]>
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 | { | |
279c3393 RG |
1069 | if (memcg_kmem_bypass()) |
1070 | return NULL; | |
1071 | ||
d46eb14b | 1072 | if (unlikely(current->active_memcg)) { |
8965aa28 | 1073 | struct mem_cgroup *memcg; |
d46eb14b SB |
1074 | |
1075 | rcu_read_lock(); | |
8965aa28 SB |
1076 | /* current->active_memcg must hold a ref. */ |
1077 | if (WARN_ON_ONCE(!css_tryget(¤t->active_memcg->css))) | |
1078 | memcg = root_mem_cgroup; | |
1079 | else | |
d46eb14b SB |
1080 | memcg = current->active_memcg; |
1081 | rcu_read_unlock(); | |
1082 | return memcg; | |
1083 | } | |
1084 | return get_mem_cgroup_from_mm(current->mm); | |
1085 | } | |
54595fe2 | 1086 | |
5660048c JW |
1087 | /** |
1088 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1089 | * @root: hierarchy root | |
1090 | * @prev: previously returned memcg, NULL on first invocation | |
1091 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1092 | * | |
1093 | * Returns references to children of the hierarchy below @root, or | |
1094 | * @root itself, or %NULL after a full round-trip. | |
1095 | * | |
1096 | * Caller must pass the return value in @prev on subsequent | |
1097 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1098 | * to cancel a hierarchy walk before the round-trip is complete. | |
1099 | * | |
05bdc520 ML |
1100 | * Reclaimers can specify a node in @reclaim to divide up the memcgs |
1101 | * in the hierarchy among all concurrent reclaimers operating on the | |
1102 | * same node. | |
5660048c | 1103 | */ |
694fbc0f | 1104 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1105 | struct mem_cgroup *prev, |
694fbc0f | 1106 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1107 | { |
3f649ab7 | 1108 | struct mem_cgroup_reclaim_iter *iter; |
5ac8fb31 | 1109 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 1110 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1111 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1112 | |
694fbc0f AM |
1113 | if (mem_cgroup_disabled()) |
1114 | return NULL; | |
5660048c | 1115 | |
9f3a0d09 JW |
1116 | if (!root) |
1117 | root = root_mem_cgroup; | |
7d74b06f | 1118 | |
9f3a0d09 | 1119 | if (prev && !reclaim) |
5ac8fb31 | 1120 | pos = prev; |
14067bb3 | 1121 | |
9f3a0d09 JW |
1122 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1123 | if (prev) | |
5ac8fb31 | 1124 | goto out; |
694fbc0f | 1125 | return root; |
9f3a0d09 | 1126 | } |
14067bb3 | 1127 | |
542f85f9 | 1128 | rcu_read_lock(); |
5f578161 | 1129 | |
5ac8fb31 | 1130 | if (reclaim) { |
ef8f2327 | 1131 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 1132 | |
ef8f2327 | 1133 | mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id); |
9da83f3f | 1134 | iter = &mz->iter; |
5ac8fb31 JW |
1135 | |
1136 | if (prev && reclaim->generation != iter->generation) | |
1137 | goto out_unlock; | |
1138 | ||
6df38689 | 1139 | while (1) { |
4db0c3c2 | 1140 | pos = READ_ONCE(iter->position); |
6df38689 VD |
1141 | if (!pos || css_tryget(&pos->css)) |
1142 | break; | |
5ac8fb31 | 1143 | /* |
6df38689 VD |
1144 | * css reference reached zero, so iter->position will |
1145 | * be cleared by ->css_released. However, we should not | |
1146 | * rely on this happening soon, because ->css_released | |
1147 | * is called from a work queue, and by busy-waiting we | |
1148 | * might block it. So we clear iter->position right | |
1149 | * away. | |
5ac8fb31 | 1150 | */ |
6df38689 VD |
1151 | (void)cmpxchg(&iter->position, pos, NULL); |
1152 | } | |
5ac8fb31 JW |
1153 | } |
1154 | ||
1155 | if (pos) | |
1156 | css = &pos->css; | |
1157 | ||
1158 | for (;;) { | |
1159 | css = css_next_descendant_pre(css, &root->css); | |
1160 | if (!css) { | |
1161 | /* | |
1162 | * Reclaimers share the hierarchy walk, and a | |
1163 | * new one might jump in right at the end of | |
1164 | * the hierarchy - make sure they see at least | |
1165 | * one group and restart from the beginning. | |
1166 | */ | |
1167 | if (!prev) | |
1168 | continue; | |
1169 | break; | |
527a5ec9 | 1170 | } |
7d74b06f | 1171 | |
5ac8fb31 JW |
1172 | /* |
1173 | * Verify the css and acquire a reference. The root | |
1174 | * is provided by the caller, so we know it's alive | |
1175 | * and kicking, and don't take an extra reference. | |
1176 | */ | |
1177 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 1178 | |
5ac8fb31 JW |
1179 | if (css == &root->css) |
1180 | break; | |
14067bb3 | 1181 | |
0b8f73e1 JW |
1182 | if (css_tryget(css)) |
1183 | break; | |
9f3a0d09 | 1184 | |
5ac8fb31 | 1185 | memcg = NULL; |
9f3a0d09 | 1186 | } |
5ac8fb31 JW |
1187 | |
1188 | if (reclaim) { | |
5ac8fb31 | 1189 | /* |
6df38689 VD |
1190 | * The position could have already been updated by a competing |
1191 | * thread, so check that the value hasn't changed since we read | |
1192 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 1193 | */ |
6df38689 VD |
1194 | (void)cmpxchg(&iter->position, pos, memcg); |
1195 | ||
5ac8fb31 JW |
1196 | if (pos) |
1197 | css_put(&pos->css); | |
1198 | ||
1199 | if (!memcg) | |
1200 | iter->generation++; | |
1201 | else if (!prev) | |
1202 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1203 | } |
5ac8fb31 | 1204 | |
542f85f9 MH |
1205 | out_unlock: |
1206 | rcu_read_unlock(); | |
5ac8fb31 | 1207 | out: |
c40046f3 MH |
1208 | if (prev && prev != root) |
1209 | css_put(&prev->css); | |
1210 | ||
9f3a0d09 | 1211 | return memcg; |
14067bb3 | 1212 | } |
7d74b06f | 1213 | |
5660048c JW |
1214 | /** |
1215 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1216 | * @root: hierarchy root | |
1217 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1218 | */ | |
1219 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1220 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1221 | { |
1222 | if (!root) | |
1223 | root = root_mem_cgroup; | |
1224 | if (prev && prev != root) | |
1225 | css_put(&prev->css); | |
1226 | } | |
7d74b06f | 1227 | |
54a83d6b MC |
1228 | static void __invalidate_reclaim_iterators(struct mem_cgroup *from, |
1229 | struct mem_cgroup *dead_memcg) | |
6df38689 | 1230 | { |
6df38689 | 1231 | struct mem_cgroup_reclaim_iter *iter; |
ef8f2327 MG |
1232 | struct mem_cgroup_per_node *mz; |
1233 | int nid; | |
6df38689 | 1234 | |
54a83d6b MC |
1235 | for_each_node(nid) { |
1236 | mz = mem_cgroup_nodeinfo(from, nid); | |
9da83f3f YS |
1237 | iter = &mz->iter; |
1238 | cmpxchg(&iter->position, dead_memcg, NULL); | |
6df38689 VD |
1239 | } |
1240 | } | |
1241 | ||
54a83d6b MC |
1242 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1243 | { | |
1244 | struct mem_cgroup *memcg = dead_memcg; | |
1245 | struct mem_cgroup *last; | |
1246 | ||
1247 | do { | |
1248 | __invalidate_reclaim_iterators(memcg, dead_memcg); | |
1249 | last = memcg; | |
1250 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1251 | ||
1252 | /* | |
1253 | * When cgruop1 non-hierarchy mode is used, | |
1254 | * parent_mem_cgroup() does not walk all the way up to the | |
1255 | * cgroup root (root_mem_cgroup). So we have to handle | |
1256 | * dead_memcg from cgroup root separately. | |
1257 | */ | |
1258 | if (last != root_mem_cgroup) | |
1259 | __invalidate_reclaim_iterators(root_mem_cgroup, | |
1260 | dead_memcg); | |
1261 | } | |
1262 | ||
7c5f64f8 VD |
1263 | /** |
1264 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1265 | * @memcg: hierarchy root | |
1266 | * @fn: function to call for each task | |
1267 | * @arg: argument passed to @fn | |
1268 | * | |
1269 | * This function iterates over tasks attached to @memcg or to any of its | |
1270 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
1271 | * value, the function breaks the iteration loop and returns the value. | |
1272 | * Otherwise, it will iterate over all tasks and return 0. | |
1273 | * | |
1274 | * This function must not be called for the root memory cgroup. | |
1275 | */ | |
1276 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
1277 | int (*fn)(struct task_struct *, void *), void *arg) | |
1278 | { | |
1279 | struct mem_cgroup *iter; | |
1280 | int ret = 0; | |
1281 | ||
1282 | BUG_ON(memcg == root_mem_cgroup); | |
1283 | ||
1284 | for_each_mem_cgroup_tree(iter, memcg) { | |
1285 | struct css_task_iter it; | |
1286 | struct task_struct *task; | |
1287 | ||
f168a9a5 | 1288 | css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it); |
7c5f64f8 VD |
1289 | while (!ret && (task = css_task_iter_next(&it))) |
1290 | ret = fn(task, arg); | |
1291 | css_task_iter_end(&it); | |
1292 | if (ret) { | |
1293 | mem_cgroup_iter_break(memcg, iter); | |
1294 | break; | |
1295 | } | |
1296 | } | |
1297 | return ret; | |
1298 | } | |
1299 | ||
925b7673 | 1300 | /** |
dfe0e773 | 1301 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1302 | * @page: the page |
f144c390 | 1303 | * @pgdat: pgdat of the page |
dfe0e773 | 1304 | * |
a0b5b414 JW |
1305 | * This function relies on page->mem_cgroup being stable - see the |
1306 | * access rules in commit_charge(). | |
925b7673 | 1307 | */ |
599d0c95 | 1308 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat) |
08e552c6 | 1309 | { |
ef8f2327 | 1310 | struct mem_cgroup_per_node *mz; |
925b7673 | 1311 | struct mem_cgroup *memcg; |
bea8c150 | 1312 | struct lruvec *lruvec; |
6d12e2d8 | 1313 | |
bea8c150 | 1314 | if (mem_cgroup_disabled()) { |
867e5e1d | 1315 | lruvec = &pgdat->__lruvec; |
bea8c150 HD |
1316 | goto out; |
1317 | } | |
925b7673 | 1318 | |
1306a85a | 1319 | memcg = page->mem_cgroup; |
7512102c | 1320 | /* |
dfe0e773 | 1321 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1322 | * possibly migrated - before they are charged. |
7512102c | 1323 | */ |
29833315 JW |
1324 | if (!memcg) |
1325 | memcg = root_mem_cgroup; | |
7512102c | 1326 | |
ef8f2327 | 1327 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
bea8c150 HD |
1328 | lruvec = &mz->lruvec; |
1329 | out: | |
1330 | /* | |
1331 | * Since a node can be onlined after the mem_cgroup was created, | |
1332 | * we have to be prepared to initialize lruvec->zone here; | |
1333 | * and if offlined then reonlined, we need to reinitialize it. | |
1334 | */ | |
599d0c95 MG |
1335 | if (unlikely(lruvec->pgdat != pgdat)) |
1336 | lruvec->pgdat = pgdat; | |
bea8c150 | 1337 | return lruvec; |
08e552c6 | 1338 | } |
b69408e8 | 1339 | |
925b7673 | 1340 | /** |
fa9add64 HD |
1341 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1342 | * @lruvec: mem_cgroup per zone lru vector | |
1343 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1344 | * @zid: zone id of the accounted pages |
fa9add64 | 1345 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1346 | * |
ca707239 HD |
1347 | * This function must be called under lru_lock, just before a page is added |
1348 | * to or just after a page is removed from an lru list (that ordering being | |
1349 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 1350 | */ |
fa9add64 | 1351 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1352 | int zid, int nr_pages) |
3f58a829 | 1353 | { |
ef8f2327 | 1354 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1355 | unsigned long *lru_size; |
ca707239 | 1356 | long size; |
3f58a829 MK |
1357 | |
1358 | if (mem_cgroup_disabled()) | |
1359 | return; | |
1360 | ||
ef8f2327 | 1361 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1362 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1363 | |
1364 | if (nr_pages < 0) | |
1365 | *lru_size += nr_pages; | |
1366 | ||
1367 | size = *lru_size; | |
b4536f0c MH |
1368 | if (WARN_ONCE(size < 0, |
1369 | "%s(%p, %d, %d): lru_size %ld\n", | |
1370 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1371 | VM_BUG_ON(1); |
1372 | *lru_size = 0; | |
1373 | } | |
1374 | ||
1375 | if (nr_pages > 0) | |
1376 | *lru_size += nr_pages; | |
08e552c6 | 1377 | } |
544122e5 | 1378 | |
19942822 | 1379 | /** |
9d11ea9f | 1380 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1381 | * @memcg: the memory cgroup |
19942822 | 1382 | * |
9d11ea9f | 1383 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1384 | * pages. |
19942822 | 1385 | */ |
c0ff4b85 | 1386 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1387 | { |
3e32cb2e JW |
1388 | unsigned long margin = 0; |
1389 | unsigned long count; | |
1390 | unsigned long limit; | |
9d11ea9f | 1391 | |
3e32cb2e | 1392 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1393 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1394 | if (count < limit) |
1395 | margin = limit - count; | |
1396 | ||
7941d214 | 1397 | if (do_memsw_account()) { |
3e32cb2e | 1398 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1399 | limit = READ_ONCE(memcg->memsw.max); |
1c4448ed | 1400 | if (count < limit) |
3e32cb2e | 1401 | margin = min(margin, limit - count); |
cbedbac3 LR |
1402 | else |
1403 | margin = 0; | |
3e32cb2e JW |
1404 | } |
1405 | ||
1406 | return margin; | |
19942822 JW |
1407 | } |
1408 | ||
32047e2a | 1409 | /* |
bdcbb659 | 1410 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1411 | * |
bdcbb659 QH |
1412 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1413 | * moving cgroups. This is for waiting at high-memory pressure | |
1414 | * caused by "move". | |
32047e2a | 1415 | */ |
c0ff4b85 | 1416 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1417 | { |
2bd9bb20 KH |
1418 | struct mem_cgroup *from; |
1419 | struct mem_cgroup *to; | |
4b534334 | 1420 | bool ret = false; |
2bd9bb20 KH |
1421 | /* |
1422 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1423 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1424 | */ | |
1425 | spin_lock(&mc.lock); | |
1426 | from = mc.from; | |
1427 | to = mc.to; | |
1428 | if (!from) | |
1429 | goto unlock; | |
3e92041d | 1430 | |
2314b42d JW |
1431 | ret = mem_cgroup_is_descendant(from, memcg) || |
1432 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1433 | unlock: |
1434 | spin_unlock(&mc.lock); | |
4b534334 KH |
1435 | return ret; |
1436 | } | |
1437 | ||
c0ff4b85 | 1438 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1439 | { |
1440 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1441 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1442 | DEFINE_WAIT(wait); |
1443 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1444 | /* moving charge context might have finished. */ | |
1445 | if (mc.moving_task) | |
1446 | schedule(); | |
1447 | finish_wait(&mc.waitq, &wait); | |
1448 | return true; | |
1449 | } | |
1450 | } | |
1451 | return false; | |
1452 | } | |
1453 | ||
5f9a4f4a MS |
1454 | struct memory_stat { |
1455 | const char *name; | |
1456 | unsigned int ratio; | |
1457 | unsigned int idx; | |
1458 | }; | |
1459 | ||
1460 | static struct memory_stat memory_stats[] = { | |
1461 | { "anon", PAGE_SIZE, NR_ANON_MAPPED }, | |
1462 | { "file", PAGE_SIZE, NR_FILE_PAGES }, | |
1463 | { "kernel_stack", 1024, NR_KERNEL_STACK_KB }, | |
1464 | { "percpu", 1, MEMCG_PERCPU_B }, | |
1465 | { "sock", PAGE_SIZE, MEMCG_SOCK }, | |
1466 | { "shmem", PAGE_SIZE, NR_SHMEM }, | |
1467 | { "file_mapped", PAGE_SIZE, NR_FILE_MAPPED }, | |
1468 | { "file_dirty", PAGE_SIZE, NR_FILE_DIRTY }, | |
1469 | { "file_writeback", PAGE_SIZE, NR_WRITEBACK }, | |
1470 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
1471 | /* | |
1472 | * The ratio will be initialized in memory_stats_init(). Because | |
1473 | * on some architectures, the macro of HPAGE_PMD_SIZE is not | |
1474 | * constant(e.g. powerpc). | |
1475 | */ | |
1476 | { "anon_thp", 0, NR_ANON_THPS }, | |
1477 | #endif | |
1478 | { "inactive_anon", PAGE_SIZE, NR_INACTIVE_ANON }, | |
1479 | { "active_anon", PAGE_SIZE, NR_ACTIVE_ANON }, | |
1480 | { "inactive_file", PAGE_SIZE, NR_INACTIVE_FILE }, | |
1481 | { "active_file", PAGE_SIZE, NR_ACTIVE_FILE }, | |
1482 | { "unevictable", PAGE_SIZE, NR_UNEVICTABLE }, | |
1483 | ||
1484 | /* | |
1485 | * Note: The slab_reclaimable and slab_unreclaimable must be | |
1486 | * together and slab_reclaimable must be in front. | |
1487 | */ | |
1488 | { "slab_reclaimable", 1, NR_SLAB_RECLAIMABLE_B }, | |
1489 | { "slab_unreclaimable", 1, NR_SLAB_UNRECLAIMABLE_B }, | |
1490 | ||
1491 | /* The memory events */ | |
1492 | { "workingset_refault_anon", 1, WORKINGSET_REFAULT_ANON }, | |
1493 | { "workingset_refault_file", 1, WORKINGSET_REFAULT_FILE }, | |
1494 | { "workingset_activate_anon", 1, WORKINGSET_ACTIVATE_ANON }, | |
1495 | { "workingset_activate_file", 1, WORKINGSET_ACTIVATE_FILE }, | |
1496 | { "workingset_restore_anon", 1, WORKINGSET_RESTORE_ANON }, | |
1497 | { "workingset_restore_file", 1, WORKINGSET_RESTORE_FILE }, | |
1498 | { "workingset_nodereclaim", 1, WORKINGSET_NODERECLAIM }, | |
1499 | }; | |
1500 | ||
1501 | static int __init memory_stats_init(void) | |
1502 | { | |
1503 | int i; | |
1504 | ||
1505 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { | |
1506 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
1507 | if (memory_stats[i].idx == NR_ANON_THPS) | |
1508 | memory_stats[i].ratio = HPAGE_PMD_SIZE; | |
1509 | #endif | |
1510 | VM_BUG_ON(!memory_stats[i].ratio); | |
1511 | VM_BUG_ON(memory_stats[i].idx >= MEMCG_NR_STAT); | |
1512 | } | |
1513 | ||
1514 | return 0; | |
1515 | } | |
1516 | pure_initcall(memory_stats_init); | |
1517 | ||
c8713d0b JW |
1518 | static char *memory_stat_format(struct mem_cgroup *memcg) |
1519 | { | |
1520 | struct seq_buf s; | |
1521 | int i; | |
71cd3113 | 1522 | |
c8713d0b JW |
1523 | seq_buf_init(&s, kmalloc(PAGE_SIZE, GFP_KERNEL), PAGE_SIZE); |
1524 | if (!s.buffer) | |
1525 | return NULL; | |
1526 | ||
1527 | /* | |
1528 | * Provide statistics on the state of the memory subsystem as | |
1529 | * well as cumulative event counters that show past behavior. | |
1530 | * | |
1531 | * This list is ordered following a combination of these gradients: | |
1532 | * 1) generic big picture -> specifics and details | |
1533 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
1534 | * | |
1535 | * Current memory state: | |
1536 | */ | |
1537 | ||
5f9a4f4a MS |
1538 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { |
1539 | u64 size; | |
c8713d0b | 1540 | |
5f9a4f4a MS |
1541 | size = memcg_page_state(memcg, memory_stats[i].idx); |
1542 | size *= memory_stats[i].ratio; | |
1543 | seq_buf_printf(&s, "%s %llu\n", memory_stats[i].name, size); | |
c8713d0b | 1544 | |
5f9a4f4a MS |
1545 | if (unlikely(memory_stats[i].idx == NR_SLAB_UNRECLAIMABLE_B)) { |
1546 | size = memcg_page_state(memcg, NR_SLAB_RECLAIMABLE_B) + | |
1547 | memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE_B); | |
1548 | seq_buf_printf(&s, "slab %llu\n", size); | |
1549 | } | |
1550 | } | |
c8713d0b JW |
1551 | |
1552 | /* Accumulated memory events */ | |
1553 | ||
ebc5d83d KK |
1554 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGFAULT), |
1555 | memcg_events(memcg, PGFAULT)); | |
1556 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGMAJFAULT), | |
1557 | memcg_events(memcg, PGMAJFAULT)); | |
ebc5d83d KK |
1558 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGREFILL), |
1559 | memcg_events(memcg, PGREFILL)); | |
c8713d0b JW |
1560 | seq_buf_printf(&s, "pgscan %lu\n", |
1561 | memcg_events(memcg, PGSCAN_KSWAPD) + | |
1562 | memcg_events(memcg, PGSCAN_DIRECT)); | |
1563 | seq_buf_printf(&s, "pgsteal %lu\n", | |
1564 | memcg_events(memcg, PGSTEAL_KSWAPD) + | |
1565 | memcg_events(memcg, PGSTEAL_DIRECT)); | |
ebc5d83d KK |
1566 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGACTIVATE), |
1567 | memcg_events(memcg, PGACTIVATE)); | |
1568 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGDEACTIVATE), | |
1569 | memcg_events(memcg, PGDEACTIVATE)); | |
1570 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREE), | |
1571 | memcg_events(memcg, PGLAZYFREE)); | |
1572 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREED), | |
1573 | memcg_events(memcg, PGLAZYFREED)); | |
c8713d0b JW |
1574 | |
1575 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
ebc5d83d | 1576 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_FAULT_ALLOC), |
c8713d0b | 1577 | memcg_events(memcg, THP_FAULT_ALLOC)); |
ebc5d83d | 1578 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_COLLAPSE_ALLOC), |
c8713d0b JW |
1579 | memcg_events(memcg, THP_COLLAPSE_ALLOC)); |
1580 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ | |
1581 | ||
1582 | /* The above should easily fit into one page */ | |
1583 | WARN_ON_ONCE(seq_buf_has_overflowed(&s)); | |
1584 | ||
1585 | return s.buffer; | |
1586 | } | |
71cd3113 | 1587 | |
58cf188e | 1588 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1589 | /** |
f0c867d9 | 1590 | * mem_cgroup_print_oom_context: Print OOM information relevant to |
1591 | * memory controller. | |
e222432b BS |
1592 | * @memcg: The memory cgroup that went over limit |
1593 | * @p: Task that is going to be killed | |
1594 | * | |
1595 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1596 | * enabled | |
1597 | */ | |
f0c867d9 | 1598 | void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) |
e222432b | 1599 | { |
e222432b BS |
1600 | rcu_read_lock(); |
1601 | ||
f0c867d9 | 1602 | if (memcg) { |
1603 | pr_cont(",oom_memcg="); | |
1604 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1605 | } else | |
1606 | pr_cont(",global_oom"); | |
2415b9f5 | 1607 | if (p) { |
f0c867d9 | 1608 | pr_cont(",task_memcg="); |
2415b9f5 | 1609 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
2415b9f5 | 1610 | } |
e222432b | 1611 | rcu_read_unlock(); |
f0c867d9 | 1612 | } |
1613 | ||
1614 | /** | |
1615 | * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to | |
1616 | * memory controller. | |
1617 | * @memcg: The memory cgroup that went over limit | |
1618 | */ | |
1619 | void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) | |
1620 | { | |
c8713d0b | 1621 | char *buf; |
e222432b | 1622 | |
3e32cb2e JW |
1623 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1624 | K((u64)page_counter_read(&memcg->memory)), | |
15b42562 | 1625 | K((u64)READ_ONCE(memcg->memory.max)), memcg->memory.failcnt); |
c8713d0b JW |
1626 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
1627 | pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1628 | K((u64)page_counter_read(&memcg->swap)), | |
32d087cd | 1629 | K((u64)READ_ONCE(memcg->swap.max)), memcg->swap.failcnt); |
c8713d0b JW |
1630 | else { |
1631 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1632 | K((u64)page_counter_read(&memcg->memsw)), | |
1633 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); | |
1634 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1635 | K((u64)page_counter_read(&memcg->kmem)), | |
1636 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); | |
58cf188e | 1637 | } |
c8713d0b JW |
1638 | |
1639 | pr_info("Memory cgroup stats for "); | |
1640 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1641 | pr_cont(":"); | |
1642 | buf = memory_stat_format(memcg); | |
1643 | if (!buf) | |
1644 | return; | |
1645 | pr_info("%s", buf); | |
1646 | kfree(buf); | |
e222432b BS |
1647 | } |
1648 | ||
a63d83f4 DR |
1649 | /* |
1650 | * Return the memory (and swap, if configured) limit for a memcg. | |
1651 | */ | |
bbec2e15 | 1652 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1653 | { |
8d387a5f WL |
1654 | unsigned long max = READ_ONCE(memcg->memory.max); |
1655 | ||
1656 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) { | |
1657 | if (mem_cgroup_swappiness(memcg)) | |
1658 | max += min(READ_ONCE(memcg->swap.max), | |
1659 | (unsigned long)total_swap_pages); | |
1660 | } else { /* v1 */ | |
1661 | if (mem_cgroup_swappiness(memcg)) { | |
1662 | /* Calculate swap excess capacity from memsw limit */ | |
1663 | unsigned long swap = READ_ONCE(memcg->memsw.max) - max; | |
1664 | ||
1665 | max += min(swap, (unsigned long)total_swap_pages); | |
1666 | } | |
9a5a8f19 | 1667 | } |
bbec2e15 | 1668 | return max; |
a63d83f4 DR |
1669 | } |
1670 | ||
9783aa99 CD |
1671 | unsigned long mem_cgroup_size(struct mem_cgroup *memcg) |
1672 | { | |
1673 | return page_counter_read(&memcg->memory); | |
1674 | } | |
1675 | ||
b6e6edcf | 1676 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1677 | int order) |
9cbb78bb | 1678 | { |
6e0fc46d DR |
1679 | struct oom_control oc = { |
1680 | .zonelist = NULL, | |
1681 | .nodemask = NULL, | |
2a966b77 | 1682 | .memcg = memcg, |
6e0fc46d DR |
1683 | .gfp_mask = gfp_mask, |
1684 | .order = order, | |
6e0fc46d | 1685 | }; |
1378b37d | 1686 | bool ret = true; |
9cbb78bb | 1687 | |
7775face TH |
1688 | if (mutex_lock_killable(&oom_lock)) |
1689 | return true; | |
1378b37d YS |
1690 | |
1691 | if (mem_cgroup_margin(memcg) >= (1 << order)) | |
1692 | goto unlock; | |
1693 | ||
7775face TH |
1694 | /* |
1695 | * A few threads which were not waiting at mutex_lock_killable() can | |
1696 | * fail to bail out. Therefore, check again after holding oom_lock. | |
1697 | */ | |
1698 | ret = should_force_charge() || out_of_memory(&oc); | |
1378b37d YS |
1699 | |
1700 | unlock: | |
dc56401f | 1701 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1702 | return ret; |
9cbb78bb DR |
1703 | } |
1704 | ||
0608f43d | 1705 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1706 | pg_data_t *pgdat, |
0608f43d AM |
1707 | gfp_t gfp_mask, |
1708 | unsigned long *total_scanned) | |
1709 | { | |
1710 | struct mem_cgroup *victim = NULL; | |
1711 | int total = 0; | |
1712 | int loop = 0; | |
1713 | unsigned long excess; | |
1714 | unsigned long nr_scanned; | |
1715 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1716 | .pgdat = pgdat, |
0608f43d AM |
1717 | }; |
1718 | ||
3e32cb2e | 1719 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1720 | |
1721 | while (1) { | |
1722 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1723 | if (!victim) { | |
1724 | loop++; | |
1725 | if (loop >= 2) { | |
1726 | /* | |
1727 | * If we have not been able to reclaim | |
1728 | * anything, it might because there are | |
1729 | * no reclaimable pages under this hierarchy | |
1730 | */ | |
1731 | if (!total) | |
1732 | break; | |
1733 | /* | |
1734 | * We want to do more targeted reclaim. | |
1735 | * excess >> 2 is not to excessive so as to | |
1736 | * reclaim too much, nor too less that we keep | |
1737 | * coming back to reclaim from this cgroup | |
1738 | */ | |
1739 | if (total >= (excess >> 2) || | |
1740 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1741 | break; | |
1742 | } | |
1743 | continue; | |
1744 | } | |
a9dd0a83 | 1745 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1746 | pgdat, &nr_scanned); |
0608f43d | 1747 | *total_scanned += nr_scanned; |
3e32cb2e | 1748 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1749 | break; |
6d61ef40 | 1750 | } |
0608f43d AM |
1751 | mem_cgroup_iter_break(root_memcg, victim); |
1752 | return total; | |
6d61ef40 BS |
1753 | } |
1754 | ||
0056f4e6 JW |
1755 | #ifdef CONFIG_LOCKDEP |
1756 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1757 | .name = "memcg_oom_lock", | |
1758 | }; | |
1759 | #endif | |
1760 | ||
fb2a6fc5 JW |
1761 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1762 | ||
867578cb KH |
1763 | /* |
1764 | * Check OOM-Killer is already running under our hierarchy. | |
1765 | * If someone is running, return false. | |
1766 | */ | |
fb2a6fc5 | 1767 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1768 | { |
79dfdacc | 1769 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1770 | |
fb2a6fc5 JW |
1771 | spin_lock(&memcg_oom_lock); |
1772 | ||
9f3a0d09 | 1773 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1774 | if (iter->oom_lock) { |
79dfdacc MH |
1775 | /* |
1776 | * this subtree of our hierarchy is already locked | |
1777 | * so we cannot give a lock. | |
1778 | */ | |
79dfdacc | 1779 | failed = iter; |
9f3a0d09 JW |
1780 | mem_cgroup_iter_break(memcg, iter); |
1781 | break; | |
23751be0 JW |
1782 | } else |
1783 | iter->oom_lock = true; | |
7d74b06f | 1784 | } |
867578cb | 1785 | |
fb2a6fc5 JW |
1786 | if (failed) { |
1787 | /* | |
1788 | * OK, we failed to lock the whole subtree so we have | |
1789 | * to clean up what we set up to the failing subtree | |
1790 | */ | |
1791 | for_each_mem_cgroup_tree(iter, memcg) { | |
1792 | if (iter == failed) { | |
1793 | mem_cgroup_iter_break(memcg, iter); | |
1794 | break; | |
1795 | } | |
1796 | iter->oom_lock = false; | |
79dfdacc | 1797 | } |
0056f4e6 JW |
1798 | } else |
1799 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1800 | |
1801 | spin_unlock(&memcg_oom_lock); | |
1802 | ||
1803 | return !failed; | |
a636b327 | 1804 | } |
0b7f569e | 1805 | |
fb2a6fc5 | 1806 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1807 | { |
7d74b06f KH |
1808 | struct mem_cgroup *iter; |
1809 | ||
fb2a6fc5 | 1810 | spin_lock(&memcg_oom_lock); |
5facae4f | 1811 | mutex_release(&memcg_oom_lock_dep_map, _RET_IP_); |
c0ff4b85 | 1812 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1813 | iter->oom_lock = false; |
fb2a6fc5 | 1814 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1815 | } |
1816 | ||
c0ff4b85 | 1817 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1818 | { |
1819 | struct mem_cgroup *iter; | |
1820 | ||
c2b42d3c | 1821 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1822 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1823 | iter->under_oom++; |
1824 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1825 | } |
1826 | ||
c0ff4b85 | 1827 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1828 | { |
1829 | struct mem_cgroup *iter; | |
1830 | ||
867578cb | 1831 | /* |
7a52d4d8 ML |
1832 | * Be careful about under_oom underflows becase a child memcg |
1833 | * could have been added after mem_cgroup_mark_under_oom. | |
867578cb | 1834 | */ |
c2b42d3c | 1835 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1836 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1837 | if (iter->under_oom > 0) |
1838 | iter->under_oom--; | |
1839 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1840 | } |
1841 | ||
867578cb KH |
1842 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1843 | ||
dc98df5a | 1844 | struct oom_wait_info { |
d79154bb | 1845 | struct mem_cgroup *memcg; |
ac6424b9 | 1846 | wait_queue_entry_t wait; |
dc98df5a KH |
1847 | }; |
1848 | ||
ac6424b9 | 1849 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1850 | unsigned mode, int sync, void *arg) |
1851 | { | |
d79154bb HD |
1852 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1853 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1854 | struct oom_wait_info *oom_wait_info; |
1855 | ||
1856 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1857 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1858 | |
2314b42d JW |
1859 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1860 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1861 | return 0; |
dc98df5a KH |
1862 | return autoremove_wake_function(wait, mode, sync, arg); |
1863 | } | |
1864 | ||
c0ff4b85 | 1865 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1866 | { |
c2b42d3c TH |
1867 | /* |
1868 | * For the following lockless ->under_oom test, the only required | |
1869 | * guarantee is that it must see the state asserted by an OOM when | |
1870 | * this function is called as a result of userland actions | |
1871 | * triggered by the notification of the OOM. This is trivially | |
1872 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1873 | * triggering notification. | |
1874 | */ | |
1875 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1876 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1877 | } |
1878 | ||
29ef680a MH |
1879 | enum oom_status { |
1880 | OOM_SUCCESS, | |
1881 | OOM_FAILED, | |
1882 | OOM_ASYNC, | |
1883 | OOM_SKIPPED | |
1884 | }; | |
1885 | ||
1886 | static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 1887 | { |
7056d3a3 MH |
1888 | enum oom_status ret; |
1889 | bool locked; | |
1890 | ||
29ef680a MH |
1891 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
1892 | return OOM_SKIPPED; | |
1893 | ||
7a1adfdd RG |
1894 | memcg_memory_event(memcg, MEMCG_OOM); |
1895 | ||
867578cb | 1896 | /* |
49426420 JW |
1897 | * We are in the middle of the charge context here, so we |
1898 | * don't want to block when potentially sitting on a callstack | |
1899 | * that holds all kinds of filesystem and mm locks. | |
1900 | * | |
29ef680a MH |
1901 | * cgroup1 allows disabling the OOM killer and waiting for outside |
1902 | * handling until the charge can succeed; remember the context and put | |
1903 | * the task to sleep at the end of the page fault when all locks are | |
1904 | * released. | |
49426420 | 1905 | * |
29ef680a MH |
1906 | * On the other hand, in-kernel OOM killer allows for an async victim |
1907 | * memory reclaim (oom_reaper) and that means that we are not solely | |
1908 | * relying on the oom victim to make a forward progress and we can | |
1909 | * invoke the oom killer here. | |
1910 | * | |
1911 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
1912 | * victim and then we have to bail out from the charge path. | |
867578cb | 1913 | */ |
29ef680a MH |
1914 | if (memcg->oom_kill_disable) { |
1915 | if (!current->in_user_fault) | |
1916 | return OOM_SKIPPED; | |
1917 | css_get(&memcg->css); | |
1918 | current->memcg_in_oom = memcg; | |
1919 | current->memcg_oom_gfp_mask = mask; | |
1920 | current->memcg_oom_order = order; | |
1921 | ||
1922 | return OOM_ASYNC; | |
1923 | } | |
1924 | ||
7056d3a3 MH |
1925 | mem_cgroup_mark_under_oom(memcg); |
1926 | ||
1927 | locked = mem_cgroup_oom_trylock(memcg); | |
1928 | ||
1929 | if (locked) | |
1930 | mem_cgroup_oom_notify(memcg); | |
1931 | ||
1932 | mem_cgroup_unmark_under_oom(memcg); | |
29ef680a | 1933 | if (mem_cgroup_out_of_memory(memcg, mask, order)) |
7056d3a3 MH |
1934 | ret = OOM_SUCCESS; |
1935 | else | |
1936 | ret = OOM_FAILED; | |
1937 | ||
1938 | if (locked) | |
1939 | mem_cgroup_oom_unlock(memcg); | |
29ef680a | 1940 | |
7056d3a3 | 1941 | return ret; |
3812c8c8 JW |
1942 | } |
1943 | ||
1944 | /** | |
1945 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1946 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1947 | * |
49426420 JW |
1948 | * This has to be called at the end of a page fault if the memcg OOM |
1949 | * handler was enabled. | |
3812c8c8 | 1950 | * |
49426420 | 1951 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1952 | * sleep on a waitqueue until the userspace task resolves the |
1953 | * situation. Sleeping directly in the charge context with all kinds | |
1954 | * of locks held is not a good idea, instead we remember an OOM state | |
1955 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1956 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1957 | * |
1958 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1959 | * completed, %false otherwise. |
3812c8c8 | 1960 | */ |
49426420 | 1961 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1962 | { |
626ebc41 | 1963 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1964 | struct oom_wait_info owait; |
49426420 | 1965 | bool locked; |
3812c8c8 JW |
1966 | |
1967 | /* OOM is global, do not handle */ | |
3812c8c8 | 1968 | if (!memcg) |
49426420 | 1969 | return false; |
3812c8c8 | 1970 | |
7c5f64f8 | 1971 | if (!handle) |
49426420 | 1972 | goto cleanup; |
3812c8c8 JW |
1973 | |
1974 | owait.memcg = memcg; | |
1975 | owait.wait.flags = 0; | |
1976 | owait.wait.func = memcg_oom_wake_function; | |
1977 | owait.wait.private = current; | |
2055da97 | 1978 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 1979 | |
3812c8c8 | 1980 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1981 | mem_cgroup_mark_under_oom(memcg); |
1982 | ||
1983 | locked = mem_cgroup_oom_trylock(memcg); | |
1984 | ||
1985 | if (locked) | |
1986 | mem_cgroup_oom_notify(memcg); | |
1987 | ||
1988 | if (locked && !memcg->oom_kill_disable) { | |
1989 | mem_cgroup_unmark_under_oom(memcg); | |
1990 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1991 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1992 | current->memcg_oom_order); | |
49426420 | 1993 | } else { |
3812c8c8 | 1994 | schedule(); |
49426420 JW |
1995 | mem_cgroup_unmark_under_oom(memcg); |
1996 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1997 | } | |
1998 | ||
1999 | if (locked) { | |
fb2a6fc5 JW |
2000 | mem_cgroup_oom_unlock(memcg); |
2001 | /* | |
2002 | * There is no guarantee that an OOM-lock contender | |
2003 | * sees the wakeups triggered by the OOM kill | |
2004 | * uncharges. Wake any sleepers explicitely. | |
2005 | */ | |
2006 | memcg_oom_recover(memcg); | |
2007 | } | |
49426420 | 2008 | cleanup: |
626ebc41 | 2009 | current->memcg_in_oom = NULL; |
3812c8c8 | 2010 | css_put(&memcg->css); |
867578cb | 2011 | return true; |
0b7f569e KH |
2012 | } |
2013 | ||
3d8b38eb RG |
2014 | /** |
2015 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
2016 | * @victim: task to be killed by the OOM killer | |
2017 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
2018 | * | |
2019 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
2020 | * by killing all belonging OOM-killable tasks. | |
2021 | * | |
2022 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
2023 | */ | |
2024 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
2025 | struct mem_cgroup *oom_domain) | |
2026 | { | |
2027 | struct mem_cgroup *oom_group = NULL; | |
2028 | struct mem_cgroup *memcg; | |
2029 | ||
2030 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
2031 | return NULL; | |
2032 | ||
2033 | if (!oom_domain) | |
2034 | oom_domain = root_mem_cgroup; | |
2035 | ||
2036 | rcu_read_lock(); | |
2037 | ||
2038 | memcg = mem_cgroup_from_task(victim); | |
2039 | if (memcg == root_mem_cgroup) | |
2040 | goto out; | |
2041 | ||
48fe267c RG |
2042 | /* |
2043 | * If the victim task has been asynchronously moved to a different | |
2044 | * memory cgroup, we might end up killing tasks outside oom_domain. | |
2045 | * In this case it's better to ignore memory.group.oom. | |
2046 | */ | |
2047 | if (unlikely(!mem_cgroup_is_descendant(memcg, oom_domain))) | |
2048 | goto out; | |
2049 | ||
3d8b38eb RG |
2050 | /* |
2051 | * Traverse the memory cgroup hierarchy from the victim task's | |
2052 | * cgroup up to the OOMing cgroup (or root) to find the | |
2053 | * highest-level memory cgroup with oom.group set. | |
2054 | */ | |
2055 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
2056 | if (memcg->oom_group) | |
2057 | oom_group = memcg; | |
2058 | ||
2059 | if (memcg == oom_domain) | |
2060 | break; | |
2061 | } | |
2062 | ||
2063 | if (oom_group) | |
2064 | css_get(&oom_group->css); | |
2065 | out: | |
2066 | rcu_read_unlock(); | |
2067 | ||
2068 | return oom_group; | |
2069 | } | |
2070 | ||
2071 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
2072 | { | |
2073 | pr_info("Tasks in "); | |
2074 | pr_cont_cgroup_path(memcg->css.cgroup); | |
2075 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
2076 | } | |
2077 | ||
d7365e78 | 2078 | /** |
81f8c3a4 JW |
2079 | * lock_page_memcg - lock a page->mem_cgroup binding |
2080 | * @page: the page | |
32047e2a | 2081 | * |
81f8c3a4 | 2082 | * This function protects unlocked LRU pages from being moved to |
739f79fc JW |
2083 | * another cgroup. |
2084 | * | |
2085 | * It ensures lifetime of the returned memcg. Caller is responsible | |
2086 | * for the lifetime of the page; __unlock_page_memcg() is available | |
2087 | * when @page might get freed inside the locked section. | |
d69b042f | 2088 | */ |
739f79fc | 2089 | struct mem_cgroup *lock_page_memcg(struct page *page) |
89c06bd5 | 2090 | { |
9da7b521 | 2091 | struct page *head = compound_head(page); /* rmap on tail pages */ |
89c06bd5 | 2092 | struct mem_cgroup *memcg; |
6de22619 | 2093 | unsigned long flags; |
89c06bd5 | 2094 | |
6de22619 JW |
2095 | /* |
2096 | * The RCU lock is held throughout the transaction. The fast | |
2097 | * path can get away without acquiring the memcg->move_lock | |
2098 | * because page moving starts with an RCU grace period. | |
739f79fc JW |
2099 | * |
2100 | * The RCU lock also protects the memcg from being freed when | |
2101 | * the page state that is going to change is the only thing | |
2102 | * preventing the page itself from being freed. E.g. writeback | |
2103 | * doesn't hold a page reference and relies on PG_writeback to | |
2104 | * keep off truncation, migration and so forth. | |
2105 | */ | |
d7365e78 JW |
2106 | rcu_read_lock(); |
2107 | ||
2108 | if (mem_cgroup_disabled()) | |
739f79fc | 2109 | return NULL; |
89c06bd5 | 2110 | again: |
9da7b521 | 2111 | memcg = head->mem_cgroup; |
29833315 | 2112 | if (unlikely(!memcg)) |
739f79fc | 2113 | return NULL; |
d7365e78 | 2114 | |
bdcbb659 | 2115 | if (atomic_read(&memcg->moving_account) <= 0) |
739f79fc | 2116 | return memcg; |
89c06bd5 | 2117 | |
6de22619 | 2118 | spin_lock_irqsave(&memcg->move_lock, flags); |
9da7b521 | 2119 | if (memcg != head->mem_cgroup) { |
6de22619 | 2120 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
2121 | goto again; |
2122 | } | |
6de22619 JW |
2123 | |
2124 | /* | |
2125 | * When charge migration first begins, we can have locked and | |
2126 | * unlocked page stat updates happening concurrently. Track | |
81f8c3a4 | 2127 | * the task who has the lock for unlock_page_memcg(). |
6de22619 JW |
2128 | */ |
2129 | memcg->move_lock_task = current; | |
2130 | memcg->move_lock_flags = flags; | |
d7365e78 | 2131 | |
739f79fc | 2132 | return memcg; |
89c06bd5 | 2133 | } |
81f8c3a4 | 2134 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 2135 | |
d7365e78 | 2136 | /** |
739f79fc JW |
2137 | * __unlock_page_memcg - unlock and unpin a memcg |
2138 | * @memcg: the memcg | |
2139 | * | |
2140 | * Unlock and unpin a memcg returned by lock_page_memcg(). | |
d7365e78 | 2141 | */ |
739f79fc | 2142 | void __unlock_page_memcg(struct mem_cgroup *memcg) |
89c06bd5 | 2143 | { |
6de22619 JW |
2144 | if (memcg && memcg->move_lock_task == current) { |
2145 | unsigned long flags = memcg->move_lock_flags; | |
2146 | ||
2147 | memcg->move_lock_task = NULL; | |
2148 | memcg->move_lock_flags = 0; | |
2149 | ||
2150 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2151 | } | |
89c06bd5 | 2152 | |
d7365e78 | 2153 | rcu_read_unlock(); |
89c06bd5 | 2154 | } |
739f79fc JW |
2155 | |
2156 | /** | |
2157 | * unlock_page_memcg - unlock a page->mem_cgroup binding | |
2158 | * @page: the page | |
2159 | */ | |
2160 | void unlock_page_memcg(struct page *page) | |
2161 | { | |
9da7b521 JW |
2162 | struct page *head = compound_head(page); |
2163 | ||
2164 | __unlock_page_memcg(head->mem_cgroup); | |
739f79fc | 2165 | } |
81f8c3a4 | 2166 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 2167 | |
cdec2e42 KH |
2168 | struct memcg_stock_pcp { |
2169 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2170 | unsigned int nr_pages; |
bf4f0599 RG |
2171 | |
2172 | #ifdef CONFIG_MEMCG_KMEM | |
2173 | struct obj_cgroup *cached_objcg; | |
2174 | unsigned int nr_bytes; | |
2175 | #endif | |
2176 | ||
cdec2e42 | 2177 | struct work_struct work; |
26fe6168 | 2178 | unsigned long flags; |
a0db00fc | 2179 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2180 | }; |
2181 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2182 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2183 | |
bf4f0599 RG |
2184 | #ifdef CONFIG_MEMCG_KMEM |
2185 | static void drain_obj_stock(struct memcg_stock_pcp *stock); | |
2186 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2187 | struct mem_cgroup *root_memcg); | |
2188 | ||
2189 | #else | |
2190 | static inline void drain_obj_stock(struct memcg_stock_pcp *stock) | |
2191 | { | |
2192 | } | |
2193 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2194 | struct mem_cgroup *root_memcg) | |
2195 | { | |
2196 | return false; | |
2197 | } | |
2198 | #endif | |
2199 | ||
a0956d54 SS |
2200 | /** |
2201 | * consume_stock: Try to consume stocked charge on this cpu. | |
2202 | * @memcg: memcg to consume from. | |
2203 | * @nr_pages: how many pages to charge. | |
2204 | * | |
2205 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2206 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2207 | * service an allocation will refill the stock. | |
2208 | * | |
2209 | * returns true if successful, false otherwise. | |
cdec2e42 | 2210 | */ |
a0956d54 | 2211 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2212 | { |
2213 | struct memcg_stock_pcp *stock; | |
db2ba40c | 2214 | unsigned long flags; |
3e32cb2e | 2215 | bool ret = false; |
cdec2e42 | 2216 | |
a983b5eb | 2217 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 2218 | return ret; |
a0956d54 | 2219 | |
db2ba40c JW |
2220 | local_irq_save(flags); |
2221 | ||
2222 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 2223 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2224 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2225 | ret = true; |
2226 | } | |
db2ba40c JW |
2227 | |
2228 | local_irq_restore(flags); | |
2229 | ||
cdec2e42 KH |
2230 | return ret; |
2231 | } | |
2232 | ||
2233 | /* | |
3e32cb2e | 2234 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2235 | */ |
2236 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2237 | { | |
2238 | struct mem_cgroup *old = stock->cached; | |
2239 | ||
1a3e1f40 JW |
2240 | if (!old) |
2241 | return; | |
2242 | ||
11c9ea4e | 2243 | if (stock->nr_pages) { |
3e32cb2e | 2244 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 2245 | if (do_memsw_account()) |
3e32cb2e | 2246 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
11c9ea4e | 2247 | stock->nr_pages = 0; |
cdec2e42 | 2248 | } |
1a3e1f40 JW |
2249 | |
2250 | css_put(&old->css); | |
cdec2e42 | 2251 | stock->cached = NULL; |
cdec2e42 KH |
2252 | } |
2253 | ||
cdec2e42 KH |
2254 | static void drain_local_stock(struct work_struct *dummy) |
2255 | { | |
db2ba40c JW |
2256 | struct memcg_stock_pcp *stock; |
2257 | unsigned long flags; | |
2258 | ||
72f0184c MH |
2259 | /* |
2260 | * The only protection from memory hotplug vs. drain_stock races is | |
2261 | * that we always operate on local CPU stock here with IRQ disabled | |
2262 | */ | |
db2ba40c JW |
2263 | local_irq_save(flags); |
2264 | ||
2265 | stock = this_cpu_ptr(&memcg_stock); | |
bf4f0599 | 2266 | drain_obj_stock(stock); |
cdec2e42 | 2267 | drain_stock(stock); |
26fe6168 | 2268 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c JW |
2269 | |
2270 | local_irq_restore(flags); | |
cdec2e42 KH |
2271 | } |
2272 | ||
2273 | /* | |
3e32cb2e | 2274 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2275 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2276 | */ |
c0ff4b85 | 2277 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2278 | { |
db2ba40c JW |
2279 | struct memcg_stock_pcp *stock; |
2280 | unsigned long flags; | |
2281 | ||
2282 | local_irq_save(flags); | |
cdec2e42 | 2283 | |
db2ba40c | 2284 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 2285 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2286 | drain_stock(stock); |
1a3e1f40 | 2287 | css_get(&memcg->css); |
c0ff4b85 | 2288 | stock->cached = memcg; |
cdec2e42 | 2289 | } |
11c9ea4e | 2290 | stock->nr_pages += nr_pages; |
db2ba40c | 2291 | |
a983b5eb | 2292 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 RG |
2293 | drain_stock(stock); |
2294 | ||
db2ba40c | 2295 | local_irq_restore(flags); |
cdec2e42 KH |
2296 | } |
2297 | ||
2298 | /* | |
c0ff4b85 | 2299 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2300 | * of the hierarchy under it. |
cdec2e42 | 2301 | */ |
6d3d6aa2 | 2302 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2303 | { |
26fe6168 | 2304 | int cpu, curcpu; |
d38144b7 | 2305 | |
6d3d6aa2 JW |
2306 | /* If someone's already draining, avoid adding running more workers. */ |
2307 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2308 | return; | |
72f0184c MH |
2309 | /* |
2310 | * Notify other cpus that system-wide "drain" is running | |
2311 | * We do not care about races with the cpu hotplug because cpu down | |
2312 | * as well as workers from this path always operate on the local | |
2313 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2314 | */ | |
5af12d0e | 2315 | curcpu = get_cpu(); |
cdec2e42 KH |
2316 | for_each_online_cpu(cpu) { |
2317 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2318 | struct mem_cgroup *memcg; |
e1a366be | 2319 | bool flush = false; |
26fe6168 | 2320 | |
e1a366be | 2321 | rcu_read_lock(); |
c0ff4b85 | 2322 | memcg = stock->cached; |
e1a366be RG |
2323 | if (memcg && stock->nr_pages && |
2324 | mem_cgroup_is_descendant(memcg, root_memcg)) | |
2325 | flush = true; | |
bf4f0599 RG |
2326 | if (obj_stock_flush_required(stock, root_memcg)) |
2327 | flush = true; | |
e1a366be RG |
2328 | rcu_read_unlock(); |
2329 | ||
2330 | if (flush && | |
2331 | !test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { | |
d1a05b69 MH |
2332 | if (cpu == curcpu) |
2333 | drain_local_stock(&stock->work); | |
2334 | else | |
2335 | schedule_work_on(cpu, &stock->work); | |
2336 | } | |
cdec2e42 | 2337 | } |
5af12d0e | 2338 | put_cpu(); |
9f50fad6 | 2339 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2340 | } |
2341 | ||
308167fc | 2342 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
cdec2e42 | 2343 | { |
cdec2e42 | 2344 | struct memcg_stock_pcp *stock; |
42a30035 | 2345 | struct mem_cgroup *memcg, *mi; |
cdec2e42 | 2346 | |
cdec2e42 KH |
2347 | stock = &per_cpu(memcg_stock, cpu); |
2348 | drain_stock(stock); | |
a983b5eb JW |
2349 | |
2350 | for_each_mem_cgroup(memcg) { | |
2351 | int i; | |
2352 | ||
2353 | for (i = 0; i < MEMCG_NR_STAT; i++) { | |
2354 | int nid; | |
2355 | long x; | |
2356 | ||
871789d4 | 2357 | x = this_cpu_xchg(memcg->vmstats_percpu->stat[i], 0); |
815744d7 | 2358 | if (x) |
42a30035 JW |
2359 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2360 | atomic_long_add(x, &memcg->vmstats[i]); | |
a983b5eb JW |
2361 | |
2362 | if (i >= NR_VM_NODE_STAT_ITEMS) | |
2363 | continue; | |
2364 | ||
2365 | for_each_node(nid) { | |
2366 | struct mem_cgroup_per_node *pn; | |
2367 | ||
2368 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
2369 | x = this_cpu_xchg(pn->lruvec_stat_cpu->count[i], 0); | |
815744d7 | 2370 | if (x) |
42a30035 JW |
2371 | do { |
2372 | atomic_long_add(x, &pn->lruvec_stat[i]); | |
2373 | } while ((pn = parent_nodeinfo(pn, nid))); | |
a983b5eb JW |
2374 | } |
2375 | } | |
2376 | ||
e27be240 | 2377 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { |
a983b5eb JW |
2378 | long x; |
2379 | ||
871789d4 | 2380 | x = this_cpu_xchg(memcg->vmstats_percpu->events[i], 0); |
815744d7 | 2381 | if (x) |
42a30035 JW |
2382 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2383 | atomic_long_add(x, &memcg->vmevents[i]); | |
a983b5eb JW |
2384 | } |
2385 | } | |
2386 | ||
308167fc | 2387 | return 0; |
cdec2e42 KH |
2388 | } |
2389 | ||
b3ff9291 CD |
2390 | static unsigned long reclaim_high(struct mem_cgroup *memcg, |
2391 | unsigned int nr_pages, | |
2392 | gfp_t gfp_mask) | |
f7e1cb6e | 2393 | { |
b3ff9291 CD |
2394 | unsigned long nr_reclaimed = 0; |
2395 | ||
f7e1cb6e | 2396 | do { |
e22c6ed9 JW |
2397 | unsigned long pflags; |
2398 | ||
d1663a90 JK |
2399 | if (page_counter_read(&memcg->memory) <= |
2400 | READ_ONCE(memcg->memory.high)) | |
f7e1cb6e | 2401 | continue; |
e22c6ed9 | 2402 | |
e27be240 | 2403 | memcg_memory_event(memcg, MEMCG_HIGH); |
e22c6ed9 JW |
2404 | |
2405 | psi_memstall_enter(&pflags); | |
b3ff9291 CD |
2406 | nr_reclaimed += try_to_free_mem_cgroup_pages(memcg, nr_pages, |
2407 | gfp_mask, true); | |
e22c6ed9 | 2408 | psi_memstall_leave(&pflags); |
4bf17307 CD |
2409 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2410 | !mem_cgroup_is_root(memcg)); | |
b3ff9291 CD |
2411 | |
2412 | return nr_reclaimed; | |
f7e1cb6e JW |
2413 | } |
2414 | ||
2415 | static void high_work_func(struct work_struct *work) | |
2416 | { | |
2417 | struct mem_cgroup *memcg; | |
2418 | ||
2419 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2420 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2421 | } |
2422 | ||
0e4b01df CD |
2423 | /* |
2424 | * Clamp the maximum sleep time per allocation batch to 2 seconds. This is | |
2425 | * enough to still cause a significant slowdown in most cases, while still | |
2426 | * allowing diagnostics and tracing to proceed without becoming stuck. | |
2427 | */ | |
2428 | #define MEMCG_MAX_HIGH_DELAY_JIFFIES (2UL*HZ) | |
2429 | ||
2430 | /* | |
2431 | * When calculating the delay, we use these either side of the exponentiation to | |
2432 | * maintain precision and scale to a reasonable number of jiffies (see the table | |
2433 | * below. | |
2434 | * | |
2435 | * - MEMCG_DELAY_PRECISION_SHIFT: Extra precision bits while translating the | |
2436 | * overage ratio to a delay. | |
ac5ddd0f | 2437 | * - MEMCG_DELAY_SCALING_SHIFT: The number of bits to scale down the |
0e4b01df CD |
2438 | * proposed penalty in order to reduce to a reasonable number of jiffies, and |
2439 | * to produce a reasonable delay curve. | |
2440 | * | |
2441 | * MEMCG_DELAY_SCALING_SHIFT just happens to be a number that produces a | |
2442 | * reasonable delay curve compared to precision-adjusted overage, not | |
2443 | * penalising heavily at first, but still making sure that growth beyond the | |
2444 | * limit penalises misbehaviour cgroups by slowing them down exponentially. For | |
2445 | * example, with a high of 100 megabytes: | |
2446 | * | |
2447 | * +-------+------------------------+ | |
2448 | * | usage | time to allocate in ms | | |
2449 | * +-------+------------------------+ | |
2450 | * | 100M | 0 | | |
2451 | * | 101M | 6 | | |
2452 | * | 102M | 25 | | |
2453 | * | 103M | 57 | | |
2454 | * | 104M | 102 | | |
2455 | * | 105M | 159 | | |
2456 | * | 106M | 230 | | |
2457 | * | 107M | 313 | | |
2458 | * | 108M | 409 | | |
2459 | * | 109M | 518 | | |
2460 | * | 110M | 639 | | |
2461 | * | 111M | 774 | | |
2462 | * | 112M | 921 | | |
2463 | * | 113M | 1081 | | |
2464 | * | 114M | 1254 | | |
2465 | * | 115M | 1439 | | |
2466 | * | 116M | 1638 | | |
2467 | * | 117M | 1849 | | |
2468 | * | 118M | 2000 | | |
2469 | * | 119M | 2000 | | |
2470 | * | 120M | 2000 | | |
2471 | * +-------+------------------------+ | |
2472 | */ | |
2473 | #define MEMCG_DELAY_PRECISION_SHIFT 20 | |
2474 | #define MEMCG_DELAY_SCALING_SHIFT 14 | |
2475 | ||
8a5dbc65 | 2476 | static u64 calculate_overage(unsigned long usage, unsigned long high) |
b23afb93 | 2477 | { |
8a5dbc65 | 2478 | u64 overage; |
b23afb93 | 2479 | |
8a5dbc65 JK |
2480 | if (usage <= high) |
2481 | return 0; | |
e26733e0 | 2482 | |
8a5dbc65 JK |
2483 | /* |
2484 | * Prevent division by 0 in overage calculation by acting as if | |
2485 | * it was a threshold of 1 page | |
2486 | */ | |
2487 | high = max(high, 1UL); | |
9b8b1754 | 2488 | |
8a5dbc65 JK |
2489 | overage = usage - high; |
2490 | overage <<= MEMCG_DELAY_PRECISION_SHIFT; | |
2491 | return div64_u64(overage, high); | |
2492 | } | |
e26733e0 | 2493 | |
8a5dbc65 JK |
2494 | static u64 mem_find_max_overage(struct mem_cgroup *memcg) |
2495 | { | |
2496 | u64 overage, max_overage = 0; | |
e26733e0 | 2497 | |
8a5dbc65 JK |
2498 | do { |
2499 | overage = calculate_overage(page_counter_read(&memcg->memory), | |
d1663a90 | 2500 | READ_ONCE(memcg->memory.high)); |
8a5dbc65 | 2501 | max_overage = max(overage, max_overage); |
e26733e0 CD |
2502 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2503 | !mem_cgroup_is_root(memcg)); | |
2504 | ||
8a5dbc65 JK |
2505 | return max_overage; |
2506 | } | |
2507 | ||
4b82ab4f JK |
2508 | static u64 swap_find_max_overage(struct mem_cgroup *memcg) |
2509 | { | |
2510 | u64 overage, max_overage = 0; | |
2511 | ||
2512 | do { | |
2513 | overage = calculate_overage(page_counter_read(&memcg->swap), | |
2514 | READ_ONCE(memcg->swap.high)); | |
2515 | if (overage) | |
2516 | memcg_memory_event(memcg, MEMCG_SWAP_HIGH); | |
2517 | max_overage = max(overage, max_overage); | |
2518 | } while ((memcg = parent_mem_cgroup(memcg)) && | |
2519 | !mem_cgroup_is_root(memcg)); | |
2520 | ||
2521 | return max_overage; | |
2522 | } | |
2523 | ||
8a5dbc65 JK |
2524 | /* |
2525 | * Get the number of jiffies that we should penalise a mischievous cgroup which | |
2526 | * is exceeding its memory.high by checking both it and its ancestors. | |
2527 | */ | |
2528 | static unsigned long calculate_high_delay(struct mem_cgroup *memcg, | |
2529 | unsigned int nr_pages, | |
2530 | u64 max_overage) | |
2531 | { | |
2532 | unsigned long penalty_jiffies; | |
2533 | ||
e26733e0 CD |
2534 | if (!max_overage) |
2535 | return 0; | |
0e4b01df CD |
2536 | |
2537 | /* | |
0e4b01df CD |
2538 | * We use overage compared to memory.high to calculate the number of |
2539 | * jiffies to sleep (penalty_jiffies). Ideally this value should be | |
2540 | * fairly lenient on small overages, and increasingly harsh when the | |
2541 | * memcg in question makes it clear that it has no intention of stopping | |
2542 | * its crazy behaviour, so we exponentially increase the delay based on | |
2543 | * overage amount. | |
2544 | */ | |
e26733e0 CD |
2545 | penalty_jiffies = max_overage * max_overage * HZ; |
2546 | penalty_jiffies >>= MEMCG_DELAY_PRECISION_SHIFT; | |
2547 | penalty_jiffies >>= MEMCG_DELAY_SCALING_SHIFT; | |
0e4b01df CD |
2548 | |
2549 | /* | |
2550 | * Factor in the task's own contribution to the overage, such that four | |
2551 | * N-sized allocations are throttled approximately the same as one | |
2552 | * 4N-sized allocation. | |
2553 | * | |
2554 | * MEMCG_CHARGE_BATCH pages is nominal, so work out how much smaller or | |
2555 | * larger the current charge patch is than that. | |
2556 | */ | |
ff144e69 | 2557 | return penalty_jiffies * nr_pages / MEMCG_CHARGE_BATCH; |
e26733e0 CD |
2558 | } |
2559 | ||
2560 | /* | |
2561 | * Scheduled by try_charge() to be executed from the userland return path | |
2562 | * and reclaims memory over the high limit. | |
2563 | */ | |
2564 | void mem_cgroup_handle_over_high(void) | |
2565 | { | |
2566 | unsigned long penalty_jiffies; | |
2567 | unsigned long pflags; | |
b3ff9291 | 2568 | unsigned long nr_reclaimed; |
e26733e0 | 2569 | unsigned int nr_pages = current->memcg_nr_pages_over_high; |
d977aa93 | 2570 | int nr_retries = MAX_RECLAIM_RETRIES; |
e26733e0 | 2571 | struct mem_cgroup *memcg; |
b3ff9291 | 2572 | bool in_retry = false; |
e26733e0 CD |
2573 | |
2574 | if (likely(!nr_pages)) | |
2575 | return; | |
2576 | ||
2577 | memcg = get_mem_cgroup_from_mm(current->mm); | |
e26733e0 CD |
2578 | current->memcg_nr_pages_over_high = 0; |
2579 | ||
b3ff9291 CD |
2580 | retry_reclaim: |
2581 | /* | |
2582 | * The allocating task should reclaim at least the batch size, but for | |
2583 | * subsequent retries we only want to do what's necessary to prevent oom | |
2584 | * or breaching resource isolation. | |
2585 | * | |
2586 | * This is distinct from memory.max or page allocator behaviour because | |
2587 | * memory.high is currently batched, whereas memory.max and the page | |
2588 | * allocator run every time an allocation is made. | |
2589 | */ | |
2590 | nr_reclaimed = reclaim_high(memcg, | |
2591 | in_retry ? SWAP_CLUSTER_MAX : nr_pages, | |
2592 | GFP_KERNEL); | |
2593 | ||
e26733e0 CD |
2594 | /* |
2595 | * memory.high is breached and reclaim is unable to keep up. Throttle | |
2596 | * allocators proactively to slow down excessive growth. | |
2597 | */ | |
8a5dbc65 JK |
2598 | penalty_jiffies = calculate_high_delay(memcg, nr_pages, |
2599 | mem_find_max_overage(memcg)); | |
0e4b01df | 2600 | |
4b82ab4f JK |
2601 | penalty_jiffies += calculate_high_delay(memcg, nr_pages, |
2602 | swap_find_max_overage(memcg)); | |
2603 | ||
ff144e69 JK |
2604 | /* |
2605 | * Clamp the max delay per usermode return so as to still keep the | |
2606 | * application moving forwards and also permit diagnostics, albeit | |
2607 | * extremely slowly. | |
2608 | */ | |
2609 | penalty_jiffies = min(penalty_jiffies, MEMCG_MAX_HIGH_DELAY_JIFFIES); | |
2610 | ||
0e4b01df CD |
2611 | /* |
2612 | * Don't sleep if the amount of jiffies this memcg owes us is so low | |
2613 | * that it's not even worth doing, in an attempt to be nice to those who | |
2614 | * go only a small amount over their memory.high value and maybe haven't | |
2615 | * been aggressively reclaimed enough yet. | |
2616 | */ | |
2617 | if (penalty_jiffies <= HZ / 100) | |
2618 | goto out; | |
2619 | ||
b3ff9291 CD |
2620 | /* |
2621 | * If reclaim is making forward progress but we're still over | |
2622 | * memory.high, we want to encourage that rather than doing allocator | |
2623 | * throttling. | |
2624 | */ | |
2625 | if (nr_reclaimed || nr_retries--) { | |
2626 | in_retry = true; | |
2627 | goto retry_reclaim; | |
2628 | } | |
2629 | ||
0e4b01df CD |
2630 | /* |
2631 | * If we exit early, we're guaranteed to die (since | |
2632 | * schedule_timeout_killable sets TASK_KILLABLE). This means we don't | |
2633 | * need to account for any ill-begotten jiffies to pay them off later. | |
2634 | */ | |
2635 | psi_memstall_enter(&pflags); | |
2636 | schedule_timeout_killable(penalty_jiffies); | |
2637 | psi_memstall_leave(&pflags); | |
2638 | ||
2639 | out: | |
2640 | css_put(&memcg->css); | |
b23afb93 TH |
2641 | } |
2642 | ||
00501b53 JW |
2643 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2644 | unsigned int nr_pages) | |
8a9f3ccd | 2645 | { |
a983b5eb | 2646 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
d977aa93 | 2647 | int nr_retries = MAX_RECLAIM_RETRIES; |
6539cc05 | 2648 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2649 | struct page_counter *counter; |
e22c6ed9 | 2650 | enum oom_status oom_status; |
6539cc05 | 2651 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2652 | bool may_swap = true; |
2653 | bool drained = false; | |
e22c6ed9 | 2654 | unsigned long pflags; |
a636b327 | 2655 | |
ce00a967 | 2656 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 2657 | return 0; |
6539cc05 | 2658 | retry: |
b6b6cc72 | 2659 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2660 | return 0; |
8a9f3ccd | 2661 | |
7941d214 | 2662 | if (!do_memsw_account() || |
6071ca52 JW |
2663 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2664 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2665 | goto done_restock; |
7941d214 | 2666 | if (do_memsw_account()) |
3e32cb2e JW |
2667 | page_counter_uncharge(&memcg->memsw, batch); |
2668 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2669 | } else { |
3e32cb2e | 2670 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2671 | may_swap = false; |
3fbe7244 | 2672 | } |
7a81b88c | 2673 | |
6539cc05 JW |
2674 | if (batch > nr_pages) { |
2675 | batch = nr_pages; | |
2676 | goto retry; | |
2677 | } | |
6d61ef40 | 2678 | |
869712fd JW |
2679 | /* |
2680 | * Memcg doesn't have a dedicated reserve for atomic | |
2681 | * allocations. But like the global atomic pool, we need to | |
2682 | * put the burden of reclaim on regular allocation requests | |
2683 | * and let these go through as privileged allocations. | |
2684 | */ | |
2685 | if (gfp_mask & __GFP_ATOMIC) | |
2686 | goto force; | |
2687 | ||
06b078fc JW |
2688 | /* |
2689 | * Unlike in global OOM situations, memcg is not in a physical | |
2690 | * memory shortage. Allow dying and OOM-killed tasks to | |
2691 | * bypass the last charges so that they can exit quickly and | |
2692 | * free their memory. | |
2693 | */ | |
7775face | 2694 | if (unlikely(should_force_charge())) |
10d53c74 | 2695 | goto force; |
06b078fc | 2696 | |
89a28483 JW |
2697 | /* |
2698 | * Prevent unbounded recursion when reclaim operations need to | |
2699 | * allocate memory. This might exceed the limits temporarily, | |
2700 | * but we prefer facilitating memory reclaim and getting back | |
2701 | * under the limit over triggering OOM kills in these cases. | |
2702 | */ | |
2703 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2704 | goto force; | |
2705 | ||
06b078fc JW |
2706 | if (unlikely(task_in_memcg_oom(current))) |
2707 | goto nomem; | |
2708 | ||
d0164adc | 2709 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2710 | goto nomem; |
4b534334 | 2711 | |
e27be240 | 2712 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
241994ed | 2713 | |
e22c6ed9 | 2714 | psi_memstall_enter(&pflags); |
b70a2a21 JW |
2715 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2716 | gfp_mask, may_swap); | |
e22c6ed9 | 2717 | psi_memstall_leave(&pflags); |
6539cc05 | 2718 | |
61e02c74 | 2719 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2720 | goto retry; |
28c34c29 | 2721 | |
b70a2a21 | 2722 | if (!drained) { |
6d3d6aa2 | 2723 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2724 | drained = true; |
2725 | goto retry; | |
2726 | } | |
2727 | ||
28c34c29 JW |
2728 | if (gfp_mask & __GFP_NORETRY) |
2729 | goto nomem; | |
6539cc05 JW |
2730 | /* |
2731 | * Even though the limit is exceeded at this point, reclaim | |
2732 | * may have been able to free some pages. Retry the charge | |
2733 | * before killing the task. | |
2734 | * | |
2735 | * Only for regular pages, though: huge pages are rather | |
2736 | * unlikely to succeed so close to the limit, and we fall back | |
2737 | * to regular pages anyway in case of failure. | |
2738 | */ | |
61e02c74 | 2739 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2740 | goto retry; |
2741 | /* | |
2742 | * At task move, charge accounts can be doubly counted. So, it's | |
2743 | * better to wait until the end of task_move if something is going on. | |
2744 | */ | |
2745 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2746 | goto retry; | |
2747 | ||
9b130619 JW |
2748 | if (nr_retries--) |
2749 | goto retry; | |
2750 | ||
38d38493 | 2751 | if (gfp_mask & __GFP_RETRY_MAYFAIL) |
29ef680a MH |
2752 | goto nomem; |
2753 | ||
06b078fc | 2754 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2755 | goto force; |
06b078fc | 2756 | |
6539cc05 | 2757 | if (fatal_signal_pending(current)) |
10d53c74 | 2758 | goto force; |
6539cc05 | 2759 | |
29ef680a MH |
2760 | /* |
2761 | * keep retrying as long as the memcg oom killer is able to make | |
2762 | * a forward progress or bypass the charge if the oom killer | |
2763 | * couldn't make any progress. | |
2764 | */ | |
2765 | oom_status = mem_cgroup_oom(mem_over_limit, gfp_mask, | |
3608de07 | 2766 | get_order(nr_pages * PAGE_SIZE)); |
29ef680a MH |
2767 | switch (oom_status) { |
2768 | case OOM_SUCCESS: | |
d977aa93 | 2769 | nr_retries = MAX_RECLAIM_RETRIES; |
29ef680a MH |
2770 | goto retry; |
2771 | case OOM_FAILED: | |
2772 | goto force; | |
2773 | default: | |
2774 | goto nomem; | |
2775 | } | |
7a81b88c | 2776 | nomem: |
6d1fdc48 | 2777 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2778 | return -ENOMEM; |
10d53c74 TH |
2779 | force: |
2780 | /* | |
2781 | * The allocation either can't fail or will lead to more memory | |
2782 | * being freed very soon. Allow memory usage go over the limit | |
2783 | * temporarily by force charging it. | |
2784 | */ | |
2785 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2786 | if (do_memsw_account()) |
10d53c74 | 2787 | page_counter_charge(&memcg->memsw, nr_pages); |
10d53c74 TH |
2788 | |
2789 | return 0; | |
6539cc05 JW |
2790 | |
2791 | done_restock: | |
2792 | if (batch > nr_pages) | |
2793 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2794 | |
241994ed | 2795 | /* |
b23afb93 TH |
2796 | * If the hierarchy is above the normal consumption range, schedule |
2797 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2798 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2799 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2800 | * not recorded as it most likely matches current's and won't | |
2801 | * change in the meantime. As high limit is checked again before | |
2802 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2803 | */ |
2804 | do { | |
4b82ab4f JK |
2805 | bool mem_high, swap_high; |
2806 | ||
2807 | mem_high = page_counter_read(&memcg->memory) > | |
2808 | READ_ONCE(memcg->memory.high); | |
2809 | swap_high = page_counter_read(&memcg->swap) > | |
2810 | READ_ONCE(memcg->swap.high); | |
2811 | ||
2812 | /* Don't bother a random interrupted task */ | |
2813 | if (in_interrupt()) { | |
2814 | if (mem_high) { | |
f7e1cb6e JW |
2815 | schedule_work(&memcg->high_work); |
2816 | break; | |
2817 | } | |
4b82ab4f JK |
2818 | continue; |
2819 | } | |
2820 | ||
2821 | if (mem_high || swap_high) { | |
2822 | /* | |
2823 | * The allocating tasks in this cgroup will need to do | |
2824 | * reclaim or be throttled to prevent further growth | |
2825 | * of the memory or swap footprints. | |
2826 | * | |
2827 | * Target some best-effort fairness between the tasks, | |
2828 | * and distribute reclaim work and delay penalties | |
2829 | * based on how much each task is actually allocating. | |
2830 | */ | |
9516a18a | 2831 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2832 | set_notify_resume(current); |
2833 | break; | |
2834 | } | |
241994ed | 2835 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2836 | |
2837 | return 0; | |
7a81b88c | 2838 | } |
8a9f3ccd | 2839 | |
f0e45fb4 | 2840 | #if defined(CONFIG_MEMCG_KMEM) || defined(CONFIG_MMU) |
00501b53 | 2841 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2842 | { |
ce00a967 JW |
2843 | if (mem_cgroup_is_root(memcg)) |
2844 | return; | |
2845 | ||
3e32cb2e | 2846 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2847 | if (do_memsw_account()) |
3e32cb2e | 2848 | page_counter_uncharge(&memcg->memsw, nr_pages); |
d01dd17f | 2849 | } |
f0e45fb4 | 2850 | #endif |
d01dd17f | 2851 | |
d9eb1ea2 | 2852 | static void commit_charge(struct page *page, struct mem_cgroup *memcg) |
0a31bc97 | 2853 | { |
1306a85a | 2854 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
0a31bc97 | 2855 | /* |
a0b5b414 | 2856 | * Any of the following ensures page->mem_cgroup stability: |
0a31bc97 | 2857 | * |
a0b5b414 JW |
2858 | * - the page lock |
2859 | * - LRU isolation | |
2860 | * - lock_page_memcg() | |
2861 | * - exclusive reference | |
0a31bc97 | 2862 | */ |
1306a85a | 2863 | page->mem_cgroup = memcg; |
7a81b88c | 2864 | } |
66e1707b | 2865 | |
84c07d11 | 2866 | #ifdef CONFIG_MEMCG_KMEM |
10befea9 RG |
2867 | int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s, |
2868 | gfp_t gfp) | |
2869 | { | |
2870 | unsigned int objects = objs_per_slab_page(s, page); | |
2871 | void *vec; | |
2872 | ||
2873 | vec = kcalloc_node(objects, sizeof(struct obj_cgroup *), gfp, | |
2874 | page_to_nid(page)); | |
2875 | if (!vec) | |
2876 | return -ENOMEM; | |
2877 | ||
2878 | if (cmpxchg(&page->obj_cgroups, NULL, | |
2879 | (struct obj_cgroup **) ((unsigned long)vec | 0x1UL))) | |
2880 | kfree(vec); | |
2881 | else | |
2882 | kmemleak_not_leak(vec); | |
2883 | ||
2884 | return 0; | |
2885 | } | |
2886 | ||
8380ce47 RG |
2887 | /* |
2888 | * Returns a pointer to the memory cgroup to which the kernel object is charged. | |
2889 | * | |
2890 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), | |
2891 | * cgroup_mutex, etc. | |
2892 | */ | |
2893 | struct mem_cgroup *mem_cgroup_from_obj(void *p) | |
2894 | { | |
2895 | struct page *page; | |
2896 | ||
2897 | if (mem_cgroup_disabled()) | |
2898 | return NULL; | |
2899 | ||
2900 | page = virt_to_head_page(p); | |
2901 | ||
19b629c9 RG |
2902 | /* |
2903 | * If page->mem_cgroup is set, it's either a simple mem_cgroup pointer | |
2904 | * or a pointer to obj_cgroup vector. In the latter case the lowest | |
2905 | * bit of the pointer is set. | |
2906 | * The page->mem_cgroup pointer can be asynchronously changed | |
2907 | * from NULL to (obj_cgroup_vec | 0x1UL), but can't be changed | |
2908 | * from a valid memcg pointer to objcg vector or back. | |
2909 | */ | |
2910 | if (!page->mem_cgroup) | |
2911 | return NULL; | |
2912 | ||
8380ce47 | 2913 | /* |
9855609b RG |
2914 | * Slab objects are accounted individually, not per-page. |
2915 | * Memcg membership data for each individual object is saved in | |
2916 | * the page->obj_cgroups. | |
8380ce47 | 2917 | */ |
9855609b RG |
2918 | if (page_has_obj_cgroups(page)) { |
2919 | struct obj_cgroup *objcg; | |
2920 | unsigned int off; | |
2921 | ||
2922 | off = obj_to_index(page->slab_cache, page, p); | |
2923 | objcg = page_obj_cgroups(page)[off]; | |
10befea9 RG |
2924 | if (objcg) |
2925 | return obj_cgroup_memcg(objcg); | |
2926 | ||
2927 | return NULL; | |
9855609b | 2928 | } |
8380ce47 RG |
2929 | |
2930 | /* All other pages use page->mem_cgroup */ | |
2931 | return page->mem_cgroup; | |
2932 | } | |
2933 | ||
bf4f0599 RG |
2934 | __always_inline struct obj_cgroup *get_obj_cgroup_from_current(void) |
2935 | { | |
2936 | struct obj_cgroup *objcg = NULL; | |
2937 | struct mem_cgroup *memcg; | |
2938 | ||
279c3393 RG |
2939 | if (memcg_kmem_bypass()) |
2940 | return NULL; | |
2941 | ||
bf4f0599 RG |
2942 | rcu_read_lock(); |
2943 | if (unlikely(current->active_memcg)) | |
2944 | memcg = rcu_dereference(current->active_memcg); | |
2945 | else | |
2946 | memcg = mem_cgroup_from_task(current); | |
2947 | ||
2948 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { | |
2949 | objcg = rcu_dereference(memcg->objcg); | |
2950 | if (objcg && obj_cgroup_tryget(objcg)) | |
2951 | break; | |
2952 | } | |
2953 | rcu_read_unlock(); | |
2954 | ||
2955 | return objcg; | |
2956 | } | |
2957 | ||
f3bb3043 | 2958 | static int memcg_alloc_cache_id(void) |
55007d84 | 2959 | { |
f3bb3043 VD |
2960 | int id, size; |
2961 | int err; | |
2962 | ||
dbcf73e2 | 2963 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2964 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2965 | if (id < 0) | |
2966 | return id; | |
55007d84 | 2967 | |
dbcf73e2 | 2968 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2969 | return id; |
2970 | ||
2971 | /* | |
2972 | * There's no space for the new id in memcg_caches arrays, | |
2973 | * so we have to grow them. | |
2974 | */ | |
05257a1a | 2975 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2976 | |
2977 | size = 2 * (id + 1); | |
55007d84 GC |
2978 | if (size < MEMCG_CACHES_MIN_SIZE) |
2979 | size = MEMCG_CACHES_MIN_SIZE; | |
2980 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2981 | size = MEMCG_CACHES_MAX_SIZE; | |
2982 | ||
9855609b | 2983 | err = memcg_update_all_list_lrus(size); |
05257a1a VD |
2984 | if (!err) |
2985 | memcg_nr_cache_ids = size; | |
2986 | ||
2987 | up_write(&memcg_cache_ids_sem); | |
2988 | ||
f3bb3043 | 2989 | if (err) { |
dbcf73e2 | 2990 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2991 | return err; |
2992 | } | |
2993 | return id; | |
2994 | } | |
2995 | ||
2996 | static void memcg_free_cache_id(int id) | |
2997 | { | |
dbcf73e2 | 2998 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2999 | } |
3000 | ||
45264778 | 3001 | /** |
4b13f64d | 3002 | * __memcg_kmem_charge: charge a number of kernel pages to a memcg |
10eaec2f | 3003 | * @memcg: memory cgroup to charge |
45264778 | 3004 | * @gfp: reclaim mode |
92d0510c | 3005 | * @nr_pages: number of pages to charge |
45264778 VD |
3006 | * |
3007 | * Returns 0 on success, an error code on failure. | |
3008 | */ | |
4b13f64d RG |
3009 | int __memcg_kmem_charge(struct mem_cgroup *memcg, gfp_t gfp, |
3010 | unsigned int nr_pages) | |
7ae1e1d0 | 3011 | { |
f3ccb2c4 | 3012 | struct page_counter *counter; |
7ae1e1d0 GC |
3013 | int ret; |
3014 | ||
f3ccb2c4 | 3015 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 3016 | if (ret) |
f3ccb2c4 | 3017 | return ret; |
52c29b04 JW |
3018 | |
3019 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
3020 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
e55d9d9b MH |
3021 | |
3022 | /* | |
3023 | * Enforce __GFP_NOFAIL allocation because callers are not | |
3024 | * prepared to see failures and likely do not have any failure | |
3025 | * handling code. | |
3026 | */ | |
3027 | if (gfp & __GFP_NOFAIL) { | |
3028 | page_counter_charge(&memcg->kmem, nr_pages); | |
3029 | return 0; | |
3030 | } | |
52c29b04 JW |
3031 | cancel_charge(memcg, nr_pages); |
3032 | return -ENOMEM; | |
7ae1e1d0 | 3033 | } |
f3ccb2c4 | 3034 | return 0; |
7ae1e1d0 GC |
3035 | } |
3036 | ||
4b13f64d RG |
3037 | /** |
3038 | * __memcg_kmem_uncharge: uncharge a number of kernel pages from a memcg | |
3039 | * @memcg: memcg to uncharge | |
3040 | * @nr_pages: number of pages to uncharge | |
3041 | */ | |
3042 | void __memcg_kmem_uncharge(struct mem_cgroup *memcg, unsigned int nr_pages) | |
3043 | { | |
3044 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
3045 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
3046 | ||
3047 | page_counter_uncharge(&memcg->memory, nr_pages); | |
3048 | if (do_memsw_account()) | |
3049 | page_counter_uncharge(&memcg->memsw, nr_pages); | |
3050 | } | |
3051 | ||
45264778 | 3052 | /** |
f4b00eab | 3053 | * __memcg_kmem_charge_page: charge a kmem page to the current memory cgroup |
45264778 VD |
3054 | * @page: page to charge |
3055 | * @gfp: reclaim mode | |
3056 | * @order: allocation order | |
3057 | * | |
3058 | * Returns 0 on success, an error code on failure. | |
3059 | */ | |
f4b00eab | 3060 | int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 3061 | { |
f3ccb2c4 | 3062 | struct mem_cgroup *memcg; |
fcff7d7e | 3063 | int ret = 0; |
7ae1e1d0 | 3064 | |
d46eb14b | 3065 | memcg = get_mem_cgroup_from_current(); |
279c3393 | 3066 | if (memcg && !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 | } |
279c3393 | 3073 | css_put(&memcg->css); |
c4159a75 | 3074 | } |
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 | 5292 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
b87d8cef | 5293 | struct mem_cgroup *memcg, *old_memcg; |
0b8f73e1 | 5294 | long error = -ENOMEM; |
d142e3e6 | 5295 | |
b87d8cef | 5296 | old_memcg = set_active_memcg(parent); |
0b8f73e1 | 5297 | memcg = mem_cgroup_alloc(); |
b87d8cef | 5298 | set_active_memcg(old_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 | ||
9a137153 | 5519 | if (!(mc.flags & MOVE_ANON)) |
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 | ||
9a137153 RC |
5538 | if (non_swap_entry(ent)) |
5539 | return NULL; | |
5540 | ||
4b91355e KH |
5541 | /* |
5542 | * Because lookup_swap_cache() updates some statistics counter, | |
5543 | * we call find_get_page() with swapper_space directly. | |
5544 | */ | |
f6ab1f7f | 5545 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
2d1c4980 | 5546 | entry->val = ent.val; |
90254a65 DN |
5547 | |
5548 | return page; | |
5549 | } | |
4b91355e KH |
5550 | #else |
5551 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 5552 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
5553 | { |
5554 | return NULL; | |
5555 | } | |
5556 | #endif | |
90254a65 | 5557 | |
87946a72 DN |
5558 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5559 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5560 | { | |
87946a72 DN |
5561 | if (!vma->vm_file) /* anonymous vma */ |
5562 | return NULL; | |
1dfab5ab | 5563 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
5564 | return NULL; |
5565 | ||
87946a72 | 5566 | /* page is moved even if it's not RSS of this task(page-faulted). */ |
aa3b1895 | 5567 | /* shmem/tmpfs may report page out on swap: account for that too. */ |
f5df8635 MWO |
5568 | return find_get_incore_page(vma->vm_file->f_mapping, |
5569 | linear_page_index(vma, addr)); | |
87946a72 DN |
5570 | } |
5571 | ||
b1b0deab CG |
5572 | /** |
5573 | * mem_cgroup_move_account - move account of the page | |
5574 | * @page: the page | |
25843c2b | 5575 | * @compound: charge the page as compound or small page |
b1b0deab CG |
5576 | * @from: mem_cgroup which the page is moved from. |
5577 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
5578 | * | |
3ac808fd | 5579 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
5580 | * |
5581 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
5582 | * from old cgroup. | |
5583 | */ | |
5584 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 5585 | bool compound, |
b1b0deab CG |
5586 | struct mem_cgroup *from, |
5587 | struct mem_cgroup *to) | |
5588 | { | |
ae8af438 KK |
5589 | struct lruvec *from_vec, *to_vec; |
5590 | struct pglist_data *pgdat; | |
6c357848 | 5591 | unsigned int nr_pages = compound ? thp_nr_pages(page) : 1; |
b1b0deab CG |
5592 | int ret; |
5593 | ||
5594 | VM_BUG_ON(from == to); | |
5595 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 5596 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
5597 | |
5598 | /* | |
6a93ca8f | 5599 | * Prevent mem_cgroup_migrate() from looking at |
45637bab | 5600 | * page->mem_cgroup of its source page while we change it. |
b1b0deab | 5601 | */ |
f627c2f5 | 5602 | ret = -EBUSY; |
b1b0deab CG |
5603 | if (!trylock_page(page)) |
5604 | goto out; | |
5605 | ||
5606 | ret = -EINVAL; | |
5607 | if (page->mem_cgroup != from) | |
5608 | goto out_unlock; | |
5609 | ||
ae8af438 | 5610 | pgdat = page_pgdat(page); |
867e5e1d JW |
5611 | from_vec = mem_cgroup_lruvec(from, pgdat); |
5612 | to_vec = mem_cgroup_lruvec(to, pgdat); | |
ae8af438 | 5613 | |
abb242f5 | 5614 | lock_page_memcg(page); |
b1b0deab | 5615 | |
be5d0a74 JW |
5616 | if (PageAnon(page)) { |
5617 | if (page_mapped(page)) { | |
5618 | __mod_lruvec_state(from_vec, NR_ANON_MAPPED, -nr_pages); | |
5619 | __mod_lruvec_state(to_vec, NR_ANON_MAPPED, nr_pages); | |
468c3982 JW |
5620 | if (PageTransHuge(page)) { |
5621 | __mod_lruvec_state(from_vec, NR_ANON_THPS, | |
5622 | -nr_pages); | |
5623 | __mod_lruvec_state(to_vec, NR_ANON_THPS, | |
5624 | nr_pages); | |
5625 | } | |
5626 | ||
be5d0a74 JW |
5627 | } |
5628 | } else { | |
0d1c2072 JW |
5629 | __mod_lruvec_state(from_vec, NR_FILE_PAGES, -nr_pages); |
5630 | __mod_lruvec_state(to_vec, NR_FILE_PAGES, nr_pages); | |
5631 | ||
5632 | if (PageSwapBacked(page)) { | |
5633 | __mod_lruvec_state(from_vec, NR_SHMEM, -nr_pages); | |
5634 | __mod_lruvec_state(to_vec, NR_SHMEM, nr_pages); | |
5635 | } | |
5636 | ||
49e50d27 JW |
5637 | if (page_mapped(page)) { |
5638 | __mod_lruvec_state(from_vec, NR_FILE_MAPPED, -nr_pages); | |
5639 | __mod_lruvec_state(to_vec, NR_FILE_MAPPED, nr_pages); | |
5640 | } | |
b1b0deab | 5641 | |
49e50d27 JW |
5642 | if (PageDirty(page)) { |
5643 | struct address_space *mapping = page_mapping(page); | |
c4843a75 | 5644 | |
f56753ac | 5645 | if (mapping_can_writeback(mapping)) { |
49e50d27 JW |
5646 | __mod_lruvec_state(from_vec, NR_FILE_DIRTY, |
5647 | -nr_pages); | |
5648 | __mod_lruvec_state(to_vec, NR_FILE_DIRTY, | |
5649 | nr_pages); | |
5650 | } | |
c4843a75 GT |
5651 | } |
5652 | } | |
5653 | ||
b1b0deab | 5654 | if (PageWriteback(page)) { |
ae8af438 KK |
5655 | __mod_lruvec_state(from_vec, NR_WRITEBACK, -nr_pages); |
5656 | __mod_lruvec_state(to_vec, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
5657 | } |
5658 | ||
5659 | /* | |
abb242f5 JW |
5660 | * All state has been migrated, let's switch to the new memcg. |
5661 | * | |
b1b0deab | 5662 | * It is safe to change page->mem_cgroup here because the page |
abb242f5 JW |
5663 | * is referenced, charged, isolated, and locked: we can't race |
5664 | * with (un)charging, migration, LRU putback, or anything else | |
5665 | * that would rely on a stable page->mem_cgroup. | |
5666 | * | |
5667 | * Note that lock_page_memcg is a memcg lock, not a page lock, | |
5668 | * to save space. As soon as we switch page->mem_cgroup to a | |
5669 | * new memcg that isn't locked, the above state can change | |
5670 | * concurrently again. Make sure we're truly done with it. | |
b1b0deab | 5671 | */ |
abb242f5 | 5672 | smp_mb(); |
b1b0deab | 5673 | |
1a3e1f40 JW |
5674 | css_get(&to->css); |
5675 | css_put(&from->css); | |
5676 | ||
5677 | page->mem_cgroup = to; | |
87eaceb3 | 5678 | |
abb242f5 | 5679 | __unlock_page_memcg(from); |
b1b0deab CG |
5680 | |
5681 | ret = 0; | |
5682 | ||
5683 | local_irq_disable(); | |
3fba69a5 | 5684 | mem_cgroup_charge_statistics(to, page, nr_pages); |
b1b0deab | 5685 | memcg_check_events(to, page); |
3fba69a5 | 5686 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
b1b0deab CG |
5687 | memcg_check_events(from, page); |
5688 | local_irq_enable(); | |
5689 | out_unlock: | |
5690 | unlock_page(page); | |
5691 | out: | |
5692 | return ret; | |
5693 | } | |
5694 | ||
7cf7806c LR |
5695 | /** |
5696 | * get_mctgt_type - get target type of moving charge | |
5697 | * @vma: the vma the pte to be checked belongs | |
5698 | * @addr: the address corresponding to the pte to be checked | |
5699 | * @ptent: the pte to be checked | |
5700 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
5701 | * | |
5702 | * Returns | |
5703 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5704 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5705 | * move charge. if @target is not NULL, the page is stored in target->page | |
5706 | * with extra refcnt got(Callers should handle it). | |
5707 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
5708 | * target for charge migration. if @target is not NULL, the entry is stored | |
5709 | * in target->ent. | |
25b2995a CH |
5710 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PRIVATE |
5711 | * (so ZONE_DEVICE page and thus not on the lru). | |
df6ad698 JG |
5712 | * For now we such page is charge like a regular page would be as for all |
5713 | * intent and purposes it is just special memory taking the place of a | |
5714 | * regular page. | |
c733a828 JG |
5715 | * |
5716 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
5717 | * |
5718 | * Called with pte lock held. | |
5719 | */ | |
5720 | ||
8d32ff84 | 5721 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5722 | unsigned long addr, pte_t ptent, union mc_target *target) |
5723 | { | |
5724 | struct page *page = NULL; | |
8d32ff84 | 5725 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5726 | swp_entry_t ent = { .val = 0 }; |
5727 | ||
5728 | if (pte_present(ptent)) | |
5729 | page = mc_handle_present_pte(vma, addr, ptent); | |
5730 | else if (is_swap_pte(ptent)) | |
48406ef8 | 5731 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 5732 | else if (pte_none(ptent)) |
87946a72 | 5733 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
5734 | |
5735 | if (!page && !ent.val) | |
8d32ff84 | 5736 | return ret; |
02491447 | 5737 | if (page) { |
02491447 | 5738 | /* |
0a31bc97 | 5739 | * Do only loose check w/o serialization. |
1306a85a | 5740 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 5741 | * not under LRU exclusion. |
02491447 | 5742 | */ |
1306a85a | 5743 | if (page->mem_cgroup == mc.from) { |
02491447 | 5744 | ret = MC_TARGET_PAGE; |
25b2995a | 5745 | if (is_device_private_page(page)) |
c733a828 | 5746 | ret = MC_TARGET_DEVICE; |
02491447 DN |
5747 | if (target) |
5748 | target->page = page; | |
5749 | } | |
5750 | if (!ret || !target) | |
5751 | put_page(page); | |
5752 | } | |
3e14a57b HY |
5753 | /* |
5754 | * There is a swap entry and a page doesn't exist or isn't charged. | |
5755 | * But we cannot move a tail-page in a THP. | |
5756 | */ | |
5757 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 5758 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5759 | ret = MC_TARGET_SWAP; |
5760 | if (target) | |
5761 | target->ent = ent; | |
4ffef5fe | 5762 | } |
4ffef5fe DN |
5763 | return ret; |
5764 | } | |
5765 | ||
12724850 NH |
5766 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5767 | /* | |
d6810d73 HY |
5768 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
5769 | * not support them for now. | |
12724850 NH |
5770 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
5771 | */ | |
5772 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5773 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5774 | { | |
5775 | struct page *page = NULL; | |
12724850 NH |
5776 | enum mc_target_type ret = MC_TARGET_NONE; |
5777 | ||
84c3fc4e ZY |
5778 | if (unlikely(is_swap_pmd(pmd))) { |
5779 | VM_BUG_ON(thp_migration_supported() && | |
5780 | !is_pmd_migration_entry(pmd)); | |
5781 | return ret; | |
5782 | } | |
12724850 | 5783 | page = pmd_page(pmd); |
309381fe | 5784 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 5785 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 5786 | return ret; |
1306a85a | 5787 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
5788 | ret = MC_TARGET_PAGE; |
5789 | if (target) { | |
5790 | get_page(page); | |
5791 | target->page = page; | |
5792 | } | |
5793 | } | |
5794 | return ret; | |
5795 | } | |
5796 | #else | |
5797 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5798 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5799 | { | |
5800 | return MC_TARGET_NONE; | |
5801 | } | |
5802 | #endif | |
5803 | ||
4ffef5fe DN |
5804 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5805 | unsigned long addr, unsigned long end, | |
5806 | struct mm_walk *walk) | |
5807 | { | |
26bcd64a | 5808 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5809 | pte_t *pte; |
5810 | spinlock_t *ptl; | |
5811 | ||
b6ec57f4 KS |
5812 | ptl = pmd_trans_huge_lock(pmd, vma); |
5813 | if (ptl) { | |
c733a828 JG |
5814 | /* |
5815 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
25b2995a CH |
5816 | * support transparent huge page with MEMORY_DEVICE_PRIVATE but |
5817 | * this might change. | |
c733a828 | 5818 | */ |
12724850 NH |
5819 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5820 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5821 | spin_unlock(ptl); |
1a5a9906 | 5822 | return 0; |
12724850 | 5823 | } |
03319327 | 5824 | |
45f83cef AA |
5825 | if (pmd_trans_unstable(pmd)) |
5826 | return 0; | |
4ffef5fe DN |
5827 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5828 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5829 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5830 | mc.precharge++; /* increment precharge temporarily */ |
5831 | pte_unmap_unlock(pte - 1, ptl); | |
5832 | cond_resched(); | |
5833 | ||
7dc74be0 DN |
5834 | return 0; |
5835 | } | |
5836 | ||
7b86ac33 CH |
5837 | static const struct mm_walk_ops precharge_walk_ops = { |
5838 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5839 | }; | |
5840 | ||
4ffef5fe DN |
5841 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5842 | { | |
5843 | unsigned long precharge; | |
4ffef5fe | 5844 | |
d8ed45c5 | 5845 | mmap_read_lock(mm); |
7b86ac33 | 5846 | walk_page_range(mm, 0, mm->highest_vm_end, &precharge_walk_ops, NULL); |
d8ed45c5 | 5847 | mmap_read_unlock(mm); |
4ffef5fe DN |
5848 | |
5849 | precharge = mc.precharge; | |
5850 | mc.precharge = 0; | |
5851 | ||
5852 | return precharge; | |
5853 | } | |
5854 | ||
4ffef5fe DN |
5855 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5856 | { | |
dfe076b0 DN |
5857 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5858 | ||
5859 | VM_BUG_ON(mc.moving_task); | |
5860 | mc.moving_task = current; | |
5861 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5862 | } |
5863 | ||
dfe076b0 DN |
5864 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5865 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5866 | { |
2bd9bb20 KH |
5867 | struct mem_cgroup *from = mc.from; |
5868 | struct mem_cgroup *to = mc.to; | |
5869 | ||
4ffef5fe | 5870 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5871 | if (mc.precharge) { |
00501b53 | 5872 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5873 | mc.precharge = 0; |
5874 | } | |
5875 | /* | |
5876 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5877 | * we must uncharge here. | |
5878 | */ | |
5879 | if (mc.moved_charge) { | |
00501b53 | 5880 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5881 | mc.moved_charge = 0; |
4ffef5fe | 5882 | } |
483c30b5 DN |
5883 | /* we must fixup refcnts and charges */ |
5884 | if (mc.moved_swap) { | |
483c30b5 | 5885 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5886 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5887 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5888 | |
615d66c3 VD |
5889 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
5890 | ||
05b84301 | 5891 | /* |
3e32cb2e JW |
5892 | * we charged both to->memory and to->memsw, so we |
5893 | * should uncharge to->memory. | |
05b84301 | 5894 | */ |
ce00a967 | 5895 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5896 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5897 | ||
483c30b5 DN |
5898 | mc.moved_swap = 0; |
5899 | } | |
dfe076b0 DN |
5900 | memcg_oom_recover(from); |
5901 | memcg_oom_recover(to); | |
5902 | wake_up_all(&mc.waitq); | |
5903 | } | |
5904 | ||
5905 | static void mem_cgroup_clear_mc(void) | |
5906 | { | |
264a0ae1 TH |
5907 | struct mm_struct *mm = mc.mm; |
5908 | ||
dfe076b0 DN |
5909 | /* |
5910 | * we must clear moving_task before waking up waiters at the end of | |
5911 | * task migration. | |
5912 | */ | |
5913 | mc.moving_task = NULL; | |
5914 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5915 | spin_lock(&mc.lock); |
4ffef5fe DN |
5916 | mc.from = NULL; |
5917 | mc.to = NULL; | |
264a0ae1 | 5918 | mc.mm = NULL; |
2bd9bb20 | 5919 | spin_unlock(&mc.lock); |
264a0ae1 TH |
5920 | |
5921 | mmput(mm); | |
4ffef5fe DN |
5922 | } |
5923 | ||
1f7dd3e5 | 5924 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5925 | { |
1f7dd3e5 | 5926 | struct cgroup_subsys_state *css; |
eed67d75 | 5927 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 5928 | struct mem_cgroup *from; |
4530eddb | 5929 | struct task_struct *leader, *p; |
9f2115f9 | 5930 | struct mm_struct *mm; |
1dfab5ab | 5931 | unsigned long move_flags; |
9f2115f9 | 5932 | int ret = 0; |
7dc74be0 | 5933 | |
1f7dd3e5 TH |
5934 | /* charge immigration isn't supported on the default hierarchy */ |
5935 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
5936 | return 0; |
5937 | ||
4530eddb TH |
5938 | /* |
5939 | * Multi-process migrations only happen on the default hierarchy | |
5940 | * where charge immigration is not used. Perform charge | |
5941 | * immigration if @tset contains a leader and whine if there are | |
5942 | * multiple. | |
5943 | */ | |
5944 | p = NULL; | |
1f7dd3e5 | 5945 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
5946 | WARN_ON_ONCE(p); |
5947 | p = leader; | |
1f7dd3e5 | 5948 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
5949 | } |
5950 | if (!p) | |
5951 | return 0; | |
5952 | ||
1f7dd3e5 TH |
5953 | /* |
5954 | * We are now commited to this value whatever it is. Changes in this | |
5955 | * tunable will only affect upcoming migrations, not the current one. | |
5956 | * So we need to save it, and keep it going. | |
5957 | */ | |
5958 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
5959 | if (!move_flags) | |
5960 | return 0; | |
5961 | ||
9f2115f9 TH |
5962 | from = mem_cgroup_from_task(p); |
5963 | ||
5964 | VM_BUG_ON(from == memcg); | |
5965 | ||
5966 | mm = get_task_mm(p); | |
5967 | if (!mm) | |
5968 | return 0; | |
5969 | /* We move charges only when we move a owner of the mm */ | |
5970 | if (mm->owner == p) { | |
5971 | VM_BUG_ON(mc.from); | |
5972 | VM_BUG_ON(mc.to); | |
5973 | VM_BUG_ON(mc.precharge); | |
5974 | VM_BUG_ON(mc.moved_charge); | |
5975 | VM_BUG_ON(mc.moved_swap); | |
5976 | ||
5977 | spin_lock(&mc.lock); | |
264a0ae1 | 5978 | mc.mm = mm; |
9f2115f9 TH |
5979 | mc.from = from; |
5980 | mc.to = memcg; | |
5981 | mc.flags = move_flags; | |
5982 | spin_unlock(&mc.lock); | |
5983 | /* We set mc.moving_task later */ | |
5984 | ||
5985 | ret = mem_cgroup_precharge_mc(mm); | |
5986 | if (ret) | |
5987 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
5988 | } else { |
5989 | mmput(mm); | |
7dc74be0 DN |
5990 | } |
5991 | return ret; | |
5992 | } | |
5993 | ||
1f7dd3e5 | 5994 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5995 | { |
4e2f245d JW |
5996 | if (mc.to) |
5997 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5998 | } |
5999 | ||
4ffef5fe DN |
6000 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
6001 | unsigned long addr, unsigned long end, | |
6002 | struct mm_walk *walk) | |
7dc74be0 | 6003 | { |
4ffef5fe | 6004 | int ret = 0; |
26bcd64a | 6005 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
6006 | pte_t *pte; |
6007 | spinlock_t *ptl; | |
12724850 NH |
6008 | enum mc_target_type target_type; |
6009 | union mc_target target; | |
6010 | struct page *page; | |
4ffef5fe | 6011 | |
b6ec57f4 KS |
6012 | ptl = pmd_trans_huge_lock(pmd, vma); |
6013 | if (ptl) { | |
62ade86a | 6014 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 6015 | spin_unlock(ptl); |
12724850 NH |
6016 | return 0; |
6017 | } | |
6018 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
6019 | if (target_type == MC_TARGET_PAGE) { | |
6020 | page = target.page; | |
6021 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 6022 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 6023 | mc.from, mc.to)) { |
12724850 NH |
6024 | mc.precharge -= HPAGE_PMD_NR; |
6025 | mc.moved_charge += HPAGE_PMD_NR; | |
6026 | } | |
6027 | putback_lru_page(page); | |
6028 | } | |
6029 | put_page(page); | |
c733a828 JG |
6030 | } else if (target_type == MC_TARGET_DEVICE) { |
6031 | page = target.page; | |
6032 | if (!mem_cgroup_move_account(page, true, | |
6033 | mc.from, mc.to)) { | |
6034 | mc.precharge -= HPAGE_PMD_NR; | |
6035 | mc.moved_charge += HPAGE_PMD_NR; | |
6036 | } | |
6037 | put_page(page); | |
12724850 | 6038 | } |
bf929152 | 6039 | spin_unlock(ptl); |
1a5a9906 | 6040 | return 0; |
12724850 NH |
6041 | } |
6042 | ||
45f83cef AA |
6043 | if (pmd_trans_unstable(pmd)) |
6044 | return 0; | |
4ffef5fe DN |
6045 | retry: |
6046 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
6047 | for (; addr != end; addr += PAGE_SIZE) { | |
6048 | pte_t ptent = *(pte++); | |
c733a828 | 6049 | bool device = false; |
02491447 | 6050 | swp_entry_t ent; |
4ffef5fe DN |
6051 | |
6052 | if (!mc.precharge) | |
6053 | break; | |
6054 | ||
8d32ff84 | 6055 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
6056 | case MC_TARGET_DEVICE: |
6057 | device = true; | |
e4a9bc58 | 6058 | fallthrough; |
4ffef5fe DN |
6059 | case MC_TARGET_PAGE: |
6060 | page = target.page; | |
53f9263b KS |
6061 | /* |
6062 | * We can have a part of the split pmd here. Moving it | |
6063 | * can be done but it would be too convoluted so simply | |
6064 | * ignore such a partial THP and keep it in original | |
6065 | * memcg. There should be somebody mapping the head. | |
6066 | */ | |
6067 | if (PageTransCompound(page)) | |
6068 | goto put; | |
c733a828 | 6069 | if (!device && isolate_lru_page(page)) |
4ffef5fe | 6070 | goto put; |
f627c2f5 KS |
6071 | if (!mem_cgroup_move_account(page, false, |
6072 | mc.from, mc.to)) { | |
4ffef5fe | 6073 | mc.precharge--; |
854ffa8d DN |
6074 | /* we uncharge from mc.from later. */ |
6075 | mc.moved_charge++; | |
4ffef5fe | 6076 | } |
c733a828 JG |
6077 | if (!device) |
6078 | putback_lru_page(page); | |
8d32ff84 | 6079 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
6080 | put_page(page); |
6081 | break; | |
02491447 DN |
6082 | case MC_TARGET_SWAP: |
6083 | ent = target.ent; | |
e91cbb42 | 6084 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6085 | mc.precharge--; |
8d22a935 HD |
6086 | mem_cgroup_id_get_many(mc.to, 1); |
6087 | /* we fixup other refcnts and charges later. */ | |
483c30b5 DN |
6088 | mc.moved_swap++; |
6089 | } | |
02491447 | 6090 | break; |
4ffef5fe DN |
6091 | default: |
6092 | break; | |
6093 | } | |
6094 | } | |
6095 | pte_unmap_unlock(pte - 1, ptl); | |
6096 | cond_resched(); | |
6097 | ||
6098 | if (addr != end) { | |
6099 | /* | |
6100 | * We have consumed all precharges we got in can_attach(). | |
6101 | * We try charge one by one, but don't do any additional | |
6102 | * charges to mc.to if we have failed in charge once in attach() | |
6103 | * phase. | |
6104 | */ | |
854ffa8d | 6105 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6106 | if (!ret) |
6107 | goto retry; | |
6108 | } | |
6109 | ||
6110 | return ret; | |
6111 | } | |
6112 | ||
7b86ac33 CH |
6113 | static const struct mm_walk_ops charge_walk_ops = { |
6114 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6115 | }; | |
6116 | ||
264a0ae1 | 6117 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 6118 | { |
4ffef5fe | 6119 | lru_add_drain_all(); |
312722cb | 6120 | /* |
81f8c3a4 JW |
6121 | * Signal lock_page_memcg() to take the memcg's move_lock |
6122 | * while we're moving its pages to another memcg. Then wait | |
6123 | * for already started RCU-only updates to finish. | |
312722cb JW |
6124 | */ |
6125 | atomic_inc(&mc.from->moving_account); | |
6126 | synchronize_rcu(); | |
dfe076b0 | 6127 | retry: |
d8ed45c5 | 6128 | if (unlikely(!mmap_read_trylock(mc.mm))) { |
dfe076b0 | 6129 | /* |
c1e8d7c6 | 6130 | * Someone who are holding the mmap_lock might be waiting in |
dfe076b0 DN |
6131 | * waitq. So we cancel all extra charges, wake up all waiters, |
6132 | * and retry. Because we cancel precharges, we might not be able | |
6133 | * to move enough charges, but moving charge is a best-effort | |
6134 | * feature anyway, so it wouldn't be a big problem. | |
6135 | */ | |
6136 | __mem_cgroup_clear_mc(); | |
6137 | cond_resched(); | |
6138 | goto retry; | |
6139 | } | |
26bcd64a NH |
6140 | /* |
6141 | * When we have consumed all precharges and failed in doing | |
6142 | * additional charge, the page walk just aborts. | |
6143 | */ | |
7b86ac33 CH |
6144 | walk_page_range(mc.mm, 0, mc.mm->highest_vm_end, &charge_walk_ops, |
6145 | NULL); | |
0247f3f4 | 6146 | |
d8ed45c5 | 6147 | mmap_read_unlock(mc.mm); |
312722cb | 6148 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
6149 | } |
6150 | ||
264a0ae1 | 6151 | static void mem_cgroup_move_task(void) |
67e465a7 | 6152 | { |
264a0ae1 TH |
6153 | if (mc.to) { |
6154 | mem_cgroup_move_charge(); | |
a433658c | 6155 | mem_cgroup_clear_mc(); |
264a0ae1 | 6156 | } |
67e465a7 | 6157 | } |
5cfb80a7 | 6158 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 6159 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6160 | { |
6161 | return 0; | |
6162 | } | |
1f7dd3e5 | 6163 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6164 | { |
6165 | } | |
264a0ae1 | 6166 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
6167 | { |
6168 | } | |
6169 | #endif | |
67e465a7 | 6170 | |
f00baae7 TH |
6171 | /* |
6172 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
6173 | * to verify whether we're attached to the default hierarchy on each mount |
6174 | * attempt. | |
f00baae7 | 6175 | */ |
eb95419b | 6176 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
6177 | { |
6178 | /* | |
aa6ec29b | 6179 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
6180 | * guarantees that @root doesn't have any children, so turning it |
6181 | * on for the root memcg is enough. | |
6182 | */ | |
9e10a130 | 6183 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
6184 | root_mem_cgroup->use_hierarchy = true; |
6185 | else | |
6186 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
6187 | } |
6188 | ||
677dc973 CD |
6189 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
6190 | { | |
6191 | if (value == PAGE_COUNTER_MAX) | |
6192 | seq_puts(m, "max\n"); | |
6193 | else | |
6194 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); | |
6195 | ||
6196 | return 0; | |
6197 | } | |
6198 | ||
241994ed JW |
6199 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
6200 | struct cftype *cft) | |
6201 | { | |
f5fc3c5d JW |
6202 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
6203 | ||
6204 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
6205 | } |
6206 | ||
bf8d5d52 RG |
6207 | static int memory_min_show(struct seq_file *m, void *v) |
6208 | { | |
677dc973 CD |
6209 | return seq_puts_memcg_tunable(m, |
6210 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); | |
bf8d5d52 RG |
6211 | } |
6212 | ||
6213 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
6214 | char *buf, size_t nbytes, loff_t off) | |
6215 | { | |
6216 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6217 | unsigned long min; | |
6218 | int err; | |
6219 | ||
6220 | buf = strstrip(buf); | |
6221 | err = page_counter_memparse(buf, "max", &min); | |
6222 | if (err) | |
6223 | return err; | |
6224 | ||
6225 | page_counter_set_min(&memcg->memory, min); | |
6226 | ||
6227 | return nbytes; | |
6228 | } | |
6229 | ||
241994ed JW |
6230 | static int memory_low_show(struct seq_file *m, void *v) |
6231 | { | |
677dc973 CD |
6232 | return seq_puts_memcg_tunable(m, |
6233 | READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); | |
241994ed JW |
6234 | } |
6235 | ||
6236 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
6237 | char *buf, size_t nbytes, loff_t off) | |
6238 | { | |
6239 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6240 | unsigned long low; | |
6241 | int err; | |
6242 | ||
6243 | buf = strstrip(buf); | |
d2973697 | 6244 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
6245 | if (err) |
6246 | return err; | |
6247 | ||
23067153 | 6248 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
6249 | |
6250 | return nbytes; | |
6251 | } | |
6252 | ||
6253 | static int memory_high_show(struct seq_file *m, void *v) | |
6254 | { | |
d1663a90 JK |
6255 | return seq_puts_memcg_tunable(m, |
6256 | READ_ONCE(mem_cgroup_from_seq(m)->memory.high)); | |
241994ed JW |
6257 | } |
6258 | ||
6259 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
6260 | char *buf, size_t nbytes, loff_t off) | |
6261 | { | |
6262 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6263 | unsigned int nr_retries = MAX_RECLAIM_RETRIES; |
8c8c383c | 6264 | bool drained = false; |
241994ed JW |
6265 | unsigned long high; |
6266 | int err; | |
6267 | ||
6268 | buf = strstrip(buf); | |
d2973697 | 6269 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
6270 | if (err) |
6271 | return err; | |
6272 | ||
8c8c383c JW |
6273 | for (;;) { |
6274 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6275 | unsigned long reclaimed; | |
6276 | ||
6277 | if (nr_pages <= high) | |
6278 | break; | |
6279 | ||
6280 | if (signal_pending(current)) | |
6281 | break; | |
6282 | ||
6283 | if (!drained) { | |
6284 | drain_all_stock(memcg); | |
6285 | drained = true; | |
6286 | continue; | |
6287 | } | |
6288 | ||
6289 | reclaimed = try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
6290 | GFP_KERNEL, true); | |
6291 | ||
6292 | if (!reclaimed && !nr_retries--) | |
6293 | break; | |
6294 | } | |
588083bb | 6295 | |
536d3bf2 RG |
6296 | page_counter_set_high(&memcg->memory, high); |
6297 | ||
19ce33ac JW |
6298 | memcg_wb_domain_size_changed(memcg); |
6299 | ||
241994ed JW |
6300 | return nbytes; |
6301 | } | |
6302 | ||
6303 | static int memory_max_show(struct seq_file *m, void *v) | |
6304 | { | |
677dc973 CD |
6305 | return seq_puts_memcg_tunable(m, |
6306 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); | |
241994ed JW |
6307 | } |
6308 | ||
6309 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
6310 | char *buf, size_t nbytes, loff_t off) | |
6311 | { | |
6312 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6313 | unsigned int nr_reclaims = MAX_RECLAIM_RETRIES; |
b6e6edcf | 6314 | bool drained = false; |
241994ed JW |
6315 | unsigned long max; |
6316 | int err; | |
6317 | ||
6318 | buf = strstrip(buf); | |
d2973697 | 6319 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
6320 | if (err) |
6321 | return err; | |
6322 | ||
bbec2e15 | 6323 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
6324 | |
6325 | for (;;) { | |
6326 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6327 | ||
6328 | if (nr_pages <= max) | |
6329 | break; | |
6330 | ||
7249c9f0 | 6331 | if (signal_pending(current)) |
b6e6edcf | 6332 | break; |
b6e6edcf JW |
6333 | |
6334 | if (!drained) { | |
6335 | drain_all_stock(memcg); | |
6336 | drained = true; | |
6337 | continue; | |
6338 | } | |
6339 | ||
6340 | if (nr_reclaims) { | |
6341 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
6342 | GFP_KERNEL, true)) | |
6343 | nr_reclaims--; | |
6344 | continue; | |
6345 | } | |
6346 | ||
e27be240 | 6347 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
6348 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
6349 | break; | |
6350 | } | |
241994ed | 6351 | |
2529bb3a | 6352 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
6353 | return nbytes; |
6354 | } | |
6355 | ||
1e577f97 SB |
6356 | static void __memory_events_show(struct seq_file *m, atomic_long_t *events) |
6357 | { | |
6358 | seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW])); | |
6359 | seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH])); | |
6360 | seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX])); | |
6361 | seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM])); | |
6362 | seq_printf(m, "oom_kill %lu\n", | |
6363 | atomic_long_read(&events[MEMCG_OOM_KILL])); | |
6364 | } | |
6365 | ||
241994ed JW |
6366 | static int memory_events_show(struct seq_file *m, void *v) |
6367 | { | |
aa9694bb | 6368 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
241994ed | 6369 | |
1e577f97 SB |
6370 | __memory_events_show(m, memcg->memory_events); |
6371 | return 0; | |
6372 | } | |
6373 | ||
6374 | static int memory_events_local_show(struct seq_file *m, void *v) | |
6375 | { | |
6376 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
241994ed | 6377 | |
1e577f97 | 6378 | __memory_events_show(m, memcg->memory_events_local); |
241994ed JW |
6379 | return 0; |
6380 | } | |
6381 | ||
587d9f72 JW |
6382 | static int memory_stat_show(struct seq_file *m, void *v) |
6383 | { | |
aa9694bb | 6384 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
c8713d0b | 6385 | char *buf; |
1ff9e6e1 | 6386 | |
c8713d0b JW |
6387 | buf = memory_stat_format(memcg); |
6388 | if (!buf) | |
6389 | return -ENOMEM; | |
6390 | seq_puts(m, buf); | |
6391 | kfree(buf); | |
587d9f72 JW |
6392 | return 0; |
6393 | } | |
6394 | ||
5f9a4f4a MS |
6395 | #ifdef CONFIG_NUMA |
6396 | static int memory_numa_stat_show(struct seq_file *m, void *v) | |
6397 | { | |
6398 | int i; | |
6399 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
6400 | ||
6401 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { | |
6402 | int nid; | |
6403 | ||
6404 | if (memory_stats[i].idx >= NR_VM_NODE_STAT_ITEMS) | |
6405 | continue; | |
6406 | ||
6407 | seq_printf(m, "%s", memory_stats[i].name); | |
6408 | for_each_node_state(nid, N_MEMORY) { | |
6409 | u64 size; | |
6410 | struct lruvec *lruvec; | |
6411 | ||
6412 | lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); | |
6413 | size = lruvec_page_state(lruvec, memory_stats[i].idx); | |
6414 | size *= memory_stats[i].ratio; | |
6415 | seq_printf(m, " N%d=%llu", nid, size); | |
6416 | } | |
6417 | seq_putc(m, '\n'); | |
6418 | } | |
6419 | ||
6420 | return 0; | |
6421 | } | |
6422 | #endif | |
6423 | ||
3d8b38eb RG |
6424 | static int memory_oom_group_show(struct seq_file *m, void *v) |
6425 | { | |
aa9694bb | 6426 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3d8b38eb RG |
6427 | |
6428 | seq_printf(m, "%d\n", memcg->oom_group); | |
6429 | ||
6430 | return 0; | |
6431 | } | |
6432 | ||
6433 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
6434 | char *buf, size_t nbytes, loff_t off) | |
6435 | { | |
6436 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6437 | int ret, oom_group; | |
6438 | ||
6439 | buf = strstrip(buf); | |
6440 | if (!buf) | |
6441 | return -EINVAL; | |
6442 | ||
6443 | ret = kstrtoint(buf, 0, &oom_group); | |
6444 | if (ret) | |
6445 | return ret; | |
6446 | ||
6447 | if (oom_group != 0 && oom_group != 1) | |
6448 | return -EINVAL; | |
6449 | ||
6450 | memcg->oom_group = oom_group; | |
6451 | ||
6452 | return nbytes; | |
6453 | } | |
6454 | ||
241994ed JW |
6455 | static struct cftype memory_files[] = { |
6456 | { | |
6457 | .name = "current", | |
f5fc3c5d | 6458 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
6459 | .read_u64 = memory_current_read, |
6460 | }, | |
bf8d5d52 RG |
6461 | { |
6462 | .name = "min", | |
6463 | .flags = CFTYPE_NOT_ON_ROOT, | |
6464 | .seq_show = memory_min_show, | |
6465 | .write = memory_min_write, | |
6466 | }, | |
241994ed JW |
6467 | { |
6468 | .name = "low", | |
6469 | .flags = CFTYPE_NOT_ON_ROOT, | |
6470 | .seq_show = memory_low_show, | |
6471 | .write = memory_low_write, | |
6472 | }, | |
6473 | { | |
6474 | .name = "high", | |
6475 | .flags = CFTYPE_NOT_ON_ROOT, | |
6476 | .seq_show = memory_high_show, | |
6477 | .write = memory_high_write, | |
6478 | }, | |
6479 | { | |
6480 | .name = "max", | |
6481 | .flags = CFTYPE_NOT_ON_ROOT, | |
6482 | .seq_show = memory_max_show, | |
6483 | .write = memory_max_write, | |
6484 | }, | |
6485 | { | |
6486 | .name = "events", | |
6487 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 6488 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
6489 | .seq_show = memory_events_show, |
6490 | }, | |
1e577f97 SB |
6491 | { |
6492 | .name = "events.local", | |
6493 | .flags = CFTYPE_NOT_ON_ROOT, | |
6494 | .file_offset = offsetof(struct mem_cgroup, events_local_file), | |
6495 | .seq_show = memory_events_local_show, | |
6496 | }, | |
587d9f72 JW |
6497 | { |
6498 | .name = "stat", | |
587d9f72 JW |
6499 | .seq_show = memory_stat_show, |
6500 | }, | |
5f9a4f4a MS |
6501 | #ifdef CONFIG_NUMA |
6502 | { | |
6503 | .name = "numa_stat", | |
6504 | .seq_show = memory_numa_stat_show, | |
6505 | }, | |
6506 | #endif | |
3d8b38eb RG |
6507 | { |
6508 | .name = "oom.group", | |
6509 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
6510 | .seq_show = memory_oom_group_show, | |
6511 | .write = memory_oom_group_write, | |
6512 | }, | |
241994ed JW |
6513 | { } /* terminate */ |
6514 | }; | |
6515 | ||
073219e9 | 6516 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 6517 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6518 | .css_online = mem_cgroup_css_online, |
92fb9748 | 6519 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 6520 | .css_released = mem_cgroup_css_released, |
92fb9748 | 6521 | .css_free = mem_cgroup_css_free, |
1ced953b | 6522 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
6523 | .can_attach = mem_cgroup_can_attach, |
6524 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 6525 | .post_attach = mem_cgroup_move_task, |
f00baae7 | 6526 | .bind = mem_cgroup_bind, |
241994ed JW |
6527 | .dfl_cftypes = memory_files, |
6528 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 6529 | .early_init = 0, |
8cdea7c0 | 6530 | }; |
c077719b | 6531 | |
bc50bcc6 JW |
6532 | /* |
6533 | * This function calculates an individual cgroup's effective | |
6534 | * protection which is derived from its own memory.min/low, its | |
6535 | * parent's and siblings' settings, as well as the actual memory | |
6536 | * distribution in the tree. | |
6537 | * | |
6538 | * The following rules apply to the effective protection values: | |
6539 | * | |
6540 | * 1. At the first level of reclaim, effective protection is equal to | |
6541 | * the declared protection in memory.min and memory.low. | |
6542 | * | |
6543 | * 2. To enable safe delegation of the protection configuration, at | |
6544 | * subsequent levels the effective protection is capped to the | |
6545 | * parent's effective protection. | |
6546 | * | |
6547 | * 3. To make complex and dynamic subtrees easier to configure, the | |
6548 | * user is allowed to overcommit the declared protection at a given | |
6549 | * level. If that is the case, the parent's effective protection is | |
6550 | * distributed to the children in proportion to how much protection | |
6551 | * they have declared and how much of it they are utilizing. | |
6552 | * | |
6553 | * This makes distribution proportional, but also work-conserving: | |
6554 | * if one cgroup claims much more protection than it uses memory, | |
6555 | * the unused remainder is available to its siblings. | |
6556 | * | |
6557 | * 4. Conversely, when the declared protection is undercommitted at a | |
6558 | * given level, the distribution of the larger parental protection | |
6559 | * budget is NOT proportional. A cgroup's protection from a sibling | |
6560 | * is capped to its own memory.min/low setting. | |
6561 | * | |
8a931f80 JW |
6562 | * 5. However, to allow protecting recursive subtrees from each other |
6563 | * without having to declare each individual cgroup's fixed share | |
6564 | * of the ancestor's claim to protection, any unutilized - | |
6565 | * "floating" - protection from up the tree is distributed in | |
6566 | * proportion to each cgroup's *usage*. This makes the protection | |
6567 | * neutral wrt sibling cgroups and lets them compete freely over | |
6568 | * the shared parental protection budget, but it protects the | |
6569 | * subtree as a whole from neighboring subtrees. | |
6570 | * | |
6571 | * Note that 4. and 5. are not in conflict: 4. is about protecting | |
6572 | * against immediate siblings whereas 5. is about protecting against | |
6573 | * neighboring subtrees. | |
bc50bcc6 JW |
6574 | */ |
6575 | static unsigned long effective_protection(unsigned long usage, | |
8a931f80 | 6576 | unsigned long parent_usage, |
bc50bcc6 JW |
6577 | unsigned long setting, |
6578 | unsigned long parent_effective, | |
6579 | unsigned long siblings_protected) | |
6580 | { | |
6581 | unsigned long protected; | |
8a931f80 | 6582 | unsigned long ep; |
bc50bcc6 JW |
6583 | |
6584 | protected = min(usage, setting); | |
6585 | /* | |
6586 | * If all cgroups at this level combined claim and use more | |
6587 | * protection then what the parent affords them, distribute | |
6588 | * shares in proportion to utilization. | |
6589 | * | |
6590 | * We are using actual utilization rather than the statically | |
6591 | * claimed protection in order to be work-conserving: claimed | |
6592 | * but unused protection is available to siblings that would | |
6593 | * otherwise get a smaller chunk than what they claimed. | |
6594 | */ | |
6595 | if (siblings_protected > parent_effective) | |
6596 | return protected * parent_effective / siblings_protected; | |
6597 | ||
6598 | /* | |
6599 | * Ok, utilized protection of all children is within what the | |
6600 | * parent affords them, so we know whatever this child claims | |
6601 | * and utilizes is effectively protected. | |
6602 | * | |
6603 | * If there is unprotected usage beyond this value, reclaim | |
6604 | * will apply pressure in proportion to that amount. | |
6605 | * | |
6606 | * If there is unutilized protection, the cgroup will be fully | |
6607 | * shielded from reclaim, but we do return a smaller value for | |
6608 | * protection than what the group could enjoy in theory. This | |
6609 | * is okay. With the overcommit distribution above, effective | |
6610 | * protection is always dependent on how memory is actually | |
6611 | * consumed among the siblings anyway. | |
6612 | */ | |
8a931f80 JW |
6613 | ep = protected; |
6614 | ||
6615 | /* | |
6616 | * If the children aren't claiming (all of) the protection | |
6617 | * afforded to them by the parent, distribute the remainder in | |
6618 | * proportion to the (unprotected) memory of each cgroup. That | |
6619 | * way, cgroups that aren't explicitly prioritized wrt each | |
6620 | * other compete freely over the allowance, but they are | |
6621 | * collectively protected from neighboring trees. | |
6622 | * | |
6623 | * We're using unprotected memory for the weight so that if | |
6624 | * some cgroups DO claim explicit protection, we don't protect | |
6625 | * the same bytes twice. | |
cd324edc JW |
6626 | * |
6627 | * Check both usage and parent_usage against the respective | |
6628 | * protected values. One should imply the other, but they | |
6629 | * aren't read atomically - make sure the division is sane. | |
8a931f80 JW |
6630 | */ |
6631 | if (!(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)) | |
6632 | return ep; | |
cd324edc JW |
6633 | if (parent_effective > siblings_protected && |
6634 | parent_usage > siblings_protected && | |
6635 | usage > protected) { | |
8a931f80 JW |
6636 | unsigned long unclaimed; |
6637 | ||
6638 | unclaimed = parent_effective - siblings_protected; | |
6639 | unclaimed *= usage - protected; | |
6640 | unclaimed /= parent_usage - siblings_protected; | |
6641 | ||
6642 | ep += unclaimed; | |
6643 | } | |
6644 | ||
6645 | return ep; | |
bc50bcc6 JW |
6646 | } |
6647 | ||
241994ed | 6648 | /** |
bf8d5d52 | 6649 | * mem_cgroup_protected - check if memory consumption is in the normal range |
34c81057 | 6650 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
6651 | * @memcg: the memory cgroup to check |
6652 | * | |
23067153 RG |
6653 | * WARNING: This function is not stateless! It can only be used as part |
6654 | * of a top-down tree iteration, not for isolated queries. | |
241994ed | 6655 | */ |
45c7f7e1 CD |
6656 | void mem_cgroup_calculate_protection(struct mem_cgroup *root, |
6657 | struct mem_cgroup *memcg) | |
241994ed | 6658 | { |
8a931f80 | 6659 | unsigned long usage, parent_usage; |
23067153 RG |
6660 | struct mem_cgroup *parent; |
6661 | ||
241994ed | 6662 | if (mem_cgroup_disabled()) |
45c7f7e1 | 6663 | return; |
241994ed | 6664 | |
34c81057 SC |
6665 | if (!root) |
6666 | root = root_mem_cgroup; | |
22f7496f YS |
6667 | |
6668 | /* | |
6669 | * Effective values of the reclaim targets are ignored so they | |
6670 | * can be stale. Have a look at mem_cgroup_protection for more | |
6671 | * details. | |
6672 | * TODO: calculation should be more robust so that we do not need | |
6673 | * that special casing. | |
6674 | */ | |
34c81057 | 6675 | if (memcg == root) |
45c7f7e1 | 6676 | return; |
241994ed | 6677 | |
23067153 | 6678 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 | 6679 | if (!usage) |
45c7f7e1 | 6680 | return; |
bf8d5d52 | 6681 | |
bf8d5d52 | 6682 | parent = parent_mem_cgroup(memcg); |
df2a4196 RG |
6683 | /* No parent means a non-hierarchical mode on v1 memcg */ |
6684 | if (!parent) | |
45c7f7e1 | 6685 | return; |
df2a4196 | 6686 | |
bc50bcc6 | 6687 | if (parent == root) { |
c3d53200 | 6688 | memcg->memory.emin = READ_ONCE(memcg->memory.min); |
03960e33 | 6689 | memcg->memory.elow = READ_ONCE(memcg->memory.low); |
45c7f7e1 | 6690 | return; |
bf8d5d52 RG |
6691 | } |
6692 | ||
8a931f80 JW |
6693 | parent_usage = page_counter_read(&parent->memory); |
6694 | ||
b3a7822e | 6695 | WRITE_ONCE(memcg->memory.emin, effective_protection(usage, parent_usage, |
c3d53200 CD |
6696 | READ_ONCE(memcg->memory.min), |
6697 | READ_ONCE(parent->memory.emin), | |
b3a7822e | 6698 | atomic_long_read(&parent->memory.children_min_usage))); |
23067153 | 6699 | |
b3a7822e | 6700 | WRITE_ONCE(memcg->memory.elow, effective_protection(usage, parent_usage, |
03960e33 CD |
6701 | READ_ONCE(memcg->memory.low), |
6702 | READ_ONCE(parent->memory.elow), | |
b3a7822e | 6703 | atomic_long_read(&parent->memory.children_low_usage))); |
241994ed JW |
6704 | } |
6705 | ||
00501b53 | 6706 | /** |
f0e45fb4 | 6707 | * mem_cgroup_charge - charge a newly allocated page to a cgroup |
00501b53 JW |
6708 | * @page: page to charge |
6709 | * @mm: mm context of the victim | |
6710 | * @gfp_mask: reclaim mode | |
00501b53 JW |
6711 | * |
6712 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
6713 | * pages according to @gfp_mask if necessary. | |
6714 | * | |
f0e45fb4 | 6715 | * Returns 0 on success. Otherwise, an error code is returned. |
00501b53 | 6716 | */ |
d9eb1ea2 | 6717 | int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask) |
00501b53 | 6718 | { |
6c357848 | 6719 | unsigned int nr_pages = thp_nr_pages(page); |
00501b53 | 6720 | struct mem_cgroup *memcg = NULL; |
00501b53 JW |
6721 | int ret = 0; |
6722 | ||
6723 | if (mem_cgroup_disabled()) | |
6724 | goto out; | |
6725 | ||
6726 | if (PageSwapCache(page)) { | |
2d1c4980 JW |
6727 | swp_entry_t ent = { .val = page_private(page), }; |
6728 | unsigned short id; | |
6729 | ||
00501b53 JW |
6730 | /* |
6731 | * Every swap fault against a single page tries to charge the | |
6732 | * page, bail as early as possible. shmem_unuse() encounters | |
eccb52e7 JW |
6733 | * already charged pages, too. page->mem_cgroup is protected |
6734 | * by the page lock, which serializes swap cache removal, which | |
00501b53 JW |
6735 | * in turn serializes uncharging. |
6736 | */ | |
e993d905 | 6737 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
abe2895b | 6738 | if (compound_head(page)->mem_cgroup) |
00501b53 | 6739 | goto out; |
e993d905 | 6740 | |
2d1c4980 JW |
6741 | id = lookup_swap_cgroup_id(ent); |
6742 | rcu_read_lock(); | |
6743 | memcg = mem_cgroup_from_id(id); | |
6744 | if (memcg && !css_tryget_online(&memcg->css)) | |
6745 | memcg = NULL; | |
6746 | rcu_read_unlock(); | |
00501b53 JW |
6747 | } |
6748 | ||
00501b53 JW |
6749 | if (!memcg) |
6750 | memcg = get_mem_cgroup_from_mm(mm); | |
6751 | ||
6752 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
f0e45fb4 JW |
6753 | if (ret) |
6754 | goto out_put; | |
00501b53 | 6755 | |
1a3e1f40 | 6756 | css_get(&memcg->css); |
d9eb1ea2 | 6757 | commit_charge(page, memcg); |
6abb5a86 | 6758 | |
6abb5a86 | 6759 | local_irq_disable(); |
3fba69a5 | 6760 | mem_cgroup_charge_statistics(memcg, page, nr_pages); |
6abb5a86 JW |
6761 | memcg_check_events(memcg, page); |
6762 | local_irq_enable(); | |
00501b53 | 6763 | |
2d1c4980 | 6764 | if (PageSwapCache(page)) { |
00501b53 JW |
6765 | swp_entry_t entry = { .val = page_private(page) }; |
6766 | /* | |
6767 | * The swap entry might not get freed for a long time, | |
6768 | * let's not wait for it. The page already received a | |
6769 | * memory+swap charge, drop the swap entry duplicate. | |
6770 | */ | |
38d8b4e6 | 6771 | mem_cgroup_uncharge_swap(entry, nr_pages); |
00501b53 | 6772 | } |
00501b53 | 6773 | |
f0e45fb4 JW |
6774 | out_put: |
6775 | css_put(&memcg->css); | |
6776 | out: | |
6777 | return ret; | |
3fea5a49 JW |
6778 | } |
6779 | ||
a9d5adee JG |
6780 | struct uncharge_gather { |
6781 | struct mem_cgroup *memcg; | |
9f762dbe | 6782 | unsigned long nr_pages; |
a9d5adee | 6783 | unsigned long pgpgout; |
a9d5adee | 6784 | unsigned long nr_kmem; |
a9d5adee JG |
6785 | struct page *dummy_page; |
6786 | }; | |
6787 | ||
6788 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 6789 | { |
a9d5adee JG |
6790 | memset(ug, 0, sizeof(*ug)); |
6791 | } | |
6792 | ||
6793 | static void uncharge_batch(const struct uncharge_gather *ug) | |
6794 | { | |
747db954 JW |
6795 | unsigned long flags; |
6796 | ||
a9d5adee | 6797 | if (!mem_cgroup_is_root(ug->memcg)) { |
9f762dbe | 6798 | page_counter_uncharge(&ug->memcg->memory, ug->nr_pages); |
7941d214 | 6799 | if (do_memsw_account()) |
9f762dbe | 6800 | page_counter_uncharge(&ug->memcg->memsw, ug->nr_pages); |
a9d5adee JG |
6801 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem) |
6802 | page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem); | |
6803 | memcg_oom_recover(ug->memcg); | |
ce00a967 | 6804 | } |
747db954 JW |
6805 | |
6806 | local_irq_save(flags); | |
c9019e9b | 6807 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); |
9f762dbe | 6808 | __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, ug->nr_pages); |
a9d5adee | 6809 | memcg_check_events(ug->memcg, ug->dummy_page); |
747db954 | 6810 | local_irq_restore(flags); |
f1796544 MH |
6811 | |
6812 | /* drop reference from uncharge_page */ | |
6813 | css_put(&ug->memcg->css); | |
a9d5adee JG |
6814 | } |
6815 | ||
6816 | static void uncharge_page(struct page *page, struct uncharge_gather *ug) | |
6817 | { | |
9f762dbe JW |
6818 | unsigned long nr_pages; |
6819 | ||
a9d5adee | 6820 | VM_BUG_ON_PAGE(PageLRU(page), page); |
a9d5adee JG |
6821 | |
6822 | if (!page->mem_cgroup) | |
6823 | return; | |
6824 | ||
6825 | /* | |
6826 | * Nobody should be changing or seriously looking at | |
6827 | * page->mem_cgroup at this point, we have fully | |
6828 | * exclusive access to the page. | |
6829 | */ | |
6830 | ||
6831 | if (ug->memcg != page->mem_cgroup) { | |
6832 | if (ug->memcg) { | |
6833 | uncharge_batch(ug); | |
6834 | uncharge_gather_clear(ug); | |
6835 | } | |
6836 | ug->memcg = page->mem_cgroup; | |
f1796544 MH |
6837 | |
6838 | /* pairs with css_put in uncharge_batch */ | |
6839 | css_get(&ug->memcg->css); | |
a9d5adee JG |
6840 | } |
6841 | ||
9f762dbe JW |
6842 | nr_pages = compound_nr(page); |
6843 | ug->nr_pages += nr_pages; | |
a9d5adee | 6844 | |
9f762dbe | 6845 | if (!PageKmemcg(page)) { |
a9d5adee JG |
6846 | ug->pgpgout++; |
6847 | } else { | |
9f762dbe | 6848 | ug->nr_kmem += nr_pages; |
a9d5adee JG |
6849 | __ClearPageKmemcg(page); |
6850 | } | |
6851 | ||
6852 | ug->dummy_page = page; | |
6853 | page->mem_cgroup = NULL; | |
1a3e1f40 | 6854 | css_put(&ug->memcg->css); |
747db954 JW |
6855 | } |
6856 | ||
6857 | static void uncharge_list(struct list_head *page_list) | |
6858 | { | |
a9d5adee | 6859 | struct uncharge_gather ug; |
747db954 | 6860 | struct list_head *next; |
a9d5adee JG |
6861 | |
6862 | uncharge_gather_clear(&ug); | |
747db954 | 6863 | |
8b592656 JW |
6864 | /* |
6865 | * Note that the list can be a single page->lru; hence the | |
6866 | * do-while loop instead of a simple list_for_each_entry(). | |
6867 | */ | |
747db954 JW |
6868 | next = page_list->next; |
6869 | do { | |
a9d5adee JG |
6870 | struct page *page; |
6871 | ||
747db954 JW |
6872 | page = list_entry(next, struct page, lru); |
6873 | next = page->lru.next; | |
6874 | ||
a9d5adee | 6875 | uncharge_page(page, &ug); |
747db954 JW |
6876 | } while (next != page_list); |
6877 | ||
a9d5adee JG |
6878 | if (ug.memcg) |
6879 | uncharge_batch(&ug); | |
747db954 JW |
6880 | } |
6881 | ||
0a31bc97 JW |
6882 | /** |
6883 | * mem_cgroup_uncharge - uncharge a page | |
6884 | * @page: page to uncharge | |
6885 | * | |
f0e45fb4 | 6886 | * Uncharge a page previously charged with mem_cgroup_charge(). |
0a31bc97 JW |
6887 | */ |
6888 | void mem_cgroup_uncharge(struct page *page) | |
6889 | { | |
a9d5adee JG |
6890 | struct uncharge_gather ug; |
6891 | ||
0a31bc97 JW |
6892 | if (mem_cgroup_disabled()) |
6893 | return; | |
6894 | ||
747db954 | 6895 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 6896 | if (!page->mem_cgroup) |
0a31bc97 JW |
6897 | return; |
6898 | ||
a9d5adee JG |
6899 | uncharge_gather_clear(&ug); |
6900 | uncharge_page(page, &ug); | |
6901 | uncharge_batch(&ug); | |
747db954 | 6902 | } |
0a31bc97 | 6903 | |
747db954 JW |
6904 | /** |
6905 | * mem_cgroup_uncharge_list - uncharge a list of page | |
6906 | * @page_list: list of pages to uncharge | |
6907 | * | |
6908 | * Uncharge a list of pages previously charged with | |
f0e45fb4 | 6909 | * mem_cgroup_charge(). |
747db954 JW |
6910 | */ |
6911 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
6912 | { | |
6913 | if (mem_cgroup_disabled()) | |
6914 | return; | |
0a31bc97 | 6915 | |
747db954 JW |
6916 | if (!list_empty(page_list)) |
6917 | uncharge_list(page_list); | |
0a31bc97 JW |
6918 | } |
6919 | ||
6920 | /** | |
6a93ca8f JW |
6921 | * mem_cgroup_migrate - charge a page's replacement |
6922 | * @oldpage: currently circulating page | |
6923 | * @newpage: replacement page | |
0a31bc97 | 6924 | * |
6a93ca8f JW |
6925 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
6926 | * be uncharged upon free. | |
0a31bc97 JW |
6927 | * |
6928 | * Both pages must be locked, @newpage->mapping must be set up. | |
6929 | */ | |
6a93ca8f | 6930 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 6931 | { |
29833315 | 6932 | struct mem_cgroup *memcg; |
44b7a8d3 | 6933 | unsigned int nr_pages; |
d93c4130 | 6934 | unsigned long flags; |
0a31bc97 JW |
6935 | |
6936 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6937 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 6938 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
6939 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6940 | newpage); | |
0a31bc97 JW |
6941 | |
6942 | if (mem_cgroup_disabled()) | |
6943 | return; | |
6944 | ||
6945 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 6946 | if (newpage->mem_cgroup) |
0a31bc97 JW |
6947 | return; |
6948 | ||
45637bab | 6949 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 6950 | memcg = oldpage->mem_cgroup; |
29833315 | 6951 | if (!memcg) |
0a31bc97 JW |
6952 | return; |
6953 | ||
44b7a8d3 | 6954 | /* Force-charge the new page. The old one will be freed soon */ |
6c357848 | 6955 | nr_pages = thp_nr_pages(newpage); |
44b7a8d3 JW |
6956 | |
6957 | page_counter_charge(&memcg->memory, nr_pages); | |
6958 | if (do_memsw_account()) | |
6959 | page_counter_charge(&memcg->memsw, nr_pages); | |
0a31bc97 | 6960 | |
1a3e1f40 | 6961 | css_get(&memcg->css); |
d9eb1ea2 | 6962 | commit_charge(newpage, memcg); |
44b7a8d3 | 6963 | |
d93c4130 | 6964 | local_irq_save(flags); |
3fba69a5 | 6965 | mem_cgroup_charge_statistics(memcg, newpage, nr_pages); |
44b7a8d3 | 6966 | memcg_check_events(memcg, newpage); |
d93c4130 | 6967 | local_irq_restore(flags); |
0a31bc97 JW |
6968 | } |
6969 | ||
ef12947c | 6970 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
6971 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
6972 | ||
2d758073 | 6973 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
6974 | { |
6975 | struct mem_cgroup *memcg; | |
6976 | ||
2d758073 JW |
6977 | if (!mem_cgroup_sockets_enabled) |
6978 | return; | |
6979 | ||
e876ecc6 SB |
6980 | /* Do not associate the sock with unrelated interrupted task's memcg. */ |
6981 | if (in_interrupt()) | |
6982 | return; | |
6983 | ||
11092087 JW |
6984 | rcu_read_lock(); |
6985 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
6986 | if (memcg == root_mem_cgroup) |
6987 | goto out; | |
0db15298 | 6988 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 6989 | goto out; |
8965aa28 | 6990 | if (css_tryget(&memcg->css)) |
11092087 | 6991 | sk->sk_memcg = memcg; |
f7e1cb6e | 6992 | out: |
11092087 JW |
6993 | rcu_read_unlock(); |
6994 | } | |
11092087 | 6995 | |
2d758073 | 6996 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 6997 | { |
2d758073 JW |
6998 | if (sk->sk_memcg) |
6999 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
7000 | } |
7001 | ||
7002 | /** | |
7003 | * mem_cgroup_charge_skmem - charge socket memory | |
7004 | * @memcg: memcg to charge | |
7005 | * @nr_pages: number of pages to charge | |
7006 | * | |
7007 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
7008 | * @memcg's configured limit, %false if the charge had to be forced. | |
7009 | */ | |
7010 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
7011 | { | |
f7e1cb6e | 7012 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 7013 | |
f7e1cb6e | 7014 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 7015 | struct page_counter *fail; |
f7e1cb6e | 7016 | |
0db15298 JW |
7017 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
7018 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
7019 | return true; |
7020 | } | |
0db15298 JW |
7021 | page_counter_charge(&memcg->tcpmem, nr_pages); |
7022 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 7023 | return false; |
11092087 | 7024 | } |
d886f4e4 | 7025 | |
f7e1cb6e JW |
7026 | /* Don't block in the packet receive path */ |
7027 | if (in_softirq()) | |
7028 | gfp_mask = GFP_NOWAIT; | |
7029 | ||
c9019e9b | 7030 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); |
b2807f07 | 7031 | |
f7e1cb6e JW |
7032 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
7033 | return true; | |
7034 | ||
7035 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
7036 | return false; |
7037 | } | |
7038 | ||
7039 | /** | |
7040 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
7041 | * @memcg: memcg to uncharge |
7042 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
7043 | */ |
7044 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
7045 | { | |
f7e1cb6e | 7046 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 7047 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
7048 | return; |
7049 | } | |
d886f4e4 | 7050 | |
c9019e9b | 7051 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 7052 | |
475d0487 | 7053 | refill_stock(memcg, nr_pages); |
11092087 JW |
7054 | } |
7055 | ||
f7e1cb6e JW |
7056 | static int __init cgroup_memory(char *s) |
7057 | { | |
7058 | char *token; | |
7059 | ||
7060 | while ((token = strsep(&s, ",")) != NULL) { | |
7061 | if (!*token) | |
7062 | continue; | |
7063 | if (!strcmp(token, "nosocket")) | |
7064 | cgroup_memory_nosocket = true; | |
04823c83 VD |
7065 | if (!strcmp(token, "nokmem")) |
7066 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
7067 | } |
7068 | return 0; | |
7069 | } | |
7070 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 7071 | |
2d11085e | 7072 | /* |
1081312f MH |
7073 | * subsys_initcall() for memory controller. |
7074 | * | |
308167fc SAS |
7075 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
7076 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
7077 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
7078 | * should be initialized from here. | |
2d11085e MH |
7079 | */ |
7080 | static int __init mem_cgroup_init(void) | |
7081 | { | |
95a045f6 JW |
7082 | int cpu, node; |
7083 | ||
308167fc SAS |
7084 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
7085 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
7086 | |
7087 | for_each_possible_cpu(cpu) | |
7088 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
7089 | drain_local_stock); | |
7090 | ||
7091 | for_each_node(node) { | |
7092 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
7093 | |
7094 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
7095 | node_online(node) ? node : NUMA_NO_NODE); | |
7096 | ||
ef8f2327 | 7097 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 7098 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 7099 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
7100 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
7101 | } | |
7102 | ||
2d11085e MH |
7103 | return 0; |
7104 | } | |
7105 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
7106 | |
7107 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
7108 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
7109 | { | |
1c2d479a | 7110 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
7111 | /* |
7112 | * The root cgroup cannot be destroyed, so it's refcount must | |
7113 | * always be >= 1. | |
7114 | */ | |
7115 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
7116 | VM_BUG_ON(1); | |
7117 | break; | |
7118 | } | |
7119 | memcg = parent_mem_cgroup(memcg); | |
7120 | if (!memcg) | |
7121 | memcg = root_mem_cgroup; | |
7122 | } | |
7123 | return memcg; | |
7124 | } | |
7125 | ||
21afa38e JW |
7126 | /** |
7127 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
7128 | * @page: page whose memsw charge to transfer | |
7129 | * @entry: swap entry to move the charge to | |
7130 | * | |
7131 | * Transfer the memsw charge of @page to @entry. | |
7132 | */ | |
7133 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
7134 | { | |
1f47b61f | 7135 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 7136 | unsigned int nr_entries; |
21afa38e JW |
7137 | unsigned short oldid; |
7138 | ||
7139 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
7140 | VM_BUG_ON_PAGE(page_count(page), page); | |
7141 | ||
2d1c4980 | 7142 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
21afa38e JW |
7143 | return; |
7144 | ||
7145 | memcg = page->mem_cgroup; | |
7146 | ||
7147 | /* Readahead page, never charged */ | |
7148 | if (!memcg) | |
7149 | return; | |
7150 | ||
1f47b61f VD |
7151 | /* |
7152 | * In case the memcg owning these pages has been offlined and doesn't | |
7153 | * have an ID allocated to it anymore, charge the closest online | |
7154 | * ancestor for the swap instead and transfer the memory+swap charge. | |
7155 | */ | |
7156 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
6c357848 | 7157 | nr_entries = thp_nr_pages(page); |
d6810d73 HY |
7158 | /* Get references for the tail pages, too */ |
7159 | if (nr_entries > 1) | |
7160 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
7161 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
7162 | nr_entries); | |
21afa38e | 7163 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7164 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e JW |
7165 | |
7166 | page->mem_cgroup = NULL; | |
7167 | ||
7168 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 7169 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 7170 | |
2d1c4980 | 7171 | if (!cgroup_memory_noswap && memcg != swap_memcg) { |
1f47b61f | 7172 | if (!mem_cgroup_is_root(swap_memcg)) |
d6810d73 HY |
7173 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
7174 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
7175 | } |
7176 | ||
ce9ce665 SAS |
7177 | /* |
7178 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 7179 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 7180 | * important here to have the interrupts disabled because it is the |
b93b0163 | 7181 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 SAS |
7182 | */ |
7183 | VM_BUG_ON(!irqs_disabled()); | |
3fba69a5 | 7184 | mem_cgroup_charge_statistics(memcg, page, -nr_entries); |
21afa38e | 7185 | memcg_check_events(memcg, page); |
73f576c0 | 7186 | |
1a3e1f40 | 7187 | css_put(&memcg->css); |
21afa38e JW |
7188 | } |
7189 | ||
38d8b4e6 HY |
7190 | /** |
7191 | * mem_cgroup_try_charge_swap - try charging swap space for a page | |
37e84351 VD |
7192 | * @page: page being added to swap |
7193 | * @entry: swap entry to charge | |
7194 | * | |
38d8b4e6 | 7195 | * Try to charge @page's memcg for the swap space at @entry. |
37e84351 VD |
7196 | * |
7197 | * Returns 0 on success, -ENOMEM on failure. | |
7198 | */ | |
7199 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
7200 | { | |
6c357848 | 7201 | unsigned int nr_pages = thp_nr_pages(page); |
37e84351 | 7202 | struct page_counter *counter; |
38d8b4e6 | 7203 | struct mem_cgroup *memcg; |
37e84351 VD |
7204 | unsigned short oldid; |
7205 | ||
2d1c4980 | 7206 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
37e84351 VD |
7207 | return 0; |
7208 | ||
7209 | memcg = page->mem_cgroup; | |
7210 | ||
7211 | /* Readahead page, never charged */ | |
7212 | if (!memcg) | |
7213 | return 0; | |
7214 | ||
f3a53a3a TH |
7215 | if (!entry.val) { |
7216 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 7217 | return 0; |
f3a53a3a | 7218 | } |
bb98f2c5 | 7219 | |
1f47b61f VD |
7220 | memcg = mem_cgroup_id_get_online(memcg); |
7221 | ||
2d1c4980 | 7222 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg) && |
38d8b4e6 | 7223 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
7224 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
7225 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 7226 | mem_cgroup_id_put(memcg); |
37e84351 | 7227 | return -ENOMEM; |
1f47b61f | 7228 | } |
37e84351 | 7229 | |
38d8b4e6 HY |
7230 | /* Get references for the tail pages, too */ |
7231 | if (nr_pages > 1) | |
7232 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
7233 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
37e84351 | 7234 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7235 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 7236 | |
37e84351 VD |
7237 | return 0; |
7238 | } | |
7239 | ||
21afa38e | 7240 | /** |
38d8b4e6 | 7241 | * mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 7242 | * @entry: swap entry to uncharge |
38d8b4e6 | 7243 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 7244 | */ |
38d8b4e6 | 7245 | void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
7246 | { |
7247 | struct mem_cgroup *memcg; | |
7248 | unsigned short id; | |
7249 | ||
38d8b4e6 | 7250 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 7251 | rcu_read_lock(); |
adbe427b | 7252 | memcg = mem_cgroup_from_id(id); |
21afa38e | 7253 | if (memcg) { |
2d1c4980 | 7254 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg)) { |
37e84351 | 7255 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
38d8b4e6 | 7256 | page_counter_uncharge(&memcg->swap, nr_pages); |
37e84351 | 7257 | else |
38d8b4e6 | 7258 | page_counter_uncharge(&memcg->memsw, nr_pages); |
37e84351 | 7259 | } |
c9019e9b | 7260 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 7261 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
7262 | } |
7263 | rcu_read_unlock(); | |
7264 | } | |
7265 | ||
d8b38438 VD |
7266 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
7267 | { | |
7268 | long nr_swap_pages = get_nr_swap_pages(); | |
7269 | ||
eccb52e7 | 7270 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
d8b38438 VD |
7271 | return nr_swap_pages; |
7272 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
7273 | nr_swap_pages = min_t(long, nr_swap_pages, | |
bbec2e15 | 7274 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
7275 | page_counter_read(&memcg->swap)); |
7276 | return nr_swap_pages; | |
7277 | } | |
7278 | ||
5ccc5aba VD |
7279 | bool mem_cgroup_swap_full(struct page *page) |
7280 | { | |
7281 | struct mem_cgroup *memcg; | |
7282 | ||
7283 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
7284 | ||
7285 | if (vm_swap_full()) | |
7286 | return true; | |
eccb52e7 | 7287 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
5ccc5aba VD |
7288 | return false; |
7289 | ||
7290 | memcg = page->mem_cgroup; | |
7291 | if (!memcg) | |
7292 | return false; | |
7293 | ||
4b82ab4f JK |
7294 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { |
7295 | unsigned long usage = page_counter_read(&memcg->swap); | |
7296 | ||
7297 | if (usage * 2 >= READ_ONCE(memcg->swap.high) || | |
7298 | usage * 2 >= READ_ONCE(memcg->swap.max)) | |
5ccc5aba | 7299 | return true; |
4b82ab4f | 7300 | } |
5ccc5aba VD |
7301 | |
7302 | return false; | |
7303 | } | |
7304 | ||
eccb52e7 | 7305 | static int __init setup_swap_account(char *s) |
21afa38e JW |
7306 | { |
7307 | if (!strcmp(s, "1")) | |
eccb52e7 | 7308 | cgroup_memory_noswap = 0; |
21afa38e | 7309 | else if (!strcmp(s, "0")) |
eccb52e7 | 7310 | cgroup_memory_noswap = 1; |
21afa38e JW |
7311 | return 1; |
7312 | } | |
eccb52e7 | 7313 | __setup("swapaccount=", setup_swap_account); |
21afa38e | 7314 | |
37e84351 VD |
7315 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
7316 | struct cftype *cft) | |
7317 | { | |
7318 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
7319 | ||
7320 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
7321 | } | |
7322 | ||
4b82ab4f JK |
7323 | static int swap_high_show(struct seq_file *m, void *v) |
7324 | { | |
7325 | return seq_puts_memcg_tunable(m, | |
7326 | READ_ONCE(mem_cgroup_from_seq(m)->swap.high)); | |
7327 | } | |
7328 | ||
7329 | static ssize_t swap_high_write(struct kernfs_open_file *of, | |
7330 | char *buf, size_t nbytes, loff_t off) | |
7331 | { | |
7332 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7333 | unsigned long high; | |
7334 | int err; | |
7335 | ||
7336 | buf = strstrip(buf); | |
7337 | err = page_counter_memparse(buf, "max", &high); | |
7338 | if (err) | |
7339 | return err; | |
7340 | ||
7341 | page_counter_set_high(&memcg->swap, high); | |
7342 | ||
7343 | return nbytes; | |
7344 | } | |
7345 | ||
37e84351 VD |
7346 | static int swap_max_show(struct seq_file *m, void *v) |
7347 | { | |
677dc973 CD |
7348 | return seq_puts_memcg_tunable(m, |
7349 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); | |
37e84351 VD |
7350 | } |
7351 | ||
7352 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
7353 | char *buf, size_t nbytes, loff_t off) | |
7354 | { | |
7355 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7356 | unsigned long max; | |
7357 | int err; | |
7358 | ||
7359 | buf = strstrip(buf); | |
7360 | err = page_counter_memparse(buf, "max", &max); | |
7361 | if (err) | |
7362 | return err; | |
7363 | ||
be09102b | 7364 | xchg(&memcg->swap.max, max); |
37e84351 VD |
7365 | |
7366 | return nbytes; | |
7367 | } | |
7368 | ||
f3a53a3a TH |
7369 | static int swap_events_show(struct seq_file *m, void *v) |
7370 | { | |
aa9694bb | 7371 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
f3a53a3a | 7372 | |
4b82ab4f JK |
7373 | seq_printf(m, "high %lu\n", |
7374 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_HIGH])); | |
f3a53a3a TH |
7375 | seq_printf(m, "max %lu\n", |
7376 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
7377 | seq_printf(m, "fail %lu\n", | |
7378 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
7379 | ||
7380 | return 0; | |
7381 | } | |
7382 | ||
37e84351 VD |
7383 | static struct cftype swap_files[] = { |
7384 | { | |
7385 | .name = "swap.current", | |
7386 | .flags = CFTYPE_NOT_ON_ROOT, | |
7387 | .read_u64 = swap_current_read, | |
7388 | }, | |
4b82ab4f JK |
7389 | { |
7390 | .name = "swap.high", | |
7391 | .flags = CFTYPE_NOT_ON_ROOT, | |
7392 | .seq_show = swap_high_show, | |
7393 | .write = swap_high_write, | |
7394 | }, | |
37e84351 VD |
7395 | { |
7396 | .name = "swap.max", | |
7397 | .flags = CFTYPE_NOT_ON_ROOT, | |
7398 | .seq_show = swap_max_show, | |
7399 | .write = swap_max_write, | |
7400 | }, | |
f3a53a3a TH |
7401 | { |
7402 | .name = "swap.events", | |
7403 | .flags = CFTYPE_NOT_ON_ROOT, | |
7404 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
7405 | .seq_show = swap_events_show, | |
7406 | }, | |
37e84351 VD |
7407 | { } /* terminate */ |
7408 | }; | |
7409 | ||
eccb52e7 | 7410 | static struct cftype memsw_files[] = { |
21afa38e JW |
7411 | { |
7412 | .name = "memsw.usage_in_bytes", | |
7413 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
7414 | .read_u64 = mem_cgroup_read_u64, | |
7415 | }, | |
7416 | { | |
7417 | .name = "memsw.max_usage_in_bytes", | |
7418 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
7419 | .write = mem_cgroup_reset, | |
7420 | .read_u64 = mem_cgroup_read_u64, | |
7421 | }, | |
7422 | { | |
7423 | .name = "memsw.limit_in_bytes", | |
7424 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
7425 | .write = mem_cgroup_write, | |
7426 | .read_u64 = mem_cgroup_read_u64, | |
7427 | }, | |
7428 | { | |
7429 | .name = "memsw.failcnt", | |
7430 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
7431 | .write = mem_cgroup_reset, | |
7432 | .read_u64 = mem_cgroup_read_u64, | |
7433 | }, | |
7434 | { }, /* terminate */ | |
7435 | }; | |
7436 | ||
82ff165c BS |
7437 | /* |
7438 | * If mem_cgroup_swap_init() is implemented as a subsys_initcall() | |
7439 | * instead of a core_initcall(), this could mean cgroup_memory_noswap still | |
7440 | * remains set to false even when memcg is disabled via "cgroup_disable=memory" | |
7441 | * boot parameter. This may result in premature OOPS inside | |
7442 | * mem_cgroup_get_nr_swap_pages() function in corner cases. | |
7443 | */ | |
21afa38e JW |
7444 | static int __init mem_cgroup_swap_init(void) |
7445 | { | |
2d1c4980 JW |
7446 | /* No memory control -> no swap control */ |
7447 | if (mem_cgroup_disabled()) | |
7448 | cgroup_memory_noswap = true; | |
7449 | ||
7450 | if (cgroup_memory_noswap) | |
eccb52e7 JW |
7451 | return 0; |
7452 | ||
7453 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, swap_files)); | |
7454 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, memsw_files)); | |
7455 | ||
21afa38e JW |
7456 | return 0; |
7457 | } | |
82ff165c | 7458 | core_initcall(mem_cgroup_swap_init); |
21afa38e JW |
7459 | |
7460 | #endif /* CONFIG_MEMCG_SWAP */ |