]>
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 | ||
f7e1cb6e JW |
76 | /* Socket memory accounting disabled? */ |
77 | static bool cgroup_memory_nosocket; | |
78 | ||
04823c83 VD |
79 | /* Kernel memory accounting disabled? */ |
80 | static bool cgroup_memory_nokmem; | |
81 | ||
21afa38e | 82 | /* Whether the swap controller is active */ |
c255a458 | 83 | #ifdef CONFIG_MEMCG_SWAP |
eccb52e7 | 84 | bool cgroup_memory_noswap __read_mostly; |
c077719b | 85 | #else |
eccb52e7 | 86 | #define cgroup_memory_noswap 1 |
2d1c4980 | 87 | #endif |
c077719b | 88 | |
97b27821 TH |
89 | #ifdef CONFIG_CGROUP_WRITEBACK |
90 | static DECLARE_WAIT_QUEUE_HEAD(memcg_cgwb_frn_waitq); | |
91 | #endif | |
92 | ||
7941d214 JW |
93 | /* Whether legacy memory+swap accounting is active */ |
94 | static bool do_memsw_account(void) | |
95 | { | |
eccb52e7 | 96 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_noswap; |
7941d214 JW |
97 | } |
98 | ||
a0db00fc KS |
99 | #define THRESHOLDS_EVENTS_TARGET 128 |
100 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
e9f8974f | 101 | |
bb4cc1a8 AM |
102 | /* |
103 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
104 | * their hierarchy representation | |
105 | */ | |
106 | ||
ef8f2327 | 107 | struct mem_cgroup_tree_per_node { |
bb4cc1a8 | 108 | struct rb_root rb_root; |
fa90b2fd | 109 | struct rb_node *rb_rightmost; |
bb4cc1a8 AM |
110 | spinlock_t lock; |
111 | }; | |
112 | ||
bb4cc1a8 AM |
113 | struct mem_cgroup_tree { |
114 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
115 | }; | |
116 | ||
117 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
118 | ||
9490ff27 KH |
119 | /* for OOM */ |
120 | struct mem_cgroup_eventfd_list { | |
121 | struct list_head list; | |
122 | struct eventfd_ctx *eventfd; | |
123 | }; | |
2e72b634 | 124 | |
79bd9814 TH |
125 | /* |
126 | * cgroup_event represents events which userspace want to receive. | |
127 | */ | |
3bc942f3 | 128 | struct mem_cgroup_event { |
79bd9814 | 129 | /* |
59b6f873 | 130 | * memcg which the event belongs to. |
79bd9814 | 131 | */ |
59b6f873 | 132 | struct mem_cgroup *memcg; |
79bd9814 TH |
133 | /* |
134 | * eventfd to signal userspace about the event. | |
135 | */ | |
136 | struct eventfd_ctx *eventfd; | |
137 | /* | |
138 | * Each of these stored in a list by the cgroup. | |
139 | */ | |
140 | struct list_head list; | |
fba94807 TH |
141 | /* |
142 | * register_event() callback will be used to add new userspace | |
143 | * waiter for changes related to this event. Use eventfd_signal() | |
144 | * on eventfd to send notification to userspace. | |
145 | */ | |
59b6f873 | 146 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 147 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
148 | /* |
149 | * unregister_event() callback will be called when userspace closes | |
150 | * the eventfd or on cgroup removing. This callback must be set, | |
151 | * if you want provide notification functionality. | |
152 | */ | |
59b6f873 | 153 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 154 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
155 | /* |
156 | * All fields below needed to unregister event when | |
157 | * userspace closes eventfd. | |
158 | */ | |
159 | poll_table pt; | |
160 | wait_queue_head_t *wqh; | |
ac6424b9 | 161 | wait_queue_entry_t wait; |
79bd9814 TH |
162 | struct work_struct remove; |
163 | }; | |
164 | ||
c0ff4b85 R |
165 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
166 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 167 | |
7dc74be0 DN |
168 | /* Stuffs for move charges at task migration. */ |
169 | /* | |
1dfab5ab | 170 | * Types of charges to be moved. |
7dc74be0 | 171 | */ |
1dfab5ab JW |
172 | #define MOVE_ANON 0x1U |
173 | #define MOVE_FILE 0x2U | |
174 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 175 | |
4ffef5fe DN |
176 | /* "mc" and its members are protected by cgroup_mutex */ |
177 | static struct move_charge_struct { | |
b1dd693e | 178 | spinlock_t lock; /* for from, to */ |
264a0ae1 | 179 | struct mm_struct *mm; |
4ffef5fe DN |
180 | struct mem_cgroup *from; |
181 | struct mem_cgroup *to; | |
1dfab5ab | 182 | unsigned long flags; |
4ffef5fe | 183 | unsigned long precharge; |
854ffa8d | 184 | unsigned long moved_charge; |
483c30b5 | 185 | unsigned long moved_swap; |
8033b97c DN |
186 | struct task_struct *moving_task; /* a task moving charges */ |
187 | wait_queue_head_t waitq; /* a waitq for other context */ | |
188 | } mc = { | |
2bd9bb20 | 189 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
190 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
191 | }; | |
4ffef5fe | 192 | |
4e416953 BS |
193 | /* |
194 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
195 | * limit reclaim to prevent infinite loops, if they ever occur. | |
196 | */ | |
a0db00fc | 197 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 198 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 199 | |
8c7c6e34 | 200 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
201 | enum res_type { |
202 | _MEM, | |
203 | _MEMSWAP, | |
204 | _OOM_TYPE, | |
510fc4e1 | 205 | _KMEM, |
d55f90bf | 206 | _TCP, |
86ae53e1 GC |
207 | }; |
208 | ||
a0db00fc KS |
209 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
210 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 211 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
212 | /* Used for OOM nofiier */ |
213 | #define OOM_CONTROL (0) | |
8c7c6e34 | 214 | |
b05706f1 KT |
215 | /* |
216 | * Iteration constructs for visiting all cgroups (under a tree). If | |
217 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
218 | * be used for reference counting. | |
219 | */ | |
220 | #define for_each_mem_cgroup_tree(iter, root) \ | |
221 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ | |
222 | iter != NULL; \ | |
223 | iter = mem_cgroup_iter(root, iter, NULL)) | |
224 | ||
225 | #define for_each_mem_cgroup(iter) \ | |
226 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ | |
227 | iter != NULL; \ | |
228 | iter = mem_cgroup_iter(NULL, iter, NULL)) | |
229 | ||
7775face TH |
230 | static inline bool should_force_charge(void) |
231 | { | |
232 | return tsk_is_oom_victim(current) || fatal_signal_pending(current) || | |
233 | (current->flags & PF_EXITING); | |
234 | } | |
235 | ||
70ddf637 AV |
236 | /* Some nice accessors for the vmpressure. */ |
237 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
238 | { | |
239 | if (!memcg) | |
240 | memcg = root_mem_cgroup; | |
241 | return &memcg->vmpressure; | |
242 | } | |
243 | ||
244 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
245 | { | |
246 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
247 | } | |
248 | ||
84c07d11 | 249 | #ifdef CONFIG_MEMCG_KMEM |
bf4f0599 RG |
250 | extern spinlock_t css_set_lock; |
251 | ||
252 | static void obj_cgroup_release(struct percpu_ref *ref) | |
253 | { | |
254 | struct obj_cgroup *objcg = container_of(ref, struct obj_cgroup, refcnt); | |
255 | struct mem_cgroup *memcg; | |
256 | unsigned int nr_bytes; | |
257 | unsigned int nr_pages; | |
258 | unsigned long flags; | |
259 | ||
260 | /* | |
261 | * At this point all allocated objects are freed, and | |
262 | * objcg->nr_charged_bytes can't have an arbitrary byte value. | |
263 | * However, it can be PAGE_SIZE or (x * PAGE_SIZE). | |
264 | * | |
265 | * The following sequence can lead to it: | |
266 | * 1) CPU0: objcg == stock->cached_objcg | |
267 | * 2) CPU1: we do a small allocation (e.g. 92 bytes), | |
268 | * PAGE_SIZE bytes are charged | |
269 | * 3) CPU1: a process from another memcg is allocating something, | |
270 | * the stock if flushed, | |
271 | * objcg->nr_charged_bytes = PAGE_SIZE - 92 | |
272 | * 5) CPU0: we do release this object, | |
273 | * 92 bytes are added to stock->nr_bytes | |
274 | * 6) CPU0: stock is flushed, | |
275 | * 92 bytes are added to objcg->nr_charged_bytes | |
276 | * | |
277 | * In the result, nr_charged_bytes == PAGE_SIZE. | |
278 | * This page will be uncharged in obj_cgroup_release(). | |
279 | */ | |
280 | nr_bytes = atomic_read(&objcg->nr_charged_bytes); | |
281 | WARN_ON_ONCE(nr_bytes & (PAGE_SIZE - 1)); | |
282 | nr_pages = nr_bytes >> PAGE_SHIFT; | |
283 | ||
284 | spin_lock_irqsave(&css_set_lock, flags); | |
285 | memcg = obj_cgroup_memcg(objcg); | |
286 | if (nr_pages) | |
287 | __memcg_kmem_uncharge(memcg, nr_pages); | |
288 | list_del(&objcg->list); | |
289 | mem_cgroup_put(memcg); | |
290 | spin_unlock_irqrestore(&css_set_lock, flags); | |
291 | ||
292 | percpu_ref_exit(ref); | |
293 | kfree_rcu(objcg, rcu); | |
294 | } | |
295 | ||
296 | static struct obj_cgroup *obj_cgroup_alloc(void) | |
297 | { | |
298 | struct obj_cgroup *objcg; | |
299 | int ret; | |
300 | ||
301 | objcg = kzalloc(sizeof(struct obj_cgroup), GFP_KERNEL); | |
302 | if (!objcg) | |
303 | return NULL; | |
304 | ||
305 | ret = percpu_ref_init(&objcg->refcnt, obj_cgroup_release, 0, | |
306 | GFP_KERNEL); | |
307 | if (ret) { | |
308 | kfree(objcg); | |
309 | return NULL; | |
310 | } | |
311 | INIT_LIST_HEAD(&objcg->list); | |
312 | return objcg; | |
313 | } | |
314 | ||
315 | static void memcg_reparent_objcgs(struct mem_cgroup *memcg, | |
316 | struct mem_cgroup *parent) | |
317 | { | |
318 | struct obj_cgroup *objcg, *iter; | |
319 | ||
320 | objcg = rcu_replace_pointer(memcg->objcg, NULL, true); | |
321 | ||
322 | spin_lock_irq(&css_set_lock); | |
323 | ||
324 | /* Move active objcg to the parent's list */ | |
325 | xchg(&objcg->memcg, parent); | |
326 | css_get(&parent->css); | |
327 | list_add(&objcg->list, &parent->objcg_list); | |
328 | ||
329 | /* Move already reparented objcgs to the parent's list */ | |
330 | list_for_each_entry(iter, &memcg->objcg_list, list) { | |
331 | css_get(&parent->css); | |
332 | xchg(&iter->memcg, parent); | |
333 | css_put(&memcg->css); | |
334 | } | |
335 | list_splice(&memcg->objcg_list, &parent->objcg_list); | |
336 | ||
337 | spin_unlock_irq(&css_set_lock); | |
338 | ||
339 | percpu_ref_kill(&objcg->refcnt); | |
340 | } | |
341 | ||
55007d84 | 342 | /* |
9855609b | 343 | * This will be used as a shrinker list's index. |
b8627835 LZ |
344 | * The main reason for not using cgroup id for this: |
345 | * this works better in sparse environments, where we have a lot of memcgs, | |
346 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
347 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
348 | * 200 entry array for that. | |
55007d84 | 349 | * |
dbcf73e2 VD |
350 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
351 | * will double each time we have to increase it. | |
55007d84 | 352 | */ |
dbcf73e2 VD |
353 | static DEFINE_IDA(memcg_cache_ida); |
354 | int memcg_nr_cache_ids; | |
749c5415 | 355 | |
05257a1a VD |
356 | /* Protects memcg_nr_cache_ids */ |
357 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
358 | ||
359 | void memcg_get_cache_ids(void) | |
360 | { | |
361 | down_read(&memcg_cache_ids_sem); | |
362 | } | |
363 | ||
364 | void memcg_put_cache_ids(void) | |
365 | { | |
366 | up_read(&memcg_cache_ids_sem); | |
367 | } | |
368 | ||
55007d84 GC |
369 | /* |
370 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
371 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
372 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
373 | * tunable, but that is strictly not necessary. | |
374 | * | |
b8627835 | 375 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
376 | * this constant directly from cgroup, but it is understandable that this is |
377 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 378 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
379 | * increase ours as well if it increases. |
380 | */ | |
381 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 382 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 383 | |
d7f25f8a GC |
384 | /* |
385 | * A lot of the calls to the cache allocation functions are expected to be | |
272911a4 | 386 | * inlined by the compiler. Since the calls to memcg_slab_pre_alloc_hook() are |
d7f25f8a GC |
387 | * conditional to this static branch, we'll have to allow modules that does |
388 | * kmem_cache_alloc and the such to see this symbol as well | |
389 | */ | |
ef12947c | 390 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
d7f25f8a | 391 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
0a432dcb | 392 | #endif |
17cc4dfe | 393 | |
0a4465d3 KT |
394 | static int memcg_shrinker_map_size; |
395 | static DEFINE_MUTEX(memcg_shrinker_map_mutex); | |
396 | ||
397 | static void memcg_free_shrinker_map_rcu(struct rcu_head *head) | |
398 | { | |
399 | kvfree(container_of(head, struct memcg_shrinker_map, rcu)); | |
400 | } | |
401 | ||
402 | static int memcg_expand_one_shrinker_map(struct mem_cgroup *memcg, | |
403 | int size, int old_size) | |
404 | { | |
405 | struct memcg_shrinker_map *new, *old; | |
406 | int nid; | |
407 | ||
408 | lockdep_assert_held(&memcg_shrinker_map_mutex); | |
409 | ||
410 | for_each_node(nid) { | |
411 | old = rcu_dereference_protected( | |
412 | mem_cgroup_nodeinfo(memcg, nid)->shrinker_map, true); | |
413 | /* Not yet online memcg */ | |
414 | if (!old) | |
415 | return 0; | |
416 | ||
86daf94e | 417 | new = kvmalloc_node(sizeof(*new) + size, GFP_KERNEL, nid); |
0a4465d3 KT |
418 | if (!new) |
419 | return -ENOMEM; | |
420 | ||
421 | /* Set all old bits, clear all new bits */ | |
422 | memset(new->map, (int)0xff, old_size); | |
423 | memset((void *)new->map + old_size, 0, size - old_size); | |
424 | ||
425 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, new); | |
426 | call_rcu(&old->rcu, memcg_free_shrinker_map_rcu); | |
427 | } | |
428 | ||
429 | return 0; | |
430 | } | |
431 | ||
432 | static void memcg_free_shrinker_maps(struct mem_cgroup *memcg) | |
433 | { | |
434 | struct mem_cgroup_per_node *pn; | |
435 | struct memcg_shrinker_map *map; | |
436 | int nid; | |
437 | ||
438 | if (mem_cgroup_is_root(memcg)) | |
439 | return; | |
440 | ||
441 | for_each_node(nid) { | |
442 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
443 | map = rcu_dereference_protected(pn->shrinker_map, true); | |
444 | if (map) | |
445 | kvfree(map); | |
446 | rcu_assign_pointer(pn->shrinker_map, NULL); | |
447 | } | |
448 | } | |
449 | ||
450 | static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg) | |
451 | { | |
452 | struct memcg_shrinker_map *map; | |
453 | int nid, size, ret = 0; | |
454 | ||
455 | if (mem_cgroup_is_root(memcg)) | |
456 | return 0; | |
457 | ||
458 | mutex_lock(&memcg_shrinker_map_mutex); | |
459 | size = memcg_shrinker_map_size; | |
460 | for_each_node(nid) { | |
86daf94e | 461 | map = kvzalloc_node(sizeof(*map) + size, GFP_KERNEL, nid); |
0a4465d3 KT |
462 | if (!map) { |
463 | memcg_free_shrinker_maps(memcg); | |
464 | ret = -ENOMEM; | |
465 | break; | |
466 | } | |
467 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, map); | |
468 | } | |
469 | mutex_unlock(&memcg_shrinker_map_mutex); | |
470 | ||
471 | return ret; | |
472 | } | |
473 | ||
474 | int memcg_expand_shrinker_maps(int new_id) | |
475 | { | |
476 | int size, old_size, ret = 0; | |
477 | struct mem_cgroup *memcg; | |
478 | ||
479 | size = DIV_ROUND_UP(new_id + 1, BITS_PER_LONG) * sizeof(unsigned long); | |
480 | old_size = memcg_shrinker_map_size; | |
481 | if (size <= old_size) | |
482 | return 0; | |
483 | ||
484 | mutex_lock(&memcg_shrinker_map_mutex); | |
485 | if (!root_mem_cgroup) | |
486 | goto unlock; | |
487 | ||
488 | for_each_mem_cgroup(memcg) { | |
489 | if (mem_cgroup_is_root(memcg)) | |
490 | continue; | |
491 | ret = memcg_expand_one_shrinker_map(memcg, size, old_size); | |
75866af6 VA |
492 | if (ret) { |
493 | mem_cgroup_iter_break(NULL, memcg); | |
0a4465d3 | 494 | goto unlock; |
75866af6 | 495 | } |
0a4465d3 KT |
496 | } |
497 | unlock: | |
498 | if (!ret) | |
499 | memcg_shrinker_map_size = size; | |
500 | mutex_unlock(&memcg_shrinker_map_mutex); | |
501 | return ret; | |
502 | } | |
fae91d6d KT |
503 | |
504 | void memcg_set_shrinker_bit(struct mem_cgroup *memcg, int nid, int shrinker_id) | |
505 | { | |
506 | if (shrinker_id >= 0 && memcg && !mem_cgroup_is_root(memcg)) { | |
507 | struct memcg_shrinker_map *map; | |
508 | ||
509 | rcu_read_lock(); | |
510 | map = rcu_dereference(memcg->nodeinfo[nid]->shrinker_map); | |
f90280d6 KT |
511 | /* Pairs with smp mb in shrink_slab() */ |
512 | smp_mb__before_atomic(); | |
fae91d6d KT |
513 | set_bit(shrinker_id, map->map); |
514 | rcu_read_unlock(); | |
515 | } | |
516 | } | |
517 | ||
ad7fa852 TH |
518 | /** |
519 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
520 | * @page: page of interest | |
521 | * | |
522 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
523 | * with @page is returned. The returned css remains associated with @page | |
524 | * until it is released. | |
525 | * | |
526 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
527 | * is returned. | |
ad7fa852 TH |
528 | */ |
529 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
530 | { | |
531 | struct mem_cgroup *memcg; | |
532 | ||
ad7fa852 TH |
533 | memcg = page->mem_cgroup; |
534 | ||
9e10a130 | 535 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
536 | memcg = root_mem_cgroup; |
537 | ||
ad7fa852 TH |
538 | return &memcg->css; |
539 | } | |
540 | ||
2fc04524 VD |
541 | /** |
542 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
543 | * @page: the page | |
544 | * | |
545 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
546 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
547 | * is safe to call this function without holding a reference to @page. | |
548 | * | |
549 | * Note, this function is inherently racy, because there is nothing to prevent | |
550 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
551 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
552 | * do not care (such as procfs interfaces). | |
553 | */ | |
554 | ino_t page_cgroup_ino(struct page *page) | |
555 | { | |
556 | struct mem_cgroup *memcg; | |
557 | unsigned long ino = 0; | |
558 | ||
559 | rcu_read_lock(); | |
9855609b | 560 | memcg = page->mem_cgroup; |
286e04b8 | 561 | |
9855609b RG |
562 | /* |
563 | * The lowest bit set means that memcg isn't a valid | |
564 | * memcg pointer, but a obj_cgroups pointer. | |
565 | * In this case the page is shared and doesn't belong | |
566 | * to any specific memory cgroup. | |
567 | */ | |
568 | if ((unsigned long) memcg & 0x1UL) | |
569 | memcg = NULL; | |
286e04b8 | 570 | |
2fc04524 VD |
571 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) |
572 | memcg = parent_mem_cgroup(memcg); | |
573 | if (memcg) | |
574 | ino = cgroup_ino(memcg->css.cgroup); | |
575 | rcu_read_unlock(); | |
576 | return ino; | |
577 | } | |
578 | ||
ef8f2327 MG |
579 | static struct mem_cgroup_per_node * |
580 | mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page) | |
f64c3f54 | 581 | { |
97a6c37b | 582 | int nid = page_to_nid(page); |
f64c3f54 | 583 | |
ef8f2327 | 584 | return memcg->nodeinfo[nid]; |
f64c3f54 BS |
585 | } |
586 | ||
ef8f2327 MG |
587 | static struct mem_cgroup_tree_per_node * |
588 | soft_limit_tree_node(int nid) | |
bb4cc1a8 | 589 | { |
ef8f2327 | 590 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
591 | } |
592 | ||
ef8f2327 | 593 | static struct mem_cgroup_tree_per_node * |
bb4cc1a8 AM |
594 | soft_limit_tree_from_page(struct page *page) |
595 | { | |
596 | int nid = page_to_nid(page); | |
bb4cc1a8 | 597 | |
ef8f2327 | 598 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
599 | } |
600 | ||
ef8f2327 MG |
601 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
602 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 603 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
604 | { |
605 | struct rb_node **p = &mctz->rb_root.rb_node; | |
606 | struct rb_node *parent = NULL; | |
ef8f2327 | 607 | struct mem_cgroup_per_node *mz_node; |
fa90b2fd | 608 | bool rightmost = true; |
bb4cc1a8 AM |
609 | |
610 | if (mz->on_tree) | |
611 | return; | |
612 | ||
613 | mz->usage_in_excess = new_usage_in_excess; | |
614 | if (!mz->usage_in_excess) | |
615 | return; | |
616 | while (*p) { | |
617 | parent = *p; | |
ef8f2327 | 618 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 | 619 | tree_node); |
fa90b2fd | 620 | if (mz->usage_in_excess < mz_node->usage_in_excess) { |
bb4cc1a8 | 621 | p = &(*p)->rb_left; |
fa90b2fd DB |
622 | rightmost = false; |
623 | } | |
624 | ||
bb4cc1a8 AM |
625 | /* |
626 | * We can't avoid mem cgroups that are over their soft | |
627 | * limit by the same amount | |
628 | */ | |
629 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
630 | p = &(*p)->rb_right; | |
631 | } | |
fa90b2fd DB |
632 | |
633 | if (rightmost) | |
634 | mctz->rb_rightmost = &mz->tree_node; | |
635 | ||
bb4cc1a8 AM |
636 | rb_link_node(&mz->tree_node, parent, p); |
637 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
638 | mz->on_tree = true; | |
639 | } | |
640 | ||
ef8f2327 MG |
641 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
642 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
643 | { |
644 | if (!mz->on_tree) | |
645 | return; | |
fa90b2fd DB |
646 | |
647 | if (&mz->tree_node == mctz->rb_rightmost) | |
648 | mctz->rb_rightmost = rb_prev(&mz->tree_node); | |
649 | ||
bb4cc1a8 AM |
650 | rb_erase(&mz->tree_node, &mctz->rb_root); |
651 | mz->on_tree = false; | |
652 | } | |
653 | ||
ef8f2327 MG |
654 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
655 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 656 | { |
0a31bc97 JW |
657 | unsigned long flags; |
658 | ||
659 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 660 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 661 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
662 | } |
663 | ||
3e32cb2e JW |
664 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
665 | { | |
666 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 667 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
668 | unsigned long excess = 0; |
669 | ||
670 | if (nr_pages > soft_limit) | |
671 | excess = nr_pages - soft_limit; | |
672 | ||
673 | return excess; | |
674 | } | |
bb4cc1a8 AM |
675 | |
676 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
677 | { | |
3e32cb2e | 678 | unsigned long excess; |
ef8f2327 MG |
679 | struct mem_cgroup_per_node *mz; |
680 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 681 | |
e231875b | 682 | mctz = soft_limit_tree_from_page(page); |
bfc7228b LD |
683 | if (!mctz) |
684 | return; | |
bb4cc1a8 AM |
685 | /* |
686 | * Necessary to update all ancestors when hierarchy is used. | |
687 | * because their event counter is not touched. | |
688 | */ | |
689 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
ef8f2327 | 690 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
3e32cb2e | 691 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
692 | /* |
693 | * We have to update the tree if mz is on RB-tree or | |
694 | * mem is over its softlimit. | |
695 | */ | |
696 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
697 | unsigned long flags; |
698 | ||
699 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
700 | /* if on-tree, remove it */ |
701 | if (mz->on_tree) | |
cf2c8127 | 702 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
703 | /* |
704 | * Insert again. mz->usage_in_excess will be updated. | |
705 | * If excess is 0, no tree ops. | |
706 | */ | |
cf2c8127 | 707 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 708 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
709 | } |
710 | } | |
711 | } | |
712 | ||
713 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
714 | { | |
ef8f2327 MG |
715 | struct mem_cgroup_tree_per_node *mctz; |
716 | struct mem_cgroup_per_node *mz; | |
717 | int nid; | |
bb4cc1a8 | 718 | |
e231875b | 719 | for_each_node(nid) { |
ef8f2327 MG |
720 | mz = mem_cgroup_nodeinfo(memcg, nid); |
721 | mctz = soft_limit_tree_node(nid); | |
bfc7228b LD |
722 | if (mctz) |
723 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
724 | } |
725 | } | |
726 | ||
ef8f2327 MG |
727 | static struct mem_cgroup_per_node * |
728 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 729 | { |
ef8f2327 | 730 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
731 | |
732 | retry: | |
733 | mz = NULL; | |
fa90b2fd | 734 | if (!mctz->rb_rightmost) |
bb4cc1a8 AM |
735 | goto done; /* Nothing to reclaim from */ |
736 | ||
fa90b2fd DB |
737 | mz = rb_entry(mctz->rb_rightmost, |
738 | struct mem_cgroup_per_node, tree_node); | |
bb4cc1a8 AM |
739 | /* |
740 | * Remove the node now but someone else can add it back, | |
741 | * we will to add it back at the end of reclaim to its correct | |
742 | * position in the tree. | |
743 | */ | |
cf2c8127 | 744 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 745 | if (!soft_limit_excess(mz->memcg) || |
8965aa28 | 746 | !css_tryget(&mz->memcg->css)) |
bb4cc1a8 AM |
747 | goto retry; |
748 | done: | |
749 | return mz; | |
750 | } | |
751 | ||
ef8f2327 MG |
752 | static struct mem_cgroup_per_node * |
753 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 754 | { |
ef8f2327 | 755 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 756 | |
0a31bc97 | 757 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 758 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 759 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
760 | return mz; |
761 | } | |
762 | ||
db9adbcb JW |
763 | /** |
764 | * __mod_memcg_state - update cgroup memory statistics | |
765 | * @memcg: the memory cgroup | |
766 | * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item | |
767 | * @val: delta to add to the counter, can be negative | |
768 | */ | |
769 | void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val) | |
770 | { | |
ea426c2a | 771 | long x, threshold = MEMCG_CHARGE_BATCH; |
db9adbcb JW |
772 | |
773 | if (mem_cgroup_disabled()) | |
774 | return; | |
775 | ||
772616b0 | 776 | if (memcg_stat_item_in_bytes(idx)) |
ea426c2a RG |
777 | threshold <<= PAGE_SHIFT; |
778 | ||
db9adbcb | 779 | x = val + __this_cpu_read(memcg->vmstats_percpu->stat[idx]); |
ea426c2a | 780 | if (unlikely(abs(x) > threshold)) { |
42a30035 JW |
781 | struct mem_cgroup *mi; |
782 | ||
766a4c19 YS |
783 | /* |
784 | * Batch local counters to keep them in sync with | |
785 | * the hierarchical ones. | |
786 | */ | |
787 | __this_cpu_add(memcg->vmstats_local->stat[idx], x); | |
42a30035 JW |
788 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
789 | atomic_long_add(x, &mi->vmstats[idx]); | |
db9adbcb JW |
790 | x = 0; |
791 | } | |
792 | __this_cpu_write(memcg->vmstats_percpu->stat[idx], x); | |
793 | } | |
794 | ||
42a30035 JW |
795 | static struct mem_cgroup_per_node * |
796 | parent_nodeinfo(struct mem_cgroup_per_node *pn, int nid) | |
797 | { | |
798 | struct mem_cgroup *parent; | |
799 | ||
800 | parent = parent_mem_cgroup(pn->memcg); | |
801 | if (!parent) | |
802 | return NULL; | |
803 | return mem_cgroup_nodeinfo(parent, nid); | |
804 | } | |
805 | ||
eedc4e5a RG |
806 | void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, |
807 | int val) | |
db9adbcb JW |
808 | { |
809 | struct mem_cgroup_per_node *pn; | |
42a30035 | 810 | struct mem_cgroup *memcg; |
ea426c2a | 811 | long x, threshold = MEMCG_CHARGE_BATCH; |
db9adbcb | 812 | |
db9adbcb | 813 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
42a30035 | 814 | memcg = pn->memcg; |
db9adbcb JW |
815 | |
816 | /* Update memcg */ | |
42a30035 | 817 | __mod_memcg_state(memcg, idx, val); |
db9adbcb | 818 | |
b4c46484 RG |
819 | /* Update lruvec */ |
820 | __this_cpu_add(pn->lruvec_stat_local->count[idx], val); | |
821 | ||
ea426c2a RG |
822 | if (vmstat_item_in_bytes(idx)) |
823 | threshold <<= PAGE_SHIFT; | |
824 | ||
db9adbcb | 825 | x = val + __this_cpu_read(pn->lruvec_stat_cpu->count[idx]); |
ea426c2a | 826 | if (unlikely(abs(x) > threshold)) { |
eedc4e5a | 827 | pg_data_t *pgdat = lruvec_pgdat(lruvec); |
42a30035 JW |
828 | struct mem_cgroup_per_node *pi; |
829 | ||
42a30035 JW |
830 | for (pi = pn; pi; pi = parent_nodeinfo(pi, pgdat->node_id)) |
831 | atomic_long_add(x, &pi->lruvec_stat[idx]); | |
db9adbcb JW |
832 | x = 0; |
833 | } | |
834 | __this_cpu_write(pn->lruvec_stat_cpu->count[idx], x); | |
835 | } | |
836 | ||
eedc4e5a RG |
837 | /** |
838 | * __mod_lruvec_state - update lruvec memory statistics | |
839 | * @lruvec: the lruvec | |
840 | * @idx: the stat item | |
841 | * @val: delta to add to the counter, can be negative | |
842 | * | |
843 | * The lruvec is the intersection of the NUMA node and a cgroup. This | |
844 | * function updates the all three counters that are affected by a | |
845 | * change of state at this level: per-node, per-cgroup, per-lruvec. | |
846 | */ | |
847 | void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, | |
848 | int val) | |
849 | { | |
850 | /* Update node */ | |
851 | __mod_node_page_state(lruvec_pgdat(lruvec), idx, val); | |
852 | ||
853 | /* Update memcg and lruvec */ | |
854 | if (!mem_cgroup_disabled()) | |
855 | __mod_memcg_lruvec_state(lruvec, idx, val); | |
856 | } | |
857 | ||
ec9f0238 RG |
858 | void __mod_lruvec_slab_state(void *p, enum node_stat_item idx, int val) |
859 | { | |
4f103c63 | 860 | pg_data_t *pgdat = page_pgdat(virt_to_page(p)); |
ec9f0238 RG |
861 | struct mem_cgroup *memcg; |
862 | struct lruvec *lruvec; | |
863 | ||
864 | rcu_read_lock(); | |
4f103c63 | 865 | memcg = mem_cgroup_from_obj(p); |
ec9f0238 RG |
866 | |
867 | /* Untracked pages have no memcg, no lruvec. Update only the node */ | |
868 | if (!memcg || memcg == root_mem_cgroup) { | |
869 | __mod_node_page_state(pgdat, idx, val); | |
870 | } else { | |
867e5e1d | 871 | lruvec = mem_cgroup_lruvec(memcg, pgdat); |
ec9f0238 RG |
872 | __mod_lruvec_state(lruvec, idx, val); |
873 | } | |
874 | rcu_read_unlock(); | |
875 | } | |
876 | ||
8380ce47 RG |
877 | void mod_memcg_obj_state(void *p, int idx, int val) |
878 | { | |
879 | struct mem_cgroup *memcg; | |
880 | ||
881 | rcu_read_lock(); | |
882 | memcg = mem_cgroup_from_obj(p); | |
883 | if (memcg) | |
884 | mod_memcg_state(memcg, idx, val); | |
885 | rcu_read_unlock(); | |
886 | } | |
887 | ||
db9adbcb JW |
888 | /** |
889 | * __count_memcg_events - account VM events in a cgroup | |
890 | * @memcg: the memory cgroup | |
891 | * @idx: the event item | |
892 | * @count: the number of events that occured | |
893 | */ | |
894 | void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, | |
895 | unsigned long count) | |
896 | { | |
897 | unsigned long x; | |
898 | ||
899 | if (mem_cgroup_disabled()) | |
900 | return; | |
901 | ||
902 | x = count + __this_cpu_read(memcg->vmstats_percpu->events[idx]); | |
903 | if (unlikely(x > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
904 | struct mem_cgroup *mi; |
905 | ||
766a4c19 YS |
906 | /* |
907 | * Batch local counters to keep them in sync with | |
908 | * the hierarchical ones. | |
909 | */ | |
910 | __this_cpu_add(memcg->vmstats_local->events[idx], x); | |
42a30035 JW |
911 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
912 | atomic_long_add(x, &mi->vmevents[idx]); | |
db9adbcb JW |
913 | x = 0; |
914 | } | |
915 | __this_cpu_write(memcg->vmstats_percpu->events[idx], x); | |
916 | } | |
917 | ||
42a30035 | 918 | static unsigned long memcg_events(struct mem_cgroup *memcg, int event) |
e9f8974f | 919 | { |
871789d4 | 920 | return atomic_long_read(&memcg->vmevents[event]); |
e9f8974f JW |
921 | } |
922 | ||
42a30035 JW |
923 | static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event) |
924 | { | |
815744d7 JW |
925 | long x = 0; |
926 | int cpu; | |
927 | ||
928 | for_each_possible_cpu(cpu) | |
929 | x += per_cpu(memcg->vmstats_local->events[event], cpu); | |
930 | return x; | |
42a30035 JW |
931 | } |
932 | ||
c0ff4b85 | 933 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 934 | struct page *page, |
3fba69a5 | 935 | int nr_pages) |
d52aa412 | 936 | { |
e401f176 KH |
937 | /* pagein of a big page is an event. So, ignore page size */ |
938 | if (nr_pages > 0) | |
c9019e9b | 939 | __count_memcg_events(memcg, PGPGIN, 1); |
3751d604 | 940 | else { |
c9019e9b | 941 | __count_memcg_events(memcg, PGPGOUT, 1); |
3751d604 KH |
942 | nr_pages = -nr_pages; /* for event */ |
943 | } | |
e401f176 | 944 | |
871789d4 | 945 | __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages); |
6d12e2d8 KH |
946 | } |
947 | ||
f53d7ce3 JW |
948 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
949 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
950 | { |
951 | unsigned long val, next; | |
952 | ||
871789d4 CD |
953 | val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events); |
954 | next = __this_cpu_read(memcg->vmstats_percpu->targets[target]); | |
7a159cc9 | 955 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 956 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
957 | switch (target) { |
958 | case MEM_CGROUP_TARGET_THRESH: | |
959 | next = val + THRESHOLDS_EVENTS_TARGET; | |
960 | break; | |
bb4cc1a8 AM |
961 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
962 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
963 | break; | |
f53d7ce3 JW |
964 | default: |
965 | break; | |
966 | } | |
871789d4 | 967 | __this_cpu_write(memcg->vmstats_percpu->targets[target], next); |
f53d7ce3 | 968 | return true; |
7a159cc9 | 969 | } |
f53d7ce3 | 970 | return false; |
d2265e6f KH |
971 | } |
972 | ||
973 | /* | |
974 | * Check events in order. | |
975 | * | |
976 | */ | |
c0ff4b85 | 977 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
978 | { |
979 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
980 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
981 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 982 | bool do_softlimit; |
f53d7ce3 | 983 | |
bb4cc1a8 AM |
984 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
985 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
c0ff4b85 | 986 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
987 | if (unlikely(do_softlimit)) |
988 | mem_cgroup_update_tree(memcg, page); | |
0a31bc97 | 989 | } |
d2265e6f KH |
990 | } |
991 | ||
cf475ad2 | 992 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 993 | { |
31a78f23 BS |
994 | /* |
995 | * mm_update_next_owner() may clear mm->owner to NULL | |
996 | * if it races with swapoff, page migration, etc. | |
997 | * So this can be called with p == NULL. | |
998 | */ | |
999 | if (unlikely(!p)) | |
1000 | return NULL; | |
1001 | ||
073219e9 | 1002 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 1003 | } |
33398cf2 | 1004 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 1005 | |
d46eb14b SB |
1006 | /** |
1007 | * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg. | |
1008 | * @mm: mm from which memcg should be extracted. It can be NULL. | |
1009 | * | |
1010 | * Obtain a reference on mm->memcg and returns it if successful. Otherwise | |
1011 | * root_mem_cgroup is returned. However if mem_cgroup is disabled, NULL is | |
1012 | * returned. | |
1013 | */ | |
1014 | struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) | |
54595fe2 | 1015 | { |
d46eb14b SB |
1016 | struct mem_cgroup *memcg; |
1017 | ||
1018 | if (mem_cgroup_disabled()) | |
1019 | return NULL; | |
0b7f569e | 1020 | |
54595fe2 KH |
1021 | rcu_read_lock(); |
1022 | do { | |
6f6acb00 MH |
1023 | /* |
1024 | * Page cache insertions can happen withou an | |
1025 | * actual mm context, e.g. during disk probing | |
1026 | * on boot, loopback IO, acct() writes etc. | |
1027 | */ | |
1028 | if (unlikely(!mm)) | |
df381975 | 1029 | memcg = root_mem_cgroup; |
6f6acb00 MH |
1030 | else { |
1031 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
1032 | if (unlikely(!memcg)) | |
1033 | memcg = root_mem_cgroup; | |
1034 | } | |
00d484f3 | 1035 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 1036 | rcu_read_unlock(); |
c0ff4b85 | 1037 | return memcg; |
54595fe2 | 1038 | } |
d46eb14b SB |
1039 | EXPORT_SYMBOL(get_mem_cgroup_from_mm); |
1040 | ||
f745c6f5 SB |
1041 | /** |
1042 | * get_mem_cgroup_from_page: Obtain a reference on given page's memcg. | |
1043 | * @page: page from which memcg should be extracted. | |
1044 | * | |
1045 | * Obtain a reference on page->memcg and returns it if successful. Otherwise | |
1046 | * root_mem_cgroup is returned. | |
1047 | */ | |
1048 | struct mem_cgroup *get_mem_cgroup_from_page(struct page *page) | |
1049 | { | |
1050 | struct mem_cgroup *memcg = page->mem_cgroup; | |
1051 | ||
1052 | if (mem_cgroup_disabled()) | |
1053 | return NULL; | |
1054 | ||
1055 | rcu_read_lock(); | |
8965aa28 SB |
1056 | /* Page should not get uncharged and freed memcg under us. */ |
1057 | if (!memcg || WARN_ON_ONCE(!css_tryget(&memcg->css))) | |
f745c6f5 SB |
1058 | memcg = root_mem_cgroup; |
1059 | rcu_read_unlock(); | |
1060 | return memcg; | |
1061 | } | |
1062 | EXPORT_SYMBOL(get_mem_cgroup_from_page); | |
1063 | ||
d46eb14b SB |
1064 | /** |
1065 | * If current->active_memcg is non-NULL, do not fallback to current->mm->memcg. | |
1066 | */ | |
1067 | static __always_inline struct mem_cgroup *get_mem_cgroup_from_current(void) | |
1068 | { | |
1069 | if (unlikely(current->active_memcg)) { | |
8965aa28 | 1070 | struct mem_cgroup *memcg; |
d46eb14b SB |
1071 | |
1072 | rcu_read_lock(); | |
8965aa28 SB |
1073 | /* current->active_memcg must hold a ref. */ |
1074 | if (WARN_ON_ONCE(!css_tryget(¤t->active_memcg->css))) | |
1075 | memcg = root_mem_cgroup; | |
1076 | else | |
d46eb14b SB |
1077 | memcg = current->active_memcg; |
1078 | rcu_read_unlock(); | |
1079 | return memcg; | |
1080 | } | |
1081 | return get_mem_cgroup_from_mm(current->mm); | |
1082 | } | |
54595fe2 | 1083 | |
5660048c JW |
1084 | /** |
1085 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1086 | * @root: hierarchy root | |
1087 | * @prev: previously returned memcg, NULL on first invocation | |
1088 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1089 | * | |
1090 | * Returns references to children of the hierarchy below @root, or | |
1091 | * @root itself, or %NULL after a full round-trip. | |
1092 | * | |
1093 | * Caller must pass the return value in @prev on subsequent | |
1094 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1095 | * to cancel a hierarchy walk before the round-trip is complete. | |
1096 | * | |
05bdc520 ML |
1097 | * Reclaimers can specify a node in @reclaim to divide up the memcgs |
1098 | * in the hierarchy among all concurrent reclaimers operating on the | |
1099 | * same node. | |
5660048c | 1100 | */ |
694fbc0f | 1101 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1102 | struct mem_cgroup *prev, |
694fbc0f | 1103 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1104 | { |
3f649ab7 | 1105 | struct mem_cgroup_reclaim_iter *iter; |
5ac8fb31 | 1106 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 1107 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1108 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1109 | |
694fbc0f AM |
1110 | if (mem_cgroup_disabled()) |
1111 | return NULL; | |
5660048c | 1112 | |
9f3a0d09 JW |
1113 | if (!root) |
1114 | root = root_mem_cgroup; | |
7d74b06f | 1115 | |
9f3a0d09 | 1116 | if (prev && !reclaim) |
5ac8fb31 | 1117 | pos = prev; |
14067bb3 | 1118 | |
9f3a0d09 JW |
1119 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1120 | if (prev) | |
5ac8fb31 | 1121 | goto out; |
694fbc0f | 1122 | return root; |
9f3a0d09 | 1123 | } |
14067bb3 | 1124 | |
542f85f9 | 1125 | rcu_read_lock(); |
5f578161 | 1126 | |
5ac8fb31 | 1127 | if (reclaim) { |
ef8f2327 | 1128 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 1129 | |
ef8f2327 | 1130 | mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id); |
9da83f3f | 1131 | iter = &mz->iter; |
5ac8fb31 JW |
1132 | |
1133 | if (prev && reclaim->generation != iter->generation) | |
1134 | goto out_unlock; | |
1135 | ||
6df38689 | 1136 | while (1) { |
4db0c3c2 | 1137 | pos = READ_ONCE(iter->position); |
6df38689 VD |
1138 | if (!pos || css_tryget(&pos->css)) |
1139 | break; | |
5ac8fb31 | 1140 | /* |
6df38689 VD |
1141 | * css reference reached zero, so iter->position will |
1142 | * be cleared by ->css_released. However, we should not | |
1143 | * rely on this happening soon, because ->css_released | |
1144 | * is called from a work queue, and by busy-waiting we | |
1145 | * might block it. So we clear iter->position right | |
1146 | * away. | |
5ac8fb31 | 1147 | */ |
6df38689 VD |
1148 | (void)cmpxchg(&iter->position, pos, NULL); |
1149 | } | |
5ac8fb31 JW |
1150 | } |
1151 | ||
1152 | if (pos) | |
1153 | css = &pos->css; | |
1154 | ||
1155 | for (;;) { | |
1156 | css = css_next_descendant_pre(css, &root->css); | |
1157 | if (!css) { | |
1158 | /* | |
1159 | * Reclaimers share the hierarchy walk, and a | |
1160 | * new one might jump in right at the end of | |
1161 | * the hierarchy - make sure they see at least | |
1162 | * one group and restart from the beginning. | |
1163 | */ | |
1164 | if (!prev) | |
1165 | continue; | |
1166 | break; | |
527a5ec9 | 1167 | } |
7d74b06f | 1168 | |
5ac8fb31 JW |
1169 | /* |
1170 | * Verify the css and acquire a reference. The root | |
1171 | * is provided by the caller, so we know it's alive | |
1172 | * and kicking, and don't take an extra reference. | |
1173 | */ | |
1174 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 1175 | |
5ac8fb31 JW |
1176 | if (css == &root->css) |
1177 | break; | |
14067bb3 | 1178 | |
0b8f73e1 JW |
1179 | if (css_tryget(css)) |
1180 | break; | |
9f3a0d09 | 1181 | |
5ac8fb31 | 1182 | memcg = NULL; |
9f3a0d09 | 1183 | } |
5ac8fb31 JW |
1184 | |
1185 | if (reclaim) { | |
5ac8fb31 | 1186 | /* |
6df38689 VD |
1187 | * The position could have already been updated by a competing |
1188 | * thread, so check that the value hasn't changed since we read | |
1189 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 1190 | */ |
6df38689 VD |
1191 | (void)cmpxchg(&iter->position, pos, memcg); |
1192 | ||
5ac8fb31 JW |
1193 | if (pos) |
1194 | css_put(&pos->css); | |
1195 | ||
1196 | if (!memcg) | |
1197 | iter->generation++; | |
1198 | else if (!prev) | |
1199 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1200 | } |
5ac8fb31 | 1201 | |
542f85f9 MH |
1202 | out_unlock: |
1203 | rcu_read_unlock(); | |
5ac8fb31 | 1204 | out: |
c40046f3 MH |
1205 | if (prev && prev != root) |
1206 | css_put(&prev->css); | |
1207 | ||
9f3a0d09 | 1208 | return memcg; |
14067bb3 | 1209 | } |
7d74b06f | 1210 | |
5660048c JW |
1211 | /** |
1212 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1213 | * @root: hierarchy root | |
1214 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1215 | */ | |
1216 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1217 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1218 | { |
1219 | if (!root) | |
1220 | root = root_mem_cgroup; | |
1221 | if (prev && prev != root) | |
1222 | css_put(&prev->css); | |
1223 | } | |
7d74b06f | 1224 | |
54a83d6b MC |
1225 | static void __invalidate_reclaim_iterators(struct mem_cgroup *from, |
1226 | struct mem_cgroup *dead_memcg) | |
6df38689 | 1227 | { |
6df38689 | 1228 | struct mem_cgroup_reclaim_iter *iter; |
ef8f2327 MG |
1229 | struct mem_cgroup_per_node *mz; |
1230 | int nid; | |
6df38689 | 1231 | |
54a83d6b MC |
1232 | for_each_node(nid) { |
1233 | mz = mem_cgroup_nodeinfo(from, nid); | |
9da83f3f YS |
1234 | iter = &mz->iter; |
1235 | cmpxchg(&iter->position, dead_memcg, NULL); | |
6df38689 VD |
1236 | } |
1237 | } | |
1238 | ||
54a83d6b MC |
1239 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1240 | { | |
1241 | struct mem_cgroup *memcg = dead_memcg; | |
1242 | struct mem_cgroup *last; | |
1243 | ||
1244 | do { | |
1245 | __invalidate_reclaim_iterators(memcg, dead_memcg); | |
1246 | last = memcg; | |
1247 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1248 | ||
1249 | /* | |
1250 | * When cgruop1 non-hierarchy mode is used, | |
1251 | * parent_mem_cgroup() does not walk all the way up to the | |
1252 | * cgroup root (root_mem_cgroup). So we have to handle | |
1253 | * dead_memcg from cgroup root separately. | |
1254 | */ | |
1255 | if (last != root_mem_cgroup) | |
1256 | __invalidate_reclaim_iterators(root_mem_cgroup, | |
1257 | dead_memcg); | |
1258 | } | |
1259 | ||
7c5f64f8 VD |
1260 | /** |
1261 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1262 | * @memcg: hierarchy root | |
1263 | * @fn: function to call for each task | |
1264 | * @arg: argument passed to @fn | |
1265 | * | |
1266 | * This function iterates over tasks attached to @memcg or to any of its | |
1267 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
1268 | * value, the function breaks the iteration loop and returns the value. | |
1269 | * Otherwise, it will iterate over all tasks and return 0. | |
1270 | * | |
1271 | * This function must not be called for the root memory cgroup. | |
1272 | */ | |
1273 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
1274 | int (*fn)(struct task_struct *, void *), void *arg) | |
1275 | { | |
1276 | struct mem_cgroup *iter; | |
1277 | int ret = 0; | |
1278 | ||
1279 | BUG_ON(memcg == root_mem_cgroup); | |
1280 | ||
1281 | for_each_mem_cgroup_tree(iter, memcg) { | |
1282 | struct css_task_iter it; | |
1283 | struct task_struct *task; | |
1284 | ||
f168a9a5 | 1285 | css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it); |
7c5f64f8 VD |
1286 | while (!ret && (task = css_task_iter_next(&it))) |
1287 | ret = fn(task, arg); | |
1288 | css_task_iter_end(&it); | |
1289 | if (ret) { | |
1290 | mem_cgroup_iter_break(memcg, iter); | |
1291 | break; | |
1292 | } | |
1293 | } | |
1294 | return ret; | |
1295 | } | |
1296 | ||
925b7673 | 1297 | /** |
dfe0e773 | 1298 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1299 | * @page: the page |
f144c390 | 1300 | * @pgdat: pgdat of the page |
dfe0e773 | 1301 | * |
a0b5b414 JW |
1302 | * This function relies on page->mem_cgroup being stable - see the |
1303 | * access rules in commit_charge(). | |
925b7673 | 1304 | */ |
599d0c95 | 1305 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat) |
08e552c6 | 1306 | { |
ef8f2327 | 1307 | struct mem_cgroup_per_node *mz; |
925b7673 | 1308 | struct mem_cgroup *memcg; |
bea8c150 | 1309 | struct lruvec *lruvec; |
6d12e2d8 | 1310 | |
bea8c150 | 1311 | if (mem_cgroup_disabled()) { |
867e5e1d | 1312 | lruvec = &pgdat->__lruvec; |
bea8c150 HD |
1313 | goto out; |
1314 | } | |
925b7673 | 1315 | |
1306a85a | 1316 | memcg = page->mem_cgroup; |
7512102c | 1317 | /* |
dfe0e773 | 1318 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1319 | * possibly migrated - before they are charged. |
7512102c | 1320 | */ |
29833315 JW |
1321 | if (!memcg) |
1322 | memcg = root_mem_cgroup; | |
7512102c | 1323 | |
ef8f2327 | 1324 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
bea8c150 HD |
1325 | lruvec = &mz->lruvec; |
1326 | out: | |
1327 | /* | |
1328 | * Since a node can be onlined after the mem_cgroup was created, | |
1329 | * we have to be prepared to initialize lruvec->zone here; | |
1330 | * and if offlined then reonlined, we need to reinitialize it. | |
1331 | */ | |
599d0c95 MG |
1332 | if (unlikely(lruvec->pgdat != pgdat)) |
1333 | lruvec->pgdat = pgdat; | |
bea8c150 | 1334 | return lruvec; |
08e552c6 | 1335 | } |
b69408e8 | 1336 | |
925b7673 | 1337 | /** |
fa9add64 HD |
1338 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1339 | * @lruvec: mem_cgroup per zone lru vector | |
1340 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1341 | * @zid: zone id of the accounted pages |
fa9add64 | 1342 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1343 | * |
ca707239 HD |
1344 | * This function must be called under lru_lock, just before a page is added |
1345 | * to or just after a page is removed from an lru list (that ordering being | |
1346 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 1347 | */ |
fa9add64 | 1348 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1349 | int zid, int nr_pages) |
3f58a829 | 1350 | { |
ef8f2327 | 1351 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1352 | unsigned long *lru_size; |
ca707239 | 1353 | long size; |
3f58a829 MK |
1354 | |
1355 | if (mem_cgroup_disabled()) | |
1356 | return; | |
1357 | ||
ef8f2327 | 1358 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1359 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1360 | |
1361 | if (nr_pages < 0) | |
1362 | *lru_size += nr_pages; | |
1363 | ||
1364 | size = *lru_size; | |
b4536f0c MH |
1365 | if (WARN_ONCE(size < 0, |
1366 | "%s(%p, %d, %d): lru_size %ld\n", | |
1367 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1368 | VM_BUG_ON(1); |
1369 | *lru_size = 0; | |
1370 | } | |
1371 | ||
1372 | if (nr_pages > 0) | |
1373 | *lru_size += nr_pages; | |
08e552c6 | 1374 | } |
544122e5 | 1375 | |
19942822 | 1376 | /** |
9d11ea9f | 1377 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1378 | * @memcg: the memory cgroup |
19942822 | 1379 | * |
9d11ea9f | 1380 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1381 | * pages. |
19942822 | 1382 | */ |
c0ff4b85 | 1383 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1384 | { |
3e32cb2e JW |
1385 | unsigned long margin = 0; |
1386 | unsigned long count; | |
1387 | unsigned long limit; | |
9d11ea9f | 1388 | |
3e32cb2e | 1389 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1390 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1391 | if (count < limit) |
1392 | margin = limit - count; | |
1393 | ||
7941d214 | 1394 | if (do_memsw_account()) { |
3e32cb2e | 1395 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1396 | limit = READ_ONCE(memcg->memsw.max); |
1c4448ed | 1397 | if (count < limit) |
3e32cb2e | 1398 | margin = min(margin, limit - count); |
cbedbac3 LR |
1399 | else |
1400 | margin = 0; | |
3e32cb2e JW |
1401 | } |
1402 | ||
1403 | return margin; | |
19942822 JW |
1404 | } |
1405 | ||
32047e2a | 1406 | /* |
bdcbb659 | 1407 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1408 | * |
bdcbb659 QH |
1409 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1410 | * moving cgroups. This is for waiting at high-memory pressure | |
1411 | * caused by "move". | |
32047e2a | 1412 | */ |
c0ff4b85 | 1413 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1414 | { |
2bd9bb20 KH |
1415 | struct mem_cgroup *from; |
1416 | struct mem_cgroup *to; | |
4b534334 | 1417 | bool ret = false; |
2bd9bb20 KH |
1418 | /* |
1419 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1420 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1421 | */ | |
1422 | spin_lock(&mc.lock); | |
1423 | from = mc.from; | |
1424 | to = mc.to; | |
1425 | if (!from) | |
1426 | goto unlock; | |
3e92041d | 1427 | |
2314b42d JW |
1428 | ret = mem_cgroup_is_descendant(from, memcg) || |
1429 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1430 | unlock: |
1431 | spin_unlock(&mc.lock); | |
4b534334 KH |
1432 | return ret; |
1433 | } | |
1434 | ||
c0ff4b85 | 1435 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1436 | { |
1437 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1438 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1439 | DEFINE_WAIT(wait); |
1440 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1441 | /* moving charge context might have finished. */ | |
1442 | if (mc.moving_task) | |
1443 | schedule(); | |
1444 | finish_wait(&mc.waitq, &wait); | |
1445 | return true; | |
1446 | } | |
1447 | } | |
1448 | return false; | |
1449 | } | |
1450 | ||
5f9a4f4a MS |
1451 | struct memory_stat { |
1452 | const char *name; | |
1453 | unsigned int ratio; | |
1454 | unsigned int idx; | |
1455 | }; | |
1456 | ||
1457 | static struct memory_stat memory_stats[] = { | |
1458 | { "anon", PAGE_SIZE, NR_ANON_MAPPED }, | |
1459 | { "file", PAGE_SIZE, NR_FILE_PAGES }, | |
1460 | { "kernel_stack", 1024, NR_KERNEL_STACK_KB }, | |
1461 | { "percpu", 1, MEMCG_PERCPU_B }, | |
1462 | { "sock", PAGE_SIZE, MEMCG_SOCK }, | |
1463 | { "shmem", PAGE_SIZE, NR_SHMEM }, | |
1464 | { "file_mapped", PAGE_SIZE, NR_FILE_MAPPED }, | |
1465 | { "file_dirty", PAGE_SIZE, NR_FILE_DIRTY }, | |
1466 | { "file_writeback", PAGE_SIZE, NR_WRITEBACK }, | |
1467 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
1468 | /* | |
1469 | * The ratio will be initialized in memory_stats_init(). Because | |
1470 | * on some architectures, the macro of HPAGE_PMD_SIZE is not | |
1471 | * constant(e.g. powerpc). | |
1472 | */ | |
1473 | { "anon_thp", 0, NR_ANON_THPS }, | |
1474 | #endif | |
1475 | { "inactive_anon", PAGE_SIZE, NR_INACTIVE_ANON }, | |
1476 | { "active_anon", PAGE_SIZE, NR_ACTIVE_ANON }, | |
1477 | { "inactive_file", PAGE_SIZE, NR_INACTIVE_FILE }, | |
1478 | { "active_file", PAGE_SIZE, NR_ACTIVE_FILE }, | |
1479 | { "unevictable", PAGE_SIZE, NR_UNEVICTABLE }, | |
1480 | ||
1481 | /* | |
1482 | * Note: The slab_reclaimable and slab_unreclaimable must be | |
1483 | * together and slab_reclaimable must be in front. | |
1484 | */ | |
1485 | { "slab_reclaimable", 1, NR_SLAB_RECLAIMABLE_B }, | |
1486 | { "slab_unreclaimable", 1, NR_SLAB_UNRECLAIMABLE_B }, | |
1487 | ||
1488 | /* The memory events */ | |
1489 | { "workingset_refault_anon", 1, WORKINGSET_REFAULT_ANON }, | |
1490 | { "workingset_refault_file", 1, WORKINGSET_REFAULT_FILE }, | |
1491 | { "workingset_activate_anon", 1, WORKINGSET_ACTIVATE_ANON }, | |
1492 | { "workingset_activate_file", 1, WORKINGSET_ACTIVATE_FILE }, | |
1493 | { "workingset_restore_anon", 1, WORKINGSET_RESTORE_ANON }, | |
1494 | { "workingset_restore_file", 1, WORKINGSET_RESTORE_FILE }, | |
1495 | { "workingset_nodereclaim", 1, WORKINGSET_NODERECLAIM }, | |
1496 | }; | |
1497 | ||
1498 | static int __init memory_stats_init(void) | |
1499 | { | |
1500 | int i; | |
1501 | ||
1502 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { | |
1503 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
1504 | if (memory_stats[i].idx == NR_ANON_THPS) | |
1505 | memory_stats[i].ratio = HPAGE_PMD_SIZE; | |
1506 | #endif | |
1507 | VM_BUG_ON(!memory_stats[i].ratio); | |
1508 | VM_BUG_ON(memory_stats[i].idx >= MEMCG_NR_STAT); | |
1509 | } | |
1510 | ||
1511 | return 0; | |
1512 | } | |
1513 | pure_initcall(memory_stats_init); | |
1514 | ||
c8713d0b JW |
1515 | static char *memory_stat_format(struct mem_cgroup *memcg) |
1516 | { | |
1517 | struct seq_buf s; | |
1518 | int i; | |
71cd3113 | 1519 | |
c8713d0b JW |
1520 | seq_buf_init(&s, kmalloc(PAGE_SIZE, GFP_KERNEL), PAGE_SIZE); |
1521 | if (!s.buffer) | |
1522 | return NULL; | |
1523 | ||
1524 | /* | |
1525 | * Provide statistics on the state of the memory subsystem as | |
1526 | * well as cumulative event counters that show past behavior. | |
1527 | * | |
1528 | * This list is ordered following a combination of these gradients: | |
1529 | * 1) generic big picture -> specifics and details | |
1530 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
1531 | * | |
1532 | * Current memory state: | |
1533 | */ | |
1534 | ||
5f9a4f4a MS |
1535 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { |
1536 | u64 size; | |
c8713d0b | 1537 | |
5f9a4f4a MS |
1538 | size = memcg_page_state(memcg, memory_stats[i].idx); |
1539 | size *= memory_stats[i].ratio; | |
1540 | seq_buf_printf(&s, "%s %llu\n", memory_stats[i].name, size); | |
c8713d0b | 1541 | |
5f9a4f4a MS |
1542 | if (unlikely(memory_stats[i].idx == NR_SLAB_UNRECLAIMABLE_B)) { |
1543 | size = memcg_page_state(memcg, NR_SLAB_RECLAIMABLE_B) + | |
1544 | memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE_B); | |
1545 | seq_buf_printf(&s, "slab %llu\n", size); | |
1546 | } | |
1547 | } | |
c8713d0b JW |
1548 | |
1549 | /* Accumulated memory events */ | |
1550 | ||
ebc5d83d KK |
1551 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGFAULT), |
1552 | memcg_events(memcg, PGFAULT)); | |
1553 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGMAJFAULT), | |
1554 | memcg_events(memcg, PGMAJFAULT)); | |
ebc5d83d KK |
1555 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGREFILL), |
1556 | memcg_events(memcg, PGREFILL)); | |
c8713d0b JW |
1557 | seq_buf_printf(&s, "pgscan %lu\n", |
1558 | memcg_events(memcg, PGSCAN_KSWAPD) + | |
1559 | memcg_events(memcg, PGSCAN_DIRECT)); | |
1560 | seq_buf_printf(&s, "pgsteal %lu\n", | |
1561 | memcg_events(memcg, PGSTEAL_KSWAPD) + | |
1562 | memcg_events(memcg, PGSTEAL_DIRECT)); | |
ebc5d83d KK |
1563 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGACTIVATE), |
1564 | memcg_events(memcg, PGACTIVATE)); | |
1565 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGDEACTIVATE), | |
1566 | memcg_events(memcg, PGDEACTIVATE)); | |
1567 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREE), | |
1568 | memcg_events(memcg, PGLAZYFREE)); | |
1569 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREED), | |
1570 | memcg_events(memcg, PGLAZYFREED)); | |
c8713d0b JW |
1571 | |
1572 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
ebc5d83d | 1573 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_FAULT_ALLOC), |
c8713d0b | 1574 | memcg_events(memcg, THP_FAULT_ALLOC)); |
ebc5d83d | 1575 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_COLLAPSE_ALLOC), |
c8713d0b JW |
1576 | memcg_events(memcg, THP_COLLAPSE_ALLOC)); |
1577 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ | |
1578 | ||
1579 | /* The above should easily fit into one page */ | |
1580 | WARN_ON_ONCE(seq_buf_has_overflowed(&s)); | |
1581 | ||
1582 | return s.buffer; | |
1583 | } | |
71cd3113 | 1584 | |
58cf188e | 1585 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1586 | /** |
f0c867d9 | 1587 | * mem_cgroup_print_oom_context: Print OOM information relevant to |
1588 | * memory controller. | |
e222432b BS |
1589 | * @memcg: The memory cgroup that went over limit |
1590 | * @p: Task that is going to be killed | |
1591 | * | |
1592 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1593 | * enabled | |
1594 | */ | |
f0c867d9 | 1595 | void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) |
e222432b | 1596 | { |
e222432b BS |
1597 | rcu_read_lock(); |
1598 | ||
f0c867d9 | 1599 | if (memcg) { |
1600 | pr_cont(",oom_memcg="); | |
1601 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1602 | } else | |
1603 | pr_cont(",global_oom"); | |
2415b9f5 | 1604 | if (p) { |
f0c867d9 | 1605 | pr_cont(",task_memcg="); |
2415b9f5 | 1606 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
2415b9f5 | 1607 | } |
e222432b | 1608 | rcu_read_unlock(); |
f0c867d9 | 1609 | } |
1610 | ||
1611 | /** | |
1612 | * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to | |
1613 | * memory controller. | |
1614 | * @memcg: The memory cgroup that went over limit | |
1615 | */ | |
1616 | void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) | |
1617 | { | |
c8713d0b | 1618 | char *buf; |
e222432b | 1619 | |
3e32cb2e JW |
1620 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1621 | K((u64)page_counter_read(&memcg->memory)), | |
15b42562 | 1622 | K((u64)READ_ONCE(memcg->memory.max)), memcg->memory.failcnt); |
c8713d0b JW |
1623 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
1624 | pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1625 | K((u64)page_counter_read(&memcg->swap)), | |
32d087cd | 1626 | K((u64)READ_ONCE(memcg->swap.max)), memcg->swap.failcnt); |
c8713d0b JW |
1627 | else { |
1628 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1629 | K((u64)page_counter_read(&memcg->memsw)), | |
1630 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); | |
1631 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1632 | K((u64)page_counter_read(&memcg->kmem)), | |
1633 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); | |
58cf188e | 1634 | } |
c8713d0b JW |
1635 | |
1636 | pr_info("Memory cgroup stats for "); | |
1637 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1638 | pr_cont(":"); | |
1639 | buf = memory_stat_format(memcg); | |
1640 | if (!buf) | |
1641 | return; | |
1642 | pr_info("%s", buf); | |
1643 | kfree(buf); | |
e222432b BS |
1644 | } |
1645 | ||
a63d83f4 DR |
1646 | /* |
1647 | * Return the memory (and swap, if configured) limit for a memcg. | |
1648 | */ | |
bbec2e15 | 1649 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1650 | { |
8d387a5f WL |
1651 | unsigned long max = READ_ONCE(memcg->memory.max); |
1652 | ||
1653 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) { | |
1654 | if (mem_cgroup_swappiness(memcg)) | |
1655 | max += min(READ_ONCE(memcg->swap.max), | |
1656 | (unsigned long)total_swap_pages); | |
1657 | } else { /* v1 */ | |
1658 | if (mem_cgroup_swappiness(memcg)) { | |
1659 | /* Calculate swap excess capacity from memsw limit */ | |
1660 | unsigned long swap = READ_ONCE(memcg->memsw.max) - max; | |
1661 | ||
1662 | max += min(swap, (unsigned long)total_swap_pages); | |
1663 | } | |
9a5a8f19 | 1664 | } |
bbec2e15 | 1665 | return max; |
a63d83f4 DR |
1666 | } |
1667 | ||
9783aa99 CD |
1668 | unsigned long mem_cgroup_size(struct mem_cgroup *memcg) |
1669 | { | |
1670 | return page_counter_read(&memcg->memory); | |
1671 | } | |
1672 | ||
b6e6edcf | 1673 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1674 | int order) |
9cbb78bb | 1675 | { |
6e0fc46d DR |
1676 | struct oom_control oc = { |
1677 | .zonelist = NULL, | |
1678 | .nodemask = NULL, | |
2a966b77 | 1679 | .memcg = memcg, |
6e0fc46d DR |
1680 | .gfp_mask = gfp_mask, |
1681 | .order = order, | |
6e0fc46d | 1682 | }; |
1378b37d | 1683 | bool ret = true; |
9cbb78bb | 1684 | |
7775face TH |
1685 | if (mutex_lock_killable(&oom_lock)) |
1686 | return true; | |
1378b37d YS |
1687 | |
1688 | if (mem_cgroup_margin(memcg) >= (1 << order)) | |
1689 | goto unlock; | |
1690 | ||
7775face TH |
1691 | /* |
1692 | * A few threads which were not waiting at mutex_lock_killable() can | |
1693 | * fail to bail out. Therefore, check again after holding oom_lock. | |
1694 | */ | |
1695 | ret = should_force_charge() || out_of_memory(&oc); | |
1378b37d YS |
1696 | |
1697 | unlock: | |
dc56401f | 1698 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1699 | return ret; |
9cbb78bb DR |
1700 | } |
1701 | ||
0608f43d | 1702 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1703 | pg_data_t *pgdat, |
0608f43d AM |
1704 | gfp_t gfp_mask, |
1705 | unsigned long *total_scanned) | |
1706 | { | |
1707 | struct mem_cgroup *victim = NULL; | |
1708 | int total = 0; | |
1709 | int loop = 0; | |
1710 | unsigned long excess; | |
1711 | unsigned long nr_scanned; | |
1712 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1713 | .pgdat = pgdat, |
0608f43d AM |
1714 | }; |
1715 | ||
3e32cb2e | 1716 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1717 | |
1718 | while (1) { | |
1719 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1720 | if (!victim) { | |
1721 | loop++; | |
1722 | if (loop >= 2) { | |
1723 | /* | |
1724 | * If we have not been able to reclaim | |
1725 | * anything, it might because there are | |
1726 | * no reclaimable pages under this hierarchy | |
1727 | */ | |
1728 | if (!total) | |
1729 | break; | |
1730 | /* | |
1731 | * We want to do more targeted reclaim. | |
1732 | * excess >> 2 is not to excessive so as to | |
1733 | * reclaim too much, nor too less that we keep | |
1734 | * coming back to reclaim from this cgroup | |
1735 | */ | |
1736 | if (total >= (excess >> 2) || | |
1737 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1738 | break; | |
1739 | } | |
1740 | continue; | |
1741 | } | |
a9dd0a83 | 1742 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1743 | pgdat, &nr_scanned); |
0608f43d | 1744 | *total_scanned += nr_scanned; |
3e32cb2e | 1745 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1746 | break; |
6d61ef40 | 1747 | } |
0608f43d AM |
1748 | mem_cgroup_iter_break(root_memcg, victim); |
1749 | return total; | |
6d61ef40 BS |
1750 | } |
1751 | ||
0056f4e6 JW |
1752 | #ifdef CONFIG_LOCKDEP |
1753 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1754 | .name = "memcg_oom_lock", | |
1755 | }; | |
1756 | #endif | |
1757 | ||
fb2a6fc5 JW |
1758 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1759 | ||
867578cb KH |
1760 | /* |
1761 | * Check OOM-Killer is already running under our hierarchy. | |
1762 | * If someone is running, return false. | |
1763 | */ | |
fb2a6fc5 | 1764 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1765 | { |
79dfdacc | 1766 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1767 | |
fb2a6fc5 JW |
1768 | spin_lock(&memcg_oom_lock); |
1769 | ||
9f3a0d09 | 1770 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1771 | if (iter->oom_lock) { |
79dfdacc MH |
1772 | /* |
1773 | * this subtree of our hierarchy is already locked | |
1774 | * so we cannot give a lock. | |
1775 | */ | |
79dfdacc | 1776 | failed = iter; |
9f3a0d09 JW |
1777 | mem_cgroup_iter_break(memcg, iter); |
1778 | break; | |
23751be0 JW |
1779 | } else |
1780 | iter->oom_lock = true; | |
7d74b06f | 1781 | } |
867578cb | 1782 | |
fb2a6fc5 JW |
1783 | if (failed) { |
1784 | /* | |
1785 | * OK, we failed to lock the whole subtree so we have | |
1786 | * to clean up what we set up to the failing subtree | |
1787 | */ | |
1788 | for_each_mem_cgroup_tree(iter, memcg) { | |
1789 | if (iter == failed) { | |
1790 | mem_cgroup_iter_break(memcg, iter); | |
1791 | break; | |
1792 | } | |
1793 | iter->oom_lock = false; | |
79dfdacc | 1794 | } |
0056f4e6 JW |
1795 | } else |
1796 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1797 | |
1798 | spin_unlock(&memcg_oom_lock); | |
1799 | ||
1800 | return !failed; | |
a636b327 | 1801 | } |
0b7f569e | 1802 | |
fb2a6fc5 | 1803 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1804 | { |
7d74b06f KH |
1805 | struct mem_cgroup *iter; |
1806 | ||
fb2a6fc5 | 1807 | spin_lock(&memcg_oom_lock); |
5facae4f | 1808 | mutex_release(&memcg_oom_lock_dep_map, _RET_IP_); |
c0ff4b85 | 1809 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1810 | iter->oom_lock = false; |
fb2a6fc5 | 1811 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1812 | } |
1813 | ||
c0ff4b85 | 1814 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1815 | { |
1816 | struct mem_cgroup *iter; | |
1817 | ||
c2b42d3c | 1818 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1819 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1820 | iter->under_oom++; |
1821 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1822 | } |
1823 | ||
c0ff4b85 | 1824 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1825 | { |
1826 | struct mem_cgroup *iter; | |
1827 | ||
867578cb | 1828 | /* |
7a52d4d8 ML |
1829 | * Be careful about under_oom underflows becase a child memcg |
1830 | * could have been added after mem_cgroup_mark_under_oom. | |
867578cb | 1831 | */ |
c2b42d3c | 1832 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1833 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1834 | if (iter->under_oom > 0) |
1835 | iter->under_oom--; | |
1836 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1837 | } |
1838 | ||
867578cb KH |
1839 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1840 | ||
dc98df5a | 1841 | struct oom_wait_info { |
d79154bb | 1842 | struct mem_cgroup *memcg; |
ac6424b9 | 1843 | wait_queue_entry_t wait; |
dc98df5a KH |
1844 | }; |
1845 | ||
ac6424b9 | 1846 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1847 | unsigned mode, int sync, void *arg) |
1848 | { | |
d79154bb HD |
1849 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1850 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1851 | struct oom_wait_info *oom_wait_info; |
1852 | ||
1853 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1854 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1855 | |
2314b42d JW |
1856 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1857 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1858 | return 0; |
dc98df5a KH |
1859 | return autoremove_wake_function(wait, mode, sync, arg); |
1860 | } | |
1861 | ||
c0ff4b85 | 1862 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1863 | { |
c2b42d3c TH |
1864 | /* |
1865 | * For the following lockless ->under_oom test, the only required | |
1866 | * guarantee is that it must see the state asserted by an OOM when | |
1867 | * this function is called as a result of userland actions | |
1868 | * triggered by the notification of the OOM. This is trivially | |
1869 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1870 | * triggering notification. | |
1871 | */ | |
1872 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1873 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1874 | } |
1875 | ||
29ef680a MH |
1876 | enum oom_status { |
1877 | OOM_SUCCESS, | |
1878 | OOM_FAILED, | |
1879 | OOM_ASYNC, | |
1880 | OOM_SKIPPED | |
1881 | }; | |
1882 | ||
1883 | static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 1884 | { |
7056d3a3 MH |
1885 | enum oom_status ret; |
1886 | bool locked; | |
1887 | ||
29ef680a MH |
1888 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
1889 | return OOM_SKIPPED; | |
1890 | ||
7a1adfdd RG |
1891 | memcg_memory_event(memcg, MEMCG_OOM); |
1892 | ||
867578cb | 1893 | /* |
49426420 JW |
1894 | * We are in the middle of the charge context here, so we |
1895 | * don't want to block when potentially sitting on a callstack | |
1896 | * that holds all kinds of filesystem and mm locks. | |
1897 | * | |
29ef680a MH |
1898 | * cgroup1 allows disabling the OOM killer and waiting for outside |
1899 | * handling until the charge can succeed; remember the context and put | |
1900 | * the task to sleep at the end of the page fault when all locks are | |
1901 | * released. | |
49426420 | 1902 | * |
29ef680a MH |
1903 | * On the other hand, in-kernel OOM killer allows for an async victim |
1904 | * memory reclaim (oom_reaper) and that means that we are not solely | |
1905 | * relying on the oom victim to make a forward progress and we can | |
1906 | * invoke the oom killer here. | |
1907 | * | |
1908 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
1909 | * victim and then we have to bail out from the charge path. | |
867578cb | 1910 | */ |
29ef680a MH |
1911 | if (memcg->oom_kill_disable) { |
1912 | if (!current->in_user_fault) | |
1913 | return OOM_SKIPPED; | |
1914 | css_get(&memcg->css); | |
1915 | current->memcg_in_oom = memcg; | |
1916 | current->memcg_oom_gfp_mask = mask; | |
1917 | current->memcg_oom_order = order; | |
1918 | ||
1919 | return OOM_ASYNC; | |
1920 | } | |
1921 | ||
7056d3a3 MH |
1922 | mem_cgroup_mark_under_oom(memcg); |
1923 | ||
1924 | locked = mem_cgroup_oom_trylock(memcg); | |
1925 | ||
1926 | if (locked) | |
1927 | mem_cgroup_oom_notify(memcg); | |
1928 | ||
1929 | mem_cgroup_unmark_under_oom(memcg); | |
29ef680a | 1930 | if (mem_cgroup_out_of_memory(memcg, mask, order)) |
7056d3a3 MH |
1931 | ret = OOM_SUCCESS; |
1932 | else | |
1933 | ret = OOM_FAILED; | |
1934 | ||
1935 | if (locked) | |
1936 | mem_cgroup_oom_unlock(memcg); | |
29ef680a | 1937 | |
7056d3a3 | 1938 | return ret; |
3812c8c8 JW |
1939 | } |
1940 | ||
1941 | /** | |
1942 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1943 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1944 | * |
49426420 JW |
1945 | * This has to be called at the end of a page fault if the memcg OOM |
1946 | * handler was enabled. | |
3812c8c8 | 1947 | * |
49426420 | 1948 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1949 | * sleep on a waitqueue until the userspace task resolves the |
1950 | * situation. Sleeping directly in the charge context with all kinds | |
1951 | * of locks held is not a good idea, instead we remember an OOM state | |
1952 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1953 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1954 | * |
1955 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1956 | * completed, %false otherwise. |
3812c8c8 | 1957 | */ |
49426420 | 1958 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1959 | { |
626ebc41 | 1960 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1961 | struct oom_wait_info owait; |
49426420 | 1962 | bool locked; |
3812c8c8 JW |
1963 | |
1964 | /* OOM is global, do not handle */ | |
3812c8c8 | 1965 | if (!memcg) |
49426420 | 1966 | return false; |
3812c8c8 | 1967 | |
7c5f64f8 | 1968 | if (!handle) |
49426420 | 1969 | goto cleanup; |
3812c8c8 JW |
1970 | |
1971 | owait.memcg = memcg; | |
1972 | owait.wait.flags = 0; | |
1973 | owait.wait.func = memcg_oom_wake_function; | |
1974 | owait.wait.private = current; | |
2055da97 | 1975 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 1976 | |
3812c8c8 | 1977 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1978 | mem_cgroup_mark_under_oom(memcg); |
1979 | ||
1980 | locked = mem_cgroup_oom_trylock(memcg); | |
1981 | ||
1982 | if (locked) | |
1983 | mem_cgroup_oom_notify(memcg); | |
1984 | ||
1985 | if (locked && !memcg->oom_kill_disable) { | |
1986 | mem_cgroup_unmark_under_oom(memcg); | |
1987 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1988 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1989 | current->memcg_oom_order); | |
49426420 | 1990 | } else { |
3812c8c8 | 1991 | schedule(); |
49426420 JW |
1992 | mem_cgroup_unmark_under_oom(memcg); |
1993 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1994 | } | |
1995 | ||
1996 | if (locked) { | |
fb2a6fc5 JW |
1997 | mem_cgroup_oom_unlock(memcg); |
1998 | /* | |
1999 | * There is no guarantee that an OOM-lock contender | |
2000 | * sees the wakeups triggered by the OOM kill | |
2001 | * uncharges. Wake any sleepers explicitely. | |
2002 | */ | |
2003 | memcg_oom_recover(memcg); | |
2004 | } | |
49426420 | 2005 | cleanup: |
626ebc41 | 2006 | current->memcg_in_oom = NULL; |
3812c8c8 | 2007 | css_put(&memcg->css); |
867578cb | 2008 | return true; |
0b7f569e KH |
2009 | } |
2010 | ||
3d8b38eb RG |
2011 | /** |
2012 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
2013 | * @victim: task to be killed by the OOM killer | |
2014 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
2015 | * | |
2016 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
2017 | * by killing all belonging OOM-killable tasks. | |
2018 | * | |
2019 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
2020 | */ | |
2021 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
2022 | struct mem_cgroup *oom_domain) | |
2023 | { | |
2024 | struct mem_cgroup *oom_group = NULL; | |
2025 | struct mem_cgroup *memcg; | |
2026 | ||
2027 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
2028 | return NULL; | |
2029 | ||
2030 | if (!oom_domain) | |
2031 | oom_domain = root_mem_cgroup; | |
2032 | ||
2033 | rcu_read_lock(); | |
2034 | ||
2035 | memcg = mem_cgroup_from_task(victim); | |
2036 | if (memcg == root_mem_cgroup) | |
2037 | goto out; | |
2038 | ||
48fe267c RG |
2039 | /* |
2040 | * If the victim task has been asynchronously moved to a different | |
2041 | * memory cgroup, we might end up killing tasks outside oom_domain. | |
2042 | * In this case it's better to ignore memory.group.oom. | |
2043 | */ | |
2044 | if (unlikely(!mem_cgroup_is_descendant(memcg, oom_domain))) | |
2045 | goto out; | |
2046 | ||
3d8b38eb RG |
2047 | /* |
2048 | * Traverse the memory cgroup hierarchy from the victim task's | |
2049 | * cgroup up to the OOMing cgroup (or root) to find the | |
2050 | * highest-level memory cgroup with oom.group set. | |
2051 | */ | |
2052 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
2053 | if (memcg->oom_group) | |
2054 | oom_group = memcg; | |
2055 | ||
2056 | if (memcg == oom_domain) | |
2057 | break; | |
2058 | } | |
2059 | ||
2060 | if (oom_group) | |
2061 | css_get(&oom_group->css); | |
2062 | out: | |
2063 | rcu_read_unlock(); | |
2064 | ||
2065 | return oom_group; | |
2066 | } | |
2067 | ||
2068 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
2069 | { | |
2070 | pr_info("Tasks in "); | |
2071 | pr_cont_cgroup_path(memcg->css.cgroup); | |
2072 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
2073 | } | |
2074 | ||
d7365e78 | 2075 | /** |
81f8c3a4 JW |
2076 | * lock_page_memcg - lock a page->mem_cgroup binding |
2077 | * @page: the page | |
32047e2a | 2078 | * |
81f8c3a4 | 2079 | * This function protects unlocked LRU pages from being moved to |
739f79fc JW |
2080 | * another cgroup. |
2081 | * | |
2082 | * It ensures lifetime of the returned memcg. Caller is responsible | |
2083 | * for the lifetime of the page; __unlock_page_memcg() is available | |
2084 | * when @page might get freed inside the locked section. | |
d69b042f | 2085 | */ |
739f79fc | 2086 | struct mem_cgroup *lock_page_memcg(struct page *page) |
89c06bd5 | 2087 | { |
9da7b521 | 2088 | struct page *head = compound_head(page); /* rmap on tail pages */ |
89c06bd5 | 2089 | struct mem_cgroup *memcg; |
6de22619 | 2090 | unsigned long flags; |
89c06bd5 | 2091 | |
6de22619 JW |
2092 | /* |
2093 | * The RCU lock is held throughout the transaction. The fast | |
2094 | * path can get away without acquiring the memcg->move_lock | |
2095 | * because page moving starts with an RCU grace period. | |
739f79fc JW |
2096 | * |
2097 | * The RCU lock also protects the memcg from being freed when | |
2098 | * the page state that is going to change is the only thing | |
2099 | * preventing the page itself from being freed. E.g. writeback | |
2100 | * doesn't hold a page reference and relies on PG_writeback to | |
2101 | * keep off truncation, migration and so forth. | |
2102 | */ | |
d7365e78 JW |
2103 | rcu_read_lock(); |
2104 | ||
2105 | if (mem_cgroup_disabled()) | |
739f79fc | 2106 | return NULL; |
89c06bd5 | 2107 | again: |
9da7b521 | 2108 | memcg = head->mem_cgroup; |
29833315 | 2109 | if (unlikely(!memcg)) |
739f79fc | 2110 | return NULL; |
d7365e78 | 2111 | |
bdcbb659 | 2112 | if (atomic_read(&memcg->moving_account) <= 0) |
739f79fc | 2113 | return memcg; |
89c06bd5 | 2114 | |
6de22619 | 2115 | spin_lock_irqsave(&memcg->move_lock, flags); |
9da7b521 | 2116 | if (memcg != head->mem_cgroup) { |
6de22619 | 2117 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
2118 | goto again; |
2119 | } | |
6de22619 JW |
2120 | |
2121 | /* | |
2122 | * When charge migration first begins, we can have locked and | |
2123 | * unlocked page stat updates happening concurrently. Track | |
81f8c3a4 | 2124 | * the task who has the lock for unlock_page_memcg(). |
6de22619 JW |
2125 | */ |
2126 | memcg->move_lock_task = current; | |
2127 | memcg->move_lock_flags = flags; | |
d7365e78 | 2128 | |
739f79fc | 2129 | return memcg; |
89c06bd5 | 2130 | } |
81f8c3a4 | 2131 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 2132 | |
d7365e78 | 2133 | /** |
739f79fc JW |
2134 | * __unlock_page_memcg - unlock and unpin a memcg |
2135 | * @memcg: the memcg | |
2136 | * | |
2137 | * Unlock and unpin a memcg returned by lock_page_memcg(). | |
d7365e78 | 2138 | */ |
739f79fc | 2139 | void __unlock_page_memcg(struct mem_cgroup *memcg) |
89c06bd5 | 2140 | { |
6de22619 JW |
2141 | if (memcg && memcg->move_lock_task == current) { |
2142 | unsigned long flags = memcg->move_lock_flags; | |
2143 | ||
2144 | memcg->move_lock_task = NULL; | |
2145 | memcg->move_lock_flags = 0; | |
2146 | ||
2147 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2148 | } | |
89c06bd5 | 2149 | |
d7365e78 | 2150 | rcu_read_unlock(); |
89c06bd5 | 2151 | } |
739f79fc JW |
2152 | |
2153 | /** | |
2154 | * unlock_page_memcg - unlock a page->mem_cgroup binding | |
2155 | * @page: the page | |
2156 | */ | |
2157 | void unlock_page_memcg(struct page *page) | |
2158 | { | |
9da7b521 JW |
2159 | struct page *head = compound_head(page); |
2160 | ||
2161 | __unlock_page_memcg(head->mem_cgroup); | |
739f79fc | 2162 | } |
81f8c3a4 | 2163 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 2164 | |
cdec2e42 KH |
2165 | struct memcg_stock_pcp { |
2166 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2167 | unsigned int nr_pages; |
bf4f0599 RG |
2168 | |
2169 | #ifdef CONFIG_MEMCG_KMEM | |
2170 | struct obj_cgroup *cached_objcg; | |
2171 | unsigned int nr_bytes; | |
2172 | #endif | |
2173 | ||
cdec2e42 | 2174 | struct work_struct work; |
26fe6168 | 2175 | unsigned long flags; |
a0db00fc | 2176 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2177 | }; |
2178 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2179 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2180 | |
bf4f0599 RG |
2181 | #ifdef CONFIG_MEMCG_KMEM |
2182 | static void drain_obj_stock(struct memcg_stock_pcp *stock); | |
2183 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2184 | struct mem_cgroup *root_memcg); | |
2185 | ||
2186 | #else | |
2187 | static inline void drain_obj_stock(struct memcg_stock_pcp *stock) | |
2188 | { | |
2189 | } | |
2190 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2191 | struct mem_cgroup *root_memcg) | |
2192 | { | |
2193 | return false; | |
2194 | } | |
2195 | #endif | |
2196 | ||
a0956d54 SS |
2197 | /** |
2198 | * consume_stock: Try to consume stocked charge on this cpu. | |
2199 | * @memcg: memcg to consume from. | |
2200 | * @nr_pages: how many pages to charge. | |
2201 | * | |
2202 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2203 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2204 | * service an allocation will refill the stock. | |
2205 | * | |
2206 | * returns true if successful, false otherwise. | |
cdec2e42 | 2207 | */ |
a0956d54 | 2208 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2209 | { |
2210 | struct memcg_stock_pcp *stock; | |
db2ba40c | 2211 | unsigned long flags; |
3e32cb2e | 2212 | bool ret = false; |
cdec2e42 | 2213 | |
a983b5eb | 2214 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 2215 | return ret; |
a0956d54 | 2216 | |
db2ba40c JW |
2217 | local_irq_save(flags); |
2218 | ||
2219 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 2220 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2221 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2222 | ret = true; |
2223 | } | |
db2ba40c JW |
2224 | |
2225 | local_irq_restore(flags); | |
2226 | ||
cdec2e42 KH |
2227 | return ret; |
2228 | } | |
2229 | ||
2230 | /* | |
3e32cb2e | 2231 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2232 | */ |
2233 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2234 | { | |
2235 | struct mem_cgroup *old = stock->cached; | |
2236 | ||
1a3e1f40 JW |
2237 | if (!old) |
2238 | return; | |
2239 | ||
11c9ea4e | 2240 | if (stock->nr_pages) { |
3e32cb2e | 2241 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 2242 | if (do_memsw_account()) |
3e32cb2e | 2243 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
11c9ea4e | 2244 | stock->nr_pages = 0; |
cdec2e42 | 2245 | } |
1a3e1f40 JW |
2246 | |
2247 | css_put(&old->css); | |
cdec2e42 | 2248 | stock->cached = NULL; |
cdec2e42 KH |
2249 | } |
2250 | ||
cdec2e42 KH |
2251 | static void drain_local_stock(struct work_struct *dummy) |
2252 | { | |
db2ba40c JW |
2253 | struct memcg_stock_pcp *stock; |
2254 | unsigned long flags; | |
2255 | ||
72f0184c MH |
2256 | /* |
2257 | * The only protection from memory hotplug vs. drain_stock races is | |
2258 | * that we always operate on local CPU stock here with IRQ disabled | |
2259 | */ | |
db2ba40c JW |
2260 | local_irq_save(flags); |
2261 | ||
2262 | stock = this_cpu_ptr(&memcg_stock); | |
bf4f0599 | 2263 | drain_obj_stock(stock); |
cdec2e42 | 2264 | drain_stock(stock); |
26fe6168 | 2265 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c JW |
2266 | |
2267 | local_irq_restore(flags); | |
cdec2e42 KH |
2268 | } |
2269 | ||
2270 | /* | |
3e32cb2e | 2271 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2272 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2273 | */ |
c0ff4b85 | 2274 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2275 | { |
db2ba40c JW |
2276 | struct memcg_stock_pcp *stock; |
2277 | unsigned long flags; | |
2278 | ||
2279 | local_irq_save(flags); | |
cdec2e42 | 2280 | |
db2ba40c | 2281 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 2282 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2283 | drain_stock(stock); |
1a3e1f40 | 2284 | css_get(&memcg->css); |
c0ff4b85 | 2285 | stock->cached = memcg; |
cdec2e42 | 2286 | } |
11c9ea4e | 2287 | stock->nr_pages += nr_pages; |
db2ba40c | 2288 | |
a983b5eb | 2289 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 RG |
2290 | drain_stock(stock); |
2291 | ||
db2ba40c | 2292 | local_irq_restore(flags); |
cdec2e42 KH |
2293 | } |
2294 | ||
2295 | /* | |
c0ff4b85 | 2296 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2297 | * of the hierarchy under it. |
cdec2e42 | 2298 | */ |
6d3d6aa2 | 2299 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2300 | { |
26fe6168 | 2301 | int cpu, curcpu; |
d38144b7 | 2302 | |
6d3d6aa2 JW |
2303 | /* If someone's already draining, avoid adding running more workers. */ |
2304 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2305 | return; | |
72f0184c MH |
2306 | /* |
2307 | * Notify other cpus that system-wide "drain" is running | |
2308 | * We do not care about races with the cpu hotplug because cpu down | |
2309 | * as well as workers from this path always operate on the local | |
2310 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2311 | */ | |
5af12d0e | 2312 | curcpu = get_cpu(); |
cdec2e42 KH |
2313 | for_each_online_cpu(cpu) { |
2314 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2315 | struct mem_cgroup *memcg; |
e1a366be | 2316 | bool flush = false; |
26fe6168 | 2317 | |
e1a366be | 2318 | rcu_read_lock(); |
c0ff4b85 | 2319 | memcg = stock->cached; |
e1a366be RG |
2320 | if (memcg && stock->nr_pages && |
2321 | mem_cgroup_is_descendant(memcg, root_memcg)) | |
2322 | flush = true; | |
bf4f0599 RG |
2323 | if (obj_stock_flush_required(stock, root_memcg)) |
2324 | flush = true; | |
e1a366be RG |
2325 | rcu_read_unlock(); |
2326 | ||
2327 | if (flush && | |
2328 | !test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { | |
d1a05b69 MH |
2329 | if (cpu == curcpu) |
2330 | drain_local_stock(&stock->work); | |
2331 | else | |
2332 | schedule_work_on(cpu, &stock->work); | |
2333 | } | |
cdec2e42 | 2334 | } |
5af12d0e | 2335 | put_cpu(); |
9f50fad6 | 2336 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2337 | } |
2338 | ||
308167fc | 2339 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
cdec2e42 | 2340 | { |
cdec2e42 | 2341 | struct memcg_stock_pcp *stock; |
42a30035 | 2342 | struct mem_cgroup *memcg, *mi; |
cdec2e42 | 2343 | |
cdec2e42 KH |
2344 | stock = &per_cpu(memcg_stock, cpu); |
2345 | drain_stock(stock); | |
a983b5eb JW |
2346 | |
2347 | for_each_mem_cgroup(memcg) { | |
2348 | int i; | |
2349 | ||
2350 | for (i = 0; i < MEMCG_NR_STAT; i++) { | |
2351 | int nid; | |
2352 | long x; | |
2353 | ||
871789d4 | 2354 | x = this_cpu_xchg(memcg->vmstats_percpu->stat[i], 0); |
815744d7 | 2355 | if (x) |
42a30035 JW |
2356 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2357 | atomic_long_add(x, &memcg->vmstats[i]); | |
a983b5eb JW |
2358 | |
2359 | if (i >= NR_VM_NODE_STAT_ITEMS) | |
2360 | continue; | |
2361 | ||
2362 | for_each_node(nid) { | |
2363 | struct mem_cgroup_per_node *pn; | |
2364 | ||
2365 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
2366 | x = this_cpu_xchg(pn->lruvec_stat_cpu->count[i], 0); | |
815744d7 | 2367 | if (x) |
42a30035 JW |
2368 | do { |
2369 | atomic_long_add(x, &pn->lruvec_stat[i]); | |
2370 | } while ((pn = parent_nodeinfo(pn, nid))); | |
a983b5eb JW |
2371 | } |
2372 | } | |
2373 | ||
e27be240 | 2374 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { |
a983b5eb JW |
2375 | long x; |
2376 | ||
871789d4 | 2377 | x = this_cpu_xchg(memcg->vmstats_percpu->events[i], 0); |
815744d7 | 2378 | if (x) |
42a30035 JW |
2379 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2380 | atomic_long_add(x, &memcg->vmevents[i]); | |
a983b5eb JW |
2381 | } |
2382 | } | |
2383 | ||
308167fc | 2384 | return 0; |
cdec2e42 KH |
2385 | } |
2386 | ||
b3ff9291 CD |
2387 | static unsigned long reclaim_high(struct mem_cgroup *memcg, |
2388 | unsigned int nr_pages, | |
2389 | gfp_t gfp_mask) | |
f7e1cb6e | 2390 | { |
b3ff9291 CD |
2391 | unsigned long nr_reclaimed = 0; |
2392 | ||
f7e1cb6e | 2393 | do { |
e22c6ed9 JW |
2394 | unsigned long pflags; |
2395 | ||
d1663a90 JK |
2396 | if (page_counter_read(&memcg->memory) <= |
2397 | READ_ONCE(memcg->memory.high)) | |
f7e1cb6e | 2398 | continue; |
e22c6ed9 | 2399 | |
e27be240 | 2400 | memcg_memory_event(memcg, MEMCG_HIGH); |
e22c6ed9 JW |
2401 | |
2402 | psi_memstall_enter(&pflags); | |
b3ff9291 CD |
2403 | nr_reclaimed += try_to_free_mem_cgroup_pages(memcg, nr_pages, |
2404 | gfp_mask, true); | |
e22c6ed9 | 2405 | psi_memstall_leave(&pflags); |
4bf17307 CD |
2406 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2407 | !mem_cgroup_is_root(memcg)); | |
b3ff9291 CD |
2408 | |
2409 | return nr_reclaimed; | |
f7e1cb6e JW |
2410 | } |
2411 | ||
2412 | static void high_work_func(struct work_struct *work) | |
2413 | { | |
2414 | struct mem_cgroup *memcg; | |
2415 | ||
2416 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2417 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2418 | } |
2419 | ||
0e4b01df CD |
2420 | /* |
2421 | * Clamp the maximum sleep time per allocation batch to 2 seconds. This is | |
2422 | * enough to still cause a significant slowdown in most cases, while still | |
2423 | * allowing diagnostics and tracing to proceed without becoming stuck. | |
2424 | */ | |
2425 | #define MEMCG_MAX_HIGH_DELAY_JIFFIES (2UL*HZ) | |
2426 | ||
2427 | /* | |
2428 | * When calculating the delay, we use these either side of the exponentiation to | |
2429 | * maintain precision and scale to a reasonable number of jiffies (see the table | |
2430 | * below. | |
2431 | * | |
2432 | * - MEMCG_DELAY_PRECISION_SHIFT: Extra precision bits while translating the | |
2433 | * overage ratio to a delay. | |
ac5ddd0f | 2434 | * - MEMCG_DELAY_SCALING_SHIFT: The number of bits to scale down the |
0e4b01df CD |
2435 | * proposed penalty in order to reduce to a reasonable number of jiffies, and |
2436 | * to produce a reasonable delay curve. | |
2437 | * | |
2438 | * MEMCG_DELAY_SCALING_SHIFT just happens to be a number that produces a | |
2439 | * reasonable delay curve compared to precision-adjusted overage, not | |
2440 | * penalising heavily at first, but still making sure that growth beyond the | |
2441 | * limit penalises misbehaviour cgroups by slowing them down exponentially. For | |
2442 | * example, with a high of 100 megabytes: | |
2443 | * | |
2444 | * +-------+------------------------+ | |
2445 | * | usage | time to allocate in ms | | |
2446 | * +-------+------------------------+ | |
2447 | * | 100M | 0 | | |
2448 | * | 101M | 6 | | |
2449 | * | 102M | 25 | | |
2450 | * | 103M | 57 | | |
2451 | * | 104M | 102 | | |
2452 | * | 105M | 159 | | |
2453 | * | 106M | 230 | | |
2454 | * | 107M | 313 | | |
2455 | * | 108M | 409 | | |
2456 | * | 109M | 518 | | |
2457 | * | 110M | 639 | | |
2458 | * | 111M | 774 | | |
2459 | * | 112M | 921 | | |
2460 | * | 113M | 1081 | | |
2461 | * | 114M | 1254 | | |
2462 | * | 115M | 1439 | | |
2463 | * | 116M | 1638 | | |
2464 | * | 117M | 1849 | | |
2465 | * | 118M | 2000 | | |
2466 | * | 119M | 2000 | | |
2467 | * | 120M | 2000 | | |
2468 | * +-------+------------------------+ | |
2469 | */ | |
2470 | #define MEMCG_DELAY_PRECISION_SHIFT 20 | |
2471 | #define MEMCG_DELAY_SCALING_SHIFT 14 | |
2472 | ||
8a5dbc65 | 2473 | static u64 calculate_overage(unsigned long usage, unsigned long high) |
b23afb93 | 2474 | { |
8a5dbc65 | 2475 | u64 overage; |
b23afb93 | 2476 | |
8a5dbc65 JK |
2477 | if (usage <= high) |
2478 | return 0; | |
e26733e0 | 2479 | |
8a5dbc65 JK |
2480 | /* |
2481 | * Prevent division by 0 in overage calculation by acting as if | |
2482 | * it was a threshold of 1 page | |
2483 | */ | |
2484 | high = max(high, 1UL); | |
9b8b1754 | 2485 | |
8a5dbc65 JK |
2486 | overage = usage - high; |
2487 | overage <<= MEMCG_DELAY_PRECISION_SHIFT; | |
2488 | return div64_u64(overage, high); | |
2489 | } | |
e26733e0 | 2490 | |
8a5dbc65 JK |
2491 | static u64 mem_find_max_overage(struct mem_cgroup *memcg) |
2492 | { | |
2493 | u64 overage, max_overage = 0; | |
e26733e0 | 2494 | |
8a5dbc65 JK |
2495 | do { |
2496 | overage = calculate_overage(page_counter_read(&memcg->memory), | |
d1663a90 | 2497 | READ_ONCE(memcg->memory.high)); |
8a5dbc65 | 2498 | max_overage = max(overage, max_overage); |
e26733e0 CD |
2499 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2500 | !mem_cgroup_is_root(memcg)); | |
2501 | ||
8a5dbc65 JK |
2502 | return max_overage; |
2503 | } | |
2504 | ||
4b82ab4f JK |
2505 | static u64 swap_find_max_overage(struct mem_cgroup *memcg) |
2506 | { | |
2507 | u64 overage, max_overage = 0; | |
2508 | ||
2509 | do { | |
2510 | overage = calculate_overage(page_counter_read(&memcg->swap), | |
2511 | READ_ONCE(memcg->swap.high)); | |
2512 | if (overage) | |
2513 | memcg_memory_event(memcg, MEMCG_SWAP_HIGH); | |
2514 | max_overage = max(overage, max_overage); | |
2515 | } while ((memcg = parent_mem_cgroup(memcg)) && | |
2516 | !mem_cgroup_is_root(memcg)); | |
2517 | ||
2518 | return max_overage; | |
2519 | } | |
2520 | ||
8a5dbc65 JK |
2521 | /* |
2522 | * Get the number of jiffies that we should penalise a mischievous cgroup which | |
2523 | * is exceeding its memory.high by checking both it and its ancestors. | |
2524 | */ | |
2525 | static unsigned long calculate_high_delay(struct mem_cgroup *memcg, | |
2526 | unsigned int nr_pages, | |
2527 | u64 max_overage) | |
2528 | { | |
2529 | unsigned long penalty_jiffies; | |
2530 | ||
e26733e0 CD |
2531 | if (!max_overage) |
2532 | return 0; | |
0e4b01df CD |
2533 | |
2534 | /* | |
0e4b01df CD |
2535 | * We use overage compared to memory.high to calculate the number of |
2536 | * jiffies to sleep (penalty_jiffies). Ideally this value should be | |
2537 | * fairly lenient on small overages, and increasingly harsh when the | |
2538 | * memcg in question makes it clear that it has no intention of stopping | |
2539 | * its crazy behaviour, so we exponentially increase the delay based on | |
2540 | * overage amount. | |
2541 | */ | |
e26733e0 CD |
2542 | penalty_jiffies = max_overage * max_overage * HZ; |
2543 | penalty_jiffies >>= MEMCG_DELAY_PRECISION_SHIFT; | |
2544 | penalty_jiffies >>= MEMCG_DELAY_SCALING_SHIFT; | |
0e4b01df CD |
2545 | |
2546 | /* | |
2547 | * Factor in the task's own contribution to the overage, such that four | |
2548 | * N-sized allocations are throttled approximately the same as one | |
2549 | * 4N-sized allocation. | |
2550 | * | |
2551 | * MEMCG_CHARGE_BATCH pages is nominal, so work out how much smaller or | |
2552 | * larger the current charge patch is than that. | |
2553 | */ | |
ff144e69 | 2554 | return penalty_jiffies * nr_pages / MEMCG_CHARGE_BATCH; |
e26733e0 CD |
2555 | } |
2556 | ||
2557 | /* | |
2558 | * Scheduled by try_charge() to be executed from the userland return path | |
2559 | * and reclaims memory over the high limit. | |
2560 | */ | |
2561 | void mem_cgroup_handle_over_high(void) | |
2562 | { | |
2563 | unsigned long penalty_jiffies; | |
2564 | unsigned long pflags; | |
b3ff9291 | 2565 | unsigned long nr_reclaimed; |
e26733e0 | 2566 | unsigned int nr_pages = current->memcg_nr_pages_over_high; |
d977aa93 | 2567 | int nr_retries = MAX_RECLAIM_RETRIES; |
e26733e0 | 2568 | struct mem_cgroup *memcg; |
b3ff9291 | 2569 | bool in_retry = false; |
e26733e0 CD |
2570 | |
2571 | if (likely(!nr_pages)) | |
2572 | return; | |
2573 | ||
2574 | memcg = get_mem_cgroup_from_mm(current->mm); | |
e26733e0 CD |
2575 | current->memcg_nr_pages_over_high = 0; |
2576 | ||
b3ff9291 CD |
2577 | retry_reclaim: |
2578 | /* | |
2579 | * The allocating task should reclaim at least the batch size, but for | |
2580 | * subsequent retries we only want to do what's necessary to prevent oom | |
2581 | * or breaching resource isolation. | |
2582 | * | |
2583 | * This is distinct from memory.max or page allocator behaviour because | |
2584 | * memory.high is currently batched, whereas memory.max and the page | |
2585 | * allocator run every time an allocation is made. | |
2586 | */ | |
2587 | nr_reclaimed = reclaim_high(memcg, | |
2588 | in_retry ? SWAP_CLUSTER_MAX : nr_pages, | |
2589 | GFP_KERNEL); | |
2590 | ||
e26733e0 CD |
2591 | /* |
2592 | * memory.high is breached and reclaim is unable to keep up. Throttle | |
2593 | * allocators proactively to slow down excessive growth. | |
2594 | */ | |
8a5dbc65 JK |
2595 | penalty_jiffies = calculate_high_delay(memcg, nr_pages, |
2596 | mem_find_max_overage(memcg)); | |
0e4b01df | 2597 | |
4b82ab4f JK |
2598 | penalty_jiffies += calculate_high_delay(memcg, nr_pages, |
2599 | swap_find_max_overage(memcg)); | |
2600 | ||
ff144e69 JK |
2601 | /* |
2602 | * Clamp the max delay per usermode return so as to still keep the | |
2603 | * application moving forwards and also permit diagnostics, albeit | |
2604 | * extremely slowly. | |
2605 | */ | |
2606 | penalty_jiffies = min(penalty_jiffies, MEMCG_MAX_HIGH_DELAY_JIFFIES); | |
2607 | ||
0e4b01df CD |
2608 | /* |
2609 | * Don't sleep if the amount of jiffies this memcg owes us is so low | |
2610 | * that it's not even worth doing, in an attempt to be nice to those who | |
2611 | * go only a small amount over their memory.high value and maybe haven't | |
2612 | * been aggressively reclaimed enough yet. | |
2613 | */ | |
2614 | if (penalty_jiffies <= HZ / 100) | |
2615 | goto out; | |
2616 | ||
b3ff9291 CD |
2617 | /* |
2618 | * If reclaim is making forward progress but we're still over | |
2619 | * memory.high, we want to encourage that rather than doing allocator | |
2620 | * throttling. | |
2621 | */ | |
2622 | if (nr_reclaimed || nr_retries--) { | |
2623 | in_retry = true; | |
2624 | goto retry_reclaim; | |
2625 | } | |
2626 | ||
0e4b01df CD |
2627 | /* |
2628 | * If we exit early, we're guaranteed to die (since | |
2629 | * schedule_timeout_killable sets TASK_KILLABLE). This means we don't | |
2630 | * need to account for any ill-begotten jiffies to pay them off later. | |
2631 | */ | |
2632 | psi_memstall_enter(&pflags); | |
2633 | schedule_timeout_killable(penalty_jiffies); | |
2634 | psi_memstall_leave(&pflags); | |
2635 | ||
2636 | out: | |
2637 | css_put(&memcg->css); | |
b23afb93 TH |
2638 | } |
2639 | ||
00501b53 JW |
2640 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2641 | unsigned int nr_pages) | |
8a9f3ccd | 2642 | { |
a983b5eb | 2643 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
d977aa93 | 2644 | int nr_retries = MAX_RECLAIM_RETRIES; |
6539cc05 | 2645 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2646 | struct page_counter *counter; |
e22c6ed9 | 2647 | enum oom_status oom_status; |
6539cc05 | 2648 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2649 | bool may_swap = true; |
2650 | bool drained = false; | |
e22c6ed9 | 2651 | unsigned long pflags; |
a636b327 | 2652 | |
ce00a967 | 2653 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 2654 | return 0; |
6539cc05 | 2655 | retry: |
b6b6cc72 | 2656 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2657 | return 0; |
8a9f3ccd | 2658 | |
7941d214 | 2659 | if (!do_memsw_account() || |
6071ca52 JW |
2660 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2661 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2662 | goto done_restock; |
7941d214 | 2663 | if (do_memsw_account()) |
3e32cb2e JW |
2664 | page_counter_uncharge(&memcg->memsw, batch); |
2665 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2666 | } else { |
3e32cb2e | 2667 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2668 | may_swap = false; |
3fbe7244 | 2669 | } |
7a81b88c | 2670 | |
6539cc05 JW |
2671 | if (batch > nr_pages) { |
2672 | batch = nr_pages; | |
2673 | goto retry; | |
2674 | } | |
6d61ef40 | 2675 | |
869712fd JW |
2676 | /* |
2677 | * Memcg doesn't have a dedicated reserve for atomic | |
2678 | * allocations. But like the global atomic pool, we need to | |
2679 | * put the burden of reclaim on regular allocation requests | |
2680 | * and let these go through as privileged allocations. | |
2681 | */ | |
2682 | if (gfp_mask & __GFP_ATOMIC) | |
2683 | goto force; | |
2684 | ||
06b078fc JW |
2685 | /* |
2686 | * Unlike in global OOM situations, memcg is not in a physical | |
2687 | * memory shortage. Allow dying and OOM-killed tasks to | |
2688 | * bypass the last charges so that they can exit quickly and | |
2689 | * free their memory. | |
2690 | */ | |
7775face | 2691 | if (unlikely(should_force_charge())) |
10d53c74 | 2692 | goto force; |
06b078fc | 2693 | |
89a28483 JW |
2694 | /* |
2695 | * Prevent unbounded recursion when reclaim operations need to | |
2696 | * allocate memory. This might exceed the limits temporarily, | |
2697 | * but we prefer facilitating memory reclaim and getting back | |
2698 | * under the limit over triggering OOM kills in these cases. | |
2699 | */ | |
2700 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2701 | goto force; | |
2702 | ||
06b078fc JW |
2703 | if (unlikely(task_in_memcg_oom(current))) |
2704 | goto nomem; | |
2705 | ||
d0164adc | 2706 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2707 | goto nomem; |
4b534334 | 2708 | |
e27be240 | 2709 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
241994ed | 2710 | |
e22c6ed9 | 2711 | psi_memstall_enter(&pflags); |
b70a2a21 JW |
2712 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2713 | gfp_mask, may_swap); | |
e22c6ed9 | 2714 | psi_memstall_leave(&pflags); |
6539cc05 | 2715 | |
61e02c74 | 2716 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2717 | goto retry; |
28c34c29 | 2718 | |
b70a2a21 | 2719 | if (!drained) { |
6d3d6aa2 | 2720 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2721 | drained = true; |
2722 | goto retry; | |
2723 | } | |
2724 | ||
28c34c29 JW |
2725 | if (gfp_mask & __GFP_NORETRY) |
2726 | goto nomem; | |
6539cc05 JW |
2727 | /* |
2728 | * Even though the limit is exceeded at this point, reclaim | |
2729 | * may have been able to free some pages. Retry the charge | |
2730 | * before killing the task. | |
2731 | * | |
2732 | * Only for regular pages, though: huge pages are rather | |
2733 | * unlikely to succeed so close to the limit, and we fall back | |
2734 | * to regular pages anyway in case of failure. | |
2735 | */ | |
61e02c74 | 2736 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2737 | goto retry; |
2738 | /* | |
2739 | * At task move, charge accounts can be doubly counted. So, it's | |
2740 | * better to wait until the end of task_move if something is going on. | |
2741 | */ | |
2742 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2743 | goto retry; | |
2744 | ||
9b130619 JW |
2745 | if (nr_retries--) |
2746 | goto retry; | |
2747 | ||
38d38493 | 2748 | if (gfp_mask & __GFP_RETRY_MAYFAIL) |
29ef680a MH |
2749 | goto nomem; |
2750 | ||
06b078fc | 2751 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2752 | goto force; |
06b078fc | 2753 | |
6539cc05 | 2754 | if (fatal_signal_pending(current)) |
10d53c74 | 2755 | goto force; |
6539cc05 | 2756 | |
29ef680a MH |
2757 | /* |
2758 | * keep retrying as long as the memcg oom killer is able to make | |
2759 | * a forward progress or bypass the charge if the oom killer | |
2760 | * couldn't make any progress. | |
2761 | */ | |
2762 | oom_status = mem_cgroup_oom(mem_over_limit, gfp_mask, | |
3608de07 | 2763 | get_order(nr_pages * PAGE_SIZE)); |
29ef680a MH |
2764 | switch (oom_status) { |
2765 | case OOM_SUCCESS: | |
d977aa93 | 2766 | nr_retries = MAX_RECLAIM_RETRIES; |
29ef680a MH |
2767 | goto retry; |
2768 | case OOM_FAILED: | |
2769 | goto force; | |
2770 | default: | |
2771 | goto nomem; | |
2772 | } | |
7a81b88c | 2773 | nomem: |
6d1fdc48 | 2774 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2775 | return -ENOMEM; |
10d53c74 TH |
2776 | force: |
2777 | /* | |
2778 | * The allocation either can't fail or will lead to more memory | |
2779 | * being freed very soon. Allow memory usage go over the limit | |
2780 | * temporarily by force charging it. | |
2781 | */ | |
2782 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2783 | if (do_memsw_account()) |
10d53c74 | 2784 | page_counter_charge(&memcg->memsw, nr_pages); |
10d53c74 TH |
2785 | |
2786 | return 0; | |
6539cc05 JW |
2787 | |
2788 | done_restock: | |
2789 | if (batch > nr_pages) | |
2790 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2791 | |
241994ed | 2792 | /* |
b23afb93 TH |
2793 | * If the hierarchy is above the normal consumption range, schedule |
2794 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2795 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2796 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2797 | * not recorded as it most likely matches current's and won't | |
2798 | * change in the meantime. As high limit is checked again before | |
2799 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2800 | */ |
2801 | do { | |
4b82ab4f JK |
2802 | bool mem_high, swap_high; |
2803 | ||
2804 | mem_high = page_counter_read(&memcg->memory) > | |
2805 | READ_ONCE(memcg->memory.high); | |
2806 | swap_high = page_counter_read(&memcg->swap) > | |
2807 | READ_ONCE(memcg->swap.high); | |
2808 | ||
2809 | /* Don't bother a random interrupted task */ | |
2810 | if (in_interrupt()) { | |
2811 | if (mem_high) { | |
f7e1cb6e JW |
2812 | schedule_work(&memcg->high_work); |
2813 | break; | |
2814 | } | |
4b82ab4f JK |
2815 | continue; |
2816 | } | |
2817 | ||
2818 | if (mem_high || swap_high) { | |
2819 | /* | |
2820 | * The allocating tasks in this cgroup will need to do | |
2821 | * reclaim or be throttled to prevent further growth | |
2822 | * of the memory or swap footprints. | |
2823 | * | |
2824 | * Target some best-effort fairness between the tasks, | |
2825 | * and distribute reclaim work and delay penalties | |
2826 | * based on how much each task is actually allocating. | |
2827 | */ | |
9516a18a | 2828 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2829 | set_notify_resume(current); |
2830 | break; | |
2831 | } | |
241994ed | 2832 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2833 | |
2834 | return 0; | |
7a81b88c | 2835 | } |
8a9f3ccd | 2836 | |
f0e45fb4 | 2837 | #if defined(CONFIG_MEMCG_KMEM) || defined(CONFIG_MMU) |
00501b53 | 2838 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2839 | { |
ce00a967 JW |
2840 | if (mem_cgroup_is_root(memcg)) |
2841 | return; | |
2842 | ||
3e32cb2e | 2843 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2844 | if (do_memsw_account()) |
3e32cb2e | 2845 | page_counter_uncharge(&memcg->memsw, nr_pages); |
d01dd17f | 2846 | } |
f0e45fb4 | 2847 | #endif |
d01dd17f | 2848 | |
d9eb1ea2 | 2849 | static void commit_charge(struct page *page, struct mem_cgroup *memcg) |
0a31bc97 | 2850 | { |
1306a85a | 2851 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
0a31bc97 | 2852 | /* |
a0b5b414 | 2853 | * Any of the following ensures page->mem_cgroup stability: |
0a31bc97 | 2854 | * |
a0b5b414 JW |
2855 | * - the page lock |
2856 | * - LRU isolation | |
2857 | * - lock_page_memcg() | |
2858 | * - exclusive reference | |
0a31bc97 | 2859 | */ |
1306a85a | 2860 | page->mem_cgroup = memcg; |
7a81b88c | 2861 | } |
66e1707b | 2862 | |
84c07d11 | 2863 | #ifdef CONFIG_MEMCG_KMEM |
10befea9 RG |
2864 | int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s, |
2865 | gfp_t gfp) | |
2866 | { | |
2867 | unsigned int objects = objs_per_slab_page(s, page); | |
2868 | void *vec; | |
2869 | ||
2870 | vec = kcalloc_node(objects, sizeof(struct obj_cgroup *), gfp, | |
2871 | page_to_nid(page)); | |
2872 | if (!vec) | |
2873 | return -ENOMEM; | |
2874 | ||
2875 | if (cmpxchg(&page->obj_cgroups, NULL, | |
2876 | (struct obj_cgroup **) ((unsigned long)vec | 0x1UL))) | |
2877 | kfree(vec); | |
2878 | else | |
2879 | kmemleak_not_leak(vec); | |
2880 | ||
2881 | return 0; | |
2882 | } | |
2883 | ||
8380ce47 RG |
2884 | /* |
2885 | * Returns a pointer to the memory cgroup to which the kernel object is charged. | |
2886 | * | |
2887 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), | |
2888 | * cgroup_mutex, etc. | |
2889 | */ | |
2890 | struct mem_cgroup *mem_cgroup_from_obj(void *p) | |
2891 | { | |
2892 | struct page *page; | |
2893 | ||
2894 | if (mem_cgroup_disabled()) | |
2895 | return NULL; | |
2896 | ||
2897 | page = virt_to_head_page(p); | |
2898 | ||
19b629c9 RG |
2899 | /* |
2900 | * If page->mem_cgroup is set, it's either a simple mem_cgroup pointer | |
2901 | * or a pointer to obj_cgroup vector. In the latter case the lowest | |
2902 | * bit of the pointer is set. | |
2903 | * The page->mem_cgroup pointer can be asynchronously changed | |
2904 | * from NULL to (obj_cgroup_vec | 0x1UL), but can't be changed | |
2905 | * from a valid memcg pointer to objcg vector or back. | |
2906 | */ | |
2907 | if (!page->mem_cgroup) | |
2908 | return NULL; | |
2909 | ||
8380ce47 | 2910 | /* |
9855609b RG |
2911 | * Slab objects are accounted individually, not per-page. |
2912 | * Memcg membership data for each individual object is saved in | |
2913 | * the page->obj_cgroups. | |
8380ce47 | 2914 | */ |
9855609b RG |
2915 | if (page_has_obj_cgroups(page)) { |
2916 | struct obj_cgroup *objcg; | |
2917 | unsigned int off; | |
2918 | ||
2919 | off = obj_to_index(page->slab_cache, page, p); | |
2920 | objcg = page_obj_cgroups(page)[off]; | |
10befea9 RG |
2921 | if (objcg) |
2922 | return obj_cgroup_memcg(objcg); | |
2923 | ||
2924 | return NULL; | |
9855609b | 2925 | } |
8380ce47 RG |
2926 | |
2927 | /* All other pages use page->mem_cgroup */ | |
2928 | return page->mem_cgroup; | |
2929 | } | |
2930 | ||
bf4f0599 RG |
2931 | __always_inline struct obj_cgroup *get_obj_cgroup_from_current(void) |
2932 | { | |
2933 | struct obj_cgroup *objcg = NULL; | |
2934 | struct mem_cgroup *memcg; | |
2935 | ||
2936 | if (unlikely(!current->mm && !current->active_memcg)) | |
2937 | return NULL; | |
2938 | ||
2939 | rcu_read_lock(); | |
2940 | if (unlikely(current->active_memcg)) | |
2941 | memcg = rcu_dereference(current->active_memcg); | |
2942 | else | |
2943 | memcg = mem_cgroup_from_task(current); | |
2944 | ||
2945 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { | |
2946 | objcg = rcu_dereference(memcg->objcg); | |
2947 | if (objcg && obj_cgroup_tryget(objcg)) | |
2948 | break; | |
2949 | } | |
2950 | rcu_read_unlock(); | |
2951 | ||
2952 | return objcg; | |
2953 | } | |
2954 | ||
f3bb3043 | 2955 | static int memcg_alloc_cache_id(void) |
55007d84 | 2956 | { |
f3bb3043 VD |
2957 | int id, size; |
2958 | int err; | |
2959 | ||
dbcf73e2 | 2960 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2961 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2962 | if (id < 0) | |
2963 | return id; | |
55007d84 | 2964 | |
dbcf73e2 | 2965 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2966 | return id; |
2967 | ||
2968 | /* | |
2969 | * There's no space for the new id in memcg_caches arrays, | |
2970 | * so we have to grow them. | |
2971 | */ | |
05257a1a | 2972 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2973 | |
2974 | size = 2 * (id + 1); | |
55007d84 GC |
2975 | if (size < MEMCG_CACHES_MIN_SIZE) |
2976 | size = MEMCG_CACHES_MIN_SIZE; | |
2977 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2978 | size = MEMCG_CACHES_MAX_SIZE; | |
2979 | ||
9855609b | 2980 | err = memcg_update_all_list_lrus(size); |
05257a1a VD |
2981 | if (!err) |
2982 | memcg_nr_cache_ids = size; | |
2983 | ||
2984 | up_write(&memcg_cache_ids_sem); | |
2985 | ||
f3bb3043 | 2986 | if (err) { |
dbcf73e2 | 2987 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2988 | return err; |
2989 | } | |
2990 | return id; | |
2991 | } | |
2992 | ||
2993 | static void memcg_free_cache_id(int id) | |
2994 | { | |
dbcf73e2 | 2995 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2996 | } |
2997 | ||
45264778 | 2998 | /** |
4b13f64d | 2999 | * __memcg_kmem_charge: charge a number of kernel pages to a memcg |
10eaec2f | 3000 | * @memcg: memory cgroup to charge |
45264778 | 3001 | * @gfp: reclaim mode |
92d0510c | 3002 | * @nr_pages: number of pages to charge |
45264778 VD |
3003 | * |
3004 | * Returns 0 on success, an error code on failure. | |
3005 | */ | |
4b13f64d RG |
3006 | int __memcg_kmem_charge(struct mem_cgroup *memcg, gfp_t gfp, |
3007 | unsigned int nr_pages) | |
7ae1e1d0 | 3008 | { |
f3ccb2c4 | 3009 | struct page_counter *counter; |
7ae1e1d0 GC |
3010 | int ret; |
3011 | ||
f3ccb2c4 | 3012 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 3013 | if (ret) |
f3ccb2c4 | 3014 | return ret; |
52c29b04 JW |
3015 | |
3016 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
3017 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
e55d9d9b MH |
3018 | |
3019 | /* | |
3020 | * Enforce __GFP_NOFAIL allocation because callers are not | |
3021 | * prepared to see failures and likely do not have any failure | |
3022 | * handling code. | |
3023 | */ | |
3024 | if (gfp & __GFP_NOFAIL) { | |
3025 | page_counter_charge(&memcg->kmem, nr_pages); | |
3026 | return 0; | |
3027 | } | |
52c29b04 JW |
3028 | cancel_charge(memcg, nr_pages); |
3029 | return -ENOMEM; | |
7ae1e1d0 | 3030 | } |
f3ccb2c4 | 3031 | return 0; |
7ae1e1d0 GC |
3032 | } |
3033 | ||
4b13f64d RG |
3034 | /** |
3035 | * __memcg_kmem_uncharge: uncharge a number of kernel pages from a memcg | |
3036 | * @memcg: memcg to uncharge | |
3037 | * @nr_pages: number of pages to uncharge | |
3038 | */ | |
3039 | void __memcg_kmem_uncharge(struct mem_cgroup *memcg, unsigned int nr_pages) | |
3040 | { | |
3041 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
3042 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
3043 | ||
3044 | page_counter_uncharge(&memcg->memory, nr_pages); | |
3045 | if (do_memsw_account()) | |
3046 | page_counter_uncharge(&memcg->memsw, nr_pages); | |
3047 | } | |
3048 | ||
45264778 | 3049 | /** |
f4b00eab | 3050 | * __memcg_kmem_charge_page: charge a kmem page to the current memory cgroup |
45264778 VD |
3051 | * @page: page to charge |
3052 | * @gfp: reclaim mode | |
3053 | * @order: allocation order | |
3054 | * | |
3055 | * Returns 0 on success, an error code on failure. | |
3056 | */ | |
f4b00eab | 3057 | int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 3058 | { |
f3ccb2c4 | 3059 | struct mem_cgroup *memcg; |
fcff7d7e | 3060 | int ret = 0; |
7ae1e1d0 | 3061 | |
60cd4bcd | 3062 | if (memcg_kmem_bypass()) |
45264778 VD |
3063 | return 0; |
3064 | ||
d46eb14b | 3065 | memcg = get_mem_cgroup_from_current(); |
c4159a75 | 3066 | if (!mem_cgroup_is_root(memcg)) { |
4b13f64d | 3067 | ret = __memcg_kmem_charge(memcg, gfp, 1 << order); |
4d96ba35 RG |
3068 | if (!ret) { |
3069 | page->mem_cgroup = memcg; | |
c4159a75 | 3070 | __SetPageKmemcg(page); |
1a3e1f40 | 3071 | return 0; |
4d96ba35 | 3072 | } |
c4159a75 | 3073 | } |
7ae1e1d0 | 3074 | css_put(&memcg->css); |
d05e83a6 | 3075 | return ret; |
7ae1e1d0 | 3076 | } |
49a18eae | 3077 | |
45264778 | 3078 | /** |
f4b00eab | 3079 | * __memcg_kmem_uncharge_page: uncharge a kmem page |
45264778 VD |
3080 | * @page: page to uncharge |
3081 | * @order: allocation order | |
3082 | */ | |
f4b00eab | 3083 | void __memcg_kmem_uncharge_page(struct page *page, int order) |
7ae1e1d0 | 3084 | { |
1306a85a | 3085 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 3086 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 3087 | |
7ae1e1d0 GC |
3088 | if (!memcg) |
3089 | return; | |
3090 | ||
309381fe | 3091 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
4b13f64d | 3092 | __memcg_kmem_uncharge(memcg, nr_pages); |
1306a85a | 3093 | page->mem_cgroup = NULL; |
1a3e1f40 | 3094 | css_put(&memcg->css); |
c4159a75 VD |
3095 | |
3096 | /* slab pages do not have PageKmemcg flag set */ | |
3097 | if (PageKmemcg(page)) | |
3098 | __ClearPageKmemcg(page); | |
60d3fd32 | 3099 | } |
bf4f0599 RG |
3100 | |
3101 | static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) | |
3102 | { | |
3103 | struct memcg_stock_pcp *stock; | |
3104 | unsigned long flags; | |
3105 | bool ret = false; | |
3106 | ||
3107 | local_irq_save(flags); | |
3108 | ||
3109 | stock = this_cpu_ptr(&memcg_stock); | |
3110 | if (objcg == stock->cached_objcg && stock->nr_bytes >= nr_bytes) { | |
3111 | stock->nr_bytes -= nr_bytes; | |
3112 | ret = true; | |
3113 | } | |
3114 | ||
3115 | local_irq_restore(flags); | |
3116 | ||
3117 | return ret; | |
3118 | } | |
3119 | ||
3120 | static void drain_obj_stock(struct memcg_stock_pcp *stock) | |
3121 | { | |
3122 | struct obj_cgroup *old = stock->cached_objcg; | |
3123 | ||
3124 | if (!old) | |
3125 | return; | |
3126 | ||
3127 | if (stock->nr_bytes) { | |
3128 | unsigned int nr_pages = stock->nr_bytes >> PAGE_SHIFT; | |
3129 | unsigned int nr_bytes = stock->nr_bytes & (PAGE_SIZE - 1); | |
3130 | ||
3131 | if (nr_pages) { | |
3132 | rcu_read_lock(); | |
3133 | __memcg_kmem_uncharge(obj_cgroup_memcg(old), nr_pages); | |
3134 | rcu_read_unlock(); | |
3135 | } | |
3136 | ||
3137 | /* | |
3138 | * The leftover is flushed to the centralized per-memcg value. | |
3139 | * On the next attempt to refill obj stock it will be moved | |
3140 | * to a per-cpu stock (probably, on an other CPU), see | |
3141 | * refill_obj_stock(). | |
3142 | * | |
3143 | * How often it's flushed is a trade-off between the memory | |
3144 | * limit enforcement accuracy and potential CPU contention, | |
3145 | * so it might be changed in the future. | |
3146 | */ | |
3147 | atomic_add(nr_bytes, &old->nr_charged_bytes); | |
3148 | stock->nr_bytes = 0; | |
3149 | } | |
3150 | ||
3151 | obj_cgroup_put(old); | |
3152 | stock->cached_objcg = NULL; | |
3153 | } | |
3154 | ||
3155 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
3156 | struct mem_cgroup *root_memcg) | |
3157 | { | |
3158 | struct mem_cgroup *memcg; | |
3159 | ||
3160 | if (stock->cached_objcg) { | |
3161 | memcg = obj_cgroup_memcg(stock->cached_objcg); | |
3162 | if (memcg && mem_cgroup_is_descendant(memcg, root_memcg)) | |
3163 | return true; | |
3164 | } | |
3165 | ||
3166 | return false; | |
3167 | } | |
3168 | ||
3169 | static void refill_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) | |
3170 | { | |
3171 | struct memcg_stock_pcp *stock; | |
3172 | unsigned long flags; | |
3173 | ||
3174 | local_irq_save(flags); | |
3175 | ||
3176 | stock = this_cpu_ptr(&memcg_stock); | |
3177 | if (stock->cached_objcg != objcg) { /* reset if necessary */ | |
3178 | drain_obj_stock(stock); | |
3179 | obj_cgroup_get(objcg); | |
3180 | stock->cached_objcg = objcg; | |
3181 | stock->nr_bytes = atomic_xchg(&objcg->nr_charged_bytes, 0); | |
3182 | } | |
3183 | stock->nr_bytes += nr_bytes; | |
3184 | ||
3185 | if (stock->nr_bytes > PAGE_SIZE) | |
3186 | drain_obj_stock(stock); | |
3187 | ||
3188 | local_irq_restore(flags); | |
3189 | } | |
3190 | ||
3191 | int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size) | |
3192 | { | |
3193 | struct mem_cgroup *memcg; | |
3194 | unsigned int nr_pages, nr_bytes; | |
3195 | int ret; | |
3196 | ||
3197 | if (consume_obj_stock(objcg, size)) | |
3198 | return 0; | |
3199 | ||
3200 | /* | |
3201 | * In theory, memcg->nr_charged_bytes can have enough | |
3202 | * pre-charged bytes to satisfy the allocation. However, | |
3203 | * flushing memcg->nr_charged_bytes requires two atomic | |
3204 | * operations, and memcg->nr_charged_bytes can't be big, | |
3205 | * so it's better to ignore it and try grab some new pages. | |
3206 | * memcg->nr_charged_bytes will be flushed in | |
3207 | * refill_obj_stock(), called from this function or | |
3208 | * independently later. | |
3209 | */ | |
3210 | rcu_read_lock(); | |
3211 | memcg = obj_cgroup_memcg(objcg); | |
3212 | css_get(&memcg->css); | |
3213 | rcu_read_unlock(); | |
3214 | ||
3215 | nr_pages = size >> PAGE_SHIFT; | |
3216 | nr_bytes = size & (PAGE_SIZE - 1); | |
3217 | ||
3218 | if (nr_bytes) | |
3219 | nr_pages += 1; | |
3220 | ||
3221 | ret = __memcg_kmem_charge(memcg, gfp, nr_pages); | |
3222 | if (!ret && nr_bytes) | |
3223 | refill_obj_stock(objcg, PAGE_SIZE - nr_bytes); | |
3224 | ||
3225 | css_put(&memcg->css); | |
3226 | return ret; | |
3227 | } | |
3228 | ||
3229 | void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size) | |
3230 | { | |
3231 | refill_obj_stock(objcg, size); | |
3232 | } | |
3233 | ||
84c07d11 | 3234 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 3235 | |
ca3e0214 KH |
3236 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3237 | ||
ca3e0214 KH |
3238 | /* |
3239 | * Because tail pages are not marked as "used", set it. We're under | |
f4b7e272 | 3240 | * pgdat->lru_lock and migration entries setup in all page mappings. |
ca3e0214 | 3241 | */ |
e94c8a9c | 3242 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 3243 | { |
1a3e1f40 | 3244 | struct mem_cgroup *memcg = head->mem_cgroup; |
e94c8a9c | 3245 | int i; |
ca3e0214 | 3246 | |
3d37c4a9 KH |
3247 | if (mem_cgroup_disabled()) |
3248 | return; | |
b070e65c | 3249 | |
1a3e1f40 JW |
3250 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
3251 | css_get(&memcg->css); | |
3252 | head[i].mem_cgroup = memcg; | |
3253 | } | |
ca3e0214 | 3254 | } |
12d27107 | 3255 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3256 | |
c255a458 | 3257 | #ifdef CONFIG_MEMCG_SWAP |
02491447 DN |
3258 | /** |
3259 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3260 | * @entry: swap entry to be moved | |
3261 | * @from: mem_cgroup which the entry is moved from | |
3262 | * @to: mem_cgroup which the entry is moved to | |
3263 | * | |
3264 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3265 | * as the mem_cgroup's id of @from. | |
3266 | * | |
3267 | * Returns 0 on success, -EINVAL on failure. | |
3268 | * | |
3e32cb2e | 3269 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
3270 | * both res and memsw, and called css_get(). |
3271 | */ | |
3272 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3273 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3274 | { |
3275 | unsigned short old_id, new_id; | |
3276 | ||
34c00c31 LZ |
3277 | old_id = mem_cgroup_id(from); |
3278 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3279 | |
3280 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
3281 | mod_memcg_state(from, MEMCG_SWAP, -1); |
3282 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
3283 | return 0; |
3284 | } | |
3285 | return -EINVAL; | |
3286 | } | |
3287 | #else | |
3288 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3289 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3290 | { |
3291 | return -EINVAL; | |
3292 | } | |
8c7c6e34 | 3293 | #endif |
d13d1443 | 3294 | |
bbec2e15 | 3295 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 3296 | |
bbec2e15 RG |
3297 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
3298 | unsigned long max, bool memsw) | |
628f4235 | 3299 | { |
3e32cb2e | 3300 | bool enlarge = false; |
bb4a7ea2 | 3301 | bool drained = false; |
3e32cb2e | 3302 | int ret; |
c054a78c YZ |
3303 | bool limits_invariant; |
3304 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 3305 | |
3e32cb2e | 3306 | do { |
628f4235 KH |
3307 | if (signal_pending(current)) { |
3308 | ret = -EINTR; | |
3309 | break; | |
3310 | } | |
3e32cb2e | 3311 | |
bbec2e15 | 3312 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
3313 | /* |
3314 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 3315 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 3316 | */ |
15b42562 | 3317 | limits_invariant = memsw ? max >= READ_ONCE(memcg->memory.max) : |
bbec2e15 | 3318 | max <= memcg->memsw.max; |
c054a78c | 3319 | if (!limits_invariant) { |
bbec2e15 | 3320 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 3321 | ret = -EINVAL; |
8c7c6e34 KH |
3322 | break; |
3323 | } | |
bbec2e15 | 3324 | if (max > counter->max) |
3e32cb2e | 3325 | enlarge = true; |
bbec2e15 RG |
3326 | ret = page_counter_set_max(counter, max); |
3327 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
3328 | |
3329 | if (!ret) | |
3330 | break; | |
3331 | ||
bb4a7ea2 SB |
3332 | if (!drained) { |
3333 | drain_all_stock(memcg); | |
3334 | drained = true; | |
3335 | continue; | |
3336 | } | |
3337 | ||
1ab5c056 AR |
3338 | if (!try_to_free_mem_cgroup_pages(memcg, 1, |
3339 | GFP_KERNEL, !memsw)) { | |
3340 | ret = -EBUSY; | |
3341 | break; | |
3342 | } | |
3343 | } while (true); | |
3e32cb2e | 3344 | |
3c11ecf4 KH |
3345 | if (!ret && enlarge) |
3346 | memcg_oom_recover(memcg); | |
3e32cb2e | 3347 | |
628f4235 KH |
3348 | return ret; |
3349 | } | |
3350 | ||
ef8f2327 | 3351 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
3352 | gfp_t gfp_mask, |
3353 | unsigned long *total_scanned) | |
3354 | { | |
3355 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 3356 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
3357 | unsigned long reclaimed; |
3358 | int loop = 0; | |
ef8f2327 | 3359 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 3360 | unsigned long excess; |
0608f43d AM |
3361 | unsigned long nr_scanned; |
3362 | ||
3363 | if (order > 0) | |
3364 | return 0; | |
3365 | ||
ef8f2327 | 3366 | mctz = soft_limit_tree_node(pgdat->node_id); |
d6507ff5 MH |
3367 | |
3368 | /* | |
3369 | * Do not even bother to check the largest node if the root | |
3370 | * is empty. Do it lockless to prevent lock bouncing. Races | |
3371 | * are acceptable as soft limit is best effort anyway. | |
3372 | */ | |
bfc7228b | 3373 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
3374 | return 0; |
3375 | ||
0608f43d AM |
3376 | /* |
3377 | * This loop can run a while, specially if mem_cgroup's continuously | |
3378 | * keep exceeding their soft limit and putting the system under | |
3379 | * pressure | |
3380 | */ | |
3381 | do { | |
3382 | if (next_mz) | |
3383 | mz = next_mz; | |
3384 | else | |
3385 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3386 | if (!mz) | |
3387 | break; | |
3388 | ||
3389 | nr_scanned = 0; | |
ef8f2327 | 3390 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
0608f43d AM |
3391 | gfp_mask, &nr_scanned); |
3392 | nr_reclaimed += reclaimed; | |
3393 | *total_scanned += nr_scanned; | |
0a31bc97 | 3394 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 3395 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
3396 | |
3397 | /* | |
3398 | * If we failed to reclaim anything from this memory cgroup | |
3399 | * it is time to move on to the next cgroup | |
3400 | */ | |
3401 | next_mz = NULL; | |
bc2f2e7f VD |
3402 | if (!reclaimed) |
3403 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3404 | ||
3e32cb2e | 3405 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3406 | /* |
3407 | * One school of thought says that we should not add | |
3408 | * back the node to the tree if reclaim returns 0. | |
3409 | * But our reclaim could return 0, simply because due | |
3410 | * to priority we are exposing a smaller subset of | |
3411 | * memory to reclaim from. Consider this as a longer | |
3412 | * term TODO. | |
3413 | */ | |
3414 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3415 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3416 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3417 | css_put(&mz->memcg->css); |
3418 | loop++; | |
3419 | /* | |
3420 | * Could not reclaim anything and there are no more | |
3421 | * mem cgroups to try or we seem to be looping without | |
3422 | * reclaiming anything. | |
3423 | */ | |
3424 | if (!nr_reclaimed && | |
3425 | (next_mz == NULL || | |
3426 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3427 | break; | |
3428 | } while (!nr_reclaimed); | |
3429 | if (next_mz) | |
3430 | css_put(&next_mz->memcg->css); | |
3431 | return nr_reclaimed; | |
3432 | } | |
3433 | ||
ea280e7b TH |
3434 | /* |
3435 | * Test whether @memcg has children, dead or alive. Note that this | |
3436 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
3437 | * returns %true if there are child csses according to the cgroup | |
b8f2935f | 3438 | * hierarchy. Testing use_hierarchy is the caller's responsibility. |
ea280e7b | 3439 | */ |
b5f99b53 GC |
3440 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
3441 | { | |
ea280e7b TH |
3442 | bool ret; |
3443 | ||
ea280e7b TH |
3444 | rcu_read_lock(); |
3445 | ret = css_next_child(NULL, &memcg->css); | |
3446 | rcu_read_unlock(); | |
3447 | return ret; | |
b5f99b53 GC |
3448 | } |
3449 | ||
c26251f9 | 3450 | /* |
51038171 | 3451 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
3452 | * |
3453 | * Caller is responsible for holding css reference for memcg. | |
3454 | */ | |
3455 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3456 | { | |
d977aa93 | 3457 | int nr_retries = MAX_RECLAIM_RETRIES; |
c26251f9 | 3458 | |
c1e862c1 KH |
3459 | /* we call try-to-free pages for make this cgroup empty */ |
3460 | lru_add_drain_all(); | |
d12c60f6 JS |
3461 | |
3462 | drain_all_stock(memcg); | |
3463 | ||
f817ed48 | 3464 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3465 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 3466 | int progress; |
c1e862c1 | 3467 | |
c26251f9 MH |
3468 | if (signal_pending(current)) |
3469 | return -EINTR; | |
3470 | ||
b70a2a21 JW |
3471 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3472 | GFP_KERNEL, true); | |
c1e862c1 | 3473 | if (!progress) { |
f817ed48 | 3474 | nr_retries--; |
c1e862c1 | 3475 | /* maybe some writeback is necessary */ |
8aa7e847 | 3476 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3477 | } |
f817ed48 KH |
3478 | |
3479 | } | |
ab5196c2 MH |
3480 | |
3481 | return 0; | |
cc847582 KH |
3482 | } |
3483 | ||
6770c64e TH |
3484 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3485 | char *buf, size_t nbytes, | |
3486 | loff_t off) | |
c1e862c1 | 3487 | { |
6770c64e | 3488 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3489 | |
d8423011 MH |
3490 | if (mem_cgroup_is_root(memcg)) |
3491 | return -EINVAL; | |
6770c64e | 3492 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3493 | } |
3494 | ||
182446d0 TH |
3495 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3496 | struct cftype *cft) | |
18f59ea7 | 3497 | { |
182446d0 | 3498 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
3499 | } |
3500 | ||
182446d0 TH |
3501 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3502 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
3503 | { |
3504 | int retval = 0; | |
182446d0 | 3505 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 3506 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 3507 | |
567fb435 | 3508 | if (memcg->use_hierarchy == val) |
0b8f73e1 | 3509 | return 0; |
567fb435 | 3510 | |
18f59ea7 | 3511 | /* |
af901ca1 | 3512 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3513 | * in the child subtrees. If it is unset, then the change can |
3514 | * occur, provided the current cgroup has no children. | |
3515 | * | |
3516 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3517 | * set if there are no children. | |
3518 | */ | |
c0ff4b85 | 3519 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 3520 | (val == 1 || val == 0)) { |
ea280e7b | 3521 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 3522 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3523 | else |
3524 | retval = -EBUSY; | |
3525 | } else | |
3526 | retval = -EINVAL; | |
567fb435 | 3527 | |
18f59ea7 BS |
3528 | return retval; |
3529 | } | |
3530 | ||
6f646156 | 3531 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 3532 | { |
42a30035 | 3533 | unsigned long val; |
ce00a967 | 3534 | |
3e32cb2e | 3535 | if (mem_cgroup_is_root(memcg)) { |
0d1c2072 | 3536 | val = memcg_page_state(memcg, NR_FILE_PAGES) + |
be5d0a74 | 3537 | memcg_page_state(memcg, NR_ANON_MAPPED); |
42a30035 JW |
3538 | if (swap) |
3539 | val += memcg_page_state(memcg, MEMCG_SWAP); | |
3e32cb2e | 3540 | } else { |
ce00a967 | 3541 | if (!swap) |
3e32cb2e | 3542 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3543 | else |
3e32cb2e | 3544 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3545 | } |
c12176d3 | 3546 | return val; |
ce00a967 JW |
3547 | } |
3548 | ||
3e32cb2e JW |
3549 | enum { |
3550 | RES_USAGE, | |
3551 | RES_LIMIT, | |
3552 | RES_MAX_USAGE, | |
3553 | RES_FAILCNT, | |
3554 | RES_SOFT_LIMIT, | |
3555 | }; | |
ce00a967 | 3556 | |
791badbd | 3557 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3558 | struct cftype *cft) |
8cdea7c0 | 3559 | { |
182446d0 | 3560 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3561 | struct page_counter *counter; |
af36f906 | 3562 | |
3e32cb2e | 3563 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3564 | case _MEM: |
3e32cb2e JW |
3565 | counter = &memcg->memory; |
3566 | break; | |
8c7c6e34 | 3567 | case _MEMSWAP: |
3e32cb2e JW |
3568 | counter = &memcg->memsw; |
3569 | break; | |
510fc4e1 | 3570 | case _KMEM: |
3e32cb2e | 3571 | counter = &memcg->kmem; |
510fc4e1 | 3572 | break; |
d55f90bf | 3573 | case _TCP: |
0db15298 | 3574 | counter = &memcg->tcpmem; |
d55f90bf | 3575 | break; |
8c7c6e34 KH |
3576 | default: |
3577 | BUG(); | |
8c7c6e34 | 3578 | } |
3e32cb2e JW |
3579 | |
3580 | switch (MEMFILE_ATTR(cft->private)) { | |
3581 | case RES_USAGE: | |
3582 | if (counter == &memcg->memory) | |
c12176d3 | 3583 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 3584 | if (counter == &memcg->memsw) |
c12176d3 | 3585 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
3586 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
3587 | case RES_LIMIT: | |
bbec2e15 | 3588 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
3589 | case RES_MAX_USAGE: |
3590 | return (u64)counter->watermark * PAGE_SIZE; | |
3591 | case RES_FAILCNT: | |
3592 | return counter->failcnt; | |
3593 | case RES_SOFT_LIMIT: | |
3594 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3595 | default: | |
3596 | BUG(); | |
3597 | } | |
8cdea7c0 | 3598 | } |
510fc4e1 | 3599 | |
4a87e2a2 | 3600 | static void memcg_flush_percpu_vmstats(struct mem_cgroup *memcg) |
c350a99e | 3601 | { |
4a87e2a2 | 3602 | unsigned long stat[MEMCG_NR_STAT] = {0}; |
c350a99e RG |
3603 | struct mem_cgroup *mi; |
3604 | int node, cpu, i; | |
c350a99e RG |
3605 | |
3606 | for_each_online_cpu(cpu) | |
4a87e2a2 | 3607 | for (i = 0; i < MEMCG_NR_STAT; i++) |
6c1c2808 | 3608 | stat[i] += per_cpu(memcg->vmstats_percpu->stat[i], cpu); |
c350a99e RG |
3609 | |
3610 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
4a87e2a2 | 3611 | for (i = 0; i < MEMCG_NR_STAT; i++) |
c350a99e RG |
3612 | atomic_long_add(stat[i], &mi->vmstats[i]); |
3613 | ||
3614 | for_each_node(node) { | |
3615 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; | |
3616 | struct mem_cgroup_per_node *pi; | |
3617 | ||
4a87e2a2 | 3618 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
c350a99e RG |
3619 | stat[i] = 0; |
3620 | ||
3621 | for_each_online_cpu(cpu) | |
4a87e2a2 | 3622 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
6c1c2808 SB |
3623 | stat[i] += per_cpu( |
3624 | pn->lruvec_stat_cpu->count[i], cpu); | |
c350a99e RG |
3625 | |
3626 | for (pi = pn; pi; pi = parent_nodeinfo(pi, node)) | |
4a87e2a2 | 3627 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
c350a99e RG |
3628 | atomic_long_add(stat[i], &pi->lruvec_stat[i]); |
3629 | } | |
3630 | } | |
3631 | ||
bb65f89b RG |
3632 | static void memcg_flush_percpu_vmevents(struct mem_cgroup *memcg) |
3633 | { | |
3634 | unsigned long events[NR_VM_EVENT_ITEMS]; | |
3635 | struct mem_cgroup *mi; | |
3636 | int cpu, i; | |
3637 | ||
3638 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
3639 | events[i] = 0; | |
3640 | ||
3641 | for_each_online_cpu(cpu) | |
3642 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
6c1c2808 SB |
3643 | events[i] += per_cpu(memcg->vmstats_percpu->events[i], |
3644 | cpu); | |
bb65f89b RG |
3645 | |
3646 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
3647 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
3648 | atomic_long_add(events[i], &mi->vmevents[i]); | |
3649 | } | |
3650 | ||
84c07d11 | 3651 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 3652 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 3653 | { |
bf4f0599 | 3654 | struct obj_cgroup *objcg; |
d6441637 VD |
3655 | int memcg_id; |
3656 | ||
b313aeee VD |
3657 | if (cgroup_memory_nokmem) |
3658 | return 0; | |
3659 | ||
2a4db7eb | 3660 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 3661 | BUG_ON(memcg->kmem_state); |
d6441637 | 3662 | |
f3bb3043 | 3663 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
3664 | if (memcg_id < 0) |
3665 | return memcg_id; | |
d6441637 | 3666 | |
bf4f0599 RG |
3667 | objcg = obj_cgroup_alloc(); |
3668 | if (!objcg) { | |
3669 | memcg_free_cache_id(memcg_id); | |
3670 | return -ENOMEM; | |
3671 | } | |
3672 | objcg->memcg = memcg; | |
3673 | rcu_assign_pointer(memcg->objcg, objcg); | |
3674 | ||
d648bcc7 RG |
3675 | static_branch_enable(&memcg_kmem_enabled_key); |
3676 | ||
d6441637 | 3677 | /* |
567e9ab2 | 3678 | * A memory cgroup is considered kmem-online as soon as it gets |
900a38f0 | 3679 | * kmemcg_id. Setting the id after enabling static branching will |
d6441637 VD |
3680 | * guarantee no one starts accounting before all call sites are |
3681 | * patched. | |
3682 | */ | |
900a38f0 | 3683 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 3684 | memcg->kmem_state = KMEM_ONLINE; |
0b8f73e1 JW |
3685 | |
3686 | return 0; | |
d6441637 VD |
3687 | } |
3688 | ||
8e0a8912 JW |
3689 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
3690 | { | |
3691 | struct cgroup_subsys_state *css; | |
3692 | struct mem_cgroup *parent, *child; | |
3693 | int kmemcg_id; | |
3694 | ||
3695 | if (memcg->kmem_state != KMEM_ONLINE) | |
3696 | return; | |
9855609b | 3697 | |
8e0a8912 JW |
3698 | memcg->kmem_state = KMEM_ALLOCATED; |
3699 | ||
8e0a8912 JW |
3700 | parent = parent_mem_cgroup(memcg); |
3701 | if (!parent) | |
3702 | parent = root_mem_cgroup; | |
3703 | ||
bf4f0599 | 3704 | memcg_reparent_objcgs(memcg, parent); |
fb2f2b0a RG |
3705 | |
3706 | kmemcg_id = memcg->kmemcg_id; | |
3707 | BUG_ON(kmemcg_id < 0); | |
3708 | ||
8e0a8912 JW |
3709 | /* |
3710 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3711 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3712 | * to ones of the parent. After we have finished, all list_lrus | |
3713 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3714 | * ordering is imposed by list_lru_node->lock taken by | |
3715 | * memcg_drain_all_list_lrus(). | |
3716 | */ | |
3a06bb78 | 3717 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
8e0a8912 JW |
3718 | css_for_each_descendant_pre(css, &memcg->css) { |
3719 | child = mem_cgroup_from_css(css); | |
3720 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3721 | child->kmemcg_id = parent->kmemcg_id; | |
3722 | if (!memcg->use_hierarchy) | |
3723 | break; | |
3724 | } | |
3a06bb78 TH |
3725 | rcu_read_unlock(); |
3726 | ||
9bec5c35 | 3727 | memcg_drain_all_list_lrus(kmemcg_id, parent); |
8e0a8912 JW |
3728 | |
3729 | memcg_free_cache_id(kmemcg_id); | |
3730 | } | |
3731 | ||
3732 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3733 | { | |
0b8f73e1 JW |
3734 | /* css_alloc() failed, offlining didn't happen */ |
3735 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
3736 | memcg_offline_kmem(memcg); | |
8e0a8912 | 3737 | } |
d6441637 | 3738 | #else |
0b8f73e1 | 3739 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
3740 | { |
3741 | return 0; | |
3742 | } | |
3743 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
3744 | { | |
3745 | } | |
3746 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3747 | { | |
3748 | } | |
84c07d11 | 3749 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 3750 | |
bbec2e15 RG |
3751 | static int memcg_update_kmem_max(struct mem_cgroup *memcg, |
3752 | unsigned long max) | |
d6441637 | 3753 | { |
b313aeee | 3754 | int ret; |
127424c8 | 3755 | |
bbec2e15 RG |
3756 | mutex_lock(&memcg_max_mutex); |
3757 | ret = page_counter_set_max(&memcg->kmem, max); | |
3758 | mutex_unlock(&memcg_max_mutex); | |
127424c8 | 3759 | return ret; |
d6441637 | 3760 | } |
510fc4e1 | 3761 | |
bbec2e15 | 3762 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
3763 | { |
3764 | int ret; | |
3765 | ||
bbec2e15 | 3766 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 3767 | |
bbec2e15 | 3768 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
3769 | if (ret) |
3770 | goto out; | |
3771 | ||
0db15298 | 3772 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
3773 | /* |
3774 | * The active flag needs to be written after the static_key | |
3775 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
3776 | * function is the last one to run. See mem_cgroup_sk_alloc() |
3777 | * for details, and note that we don't mark any socket as | |
3778 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
3779 | * |
3780 | * We need to do this, because static_keys will span multiple | |
3781 | * sites, but we can't control their order. If we mark a socket | |
3782 | * as accounted, but the accounting functions are not patched in | |
3783 | * yet, we'll lose accounting. | |
3784 | * | |
2d758073 | 3785 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
3786 | * because when this value change, the code to process it is not |
3787 | * patched in yet. | |
3788 | */ | |
3789 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 3790 | memcg->tcpmem_active = true; |
d55f90bf VD |
3791 | } |
3792 | out: | |
bbec2e15 | 3793 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
3794 | return ret; |
3795 | } | |
d55f90bf | 3796 | |
628f4235 KH |
3797 | /* |
3798 | * The user of this function is... | |
3799 | * RES_LIMIT. | |
3800 | */ | |
451af504 TH |
3801 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3802 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3803 | { |
451af504 | 3804 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3805 | unsigned long nr_pages; |
628f4235 KH |
3806 | int ret; |
3807 | ||
451af504 | 3808 | buf = strstrip(buf); |
650c5e56 | 3809 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3810 | if (ret) |
3811 | return ret; | |
af36f906 | 3812 | |
3e32cb2e | 3813 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3814 | case RES_LIMIT: |
4b3bde4c BS |
3815 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3816 | ret = -EINVAL; | |
3817 | break; | |
3818 | } | |
3e32cb2e JW |
3819 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3820 | case _MEM: | |
bbec2e15 | 3821 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 3822 | break; |
3e32cb2e | 3823 | case _MEMSWAP: |
bbec2e15 | 3824 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 3825 | break; |
3e32cb2e | 3826 | case _KMEM: |
0158115f MH |
3827 | pr_warn_once("kmem.limit_in_bytes is deprecated and will be removed. " |
3828 | "Please report your usecase to linux-mm@kvack.org if you " | |
3829 | "depend on this functionality.\n"); | |
bbec2e15 | 3830 | ret = memcg_update_kmem_max(memcg, nr_pages); |
3e32cb2e | 3831 | break; |
d55f90bf | 3832 | case _TCP: |
bbec2e15 | 3833 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 3834 | break; |
3e32cb2e | 3835 | } |
296c81d8 | 3836 | break; |
3e32cb2e JW |
3837 | case RES_SOFT_LIMIT: |
3838 | memcg->soft_limit = nr_pages; | |
3839 | ret = 0; | |
628f4235 KH |
3840 | break; |
3841 | } | |
451af504 | 3842 | return ret ?: nbytes; |
8cdea7c0 BS |
3843 | } |
3844 | ||
6770c64e TH |
3845 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3846 | size_t nbytes, loff_t off) | |
c84872e1 | 3847 | { |
6770c64e | 3848 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3849 | struct page_counter *counter; |
c84872e1 | 3850 | |
3e32cb2e JW |
3851 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3852 | case _MEM: | |
3853 | counter = &memcg->memory; | |
3854 | break; | |
3855 | case _MEMSWAP: | |
3856 | counter = &memcg->memsw; | |
3857 | break; | |
3858 | case _KMEM: | |
3859 | counter = &memcg->kmem; | |
3860 | break; | |
d55f90bf | 3861 | case _TCP: |
0db15298 | 3862 | counter = &memcg->tcpmem; |
d55f90bf | 3863 | break; |
3e32cb2e JW |
3864 | default: |
3865 | BUG(); | |
3866 | } | |
af36f906 | 3867 | |
3e32cb2e | 3868 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3869 | case RES_MAX_USAGE: |
3e32cb2e | 3870 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3871 | break; |
3872 | case RES_FAILCNT: | |
3e32cb2e | 3873 | counter->failcnt = 0; |
29f2a4da | 3874 | break; |
3e32cb2e JW |
3875 | default: |
3876 | BUG(); | |
29f2a4da | 3877 | } |
f64c3f54 | 3878 | |
6770c64e | 3879 | return nbytes; |
c84872e1 PE |
3880 | } |
3881 | ||
182446d0 | 3882 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3883 | struct cftype *cft) |
3884 | { | |
182446d0 | 3885 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3886 | } |
3887 | ||
02491447 | 3888 | #ifdef CONFIG_MMU |
182446d0 | 3889 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3890 | struct cftype *cft, u64 val) |
3891 | { | |
182446d0 | 3892 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3893 | |
1dfab5ab | 3894 | if (val & ~MOVE_MASK) |
7dc74be0 | 3895 | return -EINVAL; |
ee5e8472 | 3896 | |
7dc74be0 | 3897 | /* |
ee5e8472 GC |
3898 | * No kind of locking is needed in here, because ->can_attach() will |
3899 | * check this value once in the beginning of the process, and then carry | |
3900 | * on with stale data. This means that changes to this value will only | |
3901 | * affect task migrations starting after the change. | |
7dc74be0 | 3902 | */ |
c0ff4b85 | 3903 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3904 | return 0; |
3905 | } | |
02491447 | 3906 | #else |
182446d0 | 3907 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3908 | struct cftype *cft, u64 val) |
3909 | { | |
3910 | return -ENOSYS; | |
3911 | } | |
3912 | #endif | |
7dc74be0 | 3913 | |
406eb0c9 | 3914 | #ifdef CONFIG_NUMA |
113b7dfd JW |
3915 | |
3916 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
3917 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
3918 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) | |
3919 | ||
3920 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 | 3921 | int nid, unsigned int lru_mask, bool tree) |
113b7dfd | 3922 | { |
867e5e1d | 3923 | struct lruvec *lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); |
113b7dfd JW |
3924 | unsigned long nr = 0; |
3925 | enum lru_list lru; | |
3926 | ||
3927 | VM_BUG_ON((unsigned)nid >= nr_node_ids); | |
3928 | ||
3929 | for_each_lru(lru) { | |
3930 | if (!(BIT(lru) & lru_mask)) | |
3931 | continue; | |
dd8657b6 SB |
3932 | if (tree) |
3933 | nr += lruvec_page_state(lruvec, NR_LRU_BASE + lru); | |
3934 | else | |
3935 | nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru); | |
113b7dfd JW |
3936 | } |
3937 | return nr; | |
3938 | } | |
3939 | ||
3940 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 SB |
3941 | unsigned int lru_mask, |
3942 | bool tree) | |
113b7dfd JW |
3943 | { |
3944 | unsigned long nr = 0; | |
3945 | enum lru_list lru; | |
3946 | ||
3947 | for_each_lru(lru) { | |
3948 | if (!(BIT(lru) & lru_mask)) | |
3949 | continue; | |
dd8657b6 SB |
3950 | if (tree) |
3951 | nr += memcg_page_state(memcg, NR_LRU_BASE + lru); | |
3952 | else | |
3953 | nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru); | |
113b7dfd JW |
3954 | } |
3955 | return nr; | |
3956 | } | |
3957 | ||
2da8ca82 | 3958 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3959 | { |
25485de6 GT |
3960 | struct numa_stat { |
3961 | const char *name; | |
3962 | unsigned int lru_mask; | |
3963 | }; | |
3964 | ||
3965 | static const struct numa_stat stats[] = { | |
3966 | { "total", LRU_ALL }, | |
3967 | { "file", LRU_ALL_FILE }, | |
3968 | { "anon", LRU_ALL_ANON }, | |
3969 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3970 | }; | |
3971 | const struct numa_stat *stat; | |
406eb0c9 | 3972 | int nid; |
aa9694bb | 3973 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
406eb0c9 | 3974 | |
25485de6 | 3975 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
3976 | seq_printf(m, "%s=%lu", stat->name, |
3977 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
3978 | false)); | |
3979 | for_each_node_state(nid, N_MEMORY) | |
3980 | seq_printf(m, " N%d=%lu", nid, | |
3981 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3982 | stat->lru_mask, false)); | |
25485de6 | 3983 | seq_putc(m, '\n'); |
406eb0c9 | 3984 | } |
406eb0c9 | 3985 | |
071aee13 | 3986 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
3987 | |
3988 | seq_printf(m, "hierarchical_%s=%lu", stat->name, | |
3989 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
3990 | true)); | |
3991 | for_each_node_state(nid, N_MEMORY) | |
3992 | seq_printf(m, " N%d=%lu", nid, | |
3993 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3994 | stat->lru_mask, true)); | |
071aee13 | 3995 | seq_putc(m, '\n'); |
406eb0c9 | 3996 | } |
406eb0c9 | 3997 | |
406eb0c9 YH |
3998 | return 0; |
3999 | } | |
4000 | #endif /* CONFIG_NUMA */ | |
4001 | ||
c8713d0b | 4002 | static const unsigned int memcg1_stats[] = { |
0d1c2072 | 4003 | NR_FILE_PAGES, |
be5d0a74 | 4004 | NR_ANON_MAPPED, |
468c3982 JW |
4005 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4006 | NR_ANON_THPS, | |
4007 | #endif | |
c8713d0b JW |
4008 | NR_SHMEM, |
4009 | NR_FILE_MAPPED, | |
4010 | NR_FILE_DIRTY, | |
4011 | NR_WRITEBACK, | |
4012 | MEMCG_SWAP, | |
4013 | }; | |
4014 | ||
4015 | static const char *const memcg1_stat_names[] = { | |
4016 | "cache", | |
4017 | "rss", | |
468c3982 | 4018 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
c8713d0b | 4019 | "rss_huge", |
468c3982 | 4020 | #endif |
c8713d0b JW |
4021 | "shmem", |
4022 | "mapped_file", | |
4023 | "dirty", | |
4024 | "writeback", | |
4025 | "swap", | |
4026 | }; | |
4027 | ||
df0e53d0 | 4028 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 4029 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
4030 | PGPGIN, |
4031 | PGPGOUT, | |
4032 | PGFAULT, | |
4033 | PGMAJFAULT, | |
4034 | }; | |
4035 | ||
2da8ca82 | 4036 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 4037 | { |
aa9694bb | 4038 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3e32cb2e | 4039 | unsigned long memory, memsw; |
af7c4b0e JW |
4040 | struct mem_cgroup *mi; |
4041 | unsigned int i; | |
406eb0c9 | 4042 | |
71cd3113 | 4043 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c | 4044 | |
71cd3113 | 4045 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
468c3982 JW |
4046 | unsigned long nr; |
4047 | ||
71cd3113 | 4048 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 4049 | continue; |
468c3982 JW |
4050 | nr = memcg_page_state_local(memcg, memcg1_stats[i]); |
4051 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
4052 | if (memcg1_stats[i] == NR_ANON_THPS) | |
4053 | nr *= HPAGE_PMD_NR; | |
4054 | #endif | |
4055 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], nr * PAGE_SIZE); | |
1dd3a273 | 4056 | } |
7b854121 | 4057 | |
df0e53d0 | 4058 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d | 4059 | seq_printf(m, "%s %lu\n", vm_event_name(memcg1_events[i]), |
205b20cc | 4060 | memcg_events_local(memcg, memcg1_events[i])); |
af7c4b0e JW |
4061 | |
4062 | for (i = 0; i < NR_LRU_LISTS; i++) | |
ebc5d83d | 4063 | seq_printf(m, "%s %lu\n", lru_list_name(i), |
205b20cc | 4064 | memcg_page_state_local(memcg, NR_LRU_BASE + i) * |
21d89d15 | 4065 | PAGE_SIZE); |
af7c4b0e | 4066 | |
14067bb3 | 4067 | /* Hierarchical information */ |
3e32cb2e JW |
4068 | memory = memsw = PAGE_COUNTER_MAX; |
4069 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
15b42562 CD |
4070 | memory = min(memory, READ_ONCE(mi->memory.max)); |
4071 | memsw = min(memsw, READ_ONCE(mi->memsw.max)); | |
fee7b548 | 4072 | } |
3e32cb2e JW |
4073 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
4074 | (u64)memory * PAGE_SIZE); | |
7941d214 | 4075 | if (do_memsw_account()) |
3e32cb2e JW |
4076 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
4077 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 4078 | |
8de7ecc6 | 4079 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
71cd3113 | 4080 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 4081 | continue; |
8de7ecc6 | 4082 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], |
dd923990 YS |
4083 | (u64)memcg_page_state(memcg, memcg1_stats[i]) * |
4084 | PAGE_SIZE); | |
af7c4b0e JW |
4085 | } |
4086 | ||
8de7ecc6 | 4087 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d KK |
4088 | seq_printf(m, "total_%s %llu\n", |
4089 | vm_event_name(memcg1_events[i]), | |
dd923990 | 4090 | (u64)memcg_events(memcg, memcg1_events[i])); |
af7c4b0e | 4091 | |
8de7ecc6 | 4092 | for (i = 0; i < NR_LRU_LISTS; i++) |
ebc5d83d | 4093 | seq_printf(m, "total_%s %llu\n", lru_list_name(i), |
42a30035 JW |
4094 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
4095 | PAGE_SIZE); | |
14067bb3 | 4096 | |
7f016ee8 | 4097 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 4098 | { |
ef8f2327 MG |
4099 | pg_data_t *pgdat; |
4100 | struct mem_cgroup_per_node *mz; | |
1431d4d1 JW |
4101 | unsigned long anon_cost = 0; |
4102 | unsigned long file_cost = 0; | |
7f016ee8 | 4103 | |
ef8f2327 MG |
4104 | for_each_online_pgdat(pgdat) { |
4105 | mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); | |
7f016ee8 | 4106 | |
1431d4d1 JW |
4107 | anon_cost += mz->lruvec.anon_cost; |
4108 | file_cost += mz->lruvec.file_cost; | |
ef8f2327 | 4109 | } |
1431d4d1 JW |
4110 | seq_printf(m, "anon_cost %lu\n", anon_cost); |
4111 | seq_printf(m, "file_cost %lu\n", file_cost); | |
7f016ee8 KM |
4112 | } |
4113 | #endif | |
4114 | ||
d2ceb9b7 KH |
4115 | return 0; |
4116 | } | |
4117 | ||
182446d0 TH |
4118 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
4119 | struct cftype *cft) | |
a7885eb8 | 4120 | { |
182446d0 | 4121 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4122 | |
1f4c025b | 4123 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4124 | } |
4125 | ||
182446d0 TH |
4126 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
4127 | struct cftype *cft, u64 val) | |
a7885eb8 | 4128 | { |
182446d0 | 4129 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4130 | |
3dae7fec | 4131 | if (val > 100) |
a7885eb8 KM |
4132 | return -EINVAL; |
4133 | ||
14208b0e | 4134 | if (css->parent) |
3dae7fec JW |
4135 | memcg->swappiness = val; |
4136 | else | |
4137 | vm_swappiness = val; | |
068b38c1 | 4138 | |
a7885eb8 KM |
4139 | return 0; |
4140 | } | |
4141 | ||
2e72b634 KS |
4142 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4143 | { | |
4144 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 4145 | unsigned long usage; |
2e72b634 KS |
4146 | int i; |
4147 | ||
4148 | rcu_read_lock(); | |
4149 | if (!swap) | |
2c488db2 | 4150 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4151 | else |
2c488db2 | 4152 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4153 | |
4154 | if (!t) | |
4155 | goto unlock; | |
4156 | ||
ce00a967 | 4157 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
4158 | |
4159 | /* | |
748dad36 | 4160 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4161 | * If it's not true, a threshold was crossed after last |
4162 | * call of __mem_cgroup_threshold(). | |
4163 | */ | |
5407a562 | 4164 | i = t->current_threshold; |
2e72b634 KS |
4165 | |
4166 | /* | |
4167 | * Iterate backward over array of thresholds starting from | |
4168 | * current_threshold and check if a threshold is crossed. | |
4169 | * If none of thresholds below usage is crossed, we read | |
4170 | * only one element of the array here. | |
4171 | */ | |
4172 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4173 | eventfd_signal(t->entries[i].eventfd, 1); | |
4174 | ||
4175 | /* i = current_threshold + 1 */ | |
4176 | i++; | |
4177 | ||
4178 | /* | |
4179 | * Iterate forward over array of thresholds starting from | |
4180 | * current_threshold+1 and check if a threshold is crossed. | |
4181 | * If none of thresholds above usage is crossed, we read | |
4182 | * only one element of the array here. | |
4183 | */ | |
4184 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4185 | eventfd_signal(t->entries[i].eventfd, 1); | |
4186 | ||
4187 | /* Update current_threshold */ | |
5407a562 | 4188 | t->current_threshold = i - 1; |
2e72b634 KS |
4189 | unlock: |
4190 | rcu_read_unlock(); | |
4191 | } | |
4192 | ||
4193 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4194 | { | |
ad4ca5f4 KS |
4195 | while (memcg) { |
4196 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 4197 | if (do_memsw_account()) |
ad4ca5f4 KS |
4198 | __mem_cgroup_threshold(memcg, true); |
4199 | ||
4200 | memcg = parent_mem_cgroup(memcg); | |
4201 | } | |
2e72b634 KS |
4202 | } |
4203 | ||
4204 | static int compare_thresholds(const void *a, const void *b) | |
4205 | { | |
4206 | const struct mem_cgroup_threshold *_a = a; | |
4207 | const struct mem_cgroup_threshold *_b = b; | |
4208 | ||
2bff24a3 GT |
4209 | if (_a->threshold > _b->threshold) |
4210 | return 1; | |
4211 | ||
4212 | if (_a->threshold < _b->threshold) | |
4213 | return -1; | |
4214 | ||
4215 | return 0; | |
2e72b634 KS |
4216 | } |
4217 | ||
c0ff4b85 | 4218 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4219 | { |
4220 | struct mem_cgroup_eventfd_list *ev; | |
4221 | ||
2bcf2e92 MH |
4222 | spin_lock(&memcg_oom_lock); |
4223 | ||
c0ff4b85 | 4224 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 4225 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
4226 | |
4227 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
4228 | return 0; |
4229 | } | |
4230 | ||
c0ff4b85 | 4231 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4232 | { |
7d74b06f KH |
4233 | struct mem_cgroup *iter; |
4234 | ||
c0ff4b85 | 4235 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4236 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4237 | } |
4238 | ||
59b6f873 | 4239 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4240 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4241 | { |
2c488db2 KS |
4242 | struct mem_cgroup_thresholds *thresholds; |
4243 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
4244 | unsigned long threshold; |
4245 | unsigned long usage; | |
2c488db2 | 4246 | int i, size, ret; |
2e72b634 | 4247 | |
650c5e56 | 4248 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
4249 | if (ret) |
4250 | return ret; | |
4251 | ||
4252 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4253 | |
05b84301 | 4254 | if (type == _MEM) { |
2c488db2 | 4255 | thresholds = &memcg->thresholds; |
ce00a967 | 4256 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4257 | } else if (type == _MEMSWAP) { |
2c488db2 | 4258 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4259 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4260 | } else |
2e72b634 KS |
4261 | BUG(); |
4262 | ||
2e72b634 | 4263 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4264 | if (thresholds->primary) |
2e72b634 KS |
4265 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4266 | ||
2c488db2 | 4267 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4268 | |
4269 | /* Allocate memory for new array of thresholds */ | |
67b8046f | 4270 | new = kmalloc(struct_size(new, entries, size), GFP_KERNEL); |
2c488db2 | 4271 | if (!new) { |
2e72b634 KS |
4272 | ret = -ENOMEM; |
4273 | goto unlock; | |
4274 | } | |
2c488db2 | 4275 | new->size = size; |
2e72b634 KS |
4276 | |
4277 | /* Copy thresholds (if any) to new array */ | |
e90342e6 GS |
4278 | if (thresholds->primary) |
4279 | memcpy(new->entries, thresholds->primary->entries, | |
4280 | flex_array_size(new, entries, size - 1)); | |
2c488db2 | 4281 | |
2e72b634 | 4282 | /* Add new threshold */ |
2c488db2 KS |
4283 | new->entries[size - 1].eventfd = eventfd; |
4284 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4285 | |
4286 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
61e604e6 | 4287 | sort(new->entries, size, sizeof(*new->entries), |
2e72b634 KS |
4288 | compare_thresholds, NULL); |
4289 | ||
4290 | /* Find current threshold */ | |
2c488db2 | 4291 | new->current_threshold = -1; |
2e72b634 | 4292 | for (i = 0; i < size; i++) { |
748dad36 | 4293 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4294 | /* |
2c488db2 KS |
4295 | * new->current_threshold will not be used until |
4296 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4297 | * it here. |
4298 | */ | |
2c488db2 | 4299 | ++new->current_threshold; |
748dad36 SZ |
4300 | } else |
4301 | break; | |
2e72b634 KS |
4302 | } |
4303 | ||
2c488db2 KS |
4304 | /* Free old spare buffer and save old primary buffer as spare */ |
4305 | kfree(thresholds->spare); | |
4306 | thresholds->spare = thresholds->primary; | |
4307 | ||
4308 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4309 | |
907860ed | 4310 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4311 | synchronize_rcu(); |
4312 | ||
2e72b634 KS |
4313 | unlock: |
4314 | mutex_unlock(&memcg->thresholds_lock); | |
4315 | ||
4316 | return ret; | |
4317 | } | |
4318 | ||
59b6f873 | 4319 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4320 | struct eventfd_ctx *eventfd, const char *args) |
4321 | { | |
59b6f873 | 4322 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4323 | } |
4324 | ||
59b6f873 | 4325 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4326 | struct eventfd_ctx *eventfd, const char *args) |
4327 | { | |
59b6f873 | 4328 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4329 | } |
4330 | ||
59b6f873 | 4331 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4332 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4333 | { |
2c488db2 KS |
4334 | struct mem_cgroup_thresholds *thresholds; |
4335 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 4336 | unsigned long usage; |
7d36665a | 4337 | int i, j, size, entries; |
2e72b634 KS |
4338 | |
4339 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4340 | |
4341 | if (type == _MEM) { | |
2c488db2 | 4342 | thresholds = &memcg->thresholds; |
ce00a967 | 4343 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4344 | } else if (type == _MEMSWAP) { |
2c488db2 | 4345 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4346 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4347 | } else |
2e72b634 KS |
4348 | BUG(); |
4349 | ||
371528ca AV |
4350 | if (!thresholds->primary) |
4351 | goto unlock; | |
4352 | ||
2e72b634 KS |
4353 | /* Check if a threshold crossed before removing */ |
4354 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4355 | ||
4356 | /* Calculate new number of threshold */ | |
7d36665a | 4357 | size = entries = 0; |
2c488db2 KS |
4358 | for (i = 0; i < thresholds->primary->size; i++) { |
4359 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 | 4360 | size++; |
7d36665a CX |
4361 | else |
4362 | entries++; | |
2e72b634 KS |
4363 | } |
4364 | ||
2c488db2 | 4365 | new = thresholds->spare; |
907860ed | 4366 | |
7d36665a CX |
4367 | /* If no items related to eventfd have been cleared, nothing to do */ |
4368 | if (!entries) | |
4369 | goto unlock; | |
4370 | ||
2e72b634 KS |
4371 | /* Set thresholds array to NULL if we don't have thresholds */ |
4372 | if (!size) { | |
2c488db2 KS |
4373 | kfree(new); |
4374 | new = NULL; | |
907860ed | 4375 | goto swap_buffers; |
2e72b634 KS |
4376 | } |
4377 | ||
2c488db2 | 4378 | new->size = size; |
2e72b634 KS |
4379 | |
4380 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4381 | new->current_threshold = -1; |
4382 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4383 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4384 | continue; |
4385 | ||
2c488db2 | 4386 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4387 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4388 | /* |
2c488db2 | 4389 | * new->current_threshold will not be used |
2e72b634 KS |
4390 | * until rcu_assign_pointer(), so it's safe to increment |
4391 | * it here. | |
4392 | */ | |
2c488db2 | 4393 | ++new->current_threshold; |
2e72b634 KS |
4394 | } |
4395 | j++; | |
4396 | } | |
4397 | ||
907860ed | 4398 | swap_buffers: |
2c488db2 KS |
4399 | /* Swap primary and spare array */ |
4400 | thresholds->spare = thresholds->primary; | |
8c757763 | 4401 | |
2c488db2 | 4402 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4403 | |
907860ed | 4404 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4405 | synchronize_rcu(); |
6611d8d7 MC |
4406 | |
4407 | /* If all events are unregistered, free the spare array */ | |
4408 | if (!new) { | |
4409 | kfree(thresholds->spare); | |
4410 | thresholds->spare = NULL; | |
4411 | } | |
371528ca | 4412 | unlock: |
2e72b634 | 4413 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4414 | } |
c1e862c1 | 4415 | |
59b6f873 | 4416 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4417 | struct eventfd_ctx *eventfd) |
4418 | { | |
59b6f873 | 4419 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4420 | } |
4421 | ||
59b6f873 | 4422 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4423 | struct eventfd_ctx *eventfd) |
4424 | { | |
59b6f873 | 4425 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4426 | } |
4427 | ||
59b6f873 | 4428 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4429 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4430 | { |
9490ff27 | 4431 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4432 | |
9490ff27 KH |
4433 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4434 | if (!event) | |
4435 | return -ENOMEM; | |
4436 | ||
1af8efe9 | 4437 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4438 | |
4439 | event->eventfd = eventfd; | |
4440 | list_add(&event->list, &memcg->oom_notify); | |
4441 | ||
4442 | /* already in OOM ? */ | |
c2b42d3c | 4443 | if (memcg->under_oom) |
9490ff27 | 4444 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4445 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4446 | |
4447 | return 0; | |
4448 | } | |
4449 | ||
59b6f873 | 4450 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4451 | struct eventfd_ctx *eventfd) |
9490ff27 | 4452 | { |
9490ff27 | 4453 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4454 | |
1af8efe9 | 4455 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4456 | |
c0ff4b85 | 4457 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4458 | if (ev->eventfd == eventfd) { |
4459 | list_del(&ev->list); | |
4460 | kfree(ev); | |
4461 | } | |
4462 | } | |
4463 | ||
1af8efe9 | 4464 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4465 | } |
4466 | ||
2da8ca82 | 4467 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4468 | { |
aa9694bb | 4469 | struct mem_cgroup *memcg = mem_cgroup_from_seq(sf); |
3c11ecf4 | 4470 | |
791badbd | 4471 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 4472 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
4473 | seq_printf(sf, "oom_kill %lu\n", |
4474 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
4475 | return 0; |
4476 | } | |
4477 | ||
182446d0 | 4478 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4479 | struct cftype *cft, u64 val) |
4480 | { | |
182446d0 | 4481 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4482 | |
4483 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 4484 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4485 | return -EINVAL; |
4486 | ||
c0ff4b85 | 4487 | memcg->oom_kill_disable = val; |
4d845ebf | 4488 | if (!val) |
c0ff4b85 | 4489 | memcg_oom_recover(memcg); |
3dae7fec | 4490 | |
3c11ecf4 KH |
4491 | return 0; |
4492 | } | |
4493 | ||
52ebea74 TH |
4494 | #ifdef CONFIG_CGROUP_WRITEBACK |
4495 | ||
3a8e9ac8 TH |
4496 | #include <trace/events/writeback.h> |
4497 | ||
841710aa TH |
4498 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
4499 | { | |
4500 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
4501 | } | |
4502 | ||
4503 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4504 | { | |
4505 | wb_domain_exit(&memcg->cgwb_domain); | |
4506 | } | |
4507 | ||
2529bb3a TH |
4508 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4509 | { | |
4510 | wb_domain_size_changed(&memcg->cgwb_domain); | |
4511 | } | |
4512 | ||
841710aa TH |
4513 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
4514 | { | |
4515 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4516 | ||
4517 | if (!memcg->css.parent) | |
4518 | return NULL; | |
4519 | ||
4520 | return &memcg->cgwb_domain; | |
4521 | } | |
4522 | ||
0b3d6e6f GT |
4523 | /* |
4524 | * idx can be of type enum memcg_stat_item or node_stat_item. | |
4525 | * Keep in sync with memcg_exact_page(). | |
4526 | */ | |
4527 | static unsigned long memcg_exact_page_state(struct mem_cgroup *memcg, int idx) | |
4528 | { | |
871789d4 | 4529 | long x = atomic_long_read(&memcg->vmstats[idx]); |
0b3d6e6f GT |
4530 | int cpu; |
4531 | ||
4532 | for_each_online_cpu(cpu) | |
871789d4 | 4533 | x += per_cpu_ptr(memcg->vmstats_percpu, cpu)->stat[idx]; |
0b3d6e6f GT |
4534 | if (x < 0) |
4535 | x = 0; | |
4536 | return x; | |
4537 | } | |
4538 | ||
c2aa723a TH |
4539 | /** |
4540 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
4541 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
4542 | * @pfilepages: out parameter for number of file pages |
4543 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
4544 | * @pdirty: out parameter for number of dirty pages |
4545 | * @pwriteback: out parameter for number of pages under writeback | |
4546 | * | |
c5edf9cd TH |
4547 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
4548 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
4549 | * is a bit more involved. | |
c2aa723a | 4550 | * |
c5edf9cd TH |
4551 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
4552 | * headroom is calculated as the lowest headroom of itself and the | |
4553 | * ancestors. Note that this doesn't consider the actual amount of | |
4554 | * available memory in the system. The caller should further cap | |
4555 | * *@pheadroom accordingly. | |
c2aa723a | 4556 | */ |
c5edf9cd TH |
4557 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
4558 | unsigned long *pheadroom, unsigned long *pdirty, | |
4559 | unsigned long *pwriteback) | |
c2aa723a TH |
4560 | { |
4561 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4562 | struct mem_cgroup *parent; | |
c2aa723a | 4563 | |
0b3d6e6f | 4564 | *pdirty = memcg_exact_page_state(memcg, NR_FILE_DIRTY); |
c2aa723a | 4565 | |
0b3d6e6f | 4566 | *pwriteback = memcg_exact_page_state(memcg, NR_WRITEBACK); |
21d89d15 JW |
4567 | *pfilepages = memcg_exact_page_state(memcg, NR_INACTIVE_FILE) + |
4568 | memcg_exact_page_state(memcg, NR_ACTIVE_FILE); | |
c5edf9cd | 4569 | *pheadroom = PAGE_COUNTER_MAX; |
c2aa723a | 4570 | |
c2aa723a | 4571 | while ((parent = parent_mem_cgroup(memcg))) { |
15b42562 | 4572 | unsigned long ceiling = min(READ_ONCE(memcg->memory.max), |
d1663a90 | 4573 | READ_ONCE(memcg->memory.high)); |
c2aa723a TH |
4574 | unsigned long used = page_counter_read(&memcg->memory); |
4575 | ||
c5edf9cd | 4576 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
4577 | memcg = parent; |
4578 | } | |
c2aa723a TH |
4579 | } |
4580 | ||
97b27821 TH |
4581 | /* |
4582 | * Foreign dirty flushing | |
4583 | * | |
4584 | * There's an inherent mismatch between memcg and writeback. The former | |
4585 | * trackes ownership per-page while the latter per-inode. This was a | |
4586 | * deliberate design decision because honoring per-page ownership in the | |
4587 | * writeback path is complicated, may lead to higher CPU and IO overheads | |
4588 | * and deemed unnecessary given that write-sharing an inode across | |
4589 | * different cgroups isn't a common use-case. | |
4590 | * | |
4591 | * Combined with inode majority-writer ownership switching, this works well | |
4592 | * enough in most cases but there are some pathological cases. For | |
4593 | * example, let's say there are two cgroups A and B which keep writing to | |
4594 | * different but confined parts of the same inode. B owns the inode and | |
4595 | * A's memory is limited far below B's. A's dirty ratio can rise enough to | |
4596 | * trigger balance_dirty_pages() sleeps but B's can be low enough to avoid | |
4597 | * triggering background writeback. A will be slowed down without a way to | |
4598 | * make writeback of the dirty pages happen. | |
4599 | * | |
4600 | * Conditions like the above can lead to a cgroup getting repatedly and | |
4601 | * severely throttled after making some progress after each | |
4602 | * dirty_expire_interval while the underyling IO device is almost | |
4603 | * completely idle. | |
4604 | * | |
4605 | * Solving this problem completely requires matching the ownership tracking | |
4606 | * granularities between memcg and writeback in either direction. However, | |
4607 | * the more egregious behaviors can be avoided by simply remembering the | |
4608 | * most recent foreign dirtying events and initiating remote flushes on | |
4609 | * them when local writeback isn't enough to keep the memory clean enough. | |
4610 | * | |
4611 | * The following two functions implement such mechanism. When a foreign | |
4612 | * page - a page whose memcg and writeback ownerships don't match - is | |
4613 | * dirtied, mem_cgroup_track_foreign_dirty() records the inode owning | |
4614 | * bdi_writeback on the page owning memcg. When balance_dirty_pages() | |
4615 | * decides that the memcg needs to sleep due to high dirty ratio, it calls | |
4616 | * mem_cgroup_flush_foreign() which queues writeback on the recorded | |
4617 | * foreign bdi_writebacks which haven't expired. Both the numbers of | |
4618 | * recorded bdi_writebacks and concurrent in-flight foreign writebacks are | |
4619 | * limited to MEMCG_CGWB_FRN_CNT. | |
4620 | * | |
4621 | * The mechanism only remembers IDs and doesn't hold any object references. | |
4622 | * As being wrong occasionally doesn't matter, updates and accesses to the | |
4623 | * records are lockless and racy. | |
4624 | */ | |
4625 | void mem_cgroup_track_foreign_dirty_slowpath(struct page *page, | |
4626 | struct bdi_writeback *wb) | |
4627 | { | |
4628 | struct mem_cgroup *memcg = page->mem_cgroup; | |
4629 | struct memcg_cgwb_frn *frn; | |
4630 | u64 now = get_jiffies_64(); | |
4631 | u64 oldest_at = now; | |
4632 | int oldest = -1; | |
4633 | int i; | |
4634 | ||
3a8e9ac8 TH |
4635 | trace_track_foreign_dirty(page, wb); |
4636 | ||
97b27821 TH |
4637 | /* |
4638 | * Pick the slot to use. If there is already a slot for @wb, keep | |
4639 | * using it. If not replace the oldest one which isn't being | |
4640 | * written out. | |
4641 | */ | |
4642 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4643 | frn = &memcg->cgwb_frn[i]; | |
4644 | if (frn->bdi_id == wb->bdi->id && | |
4645 | frn->memcg_id == wb->memcg_css->id) | |
4646 | break; | |
4647 | if (time_before64(frn->at, oldest_at) && | |
4648 | atomic_read(&frn->done.cnt) == 1) { | |
4649 | oldest = i; | |
4650 | oldest_at = frn->at; | |
4651 | } | |
4652 | } | |
4653 | ||
4654 | if (i < MEMCG_CGWB_FRN_CNT) { | |
4655 | /* | |
4656 | * Re-using an existing one. Update timestamp lazily to | |
4657 | * avoid making the cacheline hot. We want them to be | |
4658 | * reasonably up-to-date and significantly shorter than | |
4659 | * dirty_expire_interval as that's what expires the record. | |
4660 | * Use the shorter of 1s and dirty_expire_interval / 8. | |
4661 | */ | |
4662 | unsigned long update_intv = | |
4663 | min_t(unsigned long, HZ, | |
4664 | msecs_to_jiffies(dirty_expire_interval * 10) / 8); | |
4665 | ||
4666 | if (time_before64(frn->at, now - update_intv)) | |
4667 | frn->at = now; | |
4668 | } else if (oldest >= 0) { | |
4669 | /* replace the oldest free one */ | |
4670 | frn = &memcg->cgwb_frn[oldest]; | |
4671 | frn->bdi_id = wb->bdi->id; | |
4672 | frn->memcg_id = wb->memcg_css->id; | |
4673 | frn->at = now; | |
4674 | } | |
4675 | } | |
4676 | ||
4677 | /* issue foreign writeback flushes for recorded foreign dirtying events */ | |
4678 | void mem_cgroup_flush_foreign(struct bdi_writeback *wb) | |
4679 | { | |
4680 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4681 | unsigned long intv = msecs_to_jiffies(dirty_expire_interval * 10); | |
4682 | u64 now = jiffies_64; | |
4683 | int i; | |
4684 | ||
4685 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4686 | struct memcg_cgwb_frn *frn = &memcg->cgwb_frn[i]; | |
4687 | ||
4688 | /* | |
4689 | * If the record is older than dirty_expire_interval, | |
4690 | * writeback on it has already started. No need to kick it | |
4691 | * off again. Also, don't start a new one if there's | |
4692 | * already one in flight. | |
4693 | */ | |
4694 | if (time_after64(frn->at, now - intv) && | |
4695 | atomic_read(&frn->done.cnt) == 1) { | |
4696 | frn->at = 0; | |
3a8e9ac8 | 4697 | trace_flush_foreign(wb, frn->bdi_id, frn->memcg_id); |
97b27821 TH |
4698 | cgroup_writeback_by_id(frn->bdi_id, frn->memcg_id, 0, |
4699 | WB_REASON_FOREIGN_FLUSH, | |
4700 | &frn->done); | |
4701 | } | |
4702 | } | |
4703 | } | |
4704 | ||
841710aa TH |
4705 | #else /* CONFIG_CGROUP_WRITEBACK */ |
4706 | ||
4707 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
4708 | { | |
4709 | return 0; | |
4710 | } | |
4711 | ||
4712 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4713 | { | |
4714 | } | |
4715 | ||
2529bb3a TH |
4716 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4717 | { | |
4718 | } | |
4719 | ||
52ebea74 TH |
4720 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
4721 | ||
3bc942f3 TH |
4722 | /* |
4723 | * DO NOT USE IN NEW FILES. | |
4724 | * | |
4725 | * "cgroup.event_control" implementation. | |
4726 | * | |
4727 | * This is way over-engineered. It tries to support fully configurable | |
4728 | * events for each user. Such level of flexibility is completely | |
4729 | * unnecessary especially in the light of the planned unified hierarchy. | |
4730 | * | |
4731 | * Please deprecate this and replace with something simpler if at all | |
4732 | * possible. | |
4733 | */ | |
4734 | ||
79bd9814 TH |
4735 | /* |
4736 | * Unregister event and free resources. | |
4737 | * | |
4738 | * Gets called from workqueue. | |
4739 | */ | |
3bc942f3 | 4740 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4741 | { |
3bc942f3 TH |
4742 | struct mem_cgroup_event *event = |
4743 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4744 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4745 | |
4746 | remove_wait_queue(event->wqh, &event->wait); | |
4747 | ||
59b6f873 | 4748 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4749 | |
4750 | /* Notify userspace the event is going away. */ | |
4751 | eventfd_signal(event->eventfd, 1); | |
4752 | ||
4753 | eventfd_ctx_put(event->eventfd); | |
4754 | kfree(event); | |
59b6f873 | 4755 | css_put(&memcg->css); |
79bd9814 TH |
4756 | } |
4757 | ||
4758 | /* | |
a9a08845 | 4759 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
4760 | * |
4761 | * Called with wqh->lock held and interrupts disabled. | |
4762 | */ | |
ac6424b9 | 4763 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 4764 | int sync, void *key) |
79bd9814 | 4765 | { |
3bc942f3 TH |
4766 | struct mem_cgroup_event *event = |
4767 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4768 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 4769 | __poll_t flags = key_to_poll(key); |
79bd9814 | 4770 | |
a9a08845 | 4771 | if (flags & EPOLLHUP) { |
79bd9814 TH |
4772 | /* |
4773 | * If the event has been detached at cgroup removal, we | |
4774 | * can simply return knowing the other side will cleanup | |
4775 | * for us. | |
4776 | * | |
4777 | * We can't race against event freeing since the other | |
4778 | * side will require wqh->lock via remove_wait_queue(), | |
4779 | * which we hold. | |
4780 | */ | |
fba94807 | 4781 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4782 | if (!list_empty(&event->list)) { |
4783 | list_del_init(&event->list); | |
4784 | /* | |
4785 | * We are in atomic context, but cgroup_event_remove() | |
4786 | * may sleep, so we have to call it in workqueue. | |
4787 | */ | |
4788 | schedule_work(&event->remove); | |
4789 | } | |
fba94807 | 4790 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4791 | } |
4792 | ||
4793 | return 0; | |
4794 | } | |
4795 | ||
3bc942f3 | 4796 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4797 | wait_queue_head_t *wqh, poll_table *pt) |
4798 | { | |
3bc942f3 TH |
4799 | struct mem_cgroup_event *event = |
4800 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4801 | |
4802 | event->wqh = wqh; | |
4803 | add_wait_queue(wqh, &event->wait); | |
4804 | } | |
4805 | ||
4806 | /* | |
3bc942f3 TH |
4807 | * DO NOT USE IN NEW FILES. |
4808 | * | |
79bd9814 TH |
4809 | * Parse input and register new cgroup event handler. |
4810 | * | |
4811 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4812 | * Interpretation of args is defined by control file implementation. | |
4813 | */ | |
451af504 TH |
4814 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4815 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4816 | { |
451af504 | 4817 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4818 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4819 | struct mem_cgroup_event *event; |
79bd9814 TH |
4820 | struct cgroup_subsys_state *cfile_css; |
4821 | unsigned int efd, cfd; | |
4822 | struct fd efile; | |
4823 | struct fd cfile; | |
fba94807 | 4824 | const char *name; |
79bd9814 TH |
4825 | char *endp; |
4826 | int ret; | |
4827 | ||
451af504 TH |
4828 | buf = strstrip(buf); |
4829 | ||
4830 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4831 | if (*endp != ' ') |
4832 | return -EINVAL; | |
451af504 | 4833 | buf = endp + 1; |
79bd9814 | 4834 | |
451af504 | 4835 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4836 | if ((*endp != ' ') && (*endp != '\0')) |
4837 | return -EINVAL; | |
451af504 | 4838 | buf = endp + 1; |
79bd9814 TH |
4839 | |
4840 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4841 | if (!event) | |
4842 | return -ENOMEM; | |
4843 | ||
59b6f873 | 4844 | event->memcg = memcg; |
79bd9814 | 4845 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4846 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4847 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4848 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4849 | |
4850 | efile = fdget(efd); | |
4851 | if (!efile.file) { | |
4852 | ret = -EBADF; | |
4853 | goto out_kfree; | |
4854 | } | |
4855 | ||
4856 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4857 | if (IS_ERR(event->eventfd)) { | |
4858 | ret = PTR_ERR(event->eventfd); | |
4859 | goto out_put_efile; | |
4860 | } | |
4861 | ||
4862 | cfile = fdget(cfd); | |
4863 | if (!cfile.file) { | |
4864 | ret = -EBADF; | |
4865 | goto out_put_eventfd; | |
4866 | } | |
4867 | ||
4868 | /* the process need read permission on control file */ | |
4869 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
4870 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
4871 | if (ret < 0) | |
4872 | goto out_put_cfile; | |
4873 | ||
fba94807 TH |
4874 | /* |
4875 | * Determine the event callbacks and set them in @event. This used | |
4876 | * to be done via struct cftype but cgroup core no longer knows | |
4877 | * about these events. The following is crude but the whole thing | |
4878 | * is for compatibility anyway. | |
3bc942f3 TH |
4879 | * |
4880 | * DO NOT ADD NEW FILES. | |
fba94807 | 4881 | */ |
b583043e | 4882 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4883 | |
4884 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4885 | event->register_event = mem_cgroup_usage_register_event; | |
4886 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4887 | } else if (!strcmp(name, "memory.oom_control")) { | |
4888 | event->register_event = mem_cgroup_oom_register_event; | |
4889 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4890 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4891 | event->register_event = vmpressure_register_event; | |
4892 | event->unregister_event = vmpressure_unregister_event; | |
4893 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4894 | event->register_event = memsw_cgroup_usage_register_event; |
4895 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4896 | } else { |
4897 | ret = -EINVAL; | |
4898 | goto out_put_cfile; | |
4899 | } | |
4900 | ||
79bd9814 | 4901 | /* |
b5557c4c TH |
4902 | * Verify @cfile should belong to @css. Also, remaining events are |
4903 | * automatically removed on cgroup destruction but the removal is | |
4904 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4905 | */ |
b583043e | 4906 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4907 | &memory_cgrp_subsys); |
79bd9814 | 4908 | ret = -EINVAL; |
5a17f543 | 4909 | if (IS_ERR(cfile_css)) |
79bd9814 | 4910 | goto out_put_cfile; |
5a17f543 TH |
4911 | if (cfile_css != css) { |
4912 | css_put(cfile_css); | |
79bd9814 | 4913 | goto out_put_cfile; |
5a17f543 | 4914 | } |
79bd9814 | 4915 | |
451af504 | 4916 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4917 | if (ret) |
4918 | goto out_put_css; | |
4919 | ||
9965ed17 | 4920 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 4921 | |
fba94807 TH |
4922 | spin_lock(&memcg->event_list_lock); |
4923 | list_add(&event->list, &memcg->event_list); | |
4924 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4925 | |
4926 | fdput(cfile); | |
4927 | fdput(efile); | |
4928 | ||
451af504 | 4929 | return nbytes; |
79bd9814 TH |
4930 | |
4931 | out_put_css: | |
b5557c4c | 4932 | css_put(css); |
79bd9814 TH |
4933 | out_put_cfile: |
4934 | fdput(cfile); | |
4935 | out_put_eventfd: | |
4936 | eventfd_ctx_put(event->eventfd); | |
4937 | out_put_efile: | |
4938 | fdput(efile); | |
4939 | out_kfree: | |
4940 | kfree(event); | |
4941 | ||
4942 | return ret; | |
4943 | } | |
4944 | ||
241994ed | 4945 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4946 | { |
0eea1030 | 4947 | .name = "usage_in_bytes", |
8c7c6e34 | 4948 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4949 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4950 | }, |
c84872e1 PE |
4951 | { |
4952 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4953 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4954 | .write = mem_cgroup_reset, |
791badbd | 4955 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4956 | }, |
8cdea7c0 | 4957 | { |
0eea1030 | 4958 | .name = "limit_in_bytes", |
8c7c6e34 | 4959 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4960 | .write = mem_cgroup_write, |
791badbd | 4961 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4962 | }, |
296c81d8 BS |
4963 | { |
4964 | .name = "soft_limit_in_bytes", | |
4965 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4966 | .write = mem_cgroup_write, |
791badbd | 4967 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4968 | }, |
8cdea7c0 BS |
4969 | { |
4970 | .name = "failcnt", | |
8c7c6e34 | 4971 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4972 | .write = mem_cgroup_reset, |
791badbd | 4973 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4974 | }, |
d2ceb9b7 KH |
4975 | { |
4976 | .name = "stat", | |
2da8ca82 | 4977 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4978 | }, |
c1e862c1 KH |
4979 | { |
4980 | .name = "force_empty", | |
6770c64e | 4981 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4982 | }, |
18f59ea7 BS |
4983 | { |
4984 | .name = "use_hierarchy", | |
4985 | .write_u64 = mem_cgroup_hierarchy_write, | |
4986 | .read_u64 = mem_cgroup_hierarchy_read, | |
4987 | }, | |
79bd9814 | 4988 | { |
3bc942f3 | 4989 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4990 | .write = memcg_write_event_control, |
7dbdb199 | 4991 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 4992 | }, |
a7885eb8 KM |
4993 | { |
4994 | .name = "swappiness", | |
4995 | .read_u64 = mem_cgroup_swappiness_read, | |
4996 | .write_u64 = mem_cgroup_swappiness_write, | |
4997 | }, | |
7dc74be0 DN |
4998 | { |
4999 | .name = "move_charge_at_immigrate", | |
5000 | .read_u64 = mem_cgroup_move_charge_read, | |
5001 | .write_u64 = mem_cgroup_move_charge_write, | |
5002 | }, | |
9490ff27 KH |
5003 | { |
5004 | .name = "oom_control", | |
2da8ca82 | 5005 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 5006 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
5007 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
5008 | }, | |
70ddf637 AV |
5009 | { |
5010 | .name = "pressure_level", | |
70ddf637 | 5011 | }, |
406eb0c9 YH |
5012 | #ifdef CONFIG_NUMA |
5013 | { | |
5014 | .name = "numa_stat", | |
2da8ca82 | 5015 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
5016 | }, |
5017 | #endif | |
510fc4e1 GC |
5018 | { |
5019 | .name = "kmem.limit_in_bytes", | |
5020 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 5021 | .write = mem_cgroup_write, |
791badbd | 5022 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5023 | }, |
5024 | { | |
5025 | .name = "kmem.usage_in_bytes", | |
5026 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 5027 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5028 | }, |
5029 | { | |
5030 | .name = "kmem.failcnt", | |
5031 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 5032 | .write = mem_cgroup_reset, |
791badbd | 5033 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5034 | }, |
5035 | { | |
5036 | .name = "kmem.max_usage_in_bytes", | |
5037 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 5038 | .write = mem_cgroup_reset, |
791badbd | 5039 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 5040 | }, |
a87425a3 YS |
5041 | #if defined(CONFIG_MEMCG_KMEM) && \ |
5042 | (defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)) | |
749c5415 GC |
5043 | { |
5044 | .name = "kmem.slabinfo", | |
b047501c | 5045 | .seq_show = memcg_slab_show, |
749c5415 GC |
5046 | }, |
5047 | #endif | |
d55f90bf VD |
5048 | { |
5049 | .name = "kmem.tcp.limit_in_bytes", | |
5050 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
5051 | .write = mem_cgroup_write, | |
5052 | .read_u64 = mem_cgroup_read_u64, | |
5053 | }, | |
5054 | { | |
5055 | .name = "kmem.tcp.usage_in_bytes", | |
5056 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
5057 | .read_u64 = mem_cgroup_read_u64, | |
5058 | }, | |
5059 | { | |
5060 | .name = "kmem.tcp.failcnt", | |
5061 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
5062 | .write = mem_cgroup_reset, | |
5063 | .read_u64 = mem_cgroup_read_u64, | |
5064 | }, | |
5065 | { | |
5066 | .name = "kmem.tcp.max_usage_in_bytes", | |
5067 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
5068 | .write = mem_cgroup_reset, | |
5069 | .read_u64 = mem_cgroup_read_u64, | |
5070 | }, | |
6bc10349 | 5071 | { }, /* terminate */ |
af36f906 | 5072 | }; |
8c7c6e34 | 5073 | |
73f576c0 JW |
5074 | /* |
5075 | * Private memory cgroup IDR | |
5076 | * | |
5077 | * Swap-out records and page cache shadow entries need to store memcg | |
5078 | * references in constrained space, so we maintain an ID space that is | |
5079 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
5080 | * memory-controlled cgroups to 64k. | |
5081 | * | |
b8f2935f | 5082 | * However, there usually are many references to the offline CSS after |
73f576c0 JW |
5083 | * the cgroup has been destroyed, such as page cache or reclaimable |
5084 | * slab objects, that don't need to hang on to the ID. We want to keep | |
5085 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
5086 | * relatively small ID space and prevent the creation of new cgroups | |
5087 | * even when there are much fewer than 64k cgroups - possibly none. | |
5088 | * | |
5089 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
5090 | * be freed and recycled when it's no longer needed, which is usually | |
5091 | * when the CSS is offlined. | |
5092 | * | |
5093 | * The only exception to that are records of swapped out tmpfs/shmem | |
5094 | * pages that need to be attributed to live ancestors on swapin. But | |
5095 | * those references are manageable from userspace. | |
5096 | */ | |
5097 | ||
5098 | static DEFINE_IDR(mem_cgroup_idr); | |
5099 | ||
7e97de0b KT |
5100 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
5101 | { | |
5102 | if (memcg->id.id > 0) { | |
5103 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
5104 | memcg->id.id = 0; | |
5105 | } | |
5106 | } | |
5107 | ||
c1514c0a VF |
5108 | static void __maybe_unused mem_cgroup_id_get_many(struct mem_cgroup *memcg, |
5109 | unsigned int n) | |
73f576c0 | 5110 | { |
1c2d479a | 5111 | refcount_add(n, &memcg->id.ref); |
73f576c0 JW |
5112 | } |
5113 | ||
615d66c3 | 5114 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 5115 | { |
1c2d479a | 5116 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 5117 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
5118 | |
5119 | /* Memcg ID pins CSS */ | |
5120 | css_put(&memcg->css); | |
5121 | } | |
5122 | } | |
5123 | ||
615d66c3 VD |
5124 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) |
5125 | { | |
5126 | mem_cgroup_id_put_many(memcg, 1); | |
5127 | } | |
5128 | ||
73f576c0 JW |
5129 | /** |
5130 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
5131 | * @id: the memcg id to look up | |
5132 | * | |
5133 | * Caller must hold rcu_read_lock(). | |
5134 | */ | |
5135 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
5136 | { | |
5137 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5138 | return idr_find(&mem_cgroup_idr, id); | |
5139 | } | |
5140 | ||
ef8f2327 | 5141 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
5142 | { |
5143 | struct mem_cgroup_per_node *pn; | |
ef8f2327 | 5144 | int tmp = node; |
1ecaab2b KH |
5145 | /* |
5146 | * This routine is called against possible nodes. | |
5147 | * But it's BUG to call kmalloc() against offline node. | |
5148 | * | |
5149 | * TODO: this routine can waste much memory for nodes which will | |
5150 | * never be onlined. It's better to use memory hotplug callback | |
5151 | * function. | |
5152 | */ | |
41e3355d KH |
5153 | if (!node_state(node, N_NORMAL_MEMORY)) |
5154 | tmp = -1; | |
17295c88 | 5155 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
5156 | if (!pn) |
5157 | return 1; | |
1ecaab2b | 5158 | |
3e38e0aa RG |
5159 | pn->lruvec_stat_local = alloc_percpu_gfp(struct lruvec_stat, |
5160 | GFP_KERNEL_ACCOUNT); | |
815744d7 JW |
5161 | if (!pn->lruvec_stat_local) { |
5162 | kfree(pn); | |
5163 | return 1; | |
5164 | } | |
5165 | ||
3e38e0aa RG |
5166 | pn->lruvec_stat_cpu = alloc_percpu_gfp(struct lruvec_stat, |
5167 | GFP_KERNEL_ACCOUNT); | |
a983b5eb | 5168 | if (!pn->lruvec_stat_cpu) { |
815744d7 | 5169 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c JW |
5170 | kfree(pn); |
5171 | return 1; | |
5172 | } | |
5173 | ||
ef8f2327 MG |
5174 | lruvec_init(&pn->lruvec); |
5175 | pn->usage_in_excess = 0; | |
5176 | pn->on_tree = false; | |
5177 | pn->memcg = memcg; | |
5178 | ||
54f72fe0 | 5179 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
5180 | return 0; |
5181 | } | |
5182 | ||
ef8f2327 | 5183 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 5184 | { |
00f3ca2c JW |
5185 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
5186 | ||
4eaf431f MH |
5187 | if (!pn) |
5188 | return; | |
5189 | ||
a983b5eb | 5190 | free_percpu(pn->lruvec_stat_cpu); |
815744d7 | 5191 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c | 5192 | kfree(pn); |
1ecaab2b KH |
5193 | } |
5194 | ||
40e952f9 | 5195 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 5196 | { |
c8b2a36f | 5197 | int node; |
59927fb9 | 5198 | |
c8b2a36f | 5199 | for_each_node(node) |
ef8f2327 | 5200 | free_mem_cgroup_per_node_info(memcg, node); |
871789d4 | 5201 | free_percpu(memcg->vmstats_percpu); |
815744d7 | 5202 | free_percpu(memcg->vmstats_local); |
8ff69e2c | 5203 | kfree(memcg); |
59927fb9 | 5204 | } |
3afe36b1 | 5205 | |
40e952f9 TE |
5206 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
5207 | { | |
5208 | memcg_wb_domain_exit(memcg); | |
7961eee3 SB |
5209 | /* |
5210 | * Flush percpu vmstats and vmevents to guarantee the value correctness | |
5211 | * on parent's and all ancestor levels. | |
5212 | */ | |
4a87e2a2 | 5213 | memcg_flush_percpu_vmstats(memcg); |
7961eee3 | 5214 | memcg_flush_percpu_vmevents(memcg); |
40e952f9 TE |
5215 | __mem_cgroup_free(memcg); |
5216 | } | |
5217 | ||
0b8f73e1 | 5218 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 5219 | { |
d142e3e6 | 5220 | struct mem_cgroup *memcg; |
b9726c26 | 5221 | unsigned int size; |
6d12e2d8 | 5222 | int node; |
97b27821 | 5223 | int __maybe_unused i; |
11d67612 | 5224 | long error = -ENOMEM; |
8cdea7c0 | 5225 | |
0b8f73e1 JW |
5226 | size = sizeof(struct mem_cgroup); |
5227 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
5228 | ||
5229 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 5230 | if (!memcg) |
11d67612 | 5231 | return ERR_PTR(error); |
0b8f73e1 | 5232 | |
73f576c0 JW |
5233 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
5234 | 1, MEM_CGROUP_ID_MAX, | |
5235 | GFP_KERNEL); | |
11d67612 YS |
5236 | if (memcg->id.id < 0) { |
5237 | error = memcg->id.id; | |
73f576c0 | 5238 | goto fail; |
11d67612 | 5239 | } |
73f576c0 | 5240 | |
3e38e0aa RG |
5241 | memcg->vmstats_local = alloc_percpu_gfp(struct memcg_vmstats_percpu, |
5242 | GFP_KERNEL_ACCOUNT); | |
815744d7 JW |
5243 | if (!memcg->vmstats_local) |
5244 | goto fail; | |
5245 | ||
3e38e0aa RG |
5246 | memcg->vmstats_percpu = alloc_percpu_gfp(struct memcg_vmstats_percpu, |
5247 | GFP_KERNEL_ACCOUNT); | |
871789d4 | 5248 | if (!memcg->vmstats_percpu) |
0b8f73e1 | 5249 | goto fail; |
78fb7466 | 5250 | |
3ed28fa1 | 5251 | for_each_node(node) |
ef8f2327 | 5252 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 5253 | goto fail; |
f64c3f54 | 5254 | |
0b8f73e1 JW |
5255 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
5256 | goto fail; | |
28dbc4b6 | 5257 | |
f7e1cb6e | 5258 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 | 5259 | INIT_LIST_HEAD(&memcg->oom_notify); |
d142e3e6 GC |
5260 | mutex_init(&memcg->thresholds_lock); |
5261 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 5262 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
5263 | INIT_LIST_HEAD(&memcg->event_list); |
5264 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 5265 | memcg->socket_pressure = jiffies; |
84c07d11 | 5266 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 5267 | memcg->kmemcg_id = -1; |
bf4f0599 | 5268 | INIT_LIST_HEAD(&memcg->objcg_list); |
900a38f0 | 5269 | #endif |
52ebea74 TH |
5270 | #ifdef CONFIG_CGROUP_WRITEBACK |
5271 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
97b27821 TH |
5272 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) |
5273 | memcg->cgwb_frn[i].done = | |
5274 | __WB_COMPLETION_INIT(&memcg_cgwb_frn_waitq); | |
87eaceb3 YS |
5275 | #endif |
5276 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5277 | spin_lock_init(&memcg->deferred_split_queue.split_queue_lock); | |
5278 | INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue); | |
5279 | memcg->deferred_split_queue.split_queue_len = 0; | |
52ebea74 | 5280 | #endif |
73f576c0 | 5281 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
5282 | return memcg; |
5283 | fail: | |
7e97de0b | 5284 | mem_cgroup_id_remove(memcg); |
40e952f9 | 5285 | __mem_cgroup_free(memcg); |
11d67612 | 5286 | return ERR_PTR(error); |
d142e3e6 GC |
5287 | } |
5288 | ||
0b8f73e1 JW |
5289 | static struct cgroup_subsys_state * __ref |
5290 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 5291 | { |
0b8f73e1 JW |
5292 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
5293 | struct mem_cgroup *memcg; | |
5294 | long error = -ENOMEM; | |
d142e3e6 | 5295 | |
3e38e0aa | 5296 | memalloc_use_memcg(parent); |
0b8f73e1 | 5297 | memcg = mem_cgroup_alloc(); |
3e38e0aa | 5298 | memalloc_unuse_memcg(); |
11d67612 YS |
5299 | if (IS_ERR(memcg)) |
5300 | return ERR_CAST(memcg); | |
d142e3e6 | 5301 | |
d1663a90 | 5302 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
0b8f73e1 | 5303 | memcg->soft_limit = PAGE_COUNTER_MAX; |
4b82ab4f | 5304 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
0b8f73e1 JW |
5305 | if (parent) { |
5306 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
5307 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
5308 | } | |
5309 | if (parent && parent->use_hierarchy) { | |
5310 | memcg->use_hierarchy = true; | |
3e32cb2e | 5311 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 5312 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e | 5313 | page_counter_init(&memcg->kmem, &parent->kmem); |
0db15298 | 5314 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 5315 | } else { |
3e32cb2e | 5316 | page_counter_init(&memcg->memory, NULL); |
37e84351 | 5317 | page_counter_init(&memcg->swap, NULL); |
3e32cb2e | 5318 | page_counter_init(&memcg->kmem, NULL); |
0db15298 | 5319 | page_counter_init(&memcg->tcpmem, NULL); |
8c7f6edb TH |
5320 | /* |
5321 | * Deeper hierachy with use_hierarchy == false doesn't make | |
5322 | * much sense so let cgroup subsystem know about this | |
5323 | * unfortunate state in our controller. | |
5324 | */ | |
d142e3e6 | 5325 | if (parent != root_mem_cgroup) |
073219e9 | 5326 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 5327 | } |
d6441637 | 5328 | |
0b8f73e1 JW |
5329 | /* The following stuff does not apply to the root */ |
5330 | if (!parent) { | |
5331 | root_mem_cgroup = memcg; | |
5332 | return &memcg->css; | |
5333 | } | |
5334 | ||
b313aeee | 5335 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
5336 | if (error) |
5337 | goto fail; | |
127424c8 | 5338 | |
f7e1cb6e | 5339 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5340 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 5341 | |
0b8f73e1 JW |
5342 | return &memcg->css; |
5343 | fail: | |
7e97de0b | 5344 | mem_cgroup_id_remove(memcg); |
0b8f73e1 | 5345 | mem_cgroup_free(memcg); |
11d67612 | 5346 | return ERR_PTR(error); |
0b8f73e1 JW |
5347 | } |
5348 | ||
73f576c0 | 5349 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 5350 | { |
58fa2a55 VD |
5351 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5352 | ||
0a4465d3 KT |
5353 | /* |
5354 | * A memcg must be visible for memcg_expand_shrinker_maps() | |
5355 | * by the time the maps are allocated. So, we allocate maps | |
5356 | * here, when for_each_mem_cgroup() can't skip it. | |
5357 | */ | |
5358 | if (memcg_alloc_shrinker_maps(memcg)) { | |
5359 | mem_cgroup_id_remove(memcg); | |
5360 | return -ENOMEM; | |
5361 | } | |
5362 | ||
73f576c0 | 5363 | /* Online state pins memcg ID, memcg ID pins CSS */ |
1c2d479a | 5364 | refcount_set(&memcg->id.ref, 1); |
73f576c0 | 5365 | css_get(css); |
2f7dd7a4 | 5366 | return 0; |
8cdea7c0 BS |
5367 | } |
5368 | ||
eb95419b | 5369 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 5370 | { |
eb95419b | 5371 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5372 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
5373 | |
5374 | /* | |
5375 | * Unregister events and notify userspace. | |
5376 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
5377 | * directory to avoid race between userspace and kernelspace. | |
5378 | */ | |
fba94807 TH |
5379 | spin_lock(&memcg->event_list_lock); |
5380 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
5381 | list_del_init(&event->list); |
5382 | schedule_work(&event->remove); | |
5383 | } | |
fba94807 | 5384 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 5385 | |
bf8d5d52 | 5386 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5387 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 5388 | |
567e9ab2 | 5389 | memcg_offline_kmem(memcg); |
52ebea74 | 5390 | wb_memcg_offline(memcg); |
73f576c0 | 5391 | |
591edfb1 RG |
5392 | drain_all_stock(memcg); |
5393 | ||
73f576c0 | 5394 | mem_cgroup_id_put(memcg); |
df878fb0 KH |
5395 | } |
5396 | ||
6df38689 VD |
5397 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
5398 | { | |
5399 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5400 | ||
5401 | invalidate_reclaim_iterators(memcg); | |
5402 | } | |
5403 | ||
eb95419b | 5404 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 5405 | { |
eb95419b | 5406 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
97b27821 | 5407 | int __maybe_unused i; |
c268e994 | 5408 | |
97b27821 TH |
5409 | #ifdef CONFIG_CGROUP_WRITEBACK |
5410 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) | |
5411 | wb_wait_for_completion(&memcg->cgwb_frn[i].done); | |
5412 | #endif | |
f7e1cb6e | 5413 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5414 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 5415 | |
0db15298 | 5416 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 5417 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 5418 | |
0b8f73e1 JW |
5419 | vmpressure_cleanup(&memcg->vmpressure); |
5420 | cancel_work_sync(&memcg->high_work); | |
5421 | mem_cgroup_remove_from_trees(memcg); | |
0a4465d3 | 5422 | memcg_free_shrinker_maps(memcg); |
d886f4e4 | 5423 | memcg_free_kmem(memcg); |
0b8f73e1 | 5424 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
5425 | } |
5426 | ||
1ced953b TH |
5427 | /** |
5428 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
5429 | * @css: the target css | |
5430 | * | |
5431 | * Reset the states of the mem_cgroup associated with @css. This is | |
5432 | * invoked when the userland requests disabling on the default hierarchy | |
5433 | * but the memcg is pinned through dependency. The memcg should stop | |
5434 | * applying policies and should revert to the vanilla state as it may be | |
5435 | * made visible again. | |
5436 | * | |
5437 | * The current implementation only resets the essential configurations. | |
5438 | * This needs to be expanded to cover all the visible parts. | |
5439 | */ | |
5440 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
5441 | { | |
5442 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5443 | ||
bbec2e15 RG |
5444 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
5445 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
bbec2e15 RG |
5446 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); |
5447 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 5448 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5449 | page_counter_set_low(&memcg->memory, 0); |
d1663a90 | 5450 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
24d404dc | 5451 | memcg->soft_limit = PAGE_COUNTER_MAX; |
4b82ab4f | 5452 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
2529bb3a | 5453 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
5454 | } |
5455 | ||
02491447 | 5456 | #ifdef CONFIG_MMU |
7dc74be0 | 5457 | /* Handlers for move charge at task migration. */ |
854ffa8d | 5458 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 5459 | { |
05b84301 | 5460 | int ret; |
9476db97 | 5461 | |
d0164adc MG |
5462 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
5463 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 5464 | if (!ret) { |
854ffa8d | 5465 | mc.precharge += count; |
854ffa8d DN |
5466 | return ret; |
5467 | } | |
9476db97 | 5468 | |
3674534b | 5469 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 5470 | while (count--) { |
3674534b | 5471 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 5472 | if (ret) |
38c5d72f | 5473 | return ret; |
854ffa8d | 5474 | mc.precharge++; |
9476db97 | 5475 | cond_resched(); |
854ffa8d | 5476 | } |
9476db97 | 5477 | return 0; |
4ffef5fe DN |
5478 | } |
5479 | ||
4ffef5fe DN |
5480 | union mc_target { |
5481 | struct page *page; | |
02491447 | 5482 | swp_entry_t ent; |
4ffef5fe DN |
5483 | }; |
5484 | ||
4ffef5fe | 5485 | enum mc_target_type { |
8d32ff84 | 5486 | MC_TARGET_NONE = 0, |
4ffef5fe | 5487 | MC_TARGET_PAGE, |
02491447 | 5488 | MC_TARGET_SWAP, |
c733a828 | 5489 | MC_TARGET_DEVICE, |
4ffef5fe DN |
5490 | }; |
5491 | ||
90254a65 DN |
5492 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5493 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5494 | { |
25b2995a | 5495 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5496 | |
90254a65 DN |
5497 | if (!page || !page_mapped(page)) |
5498 | return NULL; | |
5499 | if (PageAnon(page)) { | |
1dfab5ab | 5500 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5501 | return NULL; |
1dfab5ab JW |
5502 | } else { |
5503 | if (!(mc.flags & MOVE_FILE)) | |
5504 | return NULL; | |
5505 | } | |
90254a65 DN |
5506 | if (!get_page_unless_zero(page)) |
5507 | return NULL; | |
5508 | ||
5509 | return page; | |
5510 | } | |
5511 | ||
c733a828 | 5512 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 5513 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 5514 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 5515 | { |
90254a65 DN |
5516 | struct page *page = NULL; |
5517 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5518 | ||
1dfab5ab | 5519 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 5520 | return NULL; |
c733a828 JG |
5521 | |
5522 | /* | |
5523 | * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to | |
5524 | * a device and because they are not accessible by CPU they are store | |
5525 | * as special swap entry in the CPU page table. | |
5526 | */ | |
5527 | if (is_device_private_entry(ent)) { | |
5528 | page = device_private_entry_to_page(ent); | |
5529 | /* | |
5530 | * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have | |
5531 | * a refcount of 1 when free (unlike normal page) | |
5532 | */ | |
5533 | if (!page_ref_add_unless(page, 1, 1)) | |
5534 | return NULL; | |
5535 | return page; | |
5536 | } | |
5537 | ||
4b91355e KH |
5538 | /* |
5539 | * Because lookup_swap_cache() updates some statistics counter, | |
5540 | * we call find_get_page() with swapper_space directly. | |
5541 | */ | |
f6ab1f7f | 5542 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
2d1c4980 | 5543 | entry->val = ent.val; |
90254a65 DN |
5544 | |
5545 | return page; | |
5546 | } | |
4b91355e KH |
5547 | #else |
5548 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 5549 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
5550 | { |
5551 | return NULL; | |
5552 | } | |
5553 | #endif | |
90254a65 | 5554 | |
87946a72 DN |
5555 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5556 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5557 | { | |
87946a72 DN |
5558 | if (!vma->vm_file) /* anonymous vma */ |
5559 | return NULL; | |
1dfab5ab | 5560 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
5561 | return NULL; |
5562 | ||
87946a72 | 5563 | /* page is moved even if it's not RSS of this task(page-faulted). */ |
aa3b1895 | 5564 | /* shmem/tmpfs may report page out on swap: account for that too. */ |
f5df8635 MWO |
5565 | return find_get_incore_page(vma->vm_file->f_mapping, |
5566 | linear_page_index(vma, addr)); | |
87946a72 DN |
5567 | } |
5568 | ||
b1b0deab CG |
5569 | /** |
5570 | * mem_cgroup_move_account - move account of the page | |
5571 | * @page: the page | |
25843c2b | 5572 | * @compound: charge the page as compound or small page |
b1b0deab CG |
5573 | * @from: mem_cgroup which the page is moved from. |
5574 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
5575 | * | |
3ac808fd | 5576 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
5577 | * |
5578 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
5579 | * from old cgroup. | |
5580 | */ | |
5581 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 5582 | bool compound, |
b1b0deab CG |
5583 | struct mem_cgroup *from, |
5584 | struct mem_cgroup *to) | |
5585 | { | |
ae8af438 KK |
5586 | struct lruvec *from_vec, *to_vec; |
5587 | struct pglist_data *pgdat; | |
6c357848 | 5588 | unsigned int nr_pages = compound ? thp_nr_pages(page) : 1; |
b1b0deab CG |
5589 | int ret; |
5590 | ||
5591 | VM_BUG_ON(from == to); | |
5592 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 5593 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
5594 | |
5595 | /* | |
6a93ca8f | 5596 | * Prevent mem_cgroup_migrate() from looking at |
45637bab | 5597 | * page->mem_cgroup of its source page while we change it. |
b1b0deab | 5598 | */ |
f627c2f5 | 5599 | ret = -EBUSY; |
b1b0deab CG |
5600 | if (!trylock_page(page)) |
5601 | goto out; | |
5602 | ||
5603 | ret = -EINVAL; | |
5604 | if (page->mem_cgroup != from) | |
5605 | goto out_unlock; | |
5606 | ||
ae8af438 | 5607 | pgdat = page_pgdat(page); |
867e5e1d JW |
5608 | from_vec = mem_cgroup_lruvec(from, pgdat); |
5609 | to_vec = mem_cgroup_lruvec(to, pgdat); | |
ae8af438 | 5610 | |
abb242f5 | 5611 | lock_page_memcg(page); |
b1b0deab | 5612 | |
be5d0a74 JW |
5613 | if (PageAnon(page)) { |
5614 | if (page_mapped(page)) { | |
5615 | __mod_lruvec_state(from_vec, NR_ANON_MAPPED, -nr_pages); | |
5616 | __mod_lruvec_state(to_vec, NR_ANON_MAPPED, nr_pages); | |
468c3982 JW |
5617 | if (PageTransHuge(page)) { |
5618 | __mod_lruvec_state(from_vec, NR_ANON_THPS, | |
5619 | -nr_pages); | |
5620 | __mod_lruvec_state(to_vec, NR_ANON_THPS, | |
5621 | nr_pages); | |
5622 | } | |
5623 | ||
be5d0a74 JW |
5624 | } |
5625 | } else { | |
0d1c2072 JW |
5626 | __mod_lruvec_state(from_vec, NR_FILE_PAGES, -nr_pages); |
5627 | __mod_lruvec_state(to_vec, NR_FILE_PAGES, nr_pages); | |
5628 | ||
5629 | if (PageSwapBacked(page)) { | |
5630 | __mod_lruvec_state(from_vec, NR_SHMEM, -nr_pages); | |
5631 | __mod_lruvec_state(to_vec, NR_SHMEM, nr_pages); | |
5632 | } | |
5633 | ||
49e50d27 JW |
5634 | if (page_mapped(page)) { |
5635 | __mod_lruvec_state(from_vec, NR_FILE_MAPPED, -nr_pages); | |
5636 | __mod_lruvec_state(to_vec, NR_FILE_MAPPED, nr_pages); | |
5637 | } | |
b1b0deab | 5638 | |
49e50d27 JW |
5639 | if (PageDirty(page)) { |
5640 | struct address_space *mapping = page_mapping(page); | |
c4843a75 | 5641 | |
f56753ac | 5642 | if (mapping_can_writeback(mapping)) { |
49e50d27 JW |
5643 | __mod_lruvec_state(from_vec, NR_FILE_DIRTY, |
5644 | -nr_pages); | |
5645 | __mod_lruvec_state(to_vec, NR_FILE_DIRTY, | |
5646 | nr_pages); | |
5647 | } | |
c4843a75 GT |
5648 | } |
5649 | } | |
5650 | ||
b1b0deab | 5651 | if (PageWriteback(page)) { |
ae8af438 KK |
5652 | __mod_lruvec_state(from_vec, NR_WRITEBACK, -nr_pages); |
5653 | __mod_lruvec_state(to_vec, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
5654 | } |
5655 | ||
5656 | /* | |
abb242f5 JW |
5657 | * All state has been migrated, let's switch to the new memcg. |
5658 | * | |
b1b0deab | 5659 | * It is safe to change page->mem_cgroup here because the page |
abb242f5 JW |
5660 | * is referenced, charged, isolated, and locked: we can't race |
5661 | * with (un)charging, migration, LRU putback, or anything else | |
5662 | * that would rely on a stable page->mem_cgroup. | |
5663 | * | |
5664 | * Note that lock_page_memcg is a memcg lock, not a page lock, | |
5665 | * to save space. As soon as we switch page->mem_cgroup to a | |
5666 | * new memcg that isn't locked, the above state can change | |
5667 | * concurrently again. Make sure we're truly done with it. | |
b1b0deab | 5668 | */ |
abb242f5 | 5669 | smp_mb(); |
b1b0deab | 5670 | |
1a3e1f40 JW |
5671 | css_get(&to->css); |
5672 | css_put(&from->css); | |
5673 | ||
5674 | page->mem_cgroup = to; | |
87eaceb3 | 5675 | |
abb242f5 | 5676 | __unlock_page_memcg(from); |
b1b0deab CG |
5677 | |
5678 | ret = 0; | |
5679 | ||
5680 | local_irq_disable(); | |
3fba69a5 | 5681 | mem_cgroup_charge_statistics(to, page, nr_pages); |
b1b0deab | 5682 | memcg_check_events(to, page); |
3fba69a5 | 5683 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
b1b0deab CG |
5684 | memcg_check_events(from, page); |
5685 | local_irq_enable(); | |
5686 | out_unlock: | |
5687 | unlock_page(page); | |
5688 | out: | |
5689 | return ret; | |
5690 | } | |
5691 | ||
7cf7806c LR |
5692 | /** |
5693 | * get_mctgt_type - get target type of moving charge | |
5694 | * @vma: the vma the pte to be checked belongs | |
5695 | * @addr: the address corresponding to the pte to be checked | |
5696 | * @ptent: the pte to be checked | |
5697 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
5698 | * | |
5699 | * Returns | |
5700 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5701 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5702 | * move charge. if @target is not NULL, the page is stored in target->page | |
5703 | * with extra refcnt got(Callers should handle it). | |
5704 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
5705 | * target for charge migration. if @target is not NULL, the entry is stored | |
5706 | * in target->ent. | |
25b2995a CH |
5707 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PRIVATE |
5708 | * (so ZONE_DEVICE page and thus not on the lru). | |
df6ad698 JG |
5709 | * For now we such page is charge like a regular page would be as for all |
5710 | * intent and purposes it is just special memory taking the place of a | |
5711 | * regular page. | |
c733a828 JG |
5712 | * |
5713 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
5714 | * |
5715 | * Called with pte lock held. | |
5716 | */ | |
5717 | ||
8d32ff84 | 5718 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5719 | unsigned long addr, pte_t ptent, union mc_target *target) |
5720 | { | |
5721 | struct page *page = NULL; | |
8d32ff84 | 5722 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5723 | swp_entry_t ent = { .val = 0 }; |
5724 | ||
5725 | if (pte_present(ptent)) | |
5726 | page = mc_handle_present_pte(vma, addr, ptent); | |
5727 | else if (is_swap_pte(ptent)) | |
48406ef8 | 5728 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 5729 | else if (pte_none(ptent)) |
87946a72 | 5730 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
5731 | |
5732 | if (!page && !ent.val) | |
8d32ff84 | 5733 | return ret; |
02491447 | 5734 | if (page) { |
02491447 | 5735 | /* |
0a31bc97 | 5736 | * Do only loose check w/o serialization. |
1306a85a | 5737 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 5738 | * not under LRU exclusion. |
02491447 | 5739 | */ |
1306a85a | 5740 | if (page->mem_cgroup == mc.from) { |
02491447 | 5741 | ret = MC_TARGET_PAGE; |
25b2995a | 5742 | if (is_device_private_page(page)) |
c733a828 | 5743 | ret = MC_TARGET_DEVICE; |
02491447 DN |
5744 | if (target) |
5745 | target->page = page; | |
5746 | } | |
5747 | if (!ret || !target) | |
5748 | put_page(page); | |
5749 | } | |
3e14a57b HY |
5750 | /* |
5751 | * There is a swap entry and a page doesn't exist or isn't charged. | |
5752 | * But we cannot move a tail-page in a THP. | |
5753 | */ | |
5754 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 5755 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5756 | ret = MC_TARGET_SWAP; |
5757 | if (target) | |
5758 | target->ent = ent; | |
4ffef5fe | 5759 | } |
4ffef5fe DN |
5760 | return ret; |
5761 | } | |
5762 | ||
12724850 NH |
5763 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5764 | /* | |
d6810d73 HY |
5765 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
5766 | * not support them for now. | |
12724850 NH |
5767 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
5768 | */ | |
5769 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5770 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5771 | { | |
5772 | struct page *page = NULL; | |
12724850 NH |
5773 | enum mc_target_type ret = MC_TARGET_NONE; |
5774 | ||
84c3fc4e ZY |
5775 | if (unlikely(is_swap_pmd(pmd))) { |
5776 | VM_BUG_ON(thp_migration_supported() && | |
5777 | !is_pmd_migration_entry(pmd)); | |
5778 | return ret; | |
5779 | } | |
12724850 | 5780 | page = pmd_page(pmd); |
309381fe | 5781 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 5782 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 5783 | return ret; |
1306a85a | 5784 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
5785 | ret = MC_TARGET_PAGE; |
5786 | if (target) { | |
5787 | get_page(page); | |
5788 | target->page = page; | |
5789 | } | |
5790 | } | |
5791 | return ret; | |
5792 | } | |
5793 | #else | |
5794 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5795 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5796 | { | |
5797 | return MC_TARGET_NONE; | |
5798 | } | |
5799 | #endif | |
5800 | ||
4ffef5fe DN |
5801 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5802 | unsigned long addr, unsigned long end, | |
5803 | struct mm_walk *walk) | |
5804 | { | |
26bcd64a | 5805 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5806 | pte_t *pte; |
5807 | spinlock_t *ptl; | |
5808 | ||
b6ec57f4 KS |
5809 | ptl = pmd_trans_huge_lock(pmd, vma); |
5810 | if (ptl) { | |
c733a828 JG |
5811 | /* |
5812 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
25b2995a CH |
5813 | * support transparent huge page with MEMORY_DEVICE_PRIVATE but |
5814 | * this might change. | |
c733a828 | 5815 | */ |
12724850 NH |
5816 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5817 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5818 | spin_unlock(ptl); |
1a5a9906 | 5819 | return 0; |
12724850 | 5820 | } |
03319327 | 5821 | |
45f83cef AA |
5822 | if (pmd_trans_unstable(pmd)) |
5823 | return 0; | |
4ffef5fe DN |
5824 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5825 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5826 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5827 | mc.precharge++; /* increment precharge temporarily */ |
5828 | pte_unmap_unlock(pte - 1, ptl); | |
5829 | cond_resched(); | |
5830 | ||
7dc74be0 DN |
5831 | return 0; |
5832 | } | |
5833 | ||
7b86ac33 CH |
5834 | static const struct mm_walk_ops precharge_walk_ops = { |
5835 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5836 | }; | |
5837 | ||
4ffef5fe DN |
5838 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5839 | { | |
5840 | unsigned long precharge; | |
4ffef5fe | 5841 | |
d8ed45c5 | 5842 | mmap_read_lock(mm); |
7b86ac33 | 5843 | walk_page_range(mm, 0, mm->highest_vm_end, &precharge_walk_ops, NULL); |
d8ed45c5 | 5844 | mmap_read_unlock(mm); |
4ffef5fe DN |
5845 | |
5846 | precharge = mc.precharge; | |
5847 | mc.precharge = 0; | |
5848 | ||
5849 | return precharge; | |
5850 | } | |
5851 | ||
4ffef5fe DN |
5852 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5853 | { | |
dfe076b0 DN |
5854 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5855 | ||
5856 | VM_BUG_ON(mc.moving_task); | |
5857 | mc.moving_task = current; | |
5858 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5859 | } |
5860 | ||
dfe076b0 DN |
5861 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5862 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5863 | { |
2bd9bb20 KH |
5864 | struct mem_cgroup *from = mc.from; |
5865 | struct mem_cgroup *to = mc.to; | |
5866 | ||
4ffef5fe | 5867 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5868 | if (mc.precharge) { |
00501b53 | 5869 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5870 | mc.precharge = 0; |
5871 | } | |
5872 | /* | |
5873 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5874 | * we must uncharge here. | |
5875 | */ | |
5876 | if (mc.moved_charge) { | |
00501b53 | 5877 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5878 | mc.moved_charge = 0; |
4ffef5fe | 5879 | } |
483c30b5 DN |
5880 | /* we must fixup refcnts and charges */ |
5881 | if (mc.moved_swap) { | |
483c30b5 | 5882 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5883 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5884 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5885 | |
615d66c3 VD |
5886 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
5887 | ||
05b84301 | 5888 | /* |
3e32cb2e JW |
5889 | * we charged both to->memory and to->memsw, so we |
5890 | * should uncharge to->memory. | |
05b84301 | 5891 | */ |
ce00a967 | 5892 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5893 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5894 | ||
483c30b5 DN |
5895 | mc.moved_swap = 0; |
5896 | } | |
dfe076b0 DN |
5897 | memcg_oom_recover(from); |
5898 | memcg_oom_recover(to); | |
5899 | wake_up_all(&mc.waitq); | |
5900 | } | |
5901 | ||
5902 | static void mem_cgroup_clear_mc(void) | |
5903 | { | |
264a0ae1 TH |
5904 | struct mm_struct *mm = mc.mm; |
5905 | ||
dfe076b0 DN |
5906 | /* |
5907 | * we must clear moving_task before waking up waiters at the end of | |
5908 | * task migration. | |
5909 | */ | |
5910 | mc.moving_task = NULL; | |
5911 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5912 | spin_lock(&mc.lock); |
4ffef5fe DN |
5913 | mc.from = NULL; |
5914 | mc.to = NULL; | |
264a0ae1 | 5915 | mc.mm = NULL; |
2bd9bb20 | 5916 | spin_unlock(&mc.lock); |
264a0ae1 TH |
5917 | |
5918 | mmput(mm); | |
4ffef5fe DN |
5919 | } |
5920 | ||
1f7dd3e5 | 5921 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5922 | { |
1f7dd3e5 | 5923 | struct cgroup_subsys_state *css; |
eed67d75 | 5924 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 5925 | struct mem_cgroup *from; |
4530eddb | 5926 | struct task_struct *leader, *p; |
9f2115f9 | 5927 | struct mm_struct *mm; |
1dfab5ab | 5928 | unsigned long move_flags; |
9f2115f9 | 5929 | int ret = 0; |
7dc74be0 | 5930 | |
1f7dd3e5 TH |
5931 | /* charge immigration isn't supported on the default hierarchy */ |
5932 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
5933 | return 0; |
5934 | ||
4530eddb TH |
5935 | /* |
5936 | * Multi-process migrations only happen on the default hierarchy | |
5937 | * where charge immigration is not used. Perform charge | |
5938 | * immigration if @tset contains a leader and whine if there are | |
5939 | * multiple. | |
5940 | */ | |
5941 | p = NULL; | |
1f7dd3e5 | 5942 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
5943 | WARN_ON_ONCE(p); |
5944 | p = leader; | |
1f7dd3e5 | 5945 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
5946 | } |
5947 | if (!p) | |
5948 | return 0; | |
5949 | ||
1f7dd3e5 TH |
5950 | /* |
5951 | * We are now commited to this value whatever it is. Changes in this | |
5952 | * tunable will only affect upcoming migrations, not the current one. | |
5953 | * So we need to save it, and keep it going. | |
5954 | */ | |
5955 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
5956 | if (!move_flags) | |
5957 | return 0; | |
5958 | ||
9f2115f9 TH |
5959 | from = mem_cgroup_from_task(p); |
5960 | ||
5961 | VM_BUG_ON(from == memcg); | |
5962 | ||
5963 | mm = get_task_mm(p); | |
5964 | if (!mm) | |
5965 | return 0; | |
5966 | /* We move charges only when we move a owner of the mm */ | |
5967 | if (mm->owner == p) { | |
5968 | VM_BUG_ON(mc.from); | |
5969 | VM_BUG_ON(mc.to); | |
5970 | VM_BUG_ON(mc.precharge); | |
5971 | VM_BUG_ON(mc.moved_charge); | |
5972 | VM_BUG_ON(mc.moved_swap); | |
5973 | ||
5974 | spin_lock(&mc.lock); | |
264a0ae1 | 5975 | mc.mm = mm; |
9f2115f9 TH |
5976 | mc.from = from; |
5977 | mc.to = memcg; | |
5978 | mc.flags = move_flags; | |
5979 | spin_unlock(&mc.lock); | |
5980 | /* We set mc.moving_task later */ | |
5981 | ||
5982 | ret = mem_cgroup_precharge_mc(mm); | |
5983 | if (ret) | |
5984 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
5985 | } else { |
5986 | mmput(mm); | |
7dc74be0 DN |
5987 | } |
5988 | return ret; | |
5989 | } | |
5990 | ||
1f7dd3e5 | 5991 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5992 | { |
4e2f245d JW |
5993 | if (mc.to) |
5994 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5995 | } |
5996 | ||
4ffef5fe DN |
5997 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5998 | unsigned long addr, unsigned long end, | |
5999 | struct mm_walk *walk) | |
7dc74be0 | 6000 | { |
4ffef5fe | 6001 | int ret = 0; |
26bcd64a | 6002 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
6003 | pte_t *pte; |
6004 | spinlock_t *ptl; | |
12724850 NH |
6005 | enum mc_target_type target_type; |
6006 | union mc_target target; | |
6007 | struct page *page; | |
4ffef5fe | 6008 | |
b6ec57f4 KS |
6009 | ptl = pmd_trans_huge_lock(pmd, vma); |
6010 | if (ptl) { | |
62ade86a | 6011 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 6012 | spin_unlock(ptl); |
12724850 NH |
6013 | return 0; |
6014 | } | |
6015 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
6016 | if (target_type == MC_TARGET_PAGE) { | |
6017 | page = target.page; | |
6018 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 6019 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 6020 | mc.from, mc.to)) { |
12724850 NH |
6021 | mc.precharge -= HPAGE_PMD_NR; |
6022 | mc.moved_charge += HPAGE_PMD_NR; | |
6023 | } | |
6024 | putback_lru_page(page); | |
6025 | } | |
6026 | put_page(page); | |
c733a828 JG |
6027 | } else if (target_type == MC_TARGET_DEVICE) { |
6028 | page = target.page; | |
6029 | if (!mem_cgroup_move_account(page, true, | |
6030 | mc.from, mc.to)) { | |
6031 | mc.precharge -= HPAGE_PMD_NR; | |
6032 | mc.moved_charge += HPAGE_PMD_NR; | |
6033 | } | |
6034 | put_page(page); | |
12724850 | 6035 | } |
bf929152 | 6036 | spin_unlock(ptl); |
1a5a9906 | 6037 | return 0; |
12724850 NH |
6038 | } |
6039 | ||
45f83cef AA |
6040 | if (pmd_trans_unstable(pmd)) |
6041 | return 0; | |
4ffef5fe DN |
6042 | retry: |
6043 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
6044 | for (; addr != end; addr += PAGE_SIZE) { | |
6045 | pte_t ptent = *(pte++); | |
c733a828 | 6046 | bool device = false; |
02491447 | 6047 | swp_entry_t ent; |
4ffef5fe DN |
6048 | |
6049 | if (!mc.precharge) | |
6050 | break; | |
6051 | ||
8d32ff84 | 6052 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
6053 | case MC_TARGET_DEVICE: |
6054 | device = true; | |
e4a9bc58 | 6055 | fallthrough; |
4ffef5fe DN |
6056 | case MC_TARGET_PAGE: |
6057 | page = target.page; | |
53f9263b KS |
6058 | /* |
6059 | * We can have a part of the split pmd here. Moving it | |
6060 | * can be done but it would be too convoluted so simply | |
6061 | * ignore such a partial THP and keep it in original | |
6062 | * memcg. There should be somebody mapping the head. | |
6063 | */ | |
6064 | if (PageTransCompound(page)) | |
6065 | goto put; | |
c733a828 | 6066 | if (!device && isolate_lru_page(page)) |
4ffef5fe | 6067 | goto put; |
f627c2f5 KS |
6068 | if (!mem_cgroup_move_account(page, false, |
6069 | mc.from, mc.to)) { | |
4ffef5fe | 6070 | mc.precharge--; |
854ffa8d DN |
6071 | /* we uncharge from mc.from later. */ |
6072 | mc.moved_charge++; | |
4ffef5fe | 6073 | } |
c733a828 JG |
6074 | if (!device) |
6075 | putback_lru_page(page); | |
8d32ff84 | 6076 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
6077 | put_page(page); |
6078 | break; | |
02491447 DN |
6079 | case MC_TARGET_SWAP: |
6080 | ent = target.ent; | |
e91cbb42 | 6081 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6082 | mc.precharge--; |
8d22a935 HD |
6083 | mem_cgroup_id_get_many(mc.to, 1); |
6084 | /* we fixup other refcnts and charges later. */ | |
483c30b5 DN |
6085 | mc.moved_swap++; |
6086 | } | |
02491447 | 6087 | break; |
4ffef5fe DN |
6088 | default: |
6089 | break; | |
6090 | } | |
6091 | } | |
6092 | pte_unmap_unlock(pte - 1, ptl); | |
6093 | cond_resched(); | |
6094 | ||
6095 | if (addr != end) { | |
6096 | /* | |
6097 | * We have consumed all precharges we got in can_attach(). | |
6098 | * We try charge one by one, but don't do any additional | |
6099 | * charges to mc.to if we have failed in charge once in attach() | |
6100 | * phase. | |
6101 | */ | |
854ffa8d | 6102 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6103 | if (!ret) |
6104 | goto retry; | |
6105 | } | |
6106 | ||
6107 | return ret; | |
6108 | } | |
6109 | ||
7b86ac33 CH |
6110 | static const struct mm_walk_ops charge_walk_ops = { |
6111 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6112 | }; | |
6113 | ||
264a0ae1 | 6114 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 6115 | { |
4ffef5fe | 6116 | lru_add_drain_all(); |
312722cb | 6117 | /* |
81f8c3a4 JW |
6118 | * Signal lock_page_memcg() to take the memcg's move_lock |
6119 | * while we're moving its pages to another memcg. Then wait | |
6120 | * for already started RCU-only updates to finish. | |
312722cb JW |
6121 | */ |
6122 | atomic_inc(&mc.from->moving_account); | |
6123 | synchronize_rcu(); | |
dfe076b0 | 6124 | retry: |
d8ed45c5 | 6125 | if (unlikely(!mmap_read_trylock(mc.mm))) { |
dfe076b0 | 6126 | /* |
c1e8d7c6 | 6127 | * Someone who are holding the mmap_lock might be waiting in |
dfe076b0 DN |
6128 | * waitq. So we cancel all extra charges, wake up all waiters, |
6129 | * and retry. Because we cancel precharges, we might not be able | |
6130 | * to move enough charges, but moving charge is a best-effort | |
6131 | * feature anyway, so it wouldn't be a big problem. | |
6132 | */ | |
6133 | __mem_cgroup_clear_mc(); | |
6134 | cond_resched(); | |
6135 | goto retry; | |
6136 | } | |
26bcd64a NH |
6137 | /* |
6138 | * When we have consumed all precharges and failed in doing | |
6139 | * additional charge, the page walk just aborts. | |
6140 | */ | |
7b86ac33 CH |
6141 | walk_page_range(mc.mm, 0, mc.mm->highest_vm_end, &charge_walk_ops, |
6142 | NULL); | |
0247f3f4 | 6143 | |
d8ed45c5 | 6144 | mmap_read_unlock(mc.mm); |
312722cb | 6145 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
6146 | } |
6147 | ||
264a0ae1 | 6148 | static void mem_cgroup_move_task(void) |
67e465a7 | 6149 | { |
264a0ae1 TH |
6150 | if (mc.to) { |
6151 | mem_cgroup_move_charge(); | |
a433658c | 6152 | mem_cgroup_clear_mc(); |
264a0ae1 | 6153 | } |
67e465a7 | 6154 | } |
5cfb80a7 | 6155 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 6156 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6157 | { |
6158 | return 0; | |
6159 | } | |
1f7dd3e5 | 6160 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6161 | { |
6162 | } | |
264a0ae1 | 6163 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
6164 | { |
6165 | } | |
6166 | #endif | |
67e465a7 | 6167 | |
f00baae7 TH |
6168 | /* |
6169 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
6170 | * to verify whether we're attached to the default hierarchy on each mount |
6171 | * attempt. | |
f00baae7 | 6172 | */ |
eb95419b | 6173 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
6174 | { |
6175 | /* | |
aa6ec29b | 6176 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
6177 | * guarantees that @root doesn't have any children, so turning it |
6178 | * on for the root memcg is enough. | |
6179 | */ | |
9e10a130 | 6180 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
6181 | root_mem_cgroup->use_hierarchy = true; |
6182 | else | |
6183 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
6184 | } |
6185 | ||
677dc973 CD |
6186 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
6187 | { | |
6188 | if (value == PAGE_COUNTER_MAX) | |
6189 | seq_puts(m, "max\n"); | |
6190 | else | |
6191 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); | |
6192 | ||
6193 | return 0; | |
6194 | } | |
6195 | ||
241994ed JW |
6196 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
6197 | struct cftype *cft) | |
6198 | { | |
f5fc3c5d JW |
6199 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
6200 | ||
6201 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
6202 | } |
6203 | ||
bf8d5d52 RG |
6204 | static int memory_min_show(struct seq_file *m, void *v) |
6205 | { | |
677dc973 CD |
6206 | return seq_puts_memcg_tunable(m, |
6207 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); | |
bf8d5d52 RG |
6208 | } |
6209 | ||
6210 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
6211 | char *buf, size_t nbytes, loff_t off) | |
6212 | { | |
6213 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6214 | unsigned long min; | |
6215 | int err; | |
6216 | ||
6217 | buf = strstrip(buf); | |
6218 | err = page_counter_memparse(buf, "max", &min); | |
6219 | if (err) | |
6220 | return err; | |
6221 | ||
6222 | page_counter_set_min(&memcg->memory, min); | |
6223 | ||
6224 | return nbytes; | |
6225 | } | |
6226 | ||
241994ed JW |
6227 | static int memory_low_show(struct seq_file *m, void *v) |
6228 | { | |
677dc973 CD |
6229 | return seq_puts_memcg_tunable(m, |
6230 | READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); | |
241994ed JW |
6231 | } |
6232 | ||
6233 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
6234 | char *buf, size_t nbytes, loff_t off) | |
6235 | { | |
6236 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6237 | unsigned long low; | |
6238 | int err; | |
6239 | ||
6240 | buf = strstrip(buf); | |
d2973697 | 6241 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
6242 | if (err) |
6243 | return err; | |
6244 | ||
23067153 | 6245 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
6246 | |
6247 | return nbytes; | |
6248 | } | |
6249 | ||
6250 | static int memory_high_show(struct seq_file *m, void *v) | |
6251 | { | |
d1663a90 JK |
6252 | return seq_puts_memcg_tunable(m, |
6253 | READ_ONCE(mem_cgroup_from_seq(m)->memory.high)); | |
241994ed JW |
6254 | } |
6255 | ||
6256 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
6257 | char *buf, size_t nbytes, loff_t off) | |
6258 | { | |
6259 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6260 | unsigned int nr_retries = MAX_RECLAIM_RETRIES; |
8c8c383c | 6261 | bool drained = false; |
241994ed JW |
6262 | unsigned long high; |
6263 | int err; | |
6264 | ||
6265 | buf = strstrip(buf); | |
d2973697 | 6266 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
6267 | if (err) |
6268 | return err; | |
6269 | ||
8c8c383c JW |
6270 | for (;;) { |
6271 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6272 | unsigned long reclaimed; | |
6273 | ||
6274 | if (nr_pages <= high) | |
6275 | break; | |
6276 | ||
6277 | if (signal_pending(current)) | |
6278 | break; | |
6279 | ||
6280 | if (!drained) { | |
6281 | drain_all_stock(memcg); | |
6282 | drained = true; | |
6283 | continue; | |
6284 | } | |
6285 | ||
6286 | reclaimed = try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
6287 | GFP_KERNEL, true); | |
6288 | ||
6289 | if (!reclaimed && !nr_retries--) | |
6290 | break; | |
6291 | } | |
588083bb | 6292 | |
536d3bf2 RG |
6293 | page_counter_set_high(&memcg->memory, high); |
6294 | ||
19ce33ac JW |
6295 | memcg_wb_domain_size_changed(memcg); |
6296 | ||
241994ed JW |
6297 | return nbytes; |
6298 | } | |
6299 | ||
6300 | static int memory_max_show(struct seq_file *m, void *v) | |
6301 | { | |
677dc973 CD |
6302 | return seq_puts_memcg_tunable(m, |
6303 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); | |
241994ed JW |
6304 | } |
6305 | ||
6306 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
6307 | char *buf, size_t nbytes, loff_t off) | |
6308 | { | |
6309 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6310 | unsigned int nr_reclaims = MAX_RECLAIM_RETRIES; |
b6e6edcf | 6311 | bool drained = false; |
241994ed JW |
6312 | unsigned long max; |
6313 | int err; | |
6314 | ||
6315 | buf = strstrip(buf); | |
d2973697 | 6316 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
6317 | if (err) |
6318 | return err; | |
6319 | ||
bbec2e15 | 6320 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
6321 | |
6322 | for (;;) { | |
6323 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6324 | ||
6325 | if (nr_pages <= max) | |
6326 | break; | |
6327 | ||
7249c9f0 | 6328 | if (signal_pending(current)) |
b6e6edcf | 6329 | break; |
b6e6edcf JW |
6330 | |
6331 | if (!drained) { | |
6332 | drain_all_stock(memcg); | |
6333 | drained = true; | |
6334 | continue; | |
6335 | } | |
6336 | ||
6337 | if (nr_reclaims) { | |
6338 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
6339 | GFP_KERNEL, true)) | |
6340 | nr_reclaims--; | |
6341 | continue; | |
6342 | } | |
6343 | ||
e27be240 | 6344 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
6345 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
6346 | break; | |
6347 | } | |
241994ed | 6348 | |
2529bb3a | 6349 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
6350 | return nbytes; |
6351 | } | |
6352 | ||
1e577f97 SB |
6353 | static void __memory_events_show(struct seq_file *m, atomic_long_t *events) |
6354 | { | |
6355 | seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW])); | |
6356 | seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH])); | |
6357 | seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX])); | |
6358 | seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM])); | |
6359 | seq_printf(m, "oom_kill %lu\n", | |
6360 | atomic_long_read(&events[MEMCG_OOM_KILL])); | |
6361 | } | |
6362 | ||
241994ed JW |
6363 | static int memory_events_show(struct seq_file *m, void *v) |
6364 | { | |
aa9694bb | 6365 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
241994ed | 6366 | |
1e577f97 SB |
6367 | __memory_events_show(m, memcg->memory_events); |
6368 | return 0; | |
6369 | } | |
6370 | ||
6371 | static int memory_events_local_show(struct seq_file *m, void *v) | |
6372 | { | |
6373 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
241994ed | 6374 | |
1e577f97 | 6375 | __memory_events_show(m, memcg->memory_events_local); |
241994ed JW |
6376 | return 0; |
6377 | } | |
6378 | ||
587d9f72 JW |
6379 | static int memory_stat_show(struct seq_file *m, void *v) |
6380 | { | |
aa9694bb | 6381 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
c8713d0b | 6382 | char *buf; |
1ff9e6e1 | 6383 | |
c8713d0b JW |
6384 | buf = memory_stat_format(memcg); |
6385 | if (!buf) | |
6386 | return -ENOMEM; | |
6387 | seq_puts(m, buf); | |
6388 | kfree(buf); | |
587d9f72 JW |
6389 | return 0; |
6390 | } | |
6391 | ||
5f9a4f4a MS |
6392 | #ifdef CONFIG_NUMA |
6393 | static int memory_numa_stat_show(struct seq_file *m, void *v) | |
6394 | { | |
6395 | int i; | |
6396 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
6397 | ||
6398 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { | |
6399 | int nid; | |
6400 | ||
6401 | if (memory_stats[i].idx >= NR_VM_NODE_STAT_ITEMS) | |
6402 | continue; | |
6403 | ||
6404 | seq_printf(m, "%s", memory_stats[i].name); | |
6405 | for_each_node_state(nid, N_MEMORY) { | |
6406 | u64 size; | |
6407 | struct lruvec *lruvec; | |
6408 | ||
6409 | lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); | |
6410 | size = lruvec_page_state(lruvec, memory_stats[i].idx); | |
6411 | size *= memory_stats[i].ratio; | |
6412 | seq_printf(m, " N%d=%llu", nid, size); | |
6413 | } | |
6414 | seq_putc(m, '\n'); | |
6415 | } | |
6416 | ||
6417 | return 0; | |
6418 | } | |
6419 | #endif | |
6420 | ||
3d8b38eb RG |
6421 | static int memory_oom_group_show(struct seq_file *m, void *v) |
6422 | { | |
aa9694bb | 6423 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3d8b38eb RG |
6424 | |
6425 | seq_printf(m, "%d\n", memcg->oom_group); | |
6426 | ||
6427 | return 0; | |
6428 | } | |
6429 | ||
6430 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
6431 | char *buf, size_t nbytes, loff_t off) | |
6432 | { | |
6433 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6434 | int ret, oom_group; | |
6435 | ||
6436 | buf = strstrip(buf); | |
6437 | if (!buf) | |
6438 | return -EINVAL; | |
6439 | ||
6440 | ret = kstrtoint(buf, 0, &oom_group); | |
6441 | if (ret) | |
6442 | return ret; | |
6443 | ||
6444 | if (oom_group != 0 && oom_group != 1) | |
6445 | return -EINVAL; | |
6446 | ||
6447 | memcg->oom_group = oom_group; | |
6448 | ||
6449 | return nbytes; | |
6450 | } | |
6451 | ||
241994ed JW |
6452 | static struct cftype memory_files[] = { |
6453 | { | |
6454 | .name = "current", | |
f5fc3c5d | 6455 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
6456 | .read_u64 = memory_current_read, |
6457 | }, | |
bf8d5d52 RG |
6458 | { |
6459 | .name = "min", | |
6460 | .flags = CFTYPE_NOT_ON_ROOT, | |
6461 | .seq_show = memory_min_show, | |
6462 | .write = memory_min_write, | |
6463 | }, | |
241994ed JW |
6464 | { |
6465 | .name = "low", | |
6466 | .flags = CFTYPE_NOT_ON_ROOT, | |
6467 | .seq_show = memory_low_show, | |
6468 | .write = memory_low_write, | |
6469 | }, | |
6470 | { | |
6471 | .name = "high", | |
6472 | .flags = CFTYPE_NOT_ON_ROOT, | |
6473 | .seq_show = memory_high_show, | |
6474 | .write = memory_high_write, | |
6475 | }, | |
6476 | { | |
6477 | .name = "max", | |
6478 | .flags = CFTYPE_NOT_ON_ROOT, | |
6479 | .seq_show = memory_max_show, | |
6480 | .write = memory_max_write, | |
6481 | }, | |
6482 | { | |
6483 | .name = "events", | |
6484 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 6485 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
6486 | .seq_show = memory_events_show, |
6487 | }, | |
1e577f97 SB |
6488 | { |
6489 | .name = "events.local", | |
6490 | .flags = CFTYPE_NOT_ON_ROOT, | |
6491 | .file_offset = offsetof(struct mem_cgroup, events_local_file), | |
6492 | .seq_show = memory_events_local_show, | |
6493 | }, | |
587d9f72 JW |
6494 | { |
6495 | .name = "stat", | |
587d9f72 JW |
6496 | .seq_show = memory_stat_show, |
6497 | }, | |
5f9a4f4a MS |
6498 | #ifdef CONFIG_NUMA |
6499 | { | |
6500 | .name = "numa_stat", | |
6501 | .seq_show = memory_numa_stat_show, | |
6502 | }, | |
6503 | #endif | |
3d8b38eb RG |
6504 | { |
6505 | .name = "oom.group", | |
6506 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
6507 | .seq_show = memory_oom_group_show, | |
6508 | .write = memory_oom_group_write, | |
6509 | }, | |
241994ed JW |
6510 | { } /* terminate */ |
6511 | }; | |
6512 | ||
073219e9 | 6513 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 6514 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6515 | .css_online = mem_cgroup_css_online, |
92fb9748 | 6516 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 6517 | .css_released = mem_cgroup_css_released, |
92fb9748 | 6518 | .css_free = mem_cgroup_css_free, |
1ced953b | 6519 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
6520 | .can_attach = mem_cgroup_can_attach, |
6521 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 6522 | .post_attach = mem_cgroup_move_task, |
f00baae7 | 6523 | .bind = mem_cgroup_bind, |
241994ed JW |
6524 | .dfl_cftypes = memory_files, |
6525 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 6526 | .early_init = 0, |
8cdea7c0 | 6527 | }; |
c077719b | 6528 | |
bc50bcc6 JW |
6529 | /* |
6530 | * This function calculates an individual cgroup's effective | |
6531 | * protection which is derived from its own memory.min/low, its | |
6532 | * parent's and siblings' settings, as well as the actual memory | |
6533 | * distribution in the tree. | |
6534 | * | |
6535 | * The following rules apply to the effective protection values: | |
6536 | * | |
6537 | * 1. At the first level of reclaim, effective protection is equal to | |
6538 | * the declared protection in memory.min and memory.low. | |
6539 | * | |
6540 | * 2. To enable safe delegation of the protection configuration, at | |
6541 | * subsequent levels the effective protection is capped to the | |
6542 | * parent's effective protection. | |
6543 | * | |
6544 | * 3. To make complex and dynamic subtrees easier to configure, the | |
6545 | * user is allowed to overcommit the declared protection at a given | |
6546 | * level. If that is the case, the parent's effective protection is | |
6547 | * distributed to the children in proportion to how much protection | |
6548 | * they have declared and how much of it they are utilizing. | |
6549 | * | |
6550 | * This makes distribution proportional, but also work-conserving: | |
6551 | * if one cgroup claims much more protection than it uses memory, | |
6552 | * the unused remainder is available to its siblings. | |
6553 | * | |
6554 | * 4. Conversely, when the declared protection is undercommitted at a | |
6555 | * given level, the distribution of the larger parental protection | |
6556 | * budget is NOT proportional. A cgroup's protection from a sibling | |
6557 | * is capped to its own memory.min/low setting. | |
6558 | * | |
8a931f80 JW |
6559 | * 5. However, to allow protecting recursive subtrees from each other |
6560 | * without having to declare each individual cgroup's fixed share | |
6561 | * of the ancestor's claim to protection, any unutilized - | |
6562 | * "floating" - protection from up the tree is distributed in | |
6563 | * proportion to each cgroup's *usage*. This makes the protection | |
6564 | * neutral wrt sibling cgroups and lets them compete freely over | |
6565 | * the shared parental protection budget, but it protects the | |
6566 | * subtree as a whole from neighboring subtrees. | |
6567 | * | |
6568 | * Note that 4. and 5. are not in conflict: 4. is about protecting | |
6569 | * against immediate siblings whereas 5. is about protecting against | |
6570 | * neighboring subtrees. | |
bc50bcc6 JW |
6571 | */ |
6572 | static unsigned long effective_protection(unsigned long usage, | |
8a931f80 | 6573 | unsigned long parent_usage, |
bc50bcc6 JW |
6574 | unsigned long setting, |
6575 | unsigned long parent_effective, | |
6576 | unsigned long siblings_protected) | |
6577 | { | |
6578 | unsigned long protected; | |
8a931f80 | 6579 | unsigned long ep; |
bc50bcc6 JW |
6580 | |
6581 | protected = min(usage, setting); | |
6582 | /* | |
6583 | * If all cgroups at this level combined claim and use more | |
6584 | * protection then what the parent affords them, distribute | |
6585 | * shares in proportion to utilization. | |
6586 | * | |
6587 | * We are using actual utilization rather than the statically | |
6588 | * claimed protection in order to be work-conserving: claimed | |
6589 | * but unused protection is available to siblings that would | |
6590 | * otherwise get a smaller chunk than what they claimed. | |
6591 | */ | |
6592 | if (siblings_protected > parent_effective) | |
6593 | return protected * parent_effective / siblings_protected; | |
6594 | ||
6595 | /* | |
6596 | * Ok, utilized protection of all children is within what the | |
6597 | * parent affords them, so we know whatever this child claims | |
6598 | * and utilizes is effectively protected. | |
6599 | * | |
6600 | * If there is unprotected usage beyond this value, reclaim | |
6601 | * will apply pressure in proportion to that amount. | |
6602 | * | |
6603 | * If there is unutilized protection, the cgroup will be fully | |
6604 | * shielded from reclaim, but we do return a smaller value for | |
6605 | * protection than what the group could enjoy in theory. This | |
6606 | * is okay. With the overcommit distribution above, effective | |
6607 | * protection is always dependent on how memory is actually | |
6608 | * consumed among the siblings anyway. | |
6609 | */ | |
8a931f80 JW |
6610 | ep = protected; |
6611 | ||
6612 | /* | |
6613 | * If the children aren't claiming (all of) the protection | |
6614 | * afforded to them by the parent, distribute the remainder in | |
6615 | * proportion to the (unprotected) memory of each cgroup. That | |
6616 | * way, cgroups that aren't explicitly prioritized wrt each | |
6617 | * other compete freely over the allowance, but they are | |
6618 | * collectively protected from neighboring trees. | |
6619 | * | |
6620 | * We're using unprotected memory for the weight so that if | |
6621 | * some cgroups DO claim explicit protection, we don't protect | |
6622 | * the same bytes twice. | |
cd324edc JW |
6623 | * |
6624 | * Check both usage and parent_usage against the respective | |
6625 | * protected values. One should imply the other, but they | |
6626 | * aren't read atomically - make sure the division is sane. | |
8a931f80 JW |
6627 | */ |
6628 | if (!(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)) | |
6629 | return ep; | |
cd324edc JW |
6630 | if (parent_effective > siblings_protected && |
6631 | parent_usage > siblings_protected && | |
6632 | usage > protected) { | |
8a931f80 JW |
6633 | unsigned long unclaimed; |
6634 | ||
6635 | unclaimed = parent_effective - siblings_protected; | |
6636 | unclaimed *= usage - protected; | |
6637 | unclaimed /= parent_usage - siblings_protected; | |
6638 | ||
6639 | ep += unclaimed; | |
6640 | } | |
6641 | ||
6642 | return ep; | |
bc50bcc6 JW |
6643 | } |
6644 | ||
241994ed | 6645 | /** |
bf8d5d52 | 6646 | * mem_cgroup_protected - check if memory consumption is in the normal range |
34c81057 | 6647 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
6648 | * @memcg: the memory cgroup to check |
6649 | * | |
23067153 RG |
6650 | * WARNING: This function is not stateless! It can only be used as part |
6651 | * of a top-down tree iteration, not for isolated queries. | |
241994ed | 6652 | */ |
45c7f7e1 CD |
6653 | void mem_cgroup_calculate_protection(struct mem_cgroup *root, |
6654 | struct mem_cgroup *memcg) | |
241994ed | 6655 | { |
8a931f80 | 6656 | unsigned long usage, parent_usage; |
23067153 RG |
6657 | struct mem_cgroup *parent; |
6658 | ||
241994ed | 6659 | if (mem_cgroup_disabled()) |
45c7f7e1 | 6660 | return; |
241994ed | 6661 | |
34c81057 SC |
6662 | if (!root) |
6663 | root = root_mem_cgroup; | |
22f7496f YS |
6664 | |
6665 | /* | |
6666 | * Effective values of the reclaim targets are ignored so they | |
6667 | * can be stale. Have a look at mem_cgroup_protection for more | |
6668 | * details. | |
6669 | * TODO: calculation should be more robust so that we do not need | |
6670 | * that special casing. | |
6671 | */ | |
34c81057 | 6672 | if (memcg == root) |
45c7f7e1 | 6673 | return; |
241994ed | 6674 | |
23067153 | 6675 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 | 6676 | if (!usage) |
45c7f7e1 | 6677 | return; |
bf8d5d52 | 6678 | |
bf8d5d52 | 6679 | parent = parent_mem_cgroup(memcg); |
df2a4196 RG |
6680 | /* No parent means a non-hierarchical mode on v1 memcg */ |
6681 | if (!parent) | |
45c7f7e1 | 6682 | return; |
df2a4196 | 6683 | |
bc50bcc6 | 6684 | if (parent == root) { |
c3d53200 | 6685 | memcg->memory.emin = READ_ONCE(memcg->memory.min); |
03960e33 | 6686 | memcg->memory.elow = READ_ONCE(memcg->memory.low); |
45c7f7e1 | 6687 | return; |
bf8d5d52 RG |
6688 | } |
6689 | ||
8a931f80 JW |
6690 | parent_usage = page_counter_read(&parent->memory); |
6691 | ||
b3a7822e | 6692 | WRITE_ONCE(memcg->memory.emin, effective_protection(usage, parent_usage, |
c3d53200 CD |
6693 | READ_ONCE(memcg->memory.min), |
6694 | READ_ONCE(parent->memory.emin), | |
b3a7822e | 6695 | atomic_long_read(&parent->memory.children_min_usage))); |
23067153 | 6696 | |
b3a7822e | 6697 | WRITE_ONCE(memcg->memory.elow, effective_protection(usage, parent_usage, |
03960e33 CD |
6698 | READ_ONCE(memcg->memory.low), |
6699 | READ_ONCE(parent->memory.elow), | |
b3a7822e | 6700 | atomic_long_read(&parent->memory.children_low_usage))); |
241994ed JW |
6701 | } |
6702 | ||
00501b53 | 6703 | /** |
f0e45fb4 | 6704 | * mem_cgroup_charge - charge a newly allocated page to a cgroup |
00501b53 JW |
6705 | * @page: page to charge |
6706 | * @mm: mm context of the victim | |
6707 | * @gfp_mask: reclaim mode | |
00501b53 JW |
6708 | * |
6709 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
6710 | * pages according to @gfp_mask if necessary. | |
6711 | * | |
f0e45fb4 | 6712 | * Returns 0 on success. Otherwise, an error code is returned. |
00501b53 | 6713 | */ |
d9eb1ea2 | 6714 | int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask) |
00501b53 | 6715 | { |
6c357848 | 6716 | unsigned int nr_pages = thp_nr_pages(page); |
00501b53 | 6717 | struct mem_cgroup *memcg = NULL; |
00501b53 JW |
6718 | int ret = 0; |
6719 | ||
6720 | if (mem_cgroup_disabled()) | |
6721 | goto out; | |
6722 | ||
6723 | if (PageSwapCache(page)) { | |
2d1c4980 JW |
6724 | swp_entry_t ent = { .val = page_private(page), }; |
6725 | unsigned short id; | |
6726 | ||
00501b53 JW |
6727 | /* |
6728 | * Every swap fault against a single page tries to charge the | |
6729 | * page, bail as early as possible. shmem_unuse() encounters | |
eccb52e7 JW |
6730 | * already charged pages, too. page->mem_cgroup is protected |
6731 | * by the page lock, which serializes swap cache removal, which | |
00501b53 JW |
6732 | * in turn serializes uncharging. |
6733 | */ | |
e993d905 | 6734 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
abe2895b | 6735 | if (compound_head(page)->mem_cgroup) |
00501b53 | 6736 | goto out; |
e993d905 | 6737 | |
2d1c4980 JW |
6738 | id = lookup_swap_cgroup_id(ent); |
6739 | rcu_read_lock(); | |
6740 | memcg = mem_cgroup_from_id(id); | |
6741 | if (memcg && !css_tryget_online(&memcg->css)) | |
6742 | memcg = NULL; | |
6743 | rcu_read_unlock(); | |
00501b53 JW |
6744 | } |
6745 | ||
00501b53 JW |
6746 | if (!memcg) |
6747 | memcg = get_mem_cgroup_from_mm(mm); | |
6748 | ||
6749 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
f0e45fb4 JW |
6750 | if (ret) |
6751 | goto out_put; | |
00501b53 | 6752 | |
1a3e1f40 | 6753 | css_get(&memcg->css); |
d9eb1ea2 | 6754 | commit_charge(page, memcg); |
6abb5a86 | 6755 | |
6abb5a86 | 6756 | local_irq_disable(); |
3fba69a5 | 6757 | mem_cgroup_charge_statistics(memcg, page, nr_pages); |
6abb5a86 JW |
6758 | memcg_check_events(memcg, page); |
6759 | local_irq_enable(); | |
00501b53 | 6760 | |
2d1c4980 | 6761 | if (PageSwapCache(page)) { |
00501b53 JW |
6762 | swp_entry_t entry = { .val = page_private(page) }; |
6763 | /* | |
6764 | * The swap entry might not get freed for a long time, | |
6765 | * let's not wait for it. The page already received a | |
6766 | * memory+swap charge, drop the swap entry duplicate. | |
6767 | */ | |
38d8b4e6 | 6768 | mem_cgroup_uncharge_swap(entry, nr_pages); |
00501b53 | 6769 | } |
00501b53 | 6770 | |
f0e45fb4 JW |
6771 | out_put: |
6772 | css_put(&memcg->css); | |
6773 | out: | |
6774 | return ret; | |
3fea5a49 JW |
6775 | } |
6776 | ||
a9d5adee JG |
6777 | struct uncharge_gather { |
6778 | struct mem_cgroup *memcg; | |
9f762dbe | 6779 | unsigned long nr_pages; |
a9d5adee | 6780 | unsigned long pgpgout; |
a9d5adee | 6781 | unsigned long nr_kmem; |
a9d5adee JG |
6782 | struct page *dummy_page; |
6783 | }; | |
6784 | ||
6785 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 6786 | { |
a9d5adee JG |
6787 | memset(ug, 0, sizeof(*ug)); |
6788 | } | |
6789 | ||
6790 | static void uncharge_batch(const struct uncharge_gather *ug) | |
6791 | { | |
747db954 JW |
6792 | unsigned long flags; |
6793 | ||
a9d5adee | 6794 | if (!mem_cgroup_is_root(ug->memcg)) { |
9f762dbe | 6795 | page_counter_uncharge(&ug->memcg->memory, ug->nr_pages); |
7941d214 | 6796 | if (do_memsw_account()) |
9f762dbe | 6797 | page_counter_uncharge(&ug->memcg->memsw, ug->nr_pages); |
a9d5adee JG |
6798 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem) |
6799 | page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem); | |
6800 | memcg_oom_recover(ug->memcg); | |
ce00a967 | 6801 | } |
747db954 JW |
6802 | |
6803 | local_irq_save(flags); | |
c9019e9b | 6804 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); |
9f762dbe | 6805 | __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, ug->nr_pages); |
a9d5adee | 6806 | memcg_check_events(ug->memcg, ug->dummy_page); |
747db954 | 6807 | local_irq_restore(flags); |
f1796544 MH |
6808 | |
6809 | /* drop reference from uncharge_page */ | |
6810 | css_put(&ug->memcg->css); | |
a9d5adee JG |
6811 | } |
6812 | ||
6813 | static void uncharge_page(struct page *page, struct uncharge_gather *ug) | |
6814 | { | |
9f762dbe JW |
6815 | unsigned long nr_pages; |
6816 | ||
a9d5adee | 6817 | VM_BUG_ON_PAGE(PageLRU(page), page); |
a9d5adee JG |
6818 | |
6819 | if (!page->mem_cgroup) | |
6820 | return; | |
6821 | ||
6822 | /* | |
6823 | * Nobody should be changing or seriously looking at | |
6824 | * page->mem_cgroup at this point, we have fully | |
6825 | * exclusive access to the page. | |
6826 | */ | |
6827 | ||
6828 | if (ug->memcg != page->mem_cgroup) { | |
6829 | if (ug->memcg) { | |
6830 | uncharge_batch(ug); | |
6831 | uncharge_gather_clear(ug); | |
6832 | } | |
6833 | ug->memcg = page->mem_cgroup; | |
f1796544 MH |
6834 | |
6835 | /* pairs with css_put in uncharge_batch */ | |
6836 | css_get(&ug->memcg->css); | |
a9d5adee JG |
6837 | } |
6838 | ||
9f762dbe JW |
6839 | nr_pages = compound_nr(page); |
6840 | ug->nr_pages += nr_pages; | |
a9d5adee | 6841 | |
9f762dbe | 6842 | if (!PageKmemcg(page)) { |
a9d5adee JG |
6843 | ug->pgpgout++; |
6844 | } else { | |
9f762dbe | 6845 | ug->nr_kmem += nr_pages; |
a9d5adee JG |
6846 | __ClearPageKmemcg(page); |
6847 | } | |
6848 | ||
6849 | ug->dummy_page = page; | |
6850 | page->mem_cgroup = NULL; | |
1a3e1f40 | 6851 | css_put(&ug->memcg->css); |
747db954 JW |
6852 | } |
6853 | ||
6854 | static void uncharge_list(struct list_head *page_list) | |
6855 | { | |
a9d5adee | 6856 | struct uncharge_gather ug; |
747db954 | 6857 | struct list_head *next; |
a9d5adee JG |
6858 | |
6859 | uncharge_gather_clear(&ug); | |
747db954 | 6860 | |
8b592656 JW |
6861 | /* |
6862 | * Note that the list can be a single page->lru; hence the | |
6863 | * do-while loop instead of a simple list_for_each_entry(). | |
6864 | */ | |
747db954 JW |
6865 | next = page_list->next; |
6866 | do { | |
a9d5adee JG |
6867 | struct page *page; |
6868 | ||
747db954 JW |
6869 | page = list_entry(next, struct page, lru); |
6870 | next = page->lru.next; | |
6871 | ||
a9d5adee | 6872 | uncharge_page(page, &ug); |
747db954 JW |
6873 | } while (next != page_list); |
6874 | ||
a9d5adee JG |
6875 | if (ug.memcg) |
6876 | uncharge_batch(&ug); | |
747db954 JW |
6877 | } |
6878 | ||
0a31bc97 JW |
6879 | /** |
6880 | * mem_cgroup_uncharge - uncharge a page | |
6881 | * @page: page to uncharge | |
6882 | * | |
f0e45fb4 | 6883 | * Uncharge a page previously charged with mem_cgroup_charge(). |
0a31bc97 JW |
6884 | */ |
6885 | void mem_cgroup_uncharge(struct page *page) | |
6886 | { | |
a9d5adee JG |
6887 | struct uncharge_gather ug; |
6888 | ||
0a31bc97 JW |
6889 | if (mem_cgroup_disabled()) |
6890 | return; | |
6891 | ||
747db954 | 6892 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 6893 | if (!page->mem_cgroup) |
0a31bc97 JW |
6894 | return; |
6895 | ||
a9d5adee JG |
6896 | uncharge_gather_clear(&ug); |
6897 | uncharge_page(page, &ug); | |
6898 | uncharge_batch(&ug); | |
747db954 | 6899 | } |
0a31bc97 | 6900 | |
747db954 JW |
6901 | /** |
6902 | * mem_cgroup_uncharge_list - uncharge a list of page | |
6903 | * @page_list: list of pages to uncharge | |
6904 | * | |
6905 | * Uncharge a list of pages previously charged with | |
f0e45fb4 | 6906 | * mem_cgroup_charge(). |
747db954 JW |
6907 | */ |
6908 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
6909 | { | |
6910 | if (mem_cgroup_disabled()) | |
6911 | return; | |
0a31bc97 | 6912 | |
747db954 JW |
6913 | if (!list_empty(page_list)) |
6914 | uncharge_list(page_list); | |
0a31bc97 JW |
6915 | } |
6916 | ||
6917 | /** | |
6a93ca8f JW |
6918 | * mem_cgroup_migrate - charge a page's replacement |
6919 | * @oldpage: currently circulating page | |
6920 | * @newpage: replacement page | |
0a31bc97 | 6921 | * |
6a93ca8f JW |
6922 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
6923 | * be uncharged upon free. | |
0a31bc97 JW |
6924 | * |
6925 | * Both pages must be locked, @newpage->mapping must be set up. | |
6926 | */ | |
6a93ca8f | 6927 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 6928 | { |
29833315 | 6929 | struct mem_cgroup *memcg; |
44b7a8d3 | 6930 | unsigned int nr_pages; |
d93c4130 | 6931 | unsigned long flags; |
0a31bc97 JW |
6932 | |
6933 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6934 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 6935 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
6936 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6937 | newpage); | |
0a31bc97 JW |
6938 | |
6939 | if (mem_cgroup_disabled()) | |
6940 | return; | |
6941 | ||
6942 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 6943 | if (newpage->mem_cgroup) |
0a31bc97 JW |
6944 | return; |
6945 | ||
45637bab | 6946 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 6947 | memcg = oldpage->mem_cgroup; |
29833315 | 6948 | if (!memcg) |
0a31bc97 JW |
6949 | return; |
6950 | ||
44b7a8d3 | 6951 | /* Force-charge the new page. The old one will be freed soon */ |
6c357848 | 6952 | nr_pages = thp_nr_pages(newpage); |
44b7a8d3 JW |
6953 | |
6954 | page_counter_charge(&memcg->memory, nr_pages); | |
6955 | if (do_memsw_account()) | |
6956 | page_counter_charge(&memcg->memsw, nr_pages); | |
0a31bc97 | 6957 | |
1a3e1f40 | 6958 | css_get(&memcg->css); |
d9eb1ea2 | 6959 | commit_charge(newpage, memcg); |
44b7a8d3 | 6960 | |
d93c4130 | 6961 | local_irq_save(flags); |
3fba69a5 | 6962 | mem_cgroup_charge_statistics(memcg, newpage, nr_pages); |
44b7a8d3 | 6963 | memcg_check_events(memcg, newpage); |
d93c4130 | 6964 | local_irq_restore(flags); |
0a31bc97 JW |
6965 | } |
6966 | ||
ef12947c | 6967 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
6968 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
6969 | ||
2d758073 | 6970 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
6971 | { |
6972 | struct mem_cgroup *memcg; | |
6973 | ||
2d758073 JW |
6974 | if (!mem_cgroup_sockets_enabled) |
6975 | return; | |
6976 | ||
e876ecc6 SB |
6977 | /* Do not associate the sock with unrelated interrupted task's memcg. */ |
6978 | if (in_interrupt()) | |
6979 | return; | |
6980 | ||
11092087 JW |
6981 | rcu_read_lock(); |
6982 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
6983 | if (memcg == root_mem_cgroup) |
6984 | goto out; | |
0db15298 | 6985 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 6986 | goto out; |
8965aa28 | 6987 | if (css_tryget(&memcg->css)) |
11092087 | 6988 | sk->sk_memcg = memcg; |
f7e1cb6e | 6989 | out: |
11092087 JW |
6990 | rcu_read_unlock(); |
6991 | } | |
11092087 | 6992 | |
2d758073 | 6993 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 6994 | { |
2d758073 JW |
6995 | if (sk->sk_memcg) |
6996 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
6997 | } |
6998 | ||
6999 | /** | |
7000 | * mem_cgroup_charge_skmem - charge socket memory | |
7001 | * @memcg: memcg to charge | |
7002 | * @nr_pages: number of pages to charge | |
7003 | * | |
7004 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
7005 | * @memcg's configured limit, %false if the charge had to be forced. | |
7006 | */ | |
7007 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
7008 | { | |
f7e1cb6e | 7009 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 7010 | |
f7e1cb6e | 7011 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 7012 | struct page_counter *fail; |
f7e1cb6e | 7013 | |
0db15298 JW |
7014 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
7015 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
7016 | return true; |
7017 | } | |
0db15298 JW |
7018 | page_counter_charge(&memcg->tcpmem, nr_pages); |
7019 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 7020 | return false; |
11092087 | 7021 | } |
d886f4e4 | 7022 | |
f7e1cb6e JW |
7023 | /* Don't block in the packet receive path */ |
7024 | if (in_softirq()) | |
7025 | gfp_mask = GFP_NOWAIT; | |
7026 | ||
c9019e9b | 7027 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); |
b2807f07 | 7028 | |
f7e1cb6e JW |
7029 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
7030 | return true; | |
7031 | ||
7032 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
7033 | return false; |
7034 | } | |
7035 | ||
7036 | /** | |
7037 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
7038 | * @memcg: memcg to uncharge |
7039 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
7040 | */ |
7041 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
7042 | { | |
f7e1cb6e | 7043 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 7044 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
7045 | return; |
7046 | } | |
d886f4e4 | 7047 | |
c9019e9b | 7048 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 7049 | |
475d0487 | 7050 | refill_stock(memcg, nr_pages); |
11092087 JW |
7051 | } |
7052 | ||
f7e1cb6e JW |
7053 | static int __init cgroup_memory(char *s) |
7054 | { | |
7055 | char *token; | |
7056 | ||
7057 | while ((token = strsep(&s, ",")) != NULL) { | |
7058 | if (!*token) | |
7059 | continue; | |
7060 | if (!strcmp(token, "nosocket")) | |
7061 | cgroup_memory_nosocket = true; | |
04823c83 VD |
7062 | if (!strcmp(token, "nokmem")) |
7063 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
7064 | } |
7065 | return 0; | |
7066 | } | |
7067 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 7068 | |
2d11085e | 7069 | /* |
1081312f MH |
7070 | * subsys_initcall() for memory controller. |
7071 | * | |
308167fc SAS |
7072 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
7073 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
7074 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
7075 | * should be initialized from here. | |
2d11085e MH |
7076 | */ |
7077 | static int __init mem_cgroup_init(void) | |
7078 | { | |
95a045f6 JW |
7079 | int cpu, node; |
7080 | ||
308167fc SAS |
7081 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
7082 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
7083 | |
7084 | for_each_possible_cpu(cpu) | |
7085 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
7086 | drain_local_stock); | |
7087 | ||
7088 | for_each_node(node) { | |
7089 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
7090 | |
7091 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
7092 | node_online(node) ? node : NUMA_NO_NODE); | |
7093 | ||
ef8f2327 | 7094 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 7095 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 7096 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
7097 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
7098 | } | |
7099 | ||
2d11085e MH |
7100 | return 0; |
7101 | } | |
7102 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
7103 | |
7104 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
7105 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
7106 | { | |
1c2d479a | 7107 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
7108 | /* |
7109 | * The root cgroup cannot be destroyed, so it's refcount must | |
7110 | * always be >= 1. | |
7111 | */ | |
7112 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
7113 | VM_BUG_ON(1); | |
7114 | break; | |
7115 | } | |
7116 | memcg = parent_mem_cgroup(memcg); | |
7117 | if (!memcg) | |
7118 | memcg = root_mem_cgroup; | |
7119 | } | |
7120 | return memcg; | |
7121 | } | |
7122 | ||
21afa38e JW |
7123 | /** |
7124 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
7125 | * @page: page whose memsw charge to transfer | |
7126 | * @entry: swap entry to move the charge to | |
7127 | * | |
7128 | * Transfer the memsw charge of @page to @entry. | |
7129 | */ | |
7130 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
7131 | { | |
1f47b61f | 7132 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 7133 | unsigned int nr_entries; |
21afa38e JW |
7134 | unsigned short oldid; |
7135 | ||
7136 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
7137 | VM_BUG_ON_PAGE(page_count(page), page); | |
7138 | ||
2d1c4980 | 7139 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
21afa38e JW |
7140 | return; |
7141 | ||
7142 | memcg = page->mem_cgroup; | |
7143 | ||
7144 | /* Readahead page, never charged */ | |
7145 | if (!memcg) | |
7146 | return; | |
7147 | ||
1f47b61f VD |
7148 | /* |
7149 | * In case the memcg owning these pages has been offlined and doesn't | |
7150 | * have an ID allocated to it anymore, charge the closest online | |
7151 | * ancestor for the swap instead and transfer the memory+swap charge. | |
7152 | */ | |
7153 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
6c357848 | 7154 | nr_entries = thp_nr_pages(page); |
d6810d73 HY |
7155 | /* Get references for the tail pages, too */ |
7156 | if (nr_entries > 1) | |
7157 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
7158 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
7159 | nr_entries); | |
21afa38e | 7160 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7161 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e JW |
7162 | |
7163 | page->mem_cgroup = NULL; | |
7164 | ||
7165 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 7166 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 7167 | |
2d1c4980 | 7168 | if (!cgroup_memory_noswap && memcg != swap_memcg) { |
1f47b61f | 7169 | if (!mem_cgroup_is_root(swap_memcg)) |
d6810d73 HY |
7170 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
7171 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
7172 | } |
7173 | ||
ce9ce665 SAS |
7174 | /* |
7175 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 7176 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 7177 | * important here to have the interrupts disabled because it is the |
b93b0163 | 7178 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 SAS |
7179 | */ |
7180 | VM_BUG_ON(!irqs_disabled()); | |
3fba69a5 | 7181 | mem_cgroup_charge_statistics(memcg, page, -nr_entries); |
21afa38e | 7182 | memcg_check_events(memcg, page); |
73f576c0 | 7183 | |
1a3e1f40 | 7184 | css_put(&memcg->css); |
21afa38e JW |
7185 | } |
7186 | ||
38d8b4e6 HY |
7187 | /** |
7188 | * mem_cgroup_try_charge_swap - try charging swap space for a page | |
37e84351 VD |
7189 | * @page: page being added to swap |
7190 | * @entry: swap entry to charge | |
7191 | * | |
38d8b4e6 | 7192 | * Try to charge @page's memcg for the swap space at @entry. |
37e84351 VD |
7193 | * |
7194 | * Returns 0 on success, -ENOMEM on failure. | |
7195 | */ | |
7196 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
7197 | { | |
6c357848 | 7198 | unsigned int nr_pages = thp_nr_pages(page); |
37e84351 | 7199 | struct page_counter *counter; |
38d8b4e6 | 7200 | struct mem_cgroup *memcg; |
37e84351 VD |
7201 | unsigned short oldid; |
7202 | ||
2d1c4980 | 7203 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
37e84351 VD |
7204 | return 0; |
7205 | ||
7206 | memcg = page->mem_cgroup; | |
7207 | ||
7208 | /* Readahead page, never charged */ | |
7209 | if (!memcg) | |
7210 | return 0; | |
7211 | ||
f3a53a3a TH |
7212 | if (!entry.val) { |
7213 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 7214 | return 0; |
f3a53a3a | 7215 | } |
bb98f2c5 | 7216 | |
1f47b61f VD |
7217 | memcg = mem_cgroup_id_get_online(memcg); |
7218 | ||
2d1c4980 | 7219 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg) && |
38d8b4e6 | 7220 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
7221 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
7222 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 7223 | mem_cgroup_id_put(memcg); |
37e84351 | 7224 | return -ENOMEM; |
1f47b61f | 7225 | } |
37e84351 | 7226 | |
38d8b4e6 HY |
7227 | /* Get references for the tail pages, too */ |
7228 | if (nr_pages > 1) | |
7229 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
7230 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
37e84351 | 7231 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7232 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 7233 | |
37e84351 VD |
7234 | return 0; |
7235 | } | |
7236 | ||
21afa38e | 7237 | /** |
38d8b4e6 | 7238 | * mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 7239 | * @entry: swap entry to uncharge |
38d8b4e6 | 7240 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 7241 | */ |
38d8b4e6 | 7242 | void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
7243 | { |
7244 | struct mem_cgroup *memcg; | |
7245 | unsigned short id; | |
7246 | ||
38d8b4e6 | 7247 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 7248 | rcu_read_lock(); |
adbe427b | 7249 | memcg = mem_cgroup_from_id(id); |
21afa38e | 7250 | if (memcg) { |
2d1c4980 | 7251 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg)) { |
37e84351 | 7252 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
38d8b4e6 | 7253 | page_counter_uncharge(&memcg->swap, nr_pages); |
37e84351 | 7254 | else |
38d8b4e6 | 7255 | page_counter_uncharge(&memcg->memsw, nr_pages); |
37e84351 | 7256 | } |
c9019e9b | 7257 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 7258 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
7259 | } |
7260 | rcu_read_unlock(); | |
7261 | } | |
7262 | ||
d8b38438 VD |
7263 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
7264 | { | |
7265 | long nr_swap_pages = get_nr_swap_pages(); | |
7266 | ||
eccb52e7 | 7267 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
d8b38438 VD |
7268 | return nr_swap_pages; |
7269 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
7270 | nr_swap_pages = min_t(long, nr_swap_pages, | |
bbec2e15 | 7271 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
7272 | page_counter_read(&memcg->swap)); |
7273 | return nr_swap_pages; | |
7274 | } | |
7275 | ||
5ccc5aba VD |
7276 | bool mem_cgroup_swap_full(struct page *page) |
7277 | { | |
7278 | struct mem_cgroup *memcg; | |
7279 | ||
7280 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
7281 | ||
7282 | if (vm_swap_full()) | |
7283 | return true; | |
eccb52e7 | 7284 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
5ccc5aba VD |
7285 | return false; |
7286 | ||
7287 | memcg = page->mem_cgroup; | |
7288 | if (!memcg) | |
7289 | return false; | |
7290 | ||
4b82ab4f JK |
7291 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { |
7292 | unsigned long usage = page_counter_read(&memcg->swap); | |
7293 | ||
7294 | if (usage * 2 >= READ_ONCE(memcg->swap.high) || | |
7295 | usage * 2 >= READ_ONCE(memcg->swap.max)) | |
5ccc5aba | 7296 | return true; |
4b82ab4f | 7297 | } |
5ccc5aba VD |
7298 | |
7299 | return false; | |
7300 | } | |
7301 | ||
eccb52e7 | 7302 | static int __init setup_swap_account(char *s) |
21afa38e JW |
7303 | { |
7304 | if (!strcmp(s, "1")) | |
eccb52e7 | 7305 | cgroup_memory_noswap = 0; |
21afa38e | 7306 | else if (!strcmp(s, "0")) |
eccb52e7 | 7307 | cgroup_memory_noswap = 1; |
21afa38e JW |
7308 | return 1; |
7309 | } | |
eccb52e7 | 7310 | __setup("swapaccount=", setup_swap_account); |
21afa38e | 7311 | |
37e84351 VD |
7312 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
7313 | struct cftype *cft) | |
7314 | { | |
7315 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
7316 | ||
7317 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
7318 | } | |
7319 | ||
4b82ab4f JK |
7320 | static int swap_high_show(struct seq_file *m, void *v) |
7321 | { | |
7322 | return seq_puts_memcg_tunable(m, | |
7323 | READ_ONCE(mem_cgroup_from_seq(m)->swap.high)); | |
7324 | } | |
7325 | ||
7326 | static ssize_t swap_high_write(struct kernfs_open_file *of, | |
7327 | char *buf, size_t nbytes, loff_t off) | |
7328 | { | |
7329 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7330 | unsigned long high; | |
7331 | int err; | |
7332 | ||
7333 | buf = strstrip(buf); | |
7334 | err = page_counter_memparse(buf, "max", &high); | |
7335 | if (err) | |
7336 | return err; | |
7337 | ||
7338 | page_counter_set_high(&memcg->swap, high); | |
7339 | ||
7340 | return nbytes; | |
7341 | } | |
7342 | ||
37e84351 VD |
7343 | static int swap_max_show(struct seq_file *m, void *v) |
7344 | { | |
677dc973 CD |
7345 | return seq_puts_memcg_tunable(m, |
7346 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); | |
37e84351 VD |
7347 | } |
7348 | ||
7349 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
7350 | char *buf, size_t nbytes, loff_t off) | |
7351 | { | |
7352 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7353 | unsigned long max; | |
7354 | int err; | |
7355 | ||
7356 | buf = strstrip(buf); | |
7357 | err = page_counter_memparse(buf, "max", &max); | |
7358 | if (err) | |
7359 | return err; | |
7360 | ||
be09102b | 7361 | xchg(&memcg->swap.max, max); |
37e84351 VD |
7362 | |
7363 | return nbytes; | |
7364 | } | |
7365 | ||
f3a53a3a TH |
7366 | static int swap_events_show(struct seq_file *m, void *v) |
7367 | { | |
aa9694bb | 7368 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
f3a53a3a | 7369 | |
4b82ab4f JK |
7370 | seq_printf(m, "high %lu\n", |
7371 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_HIGH])); | |
f3a53a3a TH |
7372 | seq_printf(m, "max %lu\n", |
7373 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
7374 | seq_printf(m, "fail %lu\n", | |
7375 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
7376 | ||
7377 | return 0; | |
7378 | } | |
7379 | ||
37e84351 VD |
7380 | static struct cftype swap_files[] = { |
7381 | { | |
7382 | .name = "swap.current", | |
7383 | .flags = CFTYPE_NOT_ON_ROOT, | |
7384 | .read_u64 = swap_current_read, | |
7385 | }, | |
4b82ab4f JK |
7386 | { |
7387 | .name = "swap.high", | |
7388 | .flags = CFTYPE_NOT_ON_ROOT, | |
7389 | .seq_show = swap_high_show, | |
7390 | .write = swap_high_write, | |
7391 | }, | |
37e84351 VD |
7392 | { |
7393 | .name = "swap.max", | |
7394 | .flags = CFTYPE_NOT_ON_ROOT, | |
7395 | .seq_show = swap_max_show, | |
7396 | .write = swap_max_write, | |
7397 | }, | |
f3a53a3a TH |
7398 | { |
7399 | .name = "swap.events", | |
7400 | .flags = CFTYPE_NOT_ON_ROOT, | |
7401 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
7402 | .seq_show = swap_events_show, | |
7403 | }, | |
37e84351 VD |
7404 | { } /* terminate */ |
7405 | }; | |
7406 | ||
eccb52e7 | 7407 | static struct cftype memsw_files[] = { |
21afa38e JW |
7408 | { |
7409 | .name = "memsw.usage_in_bytes", | |
7410 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
7411 | .read_u64 = mem_cgroup_read_u64, | |
7412 | }, | |
7413 | { | |
7414 | .name = "memsw.max_usage_in_bytes", | |
7415 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
7416 | .write = mem_cgroup_reset, | |
7417 | .read_u64 = mem_cgroup_read_u64, | |
7418 | }, | |
7419 | { | |
7420 | .name = "memsw.limit_in_bytes", | |
7421 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
7422 | .write = mem_cgroup_write, | |
7423 | .read_u64 = mem_cgroup_read_u64, | |
7424 | }, | |
7425 | { | |
7426 | .name = "memsw.failcnt", | |
7427 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
7428 | .write = mem_cgroup_reset, | |
7429 | .read_u64 = mem_cgroup_read_u64, | |
7430 | }, | |
7431 | { }, /* terminate */ | |
7432 | }; | |
7433 | ||
82ff165c BS |
7434 | /* |
7435 | * If mem_cgroup_swap_init() is implemented as a subsys_initcall() | |
7436 | * instead of a core_initcall(), this could mean cgroup_memory_noswap still | |
7437 | * remains set to false even when memcg is disabled via "cgroup_disable=memory" | |
7438 | * boot parameter. This may result in premature OOPS inside | |
7439 | * mem_cgroup_get_nr_swap_pages() function in corner cases. | |
7440 | */ | |
21afa38e JW |
7441 | static int __init mem_cgroup_swap_init(void) |
7442 | { | |
2d1c4980 JW |
7443 | /* No memory control -> no swap control */ |
7444 | if (mem_cgroup_disabled()) | |
7445 | cgroup_memory_noswap = true; | |
7446 | ||
7447 | if (cgroup_memory_noswap) | |
eccb52e7 JW |
7448 | return 0; |
7449 | ||
7450 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, swap_files)); | |
7451 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, memsw_files)); | |
7452 | ||
21afa38e JW |
7453 | return 0; |
7454 | } | |
82ff165c | 7455 | core_initcall(mem_cgroup_swap_init); |
21afa38e JW |
7456 | |
7457 | #endif /* CONFIG_MEMCG_SWAP */ |