]>
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 | ||
b3ff9291 | 76 | /* The number of times we should retry reclaim failures before giving up. */ |
a181b0e8 | 77 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
8cdea7c0 | 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 | |
217bc319 KH |
203 | enum charge_type { |
204 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 205 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 206 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 207 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
208 | NR_CHARGE_TYPE, |
209 | }; | |
210 | ||
8c7c6e34 | 211 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
212 | enum res_type { |
213 | _MEM, | |
214 | _MEMSWAP, | |
215 | _OOM_TYPE, | |
510fc4e1 | 216 | _KMEM, |
d55f90bf | 217 | _TCP, |
86ae53e1 GC |
218 | }; |
219 | ||
a0db00fc KS |
220 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
221 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 222 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
223 | /* Used for OOM nofiier */ |
224 | #define OOM_CONTROL (0) | |
8c7c6e34 | 225 | |
b05706f1 KT |
226 | /* |
227 | * Iteration constructs for visiting all cgroups (under a tree). If | |
228 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
229 | * be used for reference counting. | |
230 | */ | |
231 | #define for_each_mem_cgroup_tree(iter, root) \ | |
232 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ | |
233 | iter != NULL; \ | |
234 | iter = mem_cgroup_iter(root, iter, NULL)) | |
235 | ||
236 | #define for_each_mem_cgroup(iter) \ | |
237 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ | |
238 | iter != NULL; \ | |
239 | iter = mem_cgroup_iter(NULL, iter, NULL)) | |
240 | ||
7775face TH |
241 | static inline bool should_force_charge(void) |
242 | { | |
243 | return tsk_is_oom_victim(current) || fatal_signal_pending(current) || | |
244 | (current->flags & PF_EXITING); | |
245 | } | |
246 | ||
70ddf637 AV |
247 | /* Some nice accessors for the vmpressure. */ |
248 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
249 | { | |
250 | if (!memcg) | |
251 | memcg = root_mem_cgroup; | |
252 | return &memcg->vmpressure; | |
253 | } | |
254 | ||
255 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
256 | { | |
257 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
258 | } | |
259 | ||
84c07d11 | 260 | #ifdef CONFIG_MEMCG_KMEM |
bf4f0599 RG |
261 | extern spinlock_t css_set_lock; |
262 | ||
263 | static void obj_cgroup_release(struct percpu_ref *ref) | |
264 | { | |
265 | struct obj_cgroup *objcg = container_of(ref, struct obj_cgroup, refcnt); | |
266 | struct mem_cgroup *memcg; | |
267 | unsigned int nr_bytes; | |
268 | unsigned int nr_pages; | |
269 | unsigned long flags; | |
270 | ||
271 | /* | |
272 | * At this point all allocated objects are freed, and | |
273 | * objcg->nr_charged_bytes can't have an arbitrary byte value. | |
274 | * However, it can be PAGE_SIZE or (x * PAGE_SIZE). | |
275 | * | |
276 | * The following sequence can lead to it: | |
277 | * 1) CPU0: objcg == stock->cached_objcg | |
278 | * 2) CPU1: we do a small allocation (e.g. 92 bytes), | |
279 | * PAGE_SIZE bytes are charged | |
280 | * 3) CPU1: a process from another memcg is allocating something, | |
281 | * the stock if flushed, | |
282 | * objcg->nr_charged_bytes = PAGE_SIZE - 92 | |
283 | * 5) CPU0: we do release this object, | |
284 | * 92 bytes are added to stock->nr_bytes | |
285 | * 6) CPU0: stock is flushed, | |
286 | * 92 bytes are added to objcg->nr_charged_bytes | |
287 | * | |
288 | * In the result, nr_charged_bytes == PAGE_SIZE. | |
289 | * This page will be uncharged in obj_cgroup_release(). | |
290 | */ | |
291 | nr_bytes = atomic_read(&objcg->nr_charged_bytes); | |
292 | WARN_ON_ONCE(nr_bytes & (PAGE_SIZE - 1)); | |
293 | nr_pages = nr_bytes >> PAGE_SHIFT; | |
294 | ||
295 | spin_lock_irqsave(&css_set_lock, flags); | |
296 | memcg = obj_cgroup_memcg(objcg); | |
297 | if (nr_pages) | |
298 | __memcg_kmem_uncharge(memcg, nr_pages); | |
299 | list_del(&objcg->list); | |
300 | mem_cgroup_put(memcg); | |
301 | spin_unlock_irqrestore(&css_set_lock, flags); | |
302 | ||
303 | percpu_ref_exit(ref); | |
304 | kfree_rcu(objcg, rcu); | |
305 | } | |
306 | ||
307 | static struct obj_cgroup *obj_cgroup_alloc(void) | |
308 | { | |
309 | struct obj_cgroup *objcg; | |
310 | int ret; | |
311 | ||
312 | objcg = kzalloc(sizeof(struct obj_cgroup), GFP_KERNEL); | |
313 | if (!objcg) | |
314 | return NULL; | |
315 | ||
316 | ret = percpu_ref_init(&objcg->refcnt, obj_cgroup_release, 0, | |
317 | GFP_KERNEL); | |
318 | if (ret) { | |
319 | kfree(objcg); | |
320 | return NULL; | |
321 | } | |
322 | INIT_LIST_HEAD(&objcg->list); | |
323 | return objcg; | |
324 | } | |
325 | ||
326 | static void memcg_reparent_objcgs(struct mem_cgroup *memcg, | |
327 | struct mem_cgroup *parent) | |
328 | { | |
329 | struct obj_cgroup *objcg, *iter; | |
330 | ||
331 | objcg = rcu_replace_pointer(memcg->objcg, NULL, true); | |
332 | ||
333 | spin_lock_irq(&css_set_lock); | |
334 | ||
335 | /* Move active objcg to the parent's list */ | |
336 | xchg(&objcg->memcg, parent); | |
337 | css_get(&parent->css); | |
338 | list_add(&objcg->list, &parent->objcg_list); | |
339 | ||
340 | /* Move already reparented objcgs to the parent's list */ | |
341 | list_for_each_entry(iter, &memcg->objcg_list, list) { | |
342 | css_get(&parent->css); | |
343 | xchg(&iter->memcg, parent); | |
344 | css_put(&memcg->css); | |
345 | } | |
346 | list_splice(&memcg->objcg_list, &parent->objcg_list); | |
347 | ||
348 | spin_unlock_irq(&css_set_lock); | |
349 | ||
350 | percpu_ref_kill(&objcg->refcnt); | |
351 | } | |
352 | ||
55007d84 | 353 | /* |
9855609b | 354 | * This will be used as a shrinker list's index. |
b8627835 LZ |
355 | * The main reason for not using cgroup id for this: |
356 | * this works better in sparse environments, where we have a lot of memcgs, | |
357 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
358 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
359 | * 200 entry array for that. | |
55007d84 | 360 | * |
dbcf73e2 VD |
361 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
362 | * will double each time we have to increase it. | |
55007d84 | 363 | */ |
dbcf73e2 VD |
364 | static DEFINE_IDA(memcg_cache_ida); |
365 | int memcg_nr_cache_ids; | |
749c5415 | 366 | |
05257a1a VD |
367 | /* Protects memcg_nr_cache_ids */ |
368 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
369 | ||
370 | void memcg_get_cache_ids(void) | |
371 | { | |
372 | down_read(&memcg_cache_ids_sem); | |
373 | } | |
374 | ||
375 | void memcg_put_cache_ids(void) | |
376 | { | |
377 | up_read(&memcg_cache_ids_sem); | |
378 | } | |
379 | ||
55007d84 GC |
380 | /* |
381 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
382 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
383 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
384 | * tunable, but that is strictly not necessary. | |
385 | * | |
b8627835 | 386 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
387 | * this constant directly from cgroup, but it is understandable that this is |
388 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 389 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
390 | * increase ours as well if it increases. |
391 | */ | |
392 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 393 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 394 | |
d7f25f8a GC |
395 | /* |
396 | * A lot of the calls to the cache allocation functions are expected to be | |
272911a4 | 397 | * inlined by the compiler. Since the calls to memcg_slab_pre_alloc_hook() are |
d7f25f8a GC |
398 | * conditional to this static branch, we'll have to allow modules that does |
399 | * kmem_cache_alloc and the such to see this symbol as well | |
400 | */ | |
ef12947c | 401 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
d7f25f8a | 402 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
0a432dcb | 403 | #endif |
17cc4dfe | 404 | |
0a4465d3 KT |
405 | static int memcg_shrinker_map_size; |
406 | static DEFINE_MUTEX(memcg_shrinker_map_mutex); | |
407 | ||
408 | static void memcg_free_shrinker_map_rcu(struct rcu_head *head) | |
409 | { | |
410 | kvfree(container_of(head, struct memcg_shrinker_map, rcu)); | |
411 | } | |
412 | ||
413 | static int memcg_expand_one_shrinker_map(struct mem_cgroup *memcg, | |
414 | int size, int old_size) | |
415 | { | |
416 | struct memcg_shrinker_map *new, *old; | |
417 | int nid; | |
418 | ||
419 | lockdep_assert_held(&memcg_shrinker_map_mutex); | |
420 | ||
421 | for_each_node(nid) { | |
422 | old = rcu_dereference_protected( | |
423 | mem_cgroup_nodeinfo(memcg, nid)->shrinker_map, true); | |
424 | /* Not yet online memcg */ | |
425 | if (!old) | |
426 | return 0; | |
427 | ||
86daf94e | 428 | new = kvmalloc_node(sizeof(*new) + size, GFP_KERNEL, nid); |
0a4465d3 KT |
429 | if (!new) |
430 | return -ENOMEM; | |
431 | ||
432 | /* Set all old bits, clear all new bits */ | |
433 | memset(new->map, (int)0xff, old_size); | |
434 | memset((void *)new->map + old_size, 0, size - old_size); | |
435 | ||
436 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, new); | |
437 | call_rcu(&old->rcu, memcg_free_shrinker_map_rcu); | |
438 | } | |
439 | ||
440 | return 0; | |
441 | } | |
442 | ||
443 | static void memcg_free_shrinker_maps(struct mem_cgroup *memcg) | |
444 | { | |
445 | struct mem_cgroup_per_node *pn; | |
446 | struct memcg_shrinker_map *map; | |
447 | int nid; | |
448 | ||
449 | if (mem_cgroup_is_root(memcg)) | |
450 | return; | |
451 | ||
452 | for_each_node(nid) { | |
453 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
454 | map = rcu_dereference_protected(pn->shrinker_map, true); | |
455 | if (map) | |
456 | kvfree(map); | |
457 | rcu_assign_pointer(pn->shrinker_map, NULL); | |
458 | } | |
459 | } | |
460 | ||
461 | static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg) | |
462 | { | |
463 | struct memcg_shrinker_map *map; | |
464 | int nid, size, ret = 0; | |
465 | ||
466 | if (mem_cgroup_is_root(memcg)) | |
467 | return 0; | |
468 | ||
469 | mutex_lock(&memcg_shrinker_map_mutex); | |
470 | size = memcg_shrinker_map_size; | |
471 | for_each_node(nid) { | |
86daf94e | 472 | map = kvzalloc_node(sizeof(*map) + size, GFP_KERNEL, nid); |
0a4465d3 KT |
473 | if (!map) { |
474 | memcg_free_shrinker_maps(memcg); | |
475 | ret = -ENOMEM; | |
476 | break; | |
477 | } | |
478 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, map); | |
479 | } | |
480 | mutex_unlock(&memcg_shrinker_map_mutex); | |
481 | ||
482 | return ret; | |
483 | } | |
484 | ||
485 | int memcg_expand_shrinker_maps(int new_id) | |
486 | { | |
487 | int size, old_size, ret = 0; | |
488 | struct mem_cgroup *memcg; | |
489 | ||
490 | size = DIV_ROUND_UP(new_id + 1, BITS_PER_LONG) * sizeof(unsigned long); | |
491 | old_size = memcg_shrinker_map_size; | |
492 | if (size <= old_size) | |
493 | return 0; | |
494 | ||
495 | mutex_lock(&memcg_shrinker_map_mutex); | |
496 | if (!root_mem_cgroup) | |
497 | goto unlock; | |
498 | ||
499 | for_each_mem_cgroup(memcg) { | |
500 | if (mem_cgroup_is_root(memcg)) | |
501 | continue; | |
502 | ret = memcg_expand_one_shrinker_map(memcg, size, old_size); | |
75866af6 VA |
503 | if (ret) { |
504 | mem_cgroup_iter_break(NULL, memcg); | |
0a4465d3 | 505 | goto unlock; |
75866af6 | 506 | } |
0a4465d3 KT |
507 | } |
508 | unlock: | |
509 | if (!ret) | |
510 | memcg_shrinker_map_size = size; | |
511 | mutex_unlock(&memcg_shrinker_map_mutex); | |
512 | return ret; | |
513 | } | |
fae91d6d KT |
514 | |
515 | void memcg_set_shrinker_bit(struct mem_cgroup *memcg, int nid, int shrinker_id) | |
516 | { | |
517 | if (shrinker_id >= 0 && memcg && !mem_cgroup_is_root(memcg)) { | |
518 | struct memcg_shrinker_map *map; | |
519 | ||
520 | rcu_read_lock(); | |
521 | map = rcu_dereference(memcg->nodeinfo[nid]->shrinker_map); | |
f90280d6 KT |
522 | /* Pairs with smp mb in shrink_slab() */ |
523 | smp_mb__before_atomic(); | |
fae91d6d KT |
524 | set_bit(shrinker_id, map->map); |
525 | rcu_read_unlock(); | |
526 | } | |
527 | } | |
528 | ||
ad7fa852 TH |
529 | /** |
530 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
531 | * @page: page of interest | |
532 | * | |
533 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
534 | * with @page is returned. The returned css remains associated with @page | |
535 | * until it is released. | |
536 | * | |
537 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
538 | * is returned. | |
ad7fa852 TH |
539 | */ |
540 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
541 | { | |
542 | struct mem_cgroup *memcg; | |
543 | ||
ad7fa852 TH |
544 | memcg = page->mem_cgroup; |
545 | ||
9e10a130 | 546 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
547 | memcg = root_mem_cgroup; |
548 | ||
ad7fa852 TH |
549 | return &memcg->css; |
550 | } | |
551 | ||
2fc04524 VD |
552 | /** |
553 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
554 | * @page: the page | |
555 | * | |
556 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
557 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
558 | * is safe to call this function without holding a reference to @page. | |
559 | * | |
560 | * Note, this function is inherently racy, because there is nothing to prevent | |
561 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
562 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
563 | * do not care (such as procfs interfaces). | |
564 | */ | |
565 | ino_t page_cgroup_ino(struct page *page) | |
566 | { | |
567 | struct mem_cgroup *memcg; | |
568 | unsigned long ino = 0; | |
569 | ||
570 | rcu_read_lock(); | |
9855609b | 571 | memcg = page->mem_cgroup; |
286e04b8 | 572 | |
9855609b RG |
573 | /* |
574 | * The lowest bit set means that memcg isn't a valid | |
575 | * memcg pointer, but a obj_cgroups pointer. | |
576 | * In this case the page is shared and doesn't belong | |
577 | * to any specific memory cgroup. | |
578 | */ | |
579 | if ((unsigned long) memcg & 0x1UL) | |
580 | memcg = NULL; | |
286e04b8 | 581 | |
2fc04524 VD |
582 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) |
583 | memcg = parent_mem_cgroup(memcg); | |
584 | if (memcg) | |
585 | ino = cgroup_ino(memcg->css.cgroup); | |
586 | rcu_read_unlock(); | |
587 | return ino; | |
588 | } | |
589 | ||
ef8f2327 MG |
590 | static struct mem_cgroup_per_node * |
591 | mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page) | |
f64c3f54 | 592 | { |
97a6c37b | 593 | int nid = page_to_nid(page); |
f64c3f54 | 594 | |
ef8f2327 | 595 | return memcg->nodeinfo[nid]; |
f64c3f54 BS |
596 | } |
597 | ||
ef8f2327 MG |
598 | static struct mem_cgroup_tree_per_node * |
599 | soft_limit_tree_node(int nid) | |
bb4cc1a8 | 600 | { |
ef8f2327 | 601 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
602 | } |
603 | ||
ef8f2327 | 604 | static struct mem_cgroup_tree_per_node * |
bb4cc1a8 AM |
605 | soft_limit_tree_from_page(struct page *page) |
606 | { | |
607 | int nid = page_to_nid(page); | |
bb4cc1a8 | 608 | |
ef8f2327 | 609 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
610 | } |
611 | ||
ef8f2327 MG |
612 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
613 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 614 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
615 | { |
616 | struct rb_node **p = &mctz->rb_root.rb_node; | |
617 | struct rb_node *parent = NULL; | |
ef8f2327 | 618 | struct mem_cgroup_per_node *mz_node; |
fa90b2fd | 619 | bool rightmost = true; |
bb4cc1a8 AM |
620 | |
621 | if (mz->on_tree) | |
622 | return; | |
623 | ||
624 | mz->usage_in_excess = new_usage_in_excess; | |
625 | if (!mz->usage_in_excess) | |
626 | return; | |
627 | while (*p) { | |
628 | parent = *p; | |
ef8f2327 | 629 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 | 630 | tree_node); |
fa90b2fd | 631 | if (mz->usage_in_excess < mz_node->usage_in_excess) { |
bb4cc1a8 | 632 | p = &(*p)->rb_left; |
fa90b2fd DB |
633 | rightmost = false; |
634 | } | |
635 | ||
bb4cc1a8 AM |
636 | /* |
637 | * We can't avoid mem cgroups that are over their soft | |
638 | * limit by the same amount | |
639 | */ | |
640 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
641 | p = &(*p)->rb_right; | |
642 | } | |
fa90b2fd DB |
643 | |
644 | if (rightmost) | |
645 | mctz->rb_rightmost = &mz->tree_node; | |
646 | ||
bb4cc1a8 AM |
647 | rb_link_node(&mz->tree_node, parent, p); |
648 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
649 | mz->on_tree = true; | |
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 AM |
654 | { |
655 | if (!mz->on_tree) | |
656 | return; | |
fa90b2fd DB |
657 | |
658 | if (&mz->tree_node == mctz->rb_rightmost) | |
659 | mctz->rb_rightmost = rb_prev(&mz->tree_node); | |
660 | ||
bb4cc1a8 AM |
661 | rb_erase(&mz->tree_node, &mctz->rb_root); |
662 | mz->on_tree = false; | |
663 | } | |
664 | ||
ef8f2327 MG |
665 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
666 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 667 | { |
0a31bc97 JW |
668 | unsigned long flags; |
669 | ||
670 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 671 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 672 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
673 | } |
674 | ||
3e32cb2e JW |
675 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
676 | { | |
677 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 678 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
679 | unsigned long excess = 0; |
680 | ||
681 | if (nr_pages > soft_limit) | |
682 | excess = nr_pages - soft_limit; | |
683 | ||
684 | return excess; | |
685 | } | |
bb4cc1a8 AM |
686 | |
687 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
688 | { | |
3e32cb2e | 689 | unsigned long excess; |
ef8f2327 MG |
690 | struct mem_cgroup_per_node *mz; |
691 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 692 | |
e231875b | 693 | mctz = soft_limit_tree_from_page(page); |
bfc7228b LD |
694 | if (!mctz) |
695 | return; | |
bb4cc1a8 AM |
696 | /* |
697 | * Necessary to update all ancestors when hierarchy is used. | |
698 | * because their event counter is not touched. | |
699 | */ | |
700 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
ef8f2327 | 701 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
3e32cb2e | 702 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
703 | /* |
704 | * We have to update the tree if mz is on RB-tree or | |
705 | * mem is over its softlimit. | |
706 | */ | |
707 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
708 | unsigned long flags; |
709 | ||
710 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
711 | /* if on-tree, remove it */ |
712 | if (mz->on_tree) | |
cf2c8127 | 713 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
714 | /* |
715 | * Insert again. mz->usage_in_excess will be updated. | |
716 | * If excess is 0, no tree ops. | |
717 | */ | |
cf2c8127 | 718 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 719 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
720 | } |
721 | } | |
722 | } | |
723 | ||
724 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
725 | { | |
ef8f2327 MG |
726 | struct mem_cgroup_tree_per_node *mctz; |
727 | struct mem_cgroup_per_node *mz; | |
728 | int nid; | |
bb4cc1a8 | 729 | |
e231875b | 730 | for_each_node(nid) { |
ef8f2327 MG |
731 | mz = mem_cgroup_nodeinfo(memcg, nid); |
732 | mctz = soft_limit_tree_node(nid); | |
bfc7228b LD |
733 | if (mctz) |
734 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
735 | } |
736 | } | |
737 | ||
ef8f2327 MG |
738 | static struct mem_cgroup_per_node * |
739 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 740 | { |
ef8f2327 | 741 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
742 | |
743 | retry: | |
744 | mz = NULL; | |
fa90b2fd | 745 | if (!mctz->rb_rightmost) |
bb4cc1a8 AM |
746 | goto done; /* Nothing to reclaim from */ |
747 | ||
fa90b2fd DB |
748 | mz = rb_entry(mctz->rb_rightmost, |
749 | struct mem_cgroup_per_node, tree_node); | |
bb4cc1a8 AM |
750 | /* |
751 | * Remove the node now but someone else can add it back, | |
752 | * we will to add it back at the end of reclaim to its correct | |
753 | * position in the tree. | |
754 | */ | |
cf2c8127 | 755 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 756 | if (!soft_limit_excess(mz->memcg) || |
8965aa28 | 757 | !css_tryget(&mz->memcg->css)) |
bb4cc1a8 AM |
758 | goto retry; |
759 | done: | |
760 | return mz; | |
761 | } | |
762 | ||
ef8f2327 MG |
763 | static struct mem_cgroup_per_node * |
764 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 765 | { |
ef8f2327 | 766 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 767 | |
0a31bc97 | 768 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 769 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 770 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
771 | return mz; |
772 | } | |
773 | ||
db9adbcb JW |
774 | /** |
775 | * __mod_memcg_state - update cgroup memory statistics | |
776 | * @memcg: the memory cgroup | |
777 | * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item | |
778 | * @val: delta to add to the counter, can be negative | |
779 | */ | |
780 | void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val) | |
781 | { | |
ea426c2a | 782 | long x, threshold = MEMCG_CHARGE_BATCH; |
db9adbcb JW |
783 | |
784 | if (mem_cgroup_disabled()) | |
785 | return; | |
786 | ||
ea426c2a RG |
787 | if (vmstat_item_in_bytes(idx)) |
788 | threshold <<= PAGE_SHIFT; | |
789 | ||
db9adbcb | 790 | x = val + __this_cpu_read(memcg->vmstats_percpu->stat[idx]); |
ea426c2a | 791 | if (unlikely(abs(x) > threshold)) { |
42a30035 JW |
792 | struct mem_cgroup *mi; |
793 | ||
766a4c19 YS |
794 | /* |
795 | * Batch local counters to keep them in sync with | |
796 | * the hierarchical ones. | |
797 | */ | |
798 | __this_cpu_add(memcg->vmstats_local->stat[idx], x); | |
42a30035 JW |
799 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
800 | atomic_long_add(x, &mi->vmstats[idx]); | |
db9adbcb JW |
801 | x = 0; |
802 | } | |
803 | __this_cpu_write(memcg->vmstats_percpu->stat[idx], x); | |
804 | } | |
805 | ||
42a30035 JW |
806 | static struct mem_cgroup_per_node * |
807 | parent_nodeinfo(struct mem_cgroup_per_node *pn, int nid) | |
808 | { | |
809 | struct mem_cgroup *parent; | |
810 | ||
811 | parent = parent_mem_cgroup(pn->memcg); | |
812 | if (!parent) | |
813 | return NULL; | |
814 | return mem_cgroup_nodeinfo(parent, nid); | |
815 | } | |
816 | ||
eedc4e5a RG |
817 | void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, |
818 | int val) | |
db9adbcb JW |
819 | { |
820 | struct mem_cgroup_per_node *pn; | |
42a30035 | 821 | struct mem_cgroup *memcg; |
ea426c2a | 822 | long x, threshold = MEMCG_CHARGE_BATCH; |
db9adbcb | 823 | |
db9adbcb | 824 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
42a30035 | 825 | memcg = pn->memcg; |
db9adbcb JW |
826 | |
827 | /* Update memcg */ | |
42a30035 | 828 | __mod_memcg_state(memcg, idx, val); |
db9adbcb | 829 | |
b4c46484 RG |
830 | /* Update lruvec */ |
831 | __this_cpu_add(pn->lruvec_stat_local->count[idx], val); | |
832 | ||
ea426c2a RG |
833 | if (vmstat_item_in_bytes(idx)) |
834 | threshold <<= PAGE_SHIFT; | |
835 | ||
db9adbcb | 836 | x = val + __this_cpu_read(pn->lruvec_stat_cpu->count[idx]); |
ea426c2a | 837 | if (unlikely(abs(x) > threshold)) { |
eedc4e5a | 838 | pg_data_t *pgdat = lruvec_pgdat(lruvec); |
42a30035 JW |
839 | struct mem_cgroup_per_node *pi; |
840 | ||
42a30035 JW |
841 | for (pi = pn; pi; pi = parent_nodeinfo(pi, pgdat->node_id)) |
842 | atomic_long_add(x, &pi->lruvec_stat[idx]); | |
db9adbcb JW |
843 | x = 0; |
844 | } | |
845 | __this_cpu_write(pn->lruvec_stat_cpu->count[idx], x); | |
846 | } | |
847 | ||
eedc4e5a RG |
848 | /** |
849 | * __mod_lruvec_state - update lruvec memory statistics | |
850 | * @lruvec: the lruvec | |
851 | * @idx: the stat item | |
852 | * @val: delta to add to the counter, can be negative | |
853 | * | |
854 | * The lruvec is the intersection of the NUMA node and a cgroup. This | |
855 | * function updates the all three counters that are affected by a | |
856 | * change of state at this level: per-node, per-cgroup, per-lruvec. | |
857 | */ | |
858 | void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, | |
859 | int val) | |
860 | { | |
861 | /* Update node */ | |
862 | __mod_node_page_state(lruvec_pgdat(lruvec), idx, val); | |
863 | ||
864 | /* Update memcg and lruvec */ | |
865 | if (!mem_cgroup_disabled()) | |
866 | __mod_memcg_lruvec_state(lruvec, idx, val); | |
867 | } | |
868 | ||
ec9f0238 RG |
869 | void __mod_lruvec_slab_state(void *p, enum node_stat_item idx, int val) |
870 | { | |
4f103c63 | 871 | pg_data_t *pgdat = page_pgdat(virt_to_page(p)); |
ec9f0238 RG |
872 | struct mem_cgroup *memcg; |
873 | struct lruvec *lruvec; | |
874 | ||
875 | rcu_read_lock(); | |
4f103c63 | 876 | memcg = mem_cgroup_from_obj(p); |
ec9f0238 RG |
877 | |
878 | /* Untracked pages have no memcg, no lruvec. Update only the node */ | |
879 | if (!memcg || memcg == root_mem_cgroup) { | |
880 | __mod_node_page_state(pgdat, idx, val); | |
881 | } else { | |
867e5e1d | 882 | lruvec = mem_cgroup_lruvec(memcg, pgdat); |
ec9f0238 RG |
883 | __mod_lruvec_state(lruvec, idx, val); |
884 | } | |
885 | rcu_read_unlock(); | |
886 | } | |
887 | ||
8380ce47 RG |
888 | void mod_memcg_obj_state(void *p, int idx, int val) |
889 | { | |
890 | struct mem_cgroup *memcg; | |
891 | ||
892 | rcu_read_lock(); | |
893 | memcg = mem_cgroup_from_obj(p); | |
894 | if (memcg) | |
895 | mod_memcg_state(memcg, idx, val); | |
896 | rcu_read_unlock(); | |
897 | } | |
898 | ||
db9adbcb JW |
899 | /** |
900 | * __count_memcg_events - account VM events in a cgroup | |
901 | * @memcg: the memory cgroup | |
902 | * @idx: the event item | |
903 | * @count: the number of events that occured | |
904 | */ | |
905 | void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, | |
906 | unsigned long count) | |
907 | { | |
908 | unsigned long x; | |
909 | ||
910 | if (mem_cgroup_disabled()) | |
911 | return; | |
912 | ||
913 | x = count + __this_cpu_read(memcg->vmstats_percpu->events[idx]); | |
914 | if (unlikely(x > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
915 | struct mem_cgroup *mi; |
916 | ||
766a4c19 YS |
917 | /* |
918 | * Batch local counters to keep them in sync with | |
919 | * the hierarchical ones. | |
920 | */ | |
921 | __this_cpu_add(memcg->vmstats_local->events[idx], x); | |
42a30035 JW |
922 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
923 | atomic_long_add(x, &mi->vmevents[idx]); | |
db9adbcb JW |
924 | x = 0; |
925 | } | |
926 | __this_cpu_write(memcg->vmstats_percpu->events[idx], x); | |
927 | } | |
928 | ||
42a30035 | 929 | static unsigned long memcg_events(struct mem_cgroup *memcg, int event) |
e9f8974f | 930 | { |
871789d4 | 931 | return atomic_long_read(&memcg->vmevents[event]); |
e9f8974f JW |
932 | } |
933 | ||
42a30035 JW |
934 | static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event) |
935 | { | |
815744d7 JW |
936 | long x = 0; |
937 | int cpu; | |
938 | ||
939 | for_each_possible_cpu(cpu) | |
940 | x += per_cpu(memcg->vmstats_local->events[event], cpu); | |
941 | return x; | |
42a30035 JW |
942 | } |
943 | ||
c0ff4b85 | 944 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 945 | struct page *page, |
3fba69a5 | 946 | int nr_pages) |
d52aa412 | 947 | { |
e401f176 KH |
948 | /* pagein of a big page is an event. So, ignore page size */ |
949 | if (nr_pages > 0) | |
c9019e9b | 950 | __count_memcg_events(memcg, PGPGIN, 1); |
3751d604 | 951 | else { |
c9019e9b | 952 | __count_memcg_events(memcg, PGPGOUT, 1); |
3751d604 KH |
953 | nr_pages = -nr_pages; /* for event */ |
954 | } | |
e401f176 | 955 | |
871789d4 | 956 | __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages); |
6d12e2d8 KH |
957 | } |
958 | ||
f53d7ce3 JW |
959 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
960 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
961 | { |
962 | unsigned long val, next; | |
963 | ||
871789d4 CD |
964 | val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events); |
965 | next = __this_cpu_read(memcg->vmstats_percpu->targets[target]); | |
7a159cc9 | 966 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 967 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
968 | switch (target) { |
969 | case MEM_CGROUP_TARGET_THRESH: | |
970 | next = val + THRESHOLDS_EVENTS_TARGET; | |
971 | break; | |
bb4cc1a8 AM |
972 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
973 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
974 | break; | |
f53d7ce3 JW |
975 | default: |
976 | break; | |
977 | } | |
871789d4 | 978 | __this_cpu_write(memcg->vmstats_percpu->targets[target], next); |
f53d7ce3 | 979 | return true; |
7a159cc9 | 980 | } |
f53d7ce3 | 981 | return false; |
d2265e6f KH |
982 | } |
983 | ||
984 | /* | |
985 | * Check events in order. | |
986 | * | |
987 | */ | |
c0ff4b85 | 988 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
989 | { |
990 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
991 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
992 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 993 | bool do_softlimit; |
f53d7ce3 | 994 | |
bb4cc1a8 AM |
995 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
996 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
c0ff4b85 | 997 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
998 | if (unlikely(do_softlimit)) |
999 | mem_cgroup_update_tree(memcg, page); | |
0a31bc97 | 1000 | } |
d2265e6f KH |
1001 | } |
1002 | ||
cf475ad2 | 1003 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 1004 | { |
31a78f23 BS |
1005 | /* |
1006 | * mm_update_next_owner() may clear mm->owner to NULL | |
1007 | * if it races with swapoff, page migration, etc. | |
1008 | * So this can be called with p == NULL. | |
1009 | */ | |
1010 | if (unlikely(!p)) | |
1011 | return NULL; | |
1012 | ||
073219e9 | 1013 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 1014 | } |
33398cf2 | 1015 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 1016 | |
d46eb14b SB |
1017 | /** |
1018 | * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg. | |
1019 | * @mm: mm from which memcg should be extracted. It can be NULL. | |
1020 | * | |
1021 | * Obtain a reference on mm->memcg and returns it if successful. Otherwise | |
1022 | * root_mem_cgroup is returned. However if mem_cgroup is disabled, NULL is | |
1023 | * returned. | |
1024 | */ | |
1025 | struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) | |
54595fe2 | 1026 | { |
d46eb14b SB |
1027 | struct mem_cgroup *memcg; |
1028 | ||
1029 | if (mem_cgroup_disabled()) | |
1030 | return NULL; | |
0b7f569e | 1031 | |
54595fe2 KH |
1032 | rcu_read_lock(); |
1033 | do { | |
6f6acb00 MH |
1034 | /* |
1035 | * Page cache insertions can happen withou an | |
1036 | * actual mm context, e.g. during disk probing | |
1037 | * on boot, loopback IO, acct() writes etc. | |
1038 | */ | |
1039 | if (unlikely(!mm)) | |
df381975 | 1040 | memcg = root_mem_cgroup; |
6f6acb00 MH |
1041 | else { |
1042 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
1043 | if (unlikely(!memcg)) | |
1044 | memcg = root_mem_cgroup; | |
1045 | } | |
00d484f3 | 1046 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 1047 | rcu_read_unlock(); |
c0ff4b85 | 1048 | return memcg; |
54595fe2 | 1049 | } |
d46eb14b SB |
1050 | EXPORT_SYMBOL(get_mem_cgroup_from_mm); |
1051 | ||
f745c6f5 SB |
1052 | /** |
1053 | * get_mem_cgroup_from_page: Obtain a reference on given page's memcg. | |
1054 | * @page: page from which memcg should be extracted. | |
1055 | * | |
1056 | * Obtain a reference on page->memcg and returns it if successful. Otherwise | |
1057 | * root_mem_cgroup is returned. | |
1058 | */ | |
1059 | struct mem_cgroup *get_mem_cgroup_from_page(struct page *page) | |
1060 | { | |
1061 | struct mem_cgroup *memcg = page->mem_cgroup; | |
1062 | ||
1063 | if (mem_cgroup_disabled()) | |
1064 | return NULL; | |
1065 | ||
1066 | rcu_read_lock(); | |
8965aa28 SB |
1067 | /* Page should not get uncharged and freed memcg under us. */ |
1068 | if (!memcg || WARN_ON_ONCE(!css_tryget(&memcg->css))) | |
f745c6f5 SB |
1069 | memcg = root_mem_cgroup; |
1070 | rcu_read_unlock(); | |
1071 | return memcg; | |
1072 | } | |
1073 | EXPORT_SYMBOL(get_mem_cgroup_from_page); | |
1074 | ||
d46eb14b SB |
1075 | /** |
1076 | * If current->active_memcg is non-NULL, do not fallback to current->mm->memcg. | |
1077 | */ | |
1078 | static __always_inline struct mem_cgroup *get_mem_cgroup_from_current(void) | |
1079 | { | |
1080 | if (unlikely(current->active_memcg)) { | |
8965aa28 | 1081 | struct mem_cgroup *memcg; |
d46eb14b SB |
1082 | |
1083 | rcu_read_lock(); | |
8965aa28 SB |
1084 | /* current->active_memcg must hold a ref. */ |
1085 | if (WARN_ON_ONCE(!css_tryget(¤t->active_memcg->css))) | |
1086 | memcg = root_mem_cgroup; | |
1087 | else | |
d46eb14b SB |
1088 | memcg = current->active_memcg; |
1089 | rcu_read_unlock(); | |
1090 | return memcg; | |
1091 | } | |
1092 | return get_mem_cgroup_from_mm(current->mm); | |
1093 | } | |
54595fe2 | 1094 | |
5660048c JW |
1095 | /** |
1096 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1097 | * @root: hierarchy root | |
1098 | * @prev: previously returned memcg, NULL on first invocation | |
1099 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1100 | * | |
1101 | * Returns references to children of the hierarchy below @root, or | |
1102 | * @root itself, or %NULL after a full round-trip. | |
1103 | * | |
1104 | * Caller must pass the return value in @prev on subsequent | |
1105 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1106 | * to cancel a hierarchy walk before the round-trip is complete. | |
1107 | * | |
b213b54f | 1108 | * Reclaimers can specify a node and a priority level in @reclaim to |
5660048c | 1109 | * divide up the memcgs in the hierarchy among all concurrent |
b213b54f | 1110 | * reclaimers operating on the same node and priority. |
5660048c | 1111 | */ |
694fbc0f | 1112 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1113 | struct mem_cgroup *prev, |
694fbc0f | 1114 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1115 | { |
3f649ab7 | 1116 | struct mem_cgroup_reclaim_iter *iter; |
5ac8fb31 | 1117 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 1118 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1119 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1120 | |
694fbc0f AM |
1121 | if (mem_cgroup_disabled()) |
1122 | return NULL; | |
5660048c | 1123 | |
9f3a0d09 JW |
1124 | if (!root) |
1125 | root = root_mem_cgroup; | |
7d74b06f | 1126 | |
9f3a0d09 | 1127 | if (prev && !reclaim) |
5ac8fb31 | 1128 | pos = prev; |
14067bb3 | 1129 | |
9f3a0d09 JW |
1130 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1131 | if (prev) | |
5ac8fb31 | 1132 | goto out; |
694fbc0f | 1133 | return root; |
9f3a0d09 | 1134 | } |
14067bb3 | 1135 | |
542f85f9 | 1136 | rcu_read_lock(); |
5f578161 | 1137 | |
5ac8fb31 | 1138 | if (reclaim) { |
ef8f2327 | 1139 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 1140 | |
ef8f2327 | 1141 | mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id); |
9da83f3f | 1142 | iter = &mz->iter; |
5ac8fb31 JW |
1143 | |
1144 | if (prev && reclaim->generation != iter->generation) | |
1145 | goto out_unlock; | |
1146 | ||
6df38689 | 1147 | while (1) { |
4db0c3c2 | 1148 | pos = READ_ONCE(iter->position); |
6df38689 VD |
1149 | if (!pos || css_tryget(&pos->css)) |
1150 | break; | |
5ac8fb31 | 1151 | /* |
6df38689 VD |
1152 | * css reference reached zero, so iter->position will |
1153 | * be cleared by ->css_released. However, we should not | |
1154 | * rely on this happening soon, because ->css_released | |
1155 | * is called from a work queue, and by busy-waiting we | |
1156 | * might block it. So we clear iter->position right | |
1157 | * away. | |
5ac8fb31 | 1158 | */ |
6df38689 VD |
1159 | (void)cmpxchg(&iter->position, pos, NULL); |
1160 | } | |
5ac8fb31 JW |
1161 | } |
1162 | ||
1163 | if (pos) | |
1164 | css = &pos->css; | |
1165 | ||
1166 | for (;;) { | |
1167 | css = css_next_descendant_pre(css, &root->css); | |
1168 | if (!css) { | |
1169 | /* | |
1170 | * Reclaimers share the hierarchy walk, and a | |
1171 | * new one might jump in right at the end of | |
1172 | * the hierarchy - make sure they see at least | |
1173 | * one group and restart from the beginning. | |
1174 | */ | |
1175 | if (!prev) | |
1176 | continue; | |
1177 | break; | |
527a5ec9 | 1178 | } |
7d74b06f | 1179 | |
5ac8fb31 JW |
1180 | /* |
1181 | * Verify the css and acquire a reference. The root | |
1182 | * is provided by the caller, so we know it's alive | |
1183 | * and kicking, and don't take an extra reference. | |
1184 | */ | |
1185 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 1186 | |
5ac8fb31 JW |
1187 | if (css == &root->css) |
1188 | break; | |
14067bb3 | 1189 | |
0b8f73e1 JW |
1190 | if (css_tryget(css)) |
1191 | break; | |
9f3a0d09 | 1192 | |
5ac8fb31 | 1193 | memcg = NULL; |
9f3a0d09 | 1194 | } |
5ac8fb31 JW |
1195 | |
1196 | if (reclaim) { | |
5ac8fb31 | 1197 | /* |
6df38689 VD |
1198 | * The position could have already been updated by a competing |
1199 | * thread, so check that the value hasn't changed since we read | |
1200 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 1201 | */ |
6df38689 VD |
1202 | (void)cmpxchg(&iter->position, pos, memcg); |
1203 | ||
5ac8fb31 JW |
1204 | if (pos) |
1205 | css_put(&pos->css); | |
1206 | ||
1207 | if (!memcg) | |
1208 | iter->generation++; | |
1209 | else if (!prev) | |
1210 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1211 | } |
5ac8fb31 | 1212 | |
542f85f9 MH |
1213 | out_unlock: |
1214 | rcu_read_unlock(); | |
5ac8fb31 | 1215 | out: |
c40046f3 MH |
1216 | if (prev && prev != root) |
1217 | css_put(&prev->css); | |
1218 | ||
9f3a0d09 | 1219 | return memcg; |
14067bb3 | 1220 | } |
7d74b06f | 1221 | |
5660048c JW |
1222 | /** |
1223 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1224 | * @root: hierarchy root | |
1225 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1226 | */ | |
1227 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1228 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1229 | { |
1230 | if (!root) | |
1231 | root = root_mem_cgroup; | |
1232 | if (prev && prev != root) | |
1233 | css_put(&prev->css); | |
1234 | } | |
7d74b06f | 1235 | |
54a83d6b MC |
1236 | static void __invalidate_reclaim_iterators(struct mem_cgroup *from, |
1237 | struct mem_cgroup *dead_memcg) | |
6df38689 | 1238 | { |
6df38689 | 1239 | struct mem_cgroup_reclaim_iter *iter; |
ef8f2327 MG |
1240 | struct mem_cgroup_per_node *mz; |
1241 | int nid; | |
6df38689 | 1242 | |
54a83d6b MC |
1243 | for_each_node(nid) { |
1244 | mz = mem_cgroup_nodeinfo(from, nid); | |
9da83f3f YS |
1245 | iter = &mz->iter; |
1246 | cmpxchg(&iter->position, dead_memcg, NULL); | |
6df38689 VD |
1247 | } |
1248 | } | |
1249 | ||
54a83d6b MC |
1250 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1251 | { | |
1252 | struct mem_cgroup *memcg = dead_memcg; | |
1253 | struct mem_cgroup *last; | |
1254 | ||
1255 | do { | |
1256 | __invalidate_reclaim_iterators(memcg, dead_memcg); | |
1257 | last = memcg; | |
1258 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1259 | ||
1260 | /* | |
1261 | * When cgruop1 non-hierarchy mode is used, | |
1262 | * parent_mem_cgroup() does not walk all the way up to the | |
1263 | * cgroup root (root_mem_cgroup). So we have to handle | |
1264 | * dead_memcg from cgroup root separately. | |
1265 | */ | |
1266 | if (last != root_mem_cgroup) | |
1267 | __invalidate_reclaim_iterators(root_mem_cgroup, | |
1268 | dead_memcg); | |
1269 | } | |
1270 | ||
7c5f64f8 VD |
1271 | /** |
1272 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1273 | * @memcg: hierarchy root | |
1274 | * @fn: function to call for each task | |
1275 | * @arg: argument passed to @fn | |
1276 | * | |
1277 | * This function iterates over tasks attached to @memcg or to any of its | |
1278 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
1279 | * value, the function breaks the iteration loop and returns the value. | |
1280 | * Otherwise, it will iterate over all tasks and return 0. | |
1281 | * | |
1282 | * This function must not be called for the root memory cgroup. | |
1283 | */ | |
1284 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
1285 | int (*fn)(struct task_struct *, void *), void *arg) | |
1286 | { | |
1287 | struct mem_cgroup *iter; | |
1288 | int ret = 0; | |
1289 | ||
1290 | BUG_ON(memcg == root_mem_cgroup); | |
1291 | ||
1292 | for_each_mem_cgroup_tree(iter, memcg) { | |
1293 | struct css_task_iter it; | |
1294 | struct task_struct *task; | |
1295 | ||
f168a9a5 | 1296 | css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it); |
7c5f64f8 VD |
1297 | while (!ret && (task = css_task_iter_next(&it))) |
1298 | ret = fn(task, arg); | |
1299 | css_task_iter_end(&it); | |
1300 | if (ret) { | |
1301 | mem_cgroup_iter_break(memcg, iter); | |
1302 | break; | |
1303 | } | |
1304 | } | |
1305 | return ret; | |
1306 | } | |
1307 | ||
925b7673 | 1308 | /** |
dfe0e773 | 1309 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1310 | * @page: the page |
f144c390 | 1311 | * @pgdat: pgdat of the page |
dfe0e773 | 1312 | * |
a0b5b414 JW |
1313 | * This function relies on page->mem_cgroup being stable - see the |
1314 | * access rules in commit_charge(). | |
925b7673 | 1315 | */ |
599d0c95 | 1316 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat) |
08e552c6 | 1317 | { |
ef8f2327 | 1318 | struct mem_cgroup_per_node *mz; |
925b7673 | 1319 | struct mem_cgroup *memcg; |
bea8c150 | 1320 | struct lruvec *lruvec; |
6d12e2d8 | 1321 | |
bea8c150 | 1322 | if (mem_cgroup_disabled()) { |
867e5e1d | 1323 | lruvec = &pgdat->__lruvec; |
bea8c150 HD |
1324 | goto out; |
1325 | } | |
925b7673 | 1326 | |
1306a85a | 1327 | memcg = page->mem_cgroup; |
7512102c | 1328 | /* |
dfe0e773 | 1329 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1330 | * possibly migrated - before they are charged. |
7512102c | 1331 | */ |
29833315 JW |
1332 | if (!memcg) |
1333 | memcg = root_mem_cgroup; | |
7512102c | 1334 | |
ef8f2327 | 1335 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
bea8c150 HD |
1336 | lruvec = &mz->lruvec; |
1337 | out: | |
1338 | /* | |
1339 | * Since a node can be onlined after the mem_cgroup was created, | |
1340 | * we have to be prepared to initialize lruvec->zone here; | |
1341 | * and if offlined then reonlined, we need to reinitialize it. | |
1342 | */ | |
599d0c95 MG |
1343 | if (unlikely(lruvec->pgdat != pgdat)) |
1344 | lruvec->pgdat = pgdat; | |
bea8c150 | 1345 | return lruvec; |
08e552c6 | 1346 | } |
b69408e8 | 1347 | |
925b7673 | 1348 | /** |
fa9add64 HD |
1349 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1350 | * @lruvec: mem_cgroup per zone lru vector | |
1351 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1352 | * @zid: zone id of the accounted pages |
fa9add64 | 1353 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1354 | * |
ca707239 HD |
1355 | * This function must be called under lru_lock, just before a page is added |
1356 | * to or just after a page is removed from an lru list (that ordering being | |
1357 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 1358 | */ |
fa9add64 | 1359 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1360 | int zid, int nr_pages) |
3f58a829 | 1361 | { |
ef8f2327 | 1362 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1363 | unsigned long *lru_size; |
ca707239 | 1364 | long size; |
3f58a829 MK |
1365 | |
1366 | if (mem_cgroup_disabled()) | |
1367 | return; | |
1368 | ||
ef8f2327 | 1369 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1370 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1371 | |
1372 | if (nr_pages < 0) | |
1373 | *lru_size += nr_pages; | |
1374 | ||
1375 | size = *lru_size; | |
b4536f0c MH |
1376 | if (WARN_ONCE(size < 0, |
1377 | "%s(%p, %d, %d): lru_size %ld\n", | |
1378 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1379 | VM_BUG_ON(1); |
1380 | *lru_size = 0; | |
1381 | } | |
1382 | ||
1383 | if (nr_pages > 0) | |
1384 | *lru_size += nr_pages; | |
08e552c6 | 1385 | } |
544122e5 | 1386 | |
19942822 | 1387 | /** |
9d11ea9f | 1388 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1389 | * @memcg: the memory cgroup |
19942822 | 1390 | * |
9d11ea9f | 1391 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1392 | * pages. |
19942822 | 1393 | */ |
c0ff4b85 | 1394 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1395 | { |
3e32cb2e JW |
1396 | unsigned long margin = 0; |
1397 | unsigned long count; | |
1398 | unsigned long limit; | |
9d11ea9f | 1399 | |
3e32cb2e | 1400 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1401 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1402 | if (count < limit) |
1403 | margin = limit - count; | |
1404 | ||
7941d214 | 1405 | if (do_memsw_account()) { |
3e32cb2e | 1406 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1407 | limit = READ_ONCE(memcg->memsw.max); |
1c4448ed | 1408 | if (count < limit) |
3e32cb2e | 1409 | margin = min(margin, limit - count); |
cbedbac3 LR |
1410 | else |
1411 | margin = 0; | |
3e32cb2e JW |
1412 | } |
1413 | ||
1414 | return margin; | |
19942822 JW |
1415 | } |
1416 | ||
32047e2a | 1417 | /* |
bdcbb659 | 1418 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1419 | * |
bdcbb659 QH |
1420 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1421 | * moving cgroups. This is for waiting at high-memory pressure | |
1422 | * caused by "move". | |
32047e2a | 1423 | */ |
c0ff4b85 | 1424 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1425 | { |
2bd9bb20 KH |
1426 | struct mem_cgroup *from; |
1427 | struct mem_cgroup *to; | |
4b534334 | 1428 | bool ret = false; |
2bd9bb20 KH |
1429 | /* |
1430 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1431 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1432 | */ | |
1433 | spin_lock(&mc.lock); | |
1434 | from = mc.from; | |
1435 | to = mc.to; | |
1436 | if (!from) | |
1437 | goto unlock; | |
3e92041d | 1438 | |
2314b42d JW |
1439 | ret = mem_cgroup_is_descendant(from, memcg) || |
1440 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1441 | unlock: |
1442 | spin_unlock(&mc.lock); | |
4b534334 KH |
1443 | return ret; |
1444 | } | |
1445 | ||
c0ff4b85 | 1446 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1447 | { |
1448 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1449 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1450 | DEFINE_WAIT(wait); |
1451 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1452 | /* moving charge context might have finished. */ | |
1453 | if (mc.moving_task) | |
1454 | schedule(); | |
1455 | finish_wait(&mc.waitq, &wait); | |
1456 | return true; | |
1457 | } | |
1458 | } | |
1459 | return false; | |
1460 | } | |
1461 | ||
c8713d0b JW |
1462 | static char *memory_stat_format(struct mem_cgroup *memcg) |
1463 | { | |
1464 | struct seq_buf s; | |
1465 | int i; | |
71cd3113 | 1466 | |
c8713d0b JW |
1467 | seq_buf_init(&s, kmalloc(PAGE_SIZE, GFP_KERNEL), PAGE_SIZE); |
1468 | if (!s.buffer) | |
1469 | return NULL; | |
1470 | ||
1471 | /* | |
1472 | * Provide statistics on the state of the memory subsystem as | |
1473 | * well as cumulative event counters that show past behavior. | |
1474 | * | |
1475 | * This list is ordered following a combination of these gradients: | |
1476 | * 1) generic big picture -> specifics and details | |
1477 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
1478 | * | |
1479 | * Current memory state: | |
1480 | */ | |
1481 | ||
1482 | seq_buf_printf(&s, "anon %llu\n", | |
be5d0a74 | 1483 | (u64)memcg_page_state(memcg, NR_ANON_MAPPED) * |
c8713d0b JW |
1484 | PAGE_SIZE); |
1485 | seq_buf_printf(&s, "file %llu\n", | |
0d1c2072 | 1486 | (u64)memcg_page_state(memcg, NR_FILE_PAGES) * |
c8713d0b JW |
1487 | PAGE_SIZE); |
1488 | seq_buf_printf(&s, "kernel_stack %llu\n", | |
991e7673 | 1489 | (u64)memcg_page_state(memcg, NR_KERNEL_STACK_KB) * |
c8713d0b JW |
1490 | 1024); |
1491 | seq_buf_printf(&s, "slab %llu\n", | |
d42f3245 RG |
1492 | (u64)(memcg_page_state(memcg, NR_SLAB_RECLAIMABLE_B) + |
1493 | memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE_B))); | |
c8713d0b JW |
1494 | seq_buf_printf(&s, "sock %llu\n", |
1495 | (u64)memcg_page_state(memcg, MEMCG_SOCK) * | |
1496 | PAGE_SIZE); | |
1497 | ||
1498 | seq_buf_printf(&s, "shmem %llu\n", | |
1499 | (u64)memcg_page_state(memcg, NR_SHMEM) * | |
1500 | PAGE_SIZE); | |
1501 | seq_buf_printf(&s, "file_mapped %llu\n", | |
1502 | (u64)memcg_page_state(memcg, NR_FILE_MAPPED) * | |
1503 | PAGE_SIZE); | |
1504 | seq_buf_printf(&s, "file_dirty %llu\n", | |
1505 | (u64)memcg_page_state(memcg, NR_FILE_DIRTY) * | |
1506 | PAGE_SIZE); | |
1507 | seq_buf_printf(&s, "file_writeback %llu\n", | |
1508 | (u64)memcg_page_state(memcg, NR_WRITEBACK) * | |
1509 | PAGE_SIZE); | |
1510 | ||
468c3982 | 1511 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
c8713d0b | 1512 | seq_buf_printf(&s, "anon_thp %llu\n", |
468c3982 JW |
1513 | (u64)memcg_page_state(memcg, NR_ANON_THPS) * |
1514 | HPAGE_PMD_SIZE); | |
1515 | #endif | |
c8713d0b JW |
1516 | |
1517 | for (i = 0; i < NR_LRU_LISTS; i++) | |
ebc5d83d | 1518 | seq_buf_printf(&s, "%s %llu\n", lru_list_name(i), |
c8713d0b JW |
1519 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
1520 | PAGE_SIZE); | |
1521 | ||
1522 | seq_buf_printf(&s, "slab_reclaimable %llu\n", | |
d42f3245 | 1523 | (u64)memcg_page_state(memcg, NR_SLAB_RECLAIMABLE_B)); |
c8713d0b | 1524 | seq_buf_printf(&s, "slab_unreclaimable %llu\n", |
d42f3245 | 1525 | (u64)memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE_B)); |
c8713d0b JW |
1526 | |
1527 | /* Accumulated memory events */ | |
1528 | ||
ebc5d83d KK |
1529 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGFAULT), |
1530 | memcg_events(memcg, PGFAULT)); | |
1531 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGMAJFAULT), | |
1532 | memcg_events(memcg, PGMAJFAULT)); | |
c8713d0b JW |
1533 | |
1534 | seq_buf_printf(&s, "workingset_refault %lu\n", | |
1535 | memcg_page_state(memcg, WORKINGSET_REFAULT)); | |
1536 | seq_buf_printf(&s, "workingset_activate %lu\n", | |
1537 | memcg_page_state(memcg, WORKINGSET_ACTIVATE)); | |
a6f5576b YS |
1538 | seq_buf_printf(&s, "workingset_restore %lu\n", |
1539 | memcg_page_state(memcg, WORKINGSET_RESTORE)); | |
c8713d0b JW |
1540 | seq_buf_printf(&s, "workingset_nodereclaim %lu\n", |
1541 | memcg_page_state(memcg, WORKINGSET_NODERECLAIM)); | |
1542 | ||
ebc5d83d KK |
1543 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGREFILL), |
1544 | memcg_events(memcg, PGREFILL)); | |
c8713d0b JW |
1545 | seq_buf_printf(&s, "pgscan %lu\n", |
1546 | memcg_events(memcg, PGSCAN_KSWAPD) + | |
1547 | memcg_events(memcg, PGSCAN_DIRECT)); | |
1548 | seq_buf_printf(&s, "pgsteal %lu\n", | |
1549 | memcg_events(memcg, PGSTEAL_KSWAPD) + | |
1550 | memcg_events(memcg, PGSTEAL_DIRECT)); | |
ebc5d83d KK |
1551 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGACTIVATE), |
1552 | memcg_events(memcg, PGACTIVATE)); | |
1553 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGDEACTIVATE), | |
1554 | memcg_events(memcg, PGDEACTIVATE)); | |
1555 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREE), | |
1556 | memcg_events(memcg, PGLAZYFREE)); | |
1557 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREED), | |
1558 | memcg_events(memcg, PGLAZYFREED)); | |
c8713d0b JW |
1559 | |
1560 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
ebc5d83d | 1561 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_FAULT_ALLOC), |
c8713d0b | 1562 | memcg_events(memcg, THP_FAULT_ALLOC)); |
ebc5d83d | 1563 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_COLLAPSE_ALLOC), |
c8713d0b JW |
1564 | memcg_events(memcg, THP_COLLAPSE_ALLOC)); |
1565 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ | |
1566 | ||
1567 | /* The above should easily fit into one page */ | |
1568 | WARN_ON_ONCE(seq_buf_has_overflowed(&s)); | |
1569 | ||
1570 | return s.buffer; | |
1571 | } | |
71cd3113 | 1572 | |
58cf188e | 1573 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1574 | /** |
f0c867d9 | 1575 | * mem_cgroup_print_oom_context: Print OOM information relevant to |
1576 | * memory controller. | |
e222432b BS |
1577 | * @memcg: The memory cgroup that went over limit |
1578 | * @p: Task that is going to be killed | |
1579 | * | |
1580 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1581 | * enabled | |
1582 | */ | |
f0c867d9 | 1583 | void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) |
e222432b | 1584 | { |
e222432b BS |
1585 | rcu_read_lock(); |
1586 | ||
f0c867d9 | 1587 | if (memcg) { |
1588 | pr_cont(",oom_memcg="); | |
1589 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1590 | } else | |
1591 | pr_cont(",global_oom"); | |
2415b9f5 | 1592 | if (p) { |
f0c867d9 | 1593 | pr_cont(",task_memcg="); |
2415b9f5 | 1594 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
2415b9f5 | 1595 | } |
e222432b | 1596 | rcu_read_unlock(); |
f0c867d9 | 1597 | } |
1598 | ||
1599 | /** | |
1600 | * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to | |
1601 | * memory controller. | |
1602 | * @memcg: The memory cgroup that went over limit | |
1603 | */ | |
1604 | void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) | |
1605 | { | |
c8713d0b | 1606 | char *buf; |
e222432b | 1607 | |
3e32cb2e JW |
1608 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1609 | K((u64)page_counter_read(&memcg->memory)), | |
15b42562 | 1610 | K((u64)READ_ONCE(memcg->memory.max)), memcg->memory.failcnt); |
c8713d0b JW |
1611 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
1612 | pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1613 | K((u64)page_counter_read(&memcg->swap)), | |
32d087cd | 1614 | K((u64)READ_ONCE(memcg->swap.max)), memcg->swap.failcnt); |
c8713d0b JW |
1615 | else { |
1616 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1617 | K((u64)page_counter_read(&memcg->memsw)), | |
1618 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); | |
1619 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1620 | K((u64)page_counter_read(&memcg->kmem)), | |
1621 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); | |
58cf188e | 1622 | } |
c8713d0b JW |
1623 | |
1624 | pr_info("Memory cgroup stats for "); | |
1625 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1626 | pr_cont(":"); | |
1627 | buf = memory_stat_format(memcg); | |
1628 | if (!buf) | |
1629 | return; | |
1630 | pr_info("%s", buf); | |
1631 | kfree(buf); | |
e222432b BS |
1632 | } |
1633 | ||
a63d83f4 DR |
1634 | /* |
1635 | * Return the memory (and swap, if configured) limit for a memcg. | |
1636 | */ | |
bbec2e15 | 1637 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1638 | { |
bbec2e15 | 1639 | unsigned long max; |
f3e8eb70 | 1640 | |
15b42562 | 1641 | max = READ_ONCE(memcg->memory.max); |
9a5a8f19 | 1642 | if (mem_cgroup_swappiness(memcg)) { |
bbec2e15 RG |
1643 | unsigned long memsw_max; |
1644 | unsigned long swap_max; | |
9a5a8f19 | 1645 | |
bbec2e15 | 1646 | memsw_max = memcg->memsw.max; |
32d087cd | 1647 | swap_max = READ_ONCE(memcg->swap.max); |
bbec2e15 RG |
1648 | swap_max = min(swap_max, (unsigned long)total_swap_pages); |
1649 | max = min(max + swap_max, memsw_max); | |
9a5a8f19 | 1650 | } |
bbec2e15 | 1651 | return max; |
a63d83f4 DR |
1652 | } |
1653 | ||
9783aa99 CD |
1654 | unsigned long mem_cgroup_size(struct mem_cgroup *memcg) |
1655 | { | |
1656 | return page_counter_read(&memcg->memory); | |
1657 | } | |
1658 | ||
b6e6edcf | 1659 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1660 | int order) |
9cbb78bb | 1661 | { |
6e0fc46d DR |
1662 | struct oom_control oc = { |
1663 | .zonelist = NULL, | |
1664 | .nodemask = NULL, | |
2a966b77 | 1665 | .memcg = memcg, |
6e0fc46d DR |
1666 | .gfp_mask = gfp_mask, |
1667 | .order = order, | |
6e0fc46d | 1668 | }; |
7c5f64f8 | 1669 | bool ret; |
9cbb78bb | 1670 | |
7775face TH |
1671 | if (mutex_lock_killable(&oom_lock)) |
1672 | return true; | |
1673 | /* | |
1674 | * A few threads which were not waiting at mutex_lock_killable() can | |
1675 | * fail to bail out. Therefore, check again after holding oom_lock. | |
1676 | */ | |
1677 | ret = should_force_charge() || out_of_memory(&oc); | |
dc56401f | 1678 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1679 | return ret; |
9cbb78bb DR |
1680 | } |
1681 | ||
0608f43d | 1682 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1683 | pg_data_t *pgdat, |
0608f43d AM |
1684 | gfp_t gfp_mask, |
1685 | unsigned long *total_scanned) | |
1686 | { | |
1687 | struct mem_cgroup *victim = NULL; | |
1688 | int total = 0; | |
1689 | int loop = 0; | |
1690 | unsigned long excess; | |
1691 | unsigned long nr_scanned; | |
1692 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1693 | .pgdat = pgdat, |
0608f43d AM |
1694 | }; |
1695 | ||
3e32cb2e | 1696 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1697 | |
1698 | while (1) { | |
1699 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1700 | if (!victim) { | |
1701 | loop++; | |
1702 | if (loop >= 2) { | |
1703 | /* | |
1704 | * If we have not been able to reclaim | |
1705 | * anything, it might because there are | |
1706 | * no reclaimable pages under this hierarchy | |
1707 | */ | |
1708 | if (!total) | |
1709 | break; | |
1710 | /* | |
1711 | * We want to do more targeted reclaim. | |
1712 | * excess >> 2 is not to excessive so as to | |
1713 | * reclaim too much, nor too less that we keep | |
1714 | * coming back to reclaim from this cgroup | |
1715 | */ | |
1716 | if (total >= (excess >> 2) || | |
1717 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1718 | break; | |
1719 | } | |
1720 | continue; | |
1721 | } | |
a9dd0a83 | 1722 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1723 | pgdat, &nr_scanned); |
0608f43d | 1724 | *total_scanned += nr_scanned; |
3e32cb2e | 1725 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1726 | break; |
6d61ef40 | 1727 | } |
0608f43d AM |
1728 | mem_cgroup_iter_break(root_memcg, victim); |
1729 | return total; | |
6d61ef40 BS |
1730 | } |
1731 | ||
0056f4e6 JW |
1732 | #ifdef CONFIG_LOCKDEP |
1733 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1734 | .name = "memcg_oom_lock", | |
1735 | }; | |
1736 | #endif | |
1737 | ||
fb2a6fc5 JW |
1738 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1739 | ||
867578cb KH |
1740 | /* |
1741 | * Check OOM-Killer is already running under our hierarchy. | |
1742 | * If someone is running, return false. | |
1743 | */ | |
fb2a6fc5 | 1744 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1745 | { |
79dfdacc | 1746 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1747 | |
fb2a6fc5 JW |
1748 | spin_lock(&memcg_oom_lock); |
1749 | ||
9f3a0d09 | 1750 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1751 | if (iter->oom_lock) { |
79dfdacc MH |
1752 | /* |
1753 | * this subtree of our hierarchy is already locked | |
1754 | * so we cannot give a lock. | |
1755 | */ | |
79dfdacc | 1756 | failed = iter; |
9f3a0d09 JW |
1757 | mem_cgroup_iter_break(memcg, iter); |
1758 | break; | |
23751be0 JW |
1759 | } else |
1760 | iter->oom_lock = true; | |
7d74b06f | 1761 | } |
867578cb | 1762 | |
fb2a6fc5 JW |
1763 | if (failed) { |
1764 | /* | |
1765 | * OK, we failed to lock the whole subtree so we have | |
1766 | * to clean up what we set up to the failing subtree | |
1767 | */ | |
1768 | for_each_mem_cgroup_tree(iter, memcg) { | |
1769 | if (iter == failed) { | |
1770 | mem_cgroup_iter_break(memcg, iter); | |
1771 | break; | |
1772 | } | |
1773 | iter->oom_lock = false; | |
79dfdacc | 1774 | } |
0056f4e6 JW |
1775 | } else |
1776 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1777 | |
1778 | spin_unlock(&memcg_oom_lock); | |
1779 | ||
1780 | return !failed; | |
a636b327 | 1781 | } |
0b7f569e | 1782 | |
fb2a6fc5 | 1783 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1784 | { |
7d74b06f KH |
1785 | struct mem_cgroup *iter; |
1786 | ||
fb2a6fc5 | 1787 | spin_lock(&memcg_oom_lock); |
5facae4f | 1788 | mutex_release(&memcg_oom_lock_dep_map, _RET_IP_); |
c0ff4b85 | 1789 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1790 | iter->oom_lock = false; |
fb2a6fc5 | 1791 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1792 | } |
1793 | ||
c0ff4b85 | 1794 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1795 | { |
1796 | struct mem_cgroup *iter; | |
1797 | ||
c2b42d3c | 1798 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1799 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1800 | iter->under_oom++; |
1801 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1802 | } |
1803 | ||
c0ff4b85 | 1804 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1805 | { |
1806 | struct mem_cgroup *iter; | |
1807 | ||
867578cb KH |
1808 | /* |
1809 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1810 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1811 | */ |
c2b42d3c | 1812 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1813 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1814 | if (iter->under_oom > 0) |
1815 | iter->under_oom--; | |
1816 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1817 | } |
1818 | ||
867578cb KH |
1819 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1820 | ||
dc98df5a | 1821 | struct oom_wait_info { |
d79154bb | 1822 | struct mem_cgroup *memcg; |
ac6424b9 | 1823 | wait_queue_entry_t wait; |
dc98df5a KH |
1824 | }; |
1825 | ||
ac6424b9 | 1826 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1827 | unsigned mode, int sync, void *arg) |
1828 | { | |
d79154bb HD |
1829 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1830 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1831 | struct oom_wait_info *oom_wait_info; |
1832 | ||
1833 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1834 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1835 | |
2314b42d JW |
1836 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1837 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1838 | return 0; |
dc98df5a KH |
1839 | return autoremove_wake_function(wait, mode, sync, arg); |
1840 | } | |
1841 | ||
c0ff4b85 | 1842 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1843 | { |
c2b42d3c TH |
1844 | /* |
1845 | * For the following lockless ->under_oom test, the only required | |
1846 | * guarantee is that it must see the state asserted by an OOM when | |
1847 | * this function is called as a result of userland actions | |
1848 | * triggered by the notification of the OOM. This is trivially | |
1849 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1850 | * triggering notification. | |
1851 | */ | |
1852 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1853 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1854 | } |
1855 | ||
29ef680a MH |
1856 | enum oom_status { |
1857 | OOM_SUCCESS, | |
1858 | OOM_FAILED, | |
1859 | OOM_ASYNC, | |
1860 | OOM_SKIPPED | |
1861 | }; | |
1862 | ||
1863 | static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 1864 | { |
7056d3a3 MH |
1865 | enum oom_status ret; |
1866 | bool locked; | |
1867 | ||
29ef680a MH |
1868 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
1869 | return OOM_SKIPPED; | |
1870 | ||
7a1adfdd RG |
1871 | memcg_memory_event(memcg, MEMCG_OOM); |
1872 | ||
867578cb | 1873 | /* |
49426420 JW |
1874 | * We are in the middle of the charge context here, so we |
1875 | * don't want to block when potentially sitting on a callstack | |
1876 | * that holds all kinds of filesystem and mm locks. | |
1877 | * | |
29ef680a MH |
1878 | * cgroup1 allows disabling the OOM killer and waiting for outside |
1879 | * handling until the charge can succeed; remember the context and put | |
1880 | * the task to sleep at the end of the page fault when all locks are | |
1881 | * released. | |
49426420 | 1882 | * |
29ef680a MH |
1883 | * On the other hand, in-kernel OOM killer allows for an async victim |
1884 | * memory reclaim (oom_reaper) and that means that we are not solely | |
1885 | * relying on the oom victim to make a forward progress and we can | |
1886 | * invoke the oom killer here. | |
1887 | * | |
1888 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
1889 | * victim and then we have to bail out from the charge path. | |
867578cb | 1890 | */ |
29ef680a MH |
1891 | if (memcg->oom_kill_disable) { |
1892 | if (!current->in_user_fault) | |
1893 | return OOM_SKIPPED; | |
1894 | css_get(&memcg->css); | |
1895 | current->memcg_in_oom = memcg; | |
1896 | current->memcg_oom_gfp_mask = mask; | |
1897 | current->memcg_oom_order = order; | |
1898 | ||
1899 | return OOM_ASYNC; | |
1900 | } | |
1901 | ||
7056d3a3 MH |
1902 | mem_cgroup_mark_under_oom(memcg); |
1903 | ||
1904 | locked = mem_cgroup_oom_trylock(memcg); | |
1905 | ||
1906 | if (locked) | |
1907 | mem_cgroup_oom_notify(memcg); | |
1908 | ||
1909 | mem_cgroup_unmark_under_oom(memcg); | |
29ef680a | 1910 | if (mem_cgroup_out_of_memory(memcg, mask, order)) |
7056d3a3 MH |
1911 | ret = OOM_SUCCESS; |
1912 | else | |
1913 | ret = OOM_FAILED; | |
1914 | ||
1915 | if (locked) | |
1916 | mem_cgroup_oom_unlock(memcg); | |
29ef680a | 1917 | |
7056d3a3 | 1918 | return ret; |
3812c8c8 JW |
1919 | } |
1920 | ||
1921 | /** | |
1922 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1923 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1924 | * |
49426420 JW |
1925 | * This has to be called at the end of a page fault if the memcg OOM |
1926 | * handler was enabled. | |
3812c8c8 | 1927 | * |
49426420 | 1928 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1929 | * sleep on a waitqueue until the userspace task resolves the |
1930 | * situation. Sleeping directly in the charge context with all kinds | |
1931 | * of locks held is not a good idea, instead we remember an OOM state | |
1932 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1933 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1934 | * |
1935 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1936 | * completed, %false otherwise. |
3812c8c8 | 1937 | */ |
49426420 | 1938 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1939 | { |
626ebc41 | 1940 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1941 | struct oom_wait_info owait; |
49426420 | 1942 | bool locked; |
3812c8c8 JW |
1943 | |
1944 | /* OOM is global, do not handle */ | |
3812c8c8 | 1945 | if (!memcg) |
49426420 | 1946 | return false; |
3812c8c8 | 1947 | |
7c5f64f8 | 1948 | if (!handle) |
49426420 | 1949 | goto cleanup; |
3812c8c8 JW |
1950 | |
1951 | owait.memcg = memcg; | |
1952 | owait.wait.flags = 0; | |
1953 | owait.wait.func = memcg_oom_wake_function; | |
1954 | owait.wait.private = current; | |
2055da97 | 1955 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 1956 | |
3812c8c8 | 1957 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1958 | mem_cgroup_mark_under_oom(memcg); |
1959 | ||
1960 | locked = mem_cgroup_oom_trylock(memcg); | |
1961 | ||
1962 | if (locked) | |
1963 | mem_cgroup_oom_notify(memcg); | |
1964 | ||
1965 | if (locked && !memcg->oom_kill_disable) { | |
1966 | mem_cgroup_unmark_under_oom(memcg); | |
1967 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1968 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1969 | current->memcg_oom_order); | |
49426420 | 1970 | } else { |
3812c8c8 | 1971 | schedule(); |
49426420 JW |
1972 | mem_cgroup_unmark_under_oom(memcg); |
1973 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1974 | } | |
1975 | ||
1976 | if (locked) { | |
fb2a6fc5 JW |
1977 | mem_cgroup_oom_unlock(memcg); |
1978 | /* | |
1979 | * There is no guarantee that an OOM-lock contender | |
1980 | * sees the wakeups triggered by the OOM kill | |
1981 | * uncharges. Wake any sleepers explicitely. | |
1982 | */ | |
1983 | memcg_oom_recover(memcg); | |
1984 | } | |
49426420 | 1985 | cleanup: |
626ebc41 | 1986 | current->memcg_in_oom = NULL; |
3812c8c8 | 1987 | css_put(&memcg->css); |
867578cb | 1988 | return true; |
0b7f569e KH |
1989 | } |
1990 | ||
3d8b38eb RG |
1991 | /** |
1992 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
1993 | * @victim: task to be killed by the OOM killer | |
1994 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
1995 | * | |
1996 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
1997 | * by killing all belonging OOM-killable tasks. | |
1998 | * | |
1999 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
2000 | */ | |
2001 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
2002 | struct mem_cgroup *oom_domain) | |
2003 | { | |
2004 | struct mem_cgroup *oom_group = NULL; | |
2005 | struct mem_cgroup *memcg; | |
2006 | ||
2007 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
2008 | return NULL; | |
2009 | ||
2010 | if (!oom_domain) | |
2011 | oom_domain = root_mem_cgroup; | |
2012 | ||
2013 | rcu_read_lock(); | |
2014 | ||
2015 | memcg = mem_cgroup_from_task(victim); | |
2016 | if (memcg == root_mem_cgroup) | |
2017 | goto out; | |
2018 | ||
48fe267c RG |
2019 | /* |
2020 | * If the victim task has been asynchronously moved to a different | |
2021 | * memory cgroup, we might end up killing tasks outside oom_domain. | |
2022 | * In this case it's better to ignore memory.group.oom. | |
2023 | */ | |
2024 | if (unlikely(!mem_cgroup_is_descendant(memcg, oom_domain))) | |
2025 | goto out; | |
2026 | ||
3d8b38eb RG |
2027 | /* |
2028 | * Traverse the memory cgroup hierarchy from the victim task's | |
2029 | * cgroup up to the OOMing cgroup (or root) to find the | |
2030 | * highest-level memory cgroup with oom.group set. | |
2031 | */ | |
2032 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
2033 | if (memcg->oom_group) | |
2034 | oom_group = memcg; | |
2035 | ||
2036 | if (memcg == oom_domain) | |
2037 | break; | |
2038 | } | |
2039 | ||
2040 | if (oom_group) | |
2041 | css_get(&oom_group->css); | |
2042 | out: | |
2043 | rcu_read_unlock(); | |
2044 | ||
2045 | return oom_group; | |
2046 | } | |
2047 | ||
2048 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
2049 | { | |
2050 | pr_info("Tasks in "); | |
2051 | pr_cont_cgroup_path(memcg->css.cgroup); | |
2052 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
2053 | } | |
2054 | ||
d7365e78 | 2055 | /** |
81f8c3a4 JW |
2056 | * lock_page_memcg - lock a page->mem_cgroup binding |
2057 | * @page: the page | |
32047e2a | 2058 | * |
81f8c3a4 | 2059 | * This function protects unlocked LRU pages from being moved to |
739f79fc JW |
2060 | * another cgroup. |
2061 | * | |
2062 | * It ensures lifetime of the returned memcg. Caller is responsible | |
2063 | * for the lifetime of the page; __unlock_page_memcg() is available | |
2064 | * when @page might get freed inside the locked section. | |
d69b042f | 2065 | */ |
739f79fc | 2066 | struct mem_cgroup *lock_page_memcg(struct page *page) |
89c06bd5 | 2067 | { |
9da7b521 | 2068 | struct page *head = compound_head(page); /* rmap on tail pages */ |
89c06bd5 | 2069 | struct mem_cgroup *memcg; |
6de22619 | 2070 | unsigned long flags; |
89c06bd5 | 2071 | |
6de22619 JW |
2072 | /* |
2073 | * The RCU lock is held throughout the transaction. The fast | |
2074 | * path can get away without acquiring the memcg->move_lock | |
2075 | * because page moving starts with an RCU grace period. | |
739f79fc JW |
2076 | * |
2077 | * The RCU lock also protects the memcg from being freed when | |
2078 | * the page state that is going to change is the only thing | |
2079 | * preventing the page itself from being freed. E.g. writeback | |
2080 | * doesn't hold a page reference and relies on PG_writeback to | |
2081 | * keep off truncation, migration and so forth. | |
2082 | */ | |
d7365e78 JW |
2083 | rcu_read_lock(); |
2084 | ||
2085 | if (mem_cgroup_disabled()) | |
739f79fc | 2086 | return NULL; |
89c06bd5 | 2087 | again: |
9da7b521 | 2088 | memcg = head->mem_cgroup; |
29833315 | 2089 | if (unlikely(!memcg)) |
739f79fc | 2090 | return NULL; |
d7365e78 | 2091 | |
bdcbb659 | 2092 | if (atomic_read(&memcg->moving_account) <= 0) |
739f79fc | 2093 | return memcg; |
89c06bd5 | 2094 | |
6de22619 | 2095 | spin_lock_irqsave(&memcg->move_lock, flags); |
9da7b521 | 2096 | if (memcg != head->mem_cgroup) { |
6de22619 | 2097 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
2098 | goto again; |
2099 | } | |
6de22619 JW |
2100 | |
2101 | /* | |
2102 | * When charge migration first begins, we can have locked and | |
2103 | * unlocked page stat updates happening concurrently. Track | |
81f8c3a4 | 2104 | * the task who has the lock for unlock_page_memcg(). |
6de22619 JW |
2105 | */ |
2106 | memcg->move_lock_task = current; | |
2107 | memcg->move_lock_flags = flags; | |
d7365e78 | 2108 | |
739f79fc | 2109 | return memcg; |
89c06bd5 | 2110 | } |
81f8c3a4 | 2111 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 2112 | |
d7365e78 | 2113 | /** |
739f79fc JW |
2114 | * __unlock_page_memcg - unlock and unpin a memcg |
2115 | * @memcg: the memcg | |
2116 | * | |
2117 | * Unlock and unpin a memcg returned by lock_page_memcg(). | |
d7365e78 | 2118 | */ |
739f79fc | 2119 | void __unlock_page_memcg(struct mem_cgroup *memcg) |
89c06bd5 | 2120 | { |
6de22619 JW |
2121 | if (memcg && memcg->move_lock_task == current) { |
2122 | unsigned long flags = memcg->move_lock_flags; | |
2123 | ||
2124 | memcg->move_lock_task = NULL; | |
2125 | memcg->move_lock_flags = 0; | |
2126 | ||
2127 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2128 | } | |
89c06bd5 | 2129 | |
d7365e78 | 2130 | rcu_read_unlock(); |
89c06bd5 | 2131 | } |
739f79fc JW |
2132 | |
2133 | /** | |
2134 | * unlock_page_memcg - unlock a page->mem_cgroup binding | |
2135 | * @page: the page | |
2136 | */ | |
2137 | void unlock_page_memcg(struct page *page) | |
2138 | { | |
9da7b521 JW |
2139 | struct page *head = compound_head(page); |
2140 | ||
2141 | __unlock_page_memcg(head->mem_cgroup); | |
739f79fc | 2142 | } |
81f8c3a4 | 2143 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 2144 | |
cdec2e42 KH |
2145 | struct memcg_stock_pcp { |
2146 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2147 | unsigned int nr_pages; |
bf4f0599 RG |
2148 | |
2149 | #ifdef CONFIG_MEMCG_KMEM | |
2150 | struct obj_cgroup *cached_objcg; | |
2151 | unsigned int nr_bytes; | |
2152 | #endif | |
2153 | ||
cdec2e42 | 2154 | struct work_struct work; |
26fe6168 | 2155 | unsigned long flags; |
a0db00fc | 2156 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2157 | }; |
2158 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2159 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2160 | |
bf4f0599 RG |
2161 | #ifdef CONFIG_MEMCG_KMEM |
2162 | static void drain_obj_stock(struct memcg_stock_pcp *stock); | |
2163 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2164 | struct mem_cgroup *root_memcg); | |
2165 | ||
2166 | #else | |
2167 | static inline void drain_obj_stock(struct memcg_stock_pcp *stock) | |
2168 | { | |
2169 | } | |
2170 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2171 | struct mem_cgroup *root_memcg) | |
2172 | { | |
2173 | return false; | |
2174 | } | |
2175 | #endif | |
2176 | ||
a0956d54 SS |
2177 | /** |
2178 | * consume_stock: Try to consume stocked charge on this cpu. | |
2179 | * @memcg: memcg to consume from. | |
2180 | * @nr_pages: how many pages to charge. | |
2181 | * | |
2182 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2183 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2184 | * service an allocation will refill the stock. | |
2185 | * | |
2186 | * returns true if successful, false otherwise. | |
cdec2e42 | 2187 | */ |
a0956d54 | 2188 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2189 | { |
2190 | struct memcg_stock_pcp *stock; | |
db2ba40c | 2191 | unsigned long flags; |
3e32cb2e | 2192 | bool ret = false; |
cdec2e42 | 2193 | |
a983b5eb | 2194 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 2195 | return ret; |
a0956d54 | 2196 | |
db2ba40c JW |
2197 | local_irq_save(flags); |
2198 | ||
2199 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 2200 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2201 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2202 | ret = true; |
2203 | } | |
db2ba40c JW |
2204 | |
2205 | local_irq_restore(flags); | |
2206 | ||
cdec2e42 KH |
2207 | return ret; |
2208 | } | |
2209 | ||
2210 | /* | |
3e32cb2e | 2211 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2212 | */ |
2213 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2214 | { | |
2215 | struct mem_cgroup *old = stock->cached; | |
2216 | ||
1a3e1f40 JW |
2217 | if (!old) |
2218 | return; | |
2219 | ||
11c9ea4e | 2220 | if (stock->nr_pages) { |
3e32cb2e | 2221 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 2222 | if (do_memsw_account()) |
3e32cb2e | 2223 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
11c9ea4e | 2224 | stock->nr_pages = 0; |
cdec2e42 | 2225 | } |
1a3e1f40 JW |
2226 | |
2227 | css_put(&old->css); | |
cdec2e42 | 2228 | stock->cached = NULL; |
cdec2e42 KH |
2229 | } |
2230 | ||
cdec2e42 KH |
2231 | static void drain_local_stock(struct work_struct *dummy) |
2232 | { | |
db2ba40c JW |
2233 | struct memcg_stock_pcp *stock; |
2234 | unsigned long flags; | |
2235 | ||
72f0184c MH |
2236 | /* |
2237 | * The only protection from memory hotplug vs. drain_stock races is | |
2238 | * that we always operate on local CPU stock here with IRQ disabled | |
2239 | */ | |
db2ba40c JW |
2240 | local_irq_save(flags); |
2241 | ||
2242 | stock = this_cpu_ptr(&memcg_stock); | |
bf4f0599 | 2243 | drain_obj_stock(stock); |
cdec2e42 | 2244 | drain_stock(stock); |
26fe6168 | 2245 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c JW |
2246 | |
2247 | local_irq_restore(flags); | |
cdec2e42 KH |
2248 | } |
2249 | ||
2250 | /* | |
3e32cb2e | 2251 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2252 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2253 | */ |
c0ff4b85 | 2254 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2255 | { |
db2ba40c JW |
2256 | struct memcg_stock_pcp *stock; |
2257 | unsigned long flags; | |
2258 | ||
2259 | local_irq_save(flags); | |
cdec2e42 | 2260 | |
db2ba40c | 2261 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 2262 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2263 | drain_stock(stock); |
1a3e1f40 | 2264 | css_get(&memcg->css); |
c0ff4b85 | 2265 | stock->cached = memcg; |
cdec2e42 | 2266 | } |
11c9ea4e | 2267 | stock->nr_pages += nr_pages; |
db2ba40c | 2268 | |
a983b5eb | 2269 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 RG |
2270 | drain_stock(stock); |
2271 | ||
db2ba40c | 2272 | local_irq_restore(flags); |
cdec2e42 KH |
2273 | } |
2274 | ||
2275 | /* | |
c0ff4b85 | 2276 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2277 | * of the hierarchy under it. |
cdec2e42 | 2278 | */ |
6d3d6aa2 | 2279 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2280 | { |
26fe6168 | 2281 | int cpu, curcpu; |
d38144b7 | 2282 | |
6d3d6aa2 JW |
2283 | /* If someone's already draining, avoid adding running more workers. */ |
2284 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2285 | return; | |
72f0184c MH |
2286 | /* |
2287 | * Notify other cpus that system-wide "drain" is running | |
2288 | * We do not care about races with the cpu hotplug because cpu down | |
2289 | * as well as workers from this path always operate on the local | |
2290 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2291 | */ | |
5af12d0e | 2292 | curcpu = get_cpu(); |
cdec2e42 KH |
2293 | for_each_online_cpu(cpu) { |
2294 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2295 | struct mem_cgroup *memcg; |
e1a366be | 2296 | bool flush = false; |
26fe6168 | 2297 | |
e1a366be | 2298 | rcu_read_lock(); |
c0ff4b85 | 2299 | memcg = stock->cached; |
e1a366be RG |
2300 | if (memcg && stock->nr_pages && |
2301 | mem_cgroup_is_descendant(memcg, root_memcg)) | |
2302 | flush = true; | |
bf4f0599 RG |
2303 | if (obj_stock_flush_required(stock, root_memcg)) |
2304 | flush = true; | |
e1a366be RG |
2305 | rcu_read_unlock(); |
2306 | ||
2307 | if (flush && | |
2308 | !test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { | |
d1a05b69 MH |
2309 | if (cpu == curcpu) |
2310 | drain_local_stock(&stock->work); | |
2311 | else | |
2312 | schedule_work_on(cpu, &stock->work); | |
2313 | } | |
cdec2e42 | 2314 | } |
5af12d0e | 2315 | put_cpu(); |
9f50fad6 | 2316 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2317 | } |
2318 | ||
308167fc | 2319 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
cdec2e42 | 2320 | { |
cdec2e42 | 2321 | struct memcg_stock_pcp *stock; |
42a30035 | 2322 | struct mem_cgroup *memcg, *mi; |
cdec2e42 | 2323 | |
cdec2e42 KH |
2324 | stock = &per_cpu(memcg_stock, cpu); |
2325 | drain_stock(stock); | |
a983b5eb JW |
2326 | |
2327 | for_each_mem_cgroup(memcg) { | |
2328 | int i; | |
2329 | ||
2330 | for (i = 0; i < MEMCG_NR_STAT; i++) { | |
2331 | int nid; | |
2332 | long x; | |
2333 | ||
871789d4 | 2334 | x = this_cpu_xchg(memcg->vmstats_percpu->stat[i], 0); |
815744d7 | 2335 | if (x) |
42a30035 JW |
2336 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2337 | atomic_long_add(x, &memcg->vmstats[i]); | |
a983b5eb JW |
2338 | |
2339 | if (i >= NR_VM_NODE_STAT_ITEMS) | |
2340 | continue; | |
2341 | ||
2342 | for_each_node(nid) { | |
2343 | struct mem_cgroup_per_node *pn; | |
2344 | ||
2345 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
2346 | x = this_cpu_xchg(pn->lruvec_stat_cpu->count[i], 0); | |
815744d7 | 2347 | if (x) |
42a30035 JW |
2348 | do { |
2349 | atomic_long_add(x, &pn->lruvec_stat[i]); | |
2350 | } while ((pn = parent_nodeinfo(pn, nid))); | |
a983b5eb JW |
2351 | } |
2352 | } | |
2353 | ||
e27be240 | 2354 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { |
a983b5eb JW |
2355 | long x; |
2356 | ||
871789d4 | 2357 | x = this_cpu_xchg(memcg->vmstats_percpu->events[i], 0); |
815744d7 | 2358 | if (x) |
42a30035 JW |
2359 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2360 | atomic_long_add(x, &memcg->vmevents[i]); | |
a983b5eb JW |
2361 | } |
2362 | } | |
2363 | ||
308167fc | 2364 | return 0; |
cdec2e42 KH |
2365 | } |
2366 | ||
b3ff9291 CD |
2367 | static unsigned long reclaim_high(struct mem_cgroup *memcg, |
2368 | unsigned int nr_pages, | |
2369 | gfp_t gfp_mask) | |
f7e1cb6e | 2370 | { |
b3ff9291 CD |
2371 | unsigned long nr_reclaimed = 0; |
2372 | ||
f7e1cb6e | 2373 | do { |
d1663a90 JK |
2374 | if (page_counter_read(&memcg->memory) <= |
2375 | READ_ONCE(memcg->memory.high)) | |
f7e1cb6e | 2376 | continue; |
e27be240 | 2377 | memcg_memory_event(memcg, MEMCG_HIGH); |
b3ff9291 CD |
2378 | nr_reclaimed += try_to_free_mem_cgroup_pages(memcg, nr_pages, |
2379 | gfp_mask, true); | |
4bf17307 CD |
2380 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2381 | !mem_cgroup_is_root(memcg)); | |
b3ff9291 CD |
2382 | |
2383 | return nr_reclaimed; | |
f7e1cb6e JW |
2384 | } |
2385 | ||
2386 | static void high_work_func(struct work_struct *work) | |
2387 | { | |
2388 | struct mem_cgroup *memcg; | |
2389 | ||
2390 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2391 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2392 | } |
2393 | ||
0e4b01df CD |
2394 | /* |
2395 | * Clamp the maximum sleep time per allocation batch to 2 seconds. This is | |
2396 | * enough to still cause a significant slowdown in most cases, while still | |
2397 | * allowing diagnostics and tracing to proceed without becoming stuck. | |
2398 | */ | |
2399 | #define MEMCG_MAX_HIGH_DELAY_JIFFIES (2UL*HZ) | |
2400 | ||
2401 | /* | |
2402 | * When calculating the delay, we use these either side of the exponentiation to | |
2403 | * maintain precision and scale to a reasonable number of jiffies (see the table | |
2404 | * below. | |
2405 | * | |
2406 | * - MEMCG_DELAY_PRECISION_SHIFT: Extra precision bits while translating the | |
2407 | * overage ratio to a delay. | |
2408 | * - MEMCG_DELAY_SCALING_SHIFT: The number of bits to scale down down the | |
2409 | * proposed penalty in order to reduce to a reasonable number of jiffies, and | |
2410 | * to produce a reasonable delay curve. | |
2411 | * | |
2412 | * MEMCG_DELAY_SCALING_SHIFT just happens to be a number that produces a | |
2413 | * reasonable delay curve compared to precision-adjusted overage, not | |
2414 | * penalising heavily at first, but still making sure that growth beyond the | |
2415 | * limit penalises misbehaviour cgroups by slowing them down exponentially. For | |
2416 | * example, with a high of 100 megabytes: | |
2417 | * | |
2418 | * +-------+------------------------+ | |
2419 | * | usage | time to allocate in ms | | |
2420 | * +-------+------------------------+ | |
2421 | * | 100M | 0 | | |
2422 | * | 101M | 6 | | |
2423 | * | 102M | 25 | | |
2424 | * | 103M | 57 | | |
2425 | * | 104M | 102 | | |
2426 | * | 105M | 159 | | |
2427 | * | 106M | 230 | | |
2428 | * | 107M | 313 | | |
2429 | * | 108M | 409 | | |
2430 | * | 109M | 518 | | |
2431 | * | 110M | 639 | | |
2432 | * | 111M | 774 | | |
2433 | * | 112M | 921 | | |
2434 | * | 113M | 1081 | | |
2435 | * | 114M | 1254 | | |
2436 | * | 115M | 1439 | | |
2437 | * | 116M | 1638 | | |
2438 | * | 117M | 1849 | | |
2439 | * | 118M | 2000 | | |
2440 | * | 119M | 2000 | | |
2441 | * | 120M | 2000 | | |
2442 | * +-------+------------------------+ | |
2443 | */ | |
2444 | #define MEMCG_DELAY_PRECISION_SHIFT 20 | |
2445 | #define MEMCG_DELAY_SCALING_SHIFT 14 | |
2446 | ||
8a5dbc65 | 2447 | static u64 calculate_overage(unsigned long usage, unsigned long high) |
b23afb93 | 2448 | { |
8a5dbc65 | 2449 | u64 overage; |
b23afb93 | 2450 | |
8a5dbc65 JK |
2451 | if (usage <= high) |
2452 | return 0; | |
e26733e0 | 2453 | |
8a5dbc65 JK |
2454 | /* |
2455 | * Prevent division by 0 in overage calculation by acting as if | |
2456 | * it was a threshold of 1 page | |
2457 | */ | |
2458 | high = max(high, 1UL); | |
9b8b1754 | 2459 | |
8a5dbc65 JK |
2460 | overage = usage - high; |
2461 | overage <<= MEMCG_DELAY_PRECISION_SHIFT; | |
2462 | return div64_u64(overage, high); | |
2463 | } | |
e26733e0 | 2464 | |
8a5dbc65 JK |
2465 | static u64 mem_find_max_overage(struct mem_cgroup *memcg) |
2466 | { | |
2467 | u64 overage, max_overage = 0; | |
e26733e0 | 2468 | |
8a5dbc65 JK |
2469 | do { |
2470 | overage = calculate_overage(page_counter_read(&memcg->memory), | |
d1663a90 | 2471 | READ_ONCE(memcg->memory.high)); |
8a5dbc65 | 2472 | max_overage = max(overage, max_overage); |
e26733e0 CD |
2473 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2474 | !mem_cgroup_is_root(memcg)); | |
2475 | ||
8a5dbc65 JK |
2476 | return max_overage; |
2477 | } | |
2478 | ||
4b82ab4f JK |
2479 | static u64 swap_find_max_overage(struct mem_cgroup *memcg) |
2480 | { | |
2481 | u64 overage, max_overage = 0; | |
2482 | ||
2483 | do { | |
2484 | overage = calculate_overage(page_counter_read(&memcg->swap), | |
2485 | READ_ONCE(memcg->swap.high)); | |
2486 | if (overage) | |
2487 | memcg_memory_event(memcg, MEMCG_SWAP_HIGH); | |
2488 | max_overage = max(overage, max_overage); | |
2489 | } while ((memcg = parent_mem_cgroup(memcg)) && | |
2490 | !mem_cgroup_is_root(memcg)); | |
2491 | ||
2492 | return max_overage; | |
2493 | } | |
2494 | ||
8a5dbc65 JK |
2495 | /* |
2496 | * Get the number of jiffies that we should penalise a mischievous cgroup which | |
2497 | * is exceeding its memory.high by checking both it and its ancestors. | |
2498 | */ | |
2499 | static unsigned long calculate_high_delay(struct mem_cgroup *memcg, | |
2500 | unsigned int nr_pages, | |
2501 | u64 max_overage) | |
2502 | { | |
2503 | unsigned long penalty_jiffies; | |
2504 | ||
e26733e0 CD |
2505 | if (!max_overage) |
2506 | return 0; | |
0e4b01df CD |
2507 | |
2508 | /* | |
0e4b01df CD |
2509 | * We use overage compared to memory.high to calculate the number of |
2510 | * jiffies to sleep (penalty_jiffies). Ideally this value should be | |
2511 | * fairly lenient on small overages, and increasingly harsh when the | |
2512 | * memcg in question makes it clear that it has no intention of stopping | |
2513 | * its crazy behaviour, so we exponentially increase the delay based on | |
2514 | * overage amount. | |
2515 | */ | |
e26733e0 CD |
2516 | penalty_jiffies = max_overage * max_overage * HZ; |
2517 | penalty_jiffies >>= MEMCG_DELAY_PRECISION_SHIFT; | |
2518 | penalty_jiffies >>= MEMCG_DELAY_SCALING_SHIFT; | |
0e4b01df CD |
2519 | |
2520 | /* | |
2521 | * Factor in the task's own contribution to the overage, such that four | |
2522 | * N-sized allocations are throttled approximately the same as one | |
2523 | * 4N-sized allocation. | |
2524 | * | |
2525 | * MEMCG_CHARGE_BATCH pages is nominal, so work out how much smaller or | |
2526 | * larger the current charge patch is than that. | |
2527 | */ | |
ff144e69 | 2528 | return penalty_jiffies * nr_pages / MEMCG_CHARGE_BATCH; |
e26733e0 CD |
2529 | } |
2530 | ||
2531 | /* | |
2532 | * Scheduled by try_charge() to be executed from the userland return path | |
2533 | * and reclaims memory over the high limit. | |
2534 | */ | |
2535 | void mem_cgroup_handle_over_high(void) | |
2536 | { | |
2537 | unsigned long penalty_jiffies; | |
2538 | unsigned long pflags; | |
b3ff9291 | 2539 | unsigned long nr_reclaimed; |
e26733e0 | 2540 | unsigned int nr_pages = current->memcg_nr_pages_over_high; |
b3ff9291 | 2541 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
e26733e0 | 2542 | struct mem_cgroup *memcg; |
b3ff9291 | 2543 | bool in_retry = false; |
e26733e0 CD |
2544 | |
2545 | if (likely(!nr_pages)) | |
2546 | return; | |
2547 | ||
2548 | memcg = get_mem_cgroup_from_mm(current->mm); | |
e26733e0 CD |
2549 | current->memcg_nr_pages_over_high = 0; |
2550 | ||
b3ff9291 CD |
2551 | retry_reclaim: |
2552 | /* | |
2553 | * The allocating task should reclaim at least the batch size, but for | |
2554 | * subsequent retries we only want to do what's necessary to prevent oom | |
2555 | * or breaching resource isolation. | |
2556 | * | |
2557 | * This is distinct from memory.max or page allocator behaviour because | |
2558 | * memory.high is currently batched, whereas memory.max and the page | |
2559 | * allocator run every time an allocation is made. | |
2560 | */ | |
2561 | nr_reclaimed = reclaim_high(memcg, | |
2562 | in_retry ? SWAP_CLUSTER_MAX : nr_pages, | |
2563 | GFP_KERNEL); | |
2564 | ||
e26733e0 CD |
2565 | /* |
2566 | * memory.high is breached and reclaim is unable to keep up. Throttle | |
2567 | * allocators proactively to slow down excessive growth. | |
2568 | */ | |
8a5dbc65 JK |
2569 | penalty_jiffies = calculate_high_delay(memcg, nr_pages, |
2570 | mem_find_max_overage(memcg)); | |
0e4b01df | 2571 | |
4b82ab4f JK |
2572 | penalty_jiffies += calculate_high_delay(memcg, nr_pages, |
2573 | swap_find_max_overage(memcg)); | |
2574 | ||
ff144e69 JK |
2575 | /* |
2576 | * Clamp the max delay per usermode return so as to still keep the | |
2577 | * application moving forwards and also permit diagnostics, albeit | |
2578 | * extremely slowly. | |
2579 | */ | |
2580 | penalty_jiffies = min(penalty_jiffies, MEMCG_MAX_HIGH_DELAY_JIFFIES); | |
2581 | ||
0e4b01df CD |
2582 | /* |
2583 | * Don't sleep if the amount of jiffies this memcg owes us is so low | |
2584 | * that it's not even worth doing, in an attempt to be nice to those who | |
2585 | * go only a small amount over their memory.high value and maybe haven't | |
2586 | * been aggressively reclaimed enough yet. | |
2587 | */ | |
2588 | if (penalty_jiffies <= HZ / 100) | |
2589 | goto out; | |
2590 | ||
b3ff9291 CD |
2591 | /* |
2592 | * If reclaim is making forward progress but we're still over | |
2593 | * memory.high, we want to encourage that rather than doing allocator | |
2594 | * throttling. | |
2595 | */ | |
2596 | if (nr_reclaimed || nr_retries--) { | |
2597 | in_retry = true; | |
2598 | goto retry_reclaim; | |
2599 | } | |
2600 | ||
0e4b01df CD |
2601 | /* |
2602 | * If we exit early, we're guaranteed to die (since | |
2603 | * schedule_timeout_killable sets TASK_KILLABLE). This means we don't | |
2604 | * need to account for any ill-begotten jiffies to pay them off later. | |
2605 | */ | |
2606 | psi_memstall_enter(&pflags); | |
2607 | schedule_timeout_killable(penalty_jiffies); | |
2608 | psi_memstall_leave(&pflags); | |
2609 | ||
2610 | out: | |
2611 | css_put(&memcg->css); | |
b23afb93 TH |
2612 | } |
2613 | ||
00501b53 JW |
2614 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2615 | unsigned int nr_pages) | |
8a9f3ccd | 2616 | { |
a983b5eb | 2617 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
9b130619 | 2618 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 2619 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2620 | struct page_counter *counter; |
6539cc05 | 2621 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2622 | bool may_swap = true; |
2623 | bool drained = false; | |
29ef680a | 2624 | enum oom_status oom_status; |
a636b327 | 2625 | |
ce00a967 | 2626 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 2627 | return 0; |
6539cc05 | 2628 | retry: |
b6b6cc72 | 2629 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2630 | return 0; |
8a9f3ccd | 2631 | |
7941d214 | 2632 | if (!do_memsw_account() || |
6071ca52 JW |
2633 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2634 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2635 | goto done_restock; |
7941d214 | 2636 | if (do_memsw_account()) |
3e32cb2e JW |
2637 | page_counter_uncharge(&memcg->memsw, batch); |
2638 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2639 | } else { |
3e32cb2e | 2640 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2641 | may_swap = false; |
3fbe7244 | 2642 | } |
7a81b88c | 2643 | |
6539cc05 JW |
2644 | if (batch > nr_pages) { |
2645 | batch = nr_pages; | |
2646 | goto retry; | |
2647 | } | |
6d61ef40 | 2648 | |
869712fd JW |
2649 | /* |
2650 | * Memcg doesn't have a dedicated reserve for atomic | |
2651 | * allocations. But like the global atomic pool, we need to | |
2652 | * put the burden of reclaim on regular allocation requests | |
2653 | * and let these go through as privileged allocations. | |
2654 | */ | |
2655 | if (gfp_mask & __GFP_ATOMIC) | |
2656 | goto force; | |
2657 | ||
06b078fc JW |
2658 | /* |
2659 | * Unlike in global OOM situations, memcg is not in a physical | |
2660 | * memory shortage. Allow dying and OOM-killed tasks to | |
2661 | * bypass the last charges so that they can exit quickly and | |
2662 | * free their memory. | |
2663 | */ | |
7775face | 2664 | if (unlikely(should_force_charge())) |
10d53c74 | 2665 | goto force; |
06b078fc | 2666 | |
89a28483 JW |
2667 | /* |
2668 | * Prevent unbounded recursion when reclaim operations need to | |
2669 | * allocate memory. This might exceed the limits temporarily, | |
2670 | * but we prefer facilitating memory reclaim and getting back | |
2671 | * under the limit over triggering OOM kills in these cases. | |
2672 | */ | |
2673 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2674 | goto force; | |
2675 | ||
06b078fc JW |
2676 | if (unlikely(task_in_memcg_oom(current))) |
2677 | goto nomem; | |
2678 | ||
d0164adc | 2679 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2680 | goto nomem; |
4b534334 | 2681 | |
e27be240 | 2682 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
241994ed | 2683 | |
b70a2a21 JW |
2684 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2685 | gfp_mask, may_swap); | |
6539cc05 | 2686 | |
61e02c74 | 2687 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2688 | goto retry; |
28c34c29 | 2689 | |
b70a2a21 | 2690 | if (!drained) { |
6d3d6aa2 | 2691 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2692 | drained = true; |
2693 | goto retry; | |
2694 | } | |
2695 | ||
28c34c29 JW |
2696 | if (gfp_mask & __GFP_NORETRY) |
2697 | goto nomem; | |
6539cc05 JW |
2698 | /* |
2699 | * Even though the limit is exceeded at this point, reclaim | |
2700 | * may have been able to free some pages. Retry the charge | |
2701 | * before killing the task. | |
2702 | * | |
2703 | * Only for regular pages, though: huge pages are rather | |
2704 | * unlikely to succeed so close to the limit, and we fall back | |
2705 | * to regular pages anyway in case of failure. | |
2706 | */ | |
61e02c74 | 2707 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2708 | goto retry; |
2709 | /* | |
2710 | * At task move, charge accounts can be doubly counted. So, it's | |
2711 | * better to wait until the end of task_move if something is going on. | |
2712 | */ | |
2713 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2714 | goto retry; | |
2715 | ||
9b130619 JW |
2716 | if (nr_retries--) |
2717 | goto retry; | |
2718 | ||
38d38493 | 2719 | if (gfp_mask & __GFP_RETRY_MAYFAIL) |
29ef680a MH |
2720 | goto nomem; |
2721 | ||
06b078fc | 2722 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2723 | goto force; |
06b078fc | 2724 | |
6539cc05 | 2725 | if (fatal_signal_pending(current)) |
10d53c74 | 2726 | goto force; |
6539cc05 | 2727 | |
29ef680a MH |
2728 | /* |
2729 | * keep retrying as long as the memcg oom killer is able to make | |
2730 | * a forward progress or bypass the charge if the oom killer | |
2731 | * couldn't make any progress. | |
2732 | */ | |
2733 | oom_status = mem_cgroup_oom(mem_over_limit, gfp_mask, | |
3608de07 | 2734 | get_order(nr_pages * PAGE_SIZE)); |
29ef680a MH |
2735 | switch (oom_status) { |
2736 | case OOM_SUCCESS: | |
2737 | nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
29ef680a MH |
2738 | goto retry; |
2739 | case OOM_FAILED: | |
2740 | goto force; | |
2741 | default: | |
2742 | goto nomem; | |
2743 | } | |
7a81b88c | 2744 | nomem: |
6d1fdc48 | 2745 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2746 | return -ENOMEM; |
10d53c74 TH |
2747 | force: |
2748 | /* | |
2749 | * The allocation either can't fail or will lead to more memory | |
2750 | * being freed very soon. Allow memory usage go over the limit | |
2751 | * temporarily by force charging it. | |
2752 | */ | |
2753 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2754 | if (do_memsw_account()) |
10d53c74 | 2755 | page_counter_charge(&memcg->memsw, nr_pages); |
10d53c74 TH |
2756 | |
2757 | return 0; | |
6539cc05 JW |
2758 | |
2759 | done_restock: | |
2760 | if (batch > nr_pages) | |
2761 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2762 | |
241994ed | 2763 | /* |
b23afb93 TH |
2764 | * If the hierarchy is above the normal consumption range, schedule |
2765 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2766 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2767 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2768 | * not recorded as it most likely matches current's and won't | |
2769 | * change in the meantime. As high limit is checked again before | |
2770 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2771 | */ |
2772 | do { | |
4b82ab4f JK |
2773 | bool mem_high, swap_high; |
2774 | ||
2775 | mem_high = page_counter_read(&memcg->memory) > | |
2776 | READ_ONCE(memcg->memory.high); | |
2777 | swap_high = page_counter_read(&memcg->swap) > | |
2778 | READ_ONCE(memcg->swap.high); | |
2779 | ||
2780 | /* Don't bother a random interrupted task */ | |
2781 | if (in_interrupt()) { | |
2782 | if (mem_high) { | |
f7e1cb6e JW |
2783 | schedule_work(&memcg->high_work); |
2784 | break; | |
2785 | } | |
4b82ab4f JK |
2786 | continue; |
2787 | } | |
2788 | ||
2789 | if (mem_high || swap_high) { | |
2790 | /* | |
2791 | * The allocating tasks in this cgroup will need to do | |
2792 | * reclaim or be throttled to prevent further growth | |
2793 | * of the memory or swap footprints. | |
2794 | * | |
2795 | * Target some best-effort fairness between the tasks, | |
2796 | * and distribute reclaim work and delay penalties | |
2797 | * based on how much each task is actually allocating. | |
2798 | */ | |
9516a18a | 2799 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2800 | set_notify_resume(current); |
2801 | break; | |
2802 | } | |
241994ed | 2803 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2804 | |
2805 | return 0; | |
7a81b88c | 2806 | } |
8a9f3ccd | 2807 | |
f0e45fb4 | 2808 | #if defined(CONFIG_MEMCG_KMEM) || defined(CONFIG_MMU) |
00501b53 | 2809 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2810 | { |
ce00a967 JW |
2811 | if (mem_cgroup_is_root(memcg)) |
2812 | return; | |
2813 | ||
3e32cb2e | 2814 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2815 | if (do_memsw_account()) |
3e32cb2e | 2816 | page_counter_uncharge(&memcg->memsw, nr_pages); |
d01dd17f | 2817 | } |
f0e45fb4 | 2818 | #endif |
d01dd17f | 2819 | |
d9eb1ea2 | 2820 | static void commit_charge(struct page *page, struct mem_cgroup *memcg) |
0a31bc97 | 2821 | { |
1306a85a | 2822 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
0a31bc97 | 2823 | /* |
a0b5b414 | 2824 | * Any of the following ensures page->mem_cgroup stability: |
0a31bc97 | 2825 | * |
a0b5b414 JW |
2826 | * - the page lock |
2827 | * - LRU isolation | |
2828 | * - lock_page_memcg() | |
2829 | * - exclusive reference | |
0a31bc97 | 2830 | */ |
1306a85a | 2831 | page->mem_cgroup = memcg; |
7a81b88c | 2832 | } |
66e1707b | 2833 | |
84c07d11 | 2834 | #ifdef CONFIG_MEMCG_KMEM |
10befea9 RG |
2835 | int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s, |
2836 | gfp_t gfp) | |
2837 | { | |
2838 | unsigned int objects = objs_per_slab_page(s, page); | |
2839 | void *vec; | |
2840 | ||
2841 | vec = kcalloc_node(objects, sizeof(struct obj_cgroup *), gfp, | |
2842 | page_to_nid(page)); | |
2843 | if (!vec) | |
2844 | return -ENOMEM; | |
2845 | ||
2846 | if (cmpxchg(&page->obj_cgroups, NULL, | |
2847 | (struct obj_cgroup **) ((unsigned long)vec | 0x1UL))) | |
2848 | kfree(vec); | |
2849 | else | |
2850 | kmemleak_not_leak(vec); | |
2851 | ||
2852 | return 0; | |
2853 | } | |
2854 | ||
8380ce47 RG |
2855 | /* |
2856 | * Returns a pointer to the memory cgroup to which the kernel object is charged. | |
2857 | * | |
2858 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), | |
2859 | * cgroup_mutex, etc. | |
2860 | */ | |
2861 | struct mem_cgroup *mem_cgroup_from_obj(void *p) | |
2862 | { | |
2863 | struct page *page; | |
2864 | ||
2865 | if (mem_cgroup_disabled()) | |
2866 | return NULL; | |
2867 | ||
2868 | page = virt_to_head_page(p); | |
2869 | ||
2870 | /* | |
9855609b RG |
2871 | * Slab objects are accounted individually, not per-page. |
2872 | * Memcg membership data for each individual object is saved in | |
2873 | * the page->obj_cgroups. | |
8380ce47 | 2874 | */ |
9855609b RG |
2875 | if (page_has_obj_cgroups(page)) { |
2876 | struct obj_cgroup *objcg; | |
2877 | unsigned int off; | |
2878 | ||
2879 | off = obj_to_index(page->slab_cache, page, p); | |
2880 | objcg = page_obj_cgroups(page)[off]; | |
10befea9 RG |
2881 | if (objcg) |
2882 | return obj_cgroup_memcg(objcg); | |
2883 | ||
2884 | return NULL; | |
9855609b | 2885 | } |
8380ce47 RG |
2886 | |
2887 | /* All other pages use page->mem_cgroup */ | |
2888 | return page->mem_cgroup; | |
2889 | } | |
2890 | ||
bf4f0599 RG |
2891 | __always_inline struct obj_cgroup *get_obj_cgroup_from_current(void) |
2892 | { | |
2893 | struct obj_cgroup *objcg = NULL; | |
2894 | struct mem_cgroup *memcg; | |
2895 | ||
2896 | if (unlikely(!current->mm && !current->active_memcg)) | |
2897 | return NULL; | |
2898 | ||
2899 | rcu_read_lock(); | |
2900 | if (unlikely(current->active_memcg)) | |
2901 | memcg = rcu_dereference(current->active_memcg); | |
2902 | else | |
2903 | memcg = mem_cgroup_from_task(current); | |
2904 | ||
2905 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { | |
2906 | objcg = rcu_dereference(memcg->objcg); | |
2907 | if (objcg && obj_cgroup_tryget(objcg)) | |
2908 | break; | |
2909 | } | |
2910 | rcu_read_unlock(); | |
2911 | ||
2912 | return objcg; | |
2913 | } | |
2914 | ||
f3bb3043 | 2915 | static int memcg_alloc_cache_id(void) |
55007d84 | 2916 | { |
f3bb3043 VD |
2917 | int id, size; |
2918 | int err; | |
2919 | ||
dbcf73e2 | 2920 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2921 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2922 | if (id < 0) | |
2923 | return id; | |
55007d84 | 2924 | |
dbcf73e2 | 2925 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2926 | return id; |
2927 | ||
2928 | /* | |
2929 | * There's no space for the new id in memcg_caches arrays, | |
2930 | * so we have to grow them. | |
2931 | */ | |
05257a1a | 2932 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2933 | |
2934 | size = 2 * (id + 1); | |
55007d84 GC |
2935 | if (size < MEMCG_CACHES_MIN_SIZE) |
2936 | size = MEMCG_CACHES_MIN_SIZE; | |
2937 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2938 | size = MEMCG_CACHES_MAX_SIZE; | |
2939 | ||
9855609b | 2940 | err = memcg_update_all_list_lrus(size); |
05257a1a VD |
2941 | if (!err) |
2942 | memcg_nr_cache_ids = size; | |
2943 | ||
2944 | up_write(&memcg_cache_ids_sem); | |
2945 | ||
f3bb3043 | 2946 | if (err) { |
dbcf73e2 | 2947 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2948 | return err; |
2949 | } | |
2950 | return id; | |
2951 | } | |
2952 | ||
2953 | static void memcg_free_cache_id(int id) | |
2954 | { | |
dbcf73e2 | 2955 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2956 | } |
2957 | ||
45264778 | 2958 | /** |
4b13f64d | 2959 | * __memcg_kmem_charge: charge a number of kernel pages to a memcg |
10eaec2f | 2960 | * @memcg: memory cgroup to charge |
45264778 | 2961 | * @gfp: reclaim mode |
92d0510c | 2962 | * @nr_pages: number of pages to charge |
45264778 VD |
2963 | * |
2964 | * Returns 0 on success, an error code on failure. | |
2965 | */ | |
4b13f64d RG |
2966 | int __memcg_kmem_charge(struct mem_cgroup *memcg, gfp_t gfp, |
2967 | unsigned int nr_pages) | |
7ae1e1d0 | 2968 | { |
f3ccb2c4 | 2969 | struct page_counter *counter; |
7ae1e1d0 GC |
2970 | int ret; |
2971 | ||
f3ccb2c4 | 2972 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 2973 | if (ret) |
f3ccb2c4 | 2974 | return ret; |
52c29b04 JW |
2975 | |
2976 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
2977 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
e55d9d9b MH |
2978 | |
2979 | /* | |
2980 | * Enforce __GFP_NOFAIL allocation because callers are not | |
2981 | * prepared to see failures and likely do not have any failure | |
2982 | * handling code. | |
2983 | */ | |
2984 | if (gfp & __GFP_NOFAIL) { | |
2985 | page_counter_charge(&memcg->kmem, nr_pages); | |
2986 | return 0; | |
2987 | } | |
52c29b04 JW |
2988 | cancel_charge(memcg, nr_pages); |
2989 | return -ENOMEM; | |
7ae1e1d0 | 2990 | } |
f3ccb2c4 | 2991 | return 0; |
7ae1e1d0 GC |
2992 | } |
2993 | ||
4b13f64d RG |
2994 | /** |
2995 | * __memcg_kmem_uncharge: uncharge a number of kernel pages from a memcg | |
2996 | * @memcg: memcg to uncharge | |
2997 | * @nr_pages: number of pages to uncharge | |
2998 | */ | |
2999 | void __memcg_kmem_uncharge(struct mem_cgroup *memcg, unsigned int nr_pages) | |
3000 | { | |
3001 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
3002 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
3003 | ||
3004 | page_counter_uncharge(&memcg->memory, nr_pages); | |
3005 | if (do_memsw_account()) | |
3006 | page_counter_uncharge(&memcg->memsw, nr_pages); | |
3007 | } | |
3008 | ||
45264778 | 3009 | /** |
f4b00eab | 3010 | * __memcg_kmem_charge_page: charge a kmem page to the current memory cgroup |
45264778 VD |
3011 | * @page: page to charge |
3012 | * @gfp: reclaim mode | |
3013 | * @order: allocation order | |
3014 | * | |
3015 | * Returns 0 on success, an error code on failure. | |
3016 | */ | |
f4b00eab | 3017 | int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 3018 | { |
f3ccb2c4 | 3019 | struct mem_cgroup *memcg; |
fcff7d7e | 3020 | int ret = 0; |
7ae1e1d0 | 3021 | |
60cd4bcd | 3022 | if (memcg_kmem_bypass()) |
45264778 VD |
3023 | return 0; |
3024 | ||
d46eb14b | 3025 | memcg = get_mem_cgroup_from_current(); |
c4159a75 | 3026 | if (!mem_cgroup_is_root(memcg)) { |
4b13f64d | 3027 | ret = __memcg_kmem_charge(memcg, gfp, 1 << order); |
4d96ba35 RG |
3028 | if (!ret) { |
3029 | page->mem_cgroup = memcg; | |
c4159a75 | 3030 | __SetPageKmemcg(page); |
1a3e1f40 | 3031 | return 0; |
4d96ba35 | 3032 | } |
c4159a75 | 3033 | } |
7ae1e1d0 | 3034 | css_put(&memcg->css); |
d05e83a6 | 3035 | return ret; |
7ae1e1d0 | 3036 | } |
49a18eae | 3037 | |
45264778 | 3038 | /** |
f4b00eab | 3039 | * __memcg_kmem_uncharge_page: uncharge a kmem page |
45264778 VD |
3040 | * @page: page to uncharge |
3041 | * @order: allocation order | |
3042 | */ | |
f4b00eab | 3043 | void __memcg_kmem_uncharge_page(struct page *page, int order) |
7ae1e1d0 | 3044 | { |
1306a85a | 3045 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 3046 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 3047 | |
7ae1e1d0 GC |
3048 | if (!memcg) |
3049 | return; | |
3050 | ||
309381fe | 3051 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
4b13f64d | 3052 | __memcg_kmem_uncharge(memcg, nr_pages); |
1306a85a | 3053 | page->mem_cgroup = NULL; |
1a3e1f40 | 3054 | css_put(&memcg->css); |
c4159a75 VD |
3055 | |
3056 | /* slab pages do not have PageKmemcg flag set */ | |
3057 | if (PageKmemcg(page)) | |
3058 | __ClearPageKmemcg(page); | |
60d3fd32 | 3059 | } |
bf4f0599 RG |
3060 | |
3061 | static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) | |
3062 | { | |
3063 | struct memcg_stock_pcp *stock; | |
3064 | unsigned long flags; | |
3065 | bool ret = false; | |
3066 | ||
3067 | local_irq_save(flags); | |
3068 | ||
3069 | stock = this_cpu_ptr(&memcg_stock); | |
3070 | if (objcg == stock->cached_objcg && stock->nr_bytes >= nr_bytes) { | |
3071 | stock->nr_bytes -= nr_bytes; | |
3072 | ret = true; | |
3073 | } | |
3074 | ||
3075 | local_irq_restore(flags); | |
3076 | ||
3077 | return ret; | |
3078 | } | |
3079 | ||
3080 | static void drain_obj_stock(struct memcg_stock_pcp *stock) | |
3081 | { | |
3082 | struct obj_cgroup *old = stock->cached_objcg; | |
3083 | ||
3084 | if (!old) | |
3085 | return; | |
3086 | ||
3087 | if (stock->nr_bytes) { | |
3088 | unsigned int nr_pages = stock->nr_bytes >> PAGE_SHIFT; | |
3089 | unsigned int nr_bytes = stock->nr_bytes & (PAGE_SIZE - 1); | |
3090 | ||
3091 | if (nr_pages) { | |
3092 | rcu_read_lock(); | |
3093 | __memcg_kmem_uncharge(obj_cgroup_memcg(old), nr_pages); | |
3094 | rcu_read_unlock(); | |
3095 | } | |
3096 | ||
3097 | /* | |
3098 | * The leftover is flushed to the centralized per-memcg value. | |
3099 | * On the next attempt to refill obj stock it will be moved | |
3100 | * to a per-cpu stock (probably, on an other CPU), see | |
3101 | * refill_obj_stock(). | |
3102 | * | |
3103 | * How often it's flushed is a trade-off between the memory | |
3104 | * limit enforcement accuracy and potential CPU contention, | |
3105 | * so it might be changed in the future. | |
3106 | */ | |
3107 | atomic_add(nr_bytes, &old->nr_charged_bytes); | |
3108 | stock->nr_bytes = 0; | |
3109 | } | |
3110 | ||
3111 | obj_cgroup_put(old); | |
3112 | stock->cached_objcg = NULL; | |
3113 | } | |
3114 | ||
3115 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
3116 | struct mem_cgroup *root_memcg) | |
3117 | { | |
3118 | struct mem_cgroup *memcg; | |
3119 | ||
3120 | if (stock->cached_objcg) { | |
3121 | memcg = obj_cgroup_memcg(stock->cached_objcg); | |
3122 | if (memcg && mem_cgroup_is_descendant(memcg, root_memcg)) | |
3123 | return true; | |
3124 | } | |
3125 | ||
3126 | return false; | |
3127 | } | |
3128 | ||
3129 | static void refill_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) | |
3130 | { | |
3131 | struct memcg_stock_pcp *stock; | |
3132 | unsigned long flags; | |
3133 | ||
3134 | local_irq_save(flags); | |
3135 | ||
3136 | stock = this_cpu_ptr(&memcg_stock); | |
3137 | if (stock->cached_objcg != objcg) { /* reset if necessary */ | |
3138 | drain_obj_stock(stock); | |
3139 | obj_cgroup_get(objcg); | |
3140 | stock->cached_objcg = objcg; | |
3141 | stock->nr_bytes = atomic_xchg(&objcg->nr_charged_bytes, 0); | |
3142 | } | |
3143 | stock->nr_bytes += nr_bytes; | |
3144 | ||
3145 | if (stock->nr_bytes > PAGE_SIZE) | |
3146 | drain_obj_stock(stock); | |
3147 | ||
3148 | local_irq_restore(flags); | |
3149 | } | |
3150 | ||
3151 | int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size) | |
3152 | { | |
3153 | struct mem_cgroup *memcg; | |
3154 | unsigned int nr_pages, nr_bytes; | |
3155 | int ret; | |
3156 | ||
3157 | if (consume_obj_stock(objcg, size)) | |
3158 | return 0; | |
3159 | ||
3160 | /* | |
3161 | * In theory, memcg->nr_charged_bytes can have enough | |
3162 | * pre-charged bytes to satisfy the allocation. However, | |
3163 | * flushing memcg->nr_charged_bytes requires two atomic | |
3164 | * operations, and memcg->nr_charged_bytes can't be big, | |
3165 | * so it's better to ignore it and try grab some new pages. | |
3166 | * memcg->nr_charged_bytes will be flushed in | |
3167 | * refill_obj_stock(), called from this function or | |
3168 | * independently later. | |
3169 | */ | |
3170 | rcu_read_lock(); | |
3171 | memcg = obj_cgroup_memcg(objcg); | |
3172 | css_get(&memcg->css); | |
3173 | rcu_read_unlock(); | |
3174 | ||
3175 | nr_pages = size >> PAGE_SHIFT; | |
3176 | nr_bytes = size & (PAGE_SIZE - 1); | |
3177 | ||
3178 | if (nr_bytes) | |
3179 | nr_pages += 1; | |
3180 | ||
3181 | ret = __memcg_kmem_charge(memcg, gfp, nr_pages); | |
3182 | if (!ret && nr_bytes) | |
3183 | refill_obj_stock(objcg, PAGE_SIZE - nr_bytes); | |
3184 | ||
3185 | css_put(&memcg->css); | |
3186 | return ret; | |
3187 | } | |
3188 | ||
3189 | void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size) | |
3190 | { | |
3191 | refill_obj_stock(objcg, size); | |
3192 | } | |
3193 | ||
84c07d11 | 3194 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 3195 | |
ca3e0214 KH |
3196 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3197 | ||
ca3e0214 KH |
3198 | /* |
3199 | * Because tail pages are not marked as "used", set it. We're under | |
f4b7e272 | 3200 | * pgdat->lru_lock and migration entries setup in all page mappings. |
ca3e0214 | 3201 | */ |
e94c8a9c | 3202 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 3203 | { |
1a3e1f40 | 3204 | struct mem_cgroup *memcg = head->mem_cgroup; |
e94c8a9c | 3205 | int i; |
ca3e0214 | 3206 | |
3d37c4a9 KH |
3207 | if (mem_cgroup_disabled()) |
3208 | return; | |
b070e65c | 3209 | |
1a3e1f40 JW |
3210 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
3211 | css_get(&memcg->css); | |
3212 | head[i].mem_cgroup = memcg; | |
3213 | } | |
ca3e0214 | 3214 | } |
12d27107 | 3215 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3216 | |
c255a458 | 3217 | #ifdef CONFIG_MEMCG_SWAP |
02491447 DN |
3218 | /** |
3219 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3220 | * @entry: swap entry to be moved | |
3221 | * @from: mem_cgroup which the entry is moved from | |
3222 | * @to: mem_cgroup which the entry is moved to | |
3223 | * | |
3224 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3225 | * as the mem_cgroup's id of @from. | |
3226 | * | |
3227 | * Returns 0 on success, -EINVAL on failure. | |
3228 | * | |
3e32cb2e | 3229 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
3230 | * both res and memsw, and called css_get(). |
3231 | */ | |
3232 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3233 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3234 | { |
3235 | unsigned short old_id, new_id; | |
3236 | ||
34c00c31 LZ |
3237 | old_id = mem_cgroup_id(from); |
3238 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3239 | |
3240 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
3241 | mod_memcg_state(from, MEMCG_SWAP, -1); |
3242 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
3243 | return 0; |
3244 | } | |
3245 | return -EINVAL; | |
3246 | } | |
3247 | #else | |
3248 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3249 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3250 | { |
3251 | return -EINVAL; | |
3252 | } | |
8c7c6e34 | 3253 | #endif |
d13d1443 | 3254 | |
bbec2e15 | 3255 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 3256 | |
bbec2e15 RG |
3257 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
3258 | unsigned long max, bool memsw) | |
628f4235 | 3259 | { |
3e32cb2e | 3260 | bool enlarge = false; |
bb4a7ea2 | 3261 | bool drained = false; |
3e32cb2e | 3262 | int ret; |
c054a78c YZ |
3263 | bool limits_invariant; |
3264 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 3265 | |
3e32cb2e | 3266 | do { |
628f4235 KH |
3267 | if (signal_pending(current)) { |
3268 | ret = -EINTR; | |
3269 | break; | |
3270 | } | |
3e32cb2e | 3271 | |
bbec2e15 | 3272 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
3273 | /* |
3274 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 3275 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 3276 | */ |
15b42562 | 3277 | limits_invariant = memsw ? max >= READ_ONCE(memcg->memory.max) : |
bbec2e15 | 3278 | max <= memcg->memsw.max; |
c054a78c | 3279 | if (!limits_invariant) { |
bbec2e15 | 3280 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 3281 | ret = -EINVAL; |
8c7c6e34 KH |
3282 | break; |
3283 | } | |
bbec2e15 | 3284 | if (max > counter->max) |
3e32cb2e | 3285 | enlarge = true; |
bbec2e15 RG |
3286 | ret = page_counter_set_max(counter, max); |
3287 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
3288 | |
3289 | if (!ret) | |
3290 | break; | |
3291 | ||
bb4a7ea2 SB |
3292 | if (!drained) { |
3293 | drain_all_stock(memcg); | |
3294 | drained = true; | |
3295 | continue; | |
3296 | } | |
3297 | ||
1ab5c056 AR |
3298 | if (!try_to_free_mem_cgroup_pages(memcg, 1, |
3299 | GFP_KERNEL, !memsw)) { | |
3300 | ret = -EBUSY; | |
3301 | break; | |
3302 | } | |
3303 | } while (true); | |
3e32cb2e | 3304 | |
3c11ecf4 KH |
3305 | if (!ret && enlarge) |
3306 | memcg_oom_recover(memcg); | |
3e32cb2e | 3307 | |
628f4235 KH |
3308 | return ret; |
3309 | } | |
3310 | ||
ef8f2327 | 3311 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
3312 | gfp_t gfp_mask, |
3313 | unsigned long *total_scanned) | |
3314 | { | |
3315 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 3316 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
3317 | unsigned long reclaimed; |
3318 | int loop = 0; | |
ef8f2327 | 3319 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 3320 | unsigned long excess; |
0608f43d AM |
3321 | unsigned long nr_scanned; |
3322 | ||
3323 | if (order > 0) | |
3324 | return 0; | |
3325 | ||
ef8f2327 | 3326 | mctz = soft_limit_tree_node(pgdat->node_id); |
d6507ff5 MH |
3327 | |
3328 | /* | |
3329 | * Do not even bother to check the largest node if the root | |
3330 | * is empty. Do it lockless to prevent lock bouncing. Races | |
3331 | * are acceptable as soft limit is best effort anyway. | |
3332 | */ | |
bfc7228b | 3333 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
3334 | return 0; |
3335 | ||
0608f43d AM |
3336 | /* |
3337 | * This loop can run a while, specially if mem_cgroup's continuously | |
3338 | * keep exceeding their soft limit and putting the system under | |
3339 | * pressure | |
3340 | */ | |
3341 | do { | |
3342 | if (next_mz) | |
3343 | mz = next_mz; | |
3344 | else | |
3345 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3346 | if (!mz) | |
3347 | break; | |
3348 | ||
3349 | nr_scanned = 0; | |
ef8f2327 | 3350 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
0608f43d AM |
3351 | gfp_mask, &nr_scanned); |
3352 | nr_reclaimed += reclaimed; | |
3353 | *total_scanned += nr_scanned; | |
0a31bc97 | 3354 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 3355 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
3356 | |
3357 | /* | |
3358 | * If we failed to reclaim anything from this memory cgroup | |
3359 | * it is time to move on to the next cgroup | |
3360 | */ | |
3361 | next_mz = NULL; | |
bc2f2e7f VD |
3362 | if (!reclaimed) |
3363 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3364 | ||
3e32cb2e | 3365 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3366 | /* |
3367 | * One school of thought says that we should not add | |
3368 | * back the node to the tree if reclaim returns 0. | |
3369 | * But our reclaim could return 0, simply because due | |
3370 | * to priority we are exposing a smaller subset of | |
3371 | * memory to reclaim from. Consider this as a longer | |
3372 | * term TODO. | |
3373 | */ | |
3374 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3375 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3376 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3377 | css_put(&mz->memcg->css); |
3378 | loop++; | |
3379 | /* | |
3380 | * Could not reclaim anything and there are no more | |
3381 | * mem cgroups to try or we seem to be looping without | |
3382 | * reclaiming anything. | |
3383 | */ | |
3384 | if (!nr_reclaimed && | |
3385 | (next_mz == NULL || | |
3386 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3387 | break; | |
3388 | } while (!nr_reclaimed); | |
3389 | if (next_mz) | |
3390 | css_put(&next_mz->memcg->css); | |
3391 | return nr_reclaimed; | |
3392 | } | |
3393 | ||
ea280e7b TH |
3394 | /* |
3395 | * Test whether @memcg has children, dead or alive. Note that this | |
3396 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
3397 | * returns %true if there are child csses according to the cgroup | |
b8f2935f | 3398 | * hierarchy. Testing use_hierarchy is the caller's responsibility. |
ea280e7b | 3399 | */ |
b5f99b53 GC |
3400 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
3401 | { | |
ea280e7b TH |
3402 | bool ret; |
3403 | ||
ea280e7b TH |
3404 | rcu_read_lock(); |
3405 | ret = css_next_child(NULL, &memcg->css); | |
3406 | rcu_read_unlock(); | |
3407 | return ret; | |
b5f99b53 GC |
3408 | } |
3409 | ||
c26251f9 | 3410 | /* |
51038171 | 3411 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
3412 | * |
3413 | * Caller is responsible for holding css reference for memcg. | |
3414 | */ | |
3415 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3416 | { | |
3417 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 3418 | |
c1e862c1 KH |
3419 | /* we call try-to-free pages for make this cgroup empty */ |
3420 | lru_add_drain_all(); | |
d12c60f6 JS |
3421 | |
3422 | drain_all_stock(memcg); | |
3423 | ||
f817ed48 | 3424 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3425 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 3426 | int progress; |
c1e862c1 | 3427 | |
c26251f9 MH |
3428 | if (signal_pending(current)) |
3429 | return -EINTR; | |
3430 | ||
b70a2a21 JW |
3431 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3432 | GFP_KERNEL, true); | |
c1e862c1 | 3433 | if (!progress) { |
f817ed48 | 3434 | nr_retries--; |
c1e862c1 | 3435 | /* maybe some writeback is necessary */ |
8aa7e847 | 3436 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3437 | } |
f817ed48 KH |
3438 | |
3439 | } | |
ab5196c2 MH |
3440 | |
3441 | return 0; | |
cc847582 KH |
3442 | } |
3443 | ||
6770c64e TH |
3444 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3445 | char *buf, size_t nbytes, | |
3446 | loff_t off) | |
c1e862c1 | 3447 | { |
6770c64e | 3448 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3449 | |
d8423011 MH |
3450 | if (mem_cgroup_is_root(memcg)) |
3451 | return -EINVAL; | |
6770c64e | 3452 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3453 | } |
3454 | ||
182446d0 TH |
3455 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3456 | struct cftype *cft) | |
18f59ea7 | 3457 | { |
182446d0 | 3458 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
3459 | } |
3460 | ||
182446d0 TH |
3461 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3462 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
3463 | { |
3464 | int retval = 0; | |
182446d0 | 3465 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 3466 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 3467 | |
567fb435 | 3468 | if (memcg->use_hierarchy == val) |
0b8f73e1 | 3469 | return 0; |
567fb435 | 3470 | |
18f59ea7 | 3471 | /* |
af901ca1 | 3472 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3473 | * in the child subtrees. If it is unset, then the change can |
3474 | * occur, provided the current cgroup has no children. | |
3475 | * | |
3476 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3477 | * set if there are no children. | |
3478 | */ | |
c0ff4b85 | 3479 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 3480 | (val == 1 || val == 0)) { |
ea280e7b | 3481 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 3482 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3483 | else |
3484 | retval = -EBUSY; | |
3485 | } else | |
3486 | retval = -EINVAL; | |
567fb435 | 3487 | |
18f59ea7 BS |
3488 | return retval; |
3489 | } | |
3490 | ||
6f646156 | 3491 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 3492 | { |
42a30035 | 3493 | unsigned long val; |
ce00a967 | 3494 | |
3e32cb2e | 3495 | if (mem_cgroup_is_root(memcg)) { |
0d1c2072 | 3496 | val = memcg_page_state(memcg, NR_FILE_PAGES) + |
be5d0a74 | 3497 | memcg_page_state(memcg, NR_ANON_MAPPED); |
42a30035 JW |
3498 | if (swap) |
3499 | val += memcg_page_state(memcg, MEMCG_SWAP); | |
3e32cb2e | 3500 | } else { |
ce00a967 | 3501 | if (!swap) |
3e32cb2e | 3502 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3503 | else |
3e32cb2e | 3504 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3505 | } |
c12176d3 | 3506 | return val; |
ce00a967 JW |
3507 | } |
3508 | ||
3e32cb2e JW |
3509 | enum { |
3510 | RES_USAGE, | |
3511 | RES_LIMIT, | |
3512 | RES_MAX_USAGE, | |
3513 | RES_FAILCNT, | |
3514 | RES_SOFT_LIMIT, | |
3515 | }; | |
ce00a967 | 3516 | |
791badbd | 3517 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3518 | struct cftype *cft) |
8cdea7c0 | 3519 | { |
182446d0 | 3520 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3521 | struct page_counter *counter; |
af36f906 | 3522 | |
3e32cb2e | 3523 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3524 | case _MEM: |
3e32cb2e JW |
3525 | counter = &memcg->memory; |
3526 | break; | |
8c7c6e34 | 3527 | case _MEMSWAP: |
3e32cb2e JW |
3528 | counter = &memcg->memsw; |
3529 | break; | |
510fc4e1 | 3530 | case _KMEM: |
3e32cb2e | 3531 | counter = &memcg->kmem; |
510fc4e1 | 3532 | break; |
d55f90bf | 3533 | case _TCP: |
0db15298 | 3534 | counter = &memcg->tcpmem; |
d55f90bf | 3535 | break; |
8c7c6e34 KH |
3536 | default: |
3537 | BUG(); | |
8c7c6e34 | 3538 | } |
3e32cb2e JW |
3539 | |
3540 | switch (MEMFILE_ATTR(cft->private)) { | |
3541 | case RES_USAGE: | |
3542 | if (counter == &memcg->memory) | |
c12176d3 | 3543 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 3544 | if (counter == &memcg->memsw) |
c12176d3 | 3545 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
3546 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
3547 | case RES_LIMIT: | |
bbec2e15 | 3548 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
3549 | case RES_MAX_USAGE: |
3550 | return (u64)counter->watermark * PAGE_SIZE; | |
3551 | case RES_FAILCNT: | |
3552 | return counter->failcnt; | |
3553 | case RES_SOFT_LIMIT: | |
3554 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3555 | default: | |
3556 | BUG(); | |
3557 | } | |
8cdea7c0 | 3558 | } |
510fc4e1 | 3559 | |
4a87e2a2 | 3560 | static void memcg_flush_percpu_vmstats(struct mem_cgroup *memcg) |
c350a99e | 3561 | { |
4a87e2a2 | 3562 | unsigned long stat[MEMCG_NR_STAT] = {0}; |
c350a99e RG |
3563 | struct mem_cgroup *mi; |
3564 | int node, cpu, i; | |
c350a99e RG |
3565 | |
3566 | for_each_online_cpu(cpu) | |
4a87e2a2 | 3567 | for (i = 0; i < MEMCG_NR_STAT; i++) |
6c1c2808 | 3568 | stat[i] += per_cpu(memcg->vmstats_percpu->stat[i], cpu); |
c350a99e RG |
3569 | |
3570 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
4a87e2a2 | 3571 | for (i = 0; i < MEMCG_NR_STAT; i++) |
c350a99e RG |
3572 | atomic_long_add(stat[i], &mi->vmstats[i]); |
3573 | ||
3574 | for_each_node(node) { | |
3575 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; | |
3576 | struct mem_cgroup_per_node *pi; | |
3577 | ||
4a87e2a2 | 3578 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
c350a99e RG |
3579 | stat[i] = 0; |
3580 | ||
3581 | for_each_online_cpu(cpu) | |
4a87e2a2 | 3582 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
6c1c2808 SB |
3583 | stat[i] += per_cpu( |
3584 | pn->lruvec_stat_cpu->count[i], cpu); | |
c350a99e RG |
3585 | |
3586 | for (pi = pn; pi; pi = parent_nodeinfo(pi, node)) | |
4a87e2a2 | 3587 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
c350a99e RG |
3588 | atomic_long_add(stat[i], &pi->lruvec_stat[i]); |
3589 | } | |
3590 | } | |
3591 | ||
bb65f89b RG |
3592 | static void memcg_flush_percpu_vmevents(struct mem_cgroup *memcg) |
3593 | { | |
3594 | unsigned long events[NR_VM_EVENT_ITEMS]; | |
3595 | struct mem_cgroup *mi; | |
3596 | int cpu, i; | |
3597 | ||
3598 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
3599 | events[i] = 0; | |
3600 | ||
3601 | for_each_online_cpu(cpu) | |
3602 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
6c1c2808 SB |
3603 | events[i] += per_cpu(memcg->vmstats_percpu->events[i], |
3604 | cpu); | |
bb65f89b RG |
3605 | |
3606 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
3607 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
3608 | atomic_long_add(events[i], &mi->vmevents[i]); | |
3609 | } | |
3610 | ||
84c07d11 | 3611 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 3612 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 3613 | { |
bf4f0599 | 3614 | struct obj_cgroup *objcg; |
d6441637 VD |
3615 | int memcg_id; |
3616 | ||
b313aeee VD |
3617 | if (cgroup_memory_nokmem) |
3618 | return 0; | |
3619 | ||
2a4db7eb | 3620 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 3621 | BUG_ON(memcg->kmem_state); |
d6441637 | 3622 | |
f3bb3043 | 3623 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
3624 | if (memcg_id < 0) |
3625 | return memcg_id; | |
d6441637 | 3626 | |
bf4f0599 RG |
3627 | objcg = obj_cgroup_alloc(); |
3628 | if (!objcg) { | |
3629 | memcg_free_cache_id(memcg_id); | |
3630 | return -ENOMEM; | |
3631 | } | |
3632 | objcg->memcg = memcg; | |
3633 | rcu_assign_pointer(memcg->objcg, objcg); | |
3634 | ||
d648bcc7 RG |
3635 | static_branch_enable(&memcg_kmem_enabled_key); |
3636 | ||
d6441637 | 3637 | /* |
567e9ab2 | 3638 | * A memory cgroup is considered kmem-online as soon as it gets |
900a38f0 | 3639 | * kmemcg_id. Setting the id after enabling static branching will |
d6441637 VD |
3640 | * guarantee no one starts accounting before all call sites are |
3641 | * patched. | |
3642 | */ | |
900a38f0 | 3643 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 3644 | memcg->kmem_state = KMEM_ONLINE; |
0b8f73e1 JW |
3645 | |
3646 | return 0; | |
d6441637 VD |
3647 | } |
3648 | ||
8e0a8912 JW |
3649 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
3650 | { | |
3651 | struct cgroup_subsys_state *css; | |
3652 | struct mem_cgroup *parent, *child; | |
3653 | int kmemcg_id; | |
3654 | ||
3655 | if (memcg->kmem_state != KMEM_ONLINE) | |
3656 | return; | |
9855609b | 3657 | |
8e0a8912 JW |
3658 | memcg->kmem_state = KMEM_ALLOCATED; |
3659 | ||
8e0a8912 JW |
3660 | parent = parent_mem_cgroup(memcg); |
3661 | if (!parent) | |
3662 | parent = root_mem_cgroup; | |
3663 | ||
bf4f0599 | 3664 | memcg_reparent_objcgs(memcg, parent); |
fb2f2b0a RG |
3665 | |
3666 | kmemcg_id = memcg->kmemcg_id; | |
3667 | BUG_ON(kmemcg_id < 0); | |
3668 | ||
8e0a8912 JW |
3669 | /* |
3670 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3671 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3672 | * to ones of the parent. After we have finished, all list_lrus | |
3673 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3674 | * ordering is imposed by list_lru_node->lock taken by | |
3675 | * memcg_drain_all_list_lrus(). | |
3676 | */ | |
3a06bb78 | 3677 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
8e0a8912 JW |
3678 | css_for_each_descendant_pre(css, &memcg->css) { |
3679 | child = mem_cgroup_from_css(css); | |
3680 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3681 | child->kmemcg_id = parent->kmemcg_id; | |
3682 | if (!memcg->use_hierarchy) | |
3683 | break; | |
3684 | } | |
3a06bb78 TH |
3685 | rcu_read_unlock(); |
3686 | ||
9bec5c35 | 3687 | memcg_drain_all_list_lrus(kmemcg_id, parent); |
8e0a8912 JW |
3688 | |
3689 | memcg_free_cache_id(kmemcg_id); | |
3690 | } | |
3691 | ||
3692 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3693 | { | |
0b8f73e1 JW |
3694 | /* css_alloc() failed, offlining didn't happen */ |
3695 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
3696 | memcg_offline_kmem(memcg); | |
8e0a8912 | 3697 | } |
d6441637 | 3698 | #else |
0b8f73e1 | 3699 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
3700 | { |
3701 | return 0; | |
3702 | } | |
3703 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
3704 | { | |
3705 | } | |
3706 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3707 | { | |
3708 | } | |
84c07d11 | 3709 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 3710 | |
bbec2e15 RG |
3711 | static int memcg_update_kmem_max(struct mem_cgroup *memcg, |
3712 | unsigned long max) | |
d6441637 | 3713 | { |
b313aeee | 3714 | int ret; |
127424c8 | 3715 | |
bbec2e15 RG |
3716 | mutex_lock(&memcg_max_mutex); |
3717 | ret = page_counter_set_max(&memcg->kmem, max); | |
3718 | mutex_unlock(&memcg_max_mutex); | |
127424c8 | 3719 | return ret; |
d6441637 | 3720 | } |
510fc4e1 | 3721 | |
bbec2e15 | 3722 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
3723 | { |
3724 | int ret; | |
3725 | ||
bbec2e15 | 3726 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 3727 | |
bbec2e15 | 3728 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
3729 | if (ret) |
3730 | goto out; | |
3731 | ||
0db15298 | 3732 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
3733 | /* |
3734 | * The active flag needs to be written after the static_key | |
3735 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
3736 | * function is the last one to run. See mem_cgroup_sk_alloc() |
3737 | * for details, and note that we don't mark any socket as | |
3738 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
3739 | * |
3740 | * We need to do this, because static_keys will span multiple | |
3741 | * sites, but we can't control their order. If we mark a socket | |
3742 | * as accounted, but the accounting functions are not patched in | |
3743 | * yet, we'll lose accounting. | |
3744 | * | |
2d758073 | 3745 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
3746 | * because when this value change, the code to process it is not |
3747 | * patched in yet. | |
3748 | */ | |
3749 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 3750 | memcg->tcpmem_active = true; |
d55f90bf VD |
3751 | } |
3752 | out: | |
bbec2e15 | 3753 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
3754 | return ret; |
3755 | } | |
d55f90bf | 3756 | |
628f4235 KH |
3757 | /* |
3758 | * The user of this function is... | |
3759 | * RES_LIMIT. | |
3760 | */ | |
451af504 TH |
3761 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3762 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3763 | { |
451af504 | 3764 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3765 | unsigned long nr_pages; |
628f4235 KH |
3766 | int ret; |
3767 | ||
451af504 | 3768 | buf = strstrip(buf); |
650c5e56 | 3769 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3770 | if (ret) |
3771 | return ret; | |
af36f906 | 3772 | |
3e32cb2e | 3773 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3774 | case RES_LIMIT: |
4b3bde4c BS |
3775 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3776 | ret = -EINVAL; | |
3777 | break; | |
3778 | } | |
3e32cb2e JW |
3779 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3780 | case _MEM: | |
bbec2e15 | 3781 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 3782 | break; |
3e32cb2e | 3783 | case _MEMSWAP: |
bbec2e15 | 3784 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 3785 | break; |
3e32cb2e | 3786 | case _KMEM: |
0158115f MH |
3787 | pr_warn_once("kmem.limit_in_bytes is deprecated and will be removed. " |
3788 | "Please report your usecase to linux-mm@kvack.org if you " | |
3789 | "depend on this functionality.\n"); | |
bbec2e15 | 3790 | ret = memcg_update_kmem_max(memcg, nr_pages); |
3e32cb2e | 3791 | break; |
d55f90bf | 3792 | case _TCP: |
bbec2e15 | 3793 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 3794 | break; |
3e32cb2e | 3795 | } |
296c81d8 | 3796 | break; |
3e32cb2e JW |
3797 | case RES_SOFT_LIMIT: |
3798 | memcg->soft_limit = nr_pages; | |
3799 | ret = 0; | |
628f4235 KH |
3800 | break; |
3801 | } | |
451af504 | 3802 | return ret ?: nbytes; |
8cdea7c0 BS |
3803 | } |
3804 | ||
6770c64e TH |
3805 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3806 | size_t nbytes, loff_t off) | |
c84872e1 | 3807 | { |
6770c64e | 3808 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3809 | struct page_counter *counter; |
c84872e1 | 3810 | |
3e32cb2e JW |
3811 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3812 | case _MEM: | |
3813 | counter = &memcg->memory; | |
3814 | break; | |
3815 | case _MEMSWAP: | |
3816 | counter = &memcg->memsw; | |
3817 | break; | |
3818 | case _KMEM: | |
3819 | counter = &memcg->kmem; | |
3820 | break; | |
d55f90bf | 3821 | case _TCP: |
0db15298 | 3822 | counter = &memcg->tcpmem; |
d55f90bf | 3823 | break; |
3e32cb2e JW |
3824 | default: |
3825 | BUG(); | |
3826 | } | |
af36f906 | 3827 | |
3e32cb2e | 3828 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3829 | case RES_MAX_USAGE: |
3e32cb2e | 3830 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3831 | break; |
3832 | case RES_FAILCNT: | |
3e32cb2e | 3833 | counter->failcnt = 0; |
29f2a4da | 3834 | break; |
3e32cb2e JW |
3835 | default: |
3836 | BUG(); | |
29f2a4da | 3837 | } |
f64c3f54 | 3838 | |
6770c64e | 3839 | return nbytes; |
c84872e1 PE |
3840 | } |
3841 | ||
182446d0 | 3842 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3843 | struct cftype *cft) |
3844 | { | |
182446d0 | 3845 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3846 | } |
3847 | ||
02491447 | 3848 | #ifdef CONFIG_MMU |
182446d0 | 3849 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3850 | struct cftype *cft, u64 val) |
3851 | { | |
182446d0 | 3852 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3853 | |
1dfab5ab | 3854 | if (val & ~MOVE_MASK) |
7dc74be0 | 3855 | return -EINVAL; |
ee5e8472 | 3856 | |
7dc74be0 | 3857 | /* |
ee5e8472 GC |
3858 | * No kind of locking is needed in here, because ->can_attach() will |
3859 | * check this value once in the beginning of the process, and then carry | |
3860 | * on with stale data. This means that changes to this value will only | |
3861 | * affect task migrations starting after the change. | |
7dc74be0 | 3862 | */ |
c0ff4b85 | 3863 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3864 | return 0; |
3865 | } | |
02491447 | 3866 | #else |
182446d0 | 3867 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3868 | struct cftype *cft, u64 val) |
3869 | { | |
3870 | return -ENOSYS; | |
3871 | } | |
3872 | #endif | |
7dc74be0 | 3873 | |
406eb0c9 | 3874 | #ifdef CONFIG_NUMA |
113b7dfd JW |
3875 | |
3876 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
3877 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
3878 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) | |
3879 | ||
3880 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 | 3881 | int nid, unsigned int lru_mask, bool tree) |
113b7dfd | 3882 | { |
867e5e1d | 3883 | struct lruvec *lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); |
113b7dfd JW |
3884 | unsigned long nr = 0; |
3885 | enum lru_list lru; | |
3886 | ||
3887 | VM_BUG_ON((unsigned)nid >= nr_node_ids); | |
3888 | ||
3889 | for_each_lru(lru) { | |
3890 | if (!(BIT(lru) & lru_mask)) | |
3891 | continue; | |
dd8657b6 SB |
3892 | if (tree) |
3893 | nr += lruvec_page_state(lruvec, NR_LRU_BASE + lru); | |
3894 | else | |
3895 | nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru); | |
113b7dfd JW |
3896 | } |
3897 | return nr; | |
3898 | } | |
3899 | ||
3900 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 SB |
3901 | unsigned int lru_mask, |
3902 | bool tree) | |
113b7dfd JW |
3903 | { |
3904 | unsigned long nr = 0; | |
3905 | enum lru_list lru; | |
3906 | ||
3907 | for_each_lru(lru) { | |
3908 | if (!(BIT(lru) & lru_mask)) | |
3909 | continue; | |
dd8657b6 SB |
3910 | if (tree) |
3911 | nr += memcg_page_state(memcg, NR_LRU_BASE + lru); | |
3912 | else | |
3913 | nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru); | |
113b7dfd JW |
3914 | } |
3915 | return nr; | |
3916 | } | |
3917 | ||
2da8ca82 | 3918 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3919 | { |
25485de6 GT |
3920 | struct numa_stat { |
3921 | const char *name; | |
3922 | unsigned int lru_mask; | |
3923 | }; | |
3924 | ||
3925 | static const struct numa_stat stats[] = { | |
3926 | { "total", LRU_ALL }, | |
3927 | { "file", LRU_ALL_FILE }, | |
3928 | { "anon", LRU_ALL_ANON }, | |
3929 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3930 | }; | |
3931 | const struct numa_stat *stat; | |
406eb0c9 | 3932 | int nid; |
aa9694bb | 3933 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
406eb0c9 | 3934 | |
25485de6 | 3935 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
3936 | seq_printf(m, "%s=%lu", stat->name, |
3937 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
3938 | false)); | |
3939 | for_each_node_state(nid, N_MEMORY) | |
3940 | seq_printf(m, " N%d=%lu", nid, | |
3941 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3942 | stat->lru_mask, false)); | |
25485de6 | 3943 | seq_putc(m, '\n'); |
406eb0c9 | 3944 | } |
406eb0c9 | 3945 | |
071aee13 | 3946 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
3947 | |
3948 | seq_printf(m, "hierarchical_%s=%lu", stat->name, | |
3949 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
3950 | true)); | |
3951 | for_each_node_state(nid, N_MEMORY) | |
3952 | seq_printf(m, " N%d=%lu", nid, | |
3953 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3954 | stat->lru_mask, true)); | |
071aee13 | 3955 | seq_putc(m, '\n'); |
406eb0c9 | 3956 | } |
406eb0c9 | 3957 | |
406eb0c9 YH |
3958 | return 0; |
3959 | } | |
3960 | #endif /* CONFIG_NUMA */ | |
3961 | ||
c8713d0b | 3962 | static const unsigned int memcg1_stats[] = { |
0d1c2072 | 3963 | NR_FILE_PAGES, |
be5d0a74 | 3964 | NR_ANON_MAPPED, |
468c3982 JW |
3965 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3966 | NR_ANON_THPS, | |
3967 | #endif | |
c8713d0b JW |
3968 | NR_SHMEM, |
3969 | NR_FILE_MAPPED, | |
3970 | NR_FILE_DIRTY, | |
3971 | NR_WRITEBACK, | |
3972 | MEMCG_SWAP, | |
3973 | }; | |
3974 | ||
3975 | static const char *const memcg1_stat_names[] = { | |
3976 | "cache", | |
3977 | "rss", | |
468c3982 | 3978 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
c8713d0b | 3979 | "rss_huge", |
468c3982 | 3980 | #endif |
c8713d0b JW |
3981 | "shmem", |
3982 | "mapped_file", | |
3983 | "dirty", | |
3984 | "writeback", | |
3985 | "swap", | |
3986 | }; | |
3987 | ||
df0e53d0 | 3988 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 3989 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
3990 | PGPGIN, |
3991 | PGPGOUT, | |
3992 | PGFAULT, | |
3993 | PGMAJFAULT, | |
3994 | }; | |
3995 | ||
2da8ca82 | 3996 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3997 | { |
aa9694bb | 3998 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3e32cb2e | 3999 | unsigned long memory, memsw; |
af7c4b0e JW |
4000 | struct mem_cgroup *mi; |
4001 | unsigned int i; | |
406eb0c9 | 4002 | |
71cd3113 | 4003 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c | 4004 | |
71cd3113 | 4005 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
468c3982 JW |
4006 | unsigned long nr; |
4007 | ||
71cd3113 | 4008 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 4009 | continue; |
468c3982 JW |
4010 | nr = memcg_page_state_local(memcg, memcg1_stats[i]); |
4011 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
4012 | if (memcg1_stats[i] == NR_ANON_THPS) | |
4013 | nr *= HPAGE_PMD_NR; | |
4014 | #endif | |
4015 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], nr * PAGE_SIZE); | |
1dd3a273 | 4016 | } |
7b854121 | 4017 | |
df0e53d0 | 4018 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d | 4019 | seq_printf(m, "%s %lu\n", vm_event_name(memcg1_events[i]), |
205b20cc | 4020 | memcg_events_local(memcg, memcg1_events[i])); |
af7c4b0e JW |
4021 | |
4022 | for (i = 0; i < NR_LRU_LISTS; i++) | |
ebc5d83d | 4023 | seq_printf(m, "%s %lu\n", lru_list_name(i), |
205b20cc | 4024 | memcg_page_state_local(memcg, NR_LRU_BASE + i) * |
21d89d15 | 4025 | PAGE_SIZE); |
af7c4b0e | 4026 | |
14067bb3 | 4027 | /* Hierarchical information */ |
3e32cb2e JW |
4028 | memory = memsw = PAGE_COUNTER_MAX; |
4029 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
15b42562 CD |
4030 | memory = min(memory, READ_ONCE(mi->memory.max)); |
4031 | memsw = min(memsw, READ_ONCE(mi->memsw.max)); | |
fee7b548 | 4032 | } |
3e32cb2e JW |
4033 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
4034 | (u64)memory * PAGE_SIZE); | |
7941d214 | 4035 | if (do_memsw_account()) |
3e32cb2e JW |
4036 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
4037 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 4038 | |
8de7ecc6 | 4039 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
71cd3113 | 4040 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 4041 | continue; |
8de7ecc6 | 4042 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], |
dd923990 YS |
4043 | (u64)memcg_page_state(memcg, memcg1_stats[i]) * |
4044 | PAGE_SIZE); | |
af7c4b0e JW |
4045 | } |
4046 | ||
8de7ecc6 | 4047 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d KK |
4048 | seq_printf(m, "total_%s %llu\n", |
4049 | vm_event_name(memcg1_events[i]), | |
dd923990 | 4050 | (u64)memcg_events(memcg, memcg1_events[i])); |
af7c4b0e | 4051 | |
8de7ecc6 | 4052 | for (i = 0; i < NR_LRU_LISTS; i++) |
ebc5d83d | 4053 | seq_printf(m, "total_%s %llu\n", lru_list_name(i), |
42a30035 JW |
4054 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
4055 | PAGE_SIZE); | |
14067bb3 | 4056 | |
7f016ee8 | 4057 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 4058 | { |
ef8f2327 MG |
4059 | pg_data_t *pgdat; |
4060 | struct mem_cgroup_per_node *mz; | |
1431d4d1 JW |
4061 | unsigned long anon_cost = 0; |
4062 | unsigned long file_cost = 0; | |
7f016ee8 | 4063 | |
ef8f2327 MG |
4064 | for_each_online_pgdat(pgdat) { |
4065 | mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); | |
7f016ee8 | 4066 | |
1431d4d1 JW |
4067 | anon_cost += mz->lruvec.anon_cost; |
4068 | file_cost += mz->lruvec.file_cost; | |
ef8f2327 | 4069 | } |
1431d4d1 JW |
4070 | seq_printf(m, "anon_cost %lu\n", anon_cost); |
4071 | seq_printf(m, "file_cost %lu\n", file_cost); | |
7f016ee8 KM |
4072 | } |
4073 | #endif | |
4074 | ||
d2ceb9b7 KH |
4075 | return 0; |
4076 | } | |
4077 | ||
182446d0 TH |
4078 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
4079 | struct cftype *cft) | |
a7885eb8 | 4080 | { |
182446d0 | 4081 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4082 | |
1f4c025b | 4083 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4084 | } |
4085 | ||
182446d0 TH |
4086 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
4087 | struct cftype *cft, u64 val) | |
a7885eb8 | 4088 | { |
182446d0 | 4089 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4090 | |
3dae7fec | 4091 | if (val > 100) |
a7885eb8 KM |
4092 | return -EINVAL; |
4093 | ||
14208b0e | 4094 | if (css->parent) |
3dae7fec JW |
4095 | memcg->swappiness = val; |
4096 | else | |
4097 | vm_swappiness = val; | |
068b38c1 | 4098 | |
a7885eb8 KM |
4099 | return 0; |
4100 | } | |
4101 | ||
2e72b634 KS |
4102 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4103 | { | |
4104 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 4105 | unsigned long usage; |
2e72b634 KS |
4106 | int i; |
4107 | ||
4108 | rcu_read_lock(); | |
4109 | if (!swap) | |
2c488db2 | 4110 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4111 | else |
2c488db2 | 4112 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4113 | |
4114 | if (!t) | |
4115 | goto unlock; | |
4116 | ||
ce00a967 | 4117 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
4118 | |
4119 | /* | |
748dad36 | 4120 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4121 | * If it's not true, a threshold was crossed after last |
4122 | * call of __mem_cgroup_threshold(). | |
4123 | */ | |
5407a562 | 4124 | i = t->current_threshold; |
2e72b634 KS |
4125 | |
4126 | /* | |
4127 | * Iterate backward over array of thresholds starting from | |
4128 | * current_threshold and check if a threshold is crossed. | |
4129 | * If none of thresholds below usage is crossed, we read | |
4130 | * only one element of the array here. | |
4131 | */ | |
4132 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4133 | eventfd_signal(t->entries[i].eventfd, 1); | |
4134 | ||
4135 | /* i = current_threshold + 1 */ | |
4136 | i++; | |
4137 | ||
4138 | /* | |
4139 | * Iterate forward over array of thresholds starting from | |
4140 | * current_threshold+1 and check if a threshold is crossed. | |
4141 | * If none of thresholds above usage is crossed, we read | |
4142 | * only one element of the array here. | |
4143 | */ | |
4144 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4145 | eventfd_signal(t->entries[i].eventfd, 1); | |
4146 | ||
4147 | /* Update current_threshold */ | |
5407a562 | 4148 | t->current_threshold = i - 1; |
2e72b634 KS |
4149 | unlock: |
4150 | rcu_read_unlock(); | |
4151 | } | |
4152 | ||
4153 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4154 | { | |
ad4ca5f4 KS |
4155 | while (memcg) { |
4156 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 4157 | if (do_memsw_account()) |
ad4ca5f4 KS |
4158 | __mem_cgroup_threshold(memcg, true); |
4159 | ||
4160 | memcg = parent_mem_cgroup(memcg); | |
4161 | } | |
2e72b634 KS |
4162 | } |
4163 | ||
4164 | static int compare_thresholds(const void *a, const void *b) | |
4165 | { | |
4166 | const struct mem_cgroup_threshold *_a = a; | |
4167 | const struct mem_cgroup_threshold *_b = b; | |
4168 | ||
2bff24a3 GT |
4169 | if (_a->threshold > _b->threshold) |
4170 | return 1; | |
4171 | ||
4172 | if (_a->threshold < _b->threshold) | |
4173 | return -1; | |
4174 | ||
4175 | return 0; | |
2e72b634 KS |
4176 | } |
4177 | ||
c0ff4b85 | 4178 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4179 | { |
4180 | struct mem_cgroup_eventfd_list *ev; | |
4181 | ||
2bcf2e92 MH |
4182 | spin_lock(&memcg_oom_lock); |
4183 | ||
c0ff4b85 | 4184 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 4185 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
4186 | |
4187 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
4188 | return 0; |
4189 | } | |
4190 | ||
c0ff4b85 | 4191 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4192 | { |
7d74b06f KH |
4193 | struct mem_cgroup *iter; |
4194 | ||
c0ff4b85 | 4195 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4196 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4197 | } |
4198 | ||
59b6f873 | 4199 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4200 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4201 | { |
2c488db2 KS |
4202 | struct mem_cgroup_thresholds *thresholds; |
4203 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
4204 | unsigned long threshold; |
4205 | unsigned long usage; | |
2c488db2 | 4206 | int i, size, ret; |
2e72b634 | 4207 | |
650c5e56 | 4208 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
4209 | if (ret) |
4210 | return ret; | |
4211 | ||
4212 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4213 | |
05b84301 | 4214 | if (type == _MEM) { |
2c488db2 | 4215 | thresholds = &memcg->thresholds; |
ce00a967 | 4216 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4217 | } else if (type == _MEMSWAP) { |
2c488db2 | 4218 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4219 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4220 | } else |
2e72b634 KS |
4221 | BUG(); |
4222 | ||
2e72b634 | 4223 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4224 | if (thresholds->primary) |
2e72b634 KS |
4225 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4226 | ||
2c488db2 | 4227 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4228 | |
4229 | /* Allocate memory for new array of thresholds */ | |
67b8046f | 4230 | new = kmalloc(struct_size(new, entries, size), GFP_KERNEL); |
2c488db2 | 4231 | if (!new) { |
2e72b634 KS |
4232 | ret = -ENOMEM; |
4233 | goto unlock; | |
4234 | } | |
2c488db2 | 4235 | new->size = size; |
2e72b634 KS |
4236 | |
4237 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4238 | if (thresholds->primary) { |
4239 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4240 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4241 | } |
4242 | ||
2e72b634 | 4243 | /* Add new threshold */ |
2c488db2 KS |
4244 | new->entries[size - 1].eventfd = eventfd; |
4245 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4246 | |
4247 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4248 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4249 | compare_thresholds, NULL); |
4250 | ||
4251 | /* Find current threshold */ | |
2c488db2 | 4252 | new->current_threshold = -1; |
2e72b634 | 4253 | for (i = 0; i < size; i++) { |
748dad36 | 4254 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4255 | /* |
2c488db2 KS |
4256 | * new->current_threshold will not be used until |
4257 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4258 | * it here. |
4259 | */ | |
2c488db2 | 4260 | ++new->current_threshold; |
748dad36 SZ |
4261 | } else |
4262 | break; | |
2e72b634 KS |
4263 | } |
4264 | ||
2c488db2 KS |
4265 | /* Free old spare buffer and save old primary buffer as spare */ |
4266 | kfree(thresholds->spare); | |
4267 | thresholds->spare = thresholds->primary; | |
4268 | ||
4269 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4270 | |
907860ed | 4271 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4272 | synchronize_rcu(); |
4273 | ||
2e72b634 KS |
4274 | unlock: |
4275 | mutex_unlock(&memcg->thresholds_lock); | |
4276 | ||
4277 | return ret; | |
4278 | } | |
4279 | ||
59b6f873 | 4280 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4281 | struct eventfd_ctx *eventfd, const char *args) |
4282 | { | |
59b6f873 | 4283 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4284 | } |
4285 | ||
59b6f873 | 4286 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4287 | struct eventfd_ctx *eventfd, const char *args) |
4288 | { | |
59b6f873 | 4289 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4290 | } |
4291 | ||
59b6f873 | 4292 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4293 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4294 | { |
2c488db2 KS |
4295 | struct mem_cgroup_thresholds *thresholds; |
4296 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 4297 | unsigned long usage; |
7d36665a | 4298 | int i, j, size, entries; |
2e72b634 KS |
4299 | |
4300 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4301 | |
4302 | if (type == _MEM) { | |
2c488db2 | 4303 | thresholds = &memcg->thresholds; |
ce00a967 | 4304 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4305 | } else if (type == _MEMSWAP) { |
2c488db2 | 4306 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4307 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4308 | } else |
2e72b634 KS |
4309 | BUG(); |
4310 | ||
371528ca AV |
4311 | if (!thresholds->primary) |
4312 | goto unlock; | |
4313 | ||
2e72b634 KS |
4314 | /* Check if a threshold crossed before removing */ |
4315 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4316 | ||
4317 | /* Calculate new number of threshold */ | |
7d36665a | 4318 | size = entries = 0; |
2c488db2 KS |
4319 | for (i = 0; i < thresholds->primary->size; i++) { |
4320 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 | 4321 | size++; |
7d36665a CX |
4322 | else |
4323 | entries++; | |
2e72b634 KS |
4324 | } |
4325 | ||
2c488db2 | 4326 | new = thresholds->spare; |
907860ed | 4327 | |
7d36665a CX |
4328 | /* If no items related to eventfd have been cleared, nothing to do */ |
4329 | if (!entries) | |
4330 | goto unlock; | |
4331 | ||
2e72b634 KS |
4332 | /* Set thresholds array to NULL if we don't have thresholds */ |
4333 | if (!size) { | |
2c488db2 KS |
4334 | kfree(new); |
4335 | new = NULL; | |
907860ed | 4336 | goto swap_buffers; |
2e72b634 KS |
4337 | } |
4338 | ||
2c488db2 | 4339 | new->size = size; |
2e72b634 KS |
4340 | |
4341 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4342 | new->current_threshold = -1; |
4343 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4344 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4345 | continue; |
4346 | ||
2c488db2 | 4347 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4348 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4349 | /* |
2c488db2 | 4350 | * new->current_threshold will not be used |
2e72b634 KS |
4351 | * until rcu_assign_pointer(), so it's safe to increment |
4352 | * it here. | |
4353 | */ | |
2c488db2 | 4354 | ++new->current_threshold; |
2e72b634 KS |
4355 | } |
4356 | j++; | |
4357 | } | |
4358 | ||
907860ed | 4359 | swap_buffers: |
2c488db2 KS |
4360 | /* Swap primary and spare array */ |
4361 | thresholds->spare = thresholds->primary; | |
8c757763 | 4362 | |
2c488db2 | 4363 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4364 | |
907860ed | 4365 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4366 | synchronize_rcu(); |
6611d8d7 MC |
4367 | |
4368 | /* If all events are unregistered, free the spare array */ | |
4369 | if (!new) { | |
4370 | kfree(thresholds->spare); | |
4371 | thresholds->spare = NULL; | |
4372 | } | |
371528ca | 4373 | unlock: |
2e72b634 | 4374 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4375 | } |
c1e862c1 | 4376 | |
59b6f873 | 4377 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4378 | struct eventfd_ctx *eventfd) |
4379 | { | |
59b6f873 | 4380 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4381 | } |
4382 | ||
59b6f873 | 4383 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4384 | struct eventfd_ctx *eventfd) |
4385 | { | |
59b6f873 | 4386 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4387 | } |
4388 | ||
59b6f873 | 4389 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4390 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4391 | { |
9490ff27 | 4392 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4393 | |
9490ff27 KH |
4394 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4395 | if (!event) | |
4396 | return -ENOMEM; | |
4397 | ||
1af8efe9 | 4398 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4399 | |
4400 | event->eventfd = eventfd; | |
4401 | list_add(&event->list, &memcg->oom_notify); | |
4402 | ||
4403 | /* already in OOM ? */ | |
c2b42d3c | 4404 | if (memcg->under_oom) |
9490ff27 | 4405 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4406 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4407 | |
4408 | return 0; | |
4409 | } | |
4410 | ||
59b6f873 | 4411 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4412 | struct eventfd_ctx *eventfd) |
9490ff27 | 4413 | { |
9490ff27 | 4414 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4415 | |
1af8efe9 | 4416 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4417 | |
c0ff4b85 | 4418 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4419 | if (ev->eventfd == eventfd) { |
4420 | list_del(&ev->list); | |
4421 | kfree(ev); | |
4422 | } | |
4423 | } | |
4424 | ||
1af8efe9 | 4425 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4426 | } |
4427 | ||
2da8ca82 | 4428 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4429 | { |
aa9694bb | 4430 | struct mem_cgroup *memcg = mem_cgroup_from_seq(sf); |
3c11ecf4 | 4431 | |
791badbd | 4432 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 4433 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
4434 | seq_printf(sf, "oom_kill %lu\n", |
4435 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
4436 | return 0; |
4437 | } | |
4438 | ||
182446d0 | 4439 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4440 | struct cftype *cft, u64 val) |
4441 | { | |
182446d0 | 4442 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4443 | |
4444 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 4445 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4446 | return -EINVAL; |
4447 | ||
c0ff4b85 | 4448 | memcg->oom_kill_disable = val; |
4d845ebf | 4449 | if (!val) |
c0ff4b85 | 4450 | memcg_oom_recover(memcg); |
3dae7fec | 4451 | |
3c11ecf4 KH |
4452 | return 0; |
4453 | } | |
4454 | ||
52ebea74 TH |
4455 | #ifdef CONFIG_CGROUP_WRITEBACK |
4456 | ||
3a8e9ac8 TH |
4457 | #include <trace/events/writeback.h> |
4458 | ||
841710aa TH |
4459 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
4460 | { | |
4461 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
4462 | } | |
4463 | ||
4464 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4465 | { | |
4466 | wb_domain_exit(&memcg->cgwb_domain); | |
4467 | } | |
4468 | ||
2529bb3a TH |
4469 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4470 | { | |
4471 | wb_domain_size_changed(&memcg->cgwb_domain); | |
4472 | } | |
4473 | ||
841710aa TH |
4474 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
4475 | { | |
4476 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4477 | ||
4478 | if (!memcg->css.parent) | |
4479 | return NULL; | |
4480 | ||
4481 | return &memcg->cgwb_domain; | |
4482 | } | |
4483 | ||
0b3d6e6f GT |
4484 | /* |
4485 | * idx can be of type enum memcg_stat_item or node_stat_item. | |
4486 | * Keep in sync with memcg_exact_page(). | |
4487 | */ | |
4488 | static unsigned long memcg_exact_page_state(struct mem_cgroup *memcg, int idx) | |
4489 | { | |
871789d4 | 4490 | long x = atomic_long_read(&memcg->vmstats[idx]); |
0b3d6e6f GT |
4491 | int cpu; |
4492 | ||
4493 | for_each_online_cpu(cpu) | |
871789d4 | 4494 | x += per_cpu_ptr(memcg->vmstats_percpu, cpu)->stat[idx]; |
0b3d6e6f GT |
4495 | if (x < 0) |
4496 | x = 0; | |
4497 | return x; | |
4498 | } | |
4499 | ||
c2aa723a TH |
4500 | /** |
4501 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
4502 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
4503 | * @pfilepages: out parameter for number of file pages |
4504 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
4505 | * @pdirty: out parameter for number of dirty pages |
4506 | * @pwriteback: out parameter for number of pages under writeback | |
4507 | * | |
c5edf9cd TH |
4508 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
4509 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
4510 | * is a bit more involved. | |
c2aa723a | 4511 | * |
c5edf9cd TH |
4512 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
4513 | * headroom is calculated as the lowest headroom of itself and the | |
4514 | * ancestors. Note that this doesn't consider the actual amount of | |
4515 | * available memory in the system. The caller should further cap | |
4516 | * *@pheadroom accordingly. | |
c2aa723a | 4517 | */ |
c5edf9cd TH |
4518 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
4519 | unsigned long *pheadroom, unsigned long *pdirty, | |
4520 | unsigned long *pwriteback) | |
c2aa723a TH |
4521 | { |
4522 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4523 | struct mem_cgroup *parent; | |
c2aa723a | 4524 | |
0b3d6e6f | 4525 | *pdirty = memcg_exact_page_state(memcg, NR_FILE_DIRTY); |
c2aa723a | 4526 | |
0b3d6e6f | 4527 | *pwriteback = memcg_exact_page_state(memcg, NR_WRITEBACK); |
21d89d15 JW |
4528 | *pfilepages = memcg_exact_page_state(memcg, NR_INACTIVE_FILE) + |
4529 | memcg_exact_page_state(memcg, NR_ACTIVE_FILE); | |
c5edf9cd | 4530 | *pheadroom = PAGE_COUNTER_MAX; |
c2aa723a | 4531 | |
c2aa723a | 4532 | while ((parent = parent_mem_cgroup(memcg))) { |
15b42562 | 4533 | unsigned long ceiling = min(READ_ONCE(memcg->memory.max), |
d1663a90 | 4534 | READ_ONCE(memcg->memory.high)); |
c2aa723a TH |
4535 | unsigned long used = page_counter_read(&memcg->memory); |
4536 | ||
c5edf9cd | 4537 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
4538 | memcg = parent; |
4539 | } | |
c2aa723a TH |
4540 | } |
4541 | ||
97b27821 TH |
4542 | /* |
4543 | * Foreign dirty flushing | |
4544 | * | |
4545 | * There's an inherent mismatch between memcg and writeback. The former | |
4546 | * trackes ownership per-page while the latter per-inode. This was a | |
4547 | * deliberate design decision because honoring per-page ownership in the | |
4548 | * writeback path is complicated, may lead to higher CPU and IO overheads | |
4549 | * and deemed unnecessary given that write-sharing an inode across | |
4550 | * different cgroups isn't a common use-case. | |
4551 | * | |
4552 | * Combined with inode majority-writer ownership switching, this works well | |
4553 | * enough in most cases but there are some pathological cases. For | |
4554 | * example, let's say there are two cgroups A and B which keep writing to | |
4555 | * different but confined parts of the same inode. B owns the inode and | |
4556 | * A's memory is limited far below B's. A's dirty ratio can rise enough to | |
4557 | * trigger balance_dirty_pages() sleeps but B's can be low enough to avoid | |
4558 | * triggering background writeback. A will be slowed down without a way to | |
4559 | * make writeback of the dirty pages happen. | |
4560 | * | |
4561 | * Conditions like the above can lead to a cgroup getting repatedly and | |
4562 | * severely throttled after making some progress after each | |
4563 | * dirty_expire_interval while the underyling IO device is almost | |
4564 | * completely idle. | |
4565 | * | |
4566 | * Solving this problem completely requires matching the ownership tracking | |
4567 | * granularities between memcg and writeback in either direction. However, | |
4568 | * the more egregious behaviors can be avoided by simply remembering the | |
4569 | * most recent foreign dirtying events and initiating remote flushes on | |
4570 | * them when local writeback isn't enough to keep the memory clean enough. | |
4571 | * | |
4572 | * The following two functions implement such mechanism. When a foreign | |
4573 | * page - a page whose memcg and writeback ownerships don't match - is | |
4574 | * dirtied, mem_cgroup_track_foreign_dirty() records the inode owning | |
4575 | * bdi_writeback on the page owning memcg. When balance_dirty_pages() | |
4576 | * decides that the memcg needs to sleep due to high dirty ratio, it calls | |
4577 | * mem_cgroup_flush_foreign() which queues writeback on the recorded | |
4578 | * foreign bdi_writebacks which haven't expired. Both the numbers of | |
4579 | * recorded bdi_writebacks and concurrent in-flight foreign writebacks are | |
4580 | * limited to MEMCG_CGWB_FRN_CNT. | |
4581 | * | |
4582 | * The mechanism only remembers IDs and doesn't hold any object references. | |
4583 | * As being wrong occasionally doesn't matter, updates and accesses to the | |
4584 | * records are lockless and racy. | |
4585 | */ | |
4586 | void mem_cgroup_track_foreign_dirty_slowpath(struct page *page, | |
4587 | struct bdi_writeback *wb) | |
4588 | { | |
4589 | struct mem_cgroup *memcg = page->mem_cgroup; | |
4590 | struct memcg_cgwb_frn *frn; | |
4591 | u64 now = get_jiffies_64(); | |
4592 | u64 oldest_at = now; | |
4593 | int oldest = -1; | |
4594 | int i; | |
4595 | ||
3a8e9ac8 TH |
4596 | trace_track_foreign_dirty(page, wb); |
4597 | ||
97b27821 TH |
4598 | /* |
4599 | * Pick the slot to use. If there is already a slot for @wb, keep | |
4600 | * using it. If not replace the oldest one which isn't being | |
4601 | * written out. | |
4602 | */ | |
4603 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4604 | frn = &memcg->cgwb_frn[i]; | |
4605 | if (frn->bdi_id == wb->bdi->id && | |
4606 | frn->memcg_id == wb->memcg_css->id) | |
4607 | break; | |
4608 | if (time_before64(frn->at, oldest_at) && | |
4609 | atomic_read(&frn->done.cnt) == 1) { | |
4610 | oldest = i; | |
4611 | oldest_at = frn->at; | |
4612 | } | |
4613 | } | |
4614 | ||
4615 | if (i < MEMCG_CGWB_FRN_CNT) { | |
4616 | /* | |
4617 | * Re-using an existing one. Update timestamp lazily to | |
4618 | * avoid making the cacheline hot. We want them to be | |
4619 | * reasonably up-to-date and significantly shorter than | |
4620 | * dirty_expire_interval as that's what expires the record. | |
4621 | * Use the shorter of 1s and dirty_expire_interval / 8. | |
4622 | */ | |
4623 | unsigned long update_intv = | |
4624 | min_t(unsigned long, HZ, | |
4625 | msecs_to_jiffies(dirty_expire_interval * 10) / 8); | |
4626 | ||
4627 | if (time_before64(frn->at, now - update_intv)) | |
4628 | frn->at = now; | |
4629 | } else if (oldest >= 0) { | |
4630 | /* replace the oldest free one */ | |
4631 | frn = &memcg->cgwb_frn[oldest]; | |
4632 | frn->bdi_id = wb->bdi->id; | |
4633 | frn->memcg_id = wb->memcg_css->id; | |
4634 | frn->at = now; | |
4635 | } | |
4636 | } | |
4637 | ||
4638 | /* issue foreign writeback flushes for recorded foreign dirtying events */ | |
4639 | void mem_cgroup_flush_foreign(struct bdi_writeback *wb) | |
4640 | { | |
4641 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4642 | unsigned long intv = msecs_to_jiffies(dirty_expire_interval * 10); | |
4643 | u64 now = jiffies_64; | |
4644 | int i; | |
4645 | ||
4646 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4647 | struct memcg_cgwb_frn *frn = &memcg->cgwb_frn[i]; | |
4648 | ||
4649 | /* | |
4650 | * If the record is older than dirty_expire_interval, | |
4651 | * writeback on it has already started. No need to kick it | |
4652 | * off again. Also, don't start a new one if there's | |
4653 | * already one in flight. | |
4654 | */ | |
4655 | if (time_after64(frn->at, now - intv) && | |
4656 | atomic_read(&frn->done.cnt) == 1) { | |
4657 | frn->at = 0; | |
3a8e9ac8 | 4658 | trace_flush_foreign(wb, frn->bdi_id, frn->memcg_id); |
97b27821 TH |
4659 | cgroup_writeback_by_id(frn->bdi_id, frn->memcg_id, 0, |
4660 | WB_REASON_FOREIGN_FLUSH, | |
4661 | &frn->done); | |
4662 | } | |
4663 | } | |
4664 | } | |
4665 | ||
841710aa TH |
4666 | #else /* CONFIG_CGROUP_WRITEBACK */ |
4667 | ||
4668 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
4669 | { | |
4670 | return 0; | |
4671 | } | |
4672 | ||
4673 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4674 | { | |
4675 | } | |
4676 | ||
2529bb3a TH |
4677 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4678 | { | |
4679 | } | |
4680 | ||
52ebea74 TH |
4681 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
4682 | ||
3bc942f3 TH |
4683 | /* |
4684 | * DO NOT USE IN NEW FILES. | |
4685 | * | |
4686 | * "cgroup.event_control" implementation. | |
4687 | * | |
4688 | * This is way over-engineered. It tries to support fully configurable | |
4689 | * events for each user. Such level of flexibility is completely | |
4690 | * unnecessary especially in the light of the planned unified hierarchy. | |
4691 | * | |
4692 | * Please deprecate this and replace with something simpler if at all | |
4693 | * possible. | |
4694 | */ | |
4695 | ||
79bd9814 TH |
4696 | /* |
4697 | * Unregister event and free resources. | |
4698 | * | |
4699 | * Gets called from workqueue. | |
4700 | */ | |
3bc942f3 | 4701 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4702 | { |
3bc942f3 TH |
4703 | struct mem_cgroup_event *event = |
4704 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4705 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4706 | |
4707 | remove_wait_queue(event->wqh, &event->wait); | |
4708 | ||
59b6f873 | 4709 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4710 | |
4711 | /* Notify userspace the event is going away. */ | |
4712 | eventfd_signal(event->eventfd, 1); | |
4713 | ||
4714 | eventfd_ctx_put(event->eventfd); | |
4715 | kfree(event); | |
59b6f873 | 4716 | css_put(&memcg->css); |
79bd9814 TH |
4717 | } |
4718 | ||
4719 | /* | |
a9a08845 | 4720 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
4721 | * |
4722 | * Called with wqh->lock held and interrupts disabled. | |
4723 | */ | |
ac6424b9 | 4724 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 4725 | int sync, void *key) |
79bd9814 | 4726 | { |
3bc942f3 TH |
4727 | struct mem_cgroup_event *event = |
4728 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4729 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 4730 | __poll_t flags = key_to_poll(key); |
79bd9814 | 4731 | |
a9a08845 | 4732 | if (flags & EPOLLHUP) { |
79bd9814 TH |
4733 | /* |
4734 | * If the event has been detached at cgroup removal, we | |
4735 | * can simply return knowing the other side will cleanup | |
4736 | * for us. | |
4737 | * | |
4738 | * We can't race against event freeing since the other | |
4739 | * side will require wqh->lock via remove_wait_queue(), | |
4740 | * which we hold. | |
4741 | */ | |
fba94807 | 4742 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4743 | if (!list_empty(&event->list)) { |
4744 | list_del_init(&event->list); | |
4745 | /* | |
4746 | * We are in atomic context, but cgroup_event_remove() | |
4747 | * may sleep, so we have to call it in workqueue. | |
4748 | */ | |
4749 | schedule_work(&event->remove); | |
4750 | } | |
fba94807 | 4751 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4752 | } |
4753 | ||
4754 | return 0; | |
4755 | } | |
4756 | ||
3bc942f3 | 4757 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4758 | wait_queue_head_t *wqh, poll_table *pt) |
4759 | { | |
3bc942f3 TH |
4760 | struct mem_cgroup_event *event = |
4761 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4762 | |
4763 | event->wqh = wqh; | |
4764 | add_wait_queue(wqh, &event->wait); | |
4765 | } | |
4766 | ||
4767 | /* | |
3bc942f3 TH |
4768 | * DO NOT USE IN NEW FILES. |
4769 | * | |
79bd9814 TH |
4770 | * Parse input and register new cgroup event handler. |
4771 | * | |
4772 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4773 | * Interpretation of args is defined by control file implementation. | |
4774 | */ | |
451af504 TH |
4775 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4776 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4777 | { |
451af504 | 4778 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4779 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4780 | struct mem_cgroup_event *event; |
79bd9814 TH |
4781 | struct cgroup_subsys_state *cfile_css; |
4782 | unsigned int efd, cfd; | |
4783 | struct fd efile; | |
4784 | struct fd cfile; | |
fba94807 | 4785 | const char *name; |
79bd9814 TH |
4786 | char *endp; |
4787 | int ret; | |
4788 | ||
451af504 TH |
4789 | buf = strstrip(buf); |
4790 | ||
4791 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4792 | if (*endp != ' ') |
4793 | return -EINVAL; | |
451af504 | 4794 | buf = endp + 1; |
79bd9814 | 4795 | |
451af504 | 4796 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4797 | if ((*endp != ' ') && (*endp != '\0')) |
4798 | return -EINVAL; | |
451af504 | 4799 | buf = endp + 1; |
79bd9814 TH |
4800 | |
4801 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4802 | if (!event) | |
4803 | return -ENOMEM; | |
4804 | ||
59b6f873 | 4805 | event->memcg = memcg; |
79bd9814 | 4806 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4807 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4808 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4809 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4810 | |
4811 | efile = fdget(efd); | |
4812 | if (!efile.file) { | |
4813 | ret = -EBADF; | |
4814 | goto out_kfree; | |
4815 | } | |
4816 | ||
4817 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4818 | if (IS_ERR(event->eventfd)) { | |
4819 | ret = PTR_ERR(event->eventfd); | |
4820 | goto out_put_efile; | |
4821 | } | |
4822 | ||
4823 | cfile = fdget(cfd); | |
4824 | if (!cfile.file) { | |
4825 | ret = -EBADF; | |
4826 | goto out_put_eventfd; | |
4827 | } | |
4828 | ||
4829 | /* the process need read permission on control file */ | |
4830 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
4831 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
4832 | if (ret < 0) | |
4833 | goto out_put_cfile; | |
4834 | ||
fba94807 TH |
4835 | /* |
4836 | * Determine the event callbacks and set them in @event. This used | |
4837 | * to be done via struct cftype but cgroup core no longer knows | |
4838 | * about these events. The following is crude but the whole thing | |
4839 | * is for compatibility anyway. | |
3bc942f3 TH |
4840 | * |
4841 | * DO NOT ADD NEW FILES. | |
fba94807 | 4842 | */ |
b583043e | 4843 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4844 | |
4845 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4846 | event->register_event = mem_cgroup_usage_register_event; | |
4847 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4848 | } else if (!strcmp(name, "memory.oom_control")) { | |
4849 | event->register_event = mem_cgroup_oom_register_event; | |
4850 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4851 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4852 | event->register_event = vmpressure_register_event; | |
4853 | event->unregister_event = vmpressure_unregister_event; | |
4854 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4855 | event->register_event = memsw_cgroup_usage_register_event; |
4856 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4857 | } else { |
4858 | ret = -EINVAL; | |
4859 | goto out_put_cfile; | |
4860 | } | |
4861 | ||
79bd9814 | 4862 | /* |
b5557c4c TH |
4863 | * Verify @cfile should belong to @css. Also, remaining events are |
4864 | * automatically removed on cgroup destruction but the removal is | |
4865 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4866 | */ |
b583043e | 4867 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4868 | &memory_cgrp_subsys); |
79bd9814 | 4869 | ret = -EINVAL; |
5a17f543 | 4870 | if (IS_ERR(cfile_css)) |
79bd9814 | 4871 | goto out_put_cfile; |
5a17f543 TH |
4872 | if (cfile_css != css) { |
4873 | css_put(cfile_css); | |
79bd9814 | 4874 | goto out_put_cfile; |
5a17f543 | 4875 | } |
79bd9814 | 4876 | |
451af504 | 4877 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4878 | if (ret) |
4879 | goto out_put_css; | |
4880 | ||
9965ed17 | 4881 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 4882 | |
fba94807 TH |
4883 | spin_lock(&memcg->event_list_lock); |
4884 | list_add(&event->list, &memcg->event_list); | |
4885 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4886 | |
4887 | fdput(cfile); | |
4888 | fdput(efile); | |
4889 | ||
451af504 | 4890 | return nbytes; |
79bd9814 TH |
4891 | |
4892 | out_put_css: | |
b5557c4c | 4893 | css_put(css); |
79bd9814 TH |
4894 | out_put_cfile: |
4895 | fdput(cfile); | |
4896 | out_put_eventfd: | |
4897 | eventfd_ctx_put(event->eventfd); | |
4898 | out_put_efile: | |
4899 | fdput(efile); | |
4900 | out_kfree: | |
4901 | kfree(event); | |
4902 | ||
4903 | return ret; | |
4904 | } | |
4905 | ||
241994ed | 4906 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4907 | { |
0eea1030 | 4908 | .name = "usage_in_bytes", |
8c7c6e34 | 4909 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4910 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4911 | }, |
c84872e1 PE |
4912 | { |
4913 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4914 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4915 | .write = mem_cgroup_reset, |
791badbd | 4916 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4917 | }, |
8cdea7c0 | 4918 | { |
0eea1030 | 4919 | .name = "limit_in_bytes", |
8c7c6e34 | 4920 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4921 | .write = mem_cgroup_write, |
791badbd | 4922 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4923 | }, |
296c81d8 BS |
4924 | { |
4925 | .name = "soft_limit_in_bytes", | |
4926 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4927 | .write = mem_cgroup_write, |
791badbd | 4928 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4929 | }, |
8cdea7c0 BS |
4930 | { |
4931 | .name = "failcnt", | |
8c7c6e34 | 4932 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4933 | .write = mem_cgroup_reset, |
791badbd | 4934 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4935 | }, |
d2ceb9b7 KH |
4936 | { |
4937 | .name = "stat", | |
2da8ca82 | 4938 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4939 | }, |
c1e862c1 KH |
4940 | { |
4941 | .name = "force_empty", | |
6770c64e | 4942 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4943 | }, |
18f59ea7 BS |
4944 | { |
4945 | .name = "use_hierarchy", | |
4946 | .write_u64 = mem_cgroup_hierarchy_write, | |
4947 | .read_u64 = mem_cgroup_hierarchy_read, | |
4948 | }, | |
79bd9814 | 4949 | { |
3bc942f3 | 4950 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4951 | .write = memcg_write_event_control, |
7dbdb199 | 4952 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 4953 | }, |
a7885eb8 KM |
4954 | { |
4955 | .name = "swappiness", | |
4956 | .read_u64 = mem_cgroup_swappiness_read, | |
4957 | .write_u64 = mem_cgroup_swappiness_write, | |
4958 | }, | |
7dc74be0 DN |
4959 | { |
4960 | .name = "move_charge_at_immigrate", | |
4961 | .read_u64 = mem_cgroup_move_charge_read, | |
4962 | .write_u64 = mem_cgroup_move_charge_write, | |
4963 | }, | |
9490ff27 KH |
4964 | { |
4965 | .name = "oom_control", | |
2da8ca82 | 4966 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4967 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4968 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4969 | }, | |
70ddf637 AV |
4970 | { |
4971 | .name = "pressure_level", | |
70ddf637 | 4972 | }, |
406eb0c9 YH |
4973 | #ifdef CONFIG_NUMA |
4974 | { | |
4975 | .name = "numa_stat", | |
2da8ca82 | 4976 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4977 | }, |
4978 | #endif | |
510fc4e1 GC |
4979 | { |
4980 | .name = "kmem.limit_in_bytes", | |
4981 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4982 | .write = mem_cgroup_write, |
791badbd | 4983 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4984 | }, |
4985 | { | |
4986 | .name = "kmem.usage_in_bytes", | |
4987 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4988 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4989 | }, |
4990 | { | |
4991 | .name = "kmem.failcnt", | |
4992 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4993 | .write = mem_cgroup_reset, |
791badbd | 4994 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4995 | }, |
4996 | { | |
4997 | .name = "kmem.max_usage_in_bytes", | |
4998 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4999 | .write = mem_cgroup_reset, |
791badbd | 5000 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 5001 | }, |
a87425a3 YS |
5002 | #if defined(CONFIG_MEMCG_KMEM) && \ |
5003 | (defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)) | |
749c5415 GC |
5004 | { |
5005 | .name = "kmem.slabinfo", | |
b047501c | 5006 | .seq_show = memcg_slab_show, |
749c5415 GC |
5007 | }, |
5008 | #endif | |
d55f90bf VD |
5009 | { |
5010 | .name = "kmem.tcp.limit_in_bytes", | |
5011 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
5012 | .write = mem_cgroup_write, | |
5013 | .read_u64 = mem_cgroup_read_u64, | |
5014 | }, | |
5015 | { | |
5016 | .name = "kmem.tcp.usage_in_bytes", | |
5017 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
5018 | .read_u64 = mem_cgroup_read_u64, | |
5019 | }, | |
5020 | { | |
5021 | .name = "kmem.tcp.failcnt", | |
5022 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
5023 | .write = mem_cgroup_reset, | |
5024 | .read_u64 = mem_cgroup_read_u64, | |
5025 | }, | |
5026 | { | |
5027 | .name = "kmem.tcp.max_usage_in_bytes", | |
5028 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
5029 | .write = mem_cgroup_reset, | |
5030 | .read_u64 = mem_cgroup_read_u64, | |
5031 | }, | |
6bc10349 | 5032 | { }, /* terminate */ |
af36f906 | 5033 | }; |
8c7c6e34 | 5034 | |
73f576c0 JW |
5035 | /* |
5036 | * Private memory cgroup IDR | |
5037 | * | |
5038 | * Swap-out records and page cache shadow entries need to store memcg | |
5039 | * references in constrained space, so we maintain an ID space that is | |
5040 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
5041 | * memory-controlled cgroups to 64k. | |
5042 | * | |
b8f2935f | 5043 | * However, there usually are many references to the offline CSS after |
73f576c0 JW |
5044 | * the cgroup has been destroyed, such as page cache or reclaimable |
5045 | * slab objects, that don't need to hang on to the ID. We want to keep | |
5046 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
5047 | * relatively small ID space and prevent the creation of new cgroups | |
5048 | * even when there are much fewer than 64k cgroups - possibly none. | |
5049 | * | |
5050 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
5051 | * be freed and recycled when it's no longer needed, which is usually | |
5052 | * when the CSS is offlined. | |
5053 | * | |
5054 | * The only exception to that are records of swapped out tmpfs/shmem | |
5055 | * pages that need to be attributed to live ancestors on swapin. But | |
5056 | * those references are manageable from userspace. | |
5057 | */ | |
5058 | ||
5059 | static DEFINE_IDR(mem_cgroup_idr); | |
5060 | ||
7e97de0b KT |
5061 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
5062 | { | |
5063 | if (memcg->id.id > 0) { | |
5064 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
5065 | memcg->id.id = 0; | |
5066 | } | |
5067 | } | |
5068 | ||
c1514c0a VF |
5069 | static void __maybe_unused mem_cgroup_id_get_many(struct mem_cgroup *memcg, |
5070 | unsigned int n) | |
73f576c0 | 5071 | { |
1c2d479a | 5072 | refcount_add(n, &memcg->id.ref); |
73f576c0 JW |
5073 | } |
5074 | ||
615d66c3 | 5075 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 5076 | { |
1c2d479a | 5077 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 5078 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
5079 | |
5080 | /* Memcg ID pins CSS */ | |
5081 | css_put(&memcg->css); | |
5082 | } | |
5083 | } | |
5084 | ||
615d66c3 VD |
5085 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) |
5086 | { | |
5087 | mem_cgroup_id_put_many(memcg, 1); | |
5088 | } | |
5089 | ||
73f576c0 JW |
5090 | /** |
5091 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
5092 | * @id: the memcg id to look up | |
5093 | * | |
5094 | * Caller must hold rcu_read_lock(). | |
5095 | */ | |
5096 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
5097 | { | |
5098 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5099 | return idr_find(&mem_cgroup_idr, id); | |
5100 | } | |
5101 | ||
ef8f2327 | 5102 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
5103 | { |
5104 | struct mem_cgroup_per_node *pn; | |
ef8f2327 | 5105 | int tmp = node; |
1ecaab2b KH |
5106 | /* |
5107 | * This routine is called against possible nodes. | |
5108 | * But it's BUG to call kmalloc() against offline node. | |
5109 | * | |
5110 | * TODO: this routine can waste much memory for nodes which will | |
5111 | * never be onlined. It's better to use memory hotplug callback | |
5112 | * function. | |
5113 | */ | |
41e3355d KH |
5114 | if (!node_state(node, N_NORMAL_MEMORY)) |
5115 | tmp = -1; | |
17295c88 | 5116 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
5117 | if (!pn) |
5118 | return 1; | |
1ecaab2b | 5119 | |
815744d7 JW |
5120 | pn->lruvec_stat_local = alloc_percpu(struct lruvec_stat); |
5121 | if (!pn->lruvec_stat_local) { | |
5122 | kfree(pn); | |
5123 | return 1; | |
5124 | } | |
5125 | ||
a983b5eb JW |
5126 | pn->lruvec_stat_cpu = alloc_percpu(struct lruvec_stat); |
5127 | if (!pn->lruvec_stat_cpu) { | |
815744d7 | 5128 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c JW |
5129 | kfree(pn); |
5130 | return 1; | |
5131 | } | |
5132 | ||
ef8f2327 MG |
5133 | lruvec_init(&pn->lruvec); |
5134 | pn->usage_in_excess = 0; | |
5135 | pn->on_tree = false; | |
5136 | pn->memcg = memcg; | |
5137 | ||
54f72fe0 | 5138 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
5139 | return 0; |
5140 | } | |
5141 | ||
ef8f2327 | 5142 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 5143 | { |
00f3ca2c JW |
5144 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
5145 | ||
4eaf431f MH |
5146 | if (!pn) |
5147 | return; | |
5148 | ||
a983b5eb | 5149 | free_percpu(pn->lruvec_stat_cpu); |
815744d7 | 5150 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c | 5151 | kfree(pn); |
1ecaab2b KH |
5152 | } |
5153 | ||
40e952f9 | 5154 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 5155 | { |
c8b2a36f | 5156 | int node; |
59927fb9 | 5157 | |
c8b2a36f | 5158 | for_each_node(node) |
ef8f2327 | 5159 | free_mem_cgroup_per_node_info(memcg, node); |
871789d4 | 5160 | free_percpu(memcg->vmstats_percpu); |
815744d7 | 5161 | free_percpu(memcg->vmstats_local); |
8ff69e2c | 5162 | kfree(memcg); |
59927fb9 | 5163 | } |
3afe36b1 | 5164 | |
40e952f9 TE |
5165 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
5166 | { | |
5167 | memcg_wb_domain_exit(memcg); | |
7961eee3 SB |
5168 | /* |
5169 | * Flush percpu vmstats and vmevents to guarantee the value correctness | |
5170 | * on parent's and all ancestor levels. | |
5171 | */ | |
4a87e2a2 | 5172 | memcg_flush_percpu_vmstats(memcg); |
7961eee3 | 5173 | memcg_flush_percpu_vmevents(memcg); |
40e952f9 TE |
5174 | __mem_cgroup_free(memcg); |
5175 | } | |
5176 | ||
0b8f73e1 | 5177 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 5178 | { |
d142e3e6 | 5179 | struct mem_cgroup *memcg; |
b9726c26 | 5180 | unsigned int size; |
6d12e2d8 | 5181 | int node; |
97b27821 | 5182 | int __maybe_unused i; |
11d67612 | 5183 | long error = -ENOMEM; |
8cdea7c0 | 5184 | |
0b8f73e1 JW |
5185 | size = sizeof(struct mem_cgroup); |
5186 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
5187 | ||
5188 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 5189 | if (!memcg) |
11d67612 | 5190 | return ERR_PTR(error); |
0b8f73e1 | 5191 | |
73f576c0 JW |
5192 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
5193 | 1, MEM_CGROUP_ID_MAX, | |
5194 | GFP_KERNEL); | |
11d67612 YS |
5195 | if (memcg->id.id < 0) { |
5196 | error = memcg->id.id; | |
73f576c0 | 5197 | goto fail; |
11d67612 | 5198 | } |
73f576c0 | 5199 | |
815744d7 JW |
5200 | memcg->vmstats_local = alloc_percpu(struct memcg_vmstats_percpu); |
5201 | if (!memcg->vmstats_local) | |
5202 | goto fail; | |
5203 | ||
871789d4 CD |
5204 | memcg->vmstats_percpu = alloc_percpu(struct memcg_vmstats_percpu); |
5205 | if (!memcg->vmstats_percpu) | |
0b8f73e1 | 5206 | goto fail; |
78fb7466 | 5207 | |
3ed28fa1 | 5208 | for_each_node(node) |
ef8f2327 | 5209 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 5210 | goto fail; |
f64c3f54 | 5211 | |
0b8f73e1 JW |
5212 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
5213 | goto fail; | |
28dbc4b6 | 5214 | |
f7e1cb6e | 5215 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 | 5216 | INIT_LIST_HEAD(&memcg->oom_notify); |
d142e3e6 GC |
5217 | mutex_init(&memcg->thresholds_lock); |
5218 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 5219 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
5220 | INIT_LIST_HEAD(&memcg->event_list); |
5221 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 5222 | memcg->socket_pressure = jiffies; |
84c07d11 | 5223 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 5224 | memcg->kmemcg_id = -1; |
bf4f0599 | 5225 | INIT_LIST_HEAD(&memcg->objcg_list); |
900a38f0 | 5226 | #endif |
52ebea74 TH |
5227 | #ifdef CONFIG_CGROUP_WRITEBACK |
5228 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
97b27821 TH |
5229 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) |
5230 | memcg->cgwb_frn[i].done = | |
5231 | __WB_COMPLETION_INIT(&memcg_cgwb_frn_waitq); | |
87eaceb3 YS |
5232 | #endif |
5233 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5234 | spin_lock_init(&memcg->deferred_split_queue.split_queue_lock); | |
5235 | INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue); | |
5236 | memcg->deferred_split_queue.split_queue_len = 0; | |
52ebea74 | 5237 | #endif |
73f576c0 | 5238 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
5239 | return memcg; |
5240 | fail: | |
7e97de0b | 5241 | mem_cgroup_id_remove(memcg); |
40e952f9 | 5242 | __mem_cgroup_free(memcg); |
11d67612 | 5243 | return ERR_PTR(error); |
d142e3e6 GC |
5244 | } |
5245 | ||
0b8f73e1 JW |
5246 | static struct cgroup_subsys_state * __ref |
5247 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 5248 | { |
0b8f73e1 JW |
5249 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
5250 | struct mem_cgroup *memcg; | |
5251 | long error = -ENOMEM; | |
d142e3e6 | 5252 | |
0b8f73e1 | 5253 | memcg = mem_cgroup_alloc(); |
11d67612 YS |
5254 | if (IS_ERR(memcg)) |
5255 | return ERR_CAST(memcg); | |
d142e3e6 | 5256 | |
d1663a90 | 5257 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
0b8f73e1 | 5258 | memcg->soft_limit = PAGE_COUNTER_MAX; |
4b82ab4f | 5259 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
0b8f73e1 JW |
5260 | if (parent) { |
5261 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
5262 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
5263 | } | |
5264 | if (parent && parent->use_hierarchy) { | |
5265 | memcg->use_hierarchy = true; | |
3e32cb2e | 5266 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 5267 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e JW |
5268 | page_counter_init(&memcg->memsw, &parent->memsw); |
5269 | page_counter_init(&memcg->kmem, &parent->kmem); | |
0db15298 | 5270 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 5271 | } else { |
3e32cb2e | 5272 | page_counter_init(&memcg->memory, NULL); |
37e84351 | 5273 | page_counter_init(&memcg->swap, NULL); |
3e32cb2e JW |
5274 | page_counter_init(&memcg->memsw, NULL); |
5275 | page_counter_init(&memcg->kmem, NULL); | |
0db15298 | 5276 | page_counter_init(&memcg->tcpmem, NULL); |
8c7f6edb TH |
5277 | /* |
5278 | * Deeper hierachy with use_hierarchy == false doesn't make | |
5279 | * much sense so let cgroup subsystem know about this | |
5280 | * unfortunate state in our controller. | |
5281 | */ | |
d142e3e6 | 5282 | if (parent != root_mem_cgroup) |
073219e9 | 5283 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 5284 | } |
d6441637 | 5285 | |
0b8f73e1 JW |
5286 | /* The following stuff does not apply to the root */ |
5287 | if (!parent) { | |
5288 | root_mem_cgroup = memcg; | |
5289 | return &memcg->css; | |
5290 | } | |
5291 | ||
b313aeee | 5292 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
5293 | if (error) |
5294 | goto fail; | |
127424c8 | 5295 | |
f7e1cb6e | 5296 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5297 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 5298 | |
0b8f73e1 JW |
5299 | return &memcg->css; |
5300 | fail: | |
7e97de0b | 5301 | mem_cgroup_id_remove(memcg); |
0b8f73e1 | 5302 | mem_cgroup_free(memcg); |
11d67612 | 5303 | return ERR_PTR(error); |
0b8f73e1 JW |
5304 | } |
5305 | ||
73f576c0 | 5306 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 5307 | { |
58fa2a55 VD |
5308 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5309 | ||
0a4465d3 KT |
5310 | /* |
5311 | * A memcg must be visible for memcg_expand_shrinker_maps() | |
5312 | * by the time the maps are allocated. So, we allocate maps | |
5313 | * here, when for_each_mem_cgroup() can't skip it. | |
5314 | */ | |
5315 | if (memcg_alloc_shrinker_maps(memcg)) { | |
5316 | mem_cgroup_id_remove(memcg); | |
5317 | return -ENOMEM; | |
5318 | } | |
5319 | ||
73f576c0 | 5320 | /* Online state pins memcg ID, memcg ID pins CSS */ |
1c2d479a | 5321 | refcount_set(&memcg->id.ref, 1); |
73f576c0 | 5322 | css_get(css); |
2f7dd7a4 | 5323 | return 0; |
8cdea7c0 BS |
5324 | } |
5325 | ||
eb95419b | 5326 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 5327 | { |
eb95419b | 5328 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5329 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
5330 | |
5331 | /* | |
5332 | * Unregister events and notify userspace. | |
5333 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
5334 | * directory to avoid race between userspace and kernelspace. | |
5335 | */ | |
fba94807 TH |
5336 | spin_lock(&memcg->event_list_lock); |
5337 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
5338 | list_del_init(&event->list); |
5339 | schedule_work(&event->remove); | |
5340 | } | |
fba94807 | 5341 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 5342 | |
bf8d5d52 | 5343 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5344 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 5345 | |
567e9ab2 | 5346 | memcg_offline_kmem(memcg); |
52ebea74 | 5347 | wb_memcg_offline(memcg); |
73f576c0 | 5348 | |
591edfb1 RG |
5349 | drain_all_stock(memcg); |
5350 | ||
73f576c0 | 5351 | mem_cgroup_id_put(memcg); |
df878fb0 KH |
5352 | } |
5353 | ||
6df38689 VD |
5354 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
5355 | { | |
5356 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5357 | ||
5358 | invalidate_reclaim_iterators(memcg); | |
5359 | } | |
5360 | ||
eb95419b | 5361 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 5362 | { |
eb95419b | 5363 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
97b27821 | 5364 | int __maybe_unused i; |
c268e994 | 5365 | |
97b27821 TH |
5366 | #ifdef CONFIG_CGROUP_WRITEBACK |
5367 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) | |
5368 | wb_wait_for_completion(&memcg->cgwb_frn[i].done); | |
5369 | #endif | |
f7e1cb6e | 5370 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5371 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 5372 | |
0db15298 | 5373 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 5374 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 5375 | |
0b8f73e1 JW |
5376 | vmpressure_cleanup(&memcg->vmpressure); |
5377 | cancel_work_sync(&memcg->high_work); | |
5378 | mem_cgroup_remove_from_trees(memcg); | |
0a4465d3 | 5379 | memcg_free_shrinker_maps(memcg); |
d886f4e4 | 5380 | memcg_free_kmem(memcg); |
0b8f73e1 | 5381 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
5382 | } |
5383 | ||
1ced953b TH |
5384 | /** |
5385 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
5386 | * @css: the target css | |
5387 | * | |
5388 | * Reset the states of the mem_cgroup associated with @css. This is | |
5389 | * invoked when the userland requests disabling on the default hierarchy | |
5390 | * but the memcg is pinned through dependency. The memcg should stop | |
5391 | * applying policies and should revert to the vanilla state as it may be | |
5392 | * made visible again. | |
5393 | * | |
5394 | * The current implementation only resets the essential configurations. | |
5395 | * This needs to be expanded to cover all the visible parts. | |
5396 | */ | |
5397 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
5398 | { | |
5399 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5400 | ||
bbec2e15 RG |
5401 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
5402 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
5403 | page_counter_set_max(&memcg->memsw, PAGE_COUNTER_MAX); | |
5404 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); | |
5405 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 5406 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5407 | page_counter_set_low(&memcg->memory, 0); |
d1663a90 | 5408 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
24d404dc | 5409 | memcg->soft_limit = PAGE_COUNTER_MAX; |
4b82ab4f | 5410 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
2529bb3a | 5411 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
5412 | } |
5413 | ||
02491447 | 5414 | #ifdef CONFIG_MMU |
7dc74be0 | 5415 | /* Handlers for move charge at task migration. */ |
854ffa8d | 5416 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 5417 | { |
05b84301 | 5418 | int ret; |
9476db97 | 5419 | |
d0164adc MG |
5420 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
5421 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 5422 | if (!ret) { |
854ffa8d | 5423 | mc.precharge += count; |
854ffa8d DN |
5424 | return ret; |
5425 | } | |
9476db97 | 5426 | |
3674534b | 5427 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 5428 | while (count--) { |
3674534b | 5429 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 5430 | if (ret) |
38c5d72f | 5431 | return ret; |
854ffa8d | 5432 | mc.precharge++; |
9476db97 | 5433 | cond_resched(); |
854ffa8d | 5434 | } |
9476db97 | 5435 | return 0; |
4ffef5fe DN |
5436 | } |
5437 | ||
4ffef5fe DN |
5438 | union mc_target { |
5439 | struct page *page; | |
02491447 | 5440 | swp_entry_t ent; |
4ffef5fe DN |
5441 | }; |
5442 | ||
4ffef5fe | 5443 | enum mc_target_type { |
8d32ff84 | 5444 | MC_TARGET_NONE = 0, |
4ffef5fe | 5445 | MC_TARGET_PAGE, |
02491447 | 5446 | MC_TARGET_SWAP, |
c733a828 | 5447 | MC_TARGET_DEVICE, |
4ffef5fe DN |
5448 | }; |
5449 | ||
90254a65 DN |
5450 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5451 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5452 | { |
25b2995a | 5453 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5454 | |
90254a65 DN |
5455 | if (!page || !page_mapped(page)) |
5456 | return NULL; | |
5457 | if (PageAnon(page)) { | |
1dfab5ab | 5458 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5459 | return NULL; |
1dfab5ab JW |
5460 | } else { |
5461 | if (!(mc.flags & MOVE_FILE)) | |
5462 | return NULL; | |
5463 | } | |
90254a65 DN |
5464 | if (!get_page_unless_zero(page)) |
5465 | return NULL; | |
5466 | ||
5467 | return page; | |
5468 | } | |
5469 | ||
c733a828 | 5470 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 5471 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 5472 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 5473 | { |
90254a65 DN |
5474 | struct page *page = NULL; |
5475 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5476 | ||
1dfab5ab | 5477 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 5478 | return NULL; |
c733a828 JG |
5479 | |
5480 | /* | |
5481 | * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to | |
5482 | * a device and because they are not accessible by CPU they are store | |
5483 | * as special swap entry in the CPU page table. | |
5484 | */ | |
5485 | if (is_device_private_entry(ent)) { | |
5486 | page = device_private_entry_to_page(ent); | |
5487 | /* | |
5488 | * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have | |
5489 | * a refcount of 1 when free (unlike normal page) | |
5490 | */ | |
5491 | if (!page_ref_add_unless(page, 1, 1)) | |
5492 | return NULL; | |
5493 | return page; | |
5494 | } | |
5495 | ||
4b91355e KH |
5496 | /* |
5497 | * Because lookup_swap_cache() updates some statistics counter, | |
5498 | * we call find_get_page() with swapper_space directly. | |
5499 | */ | |
f6ab1f7f | 5500 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
2d1c4980 | 5501 | entry->val = ent.val; |
90254a65 DN |
5502 | |
5503 | return page; | |
5504 | } | |
4b91355e KH |
5505 | #else |
5506 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 5507 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
5508 | { |
5509 | return NULL; | |
5510 | } | |
5511 | #endif | |
90254a65 | 5512 | |
87946a72 DN |
5513 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5514 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5515 | { | |
5516 | struct page *page = NULL; | |
87946a72 DN |
5517 | struct address_space *mapping; |
5518 | pgoff_t pgoff; | |
5519 | ||
5520 | if (!vma->vm_file) /* anonymous vma */ | |
5521 | return NULL; | |
1dfab5ab | 5522 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
5523 | return NULL; |
5524 | ||
87946a72 | 5525 | mapping = vma->vm_file->f_mapping; |
0661a336 | 5526 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
5527 | |
5528 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
5529 | #ifdef CONFIG_SWAP |
5530 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
5531 | if (shmem_mapping(mapping)) { |
5532 | page = find_get_entry(mapping, pgoff); | |
3159f943 | 5533 | if (xa_is_value(page)) { |
139b6a6f | 5534 | swp_entry_t swp = radix_to_swp_entry(page); |
2d1c4980 | 5535 | *entry = swp; |
f6ab1f7f HY |
5536 | page = find_get_page(swap_address_space(swp), |
5537 | swp_offset(swp)); | |
139b6a6f JW |
5538 | } |
5539 | } else | |
5540 | page = find_get_page(mapping, pgoff); | |
5541 | #else | |
5542 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 5543 | #endif |
87946a72 DN |
5544 | return page; |
5545 | } | |
5546 | ||
b1b0deab CG |
5547 | /** |
5548 | * mem_cgroup_move_account - move account of the page | |
5549 | * @page: the page | |
25843c2b | 5550 | * @compound: charge the page as compound or small page |
b1b0deab CG |
5551 | * @from: mem_cgroup which the page is moved from. |
5552 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
5553 | * | |
3ac808fd | 5554 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
5555 | * |
5556 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
5557 | * from old cgroup. | |
5558 | */ | |
5559 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 5560 | bool compound, |
b1b0deab CG |
5561 | struct mem_cgroup *from, |
5562 | struct mem_cgroup *to) | |
5563 | { | |
ae8af438 KK |
5564 | struct lruvec *from_vec, *to_vec; |
5565 | struct pglist_data *pgdat; | |
f627c2f5 | 5566 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
b1b0deab CG |
5567 | int ret; |
5568 | ||
5569 | VM_BUG_ON(from == to); | |
5570 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 5571 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
5572 | |
5573 | /* | |
6a93ca8f | 5574 | * Prevent mem_cgroup_migrate() from looking at |
45637bab | 5575 | * page->mem_cgroup of its source page while we change it. |
b1b0deab | 5576 | */ |
f627c2f5 | 5577 | ret = -EBUSY; |
b1b0deab CG |
5578 | if (!trylock_page(page)) |
5579 | goto out; | |
5580 | ||
5581 | ret = -EINVAL; | |
5582 | if (page->mem_cgroup != from) | |
5583 | goto out_unlock; | |
5584 | ||
ae8af438 | 5585 | pgdat = page_pgdat(page); |
867e5e1d JW |
5586 | from_vec = mem_cgroup_lruvec(from, pgdat); |
5587 | to_vec = mem_cgroup_lruvec(to, pgdat); | |
ae8af438 | 5588 | |
abb242f5 | 5589 | lock_page_memcg(page); |
b1b0deab | 5590 | |
be5d0a74 JW |
5591 | if (PageAnon(page)) { |
5592 | if (page_mapped(page)) { | |
5593 | __mod_lruvec_state(from_vec, NR_ANON_MAPPED, -nr_pages); | |
5594 | __mod_lruvec_state(to_vec, NR_ANON_MAPPED, nr_pages); | |
468c3982 JW |
5595 | if (PageTransHuge(page)) { |
5596 | __mod_lruvec_state(from_vec, NR_ANON_THPS, | |
5597 | -nr_pages); | |
5598 | __mod_lruvec_state(to_vec, NR_ANON_THPS, | |
5599 | nr_pages); | |
5600 | } | |
5601 | ||
be5d0a74 JW |
5602 | } |
5603 | } else { | |
0d1c2072 JW |
5604 | __mod_lruvec_state(from_vec, NR_FILE_PAGES, -nr_pages); |
5605 | __mod_lruvec_state(to_vec, NR_FILE_PAGES, nr_pages); | |
5606 | ||
5607 | if (PageSwapBacked(page)) { | |
5608 | __mod_lruvec_state(from_vec, NR_SHMEM, -nr_pages); | |
5609 | __mod_lruvec_state(to_vec, NR_SHMEM, nr_pages); | |
5610 | } | |
5611 | ||
49e50d27 JW |
5612 | if (page_mapped(page)) { |
5613 | __mod_lruvec_state(from_vec, NR_FILE_MAPPED, -nr_pages); | |
5614 | __mod_lruvec_state(to_vec, NR_FILE_MAPPED, nr_pages); | |
5615 | } | |
b1b0deab | 5616 | |
49e50d27 JW |
5617 | if (PageDirty(page)) { |
5618 | struct address_space *mapping = page_mapping(page); | |
c4843a75 | 5619 | |
49e50d27 JW |
5620 | if (mapping_cap_account_dirty(mapping)) { |
5621 | __mod_lruvec_state(from_vec, NR_FILE_DIRTY, | |
5622 | -nr_pages); | |
5623 | __mod_lruvec_state(to_vec, NR_FILE_DIRTY, | |
5624 | nr_pages); | |
5625 | } | |
c4843a75 GT |
5626 | } |
5627 | } | |
5628 | ||
b1b0deab | 5629 | if (PageWriteback(page)) { |
ae8af438 KK |
5630 | __mod_lruvec_state(from_vec, NR_WRITEBACK, -nr_pages); |
5631 | __mod_lruvec_state(to_vec, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
5632 | } |
5633 | ||
5634 | /* | |
abb242f5 JW |
5635 | * All state has been migrated, let's switch to the new memcg. |
5636 | * | |
b1b0deab | 5637 | * It is safe to change page->mem_cgroup here because the page |
abb242f5 JW |
5638 | * is referenced, charged, isolated, and locked: we can't race |
5639 | * with (un)charging, migration, LRU putback, or anything else | |
5640 | * that would rely on a stable page->mem_cgroup. | |
5641 | * | |
5642 | * Note that lock_page_memcg is a memcg lock, not a page lock, | |
5643 | * to save space. As soon as we switch page->mem_cgroup to a | |
5644 | * new memcg that isn't locked, the above state can change | |
5645 | * concurrently again. Make sure we're truly done with it. | |
b1b0deab | 5646 | */ |
abb242f5 | 5647 | smp_mb(); |
b1b0deab | 5648 | |
1a3e1f40 JW |
5649 | css_get(&to->css); |
5650 | css_put(&from->css); | |
5651 | ||
5652 | page->mem_cgroup = to; | |
87eaceb3 | 5653 | |
abb242f5 | 5654 | __unlock_page_memcg(from); |
b1b0deab CG |
5655 | |
5656 | ret = 0; | |
5657 | ||
5658 | local_irq_disable(); | |
3fba69a5 | 5659 | mem_cgroup_charge_statistics(to, page, nr_pages); |
b1b0deab | 5660 | memcg_check_events(to, page); |
3fba69a5 | 5661 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
b1b0deab CG |
5662 | memcg_check_events(from, page); |
5663 | local_irq_enable(); | |
5664 | out_unlock: | |
5665 | unlock_page(page); | |
5666 | out: | |
5667 | return ret; | |
5668 | } | |
5669 | ||
7cf7806c LR |
5670 | /** |
5671 | * get_mctgt_type - get target type of moving charge | |
5672 | * @vma: the vma the pte to be checked belongs | |
5673 | * @addr: the address corresponding to the pte to be checked | |
5674 | * @ptent: the pte to be checked | |
5675 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
5676 | * | |
5677 | * Returns | |
5678 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5679 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5680 | * move charge. if @target is not NULL, the page is stored in target->page | |
5681 | * with extra refcnt got(Callers should handle it). | |
5682 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
5683 | * target for charge migration. if @target is not NULL, the entry is stored | |
5684 | * in target->ent. | |
25b2995a CH |
5685 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PRIVATE |
5686 | * (so ZONE_DEVICE page and thus not on the lru). | |
df6ad698 JG |
5687 | * For now we such page is charge like a regular page would be as for all |
5688 | * intent and purposes it is just special memory taking the place of a | |
5689 | * regular page. | |
c733a828 JG |
5690 | * |
5691 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
5692 | * |
5693 | * Called with pte lock held. | |
5694 | */ | |
5695 | ||
8d32ff84 | 5696 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5697 | unsigned long addr, pte_t ptent, union mc_target *target) |
5698 | { | |
5699 | struct page *page = NULL; | |
8d32ff84 | 5700 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5701 | swp_entry_t ent = { .val = 0 }; |
5702 | ||
5703 | if (pte_present(ptent)) | |
5704 | page = mc_handle_present_pte(vma, addr, ptent); | |
5705 | else if (is_swap_pte(ptent)) | |
48406ef8 | 5706 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 5707 | else if (pte_none(ptent)) |
87946a72 | 5708 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
5709 | |
5710 | if (!page && !ent.val) | |
8d32ff84 | 5711 | return ret; |
02491447 | 5712 | if (page) { |
02491447 | 5713 | /* |
0a31bc97 | 5714 | * Do only loose check w/o serialization. |
1306a85a | 5715 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 5716 | * not under LRU exclusion. |
02491447 | 5717 | */ |
1306a85a | 5718 | if (page->mem_cgroup == mc.from) { |
02491447 | 5719 | ret = MC_TARGET_PAGE; |
25b2995a | 5720 | if (is_device_private_page(page)) |
c733a828 | 5721 | ret = MC_TARGET_DEVICE; |
02491447 DN |
5722 | if (target) |
5723 | target->page = page; | |
5724 | } | |
5725 | if (!ret || !target) | |
5726 | put_page(page); | |
5727 | } | |
3e14a57b HY |
5728 | /* |
5729 | * There is a swap entry and a page doesn't exist or isn't charged. | |
5730 | * But we cannot move a tail-page in a THP. | |
5731 | */ | |
5732 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 5733 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5734 | ret = MC_TARGET_SWAP; |
5735 | if (target) | |
5736 | target->ent = ent; | |
4ffef5fe | 5737 | } |
4ffef5fe DN |
5738 | return ret; |
5739 | } | |
5740 | ||
12724850 NH |
5741 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5742 | /* | |
d6810d73 HY |
5743 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
5744 | * not support them for now. | |
12724850 NH |
5745 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
5746 | */ | |
5747 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5748 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5749 | { | |
5750 | struct page *page = NULL; | |
12724850 NH |
5751 | enum mc_target_type ret = MC_TARGET_NONE; |
5752 | ||
84c3fc4e ZY |
5753 | if (unlikely(is_swap_pmd(pmd))) { |
5754 | VM_BUG_ON(thp_migration_supported() && | |
5755 | !is_pmd_migration_entry(pmd)); | |
5756 | return ret; | |
5757 | } | |
12724850 | 5758 | page = pmd_page(pmd); |
309381fe | 5759 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 5760 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 5761 | return ret; |
1306a85a | 5762 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
5763 | ret = MC_TARGET_PAGE; |
5764 | if (target) { | |
5765 | get_page(page); | |
5766 | target->page = page; | |
5767 | } | |
5768 | } | |
5769 | return ret; | |
5770 | } | |
5771 | #else | |
5772 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5773 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5774 | { | |
5775 | return MC_TARGET_NONE; | |
5776 | } | |
5777 | #endif | |
5778 | ||
4ffef5fe DN |
5779 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5780 | unsigned long addr, unsigned long end, | |
5781 | struct mm_walk *walk) | |
5782 | { | |
26bcd64a | 5783 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5784 | pte_t *pte; |
5785 | spinlock_t *ptl; | |
5786 | ||
b6ec57f4 KS |
5787 | ptl = pmd_trans_huge_lock(pmd, vma); |
5788 | if (ptl) { | |
c733a828 JG |
5789 | /* |
5790 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
25b2995a CH |
5791 | * support transparent huge page with MEMORY_DEVICE_PRIVATE but |
5792 | * this might change. | |
c733a828 | 5793 | */ |
12724850 NH |
5794 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5795 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5796 | spin_unlock(ptl); |
1a5a9906 | 5797 | return 0; |
12724850 | 5798 | } |
03319327 | 5799 | |
45f83cef AA |
5800 | if (pmd_trans_unstable(pmd)) |
5801 | return 0; | |
4ffef5fe DN |
5802 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5803 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5804 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5805 | mc.precharge++; /* increment precharge temporarily */ |
5806 | pte_unmap_unlock(pte - 1, ptl); | |
5807 | cond_resched(); | |
5808 | ||
7dc74be0 DN |
5809 | return 0; |
5810 | } | |
5811 | ||
7b86ac33 CH |
5812 | static const struct mm_walk_ops precharge_walk_ops = { |
5813 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5814 | }; | |
5815 | ||
4ffef5fe DN |
5816 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5817 | { | |
5818 | unsigned long precharge; | |
4ffef5fe | 5819 | |
d8ed45c5 | 5820 | mmap_read_lock(mm); |
7b86ac33 | 5821 | walk_page_range(mm, 0, mm->highest_vm_end, &precharge_walk_ops, NULL); |
d8ed45c5 | 5822 | mmap_read_unlock(mm); |
4ffef5fe DN |
5823 | |
5824 | precharge = mc.precharge; | |
5825 | mc.precharge = 0; | |
5826 | ||
5827 | return precharge; | |
5828 | } | |
5829 | ||
4ffef5fe DN |
5830 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5831 | { | |
dfe076b0 DN |
5832 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5833 | ||
5834 | VM_BUG_ON(mc.moving_task); | |
5835 | mc.moving_task = current; | |
5836 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5837 | } |
5838 | ||
dfe076b0 DN |
5839 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5840 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5841 | { |
2bd9bb20 KH |
5842 | struct mem_cgroup *from = mc.from; |
5843 | struct mem_cgroup *to = mc.to; | |
5844 | ||
4ffef5fe | 5845 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5846 | if (mc.precharge) { |
00501b53 | 5847 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5848 | mc.precharge = 0; |
5849 | } | |
5850 | /* | |
5851 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5852 | * we must uncharge here. | |
5853 | */ | |
5854 | if (mc.moved_charge) { | |
00501b53 | 5855 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5856 | mc.moved_charge = 0; |
4ffef5fe | 5857 | } |
483c30b5 DN |
5858 | /* we must fixup refcnts and charges */ |
5859 | if (mc.moved_swap) { | |
483c30b5 | 5860 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5861 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5862 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5863 | |
615d66c3 VD |
5864 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
5865 | ||
05b84301 | 5866 | /* |
3e32cb2e JW |
5867 | * we charged both to->memory and to->memsw, so we |
5868 | * should uncharge to->memory. | |
05b84301 | 5869 | */ |
ce00a967 | 5870 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5871 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5872 | ||
483c30b5 DN |
5873 | mc.moved_swap = 0; |
5874 | } | |
dfe076b0 DN |
5875 | memcg_oom_recover(from); |
5876 | memcg_oom_recover(to); | |
5877 | wake_up_all(&mc.waitq); | |
5878 | } | |
5879 | ||
5880 | static void mem_cgroup_clear_mc(void) | |
5881 | { | |
264a0ae1 TH |
5882 | struct mm_struct *mm = mc.mm; |
5883 | ||
dfe076b0 DN |
5884 | /* |
5885 | * we must clear moving_task before waking up waiters at the end of | |
5886 | * task migration. | |
5887 | */ | |
5888 | mc.moving_task = NULL; | |
5889 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5890 | spin_lock(&mc.lock); |
4ffef5fe DN |
5891 | mc.from = NULL; |
5892 | mc.to = NULL; | |
264a0ae1 | 5893 | mc.mm = NULL; |
2bd9bb20 | 5894 | spin_unlock(&mc.lock); |
264a0ae1 TH |
5895 | |
5896 | mmput(mm); | |
4ffef5fe DN |
5897 | } |
5898 | ||
1f7dd3e5 | 5899 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5900 | { |
1f7dd3e5 | 5901 | struct cgroup_subsys_state *css; |
eed67d75 | 5902 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 5903 | struct mem_cgroup *from; |
4530eddb | 5904 | struct task_struct *leader, *p; |
9f2115f9 | 5905 | struct mm_struct *mm; |
1dfab5ab | 5906 | unsigned long move_flags; |
9f2115f9 | 5907 | int ret = 0; |
7dc74be0 | 5908 | |
1f7dd3e5 TH |
5909 | /* charge immigration isn't supported on the default hierarchy */ |
5910 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
5911 | return 0; |
5912 | ||
4530eddb TH |
5913 | /* |
5914 | * Multi-process migrations only happen on the default hierarchy | |
5915 | * where charge immigration is not used. Perform charge | |
5916 | * immigration if @tset contains a leader and whine if there are | |
5917 | * multiple. | |
5918 | */ | |
5919 | p = NULL; | |
1f7dd3e5 | 5920 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
5921 | WARN_ON_ONCE(p); |
5922 | p = leader; | |
1f7dd3e5 | 5923 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
5924 | } |
5925 | if (!p) | |
5926 | return 0; | |
5927 | ||
1f7dd3e5 TH |
5928 | /* |
5929 | * We are now commited to this value whatever it is. Changes in this | |
5930 | * tunable will only affect upcoming migrations, not the current one. | |
5931 | * So we need to save it, and keep it going. | |
5932 | */ | |
5933 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
5934 | if (!move_flags) | |
5935 | return 0; | |
5936 | ||
9f2115f9 TH |
5937 | from = mem_cgroup_from_task(p); |
5938 | ||
5939 | VM_BUG_ON(from == memcg); | |
5940 | ||
5941 | mm = get_task_mm(p); | |
5942 | if (!mm) | |
5943 | return 0; | |
5944 | /* We move charges only when we move a owner of the mm */ | |
5945 | if (mm->owner == p) { | |
5946 | VM_BUG_ON(mc.from); | |
5947 | VM_BUG_ON(mc.to); | |
5948 | VM_BUG_ON(mc.precharge); | |
5949 | VM_BUG_ON(mc.moved_charge); | |
5950 | VM_BUG_ON(mc.moved_swap); | |
5951 | ||
5952 | spin_lock(&mc.lock); | |
264a0ae1 | 5953 | mc.mm = mm; |
9f2115f9 TH |
5954 | mc.from = from; |
5955 | mc.to = memcg; | |
5956 | mc.flags = move_flags; | |
5957 | spin_unlock(&mc.lock); | |
5958 | /* We set mc.moving_task later */ | |
5959 | ||
5960 | ret = mem_cgroup_precharge_mc(mm); | |
5961 | if (ret) | |
5962 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
5963 | } else { |
5964 | mmput(mm); | |
7dc74be0 DN |
5965 | } |
5966 | return ret; | |
5967 | } | |
5968 | ||
1f7dd3e5 | 5969 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5970 | { |
4e2f245d JW |
5971 | if (mc.to) |
5972 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5973 | } |
5974 | ||
4ffef5fe DN |
5975 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5976 | unsigned long addr, unsigned long end, | |
5977 | struct mm_walk *walk) | |
7dc74be0 | 5978 | { |
4ffef5fe | 5979 | int ret = 0; |
26bcd64a | 5980 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5981 | pte_t *pte; |
5982 | spinlock_t *ptl; | |
12724850 NH |
5983 | enum mc_target_type target_type; |
5984 | union mc_target target; | |
5985 | struct page *page; | |
4ffef5fe | 5986 | |
b6ec57f4 KS |
5987 | ptl = pmd_trans_huge_lock(pmd, vma); |
5988 | if (ptl) { | |
62ade86a | 5989 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5990 | spin_unlock(ptl); |
12724850 NH |
5991 | return 0; |
5992 | } | |
5993 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5994 | if (target_type == MC_TARGET_PAGE) { | |
5995 | page = target.page; | |
5996 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 5997 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 5998 | mc.from, mc.to)) { |
12724850 NH |
5999 | mc.precharge -= HPAGE_PMD_NR; |
6000 | mc.moved_charge += HPAGE_PMD_NR; | |
6001 | } | |
6002 | putback_lru_page(page); | |
6003 | } | |
6004 | put_page(page); | |
c733a828 JG |
6005 | } else if (target_type == MC_TARGET_DEVICE) { |
6006 | page = target.page; | |
6007 | if (!mem_cgroup_move_account(page, true, | |
6008 | mc.from, mc.to)) { | |
6009 | mc.precharge -= HPAGE_PMD_NR; | |
6010 | mc.moved_charge += HPAGE_PMD_NR; | |
6011 | } | |
6012 | put_page(page); | |
12724850 | 6013 | } |
bf929152 | 6014 | spin_unlock(ptl); |
1a5a9906 | 6015 | return 0; |
12724850 NH |
6016 | } |
6017 | ||
45f83cef AA |
6018 | if (pmd_trans_unstable(pmd)) |
6019 | return 0; | |
4ffef5fe DN |
6020 | retry: |
6021 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
6022 | for (; addr != end; addr += PAGE_SIZE) { | |
6023 | pte_t ptent = *(pte++); | |
c733a828 | 6024 | bool device = false; |
02491447 | 6025 | swp_entry_t ent; |
4ffef5fe DN |
6026 | |
6027 | if (!mc.precharge) | |
6028 | break; | |
6029 | ||
8d32ff84 | 6030 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
6031 | case MC_TARGET_DEVICE: |
6032 | device = true; | |
e4a9bc58 | 6033 | fallthrough; |
4ffef5fe DN |
6034 | case MC_TARGET_PAGE: |
6035 | page = target.page; | |
53f9263b KS |
6036 | /* |
6037 | * We can have a part of the split pmd here. Moving it | |
6038 | * can be done but it would be too convoluted so simply | |
6039 | * ignore such a partial THP and keep it in original | |
6040 | * memcg. There should be somebody mapping the head. | |
6041 | */ | |
6042 | if (PageTransCompound(page)) | |
6043 | goto put; | |
c733a828 | 6044 | if (!device && isolate_lru_page(page)) |
4ffef5fe | 6045 | goto put; |
f627c2f5 KS |
6046 | if (!mem_cgroup_move_account(page, false, |
6047 | mc.from, mc.to)) { | |
4ffef5fe | 6048 | mc.precharge--; |
854ffa8d DN |
6049 | /* we uncharge from mc.from later. */ |
6050 | mc.moved_charge++; | |
4ffef5fe | 6051 | } |
c733a828 JG |
6052 | if (!device) |
6053 | putback_lru_page(page); | |
8d32ff84 | 6054 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
6055 | put_page(page); |
6056 | break; | |
02491447 DN |
6057 | case MC_TARGET_SWAP: |
6058 | ent = target.ent; | |
e91cbb42 | 6059 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6060 | mc.precharge--; |
8d22a935 HD |
6061 | mem_cgroup_id_get_many(mc.to, 1); |
6062 | /* we fixup other refcnts and charges later. */ | |
483c30b5 DN |
6063 | mc.moved_swap++; |
6064 | } | |
02491447 | 6065 | break; |
4ffef5fe DN |
6066 | default: |
6067 | break; | |
6068 | } | |
6069 | } | |
6070 | pte_unmap_unlock(pte - 1, ptl); | |
6071 | cond_resched(); | |
6072 | ||
6073 | if (addr != end) { | |
6074 | /* | |
6075 | * We have consumed all precharges we got in can_attach(). | |
6076 | * We try charge one by one, but don't do any additional | |
6077 | * charges to mc.to if we have failed in charge once in attach() | |
6078 | * phase. | |
6079 | */ | |
854ffa8d | 6080 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6081 | if (!ret) |
6082 | goto retry; | |
6083 | } | |
6084 | ||
6085 | return ret; | |
6086 | } | |
6087 | ||
7b86ac33 CH |
6088 | static const struct mm_walk_ops charge_walk_ops = { |
6089 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6090 | }; | |
6091 | ||
264a0ae1 | 6092 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 6093 | { |
4ffef5fe | 6094 | lru_add_drain_all(); |
312722cb | 6095 | /* |
81f8c3a4 JW |
6096 | * Signal lock_page_memcg() to take the memcg's move_lock |
6097 | * while we're moving its pages to another memcg. Then wait | |
6098 | * for already started RCU-only updates to finish. | |
312722cb JW |
6099 | */ |
6100 | atomic_inc(&mc.from->moving_account); | |
6101 | synchronize_rcu(); | |
dfe076b0 | 6102 | retry: |
d8ed45c5 | 6103 | if (unlikely(!mmap_read_trylock(mc.mm))) { |
dfe076b0 | 6104 | /* |
c1e8d7c6 | 6105 | * Someone who are holding the mmap_lock might be waiting in |
dfe076b0 DN |
6106 | * waitq. So we cancel all extra charges, wake up all waiters, |
6107 | * and retry. Because we cancel precharges, we might not be able | |
6108 | * to move enough charges, but moving charge is a best-effort | |
6109 | * feature anyway, so it wouldn't be a big problem. | |
6110 | */ | |
6111 | __mem_cgroup_clear_mc(); | |
6112 | cond_resched(); | |
6113 | goto retry; | |
6114 | } | |
26bcd64a NH |
6115 | /* |
6116 | * When we have consumed all precharges and failed in doing | |
6117 | * additional charge, the page walk just aborts. | |
6118 | */ | |
7b86ac33 CH |
6119 | walk_page_range(mc.mm, 0, mc.mm->highest_vm_end, &charge_walk_ops, |
6120 | NULL); | |
0247f3f4 | 6121 | |
d8ed45c5 | 6122 | mmap_read_unlock(mc.mm); |
312722cb | 6123 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
6124 | } |
6125 | ||
264a0ae1 | 6126 | static void mem_cgroup_move_task(void) |
67e465a7 | 6127 | { |
264a0ae1 TH |
6128 | if (mc.to) { |
6129 | mem_cgroup_move_charge(); | |
a433658c | 6130 | mem_cgroup_clear_mc(); |
264a0ae1 | 6131 | } |
67e465a7 | 6132 | } |
5cfb80a7 | 6133 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 6134 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6135 | { |
6136 | return 0; | |
6137 | } | |
1f7dd3e5 | 6138 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6139 | { |
6140 | } | |
264a0ae1 | 6141 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
6142 | { |
6143 | } | |
6144 | #endif | |
67e465a7 | 6145 | |
f00baae7 TH |
6146 | /* |
6147 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
6148 | * to verify whether we're attached to the default hierarchy on each mount |
6149 | * attempt. | |
f00baae7 | 6150 | */ |
eb95419b | 6151 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
6152 | { |
6153 | /* | |
aa6ec29b | 6154 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
6155 | * guarantees that @root doesn't have any children, so turning it |
6156 | * on for the root memcg is enough. | |
6157 | */ | |
9e10a130 | 6158 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
6159 | root_mem_cgroup->use_hierarchy = true; |
6160 | else | |
6161 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
6162 | } |
6163 | ||
677dc973 CD |
6164 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
6165 | { | |
6166 | if (value == PAGE_COUNTER_MAX) | |
6167 | seq_puts(m, "max\n"); | |
6168 | else | |
6169 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); | |
6170 | ||
6171 | return 0; | |
6172 | } | |
6173 | ||
241994ed JW |
6174 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
6175 | struct cftype *cft) | |
6176 | { | |
f5fc3c5d JW |
6177 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
6178 | ||
6179 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
6180 | } |
6181 | ||
bf8d5d52 RG |
6182 | static int memory_min_show(struct seq_file *m, void *v) |
6183 | { | |
677dc973 CD |
6184 | return seq_puts_memcg_tunable(m, |
6185 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); | |
bf8d5d52 RG |
6186 | } |
6187 | ||
6188 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
6189 | char *buf, size_t nbytes, loff_t off) | |
6190 | { | |
6191 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6192 | unsigned long min; | |
6193 | int err; | |
6194 | ||
6195 | buf = strstrip(buf); | |
6196 | err = page_counter_memparse(buf, "max", &min); | |
6197 | if (err) | |
6198 | return err; | |
6199 | ||
6200 | page_counter_set_min(&memcg->memory, min); | |
6201 | ||
6202 | return nbytes; | |
6203 | } | |
6204 | ||
241994ed JW |
6205 | static int memory_low_show(struct seq_file *m, void *v) |
6206 | { | |
677dc973 CD |
6207 | return seq_puts_memcg_tunable(m, |
6208 | READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); | |
241994ed JW |
6209 | } |
6210 | ||
6211 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
6212 | char *buf, size_t nbytes, loff_t off) | |
6213 | { | |
6214 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6215 | unsigned long low; | |
6216 | int err; | |
6217 | ||
6218 | buf = strstrip(buf); | |
d2973697 | 6219 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
6220 | if (err) |
6221 | return err; | |
6222 | ||
23067153 | 6223 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
6224 | |
6225 | return nbytes; | |
6226 | } | |
6227 | ||
6228 | static int memory_high_show(struct seq_file *m, void *v) | |
6229 | { | |
d1663a90 JK |
6230 | return seq_puts_memcg_tunable(m, |
6231 | READ_ONCE(mem_cgroup_from_seq(m)->memory.high)); | |
241994ed JW |
6232 | } |
6233 | ||
6234 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
6235 | char *buf, size_t nbytes, loff_t off) | |
6236 | { | |
6237 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
8c8c383c JW |
6238 | unsigned int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6239 | bool drained = false; | |
241994ed JW |
6240 | unsigned long high; |
6241 | int err; | |
6242 | ||
6243 | buf = strstrip(buf); | |
d2973697 | 6244 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
6245 | if (err) |
6246 | return err; | |
6247 | ||
8c8c383c JW |
6248 | for (;;) { |
6249 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6250 | unsigned long reclaimed; | |
6251 | ||
6252 | if (nr_pages <= high) | |
6253 | break; | |
6254 | ||
6255 | if (signal_pending(current)) | |
6256 | break; | |
6257 | ||
6258 | if (!drained) { | |
6259 | drain_all_stock(memcg); | |
6260 | drained = true; | |
6261 | continue; | |
6262 | } | |
6263 | ||
6264 | reclaimed = try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
6265 | GFP_KERNEL, true); | |
6266 | ||
6267 | if (!reclaimed && !nr_retries--) | |
6268 | break; | |
6269 | } | |
588083bb | 6270 | |
536d3bf2 RG |
6271 | page_counter_set_high(&memcg->memory, high); |
6272 | ||
241994ed JW |
6273 | return nbytes; |
6274 | } | |
6275 | ||
6276 | static int memory_max_show(struct seq_file *m, void *v) | |
6277 | { | |
677dc973 CD |
6278 | return seq_puts_memcg_tunable(m, |
6279 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); | |
241994ed JW |
6280 | } |
6281 | ||
6282 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
6283 | char *buf, size_t nbytes, loff_t off) | |
6284 | { | |
6285 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
b6e6edcf JW |
6286 | unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES; |
6287 | bool drained = false; | |
241994ed JW |
6288 | unsigned long max; |
6289 | int err; | |
6290 | ||
6291 | buf = strstrip(buf); | |
d2973697 | 6292 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
6293 | if (err) |
6294 | return err; | |
6295 | ||
bbec2e15 | 6296 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
6297 | |
6298 | for (;;) { | |
6299 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6300 | ||
6301 | if (nr_pages <= max) | |
6302 | break; | |
6303 | ||
7249c9f0 | 6304 | if (signal_pending(current)) |
b6e6edcf | 6305 | break; |
b6e6edcf JW |
6306 | |
6307 | if (!drained) { | |
6308 | drain_all_stock(memcg); | |
6309 | drained = true; | |
6310 | continue; | |
6311 | } | |
6312 | ||
6313 | if (nr_reclaims) { | |
6314 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
6315 | GFP_KERNEL, true)) | |
6316 | nr_reclaims--; | |
6317 | continue; | |
6318 | } | |
6319 | ||
e27be240 | 6320 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
6321 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
6322 | break; | |
6323 | } | |
241994ed | 6324 | |
2529bb3a | 6325 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
6326 | return nbytes; |
6327 | } | |
6328 | ||
1e577f97 SB |
6329 | static void __memory_events_show(struct seq_file *m, atomic_long_t *events) |
6330 | { | |
6331 | seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW])); | |
6332 | seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH])); | |
6333 | seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX])); | |
6334 | seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM])); | |
6335 | seq_printf(m, "oom_kill %lu\n", | |
6336 | atomic_long_read(&events[MEMCG_OOM_KILL])); | |
6337 | } | |
6338 | ||
241994ed JW |
6339 | static int memory_events_show(struct seq_file *m, void *v) |
6340 | { | |
aa9694bb | 6341 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
241994ed | 6342 | |
1e577f97 SB |
6343 | __memory_events_show(m, memcg->memory_events); |
6344 | return 0; | |
6345 | } | |
6346 | ||
6347 | static int memory_events_local_show(struct seq_file *m, void *v) | |
6348 | { | |
6349 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
241994ed | 6350 | |
1e577f97 | 6351 | __memory_events_show(m, memcg->memory_events_local); |
241994ed JW |
6352 | return 0; |
6353 | } | |
6354 | ||
587d9f72 JW |
6355 | static int memory_stat_show(struct seq_file *m, void *v) |
6356 | { | |
aa9694bb | 6357 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
c8713d0b | 6358 | char *buf; |
1ff9e6e1 | 6359 | |
c8713d0b JW |
6360 | buf = memory_stat_format(memcg); |
6361 | if (!buf) | |
6362 | return -ENOMEM; | |
6363 | seq_puts(m, buf); | |
6364 | kfree(buf); | |
587d9f72 JW |
6365 | return 0; |
6366 | } | |
6367 | ||
3d8b38eb RG |
6368 | static int memory_oom_group_show(struct seq_file *m, void *v) |
6369 | { | |
aa9694bb | 6370 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3d8b38eb RG |
6371 | |
6372 | seq_printf(m, "%d\n", memcg->oom_group); | |
6373 | ||
6374 | return 0; | |
6375 | } | |
6376 | ||
6377 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
6378 | char *buf, size_t nbytes, loff_t off) | |
6379 | { | |
6380 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6381 | int ret, oom_group; | |
6382 | ||
6383 | buf = strstrip(buf); | |
6384 | if (!buf) | |
6385 | return -EINVAL; | |
6386 | ||
6387 | ret = kstrtoint(buf, 0, &oom_group); | |
6388 | if (ret) | |
6389 | return ret; | |
6390 | ||
6391 | if (oom_group != 0 && oom_group != 1) | |
6392 | return -EINVAL; | |
6393 | ||
6394 | memcg->oom_group = oom_group; | |
6395 | ||
6396 | return nbytes; | |
6397 | } | |
6398 | ||
241994ed JW |
6399 | static struct cftype memory_files[] = { |
6400 | { | |
6401 | .name = "current", | |
f5fc3c5d | 6402 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
6403 | .read_u64 = memory_current_read, |
6404 | }, | |
bf8d5d52 RG |
6405 | { |
6406 | .name = "min", | |
6407 | .flags = CFTYPE_NOT_ON_ROOT, | |
6408 | .seq_show = memory_min_show, | |
6409 | .write = memory_min_write, | |
6410 | }, | |
241994ed JW |
6411 | { |
6412 | .name = "low", | |
6413 | .flags = CFTYPE_NOT_ON_ROOT, | |
6414 | .seq_show = memory_low_show, | |
6415 | .write = memory_low_write, | |
6416 | }, | |
6417 | { | |
6418 | .name = "high", | |
6419 | .flags = CFTYPE_NOT_ON_ROOT, | |
6420 | .seq_show = memory_high_show, | |
6421 | .write = memory_high_write, | |
6422 | }, | |
6423 | { | |
6424 | .name = "max", | |
6425 | .flags = CFTYPE_NOT_ON_ROOT, | |
6426 | .seq_show = memory_max_show, | |
6427 | .write = memory_max_write, | |
6428 | }, | |
6429 | { | |
6430 | .name = "events", | |
6431 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 6432 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
6433 | .seq_show = memory_events_show, |
6434 | }, | |
1e577f97 SB |
6435 | { |
6436 | .name = "events.local", | |
6437 | .flags = CFTYPE_NOT_ON_ROOT, | |
6438 | .file_offset = offsetof(struct mem_cgroup, events_local_file), | |
6439 | .seq_show = memory_events_local_show, | |
6440 | }, | |
587d9f72 JW |
6441 | { |
6442 | .name = "stat", | |
587d9f72 JW |
6443 | .seq_show = memory_stat_show, |
6444 | }, | |
3d8b38eb RG |
6445 | { |
6446 | .name = "oom.group", | |
6447 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
6448 | .seq_show = memory_oom_group_show, | |
6449 | .write = memory_oom_group_write, | |
6450 | }, | |
241994ed JW |
6451 | { } /* terminate */ |
6452 | }; | |
6453 | ||
073219e9 | 6454 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 6455 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6456 | .css_online = mem_cgroup_css_online, |
92fb9748 | 6457 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 6458 | .css_released = mem_cgroup_css_released, |
92fb9748 | 6459 | .css_free = mem_cgroup_css_free, |
1ced953b | 6460 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
6461 | .can_attach = mem_cgroup_can_attach, |
6462 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 6463 | .post_attach = mem_cgroup_move_task, |
f00baae7 | 6464 | .bind = mem_cgroup_bind, |
241994ed JW |
6465 | .dfl_cftypes = memory_files, |
6466 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 6467 | .early_init = 0, |
8cdea7c0 | 6468 | }; |
c077719b | 6469 | |
bc50bcc6 JW |
6470 | /* |
6471 | * This function calculates an individual cgroup's effective | |
6472 | * protection which is derived from its own memory.min/low, its | |
6473 | * parent's and siblings' settings, as well as the actual memory | |
6474 | * distribution in the tree. | |
6475 | * | |
6476 | * The following rules apply to the effective protection values: | |
6477 | * | |
6478 | * 1. At the first level of reclaim, effective protection is equal to | |
6479 | * the declared protection in memory.min and memory.low. | |
6480 | * | |
6481 | * 2. To enable safe delegation of the protection configuration, at | |
6482 | * subsequent levels the effective protection is capped to the | |
6483 | * parent's effective protection. | |
6484 | * | |
6485 | * 3. To make complex and dynamic subtrees easier to configure, the | |
6486 | * user is allowed to overcommit the declared protection at a given | |
6487 | * level. If that is the case, the parent's effective protection is | |
6488 | * distributed to the children in proportion to how much protection | |
6489 | * they have declared and how much of it they are utilizing. | |
6490 | * | |
6491 | * This makes distribution proportional, but also work-conserving: | |
6492 | * if one cgroup claims much more protection than it uses memory, | |
6493 | * the unused remainder is available to its siblings. | |
6494 | * | |
6495 | * 4. Conversely, when the declared protection is undercommitted at a | |
6496 | * given level, the distribution of the larger parental protection | |
6497 | * budget is NOT proportional. A cgroup's protection from a sibling | |
6498 | * is capped to its own memory.min/low setting. | |
6499 | * | |
8a931f80 JW |
6500 | * 5. However, to allow protecting recursive subtrees from each other |
6501 | * without having to declare each individual cgroup's fixed share | |
6502 | * of the ancestor's claim to protection, any unutilized - | |
6503 | * "floating" - protection from up the tree is distributed in | |
6504 | * proportion to each cgroup's *usage*. This makes the protection | |
6505 | * neutral wrt sibling cgroups and lets them compete freely over | |
6506 | * the shared parental protection budget, but it protects the | |
6507 | * subtree as a whole from neighboring subtrees. | |
6508 | * | |
6509 | * Note that 4. and 5. are not in conflict: 4. is about protecting | |
6510 | * against immediate siblings whereas 5. is about protecting against | |
6511 | * neighboring subtrees. | |
bc50bcc6 JW |
6512 | */ |
6513 | static unsigned long effective_protection(unsigned long usage, | |
8a931f80 | 6514 | unsigned long parent_usage, |
bc50bcc6 JW |
6515 | unsigned long setting, |
6516 | unsigned long parent_effective, | |
6517 | unsigned long siblings_protected) | |
6518 | { | |
6519 | unsigned long protected; | |
8a931f80 | 6520 | unsigned long ep; |
bc50bcc6 JW |
6521 | |
6522 | protected = min(usage, setting); | |
6523 | /* | |
6524 | * If all cgroups at this level combined claim and use more | |
6525 | * protection then what the parent affords them, distribute | |
6526 | * shares in proportion to utilization. | |
6527 | * | |
6528 | * We are using actual utilization rather than the statically | |
6529 | * claimed protection in order to be work-conserving: claimed | |
6530 | * but unused protection is available to siblings that would | |
6531 | * otherwise get a smaller chunk than what they claimed. | |
6532 | */ | |
6533 | if (siblings_protected > parent_effective) | |
6534 | return protected * parent_effective / siblings_protected; | |
6535 | ||
6536 | /* | |
6537 | * Ok, utilized protection of all children is within what the | |
6538 | * parent affords them, so we know whatever this child claims | |
6539 | * and utilizes is effectively protected. | |
6540 | * | |
6541 | * If there is unprotected usage beyond this value, reclaim | |
6542 | * will apply pressure in proportion to that amount. | |
6543 | * | |
6544 | * If there is unutilized protection, the cgroup will be fully | |
6545 | * shielded from reclaim, but we do return a smaller value for | |
6546 | * protection than what the group could enjoy in theory. This | |
6547 | * is okay. With the overcommit distribution above, effective | |
6548 | * protection is always dependent on how memory is actually | |
6549 | * consumed among the siblings anyway. | |
6550 | */ | |
8a931f80 JW |
6551 | ep = protected; |
6552 | ||
6553 | /* | |
6554 | * If the children aren't claiming (all of) the protection | |
6555 | * afforded to them by the parent, distribute the remainder in | |
6556 | * proportion to the (unprotected) memory of each cgroup. That | |
6557 | * way, cgroups that aren't explicitly prioritized wrt each | |
6558 | * other compete freely over the allowance, but they are | |
6559 | * collectively protected from neighboring trees. | |
6560 | * | |
6561 | * We're using unprotected memory for the weight so that if | |
6562 | * some cgroups DO claim explicit protection, we don't protect | |
6563 | * the same bytes twice. | |
cd324edc JW |
6564 | * |
6565 | * Check both usage and parent_usage against the respective | |
6566 | * protected values. One should imply the other, but they | |
6567 | * aren't read atomically - make sure the division is sane. | |
8a931f80 JW |
6568 | */ |
6569 | if (!(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)) | |
6570 | return ep; | |
cd324edc JW |
6571 | if (parent_effective > siblings_protected && |
6572 | parent_usage > siblings_protected && | |
6573 | usage > protected) { | |
8a931f80 JW |
6574 | unsigned long unclaimed; |
6575 | ||
6576 | unclaimed = parent_effective - siblings_protected; | |
6577 | unclaimed *= usage - protected; | |
6578 | unclaimed /= parent_usage - siblings_protected; | |
6579 | ||
6580 | ep += unclaimed; | |
6581 | } | |
6582 | ||
6583 | return ep; | |
bc50bcc6 JW |
6584 | } |
6585 | ||
241994ed | 6586 | /** |
bf8d5d52 | 6587 | * mem_cgroup_protected - check if memory consumption is in the normal range |
34c81057 | 6588 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
6589 | * @memcg: the memory cgroup to check |
6590 | * | |
23067153 RG |
6591 | * WARNING: This function is not stateless! It can only be used as part |
6592 | * of a top-down tree iteration, not for isolated queries. | |
34c81057 | 6593 | * |
bf8d5d52 RG |
6594 | * Returns one of the following: |
6595 | * MEMCG_PROT_NONE: cgroup memory is not protected | |
6596 | * MEMCG_PROT_LOW: cgroup memory is protected as long there is | |
6597 | * an unprotected supply of reclaimable memory from other cgroups. | |
6598 | * MEMCG_PROT_MIN: cgroup memory is protected | |
241994ed | 6599 | */ |
bf8d5d52 RG |
6600 | enum mem_cgroup_protection mem_cgroup_protected(struct mem_cgroup *root, |
6601 | struct mem_cgroup *memcg) | |
241994ed | 6602 | { |
8a931f80 | 6603 | unsigned long usage, parent_usage; |
23067153 RG |
6604 | struct mem_cgroup *parent; |
6605 | ||
241994ed | 6606 | if (mem_cgroup_disabled()) |
bf8d5d52 | 6607 | return MEMCG_PROT_NONE; |
241994ed | 6608 | |
34c81057 SC |
6609 | if (!root) |
6610 | root = root_mem_cgroup; | |
6611 | if (memcg == root) | |
bf8d5d52 | 6612 | return MEMCG_PROT_NONE; |
241994ed | 6613 | |
23067153 | 6614 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 RG |
6615 | if (!usage) |
6616 | return MEMCG_PROT_NONE; | |
6617 | ||
bf8d5d52 | 6618 | parent = parent_mem_cgroup(memcg); |
df2a4196 RG |
6619 | /* No parent means a non-hierarchical mode on v1 memcg */ |
6620 | if (!parent) | |
6621 | return MEMCG_PROT_NONE; | |
6622 | ||
bc50bcc6 | 6623 | if (parent == root) { |
c3d53200 | 6624 | memcg->memory.emin = READ_ONCE(memcg->memory.min); |
03960e33 | 6625 | memcg->memory.elow = READ_ONCE(memcg->memory.low); |
bc50bcc6 | 6626 | goto out; |
bf8d5d52 RG |
6627 | } |
6628 | ||
8a931f80 JW |
6629 | parent_usage = page_counter_read(&parent->memory); |
6630 | ||
b3a7822e | 6631 | WRITE_ONCE(memcg->memory.emin, effective_protection(usage, parent_usage, |
c3d53200 CD |
6632 | READ_ONCE(memcg->memory.min), |
6633 | READ_ONCE(parent->memory.emin), | |
b3a7822e | 6634 | atomic_long_read(&parent->memory.children_min_usage))); |
23067153 | 6635 | |
b3a7822e | 6636 | WRITE_ONCE(memcg->memory.elow, effective_protection(usage, parent_usage, |
03960e33 CD |
6637 | READ_ONCE(memcg->memory.low), |
6638 | READ_ONCE(parent->memory.elow), | |
b3a7822e | 6639 | atomic_long_read(&parent->memory.children_low_usage))); |
23067153 | 6640 | |
bc50bcc6 JW |
6641 | out: |
6642 | if (usage <= memcg->memory.emin) | |
bf8d5d52 | 6643 | return MEMCG_PROT_MIN; |
bc50bcc6 | 6644 | else if (usage <= memcg->memory.elow) |
bf8d5d52 RG |
6645 | return MEMCG_PROT_LOW; |
6646 | else | |
6647 | return MEMCG_PROT_NONE; | |
241994ed JW |
6648 | } |
6649 | ||
00501b53 | 6650 | /** |
f0e45fb4 | 6651 | * mem_cgroup_charge - charge a newly allocated page to a cgroup |
00501b53 JW |
6652 | * @page: page to charge |
6653 | * @mm: mm context of the victim | |
6654 | * @gfp_mask: reclaim mode | |
00501b53 JW |
6655 | * |
6656 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
6657 | * pages according to @gfp_mask if necessary. | |
6658 | * | |
f0e45fb4 | 6659 | * Returns 0 on success. Otherwise, an error code is returned. |
00501b53 | 6660 | */ |
d9eb1ea2 | 6661 | int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask) |
00501b53 | 6662 | { |
3fba69a5 | 6663 | unsigned int nr_pages = hpage_nr_pages(page); |
00501b53 | 6664 | struct mem_cgroup *memcg = NULL; |
00501b53 JW |
6665 | int ret = 0; |
6666 | ||
6667 | if (mem_cgroup_disabled()) | |
6668 | goto out; | |
6669 | ||
6670 | if (PageSwapCache(page)) { | |
2d1c4980 JW |
6671 | swp_entry_t ent = { .val = page_private(page), }; |
6672 | unsigned short id; | |
6673 | ||
00501b53 JW |
6674 | /* |
6675 | * Every swap fault against a single page tries to charge the | |
6676 | * page, bail as early as possible. shmem_unuse() encounters | |
eccb52e7 JW |
6677 | * already charged pages, too. page->mem_cgroup is protected |
6678 | * by the page lock, which serializes swap cache removal, which | |
00501b53 JW |
6679 | * in turn serializes uncharging. |
6680 | */ | |
e993d905 | 6681 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
abe2895b | 6682 | if (compound_head(page)->mem_cgroup) |
00501b53 | 6683 | goto out; |
e993d905 | 6684 | |
2d1c4980 JW |
6685 | id = lookup_swap_cgroup_id(ent); |
6686 | rcu_read_lock(); | |
6687 | memcg = mem_cgroup_from_id(id); | |
6688 | if (memcg && !css_tryget_online(&memcg->css)) | |
6689 | memcg = NULL; | |
6690 | rcu_read_unlock(); | |
00501b53 JW |
6691 | } |
6692 | ||
00501b53 JW |
6693 | if (!memcg) |
6694 | memcg = get_mem_cgroup_from_mm(mm); | |
6695 | ||
6696 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
f0e45fb4 JW |
6697 | if (ret) |
6698 | goto out_put; | |
00501b53 | 6699 | |
1a3e1f40 | 6700 | css_get(&memcg->css); |
d9eb1ea2 | 6701 | commit_charge(page, memcg); |
6abb5a86 | 6702 | |
6abb5a86 | 6703 | local_irq_disable(); |
3fba69a5 | 6704 | mem_cgroup_charge_statistics(memcg, page, nr_pages); |
6abb5a86 JW |
6705 | memcg_check_events(memcg, page); |
6706 | local_irq_enable(); | |
00501b53 | 6707 | |
2d1c4980 | 6708 | if (PageSwapCache(page)) { |
00501b53 JW |
6709 | swp_entry_t entry = { .val = page_private(page) }; |
6710 | /* | |
6711 | * The swap entry might not get freed for a long time, | |
6712 | * let's not wait for it. The page already received a | |
6713 | * memory+swap charge, drop the swap entry duplicate. | |
6714 | */ | |
38d8b4e6 | 6715 | mem_cgroup_uncharge_swap(entry, nr_pages); |
00501b53 | 6716 | } |
00501b53 | 6717 | |
f0e45fb4 JW |
6718 | out_put: |
6719 | css_put(&memcg->css); | |
6720 | out: | |
6721 | return ret; | |
3fea5a49 JW |
6722 | } |
6723 | ||
a9d5adee JG |
6724 | struct uncharge_gather { |
6725 | struct mem_cgroup *memcg; | |
9f762dbe | 6726 | unsigned long nr_pages; |
a9d5adee | 6727 | unsigned long pgpgout; |
a9d5adee | 6728 | unsigned long nr_kmem; |
a9d5adee JG |
6729 | struct page *dummy_page; |
6730 | }; | |
6731 | ||
6732 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 6733 | { |
a9d5adee JG |
6734 | memset(ug, 0, sizeof(*ug)); |
6735 | } | |
6736 | ||
6737 | static void uncharge_batch(const struct uncharge_gather *ug) | |
6738 | { | |
747db954 JW |
6739 | unsigned long flags; |
6740 | ||
a9d5adee | 6741 | if (!mem_cgroup_is_root(ug->memcg)) { |
9f762dbe | 6742 | page_counter_uncharge(&ug->memcg->memory, ug->nr_pages); |
7941d214 | 6743 | if (do_memsw_account()) |
9f762dbe | 6744 | page_counter_uncharge(&ug->memcg->memsw, ug->nr_pages); |
a9d5adee JG |
6745 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem) |
6746 | page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem); | |
6747 | memcg_oom_recover(ug->memcg); | |
ce00a967 | 6748 | } |
747db954 JW |
6749 | |
6750 | local_irq_save(flags); | |
c9019e9b | 6751 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); |
9f762dbe | 6752 | __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, ug->nr_pages); |
a9d5adee | 6753 | memcg_check_events(ug->memcg, ug->dummy_page); |
747db954 | 6754 | local_irq_restore(flags); |
a9d5adee JG |
6755 | } |
6756 | ||
6757 | static void uncharge_page(struct page *page, struct uncharge_gather *ug) | |
6758 | { | |
9f762dbe JW |
6759 | unsigned long nr_pages; |
6760 | ||
a9d5adee | 6761 | VM_BUG_ON_PAGE(PageLRU(page), page); |
a9d5adee JG |
6762 | |
6763 | if (!page->mem_cgroup) | |
6764 | return; | |
6765 | ||
6766 | /* | |
6767 | * Nobody should be changing or seriously looking at | |
6768 | * page->mem_cgroup at this point, we have fully | |
6769 | * exclusive access to the page. | |
6770 | */ | |
6771 | ||
6772 | if (ug->memcg != page->mem_cgroup) { | |
6773 | if (ug->memcg) { | |
6774 | uncharge_batch(ug); | |
6775 | uncharge_gather_clear(ug); | |
6776 | } | |
6777 | ug->memcg = page->mem_cgroup; | |
6778 | } | |
6779 | ||
9f762dbe JW |
6780 | nr_pages = compound_nr(page); |
6781 | ug->nr_pages += nr_pages; | |
a9d5adee | 6782 | |
9f762dbe | 6783 | if (!PageKmemcg(page)) { |
a9d5adee JG |
6784 | ug->pgpgout++; |
6785 | } else { | |
9f762dbe | 6786 | ug->nr_kmem += nr_pages; |
a9d5adee JG |
6787 | __ClearPageKmemcg(page); |
6788 | } | |
6789 | ||
6790 | ug->dummy_page = page; | |
6791 | page->mem_cgroup = NULL; | |
1a3e1f40 | 6792 | css_put(&ug->memcg->css); |
747db954 JW |
6793 | } |
6794 | ||
6795 | static void uncharge_list(struct list_head *page_list) | |
6796 | { | |
a9d5adee | 6797 | struct uncharge_gather ug; |
747db954 | 6798 | struct list_head *next; |
a9d5adee JG |
6799 | |
6800 | uncharge_gather_clear(&ug); | |
747db954 | 6801 | |
8b592656 JW |
6802 | /* |
6803 | * Note that the list can be a single page->lru; hence the | |
6804 | * do-while loop instead of a simple list_for_each_entry(). | |
6805 | */ | |
747db954 JW |
6806 | next = page_list->next; |
6807 | do { | |
a9d5adee JG |
6808 | struct page *page; |
6809 | ||
747db954 JW |
6810 | page = list_entry(next, struct page, lru); |
6811 | next = page->lru.next; | |
6812 | ||
a9d5adee | 6813 | uncharge_page(page, &ug); |
747db954 JW |
6814 | } while (next != page_list); |
6815 | ||
a9d5adee JG |
6816 | if (ug.memcg) |
6817 | uncharge_batch(&ug); | |
747db954 JW |
6818 | } |
6819 | ||
0a31bc97 JW |
6820 | /** |
6821 | * mem_cgroup_uncharge - uncharge a page | |
6822 | * @page: page to uncharge | |
6823 | * | |
f0e45fb4 | 6824 | * Uncharge a page previously charged with mem_cgroup_charge(). |
0a31bc97 JW |
6825 | */ |
6826 | void mem_cgroup_uncharge(struct page *page) | |
6827 | { | |
a9d5adee JG |
6828 | struct uncharge_gather ug; |
6829 | ||
0a31bc97 JW |
6830 | if (mem_cgroup_disabled()) |
6831 | return; | |
6832 | ||
747db954 | 6833 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 6834 | if (!page->mem_cgroup) |
0a31bc97 JW |
6835 | return; |
6836 | ||
a9d5adee JG |
6837 | uncharge_gather_clear(&ug); |
6838 | uncharge_page(page, &ug); | |
6839 | uncharge_batch(&ug); | |
747db954 | 6840 | } |
0a31bc97 | 6841 | |
747db954 JW |
6842 | /** |
6843 | * mem_cgroup_uncharge_list - uncharge a list of page | |
6844 | * @page_list: list of pages to uncharge | |
6845 | * | |
6846 | * Uncharge a list of pages previously charged with | |
f0e45fb4 | 6847 | * mem_cgroup_charge(). |
747db954 JW |
6848 | */ |
6849 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
6850 | { | |
6851 | if (mem_cgroup_disabled()) | |
6852 | return; | |
0a31bc97 | 6853 | |
747db954 JW |
6854 | if (!list_empty(page_list)) |
6855 | uncharge_list(page_list); | |
0a31bc97 JW |
6856 | } |
6857 | ||
6858 | /** | |
6a93ca8f JW |
6859 | * mem_cgroup_migrate - charge a page's replacement |
6860 | * @oldpage: currently circulating page | |
6861 | * @newpage: replacement page | |
0a31bc97 | 6862 | * |
6a93ca8f JW |
6863 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
6864 | * be uncharged upon free. | |
0a31bc97 JW |
6865 | * |
6866 | * Both pages must be locked, @newpage->mapping must be set up. | |
6867 | */ | |
6a93ca8f | 6868 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 6869 | { |
29833315 | 6870 | struct mem_cgroup *memcg; |
44b7a8d3 | 6871 | unsigned int nr_pages; |
d93c4130 | 6872 | unsigned long flags; |
0a31bc97 JW |
6873 | |
6874 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6875 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 6876 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
6877 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6878 | newpage); | |
0a31bc97 JW |
6879 | |
6880 | if (mem_cgroup_disabled()) | |
6881 | return; | |
6882 | ||
6883 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 6884 | if (newpage->mem_cgroup) |
0a31bc97 JW |
6885 | return; |
6886 | ||
45637bab | 6887 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 6888 | memcg = oldpage->mem_cgroup; |
29833315 | 6889 | if (!memcg) |
0a31bc97 JW |
6890 | return; |
6891 | ||
44b7a8d3 | 6892 | /* Force-charge the new page. The old one will be freed soon */ |
92855270 | 6893 | nr_pages = hpage_nr_pages(newpage); |
44b7a8d3 JW |
6894 | |
6895 | page_counter_charge(&memcg->memory, nr_pages); | |
6896 | if (do_memsw_account()) | |
6897 | page_counter_charge(&memcg->memsw, nr_pages); | |
0a31bc97 | 6898 | |
1a3e1f40 | 6899 | css_get(&memcg->css); |
d9eb1ea2 | 6900 | commit_charge(newpage, memcg); |
44b7a8d3 | 6901 | |
d93c4130 | 6902 | local_irq_save(flags); |
3fba69a5 | 6903 | mem_cgroup_charge_statistics(memcg, newpage, nr_pages); |
44b7a8d3 | 6904 | memcg_check_events(memcg, newpage); |
d93c4130 | 6905 | local_irq_restore(flags); |
0a31bc97 JW |
6906 | } |
6907 | ||
ef12947c | 6908 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
6909 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
6910 | ||
2d758073 | 6911 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
6912 | { |
6913 | struct mem_cgroup *memcg; | |
6914 | ||
2d758073 JW |
6915 | if (!mem_cgroup_sockets_enabled) |
6916 | return; | |
6917 | ||
e876ecc6 SB |
6918 | /* Do not associate the sock with unrelated interrupted task's memcg. */ |
6919 | if (in_interrupt()) | |
6920 | return; | |
6921 | ||
11092087 JW |
6922 | rcu_read_lock(); |
6923 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
6924 | if (memcg == root_mem_cgroup) |
6925 | goto out; | |
0db15298 | 6926 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 6927 | goto out; |
8965aa28 | 6928 | if (css_tryget(&memcg->css)) |
11092087 | 6929 | sk->sk_memcg = memcg; |
f7e1cb6e | 6930 | out: |
11092087 JW |
6931 | rcu_read_unlock(); |
6932 | } | |
11092087 | 6933 | |
2d758073 | 6934 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 6935 | { |
2d758073 JW |
6936 | if (sk->sk_memcg) |
6937 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
6938 | } |
6939 | ||
6940 | /** | |
6941 | * mem_cgroup_charge_skmem - charge socket memory | |
6942 | * @memcg: memcg to charge | |
6943 | * @nr_pages: number of pages to charge | |
6944 | * | |
6945 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
6946 | * @memcg's configured limit, %false if the charge had to be forced. | |
6947 | */ | |
6948 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6949 | { | |
f7e1cb6e | 6950 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 6951 | |
f7e1cb6e | 6952 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6953 | struct page_counter *fail; |
f7e1cb6e | 6954 | |
0db15298 JW |
6955 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
6956 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
6957 | return true; |
6958 | } | |
0db15298 JW |
6959 | page_counter_charge(&memcg->tcpmem, nr_pages); |
6960 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 6961 | return false; |
11092087 | 6962 | } |
d886f4e4 | 6963 | |
f7e1cb6e JW |
6964 | /* Don't block in the packet receive path */ |
6965 | if (in_softirq()) | |
6966 | gfp_mask = GFP_NOWAIT; | |
6967 | ||
c9019e9b | 6968 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); |
b2807f07 | 6969 | |
f7e1cb6e JW |
6970 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
6971 | return true; | |
6972 | ||
6973 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
6974 | return false; |
6975 | } | |
6976 | ||
6977 | /** | |
6978 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
6979 | * @memcg: memcg to uncharge |
6980 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
6981 | */ |
6982 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6983 | { | |
f7e1cb6e | 6984 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6985 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
6986 | return; |
6987 | } | |
d886f4e4 | 6988 | |
c9019e9b | 6989 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 6990 | |
475d0487 | 6991 | refill_stock(memcg, nr_pages); |
11092087 JW |
6992 | } |
6993 | ||
f7e1cb6e JW |
6994 | static int __init cgroup_memory(char *s) |
6995 | { | |
6996 | char *token; | |
6997 | ||
6998 | while ((token = strsep(&s, ",")) != NULL) { | |
6999 | if (!*token) | |
7000 | continue; | |
7001 | if (!strcmp(token, "nosocket")) | |
7002 | cgroup_memory_nosocket = true; | |
04823c83 VD |
7003 | if (!strcmp(token, "nokmem")) |
7004 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
7005 | } |
7006 | return 0; | |
7007 | } | |
7008 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 7009 | |
2d11085e | 7010 | /* |
1081312f MH |
7011 | * subsys_initcall() for memory controller. |
7012 | * | |
308167fc SAS |
7013 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
7014 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
7015 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
7016 | * should be initialized from here. | |
2d11085e MH |
7017 | */ |
7018 | static int __init mem_cgroup_init(void) | |
7019 | { | |
95a045f6 JW |
7020 | int cpu, node; |
7021 | ||
308167fc SAS |
7022 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
7023 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
7024 | |
7025 | for_each_possible_cpu(cpu) | |
7026 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
7027 | drain_local_stock); | |
7028 | ||
7029 | for_each_node(node) { | |
7030 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
7031 | |
7032 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
7033 | node_online(node) ? node : NUMA_NO_NODE); | |
7034 | ||
ef8f2327 | 7035 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 7036 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 7037 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
7038 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
7039 | } | |
7040 | ||
2d11085e MH |
7041 | return 0; |
7042 | } | |
7043 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
7044 | |
7045 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
7046 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
7047 | { | |
1c2d479a | 7048 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
7049 | /* |
7050 | * The root cgroup cannot be destroyed, so it's refcount must | |
7051 | * always be >= 1. | |
7052 | */ | |
7053 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
7054 | VM_BUG_ON(1); | |
7055 | break; | |
7056 | } | |
7057 | memcg = parent_mem_cgroup(memcg); | |
7058 | if (!memcg) | |
7059 | memcg = root_mem_cgroup; | |
7060 | } | |
7061 | return memcg; | |
7062 | } | |
7063 | ||
21afa38e JW |
7064 | /** |
7065 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
7066 | * @page: page whose memsw charge to transfer | |
7067 | * @entry: swap entry to move the charge to | |
7068 | * | |
7069 | * Transfer the memsw charge of @page to @entry. | |
7070 | */ | |
7071 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
7072 | { | |
1f47b61f | 7073 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 7074 | unsigned int nr_entries; |
21afa38e JW |
7075 | unsigned short oldid; |
7076 | ||
7077 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
7078 | VM_BUG_ON_PAGE(page_count(page), page); | |
7079 | ||
2d1c4980 | 7080 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
21afa38e JW |
7081 | return; |
7082 | ||
7083 | memcg = page->mem_cgroup; | |
7084 | ||
7085 | /* Readahead page, never charged */ | |
7086 | if (!memcg) | |
7087 | return; | |
7088 | ||
1f47b61f VD |
7089 | /* |
7090 | * In case the memcg owning these pages has been offlined and doesn't | |
7091 | * have an ID allocated to it anymore, charge the closest online | |
7092 | * ancestor for the swap instead and transfer the memory+swap charge. | |
7093 | */ | |
7094 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
d6810d73 HY |
7095 | nr_entries = hpage_nr_pages(page); |
7096 | /* Get references for the tail pages, too */ | |
7097 | if (nr_entries > 1) | |
7098 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
7099 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
7100 | nr_entries); | |
21afa38e | 7101 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7102 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e JW |
7103 | |
7104 | page->mem_cgroup = NULL; | |
7105 | ||
7106 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 7107 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 7108 | |
2d1c4980 | 7109 | if (!cgroup_memory_noswap && memcg != swap_memcg) { |
1f47b61f | 7110 | if (!mem_cgroup_is_root(swap_memcg)) |
d6810d73 HY |
7111 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
7112 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
7113 | } |
7114 | ||
ce9ce665 SAS |
7115 | /* |
7116 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 7117 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 7118 | * important here to have the interrupts disabled because it is the |
b93b0163 | 7119 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 SAS |
7120 | */ |
7121 | VM_BUG_ON(!irqs_disabled()); | |
3fba69a5 | 7122 | mem_cgroup_charge_statistics(memcg, page, -nr_entries); |
21afa38e | 7123 | memcg_check_events(memcg, page); |
73f576c0 | 7124 | |
1a3e1f40 | 7125 | css_put(&memcg->css); |
21afa38e JW |
7126 | } |
7127 | ||
38d8b4e6 HY |
7128 | /** |
7129 | * mem_cgroup_try_charge_swap - try charging swap space for a page | |
37e84351 VD |
7130 | * @page: page being added to swap |
7131 | * @entry: swap entry to charge | |
7132 | * | |
38d8b4e6 | 7133 | * Try to charge @page's memcg for the swap space at @entry. |
37e84351 VD |
7134 | * |
7135 | * Returns 0 on success, -ENOMEM on failure. | |
7136 | */ | |
7137 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
7138 | { | |
38d8b4e6 | 7139 | unsigned int nr_pages = hpage_nr_pages(page); |
37e84351 | 7140 | struct page_counter *counter; |
38d8b4e6 | 7141 | struct mem_cgroup *memcg; |
37e84351 VD |
7142 | unsigned short oldid; |
7143 | ||
2d1c4980 | 7144 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
37e84351 VD |
7145 | return 0; |
7146 | ||
7147 | memcg = page->mem_cgroup; | |
7148 | ||
7149 | /* Readahead page, never charged */ | |
7150 | if (!memcg) | |
7151 | return 0; | |
7152 | ||
f3a53a3a TH |
7153 | if (!entry.val) { |
7154 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 7155 | return 0; |
f3a53a3a | 7156 | } |
bb98f2c5 | 7157 | |
1f47b61f VD |
7158 | memcg = mem_cgroup_id_get_online(memcg); |
7159 | ||
2d1c4980 | 7160 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg) && |
38d8b4e6 | 7161 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
7162 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
7163 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 7164 | mem_cgroup_id_put(memcg); |
37e84351 | 7165 | return -ENOMEM; |
1f47b61f | 7166 | } |
37e84351 | 7167 | |
38d8b4e6 HY |
7168 | /* Get references for the tail pages, too */ |
7169 | if (nr_pages > 1) | |
7170 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
7171 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
37e84351 | 7172 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7173 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 7174 | |
37e84351 VD |
7175 | return 0; |
7176 | } | |
7177 | ||
21afa38e | 7178 | /** |
38d8b4e6 | 7179 | * mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 7180 | * @entry: swap entry to uncharge |
38d8b4e6 | 7181 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 7182 | */ |
38d8b4e6 | 7183 | void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
7184 | { |
7185 | struct mem_cgroup *memcg; | |
7186 | unsigned short id; | |
7187 | ||
38d8b4e6 | 7188 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 7189 | rcu_read_lock(); |
adbe427b | 7190 | memcg = mem_cgroup_from_id(id); |
21afa38e | 7191 | if (memcg) { |
2d1c4980 | 7192 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg)) { |
37e84351 | 7193 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
38d8b4e6 | 7194 | page_counter_uncharge(&memcg->swap, nr_pages); |
37e84351 | 7195 | else |
38d8b4e6 | 7196 | page_counter_uncharge(&memcg->memsw, nr_pages); |
37e84351 | 7197 | } |
c9019e9b | 7198 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 7199 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
7200 | } |
7201 | rcu_read_unlock(); | |
7202 | } | |
7203 | ||
d8b38438 VD |
7204 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
7205 | { | |
7206 | long nr_swap_pages = get_nr_swap_pages(); | |
7207 | ||
eccb52e7 | 7208 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
d8b38438 VD |
7209 | return nr_swap_pages; |
7210 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
7211 | nr_swap_pages = min_t(long, nr_swap_pages, | |
bbec2e15 | 7212 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
7213 | page_counter_read(&memcg->swap)); |
7214 | return nr_swap_pages; | |
7215 | } | |
7216 | ||
5ccc5aba VD |
7217 | bool mem_cgroup_swap_full(struct page *page) |
7218 | { | |
7219 | struct mem_cgroup *memcg; | |
7220 | ||
7221 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
7222 | ||
7223 | if (vm_swap_full()) | |
7224 | return true; | |
eccb52e7 | 7225 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
5ccc5aba VD |
7226 | return false; |
7227 | ||
7228 | memcg = page->mem_cgroup; | |
7229 | if (!memcg) | |
7230 | return false; | |
7231 | ||
4b82ab4f JK |
7232 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { |
7233 | unsigned long usage = page_counter_read(&memcg->swap); | |
7234 | ||
7235 | if (usage * 2 >= READ_ONCE(memcg->swap.high) || | |
7236 | usage * 2 >= READ_ONCE(memcg->swap.max)) | |
5ccc5aba | 7237 | return true; |
4b82ab4f | 7238 | } |
5ccc5aba VD |
7239 | |
7240 | return false; | |
7241 | } | |
7242 | ||
eccb52e7 | 7243 | static int __init setup_swap_account(char *s) |
21afa38e JW |
7244 | { |
7245 | if (!strcmp(s, "1")) | |
eccb52e7 | 7246 | cgroup_memory_noswap = 0; |
21afa38e | 7247 | else if (!strcmp(s, "0")) |
eccb52e7 | 7248 | cgroup_memory_noswap = 1; |
21afa38e JW |
7249 | return 1; |
7250 | } | |
eccb52e7 | 7251 | __setup("swapaccount=", setup_swap_account); |
21afa38e | 7252 | |
37e84351 VD |
7253 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
7254 | struct cftype *cft) | |
7255 | { | |
7256 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
7257 | ||
7258 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
7259 | } | |
7260 | ||
4b82ab4f JK |
7261 | static int swap_high_show(struct seq_file *m, void *v) |
7262 | { | |
7263 | return seq_puts_memcg_tunable(m, | |
7264 | READ_ONCE(mem_cgroup_from_seq(m)->swap.high)); | |
7265 | } | |
7266 | ||
7267 | static ssize_t swap_high_write(struct kernfs_open_file *of, | |
7268 | char *buf, size_t nbytes, loff_t off) | |
7269 | { | |
7270 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7271 | unsigned long high; | |
7272 | int err; | |
7273 | ||
7274 | buf = strstrip(buf); | |
7275 | err = page_counter_memparse(buf, "max", &high); | |
7276 | if (err) | |
7277 | return err; | |
7278 | ||
7279 | page_counter_set_high(&memcg->swap, high); | |
7280 | ||
7281 | return nbytes; | |
7282 | } | |
7283 | ||
37e84351 VD |
7284 | static int swap_max_show(struct seq_file *m, void *v) |
7285 | { | |
677dc973 CD |
7286 | return seq_puts_memcg_tunable(m, |
7287 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); | |
37e84351 VD |
7288 | } |
7289 | ||
7290 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
7291 | char *buf, size_t nbytes, loff_t off) | |
7292 | { | |
7293 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7294 | unsigned long max; | |
7295 | int err; | |
7296 | ||
7297 | buf = strstrip(buf); | |
7298 | err = page_counter_memparse(buf, "max", &max); | |
7299 | if (err) | |
7300 | return err; | |
7301 | ||
be09102b | 7302 | xchg(&memcg->swap.max, max); |
37e84351 VD |
7303 | |
7304 | return nbytes; | |
7305 | } | |
7306 | ||
f3a53a3a TH |
7307 | static int swap_events_show(struct seq_file *m, void *v) |
7308 | { | |
aa9694bb | 7309 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
f3a53a3a | 7310 | |
4b82ab4f JK |
7311 | seq_printf(m, "high %lu\n", |
7312 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_HIGH])); | |
f3a53a3a TH |
7313 | seq_printf(m, "max %lu\n", |
7314 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
7315 | seq_printf(m, "fail %lu\n", | |
7316 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
7317 | ||
7318 | return 0; | |
7319 | } | |
7320 | ||
37e84351 VD |
7321 | static struct cftype swap_files[] = { |
7322 | { | |
7323 | .name = "swap.current", | |
7324 | .flags = CFTYPE_NOT_ON_ROOT, | |
7325 | .read_u64 = swap_current_read, | |
7326 | }, | |
4b82ab4f JK |
7327 | { |
7328 | .name = "swap.high", | |
7329 | .flags = CFTYPE_NOT_ON_ROOT, | |
7330 | .seq_show = swap_high_show, | |
7331 | .write = swap_high_write, | |
7332 | }, | |
37e84351 VD |
7333 | { |
7334 | .name = "swap.max", | |
7335 | .flags = CFTYPE_NOT_ON_ROOT, | |
7336 | .seq_show = swap_max_show, | |
7337 | .write = swap_max_write, | |
7338 | }, | |
f3a53a3a TH |
7339 | { |
7340 | .name = "swap.events", | |
7341 | .flags = CFTYPE_NOT_ON_ROOT, | |
7342 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
7343 | .seq_show = swap_events_show, | |
7344 | }, | |
37e84351 VD |
7345 | { } /* terminate */ |
7346 | }; | |
7347 | ||
eccb52e7 | 7348 | static struct cftype memsw_files[] = { |
21afa38e JW |
7349 | { |
7350 | .name = "memsw.usage_in_bytes", | |
7351 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
7352 | .read_u64 = mem_cgroup_read_u64, | |
7353 | }, | |
7354 | { | |
7355 | .name = "memsw.max_usage_in_bytes", | |
7356 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
7357 | .write = mem_cgroup_reset, | |
7358 | .read_u64 = mem_cgroup_read_u64, | |
7359 | }, | |
7360 | { | |
7361 | .name = "memsw.limit_in_bytes", | |
7362 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
7363 | .write = mem_cgroup_write, | |
7364 | .read_u64 = mem_cgroup_read_u64, | |
7365 | }, | |
7366 | { | |
7367 | .name = "memsw.failcnt", | |
7368 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
7369 | .write = mem_cgroup_reset, | |
7370 | .read_u64 = mem_cgroup_read_u64, | |
7371 | }, | |
7372 | { }, /* terminate */ | |
7373 | }; | |
7374 | ||
82ff165c BS |
7375 | /* |
7376 | * If mem_cgroup_swap_init() is implemented as a subsys_initcall() | |
7377 | * instead of a core_initcall(), this could mean cgroup_memory_noswap still | |
7378 | * remains set to false even when memcg is disabled via "cgroup_disable=memory" | |
7379 | * boot parameter. This may result in premature OOPS inside | |
7380 | * mem_cgroup_get_nr_swap_pages() function in corner cases. | |
7381 | */ | |
21afa38e JW |
7382 | static int __init mem_cgroup_swap_init(void) |
7383 | { | |
2d1c4980 JW |
7384 | /* No memory control -> no swap control */ |
7385 | if (mem_cgroup_disabled()) | |
7386 | cgroup_memory_noswap = true; | |
7387 | ||
7388 | if (cgroup_memory_noswap) | |
eccb52e7 JW |
7389 | return 0; |
7390 | ||
7391 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, swap_files)); | |
7392 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, memsw_files)); | |
7393 | ||
21afa38e JW |
7394 | return 0; |
7395 | } | |
82ff165c | 7396 | core_initcall(mem_cgroup_swap_init); |
21afa38e JW |
7397 | |
7398 | #endif /* CONFIG_MEMCG_SWAP */ |