]>
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 | |
217bc319 KH |
200 | enum charge_type { |
201 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 202 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 203 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 204 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
205 | NR_CHARGE_TYPE, |
206 | }; | |
207 | ||
8c7c6e34 | 208 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
209 | enum res_type { |
210 | _MEM, | |
211 | _MEMSWAP, | |
212 | _OOM_TYPE, | |
510fc4e1 | 213 | _KMEM, |
d55f90bf | 214 | _TCP, |
86ae53e1 GC |
215 | }; |
216 | ||
a0db00fc KS |
217 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
218 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 219 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
220 | /* Used for OOM nofiier */ |
221 | #define OOM_CONTROL (0) | |
8c7c6e34 | 222 | |
b05706f1 KT |
223 | /* |
224 | * Iteration constructs for visiting all cgroups (under a tree). If | |
225 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
226 | * be used for reference counting. | |
227 | */ | |
228 | #define for_each_mem_cgroup_tree(iter, root) \ | |
229 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ | |
230 | iter != NULL; \ | |
231 | iter = mem_cgroup_iter(root, iter, NULL)) | |
232 | ||
233 | #define for_each_mem_cgroup(iter) \ | |
234 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ | |
235 | iter != NULL; \ | |
236 | iter = mem_cgroup_iter(NULL, iter, NULL)) | |
237 | ||
7775face TH |
238 | static inline bool should_force_charge(void) |
239 | { | |
240 | return tsk_is_oom_victim(current) || fatal_signal_pending(current) || | |
241 | (current->flags & PF_EXITING); | |
242 | } | |
243 | ||
70ddf637 AV |
244 | /* Some nice accessors for the vmpressure. */ |
245 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
246 | { | |
247 | if (!memcg) | |
248 | memcg = root_mem_cgroup; | |
249 | return &memcg->vmpressure; | |
250 | } | |
251 | ||
252 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
253 | { | |
254 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
255 | } | |
256 | ||
84c07d11 | 257 | #ifdef CONFIG_MEMCG_KMEM |
bf4f0599 RG |
258 | extern spinlock_t css_set_lock; |
259 | ||
260 | static void obj_cgroup_release(struct percpu_ref *ref) | |
261 | { | |
262 | struct obj_cgroup *objcg = container_of(ref, struct obj_cgroup, refcnt); | |
263 | struct mem_cgroup *memcg; | |
264 | unsigned int nr_bytes; | |
265 | unsigned int nr_pages; | |
266 | unsigned long flags; | |
267 | ||
268 | /* | |
269 | * At this point all allocated objects are freed, and | |
270 | * objcg->nr_charged_bytes can't have an arbitrary byte value. | |
271 | * However, it can be PAGE_SIZE or (x * PAGE_SIZE). | |
272 | * | |
273 | * The following sequence can lead to it: | |
274 | * 1) CPU0: objcg == stock->cached_objcg | |
275 | * 2) CPU1: we do a small allocation (e.g. 92 bytes), | |
276 | * PAGE_SIZE bytes are charged | |
277 | * 3) CPU1: a process from another memcg is allocating something, | |
278 | * the stock if flushed, | |
279 | * objcg->nr_charged_bytes = PAGE_SIZE - 92 | |
280 | * 5) CPU0: we do release this object, | |
281 | * 92 bytes are added to stock->nr_bytes | |
282 | * 6) CPU0: stock is flushed, | |
283 | * 92 bytes are added to objcg->nr_charged_bytes | |
284 | * | |
285 | * In the result, nr_charged_bytes == PAGE_SIZE. | |
286 | * This page will be uncharged in obj_cgroup_release(). | |
287 | */ | |
288 | nr_bytes = atomic_read(&objcg->nr_charged_bytes); | |
289 | WARN_ON_ONCE(nr_bytes & (PAGE_SIZE - 1)); | |
290 | nr_pages = nr_bytes >> PAGE_SHIFT; | |
291 | ||
292 | spin_lock_irqsave(&css_set_lock, flags); | |
293 | memcg = obj_cgroup_memcg(objcg); | |
294 | if (nr_pages) | |
295 | __memcg_kmem_uncharge(memcg, nr_pages); | |
296 | list_del(&objcg->list); | |
297 | mem_cgroup_put(memcg); | |
298 | spin_unlock_irqrestore(&css_set_lock, flags); | |
299 | ||
300 | percpu_ref_exit(ref); | |
301 | kfree_rcu(objcg, rcu); | |
302 | } | |
303 | ||
304 | static struct obj_cgroup *obj_cgroup_alloc(void) | |
305 | { | |
306 | struct obj_cgroup *objcg; | |
307 | int ret; | |
308 | ||
309 | objcg = kzalloc(sizeof(struct obj_cgroup), GFP_KERNEL); | |
310 | if (!objcg) | |
311 | return NULL; | |
312 | ||
313 | ret = percpu_ref_init(&objcg->refcnt, obj_cgroup_release, 0, | |
314 | GFP_KERNEL); | |
315 | if (ret) { | |
316 | kfree(objcg); | |
317 | return NULL; | |
318 | } | |
319 | INIT_LIST_HEAD(&objcg->list); | |
320 | return objcg; | |
321 | } | |
322 | ||
323 | static void memcg_reparent_objcgs(struct mem_cgroup *memcg, | |
324 | struct mem_cgroup *parent) | |
325 | { | |
326 | struct obj_cgroup *objcg, *iter; | |
327 | ||
328 | objcg = rcu_replace_pointer(memcg->objcg, NULL, true); | |
329 | ||
330 | spin_lock_irq(&css_set_lock); | |
331 | ||
332 | /* Move active objcg to the parent's list */ | |
333 | xchg(&objcg->memcg, parent); | |
334 | css_get(&parent->css); | |
335 | list_add(&objcg->list, &parent->objcg_list); | |
336 | ||
337 | /* Move already reparented objcgs to the parent's list */ | |
338 | list_for_each_entry(iter, &memcg->objcg_list, list) { | |
339 | css_get(&parent->css); | |
340 | xchg(&iter->memcg, parent); | |
341 | css_put(&memcg->css); | |
342 | } | |
343 | list_splice(&memcg->objcg_list, &parent->objcg_list); | |
344 | ||
345 | spin_unlock_irq(&css_set_lock); | |
346 | ||
347 | percpu_ref_kill(&objcg->refcnt); | |
348 | } | |
349 | ||
55007d84 | 350 | /* |
9855609b | 351 | * This will be used as a shrinker list's index. |
b8627835 LZ |
352 | * The main reason for not using cgroup id for this: |
353 | * this works better in sparse environments, where we have a lot of memcgs, | |
354 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
355 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
356 | * 200 entry array for that. | |
55007d84 | 357 | * |
dbcf73e2 VD |
358 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
359 | * will double each time we have to increase it. | |
55007d84 | 360 | */ |
dbcf73e2 VD |
361 | static DEFINE_IDA(memcg_cache_ida); |
362 | int memcg_nr_cache_ids; | |
749c5415 | 363 | |
05257a1a VD |
364 | /* Protects memcg_nr_cache_ids */ |
365 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
366 | ||
367 | void memcg_get_cache_ids(void) | |
368 | { | |
369 | down_read(&memcg_cache_ids_sem); | |
370 | } | |
371 | ||
372 | void memcg_put_cache_ids(void) | |
373 | { | |
374 | up_read(&memcg_cache_ids_sem); | |
375 | } | |
376 | ||
55007d84 GC |
377 | /* |
378 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
379 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
380 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
381 | * tunable, but that is strictly not necessary. | |
382 | * | |
b8627835 | 383 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
384 | * this constant directly from cgroup, but it is understandable that this is |
385 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 386 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
387 | * increase ours as well if it increases. |
388 | */ | |
389 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 390 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 391 | |
d7f25f8a GC |
392 | /* |
393 | * A lot of the calls to the cache allocation functions are expected to be | |
272911a4 | 394 | * inlined by the compiler. Since the calls to memcg_slab_pre_alloc_hook() are |
d7f25f8a GC |
395 | * conditional to this static branch, we'll have to allow modules that does |
396 | * kmem_cache_alloc and the such to see this symbol as well | |
397 | */ | |
ef12947c | 398 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
d7f25f8a | 399 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
0a432dcb | 400 | #endif |
17cc4dfe | 401 | |
0a4465d3 KT |
402 | static int memcg_shrinker_map_size; |
403 | static DEFINE_MUTEX(memcg_shrinker_map_mutex); | |
404 | ||
405 | static void memcg_free_shrinker_map_rcu(struct rcu_head *head) | |
406 | { | |
407 | kvfree(container_of(head, struct memcg_shrinker_map, rcu)); | |
408 | } | |
409 | ||
410 | static int memcg_expand_one_shrinker_map(struct mem_cgroup *memcg, | |
411 | int size, int old_size) | |
412 | { | |
413 | struct memcg_shrinker_map *new, *old; | |
414 | int nid; | |
415 | ||
416 | lockdep_assert_held(&memcg_shrinker_map_mutex); | |
417 | ||
418 | for_each_node(nid) { | |
419 | old = rcu_dereference_protected( | |
420 | mem_cgroup_nodeinfo(memcg, nid)->shrinker_map, true); | |
421 | /* Not yet online memcg */ | |
422 | if (!old) | |
423 | return 0; | |
424 | ||
86daf94e | 425 | new = kvmalloc_node(sizeof(*new) + size, GFP_KERNEL, nid); |
0a4465d3 KT |
426 | if (!new) |
427 | return -ENOMEM; | |
428 | ||
429 | /* Set all old bits, clear all new bits */ | |
430 | memset(new->map, (int)0xff, old_size); | |
431 | memset((void *)new->map + old_size, 0, size - old_size); | |
432 | ||
433 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, new); | |
434 | call_rcu(&old->rcu, memcg_free_shrinker_map_rcu); | |
435 | } | |
436 | ||
437 | return 0; | |
438 | } | |
439 | ||
440 | static void memcg_free_shrinker_maps(struct mem_cgroup *memcg) | |
441 | { | |
442 | struct mem_cgroup_per_node *pn; | |
443 | struct memcg_shrinker_map *map; | |
444 | int nid; | |
445 | ||
446 | if (mem_cgroup_is_root(memcg)) | |
447 | return; | |
448 | ||
449 | for_each_node(nid) { | |
450 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
451 | map = rcu_dereference_protected(pn->shrinker_map, true); | |
452 | if (map) | |
453 | kvfree(map); | |
454 | rcu_assign_pointer(pn->shrinker_map, NULL); | |
455 | } | |
456 | } | |
457 | ||
458 | static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg) | |
459 | { | |
460 | struct memcg_shrinker_map *map; | |
461 | int nid, size, ret = 0; | |
462 | ||
463 | if (mem_cgroup_is_root(memcg)) | |
464 | return 0; | |
465 | ||
466 | mutex_lock(&memcg_shrinker_map_mutex); | |
467 | size = memcg_shrinker_map_size; | |
468 | for_each_node(nid) { | |
86daf94e | 469 | map = kvzalloc_node(sizeof(*map) + size, GFP_KERNEL, nid); |
0a4465d3 KT |
470 | if (!map) { |
471 | memcg_free_shrinker_maps(memcg); | |
472 | ret = -ENOMEM; | |
473 | break; | |
474 | } | |
475 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, map); | |
476 | } | |
477 | mutex_unlock(&memcg_shrinker_map_mutex); | |
478 | ||
479 | return ret; | |
480 | } | |
481 | ||
482 | int memcg_expand_shrinker_maps(int new_id) | |
483 | { | |
484 | int size, old_size, ret = 0; | |
485 | struct mem_cgroup *memcg; | |
486 | ||
487 | size = DIV_ROUND_UP(new_id + 1, BITS_PER_LONG) * sizeof(unsigned long); | |
488 | old_size = memcg_shrinker_map_size; | |
489 | if (size <= old_size) | |
490 | return 0; | |
491 | ||
492 | mutex_lock(&memcg_shrinker_map_mutex); | |
493 | if (!root_mem_cgroup) | |
494 | goto unlock; | |
495 | ||
496 | for_each_mem_cgroup(memcg) { | |
497 | if (mem_cgroup_is_root(memcg)) | |
498 | continue; | |
499 | ret = memcg_expand_one_shrinker_map(memcg, size, old_size); | |
75866af6 VA |
500 | if (ret) { |
501 | mem_cgroup_iter_break(NULL, memcg); | |
0a4465d3 | 502 | goto unlock; |
75866af6 | 503 | } |
0a4465d3 KT |
504 | } |
505 | unlock: | |
506 | if (!ret) | |
507 | memcg_shrinker_map_size = size; | |
508 | mutex_unlock(&memcg_shrinker_map_mutex); | |
509 | return ret; | |
510 | } | |
fae91d6d KT |
511 | |
512 | void memcg_set_shrinker_bit(struct mem_cgroup *memcg, int nid, int shrinker_id) | |
513 | { | |
514 | if (shrinker_id >= 0 && memcg && !mem_cgroup_is_root(memcg)) { | |
515 | struct memcg_shrinker_map *map; | |
516 | ||
517 | rcu_read_lock(); | |
518 | map = rcu_dereference(memcg->nodeinfo[nid]->shrinker_map); | |
f90280d6 KT |
519 | /* Pairs with smp mb in shrink_slab() */ |
520 | smp_mb__before_atomic(); | |
fae91d6d KT |
521 | set_bit(shrinker_id, map->map); |
522 | rcu_read_unlock(); | |
523 | } | |
524 | } | |
525 | ||
ad7fa852 TH |
526 | /** |
527 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
528 | * @page: page of interest | |
529 | * | |
530 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
531 | * with @page is returned. The returned css remains associated with @page | |
532 | * until it is released. | |
533 | * | |
534 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
535 | * is returned. | |
ad7fa852 TH |
536 | */ |
537 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
538 | { | |
539 | struct mem_cgroup *memcg; | |
540 | ||
ad7fa852 TH |
541 | memcg = page->mem_cgroup; |
542 | ||
9e10a130 | 543 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
544 | memcg = root_mem_cgroup; |
545 | ||
ad7fa852 TH |
546 | return &memcg->css; |
547 | } | |
548 | ||
2fc04524 VD |
549 | /** |
550 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
551 | * @page: the page | |
552 | * | |
553 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
554 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
555 | * is safe to call this function without holding a reference to @page. | |
556 | * | |
557 | * Note, this function is inherently racy, because there is nothing to prevent | |
558 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
559 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
560 | * do not care (such as procfs interfaces). | |
561 | */ | |
562 | ino_t page_cgroup_ino(struct page *page) | |
563 | { | |
564 | struct mem_cgroup *memcg; | |
565 | unsigned long ino = 0; | |
566 | ||
567 | rcu_read_lock(); | |
9855609b | 568 | memcg = page->mem_cgroup; |
286e04b8 | 569 | |
9855609b RG |
570 | /* |
571 | * The lowest bit set means that memcg isn't a valid | |
572 | * memcg pointer, but a obj_cgroups pointer. | |
573 | * In this case the page is shared and doesn't belong | |
574 | * to any specific memory cgroup. | |
575 | */ | |
576 | if ((unsigned long) memcg & 0x1UL) | |
577 | memcg = NULL; | |
286e04b8 | 578 | |
2fc04524 VD |
579 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) |
580 | memcg = parent_mem_cgroup(memcg); | |
581 | if (memcg) | |
582 | ino = cgroup_ino(memcg->css.cgroup); | |
583 | rcu_read_unlock(); | |
584 | return ino; | |
585 | } | |
586 | ||
ef8f2327 MG |
587 | static struct mem_cgroup_per_node * |
588 | mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page) | |
f64c3f54 | 589 | { |
97a6c37b | 590 | int nid = page_to_nid(page); |
f64c3f54 | 591 | |
ef8f2327 | 592 | return memcg->nodeinfo[nid]; |
f64c3f54 BS |
593 | } |
594 | ||
ef8f2327 MG |
595 | static struct mem_cgroup_tree_per_node * |
596 | soft_limit_tree_node(int nid) | |
bb4cc1a8 | 597 | { |
ef8f2327 | 598 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
599 | } |
600 | ||
ef8f2327 | 601 | static struct mem_cgroup_tree_per_node * |
bb4cc1a8 AM |
602 | soft_limit_tree_from_page(struct page *page) |
603 | { | |
604 | int nid = page_to_nid(page); | |
bb4cc1a8 | 605 | |
ef8f2327 | 606 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
607 | } |
608 | ||
ef8f2327 MG |
609 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
610 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 611 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
612 | { |
613 | struct rb_node **p = &mctz->rb_root.rb_node; | |
614 | struct rb_node *parent = NULL; | |
ef8f2327 | 615 | struct mem_cgroup_per_node *mz_node; |
fa90b2fd | 616 | bool rightmost = true; |
bb4cc1a8 AM |
617 | |
618 | if (mz->on_tree) | |
619 | return; | |
620 | ||
621 | mz->usage_in_excess = new_usage_in_excess; | |
622 | if (!mz->usage_in_excess) | |
623 | return; | |
624 | while (*p) { | |
625 | parent = *p; | |
ef8f2327 | 626 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 | 627 | tree_node); |
fa90b2fd | 628 | if (mz->usage_in_excess < mz_node->usage_in_excess) { |
bb4cc1a8 | 629 | p = &(*p)->rb_left; |
fa90b2fd DB |
630 | rightmost = false; |
631 | } | |
632 | ||
bb4cc1a8 AM |
633 | /* |
634 | * We can't avoid mem cgroups that are over their soft | |
635 | * limit by the same amount | |
636 | */ | |
637 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
638 | p = &(*p)->rb_right; | |
639 | } | |
fa90b2fd DB |
640 | |
641 | if (rightmost) | |
642 | mctz->rb_rightmost = &mz->tree_node; | |
643 | ||
bb4cc1a8 AM |
644 | rb_link_node(&mz->tree_node, parent, p); |
645 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
646 | mz->on_tree = true; | |
647 | } | |
648 | ||
ef8f2327 MG |
649 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
650 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
651 | { |
652 | if (!mz->on_tree) | |
653 | return; | |
fa90b2fd DB |
654 | |
655 | if (&mz->tree_node == mctz->rb_rightmost) | |
656 | mctz->rb_rightmost = rb_prev(&mz->tree_node); | |
657 | ||
bb4cc1a8 AM |
658 | rb_erase(&mz->tree_node, &mctz->rb_root); |
659 | mz->on_tree = false; | |
660 | } | |
661 | ||
ef8f2327 MG |
662 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
663 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 664 | { |
0a31bc97 JW |
665 | unsigned long flags; |
666 | ||
667 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 668 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 669 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
670 | } |
671 | ||
3e32cb2e JW |
672 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
673 | { | |
674 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 675 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
676 | unsigned long excess = 0; |
677 | ||
678 | if (nr_pages > soft_limit) | |
679 | excess = nr_pages - soft_limit; | |
680 | ||
681 | return excess; | |
682 | } | |
bb4cc1a8 AM |
683 | |
684 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
685 | { | |
3e32cb2e | 686 | unsigned long excess; |
ef8f2327 MG |
687 | struct mem_cgroup_per_node *mz; |
688 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 689 | |
e231875b | 690 | mctz = soft_limit_tree_from_page(page); |
bfc7228b LD |
691 | if (!mctz) |
692 | return; | |
bb4cc1a8 AM |
693 | /* |
694 | * Necessary to update all ancestors when hierarchy is used. | |
695 | * because their event counter is not touched. | |
696 | */ | |
697 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
ef8f2327 | 698 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
3e32cb2e | 699 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
700 | /* |
701 | * We have to update the tree if mz is on RB-tree or | |
702 | * mem is over its softlimit. | |
703 | */ | |
704 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
705 | unsigned long flags; |
706 | ||
707 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
708 | /* if on-tree, remove it */ |
709 | if (mz->on_tree) | |
cf2c8127 | 710 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
711 | /* |
712 | * Insert again. mz->usage_in_excess will be updated. | |
713 | * If excess is 0, no tree ops. | |
714 | */ | |
cf2c8127 | 715 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 716 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
717 | } |
718 | } | |
719 | } | |
720 | ||
721 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
722 | { | |
ef8f2327 MG |
723 | struct mem_cgroup_tree_per_node *mctz; |
724 | struct mem_cgroup_per_node *mz; | |
725 | int nid; | |
bb4cc1a8 | 726 | |
e231875b | 727 | for_each_node(nid) { |
ef8f2327 MG |
728 | mz = mem_cgroup_nodeinfo(memcg, nid); |
729 | mctz = soft_limit_tree_node(nid); | |
bfc7228b LD |
730 | if (mctz) |
731 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
732 | } |
733 | } | |
734 | ||
ef8f2327 MG |
735 | static struct mem_cgroup_per_node * |
736 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 737 | { |
ef8f2327 | 738 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
739 | |
740 | retry: | |
741 | mz = NULL; | |
fa90b2fd | 742 | if (!mctz->rb_rightmost) |
bb4cc1a8 AM |
743 | goto done; /* Nothing to reclaim from */ |
744 | ||
fa90b2fd DB |
745 | mz = rb_entry(mctz->rb_rightmost, |
746 | struct mem_cgroup_per_node, tree_node); | |
bb4cc1a8 AM |
747 | /* |
748 | * Remove the node now but someone else can add it back, | |
749 | * we will to add it back at the end of reclaim to its correct | |
750 | * position in the tree. | |
751 | */ | |
cf2c8127 | 752 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 753 | if (!soft_limit_excess(mz->memcg) || |
8965aa28 | 754 | !css_tryget(&mz->memcg->css)) |
bb4cc1a8 AM |
755 | goto retry; |
756 | done: | |
757 | return mz; | |
758 | } | |
759 | ||
ef8f2327 MG |
760 | static struct mem_cgroup_per_node * |
761 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 762 | { |
ef8f2327 | 763 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 764 | |
0a31bc97 | 765 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 766 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 767 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
768 | return mz; |
769 | } | |
770 | ||
db9adbcb JW |
771 | /** |
772 | * __mod_memcg_state - update cgroup memory statistics | |
773 | * @memcg: the memory cgroup | |
774 | * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item | |
775 | * @val: delta to add to the counter, can be negative | |
776 | */ | |
777 | void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val) | |
778 | { | |
ea426c2a | 779 | long x, threshold = MEMCG_CHARGE_BATCH; |
db9adbcb JW |
780 | |
781 | if (mem_cgroup_disabled()) | |
782 | return; | |
783 | ||
ea426c2a RG |
784 | if (vmstat_item_in_bytes(idx)) |
785 | threshold <<= PAGE_SHIFT; | |
786 | ||
db9adbcb | 787 | x = val + __this_cpu_read(memcg->vmstats_percpu->stat[idx]); |
ea426c2a | 788 | if (unlikely(abs(x) > threshold)) { |
42a30035 JW |
789 | struct mem_cgroup *mi; |
790 | ||
766a4c19 YS |
791 | /* |
792 | * Batch local counters to keep them in sync with | |
793 | * the hierarchical ones. | |
794 | */ | |
795 | __this_cpu_add(memcg->vmstats_local->stat[idx], x); | |
42a30035 JW |
796 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
797 | atomic_long_add(x, &mi->vmstats[idx]); | |
db9adbcb JW |
798 | x = 0; |
799 | } | |
800 | __this_cpu_write(memcg->vmstats_percpu->stat[idx], x); | |
801 | } | |
802 | ||
42a30035 JW |
803 | static struct mem_cgroup_per_node * |
804 | parent_nodeinfo(struct mem_cgroup_per_node *pn, int nid) | |
805 | { | |
806 | struct mem_cgroup *parent; | |
807 | ||
808 | parent = parent_mem_cgroup(pn->memcg); | |
809 | if (!parent) | |
810 | return NULL; | |
811 | return mem_cgroup_nodeinfo(parent, nid); | |
812 | } | |
813 | ||
eedc4e5a RG |
814 | void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, |
815 | int val) | |
db9adbcb JW |
816 | { |
817 | struct mem_cgroup_per_node *pn; | |
42a30035 | 818 | struct mem_cgroup *memcg; |
ea426c2a | 819 | long x, threshold = MEMCG_CHARGE_BATCH; |
db9adbcb | 820 | |
db9adbcb | 821 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
42a30035 | 822 | memcg = pn->memcg; |
db9adbcb JW |
823 | |
824 | /* Update memcg */ | |
42a30035 | 825 | __mod_memcg_state(memcg, idx, val); |
db9adbcb | 826 | |
b4c46484 RG |
827 | /* Update lruvec */ |
828 | __this_cpu_add(pn->lruvec_stat_local->count[idx], val); | |
829 | ||
ea426c2a RG |
830 | if (vmstat_item_in_bytes(idx)) |
831 | threshold <<= PAGE_SHIFT; | |
832 | ||
db9adbcb | 833 | x = val + __this_cpu_read(pn->lruvec_stat_cpu->count[idx]); |
ea426c2a | 834 | if (unlikely(abs(x) > threshold)) { |
eedc4e5a | 835 | pg_data_t *pgdat = lruvec_pgdat(lruvec); |
42a30035 JW |
836 | struct mem_cgroup_per_node *pi; |
837 | ||
42a30035 JW |
838 | for (pi = pn; pi; pi = parent_nodeinfo(pi, pgdat->node_id)) |
839 | atomic_long_add(x, &pi->lruvec_stat[idx]); | |
db9adbcb JW |
840 | x = 0; |
841 | } | |
842 | __this_cpu_write(pn->lruvec_stat_cpu->count[idx], x); | |
843 | } | |
844 | ||
eedc4e5a RG |
845 | /** |
846 | * __mod_lruvec_state - update lruvec memory statistics | |
847 | * @lruvec: the lruvec | |
848 | * @idx: the stat item | |
849 | * @val: delta to add to the counter, can be negative | |
850 | * | |
851 | * The lruvec is the intersection of the NUMA node and a cgroup. This | |
852 | * function updates the all three counters that are affected by a | |
853 | * change of state at this level: per-node, per-cgroup, per-lruvec. | |
854 | */ | |
855 | void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, | |
856 | int val) | |
857 | { | |
858 | /* Update node */ | |
859 | __mod_node_page_state(lruvec_pgdat(lruvec), idx, val); | |
860 | ||
861 | /* Update memcg and lruvec */ | |
862 | if (!mem_cgroup_disabled()) | |
863 | __mod_memcg_lruvec_state(lruvec, idx, val); | |
864 | } | |
865 | ||
ec9f0238 RG |
866 | void __mod_lruvec_slab_state(void *p, enum node_stat_item idx, int val) |
867 | { | |
4f103c63 | 868 | pg_data_t *pgdat = page_pgdat(virt_to_page(p)); |
ec9f0238 RG |
869 | struct mem_cgroup *memcg; |
870 | struct lruvec *lruvec; | |
871 | ||
872 | rcu_read_lock(); | |
4f103c63 | 873 | memcg = mem_cgroup_from_obj(p); |
ec9f0238 RG |
874 | |
875 | /* Untracked pages have no memcg, no lruvec. Update only the node */ | |
876 | if (!memcg || memcg == root_mem_cgroup) { | |
877 | __mod_node_page_state(pgdat, idx, val); | |
878 | } else { | |
867e5e1d | 879 | lruvec = mem_cgroup_lruvec(memcg, pgdat); |
ec9f0238 RG |
880 | __mod_lruvec_state(lruvec, idx, val); |
881 | } | |
882 | rcu_read_unlock(); | |
883 | } | |
884 | ||
8380ce47 RG |
885 | void mod_memcg_obj_state(void *p, int idx, int val) |
886 | { | |
887 | struct mem_cgroup *memcg; | |
888 | ||
889 | rcu_read_lock(); | |
890 | memcg = mem_cgroup_from_obj(p); | |
891 | if (memcg) | |
892 | mod_memcg_state(memcg, idx, val); | |
893 | rcu_read_unlock(); | |
894 | } | |
895 | ||
db9adbcb JW |
896 | /** |
897 | * __count_memcg_events - account VM events in a cgroup | |
898 | * @memcg: the memory cgroup | |
899 | * @idx: the event item | |
900 | * @count: the number of events that occured | |
901 | */ | |
902 | void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, | |
903 | unsigned long count) | |
904 | { | |
905 | unsigned long x; | |
906 | ||
907 | if (mem_cgroup_disabled()) | |
908 | return; | |
909 | ||
910 | x = count + __this_cpu_read(memcg->vmstats_percpu->events[idx]); | |
911 | if (unlikely(x > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
912 | struct mem_cgroup *mi; |
913 | ||
766a4c19 YS |
914 | /* |
915 | * Batch local counters to keep them in sync with | |
916 | * the hierarchical ones. | |
917 | */ | |
918 | __this_cpu_add(memcg->vmstats_local->events[idx], x); | |
42a30035 JW |
919 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
920 | atomic_long_add(x, &mi->vmevents[idx]); | |
db9adbcb JW |
921 | x = 0; |
922 | } | |
923 | __this_cpu_write(memcg->vmstats_percpu->events[idx], x); | |
924 | } | |
925 | ||
42a30035 | 926 | static unsigned long memcg_events(struct mem_cgroup *memcg, int event) |
e9f8974f | 927 | { |
871789d4 | 928 | return atomic_long_read(&memcg->vmevents[event]); |
e9f8974f JW |
929 | } |
930 | ||
42a30035 JW |
931 | static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event) |
932 | { | |
815744d7 JW |
933 | long x = 0; |
934 | int cpu; | |
935 | ||
936 | for_each_possible_cpu(cpu) | |
937 | x += per_cpu(memcg->vmstats_local->events[event], cpu); | |
938 | return x; | |
42a30035 JW |
939 | } |
940 | ||
c0ff4b85 | 941 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 942 | struct page *page, |
3fba69a5 | 943 | int nr_pages) |
d52aa412 | 944 | { |
e401f176 KH |
945 | /* pagein of a big page is an event. So, ignore page size */ |
946 | if (nr_pages > 0) | |
c9019e9b | 947 | __count_memcg_events(memcg, PGPGIN, 1); |
3751d604 | 948 | else { |
c9019e9b | 949 | __count_memcg_events(memcg, PGPGOUT, 1); |
3751d604 KH |
950 | nr_pages = -nr_pages; /* for event */ |
951 | } | |
e401f176 | 952 | |
871789d4 | 953 | __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages); |
6d12e2d8 KH |
954 | } |
955 | ||
f53d7ce3 JW |
956 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
957 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
958 | { |
959 | unsigned long val, next; | |
960 | ||
871789d4 CD |
961 | val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events); |
962 | next = __this_cpu_read(memcg->vmstats_percpu->targets[target]); | |
7a159cc9 | 963 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 964 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
965 | switch (target) { |
966 | case MEM_CGROUP_TARGET_THRESH: | |
967 | next = val + THRESHOLDS_EVENTS_TARGET; | |
968 | break; | |
bb4cc1a8 AM |
969 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
970 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
971 | break; | |
f53d7ce3 JW |
972 | default: |
973 | break; | |
974 | } | |
871789d4 | 975 | __this_cpu_write(memcg->vmstats_percpu->targets[target], next); |
f53d7ce3 | 976 | return true; |
7a159cc9 | 977 | } |
f53d7ce3 | 978 | return false; |
d2265e6f KH |
979 | } |
980 | ||
981 | /* | |
982 | * Check events in order. | |
983 | * | |
984 | */ | |
c0ff4b85 | 985 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
986 | { |
987 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
988 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
989 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 990 | bool do_softlimit; |
f53d7ce3 | 991 | |
bb4cc1a8 AM |
992 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
993 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
c0ff4b85 | 994 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
995 | if (unlikely(do_softlimit)) |
996 | mem_cgroup_update_tree(memcg, page); | |
0a31bc97 | 997 | } |
d2265e6f KH |
998 | } |
999 | ||
cf475ad2 | 1000 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 1001 | { |
31a78f23 BS |
1002 | /* |
1003 | * mm_update_next_owner() may clear mm->owner to NULL | |
1004 | * if it races with swapoff, page migration, etc. | |
1005 | * So this can be called with p == NULL. | |
1006 | */ | |
1007 | if (unlikely(!p)) | |
1008 | return NULL; | |
1009 | ||
073219e9 | 1010 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 1011 | } |
33398cf2 | 1012 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 1013 | |
d46eb14b SB |
1014 | /** |
1015 | * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg. | |
1016 | * @mm: mm from which memcg should be extracted. It can be NULL. | |
1017 | * | |
1018 | * Obtain a reference on mm->memcg and returns it if successful. Otherwise | |
1019 | * root_mem_cgroup is returned. However if mem_cgroup is disabled, NULL is | |
1020 | * returned. | |
1021 | */ | |
1022 | struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) | |
54595fe2 | 1023 | { |
d46eb14b SB |
1024 | struct mem_cgroup *memcg; |
1025 | ||
1026 | if (mem_cgroup_disabled()) | |
1027 | return NULL; | |
0b7f569e | 1028 | |
54595fe2 KH |
1029 | rcu_read_lock(); |
1030 | do { | |
6f6acb00 MH |
1031 | /* |
1032 | * Page cache insertions can happen withou an | |
1033 | * actual mm context, e.g. during disk probing | |
1034 | * on boot, loopback IO, acct() writes etc. | |
1035 | */ | |
1036 | if (unlikely(!mm)) | |
df381975 | 1037 | memcg = root_mem_cgroup; |
6f6acb00 MH |
1038 | else { |
1039 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
1040 | if (unlikely(!memcg)) | |
1041 | memcg = root_mem_cgroup; | |
1042 | } | |
00d484f3 | 1043 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 1044 | rcu_read_unlock(); |
c0ff4b85 | 1045 | return memcg; |
54595fe2 | 1046 | } |
d46eb14b SB |
1047 | EXPORT_SYMBOL(get_mem_cgroup_from_mm); |
1048 | ||
f745c6f5 SB |
1049 | /** |
1050 | * get_mem_cgroup_from_page: Obtain a reference on given page's memcg. | |
1051 | * @page: page from which memcg should be extracted. | |
1052 | * | |
1053 | * Obtain a reference on page->memcg and returns it if successful. Otherwise | |
1054 | * root_mem_cgroup is returned. | |
1055 | */ | |
1056 | struct mem_cgroup *get_mem_cgroup_from_page(struct page *page) | |
1057 | { | |
1058 | struct mem_cgroup *memcg = page->mem_cgroup; | |
1059 | ||
1060 | if (mem_cgroup_disabled()) | |
1061 | return NULL; | |
1062 | ||
1063 | rcu_read_lock(); | |
8965aa28 SB |
1064 | /* Page should not get uncharged and freed memcg under us. */ |
1065 | if (!memcg || WARN_ON_ONCE(!css_tryget(&memcg->css))) | |
f745c6f5 SB |
1066 | memcg = root_mem_cgroup; |
1067 | rcu_read_unlock(); | |
1068 | return memcg; | |
1069 | } | |
1070 | EXPORT_SYMBOL(get_mem_cgroup_from_page); | |
1071 | ||
d46eb14b SB |
1072 | /** |
1073 | * If current->active_memcg is non-NULL, do not fallback to current->mm->memcg. | |
1074 | */ | |
1075 | static __always_inline struct mem_cgroup *get_mem_cgroup_from_current(void) | |
1076 | { | |
1077 | if (unlikely(current->active_memcg)) { | |
8965aa28 | 1078 | struct mem_cgroup *memcg; |
d46eb14b SB |
1079 | |
1080 | rcu_read_lock(); | |
8965aa28 SB |
1081 | /* current->active_memcg must hold a ref. */ |
1082 | if (WARN_ON_ONCE(!css_tryget(¤t->active_memcg->css))) | |
1083 | memcg = root_mem_cgroup; | |
1084 | else | |
d46eb14b SB |
1085 | memcg = current->active_memcg; |
1086 | rcu_read_unlock(); | |
1087 | return memcg; | |
1088 | } | |
1089 | return get_mem_cgroup_from_mm(current->mm); | |
1090 | } | |
54595fe2 | 1091 | |
5660048c JW |
1092 | /** |
1093 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1094 | * @root: hierarchy root | |
1095 | * @prev: previously returned memcg, NULL on first invocation | |
1096 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1097 | * | |
1098 | * Returns references to children of the hierarchy below @root, or | |
1099 | * @root itself, or %NULL after a full round-trip. | |
1100 | * | |
1101 | * Caller must pass the return value in @prev on subsequent | |
1102 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1103 | * to cancel a hierarchy walk before the round-trip is complete. | |
1104 | * | |
b213b54f | 1105 | * Reclaimers can specify a node and a priority level in @reclaim to |
5660048c | 1106 | * divide up the memcgs in the hierarchy among all concurrent |
b213b54f | 1107 | * reclaimers operating on the same node and priority. |
5660048c | 1108 | */ |
694fbc0f | 1109 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1110 | struct mem_cgroup *prev, |
694fbc0f | 1111 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1112 | { |
3f649ab7 | 1113 | struct mem_cgroup_reclaim_iter *iter; |
5ac8fb31 | 1114 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 1115 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1116 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1117 | |
694fbc0f AM |
1118 | if (mem_cgroup_disabled()) |
1119 | return NULL; | |
5660048c | 1120 | |
9f3a0d09 JW |
1121 | if (!root) |
1122 | root = root_mem_cgroup; | |
7d74b06f | 1123 | |
9f3a0d09 | 1124 | if (prev && !reclaim) |
5ac8fb31 | 1125 | pos = prev; |
14067bb3 | 1126 | |
9f3a0d09 JW |
1127 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1128 | if (prev) | |
5ac8fb31 | 1129 | goto out; |
694fbc0f | 1130 | return root; |
9f3a0d09 | 1131 | } |
14067bb3 | 1132 | |
542f85f9 | 1133 | rcu_read_lock(); |
5f578161 | 1134 | |
5ac8fb31 | 1135 | if (reclaim) { |
ef8f2327 | 1136 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 1137 | |
ef8f2327 | 1138 | mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id); |
9da83f3f | 1139 | iter = &mz->iter; |
5ac8fb31 JW |
1140 | |
1141 | if (prev && reclaim->generation != iter->generation) | |
1142 | goto out_unlock; | |
1143 | ||
6df38689 | 1144 | while (1) { |
4db0c3c2 | 1145 | pos = READ_ONCE(iter->position); |
6df38689 VD |
1146 | if (!pos || css_tryget(&pos->css)) |
1147 | break; | |
5ac8fb31 | 1148 | /* |
6df38689 VD |
1149 | * css reference reached zero, so iter->position will |
1150 | * be cleared by ->css_released. However, we should not | |
1151 | * rely on this happening soon, because ->css_released | |
1152 | * is called from a work queue, and by busy-waiting we | |
1153 | * might block it. So we clear iter->position right | |
1154 | * away. | |
5ac8fb31 | 1155 | */ |
6df38689 VD |
1156 | (void)cmpxchg(&iter->position, pos, NULL); |
1157 | } | |
5ac8fb31 JW |
1158 | } |
1159 | ||
1160 | if (pos) | |
1161 | css = &pos->css; | |
1162 | ||
1163 | for (;;) { | |
1164 | css = css_next_descendant_pre(css, &root->css); | |
1165 | if (!css) { | |
1166 | /* | |
1167 | * Reclaimers share the hierarchy walk, and a | |
1168 | * new one might jump in right at the end of | |
1169 | * the hierarchy - make sure they see at least | |
1170 | * one group and restart from the beginning. | |
1171 | */ | |
1172 | if (!prev) | |
1173 | continue; | |
1174 | break; | |
527a5ec9 | 1175 | } |
7d74b06f | 1176 | |
5ac8fb31 JW |
1177 | /* |
1178 | * Verify the css and acquire a reference. The root | |
1179 | * is provided by the caller, so we know it's alive | |
1180 | * and kicking, and don't take an extra reference. | |
1181 | */ | |
1182 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 1183 | |
5ac8fb31 JW |
1184 | if (css == &root->css) |
1185 | break; | |
14067bb3 | 1186 | |
0b8f73e1 JW |
1187 | if (css_tryget(css)) |
1188 | break; | |
9f3a0d09 | 1189 | |
5ac8fb31 | 1190 | memcg = NULL; |
9f3a0d09 | 1191 | } |
5ac8fb31 JW |
1192 | |
1193 | if (reclaim) { | |
5ac8fb31 | 1194 | /* |
6df38689 VD |
1195 | * The position could have already been updated by a competing |
1196 | * thread, so check that the value hasn't changed since we read | |
1197 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 1198 | */ |
6df38689 VD |
1199 | (void)cmpxchg(&iter->position, pos, memcg); |
1200 | ||
5ac8fb31 JW |
1201 | if (pos) |
1202 | css_put(&pos->css); | |
1203 | ||
1204 | if (!memcg) | |
1205 | iter->generation++; | |
1206 | else if (!prev) | |
1207 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1208 | } |
5ac8fb31 | 1209 | |
542f85f9 MH |
1210 | out_unlock: |
1211 | rcu_read_unlock(); | |
5ac8fb31 | 1212 | out: |
c40046f3 MH |
1213 | if (prev && prev != root) |
1214 | css_put(&prev->css); | |
1215 | ||
9f3a0d09 | 1216 | return memcg; |
14067bb3 | 1217 | } |
7d74b06f | 1218 | |
5660048c JW |
1219 | /** |
1220 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1221 | * @root: hierarchy root | |
1222 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1223 | */ | |
1224 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1225 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1226 | { |
1227 | if (!root) | |
1228 | root = root_mem_cgroup; | |
1229 | if (prev && prev != root) | |
1230 | css_put(&prev->css); | |
1231 | } | |
7d74b06f | 1232 | |
54a83d6b MC |
1233 | static void __invalidate_reclaim_iterators(struct mem_cgroup *from, |
1234 | struct mem_cgroup *dead_memcg) | |
6df38689 | 1235 | { |
6df38689 | 1236 | struct mem_cgroup_reclaim_iter *iter; |
ef8f2327 MG |
1237 | struct mem_cgroup_per_node *mz; |
1238 | int nid; | |
6df38689 | 1239 | |
54a83d6b MC |
1240 | for_each_node(nid) { |
1241 | mz = mem_cgroup_nodeinfo(from, nid); | |
9da83f3f YS |
1242 | iter = &mz->iter; |
1243 | cmpxchg(&iter->position, dead_memcg, NULL); | |
6df38689 VD |
1244 | } |
1245 | } | |
1246 | ||
54a83d6b MC |
1247 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1248 | { | |
1249 | struct mem_cgroup *memcg = dead_memcg; | |
1250 | struct mem_cgroup *last; | |
1251 | ||
1252 | do { | |
1253 | __invalidate_reclaim_iterators(memcg, dead_memcg); | |
1254 | last = memcg; | |
1255 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1256 | ||
1257 | /* | |
1258 | * When cgruop1 non-hierarchy mode is used, | |
1259 | * parent_mem_cgroup() does not walk all the way up to the | |
1260 | * cgroup root (root_mem_cgroup). So we have to handle | |
1261 | * dead_memcg from cgroup root separately. | |
1262 | */ | |
1263 | if (last != root_mem_cgroup) | |
1264 | __invalidate_reclaim_iterators(root_mem_cgroup, | |
1265 | dead_memcg); | |
1266 | } | |
1267 | ||
7c5f64f8 VD |
1268 | /** |
1269 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1270 | * @memcg: hierarchy root | |
1271 | * @fn: function to call for each task | |
1272 | * @arg: argument passed to @fn | |
1273 | * | |
1274 | * This function iterates over tasks attached to @memcg or to any of its | |
1275 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
1276 | * value, the function breaks the iteration loop and returns the value. | |
1277 | * Otherwise, it will iterate over all tasks and return 0. | |
1278 | * | |
1279 | * This function must not be called for the root memory cgroup. | |
1280 | */ | |
1281 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
1282 | int (*fn)(struct task_struct *, void *), void *arg) | |
1283 | { | |
1284 | struct mem_cgroup *iter; | |
1285 | int ret = 0; | |
1286 | ||
1287 | BUG_ON(memcg == root_mem_cgroup); | |
1288 | ||
1289 | for_each_mem_cgroup_tree(iter, memcg) { | |
1290 | struct css_task_iter it; | |
1291 | struct task_struct *task; | |
1292 | ||
f168a9a5 | 1293 | css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it); |
7c5f64f8 VD |
1294 | while (!ret && (task = css_task_iter_next(&it))) |
1295 | ret = fn(task, arg); | |
1296 | css_task_iter_end(&it); | |
1297 | if (ret) { | |
1298 | mem_cgroup_iter_break(memcg, iter); | |
1299 | break; | |
1300 | } | |
1301 | } | |
1302 | return ret; | |
1303 | } | |
1304 | ||
925b7673 | 1305 | /** |
dfe0e773 | 1306 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1307 | * @page: the page |
f144c390 | 1308 | * @pgdat: pgdat of the page |
dfe0e773 | 1309 | * |
a0b5b414 JW |
1310 | * This function relies on page->mem_cgroup being stable - see the |
1311 | * access rules in commit_charge(). | |
925b7673 | 1312 | */ |
599d0c95 | 1313 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat) |
08e552c6 | 1314 | { |
ef8f2327 | 1315 | struct mem_cgroup_per_node *mz; |
925b7673 | 1316 | struct mem_cgroup *memcg; |
bea8c150 | 1317 | struct lruvec *lruvec; |
6d12e2d8 | 1318 | |
bea8c150 | 1319 | if (mem_cgroup_disabled()) { |
867e5e1d | 1320 | lruvec = &pgdat->__lruvec; |
bea8c150 HD |
1321 | goto out; |
1322 | } | |
925b7673 | 1323 | |
1306a85a | 1324 | memcg = page->mem_cgroup; |
7512102c | 1325 | /* |
dfe0e773 | 1326 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1327 | * possibly migrated - before they are charged. |
7512102c | 1328 | */ |
29833315 JW |
1329 | if (!memcg) |
1330 | memcg = root_mem_cgroup; | |
7512102c | 1331 | |
ef8f2327 | 1332 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
bea8c150 HD |
1333 | lruvec = &mz->lruvec; |
1334 | out: | |
1335 | /* | |
1336 | * Since a node can be onlined after the mem_cgroup was created, | |
1337 | * we have to be prepared to initialize lruvec->zone here; | |
1338 | * and if offlined then reonlined, we need to reinitialize it. | |
1339 | */ | |
599d0c95 MG |
1340 | if (unlikely(lruvec->pgdat != pgdat)) |
1341 | lruvec->pgdat = pgdat; | |
bea8c150 | 1342 | return lruvec; |
08e552c6 | 1343 | } |
b69408e8 | 1344 | |
925b7673 | 1345 | /** |
fa9add64 HD |
1346 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1347 | * @lruvec: mem_cgroup per zone lru vector | |
1348 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1349 | * @zid: zone id of the accounted pages |
fa9add64 | 1350 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1351 | * |
ca707239 HD |
1352 | * This function must be called under lru_lock, just before a page is added |
1353 | * to or just after a page is removed from an lru list (that ordering being | |
1354 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 1355 | */ |
fa9add64 | 1356 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1357 | int zid, int nr_pages) |
3f58a829 | 1358 | { |
ef8f2327 | 1359 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1360 | unsigned long *lru_size; |
ca707239 | 1361 | long size; |
3f58a829 MK |
1362 | |
1363 | if (mem_cgroup_disabled()) | |
1364 | return; | |
1365 | ||
ef8f2327 | 1366 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1367 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1368 | |
1369 | if (nr_pages < 0) | |
1370 | *lru_size += nr_pages; | |
1371 | ||
1372 | size = *lru_size; | |
b4536f0c MH |
1373 | if (WARN_ONCE(size < 0, |
1374 | "%s(%p, %d, %d): lru_size %ld\n", | |
1375 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1376 | VM_BUG_ON(1); |
1377 | *lru_size = 0; | |
1378 | } | |
1379 | ||
1380 | if (nr_pages > 0) | |
1381 | *lru_size += nr_pages; | |
08e552c6 | 1382 | } |
544122e5 | 1383 | |
19942822 | 1384 | /** |
9d11ea9f | 1385 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1386 | * @memcg: the memory cgroup |
19942822 | 1387 | * |
9d11ea9f | 1388 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1389 | * pages. |
19942822 | 1390 | */ |
c0ff4b85 | 1391 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1392 | { |
3e32cb2e JW |
1393 | unsigned long margin = 0; |
1394 | unsigned long count; | |
1395 | unsigned long limit; | |
9d11ea9f | 1396 | |
3e32cb2e | 1397 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1398 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1399 | if (count < limit) |
1400 | margin = limit - count; | |
1401 | ||
7941d214 | 1402 | if (do_memsw_account()) { |
3e32cb2e | 1403 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1404 | limit = READ_ONCE(memcg->memsw.max); |
1c4448ed | 1405 | if (count < limit) |
3e32cb2e | 1406 | margin = min(margin, limit - count); |
cbedbac3 LR |
1407 | else |
1408 | margin = 0; | |
3e32cb2e JW |
1409 | } |
1410 | ||
1411 | return margin; | |
19942822 JW |
1412 | } |
1413 | ||
32047e2a | 1414 | /* |
bdcbb659 | 1415 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1416 | * |
bdcbb659 QH |
1417 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1418 | * moving cgroups. This is for waiting at high-memory pressure | |
1419 | * caused by "move". | |
32047e2a | 1420 | */ |
c0ff4b85 | 1421 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1422 | { |
2bd9bb20 KH |
1423 | struct mem_cgroup *from; |
1424 | struct mem_cgroup *to; | |
4b534334 | 1425 | bool ret = false; |
2bd9bb20 KH |
1426 | /* |
1427 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1428 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1429 | */ | |
1430 | spin_lock(&mc.lock); | |
1431 | from = mc.from; | |
1432 | to = mc.to; | |
1433 | if (!from) | |
1434 | goto unlock; | |
3e92041d | 1435 | |
2314b42d JW |
1436 | ret = mem_cgroup_is_descendant(from, memcg) || |
1437 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1438 | unlock: |
1439 | spin_unlock(&mc.lock); | |
4b534334 KH |
1440 | return ret; |
1441 | } | |
1442 | ||
c0ff4b85 | 1443 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1444 | { |
1445 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1446 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1447 | DEFINE_WAIT(wait); |
1448 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1449 | /* moving charge context might have finished. */ | |
1450 | if (mc.moving_task) | |
1451 | schedule(); | |
1452 | finish_wait(&mc.waitq, &wait); | |
1453 | return true; | |
1454 | } | |
1455 | } | |
1456 | return false; | |
1457 | } | |
1458 | ||
c8713d0b JW |
1459 | static char *memory_stat_format(struct mem_cgroup *memcg) |
1460 | { | |
1461 | struct seq_buf s; | |
1462 | int i; | |
71cd3113 | 1463 | |
c8713d0b JW |
1464 | seq_buf_init(&s, kmalloc(PAGE_SIZE, GFP_KERNEL), PAGE_SIZE); |
1465 | if (!s.buffer) | |
1466 | return NULL; | |
1467 | ||
1468 | /* | |
1469 | * Provide statistics on the state of the memory subsystem as | |
1470 | * well as cumulative event counters that show past behavior. | |
1471 | * | |
1472 | * This list is ordered following a combination of these gradients: | |
1473 | * 1) generic big picture -> specifics and details | |
1474 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
1475 | * | |
1476 | * Current memory state: | |
1477 | */ | |
1478 | ||
1479 | seq_buf_printf(&s, "anon %llu\n", | |
be5d0a74 | 1480 | (u64)memcg_page_state(memcg, NR_ANON_MAPPED) * |
c8713d0b JW |
1481 | PAGE_SIZE); |
1482 | seq_buf_printf(&s, "file %llu\n", | |
0d1c2072 | 1483 | (u64)memcg_page_state(memcg, NR_FILE_PAGES) * |
c8713d0b JW |
1484 | PAGE_SIZE); |
1485 | seq_buf_printf(&s, "kernel_stack %llu\n", | |
991e7673 | 1486 | (u64)memcg_page_state(memcg, NR_KERNEL_STACK_KB) * |
c8713d0b JW |
1487 | 1024); |
1488 | seq_buf_printf(&s, "slab %llu\n", | |
d42f3245 RG |
1489 | (u64)(memcg_page_state(memcg, NR_SLAB_RECLAIMABLE_B) + |
1490 | memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE_B))); | |
c8713d0b JW |
1491 | seq_buf_printf(&s, "sock %llu\n", |
1492 | (u64)memcg_page_state(memcg, MEMCG_SOCK) * | |
1493 | PAGE_SIZE); | |
1494 | ||
1495 | seq_buf_printf(&s, "shmem %llu\n", | |
1496 | (u64)memcg_page_state(memcg, NR_SHMEM) * | |
1497 | PAGE_SIZE); | |
1498 | seq_buf_printf(&s, "file_mapped %llu\n", | |
1499 | (u64)memcg_page_state(memcg, NR_FILE_MAPPED) * | |
1500 | PAGE_SIZE); | |
1501 | seq_buf_printf(&s, "file_dirty %llu\n", | |
1502 | (u64)memcg_page_state(memcg, NR_FILE_DIRTY) * | |
1503 | PAGE_SIZE); | |
1504 | seq_buf_printf(&s, "file_writeback %llu\n", | |
1505 | (u64)memcg_page_state(memcg, NR_WRITEBACK) * | |
1506 | PAGE_SIZE); | |
1507 | ||
468c3982 | 1508 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
c8713d0b | 1509 | seq_buf_printf(&s, "anon_thp %llu\n", |
468c3982 JW |
1510 | (u64)memcg_page_state(memcg, NR_ANON_THPS) * |
1511 | HPAGE_PMD_SIZE); | |
1512 | #endif | |
c8713d0b JW |
1513 | |
1514 | for (i = 0; i < NR_LRU_LISTS; i++) | |
ebc5d83d | 1515 | seq_buf_printf(&s, "%s %llu\n", lru_list_name(i), |
c8713d0b JW |
1516 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
1517 | PAGE_SIZE); | |
1518 | ||
1519 | seq_buf_printf(&s, "slab_reclaimable %llu\n", | |
d42f3245 | 1520 | (u64)memcg_page_state(memcg, NR_SLAB_RECLAIMABLE_B)); |
c8713d0b | 1521 | seq_buf_printf(&s, "slab_unreclaimable %llu\n", |
d42f3245 | 1522 | (u64)memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE_B)); |
c8713d0b JW |
1523 | |
1524 | /* Accumulated memory events */ | |
1525 | ||
ebc5d83d KK |
1526 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGFAULT), |
1527 | memcg_events(memcg, PGFAULT)); | |
1528 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGMAJFAULT), | |
1529 | memcg_events(memcg, PGMAJFAULT)); | |
c8713d0b JW |
1530 | |
1531 | seq_buf_printf(&s, "workingset_refault %lu\n", | |
1532 | memcg_page_state(memcg, WORKINGSET_REFAULT)); | |
1533 | seq_buf_printf(&s, "workingset_activate %lu\n", | |
1534 | memcg_page_state(memcg, WORKINGSET_ACTIVATE)); | |
a6f5576b YS |
1535 | seq_buf_printf(&s, "workingset_restore %lu\n", |
1536 | memcg_page_state(memcg, WORKINGSET_RESTORE)); | |
c8713d0b JW |
1537 | seq_buf_printf(&s, "workingset_nodereclaim %lu\n", |
1538 | memcg_page_state(memcg, WORKINGSET_NODERECLAIM)); | |
1539 | ||
ebc5d83d KK |
1540 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGREFILL), |
1541 | memcg_events(memcg, PGREFILL)); | |
c8713d0b JW |
1542 | seq_buf_printf(&s, "pgscan %lu\n", |
1543 | memcg_events(memcg, PGSCAN_KSWAPD) + | |
1544 | memcg_events(memcg, PGSCAN_DIRECT)); | |
1545 | seq_buf_printf(&s, "pgsteal %lu\n", | |
1546 | memcg_events(memcg, PGSTEAL_KSWAPD) + | |
1547 | memcg_events(memcg, PGSTEAL_DIRECT)); | |
ebc5d83d KK |
1548 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGACTIVATE), |
1549 | memcg_events(memcg, PGACTIVATE)); | |
1550 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGDEACTIVATE), | |
1551 | memcg_events(memcg, PGDEACTIVATE)); | |
1552 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREE), | |
1553 | memcg_events(memcg, PGLAZYFREE)); | |
1554 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREED), | |
1555 | memcg_events(memcg, PGLAZYFREED)); | |
c8713d0b JW |
1556 | |
1557 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
ebc5d83d | 1558 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_FAULT_ALLOC), |
c8713d0b | 1559 | memcg_events(memcg, THP_FAULT_ALLOC)); |
ebc5d83d | 1560 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_COLLAPSE_ALLOC), |
c8713d0b JW |
1561 | memcg_events(memcg, THP_COLLAPSE_ALLOC)); |
1562 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ | |
1563 | ||
1564 | /* The above should easily fit into one page */ | |
1565 | WARN_ON_ONCE(seq_buf_has_overflowed(&s)); | |
1566 | ||
1567 | return s.buffer; | |
1568 | } | |
71cd3113 | 1569 | |
58cf188e | 1570 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1571 | /** |
f0c867d9 | 1572 | * mem_cgroup_print_oom_context: Print OOM information relevant to |
1573 | * memory controller. | |
e222432b BS |
1574 | * @memcg: The memory cgroup that went over limit |
1575 | * @p: Task that is going to be killed | |
1576 | * | |
1577 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1578 | * enabled | |
1579 | */ | |
f0c867d9 | 1580 | void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) |
e222432b | 1581 | { |
e222432b BS |
1582 | rcu_read_lock(); |
1583 | ||
f0c867d9 | 1584 | if (memcg) { |
1585 | pr_cont(",oom_memcg="); | |
1586 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1587 | } else | |
1588 | pr_cont(",global_oom"); | |
2415b9f5 | 1589 | if (p) { |
f0c867d9 | 1590 | pr_cont(",task_memcg="); |
2415b9f5 | 1591 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
2415b9f5 | 1592 | } |
e222432b | 1593 | rcu_read_unlock(); |
f0c867d9 | 1594 | } |
1595 | ||
1596 | /** | |
1597 | * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to | |
1598 | * memory controller. | |
1599 | * @memcg: The memory cgroup that went over limit | |
1600 | */ | |
1601 | void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) | |
1602 | { | |
c8713d0b | 1603 | char *buf; |
e222432b | 1604 | |
3e32cb2e JW |
1605 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1606 | K((u64)page_counter_read(&memcg->memory)), | |
15b42562 | 1607 | K((u64)READ_ONCE(memcg->memory.max)), memcg->memory.failcnt); |
c8713d0b JW |
1608 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
1609 | pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1610 | K((u64)page_counter_read(&memcg->swap)), | |
32d087cd | 1611 | K((u64)READ_ONCE(memcg->swap.max)), memcg->swap.failcnt); |
c8713d0b JW |
1612 | else { |
1613 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1614 | K((u64)page_counter_read(&memcg->memsw)), | |
1615 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); | |
1616 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1617 | K((u64)page_counter_read(&memcg->kmem)), | |
1618 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); | |
58cf188e | 1619 | } |
c8713d0b JW |
1620 | |
1621 | pr_info("Memory cgroup stats for "); | |
1622 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1623 | pr_cont(":"); | |
1624 | buf = memory_stat_format(memcg); | |
1625 | if (!buf) | |
1626 | return; | |
1627 | pr_info("%s", buf); | |
1628 | kfree(buf); | |
e222432b BS |
1629 | } |
1630 | ||
a63d83f4 DR |
1631 | /* |
1632 | * Return the memory (and swap, if configured) limit for a memcg. | |
1633 | */ | |
bbec2e15 | 1634 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1635 | { |
bbec2e15 | 1636 | unsigned long max; |
f3e8eb70 | 1637 | |
15b42562 | 1638 | max = READ_ONCE(memcg->memory.max); |
9a5a8f19 | 1639 | if (mem_cgroup_swappiness(memcg)) { |
bbec2e15 RG |
1640 | unsigned long memsw_max; |
1641 | unsigned long swap_max; | |
9a5a8f19 | 1642 | |
bbec2e15 | 1643 | memsw_max = memcg->memsw.max; |
32d087cd | 1644 | swap_max = READ_ONCE(memcg->swap.max); |
bbec2e15 RG |
1645 | swap_max = min(swap_max, (unsigned long)total_swap_pages); |
1646 | max = min(max + swap_max, memsw_max); | |
9a5a8f19 | 1647 | } |
bbec2e15 | 1648 | return max; |
a63d83f4 DR |
1649 | } |
1650 | ||
9783aa99 CD |
1651 | unsigned long mem_cgroup_size(struct mem_cgroup *memcg) |
1652 | { | |
1653 | return page_counter_read(&memcg->memory); | |
1654 | } | |
1655 | ||
b6e6edcf | 1656 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1657 | int order) |
9cbb78bb | 1658 | { |
6e0fc46d DR |
1659 | struct oom_control oc = { |
1660 | .zonelist = NULL, | |
1661 | .nodemask = NULL, | |
2a966b77 | 1662 | .memcg = memcg, |
6e0fc46d DR |
1663 | .gfp_mask = gfp_mask, |
1664 | .order = order, | |
6e0fc46d | 1665 | }; |
1378b37d | 1666 | bool ret = true; |
9cbb78bb | 1667 | |
7775face TH |
1668 | if (mutex_lock_killable(&oom_lock)) |
1669 | return true; | |
1378b37d YS |
1670 | |
1671 | if (mem_cgroup_margin(memcg) >= (1 << order)) | |
1672 | goto unlock; | |
1673 | ||
7775face TH |
1674 | /* |
1675 | * A few threads which were not waiting at mutex_lock_killable() can | |
1676 | * fail to bail out. Therefore, check again after holding oom_lock. | |
1677 | */ | |
1678 | ret = should_force_charge() || out_of_memory(&oc); | |
1378b37d YS |
1679 | |
1680 | unlock: | |
dc56401f | 1681 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1682 | return ret; |
9cbb78bb DR |
1683 | } |
1684 | ||
0608f43d | 1685 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1686 | pg_data_t *pgdat, |
0608f43d AM |
1687 | gfp_t gfp_mask, |
1688 | unsigned long *total_scanned) | |
1689 | { | |
1690 | struct mem_cgroup *victim = NULL; | |
1691 | int total = 0; | |
1692 | int loop = 0; | |
1693 | unsigned long excess; | |
1694 | unsigned long nr_scanned; | |
1695 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1696 | .pgdat = pgdat, |
0608f43d AM |
1697 | }; |
1698 | ||
3e32cb2e | 1699 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1700 | |
1701 | while (1) { | |
1702 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1703 | if (!victim) { | |
1704 | loop++; | |
1705 | if (loop >= 2) { | |
1706 | /* | |
1707 | * If we have not been able to reclaim | |
1708 | * anything, it might because there are | |
1709 | * no reclaimable pages under this hierarchy | |
1710 | */ | |
1711 | if (!total) | |
1712 | break; | |
1713 | /* | |
1714 | * We want to do more targeted reclaim. | |
1715 | * excess >> 2 is not to excessive so as to | |
1716 | * reclaim too much, nor too less that we keep | |
1717 | * coming back to reclaim from this cgroup | |
1718 | */ | |
1719 | if (total >= (excess >> 2) || | |
1720 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1721 | break; | |
1722 | } | |
1723 | continue; | |
1724 | } | |
a9dd0a83 | 1725 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1726 | pgdat, &nr_scanned); |
0608f43d | 1727 | *total_scanned += nr_scanned; |
3e32cb2e | 1728 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1729 | break; |
6d61ef40 | 1730 | } |
0608f43d AM |
1731 | mem_cgroup_iter_break(root_memcg, victim); |
1732 | return total; | |
6d61ef40 BS |
1733 | } |
1734 | ||
0056f4e6 JW |
1735 | #ifdef CONFIG_LOCKDEP |
1736 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1737 | .name = "memcg_oom_lock", | |
1738 | }; | |
1739 | #endif | |
1740 | ||
fb2a6fc5 JW |
1741 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1742 | ||
867578cb KH |
1743 | /* |
1744 | * Check OOM-Killer is already running under our hierarchy. | |
1745 | * If someone is running, return false. | |
1746 | */ | |
fb2a6fc5 | 1747 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1748 | { |
79dfdacc | 1749 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1750 | |
fb2a6fc5 JW |
1751 | spin_lock(&memcg_oom_lock); |
1752 | ||
9f3a0d09 | 1753 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1754 | if (iter->oom_lock) { |
79dfdacc MH |
1755 | /* |
1756 | * this subtree of our hierarchy is already locked | |
1757 | * so we cannot give a lock. | |
1758 | */ | |
79dfdacc | 1759 | failed = iter; |
9f3a0d09 JW |
1760 | mem_cgroup_iter_break(memcg, iter); |
1761 | break; | |
23751be0 JW |
1762 | } else |
1763 | iter->oom_lock = true; | |
7d74b06f | 1764 | } |
867578cb | 1765 | |
fb2a6fc5 JW |
1766 | if (failed) { |
1767 | /* | |
1768 | * OK, we failed to lock the whole subtree so we have | |
1769 | * to clean up what we set up to the failing subtree | |
1770 | */ | |
1771 | for_each_mem_cgroup_tree(iter, memcg) { | |
1772 | if (iter == failed) { | |
1773 | mem_cgroup_iter_break(memcg, iter); | |
1774 | break; | |
1775 | } | |
1776 | iter->oom_lock = false; | |
79dfdacc | 1777 | } |
0056f4e6 JW |
1778 | } else |
1779 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1780 | |
1781 | spin_unlock(&memcg_oom_lock); | |
1782 | ||
1783 | return !failed; | |
a636b327 | 1784 | } |
0b7f569e | 1785 | |
fb2a6fc5 | 1786 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1787 | { |
7d74b06f KH |
1788 | struct mem_cgroup *iter; |
1789 | ||
fb2a6fc5 | 1790 | spin_lock(&memcg_oom_lock); |
5facae4f | 1791 | mutex_release(&memcg_oom_lock_dep_map, _RET_IP_); |
c0ff4b85 | 1792 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1793 | iter->oom_lock = false; |
fb2a6fc5 | 1794 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1795 | } |
1796 | ||
c0ff4b85 | 1797 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1798 | { |
1799 | struct mem_cgroup *iter; | |
1800 | ||
c2b42d3c | 1801 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1802 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1803 | iter->under_oom++; |
1804 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1805 | } |
1806 | ||
c0ff4b85 | 1807 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1808 | { |
1809 | struct mem_cgroup *iter; | |
1810 | ||
867578cb KH |
1811 | /* |
1812 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1813 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1814 | */ |
c2b42d3c | 1815 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1816 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1817 | if (iter->under_oom > 0) |
1818 | iter->under_oom--; | |
1819 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1820 | } |
1821 | ||
867578cb KH |
1822 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1823 | ||
dc98df5a | 1824 | struct oom_wait_info { |
d79154bb | 1825 | struct mem_cgroup *memcg; |
ac6424b9 | 1826 | wait_queue_entry_t wait; |
dc98df5a KH |
1827 | }; |
1828 | ||
ac6424b9 | 1829 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1830 | unsigned mode, int sync, void *arg) |
1831 | { | |
d79154bb HD |
1832 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1833 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1834 | struct oom_wait_info *oom_wait_info; |
1835 | ||
1836 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1837 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1838 | |
2314b42d JW |
1839 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1840 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1841 | return 0; |
dc98df5a KH |
1842 | return autoremove_wake_function(wait, mode, sync, arg); |
1843 | } | |
1844 | ||
c0ff4b85 | 1845 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1846 | { |
c2b42d3c TH |
1847 | /* |
1848 | * For the following lockless ->under_oom test, the only required | |
1849 | * guarantee is that it must see the state asserted by an OOM when | |
1850 | * this function is called as a result of userland actions | |
1851 | * triggered by the notification of the OOM. This is trivially | |
1852 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1853 | * triggering notification. | |
1854 | */ | |
1855 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1856 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1857 | } |
1858 | ||
29ef680a MH |
1859 | enum oom_status { |
1860 | OOM_SUCCESS, | |
1861 | OOM_FAILED, | |
1862 | OOM_ASYNC, | |
1863 | OOM_SKIPPED | |
1864 | }; | |
1865 | ||
1866 | static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 1867 | { |
7056d3a3 MH |
1868 | enum oom_status ret; |
1869 | bool locked; | |
1870 | ||
29ef680a MH |
1871 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
1872 | return OOM_SKIPPED; | |
1873 | ||
7a1adfdd RG |
1874 | memcg_memory_event(memcg, MEMCG_OOM); |
1875 | ||
867578cb | 1876 | /* |
49426420 JW |
1877 | * We are in the middle of the charge context here, so we |
1878 | * don't want to block when potentially sitting on a callstack | |
1879 | * that holds all kinds of filesystem and mm locks. | |
1880 | * | |
29ef680a MH |
1881 | * cgroup1 allows disabling the OOM killer and waiting for outside |
1882 | * handling until the charge can succeed; remember the context and put | |
1883 | * the task to sleep at the end of the page fault when all locks are | |
1884 | * released. | |
49426420 | 1885 | * |
29ef680a MH |
1886 | * On the other hand, in-kernel OOM killer allows for an async victim |
1887 | * memory reclaim (oom_reaper) and that means that we are not solely | |
1888 | * relying on the oom victim to make a forward progress and we can | |
1889 | * invoke the oom killer here. | |
1890 | * | |
1891 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
1892 | * victim and then we have to bail out from the charge path. | |
867578cb | 1893 | */ |
29ef680a MH |
1894 | if (memcg->oom_kill_disable) { |
1895 | if (!current->in_user_fault) | |
1896 | return OOM_SKIPPED; | |
1897 | css_get(&memcg->css); | |
1898 | current->memcg_in_oom = memcg; | |
1899 | current->memcg_oom_gfp_mask = mask; | |
1900 | current->memcg_oom_order = order; | |
1901 | ||
1902 | return OOM_ASYNC; | |
1903 | } | |
1904 | ||
7056d3a3 MH |
1905 | mem_cgroup_mark_under_oom(memcg); |
1906 | ||
1907 | locked = mem_cgroup_oom_trylock(memcg); | |
1908 | ||
1909 | if (locked) | |
1910 | mem_cgroup_oom_notify(memcg); | |
1911 | ||
1912 | mem_cgroup_unmark_under_oom(memcg); | |
29ef680a | 1913 | if (mem_cgroup_out_of_memory(memcg, mask, order)) |
7056d3a3 MH |
1914 | ret = OOM_SUCCESS; |
1915 | else | |
1916 | ret = OOM_FAILED; | |
1917 | ||
1918 | if (locked) | |
1919 | mem_cgroup_oom_unlock(memcg); | |
29ef680a | 1920 | |
7056d3a3 | 1921 | return ret; |
3812c8c8 JW |
1922 | } |
1923 | ||
1924 | /** | |
1925 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1926 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1927 | * |
49426420 JW |
1928 | * This has to be called at the end of a page fault if the memcg OOM |
1929 | * handler was enabled. | |
3812c8c8 | 1930 | * |
49426420 | 1931 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1932 | * sleep on a waitqueue until the userspace task resolves the |
1933 | * situation. Sleeping directly in the charge context with all kinds | |
1934 | * of locks held is not a good idea, instead we remember an OOM state | |
1935 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1936 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1937 | * |
1938 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1939 | * completed, %false otherwise. |
3812c8c8 | 1940 | */ |
49426420 | 1941 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1942 | { |
626ebc41 | 1943 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1944 | struct oom_wait_info owait; |
49426420 | 1945 | bool locked; |
3812c8c8 JW |
1946 | |
1947 | /* OOM is global, do not handle */ | |
3812c8c8 | 1948 | if (!memcg) |
49426420 | 1949 | return false; |
3812c8c8 | 1950 | |
7c5f64f8 | 1951 | if (!handle) |
49426420 | 1952 | goto cleanup; |
3812c8c8 JW |
1953 | |
1954 | owait.memcg = memcg; | |
1955 | owait.wait.flags = 0; | |
1956 | owait.wait.func = memcg_oom_wake_function; | |
1957 | owait.wait.private = current; | |
2055da97 | 1958 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 1959 | |
3812c8c8 | 1960 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1961 | mem_cgroup_mark_under_oom(memcg); |
1962 | ||
1963 | locked = mem_cgroup_oom_trylock(memcg); | |
1964 | ||
1965 | if (locked) | |
1966 | mem_cgroup_oom_notify(memcg); | |
1967 | ||
1968 | if (locked && !memcg->oom_kill_disable) { | |
1969 | mem_cgroup_unmark_under_oom(memcg); | |
1970 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1971 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1972 | current->memcg_oom_order); | |
49426420 | 1973 | } else { |
3812c8c8 | 1974 | schedule(); |
49426420 JW |
1975 | mem_cgroup_unmark_under_oom(memcg); |
1976 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1977 | } | |
1978 | ||
1979 | if (locked) { | |
fb2a6fc5 JW |
1980 | mem_cgroup_oom_unlock(memcg); |
1981 | /* | |
1982 | * There is no guarantee that an OOM-lock contender | |
1983 | * sees the wakeups triggered by the OOM kill | |
1984 | * uncharges. Wake any sleepers explicitely. | |
1985 | */ | |
1986 | memcg_oom_recover(memcg); | |
1987 | } | |
49426420 | 1988 | cleanup: |
626ebc41 | 1989 | current->memcg_in_oom = NULL; |
3812c8c8 | 1990 | css_put(&memcg->css); |
867578cb | 1991 | return true; |
0b7f569e KH |
1992 | } |
1993 | ||
3d8b38eb RG |
1994 | /** |
1995 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
1996 | * @victim: task to be killed by the OOM killer | |
1997 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
1998 | * | |
1999 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
2000 | * by killing all belonging OOM-killable tasks. | |
2001 | * | |
2002 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
2003 | */ | |
2004 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
2005 | struct mem_cgroup *oom_domain) | |
2006 | { | |
2007 | struct mem_cgroup *oom_group = NULL; | |
2008 | struct mem_cgroup *memcg; | |
2009 | ||
2010 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
2011 | return NULL; | |
2012 | ||
2013 | if (!oom_domain) | |
2014 | oom_domain = root_mem_cgroup; | |
2015 | ||
2016 | rcu_read_lock(); | |
2017 | ||
2018 | memcg = mem_cgroup_from_task(victim); | |
2019 | if (memcg == root_mem_cgroup) | |
2020 | goto out; | |
2021 | ||
48fe267c RG |
2022 | /* |
2023 | * If the victim task has been asynchronously moved to a different | |
2024 | * memory cgroup, we might end up killing tasks outside oom_domain. | |
2025 | * In this case it's better to ignore memory.group.oom. | |
2026 | */ | |
2027 | if (unlikely(!mem_cgroup_is_descendant(memcg, oom_domain))) | |
2028 | goto out; | |
2029 | ||
3d8b38eb RG |
2030 | /* |
2031 | * Traverse the memory cgroup hierarchy from the victim task's | |
2032 | * cgroup up to the OOMing cgroup (or root) to find the | |
2033 | * highest-level memory cgroup with oom.group set. | |
2034 | */ | |
2035 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
2036 | if (memcg->oom_group) | |
2037 | oom_group = memcg; | |
2038 | ||
2039 | if (memcg == oom_domain) | |
2040 | break; | |
2041 | } | |
2042 | ||
2043 | if (oom_group) | |
2044 | css_get(&oom_group->css); | |
2045 | out: | |
2046 | rcu_read_unlock(); | |
2047 | ||
2048 | return oom_group; | |
2049 | } | |
2050 | ||
2051 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
2052 | { | |
2053 | pr_info("Tasks in "); | |
2054 | pr_cont_cgroup_path(memcg->css.cgroup); | |
2055 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
2056 | } | |
2057 | ||
d7365e78 | 2058 | /** |
81f8c3a4 JW |
2059 | * lock_page_memcg - lock a page->mem_cgroup binding |
2060 | * @page: the page | |
32047e2a | 2061 | * |
81f8c3a4 | 2062 | * This function protects unlocked LRU pages from being moved to |
739f79fc JW |
2063 | * another cgroup. |
2064 | * | |
2065 | * It ensures lifetime of the returned memcg. Caller is responsible | |
2066 | * for the lifetime of the page; __unlock_page_memcg() is available | |
2067 | * when @page might get freed inside the locked section. | |
d69b042f | 2068 | */ |
739f79fc | 2069 | struct mem_cgroup *lock_page_memcg(struct page *page) |
89c06bd5 | 2070 | { |
9da7b521 | 2071 | struct page *head = compound_head(page); /* rmap on tail pages */ |
89c06bd5 | 2072 | struct mem_cgroup *memcg; |
6de22619 | 2073 | unsigned long flags; |
89c06bd5 | 2074 | |
6de22619 JW |
2075 | /* |
2076 | * The RCU lock is held throughout the transaction. The fast | |
2077 | * path can get away without acquiring the memcg->move_lock | |
2078 | * because page moving starts with an RCU grace period. | |
739f79fc JW |
2079 | * |
2080 | * The RCU lock also protects the memcg from being freed when | |
2081 | * the page state that is going to change is the only thing | |
2082 | * preventing the page itself from being freed. E.g. writeback | |
2083 | * doesn't hold a page reference and relies on PG_writeback to | |
2084 | * keep off truncation, migration and so forth. | |
2085 | */ | |
d7365e78 JW |
2086 | rcu_read_lock(); |
2087 | ||
2088 | if (mem_cgroup_disabled()) | |
739f79fc | 2089 | return NULL; |
89c06bd5 | 2090 | again: |
9da7b521 | 2091 | memcg = head->mem_cgroup; |
29833315 | 2092 | if (unlikely(!memcg)) |
739f79fc | 2093 | return NULL; |
d7365e78 | 2094 | |
bdcbb659 | 2095 | if (atomic_read(&memcg->moving_account) <= 0) |
739f79fc | 2096 | return memcg; |
89c06bd5 | 2097 | |
6de22619 | 2098 | spin_lock_irqsave(&memcg->move_lock, flags); |
9da7b521 | 2099 | if (memcg != head->mem_cgroup) { |
6de22619 | 2100 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
2101 | goto again; |
2102 | } | |
6de22619 JW |
2103 | |
2104 | /* | |
2105 | * When charge migration first begins, we can have locked and | |
2106 | * unlocked page stat updates happening concurrently. Track | |
81f8c3a4 | 2107 | * the task who has the lock for unlock_page_memcg(). |
6de22619 JW |
2108 | */ |
2109 | memcg->move_lock_task = current; | |
2110 | memcg->move_lock_flags = flags; | |
d7365e78 | 2111 | |
739f79fc | 2112 | return memcg; |
89c06bd5 | 2113 | } |
81f8c3a4 | 2114 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 2115 | |
d7365e78 | 2116 | /** |
739f79fc JW |
2117 | * __unlock_page_memcg - unlock and unpin a memcg |
2118 | * @memcg: the memcg | |
2119 | * | |
2120 | * Unlock and unpin a memcg returned by lock_page_memcg(). | |
d7365e78 | 2121 | */ |
739f79fc | 2122 | void __unlock_page_memcg(struct mem_cgroup *memcg) |
89c06bd5 | 2123 | { |
6de22619 JW |
2124 | if (memcg && memcg->move_lock_task == current) { |
2125 | unsigned long flags = memcg->move_lock_flags; | |
2126 | ||
2127 | memcg->move_lock_task = NULL; | |
2128 | memcg->move_lock_flags = 0; | |
2129 | ||
2130 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2131 | } | |
89c06bd5 | 2132 | |
d7365e78 | 2133 | rcu_read_unlock(); |
89c06bd5 | 2134 | } |
739f79fc JW |
2135 | |
2136 | /** | |
2137 | * unlock_page_memcg - unlock a page->mem_cgroup binding | |
2138 | * @page: the page | |
2139 | */ | |
2140 | void unlock_page_memcg(struct page *page) | |
2141 | { | |
9da7b521 JW |
2142 | struct page *head = compound_head(page); |
2143 | ||
2144 | __unlock_page_memcg(head->mem_cgroup); | |
739f79fc | 2145 | } |
81f8c3a4 | 2146 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 2147 | |
cdec2e42 KH |
2148 | struct memcg_stock_pcp { |
2149 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2150 | unsigned int nr_pages; |
bf4f0599 RG |
2151 | |
2152 | #ifdef CONFIG_MEMCG_KMEM | |
2153 | struct obj_cgroup *cached_objcg; | |
2154 | unsigned int nr_bytes; | |
2155 | #endif | |
2156 | ||
cdec2e42 | 2157 | struct work_struct work; |
26fe6168 | 2158 | unsigned long flags; |
a0db00fc | 2159 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2160 | }; |
2161 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2162 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2163 | |
bf4f0599 RG |
2164 | #ifdef CONFIG_MEMCG_KMEM |
2165 | static void drain_obj_stock(struct memcg_stock_pcp *stock); | |
2166 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2167 | struct mem_cgroup *root_memcg); | |
2168 | ||
2169 | #else | |
2170 | static inline void drain_obj_stock(struct memcg_stock_pcp *stock) | |
2171 | { | |
2172 | } | |
2173 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2174 | struct mem_cgroup *root_memcg) | |
2175 | { | |
2176 | return false; | |
2177 | } | |
2178 | #endif | |
2179 | ||
a0956d54 SS |
2180 | /** |
2181 | * consume_stock: Try to consume stocked charge on this cpu. | |
2182 | * @memcg: memcg to consume from. | |
2183 | * @nr_pages: how many pages to charge. | |
2184 | * | |
2185 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2186 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2187 | * service an allocation will refill the stock. | |
2188 | * | |
2189 | * returns true if successful, false otherwise. | |
cdec2e42 | 2190 | */ |
a0956d54 | 2191 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2192 | { |
2193 | struct memcg_stock_pcp *stock; | |
db2ba40c | 2194 | unsigned long flags; |
3e32cb2e | 2195 | bool ret = false; |
cdec2e42 | 2196 | |
a983b5eb | 2197 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 2198 | return ret; |
a0956d54 | 2199 | |
db2ba40c JW |
2200 | local_irq_save(flags); |
2201 | ||
2202 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 2203 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2204 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2205 | ret = true; |
2206 | } | |
db2ba40c JW |
2207 | |
2208 | local_irq_restore(flags); | |
2209 | ||
cdec2e42 KH |
2210 | return ret; |
2211 | } | |
2212 | ||
2213 | /* | |
3e32cb2e | 2214 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2215 | */ |
2216 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2217 | { | |
2218 | struct mem_cgroup *old = stock->cached; | |
2219 | ||
1a3e1f40 JW |
2220 | if (!old) |
2221 | return; | |
2222 | ||
11c9ea4e | 2223 | if (stock->nr_pages) { |
3e32cb2e | 2224 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 2225 | if (do_memsw_account()) |
3e32cb2e | 2226 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
11c9ea4e | 2227 | stock->nr_pages = 0; |
cdec2e42 | 2228 | } |
1a3e1f40 JW |
2229 | |
2230 | css_put(&old->css); | |
cdec2e42 | 2231 | stock->cached = NULL; |
cdec2e42 KH |
2232 | } |
2233 | ||
cdec2e42 KH |
2234 | static void drain_local_stock(struct work_struct *dummy) |
2235 | { | |
db2ba40c JW |
2236 | struct memcg_stock_pcp *stock; |
2237 | unsigned long flags; | |
2238 | ||
72f0184c MH |
2239 | /* |
2240 | * The only protection from memory hotplug vs. drain_stock races is | |
2241 | * that we always operate on local CPU stock here with IRQ disabled | |
2242 | */ | |
db2ba40c JW |
2243 | local_irq_save(flags); |
2244 | ||
2245 | stock = this_cpu_ptr(&memcg_stock); | |
bf4f0599 | 2246 | drain_obj_stock(stock); |
cdec2e42 | 2247 | drain_stock(stock); |
26fe6168 | 2248 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c JW |
2249 | |
2250 | local_irq_restore(flags); | |
cdec2e42 KH |
2251 | } |
2252 | ||
2253 | /* | |
3e32cb2e | 2254 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2255 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2256 | */ |
c0ff4b85 | 2257 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2258 | { |
db2ba40c JW |
2259 | struct memcg_stock_pcp *stock; |
2260 | unsigned long flags; | |
2261 | ||
2262 | local_irq_save(flags); | |
cdec2e42 | 2263 | |
db2ba40c | 2264 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 2265 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2266 | drain_stock(stock); |
1a3e1f40 | 2267 | css_get(&memcg->css); |
c0ff4b85 | 2268 | stock->cached = memcg; |
cdec2e42 | 2269 | } |
11c9ea4e | 2270 | stock->nr_pages += nr_pages; |
db2ba40c | 2271 | |
a983b5eb | 2272 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 RG |
2273 | drain_stock(stock); |
2274 | ||
db2ba40c | 2275 | local_irq_restore(flags); |
cdec2e42 KH |
2276 | } |
2277 | ||
2278 | /* | |
c0ff4b85 | 2279 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2280 | * of the hierarchy under it. |
cdec2e42 | 2281 | */ |
6d3d6aa2 | 2282 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2283 | { |
26fe6168 | 2284 | int cpu, curcpu; |
d38144b7 | 2285 | |
6d3d6aa2 JW |
2286 | /* If someone's already draining, avoid adding running more workers. */ |
2287 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2288 | return; | |
72f0184c MH |
2289 | /* |
2290 | * Notify other cpus that system-wide "drain" is running | |
2291 | * We do not care about races with the cpu hotplug because cpu down | |
2292 | * as well as workers from this path always operate on the local | |
2293 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2294 | */ | |
5af12d0e | 2295 | curcpu = get_cpu(); |
cdec2e42 KH |
2296 | for_each_online_cpu(cpu) { |
2297 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2298 | struct mem_cgroup *memcg; |
e1a366be | 2299 | bool flush = false; |
26fe6168 | 2300 | |
e1a366be | 2301 | rcu_read_lock(); |
c0ff4b85 | 2302 | memcg = stock->cached; |
e1a366be RG |
2303 | if (memcg && stock->nr_pages && |
2304 | mem_cgroup_is_descendant(memcg, root_memcg)) | |
2305 | flush = true; | |
bf4f0599 RG |
2306 | if (obj_stock_flush_required(stock, root_memcg)) |
2307 | flush = true; | |
e1a366be RG |
2308 | rcu_read_unlock(); |
2309 | ||
2310 | if (flush && | |
2311 | !test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { | |
d1a05b69 MH |
2312 | if (cpu == curcpu) |
2313 | drain_local_stock(&stock->work); | |
2314 | else | |
2315 | schedule_work_on(cpu, &stock->work); | |
2316 | } | |
cdec2e42 | 2317 | } |
5af12d0e | 2318 | put_cpu(); |
9f50fad6 | 2319 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2320 | } |
2321 | ||
308167fc | 2322 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
cdec2e42 | 2323 | { |
cdec2e42 | 2324 | struct memcg_stock_pcp *stock; |
42a30035 | 2325 | struct mem_cgroup *memcg, *mi; |
cdec2e42 | 2326 | |
cdec2e42 KH |
2327 | stock = &per_cpu(memcg_stock, cpu); |
2328 | drain_stock(stock); | |
a983b5eb JW |
2329 | |
2330 | for_each_mem_cgroup(memcg) { | |
2331 | int i; | |
2332 | ||
2333 | for (i = 0; i < MEMCG_NR_STAT; i++) { | |
2334 | int nid; | |
2335 | long x; | |
2336 | ||
871789d4 | 2337 | x = this_cpu_xchg(memcg->vmstats_percpu->stat[i], 0); |
815744d7 | 2338 | if (x) |
42a30035 JW |
2339 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2340 | atomic_long_add(x, &memcg->vmstats[i]); | |
a983b5eb JW |
2341 | |
2342 | if (i >= NR_VM_NODE_STAT_ITEMS) | |
2343 | continue; | |
2344 | ||
2345 | for_each_node(nid) { | |
2346 | struct mem_cgroup_per_node *pn; | |
2347 | ||
2348 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
2349 | x = this_cpu_xchg(pn->lruvec_stat_cpu->count[i], 0); | |
815744d7 | 2350 | if (x) |
42a30035 JW |
2351 | do { |
2352 | atomic_long_add(x, &pn->lruvec_stat[i]); | |
2353 | } while ((pn = parent_nodeinfo(pn, nid))); | |
a983b5eb JW |
2354 | } |
2355 | } | |
2356 | ||
e27be240 | 2357 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { |
a983b5eb JW |
2358 | long x; |
2359 | ||
871789d4 | 2360 | x = this_cpu_xchg(memcg->vmstats_percpu->events[i], 0); |
815744d7 | 2361 | if (x) |
42a30035 JW |
2362 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2363 | atomic_long_add(x, &memcg->vmevents[i]); | |
a983b5eb JW |
2364 | } |
2365 | } | |
2366 | ||
308167fc | 2367 | return 0; |
cdec2e42 KH |
2368 | } |
2369 | ||
b3ff9291 CD |
2370 | static unsigned long reclaim_high(struct mem_cgroup *memcg, |
2371 | unsigned int nr_pages, | |
2372 | gfp_t gfp_mask) | |
f7e1cb6e | 2373 | { |
b3ff9291 CD |
2374 | unsigned long nr_reclaimed = 0; |
2375 | ||
f7e1cb6e | 2376 | do { |
d1663a90 JK |
2377 | if (page_counter_read(&memcg->memory) <= |
2378 | READ_ONCE(memcg->memory.high)) | |
f7e1cb6e | 2379 | continue; |
e27be240 | 2380 | memcg_memory_event(memcg, MEMCG_HIGH); |
b3ff9291 CD |
2381 | nr_reclaimed += try_to_free_mem_cgroup_pages(memcg, nr_pages, |
2382 | gfp_mask, true); | |
4bf17307 CD |
2383 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2384 | !mem_cgroup_is_root(memcg)); | |
b3ff9291 CD |
2385 | |
2386 | return nr_reclaimed; | |
f7e1cb6e JW |
2387 | } |
2388 | ||
2389 | static void high_work_func(struct work_struct *work) | |
2390 | { | |
2391 | struct mem_cgroup *memcg; | |
2392 | ||
2393 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2394 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2395 | } |
2396 | ||
0e4b01df CD |
2397 | /* |
2398 | * Clamp the maximum sleep time per allocation batch to 2 seconds. This is | |
2399 | * enough to still cause a significant slowdown in most cases, while still | |
2400 | * allowing diagnostics and tracing to proceed without becoming stuck. | |
2401 | */ | |
2402 | #define MEMCG_MAX_HIGH_DELAY_JIFFIES (2UL*HZ) | |
2403 | ||
2404 | /* | |
2405 | * When calculating the delay, we use these either side of the exponentiation to | |
2406 | * maintain precision and scale to a reasonable number of jiffies (see the table | |
2407 | * below. | |
2408 | * | |
2409 | * - MEMCG_DELAY_PRECISION_SHIFT: Extra precision bits while translating the | |
2410 | * overage ratio to a delay. | |
2411 | * - MEMCG_DELAY_SCALING_SHIFT: The number of bits to scale down down the | |
2412 | * proposed penalty in order to reduce to a reasonable number of jiffies, and | |
2413 | * to produce a reasonable delay curve. | |
2414 | * | |
2415 | * MEMCG_DELAY_SCALING_SHIFT just happens to be a number that produces a | |
2416 | * reasonable delay curve compared to precision-adjusted overage, not | |
2417 | * penalising heavily at first, but still making sure that growth beyond the | |
2418 | * limit penalises misbehaviour cgroups by slowing them down exponentially. For | |
2419 | * example, with a high of 100 megabytes: | |
2420 | * | |
2421 | * +-------+------------------------+ | |
2422 | * | usage | time to allocate in ms | | |
2423 | * +-------+------------------------+ | |
2424 | * | 100M | 0 | | |
2425 | * | 101M | 6 | | |
2426 | * | 102M | 25 | | |
2427 | * | 103M | 57 | | |
2428 | * | 104M | 102 | | |
2429 | * | 105M | 159 | | |
2430 | * | 106M | 230 | | |
2431 | * | 107M | 313 | | |
2432 | * | 108M | 409 | | |
2433 | * | 109M | 518 | | |
2434 | * | 110M | 639 | | |
2435 | * | 111M | 774 | | |
2436 | * | 112M | 921 | | |
2437 | * | 113M | 1081 | | |
2438 | * | 114M | 1254 | | |
2439 | * | 115M | 1439 | | |
2440 | * | 116M | 1638 | | |
2441 | * | 117M | 1849 | | |
2442 | * | 118M | 2000 | | |
2443 | * | 119M | 2000 | | |
2444 | * | 120M | 2000 | | |
2445 | * +-------+------------------------+ | |
2446 | */ | |
2447 | #define MEMCG_DELAY_PRECISION_SHIFT 20 | |
2448 | #define MEMCG_DELAY_SCALING_SHIFT 14 | |
2449 | ||
8a5dbc65 | 2450 | static u64 calculate_overage(unsigned long usage, unsigned long high) |
b23afb93 | 2451 | { |
8a5dbc65 | 2452 | u64 overage; |
b23afb93 | 2453 | |
8a5dbc65 JK |
2454 | if (usage <= high) |
2455 | return 0; | |
e26733e0 | 2456 | |
8a5dbc65 JK |
2457 | /* |
2458 | * Prevent division by 0 in overage calculation by acting as if | |
2459 | * it was a threshold of 1 page | |
2460 | */ | |
2461 | high = max(high, 1UL); | |
9b8b1754 | 2462 | |
8a5dbc65 JK |
2463 | overage = usage - high; |
2464 | overage <<= MEMCG_DELAY_PRECISION_SHIFT; | |
2465 | return div64_u64(overage, high); | |
2466 | } | |
e26733e0 | 2467 | |
8a5dbc65 JK |
2468 | static u64 mem_find_max_overage(struct mem_cgroup *memcg) |
2469 | { | |
2470 | u64 overage, max_overage = 0; | |
e26733e0 | 2471 | |
8a5dbc65 JK |
2472 | do { |
2473 | overage = calculate_overage(page_counter_read(&memcg->memory), | |
d1663a90 | 2474 | READ_ONCE(memcg->memory.high)); |
8a5dbc65 | 2475 | max_overage = max(overage, max_overage); |
e26733e0 CD |
2476 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2477 | !mem_cgroup_is_root(memcg)); | |
2478 | ||
8a5dbc65 JK |
2479 | return max_overage; |
2480 | } | |
2481 | ||
4b82ab4f JK |
2482 | static u64 swap_find_max_overage(struct mem_cgroup *memcg) |
2483 | { | |
2484 | u64 overage, max_overage = 0; | |
2485 | ||
2486 | do { | |
2487 | overage = calculate_overage(page_counter_read(&memcg->swap), | |
2488 | READ_ONCE(memcg->swap.high)); | |
2489 | if (overage) | |
2490 | memcg_memory_event(memcg, MEMCG_SWAP_HIGH); | |
2491 | max_overage = max(overage, max_overage); | |
2492 | } while ((memcg = parent_mem_cgroup(memcg)) && | |
2493 | !mem_cgroup_is_root(memcg)); | |
2494 | ||
2495 | return max_overage; | |
2496 | } | |
2497 | ||
8a5dbc65 JK |
2498 | /* |
2499 | * Get the number of jiffies that we should penalise a mischievous cgroup which | |
2500 | * is exceeding its memory.high by checking both it and its ancestors. | |
2501 | */ | |
2502 | static unsigned long calculate_high_delay(struct mem_cgroup *memcg, | |
2503 | unsigned int nr_pages, | |
2504 | u64 max_overage) | |
2505 | { | |
2506 | unsigned long penalty_jiffies; | |
2507 | ||
e26733e0 CD |
2508 | if (!max_overage) |
2509 | return 0; | |
0e4b01df CD |
2510 | |
2511 | /* | |
0e4b01df CD |
2512 | * We use overage compared to memory.high to calculate the number of |
2513 | * jiffies to sleep (penalty_jiffies). Ideally this value should be | |
2514 | * fairly lenient on small overages, and increasingly harsh when the | |
2515 | * memcg in question makes it clear that it has no intention of stopping | |
2516 | * its crazy behaviour, so we exponentially increase the delay based on | |
2517 | * overage amount. | |
2518 | */ | |
e26733e0 CD |
2519 | penalty_jiffies = max_overage * max_overage * HZ; |
2520 | penalty_jiffies >>= MEMCG_DELAY_PRECISION_SHIFT; | |
2521 | penalty_jiffies >>= MEMCG_DELAY_SCALING_SHIFT; | |
0e4b01df CD |
2522 | |
2523 | /* | |
2524 | * Factor in the task's own contribution to the overage, such that four | |
2525 | * N-sized allocations are throttled approximately the same as one | |
2526 | * 4N-sized allocation. | |
2527 | * | |
2528 | * MEMCG_CHARGE_BATCH pages is nominal, so work out how much smaller or | |
2529 | * larger the current charge patch is than that. | |
2530 | */ | |
ff144e69 | 2531 | return penalty_jiffies * nr_pages / MEMCG_CHARGE_BATCH; |
e26733e0 CD |
2532 | } |
2533 | ||
2534 | /* | |
2535 | * Scheduled by try_charge() to be executed from the userland return path | |
2536 | * and reclaims memory over the high limit. | |
2537 | */ | |
2538 | void mem_cgroup_handle_over_high(void) | |
2539 | { | |
2540 | unsigned long penalty_jiffies; | |
2541 | unsigned long pflags; | |
b3ff9291 | 2542 | unsigned long nr_reclaimed; |
e26733e0 | 2543 | unsigned int nr_pages = current->memcg_nr_pages_over_high; |
d977aa93 | 2544 | int nr_retries = MAX_RECLAIM_RETRIES; |
e26733e0 | 2545 | struct mem_cgroup *memcg; |
b3ff9291 | 2546 | bool in_retry = false; |
e26733e0 CD |
2547 | |
2548 | if (likely(!nr_pages)) | |
2549 | return; | |
2550 | ||
2551 | memcg = get_mem_cgroup_from_mm(current->mm); | |
e26733e0 CD |
2552 | current->memcg_nr_pages_over_high = 0; |
2553 | ||
b3ff9291 CD |
2554 | retry_reclaim: |
2555 | /* | |
2556 | * The allocating task should reclaim at least the batch size, but for | |
2557 | * subsequent retries we only want to do what's necessary to prevent oom | |
2558 | * or breaching resource isolation. | |
2559 | * | |
2560 | * This is distinct from memory.max or page allocator behaviour because | |
2561 | * memory.high is currently batched, whereas memory.max and the page | |
2562 | * allocator run every time an allocation is made. | |
2563 | */ | |
2564 | nr_reclaimed = reclaim_high(memcg, | |
2565 | in_retry ? SWAP_CLUSTER_MAX : nr_pages, | |
2566 | GFP_KERNEL); | |
2567 | ||
e26733e0 CD |
2568 | /* |
2569 | * memory.high is breached and reclaim is unable to keep up. Throttle | |
2570 | * allocators proactively to slow down excessive growth. | |
2571 | */ | |
8a5dbc65 JK |
2572 | penalty_jiffies = calculate_high_delay(memcg, nr_pages, |
2573 | mem_find_max_overage(memcg)); | |
0e4b01df | 2574 | |
4b82ab4f JK |
2575 | penalty_jiffies += calculate_high_delay(memcg, nr_pages, |
2576 | swap_find_max_overage(memcg)); | |
2577 | ||
ff144e69 JK |
2578 | /* |
2579 | * Clamp the max delay per usermode return so as to still keep the | |
2580 | * application moving forwards and also permit diagnostics, albeit | |
2581 | * extremely slowly. | |
2582 | */ | |
2583 | penalty_jiffies = min(penalty_jiffies, MEMCG_MAX_HIGH_DELAY_JIFFIES); | |
2584 | ||
0e4b01df CD |
2585 | /* |
2586 | * Don't sleep if the amount of jiffies this memcg owes us is so low | |
2587 | * that it's not even worth doing, in an attempt to be nice to those who | |
2588 | * go only a small amount over their memory.high value and maybe haven't | |
2589 | * been aggressively reclaimed enough yet. | |
2590 | */ | |
2591 | if (penalty_jiffies <= HZ / 100) | |
2592 | goto out; | |
2593 | ||
b3ff9291 CD |
2594 | /* |
2595 | * If reclaim is making forward progress but we're still over | |
2596 | * memory.high, we want to encourage that rather than doing allocator | |
2597 | * throttling. | |
2598 | */ | |
2599 | if (nr_reclaimed || nr_retries--) { | |
2600 | in_retry = true; | |
2601 | goto retry_reclaim; | |
2602 | } | |
2603 | ||
0e4b01df CD |
2604 | /* |
2605 | * If we exit early, we're guaranteed to die (since | |
2606 | * schedule_timeout_killable sets TASK_KILLABLE). This means we don't | |
2607 | * need to account for any ill-begotten jiffies to pay them off later. | |
2608 | */ | |
2609 | psi_memstall_enter(&pflags); | |
2610 | schedule_timeout_killable(penalty_jiffies); | |
2611 | psi_memstall_leave(&pflags); | |
2612 | ||
2613 | out: | |
2614 | css_put(&memcg->css); | |
b23afb93 TH |
2615 | } |
2616 | ||
00501b53 JW |
2617 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2618 | unsigned int nr_pages) | |
8a9f3ccd | 2619 | { |
a983b5eb | 2620 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
d977aa93 | 2621 | int nr_retries = MAX_RECLAIM_RETRIES; |
6539cc05 | 2622 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2623 | struct page_counter *counter; |
6539cc05 | 2624 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2625 | bool may_swap = true; |
2626 | bool drained = false; | |
29ef680a | 2627 | enum oom_status oom_status; |
a636b327 | 2628 | |
ce00a967 | 2629 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 2630 | return 0; |
6539cc05 | 2631 | retry: |
b6b6cc72 | 2632 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2633 | return 0; |
8a9f3ccd | 2634 | |
7941d214 | 2635 | if (!do_memsw_account() || |
6071ca52 JW |
2636 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2637 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2638 | goto done_restock; |
7941d214 | 2639 | if (do_memsw_account()) |
3e32cb2e JW |
2640 | page_counter_uncharge(&memcg->memsw, batch); |
2641 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2642 | } else { |
3e32cb2e | 2643 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2644 | may_swap = false; |
3fbe7244 | 2645 | } |
7a81b88c | 2646 | |
6539cc05 JW |
2647 | if (batch > nr_pages) { |
2648 | batch = nr_pages; | |
2649 | goto retry; | |
2650 | } | |
6d61ef40 | 2651 | |
869712fd JW |
2652 | /* |
2653 | * Memcg doesn't have a dedicated reserve for atomic | |
2654 | * allocations. But like the global atomic pool, we need to | |
2655 | * put the burden of reclaim on regular allocation requests | |
2656 | * and let these go through as privileged allocations. | |
2657 | */ | |
2658 | if (gfp_mask & __GFP_ATOMIC) | |
2659 | goto force; | |
2660 | ||
06b078fc JW |
2661 | /* |
2662 | * Unlike in global OOM situations, memcg is not in a physical | |
2663 | * memory shortage. Allow dying and OOM-killed tasks to | |
2664 | * bypass the last charges so that they can exit quickly and | |
2665 | * free their memory. | |
2666 | */ | |
7775face | 2667 | if (unlikely(should_force_charge())) |
10d53c74 | 2668 | goto force; |
06b078fc | 2669 | |
89a28483 JW |
2670 | /* |
2671 | * Prevent unbounded recursion when reclaim operations need to | |
2672 | * allocate memory. This might exceed the limits temporarily, | |
2673 | * but we prefer facilitating memory reclaim and getting back | |
2674 | * under the limit over triggering OOM kills in these cases. | |
2675 | */ | |
2676 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2677 | goto force; | |
2678 | ||
06b078fc JW |
2679 | if (unlikely(task_in_memcg_oom(current))) |
2680 | goto nomem; | |
2681 | ||
d0164adc | 2682 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2683 | goto nomem; |
4b534334 | 2684 | |
e27be240 | 2685 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
241994ed | 2686 | |
b70a2a21 JW |
2687 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2688 | gfp_mask, may_swap); | |
6539cc05 | 2689 | |
61e02c74 | 2690 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2691 | goto retry; |
28c34c29 | 2692 | |
b70a2a21 | 2693 | if (!drained) { |
6d3d6aa2 | 2694 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2695 | drained = true; |
2696 | goto retry; | |
2697 | } | |
2698 | ||
28c34c29 JW |
2699 | if (gfp_mask & __GFP_NORETRY) |
2700 | goto nomem; | |
6539cc05 JW |
2701 | /* |
2702 | * Even though the limit is exceeded at this point, reclaim | |
2703 | * may have been able to free some pages. Retry the charge | |
2704 | * before killing the task. | |
2705 | * | |
2706 | * Only for regular pages, though: huge pages are rather | |
2707 | * unlikely to succeed so close to the limit, and we fall back | |
2708 | * to regular pages anyway in case of failure. | |
2709 | */ | |
61e02c74 | 2710 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2711 | goto retry; |
2712 | /* | |
2713 | * At task move, charge accounts can be doubly counted. So, it's | |
2714 | * better to wait until the end of task_move if something is going on. | |
2715 | */ | |
2716 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2717 | goto retry; | |
2718 | ||
9b130619 JW |
2719 | if (nr_retries--) |
2720 | goto retry; | |
2721 | ||
38d38493 | 2722 | if (gfp_mask & __GFP_RETRY_MAYFAIL) |
29ef680a MH |
2723 | goto nomem; |
2724 | ||
06b078fc | 2725 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2726 | goto force; |
06b078fc | 2727 | |
6539cc05 | 2728 | if (fatal_signal_pending(current)) |
10d53c74 | 2729 | goto force; |
6539cc05 | 2730 | |
29ef680a MH |
2731 | /* |
2732 | * keep retrying as long as the memcg oom killer is able to make | |
2733 | * a forward progress or bypass the charge if the oom killer | |
2734 | * couldn't make any progress. | |
2735 | */ | |
2736 | oom_status = mem_cgroup_oom(mem_over_limit, gfp_mask, | |
3608de07 | 2737 | get_order(nr_pages * PAGE_SIZE)); |
29ef680a MH |
2738 | switch (oom_status) { |
2739 | case OOM_SUCCESS: | |
d977aa93 | 2740 | nr_retries = MAX_RECLAIM_RETRIES; |
29ef680a MH |
2741 | goto retry; |
2742 | case OOM_FAILED: | |
2743 | goto force; | |
2744 | default: | |
2745 | goto nomem; | |
2746 | } | |
7a81b88c | 2747 | nomem: |
6d1fdc48 | 2748 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2749 | return -ENOMEM; |
10d53c74 TH |
2750 | force: |
2751 | /* | |
2752 | * The allocation either can't fail or will lead to more memory | |
2753 | * being freed very soon. Allow memory usage go over the limit | |
2754 | * temporarily by force charging it. | |
2755 | */ | |
2756 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2757 | if (do_memsw_account()) |
10d53c74 | 2758 | page_counter_charge(&memcg->memsw, nr_pages); |
10d53c74 TH |
2759 | |
2760 | return 0; | |
6539cc05 JW |
2761 | |
2762 | done_restock: | |
2763 | if (batch > nr_pages) | |
2764 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2765 | |
241994ed | 2766 | /* |
b23afb93 TH |
2767 | * If the hierarchy is above the normal consumption range, schedule |
2768 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2769 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2770 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2771 | * not recorded as it most likely matches current's and won't | |
2772 | * change in the meantime. As high limit is checked again before | |
2773 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2774 | */ |
2775 | do { | |
4b82ab4f JK |
2776 | bool mem_high, swap_high; |
2777 | ||
2778 | mem_high = page_counter_read(&memcg->memory) > | |
2779 | READ_ONCE(memcg->memory.high); | |
2780 | swap_high = page_counter_read(&memcg->swap) > | |
2781 | READ_ONCE(memcg->swap.high); | |
2782 | ||
2783 | /* Don't bother a random interrupted task */ | |
2784 | if (in_interrupt()) { | |
2785 | if (mem_high) { | |
f7e1cb6e JW |
2786 | schedule_work(&memcg->high_work); |
2787 | break; | |
2788 | } | |
4b82ab4f JK |
2789 | continue; |
2790 | } | |
2791 | ||
2792 | if (mem_high || swap_high) { | |
2793 | /* | |
2794 | * The allocating tasks in this cgroup will need to do | |
2795 | * reclaim or be throttled to prevent further growth | |
2796 | * of the memory or swap footprints. | |
2797 | * | |
2798 | * Target some best-effort fairness between the tasks, | |
2799 | * and distribute reclaim work and delay penalties | |
2800 | * based on how much each task is actually allocating. | |
2801 | */ | |
9516a18a | 2802 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2803 | set_notify_resume(current); |
2804 | break; | |
2805 | } | |
241994ed | 2806 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2807 | |
2808 | return 0; | |
7a81b88c | 2809 | } |
8a9f3ccd | 2810 | |
f0e45fb4 | 2811 | #if defined(CONFIG_MEMCG_KMEM) || defined(CONFIG_MMU) |
00501b53 | 2812 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2813 | { |
ce00a967 JW |
2814 | if (mem_cgroup_is_root(memcg)) |
2815 | return; | |
2816 | ||
3e32cb2e | 2817 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2818 | if (do_memsw_account()) |
3e32cb2e | 2819 | page_counter_uncharge(&memcg->memsw, nr_pages); |
d01dd17f | 2820 | } |
f0e45fb4 | 2821 | #endif |
d01dd17f | 2822 | |
d9eb1ea2 | 2823 | static void commit_charge(struct page *page, struct mem_cgroup *memcg) |
0a31bc97 | 2824 | { |
1306a85a | 2825 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
0a31bc97 | 2826 | /* |
a0b5b414 | 2827 | * Any of the following ensures page->mem_cgroup stability: |
0a31bc97 | 2828 | * |
a0b5b414 JW |
2829 | * - the page lock |
2830 | * - LRU isolation | |
2831 | * - lock_page_memcg() | |
2832 | * - exclusive reference | |
0a31bc97 | 2833 | */ |
1306a85a | 2834 | page->mem_cgroup = memcg; |
7a81b88c | 2835 | } |
66e1707b | 2836 | |
84c07d11 | 2837 | #ifdef CONFIG_MEMCG_KMEM |
10befea9 RG |
2838 | int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s, |
2839 | gfp_t gfp) | |
2840 | { | |
2841 | unsigned int objects = objs_per_slab_page(s, page); | |
2842 | void *vec; | |
2843 | ||
2844 | vec = kcalloc_node(objects, sizeof(struct obj_cgroup *), gfp, | |
2845 | page_to_nid(page)); | |
2846 | if (!vec) | |
2847 | return -ENOMEM; | |
2848 | ||
2849 | if (cmpxchg(&page->obj_cgroups, NULL, | |
2850 | (struct obj_cgroup **) ((unsigned long)vec | 0x1UL))) | |
2851 | kfree(vec); | |
2852 | else | |
2853 | kmemleak_not_leak(vec); | |
2854 | ||
2855 | return 0; | |
2856 | } | |
2857 | ||
8380ce47 RG |
2858 | /* |
2859 | * Returns a pointer to the memory cgroup to which the kernel object is charged. | |
2860 | * | |
2861 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), | |
2862 | * cgroup_mutex, etc. | |
2863 | */ | |
2864 | struct mem_cgroup *mem_cgroup_from_obj(void *p) | |
2865 | { | |
2866 | struct page *page; | |
2867 | ||
2868 | if (mem_cgroup_disabled()) | |
2869 | return NULL; | |
2870 | ||
2871 | page = virt_to_head_page(p); | |
2872 | ||
2873 | /* | |
9855609b RG |
2874 | * Slab objects are accounted individually, not per-page. |
2875 | * Memcg membership data for each individual object is saved in | |
2876 | * the page->obj_cgroups. | |
8380ce47 | 2877 | */ |
9855609b RG |
2878 | if (page_has_obj_cgroups(page)) { |
2879 | struct obj_cgroup *objcg; | |
2880 | unsigned int off; | |
2881 | ||
2882 | off = obj_to_index(page->slab_cache, page, p); | |
2883 | objcg = page_obj_cgroups(page)[off]; | |
10befea9 RG |
2884 | if (objcg) |
2885 | return obj_cgroup_memcg(objcg); | |
2886 | ||
2887 | return NULL; | |
9855609b | 2888 | } |
8380ce47 RG |
2889 | |
2890 | /* All other pages use page->mem_cgroup */ | |
2891 | return page->mem_cgroup; | |
2892 | } | |
2893 | ||
bf4f0599 RG |
2894 | __always_inline struct obj_cgroup *get_obj_cgroup_from_current(void) |
2895 | { | |
2896 | struct obj_cgroup *objcg = NULL; | |
2897 | struct mem_cgroup *memcg; | |
2898 | ||
2899 | if (unlikely(!current->mm && !current->active_memcg)) | |
2900 | return NULL; | |
2901 | ||
2902 | rcu_read_lock(); | |
2903 | if (unlikely(current->active_memcg)) | |
2904 | memcg = rcu_dereference(current->active_memcg); | |
2905 | else | |
2906 | memcg = mem_cgroup_from_task(current); | |
2907 | ||
2908 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { | |
2909 | objcg = rcu_dereference(memcg->objcg); | |
2910 | if (objcg && obj_cgroup_tryget(objcg)) | |
2911 | break; | |
2912 | } | |
2913 | rcu_read_unlock(); | |
2914 | ||
2915 | return objcg; | |
2916 | } | |
2917 | ||
f3bb3043 | 2918 | static int memcg_alloc_cache_id(void) |
55007d84 | 2919 | { |
f3bb3043 VD |
2920 | int id, size; |
2921 | int err; | |
2922 | ||
dbcf73e2 | 2923 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2924 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2925 | if (id < 0) | |
2926 | return id; | |
55007d84 | 2927 | |
dbcf73e2 | 2928 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2929 | return id; |
2930 | ||
2931 | /* | |
2932 | * There's no space for the new id in memcg_caches arrays, | |
2933 | * so we have to grow them. | |
2934 | */ | |
05257a1a | 2935 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2936 | |
2937 | size = 2 * (id + 1); | |
55007d84 GC |
2938 | if (size < MEMCG_CACHES_MIN_SIZE) |
2939 | size = MEMCG_CACHES_MIN_SIZE; | |
2940 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2941 | size = MEMCG_CACHES_MAX_SIZE; | |
2942 | ||
9855609b | 2943 | err = memcg_update_all_list_lrus(size); |
05257a1a VD |
2944 | if (!err) |
2945 | memcg_nr_cache_ids = size; | |
2946 | ||
2947 | up_write(&memcg_cache_ids_sem); | |
2948 | ||
f3bb3043 | 2949 | if (err) { |
dbcf73e2 | 2950 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2951 | return err; |
2952 | } | |
2953 | return id; | |
2954 | } | |
2955 | ||
2956 | static void memcg_free_cache_id(int id) | |
2957 | { | |
dbcf73e2 | 2958 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2959 | } |
2960 | ||
45264778 | 2961 | /** |
4b13f64d | 2962 | * __memcg_kmem_charge: charge a number of kernel pages to a memcg |
10eaec2f | 2963 | * @memcg: memory cgroup to charge |
45264778 | 2964 | * @gfp: reclaim mode |
92d0510c | 2965 | * @nr_pages: number of pages to charge |
45264778 VD |
2966 | * |
2967 | * Returns 0 on success, an error code on failure. | |
2968 | */ | |
4b13f64d RG |
2969 | int __memcg_kmem_charge(struct mem_cgroup *memcg, gfp_t gfp, |
2970 | unsigned int nr_pages) | |
7ae1e1d0 | 2971 | { |
f3ccb2c4 | 2972 | struct page_counter *counter; |
7ae1e1d0 GC |
2973 | int ret; |
2974 | ||
f3ccb2c4 | 2975 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 2976 | if (ret) |
f3ccb2c4 | 2977 | return ret; |
52c29b04 JW |
2978 | |
2979 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
2980 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
e55d9d9b MH |
2981 | |
2982 | /* | |
2983 | * Enforce __GFP_NOFAIL allocation because callers are not | |
2984 | * prepared to see failures and likely do not have any failure | |
2985 | * handling code. | |
2986 | */ | |
2987 | if (gfp & __GFP_NOFAIL) { | |
2988 | page_counter_charge(&memcg->kmem, nr_pages); | |
2989 | return 0; | |
2990 | } | |
52c29b04 JW |
2991 | cancel_charge(memcg, nr_pages); |
2992 | return -ENOMEM; | |
7ae1e1d0 | 2993 | } |
f3ccb2c4 | 2994 | return 0; |
7ae1e1d0 GC |
2995 | } |
2996 | ||
4b13f64d RG |
2997 | /** |
2998 | * __memcg_kmem_uncharge: uncharge a number of kernel pages from a memcg | |
2999 | * @memcg: memcg to uncharge | |
3000 | * @nr_pages: number of pages to uncharge | |
3001 | */ | |
3002 | void __memcg_kmem_uncharge(struct mem_cgroup *memcg, unsigned int nr_pages) | |
3003 | { | |
3004 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
3005 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
3006 | ||
3007 | page_counter_uncharge(&memcg->memory, nr_pages); | |
3008 | if (do_memsw_account()) | |
3009 | page_counter_uncharge(&memcg->memsw, nr_pages); | |
3010 | } | |
3011 | ||
45264778 | 3012 | /** |
f4b00eab | 3013 | * __memcg_kmem_charge_page: charge a kmem page to the current memory cgroup |
45264778 VD |
3014 | * @page: page to charge |
3015 | * @gfp: reclaim mode | |
3016 | * @order: allocation order | |
3017 | * | |
3018 | * Returns 0 on success, an error code on failure. | |
3019 | */ | |
f4b00eab | 3020 | int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 3021 | { |
f3ccb2c4 | 3022 | struct mem_cgroup *memcg; |
fcff7d7e | 3023 | int ret = 0; |
7ae1e1d0 | 3024 | |
60cd4bcd | 3025 | if (memcg_kmem_bypass()) |
45264778 VD |
3026 | return 0; |
3027 | ||
d46eb14b | 3028 | memcg = get_mem_cgroup_from_current(); |
c4159a75 | 3029 | if (!mem_cgroup_is_root(memcg)) { |
4b13f64d | 3030 | ret = __memcg_kmem_charge(memcg, gfp, 1 << order); |
4d96ba35 RG |
3031 | if (!ret) { |
3032 | page->mem_cgroup = memcg; | |
c4159a75 | 3033 | __SetPageKmemcg(page); |
1a3e1f40 | 3034 | return 0; |
4d96ba35 | 3035 | } |
c4159a75 | 3036 | } |
7ae1e1d0 | 3037 | css_put(&memcg->css); |
d05e83a6 | 3038 | return ret; |
7ae1e1d0 | 3039 | } |
49a18eae | 3040 | |
45264778 | 3041 | /** |
f4b00eab | 3042 | * __memcg_kmem_uncharge_page: uncharge a kmem page |
45264778 VD |
3043 | * @page: page to uncharge |
3044 | * @order: allocation order | |
3045 | */ | |
f4b00eab | 3046 | void __memcg_kmem_uncharge_page(struct page *page, int order) |
7ae1e1d0 | 3047 | { |
1306a85a | 3048 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 3049 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 3050 | |
7ae1e1d0 GC |
3051 | if (!memcg) |
3052 | return; | |
3053 | ||
309381fe | 3054 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
4b13f64d | 3055 | __memcg_kmem_uncharge(memcg, nr_pages); |
1306a85a | 3056 | page->mem_cgroup = NULL; |
1a3e1f40 | 3057 | css_put(&memcg->css); |
c4159a75 VD |
3058 | |
3059 | /* slab pages do not have PageKmemcg flag set */ | |
3060 | if (PageKmemcg(page)) | |
3061 | __ClearPageKmemcg(page); | |
60d3fd32 | 3062 | } |
bf4f0599 RG |
3063 | |
3064 | static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) | |
3065 | { | |
3066 | struct memcg_stock_pcp *stock; | |
3067 | unsigned long flags; | |
3068 | bool ret = false; | |
3069 | ||
3070 | local_irq_save(flags); | |
3071 | ||
3072 | stock = this_cpu_ptr(&memcg_stock); | |
3073 | if (objcg == stock->cached_objcg && stock->nr_bytes >= nr_bytes) { | |
3074 | stock->nr_bytes -= nr_bytes; | |
3075 | ret = true; | |
3076 | } | |
3077 | ||
3078 | local_irq_restore(flags); | |
3079 | ||
3080 | return ret; | |
3081 | } | |
3082 | ||
3083 | static void drain_obj_stock(struct memcg_stock_pcp *stock) | |
3084 | { | |
3085 | struct obj_cgroup *old = stock->cached_objcg; | |
3086 | ||
3087 | if (!old) | |
3088 | return; | |
3089 | ||
3090 | if (stock->nr_bytes) { | |
3091 | unsigned int nr_pages = stock->nr_bytes >> PAGE_SHIFT; | |
3092 | unsigned int nr_bytes = stock->nr_bytes & (PAGE_SIZE - 1); | |
3093 | ||
3094 | if (nr_pages) { | |
3095 | rcu_read_lock(); | |
3096 | __memcg_kmem_uncharge(obj_cgroup_memcg(old), nr_pages); | |
3097 | rcu_read_unlock(); | |
3098 | } | |
3099 | ||
3100 | /* | |
3101 | * The leftover is flushed to the centralized per-memcg value. | |
3102 | * On the next attempt to refill obj stock it will be moved | |
3103 | * to a per-cpu stock (probably, on an other CPU), see | |
3104 | * refill_obj_stock(). | |
3105 | * | |
3106 | * How often it's flushed is a trade-off between the memory | |
3107 | * limit enforcement accuracy and potential CPU contention, | |
3108 | * so it might be changed in the future. | |
3109 | */ | |
3110 | atomic_add(nr_bytes, &old->nr_charged_bytes); | |
3111 | stock->nr_bytes = 0; | |
3112 | } | |
3113 | ||
3114 | obj_cgroup_put(old); | |
3115 | stock->cached_objcg = NULL; | |
3116 | } | |
3117 | ||
3118 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
3119 | struct mem_cgroup *root_memcg) | |
3120 | { | |
3121 | struct mem_cgroup *memcg; | |
3122 | ||
3123 | if (stock->cached_objcg) { | |
3124 | memcg = obj_cgroup_memcg(stock->cached_objcg); | |
3125 | if (memcg && mem_cgroup_is_descendant(memcg, root_memcg)) | |
3126 | return true; | |
3127 | } | |
3128 | ||
3129 | return false; | |
3130 | } | |
3131 | ||
3132 | static void refill_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) | |
3133 | { | |
3134 | struct memcg_stock_pcp *stock; | |
3135 | unsigned long flags; | |
3136 | ||
3137 | local_irq_save(flags); | |
3138 | ||
3139 | stock = this_cpu_ptr(&memcg_stock); | |
3140 | if (stock->cached_objcg != objcg) { /* reset if necessary */ | |
3141 | drain_obj_stock(stock); | |
3142 | obj_cgroup_get(objcg); | |
3143 | stock->cached_objcg = objcg; | |
3144 | stock->nr_bytes = atomic_xchg(&objcg->nr_charged_bytes, 0); | |
3145 | } | |
3146 | stock->nr_bytes += nr_bytes; | |
3147 | ||
3148 | if (stock->nr_bytes > PAGE_SIZE) | |
3149 | drain_obj_stock(stock); | |
3150 | ||
3151 | local_irq_restore(flags); | |
3152 | } | |
3153 | ||
3154 | int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size) | |
3155 | { | |
3156 | struct mem_cgroup *memcg; | |
3157 | unsigned int nr_pages, nr_bytes; | |
3158 | int ret; | |
3159 | ||
3160 | if (consume_obj_stock(objcg, size)) | |
3161 | return 0; | |
3162 | ||
3163 | /* | |
3164 | * In theory, memcg->nr_charged_bytes can have enough | |
3165 | * pre-charged bytes to satisfy the allocation. However, | |
3166 | * flushing memcg->nr_charged_bytes requires two atomic | |
3167 | * operations, and memcg->nr_charged_bytes can't be big, | |
3168 | * so it's better to ignore it and try grab some new pages. | |
3169 | * memcg->nr_charged_bytes will be flushed in | |
3170 | * refill_obj_stock(), called from this function or | |
3171 | * independently later. | |
3172 | */ | |
3173 | rcu_read_lock(); | |
3174 | memcg = obj_cgroup_memcg(objcg); | |
3175 | css_get(&memcg->css); | |
3176 | rcu_read_unlock(); | |
3177 | ||
3178 | nr_pages = size >> PAGE_SHIFT; | |
3179 | nr_bytes = size & (PAGE_SIZE - 1); | |
3180 | ||
3181 | if (nr_bytes) | |
3182 | nr_pages += 1; | |
3183 | ||
3184 | ret = __memcg_kmem_charge(memcg, gfp, nr_pages); | |
3185 | if (!ret && nr_bytes) | |
3186 | refill_obj_stock(objcg, PAGE_SIZE - nr_bytes); | |
3187 | ||
3188 | css_put(&memcg->css); | |
3189 | return ret; | |
3190 | } | |
3191 | ||
3192 | void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size) | |
3193 | { | |
3194 | refill_obj_stock(objcg, size); | |
3195 | } | |
3196 | ||
84c07d11 | 3197 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 3198 | |
ca3e0214 KH |
3199 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3200 | ||
ca3e0214 KH |
3201 | /* |
3202 | * Because tail pages are not marked as "used", set it. We're under | |
f4b7e272 | 3203 | * pgdat->lru_lock and migration entries setup in all page mappings. |
ca3e0214 | 3204 | */ |
e94c8a9c | 3205 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 3206 | { |
1a3e1f40 | 3207 | struct mem_cgroup *memcg = head->mem_cgroup; |
e94c8a9c | 3208 | int i; |
ca3e0214 | 3209 | |
3d37c4a9 KH |
3210 | if (mem_cgroup_disabled()) |
3211 | return; | |
b070e65c | 3212 | |
1a3e1f40 JW |
3213 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
3214 | css_get(&memcg->css); | |
3215 | head[i].mem_cgroup = memcg; | |
3216 | } | |
ca3e0214 | 3217 | } |
12d27107 | 3218 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3219 | |
c255a458 | 3220 | #ifdef CONFIG_MEMCG_SWAP |
02491447 DN |
3221 | /** |
3222 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3223 | * @entry: swap entry to be moved | |
3224 | * @from: mem_cgroup which the entry is moved from | |
3225 | * @to: mem_cgroup which the entry is moved to | |
3226 | * | |
3227 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3228 | * as the mem_cgroup's id of @from. | |
3229 | * | |
3230 | * Returns 0 on success, -EINVAL on failure. | |
3231 | * | |
3e32cb2e | 3232 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
3233 | * both res and memsw, and called css_get(). |
3234 | */ | |
3235 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3236 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3237 | { |
3238 | unsigned short old_id, new_id; | |
3239 | ||
34c00c31 LZ |
3240 | old_id = mem_cgroup_id(from); |
3241 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3242 | |
3243 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
3244 | mod_memcg_state(from, MEMCG_SWAP, -1); |
3245 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
3246 | return 0; |
3247 | } | |
3248 | return -EINVAL; | |
3249 | } | |
3250 | #else | |
3251 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3252 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3253 | { |
3254 | return -EINVAL; | |
3255 | } | |
8c7c6e34 | 3256 | #endif |
d13d1443 | 3257 | |
bbec2e15 | 3258 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 3259 | |
bbec2e15 RG |
3260 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
3261 | unsigned long max, bool memsw) | |
628f4235 | 3262 | { |
3e32cb2e | 3263 | bool enlarge = false; |
bb4a7ea2 | 3264 | bool drained = false; |
3e32cb2e | 3265 | int ret; |
c054a78c YZ |
3266 | bool limits_invariant; |
3267 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 3268 | |
3e32cb2e | 3269 | do { |
628f4235 KH |
3270 | if (signal_pending(current)) { |
3271 | ret = -EINTR; | |
3272 | break; | |
3273 | } | |
3e32cb2e | 3274 | |
bbec2e15 | 3275 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
3276 | /* |
3277 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 3278 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 3279 | */ |
15b42562 | 3280 | limits_invariant = memsw ? max >= READ_ONCE(memcg->memory.max) : |
bbec2e15 | 3281 | max <= memcg->memsw.max; |
c054a78c | 3282 | if (!limits_invariant) { |
bbec2e15 | 3283 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 3284 | ret = -EINVAL; |
8c7c6e34 KH |
3285 | break; |
3286 | } | |
bbec2e15 | 3287 | if (max > counter->max) |
3e32cb2e | 3288 | enlarge = true; |
bbec2e15 RG |
3289 | ret = page_counter_set_max(counter, max); |
3290 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
3291 | |
3292 | if (!ret) | |
3293 | break; | |
3294 | ||
bb4a7ea2 SB |
3295 | if (!drained) { |
3296 | drain_all_stock(memcg); | |
3297 | drained = true; | |
3298 | continue; | |
3299 | } | |
3300 | ||
1ab5c056 AR |
3301 | if (!try_to_free_mem_cgroup_pages(memcg, 1, |
3302 | GFP_KERNEL, !memsw)) { | |
3303 | ret = -EBUSY; | |
3304 | break; | |
3305 | } | |
3306 | } while (true); | |
3e32cb2e | 3307 | |
3c11ecf4 KH |
3308 | if (!ret && enlarge) |
3309 | memcg_oom_recover(memcg); | |
3e32cb2e | 3310 | |
628f4235 KH |
3311 | return ret; |
3312 | } | |
3313 | ||
ef8f2327 | 3314 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
3315 | gfp_t gfp_mask, |
3316 | unsigned long *total_scanned) | |
3317 | { | |
3318 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 3319 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
3320 | unsigned long reclaimed; |
3321 | int loop = 0; | |
ef8f2327 | 3322 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 3323 | unsigned long excess; |
0608f43d AM |
3324 | unsigned long nr_scanned; |
3325 | ||
3326 | if (order > 0) | |
3327 | return 0; | |
3328 | ||
ef8f2327 | 3329 | mctz = soft_limit_tree_node(pgdat->node_id); |
d6507ff5 MH |
3330 | |
3331 | /* | |
3332 | * Do not even bother to check the largest node if the root | |
3333 | * is empty. Do it lockless to prevent lock bouncing. Races | |
3334 | * are acceptable as soft limit is best effort anyway. | |
3335 | */ | |
bfc7228b | 3336 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
3337 | return 0; |
3338 | ||
0608f43d AM |
3339 | /* |
3340 | * This loop can run a while, specially if mem_cgroup's continuously | |
3341 | * keep exceeding their soft limit and putting the system under | |
3342 | * pressure | |
3343 | */ | |
3344 | do { | |
3345 | if (next_mz) | |
3346 | mz = next_mz; | |
3347 | else | |
3348 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3349 | if (!mz) | |
3350 | break; | |
3351 | ||
3352 | nr_scanned = 0; | |
ef8f2327 | 3353 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
0608f43d AM |
3354 | gfp_mask, &nr_scanned); |
3355 | nr_reclaimed += reclaimed; | |
3356 | *total_scanned += nr_scanned; | |
0a31bc97 | 3357 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 3358 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
3359 | |
3360 | /* | |
3361 | * If we failed to reclaim anything from this memory cgroup | |
3362 | * it is time to move on to the next cgroup | |
3363 | */ | |
3364 | next_mz = NULL; | |
bc2f2e7f VD |
3365 | if (!reclaimed) |
3366 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3367 | ||
3e32cb2e | 3368 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3369 | /* |
3370 | * One school of thought says that we should not add | |
3371 | * back the node to the tree if reclaim returns 0. | |
3372 | * But our reclaim could return 0, simply because due | |
3373 | * to priority we are exposing a smaller subset of | |
3374 | * memory to reclaim from. Consider this as a longer | |
3375 | * term TODO. | |
3376 | */ | |
3377 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3378 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3379 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3380 | css_put(&mz->memcg->css); |
3381 | loop++; | |
3382 | /* | |
3383 | * Could not reclaim anything and there are no more | |
3384 | * mem cgroups to try or we seem to be looping without | |
3385 | * reclaiming anything. | |
3386 | */ | |
3387 | if (!nr_reclaimed && | |
3388 | (next_mz == NULL || | |
3389 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3390 | break; | |
3391 | } while (!nr_reclaimed); | |
3392 | if (next_mz) | |
3393 | css_put(&next_mz->memcg->css); | |
3394 | return nr_reclaimed; | |
3395 | } | |
3396 | ||
ea280e7b TH |
3397 | /* |
3398 | * Test whether @memcg has children, dead or alive. Note that this | |
3399 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
3400 | * returns %true if there are child csses according to the cgroup | |
b8f2935f | 3401 | * hierarchy. Testing use_hierarchy is the caller's responsibility. |
ea280e7b | 3402 | */ |
b5f99b53 GC |
3403 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
3404 | { | |
ea280e7b TH |
3405 | bool ret; |
3406 | ||
ea280e7b TH |
3407 | rcu_read_lock(); |
3408 | ret = css_next_child(NULL, &memcg->css); | |
3409 | rcu_read_unlock(); | |
3410 | return ret; | |
b5f99b53 GC |
3411 | } |
3412 | ||
c26251f9 | 3413 | /* |
51038171 | 3414 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
3415 | * |
3416 | * Caller is responsible for holding css reference for memcg. | |
3417 | */ | |
3418 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3419 | { | |
d977aa93 | 3420 | int nr_retries = MAX_RECLAIM_RETRIES; |
c26251f9 | 3421 | |
c1e862c1 KH |
3422 | /* we call try-to-free pages for make this cgroup empty */ |
3423 | lru_add_drain_all(); | |
d12c60f6 JS |
3424 | |
3425 | drain_all_stock(memcg); | |
3426 | ||
f817ed48 | 3427 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3428 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 3429 | int progress; |
c1e862c1 | 3430 | |
c26251f9 MH |
3431 | if (signal_pending(current)) |
3432 | return -EINTR; | |
3433 | ||
b70a2a21 JW |
3434 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3435 | GFP_KERNEL, true); | |
c1e862c1 | 3436 | if (!progress) { |
f817ed48 | 3437 | nr_retries--; |
c1e862c1 | 3438 | /* maybe some writeback is necessary */ |
8aa7e847 | 3439 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3440 | } |
f817ed48 KH |
3441 | |
3442 | } | |
ab5196c2 MH |
3443 | |
3444 | return 0; | |
cc847582 KH |
3445 | } |
3446 | ||
6770c64e TH |
3447 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3448 | char *buf, size_t nbytes, | |
3449 | loff_t off) | |
c1e862c1 | 3450 | { |
6770c64e | 3451 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3452 | |
d8423011 MH |
3453 | if (mem_cgroup_is_root(memcg)) |
3454 | return -EINVAL; | |
6770c64e | 3455 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3456 | } |
3457 | ||
182446d0 TH |
3458 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3459 | struct cftype *cft) | |
18f59ea7 | 3460 | { |
182446d0 | 3461 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
3462 | } |
3463 | ||
182446d0 TH |
3464 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3465 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
3466 | { |
3467 | int retval = 0; | |
182446d0 | 3468 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 3469 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 3470 | |
567fb435 | 3471 | if (memcg->use_hierarchy == val) |
0b8f73e1 | 3472 | return 0; |
567fb435 | 3473 | |
18f59ea7 | 3474 | /* |
af901ca1 | 3475 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3476 | * in the child subtrees. If it is unset, then the change can |
3477 | * occur, provided the current cgroup has no children. | |
3478 | * | |
3479 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3480 | * set if there are no children. | |
3481 | */ | |
c0ff4b85 | 3482 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 3483 | (val == 1 || val == 0)) { |
ea280e7b | 3484 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 3485 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3486 | else |
3487 | retval = -EBUSY; | |
3488 | } else | |
3489 | retval = -EINVAL; | |
567fb435 | 3490 | |
18f59ea7 BS |
3491 | return retval; |
3492 | } | |
3493 | ||
6f646156 | 3494 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 3495 | { |
42a30035 | 3496 | unsigned long val; |
ce00a967 | 3497 | |
3e32cb2e | 3498 | if (mem_cgroup_is_root(memcg)) { |
0d1c2072 | 3499 | val = memcg_page_state(memcg, NR_FILE_PAGES) + |
be5d0a74 | 3500 | memcg_page_state(memcg, NR_ANON_MAPPED); |
42a30035 JW |
3501 | if (swap) |
3502 | val += memcg_page_state(memcg, MEMCG_SWAP); | |
3e32cb2e | 3503 | } else { |
ce00a967 | 3504 | if (!swap) |
3e32cb2e | 3505 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3506 | else |
3e32cb2e | 3507 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3508 | } |
c12176d3 | 3509 | return val; |
ce00a967 JW |
3510 | } |
3511 | ||
3e32cb2e JW |
3512 | enum { |
3513 | RES_USAGE, | |
3514 | RES_LIMIT, | |
3515 | RES_MAX_USAGE, | |
3516 | RES_FAILCNT, | |
3517 | RES_SOFT_LIMIT, | |
3518 | }; | |
ce00a967 | 3519 | |
791badbd | 3520 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3521 | struct cftype *cft) |
8cdea7c0 | 3522 | { |
182446d0 | 3523 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3524 | struct page_counter *counter; |
af36f906 | 3525 | |
3e32cb2e | 3526 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3527 | case _MEM: |
3e32cb2e JW |
3528 | counter = &memcg->memory; |
3529 | break; | |
8c7c6e34 | 3530 | case _MEMSWAP: |
3e32cb2e JW |
3531 | counter = &memcg->memsw; |
3532 | break; | |
510fc4e1 | 3533 | case _KMEM: |
3e32cb2e | 3534 | counter = &memcg->kmem; |
510fc4e1 | 3535 | break; |
d55f90bf | 3536 | case _TCP: |
0db15298 | 3537 | counter = &memcg->tcpmem; |
d55f90bf | 3538 | break; |
8c7c6e34 KH |
3539 | default: |
3540 | BUG(); | |
8c7c6e34 | 3541 | } |
3e32cb2e JW |
3542 | |
3543 | switch (MEMFILE_ATTR(cft->private)) { | |
3544 | case RES_USAGE: | |
3545 | if (counter == &memcg->memory) | |
c12176d3 | 3546 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 3547 | if (counter == &memcg->memsw) |
c12176d3 | 3548 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
3549 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
3550 | case RES_LIMIT: | |
bbec2e15 | 3551 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
3552 | case RES_MAX_USAGE: |
3553 | return (u64)counter->watermark * PAGE_SIZE; | |
3554 | case RES_FAILCNT: | |
3555 | return counter->failcnt; | |
3556 | case RES_SOFT_LIMIT: | |
3557 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3558 | default: | |
3559 | BUG(); | |
3560 | } | |
8cdea7c0 | 3561 | } |
510fc4e1 | 3562 | |
4a87e2a2 | 3563 | static void memcg_flush_percpu_vmstats(struct mem_cgroup *memcg) |
c350a99e | 3564 | { |
4a87e2a2 | 3565 | unsigned long stat[MEMCG_NR_STAT] = {0}; |
c350a99e RG |
3566 | struct mem_cgroup *mi; |
3567 | int node, cpu, i; | |
c350a99e RG |
3568 | |
3569 | for_each_online_cpu(cpu) | |
4a87e2a2 | 3570 | for (i = 0; i < MEMCG_NR_STAT; i++) |
6c1c2808 | 3571 | stat[i] += per_cpu(memcg->vmstats_percpu->stat[i], cpu); |
c350a99e RG |
3572 | |
3573 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
4a87e2a2 | 3574 | for (i = 0; i < MEMCG_NR_STAT; i++) |
c350a99e RG |
3575 | atomic_long_add(stat[i], &mi->vmstats[i]); |
3576 | ||
3577 | for_each_node(node) { | |
3578 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; | |
3579 | struct mem_cgroup_per_node *pi; | |
3580 | ||
4a87e2a2 | 3581 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
c350a99e RG |
3582 | stat[i] = 0; |
3583 | ||
3584 | for_each_online_cpu(cpu) | |
4a87e2a2 | 3585 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
6c1c2808 SB |
3586 | stat[i] += per_cpu( |
3587 | pn->lruvec_stat_cpu->count[i], cpu); | |
c350a99e RG |
3588 | |
3589 | for (pi = pn; pi; pi = parent_nodeinfo(pi, node)) | |
4a87e2a2 | 3590 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
c350a99e RG |
3591 | atomic_long_add(stat[i], &pi->lruvec_stat[i]); |
3592 | } | |
3593 | } | |
3594 | ||
bb65f89b RG |
3595 | static void memcg_flush_percpu_vmevents(struct mem_cgroup *memcg) |
3596 | { | |
3597 | unsigned long events[NR_VM_EVENT_ITEMS]; | |
3598 | struct mem_cgroup *mi; | |
3599 | int cpu, i; | |
3600 | ||
3601 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
3602 | events[i] = 0; | |
3603 | ||
3604 | for_each_online_cpu(cpu) | |
3605 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
6c1c2808 SB |
3606 | events[i] += per_cpu(memcg->vmstats_percpu->events[i], |
3607 | cpu); | |
bb65f89b RG |
3608 | |
3609 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
3610 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
3611 | atomic_long_add(events[i], &mi->vmevents[i]); | |
3612 | } | |
3613 | ||
84c07d11 | 3614 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 3615 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 3616 | { |
bf4f0599 | 3617 | struct obj_cgroup *objcg; |
d6441637 VD |
3618 | int memcg_id; |
3619 | ||
b313aeee VD |
3620 | if (cgroup_memory_nokmem) |
3621 | return 0; | |
3622 | ||
2a4db7eb | 3623 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 3624 | BUG_ON(memcg->kmem_state); |
d6441637 | 3625 | |
f3bb3043 | 3626 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
3627 | if (memcg_id < 0) |
3628 | return memcg_id; | |
d6441637 | 3629 | |
bf4f0599 RG |
3630 | objcg = obj_cgroup_alloc(); |
3631 | if (!objcg) { | |
3632 | memcg_free_cache_id(memcg_id); | |
3633 | return -ENOMEM; | |
3634 | } | |
3635 | objcg->memcg = memcg; | |
3636 | rcu_assign_pointer(memcg->objcg, objcg); | |
3637 | ||
d648bcc7 RG |
3638 | static_branch_enable(&memcg_kmem_enabled_key); |
3639 | ||
d6441637 | 3640 | /* |
567e9ab2 | 3641 | * A memory cgroup is considered kmem-online as soon as it gets |
900a38f0 | 3642 | * kmemcg_id. Setting the id after enabling static branching will |
d6441637 VD |
3643 | * guarantee no one starts accounting before all call sites are |
3644 | * patched. | |
3645 | */ | |
900a38f0 | 3646 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 3647 | memcg->kmem_state = KMEM_ONLINE; |
0b8f73e1 JW |
3648 | |
3649 | return 0; | |
d6441637 VD |
3650 | } |
3651 | ||
8e0a8912 JW |
3652 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
3653 | { | |
3654 | struct cgroup_subsys_state *css; | |
3655 | struct mem_cgroup *parent, *child; | |
3656 | int kmemcg_id; | |
3657 | ||
3658 | if (memcg->kmem_state != KMEM_ONLINE) | |
3659 | return; | |
9855609b | 3660 | |
8e0a8912 JW |
3661 | memcg->kmem_state = KMEM_ALLOCATED; |
3662 | ||
8e0a8912 JW |
3663 | parent = parent_mem_cgroup(memcg); |
3664 | if (!parent) | |
3665 | parent = root_mem_cgroup; | |
3666 | ||
bf4f0599 | 3667 | memcg_reparent_objcgs(memcg, parent); |
fb2f2b0a RG |
3668 | |
3669 | kmemcg_id = memcg->kmemcg_id; | |
3670 | BUG_ON(kmemcg_id < 0); | |
3671 | ||
8e0a8912 JW |
3672 | /* |
3673 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3674 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3675 | * to ones of the parent. After we have finished, all list_lrus | |
3676 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3677 | * ordering is imposed by list_lru_node->lock taken by | |
3678 | * memcg_drain_all_list_lrus(). | |
3679 | */ | |
3a06bb78 | 3680 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
8e0a8912 JW |
3681 | css_for_each_descendant_pre(css, &memcg->css) { |
3682 | child = mem_cgroup_from_css(css); | |
3683 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3684 | child->kmemcg_id = parent->kmemcg_id; | |
3685 | if (!memcg->use_hierarchy) | |
3686 | break; | |
3687 | } | |
3a06bb78 TH |
3688 | rcu_read_unlock(); |
3689 | ||
9bec5c35 | 3690 | memcg_drain_all_list_lrus(kmemcg_id, parent); |
8e0a8912 JW |
3691 | |
3692 | memcg_free_cache_id(kmemcg_id); | |
3693 | } | |
3694 | ||
3695 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3696 | { | |
0b8f73e1 JW |
3697 | /* css_alloc() failed, offlining didn't happen */ |
3698 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
3699 | memcg_offline_kmem(memcg); | |
8e0a8912 | 3700 | } |
d6441637 | 3701 | #else |
0b8f73e1 | 3702 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
3703 | { |
3704 | return 0; | |
3705 | } | |
3706 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
3707 | { | |
3708 | } | |
3709 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3710 | { | |
3711 | } | |
84c07d11 | 3712 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 3713 | |
bbec2e15 RG |
3714 | static int memcg_update_kmem_max(struct mem_cgroup *memcg, |
3715 | unsigned long max) | |
d6441637 | 3716 | { |
b313aeee | 3717 | int ret; |
127424c8 | 3718 | |
bbec2e15 RG |
3719 | mutex_lock(&memcg_max_mutex); |
3720 | ret = page_counter_set_max(&memcg->kmem, max); | |
3721 | mutex_unlock(&memcg_max_mutex); | |
127424c8 | 3722 | return ret; |
d6441637 | 3723 | } |
510fc4e1 | 3724 | |
bbec2e15 | 3725 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
3726 | { |
3727 | int ret; | |
3728 | ||
bbec2e15 | 3729 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 3730 | |
bbec2e15 | 3731 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
3732 | if (ret) |
3733 | goto out; | |
3734 | ||
0db15298 | 3735 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
3736 | /* |
3737 | * The active flag needs to be written after the static_key | |
3738 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
3739 | * function is the last one to run. See mem_cgroup_sk_alloc() |
3740 | * for details, and note that we don't mark any socket as | |
3741 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
3742 | * |
3743 | * We need to do this, because static_keys will span multiple | |
3744 | * sites, but we can't control their order. If we mark a socket | |
3745 | * as accounted, but the accounting functions are not patched in | |
3746 | * yet, we'll lose accounting. | |
3747 | * | |
2d758073 | 3748 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
3749 | * because when this value change, the code to process it is not |
3750 | * patched in yet. | |
3751 | */ | |
3752 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 3753 | memcg->tcpmem_active = true; |
d55f90bf VD |
3754 | } |
3755 | out: | |
bbec2e15 | 3756 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
3757 | return ret; |
3758 | } | |
d55f90bf | 3759 | |
628f4235 KH |
3760 | /* |
3761 | * The user of this function is... | |
3762 | * RES_LIMIT. | |
3763 | */ | |
451af504 TH |
3764 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3765 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3766 | { |
451af504 | 3767 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3768 | unsigned long nr_pages; |
628f4235 KH |
3769 | int ret; |
3770 | ||
451af504 | 3771 | buf = strstrip(buf); |
650c5e56 | 3772 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3773 | if (ret) |
3774 | return ret; | |
af36f906 | 3775 | |
3e32cb2e | 3776 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3777 | case RES_LIMIT: |
4b3bde4c BS |
3778 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3779 | ret = -EINVAL; | |
3780 | break; | |
3781 | } | |
3e32cb2e JW |
3782 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3783 | case _MEM: | |
bbec2e15 | 3784 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 3785 | break; |
3e32cb2e | 3786 | case _MEMSWAP: |
bbec2e15 | 3787 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 3788 | break; |
3e32cb2e | 3789 | case _KMEM: |
0158115f MH |
3790 | pr_warn_once("kmem.limit_in_bytes is deprecated and will be removed. " |
3791 | "Please report your usecase to linux-mm@kvack.org if you " | |
3792 | "depend on this functionality.\n"); | |
bbec2e15 | 3793 | ret = memcg_update_kmem_max(memcg, nr_pages); |
3e32cb2e | 3794 | break; |
d55f90bf | 3795 | case _TCP: |
bbec2e15 | 3796 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 3797 | break; |
3e32cb2e | 3798 | } |
296c81d8 | 3799 | break; |
3e32cb2e JW |
3800 | case RES_SOFT_LIMIT: |
3801 | memcg->soft_limit = nr_pages; | |
3802 | ret = 0; | |
628f4235 KH |
3803 | break; |
3804 | } | |
451af504 | 3805 | return ret ?: nbytes; |
8cdea7c0 BS |
3806 | } |
3807 | ||
6770c64e TH |
3808 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3809 | size_t nbytes, loff_t off) | |
c84872e1 | 3810 | { |
6770c64e | 3811 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3812 | struct page_counter *counter; |
c84872e1 | 3813 | |
3e32cb2e JW |
3814 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3815 | case _MEM: | |
3816 | counter = &memcg->memory; | |
3817 | break; | |
3818 | case _MEMSWAP: | |
3819 | counter = &memcg->memsw; | |
3820 | break; | |
3821 | case _KMEM: | |
3822 | counter = &memcg->kmem; | |
3823 | break; | |
d55f90bf | 3824 | case _TCP: |
0db15298 | 3825 | counter = &memcg->tcpmem; |
d55f90bf | 3826 | break; |
3e32cb2e JW |
3827 | default: |
3828 | BUG(); | |
3829 | } | |
af36f906 | 3830 | |
3e32cb2e | 3831 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3832 | case RES_MAX_USAGE: |
3e32cb2e | 3833 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3834 | break; |
3835 | case RES_FAILCNT: | |
3e32cb2e | 3836 | counter->failcnt = 0; |
29f2a4da | 3837 | break; |
3e32cb2e JW |
3838 | default: |
3839 | BUG(); | |
29f2a4da | 3840 | } |
f64c3f54 | 3841 | |
6770c64e | 3842 | return nbytes; |
c84872e1 PE |
3843 | } |
3844 | ||
182446d0 | 3845 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3846 | struct cftype *cft) |
3847 | { | |
182446d0 | 3848 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3849 | } |
3850 | ||
02491447 | 3851 | #ifdef CONFIG_MMU |
182446d0 | 3852 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3853 | struct cftype *cft, u64 val) |
3854 | { | |
182446d0 | 3855 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3856 | |
1dfab5ab | 3857 | if (val & ~MOVE_MASK) |
7dc74be0 | 3858 | return -EINVAL; |
ee5e8472 | 3859 | |
7dc74be0 | 3860 | /* |
ee5e8472 GC |
3861 | * No kind of locking is needed in here, because ->can_attach() will |
3862 | * check this value once in the beginning of the process, and then carry | |
3863 | * on with stale data. This means that changes to this value will only | |
3864 | * affect task migrations starting after the change. | |
7dc74be0 | 3865 | */ |
c0ff4b85 | 3866 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3867 | return 0; |
3868 | } | |
02491447 | 3869 | #else |
182446d0 | 3870 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3871 | struct cftype *cft, u64 val) |
3872 | { | |
3873 | return -ENOSYS; | |
3874 | } | |
3875 | #endif | |
7dc74be0 | 3876 | |
406eb0c9 | 3877 | #ifdef CONFIG_NUMA |
113b7dfd JW |
3878 | |
3879 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
3880 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
3881 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) | |
3882 | ||
3883 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 | 3884 | int nid, unsigned int lru_mask, bool tree) |
113b7dfd | 3885 | { |
867e5e1d | 3886 | struct lruvec *lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); |
113b7dfd JW |
3887 | unsigned long nr = 0; |
3888 | enum lru_list lru; | |
3889 | ||
3890 | VM_BUG_ON((unsigned)nid >= nr_node_ids); | |
3891 | ||
3892 | for_each_lru(lru) { | |
3893 | if (!(BIT(lru) & lru_mask)) | |
3894 | continue; | |
dd8657b6 SB |
3895 | if (tree) |
3896 | nr += lruvec_page_state(lruvec, NR_LRU_BASE + lru); | |
3897 | else | |
3898 | nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru); | |
113b7dfd JW |
3899 | } |
3900 | return nr; | |
3901 | } | |
3902 | ||
3903 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 SB |
3904 | unsigned int lru_mask, |
3905 | bool tree) | |
113b7dfd JW |
3906 | { |
3907 | unsigned long nr = 0; | |
3908 | enum lru_list lru; | |
3909 | ||
3910 | for_each_lru(lru) { | |
3911 | if (!(BIT(lru) & lru_mask)) | |
3912 | continue; | |
dd8657b6 SB |
3913 | if (tree) |
3914 | nr += memcg_page_state(memcg, NR_LRU_BASE + lru); | |
3915 | else | |
3916 | nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru); | |
113b7dfd JW |
3917 | } |
3918 | return nr; | |
3919 | } | |
3920 | ||
2da8ca82 | 3921 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3922 | { |
25485de6 GT |
3923 | struct numa_stat { |
3924 | const char *name; | |
3925 | unsigned int lru_mask; | |
3926 | }; | |
3927 | ||
3928 | static const struct numa_stat stats[] = { | |
3929 | { "total", LRU_ALL }, | |
3930 | { "file", LRU_ALL_FILE }, | |
3931 | { "anon", LRU_ALL_ANON }, | |
3932 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3933 | }; | |
3934 | const struct numa_stat *stat; | |
406eb0c9 | 3935 | int nid; |
aa9694bb | 3936 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
406eb0c9 | 3937 | |
25485de6 | 3938 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
3939 | seq_printf(m, "%s=%lu", stat->name, |
3940 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
3941 | false)); | |
3942 | for_each_node_state(nid, N_MEMORY) | |
3943 | seq_printf(m, " N%d=%lu", nid, | |
3944 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3945 | stat->lru_mask, false)); | |
25485de6 | 3946 | seq_putc(m, '\n'); |
406eb0c9 | 3947 | } |
406eb0c9 | 3948 | |
071aee13 | 3949 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
3950 | |
3951 | seq_printf(m, "hierarchical_%s=%lu", stat->name, | |
3952 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
3953 | true)); | |
3954 | for_each_node_state(nid, N_MEMORY) | |
3955 | seq_printf(m, " N%d=%lu", nid, | |
3956 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3957 | stat->lru_mask, true)); | |
071aee13 | 3958 | seq_putc(m, '\n'); |
406eb0c9 | 3959 | } |
406eb0c9 | 3960 | |
406eb0c9 YH |
3961 | return 0; |
3962 | } | |
3963 | #endif /* CONFIG_NUMA */ | |
3964 | ||
c8713d0b | 3965 | static const unsigned int memcg1_stats[] = { |
0d1c2072 | 3966 | NR_FILE_PAGES, |
be5d0a74 | 3967 | NR_ANON_MAPPED, |
468c3982 JW |
3968 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3969 | NR_ANON_THPS, | |
3970 | #endif | |
c8713d0b JW |
3971 | NR_SHMEM, |
3972 | NR_FILE_MAPPED, | |
3973 | NR_FILE_DIRTY, | |
3974 | NR_WRITEBACK, | |
3975 | MEMCG_SWAP, | |
3976 | }; | |
3977 | ||
3978 | static const char *const memcg1_stat_names[] = { | |
3979 | "cache", | |
3980 | "rss", | |
468c3982 | 3981 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
c8713d0b | 3982 | "rss_huge", |
468c3982 | 3983 | #endif |
c8713d0b JW |
3984 | "shmem", |
3985 | "mapped_file", | |
3986 | "dirty", | |
3987 | "writeback", | |
3988 | "swap", | |
3989 | }; | |
3990 | ||
df0e53d0 | 3991 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 3992 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
3993 | PGPGIN, |
3994 | PGPGOUT, | |
3995 | PGFAULT, | |
3996 | PGMAJFAULT, | |
3997 | }; | |
3998 | ||
2da8ca82 | 3999 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 4000 | { |
aa9694bb | 4001 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3e32cb2e | 4002 | unsigned long memory, memsw; |
af7c4b0e JW |
4003 | struct mem_cgroup *mi; |
4004 | unsigned int i; | |
406eb0c9 | 4005 | |
71cd3113 | 4006 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c | 4007 | |
71cd3113 | 4008 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
468c3982 JW |
4009 | unsigned long nr; |
4010 | ||
71cd3113 | 4011 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 4012 | continue; |
468c3982 JW |
4013 | nr = memcg_page_state_local(memcg, memcg1_stats[i]); |
4014 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
4015 | if (memcg1_stats[i] == NR_ANON_THPS) | |
4016 | nr *= HPAGE_PMD_NR; | |
4017 | #endif | |
4018 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], nr * PAGE_SIZE); | |
1dd3a273 | 4019 | } |
7b854121 | 4020 | |
df0e53d0 | 4021 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d | 4022 | seq_printf(m, "%s %lu\n", vm_event_name(memcg1_events[i]), |
205b20cc | 4023 | memcg_events_local(memcg, memcg1_events[i])); |
af7c4b0e JW |
4024 | |
4025 | for (i = 0; i < NR_LRU_LISTS; i++) | |
ebc5d83d | 4026 | seq_printf(m, "%s %lu\n", lru_list_name(i), |
205b20cc | 4027 | memcg_page_state_local(memcg, NR_LRU_BASE + i) * |
21d89d15 | 4028 | PAGE_SIZE); |
af7c4b0e | 4029 | |
14067bb3 | 4030 | /* Hierarchical information */ |
3e32cb2e JW |
4031 | memory = memsw = PAGE_COUNTER_MAX; |
4032 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
15b42562 CD |
4033 | memory = min(memory, READ_ONCE(mi->memory.max)); |
4034 | memsw = min(memsw, READ_ONCE(mi->memsw.max)); | |
fee7b548 | 4035 | } |
3e32cb2e JW |
4036 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
4037 | (u64)memory * PAGE_SIZE); | |
7941d214 | 4038 | if (do_memsw_account()) |
3e32cb2e JW |
4039 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
4040 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 4041 | |
8de7ecc6 | 4042 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
71cd3113 | 4043 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 4044 | continue; |
8de7ecc6 | 4045 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], |
dd923990 YS |
4046 | (u64)memcg_page_state(memcg, memcg1_stats[i]) * |
4047 | PAGE_SIZE); | |
af7c4b0e JW |
4048 | } |
4049 | ||
8de7ecc6 | 4050 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d KK |
4051 | seq_printf(m, "total_%s %llu\n", |
4052 | vm_event_name(memcg1_events[i]), | |
dd923990 | 4053 | (u64)memcg_events(memcg, memcg1_events[i])); |
af7c4b0e | 4054 | |
8de7ecc6 | 4055 | for (i = 0; i < NR_LRU_LISTS; i++) |
ebc5d83d | 4056 | seq_printf(m, "total_%s %llu\n", lru_list_name(i), |
42a30035 JW |
4057 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
4058 | PAGE_SIZE); | |
14067bb3 | 4059 | |
7f016ee8 | 4060 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 4061 | { |
ef8f2327 MG |
4062 | pg_data_t *pgdat; |
4063 | struct mem_cgroup_per_node *mz; | |
1431d4d1 JW |
4064 | unsigned long anon_cost = 0; |
4065 | unsigned long file_cost = 0; | |
7f016ee8 | 4066 | |
ef8f2327 MG |
4067 | for_each_online_pgdat(pgdat) { |
4068 | mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); | |
7f016ee8 | 4069 | |
1431d4d1 JW |
4070 | anon_cost += mz->lruvec.anon_cost; |
4071 | file_cost += mz->lruvec.file_cost; | |
ef8f2327 | 4072 | } |
1431d4d1 JW |
4073 | seq_printf(m, "anon_cost %lu\n", anon_cost); |
4074 | seq_printf(m, "file_cost %lu\n", file_cost); | |
7f016ee8 KM |
4075 | } |
4076 | #endif | |
4077 | ||
d2ceb9b7 KH |
4078 | return 0; |
4079 | } | |
4080 | ||
182446d0 TH |
4081 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
4082 | struct cftype *cft) | |
a7885eb8 | 4083 | { |
182446d0 | 4084 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4085 | |
1f4c025b | 4086 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4087 | } |
4088 | ||
182446d0 TH |
4089 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
4090 | struct cftype *cft, u64 val) | |
a7885eb8 | 4091 | { |
182446d0 | 4092 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4093 | |
3dae7fec | 4094 | if (val > 100) |
a7885eb8 KM |
4095 | return -EINVAL; |
4096 | ||
14208b0e | 4097 | if (css->parent) |
3dae7fec JW |
4098 | memcg->swappiness = val; |
4099 | else | |
4100 | vm_swappiness = val; | |
068b38c1 | 4101 | |
a7885eb8 KM |
4102 | return 0; |
4103 | } | |
4104 | ||
2e72b634 KS |
4105 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4106 | { | |
4107 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 4108 | unsigned long usage; |
2e72b634 KS |
4109 | int i; |
4110 | ||
4111 | rcu_read_lock(); | |
4112 | if (!swap) | |
2c488db2 | 4113 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4114 | else |
2c488db2 | 4115 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4116 | |
4117 | if (!t) | |
4118 | goto unlock; | |
4119 | ||
ce00a967 | 4120 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
4121 | |
4122 | /* | |
748dad36 | 4123 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4124 | * If it's not true, a threshold was crossed after last |
4125 | * call of __mem_cgroup_threshold(). | |
4126 | */ | |
5407a562 | 4127 | i = t->current_threshold; |
2e72b634 KS |
4128 | |
4129 | /* | |
4130 | * Iterate backward over array of thresholds starting from | |
4131 | * current_threshold and check if a threshold is crossed. | |
4132 | * If none of thresholds below usage is crossed, we read | |
4133 | * only one element of the array here. | |
4134 | */ | |
4135 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4136 | eventfd_signal(t->entries[i].eventfd, 1); | |
4137 | ||
4138 | /* i = current_threshold + 1 */ | |
4139 | i++; | |
4140 | ||
4141 | /* | |
4142 | * Iterate forward over array of thresholds starting from | |
4143 | * current_threshold+1 and check if a threshold is crossed. | |
4144 | * If none of thresholds above usage is crossed, we read | |
4145 | * only one element of the array here. | |
4146 | */ | |
4147 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4148 | eventfd_signal(t->entries[i].eventfd, 1); | |
4149 | ||
4150 | /* Update current_threshold */ | |
5407a562 | 4151 | t->current_threshold = i - 1; |
2e72b634 KS |
4152 | unlock: |
4153 | rcu_read_unlock(); | |
4154 | } | |
4155 | ||
4156 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4157 | { | |
ad4ca5f4 KS |
4158 | while (memcg) { |
4159 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 4160 | if (do_memsw_account()) |
ad4ca5f4 KS |
4161 | __mem_cgroup_threshold(memcg, true); |
4162 | ||
4163 | memcg = parent_mem_cgroup(memcg); | |
4164 | } | |
2e72b634 KS |
4165 | } |
4166 | ||
4167 | static int compare_thresholds(const void *a, const void *b) | |
4168 | { | |
4169 | const struct mem_cgroup_threshold *_a = a; | |
4170 | const struct mem_cgroup_threshold *_b = b; | |
4171 | ||
2bff24a3 GT |
4172 | if (_a->threshold > _b->threshold) |
4173 | return 1; | |
4174 | ||
4175 | if (_a->threshold < _b->threshold) | |
4176 | return -1; | |
4177 | ||
4178 | return 0; | |
2e72b634 KS |
4179 | } |
4180 | ||
c0ff4b85 | 4181 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4182 | { |
4183 | struct mem_cgroup_eventfd_list *ev; | |
4184 | ||
2bcf2e92 MH |
4185 | spin_lock(&memcg_oom_lock); |
4186 | ||
c0ff4b85 | 4187 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 4188 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
4189 | |
4190 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
4191 | return 0; |
4192 | } | |
4193 | ||
c0ff4b85 | 4194 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4195 | { |
7d74b06f KH |
4196 | struct mem_cgroup *iter; |
4197 | ||
c0ff4b85 | 4198 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4199 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4200 | } |
4201 | ||
59b6f873 | 4202 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4203 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4204 | { |
2c488db2 KS |
4205 | struct mem_cgroup_thresholds *thresholds; |
4206 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
4207 | unsigned long threshold; |
4208 | unsigned long usage; | |
2c488db2 | 4209 | int i, size, ret; |
2e72b634 | 4210 | |
650c5e56 | 4211 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
4212 | if (ret) |
4213 | return ret; | |
4214 | ||
4215 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4216 | |
05b84301 | 4217 | if (type == _MEM) { |
2c488db2 | 4218 | thresholds = &memcg->thresholds; |
ce00a967 | 4219 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4220 | } else if (type == _MEMSWAP) { |
2c488db2 | 4221 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4222 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4223 | } else |
2e72b634 KS |
4224 | BUG(); |
4225 | ||
2e72b634 | 4226 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4227 | if (thresholds->primary) |
2e72b634 KS |
4228 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4229 | ||
2c488db2 | 4230 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4231 | |
4232 | /* Allocate memory for new array of thresholds */ | |
67b8046f | 4233 | new = kmalloc(struct_size(new, entries, size), GFP_KERNEL); |
2c488db2 | 4234 | if (!new) { |
2e72b634 KS |
4235 | ret = -ENOMEM; |
4236 | goto unlock; | |
4237 | } | |
2c488db2 | 4238 | new->size = size; |
2e72b634 KS |
4239 | |
4240 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4241 | if (thresholds->primary) { |
4242 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4243 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4244 | } |
4245 | ||
2e72b634 | 4246 | /* Add new threshold */ |
2c488db2 KS |
4247 | new->entries[size - 1].eventfd = eventfd; |
4248 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4249 | |
4250 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4251 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4252 | compare_thresholds, NULL); |
4253 | ||
4254 | /* Find current threshold */ | |
2c488db2 | 4255 | new->current_threshold = -1; |
2e72b634 | 4256 | for (i = 0; i < size; i++) { |
748dad36 | 4257 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4258 | /* |
2c488db2 KS |
4259 | * new->current_threshold will not be used until |
4260 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4261 | * it here. |
4262 | */ | |
2c488db2 | 4263 | ++new->current_threshold; |
748dad36 SZ |
4264 | } else |
4265 | break; | |
2e72b634 KS |
4266 | } |
4267 | ||
2c488db2 KS |
4268 | /* Free old spare buffer and save old primary buffer as spare */ |
4269 | kfree(thresholds->spare); | |
4270 | thresholds->spare = thresholds->primary; | |
4271 | ||
4272 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4273 | |
907860ed | 4274 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4275 | synchronize_rcu(); |
4276 | ||
2e72b634 KS |
4277 | unlock: |
4278 | mutex_unlock(&memcg->thresholds_lock); | |
4279 | ||
4280 | return ret; | |
4281 | } | |
4282 | ||
59b6f873 | 4283 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4284 | struct eventfd_ctx *eventfd, const char *args) |
4285 | { | |
59b6f873 | 4286 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4287 | } |
4288 | ||
59b6f873 | 4289 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4290 | struct eventfd_ctx *eventfd, const char *args) |
4291 | { | |
59b6f873 | 4292 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4293 | } |
4294 | ||
59b6f873 | 4295 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4296 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4297 | { |
2c488db2 KS |
4298 | struct mem_cgroup_thresholds *thresholds; |
4299 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 4300 | unsigned long usage; |
7d36665a | 4301 | int i, j, size, entries; |
2e72b634 KS |
4302 | |
4303 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4304 | |
4305 | if (type == _MEM) { | |
2c488db2 | 4306 | thresholds = &memcg->thresholds; |
ce00a967 | 4307 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4308 | } else if (type == _MEMSWAP) { |
2c488db2 | 4309 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4310 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4311 | } else |
2e72b634 KS |
4312 | BUG(); |
4313 | ||
371528ca AV |
4314 | if (!thresholds->primary) |
4315 | goto unlock; | |
4316 | ||
2e72b634 KS |
4317 | /* Check if a threshold crossed before removing */ |
4318 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4319 | ||
4320 | /* Calculate new number of threshold */ | |
7d36665a | 4321 | size = entries = 0; |
2c488db2 KS |
4322 | for (i = 0; i < thresholds->primary->size; i++) { |
4323 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 | 4324 | size++; |
7d36665a CX |
4325 | else |
4326 | entries++; | |
2e72b634 KS |
4327 | } |
4328 | ||
2c488db2 | 4329 | new = thresholds->spare; |
907860ed | 4330 | |
7d36665a CX |
4331 | /* If no items related to eventfd have been cleared, nothing to do */ |
4332 | if (!entries) | |
4333 | goto unlock; | |
4334 | ||
2e72b634 KS |
4335 | /* Set thresholds array to NULL if we don't have thresholds */ |
4336 | if (!size) { | |
2c488db2 KS |
4337 | kfree(new); |
4338 | new = NULL; | |
907860ed | 4339 | goto swap_buffers; |
2e72b634 KS |
4340 | } |
4341 | ||
2c488db2 | 4342 | new->size = size; |
2e72b634 KS |
4343 | |
4344 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4345 | new->current_threshold = -1; |
4346 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4347 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4348 | continue; |
4349 | ||
2c488db2 | 4350 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4351 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4352 | /* |
2c488db2 | 4353 | * new->current_threshold will not be used |
2e72b634 KS |
4354 | * until rcu_assign_pointer(), so it's safe to increment |
4355 | * it here. | |
4356 | */ | |
2c488db2 | 4357 | ++new->current_threshold; |
2e72b634 KS |
4358 | } |
4359 | j++; | |
4360 | } | |
4361 | ||
907860ed | 4362 | swap_buffers: |
2c488db2 KS |
4363 | /* Swap primary and spare array */ |
4364 | thresholds->spare = thresholds->primary; | |
8c757763 | 4365 | |
2c488db2 | 4366 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4367 | |
907860ed | 4368 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4369 | synchronize_rcu(); |
6611d8d7 MC |
4370 | |
4371 | /* If all events are unregistered, free the spare array */ | |
4372 | if (!new) { | |
4373 | kfree(thresholds->spare); | |
4374 | thresholds->spare = NULL; | |
4375 | } | |
371528ca | 4376 | unlock: |
2e72b634 | 4377 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4378 | } |
c1e862c1 | 4379 | |
59b6f873 | 4380 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4381 | struct eventfd_ctx *eventfd) |
4382 | { | |
59b6f873 | 4383 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4384 | } |
4385 | ||
59b6f873 | 4386 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4387 | struct eventfd_ctx *eventfd) |
4388 | { | |
59b6f873 | 4389 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4390 | } |
4391 | ||
59b6f873 | 4392 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4393 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4394 | { |
9490ff27 | 4395 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4396 | |
9490ff27 KH |
4397 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4398 | if (!event) | |
4399 | return -ENOMEM; | |
4400 | ||
1af8efe9 | 4401 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4402 | |
4403 | event->eventfd = eventfd; | |
4404 | list_add(&event->list, &memcg->oom_notify); | |
4405 | ||
4406 | /* already in OOM ? */ | |
c2b42d3c | 4407 | if (memcg->under_oom) |
9490ff27 | 4408 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4409 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4410 | |
4411 | return 0; | |
4412 | } | |
4413 | ||
59b6f873 | 4414 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4415 | struct eventfd_ctx *eventfd) |
9490ff27 | 4416 | { |
9490ff27 | 4417 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4418 | |
1af8efe9 | 4419 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4420 | |
c0ff4b85 | 4421 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4422 | if (ev->eventfd == eventfd) { |
4423 | list_del(&ev->list); | |
4424 | kfree(ev); | |
4425 | } | |
4426 | } | |
4427 | ||
1af8efe9 | 4428 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4429 | } |
4430 | ||
2da8ca82 | 4431 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4432 | { |
aa9694bb | 4433 | struct mem_cgroup *memcg = mem_cgroup_from_seq(sf); |
3c11ecf4 | 4434 | |
791badbd | 4435 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 4436 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
4437 | seq_printf(sf, "oom_kill %lu\n", |
4438 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
4439 | return 0; |
4440 | } | |
4441 | ||
182446d0 | 4442 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4443 | struct cftype *cft, u64 val) |
4444 | { | |
182446d0 | 4445 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4446 | |
4447 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 4448 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4449 | return -EINVAL; |
4450 | ||
c0ff4b85 | 4451 | memcg->oom_kill_disable = val; |
4d845ebf | 4452 | if (!val) |
c0ff4b85 | 4453 | memcg_oom_recover(memcg); |
3dae7fec | 4454 | |
3c11ecf4 KH |
4455 | return 0; |
4456 | } | |
4457 | ||
52ebea74 TH |
4458 | #ifdef CONFIG_CGROUP_WRITEBACK |
4459 | ||
3a8e9ac8 TH |
4460 | #include <trace/events/writeback.h> |
4461 | ||
841710aa TH |
4462 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
4463 | { | |
4464 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
4465 | } | |
4466 | ||
4467 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4468 | { | |
4469 | wb_domain_exit(&memcg->cgwb_domain); | |
4470 | } | |
4471 | ||
2529bb3a TH |
4472 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4473 | { | |
4474 | wb_domain_size_changed(&memcg->cgwb_domain); | |
4475 | } | |
4476 | ||
841710aa TH |
4477 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
4478 | { | |
4479 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4480 | ||
4481 | if (!memcg->css.parent) | |
4482 | return NULL; | |
4483 | ||
4484 | return &memcg->cgwb_domain; | |
4485 | } | |
4486 | ||
0b3d6e6f GT |
4487 | /* |
4488 | * idx can be of type enum memcg_stat_item or node_stat_item. | |
4489 | * Keep in sync with memcg_exact_page(). | |
4490 | */ | |
4491 | static unsigned long memcg_exact_page_state(struct mem_cgroup *memcg, int idx) | |
4492 | { | |
871789d4 | 4493 | long x = atomic_long_read(&memcg->vmstats[idx]); |
0b3d6e6f GT |
4494 | int cpu; |
4495 | ||
4496 | for_each_online_cpu(cpu) | |
871789d4 | 4497 | x += per_cpu_ptr(memcg->vmstats_percpu, cpu)->stat[idx]; |
0b3d6e6f GT |
4498 | if (x < 0) |
4499 | x = 0; | |
4500 | return x; | |
4501 | } | |
4502 | ||
c2aa723a TH |
4503 | /** |
4504 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
4505 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
4506 | * @pfilepages: out parameter for number of file pages |
4507 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
4508 | * @pdirty: out parameter for number of dirty pages |
4509 | * @pwriteback: out parameter for number of pages under writeback | |
4510 | * | |
c5edf9cd TH |
4511 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
4512 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
4513 | * is a bit more involved. | |
c2aa723a | 4514 | * |
c5edf9cd TH |
4515 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
4516 | * headroom is calculated as the lowest headroom of itself and the | |
4517 | * ancestors. Note that this doesn't consider the actual amount of | |
4518 | * available memory in the system. The caller should further cap | |
4519 | * *@pheadroom accordingly. | |
c2aa723a | 4520 | */ |
c5edf9cd TH |
4521 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
4522 | unsigned long *pheadroom, unsigned long *pdirty, | |
4523 | unsigned long *pwriteback) | |
c2aa723a TH |
4524 | { |
4525 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4526 | struct mem_cgroup *parent; | |
c2aa723a | 4527 | |
0b3d6e6f | 4528 | *pdirty = memcg_exact_page_state(memcg, NR_FILE_DIRTY); |
c2aa723a | 4529 | |
0b3d6e6f | 4530 | *pwriteback = memcg_exact_page_state(memcg, NR_WRITEBACK); |
21d89d15 JW |
4531 | *pfilepages = memcg_exact_page_state(memcg, NR_INACTIVE_FILE) + |
4532 | memcg_exact_page_state(memcg, NR_ACTIVE_FILE); | |
c5edf9cd | 4533 | *pheadroom = PAGE_COUNTER_MAX; |
c2aa723a | 4534 | |
c2aa723a | 4535 | while ((parent = parent_mem_cgroup(memcg))) { |
15b42562 | 4536 | unsigned long ceiling = min(READ_ONCE(memcg->memory.max), |
d1663a90 | 4537 | READ_ONCE(memcg->memory.high)); |
c2aa723a TH |
4538 | unsigned long used = page_counter_read(&memcg->memory); |
4539 | ||
c5edf9cd | 4540 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
4541 | memcg = parent; |
4542 | } | |
c2aa723a TH |
4543 | } |
4544 | ||
97b27821 TH |
4545 | /* |
4546 | * Foreign dirty flushing | |
4547 | * | |
4548 | * There's an inherent mismatch between memcg and writeback. The former | |
4549 | * trackes ownership per-page while the latter per-inode. This was a | |
4550 | * deliberate design decision because honoring per-page ownership in the | |
4551 | * writeback path is complicated, may lead to higher CPU and IO overheads | |
4552 | * and deemed unnecessary given that write-sharing an inode across | |
4553 | * different cgroups isn't a common use-case. | |
4554 | * | |
4555 | * Combined with inode majority-writer ownership switching, this works well | |
4556 | * enough in most cases but there are some pathological cases. For | |
4557 | * example, let's say there are two cgroups A and B which keep writing to | |
4558 | * different but confined parts of the same inode. B owns the inode and | |
4559 | * A's memory is limited far below B's. A's dirty ratio can rise enough to | |
4560 | * trigger balance_dirty_pages() sleeps but B's can be low enough to avoid | |
4561 | * triggering background writeback. A will be slowed down without a way to | |
4562 | * make writeback of the dirty pages happen. | |
4563 | * | |
4564 | * Conditions like the above can lead to a cgroup getting repatedly and | |
4565 | * severely throttled after making some progress after each | |
4566 | * dirty_expire_interval while the underyling IO device is almost | |
4567 | * completely idle. | |
4568 | * | |
4569 | * Solving this problem completely requires matching the ownership tracking | |
4570 | * granularities between memcg and writeback in either direction. However, | |
4571 | * the more egregious behaviors can be avoided by simply remembering the | |
4572 | * most recent foreign dirtying events and initiating remote flushes on | |
4573 | * them when local writeback isn't enough to keep the memory clean enough. | |
4574 | * | |
4575 | * The following two functions implement such mechanism. When a foreign | |
4576 | * page - a page whose memcg and writeback ownerships don't match - is | |
4577 | * dirtied, mem_cgroup_track_foreign_dirty() records the inode owning | |
4578 | * bdi_writeback on the page owning memcg. When balance_dirty_pages() | |
4579 | * decides that the memcg needs to sleep due to high dirty ratio, it calls | |
4580 | * mem_cgroup_flush_foreign() which queues writeback on the recorded | |
4581 | * foreign bdi_writebacks which haven't expired. Both the numbers of | |
4582 | * recorded bdi_writebacks and concurrent in-flight foreign writebacks are | |
4583 | * limited to MEMCG_CGWB_FRN_CNT. | |
4584 | * | |
4585 | * The mechanism only remembers IDs and doesn't hold any object references. | |
4586 | * As being wrong occasionally doesn't matter, updates and accesses to the | |
4587 | * records are lockless and racy. | |
4588 | */ | |
4589 | void mem_cgroup_track_foreign_dirty_slowpath(struct page *page, | |
4590 | struct bdi_writeback *wb) | |
4591 | { | |
4592 | struct mem_cgroup *memcg = page->mem_cgroup; | |
4593 | struct memcg_cgwb_frn *frn; | |
4594 | u64 now = get_jiffies_64(); | |
4595 | u64 oldest_at = now; | |
4596 | int oldest = -1; | |
4597 | int i; | |
4598 | ||
3a8e9ac8 TH |
4599 | trace_track_foreign_dirty(page, wb); |
4600 | ||
97b27821 TH |
4601 | /* |
4602 | * Pick the slot to use. If there is already a slot for @wb, keep | |
4603 | * using it. If not replace the oldest one which isn't being | |
4604 | * written out. | |
4605 | */ | |
4606 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4607 | frn = &memcg->cgwb_frn[i]; | |
4608 | if (frn->bdi_id == wb->bdi->id && | |
4609 | frn->memcg_id == wb->memcg_css->id) | |
4610 | break; | |
4611 | if (time_before64(frn->at, oldest_at) && | |
4612 | atomic_read(&frn->done.cnt) == 1) { | |
4613 | oldest = i; | |
4614 | oldest_at = frn->at; | |
4615 | } | |
4616 | } | |
4617 | ||
4618 | if (i < MEMCG_CGWB_FRN_CNT) { | |
4619 | /* | |
4620 | * Re-using an existing one. Update timestamp lazily to | |
4621 | * avoid making the cacheline hot. We want them to be | |
4622 | * reasonably up-to-date and significantly shorter than | |
4623 | * dirty_expire_interval as that's what expires the record. | |
4624 | * Use the shorter of 1s and dirty_expire_interval / 8. | |
4625 | */ | |
4626 | unsigned long update_intv = | |
4627 | min_t(unsigned long, HZ, | |
4628 | msecs_to_jiffies(dirty_expire_interval * 10) / 8); | |
4629 | ||
4630 | if (time_before64(frn->at, now - update_intv)) | |
4631 | frn->at = now; | |
4632 | } else if (oldest >= 0) { | |
4633 | /* replace the oldest free one */ | |
4634 | frn = &memcg->cgwb_frn[oldest]; | |
4635 | frn->bdi_id = wb->bdi->id; | |
4636 | frn->memcg_id = wb->memcg_css->id; | |
4637 | frn->at = now; | |
4638 | } | |
4639 | } | |
4640 | ||
4641 | /* issue foreign writeback flushes for recorded foreign dirtying events */ | |
4642 | void mem_cgroup_flush_foreign(struct bdi_writeback *wb) | |
4643 | { | |
4644 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4645 | unsigned long intv = msecs_to_jiffies(dirty_expire_interval * 10); | |
4646 | u64 now = jiffies_64; | |
4647 | int i; | |
4648 | ||
4649 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4650 | struct memcg_cgwb_frn *frn = &memcg->cgwb_frn[i]; | |
4651 | ||
4652 | /* | |
4653 | * If the record is older than dirty_expire_interval, | |
4654 | * writeback on it has already started. No need to kick it | |
4655 | * off again. Also, don't start a new one if there's | |
4656 | * already one in flight. | |
4657 | */ | |
4658 | if (time_after64(frn->at, now - intv) && | |
4659 | atomic_read(&frn->done.cnt) == 1) { | |
4660 | frn->at = 0; | |
3a8e9ac8 | 4661 | trace_flush_foreign(wb, frn->bdi_id, frn->memcg_id); |
97b27821 TH |
4662 | cgroup_writeback_by_id(frn->bdi_id, frn->memcg_id, 0, |
4663 | WB_REASON_FOREIGN_FLUSH, | |
4664 | &frn->done); | |
4665 | } | |
4666 | } | |
4667 | } | |
4668 | ||
841710aa TH |
4669 | #else /* CONFIG_CGROUP_WRITEBACK */ |
4670 | ||
4671 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
4672 | { | |
4673 | return 0; | |
4674 | } | |
4675 | ||
4676 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4677 | { | |
4678 | } | |
4679 | ||
2529bb3a TH |
4680 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4681 | { | |
4682 | } | |
4683 | ||
52ebea74 TH |
4684 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
4685 | ||
3bc942f3 TH |
4686 | /* |
4687 | * DO NOT USE IN NEW FILES. | |
4688 | * | |
4689 | * "cgroup.event_control" implementation. | |
4690 | * | |
4691 | * This is way over-engineered. It tries to support fully configurable | |
4692 | * events for each user. Such level of flexibility is completely | |
4693 | * unnecessary especially in the light of the planned unified hierarchy. | |
4694 | * | |
4695 | * Please deprecate this and replace with something simpler if at all | |
4696 | * possible. | |
4697 | */ | |
4698 | ||
79bd9814 TH |
4699 | /* |
4700 | * Unregister event and free resources. | |
4701 | * | |
4702 | * Gets called from workqueue. | |
4703 | */ | |
3bc942f3 | 4704 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4705 | { |
3bc942f3 TH |
4706 | struct mem_cgroup_event *event = |
4707 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4708 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4709 | |
4710 | remove_wait_queue(event->wqh, &event->wait); | |
4711 | ||
59b6f873 | 4712 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4713 | |
4714 | /* Notify userspace the event is going away. */ | |
4715 | eventfd_signal(event->eventfd, 1); | |
4716 | ||
4717 | eventfd_ctx_put(event->eventfd); | |
4718 | kfree(event); | |
59b6f873 | 4719 | css_put(&memcg->css); |
79bd9814 TH |
4720 | } |
4721 | ||
4722 | /* | |
a9a08845 | 4723 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
4724 | * |
4725 | * Called with wqh->lock held and interrupts disabled. | |
4726 | */ | |
ac6424b9 | 4727 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 4728 | int sync, void *key) |
79bd9814 | 4729 | { |
3bc942f3 TH |
4730 | struct mem_cgroup_event *event = |
4731 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4732 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 4733 | __poll_t flags = key_to_poll(key); |
79bd9814 | 4734 | |
a9a08845 | 4735 | if (flags & EPOLLHUP) { |
79bd9814 TH |
4736 | /* |
4737 | * If the event has been detached at cgroup removal, we | |
4738 | * can simply return knowing the other side will cleanup | |
4739 | * for us. | |
4740 | * | |
4741 | * We can't race against event freeing since the other | |
4742 | * side will require wqh->lock via remove_wait_queue(), | |
4743 | * which we hold. | |
4744 | */ | |
fba94807 | 4745 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4746 | if (!list_empty(&event->list)) { |
4747 | list_del_init(&event->list); | |
4748 | /* | |
4749 | * We are in atomic context, but cgroup_event_remove() | |
4750 | * may sleep, so we have to call it in workqueue. | |
4751 | */ | |
4752 | schedule_work(&event->remove); | |
4753 | } | |
fba94807 | 4754 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4755 | } |
4756 | ||
4757 | return 0; | |
4758 | } | |
4759 | ||
3bc942f3 | 4760 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4761 | wait_queue_head_t *wqh, poll_table *pt) |
4762 | { | |
3bc942f3 TH |
4763 | struct mem_cgroup_event *event = |
4764 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4765 | |
4766 | event->wqh = wqh; | |
4767 | add_wait_queue(wqh, &event->wait); | |
4768 | } | |
4769 | ||
4770 | /* | |
3bc942f3 TH |
4771 | * DO NOT USE IN NEW FILES. |
4772 | * | |
79bd9814 TH |
4773 | * Parse input and register new cgroup event handler. |
4774 | * | |
4775 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4776 | * Interpretation of args is defined by control file implementation. | |
4777 | */ | |
451af504 TH |
4778 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4779 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4780 | { |
451af504 | 4781 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4782 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4783 | struct mem_cgroup_event *event; |
79bd9814 TH |
4784 | struct cgroup_subsys_state *cfile_css; |
4785 | unsigned int efd, cfd; | |
4786 | struct fd efile; | |
4787 | struct fd cfile; | |
fba94807 | 4788 | const char *name; |
79bd9814 TH |
4789 | char *endp; |
4790 | int ret; | |
4791 | ||
451af504 TH |
4792 | buf = strstrip(buf); |
4793 | ||
4794 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4795 | if (*endp != ' ') |
4796 | return -EINVAL; | |
451af504 | 4797 | buf = endp + 1; |
79bd9814 | 4798 | |
451af504 | 4799 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4800 | if ((*endp != ' ') && (*endp != '\0')) |
4801 | return -EINVAL; | |
451af504 | 4802 | buf = endp + 1; |
79bd9814 TH |
4803 | |
4804 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4805 | if (!event) | |
4806 | return -ENOMEM; | |
4807 | ||
59b6f873 | 4808 | event->memcg = memcg; |
79bd9814 | 4809 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4810 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4811 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4812 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4813 | |
4814 | efile = fdget(efd); | |
4815 | if (!efile.file) { | |
4816 | ret = -EBADF; | |
4817 | goto out_kfree; | |
4818 | } | |
4819 | ||
4820 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4821 | if (IS_ERR(event->eventfd)) { | |
4822 | ret = PTR_ERR(event->eventfd); | |
4823 | goto out_put_efile; | |
4824 | } | |
4825 | ||
4826 | cfile = fdget(cfd); | |
4827 | if (!cfile.file) { | |
4828 | ret = -EBADF; | |
4829 | goto out_put_eventfd; | |
4830 | } | |
4831 | ||
4832 | /* the process need read permission on control file */ | |
4833 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
4834 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
4835 | if (ret < 0) | |
4836 | goto out_put_cfile; | |
4837 | ||
fba94807 TH |
4838 | /* |
4839 | * Determine the event callbacks and set them in @event. This used | |
4840 | * to be done via struct cftype but cgroup core no longer knows | |
4841 | * about these events. The following is crude but the whole thing | |
4842 | * is for compatibility anyway. | |
3bc942f3 TH |
4843 | * |
4844 | * DO NOT ADD NEW FILES. | |
fba94807 | 4845 | */ |
b583043e | 4846 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4847 | |
4848 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4849 | event->register_event = mem_cgroup_usage_register_event; | |
4850 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4851 | } else if (!strcmp(name, "memory.oom_control")) { | |
4852 | event->register_event = mem_cgroup_oom_register_event; | |
4853 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4854 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4855 | event->register_event = vmpressure_register_event; | |
4856 | event->unregister_event = vmpressure_unregister_event; | |
4857 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4858 | event->register_event = memsw_cgroup_usage_register_event; |
4859 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4860 | } else { |
4861 | ret = -EINVAL; | |
4862 | goto out_put_cfile; | |
4863 | } | |
4864 | ||
79bd9814 | 4865 | /* |
b5557c4c TH |
4866 | * Verify @cfile should belong to @css. Also, remaining events are |
4867 | * automatically removed on cgroup destruction but the removal is | |
4868 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4869 | */ |
b583043e | 4870 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4871 | &memory_cgrp_subsys); |
79bd9814 | 4872 | ret = -EINVAL; |
5a17f543 | 4873 | if (IS_ERR(cfile_css)) |
79bd9814 | 4874 | goto out_put_cfile; |
5a17f543 TH |
4875 | if (cfile_css != css) { |
4876 | css_put(cfile_css); | |
79bd9814 | 4877 | goto out_put_cfile; |
5a17f543 | 4878 | } |
79bd9814 | 4879 | |
451af504 | 4880 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4881 | if (ret) |
4882 | goto out_put_css; | |
4883 | ||
9965ed17 | 4884 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 4885 | |
fba94807 TH |
4886 | spin_lock(&memcg->event_list_lock); |
4887 | list_add(&event->list, &memcg->event_list); | |
4888 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4889 | |
4890 | fdput(cfile); | |
4891 | fdput(efile); | |
4892 | ||
451af504 | 4893 | return nbytes; |
79bd9814 TH |
4894 | |
4895 | out_put_css: | |
b5557c4c | 4896 | css_put(css); |
79bd9814 TH |
4897 | out_put_cfile: |
4898 | fdput(cfile); | |
4899 | out_put_eventfd: | |
4900 | eventfd_ctx_put(event->eventfd); | |
4901 | out_put_efile: | |
4902 | fdput(efile); | |
4903 | out_kfree: | |
4904 | kfree(event); | |
4905 | ||
4906 | return ret; | |
4907 | } | |
4908 | ||
241994ed | 4909 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4910 | { |
0eea1030 | 4911 | .name = "usage_in_bytes", |
8c7c6e34 | 4912 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4913 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4914 | }, |
c84872e1 PE |
4915 | { |
4916 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4917 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4918 | .write = mem_cgroup_reset, |
791badbd | 4919 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4920 | }, |
8cdea7c0 | 4921 | { |
0eea1030 | 4922 | .name = "limit_in_bytes", |
8c7c6e34 | 4923 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4924 | .write = mem_cgroup_write, |
791badbd | 4925 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4926 | }, |
296c81d8 BS |
4927 | { |
4928 | .name = "soft_limit_in_bytes", | |
4929 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4930 | .write = mem_cgroup_write, |
791badbd | 4931 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4932 | }, |
8cdea7c0 BS |
4933 | { |
4934 | .name = "failcnt", | |
8c7c6e34 | 4935 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4936 | .write = mem_cgroup_reset, |
791badbd | 4937 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4938 | }, |
d2ceb9b7 KH |
4939 | { |
4940 | .name = "stat", | |
2da8ca82 | 4941 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4942 | }, |
c1e862c1 KH |
4943 | { |
4944 | .name = "force_empty", | |
6770c64e | 4945 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4946 | }, |
18f59ea7 BS |
4947 | { |
4948 | .name = "use_hierarchy", | |
4949 | .write_u64 = mem_cgroup_hierarchy_write, | |
4950 | .read_u64 = mem_cgroup_hierarchy_read, | |
4951 | }, | |
79bd9814 | 4952 | { |
3bc942f3 | 4953 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4954 | .write = memcg_write_event_control, |
7dbdb199 | 4955 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 4956 | }, |
a7885eb8 KM |
4957 | { |
4958 | .name = "swappiness", | |
4959 | .read_u64 = mem_cgroup_swappiness_read, | |
4960 | .write_u64 = mem_cgroup_swappiness_write, | |
4961 | }, | |
7dc74be0 DN |
4962 | { |
4963 | .name = "move_charge_at_immigrate", | |
4964 | .read_u64 = mem_cgroup_move_charge_read, | |
4965 | .write_u64 = mem_cgroup_move_charge_write, | |
4966 | }, | |
9490ff27 KH |
4967 | { |
4968 | .name = "oom_control", | |
2da8ca82 | 4969 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4970 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4971 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4972 | }, | |
70ddf637 AV |
4973 | { |
4974 | .name = "pressure_level", | |
70ddf637 | 4975 | }, |
406eb0c9 YH |
4976 | #ifdef CONFIG_NUMA |
4977 | { | |
4978 | .name = "numa_stat", | |
2da8ca82 | 4979 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4980 | }, |
4981 | #endif | |
510fc4e1 GC |
4982 | { |
4983 | .name = "kmem.limit_in_bytes", | |
4984 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4985 | .write = mem_cgroup_write, |
791badbd | 4986 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4987 | }, |
4988 | { | |
4989 | .name = "kmem.usage_in_bytes", | |
4990 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4991 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4992 | }, |
4993 | { | |
4994 | .name = "kmem.failcnt", | |
4995 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4996 | .write = mem_cgroup_reset, |
791badbd | 4997 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4998 | }, |
4999 | { | |
5000 | .name = "kmem.max_usage_in_bytes", | |
5001 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 5002 | .write = mem_cgroup_reset, |
791badbd | 5003 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 5004 | }, |
a87425a3 YS |
5005 | #if defined(CONFIG_MEMCG_KMEM) && \ |
5006 | (defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)) | |
749c5415 GC |
5007 | { |
5008 | .name = "kmem.slabinfo", | |
b047501c | 5009 | .seq_show = memcg_slab_show, |
749c5415 GC |
5010 | }, |
5011 | #endif | |
d55f90bf VD |
5012 | { |
5013 | .name = "kmem.tcp.limit_in_bytes", | |
5014 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
5015 | .write = mem_cgroup_write, | |
5016 | .read_u64 = mem_cgroup_read_u64, | |
5017 | }, | |
5018 | { | |
5019 | .name = "kmem.tcp.usage_in_bytes", | |
5020 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
5021 | .read_u64 = mem_cgroup_read_u64, | |
5022 | }, | |
5023 | { | |
5024 | .name = "kmem.tcp.failcnt", | |
5025 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
5026 | .write = mem_cgroup_reset, | |
5027 | .read_u64 = mem_cgroup_read_u64, | |
5028 | }, | |
5029 | { | |
5030 | .name = "kmem.tcp.max_usage_in_bytes", | |
5031 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
5032 | .write = mem_cgroup_reset, | |
5033 | .read_u64 = mem_cgroup_read_u64, | |
5034 | }, | |
6bc10349 | 5035 | { }, /* terminate */ |
af36f906 | 5036 | }; |
8c7c6e34 | 5037 | |
73f576c0 JW |
5038 | /* |
5039 | * Private memory cgroup IDR | |
5040 | * | |
5041 | * Swap-out records and page cache shadow entries need to store memcg | |
5042 | * references in constrained space, so we maintain an ID space that is | |
5043 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
5044 | * memory-controlled cgroups to 64k. | |
5045 | * | |
b8f2935f | 5046 | * However, there usually are many references to the offline CSS after |
73f576c0 JW |
5047 | * the cgroup has been destroyed, such as page cache or reclaimable |
5048 | * slab objects, that don't need to hang on to the ID. We want to keep | |
5049 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
5050 | * relatively small ID space and prevent the creation of new cgroups | |
5051 | * even when there are much fewer than 64k cgroups - possibly none. | |
5052 | * | |
5053 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
5054 | * be freed and recycled when it's no longer needed, which is usually | |
5055 | * when the CSS is offlined. | |
5056 | * | |
5057 | * The only exception to that are records of swapped out tmpfs/shmem | |
5058 | * pages that need to be attributed to live ancestors on swapin. But | |
5059 | * those references are manageable from userspace. | |
5060 | */ | |
5061 | ||
5062 | static DEFINE_IDR(mem_cgroup_idr); | |
5063 | ||
7e97de0b KT |
5064 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
5065 | { | |
5066 | if (memcg->id.id > 0) { | |
5067 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
5068 | memcg->id.id = 0; | |
5069 | } | |
5070 | } | |
5071 | ||
c1514c0a VF |
5072 | static void __maybe_unused mem_cgroup_id_get_many(struct mem_cgroup *memcg, |
5073 | unsigned int n) | |
73f576c0 | 5074 | { |
1c2d479a | 5075 | refcount_add(n, &memcg->id.ref); |
73f576c0 JW |
5076 | } |
5077 | ||
615d66c3 | 5078 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 5079 | { |
1c2d479a | 5080 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 5081 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
5082 | |
5083 | /* Memcg ID pins CSS */ | |
5084 | css_put(&memcg->css); | |
5085 | } | |
5086 | } | |
5087 | ||
615d66c3 VD |
5088 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) |
5089 | { | |
5090 | mem_cgroup_id_put_many(memcg, 1); | |
5091 | } | |
5092 | ||
73f576c0 JW |
5093 | /** |
5094 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
5095 | * @id: the memcg id to look up | |
5096 | * | |
5097 | * Caller must hold rcu_read_lock(). | |
5098 | */ | |
5099 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
5100 | { | |
5101 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5102 | return idr_find(&mem_cgroup_idr, id); | |
5103 | } | |
5104 | ||
ef8f2327 | 5105 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
5106 | { |
5107 | struct mem_cgroup_per_node *pn; | |
ef8f2327 | 5108 | int tmp = node; |
1ecaab2b KH |
5109 | /* |
5110 | * This routine is called against possible nodes. | |
5111 | * But it's BUG to call kmalloc() against offline node. | |
5112 | * | |
5113 | * TODO: this routine can waste much memory for nodes which will | |
5114 | * never be onlined. It's better to use memory hotplug callback | |
5115 | * function. | |
5116 | */ | |
41e3355d KH |
5117 | if (!node_state(node, N_NORMAL_MEMORY)) |
5118 | tmp = -1; | |
17295c88 | 5119 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
5120 | if (!pn) |
5121 | return 1; | |
1ecaab2b | 5122 | |
815744d7 JW |
5123 | pn->lruvec_stat_local = alloc_percpu(struct lruvec_stat); |
5124 | if (!pn->lruvec_stat_local) { | |
5125 | kfree(pn); | |
5126 | return 1; | |
5127 | } | |
5128 | ||
a983b5eb JW |
5129 | pn->lruvec_stat_cpu = alloc_percpu(struct lruvec_stat); |
5130 | if (!pn->lruvec_stat_cpu) { | |
815744d7 | 5131 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c JW |
5132 | kfree(pn); |
5133 | return 1; | |
5134 | } | |
5135 | ||
ef8f2327 MG |
5136 | lruvec_init(&pn->lruvec); |
5137 | pn->usage_in_excess = 0; | |
5138 | pn->on_tree = false; | |
5139 | pn->memcg = memcg; | |
5140 | ||
54f72fe0 | 5141 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
5142 | return 0; |
5143 | } | |
5144 | ||
ef8f2327 | 5145 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 5146 | { |
00f3ca2c JW |
5147 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
5148 | ||
4eaf431f MH |
5149 | if (!pn) |
5150 | return; | |
5151 | ||
a983b5eb | 5152 | free_percpu(pn->lruvec_stat_cpu); |
815744d7 | 5153 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c | 5154 | kfree(pn); |
1ecaab2b KH |
5155 | } |
5156 | ||
40e952f9 | 5157 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 5158 | { |
c8b2a36f | 5159 | int node; |
59927fb9 | 5160 | |
c8b2a36f | 5161 | for_each_node(node) |
ef8f2327 | 5162 | free_mem_cgroup_per_node_info(memcg, node); |
871789d4 | 5163 | free_percpu(memcg->vmstats_percpu); |
815744d7 | 5164 | free_percpu(memcg->vmstats_local); |
8ff69e2c | 5165 | kfree(memcg); |
59927fb9 | 5166 | } |
3afe36b1 | 5167 | |
40e952f9 TE |
5168 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
5169 | { | |
5170 | memcg_wb_domain_exit(memcg); | |
7961eee3 SB |
5171 | /* |
5172 | * Flush percpu vmstats and vmevents to guarantee the value correctness | |
5173 | * on parent's and all ancestor levels. | |
5174 | */ | |
4a87e2a2 | 5175 | memcg_flush_percpu_vmstats(memcg); |
7961eee3 | 5176 | memcg_flush_percpu_vmevents(memcg); |
40e952f9 TE |
5177 | __mem_cgroup_free(memcg); |
5178 | } | |
5179 | ||
0b8f73e1 | 5180 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 5181 | { |
d142e3e6 | 5182 | struct mem_cgroup *memcg; |
b9726c26 | 5183 | unsigned int size; |
6d12e2d8 | 5184 | int node; |
97b27821 | 5185 | int __maybe_unused i; |
11d67612 | 5186 | long error = -ENOMEM; |
8cdea7c0 | 5187 | |
0b8f73e1 JW |
5188 | size = sizeof(struct mem_cgroup); |
5189 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
5190 | ||
5191 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 5192 | if (!memcg) |
11d67612 | 5193 | return ERR_PTR(error); |
0b8f73e1 | 5194 | |
73f576c0 JW |
5195 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
5196 | 1, MEM_CGROUP_ID_MAX, | |
5197 | GFP_KERNEL); | |
11d67612 YS |
5198 | if (memcg->id.id < 0) { |
5199 | error = memcg->id.id; | |
73f576c0 | 5200 | goto fail; |
11d67612 | 5201 | } |
73f576c0 | 5202 | |
815744d7 JW |
5203 | memcg->vmstats_local = alloc_percpu(struct memcg_vmstats_percpu); |
5204 | if (!memcg->vmstats_local) | |
5205 | goto fail; | |
5206 | ||
871789d4 CD |
5207 | memcg->vmstats_percpu = alloc_percpu(struct memcg_vmstats_percpu); |
5208 | if (!memcg->vmstats_percpu) | |
0b8f73e1 | 5209 | goto fail; |
78fb7466 | 5210 | |
3ed28fa1 | 5211 | for_each_node(node) |
ef8f2327 | 5212 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 5213 | goto fail; |
f64c3f54 | 5214 | |
0b8f73e1 JW |
5215 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
5216 | goto fail; | |
28dbc4b6 | 5217 | |
f7e1cb6e | 5218 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 | 5219 | INIT_LIST_HEAD(&memcg->oom_notify); |
d142e3e6 GC |
5220 | mutex_init(&memcg->thresholds_lock); |
5221 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 5222 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
5223 | INIT_LIST_HEAD(&memcg->event_list); |
5224 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 5225 | memcg->socket_pressure = jiffies; |
84c07d11 | 5226 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 5227 | memcg->kmemcg_id = -1; |
bf4f0599 | 5228 | INIT_LIST_HEAD(&memcg->objcg_list); |
900a38f0 | 5229 | #endif |
52ebea74 TH |
5230 | #ifdef CONFIG_CGROUP_WRITEBACK |
5231 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
97b27821 TH |
5232 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) |
5233 | memcg->cgwb_frn[i].done = | |
5234 | __WB_COMPLETION_INIT(&memcg_cgwb_frn_waitq); | |
87eaceb3 YS |
5235 | #endif |
5236 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5237 | spin_lock_init(&memcg->deferred_split_queue.split_queue_lock); | |
5238 | INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue); | |
5239 | memcg->deferred_split_queue.split_queue_len = 0; | |
52ebea74 | 5240 | #endif |
73f576c0 | 5241 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
5242 | return memcg; |
5243 | fail: | |
7e97de0b | 5244 | mem_cgroup_id_remove(memcg); |
40e952f9 | 5245 | __mem_cgroup_free(memcg); |
11d67612 | 5246 | return ERR_PTR(error); |
d142e3e6 GC |
5247 | } |
5248 | ||
0b8f73e1 JW |
5249 | static struct cgroup_subsys_state * __ref |
5250 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 5251 | { |
0b8f73e1 JW |
5252 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
5253 | struct mem_cgroup *memcg; | |
5254 | long error = -ENOMEM; | |
d142e3e6 | 5255 | |
0b8f73e1 | 5256 | memcg = mem_cgroup_alloc(); |
11d67612 YS |
5257 | if (IS_ERR(memcg)) |
5258 | return ERR_CAST(memcg); | |
d142e3e6 | 5259 | |
d1663a90 | 5260 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
0b8f73e1 | 5261 | memcg->soft_limit = PAGE_COUNTER_MAX; |
4b82ab4f | 5262 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
0b8f73e1 JW |
5263 | if (parent) { |
5264 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
5265 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
5266 | } | |
5267 | if (parent && parent->use_hierarchy) { | |
5268 | memcg->use_hierarchy = true; | |
3e32cb2e | 5269 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 5270 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e JW |
5271 | page_counter_init(&memcg->memsw, &parent->memsw); |
5272 | page_counter_init(&memcg->kmem, &parent->kmem); | |
0db15298 | 5273 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 5274 | } else { |
3e32cb2e | 5275 | page_counter_init(&memcg->memory, NULL); |
37e84351 | 5276 | page_counter_init(&memcg->swap, NULL); |
3e32cb2e JW |
5277 | page_counter_init(&memcg->memsw, NULL); |
5278 | page_counter_init(&memcg->kmem, NULL); | |
0db15298 | 5279 | page_counter_init(&memcg->tcpmem, NULL); |
8c7f6edb TH |
5280 | /* |
5281 | * Deeper hierachy with use_hierarchy == false doesn't make | |
5282 | * much sense so let cgroup subsystem know about this | |
5283 | * unfortunate state in our controller. | |
5284 | */ | |
d142e3e6 | 5285 | if (parent != root_mem_cgroup) |
073219e9 | 5286 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 5287 | } |
d6441637 | 5288 | |
0b8f73e1 JW |
5289 | /* The following stuff does not apply to the root */ |
5290 | if (!parent) { | |
5291 | root_mem_cgroup = memcg; | |
5292 | return &memcg->css; | |
5293 | } | |
5294 | ||
b313aeee | 5295 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
5296 | if (error) |
5297 | goto fail; | |
127424c8 | 5298 | |
f7e1cb6e | 5299 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5300 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 5301 | |
0b8f73e1 JW |
5302 | return &memcg->css; |
5303 | fail: | |
7e97de0b | 5304 | mem_cgroup_id_remove(memcg); |
0b8f73e1 | 5305 | mem_cgroup_free(memcg); |
11d67612 | 5306 | return ERR_PTR(error); |
0b8f73e1 JW |
5307 | } |
5308 | ||
73f576c0 | 5309 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 5310 | { |
58fa2a55 VD |
5311 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5312 | ||
0a4465d3 KT |
5313 | /* |
5314 | * A memcg must be visible for memcg_expand_shrinker_maps() | |
5315 | * by the time the maps are allocated. So, we allocate maps | |
5316 | * here, when for_each_mem_cgroup() can't skip it. | |
5317 | */ | |
5318 | if (memcg_alloc_shrinker_maps(memcg)) { | |
5319 | mem_cgroup_id_remove(memcg); | |
5320 | return -ENOMEM; | |
5321 | } | |
5322 | ||
73f576c0 | 5323 | /* Online state pins memcg ID, memcg ID pins CSS */ |
1c2d479a | 5324 | refcount_set(&memcg->id.ref, 1); |
73f576c0 | 5325 | css_get(css); |
2f7dd7a4 | 5326 | return 0; |
8cdea7c0 BS |
5327 | } |
5328 | ||
eb95419b | 5329 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 5330 | { |
eb95419b | 5331 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5332 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
5333 | |
5334 | /* | |
5335 | * Unregister events and notify userspace. | |
5336 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
5337 | * directory to avoid race between userspace and kernelspace. | |
5338 | */ | |
fba94807 TH |
5339 | spin_lock(&memcg->event_list_lock); |
5340 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
5341 | list_del_init(&event->list); |
5342 | schedule_work(&event->remove); | |
5343 | } | |
fba94807 | 5344 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 5345 | |
bf8d5d52 | 5346 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5347 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 5348 | |
567e9ab2 | 5349 | memcg_offline_kmem(memcg); |
52ebea74 | 5350 | wb_memcg_offline(memcg); |
73f576c0 | 5351 | |
591edfb1 RG |
5352 | drain_all_stock(memcg); |
5353 | ||
73f576c0 | 5354 | mem_cgroup_id_put(memcg); |
df878fb0 KH |
5355 | } |
5356 | ||
6df38689 VD |
5357 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
5358 | { | |
5359 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5360 | ||
5361 | invalidate_reclaim_iterators(memcg); | |
5362 | } | |
5363 | ||
eb95419b | 5364 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 5365 | { |
eb95419b | 5366 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
97b27821 | 5367 | int __maybe_unused i; |
c268e994 | 5368 | |
97b27821 TH |
5369 | #ifdef CONFIG_CGROUP_WRITEBACK |
5370 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) | |
5371 | wb_wait_for_completion(&memcg->cgwb_frn[i].done); | |
5372 | #endif | |
f7e1cb6e | 5373 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5374 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 5375 | |
0db15298 | 5376 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 5377 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 5378 | |
0b8f73e1 JW |
5379 | vmpressure_cleanup(&memcg->vmpressure); |
5380 | cancel_work_sync(&memcg->high_work); | |
5381 | mem_cgroup_remove_from_trees(memcg); | |
0a4465d3 | 5382 | memcg_free_shrinker_maps(memcg); |
d886f4e4 | 5383 | memcg_free_kmem(memcg); |
0b8f73e1 | 5384 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
5385 | } |
5386 | ||
1ced953b TH |
5387 | /** |
5388 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
5389 | * @css: the target css | |
5390 | * | |
5391 | * Reset the states of the mem_cgroup associated with @css. This is | |
5392 | * invoked when the userland requests disabling on the default hierarchy | |
5393 | * but the memcg is pinned through dependency. The memcg should stop | |
5394 | * applying policies and should revert to the vanilla state as it may be | |
5395 | * made visible again. | |
5396 | * | |
5397 | * The current implementation only resets the essential configurations. | |
5398 | * This needs to be expanded to cover all the visible parts. | |
5399 | */ | |
5400 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
5401 | { | |
5402 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5403 | ||
bbec2e15 RG |
5404 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
5405 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
5406 | page_counter_set_max(&memcg->memsw, PAGE_COUNTER_MAX); | |
5407 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); | |
5408 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 5409 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5410 | page_counter_set_low(&memcg->memory, 0); |
d1663a90 | 5411 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
24d404dc | 5412 | memcg->soft_limit = PAGE_COUNTER_MAX; |
4b82ab4f | 5413 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
2529bb3a | 5414 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
5415 | } |
5416 | ||
02491447 | 5417 | #ifdef CONFIG_MMU |
7dc74be0 | 5418 | /* Handlers for move charge at task migration. */ |
854ffa8d | 5419 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 5420 | { |
05b84301 | 5421 | int ret; |
9476db97 | 5422 | |
d0164adc MG |
5423 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
5424 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 5425 | if (!ret) { |
854ffa8d | 5426 | mc.precharge += count; |
854ffa8d DN |
5427 | return ret; |
5428 | } | |
9476db97 | 5429 | |
3674534b | 5430 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 5431 | while (count--) { |
3674534b | 5432 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 5433 | if (ret) |
38c5d72f | 5434 | return ret; |
854ffa8d | 5435 | mc.precharge++; |
9476db97 | 5436 | cond_resched(); |
854ffa8d | 5437 | } |
9476db97 | 5438 | return 0; |
4ffef5fe DN |
5439 | } |
5440 | ||
4ffef5fe DN |
5441 | union mc_target { |
5442 | struct page *page; | |
02491447 | 5443 | swp_entry_t ent; |
4ffef5fe DN |
5444 | }; |
5445 | ||
4ffef5fe | 5446 | enum mc_target_type { |
8d32ff84 | 5447 | MC_TARGET_NONE = 0, |
4ffef5fe | 5448 | MC_TARGET_PAGE, |
02491447 | 5449 | MC_TARGET_SWAP, |
c733a828 | 5450 | MC_TARGET_DEVICE, |
4ffef5fe DN |
5451 | }; |
5452 | ||
90254a65 DN |
5453 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5454 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5455 | { |
25b2995a | 5456 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5457 | |
90254a65 DN |
5458 | if (!page || !page_mapped(page)) |
5459 | return NULL; | |
5460 | if (PageAnon(page)) { | |
1dfab5ab | 5461 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5462 | return NULL; |
1dfab5ab JW |
5463 | } else { |
5464 | if (!(mc.flags & MOVE_FILE)) | |
5465 | return NULL; | |
5466 | } | |
90254a65 DN |
5467 | if (!get_page_unless_zero(page)) |
5468 | return NULL; | |
5469 | ||
5470 | return page; | |
5471 | } | |
5472 | ||
c733a828 | 5473 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 5474 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 5475 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 5476 | { |
90254a65 DN |
5477 | struct page *page = NULL; |
5478 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5479 | ||
1dfab5ab | 5480 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 5481 | return NULL; |
c733a828 JG |
5482 | |
5483 | /* | |
5484 | * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to | |
5485 | * a device and because they are not accessible by CPU they are store | |
5486 | * as special swap entry in the CPU page table. | |
5487 | */ | |
5488 | if (is_device_private_entry(ent)) { | |
5489 | page = device_private_entry_to_page(ent); | |
5490 | /* | |
5491 | * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have | |
5492 | * a refcount of 1 when free (unlike normal page) | |
5493 | */ | |
5494 | if (!page_ref_add_unless(page, 1, 1)) | |
5495 | return NULL; | |
5496 | return page; | |
5497 | } | |
5498 | ||
4b91355e KH |
5499 | /* |
5500 | * Because lookup_swap_cache() updates some statistics counter, | |
5501 | * we call find_get_page() with swapper_space directly. | |
5502 | */ | |
f6ab1f7f | 5503 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
2d1c4980 | 5504 | entry->val = ent.val; |
90254a65 DN |
5505 | |
5506 | return page; | |
5507 | } | |
4b91355e KH |
5508 | #else |
5509 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 5510 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
5511 | { |
5512 | return NULL; | |
5513 | } | |
5514 | #endif | |
90254a65 | 5515 | |
87946a72 DN |
5516 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5517 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5518 | { | |
5519 | struct page *page = NULL; | |
87946a72 DN |
5520 | struct address_space *mapping; |
5521 | pgoff_t pgoff; | |
5522 | ||
5523 | if (!vma->vm_file) /* anonymous vma */ | |
5524 | return NULL; | |
1dfab5ab | 5525 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
5526 | return NULL; |
5527 | ||
87946a72 | 5528 | mapping = vma->vm_file->f_mapping; |
0661a336 | 5529 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
5530 | |
5531 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
5532 | #ifdef CONFIG_SWAP |
5533 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
5534 | if (shmem_mapping(mapping)) { |
5535 | page = find_get_entry(mapping, pgoff); | |
3159f943 | 5536 | if (xa_is_value(page)) { |
139b6a6f | 5537 | swp_entry_t swp = radix_to_swp_entry(page); |
2d1c4980 | 5538 | *entry = swp; |
f6ab1f7f HY |
5539 | page = find_get_page(swap_address_space(swp), |
5540 | swp_offset(swp)); | |
139b6a6f JW |
5541 | } |
5542 | } else | |
5543 | page = find_get_page(mapping, pgoff); | |
5544 | #else | |
5545 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 5546 | #endif |
87946a72 DN |
5547 | return page; |
5548 | } | |
5549 | ||
b1b0deab CG |
5550 | /** |
5551 | * mem_cgroup_move_account - move account of the page | |
5552 | * @page: the page | |
25843c2b | 5553 | * @compound: charge the page as compound or small page |
b1b0deab CG |
5554 | * @from: mem_cgroup which the page is moved from. |
5555 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
5556 | * | |
3ac808fd | 5557 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
5558 | * |
5559 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
5560 | * from old cgroup. | |
5561 | */ | |
5562 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 5563 | bool compound, |
b1b0deab CG |
5564 | struct mem_cgroup *from, |
5565 | struct mem_cgroup *to) | |
5566 | { | |
ae8af438 KK |
5567 | struct lruvec *from_vec, *to_vec; |
5568 | struct pglist_data *pgdat; | |
f627c2f5 | 5569 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
b1b0deab CG |
5570 | int ret; |
5571 | ||
5572 | VM_BUG_ON(from == to); | |
5573 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 5574 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
5575 | |
5576 | /* | |
6a93ca8f | 5577 | * Prevent mem_cgroup_migrate() from looking at |
45637bab | 5578 | * page->mem_cgroup of its source page while we change it. |
b1b0deab | 5579 | */ |
f627c2f5 | 5580 | ret = -EBUSY; |
b1b0deab CG |
5581 | if (!trylock_page(page)) |
5582 | goto out; | |
5583 | ||
5584 | ret = -EINVAL; | |
5585 | if (page->mem_cgroup != from) | |
5586 | goto out_unlock; | |
5587 | ||
ae8af438 | 5588 | pgdat = page_pgdat(page); |
867e5e1d JW |
5589 | from_vec = mem_cgroup_lruvec(from, pgdat); |
5590 | to_vec = mem_cgroup_lruvec(to, pgdat); | |
ae8af438 | 5591 | |
abb242f5 | 5592 | lock_page_memcg(page); |
b1b0deab | 5593 | |
be5d0a74 JW |
5594 | if (PageAnon(page)) { |
5595 | if (page_mapped(page)) { | |
5596 | __mod_lruvec_state(from_vec, NR_ANON_MAPPED, -nr_pages); | |
5597 | __mod_lruvec_state(to_vec, NR_ANON_MAPPED, nr_pages); | |
468c3982 JW |
5598 | if (PageTransHuge(page)) { |
5599 | __mod_lruvec_state(from_vec, NR_ANON_THPS, | |
5600 | -nr_pages); | |
5601 | __mod_lruvec_state(to_vec, NR_ANON_THPS, | |
5602 | nr_pages); | |
5603 | } | |
5604 | ||
be5d0a74 JW |
5605 | } |
5606 | } else { | |
0d1c2072 JW |
5607 | __mod_lruvec_state(from_vec, NR_FILE_PAGES, -nr_pages); |
5608 | __mod_lruvec_state(to_vec, NR_FILE_PAGES, nr_pages); | |
5609 | ||
5610 | if (PageSwapBacked(page)) { | |
5611 | __mod_lruvec_state(from_vec, NR_SHMEM, -nr_pages); | |
5612 | __mod_lruvec_state(to_vec, NR_SHMEM, nr_pages); | |
5613 | } | |
5614 | ||
49e50d27 JW |
5615 | if (page_mapped(page)) { |
5616 | __mod_lruvec_state(from_vec, NR_FILE_MAPPED, -nr_pages); | |
5617 | __mod_lruvec_state(to_vec, NR_FILE_MAPPED, nr_pages); | |
5618 | } | |
b1b0deab | 5619 | |
49e50d27 JW |
5620 | if (PageDirty(page)) { |
5621 | struct address_space *mapping = page_mapping(page); | |
c4843a75 | 5622 | |
49e50d27 JW |
5623 | if (mapping_cap_account_dirty(mapping)) { |
5624 | __mod_lruvec_state(from_vec, NR_FILE_DIRTY, | |
5625 | -nr_pages); | |
5626 | __mod_lruvec_state(to_vec, NR_FILE_DIRTY, | |
5627 | nr_pages); | |
5628 | } | |
c4843a75 GT |
5629 | } |
5630 | } | |
5631 | ||
b1b0deab | 5632 | if (PageWriteback(page)) { |
ae8af438 KK |
5633 | __mod_lruvec_state(from_vec, NR_WRITEBACK, -nr_pages); |
5634 | __mod_lruvec_state(to_vec, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
5635 | } |
5636 | ||
5637 | /* | |
abb242f5 JW |
5638 | * All state has been migrated, let's switch to the new memcg. |
5639 | * | |
b1b0deab | 5640 | * It is safe to change page->mem_cgroup here because the page |
abb242f5 JW |
5641 | * is referenced, charged, isolated, and locked: we can't race |
5642 | * with (un)charging, migration, LRU putback, or anything else | |
5643 | * that would rely on a stable page->mem_cgroup. | |
5644 | * | |
5645 | * Note that lock_page_memcg is a memcg lock, not a page lock, | |
5646 | * to save space. As soon as we switch page->mem_cgroup to a | |
5647 | * new memcg that isn't locked, the above state can change | |
5648 | * concurrently again. Make sure we're truly done with it. | |
b1b0deab | 5649 | */ |
abb242f5 | 5650 | smp_mb(); |
b1b0deab | 5651 | |
1a3e1f40 JW |
5652 | css_get(&to->css); |
5653 | css_put(&from->css); | |
5654 | ||
5655 | page->mem_cgroup = to; | |
87eaceb3 | 5656 | |
abb242f5 | 5657 | __unlock_page_memcg(from); |
b1b0deab CG |
5658 | |
5659 | ret = 0; | |
5660 | ||
5661 | local_irq_disable(); | |
3fba69a5 | 5662 | mem_cgroup_charge_statistics(to, page, nr_pages); |
b1b0deab | 5663 | memcg_check_events(to, page); |
3fba69a5 | 5664 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
b1b0deab CG |
5665 | memcg_check_events(from, page); |
5666 | local_irq_enable(); | |
5667 | out_unlock: | |
5668 | unlock_page(page); | |
5669 | out: | |
5670 | return ret; | |
5671 | } | |
5672 | ||
7cf7806c LR |
5673 | /** |
5674 | * get_mctgt_type - get target type of moving charge | |
5675 | * @vma: the vma the pte to be checked belongs | |
5676 | * @addr: the address corresponding to the pte to be checked | |
5677 | * @ptent: the pte to be checked | |
5678 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
5679 | * | |
5680 | * Returns | |
5681 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5682 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5683 | * move charge. if @target is not NULL, the page is stored in target->page | |
5684 | * with extra refcnt got(Callers should handle it). | |
5685 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
5686 | * target for charge migration. if @target is not NULL, the entry is stored | |
5687 | * in target->ent. | |
25b2995a CH |
5688 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PRIVATE |
5689 | * (so ZONE_DEVICE page and thus not on the lru). | |
df6ad698 JG |
5690 | * For now we such page is charge like a regular page would be as for all |
5691 | * intent and purposes it is just special memory taking the place of a | |
5692 | * regular page. | |
c733a828 JG |
5693 | * |
5694 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
5695 | * |
5696 | * Called with pte lock held. | |
5697 | */ | |
5698 | ||
8d32ff84 | 5699 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5700 | unsigned long addr, pte_t ptent, union mc_target *target) |
5701 | { | |
5702 | struct page *page = NULL; | |
8d32ff84 | 5703 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5704 | swp_entry_t ent = { .val = 0 }; |
5705 | ||
5706 | if (pte_present(ptent)) | |
5707 | page = mc_handle_present_pte(vma, addr, ptent); | |
5708 | else if (is_swap_pte(ptent)) | |
48406ef8 | 5709 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 5710 | else if (pte_none(ptent)) |
87946a72 | 5711 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
5712 | |
5713 | if (!page && !ent.val) | |
8d32ff84 | 5714 | return ret; |
02491447 | 5715 | if (page) { |
02491447 | 5716 | /* |
0a31bc97 | 5717 | * Do only loose check w/o serialization. |
1306a85a | 5718 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 5719 | * not under LRU exclusion. |
02491447 | 5720 | */ |
1306a85a | 5721 | if (page->mem_cgroup == mc.from) { |
02491447 | 5722 | ret = MC_TARGET_PAGE; |
25b2995a | 5723 | if (is_device_private_page(page)) |
c733a828 | 5724 | ret = MC_TARGET_DEVICE; |
02491447 DN |
5725 | if (target) |
5726 | target->page = page; | |
5727 | } | |
5728 | if (!ret || !target) | |
5729 | put_page(page); | |
5730 | } | |
3e14a57b HY |
5731 | /* |
5732 | * There is a swap entry and a page doesn't exist or isn't charged. | |
5733 | * But we cannot move a tail-page in a THP. | |
5734 | */ | |
5735 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 5736 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5737 | ret = MC_TARGET_SWAP; |
5738 | if (target) | |
5739 | target->ent = ent; | |
4ffef5fe | 5740 | } |
4ffef5fe DN |
5741 | return ret; |
5742 | } | |
5743 | ||
12724850 NH |
5744 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5745 | /* | |
d6810d73 HY |
5746 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
5747 | * not support them for now. | |
12724850 NH |
5748 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
5749 | */ | |
5750 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5751 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5752 | { | |
5753 | struct page *page = NULL; | |
12724850 NH |
5754 | enum mc_target_type ret = MC_TARGET_NONE; |
5755 | ||
84c3fc4e ZY |
5756 | if (unlikely(is_swap_pmd(pmd))) { |
5757 | VM_BUG_ON(thp_migration_supported() && | |
5758 | !is_pmd_migration_entry(pmd)); | |
5759 | return ret; | |
5760 | } | |
12724850 | 5761 | page = pmd_page(pmd); |
309381fe | 5762 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 5763 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 5764 | return ret; |
1306a85a | 5765 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
5766 | ret = MC_TARGET_PAGE; |
5767 | if (target) { | |
5768 | get_page(page); | |
5769 | target->page = page; | |
5770 | } | |
5771 | } | |
5772 | return ret; | |
5773 | } | |
5774 | #else | |
5775 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5776 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5777 | { | |
5778 | return MC_TARGET_NONE; | |
5779 | } | |
5780 | #endif | |
5781 | ||
4ffef5fe DN |
5782 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5783 | unsigned long addr, unsigned long end, | |
5784 | struct mm_walk *walk) | |
5785 | { | |
26bcd64a | 5786 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5787 | pte_t *pte; |
5788 | spinlock_t *ptl; | |
5789 | ||
b6ec57f4 KS |
5790 | ptl = pmd_trans_huge_lock(pmd, vma); |
5791 | if (ptl) { | |
c733a828 JG |
5792 | /* |
5793 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
25b2995a CH |
5794 | * support transparent huge page with MEMORY_DEVICE_PRIVATE but |
5795 | * this might change. | |
c733a828 | 5796 | */ |
12724850 NH |
5797 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5798 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5799 | spin_unlock(ptl); |
1a5a9906 | 5800 | return 0; |
12724850 | 5801 | } |
03319327 | 5802 | |
45f83cef AA |
5803 | if (pmd_trans_unstable(pmd)) |
5804 | return 0; | |
4ffef5fe DN |
5805 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5806 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5807 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5808 | mc.precharge++; /* increment precharge temporarily */ |
5809 | pte_unmap_unlock(pte - 1, ptl); | |
5810 | cond_resched(); | |
5811 | ||
7dc74be0 DN |
5812 | return 0; |
5813 | } | |
5814 | ||
7b86ac33 CH |
5815 | static const struct mm_walk_ops precharge_walk_ops = { |
5816 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5817 | }; | |
5818 | ||
4ffef5fe DN |
5819 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5820 | { | |
5821 | unsigned long precharge; | |
4ffef5fe | 5822 | |
d8ed45c5 | 5823 | mmap_read_lock(mm); |
7b86ac33 | 5824 | walk_page_range(mm, 0, mm->highest_vm_end, &precharge_walk_ops, NULL); |
d8ed45c5 | 5825 | mmap_read_unlock(mm); |
4ffef5fe DN |
5826 | |
5827 | precharge = mc.precharge; | |
5828 | mc.precharge = 0; | |
5829 | ||
5830 | return precharge; | |
5831 | } | |
5832 | ||
4ffef5fe DN |
5833 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5834 | { | |
dfe076b0 DN |
5835 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5836 | ||
5837 | VM_BUG_ON(mc.moving_task); | |
5838 | mc.moving_task = current; | |
5839 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5840 | } |
5841 | ||
dfe076b0 DN |
5842 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5843 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5844 | { |
2bd9bb20 KH |
5845 | struct mem_cgroup *from = mc.from; |
5846 | struct mem_cgroup *to = mc.to; | |
5847 | ||
4ffef5fe | 5848 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5849 | if (mc.precharge) { |
00501b53 | 5850 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5851 | mc.precharge = 0; |
5852 | } | |
5853 | /* | |
5854 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5855 | * we must uncharge here. | |
5856 | */ | |
5857 | if (mc.moved_charge) { | |
00501b53 | 5858 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5859 | mc.moved_charge = 0; |
4ffef5fe | 5860 | } |
483c30b5 DN |
5861 | /* we must fixup refcnts and charges */ |
5862 | if (mc.moved_swap) { | |
483c30b5 | 5863 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5864 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5865 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5866 | |
615d66c3 VD |
5867 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
5868 | ||
05b84301 | 5869 | /* |
3e32cb2e JW |
5870 | * we charged both to->memory and to->memsw, so we |
5871 | * should uncharge to->memory. | |
05b84301 | 5872 | */ |
ce00a967 | 5873 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5874 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5875 | ||
483c30b5 DN |
5876 | mc.moved_swap = 0; |
5877 | } | |
dfe076b0 DN |
5878 | memcg_oom_recover(from); |
5879 | memcg_oom_recover(to); | |
5880 | wake_up_all(&mc.waitq); | |
5881 | } | |
5882 | ||
5883 | static void mem_cgroup_clear_mc(void) | |
5884 | { | |
264a0ae1 TH |
5885 | struct mm_struct *mm = mc.mm; |
5886 | ||
dfe076b0 DN |
5887 | /* |
5888 | * we must clear moving_task before waking up waiters at the end of | |
5889 | * task migration. | |
5890 | */ | |
5891 | mc.moving_task = NULL; | |
5892 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5893 | spin_lock(&mc.lock); |
4ffef5fe DN |
5894 | mc.from = NULL; |
5895 | mc.to = NULL; | |
264a0ae1 | 5896 | mc.mm = NULL; |
2bd9bb20 | 5897 | spin_unlock(&mc.lock); |
264a0ae1 TH |
5898 | |
5899 | mmput(mm); | |
4ffef5fe DN |
5900 | } |
5901 | ||
1f7dd3e5 | 5902 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5903 | { |
1f7dd3e5 | 5904 | struct cgroup_subsys_state *css; |
eed67d75 | 5905 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 5906 | struct mem_cgroup *from; |
4530eddb | 5907 | struct task_struct *leader, *p; |
9f2115f9 | 5908 | struct mm_struct *mm; |
1dfab5ab | 5909 | unsigned long move_flags; |
9f2115f9 | 5910 | int ret = 0; |
7dc74be0 | 5911 | |
1f7dd3e5 TH |
5912 | /* charge immigration isn't supported on the default hierarchy */ |
5913 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
5914 | return 0; |
5915 | ||
4530eddb TH |
5916 | /* |
5917 | * Multi-process migrations only happen on the default hierarchy | |
5918 | * where charge immigration is not used. Perform charge | |
5919 | * immigration if @tset contains a leader and whine if there are | |
5920 | * multiple. | |
5921 | */ | |
5922 | p = NULL; | |
1f7dd3e5 | 5923 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
5924 | WARN_ON_ONCE(p); |
5925 | p = leader; | |
1f7dd3e5 | 5926 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
5927 | } |
5928 | if (!p) | |
5929 | return 0; | |
5930 | ||
1f7dd3e5 TH |
5931 | /* |
5932 | * We are now commited to this value whatever it is. Changes in this | |
5933 | * tunable will only affect upcoming migrations, not the current one. | |
5934 | * So we need to save it, and keep it going. | |
5935 | */ | |
5936 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
5937 | if (!move_flags) | |
5938 | return 0; | |
5939 | ||
9f2115f9 TH |
5940 | from = mem_cgroup_from_task(p); |
5941 | ||
5942 | VM_BUG_ON(from == memcg); | |
5943 | ||
5944 | mm = get_task_mm(p); | |
5945 | if (!mm) | |
5946 | return 0; | |
5947 | /* We move charges only when we move a owner of the mm */ | |
5948 | if (mm->owner == p) { | |
5949 | VM_BUG_ON(mc.from); | |
5950 | VM_BUG_ON(mc.to); | |
5951 | VM_BUG_ON(mc.precharge); | |
5952 | VM_BUG_ON(mc.moved_charge); | |
5953 | VM_BUG_ON(mc.moved_swap); | |
5954 | ||
5955 | spin_lock(&mc.lock); | |
264a0ae1 | 5956 | mc.mm = mm; |
9f2115f9 TH |
5957 | mc.from = from; |
5958 | mc.to = memcg; | |
5959 | mc.flags = move_flags; | |
5960 | spin_unlock(&mc.lock); | |
5961 | /* We set mc.moving_task later */ | |
5962 | ||
5963 | ret = mem_cgroup_precharge_mc(mm); | |
5964 | if (ret) | |
5965 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
5966 | } else { |
5967 | mmput(mm); | |
7dc74be0 DN |
5968 | } |
5969 | return ret; | |
5970 | } | |
5971 | ||
1f7dd3e5 | 5972 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5973 | { |
4e2f245d JW |
5974 | if (mc.to) |
5975 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5976 | } |
5977 | ||
4ffef5fe DN |
5978 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5979 | unsigned long addr, unsigned long end, | |
5980 | struct mm_walk *walk) | |
7dc74be0 | 5981 | { |
4ffef5fe | 5982 | int ret = 0; |
26bcd64a | 5983 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5984 | pte_t *pte; |
5985 | spinlock_t *ptl; | |
12724850 NH |
5986 | enum mc_target_type target_type; |
5987 | union mc_target target; | |
5988 | struct page *page; | |
4ffef5fe | 5989 | |
b6ec57f4 KS |
5990 | ptl = pmd_trans_huge_lock(pmd, vma); |
5991 | if (ptl) { | |
62ade86a | 5992 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5993 | spin_unlock(ptl); |
12724850 NH |
5994 | return 0; |
5995 | } | |
5996 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5997 | if (target_type == MC_TARGET_PAGE) { | |
5998 | page = target.page; | |
5999 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 6000 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 6001 | mc.from, mc.to)) { |
12724850 NH |
6002 | mc.precharge -= HPAGE_PMD_NR; |
6003 | mc.moved_charge += HPAGE_PMD_NR; | |
6004 | } | |
6005 | putback_lru_page(page); | |
6006 | } | |
6007 | put_page(page); | |
c733a828 JG |
6008 | } else if (target_type == MC_TARGET_DEVICE) { |
6009 | page = target.page; | |
6010 | if (!mem_cgroup_move_account(page, true, | |
6011 | mc.from, mc.to)) { | |
6012 | mc.precharge -= HPAGE_PMD_NR; | |
6013 | mc.moved_charge += HPAGE_PMD_NR; | |
6014 | } | |
6015 | put_page(page); | |
12724850 | 6016 | } |
bf929152 | 6017 | spin_unlock(ptl); |
1a5a9906 | 6018 | return 0; |
12724850 NH |
6019 | } |
6020 | ||
45f83cef AA |
6021 | if (pmd_trans_unstable(pmd)) |
6022 | return 0; | |
4ffef5fe DN |
6023 | retry: |
6024 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
6025 | for (; addr != end; addr += PAGE_SIZE) { | |
6026 | pte_t ptent = *(pte++); | |
c733a828 | 6027 | bool device = false; |
02491447 | 6028 | swp_entry_t ent; |
4ffef5fe DN |
6029 | |
6030 | if (!mc.precharge) | |
6031 | break; | |
6032 | ||
8d32ff84 | 6033 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
6034 | case MC_TARGET_DEVICE: |
6035 | device = true; | |
e4a9bc58 | 6036 | fallthrough; |
4ffef5fe DN |
6037 | case MC_TARGET_PAGE: |
6038 | page = target.page; | |
53f9263b KS |
6039 | /* |
6040 | * We can have a part of the split pmd here. Moving it | |
6041 | * can be done but it would be too convoluted so simply | |
6042 | * ignore such a partial THP and keep it in original | |
6043 | * memcg. There should be somebody mapping the head. | |
6044 | */ | |
6045 | if (PageTransCompound(page)) | |
6046 | goto put; | |
c733a828 | 6047 | if (!device && isolate_lru_page(page)) |
4ffef5fe | 6048 | goto put; |
f627c2f5 KS |
6049 | if (!mem_cgroup_move_account(page, false, |
6050 | mc.from, mc.to)) { | |
4ffef5fe | 6051 | mc.precharge--; |
854ffa8d DN |
6052 | /* we uncharge from mc.from later. */ |
6053 | mc.moved_charge++; | |
4ffef5fe | 6054 | } |
c733a828 JG |
6055 | if (!device) |
6056 | putback_lru_page(page); | |
8d32ff84 | 6057 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
6058 | put_page(page); |
6059 | break; | |
02491447 DN |
6060 | case MC_TARGET_SWAP: |
6061 | ent = target.ent; | |
e91cbb42 | 6062 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6063 | mc.precharge--; |
8d22a935 HD |
6064 | mem_cgroup_id_get_many(mc.to, 1); |
6065 | /* we fixup other refcnts and charges later. */ | |
483c30b5 DN |
6066 | mc.moved_swap++; |
6067 | } | |
02491447 | 6068 | break; |
4ffef5fe DN |
6069 | default: |
6070 | break; | |
6071 | } | |
6072 | } | |
6073 | pte_unmap_unlock(pte - 1, ptl); | |
6074 | cond_resched(); | |
6075 | ||
6076 | if (addr != end) { | |
6077 | /* | |
6078 | * We have consumed all precharges we got in can_attach(). | |
6079 | * We try charge one by one, but don't do any additional | |
6080 | * charges to mc.to if we have failed in charge once in attach() | |
6081 | * phase. | |
6082 | */ | |
854ffa8d | 6083 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6084 | if (!ret) |
6085 | goto retry; | |
6086 | } | |
6087 | ||
6088 | return ret; | |
6089 | } | |
6090 | ||
7b86ac33 CH |
6091 | static const struct mm_walk_ops charge_walk_ops = { |
6092 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6093 | }; | |
6094 | ||
264a0ae1 | 6095 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 6096 | { |
4ffef5fe | 6097 | lru_add_drain_all(); |
312722cb | 6098 | /* |
81f8c3a4 JW |
6099 | * Signal lock_page_memcg() to take the memcg's move_lock |
6100 | * while we're moving its pages to another memcg. Then wait | |
6101 | * for already started RCU-only updates to finish. | |
312722cb JW |
6102 | */ |
6103 | atomic_inc(&mc.from->moving_account); | |
6104 | synchronize_rcu(); | |
dfe076b0 | 6105 | retry: |
d8ed45c5 | 6106 | if (unlikely(!mmap_read_trylock(mc.mm))) { |
dfe076b0 | 6107 | /* |
c1e8d7c6 | 6108 | * Someone who are holding the mmap_lock might be waiting in |
dfe076b0 DN |
6109 | * waitq. So we cancel all extra charges, wake up all waiters, |
6110 | * and retry. Because we cancel precharges, we might not be able | |
6111 | * to move enough charges, but moving charge is a best-effort | |
6112 | * feature anyway, so it wouldn't be a big problem. | |
6113 | */ | |
6114 | __mem_cgroup_clear_mc(); | |
6115 | cond_resched(); | |
6116 | goto retry; | |
6117 | } | |
26bcd64a NH |
6118 | /* |
6119 | * When we have consumed all precharges and failed in doing | |
6120 | * additional charge, the page walk just aborts. | |
6121 | */ | |
7b86ac33 CH |
6122 | walk_page_range(mc.mm, 0, mc.mm->highest_vm_end, &charge_walk_ops, |
6123 | NULL); | |
0247f3f4 | 6124 | |
d8ed45c5 | 6125 | mmap_read_unlock(mc.mm); |
312722cb | 6126 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
6127 | } |
6128 | ||
264a0ae1 | 6129 | static void mem_cgroup_move_task(void) |
67e465a7 | 6130 | { |
264a0ae1 TH |
6131 | if (mc.to) { |
6132 | mem_cgroup_move_charge(); | |
a433658c | 6133 | mem_cgroup_clear_mc(); |
264a0ae1 | 6134 | } |
67e465a7 | 6135 | } |
5cfb80a7 | 6136 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 6137 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6138 | { |
6139 | return 0; | |
6140 | } | |
1f7dd3e5 | 6141 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6142 | { |
6143 | } | |
264a0ae1 | 6144 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
6145 | { |
6146 | } | |
6147 | #endif | |
67e465a7 | 6148 | |
f00baae7 TH |
6149 | /* |
6150 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
6151 | * to verify whether we're attached to the default hierarchy on each mount |
6152 | * attempt. | |
f00baae7 | 6153 | */ |
eb95419b | 6154 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
6155 | { |
6156 | /* | |
aa6ec29b | 6157 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
6158 | * guarantees that @root doesn't have any children, so turning it |
6159 | * on for the root memcg is enough. | |
6160 | */ | |
9e10a130 | 6161 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
6162 | root_mem_cgroup->use_hierarchy = true; |
6163 | else | |
6164 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
6165 | } |
6166 | ||
677dc973 CD |
6167 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
6168 | { | |
6169 | if (value == PAGE_COUNTER_MAX) | |
6170 | seq_puts(m, "max\n"); | |
6171 | else | |
6172 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); | |
6173 | ||
6174 | return 0; | |
6175 | } | |
6176 | ||
241994ed JW |
6177 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
6178 | struct cftype *cft) | |
6179 | { | |
f5fc3c5d JW |
6180 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
6181 | ||
6182 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
6183 | } |
6184 | ||
bf8d5d52 RG |
6185 | static int memory_min_show(struct seq_file *m, void *v) |
6186 | { | |
677dc973 CD |
6187 | return seq_puts_memcg_tunable(m, |
6188 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); | |
bf8d5d52 RG |
6189 | } |
6190 | ||
6191 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
6192 | char *buf, size_t nbytes, loff_t off) | |
6193 | { | |
6194 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6195 | unsigned long min; | |
6196 | int err; | |
6197 | ||
6198 | buf = strstrip(buf); | |
6199 | err = page_counter_memparse(buf, "max", &min); | |
6200 | if (err) | |
6201 | return err; | |
6202 | ||
6203 | page_counter_set_min(&memcg->memory, min); | |
6204 | ||
6205 | return nbytes; | |
6206 | } | |
6207 | ||
241994ed JW |
6208 | static int memory_low_show(struct seq_file *m, void *v) |
6209 | { | |
677dc973 CD |
6210 | return seq_puts_memcg_tunable(m, |
6211 | READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); | |
241994ed JW |
6212 | } |
6213 | ||
6214 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
6215 | char *buf, size_t nbytes, loff_t off) | |
6216 | { | |
6217 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6218 | unsigned long low; | |
6219 | int err; | |
6220 | ||
6221 | buf = strstrip(buf); | |
d2973697 | 6222 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
6223 | if (err) |
6224 | return err; | |
6225 | ||
23067153 | 6226 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
6227 | |
6228 | return nbytes; | |
6229 | } | |
6230 | ||
6231 | static int memory_high_show(struct seq_file *m, void *v) | |
6232 | { | |
d1663a90 JK |
6233 | return seq_puts_memcg_tunable(m, |
6234 | READ_ONCE(mem_cgroup_from_seq(m)->memory.high)); | |
241994ed JW |
6235 | } |
6236 | ||
6237 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
6238 | char *buf, size_t nbytes, loff_t off) | |
6239 | { | |
6240 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6241 | unsigned int nr_retries = MAX_RECLAIM_RETRIES; |
8c8c383c | 6242 | bool drained = false; |
241994ed JW |
6243 | unsigned long high; |
6244 | int err; | |
6245 | ||
6246 | buf = strstrip(buf); | |
d2973697 | 6247 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
6248 | if (err) |
6249 | return err; | |
6250 | ||
8c8c383c JW |
6251 | for (;;) { |
6252 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6253 | unsigned long reclaimed; | |
6254 | ||
6255 | if (nr_pages <= high) | |
6256 | break; | |
6257 | ||
6258 | if (signal_pending(current)) | |
6259 | break; | |
6260 | ||
6261 | if (!drained) { | |
6262 | drain_all_stock(memcg); | |
6263 | drained = true; | |
6264 | continue; | |
6265 | } | |
6266 | ||
6267 | reclaimed = try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
6268 | GFP_KERNEL, true); | |
6269 | ||
6270 | if (!reclaimed && !nr_retries--) | |
6271 | break; | |
6272 | } | |
588083bb | 6273 | |
536d3bf2 RG |
6274 | page_counter_set_high(&memcg->memory, high); |
6275 | ||
241994ed JW |
6276 | return nbytes; |
6277 | } | |
6278 | ||
6279 | static int memory_max_show(struct seq_file *m, void *v) | |
6280 | { | |
677dc973 CD |
6281 | return seq_puts_memcg_tunable(m, |
6282 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); | |
241994ed JW |
6283 | } |
6284 | ||
6285 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
6286 | char *buf, size_t nbytes, loff_t off) | |
6287 | { | |
6288 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6289 | unsigned int nr_reclaims = MAX_RECLAIM_RETRIES; |
b6e6edcf | 6290 | bool drained = false; |
241994ed JW |
6291 | unsigned long max; |
6292 | int err; | |
6293 | ||
6294 | buf = strstrip(buf); | |
d2973697 | 6295 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
6296 | if (err) |
6297 | return err; | |
6298 | ||
bbec2e15 | 6299 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
6300 | |
6301 | for (;;) { | |
6302 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6303 | ||
6304 | if (nr_pages <= max) | |
6305 | break; | |
6306 | ||
7249c9f0 | 6307 | if (signal_pending(current)) |
b6e6edcf | 6308 | break; |
b6e6edcf JW |
6309 | |
6310 | if (!drained) { | |
6311 | drain_all_stock(memcg); | |
6312 | drained = true; | |
6313 | continue; | |
6314 | } | |
6315 | ||
6316 | if (nr_reclaims) { | |
6317 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
6318 | GFP_KERNEL, true)) | |
6319 | nr_reclaims--; | |
6320 | continue; | |
6321 | } | |
6322 | ||
e27be240 | 6323 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
6324 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
6325 | break; | |
6326 | } | |
241994ed | 6327 | |
2529bb3a | 6328 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
6329 | return nbytes; |
6330 | } | |
6331 | ||
1e577f97 SB |
6332 | static void __memory_events_show(struct seq_file *m, atomic_long_t *events) |
6333 | { | |
6334 | seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW])); | |
6335 | seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH])); | |
6336 | seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX])); | |
6337 | seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM])); | |
6338 | seq_printf(m, "oom_kill %lu\n", | |
6339 | atomic_long_read(&events[MEMCG_OOM_KILL])); | |
6340 | } | |
6341 | ||
241994ed JW |
6342 | static int memory_events_show(struct seq_file *m, void *v) |
6343 | { | |
aa9694bb | 6344 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
241994ed | 6345 | |
1e577f97 SB |
6346 | __memory_events_show(m, memcg->memory_events); |
6347 | return 0; | |
6348 | } | |
6349 | ||
6350 | static int memory_events_local_show(struct seq_file *m, void *v) | |
6351 | { | |
6352 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
241994ed | 6353 | |
1e577f97 | 6354 | __memory_events_show(m, memcg->memory_events_local); |
241994ed JW |
6355 | return 0; |
6356 | } | |
6357 | ||
587d9f72 JW |
6358 | static int memory_stat_show(struct seq_file *m, void *v) |
6359 | { | |
aa9694bb | 6360 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
c8713d0b | 6361 | char *buf; |
1ff9e6e1 | 6362 | |
c8713d0b JW |
6363 | buf = memory_stat_format(memcg); |
6364 | if (!buf) | |
6365 | return -ENOMEM; | |
6366 | seq_puts(m, buf); | |
6367 | kfree(buf); | |
587d9f72 JW |
6368 | return 0; |
6369 | } | |
6370 | ||
3d8b38eb RG |
6371 | static int memory_oom_group_show(struct seq_file *m, void *v) |
6372 | { | |
aa9694bb | 6373 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3d8b38eb RG |
6374 | |
6375 | seq_printf(m, "%d\n", memcg->oom_group); | |
6376 | ||
6377 | return 0; | |
6378 | } | |
6379 | ||
6380 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
6381 | char *buf, size_t nbytes, loff_t off) | |
6382 | { | |
6383 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6384 | int ret, oom_group; | |
6385 | ||
6386 | buf = strstrip(buf); | |
6387 | if (!buf) | |
6388 | return -EINVAL; | |
6389 | ||
6390 | ret = kstrtoint(buf, 0, &oom_group); | |
6391 | if (ret) | |
6392 | return ret; | |
6393 | ||
6394 | if (oom_group != 0 && oom_group != 1) | |
6395 | return -EINVAL; | |
6396 | ||
6397 | memcg->oom_group = oom_group; | |
6398 | ||
6399 | return nbytes; | |
6400 | } | |
6401 | ||
241994ed JW |
6402 | static struct cftype memory_files[] = { |
6403 | { | |
6404 | .name = "current", | |
f5fc3c5d | 6405 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
6406 | .read_u64 = memory_current_read, |
6407 | }, | |
bf8d5d52 RG |
6408 | { |
6409 | .name = "min", | |
6410 | .flags = CFTYPE_NOT_ON_ROOT, | |
6411 | .seq_show = memory_min_show, | |
6412 | .write = memory_min_write, | |
6413 | }, | |
241994ed JW |
6414 | { |
6415 | .name = "low", | |
6416 | .flags = CFTYPE_NOT_ON_ROOT, | |
6417 | .seq_show = memory_low_show, | |
6418 | .write = memory_low_write, | |
6419 | }, | |
6420 | { | |
6421 | .name = "high", | |
6422 | .flags = CFTYPE_NOT_ON_ROOT, | |
6423 | .seq_show = memory_high_show, | |
6424 | .write = memory_high_write, | |
6425 | }, | |
6426 | { | |
6427 | .name = "max", | |
6428 | .flags = CFTYPE_NOT_ON_ROOT, | |
6429 | .seq_show = memory_max_show, | |
6430 | .write = memory_max_write, | |
6431 | }, | |
6432 | { | |
6433 | .name = "events", | |
6434 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 6435 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
6436 | .seq_show = memory_events_show, |
6437 | }, | |
1e577f97 SB |
6438 | { |
6439 | .name = "events.local", | |
6440 | .flags = CFTYPE_NOT_ON_ROOT, | |
6441 | .file_offset = offsetof(struct mem_cgroup, events_local_file), | |
6442 | .seq_show = memory_events_local_show, | |
6443 | }, | |
587d9f72 JW |
6444 | { |
6445 | .name = "stat", | |
587d9f72 JW |
6446 | .seq_show = memory_stat_show, |
6447 | }, | |
3d8b38eb RG |
6448 | { |
6449 | .name = "oom.group", | |
6450 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
6451 | .seq_show = memory_oom_group_show, | |
6452 | .write = memory_oom_group_write, | |
6453 | }, | |
241994ed JW |
6454 | { } /* terminate */ |
6455 | }; | |
6456 | ||
073219e9 | 6457 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 6458 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6459 | .css_online = mem_cgroup_css_online, |
92fb9748 | 6460 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 6461 | .css_released = mem_cgroup_css_released, |
92fb9748 | 6462 | .css_free = mem_cgroup_css_free, |
1ced953b | 6463 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
6464 | .can_attach = mem_cgroup_can_attach, |
6465 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 6466 | .post_attach = mem_cgroup_move_task, |
f00baae7 | 6467 | .bind = mem_cgroup_bind, |
241994ed JW |
6468 | .dfl_cftypes = memory_files, |
6469 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 6470 | .early_init = 0, |
8cdea7c0 | 6471 | }; |
c077719b | 6472 | |
bc50bcc6 JW |
6473 | /* |
6474 | * This function calculates an individual cgroup's effective | |
6475 | * protection which is derived from its own memory.min/low, its | |
6476 | * parent's and siblings' settings, as well as the actual memory | |
6477 | * distribution in the tree. | |
6478 | * | |
6479 | * The following rules apply to the effective protection values: | |
6480 | * | |
6481 | * 1. At the first level of reclaim, effective protection is equal to | |
6482 | * the declared protection in memory.min and memory.low. | |
6483 | * | |
6484 | * 2. To enable safe delegation of the protection configuration, at | |
6485 | * subsequent levels the effective protection is capped to the | |
6486 | * parent's effective protection. | |
6487 | * | |
6488 | * 3. To make complex and dynamic subtrees easier to configure, the | |
6489 | * user is allowed to overcommit the declared protection at a given | |
6490 | * level. If that is the case, the parent's effective protection is | |
6491 | * distributed to the children in proportion to how much protection | |
6492 | * they have declared and how much of it they are utilizing. | |
6493 | * | |
6494 | * This makes distribution proportional, but also work-conserving: | |
6495 | * if one cgroup claims much more protection than it uses memory, | |
6496 | * the unused remainder is available to its siblings. | |
6497 | * | |
6498 | * 4. Conversely, when the declared protection is undercommitted at a | |
6499 | * given level, the distribution of the larger parental protection | |
6500 | * budget is NOT proportional. A cgroup's protection from a sibling | |
6501 | * is capped to its own memory.min/low setting. | |
6502 | * | |
8a931f80 JW |
6503 | * 5. However, to allow protecting recursive subtrees from each other |
6504 | * without having to declare each individual cgroup's fixed share | |
6505 | * of the ancestor's claim to protection, any unutilized - | |
6506 | * "floating" - protection from up the tree is distributed in | |
6507 | * proportion to each cgroup's *usage*. This makes the protection | |
6508 | * neutral wrt sibling cgroups and lets them compete freely over | |
6509 | * the shared parental protection budget, but it protects the | |
6510 | * subtree as a whole from neighboring subtrees. | |
6511 | * | |
6512 | * Note that 4. and 5. are not in conflict: 4. is about protecting | |
6513 | * against immediate siblings whereas 5. is about protecting against | |
6514 | * neighboring subtrees. | |
bc50bcc6 JW |
6515 | */ |
6516 | static unsigned long effective_protection(unsigned long usage, | |
8a931f80 | 6517 | unsigned long parent_usage, |
bc50bcc6 JW |
6518 | unsigned long setting, |
6519 | unsigned long parent_effective, | |
6520 | unsigned long siblings_protected) | |
6521 | { | |
6522 | unsigned long protected; | |
8a931f80 | 6523 | unsigned long ep; |
bc50bcc6 JW |
6524 | |
6525 | protected = min(usage, setting); | |
6526 | /* | |
6527 | * If all cgroups at this level combined claim and use more | |
6528 | * protection then what the parent affords them, distribute | |
6529 | * shares in proportion to utilization. | |
6530 | * | |
6531 | * We are using actual utilization rather than the statically | |
6532 | * claimed protection in order to be work-conserving: claimed | |
6533 | * but unused protection is available to siblings that would | |
6534 | * otherwise get a smaller chunk than what they claimed. | |
6535 | */ | |
6536 | if (siblings_protected > parent_effective) | |
6537 | return protected * parent_effective / siblings_protected; | |
6538 | ||
6539 | /* | |
6540 | * Ok, utilized protection of all children is within what the | |
6541 | * parent affords them, so we know whatever this child claims | |
6542 | * and utilizes is effectively protected. | |
6543 | * | |
6544 | * If there is unprotected usage beyond this value, reclaim | |
6545 | * will apply pressure in proportion to that amount. | |
6546 | * | |
6547 | * If there is unutilized protection, the cgroup will be fully | |
6548 | * shielded from reclaim, but we do return a smaller value for | |
6549 | * protection than what the group could enjoy in theory. This | |
6550 | * is okay. With the overcommit distribution above, effective | |
6551 | * protection is always dependent on how memory is actually | |
6552 | * consumed among the siblings anyway. | |
6553 | */ | |
8a931f80 JW |
6554 | ep = protected; |
6555 | ||
6556 | /* | |
6557 | * If the children aren't claiming (all of) the protection | |
6558 | * afforded to them by the parent, distribute the remainder in | |
6559 | * proportion to the (unprotected) memory of each cgroup. That | |
6560 | * way, cgroups that aren't explicitly prioritized wrt each | |
6561 | * other compete freely over the allowance, but they are | |
6562 | * collectively protected from neighboring trees. | |
6563 | * | |
6564 | * We're using unprotected memory for the weight so that if | |
6565 | * some cgroups DO claim explicit protection, we don't protect | |
6566 | * the same bytes twice. | |
cd324edc JW |
6567 | * |
6568 | * Check both usage and parent_usage against the respective | |
6569 | * protected values. One should imply the other, but they | |
6570 | * aren't read atomically - make sure the division is sane. | |
8a931f80 JW |
6571 | */ |
6572 | if (!(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)) | |
6573 | return ep; | |
cd324edc JW |
6574 | if (parent_effective > siblings_protected && |
6575 | parent_usage > siblings_protected && | |
6576 | usage > protected) { | |
8a931f80 JW |
6577 | unsigned long unclaimed; |
6578 | ||
6579 | unclaimed = parent_effective - siblings_protected; | |
6580 | unclaimed *= usage - protected; | |
6581 | unclaimed /= parent_usage - siblings_protected; | |
6582 | ||
6583 | ep += unclaimed; | |
6584 | } | |
6585 | ||
6586 | return ep; | |
bc50bcc6 JW |
6587 | } |
6588 | ||
241994ed | 6589 | /** |
bf8d5d52 | 6590 | * mem_cgroup_protected - check if memory consumption is in the normal range |
34c81057 | 6591 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
6592 | * @memcg: the memory cgroup to check |
6593 | * | |
23067153 RG |
6594 | * WARNING: This function is not stateless! It can only be used as part |
6595 | * of a top-down tree iteration, not for isolated queries. | |
241994ed | 6596 | */ |
45c7f7e1 CD |
6597 | void mem_cgroup_calculate_protection(struct mem_cgroup *root, |
6598 | struct mem_cgroup *memcg) | |
241994ed | 6599 | { |
8a931f80 | 6600 | unsigned long usage, parent_usage; |
23067153 RG |
6601 | struct mem_cgroup *parent; |
6602 | ||
241994ed | 6603 | if (mem_cgroup_disabled()) |
45c7f7e1 | 6604 | return; |
241994ed | 6605 | |
34c81057 SC |
6606 | if (!root) |
6607 | root = root_mem_cgroup; | |
22f7496f YS |
6608 | |
6609 | /* | |
6610 | * Effective values of the reclaim targets are ignored so they | |
6611 | * can be stale. Have a look at mem_cgroup_protection for more | |
6612 | * details. | |
6613 | * TODO: calculation should be more robust so that we do not need | |
6614 | * that special casing. | |
6615 | */ | |
34c81057 | 6616 | if (memcg == root) |
45c7f7e1 | 6617 | return; |
241994ed | 6618 | |
23067153 | 6619 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 | 6620 | if (!usage) |
45c7f7e1 | 6621 | return; |
bf8d5d52 | 6622 | |
bf8d5d52 | 6623 | parent = parent_mem_cgroup(memcg); |
df2a4196 RG |
6624 | /* No parent means a non-hierarchical mode on v1 memcg */ |
6625 | if (!parent) | |
45c7f7e1 | 6626 | return; |
df2a4196 | 6627 | |
bc50bcc6 | 6628 | if (parent == root) { |
c3d53200 | 6629 | memcg->memory.emin = READ_ONCE(memcg->memory.min); |
03960e33 | 6630 | memcg->memory.elow = READ_ONCE(memcg->memory.low); |
45c7f7e1 | 6631 | return; |
bf8d5d52 RG |
6632 | } |
6633 | ||
8a931f80 JW |
6634 | parent_usage = page_counter_read(&parent->memory); |
6635 | ||
b3a7822e | 6636 | WRITE_ONCE(memcg->memory.emin, effective_protection(usage, parent_usage, |
c3d53200 CD |
6637 | READ_ONCE(memcg->memory.min), |
6638 | READ_ONCE(parent->memory.emin), | |
b3a7822e | 6639 | atomic_long_read(&parent->memory.children_min_usage))); |
23067153 | 6640 | |
b3a7822e | 6641 | WRITE_ONCE(memcg->memory.elow, effective_protection(usage, parent_usage, |
03960e33 CD |
6642 | READ_ONCE(memcg->memory.low), |
6643 | READ_ONCE(parent->memory.elow), | |
b3a7822e | 6644 | atomic_long_read(&parent->memory.children_low_usage))); |
241994ed JW |
6645 | } |
6646 | ||
00501b53 | 6647 | /** |
f0e45fb4 | 6648 | * mem_cgroup_charge - charge a newly allocated page to a cgroup |
00501b53 JW |
6649 | * @page: page to charge |
6650 | * @mm: mm context of the victim | |
6651 | * @gfp_mask: reclaim mode | |
00501b53 JW |
6652 | * |
6653 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
6654 | * pages according to @gfp_mask if necessary. | |
6655 | * | |
f0e45fb4 | 6656 | * Returns 0 on success. Otherwise, an error code is returned. |
00501b53 | 6657 | */ |
d9eb1ea2 | 6658 | int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask) |
00501b53 | 6659 | { |
3fba69a5 | 6660 | unsigned int nr_pages = hpage_nr_pages(page); |
00501b53 | 6661 | struct mem_cgroup *memcg = NULL; |
00501b53 JW |
6662 | int ret = 0; |
6663 | ||
6664 | if (mem_cgroup_disabled()) | |
6665 | goto out; | |
6666 | ||
6667 | if (PageSwapCache(page)) { | |
2d1c4980 JW |
6668 | swp_entry_t ent = { .val = page_private(page), }; |
6669 | unsigned short id; | |
6670 | ||
00501b53 JW |
6671 | /* |
6672 | * Every swap fault against a single page tries to charge the | |
6673 | * page, bail as early as possible. shmem_unuse() encounters | |
eccb52e7 JW |
6674 | * already charged pages, too. page->mem_cgroup is protected |
6675 | * by the page lock, which serializes swap cache removal, which | |
00501b53 JW |
6676 | * in turn serializes uncharging. |
6677 | */ | |
e993d905 | 6678 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
abe2895b | 6679 | if (compound_head(page)->mem_cgroup) |
00501b53 | 6680 | goto out; |
e993d905 | 6681 | |
2d1c4980 JW |
6682 | id = lookup_swap_cgroup_id(ent); |
6683 | rcu_read_lock(); | |
6684 | memcg = mem_cgroup_from_id(id); | |
6685 | if (memcg && !css_tryget_online(&memcg->css)) | |
6686 | memcg = NULL; | |
6687 | rcu_read_unlock(); | |
00501b53 JW |
6688 | } |
6689 | ||
00501b53 JW |
6690 | if (!memcg) |
6691 | memcg = get_mem_cgroup_from_mm(mm); | |
6692 | ||
6693 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
f0e45fb4 JW |
6694 | if (ret) |
6695 | goto out_put; | |
00501b53 | 6696 | |
1a3e1f40 | 6697 | css_get(&memcg->css); |
d9eb1ea2 | 6698 | commit_charge(page, memcg); |
6abb5a86 | 6699 | |
6abb5a86 | 6700 | local_irq_disable(); |
3fba69a5 | 6701 | mem_cgroup_charge_statistics(memcg, page, nr_pages); |
6abb5a86 JW |
6702 | memcg_check_events(memcg, page); |
6703 | local_irq_enable(); | |
00501b53 | 6704 | |
2d1c4980 | 6705 | if (PageSwapCache(page)) { |
00501b53 JW |
6706 | swp_entry_t entry = { .val = page_private(page) }; |
6707 | /* | |
6708 | * The swap entry might not get freed for a long time, | |
6709 | * let's not wait for it. The page already received a | |
6710 | * memory+swap charge, drop the swap entry duplicate. | |
6711 | */ | |
38d8b4e6 | 6712 | mem_cgroup_uncharge_swap(entry, nr_pages); |
00501b53 | 6713 | } |
00501b53 | 6714 | |
f0e45fb4 JW |
6715 | out_put: |
6716 | css_put(&memcg->css); | |
6717 | out: | |
6718 | return ret; | |
3fea5a49 JW |
6719 | } |
6720 | ||
a9d5adee JG |
6721 | struct uncharge_gather { |
6722 | struct mem_cgroup *memcg; | |
9f762dbe | 6723 | unsigned long nr_pages; |
a9d5adee | 6724 | unsigned long pgpgout; |
a9d5adee | 6725 | unsigned long nr_kmem; |
a9d5adee JG |
6726 | struct page *dummy_page; |
6727 | }; | |
6728 | ||
6729 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 6730 | { |
a9d5adee JG |
6731 | memset(ug, 0, sizeof(*ug)); |
6732 | } | |
6733 | ||
6734 | static void uncharge_batch(const struct uncharge_gather *ug) | |
6735 | { | |
747db954 JW |
6736 | unsigned long flags; |
6737 | ||
a9d5adee | 6738 | if (!mem_cgroup_is_root(ug->memcg)) { |
9f762dbe | 6739 | page_counter_uncharge(&ug->memcg->memory, ug->nr_pages); |
7941d214 | 6740 | if (do_memsw_account()) |
9f762dbe | 6741 | page_counter_uncharge(&ug->memcg->memsw, ug->nr_pages); |
a9d5adee JG |
6742 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem) |
6743 | page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem); | |
6744 | memcg_oom_recover(ug->memcg); | |
ce00a967 | 6745 | } |
747db954 JW |
6746 | |
6747 | local_irq_save(flags); | |
c9019e9b | 6748 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); |
9f762dbe | 6749 | __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, ug->nr_pages); |
a9d5adee | 6750 | memcg_check_events(ug->memcg, ug->dummy_page); |
747db954 | 6751 | local_irq_restore(flags); |
a9d5adee JG |
6752 | } |
6753 | ||
6754 | static void uncharge_page(struct page *page, struct uncharge_gather *ug) | |
6755 | { | |
9f762dbe JW |
6756 | unsigned long nr_pages; |
6757 | ||
a9d5adee | 6758 | VM_BUG_ON_PAGE(PageLRU(page), page); |
a9d5adee JG |
6759 | |
6760 | if (!page->mem_cgroup) | |
6761 | return; | |
6762 | ||
6763 | /* | |
6764 | * Nobody should be changing or seriously looking at | |
6765 | * page->mem_cgroup at this point, we have fully | |
6766 | * exclusive access to the page. | |
6767 | */ | |
6768 | ||
6769 | if (ug->memcg != page->mem_cgroup) { | |
6770 | if (ug->memcg) { | |
6771 | uncharge_batch(ug); | |
6772 | uncharge_gather_clear(ug); | |
6773 | } | |
6774 | ug->memcg = page->mem_cgroup; | |
6775 | } | |
6776 | ||
9f762dbe JW |
6777 | nr_pages = compound_nr(page); |
6778 | ug->nr_pages += nr_pages; | |
a9d5adee | 6779 | |
9f762dbe | 6780 | if (!PageKmemcg(page)) { |
a9d5adee JG |
6781 | ug->pgpgout++; |
6782 | } else { | |
9f762dbe | 6783 | ug->nr_kmem += nr_pages; |
a9d5adee JG |
6784 | __ClearPageKmemcg(page); |
6785 | } | |
6786 | ||
6787 | ug->dummy_page = page; | |
6788 | page->mem_cgroup = NULL; | |
1a3e1f40 | 6789 | css_put(&ug->memcg->css); |
747db954 JW |
6790 | } |
6791 | ||
6792 | static void uncharge_list(struct list_head *page_list) | |
6793 | { | |
a9d5adee | 6794 | struct uncharge_gather ug; |
747db954 | 6795 | struct list_head *next; |
a9d5adee JG |
6796 | |
6797 | uncharge_gather_clear(&ug); | |
747db954 | 6798 | |
8b592656 JW |
6799 | /* |
6800 | * Note that the list can be a single page->lru; hence the | |
6801 | * do-while loop instead of a simple list_for_each_entry(). | |
6802 | */ | |
747db954 JW |
6803 | next = page_list->next; |
6804 | do { | |
a9d5adee JG |
6805 | struct page *page; |
6806 | ||
747db954 JW |
6807 | page = list_entry(next, struct page, lru); |
6808 | next = page->lru.next; | |
6809 | ||
a9d5adee | 6810 | uncharge_page(page, &ug); |
747db954 JW |
6811 | } while (next != page_list); |
6812 | ||
a9d5adee JG |
6813 | if (ug.memcg) |
6814 | uncharge_batch(&ug); | |
747db954 JW |
6815 | } |
6816 | ||
0a31bc97 JW |
6817 | /** |
6818 | * mem_cgroup_uncharge - uncharge a page | |
6819 | * @page: page to uncharge | |
6820 | * | |
f0e45fb4 | 6821 | * Uncharge a page previously charged with mem_cgroup_charge(). |
0a31bc97 JW |
6822 | */ |
6823 | void mem_cgroup_uncharge(struct page *page) | |
6824 | { | |
a9d5adee JG |
6825 | struct uncharge_gather ug; |
6826 | ||
0a31bc97 JW |
6827 | if (mem_cgroup_disabled()) |
6828 | return; | |
6829 | ||
747db954 | 6830 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 6831 | if (!page->mem_cgroup) |
0a31bc97 JW |
6832 | return; |
6833 | ||
a9d5adee JG |
6834 | uncharge_gather_clear(&ug); |
6835 | uncharge_page(page, &ug); | |
6836 | uncharge_batch(&ug); | |
747db954 | 6837 | } |
0a31bc97 | 6838 | |
747db954 JW |
6839 | /** |
6840 | * mem_cgroup_uncharge_list - uncharge a list of page | |
6841 | * @page_list: list of pages to uncharge | |
6842 | * | |
6843 | * Uncharge a list of pages previously charged with | |
f0e45fb4 | 6844 | * mem_cgroup_charge(). |
747db954 JW |
6845 | */ |
6846 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
6847 | { | |
6848 | if (mem_cgroup_disabled()) | |
6849 | return; | |
0a31bc97 | 6850 | |
747db954 JW |
6851 | if (!list_empty(page_list)) |
6852 | uncharge_list(page_list); | |
0a31bc97 JW |
6853 | } |
6854 | ||
6855 | /** | |
6a93ca8f JW |
6856 | * mem_cgroup_migrate - charge a page's replacement |
6857 | * @oldpage: currently circulating page | |
6858 | * @newpage: replacement page | |
0a31bc97 | 6859 | * |
6a93ca8f JW |
6860 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
6861 | * be uncharged upon free. | |
0a31bc97 JW |
6862 | * |
6863 | * Both pages must be locked, @newpage->mapping must be set up. | |
6864 | */ | |
6a93ca8f | 6865 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 6866 | { |
29833315 | 6867 | struct mem_cgroup *memcg; |
44b7a8d3 | 6868 | unsigned int nr_pages; |
d93c4130 | 6869 | unsigned long flags; |
0a31bc97 JW |
6870 | |
6871 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6872 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 6873 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
6874 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6875 | newpage); | |
0a31bc97 JW |
6876 | |
6877 | if (mem_cgroup_disabled()) | |
6878 | return; | |
6879 | ||
6880 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 6881 | if (newpage->mem_cgroup) |
0a31bc97 JW |
6882 | return; |
6883 | ||
45637bab | 6884 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 6885 | memcg = oldpage->mem_cgroup; |
29833315 | 6886 | if (!memcg) |
0a31bc97 JW |
6887 | return; |
6888 | ||
44b7a8d3 | 6889 | /* Force-charge the new page. The old one will be freed soon */ |
92855270 | 6890 | nr_pages = hpage_nr_pages(newpage); |
44b7a8d3 JW |
6891 | |
6892 | page_counter_charge(&memcg->memory, nr_pages); | |
6893 | if (do_memsw_account()) | |
6894 | page_counter_charge(&memcg->memsw, nr_pages); | |
0a31bc97 | 6895 | |
1a3e1f40 | 6896 | css_get(&memcg->css); |
d9eb1ea2 | 6897 | commit_charge(newpage, memcg); |
44b7a8d3 | 6898 | |
d93c4130 | 6899 | local_irq_save(flags); |
3fba69a5 | 6900 | mem_cgroup_charge_statistics(memcg, newpage, nr_pages); |
44b7a8d3 | 6901 | memcg_check_events(memcg, newpage); |
d93c4130 | 6902 | local_irq_restore(flags); |
0a31bc97 JW |
6903 | } |
6904 | ||
ef12947c | 6905 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
6906 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
6907 | ||
2d758073 | 6908 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
6909 | { |
6910 | struct mem_cgroup *memcg; | |
6911 | ||
2d758073 JW |
6912 | if (!mem_cgroup_sockets_enabled) |
6913 | return; | |
6914 | ||
e876ecc6 SB |
6915 | /* Do not associate the sock with unrelated interrupted task's memcg. */ |
6916 | if (in_interrupt()) | |
6917 | return; | |
6918 | ||
11092087 JW |
6919 | rcu_read_lock(); |
6920 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
6921 | if (memcg == root_mem_cgroup) |
6922 | goto out; | |
0db15298 | 6923 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 6924 | goto out; |
8965aa28 | 6925 | if (css_tryget(&memcg->css)) |
11092087 | 6926 | sk->sk_memcg = memcg; |
f7e1cb6e | 6927 | out: |
11092087 JW |
6928 | rcu_read_unlock(); |
6929 | } | |
11092087 | 6930 | |
2d758073 | 6931 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 6932 | { |
2d758073 JW |
6933 | if (sk->sk_memcg) |
6934 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
6935 | } |
6936 | ||
6937 | /** | |
6938 | * mem_cgroup_charge_skmem - charge socket memory | |
6939 | * @memcg: memcg to charge | |
6940 | * @nr_pages: number of pages to charge | |
6941 | * | |
6942 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
6943 | * @memcg's configured limit, %false if the charge had to be forced. | |
6944 | */ | |
6945 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6946 | { | |
f7e1cb6e | 6947 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 6948 | |
f7e1cb6e | 6949 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6950 | struct page_counter *fail; |
f7e1cb6e | 6951 | |
0db15298 JW |
6952 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
6953 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
6954 | return true; |
6955 | } | |
0db15298 JW |
6956 | page_counter_charge(&memcg->tcpmem, nr_pages); |
6957 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 6958 | return false; |
11092087 | 6959 | } |
d886f4e4 | 6960 | |
f7e1cb6e JW |
6961 | /* Don't block in the packet receive path */ |
6962 | if (in_softirq()) | |
6963 | gfp_mask = GFP_NOWAIT; | |
6964 | ||
c9019e9b | 6965 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); |
b2807f07 | 6966 | |
f7e1cb6e JW |
6967 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
6968 | return true; | |
6969 | ||
6970 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
6971 | return false; |
6972 | } | |
6973 | ||
6974 | /** | |
6975 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
6976 | * @memcg: memcg to uncharge |
6977 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
6978 | */ |
6979 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6980 | { | |
f7e1cb6e | 6981 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6982 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
6983 | return; |
6984 | } | |
d886f4e4 | 6985 | |
c9019e9b | 6986 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 6987 | |
475d0487 | 6988 | refill_stock(memcg, nr_pages); |
11092087 JW |
6989 | } |
6990 | ||
f7e1cb6e JW |
6991 | static int __init cgroup_memory(char *s) |
6992 | { | |
6993 | char *token; | |
6994 | ||
6995 | while ((token = strsep(&s, ",")) != NULL) { | |
6996 | if (!*token) | |
6997 | continue; | |
6998 | if (!strcmp(token, "nosocket")) | |
6999 | cgroup_memory_nosocket = true; | |
04823c83 VD |
7000 | if (!strcmp(token, "nokmem")) |
7001 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
7002 | } |
7003 | return 0; | |
7004 | } | |
7005 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 7006 | |
2d11085e | 7007 | /* |
1081312f MH |
7008 | * subsys_initcall() for memory controller. |
7009 | * | |
308167fc SAS |
7010 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
7011 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
7012 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
7013 | * should be initialized from here. | |
2d11085e MH |
7014 | */ |
7015 | static int __init mem_cgroup_init(void) | |
7016 | { | |
95a045f6 JW |
7017 | int cpu, node; |
7018 | ||
308167fc SAS |
7019 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
7020 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
7021 | |
7022 | for_each_possible_cpu(cpu) | |
7023 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
7024 | drain_local_stock); | |
7025 | ||
7026 | for_each_node(node) { | |
7027 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
7028 | |
7029 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
7030 | node_online(node) ? node : NUMA_NO_NODE); | |
7031 | ||
ef8f2327 | 7032 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 7033 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 7034 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
7035 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
7036 | } | |
7037 | ||
2d11085e MH |
7038 | return 0; |
7039 | } | |
7040 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
7041 | |
7042 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
7043 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
7044 | { | |
1c2d479a | 7045 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
7046 | /* |
7047 | * The root cgroup cannot be destroyed, so it's refcount must | |
7048 | * always be >= 1. | |
7049 | */ | |
7050 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
7051 | VM_BUG_ON(1); | |
7052 | break; | |
7053 | } | |
7054 | memcg = parent_mem_cgroup(memcg); | |
7055 | if (!memcg) | |
7056 | memcg = root_mem_cgroup; | |
7057 | } | |
7058 | return memcg; | |
7059 | } | |
7060 | ||
21afa38e JW |
7061 | /** |
7062 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
7063 | * @page: page whose memsw charge to transfer | |
7064 | * @entry: swap entry to move the charge to | |
7065 | * | |
7066 | * Transfer the memsw charge of @page to @entry. | |
7067 | */ | |
7068 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
7069 | { | |
1f47b61f | 7070 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 7071 | unsigned int nr_entries; |
21afa38e JW |
7072 | unsigned short oldid; |
7073 | ||
7074 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
7075 | VM_BUG_ON_PAGE(page_count(page), page); | |
7076 | ||
2d1c4980 | 7077 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
21afa38e JW |
7078 | return; |
7079 | ||
7080 | memcg = page->mem_cgroup; | |
7081 | ||
7082 | /* Readahead page, never charged */ | |
7083 | if (!memcg) | |
7084 | return; | |
7085 | ||
1f47b61f VD |
7086 | /* |
7087 | * In case the memcg owning these pages has been offlined and doesn't | |
7088 | * have an ID allocated to it anymore, charge the closest online | |
7089 | * ancestor for the swap instead and transfer the memory+swap charge. | |
7090 | */ | |
7091 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
d6810d73 HY |
7092 | nr_entries = hpage_nr_pages(page); |
7093 | /* Get references for the tail pages, too */ | |
7094 | if (nr_entries > 1) | |
7095 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
7096 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
7097 | nr_entries); | |
21afa38e | 7098 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7099 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e JW |
7100 | |
7101 | page->mem_cgroup = NULL; | |
7102 | ||
7103 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 7104 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 7105 | |
2d1c4980 | 7106 | if (!cgroup_memory_noswap && memcg != swap_memcg) { |
1f47b61f | 7107 | if (!mem_cgroup_is_root(swap_memcg)) |
d6810d73 HY |
7108 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
7109 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
7110 | } |
7111 | ||
ce9ce665 SAS |
7112 | /* |
7113 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 7114 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 7115 | * important here to have the interrupts disabled because it is the |
b93b0163 | 7116 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 SAS |
7117 | */ |
7118 | VM_BUG_ON(!irqs_disabled()); | |
3fba69a5 | 7119 | mem_cgroup_charge_statistics(memcg, page, -nr_entries); |
21afa38e | 7120 | memcg_check_events(memcg, page); |
73f576c0 | 7121 | |
1a3e1f40 | 7122 | css_put(&memcg->css); |
21afa38e JW |
7123 | } |
7124 | ||
38d8b4e6 HY |
7125 | /** |
7126 | * mem_cgroup_try_charge_swap - try charging swap space for a page | |
37e84351 VD |
7127 | * @page: page being added to swap |
7128 | * @entry: swap entry to charge | |
7129 | * | |
38d8b4e6 | 7130 | * Try to charge @page's memcg for the swap space at @entry. |
37e84351 VD |
7131 | * |
7132 | * Returns 0 on success, -ENOMEM on failure. | |
7133 | */ | |
7134 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
7135 | { | |
38d8b4e6 | 7136 | unsigned int nr_pages = hpage_nr_pages(page); |
37e84351 | 7137 | struct page_counter *counter; |
38d8b4e6 | 7138 | struct mem_cgroup *memcg; |
37e84351 VD |
7139 | unsigned short oldid; |
7140 | ||
2d1c4980 | 7141 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
37e84351 VD |
7142 | return 0; |
7143 | ||
7144 | memcg = page->mem_cgroup; | |
7145 | ||
7146 | /* Readahead page, never charged */ | |
7147 | if (!memcg) | |
7148 | return 0; | |
7149 | ||
f3a53a3a TH |
7150 | if (!entry.val) { |
7151 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 7152 | return 0; |
f3a53a3a | 7153 | } |
bb98f2c5 | 7154 | |
1f47b61f VD |
7155 | memcg = mem_cgroup_id_get_online(memcg); |
7156 | ||
2d1c4980 | 7157 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg) && |
38d8b4e6 | 7158 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
7159 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
7160 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 7161 | mem_cgroup_id_put(memcg); |
37e84351 | 7162 | return -ENOMEM; |
1f47b61f | 7163 | } |
37e84351 | 7164 | |
38d8b4e6 HY |
7165 | /* Get references for the tail pages, too */ |
7166 | if (nr_pages > 1) | |
7167 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
7168 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
37e84351 | 7169 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7170 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 7171 | |
37e84351 VD |
7172 | return 0; |
7173 | } | |
7174 | ||
21afa38e | 7175 | /** |
38d8b4e6 | 7176 | * mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 7177 | * @entry: swap entry to uncharge |
38d8b4e6 | 7178 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 7179 | */ |
38d8b4e6 | 7180 | void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
7181 | { |
7182 | struct mem_cgroup *memcg; | |
7183 | unsigned short id; | |
7184 | ||
38d8b4e6 | 7185 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 7186 | rcu_read_lock(); |
adbe427b | 7187 | memcg = mem_cgroup_from_id(id); |
21afa38e | 7188 | if (memcg) { |
2d1c4980 | 7189 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg)) { |
37e84351 | 7190 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
38d8b4e6 | 7191 | page_counter_uncharge(&memcg->swap, nr_pages); |
37e84351 | 7192 | else |
38d8b4e6 | 7193 | page_counter_uncharge(&memcg->memsw, nr_pages); |
37e84351 | 7194 | } |
c9019e9b | 7195 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 7196 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
7197 | } |
7198 | rcu_read_unlock(); | |
7199 | } | |
7200 | ||
d8b38438 VD |
7201 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
7202 | { | |
7203 | long nr_swap_pages = get_nr_swap_pages(); | |
7204 | ||
eccb52e7 | 7205 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
d8b38438 VD |
7206 | return nr_swap_pages; |
7207 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
7208 | nr_swap_pages = min_t(long, nr_swap_pages, | |
bbec2e15 | 7209 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
7210 | page_counter_read(&memcg->swap)); |
7211 | return nr_swap_pages; | |
7212 | } | |
7213 | ||
5ccc5aba VD |
7214 | bool mem_cgroup_swap_full(struct page *page) |
7215 | { | |
7216 | struct mem_cgroup *memcg; | |
7217 | ||
7218 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
7219 | ||
7220 | if (vm_swap_full()) | |
7221 | return true; | |
eccb52e7 | 7222 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
5ccc5aba VD |
7223 | return false; |
7224 | ||
7225 | memcg = page->mem_cgroup; | |
7226 | if (!memcg) | |
7227 | return false; | |
7228 | ||
4b82ab4f JK |
7229 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { |
7230 | unsigned long usage = page_counter_read(&memcg->swap); | |
7231 | ||
7232 | if (usage * 2 >= READ_ONCE(memcg->swap.high) || | |
7233 | usage * 2 >= READ_ONCE(memcg->swap.max)) | |
5ccc5aba | 7234 | return true; |
4b82ab4f | 7235 | } |
5ccc5aba VD |
7236 | |
7237 | return false; | |
7238 | } | |
7239 | ||
eccb52e7 | 7240 | static int __init setup_swap_account(char *s) |
21afa38e JW |
7241 | { |
7242 | if (!strcmp(s, "1")) | |
eccb52e7 | 7243 | cgroup_memory_noswap = 0; |
21afa38e | 7244 | else if (!strcmp(s, "0")) |
eccb52e7 | 7245 | cgroup_memory_noswap = 1; |
21afa38e JW |
7246 | return 1; |
7247 | } | |
eccb52e7 | 7248 | __setup("swapaccount=", setup_swap_account); |
21afa38e | 7249 | |
37e84351 VD |
7250 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
7251 | struct cftype *cft) | |
7252 | { | |
7253 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
7254 | ||
7255 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
7256 | } | |
7257 | ||
4b82ab4f JK |
7258 | static int swap_high_show(struct seq_file *m, void *v) |
7259 | { | |
7260 | return seq_puts_memcg_tunable(m, | |
7261 | READ_ONCE(mem_cgroup_from_seq(m)->swap.high)); | |
7262 | } | |
7263 | ||
7264 | static ssize_t swap_high_write(struct kernfs_open_file *of, | |
7265 | char *buf, size_t nbytes, loff_t off) | |
7266 | { | |
7267 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7268 | unsigned long high; | |
7269 | int err; | |
7270 | ||
7271 | buf = strstrip(buf); | |
7272 | err = page_counter_memparse(buf, "max", &high); | |
7273 | if (err) | |
7274 | return err; | |
7275 | ||
7276 | page_counter_set_high(&memcg->swap, high); | |
7277 | ||
7278 | return nbytes; | |
7279 | } | |
7280 | ||
37e84351 VD |
7281 | static int swap_max_show(struct seq_file *m, void *v) |
7282 | { | |
677dc973 CD |
7283 | return seq_puts_memcg_tunable(m, |
7284 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); | |
37e84351 VD |
7285 | } |
7286 | ||
7287 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
7288 | char *buf, size_t nbytes, loff_t off) | |
7289 | { | |
7290 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7291 | unsigned long max; | |
7292 | int err; | |
7293 | ||
7294 | buf = strstrip(buf); | |
7295 | err = page_counter_memparse(buf, "max", &max); | |
7296 | if (err) | |
7297 | return err; | |
7298 | ||
be09102b | 7299 | xchg(&memcg->swap.max, max); |
37e84351 VD |
7300 | |
7301 | return nbytes; | |
7302 | } | |
7303 | ||
f3a53a3a TH |
7304 | static int swap_events_show(struct seq_file *m, void *v) |
7305 | { | |
aa9694bb | 7306 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
f3a53a3a | 7307 | |
4b82ab4f JK |
7308 | seq_printf(m, "high %lu\n", |
7309 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_HIGH])); | |
f3a53a3a TH |
7310 | seq_printf(m, "max %lu\n", |
7311 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
7312 | seq_printf(m, "fail %lu\n", | |
7313 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
7314 | ||
7315 | return 0; | |
7316 | } | |
7317 | ||
37e84351 VD |
7318 | static struct cftype swap_files[] = { |
7319 | { | |
7320 | .name = "swap.current", | |
7321 | .flags = CFTYPE_NOT_ON_ROOT, | |
7322 | .read_u64 = swap_current_read, | |
7323 | }, | |
4b82ab4f JK |
7324 | { |
7325 | .name = "swap.high", | |
7326 | .flags = CFTYPE_NOT_ON_ROOT, | |
7327 | .seq_show = swap_high_show, | |
7328 | .write = swap_high_write, | |
7329 | }, | |
37e84351 VD |
7330 | { |
7331 | .name = "swap.max", | |
7332 | .flags = CFTYPE_NOT_ON_ROOT, | |
7333 | .seq_show = swap_max_show, | |
7334 | .write = swap_max_write, | |
7335 | }, | |
f3a53a3a TH |
7336 | { |
7337 | .name = "swap.events", | |
7338 | .flags = CFTYPE_NOT_ON_ROOT, | |
7339 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
7340 | .seq_show = swap_events_show, | |
7341 | }, | |
37e84351 VD |
7342 | { } /* terminate */ |
7343 | }; | |
7344 | ||
eccb52e7 | 7345 | static struct cftype memsw_files[] = { |
21afa38e JW |
7346 | { |
7347 | .name = "memsw.usage_in_bytes", | |
7348 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
7349 | .read_u64 = mem_cgroup_read_u64, | |
7350 | }, | |
7351 | { | |
7352 | .name = "memsw.max_usage_in_bytes", | |
7353 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
7354 | .write = mem_cgroup_reset, | |
7355 | .read_u64 = mem_cgroup_read_u64, | |
7356 | }, | |
7357 | { | |
7358 | .name = "memsw.limit_in_bytes", | |
7359 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
7360 | .write = mem_cgroup_write, | |
7361 | .read_u64 = mem_cgroup_read_u64, | |
7362 | }, | |
7363 | { | |
7364 | .name = "memsw.failcnt", | |
7365 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
7366 | .write = mem_cgroup_reset, | |
7367 | .read_u64 = mem_cgroup_read_u64, | |
7368 | }, | |
7369 | { }, /* terminate */ | |
7370 | }; | |
7371 | ||
82ff165c BS |
7372 | /* |
7373 | * If mem_cgroup_swap_init() is implemented as a subsys_initcall() | |
7374 | * instead of a core_initcall(), this could mean cgroup_memory_noswap still | |
7375 | * remains set to false even when memcg is disabled via "cgroup_disable=memory" | |
7376 | * boot parameter. This may result in premature OOPS inside | |
7377 | * mem_cgroup_get_nr_swap_pages() function in corner cases. | |
7378 | */ | |
21afa38e JW |
7379 | static int __init mem_cgroup_swap_init(void) |
7380 | { | |
2d1c4980 JW |
7381 | /* No memory control -> no swap control */ |
7382 | if (mem_cgroup_disabled()) | |
7383 | cgroup_memory_noswap = true; | |
7384 | ||
7385 | if (cgroup_memory_noswap) | |
eccb52e7 JW |
7386 | return 0; |
7387 | ||
7388 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, swap_files)); | |
7389 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, memsw_files)); | |
7390 | ||
21afa38e JW |
7391 | return 0; |
7392 | } | |
82ff165c | 7393 | core_initcall(mem_cgroup_swap_init); |
21afa38e JW |
7394 | |
7395 | #endif /* CONFIG_MEMCG_SWAP */ |