now; | |
4619 | int oldest = -1; | |
4620 | int i; | |
4621 | ||
3a8e9ac8 TH |
4622 | trace_track_foreign_dirty(page, wb); |
4623 | ||
97b27821 TH |
4624 | /* |
4625 | * Pick the slot to use. If there is already a slot for @wb, keep | |
4626 | * using it. If not replace the oldest one which isn't being | |
4627 | * written out. | |
4628 | */ | |
4629 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4630 | frn = &memcg->cgwb_frn[i]; | |
4631 | if (frn->bdi_id == wb->bdi->id && | |
4632 | frn->memcg_id == wb->memcg_css->id) | |
4633 | break; | |
4634 | if (time_before64(frn->at, oldest_at) && | |
4635 | atomic_read(&frn->done.cnt) == 1) { | |
4636 | oldest = i; | |
4637 | oldest_at = frn->at; | |
4638 | } | |
4639 | } | |
4640 | ||
4641 | if (i < MEMCG_CGWB_FRN_CNT) { | |
4642 | /* | |
4643 | * Re-using an existing one. Update timestamp lazily to | |
4644 | * avoid making the cacheline hot. We want them to be | |
4645 | * reasonably up-to-date and significantly shorter than | |
4646 | * dirty_expire_interval as that's what expires the record. | |
4647 | * Use the shorter of 1s and dirty_expire_interval / 8. | |
4648 | */ | |
4649 | unsigned long update_intv = | |
4650 | min_t(unsigned long, HZ, | |
4651 | msecs_to_jiffies(dirty_expire_interval * 10) / 8); | |
4652 | ||
4653 | if (time_before64(frn->at, now - update_intv)) | |
4654 | frn->at = now; | |
4655 | } else if (oldest >= 0) { | |
4656 | /* replace the oldest free one */ | |
4657 | frn = &memcg->cgwb_frn[oldest]; | |
4658 | frn->bdi_id = wb->bdi->id; | |
4659 | frn->memcg_id = wb->memcg_css->id; | |
4660 | frn->at = now; | |
4661 | } | |
4662 | } | |
4663 | ||
4664 | /* issue foreign writeback flushes for recorded foreign dirtying events */ | |
4665 | void mem_cgroup_flush_foreign(struct bdi_writeback *wb) | |
4666 | { | |
4667 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4668 | unsigned long intv = msecs_to_jiffies(dirty_expire_interval * 10); | |
4669 | u64 now = jiffies_64; | |
4670 | int i; | |
4671 | ||
4672 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4673 | struct memcg_cgwb_frn *frn = &memcg->cgwb_frn[i]; | |
4674 | ||
4675 | /* | |
4676 | * If the record is older than dirty_expire_interval, | |
4677 | * writeback on it has already started. No need to kick it | |
4678 | * off again. Also, don't start a new one if there's | |
4679 | * already one in flight. | |
4680 | */ | |
4681 | if (time_after64(frn->at, now - intv) && | |
4682 | atomic_read(&frn->done.cnt) == 1) { | |
4683 | frn->at = 0; | |
3a8e9ac8 | 4684 | trace_flush_foreign(wb, frn->bdi_id, frn->memcg_id); |
97b27821 TH |
4685 | cgroup_writeback_by_id(frn->bdi_id, frn->memcg_id, 0, |
4686 | WB_REASON_FOREIGN_FLUSH, | |
4687 | &frn->done); | |
4688 | } | |
4689 | } | |
4690 | } | |
4691 | ||
841710aa TH |
4692 | #else /* CONFIG_CGROUP_WRITEBACK */ |
4693 | ||
4694 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
4695 | { | |
4696 | return 0; | |
4697 | } | |
4698 | ||
4699 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4700 | { | |
4701 | } | |
4702 | ||
2529bb3a TH |
4703 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4704 | { | |
4705 | } | |
4706 | ||
52ebea74 TH |
4707 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
4708 | ||
3bc942f3 TH |
4709 | /* |
4710 | * DO NOT USE IN NEW FILES. | |
4711 | * | |
4712 | * "cgroup.event_control" implementation. | |
4713 | * | |
4714 | * This is way over-engineered. It tries to support fully configurable | |
4715 | * events for each user. Such level of flexibility is completely | |
4716 | * unnecessary especially in the light of the planned unified hierarchy. | |
4717 | * | |
4718 | * Please deprecate this and replace with something simpler if at all | |
4719 | * possible. | |
4720 | */ | |
4721 | ||
79bd9814 TH |
4722 | /* |
4723 | * Unregister event and free resources. | |
4724 | * | |
4725 | * Gets called from workqueue. | |
4726 | */ | |
3bc942f3 | 4727 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4728 | { |
3bc942f3 TH |
4729 | struct mem_cgroup_event *event = |
4730 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4731 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4732 | |
4733 | remove_wait_queue(event->wqh, &event->wait); | |
4734 | ||
59b6f873 | 4735 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4736 | |
4737 | /* Notify userspace the event is going away. */ | |
4738 | eventfd_signal(event->eventfd, 1); | |
4739 | ||
4740 | eventfd_ctx_put(event->eventfd); | |
4741 | kfree(event); | |
59b6f873 | 4742 | css_put(&memcg->css); |
79bd9814 TH |
4743 | } |
4744 | ||
4745 | /* | |
a9a08845 | 4746 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
4747 | * |
4748 | * Called with wqh->lock held and interrupts disabled. | |
4749 | */ | |
ac6424b9 | 4750 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 4751 | int sync, void *key) |
79bd9814 | 4752 | { |
3bc942f3 TH |
4753 | struct mem_cgroup_event *event = |
4754 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4755 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 4756 | __poll_t flags = key_to_poll(key); |
79bd9814 | 4757 | |
a9a08845 | 4758 | if (flags & EPOLLHUP) { |
79bd9814 TH |
4759 | /* |
4760 | * If the event has been detached at cgroup removal, we | |
4761 | * can simply return knowing the other side will cleanup | |
4762 | * for us. | |
4763 | * | |
4764 | * We can't race against event freeing since the other | |
4765 | * side will require wqh->lock via remove_wait_queue(), | |
4766 | * which we hold. | |
4767 | */ | |
fba94807 | 4768 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4769 | if (!list_empty(&event->list)) { |
4770 | list_del_init(&event->list); | |
4771 | /* | |
4772 | * We are in atomic context, but cgroup_event_remove() | |
4773 | * may sleep, so we have to call it in workqueue. | |
4774 | */ | |
4775 | schedule_work(&event->remove); | |
4776 | } | |
fba94807 | 4777 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4778 | } |
4779 | ||
4780 | return 0; | |
4781 | } | |
4782 | ||
3bc942f3 | 4783 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4784 | wait_queue_head_t *wqh, poll_table *pt) |
4785 | { | |
3bc942f3 TH |
4786 | struct mem_cgroup_event *event = |
4787 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4788 | |
4789 | event->wqh = wqh; | |
4790 | add_wait_queue(wqh, &event->wait); | |
4791 | } | |
4792 | ||
4793 | /* | |
3bc942f3 TH |
4794 | * DO NOT USE IN NEW FILES. |
4795 | * | |
79bd9814 TH |
4796 | * Parse input and register new cgroup event handler. |
4797 | * | |
4798 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4799 | * Interpretation of args is defined by control file implementation. | |
4800 | */ | |
451af504 TH |
4801 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4802 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4803 | { |
451af504 | 4804 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4805 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4806 | struct mem_cgroup_event *event; |
79bd9814 TH |
4807 | struct cgroup_subsys_state *cfile_css; |
4808 | unsigned int efd, cfd; | |
4809 | struct fd efile; | |
4810 | struct fd cfile; | |
fba94807 | 4811 | const char *name; |
79bd9814 TH |
4812 | char *endp; |
4813 | int ret; | |
4814 | ||
451af504 TH |
4815 | buf = strstrip(buf); |
4816 | ||
4817 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4818 | if (*endp != ' ') |
4819 | return -EINVAL; | |
451af504 | 4820 | buf = endp + 1; |
79bd9814 | 4821 | |
451af504 | 4822 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4823 | if ((*endp != ' ') && (*endp != '\0')) |
4824 | return -EINVAL; | |
451af504 | 4825 | buf = endp + 1; |
79bd9814 TH |
4826 | |
4827 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4828 | if (!event) | |
4829 | return -ENOMEM; | |
4830 | ||
59b6f873 | 4831 | event->memcg = memcg; |
79bd9814 | 4832 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4833 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4834 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4835 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4836 | |
4837 | efile = fdget(efd); | |
4838 | if (!efile.file) { | |
4839 | ret = -EBADF; | |
4840 | goto out_kfree; | |
4841 | } | |
4842 | ||
4843 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4844 | if (IS_ERR(event->eventfd)) { | |
4845 | ret = PTR_ERR(event->eventfd); | |
4846 | goto out_put_efile; | |
4847 | } | |
4848 | ||
4849 | cfile = fdget(cfd); | |
4850 | if (!cfile.file) { | |
4851 | ret = -EBADF; | |
4852 | goto out_put_eventfd; | |
4853 | } | |
4854 | ||
4855 | /* the process need read permission on control file */ | |
4856 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
4857 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
4858 | if (ret < 0) | |
4859 | goto out_put_cfile; | |
4860 | ||
fba94807 TH |
4861 | /* |
4862 | * Determine the event callbacks and set them in @event. This used | |
4863 | * to be done via struct cftype but cgroup core no longer knows | |
4864 | * about these events. The following is crude but the whole thing | |
4865 | * is for compatibility anyway. | |
3bc942f3 TH |
4866 | * |
4867 | * DO NOT ADD NEW FILES. | |
fba94807 | 4868 | */ |
b583043e | 4869 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4870 | |
4871 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4872 | event->register_event = mem_cgroup_usage_register_event; | |
4873 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4874 | } else if (!strcmp(name, "memory.oom_control")) { | |
4875 | event->register_event = mem_cgroup_oom_register_event; | |
4876 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4877 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4878 | event->register_event = vmpressure_register_event; | |
4879 | event->unregister_event = vmpressure_unregister_event; | |
4880 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4881 | event->register_event = memsw_cgroup_usage_register_event; |
4882 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4883 | } else { |
4884 | ret = -EINVAL; | |
4885 | goto out_put_cfile; | |
4886 | } | |
4887 | ||
79bd9814 | 4888 | /* |
b5557c4c TH |
4889 | * Verify @cfile should belong to @css. Also, remaining events are |
4890 | * automatically removed on cgroup destruction but the removal is | |
4891 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4892 | */ |
b583043e | 4893 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4894 | &memory_cgrp_subsys); |
79bd9814 | 4895 | ret = -EINVAL; |
5a17f543 | 4896 | if (IS_ERR(cfile_css)) |
79bd9814 | 4897 | goto out_put_cfile; |
5a17f543 TH |
4898 | if (cfile_css != css) { |
4899 | css_put(cfile_css); | |
79bd9814 | 4900 | goto out_put_cfile; |
5a17f543 | 4901 | } |
79bd9814 | 4902 | |
451af504 | 4903 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4904 | if (ret) |
4905 | goto out_put_css; | |
4906 | ||
9965ed17 | 4907 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 4908 | |
fba94807 TH |
4909 | spin_lock(&memcg->event_list_lock); |
4910 | list_add(&event->list, &memcg->event_list); | |
4911 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4912 | |
4913 | fdput(cfile); | |
4914 | fdput(efile); | |
4915 | ||
451af504 | 4916 | return nbytes; |
79bd9814 TH |
4917 | |
4918 | out_put_css: | |
b5557c4c | 4919 | css_put(css); |
79bd9814 TH |
4920 | out_put_cfile: |
4921 | fdput(cfile); | |
4922 | out_put_eventfd: | |
4923 | eventfd_ctx_put(event->eventfd); | |
4924 | out_put_efile: | |
4925 | fdput(efile); | |
4926 | out_kfree: | |
4927 | kfree(event); | |
4928 | ||
4929 | return ret; | |
4930 | } | |
4931 | ||
241994ed | 4932 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4933 | { |
0eea1030 | 4934 | .name = "usage_in_bytes", |
8c7c6e34 | 4935 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4936 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4937 | }, |
c84872e1 PE |
4938 | { |
4939 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4940 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4941 | .write = mem_cgroup_reset, |
791badbd | 4942 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4943 | }, |
8cdea7c0 | 4944 | { |
0eea1030 | 4945 | .name = "limit_in_bytes", |
8c7c6e34 | 4946 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4947 | .write = mem_cgroup_write, |
791badbd | 4948 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4949 | }, |
296c81d8 BS |
4950 | { |
4951 | .name = "soft_limit_in_bytes", | |
4952 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4953 | .write = mem_cgroup_write, |
791badbd | 4954 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4955 | }, |
8cdea7c0 BS |
4956 | { |
4957 | .name = "failcnt", | |
8c7c6e34 | 4958 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4959 | .write = mem_cgroup_reset, |
791badbd | 4960 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4961 | }, |
d2ceb9b7 KH |
4962 | { |
4963 | .name = "stat", | |
2da8ca82 | 4964 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4965 | }, |
c1e862c1 KH |
4966 | { |
4967 | .name = "force_empty", | |
6770c64e | 4968 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4969 | }, |
18f59ea7 BS |
4970 | { |
4971 | .name = "use_hierarchy", | |
4972 | .write_u64 = mem_cgroup_hierarchy_write, | |
4973 | .read_u64 = mem_cgroup_hierarchy_read, | |
4974 | }, | |
79bd9814 | 4975 | { |
3bc942f3 | 4976 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4977 | .write = memcg_write_event_control, |
7dbdb199 | 4978 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 4979 | }, |
a7885eb8 KM |
4980 | { |
4981 | .name = "swappiness", | |
4982 | .read_u64 = mem_cgroup_swappiness_read, | |
4983 | .write_u64 = mem_cgroup_swappiness_write, | |
4984 | }, | |
7dc74be0 DN |
4985 | { |
4986 | .name = "move_charge_at_immigrate", | |
4987 | .read_u64 = mem_cgroup_move_charge_read, | |
4988 | .write_u64 = mem_cgroup_move_charge_write, | |
4989 | }, | |
9490ff27 KH |
4990 | { |
4991 | .name = "oom_control", | |
2da8ca82 | 4992 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4993 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4994 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4995 | }, | |
70ddf637 AV |
4996 | { |
4997 | .name = "pressure_level", | |
70ddf637 | 4998 | }, |
406eb0c9 YH |
4999 | #ifdef CONFIG_NUMA |
5000 | { | |
5001 | .name = "numa_stat", | |
2da8ca82 | 5002 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
5003 | }, |
5004 | #endif | |
510fc4e1 GC |
5005 | { |
5006 | .name = "kmem.limit_in_bytes", | |
5007 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 5008 | .write = mem_cgroup_write, |
791badbd | 5009 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5010 | }, |
5011 | { | |
5012 | .name = "kmem.usage_in_bytes", | |
5013 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 5014 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5015 | }, |
5016 | { | |
5017 | .name = "kmem.failcnt", | |
5018 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 5019 | .write = mem_cgroup_reset, |
791badbd | 5020 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5021 | }, |
5022 | { | |
5023 | .name = "kmem.max_usage_in_bytes", | |
5024 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 5025 | .write = mem_cgroup_reset, |
791badbd | 5026 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 5027 | }, |
a87425a3 YS |
5028 | #if defined(CONFIG_MEMCG_KMEM) && \ |
5029 | (defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)) | |
749c5415 GC |
5030 | { |
5031 | .name = "kmem.slabinfo", | |
b047501c | 5032 | .seq_show = memcg_slab_show, |
749c5415 GC |
5033 | }, |
5034 | #endif | |
d55f90bf VD |
5035 | { |
5036 | .name = "kmem.tcp.limit_in_bytes", | |
5037 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
5038 | .write = mem_cgroup_write, | |
5039 | .read_u64 = mem_cgroup_read_u64, | |
5040 | }, | |
5041 | { | |
5042 | .name = "kmem.tcp.usage_in_bytes", | |
5043 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
5044 | .read_u64 = mem_cgroup_read_u64, | |
5045 | }, | |
5046 | { | |
5047 | .name = "kmem.tcp.failcnt", | |
5048 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
5049 | .write = mem_cgroup_reset, | |
5050 | .read_u64 = mem_cgroup_read_u64, | |
5051 | }, | |
5052 | { | |
5053 | .name = "kmem.tcp.max_usage_in_bytes", | |
5054 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
5055 | .write = mem_cgroup_reset, | |
5056 | .read_u64 = mem_cgroup_read_u64, | |
5057 | }, | |
6bc10349 | 5058 | { }, /* terminate */ |
af36f906 | 5059 | }; |
8c7c6e34 | 5060 | |
73f576c0 JW |
5061 | /* |
5062 | * Private memory cgroup IDR | |
5063 | * | |
5064 | * Swap-out records and page cache shadow entries need to store memcg | |
5065 | * references in constrained space, so we maintain an ID space that is | |
5066 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
5067 | * memory-controlled cgroups to 64k. | |
5068 | * | |
b8f2935f | 5069 | * However, there usually are many references to the offline CSS after |
73f576c0 JW |
5070 | * the cgroup has been destroyed, such as page cache or reclaimable |
5071 | * slab objects, that don't need to hang on to the ID. We want to keep | |
5072 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
5073 | * relatively small ID space and prevent the creation of new cgroups | |
5074 | * even when there are much fewer than 64k cgroups - possibly none. | |
5075 | * | |
5076 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
5077 | * be freed and recycled when it's no longer needed, which is usually | |
5078 | * when the CSS is offlined. | |
5079 | * | |
5080 | * The only exception to that are records of swapped out tmpfs/shmem | |
5081 | * pages that need to be attributed to live ancestors on swapin. But | |
5082 | * those references are manageable from userspace. | |
5083 | */ | |
5084 | ||
5085 | static DEFINE_IDR(mem_cgroup_idr); | |
5086 | ||
7e97de0b KT |
5087 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
5088 | { | |
5089 | if (memcg->id.id > 0) { | |
5090 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
5091 | memcg->id.id = 0; | |
5092 | } | |
5093 | } | |
5094 | ||
c1514c0a VF |
5095 | static void __maybe_unused mem_cgroup_id_get_many(struct mem_cgroup *memcg, |
5096 | unsigned int n) | |
73f576c0 | 5097 | { |
1c2d479a | 5098 | refcount_add(n, &memcg->id.ref); |
73f576c0 JW |
5099 | } |
5100 | ||
615d66c3 | 5101 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 5102 | { |
1c2d479a | 5103 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 5104 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
5105 | |
5106 | /* Memcg ID pins CSS */ | |
5107 | css_put(&memcg->css); | |
5108 | } | |
5109 | } | |
5110 | ||
615d66c3 VD |
5111 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) |
5112 | { | |
5113 | mem_cgroup_id_put_many(memcg, 1); | |
5114 | } | |
5115 | ||
73f576c0 JW |
5116 | /** |
5117 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
5118 | * @id: the memcg id to look up | |
5119 | * | |
5120 | * Caller must hold rcu_read_lock(). | |
5121 | */ | |
5122 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
5123 | { | |
5124 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5125 | return idr_find(&mem_cgroup_idr, id); | |
5126 | } | |
5127 | ||
ef8f2327 | 5128 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
5129 | { |
5130 | struct mem_cgroup_per_node *pn; | |
ef8f2327 | 5131 | int tmp = node; |
1ecaab2b KH |
5132 | /* |
5133 | * This routine is called against possible nodes. | |
5134 | * But it's BUG to call kmalloc() against offline node. | |
5135 | * | |
5136 | * TODO: this routine can waste much memory for nodes which will | |
5137 | * never be onlined. It's better to use memory hotplug callback | |
5138 | * function. | |
5139 | */ | |
41e3355d KH |
5140 | if (!node_state(node, N_NORMAL_MEMORY)) |
5141 | tmp = -1; | |
17295c88 | 5142 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
5143 | if (!pn) |
5144 | return 1; | |
1ecaab2b | 5145 | |
3e38e0aa RG |
5146 | pn->lruvec_stat_local = alloc_percpu_gfp(struct lruvec_stat, |
5147 | GFP_KERNEL_ACCOUNT); | |
815744d7 JW |
5148 | if (!pn->lruvec_stat_local) { |
5149 | kfree(pn); | |
5150 | return 1; | |
5151 | } | |
5152 | ||
3e38e0aa RG |
5153 | pn->lruvec_stat_cpu = alloc_percpu_gfp(struct lruvec_stat, |
5154 | GFP_KERNEL_ACCOUNT); | |
a983b5eb | 5155 | if (!pn->lruvec_stat_cpu) { |
815744d7 | 5156 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c JW |
5157 | kfree(pn); |
5158 | return 1; | |
5159 | } | |
5160 | ||
ef8f2327 MG |
5161 | lruvec_init(&pn->lruvec); |
5162 | pn->usage_in_excess = 0; | |
5163 | pn->on_tree = false; | |
5164 | pn->memcg = memcg; | |
5165 | ||
54f72fe0 | 5166 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
5167 | return 0; |
5168 | } | |
5169 | ||
ef8f2327 | 5170 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 5171 | { |
00f3ca2c JW |
5172 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
5173 | ||
4eaf431f MH |
5174 | if (!pn) |
5175 | return; | |
5176 | ||
a983b5eb | 5177 | free_percpu(pn->lruvec_stat_cpu); |
815744d7 | 5178 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c | 5179 | kfree(pn); |
1ecaab2b KH |
5180 | } |
5181 | ||
40e952f9 | 5182 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 5183 | { |
c8b2a36f | 5184 | int node; |
59927fb9 | 5185 | |
c8b2a36f | 5186 | for_each_node(node) |
ef8f2327 | 5187 | free_mem_cgroup_per_node_info(memcg, node); |
871789d4 | 5188 | free_percpu(memcg->vmstats_percpu); |
815744d7 | 5189 | free_percpu(memcg->vmstats_local); |
8ff69e2c | 5190 | kfree(memcg); |
59927fb9 | 5191 | } |
3afe36b1 | 5192 | |
40e952f9 TE |
5193 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
5194 | { | |
5195 | memcg_wb_domain_exit(memcg); | |
7961eee3 SB |
5196 | /* |
5197 | * Flush percpu vmstats and vmevents to guarantee the value correctness | |
5198 | * on parent's and all ancestor levels. | |
5199 | */ | |
4a87e2a2 | 5200 | memcg_flush_percpu_vmstats(memcg); |
7961eee3 | 5201 | memcg_flush_percpu_vmevents(memcg); |
40e952f9 TE |
5202 | __mem_cgroup_free(memcg); |
5203 | } | |
5204 | ||
0b8f73e1 | 5205 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 5206 | { |
d142e3e6 | 5207 | struct mem_cgroup *memcg; |
b9726c26 | 5208 | unsigned int size; |
6d12e2d8 | 5209 | int node; |
97b27821 | 5210 | int __maybe_unused i; |
11d67612 | 5211 | long error = -ENOMEM; |
8cdea7c0 | 5212 | |
0b8f73e1 JW |
5213 | size = sizeof(struct mem_cgroup); |
5214 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
5215 | ||
5216 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 5217 | if (!memcg) |
11d67612 | 5218 | return ERR_PTR(error); |
0b8f73e1 | 5219 | |
73f576c0 JW |
5220 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
5221 | 1, MEM_CGROUP_ID_MAX, | |
5222 | GFP_KERNEL); | |
11d67612 YS |
5223 | if (memcg->id.id < 0) { |
5224 | error = memcg->id.id; | |
73f576c0 | 5225 | goto fail; |
11d67612 | 5226 | } |
73f576c0 | 5227 | |
3e38e0aa RG |
5228 | memcg->vmstats_local = alloc_percpu_gfp(struct memcg_vmstats_percpu, |
5229 | GFP_KERNEL_ACCOUNT); | |
815744d7 JW |
5230 | if (!memcg->vmstats_local) |
5231 | goto fail; | |
5232 | ||
3e38e0aa RG |
5233 | memcg->vmstats_percpu = alloc_percpu_gfp(struct memcg_vmstats_percpu, |
5234 | GFP_KERNEL_ACCOUNT); | |
871789d4 | 5235 | if (!memcg->vmstats_percpu) |
0b8f73e1 | 5236 | goto fail; |
78fb7466 | 5237 | |
3ed28fa1 | 5238 | for_each_node(node) |
ef8f2327 | 5239 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 5240 | goto fail; |
f64c3f54 | 5241 | |
0b8f73e1 JW |
5242 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
5243 | goto fail; | |
28dbc4b6 | 5244 | |
f7e1cb6e | 5245 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 | 5246 | INIT_LIST_HEAD(&memcg->oom_notify); |
d142e3e6 GC |
5247 | mutex_init(&memcg->thresholds_lock); |
5248 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 5249 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
5250 | INIT_LIST_HEAD(&memcg->event_list); |
5251 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 5252 | memcg->socket_pressure = jiffies; |
84c07d11 | 5253 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 5254 | memcg->kmemcg_id = -1; |
bf4f0599 | 5255 | INIT_LIST_HEAD(&memcg->objcg_list); |
900a38f0 | 5256 | #endif |
52ebea74 TH |
5257 | #ifdef CONFIG_CGROUP_WRITEBACK |
5258 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
97b27821 TH |
5259 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) |
5260 | memcg->cgwb_frn[i].done = | |
5261 | __WB_COMPLETION_INIT(&memcg_cgwb_frn_waitq); | |
87eaceb3 YS |
5262 | #endif |
5263 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5264 | spin_lock_init(&memcg->deferred_split_queue.split_queue_lock); | |
5265 | INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue); | |
5266 | memcg->deferred_split_queue.split_queue_len = 0; | |
52ebea74 | 5267 | #endif |
73f576c0 | 5268 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
5269 | return memcg; |
5270 | fail: | |
7e97de0b | 5271 | mem_cgroup_id_remove(memcg); |
40e952f9 | 5272 | __mem_cgroup_free(memcg); |
11d67612 | 5273 | return ERR_PTR(error); |
d142e3e6 GC |
5274 | } |
5275 | ||
0b8f73e1 JW |
5276 | static struct cgroup_subsys_state * __ref |
5277 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 5278 | { |
0b8f73e1 | 5279 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
b87d8cef | 5280 | struct mem_cgroup *memcg, *old_memcg; |
0b8f73e1 | 5281 | long error = -ENOMEM; |
d142e3e6 | 5282 | |
b87d8cef | 5283 | old_memcg = set_active_memcg(parent); |
0b8f73e1 | 5284 | memcg = mem_cgroup_alloc(); |
b87d8cef | 5285 | set_active_memcg(old_memcg); |
11d67612 YS |
5286 | if (IS_ERR(memcg)) |
5287 | return ERR_CAST(memcg); | |
d142e3e6 | 5288 | |
d1663a90 | 5289 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
0b8f73e1 | 5290 | memcg->soft_limit = PAGE_COUNTER_MAX; |
4b82ab4f | 5291 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
0b8f73e1 JW |
5292 | if (parent) { |
5293 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
5294 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
bef8620c | 5295 | |
3e32cb2e | 5296 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 5297 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e | 5298 | page_counter_init(&memcg->kmem, &parent->kmem); |
0db15298 | 5299 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 5300 | } else { |
bef8620c RG |
5301 | page_counter_init(&memcg->memory, NULL); |
5302 | page_counter_init(&memcg->swap, NULL); | |
5303 | page_counter_init(&memcg->kmem, NULL); | |
5304 | page_counter_init(&memcg->tcpmem, NULL); | |
d6441637 | 5305 | |
0b8f73e1 JW |
5306 | root_mem_cgroup = memcg; |
5307 | return &memcg->css; | |
5308 | } | |
5309 | ||
bef8620c | 5310 | /* The following stuff does not apply to the root */ |
b313aeee | 5311 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
5312 | if (error) |
5313 | goto fail; | |
127424c8 | 5314 | |
f7e1cb6e | 5315 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5316 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 5317 | |
0b8f73e1 JW |
5318 | return &memcg->css; |
5319 | fail: | |
7e97de0b | 5320 | mem_cgroup_id_remove(memcg); |
0b8f73e1 | 5321 | mem_cgroup_free(memcg); |
11d67612 | 5322 | return ERR_PTR(error); |
0b8f73e1 JW |
5323 | } |
5324 | ||
73f576c0 | 5325 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 5326 | { |
58fa2a55 VD |
5327 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5328 | ||
0a4465d3 KT |
5329 | /* |
5330 | * A memcg must be visible for memcg_expand_shrinker_maps() | |
5331 | * by the time the maps are allocated. So, we allocate maps | |
5332 | * here, when for_each_mem_cgroup() can't skip it. | |
5333 | */ | |
5334 | if (memcg_alloc_shrinker_maps(memcg)) { | |
5335 | mem_cgroup_id_remove(memcg); | |
5336 | return -ENOMEM; | |
5337 | } | |
5338 | ||
73f576c0 | 5339 | /* Online state pins memcg ID, memcg ID pins CSS */ |
1c2d479a | 5340 | refcount_set(&memcg->id.ref, 1); |
73f576c0 | 5341 | css_get(css); |
2f7dd7a4 | 5342 | return 0; |
8cdea7c0 BS |
5343 | } |
5344 | ||
eb95419b | 5345 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 5346 | { |
eb95419b | 5347 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5348 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
5349 | |
5350 | /* | |
5351 | * Unregister events and notify userspace. | |
5352 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
5353 | * directory to avoid race between userspace and kernelspace. | |
5354 | */ | |
fba94807 TH |
5355 | spin_lock(&memcg->event_list_lock); |
5356 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
5357 | list_del_init(&event->list); |
5358 | schedule_work(&event->remove); | |
5359 | } | |
fba94807 | 5360 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 5361 | |
bf8d5d52 | 5362 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5363 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 5364 | |
567e9ab2 | 5365 | memcg_offline_kmem(memcg); |
52ebea74 | 5366 | wb_memcg_offline(memcg); |
73f576c0 | 5367 | |
591edfb1 RG |
5368 | drain_all_stock(memcg); |
5369 | ||
73f576c0 | 5370 | mem_cgroup_id_put(memcg); |
df878fb0 KH |
5371 | } |
5372 | ||
6df38689 VD |
5373 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
5374 | { | |
5375 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5376 | ||
5377 | invalidate_reclaim_iterators(memcg); | |
5378 | } | |
5379 | ||
eb95419b | 5380 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 5381 | { |
eb95419b | 5382 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
97b27821 | 5383 | int __maybe_unused i; |
c268e994 | 5384 | |
97b27821 TH |
5385 | #ifdef CONFIG_CGROUP_WRITEBACK |
5386 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) | |
5387 | wb_wait_for_completion(&memcg->cgwb_frn[i].done); | |
5388 | #endif | |
f7e1cb6e | 5389 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5390 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 5391 | |
0db15298 | 5392 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 5393 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 5394 | |
0b8f73e1 JW |
5395 | vmpressure_cleanup(&memcg->vmpressure); |
5396 | cancel_work_sync(&memcg->high_work); | |
5397 | mem_cgroup_remove_from_trees(memcg); | |
0a4465d3 | 5398 | memcg_free_shrinker_maps(memcg); |
d886f4e4 | 5399 | memcg_free_kmem(memcg); |
0b8f73e1 | 5400 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
5401 | } |
5402 | ||
1ced953b TH |
5403 | /** |
5404 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
5405 | * @css: the target css | |
5406 | * | |
5407 | * Reset the states of the mem_cgroup associated with @css. This is | |
5408 | * invoked when the userland requests disabling on the default hierarchy | |
5409 | * but the memcg is pinned through dependency. The memcg should stop | |
5410 | * applying policies and should revert to the vanilla state as it may be | |
5411 | * made visible again. | |
5412 | * | |
5413 | * The current implementation only resets the essential configurations. | |
5414 | * This needs to be expanded to cover all the visible parts. | |
5415 | */ | |
5416 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
5417 | { | |
5418 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5419 | ||
bbec2e15 RG |
5420 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
5421 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
bbec2e15 RG |
5422 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); |
5423 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 5424 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5425 | page_counter_set_low(&memcg->memory, 0); |
d1663a90 | 5426 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
24d404dc | 5427 | memcg->soft_limit = PAGE_COUNTER_MAX; |
4b82ab4f | 5428 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
2529bb3a | 5429 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
5430 | } |
5431 | ||
02491447 | 5432 | #ifdef CONFIG_MMU |
7dc74be0 | 5433 | /* Handlers for move charge at task migration. */ |
854ffa8d | 5434 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 5435 | { |
05b84301 | 5436 | int ret; |
9476db97 | 5437 | |
d0164adc MG |
5438 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
5439 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 5440 | if (!ret) { |
854ffa8d | 5441 | mc.precharge += count; |
854ffa8d DN |
5442 | return ret; |
5443 | } | |
9476db97 | 5444 | |
3674534b | 5445 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 5446 | while (count--) { |
3674534b | 5447 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 5448 | if (ret) |
38c5d72f | 5449 | return ret; |
854ffa8d | 5450 | mc.precharge++; |
9476db97 | 5451 | cond_resched(); |
854ffa8d | 5452 | } |
9476db97 | 5453 | return 0; |
4ffef5fe DN |
5454 | } |
5455 | ||
4ffef5fe DN |
5456 | union mc_target { |
5457 | struct page *page; | |
02491447 | 5458 | swp_entry_t ent; |
4ffef5fe DN |
5459 | }; |
5460 | ||
4ffef5fe | 5461 | enum mc_target_type { |
8d32ff84 | 5462 | MC_TARGET_NONE = 0, |
4ffef5fe | 5463 | MC_TARGET_PAGE, |
02491447 | 5464 | MC_TARGET_SWAP, |
c733a828 | 5465 | MC_TARGET_DEVICE, |
4ffef5fe DN |
5466 | }; |
5467 | ||
90254a65 DN |
5468 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5469 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5470 | { |
25b2995a | 5471 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5472 | |
90254a65 DN |
5473 | if (!page || !page_mapped(page)) |
5474 | return NULL; | |
5475 | if (PageAnon(page)) { | |
1dfab5ab | 5476 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5477 | return NULL; |
1dfab5ab JW |
5478 | } else { |
5479 | if (!(mc.flags & MOVE_FILE)) | |
5480 | return NULL; | |
5481 | } | |
90254a65 DN |
5482 | if (!get_page_unless_zero(page)) |
5483 | return NULL; | |
5484 | ||
5485 | return page; | |
5486 | } | |
5487 | ||
c733a828 | 5488 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 5489 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 5490 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 5491 | { |
90254a65 DN |
5492 | struct page *page = NULL; |
5493 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5494 | ||
9a137153 | 5495 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5496 | return NULL; |
c733a828 JG |
5497 | |
5498 | /* | |
5499 | * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to | |
5500 | * a device and because they are not accessible by CPU they are store | |
5501 | * as special swap entry in the CPU page table. | |
5502 | */ | |
5503 | if (is_device_private_entry(ent)) { | |
5504 | page = device_private_entry_to_page(ent); | |
5505 | /* | |
5506 | * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have | |
5507 | * a refcount of 1 when free (unlike normal page) | |
5508 | */ | |
5509 | if (!page_ref_add_unless(page, 1, 1)) | |
5510 | return NULL; | |
5511 | return page; | |
5512 | } | |
5513 | ||
9a137153 RC |
5514 | if (non_swap_entry(ent)) |
5515 | return NULL; | |
5516 | ||
4b91355e KH |
5517 | /* |
5518 | * Because lookup_swap_cache() updates some statistics counter, | |
5519 | * we call find_get_page() with swapper_space directly. | |
5520 | */ | |
f6ab1f7f | 5521 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
2d1c4980 | 5522 | entry->val = ent.val; |
90254a65 DN |
5523 | |
5524 | return page; | |
5525 | } | |
4b91355e KH |
5526 | #else |
5527 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 5528 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
5529 | { |
5530 | return NULL; | |
5531 | } | |
5532 | #endif | |
90254a65 | 5533 | |
87946a72 DN |
5534 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5535 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5536 | { | |
87946a72 DN |
5537 | if (!vma->vm_file) /* anonymous vma */ |
5538 | return NULL; | |
1dfab5ab | 5539 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
5540 | return NULL; |
5541 | ||
87946a72 | 5542 | /* page is moved even if it's not RSS of this task(page-faulted). */ |
aa3b1895 | 5543 | /* shmem/tmpfs may report page out on swap: account for that too. */ |
f5df8635 MWO |
5544 | return find_get_incore_page(vma->vm_file->f_mapping, |
5545 | linear_page_index(vma, addr)); | |
87946a72 DN |
5546 | } |
5547 | ||
b1b0deab CG |
5548 | /** |
5549 | * mem_cgroup_move_account - move account of the page | |
5550 | * @page: the page | |
25843c2b | 5551 | * @compound: charge the page as compound or small page |
b1b0deab CG |
5552 | * @from: mem_cgroup which the page is moved from. |
5553 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
5554 | * | |
3ac808fd | 5555 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
5556 | * |
5557 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
5558 | * from old cgroup. | |
5559 | */ | |
5560 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 5561 | bool compound, |
b1b0deab CG |
5562 | struct mem_cgroup *from, |
5563 | struct mem_cgroup *to) | |
5564 | { | |
ae8af438 KK |
5565 | struct lruvec *from_vec, *to_vec; |
5566 | struct pglist_data *pgdat; | |
6c357848 | 5567 | unsigned int nr_pages = compound ? thp_nr_pages(page) : 1; |
b1b0deab CG |
5568 | int ret; |
5569 | ||
5570 | VM_BUG_ON(from == to); | |
5571 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 5572 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
5573 | |
5574 | /* | |
6a93ca8f | 5575 | * Prevent mem_cgroup_migrate() from looking at |
45637bab | 5576 | * page->mem_cgroup of its source page while we change it. |
b1b0deab | 5577 | */ |
f627c2f5 | 5578 | ret = -EBUSY; |
b1b0deab CG |
5579 | if (!trylock_page(page)) |
5580 | goto out; | |
5581 | ||
5582 | ret = -EINVAL; | |
5583 | if (page->mem_cgroup != from) | |
5584 | goto out_unlock; | |
5585 | ||
ae8af438 | 5586 | pgdat = page_pgdat(page); |
867e5e1d JW |
5587 | from_vec = mem_cgroup_lruvec(from, pgdat); |
5588 | to_vec = mem_cgroup_lruvec(to, pgdat); | |
ae8af438 | 5589 | |
abb242f5 | 5590 | lock_page_memcg(page); |
b1b0deab | 5591 | |
be5d0a74 JW |
5592 | if (PageAnon(page)) { |
5593 | if (page_mapped(page)) { | |
5594 | __mod_lruvec_state(from_vec, NR_ANON_MAPPED, -nr_pages); | |
5595 | __mod_lruvec_state(to_vec, NR_ANON_MAPPED, nr_pages); | |
468c3982 JW |
5596 | if (PageTransHuge(page)) { |
5597 | __mod_lruvec_state(from_vec, NR_ANON_THPS, | |
5598 | -nr_pages); | |
5599 | __mod_lruvec_state(to_vec, NR_ANON_THPS, | |
5600 | nr_pages); | |
5601 | } | |
5602 | ||
be5d0a74 JW |
5603 | } |
5604 | } else { | |
0d1c2072 JW |
5605 | __mod_lruvec_state(from_vec, NR_FILE_PAGES, -nr_pages); |
5606 | __mod_lruvec_state(to_vec, NR_FILE_PAGES, nr_pages); | |
5607 | ||
5608 | if (PageSwapBacked(page)) { | |
5609 | __mod_lruvec_state(from_vec, NR_SHMEM, -nr_pages); | |
5610 | __mod_lruvec_state(to_vec, NR_SHMEM, nr_pages); | |
5611 | } | |
5612 | ||
49e50d27 JW |
5613 | if (page_mapped(page)) { |
5614 | __mod_lruvec_state(from_vec, NR_FILE_MAPPED, -nr_pages); | |
5615 | __mod_lruvec_state(to_vec, NR_FILE_MAPPED, nr_pages); | |
5616 | } | |
b1b0deab | 5617 | |
49e50d27 JW |
5618 | if (PageDirty(page)) { |
5619 | struct address_space *mapping = page_mapping(page); | |
c4843a75 | 5620 | |
f56753ac | 5621 | if (mapping_can_writeback(mapping)) { |
49e50d27 JW |
5622 | __mod_lruvec_state(from_vec, NR_FILE_DIRTY, |
5623 | -nr_pages); | |
5624 | __mod_lruvec_state(to_vec, NR_FILE_DIRTY, | |
5625 | nr_pages); | |
5626 | } | |
c4843a75 GT |
5627 | } |
5628 | } | |
5629 | ||
b1b0deab | 5630 | if (PageWriteback(page)) { |
ae8af438 KK |
5631 | __mod_lruvec_state(from_vec, NR_WRITEBACK, -nr_pages); |
5632 | __mod_lruvec_state(to_vec, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
5633 | } |
5634 | ||
5635 | /* | |
abb242f5 JW |
5636 | * All state has been migrated, let's switch to the new memcg. |
5637 | * | |
b1b0deab | 5638 | * It is safe to change page->mem_cgroup here because the page |
abb242f5 JW |
5639 | * is referenced, charged, isolated, and locked: we can't race |
5640 | * with (un)charging, migration, LRU putback, or anything else | |
5641 | * that would rely on a stable page->mem_cgroup. | |
5642 | * | |
5643 | * Note that lock_page_memcg is a memcg lock, not a page lock, | |
5644 | * to save space. As soon as we switch page->mem_cgroup to a | |
5645 | * new memcg that isn't locked, the above state can change | |
5646 | * concurrently again. Make sure we're truly done with it. | |
b1b0deab | 5647 | */ |
abb242f5 | 5648 | smp_mb(); |
b1b0deab | 5649 | |
1a3e1f40 JW |
5650 | css_get(&to->css); |
5651 | css_put(&from->css); | |
5652 | ||
5653 | page->mem_cgroup = to; | |
87eaceb3 | 5654 | |
abb242f5 | 5655 | __unlock_page_memcg(from); |
b1b0deab CG |
5656 | |
5657 | ret = 0; | |
5658 | ||
5659 | local_irq_disable(); | |
3fba69a5 | 5660 | mem_cgroup_charge_statistics(to, page, nr_pages); |
b1b0deab | 5661 | memcg_check_events(to, page); |
3fba69a5 | 5662 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
b1b0deab CG |
5663 | memcg_check_events(from, page); |
5664 | local_irq_enable(); | |
5665 | out_unlock: | |
5666 | unlock_page(page); | |
5667 | out: | |
5668 | return ret; | |
5669 | } | |
5670 | ||
7cf7806c LR |
5671 | /** |
5672 | * get_mctgt_type - get target type of moving charge | |
5673 | * @vma: the vma the pte to be checked belongs | |
5674 | * @addr: the address corresponding to the pte to be checked | |
5675 | * @ptent: the pte to be checked | |
5676 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
5677 | * | |
5678 | * Returns | |
5679 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5680 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5681 | * move charge. if @target is not NULL, the page is stored in target->page | |
5682 | * with extra refcnt got(Callers should handle it). | |
5683 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
5684 | * target for charge migration. if @target is not NULL, the entry is stored | |
5685 | * in target->ent. | |
25b2995a CH |
5686 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PRIVATE |
5687 | * (so ZONE_DEVICE page and thus not on the lru). | |
df6ad698 JG |
5688 | * For now we such page is charge like a regular page would be as for all |
5689 | * intent and purposes it is just special memory taking the place of a | |
5690 | * regular page. | |
c733a828 JG |
5691 | * |
5692 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
5693 | * |
5694 | * Called with pte lock held. | |
5695 | */ | |
5696 | ||
8d32ff84 | 5697 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5698 | unsigned long addr, pte_t ptent, union mc_target *target) |
5699 | { | |
5700 | struct page *page = NULL; | |
8d32ff84 | 5701 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5702 | swp_entry_t ent = { .val = 0 }; |
5703 | ||
5704 | if (pte_present(ptent)) | |
5705 | page = mc_handle_present_pte(vma, addr, ptent); | |
5706 | else if (is_swap_pte(ptent)) | |
48406ef8 | 5707 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 5708 | else if (pte_none(ptent)) |
87946a72 | 5709 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
5710 | |
5711 | if (!page && !ent.val) | |
8d32ff84 | 5712 | return ret; |
02491447 | 5713 | if (page) { |
02491447 | 5714 | /* |
0a31bc97 | 5715 | * Do only loose check w/o serialization. |
1306a85a | 5716 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 5717 | * not under LRU exclusion. |
02491447 | 5718 | */ |
1306a85a | 5719 | if (page->mem_cgroup == mc.from) { |
02491447 | 5720 | ret = MC_TARGET_PAGE; |
25b2995a | 5721 | if (is_device_private_page(page)) |
c733a828 | 5722 | ret = MC_TARGET_DEVICE; |
02491447 DN |
5723 | if (target) |
5724 | target->page = page; | |
5725 | } | |
5726 | if (!ret || !target) | |
5727 | put_page(page); | |
5728 | } | |
3e14a57b HY |
5729 | /* |
5730 | * There is a swap entry and a page doesn't exist or isn't charged. | |
5731 | * But we cannot move a tail-page in a THP. | |
5732 | */ | |
5733 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 5734 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5735 | ret = MC_TARGET_SWAP; |
5736 | if (target) | |
5737 | target->ent = ent; | |
4ffef5fe | 5738 | } |
4ffef5fe DN |
5739 | return ret; |
5740 | } | |
5741 | ||
12724850 NH |
5742 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5743 | /* | |
d6810d73 HY |
5744 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
5745 | * not support them for now. | |
12724850 NH |
5746 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
5747 | */ | |
5748 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5749 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5750 | { | |
5751 | struct page *page = NULL; | |
12724850 NH |
5752 | enum mc_target_type ret = MC_TARGET_NONE; |
5753 | ||
84c3fc4e ZY |
5754 | if (unlikely(is_swap_pmd(pmd))) { |
5755 | VM_BUG_ON(thp_migration_supported() && | |
5756 | !is_pmd_migration_entry(pmd)); | |
5757 | return ret; | |
5758 | } | |
12724850 | 5759 | page = pmd_page(pmd); |
309381fe | 5760 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 5761 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 5762 | return ret; |
1306a85a | 5763 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
5764 | ret = MC_TARGET_PAGE; |
5765 | if (target) { | |
5766 | get_page(page); | |
5767 | target->page = page; | |
5768 | } | |
5769 | } | |
5770 | return ret; | |
5771 | } | |
5772 | #else | |
5773 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5774 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5775 | { | |
5776 | return MC_TARGET_NONE; | |
5777 | } | |
5778 | #endif | |
5779 | ||
4ffef5fe DN |
5780 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5781 | unsigned long addr, unsigned long end, | |
5782 | struct mm_walk *walk) | |
5783 | { | |
26bcd64a | 5784 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5785 | pte_t *pte; |
5786 | spinlock_t *ptl; | |
5787 | ||
b6ec57f4 KS |
5788 | ptl = pmd_trans_huge_lock(pmd, vma); |
5789 | if (ptl) { | |
c733a828 JG |
5790 | /* |
5791 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
25b2995a CH |
5792 | * support transparent huge page with MEMORY_DEVICE_PRIVATE but |
5793 | * this might change. | |
c733a828 | 5794 | */ |
12724850 NH |
5795 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5796 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5797 | spin_unlock(ptl); |
1a5a9906 | 5798 | return 0; |
12724850 | 5799 | } |
03319327 | 5800 | |
45f83cef AA |
5801 | if (pmd_trans_unstable(pmd)) |
5802 | return 0; | |
4ffef5fe DN |
5803 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5804 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5805 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5806 | mc.precharge++; /* increment precharge temporarily */ |
5807 | pte_unmap_unlock(pte - 1, ptl); | |
5808 | cond_resched(); | |
5809 | ||
7dc74be0 DN |
5810 | return 0; |
5811 | } | |
5812 | ||
7b86ac33 CH |
5813 | static const struct mm_walk_ops precharge_walk_ops = { |
5814 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5815 | }; | |
5816 | ||
4ffef5fe DN |
5817 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5818 | { | |
5819 | unsigned long precharge; | |
4ffef5fe | 5820 | |
d8ed45c5 | 5821 | mmap_read_lock(mm); |
7b86ac33 | 5822 | walk_page_range(mm, 0, mm->highest_vm_end, &precharge_walk_ops, NULL); |
d8ed45c5 | 5823 | mmap_read_unlock(mm); |
4ffef5fe DN |
5824 | |
5825 | precharge = mc.precharge; | |
5826 | mc.precharge = 0; | |
5827 | ||
5828 | return precharge; | |
5829 | } | |
5830 | ||
4ffef5fe DN |
5831 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5832 | { | |
dfe076b0 DN |
5833 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5834 | ||
5835 | VM_BUG_ON(mc.moving_task); | |
5836 | mc.moving_task = current; | |
5837 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5838 | } |
5839 | ||
dfe076b0 DN |
5840 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5841 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5842 | { |
2bd9bb20 KH |
5843 | struct mem_cgroup *from = mc.from; |
5844 | struct mem_cgroup *to = mc.to; | |
5845 | ||
4ffef5fe | 5846 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5847 | if (mc.precharge) { |
00501b53 | 5848 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5849 | mc.precharge = 0; |
5850 | } | |
5851 | /* | |
5852 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5853 | * we must uncharge here. | |
5854 | */ | |
5855 | if (mc.moved_charge) { | |
00501b53 | 5856 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5857 | mc.moved_charge = 0; |
4ffef5fe | 5858 | } |
483c30b5 DN |
5859 | /* we must fixup refcnts and charges */ |
5860 | if (mc.moved_swap) { | |
483c30b5 | 5861 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5862 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5863 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5864 | |
615d66c3 VD |
5865 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
5866 | ||
05b84301 | 5867 | /* |
3e32cb2e JW |
5868 | * we charged both to->memory and to->memsw, so we |
5869 | * should uncharge to->memory. | |
05b84301 | 5870 | */ |
ce00a967 | 5871 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5872 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5873 | ||
483c30b5 DN |
5874 | mc.moved_swap = 0; |
5875 | } | |
dfe076b0 DN |
5876 | memcg_oom_recover(from); |
5877 | memcg_oom_recover(to); | |
5878 | wake_up_all(&mc.waitq); | |
5879 | } | |
5880 | ||
5881 | static void mem_cgroup_clear_mc(void) | |
5882 | { | |
264a0ae1 TH |
5883 | struct mm_struct *mm = mc.mm; |
5884 | ||
dfe076b0 DN |
5885 | /* |
5886 | * we must clear moving_task before waking up waiters at the end of | |
5887 | * task migration. | |
5888 | */ | |
5889 | mc.moving_task = NULL; | |
5890 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5891 | spin_lock(&mc.lock); |
4ffef5fe DN |
5892 | mc.from = NULL; |
5893 | mc.to = NULL; | |
264a0ae1 | 5894 | mc.mm = NULL; |
2bd9bb20 | 5895 | spin_unlock(&mc.lock); |
264a0ae1 TH |
5896 | |
5897 | mmput(mm); | |
4ffef5fe DN |
5898 | } |
5899 | ||
1f7dd3e5 | 5900 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5901 | { |
1f7dd3e5 | 5902 | struct cgroup_subsys_state *css; |
eed67d75 | 5903 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 5904 | struct mem_cgroup *from; |
4530eddb | 5905 | struct task_struct *leader, *p; |
9f2115f9 | 5906 | struct mm_struct *mm; |
1dfab5ab | 5907 | unsigned long move_flags; |
9f2115f9 | 5908 | int ret = 0; |
7dc74be0 | 5909 | |
1f7dd3e5 TH |
5910 | /* charge immigration isn't supported on the default hierarchy */ |
5911 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
5912 | return 0; |
5913 | ||
4530eddb TH |
5914 | /* |
5915 | * Multi-process migrations only happen on the default hierarchy | |
5916 | * where charge immigration is not used. Perform charge | |
5917 | * immigration if @tset contains a leader and whine if there are | |
5918 | * multiple. | |
5919 | */ | |
5920 | p = NULL; | |
1f7dd3e5 | 5921 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
5922 | WARN_ON_ONCE(p); |
5923 | p = leader; | |
1f7dd3e5 | 5924 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
5925 | } |
5926 | if (!p) | |
5927 | return 0; | |
5928 | ||
1f7dd3e5 TH |
5929 | /* |
5930 | * We are now commited to this value whatever it is. Changes in this | |
5931 | * tunable will only affect upcoming migrations, not the current one. | |
5932 | * So we need to save it, and keep it going. | |
5933 | */ | |
5934 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
5935 | if (!move_flags) | |
5936 | return 0; | |
5937 | ||
9f2115f9 TH |
5938 | from = mem_cgroup_from_task(p); |
5939 | ||
5940 | VM_BUG_ON(from == memcg); | |
5941 | ||
5942 | mm = get_task_mm(p); | |
5943 | if (!mm) | |
5944 | return 0; | |
5945 | /* We move charges only when we move a owner of the mm */ | |
5946 | if (mm->owner == p) { | |
5947 | VM_BUG_ON(mc.from); | |
5948 | VM_BUG_ON(mc.to); | |
5949 | VM_BUG_ON(mc.precharge); | |
5950 | VM_BUG_ON(mc.moved_charge); | |
5951 | VM_BUG_ON(mc.moved_swap); | |
5952 | ||
5953 | spin_lock(&mc.lock); | |
264a0ae1 | 5954 | mc.mm = mm; |
9f2115f9 TH |
5955 | mc.from = from; |
5956 | mc.to = memcg; | |
5957 | mc.flags = move_flags; | |
5958 | spin_unlock(&mc.lock); | |
5959 | /* We set mc.moving_task later */ | |
5960 | ||
5961 | ret = mem_cgroup_precharge_mc(mm); | |
5962 | if (ret) | |
5963 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
5964 | } else { |
5965 | mmput(mm); | |
7dc74be0 DN |
5966 | } |
5967 | return ret; | |
5968 | } | |
5969 | ||
1f7dd3e5 | 5970 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5971 | { |
4e2f245d JW |
5972 | if (mc.to) |
5973 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5974 | } |
5975 | ||
4ffef5fe DN |
5976 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5977 | unsigned long addr, unsigned long end, | |
5978 | struct mm_walk *walk) | |
7dc74be0 | 5979 | { |
4ffef5fe | 5980 | int ret = 0; |
26bcd64a | 5981 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5982 | pte_t *pte; |
5983 | spinlock_t *ptl; | |
12724850 NH |
5984 | enum mc_target_type target_type; |
5985 | union mc_target target; | |
5986 | struct page *page; | |
4ffef5fe | 5987 | |
b6ec57f4 KS |
5988 | ptl = pmd_trans_huge_lock(pmd, vma); |
5989 | if (ptl) { | |
62ade86a | 5990 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5991 | spin_unlock(ptl); |
12724850 NH |
5992 | return 0; |
5993 | } | |
5994 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5995 | if (target_type == MC_TARGET_PAGE) { | |
5996 | page = target.page; | |
5997 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 5998 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 5999 | mc.from, mc.to)) { |
12724850 NH |
6000 | mc.precharge -= HPAGE_PMD_NR; |
6001 | mc.moved_charge += HPAGE_PMD_NR; | |
6002 | } | |
6003 | putback_lru_page(page); | |
6004 | } | |
6005 | put_page(page); | |
c733a828 JG |
6006 | } else if (target_type == MC_TARGET_DEVICE) { |
6007 | page = target.page; | |
6008 | if (!mem_cgroup_move_account(page, true, | |
6009 | mc.from, mc.to)) { | |
6010 | mc.precharge -= HPAGE_PMD_NR; | |
6011 | mc.moved_charge += HPAGE_PMD_NR; | |
6012 | } | |
6013 | put_page(page); | |
12724850 | 6014 | } |
bf929152 | 6015 | spin_unlock(ptl); |
1a5a9906 | 6016 | return 0; |
12724850 NH |
6017 | } |
6018 | ||
45f83cef AA |
6019 | if (pmd_trans_unstable(pmd)) |
6020 | return 0; | |
4ffef5fe DN |
6021 | retry: |
6022 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
6023 | for (; addr != end; addr += PAGE_SIZE) { | |
6024 | pte_t ptent = *(pte++); | |
c733a828 | 6025 | bool device = false; |
02491447 | 6026 | swp_entry_t ent; |
4ffef5fe DN |
6027 | |
6028 | if (!mc.precharge) | |
6029 | break; | |
6030 | ||
8d32ff84 | 6031 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
6032 | case MC_TARGET_DEVICE: |
6033 | device = true; | |
e4a9bc58 | 6034 | fallthrough; |
4ffef5fe DN |
6035 | case MC_TARGET_PAGE: |
6036 | page = target.page; | |
53f9263b KS |
6037 | /* |
6038 | * We can have a part of the split pmd here. Moving it | |
6039 | * can be done but it would be too convoluted so simply | |
6040 | * ignore such a partial THP and keep it in original | |
6041 | * memcg. There should be somebody mapping the head. | |
6042 | */ | |
6043 | if (PageTransCompound(page)) | |
6044 | goto put; | |
c733a828 | 6045 | if (!device && isolate_lru_page(page)) |
4ffef5fe | 6046 | goto put; |
f627c2f5 KS |
6047 | if (!mem_cgroup_move_account(page, false, |
6048 | mc.from, mc.to)) { | |
4ffef5fe | 6049 | mc.precharge--; |
854ffa8d DN |
6050 | /* we uncharge from mc.from later. */ |
6051 | mc.moved_charge++; | |
4ffef5fe | 6052 | } |
c733a828 JG |
6053 | if (!device) |
6054 | putback_lru_page(page); | |
8d32ff84 | 6055 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
6056 | put_page(page); |
6057 | break; | |
02491447 DN |
6058 | case MC_TARGET_SWAP: |
6059 | ent = target.ent; | |
e91cbb42 | 6060 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6061 | mc.precharge--; |
8d22a935 HD |
6062 | mem_cgroup_id_get_many(mc.to, 1); |
6063 | /* we fixup other refcnts and charges later. */ | |
483c30b5 DN |
6064 | mc.moved_swap++; |
6065 | } | |
02491447 | 6066 | break; |
4ffef5fe DN |
6067 | default: |
6068 | break; | |
6069 | } | |
6070 | } | |
6071 | pte_unmap_unlock(pte - 1, ptl); | |
6072 | cond_resched(); | |
6073 | ||
6074 | if (addr != end) { | |
6075 | /* | |
6076 | * We have consumed all precharges we got in can_attach(). | |
6077 | * We try charge one by one, but don't do any additional | |
6078 | * charges to mc.to if we have failed in charge once in attach() | |
6079 | * phase. | |
6080 | */ | |
854ffa8d | 6081 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6082 | if (!ret) |
6083 | goto retry; | |
6084 | } | |
6085 | ||
6086 | return ret; | |
6087 | } | |
6088 | ||
7b86ac33 CH |
6089 | static const struct mm_walk_ops charge_walk_ops = { |
6090 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6091 | }; | |
6092 | ||
264a0ae1 | 6093 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 6094 | { |
4ffef5fe | 6095 | lru_add_drain_all(); |
312722cb | 6096 | /* |
81f8c3a4 JW |
6097 | * Signal lock_page_memcg() to take the memcg's move_lock |
6098 | * while we're moving its pages to another memcg. Then wait | |
6099 | * for already started RCU-only updates to finish. | |
312722cb JW |
6100 | */ |
6101 | atomic_inc(&mc.from->moving_account); | |
6102 | synchronize_rcu(); | |
dfe076b0 | 6103 | retry: |
d8ed45c5 | 6104 | if (unlikely(!mmap_read_trylock(mc.mm))) { |
dfe076b0 | 6105 | /* |
c1e8d7c6 | 6106 | * Someone who are holding the mmap_lock might be waiting in |
dfe076b0 DN |
6107 | * waitq. So we cancel all extra charges, wake up all waiters, |
6108 | * and retry. Because we cancel precharges, we might not be able | |
6109 | * to move enough charges, but moving charge is a best-effort | |
6110 | * feature anyway, so it wouldn't be a big problem. | |
6111 | */ | |
6112 | __mem_cgroup_clear_mc(); | |
6113 | cond_resched(); | |
6114 | goto retry; | |
6115 | } | |
26bcd64a NH |
6116 | /* |
6117 | * When we have consumed all precharges and failed in doing | |
6118 | * additional charge, the page walk just aborts. | |
6119 | */ | |
7b86ac33 CH |
6120 | walk_page_range(mc.mm, 0, mc.mm->highest_vm_end, &charge_walk_ops, |
6121 | NULL); | |
0247f3f4 | 6122 | |
d8ed45c5 | 6123 | mmap_read_unlock(mc.mm); |
312722cb | 6124 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
6125 | } |
6126 | ||
264a0ae1 | 6127 | static void mem_cgroup_move_task(void) |
67e465a7 | 6128 | { |
264a0ae1 TH |
6129 | if (mc.to) { |
6130 | mem_cgroup_move_charge(); | |
a433658c | 6131 | mem_cgroup_clear_mc(); |
264a0ae1 | 6132 | } |
67e465a7 | 6133 | } |
5cfb80a7 | 6134 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 6135 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6136 | { |
6137 | return 0; | |
6138 | } | |
1f7dd3e5 | 6139 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6140 | { |
6141 | } | |
264a0ae1 | 6142 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
6143 | { |
6144 | } | |
6145 | #endif | |
67e465a7 | 6146 | |
677dc973 CD |
6147 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
6148 | { | |
6149 | if (value == PAGE_COUNTER_MAX) | |
6150 | seq_puts(m, "max\n"); | |
6151 | else | |
6152 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); | |
6153 | ||
6154 | return 0; | |
6155 | } | |
6156 | ||
241994ed JW |
6157 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
6158 | struct cftype *cft) | |
6159 | { | |
f5fc3c5d JW |
6160 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
6161 | ||
6162 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
6163 | } |
6164 | ||
bf8d5d52 RG |
6165 | static int memory_min_show(struct seq_file *m, void *v) |
6166 | { | |
677dc973 CD |
6167 | return seq_puts_memcg_tunable(m, |
6168 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); | |
bf8d5d52 RG |
6169 | } |
6170 | ||
6171 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
6172 | char *buf, size_t nbytes, loff_t off) | |
6173 | { | |
6174 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6175 | unsigned long min; | |
6176 | int err; | |
6177 | ||
6178 | buf = strstrip(buf); | |
6179 | err = page_counter_memparse(buf, "max", &min); | |
6180 | if (err) | |
6181 | return err; | |
6182 | ||
6183 | page_counter_set_min(&memcg->memory, min); | |
6184 | ||
6185 | return nbytes; | |
6186 | } | |
6187 | ||
241994ed JW |
6188 | static int memory_low_show(struct seq_file *m, void *v) |
6189 | { | |
677dc973 CD |
6190 | return seq_puts_memcg_tunable(m, |
6191 | READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); | |
241994ed JW |
6192 | } |
6193 | ||
6194 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
6195 | char *buf, size_t nbytes, loff_t off) | |
6196 | { | |
6197 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6198 | unsigned long low; | |
6199 | int err; | |
6200 | ||
6201 | buf = strstrip(buf); | |
d2973697 | 6202 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
6203 | if (err) |
6204 | return err; | |
6205 | ||
23067153 | 6206 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
6207 | |
6208 | return nbytes; | |
6209 | } | |
6210 | ||
6211 | static int memory_high_show(struct seq_file *m, void *v) | |
6212 | { | |
d1663a90 JK |
6213 | return seq_puts_memcg_tunable(m, |
6214 | READ_ONCE(mem_cgroup_from_seq(m)->memory.high)); | |
241994ed JW |
6215 | } |
6216 | ||
6217 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
6218 | char *buf, size_t nbytes, loff_t off) | |
6219 | { | |
6220 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6221 | unsigned int nr_retries = MAX_RECLAIM_RETRIES; |
8c8c383c | 6222 | bool drained = false; |
241994ed JW |
6223 | unsigned long high; |
6224 | int err; | |
6225 | ||
6226 | buf = strstrip(buf); | |
d2973697 | 6227 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
6228 | if (err) |
6229 | return err; | |
6230 | ||
8c8c383c JW |
6231 | for (;;) { |
6232 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6233 | unsigned long reclaimed; | |
6234 | ||
6235 | if (nr_pages <= high) | |
6236 | break; | |
6237 | ||
6238 | if (signal_pending(current)) | |
6239 | break; | |
6240 | ||
6241 | if (!drained) { | |
6242 | drain_all_stock(memcg); | |
6243 | drained = true; | |
6244 | continue; | |
6245 | } | |
6246 | ||
6247 | reclaimed = try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
6248 | GFP_KERNEL, true); | |
6249 | ||
6250 | if (!reclaimed && !nr_retries--) | |
6251 | break; | |
6252 | } | |
588083bb | 6253 | |
536d3bf2 RG |
6254 | page_counter_set_high(&memcg->memory, high); |
6255 | ||
19ce33ac JW |
6256 | memcg_wb_domain_size_changed(memcg); |
6257 | ||
241994ed JW |
6258 | return nbytes; |
6259 | } | |
6260 | ||
6261 | static int memory_max_show(struct seq_file *m, void *v) | |
6262 | { | |
677dc973 CD |
6263 | return seq_puts_memcg_tunable(m, |
6264 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); | |
241994ed JW |
6265 | } |
6266 | ||
6267 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
6268 | char *buf, size_t nbytes, loff_t off) | |
6269 | { | |
6270 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6271 | unsigned int nr_reclaims = MAX_RECLAIM_RETRIES; |
b6e6edcf | 6272 | bool drained = false; |
241994ed JW |
6273 | unsigned long max; |
6274 | int err; | |
6275 | ||
6276 | buf = strstrip(buf); | |
d2973697 | 6277 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
6278 | if (err) |
6279 | return err; | |
6280 | ||
bbec2e15 | 6281 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
6282 | |
6283 | for (;;) { | |
6284 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6285 | ||
6286 | if (nr_pages <= max) | |
6287 | break; | |
6288 | ||
7249c9f0 | 6289 | if (signal_pending(current)) |
b6e6edcf | 6290 | break; |
b6e6edcf JW |
6291 | |
6292 | if (!drained) { | |
6293 | drain_all_stock(memcg); | |
6294 | drained = true; | |
6295 | continue; | |
6296 | } | |
6297 | ||
6298 | if (nr_reclaims) { | |
6299 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
6300 | GFP_KERNEL, true)) | |
6301 | nr_reclaims--; | |
6302 | continue; | |
6303 | } | |
6304 | ||
e27be240 | 6305 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
6306 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
6307 | break; | |
6308 | } | |
241994ed | 6309 | |
2529bb3a | 6310 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
6311 | return nbytes; |
6312 | } | |
6313 | ||
1e577f97 SB |
6314 | static void __memory_events_show(struct seq_file *m, atomic_long_t *events) |
6315 | { | |
6316 | seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW])); | |
6317 | seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH])); | |
6318 | seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX])); | |
6319 | seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM])); | |
6320 | seq_printf(m, "oom_kill %lu\n", | |
6321 | atomic_long_read(&events[MEMCG_OOM_KILL])); | |
6322 | } | |
6323 | ||
241994ed JW |
6324 | static int memory_events_show(struct seq_file *m, void *v) |
6325 | { | |
aa9694bb | 6326 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
241994ed | 6327 | |
1e577f97 SB |
6328 | __memory_events_show(m, memcg->memory_events); |
6329 | return 0; | |
6330 | } | |
6331 | ||
6332 | static int memory_events_local_show(struct seq_file *m, void *v) | |
6333 | { | |
6334 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
241994ed | 6335 | |
1e577f97 | 6336 | __memory_events_show(m, memcg->memory_events_local); |
241994ed JW |
6337 | return 0; |
6338 | } | |
6339 | ||
587d9f72 JW |
6340 | static int memory_stat_show(struct seq_file *m, void *v) |
6341 | { | |
aa9694bb | 6342 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
c8713d0b | 6343 | char *buf; |
1ff9e6e1 | 6344 | |
c8713d0b JW |
6345 | buf = memory_stat_format(memcg); |
6346 | if (!buf) | |
6347 | return -ENOMEM; | |
6348 | seq_puts(m, buf); | |
6349 | kfree(buf); | |
587d9f72 JW |
6350 | return 0; |
6351 | } | |
6352 | ||
5f9a4f4a MS |
6353 | #ifdef CONFIG_NUMA |
6354 | static int memory_numa_stat_show(struct seq_file *m, void *v) | |
6355 | { | |
6356 | int i; | |
6357 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
6358 | ||
6359 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { | |
6360 | int nid; | |
6361 | ||
6362 | if (memory_stats[i].idx >= NR_VM_NODE_STAT_ITEMS) | |
6363 | continue; | |
6364 | ||
6365 | seq_printf(m, "%s", memory_stats[i].name); | |
6366 | for_each_node_state(nid, N_MEMORY) { | |
6367 | u64 size; | |
6368 | struct lruvec *lruvec; | |
6369 | ||
6370 | lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); | |
6371 | size = lruvec_page_state(lruvec, memory_stats[i].idx); | |
6372 | size *= memory_stats[i].ratio; | |
6373 | seq_printf(m, " N%d=%llu", nid, size); | |
6374 | } | |
6375 | seq_putc(m, '\n'); | |
6376 | } | |
6377 | ||
6378 | return 0; | |
6379 | } | |
6380 | #endif | |
6381 | ||
3d8b38eb RG |
6382 | static int memory_oom_group_show(struct seq_file *m, void *v) |
6383 | { | |
aa9694bb | 6384 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3d8b38eb RG |
6385 | |
6386 | seq_printf(m, "%d\n", memcg->oom_group); | |
6387 | ||
6388 | return 0; | |
6389 | } | |
6390 | ||
6391 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
6392 | char *buf, size_t nbytes, loff_t off) | |
6393 | { | |
6394 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6395 | int ret, oom_group; | |
6396 | ||
6397 | buf = strstrip(buf); | |
6398 | if (!buf) | |
6399 | return -EINVAL; | |
6400 | ||
6401 | ret = kstrtoint(buf, 0, &oom_group); | |
6402 | if (ret) | |
6403 | return ret; | |
6404 | ||
6405 | if (oom_group != 0 && oom_group != 1) | |
6406 | return -EINVAL; | |
6407 | ||
6408 | memcg->oom_group = oom_group; | |
6409 | ||
6410 | return nbytes; | |
6411 | } | |
6412 | ||
241994ed JW |
6413 | static struct cftype memory_files[] = { |
6414 | { | |
6415 | .name = "current", | |
f5fc3c5d | 6416 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
6417 | .read_u64 = memory_current_read, |
6418 | }, | |
bf8d5d52 RG |
6419 | { |
6420 | .name = "min", | |
6421 | .flags = CFTYPE_NOT_ON_ROOT, | |
6422 | .seq_show = memory_min_show, | |
6423 | .write = memory_min_write, | |
6424 | }, | |
241994ed JW |
6425 | { |
6426 | .name = "low", | |
6427 | .flags = CFTYPE_NOT_ON_ROOT, | |
6428 | .seq_show = memory_low_show, | |
6429 | .write = memory_low_write, | |
6430 | }, | |
6431 | { | |
6432 | .name = "high", | |
6433 | .flags = CFTYPE_NOT_ON_ROOT, | |
6434 | .seq_show = memory_high_show, | |
6435 | .write = memory_high_write, | |
6436 | }, | |
6437 | { | |
6438 | .name = "max", | |
6439 | .flags = CFTYPE_NOT_ON_ROOT, | |
6440 | .seq_show = memory_max_show, | |
6441 | .write = memory_max_write, | |
6442 | }, | |
6443 | { | |
6444 | .name = "events", | |
6445 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 6446 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
6447 | .seq_show = memory_events_show, |
6448 | }, | |
1e577f97 SB |
6449 | { |
6450 | .name = "events.local", | |
6451 | .flags = CFTYPE_NOT_ON_ROOT, | |
6452 | .file_offset = offsetof(struct mem_cgroup, events_local_file), | |
6453 | .seq_show = memory_events_local_show, | |
6454 | }, | |
587d9f72 JW |
6455 | { |
6456 | .name = "stat", | |
587d9f72 JW |
6457 | .seq_show = memory_stat_show, |
6458 | }, | |
5f9a4f4a MS |
6459 | #ifdef CONFIG_NUMA |
6460 | { | |
6461 | .name = "numa_stat", | |
6462 | .seq_show = memory_numa_stat_show, | |
6463 | }, | |
6464 | #endif | |
3d8b38eb RG |
6465 | { |
6466 | .name = "oom.group", | |
6467 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
6468 | .seq_show = memory_oom_group_show, | |
6469 | .write = memory_oom_group_write, | |
6470 | }, | |
241994ed JW |
6471 | { } /* terminate */ |
6472 | }; | |
6473 | ||
073219e9 | 6474 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 6475 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6476 | .css_online = mem_cgroup_css_online, |
92fb9748 | 6477 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 6478 | .css_released = mem_cgroup_css_released, |
92fb9748 | 6479 | .css_free = mem_cgroup_css_free, |
1ced953b | 6480 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
6481 | .can_attach = mem_cgroup_can_attach, |
6482 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 6483 | .post_attach = mem_cgroup_move_task, |
241994ed JW |
6484 | .dfl_cftypes = memory_files, |
6485 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 6486 | .early_init = 0, |
8cdea7c0 | 6487 | }; |
c077719b | 6488 | |
bc50bcc6 JW |
6489 | /* |
6490 | * This function calculates an individual cgroup's effective | |
6491 | * protection which is derived from its own memory.min/low, its | |
6492 | * parent's and siblings' settings, as well as the actual memory | |
6493 | * distribution in the tree. | |
6494 | * | |
6495 | * The following rules apply to the effective protection values: | |
6496 | * | |
6497 | * 1. At the first level of reclaim, effective protection is equal to | |
6498 | * the declared protection in memory.min and memory.low. | |
6499 | * | |
6500 | * 2. To enable safe delegation of the protection configuration, at | |
6501 | * subsequent levels the effective protection is capped to the | |
6502 | * parent's effective protection. | |
6503 | * | |
6504 | * 3. To make complex and dynamic subtrees easier to configure, the | |
6505 | * user is allowed to overcommit the declared protection at a given | |
6506 | * level. If that is the case, the parent's effective protection is | |
6507 | * distributed to the children in proportion to how much protection | |
6508 | * they have declared and how much of it they are utilizing. | |
6509 | * | |
6510 | * This makes distribution proportional, but also work-conserving: | |
6511 | * if one cgroup claims much more protection than it uses memory, | |
6512 | * the unused remainder is available to its siblings. | |
6513 | * | |
6514 | * 4. Conversely, when the declared protection is undercommitted at a | |
6515 | * given level, the distribution of the larger parental protection | |
6516 | * budget is NOT proportional. A cgroup's protection from a sibling | |
6517 | * is capped to its own memory.min/low setting. | |
6518 | * | |
8a931f80 JW |
6519 | * 5. However, to allow protecting recursive subtrees from each other |
6520 | * without having to declare each individual cgroup's fixed share | |
6521 | * of the ancestor's claim to protection, any unutilized - | |
6522 | * "floating" - protection from up the tree is distributed in | |
6523 | * proportion to each cgroup's *usage*. This makes the protection | |
6524 | * neutral wrt sibling cgroups and lets them compete freely over | |
6525 | * the shared parental protection budget, but it protects the | |
6526 | * subtree as a whole from neighboring subtrees. | |
6527 | * | |
6528 | * Note that 4. and 5. are not in conflict: 4. is about protecting | |
6529 | * against immediate siblings whereas 5. is about protecting against | |
6530 | * neighboring subtrees. | |
bc50bcc6 JW |
6531 | */ |
6532 | static unsigned long effective_protection(unsigned long usage, | |
8a931f80 | 6533 | unsigned long parent_usage, |
bc50bcc6 JW |
6534 | unsigned long setting, |
6535 | unsigned long parent_effective, | |
6536 | unsigned long siblings_protected) | |
6537 | { | |
6538 | unsigned long protected; | |
8a931f80 | 6539 | unsigned long ep; |
bc50bcc6 JW |
6540 | |
6541 | protected = min(usage, setting); | |
6542 | /* | |
6543 | * If all cgroups at this level combined claim and use more | |
6544 | * protection then what the parent affords them, distribute | |
6545 | * shares in proportion to utilization. | |
6546 | * | |
6547 | * We are using actual utilization rather than the statically | |
6548 | * claimed protection in order to be work-conserving: claimed | |
6549 | * but unused protection is available to siblings that would | |
6550 | * otherwise get a smaller chunk than what they claimed. | |
6551 | */ | |
6552 | if (siblings_protected > parent_effective) | |
6553 | return protected * parent_effective / siblings_protected; | |
6554 | ||
6555 | /* | |
6556 | * Ok, utilized protection of all children is within what the | |
6557 | * parent affords them, so we know whatever this child claims | |
6558 | * and utilizes is effectively protected. | |
6559 | * | |
6560 | * If there is unprotected usage beyond this value, reclaim | |
6561 | * will apply pressure in proportion to that amount. | |
6562 | * | |
6563 | * If there is unutilized protection, the cgroup will be fully | |
6564 | * shielded from reclaim, but we do return a smaller value for | |
6565 | * protection than what the group could enjoy in theory. This | |
6566 | * is okay. With the overcommit distribution above, effective | |
6567 | * protection is always dependent on how memory is actually | |
6568 | * consumed among the siblings anyway. | |
6569 | */ | |
8a931f80 JW |
6570 | ep = protected; |
6571 | ||
6572 | /* | |
6573 | * If the children aren't claiming (all of) the protection | |
6574 | * afforded to them by the parent, distribute the remainder in | |
6575 | * proportion to the (unprotected) memory of each cgroup. That | |
6576 | * way, cgroups that aren't explicitly prioritized wrt each | |
6577 | * other compete freely over the allowance, but they are | |
6578 | * collectively protected from neighboring trees. | |
6579 | * | |
6580 | * We're using unprotected memory for the weight so that if | |
6581 | * some cgroups DO claim explicit protection, we don't protect | |
6582 | * the same bytes twice. | |
cd324edc JW |
6583 | * |
6584 | * Check both usage and parent_usage against the respective | |
6585 | * protected values. One should imply the other, but they | |
6586 | * aren't read atomically - make sure the division is sane. | |
8a931f80 JW |
6587 | */ |
6588 | if (!(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)) | |
6589 | return ep; | |
cd324edc JW |
6590 | if (parent_effective > siblings_protected && |
6591 | parent_usage > siblings_protected && | |
6592 | usage > protected) { | |
8a931f80 JW |
6593 | unsigned long unclaimed; |
6594 | ||
6595 | unclaimed = parent_effective - siblings_protected; | |
6596 | unclaimed *= usage - protected; | |
6597 | unclaimed /= parent_usage - siblings_protected; | |
6598 | ||
6599 | ep += unclaimed; | |
6600 | } | |
6601 | ||
6602 | return ep; | |
bc50bcc6 JW |
6603 | } |
6604 | ||
241994ed | 6605 | /** |
bf8d5d52 | 6606 | * mem_cgroup_protected - check if memory consumption is in the normal range |
34c81057 | 6607 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
6608 | * @memcg: the memory cgroup to check |
6609 | * | |
23067153 RG |
6610 | * WARNING: This function is not stateless! It can only be used as part |
6611 | * of a top-down tree iteration, not for isolated queries. | |
241994ed | 6612 | */ |
45c7f7e1 CD |
6613 | void mem_cgroup_calculate_protection(struct mem_cgroup *root, |
6614 | struct mem_cgroup *memcg) | |
241994ed | 6615 | { |
8a931f80 | 6616 | unsigned long usage, parent_usage; |
23067153 RG |
6617 | struct mem_cgroup *parent; |
6618 | ||
241994ed | 6619 | if (mem_cgroup_disabled()) |
45c7f7e1 | 6620 | return; |
241994ed | 6621 | |
34c81057 SC |
6622 | if (!root) |
6623 | root = root_mem_cgroup; | |
22f7496f YS |
6624 | |
6625 | /* | |
6626 | * Effective values of the reclaim targets are ignored so they | |
6627 | * can be stale. Have a look at mem_cgroup_protection for more | |
6628 | * details. | |
6629 | * TODO: calculation should be more robust so that we do not need | |
6630 | * that special casing. | |
6631 | */ | |
34c81057 | 6632 | if (memcg == root) |
45c7f7e1 | 6633 | return; |
241994ed | 6634 | |
23067153 | 6635 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 | 6636 | if (!usage) |
45c7f7e1 | 6637 | return; |
bf8d5d52 | 6638 | |
bf8d5d52 | 6639 | parent = parent_mem_cgroup(memcg); |
df2a4196 RG |
6640 | /* No parent means a non-hierarchical mode on v1 memcg */ |
6641 | if (!parent) | |
45c7f7e1 | 6642 | return; |
df2a4196 | 6643 | |
bc50bcc6 | 6644 | if (parent == root) { |
c3d53200 | 6645 | memcg->memory.emin = READ_ONCE(memcg->memory.min); |
03960e33 | 6646 | memcg->memory.elow = READ_ONCE(memcg->memory.low); |
45c7f7e1 | 6647 | return; |
bf8d5d52 RG |
6648 | } |
6649 | ||
8a931f80 JW |
6650 | parent_usage = page_counter_read(&parent->memory); |
6651 | ||
b3a7822e | 6652 | WRITE_ONCE(memcg->memory.emin, effective_protection(usage, parent_usage, |
c3d53200 CD |
6653 | READ_ONCE(memcg->memory.min), |
6654 | READ_ONCE(parent->memory.emin), | |
b3a7822e | 6655 | atomic_long_read(&parent->memory.children_min_usage))); |
23067153 | 6656 | |
b3a7822e | 6657 | WRITE_ONCE(memcg->memory.elow, effective_protection(usage, parent_usage, |
03960e33 CD |
6658 | READ_ONCE(memcg->memory.low), |
6659 | READ_ONCE(parent->memory.elow), | |
b3a7822e | 6660 | atomic_long_read(&parent->memory.children_low_usage))); |
241994ed JW |
6661 | } |
6662 | ||
00501b53 | 6663 | /** |
f0e45fb4 | 6664 | * mem_cgroup_charge - charge a newly allocated page to a cgroup |
00501b53 JW |
6665 | * @page: page to charge |
6666 | * @mm: mm context of the victim | |
6667 | * @gfp_mask: reclaim mode | |
00501b53 JW |
6668 | * |
6669 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
6670 | * pages according to @gfp_mask if necessary. | |
6671 | * | |
f0e45fb4 | 6672 | * Returns 0 on success. Otherwise, an error code is returned. |
00501b53 | 6673 | */ |
d9eb1ea2 | 6674 | int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask) |
00501b53 | 6675 | { |
6c357848 | 6676 | unsigned int nr_pages = thp_nr_pages(page); |
00501b53 | 6677 | struct mem_cgroup *memcg = NULL; |
00501b53 JW |
6678 | int ret = 0; |
6679 | ||
6680 | if (mem_cgroup_disabled()) | |
6681 | goto out; | |
6682 | ||
6683 | if (PageSwapCache(page)) { | |
2d1c4980 JW |
6684 | swp_entry_t ent = { .val = page_private(page), }; |
6685 | unsigned short id; | |
6686 | ||
00501b53 JW |
6687 | /* |
6688 | * Every swap fault against a single page tries to charge the | |
6689 | * page, bail as early as possible. shmem_unuse() encounters | |
eccb52e7 JW |
6690 | * already charged pages, too. page->mem_cgroup is protected |
6691 | * by the page lock, which serializes swap cache removal, which | |
00501b53 JW |
6692 | * in turn serializes uncharging. |
6693 | */ | |
e993d905 | 6694 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
abe2895b | 6695 | if (compound_head(page)->mem_cgroup) |
00501b53 | 6696 | goto out; |
e993d905 | 6697 | |
2d1c4980 JW |
6698 | id = lookup_swap_cgroup_id(ent); |
6699 | rcu_read_lock(); | |
6700 | memcg = mem_cgroup_from_id(id); | |
6701 | if (memcg && !css_tryget_online(&memcg->css)) | |
6702 | memcg = NULL; | |
6703 | rcu_read_unlock(); | |
00501b53 JW |
6704 | } |
6705 | ||
00501b53 JW |
6706 | if (!memcg) |
6707 | memcg = get_mem_cgroup_from_mm(mm); | |
6708 | ||
6709 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
f0e45fb4 JW |
6710 | if (ret) |
6711 | goto out_put; | |
00501b53 | 6712 | |
1a3e1f40 | 6713 | css_get(&memcg->css); |
d9eb1ea2 | 6714 | commit_charge(page, memcg); |
6abb5a86 | 6715 | |
6abb5a86 | 6716 | local_irq_disable(); |
3fba69a5 | 6717 | mem_cgroup_charge_statistics(memcg, page, nr_pages); |
6abb5a86 JW |
6718 | memcg_check_events(memcg, page); |
6719 | local_irq_enable(); | |
00501b53 | 6720 | |
2d1c4980 | 6721 | if (PageSwapCache(page)) { |
00501b53 JW |
6722 | swp_entry_t entry = { .val = page_private(page) }; |
6723 | /* | |
6724 | * The swap entry might not get freed for a long time, | |
6725 | * let's not wait for it. The page already received a | |
6726 | * memory+swap charge, drop the swap entry duplicate. | |
6727 | */ | |
38d8b4e6 | 6728 | mem_cgroup_uncharge_swap(entry, nr_pages); |
00501b53 | 6729 | } |
00501b53 | 6730 | |
f0e45fb4 JW |
6731 | out_put: |
6732 | css_put(&memcg->css); | |
6733 | out: | |
6734 | return ret; | |
3fea5a49 JW |
6735 | } |
6736 | ||
a9d5adee JG |
6737 | struct uncharge_gather { |
6738 | struct mem_cgroup *memcg; | |
9f762dbe | 6739 | unsigned long nr_pages; |
a9d5adee | 6740 | unsigned long pgpgout; |
a9d5adee | 6741 | unsigned long nr_kmem; |
a9d5adee JG |
6742 | struct page *dummy_page; |
6743 | }; | |
6744 | ||
6745 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 6746 | { |
a9d5adee JG |
6747 | memset(ug, 0, sizeof(*ug)); |
6748 | } | |
6749 | ||
6750 | static void uncharge_batch(const struct uncharge_gather *ug) | |
6751 | { | |
747db954 JW |
6752 | unsigned long flags; |
6753 | ||
a9d5adee | 6754 | if (!mem_cgroup_is_root(ug->memcg)) { |
9f762dbe | 6755 | page_counter_uncharge(&ug->memcg->memory, ug->nr_pages); |
7941d214 | 6756 | if (do_memsw_account()) |
9f762dbe | 6757 | page_counter_uncharge(&ug->memcg->memsw, ug->nr_pages); |
a9d5adee JG |
6758 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem) |
6759 | page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem); | |
6760 | memcg_oom_recover(ug->memcg); | |
ce00a967 | 6761 | } |
747db954 JW |
6762 | |
6763 | local_irq_save(flags); | |
c9019e9b | 6764 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); |
9f762dbe | 6765 | __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, ug->nr_pages); |
a9d5adee | 6766 | memcg_check_events(ug->memcg, ug->dummy_page); |
747db954 | 6767 | local_irq_restore(flags); |
f1796544 MH |
6768 | |
6769 | /* drop reference from uncharge_page */ | |
6770 | css_put(&ug->memcg->css); | |
a9d5adee JG |
6771 | } |
6772 | ||
6773 | static void uncharge_page(struct page *page, struct uncharge_gather *ug) | |
6774 | { | |
9f762dbe JW |
6775 | unsigned long nr_pages; |
6776 | ||
a9d5adee | 6777 | VM_BUG_ON_PAGE(PageLRU(page), page); |
a9d5adee JG |
6778 | |
6779 | if (!page->mem_cgroup) | |
6780 | return; | |
6781 | ||
6782 | /* | |
6783 | * Nobody should be changing or seriously looking at | |
6784 | * page->mem_cgroup at this point, we have fully | |
6785 | * exclusive access to the page. | |
6786 | */ | |
6787 | ||
6788 | if (ug->memcg != page->mem_cgroup) { | |
6789 | if (ug->memcg) { | |
6790 | uncharge_batch(ug); | |
6791 | uncharge_gather_clear(ug); | |
6792 | } | |
6793 | ug->memcg = page->mem_cgroup; | |
f1796544 MH |
6794 | |
6795 | /* pairs with css_put in uncharge_batch */ | |
6796 | css_get(&ug->memcg->css); | |
a9d5adee JG |
6797 | } |
6798 | ||
9f762dbe JW |
6799 | nr_pages = compound_nr(page); |
6800 | ug->nr_pages += nr_pages; | |
a9d5adee | 6801 | |
9f762dbe | 6802 | if (!PageKmemcg(page)) { |
a9d5adee JG |
6803 | ug->pgpgout++; |
6804 | } else { | |
9f762dbe | 6805 | ug->nr_kmem += nr_pages; |
a9d5adee JG |
6806 | __ClearPageKmemcg(page); |
6807 | } | |
6808 | ||
6809 | ug->dummy_page = page; | |
6810 | page->mem_cgroup = NULL; | |
1a3e1f40 | 6811 | css_put(&ug->memcg->css); |
747db954 JW |
6812 | } |
6813 | ||
6814 | static void uncharge_list(struct list_head *page_list) | |
6815 | { | |
a9d5adee | 6816 | struct uncharge_gather ug; |
747db954 | 6817 | struct list_head *next; |
a9d5adee JG |
6818 | |
6819 | uncharge_gather_clear(&ug); | |
747db954 | 6820 | |
8b592656 JW |
6821 | /* |
6822 | * Note that the list can be a single page->lru; hence the | |
6823 | * do-while loop instead of a simple list_for_each_entry(). | |
6824 | */ | |
747db954 JW |
6825 | next = page_list->next; |
6826 | do { | |
a9d5adee JG |
6827 | struct page *page; |
6828 | ||
747db954 JW |
6829 | page = list_entry(next, struct page, lru); |
6830 | next = page->lru.next; | |
6831 | ||
a9d5adee | 6832 | uncharge_page(page, &ug); |
747db954 JW |
6833 | } while (next != page_list); |
6834 | ||
a9d5adee JG |
6835 | if (ug.memcg) |
6836 | uncharge_batch(&ug); | |
747db954 JW |
6837 | } |
6838 | ||
0a31bc97 JW |
6839 | /** |
6840 | * mem_cgroup_uncharge - uncharge a page | |
6841 | * @page: page to uncharge | |
6842 | * | |
f0e45fb4 | 6843 | * Uncharge a page previously charged with mem_cgroup_charge(). |
0a31bc97 JW |
6844 | */ |
6845 | void mem_cgroup_uncharge(struct page *page) | |
6846 | { | |
a9d5adee JG |
6847 | struct uncharge_gather ug; |
6848 | ||
0a31bc97 JW |
6849 | if (mem_cgroup_disabled()) |
6850 | return; | |
6851 | ||
747db954 | 6852 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 6853 | if (!page->mem_cgroup) |
0a31bc97 JW |
6854 | return; |
6855 | ||
a9d5adee JG |
6856 | uncharge_gather_clear(&ug); |
6857 | uncharge_page(page, &ug); | |
6858 | uncharge_batch(&ug); | |
747db954 | 6859 | } |
0a31bc97 | 6860 | |
747db954 JW |
6861 | /** |
6862 | * mem_cgroup_uncharge_list - uncharge a list of page | |
6863 | * @page_list: list of pages to uncharge | |
6864 | * | |
6865 | * Uncharge a list of pages previously charged with | |
f0e45fb4 | 6866 | * mem_cgroup_charge(). |
747db954 JW |
6867 | */ |
6868 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
6869 | { | |
6870 | if (mem_cgroup_disabled()) | |
6871 | return; | |
0a31bc97 | 6872 | |
747db954 JW |
6873 | if (!list_empty(page_list)) |
6874 | uncharge_list(page_list); | |
0a31bc97 JW |
6875 | } |
6876 | ||
6877 | /** | |
6a93ca8f JW |
6878 | * mem_cgroup_migrate - charge a page's replacement |
6879 | * @oldpage: currently circulating page | |
6880 | * @newpage: replacement page | |
0a31bc97 | 6881 | * |
6a93ca8f JW |
6882 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
6883 | * be uncharged upon free. | |
0a31bc97 JW |
6884 | * |
6885 | * Both pages must be locked, @newpage->mapping must be set up. | |
6886 | */ | |
6a93ca8f | 6887 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 6888 | { |
29833315 | 6889 | struct mem_cgroup *memcg; |
44b7a8d3 | 6890 | unsigned int nr_pages; |
d93c4130 | 6891 | unsigned long flags; |
0a31bc97 JW |
6892 | |
6893 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6894 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 6895 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
6896 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6897 | newpage); | |
0a31bc97 JW |
6898 | |
6899 | if (mem_cgroup_disabled()) | |
6900 | return; | |
6901 | ||
6902 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 6903 | if (newpage->mem_cgroup) |
0a31bc97 JW |
6904 | return; |
6905 | ||
1306a85a | 6906 | memcg = oldpage->mem_cgroup; |
29833315 | 6907 | if (!memcg) |
0a31bc97 JW |
6908 | return; |
6909 | ||
44b7a8d3 | 6910 | /* Force-charge the new page. The old one will be freed soon */ |
6c357848 | 6911 | nr_pages = thp_nr_pages(newpage); |
44b7a8d3 JW |
6912 | |
6913 | page_counter_charge(&memcg->memory, nr_pages); | |
6914 | if (do_memsw_account()) | |
6915 | page_counter_charge(&memcg->memsw, nr_pages); | |
0a31bc97 | 6916 | |
1a3e1f40 | 6917 | css_get(&memcg->css); |
d9eb1ea2 | 6918 | commit_charge(newpage, memcg); |
44b7a8d3 | 6919 | |
d93c4130 | 6920 | local_irq_save(flags); |
3fba69a5 | 6921 | mem_cgroup_charge_statistics(memcg, newpage, nr_pages); |
44b7a8d3 | 6922 | memcg_check_events(memcg, newpage); |
d93c4130 | 6923 | local_irq_restore(flags); |
0a31bc97 JW |
6924 | } |
6925 | ||
ef12947c | 6926 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
6927 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
6928 | ||
2d758073 | 6929 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
6930 | { |
6931 | struct mem_cgroup *memcg; | |
6932 | ||
2d758073 JW |
6933 | if (!mem_cgroup_sockets_enabled) |
6934 | return; | |
6935 | ||
e876ecc6 SB |
6936 | /* Do not associate the sock with unrelated interrupted task's memcg. */ |
6937 | if (in_interrupt()) | |
6938 | return; | |
6939 | ||
11092087 JW |
6940 | rcu_read_lock(); |
6941 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
6942 | if (memcg == root_mem_cgroup) |
6943 | goto out; | |
0db15298 | 6944 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 6945 | goto out; |
8965aa28 | 6946 | if (css_tryget(&memcg->css)) |
11092087 | 6947 | sk->sk_memcg = memcg; |
f7e1cb6e | 6948 | out: |
11092087 JW |
6949 | rcu_read_unlock(); |
6950 | } | |
11092087 | 6951 | |
2d758073 | 6952 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 6953 | { |
2d758073 JW |
6954 | if (sk->sk_memcg) |
6955 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
6956 | } |
6957 | ||
6958 | /** | |
6959 | * mem_cgroup_charge_skmem - charge socket memory | |
6960 | * @memcg: memcg to charge | |
6961 | * @nr_pages: number of pages to charge | |
6962 | * | |
6963 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
6964 | * @memcg's configured limit, %false if the charge had to be forced. | |
6965 | */ | |
6966 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6967 | { | |
f7e1cb6e | 6968 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 6969 | |
f7e1cb6e | 6970 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6971 | struct page_counter *fail; |
f7e1cb6e | 6972 | |
0db15298 JW |
6973 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
6974 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
6975 | return true; |
6976 | } | |
0db15298 JW |
6977 | page_counter_charge(&memcg->tcpmem, nr_pages); |
6978 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 6979 | return false; |
11092087 | 6980 | } |
d886f4e4 | 6981 | |
f7e1cb6e JW |
6982 | /* Don't block in the packet receive path */ |
6983 | if (in_softirq()) | |
6984 | gfp_mask = GFP_NOWAIT; | |
6985 | ||
c9019e9b | 6986 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); |
b2807f07 | 6987 | |
f7e1cb6e JW |
6988 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
6989 | return true; | |
6990 | ||
6991 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
6992 | return false; |
6993 | } | |
6994 | ||
6995 | /** | |
6996 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
6997 | * @memcg: memcg to uncharge |
6998 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
6999 | */ |
7000 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
7001 | { | |
f7e1cb6e | 7002 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 7003 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
7004 | return; |
7005 | } | |
d886f4e4 | 7006 | |
c9019e9b | 7007 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 7008 | |
475d0487 | 7009 | refill_stock(memcg, nr_pages); |
11092087 JW |
7010 | } |
7011 | ||
f7e1cb6e JW |
7012 | static int __init cgroup_memory(char *s) |
7013 | { | |
7014 | char *token; | |
7015 | ||
7016 | while ((token = strsep(&s, ",")) != NULL) { | |
7017 | if (!*token) | |
7018 | continue; | |
7019 | if (!strcmp(token, "nosocket")) | |
7020 | cgroup_memory_nosocket = true; | |
04823c83 VD |
7021 | if (!strcmp(token, "nokmem")) |
7022 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
7023 | } |
7024 | return 0; | |
7025 | } | |
7026 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 7027 | |
2d11085e | 7028 | /* |
1081312f MH |
7029 | * subsys_initcall() for memory controller. |
7030 | * | |
308167fc SAS |
7031 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
7032 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
7033 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
7034 | * should be initialized from here. | |
2d11085e MH |
7035 | */ |
7036 | static int __init mem_cgroup_init(void) | |
7037 | { | |
95a045f6 JW |
7038 | int cpu, node; |
7039 | ||
308167fc SAS |
7040 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
7041 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
7042 | |
7043 | for_each_possible_cpu(cpu) | |
7044 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
7045 | drain_local_stock); | |
7046 | ||
7047 | for_each_node(node) { | |
7048 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
7049 | |
7050 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
7051 | node_online(node) ? node : NUMA_NO_NODE); | |
7052 | ||
ef8f2327 | 7053 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 7054 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 7055 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
7056 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
7057 | } | |
7058 | ||
2d11085e MH |
7059 | return 0; |
7060 | } | |
7061 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
7062 | |
7063 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
7064 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
7065 | { | |
1c2d479a | 7066 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
7067 | /* |
7068 | * The root cgroup cannot be destroyed, so it's refcount must | |
7069 | * always be >= 1. | |
7070 | */ | |
7071 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
7072 | VM_BUG_ON(1); | |
7073 | break; | |
7074 | } | |
7075 | memcg = parent_mem_cgroup(memcg); | |
7076 | if (!memcg) | |
7077 | memcg = root_mem_cgroup; | |
7078 | } | |
7079 | return memcg; | |
7080 | } | |
7081 | ||
21afa38e JW |
7082 | /** |
7083 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
7084 | * @page: page whose memsw charge to transfer | |
7085 | * @entry: swap entry to move the charge to | |
7086 | * | |
7087 | * Transfer the memsw charge of @page to @entry. | |
7088 | */ | |
7089 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
7090 | { | |
1f47b61f | 7091 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 7092 | unsigned int nr_entries; |
21afa38e JW |
7093 | unsigned short oldid; |
7094 | ||
7095 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
7096 | VM_BUG_ON_PAGE(page_count(page), page); | |
7097 | ||
2d1c4980 | 7098 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
21afa38e JW |
7099 | return; |
7100 | ||
7101 | memcg = page->mem_cgroup; | |
7102 | ||
7103 | /* Readahead page, never charged */ | |
7104 | if (!memcg) | |
7105 | return; | |
7106 | ||
1f47b61f VD |
7107 | /* |
7108 | * In case the memcg owning these pages has been offlined and doesn't | |
7109 | * have an ID allocated to it anymore, charge the closest online | |
7110 | * ancestor for the swap instead and transfer the memory+swap charge. | |
7111 | */ | |
7112 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
6c357848 | 7113 | nr_entries = thp_nr_pages(page); |
d6810d73 HY |
7114 | /* Get references for the tail pages, too */ |
7115 | if (nr_entries > 1) | |
7116 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
7117 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
7118 | nr_entries); | |
21afa38e | 7119 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7120 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e JW |
7121 | |
7122 | page->mem_cgroup = NULL; | |
7123 | ||
7124 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 7125 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 7126 | |
2d1c4980 | 7127 | if (!cgroup_memory_noswap && memcg != swap_memcg) { |
1f47b61f | 7128 | if (!mem_cgroup_is_root(swap_memcg)) |
d6810d73 HY |
7129 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
7130 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
7131 | } |
7132 | ||
ce9ce665 SAS |
7133 | /* |
7134 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 7135 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 7136 | * important here to have the interrupts disabled because it is the |
b93b0163 | 7137 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 SAS |
7138 | */ |
7139 | VM_BUG_ON(!irqs_disabled()); | |
3fba69a5 | 7140 | mem_cgroup_charge_statistics(memcg, page, -nr_entries); |
21afa38e | 7141 | memcg_check_events(memcg, page); |
73f576c0 | 7142 | |
1a3e1f40 | 7143 | css_put(&memcg->css); |
21afa38e JW |
7144 | } |
7145 | ||
38d8b4e6 HY |
7146 | /** |
7147 | * mem_cgroup_try_charge_swap - try charging swap space for a page | |
37e84351 VD |
7148 | * @page: page being added to swap |
7149 | * @entry: swap entry to charge | |
7150 | * | |
38d8b4e6 | 7151 | * Try to charge @page's memcg for the swap space at @entry. |
37e84351 VD |
7152 | * |
7153 | * Returns 0 on success, -ENOMEM on failure. | |
7154 | */ | |
7155 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
7156 | { | |
6c357848 | 7157 | unsigned int nr_pages = thp_nr_pages(page); |
37e84351 | 7158 | struct page_counter *counter; |
38d8b4e6 | 7159 | struct mem_cgroup *memcg; |
37e84351 VD |
7160 | unsigned short oldid; |
7161 | ||
2d1c4980 | 7162 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
37e84351 VD |
7163 | return 0; |
7164 | ||
7165 | memcg = page->mem_cgroup; | |
7166 | ||
7167 | /* Readahead page, never charged */ | |
7168 | if (!memcg) | |
7169 | return 0; | |
7170 | ||
f3a53a3a TH |
7171 | if (!entry.val) { |
7172 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 7173 | return 0; |
f3a53a3a | 7174 | } |
bb98f2c5 | 7175 | |
1f47b61f VD |
7176 | memcg = mem_cgroup_id_get_online(memcg); |
7177 | ||
2d1c4980 | 7178 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg) && |
38d8b4e6 | 7179 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
7180 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
7181 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 7182 | mem_cgroup_id_put(memcg); |
37e84351 | 7183 | return -ENOMEM; |
1f47b61f | 7184 | } |
37e84351 | 7185 | |
38d8b4e6 HY |
7186 | /* Get references for the tail pages, too */ |
7187 | if (nr_pages > 1) | |
7188 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
7189 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
37e84351 | 7190 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7191 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 7192 | |
37e84351 VD |
7193 | return 0; |
7194 | } | |
7195 | ||
21afa38e | 7196 | /** |
38d8b4e6 | 7197 | * mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 7198 | * @entry: swap entry to uncharge |
38d8b4e6 | 7199 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 7200 | */ |
38d8b4e6 | 7201 | void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
7202 | { |
7203 | struct mem_cgroup *memcg; | |
7204 | unsigned short id; | |
7205 | ||
38d8b4e6 | 7206 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 7207 | rcu_read_lock(); |
adbe427b | 7208 | memcg = mem_cgroup_from_id(id); |
21afa38e | 7209 | if (memcg) { |
2d1c4980 | 7210 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg)) { |
37e84351 | 7211 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
38d8b4e6 | 7212 | page_counter_uncharge(&memcg->swap, nr_pages); |
37e84351 | 7213 | else |
38d8b4e6 | 7214 | page_counter_uncharge(&memcg->memsw, nr_pages); |
37e84351 | 7215 | } |
c9019e9b | 7216 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 7217 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
7218 | } |
7219 | rcu_read_unlock(); | |
7220 | } | |
7221 | ||
d8b38438 VD |
7222 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
7223 | { | |
7224 | long nr_swap_pages = get_nr_swap_pages(); | |
7225 | ||
eccb52e7 | 7226 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
d8b38438 VD |
7227 | return nr_swap_pages; |
7228 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
7229 | nr_swap_pages = min_t(long, nr_swap_pages, | |
bbec2e15 | 7230 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
7231 | page_counter_read(&memcg->swap)); |
7232 | return nr_swap_pages; | |
7233 | } | |
7234 | ||
5ccc5aba VD |
7235 | bool mem_cgroup_swap_full(struct page *page) |
7236 | { | |
7237 | struct mem_cgroup *memcg; | |
7238 | ||
7239 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
7240 | ||
7241 | if (vm_swap_full()) | |
7242 | return true; | |
eccb52e7 | 7243 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
5ccc5aba VD |
7244 | return false; |
7245 | ||
7246 | memcg = page->mem_cgroup; | |
7247 | if (!memcg) | |
7248 | return false; | |
7249 | ||
4b82ab4f JK |
7250 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { |
7251 | unsigned long usage = page_counter_read(&memcg->swap); | |
7252 | ||
7253 | if (usage * 2 >= READ_ONCE(memcg->swap.high) || | |
7254 | usage * 2 >= READ_ONCE(memcg->swap.max)) | |
5ccc5aba | 7255 | return true; |
4b82ab4f | 7256 | } |
5ccc5aba VD |
7257 | |
7258 | return false; | |
7259 | } | |
7260 | ||
eccb52e7 | 7261 | static int __init setup_swap_account(char *s) |
21afa38e JW |
7262 | { |
7263 | if (!strcmp(s, "1")) | |
5ab92901 | 7264 | cgroup_memory_noswap = false; |
21afa38e | 7265 | else if (!strcmp(s, "0")) |
5ab92901 | 7266 | cgroup_memory_noswap = true; |
21afa38e JW |
7267 | return 1; |
7268 | } | |
eccb52e7 | 7269 | __setup("swapaccount=", setup_swap_account); |
21afa38e | 7270 | |
37e84351 VD |
7271 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
7272 | struct cftype *cft) | |
7273 | { | |
7274 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
7275 | ||
7276 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
7277 | } | |
7278 | ||
4b82ab4f JK |
7279 | static int swap_high_show(struct seq_file *m, void *v) |
7280 | { | |
7281 | return seq_puts_memcg_tunable(m, | |
7282 | READ_ONCE(mem_cgroup_from_seq(m)->swap.high)); | |
7283 | } | |
7284 | ||
7285 | static ssize_t swap_high_write(struct kernfs_open_file *of, | |
7286 | char *buf, size_t nbytes, loff_t off) | |
7287 | { | |
7288 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7289 | unsigned long high; | |
7290 | int err; | |
7291 | ||
7292 | buf = strstrip(buf); | |
7293 | err = page_counter_memparse(buf, "max", &high); | |
7294 | if (err) | |
7295 | return err; | |
7296 | ||
7297 | page_counter_set_high(&memcg->swap, high); | |
7298 | ||
7299 | return nbytes; | |
7300 | } | |
7301 | ||
37e84351 VD |
7302 | static int swap_max_show(struct seq_file *m, void *v) |
7303 | { | |
677dc973 CD |
7304 | return seq_puts_memcg_tunable(m, |
7305 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); | |
37e84351 VD |
7306 | } |
7307 | ||
7308 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
7309 | char *buf, size_t nbytes, loff_t off) | |
7310 | { | |
7311 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7312 | unsigned long max; | |
7313 | int err; | |
7314 | ||
7315 | buf = strstrip(buf); | |
7316 | err = page_counter_memparse(buf, "max", &max); | |
7317 | if (err) | |
7318 | return err; | |
7319 | ||
be09102b | 7320 | xchg(&memcg->swap.max, max); |
37e84351 VD |
7321 | |
7322 | return nbytes; | |
7323 | } | |
7324 | ||
f3a53a3a TH |
7325 | static int swap_events_show(struct seq_file *m, void *v) |
7326 | { | |
aa9694bb | 7327 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
f3a53a3a | 7328 | |
4b82ab4f JK |
7329 | seq_printf(m, "high %lu\n", |
7330 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_HIGH])); | |
f3a53a3a TH |
7331 | seq_printf(m, "max %lu\n", |
7332 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
7333 | seq_printf(m, "fail %lu\n", | |
7334 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
7335 | ||
7336 | return 0; | |
7337 | } | |
7338 | ||
37e84351 VD |
7339 | static struct cftype swap_files[] = { |
7340 | { | |
7341 | .name = "swap.current", | |
7342 | .flags = CFTYPE_NOT_ON_ROOT, | |
7343 | .read_u64 = swap_current_read, | |
7344 | }, | |
4b82ab4f JK |
7345 | { |
7346 | .name = "swap.high", | |
7347 | .flags = CFTYPE_NOT_ON_ROOT, | |
7348 | .seq_show = swap_high_show, | |
7349 | .write = swap_high_write, | |
7350 | }, | |
37e84351 VD |
7351 | { |
7352 | .name = "swap.max", | |
7353 | .flags = CFTYPE_NOT_ON_ROOT, | |
7354 | .seq_show = swap_max_show, | |
7355 | .write = swap_max_write, | |
7356 | }, | |
f3a53a3a TH |
7357 | { |
7358 | .name = "swap.events", | |
7359 | .flags = CFTYPE_NOT_ON_ROOT, | |
7360 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
7361 | .seq_show = swap_events_show, | |
7362 | }, | |
37e84351 VD |
7363 | { } /* terminate */ |
7364 | }; | |
7365 | ||
eccb52e7 | 7366 | static struct cftype memsw_files[] = { |
21afa38e JW |
7367 | { |
7368 | .name = "memsw.usage_in_bytes", | |
7369 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
7370 | .read_u64 = mem_cgroup_read_u64, | |
7371 | }, | |
7372 | { | |
7373 | .name = "memsw.max_usage_in_bytes", | |
7374 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
7375 | .write = mem_cgroup_reset, | |
7376 | .read_u64 = mem_cgroup_read_u64, | |
7377 | }, | |
7378 | { | |
7379 | .name = "memsw.limit_in_bytes", | |
7380 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
7381 | .write = mem_cgroup_write, | |
7382 | .read_u64 = mem_cgroup_read_u64, | |
7383 | }, | |
7384 | { | |
7385 | .name = "memsw.failcnt", | |
7386 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
7387 | .write = mem_cgroup_reset, | |
7388 | .read_u64 = mem_cgroup_read_u64, | |
7389 | }, | |
7390 | { }, /* terminate */ | |
7391 | }; | |
7392 | ||
82ff165c BS |
7393 | /* |
7394 | * If mem_cgroup_swap_init() is implemented as a subsys_initcall() | |
7395 | * instead of a core_initcall(), this could mean cgroup_memory_noswap still | |
7396 | * remains set to false even when memcg is disabled via "cgroup_disable=memory" | |
7397 | * boot parameter. This may result in premature OOPS inside | |
7398 | * mem_cgroup_get_nr_swap_pages() function in corner cases. | |
7399 | */ | |
21afa38e JW |
7400 | static int __init mem_cgroup_swap_init(void) |
7401 | { | |
2d1c4980 JW |
7402 | /* No memory control -> no swap control */ |
7403 | if (mem_cgroup_disabled()) | |
7404 | cgroup_memory_noswap = true; | |
7405 | ||
7406 | if (cgroup_memory_noswap) | |
eccb52e7 JW |
7407 | return 0; |
7408 | ||
7409 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, swap_files)); | |
7410 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, memsw_files)); | |
7411 | ||
21afa38e JW |
7412 | return 0; |
7413 | } | |
82ff165c | 7414 | core_initcall(mem_cgroup_swap_init); |
21afa38e JW |
7415 | |
7416 | #endif /* CONFIG_MEMCG_SWAP */ |