]>
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 | |
6168d0da AS |
23 | * |
24 | * Per memcg lru locking | |
25 | * Copyright (C) 2020 Alibaba, Inc, Alex Shi | |
8cdea7c0 BS |
26 | */ |
27 | ||
3e32cb2e | 28 | #include <linux/page_counter.h> |
8cdea7c0 BS |
29 | #include <linux/memcontrol.h> |
30 | #include <linux/cgroup.h> | |
a520110e | 31 | #include <linux/pagewalk.h> |
6e84f315 | 32 | #include <linux/sched/mm.h> |
3a4f8a0b | 33 | #include <linux/shmem_fs.h> |
4ffef5fe | 34 | #include <linux/hugetlb.h> |
d13d1443 | 35 | #include <linux/pagemap.h> |
1ff9e6e1 | 36 | #include <linux/vm_event_item.h> |
d52aa412 | 37 | #include <linux/smp.h> |
8a9f3ccd | 38 | #include <linux/page-flags.h> |
66e1707b | 39 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
40 | #include <linux/bit_spinlock.h> |
41 | #include <linux/rcupdate.h> | |
e222432b | 42 | #include <linux/limits.h> |
b9e15baf | 43 | #include <linux/export.h> |
8c7c6e34 | 44 | #include <linux/mutex.h> |
bb4cc1a8 | 45 | #include <linux/rbtree.h> |
b6ac57d5 | 46 | #include <linux/slab.h> |
66e1707b | 47 | #include <linux/swap.h> |
02491447 | 48 | #include <linux/swapops.h> |
66e1707b | 49 | #include <linux/spinlock.h> |
2e72b634 | 50 | #include <linux/eventfd.h> |
79bd9814 | 51 | #include <linux/poll.h> |
2e72b634 | 52 | #include <linux/sort.h> |
66e1707b | 53 | #include <linux/fs.h> |
d2ceb9b7 | 54 | #include <linux/seq_file.h> |
70ddf637 | 55 | #include <linux/vmpressure.h> |
b69408e8 | 56 | #include <linux/mm_inline.h> |
5d1ea48b | 57 | #include <linux/swap_cgroup.h> |
cdec2e42 | 58 | #include <linux/cpu.h> |
158e0a2d | 59 | #include <linux/oom.h> |
0056f4e6 | 60 | #include <linux/lockdep.h> |
79bd9814 | 61 | #include <linux/file.h> |
b23afb93 | 62 | #include <linux/tracehook.h> |
0e4b01df | 63 | #include <linux/psi.h> |
c8713d0b | 64 | #include <linux/seq_buf.h> |
08e552c6 | 65 | #include "internal.h" |
d1a4c0b3 | 66 | #include <net/sock.h> |
4bd2c1ee | 67 | #include <net/ip.h> |
f35c3a8e | 68 | #include "slab.h" |
8cdea7c0 | 69 | |
7c0f6ba6 | 70 | #include <linux/uaccess.h> |
8697d331 | 71 | |
cc8e970c KM |
72 | #include <trace/events/vmscan.h> |
73 | ||
073219e9 TH |
74 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
75 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 76 | |
7d828602 JW |
77 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
78 | ||
37d5985c RG |
79 | /* Active memory cgroup to use from an interrupt context */ |
80 | DEFINE_PER_CPU(struct mem_cgroup *, int_active_memcg); | |
c74d40e8 | 81 | EXPORT_PER_CPU_SYMBOL_GPL(int_active_memcg); |
37d5985c | 82 | |
f7e1cb6e | 83 | /* Socket memory accounting disabled? */ |
0f0cace3 | 84 | static bool cgroup_memory_nosocket __ro_after_init; |
f7e1cb6e | 85 | |
04823c83 | 86 | /* Kernel memory accounting disabled? */ |
e267992f | 87 | bool cgroup_memory_nokmem __ro_after_init; |
04823c83 | 88 | |
21afa38e | 89 | /* Whether the swap controller is active */ |
c255a458 | 90 | #ifdef CONFIG_MEMCG_SWAP |
0f0cace3 | 91 | bool cgroup_memory_noswap __ro_after_init; |
c077719b | 92 | #else |
eccb52e7 | 93 | #define cgroup_memory_noswap 1 |
2d1c4980 | 94 | #endif |
c077719b | 95 | |
97b27821 TH |
96 | #ifdef CONFIG_CGROUP_WRITEBACK |
97 | static DECLARE_WAIT_QUEUE_HEAD(memcg_cgwb_frn_waitq); | |
98 | #endif | |
99 | ||
7941d214 JW |
100 | /* Whether legacy memory+swap accounting is active */ |
101 | static bool do_memsw_account(void) | |
102 | { | |
eccb52e7 | 103 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_noswap; |
7941d214 JW |
104 | } |
105 | ||
aa48e47e SB |
106 | /* memcg and lruvec stats flushing */ |
107 | static void flush_memcg_stats_dwork(struct work_struct *w); | |
108 | static DECLARE_DEFERRABLE_WORK(stats_flush_dwork, flush_memcg_stats_dwork); | |
aa48e47e SB |
109 | static DEFINE_SPINLOCK(stats_flush_lock); |
110 | ||
a0db00fc KS |
111 | #define THRESHOLDS_EVENTS_TARGET 128 |
112 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
e9f8974f | 113 | |
bb4cc1a8 AM |
114 | /* |
115 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
116 | * their hierarchy representation | |
117 | */ | |
118 | ||
ef8f2327 | 119 | struct mem_cgroup_tree_per_node { |
bb4cc1a8 | 120 | struct rb_root rb_root; |
fa90b2fd | 121 | struct rb_node *rb_rightmost; |
bb4cc1a8 AM |
122 | spinlock_t lock; |
123 | }; | |
124 | ||
bb4cc1a8 AM |
125 | struct mem_cgroup_tree { |
126 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
127 | }; | |
128 | ||
129 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
130 | ||
9490ff27 KH |
131 | /* for OOM */ |
132 | struct mem_cgroup_eventfd_list { | |
133 | struct list_head list; | |
134 | struct eventfd_ctx *eventfd; | |
135 | }; | |
2e72b634 | 136 | |
79bd9814 TH |
137 | /* |
138 | * cgroup_event represents events which userspace want to receive. | |
139 | */ | |
3bc942f3 | 140 | struct mem_cgroup_event { |
79bd9814 | 141 | /* |
59b6f873 | 142 | * memcg which the event belongs to. |
79bd9814 | 143 | */ |
59b6f873 | 144 | struct mem_cgroup *memcg; |
79bd9814 TH |
145 | /* |
146 | * eventfd to signal userspace about the event. | |
147 | */ | |
148 | struct eventfd_ctx *eventfd; | |
149 | /* | |
150 | * Each of these stored in a list by the cgroup. | |
151 | */ | |
152 | struct list_head list; | |
fba94807 TH |
153 | /* |
154 | * register_event() callback will be used to add new userspace | |
155 | * waiter for changes related to this event. Use eventfd_signal() | |
156 | * on eventfd to send notification to userspace. | |
157 | */ | |
59b6f873 | 158 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 159 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
160 | /* |
161 | * unregister_event() callback will be called when userspace closes | |
162 | * the eventfd or on cgroup removing. This callback must be set, | |
163 | * if you want provide notification functionality. | |
164 | */ | |
59b6f873 | 165 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 166 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
167 | /* |
168 | * All fields below needed to unregister event when | |
169 | * userspace closes eventfd. | |
170 | */ | |
171 | poll_table pt; | |
172 | wait_queue_head_t *wqh; | |
ac6424b9 | 173 | wait_queue_entry_t wait; |
79bd9814 TH |
174 | struct work_struct remove; |
175 | }; | |
176 | ||
c0ff4b85 R |
177 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
178 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 179 | |
7dc74be0 DN |
180 | /* Stuffs for move charges at task migration. */ |
181 | /* | |
1dfab5ab | 182 | * Types of charges to be moved. |
7dc74be0 | 183 | */ |
1dfab5ab JW |
184 | #define MOVE_ANON 0x1U |
185 | #define MOVE_FILE 0x2U | |
186 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 187 | |
4ffef5fe DN |
188 | /* "mc" and its members are protected by cgroup_mutex */ |
189 | static struct move_charge_struct { | |
b1dd693e | 190 | spinlock_t lock; /* for from, to */ |
264a0ae1 | 191 | struct mm_struct *mm; |
4ffef5fe DN |
192 | struct mem_cgroup *from; |
193 | struct mem_cgroup *to; | |
1dfab5ab | 194 | unsigned long flags; |
4ffef5fe | 195 | unsigned long precharge; |
854ffa8d | 196 | unsigned long moved_charge; |
483c30b5 | 197 | unsigned long moved_swap; |
8033b97c DN |
198 | struct task_struct *moving_task; /* a task moving charges */ |
199 | wait_queue_head_t waitq; /* a waitq for other context */ | |
200 | } mc = { | |
2bd9bb20 | 201 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
202 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
203 | }; | |
4ffef5fe | 204 | |
4e416953 BS |
205 | /* |
206 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
207 | * limit reclaim to prevent infinite loops, if they ever occur. | |
208 | */ | |
a0db00fc | 209 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 210 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 211 | |
8c7c6e34 | 212 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
213 | enum res_type { |
214 | _MEM, | |
215 | _MEMSWAP, | |
216 | _OOM_TYPE, | |
510fc4e1 | 217 | _KMEM, |
d55f90bf | 218 | _TCP, |
86ae53e1 GC |
219 | }; |
220 | ||
a0db00fc KS |
221 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
222 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 223 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
f0953a1b | 224 | /* Used for OOM notifier */ |
9490ff27 | 225 | #define OOM_CONTROL (0) |
8c7c6e34 | 226 | |
b05706f1 KT |
227 | /* |
228 | * Iteration constructs for visiting all cgroups (under a tree). If | |
229 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
230 | * be used for reference counting. | |
231 | */ | |
232 | #define for_each_mem_cgroup_tree(iter, root) \ | |
233 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ | |
234 | iter != NULL; \ | |
235 | iter = mem_cgroup_iter(root, iter, NULL)) | |
236 | ||
237 | #define for_each_mem_cgroup(iter) \ | |
238 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ | |
239 | iter != NULL; \ | |
240 | iter = mem_cgroup_iter(NULL, iter, NULL)) | |
241 | ||
7775face TH |
242 | static inline bool should_force_charge(void) |
243 | { | |
244 | return tsk_is_oom_victim(current) || fatal_signal_pending(current) || | |
245 | (current->flags & PF_EXITING); | |
246 | } | |
247 | ||
70ddf637 AV |
248 | /* Some nice accessors for the vmpressure. */ |
249 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
250 | { | |
251 | if (!memcg) | |
252 | memcg = root_mem_cgroup; | |
253 | return &memcg->vmpressure; | |
254 | } | |
255 | ||
9647875b | 256 | struct mem_cgroup *vmpressure_to_memcg(struct vmpressure *vmpr) |
70ddf637 | 257 | { |
9647875b | 258 | return container_of(vmpr, struct mem_cgroup, vmpressure); |
70ddf637 AV |
259 | } |
260 | ||
84c07d11 | 261 | #ifdef CONFIG_MEMCG_KMEM |
bf4f0599 RG |
262 | extern spinlock_t css_set_lock; |
263 | ||
4d5c8aed RG |
264 | bool mem_cgroup_kmem_disabled(void) |
265 | { | |
266 | return cgroup_memory_nokmem; | |
267 | } | |
268 | ||
f1286fae MS |
269 | static void obj_cgroup_uncharge_pages(struct obj_cgroup *objcg, |
270 | unsigned int nr_pages); | |
c1a660de | 271 | |
bf4f0599 RG |
272 | static void obj_cgroup_release(struct percpu_ref *ref) |
273 | { | |
274 | struct obj_cgroup *objcg = container_of(ref, struct obj_cgroup, refcnt); | |
bf4f0599 RG |
275 | unsigned int nr_bytes; |
276 | unsigned int nr_pages; | |
277 | unsigned long flags; | |
278 | ||
279 | /* | |
280 | * At this point all allocated objects are freed, and | |
281 | * objcg->nr_charged_bytes can't have an arbitrary byte value. | |
282 | * However, it can be PAGE_SIZE or (x * PAGE_SIZE). | |
283 | * | |
284 | * The following sequence can lead to it: | |
285 | * 1) CPU0: objcg == stock->cached_objcg | |
286 | * 2) CPU1: we do a small allocation (e.g. 92 bytes), | |
287 | * PAGE_SIZE bytes are charged | |
288 | * 3) CPU1: a process from another memcg is allocating something, | |
289 | * the stock if flushed, | |
290 | * objcg->nr_charged_bytes = PAGE_SIZE - 92 | |
291 | * 5) CPU0: we do release this object, | |
292 | * 92 bytes are added to stock->nr_bytes | |
293 | * 6) CPU0: stock is flushed, | |
294 | * 92 bytes are added to objcg->nr_charged_bytes | |
295 | * | |
296 | * In the result, nr_charged_bytes == PAGE_SIZE. | |
297 | * This page will be uncharged in obj_cgroup_release(). | |
298 | */ | |
299 | nr_bytes = atomic_read(&objcg->nr_charged_bytes); | |
300 | WARN_ON_ONCE(nr_bytes & (PAGE_SIZE - 1)); | |
301 | nr_pages = nr_bytes >> PAGE_SHIFT; | |
302 | ||
bf4f0599 | 303 | if (nr_pages) |
f1286fae | 304 | obj_cgroup_uncharge_pages(objcg, nr_pages); |
271dd6b1 MS |
305 | |
306 | spin_lock_irqsave(&css_set_lock, flags); | |
bf4f0599 | 307 | list_del(&objcg->list); |
bf4f0599 RG |
308 | spin_unlock_irqrestore(&css_set_lock, flags); |
309 | ||
310 | percpu_ref_exit(ref); | |
311 | kfree_rcu(objcg, rcu); | |
312 | } | |
313 | ||
314 | static struct obj_cgroup *obj_cgroup_alloc(void) | |
315 | { | |
316 | struct obj_cgroup *objcg; | |
317 | int ret; | |
318 | ||
319 | objcg = kzalloc(sizeof(struct obj_cgroup), GFP_KERNEL); | |
320 | if (!objcg) | |
321 | return NULL; | |
322 | ||
323 | ret = percpu_ref_init(&objcg->refcnt, obj_cgroup_release, 0, | |
324 | GFP_KERNEL); | |
325 | if (ret) { | |
326 | kfree(objcg); | |
327 | return NULL; | |
328 | } | |
329 | INIT_LIST_HEAD(&objcg->list); | |
330 | return objcg; | |
331 | } | |
332 | ||
333 | static void memcg_reparent_objcgs(struct mem_cgroup *memcg, | |
334 | struct mem_cgroup *parent) | |
335 | { | |
336 | struct obj_cgroup *objcg, *iter; | |
337 | ||
338 | objcg = rcu_replace_pointer(memcg->objcg, NULL, true); | |
339 | ||
340 | spin_lock_irq(&css_set_lock); | |
341 | ||
9838354e MS |
342 | /* 1) Ready to reparent active objcg. */ |
343 | list_add(&objcg->list, &memcg->objcg_list); | |
344 | /* 2) Reparent active objcg and already reparented objcgs to parent. */ | |
345 | list_for_each_entry(iter, &memcg->objcg_list, list) | |
346 | WRITE_ONCE(iter->memcg, parent); | |
347 | /* 3) Move already reparented objcgs to the parent's list */ | |
bf4f0599 RG |
348 | list_splice(&memcg->objcg_list, &parent->objcg_list); |
349 | ||
350 | spin_unlock_irq(&css_set_lock); | |
351 | ||
352 | percpu_ref_kill(&objcg->refcnt); | |
353 | } | |
354 | ||
55007d84 | 355 | /* |
9855609b | 356 | * This will be used as a shrinker list's index. |
b8627835 LZ |
357 | * The main reason for not using cgroup id for this: |
358 | * this works better in sparse environments, where we have a lot of memcgs, | |
359 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
360 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
361 | * 200 entry array for that. | |
55007d84 | 362 | * |
dbcf73e2 VD |
363 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
364 | * will double each time we have to increase it. | |
55007d84 | 365 | */ |
dbcf73e2 VD |
366 | static DEFINE_IDA(memcg_cache_ida); |
367 | int memcg_nr_cache_ids; | |
749c5415 | 368 | |
05257a1a VD |
369 | /* Protects memcg_nr_cache_ids */ |
370 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
371 | ||
372 | void memcg_get_cache_ids(void) | |
373 | { | |
374 | down_read(&memcg_cache_ids_sem); | |
375 | } | |
376 | ||
377 | void memcg_put_cache_ids(void) | |
378 | { | |
379 | up_read(&memcg_cache_ids_sem); | |
380 | } | |
381 | ||
55007d84 GC |
382 | /* |
383 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
384 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
385 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
386 | * tunable, but that is strictly not necessary. | |
387 | * | |
b8627835 | 388 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
389 | * this constant directly from cgroup, but it is understandable that this is |
390 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 391 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
392 | * increase ours as well if it increases. |
393 | */ | |
394 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 395 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 396 | |
d7f25f8a GC |
397 | /* |
398 | * A lot of the calls to the cache allocation functions are expected to be | |
272911a4 | 399 | * inlined by the compiler. Since the calls to memcg_slab_pre_alloc_hook() are |
d7f25f8a GC |
400 | * conditional to this static branch, we'll have to allow modules that does |
401 | * kmem_cache_alloc and the such to see this symbol as well | |
402 | */ | |
ef12947c | 403 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
d7f25f8a | 404 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
0a432dcb | 405 | #endif |
17cc4dfe | 406 | |
ad7fa852 TH |
407 | /** |
408 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
409 | * @page: page of interest | |
410 | * | |
411 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
412 | * with @page is returned. The returned css remains associated with @page | |
413 | * until it is released. | |
414 | * | |
415 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
416 | * is returned. | |
ad7fa852 TH |
417 | */ |
418 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
419 | { | |
420 | struct mem_cgroup *memcg; | |
421 | ||
bcfe06bf | 422 | memcg = page_memcg(page); |
ad7fa852 | 423 | |
9e10a130 | 424 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
425 | memcg = root_mem_cgroup; |
426 | ||
ad7fa852 TH |
427 | return &memcg->css; |
428 | } | |
429 | ||
2fc04524 VD |
430 | /** |
431 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
432 | * @page: the page | |
433 | * | |
434 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
435 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
436 | * is safe to call this function without holding a reference to @page. | |
437 | * | |
438 | * Note, this function is inherently racy, because there is nothing to prevent | |
439 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
440 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
441 | * do not care (such as procfs interfaces). | |
442 | */ | |
443 | ino_t page_cgroup_ino(struct page *page) | |
444 | { | |
445 | struct mem_cgroup *memcg; | |
446 | unsigned long ino = 0; | |
447 | ||
448 | rcu_read_lock(); | |
bcfe06bf | 449 | memcg = page_memcg_check(page); |
286e04b8 | 450 | |
2fc04524 VD |
451 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) |
452 | memcg = parent_mem_cgroup(memcg); | |
453 | if (memcg) | |
454 | ino = cgroup_ino(memcg->css.cgroup); | |
455 | rcu_read_unlock(); | |
456 | return ino; | |
457 | } | |
458 | ||
ef8f2327 MG |
459 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
460 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 461 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
462 | { |
463 | struct rb_node **p = &mctz->rb_root.rb_node; | |
464 | struct rb_node *parent = NULL; | |
ef8f2327 | 465 | struct mem_cgroup_per_node *mz_node; |
fa90b2fd | 466 | bool rightmost = true; |
bb4cc1a8 AM |
467 | |
468 | if (mz->on_tree) | |
469 | return; | |
470 | ||
471 | mz->usage_in_excess = new_usage_in_excess; | |
472 | if (!mz->usage_in_excess) | |
473 | return; | |
474 | while (*p) { | |
475 | parent = *p; | |
ef8f2327 | 476 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 | 477 | tree_node); |
fa90b2fd | 478 | if (mz->usage_in_excess < mz_node->usage_in_excess) { |
bb4cc1a8 | 479 | p = &(*p)->rb_left; |
fa90b2fd | 480 | rightmost = false; |
378876b0 | 481 | } else { |
bb4cc1a8 | 482 | p = &(*p)->rb_right; |
378876b0 | 483 | } |
bb4cc1a8 | 484 | } |
fa90b2fd DB |
485 | |
486 | if (rightmost) | |
487 | mctz->rb_rightmost = &mz->tree_node; | |
488 | ||
bb4cc1a8 AM |
489 | rb_link_node(&mz->tree_node, parent, p); |
490 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
491 | mz->on_tree = true; | |
492 | } | |
493 | ||
ef8f2327 MG |
494 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
495 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
496 | { |
497 | if (!mz->on_tree) | |
498 | return; | |
fa90b2fd DB |
499 | |
500 | if (&mz->tree_node == mctz->rb_rightmost) | |
501 | mctz->rb_rightmost = rb_prev(&mz->tree_node); | |
502 | ||
bb4cc1a8 AM |
503 | rb_erase(&mz->tree_node, &mctz->rb_root); |
504 | mz->on_tree = false; | |
505 | } | |
506 | ||
ef8f2327 MG |
507 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
508 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 509 | { |
0a31bc97 JW |
510 | unsigned long flags; |
511 | ||
512 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 513 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 514 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
515 | } |
516 | ||
3e32cb2e JW |
517 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
518 | { | |
519 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 520 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
521 | unsigned long excess = 0; |
522 | ||
523 | if (nr_pages > soft_limit) | |
524 | excess = nr_pages - soft_limit; | |
525 | ||
526 | return excess; | |
527 | } | |
bb4cc1a8 | 528 | |
658b69c9 | 529 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, int nid) |
bb4cc1a8 | 530 | { |
3e32cb2e | 531 | unsigned long excess; |
ef8f2327 MG |
532 | struct mem_cgroup_per_node *mz; |
533 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 534 | |
2ab082ba | 535 | mctz = soft_limit_tree.rb_tree_per_node[nid]; |
bfc7228b LD |
536 | if (!mctz) |
537 | return; | |
bb4cc1a8 AM |
538 | /* |
539 | * Necessary to update all ancestors when hierarchy is used. | |
540 | * because their event counter is not touched. | |
541 | */ | |
542 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
658b69c9 | 543 | mz = memcg->nodeinfo[nid]; |
3e32cb2e | 544 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
545 | /* |
546 | * We have to update the tree if mz is on RB-tree or | |
547 | * mem is over its softlimit. | |
548 | */ | |
549 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
550 | unsigned long flags; |
551 | ||
552 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
553 | /* if on-tree, remove it */ |
554 | if (mz->on_tree) | |
cf2c8127 | 555 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
556 | /* |
557 | * Insert again. mz->usage_in_excess will be updated. | |
558 | * If excess is 0, no tree ops. | |
559 | */ | |
cf2c8127 | 560 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 561 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
562 | } |
563 | } | |
564 | } | |
565 | ||
566 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
567 | { | |
ef8f2327 MG |
568 | struct mem_cgroup_tree_per_node *mctz; |
569 | struct mem_cgroup_per_node *mz; | |
570 | int nid; | |
bb4cc1a8 | 571 | |
e231875b | 572 | for_each_node(nid) { |
a3747b53 | 573 | mz = memcg->nodeinfo[nid]; |
2ab082ba | 574 | mctz = soft_limit_tree.rb_tree_per_node[nid]; |
bfc7228b LD |
575 | if (mctz) |
576 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
577 | } |
578 | } | |
579 | ||
ef8f2327 MG |
580 | static struct mem_cgroup_per_node * |
581 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 582 | { |
ef8f2327 | 583 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
584 | |
585 | retry: | |
586 | mz = NULL; | |
fa90b2fd | 587 | if (!mctz->rb_rightmost) |
bb4cc1a8 AM |
588 | goto done; /* Nothing to reclaim from */ |
589 | ||
fa90b2fd DB |
590 | mz = rb_entry(mctz->rb_rightmost, |
591 | struct mem_cgroup_per_node, tree_node); | |
bb4cc1a8 AM |
592 | /* |
593 | * Remove the node now but someone else can add it back, | |
594 | * we will to add it back at the end of reclaim to its correct | |
595 | * position in the tree. | |
596 | */ | |
cf2c8127 | 597 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 598 | if (!soft_limit_excess(mz->memcg) || |
8965aa28 | 599 | !css_tryget(&mz->memcg->css)) |
bb4cc1a8 AM |
600 | goto retry; |
601 | done: | |
602 | return mz; | |
603 | } | |
604 | ||
ef8f2327 MG |
605 | static struct mem_cgroup_per_node * |
606 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 607 | { |
ef8f2327 | 608 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 609 | |
0a31bc97 | 610 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 611 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 612 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
613 | return mz; |
614 | } | |
615 | ||
db9adbcb JW |
616 | /** |
617 | * __mod_memcg_state - update cgroup memory statistics | |
618 | * @memcg: the memory cgroup | |
619 | * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item | |
620 | * @val: delta to add to the counter, can be negative | |
621 | */ | |
622 | void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val) | |
623 | { | |
db9adbcb JW |
624 | if (mem_cgroup_disabled()) |
625 | return; | |
626 | ||
2d146aa3 JW |
627 | __this_cpu_add(memcg->vmstats_percpu->state[idx], val); |
628 | cgroup_rstat_updated(memcg->css.cgroup, smp_processor_id()); | |
db9adbcb JW |
629 | } |
630 | ||
2d146aa3 | 631 | /* idx can be of type enum memcg_stat_item or node_stat_item. */ |
a18e6e6e JW |
632 | static unsigned long memcg_page_state_local(struct mem_cgroup *memcg, int idx) |
633 | { | |
634 | long x = 0; | |
635 | int cpu; | |
636 | ||
637 | for_each_possible_cpu(cpu) | |
2d146aa3 | 638 | x += per_cpu(memcg->vmstats_percpu->state[idx], cpu); |
a18e6e6e JW |
639 | #ifdef CONFIG_SMP |
640 | if (x < 0) | |
641 | x = 0; | |
642 | #endif | |
643 | return x; | |
644 | } | |
645 | ||
eedc4e5a RG |
646 | void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, |
647 | int val) | |
db9adbcb JW |
648 | { |
649 | struct mem_cgroup_per_node *pn; | |
42a30035 | 650 | struct mem_cgroup *memcg; |
db9adbcb | 651 | |
db9adbcb | 652 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
42a30035 | 653 | memcg = pn->memcg; |
db9adbcb JW |
654 | |
655 | /* Update memcg */ | |
42a30035 | 656 | __mod_memcg_state(memcg, idx, val); |
db9adbcb | 657 | |
b4c46484 | 658 | /* Update lruvec */ |
7e1c0d6f | 659 | __this_cpu_add(pn->lruvec_stats_percpu->state[idx], val); |
db9adbcb JW |
660 | } |
661 | ||
eedc4e5a RG |
662 | /** |
663 | * __mod_lruvec_state - update lruvec memory statistics | |
664 | * @lruvec: the lruvec | |
665 | * @idx: the stat item | |
666 | * @val: delta to add to the counter, can be negative | |
667 | * | |
668 | * The lruvec is the intersection of the NUMA node and a cgroup. This | |
669 | * function updates the all three counters that are affected by a | |
670 | * change of state at this level: per-node, per-cgroup, per-lruvec. | |
671 | */ | |
672 | void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, | |
673 | int val) | |
674 | { | |
675 | /* Update node */ | |
676 | __mod_node_page_state(lruvec_pgdat(lruvec), idx, val); | |
677 | ||
678 | /* Update memcg and lruvec */ | |
679 | if (!mem_cgroup_disabled()) | |
680 | __mod_memcg_lruvec_state(lruvec, idx, val); | |
681 | } | |
682 | ||
c47d5032 SB |
683 | void __mod_lruvec_page_state(struct page *page, enum node_stat_item idx, |
684 | int val) | |
685 | { | |
686 | struct page *head = compound_head(page); /* rmap on tail pages */ | |
b4e0b68f | 687 | struct mem_cgroup *memcg; |
c47d5032 SB |
688 | pg_data_t *pgdat = page_pgdat(page); |
689 | struct lruvec *lruvec; | |
690 | ||
b4e0b68f MS |
691 | rcu_read_lock(); |
692 | memcg = page_memcg(head); | |
c47d5032 | 693 | /* Untracked pages have no memcg, no lruvec. Update only the node */ |
d635a69d | 694 | if (!memcg) { |
b4e0b68f | 695 | rcu_read_unlock(); |
c47d5032 SB |
696 | __mod_node_page_state(pgdat, idx, val); |
697 | return; | |
698 | } | |
699 | ||
d635a69d | 700 | lruvec = mem_cgroup_lruvec(memcg, pgdat); |
c47d5032 | 701 | __mod_lruvec_state(lruvec, idx, val); |
b4e0b68f | 702 | rcu_read_unlock(); |
c47d5032 | 703 | } |
f0c0c115 | 704 | EXPORT_SYMBOL(__mod_lruvec_page_state); |
c47d5032 | 705 | |
da3ceeff | 706 | void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, int val) |
ec9f0238 | 707 | { |
4f103c63 | 708 | pg_data_t *pgdat = page_pgdat(virt_to_page(p)); |
ec9f0238 RG |
709 | struct mem_cgroup *memcg; |
710 | struct lruvec *lruvec; | |
711 | ||
712 | rcu_read_lock(); | |
4f103c63 | 713 | memcg = mem_cgroup_from_obj(p); |
ec9f0238 | 714 | |
8faeb1ff MS |
715 | /* |
716 | * Untracked pages have no memcg, no lruvec. Update only the | |
717 | * node. If we reparent the slab objects to the root memcg, | |
718 | * when we free the slab object, we need to update the per-memcg | |
719 | * vmstats to keep it correct for the root memcg. | |
720 | */ | |
721 | if (!memcg) { | |
ec9f0238 RG |
722 | __mod_node_page_state(pgdat, idx, val); |
723 | } else { | |
867e5e1d | 724 | lruvec = mem_cgroup_lruvec(memcg, pgdat); |
ec9f0238 RG |
725 | __mod_lruvec_state(lruvec, idx, val); |
726 | } | |
727 | rcu_read_unlock(); | |
728 | } | |
729 | ||
55927114 WL |
730 | /* |
731 | * mod_objcg_mlstate() may be called with irq enabled, so | |
732 | * mod_memcg_lruvec_state() should be used. | |
733 | */ | |
68ac5b3c WL |
734 | static inline void mod_objcg_mlstate(struct obj_cgroup *objcg, |
735 | struct pglist_data *pgdat, | |
736 | enum node_stat_item idx, int nr) | |
fdbcb2a6 WL |
737 | { |
738 | struct mem_cgroup *memcg; | |
739 | struct lruvec *lruvec; | |
740 | ||
741 | rcu_read_lock(); | |
742 | memcg = obj_cgroup_memcg(objcg); | |
743 | lruvec = mem_cgroup_lruvec(memcg, pgdat); | |
55927114 | 744 | mod_memcg_lruvec_state(lruvec, idx, nr); |
fdbcb2a6 WL |
745 | rcu_read_unlock(); |
746 | } | |
747 | ||
db9adbcb JW |
748 | /** |
749 | * __count_memcg_events - account VM events in a cgroup | |
750 | * @memcg: the memory cgroup | |
751 | * @idx: the event item | |
f0953a1b | 752 | * @count: the number of events that occurred |
db9adbcb JW |
753 | */ |
754 | void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, | |
755 | unsigned long count) | |
756 | { | |
db9adbcb JW |
757 | if (mem_cgroup_disabled()) |
758 | return; | |
759 | ||
2d146aa3 JW |
760 | __this_cpu_add(memcg->vmstats_percpu->events[idx], count); |
761 | cgroup_rstat_updated(memcg->css.cgroup, smp_processor_id()); | |
db9adbcb JW |
762 | } |
763 | ||
42a30035 | 764 | static unsigned long memcg_events(struct mem_cgroup *memcg, int event) |
e9f8974f | 765 | { |
2d146aa3 | 766 | return READ_ONCE(memcg->vmstats.events[event]); |
e9f8974f JW |
767 | } |
768 | ||
42a30035 JW |
769 | static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event) |
770 | { | |
815744d7 JW |
771 | long x = 0; |
772 | int cpu; | |
773 | ||
774 | for_each_possible_cpu(cpu) | |
2d146aa3 | 775 | x += per_cpu(memcg->vmstats_percpu->events[event], cpu); |
815744d7 | 776 | return x; |
42a30035 JW |
777 | } |
778 | ||
c0ff4b85 | 779 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
3fba69a5 | 780 | int nr_pages) |
d52aa412 | 781 | { |
e401f176 KH |
782 | /* pagein of a big page is an event. So, ignore page size */ |
783 | if (nr_pages > 0) | |
c9019e9b | 784 | __count_memcg_events(memcg, PGPGIN, 1); |
3751d604 | 785 | else { |
c9019e9b | 786 | __count_memcg_events(memcg, PGPGOUT, 1); |
3751d604 KH |
787 | nr_pages = -nr_pages; /* for event */ |
788 | } | |
e401f176 | 789 | |
871789d4 | 790 | __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages); |
6d12e2d8 KH |
791 | } |
792 | ||
f53d7ce3 JW |
793 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
794 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
795 | { |
796 | unsigned long val, next; | |
797 | ||
871789d4 CD |
798 | val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events); |
799 | next = __this_cpu_read(memcg->vmstats_percpu->targets[target]); | |
7a159cc9 | 800 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 801 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
802 | switch (target) { |
803 | case MEM_CGROUP_TARGET_THRESH: | |
804 | next = val + THRESHOLDS_EVENTS_TARGET; | |
805 | break; | |
bb4cc1a8 AM |
806 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
807 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
808 | break; | |
f53d7ce3 JW |
809 | default: |
810 | break; | |
811 | } | |
871789d4 | 812 | __this_cpu_write(memcg->vmstats_percpu->targets[target], next); |
f53d7ce3 | 813 | return true; |
7a159cc9 | 814 | } |
f53d7ce3 | 815 | return false; |
d2265e6f KH |
816 | } |
817 | ||
818 | /* | |
819 | * Check events in order. | |
820 | * | |
821 | */ | |
8e88bd2d | 822 | static void memcg_check_events(struct mem_cgroup *memcg, int nid) |
d2265e6f KH |
823 | { |
824 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
825 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
826 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 827 | bool do_softlimit; |
f53d7ce3 | 828 | |
bb4cc1a8 AM |
829 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
830 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
c0ff4b85 | 831 | mem_cgroup_threshold(memcg); |
bb4cc1a8 | 832 | if (unlikely(do_softlimit)) |
8e88bd2d | 833 | mem_cgroup_update_tree(memcg, nid); |
0a31bc97 | 834 | } |
d2265e6f KH |
835 | } |
836 | ||
cf475ad2 | 837 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 838 | { |
31a78f23 BS |
839 | /* |
840 | * mm_update_next_owner() may clear mm->owner to NULL | |
841 | * if it races with swapoff, page migration, etc. | |
842 | * So this can be called with p == NULL. | |
843 | */ | |
844 | if (unlikely(!p)) | |
845 | return NULL; | |
846 | ||
073219e9 | 847 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 848 | } |
33398cf2 | 849 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 850 | |
04f94e3f DS |
851 | static __always_inline struct mem_cgroup *active_memcg(void) |
852 | { | |
55a68c82 | 853 | if (!in_task()) |
04f94e3f DS |
854 | return this_cpu_read(int_active_memcg); |
855 | else | |
856 | return current->active_memcg; | |
857 | } | |
858 | ||
d46eb14b SB |
859 | /** |
860 | * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg. | |
861 | * @mm: mm from which memcg should be extracted. It can be NULL. | |
862 | * | |
04f94e3f DS |
863 | * Obtain a reference on mm->memcg and returns it if successful. If mm |
864 | * is NULL, then the memcg is chosen as follows: | |
865 | * 1) The active memcg, if set. | |
866 | * 2) current->mm->memcg, if available | |
867 | * 3) root memcg | |
868 | * If mem_cgroup is disabled, NULL is returned. | |
d46eb14b SB |
869 | */ |
870 | struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) | |
54595fe2 | 871 | { |
d46eb14b SB |
872 | struct mem_cgroup *memcg; |
873 | ||
874 | if (mem_cgroup_disabled()) | |
875 | return NULL; | |
0b7f569e | 876 | |
2884b6b7 MS |
877 | /* |
878 | * Page cache insertions can happen without an | |
879 | * actual mm context, e.g. during disk probing | |
880 | * on boot, loopback IO, acct() writes etc. | |
881 | * | |
882 | * No need to css_get on root memcg as the reference | |
883 | * counting is disabled on the root level in the | |
884 | * cgroup core. See CSS_NO_REF. | |
885 | */ | |
04f94e3f DS |
886 | if (unlikely(!mm)) { |
887 | memcg = active_memcg(); | |
888 | if (unlikely(memcg)) { | |
889 | /* remote memcg must hold a ref */ | |
890 | css_get(&memcg->css); | |
891 | return memcg; | |
892 | } | |
893 | mm = current->mm; | |
894 | if (unlikely(!mm)) | |
895 | return root_mem_cgroup; | |
896 | } | |
2884b6b7 | 897 | |
54595fe2 KH |
898 | rcu_read_lock(); |
899 | do { | |
2884b6b7 MS |
900 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
901 | if (unlikely(!memcg)) | |
df381975 | 902 | memcg = root_mem_cgroup; |
00d484f3 | 903 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 904 | rcu_read_unlock(); |
c0ff4b85 | 905 | return memcg; |
54595fe2 | 906 | } |
d46eb14b SB |
907 | EXPORT_SYMBOL(get_mem_cgroup_from_mm); |
908 | ||
4127c650 RG |
909 | static __always_inline bool memcg_kmem_bypass(void) |
910 | { | |
911 | /* Allow remote memcg charging from any context. */ | |
912 | if (unlikely(active_memcg())) | |
913 | return false; | |
914 | ||
915 | /* Memcg to charge can't be determined. */ | |
6126891c | 916 | if (!in_task() || !current->mm || (current->flags & PF_KTHREAD)) |
4127c650 RG |
917 | return true; |
918 | ||
919 | return false; | |
920 | } | |
921 | ||
5660048c JW |
922 | /** |
923 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
924 | * @root: hierarchy root | |
925 | * @prev: previously returned memcg, NULL on first invocation | |
926 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
927 | * | |
928 | * Returns references to children of the hierarchy below @root, or | |
929 | * @root itself, or %NULL after a full round-trip. | |
930 | * | |
931 | * Caller must pass the return value in @prev on subsequent | |
932 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
933 | * to cancel a hierarchy walk before the round-trip is complete. | |
934 | * | |
05bdc520 ML |
935 | * Reclaimers can specify a node in @reclaim to divide up the memcgs |
936 | * in the hierarchy among all concurrent reclaimers operating on the | |
937 | * same node. | |
5660048c | 938 | */ |
694fbc0f | 939 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 940 | struct mem_cgroup *prev, |
694fbc0f | 941 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 942 | { |
3f649ab7 | 943 | struct mem_cgroup_reclaim_iter *iter; |
5ac8fb31 | 944 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 945 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 946 | struct mem_cgroup *pos = NULL; |
711d3d2c | 947 | |
694fbc0f AM |
948 | if (mem_cgroup_disabled()) |
949 | return NULL; | |
5660048c | 950 | |
9f3a0d09 JW |
951 | if (!root) |
952 | root = root_mem_cgroup; | |
7d74b06f | 953 | |
9f3a0d09 | 954 | if (prev && !reclaim) |
5ac8fb31 | 955 | pos = prev; |
14067bb3 | 956 | |
542f85f9 | 957 | rcu_read_lock(); |
5f578161 | 958 | |
5ac8fb31 | 959 | if (reclaim) { |
ef8f2327 | 960 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 961 | |
a3747b53 | 962 | mz = root->nodeinfo[reclaim->pgdat->node_id]; |
9da83f3f | 963 | iter = &mz->iter; |
5ac8fb31 JW |
964 | |
965 | if (prev && reclaim->generation != iter->generation) | |
966 | goto out_unlock; | |
967 | ||
6df38689 | 968 | while (1) { |
4db0c3c2 | 969 | pos = READ_ONCE(iter->position); |
6df38689 VD |
970 | if (!pos || css_tryget(&pos->css)) |
971 | break; | |
5ac8fb31 | 972 | /* |
6df38689 VD |
973 | * css reference reached zero, so iter->position will |
974 | * be cleared by ->css_released. However, we should not | |
975 | * rely on this happening soon, because ->css_released | |
976 | * is called from a work queue, and by busy-waiting we | |
977 | * might block it. So we clear iter->position right | |
978 | * away. | |
5ac8fb31 | 979 | */ |
6df38689 VD |
980 | (void)cmpxchg(&iter->position, pos, NULL); |
981 | } | |
5ac8fb31 JW |
982 | } |
983 | ||
984 | if (pos) | |
985 | css = &pos->css; | |
986 | ||
987 | for (;;) { | |
988 | css = css_next_descendant_pre(css, &root->css); | |
989 | if (!css) { | |
990 | /* | |
991 | * Reclaimers share the hierarchy walk, and a | |
992 | * new one might jump in right at the end of | |
993 | * the hierarchy - make sure they see at least | |
994 | * one group and restart from the beginning. | |
995 | */ | |
996 | if (!prev) | |
997 | continue; | |
998 | break; | |
527a5ec9 | 999 | } |
7d74b06f | 1000 | |
5ac8fb31 JW |
1001 | /* |
1002 | * Verify the css and acquire a reference. The root | |
1003 | * is provided by the caller, so we know it's alive | |
1004 | * and kicking, and don't take an extra reference. | |
1005 | */ | |
1006 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 1007 | |
5ac8fb31 JW |
1008 | if (css == &root->css) |
1009 | break; | |
14067bb3 | 1010 | |
0b8f73e1 JW |
1011 | if (css_tryget(css)) |
1012 | break; | |
9f3a0d09 | 1013 | |
5ac8fb31 | 1014 | memcg = NULL; |
9f3a0d09 | 1015 | } |
5ac8fb31 JW |
1016 | |
1017 | if (reclaim) { | |
5ac8fb31 | 1018 | /* |
6df38689 VD |
1019 | * The position could have already been updated by a competing |
1020 | * thread, so check that the value hasn't changed since we read | |
1021 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 1022 | */ |
6df38689 VD |
1023 | (void)cmpxchg(&iter->position, pos, memcg); |
1024 | ||
5ac8fb31 JW |
1025 | if (pos) |
1026 | css_put(&pos->css); | |
1027 | ||
1028 | if (!memcg) | |
1029 | iter->generation++; | |
1030 | else if (!prev) | |
1031 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1032 | } |
5ac8fb31 | 1033 | |
542f85f9 MH |
1034 | out_unlock: |
1035 | rcu_read_unlock(); | |
c40046f3 MH |
1036 | if (prev && prev != root) |
1037 | css_put(&prev->css); | |
1038 | ||
9f3a0d09 | 1039 | return memcg; |
14067bb3 | 1040 | } |
7d74b06f | 1041 | |
5660048c JW |
1042 | /** |
1043 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1044 | * @root: hierarchy root | |
1045 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1046 | */ | |
1047 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1048 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1049 | { |
1050 | if (!root) | |
1051 | root = root_mem_cgroup; | |
1052 | if (prev && prev != root) | |
1053 | css_put(&prev->css); | |
1054 | } | |
7d74b06f | 1055 | |
54a83d6b MC |
1056 | static void __invalidate_reclaim_iterators(struct mem_cgroup *from, |
1057 | struct mem_cgroup *dead_memcg) | |
6df38689 | 1058 | { |
6df38689 | 1059 | struct mem_cgroup_reclaim_iter *iter; |
ef8f2327 MG |
1060 | struct mem_cgroup_per_node *mz; |
1061 | int nid; | |
6df38689 | 1062 | |
54a83d6b | 1063 | for_each_node(nid) { |
a3747b53 | 1064 | mz = from->nodeinfo[nid]; |
9da83f3f YS |
1065 | iter = &mz->iter; |
1066 | cmpxchg(&iter->position, dead_memcg, NULL); | |
6df38689 VD |
1067 | } |
1068 | } | |
1069 | ||
54a83d6b MC |
1070 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1071 | { | |
1072 | struct mem_cgroup *memcg = dead_memcg; | |
1073 | struct mem_cgroup *last; | |
1074 | ||
1075 | do { | |
1076 | __invalidate_reclaim_iterators(memcg, dead_memcg); | |
1077 | last = memcg; | |
1078 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1079 | ||
1080 | /* | |
1081 | * When cgruop1 non-hierarchy mode is used, | |
1082 | * parent_mem_cgroup() does not walk all the way up to the | |
1083 | * cgroup root (root_mem_cgroup). So we have to handle | |
1084 | * dead_memcg from cgroup root separately. | |
1085 | */ | |
1086 | if (last != root_mem_cgroup) | |
1087 | __invalidate_reclaim_iterators(root_mem_cgroup, | |
1088 | dead_memcg); | |
1089 | } | |
1090 | ||
7c5f64f8 VD |
1091 | /** |
1092 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1093 | * @memcg: hierarchy root | |
1094 | * @fn: function to call for each task | |
1095 | * @arg: argument passed to @fn | |
1096 | * | |
1097 | * This function iterates over tasks attached to @memcg or to any of its | |
1098 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
1099 | * value, the function breaks the iteration loop and returns the value. | |
1100 | * Otherwise, it will iterate over all tasks and return 0. | |
1101 | * | |
1102 | * This function must not be called for the root memory cgroup. | |
1103 | */ | |
1104 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
1105 | int (*fn)(struct task_struct *, void *), void *arg) | |
1106 | { | |
1107 | struct mem_cgroup *iter; | |
1108 | int ret = 0; | |
1109 | ||
1110 | BUG_ON(memcg == root_mem_cgroup); | |
1111 | ||
1112 | for_each_mem_cgroup_tree(iter, memcg) { | |
1113 | struct css_task_iter it; | |
1114 | struct task_struct *task; | |
1115 | ||
f168a9a5 | 1116 | css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it); |
7c5f64f8 VD |
1117 | while (!ret && (task = css_task_iter_next(&it))) |
1118 | ret = fn(task, arg); | |
1119 | css_task_iter_end(&it); | |
1120 | if (ret) { | |
1121 | mem_cgroup_iter_break(memcg, iter); | |
1122 | break; | |
1123 | } | |
1124 | } | |
1125 | return ret; | |
1126 | } | |
1127 | ||
6168d0da AS |
1128 | #ifdef CONFIG_DEBUG_VM |
1129 | void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page) | |
1130 | { | |
1131 | struct mem_cgroup *memcg; | |
1132 | ||
1133 | if (mem_cgroup_disabled()) | |
1134 | return; | |
1135 | ||
1136 | memcg = page_memcg(page); | |
1137 | ||
1138 | if (!memcg) | |
1139 | VM_BUG_ON_PAGE(lruvec_memcg(lruvec) != root_mem_cgroup, page); | |
1140 | else | |
1141 | VM_BUG_ON_PAGE(lruvec_memcg(lruvec) != memcg, page); | |
1142 | } | |
1143 | #endif | |
1144 | ||
6168d0da AS |
1145 | /** |
1146 | * lock_page_lruvec - lock and return lruvec for a given page. | |
1147 | * @page: the page | |
1148 | * | |
d7e3aba5 AS |
1149 | * These functions are safe to use under any of the following conditions: |
1150 | * - page locked | |
1151 | * - PageLRU cleared | |
1152 | * - lock_page_memcg() | |
1153 | * - page->_refcount is zero | |
6168d0da AS |
1154 | */ |
1155 | struct lruvec *lock_page_lruvec(struct page *page) | |
1156 | { | |
1157 | struct lruvec *lruvec; | |
6168d0da | 1158 | |
a984226f | 1159 | lruvec = mem_cgroup_page_lruvec(page); |
6168d0da | 1160 | spin_lock(&lruvec->lru_lock); |
6168d0da AS |
1161 | |
1162 | lruvec_memcg_debug(lruvec, page); | |
1163 | ||
1164 | return lruvec; | |
1165 | } | |
1166 | ||
1167 | struct lruvec *lock_page_lruvec_irq(struct page *page) | |
1168 | { | |
1169 | struct lruvec *lruvec; | |
6168d0da | 1170 | |
a984226f | 1171 | lruvec = mem_cgroup_page_lruvec(page); |
6168d0da | 1172 | spin_lock_irq(&lruvec->lru_lock); |
6168d0da AS |
1173 | |
1174 | lruvec_memcg_debug(lruvec, page); | |
1175 | ||
1176 | return lruvec; | |
1177 | } | |
1178 | ||
1179 | struct lruvec *lock_page_lruvec_irqsave(struct page *page, unsigned long *flags) | |
1180 | { | |
1181 | struct lruvec *lruvec; | |
6168d0da | 1182 | |
a984226f | 1183 | lruvec = mem_cgroup_page_lruvec(page); |
6168d0da | 1184 | spin_lock_irqsave(&lruvec->lru_lock, *flags); |
6168d0da AS |
1185 | |
1186 | lruvec_memcg_debug(lruvec, page); | |
1187 | ||
1188 | return lruvec; | |
1189 | } | |
1190 | ||
925b7673 | 1191 | /** |
fa9add64 HD |
1192 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1193 | * @lruvec: mem_cgroup per zone lru vector | |
1194 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1195 | * @zid: zone id of the accounted pages |
fa9add64 | 1196 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1197 | * |
ca707239 HD |
1198 | * This function must be called under lru_lock, just before a page is added |
1199 | * to or just after a page is removed from an lru list (that ordering being | |
1200 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 1201 | */ |
fa9add64 | 1202 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1203 | int zid, int nr_pages) |
3f58a829 | 1204 | { |
ef8f2327 | 1205 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1206 | unsigned long *lru_size; |
ca707239 | 1207 | long size; |
3f58a829 MK |
1208 | |
1209 | if (mem_cgroup_disabled()) | |
1210 | return; | |
1211 | ||
ef8f2327 | 1212 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1213 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1214 | |
1215 | if (nr_pages < 0) | |
1216 | *lru_size += nr_pages; | |
1217 | ||
1218 | size = *lru_size; | |
b4536f0c MH |
1219 | if (WARN_ONCE(size < 0, |
1220 | "%s(%p, %d, %d): lru_size %ld\n", | |
1221 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1222 | VM_BUG_ON(1); |
1223 | *lru_size = 0; | |
1224 | } | |
1225 | ||
1226 | if (nr_pages > 0) | |
1227 | *lru_size += nr_pages; | |
08e552c6 | 1228 | } |
544122e5 | 1229 | |
19942822 | 1230 | /** |
9d11ea9f | 1231 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1232 | * @memcg: the memory cgroup |
19942822 | 1233 | * |
9d11ea9f | 1234 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1235 | * pages. |
19942822 | 1236 | */ |
c0ff4b85 | 1237 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1238 | { |
3e32cb2e JW |
1239 | unsigned long margin = 0; |
1240 | unsigned long count; | |
1241 | unsigned long limit; | |
9d11ea9f | 1242 | |
3e32cb2e | 1243 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1244 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1245 | if (count < limit) |
1246 | margin = limit - count; | |
1247 | ||
7941d214 | 1248 | if (do_memsw_account()) { |
3e32cb2e | 1249 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1250 | limit = READ_ONCE(memcg->memsw.max); |
1c4448ed | 1251 | if (count < limit) |
3e32cb2e | 1252 | margin = min(margin, limit - count); |
cbedbac3 LR |
1253 | else |
1254 | margin = 0; | |
3e32cb2e JW |
1255 | } |
1256 | ||
1257 | return margin; | |
19942822 JW |
1258 | } |
1259 | ||
32047e2a | 1260 | /* |
bdcbb659 | 1261 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1262 | * |
bdcbb659 QH |
1263 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1264 | * moving cgroups. This is for waiting at high-memory pressure | |
1265 | * caused by "move". | |
32047e2a | 1266 | */ |
c0ff4b85 | 1267 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1268 | { |
2bd9bb20 KH |
1269 | struct mem_cgroup *from; |
1270 | struct mem_cgroup *to; | |
4b534334 | 1271 | bool ret = false; |
2bd9bb20 KH |
1272 | /* |
1273 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1274 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1275 | */ | |
1276 | spin_lock(&mc.lock); | |
1277 | from = mc.from; | |
1278 | to = mc.to; | |
1279 | if (!from) | |
1280 | goto unlock; | |
3e92041d | 1281 | |
2314b42d JW |
1282 | ret = mem_cgroup_is_descendant(from, memcg) || |
1283 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1284 | unlock: |
1285 | spin_unlock(&mc.lock); | |
4b534334 KH |
1286 | return ret; |
1287 | } | |
1288 | ||
c0ff4b85 | 1289 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1290 | { |
1291 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1292 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1293 | DEFINE_WAIT(wait); |
1294 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1295 | /* moving charge context might have finished. */ | |
1296 | if (mc.moving_task) | |
1297 | schedule(); | |
1298 | finish_wait(&mc.waitq, &wait); | |
1299 | return true; | |
1300 | } | |
1301 | } | |
1302 | return false; | |
1303 | } | |
1304 | ||
5f9a4f4a MS |
1305 | struct memory_stat { |
1306 | const char *name; | |
5f9a4f4a MS |
1307 | unsigned int idx; |
1308 | }; | |
1309 | ||
57b2847d | 1310 | static const struct memory_stat memory_stats[] = { |
fff66b79 MS |
1311 | { "anon", NR_ANON_MAPPED }, |
1312 | { "file", NR_FILE_PAGES }, | |
1313 | { "kernel_stack", NR_KERNEL_STACK_KB }, | |
1314 | { "pagetables", NR_PAGETABLE }, | |
1315 | { "percpu", MEMCG_PERCPU_B }, | |
1316 | { "sock", MEMCG_SOCK }, | |
1317 | { "shmem", NR_SHMEM }, | |
1318 | { "file_mapped", NR_FILE_MAPPED }, | |
1319 | { "file_dirty", NR_FILE_DIRTY }, | |
1320 | { "file_writeback", NR_WRITEBACK }, | |
b6038942 SB |
1321 | #ifdef CONFIG_SWAP |
1322 | { "swapcached", NR_SWAPCACHE }, | |
1323 | #endif | |
5f9a4f4a | 1324 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
fff66b79 MS |
1325 | { "anon_thp", NR_ANON_THPS }, |
1326 | { "file_thp", NR_FILE_THPS }, | |
1327 | { "shmem_thp", NR_SHMEM_THPS }, | |
5f9a4f4a | 1328 | #endif |
fff66b79 MS |
1329 | { "inactive_anon", NR_INACTIVE_ANON }, |
1330 | { "active_anon", NR_ACTIVE_ANON }, | |
1331 | { "inactive_file", NR_INACTIVE_FILE }, | |
1332 | { "active_file", NR_ACTIVE_FILE }, | |
1333 | { "unevictable", NR_UNEVICTABLE }, | |
1334 | { "slab_reclaimable", NR_SLAB_RECLAIMABLE_B }, | |
1335 | { "slab_unreclaimable", NR_SLAB_UNRECLAIMABLE_B }, | |
5f9a4f4a MS |
1336 | |
1337 | /* The memory events */ | |
fff66b79 MS |
1338 | { "workingset_refault_anon", WORKINGSET_REFAULT_ANON }, |
1339 | { "workingset_refault_file", WORKINGSET_REFAULT_FILE }, | |
1340 | { "workingset_activate_anon", WORKINGSET_ACTIVATE_ANON }, | |
1341 | { "workingset_activate_file", WORKINGSET_ACTIVATE_FILE }, | |
1342 | { "workingset_restore_anon", WORKINGSET_RESTORE_ANON }, | |
1343 | { "workingset_restore_file", WORKINGSET_RESTORE_FILE }, | |
1344 | { "workingset_nodereclaim", WORKINGSET_NODERECLAIM }, | |
5f9a4f4a MS |
1345 | }; |
1346 | ||
fff66b79 MS |
1347 | /* Translate stat items to the correct unit for memory.stat output */ |
1348 | static int memcg_page_state_unit(int item) | |
1349 | { | |
1350 | switch (item) { | |
1351 | case MEMCG_PERCPU_B: | |
1352 | case NR_SLAB_RECLAIMABLE_B: | |
1353 | case NR_SLAB_UNRECLAIMABLE_B: | |
1354 | case WORKINGSET_REFAULT_ANON: | |
1355 | case WORKINGSET_REFAULT_FILE: | |
1356 | case WORKINGSET_ACTIVATE_ANON: | |
1357 | case WORKINGSET_ACTIVATE_FILE: | |
1358 | case WORKINGSET_RESTORE_ANON: | |
1359 | case WORKINGSET_RESTORE_FILE: | |
1360 | case WORKINGSET_NODERECLAIM: | |
1361 | return 1; | |
1362 | case NR_KERNEL_STACK_KB: | |
1363 | return SZ_1K; | |
1364 | default: | |
1365 | return PAGE_SIZE; | |
1366 | } | |
1367 | } | |
1368 | ||
1369 | static inline unsigned long memcg_page_state_output(struct mem_cgroup *memcg, | |
1370 | int item) | |
1371 | { | |
1372 | return memcg_page_state(memcg, item) * memcg_page_state_unit(item); | |
1373 | } | |
1374 | ||
c8713d0b JW |
1375 | static char *memory_stat_format(struct mem_cgroup *memcg) |
1376 | { | |
1377 | struct seq_buf s; | |
1378 | int i; | |
71cd3113 | 1379 | |
c8713d0b JW |
1380 | seq_buf_init(&s, kmalloc(PAGE_SIZE, GFP_KERNEL), PAGE_SIZE); |
1381 | if (!s.buffer) | |
1382 | return NULL; | |
1383 | ||
1384 | /* | |
1385 | * Provide statistics on the state of the memory subsystem as | |
1386 | * well as cumulative event counters that show past behavior. | |
1387 | * | |
1388 | * This list is ordered following a combination of these gradients: | |
1389 | * 1) generic big picture -> specifics and details | |
1390 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
1391 | * | |
1392 | * Current memory state: | |
1393 | */ | |
2d146aa3 | 1394 | cgroup_rstat_flush(memcg->css.cgroup); |
c8713d0b | 1395 | |
5f9a4f4a MS |
1396 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { |
1397 | u64 size; | |
c8713d0b | 1398 | |
fff66b79 | 1399 | size = memcg_page_state_output(memcg, memory_stats[i].idx); |
5f9a4f4a | 1400 | seq_buf_printf(&s, "%s %llu\n", memory_stats[i].name, size); |
c8713d0b | 1401 | |
5f9a4f4a | 1402 | if (unlikely(memory_stats[i].idx == NR_SLAB_UNRECLAIMABLE_B)) { |
fff66b79 MS |
1403 | size += memcg_page_state_output(memcg, |
1404 | NR_SLAB_RECLAIMABLE_B); | |
5f9a4f4a MS |
1405 | seq_buf_printf(&s, "slab %llu\n", size); |
1406 | } | |
1407 | } | |
c8713d0b JW |
1408 | |
1409 | /* Accumulated memory events */ | |
1410 | ||
ebc5d83d KK |
1411 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGFAULT), |
1412 | memcg_events(memcg, PGFAULT)); | |
1413 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGMAJFAULT), | |
1414 | memcg_events(memcg, PGMAJFAULT)); | |
ebc5d83d KK |
1415 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGREFILL), |
1416 | memcg_events(memcg, PGREFILL)); | |
c8713d0b JW |
1417 | seq_buf_printf(&s, "pgscan %lu\n", |
1418 | memcg_events(memcg, PGSCAN_KSWAPD) + | |
1419 | memcg_events(memcg, PGSCAN_DIRECT)); | |
1420 | seq_buf_printf(&s, "pgsteal %lu\n", | |
1421 | memcg_events(memcg, PGSTEAL_KSWAPD) + | |
1422 | memcg_events(memcg, PGSTEAL_DIRECT)); | |
ebc5d83d KK |
1423 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGACTIVATE), |
1424 | memcg_events(memcg, PGACTIVATE)); | |
1425 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGDEACTIVATE), | |
1426 | memcg_events(memcg, PGDEACTIVATE)); | |
1427 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREE), | |
1428 | memcg_events(memcg, PGLAZYFREE)); | |
1429 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREED), | |
1430 | memcg_events(memcg, PGLAZYFREED)); | |
c8713d0b JW |
1431 | |
1432 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
ebc5d83d | 1433 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_FAULT_ALLOC), |
c8713d0b | 1434 | memcg_events(memcg, THP_FAULT_ALLOC)); |
ebc5d83d | 1435 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_COLLAPSE_ALLOC), |
c8713d0b JW |
1436 | memcg_events(memcg, THP_COLLAPSE_ALLOC)); |
1437 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ | |
1438 | ||
1439 | /* The above should easily fit into one page */ | |
1440 | WARN_ON_ONCE(seq_buf_has_overflowed(&s)); | |
1441 | ||
1442 | return s.buffer; | |
1443 | } | |
71cd3113 | 1444 | |
58cf188e | 1445 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1446 | /** |
f0c867d9 | 1447 | * mem_cgroup_print_oom_context: Print OOM information relevant to |
1448 | * memory controller. | |
e222432b BS |
1449 | * @memcg: The memory cgroup that went over limit |
1450 | * @p: Task that is going to be killed | |
1451 | * | |
1452 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1453 | * enabled | |
1454 | */ | |
f0c867d9 | 1455 | void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) |
e222432b | 1456 | { |
e222432b BS |
1457 | rcu_read_lock(); |
1458 | ||
f0c867d9 | 1459 | if (memcg) { |
1460 | pr_cont(",oom_memcg="); | |
1461 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1462 | } else | |
1463 | pr_cont(",global_oom"); | |
2415b9f5 | 1464 | if (p) { |
f0c867d9 | 1465 | pr_cont(",task_memcg="); |
2415b9f5 | 1466 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
2415b9f5 | 1467 | } |
e222432b | 1468 | rcu_read_unlock(); |
f0c867d9 | 1469 | } |
1470 | ||
1471 | /** | |
1472 | * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to | |
1473 | * memory controller. | |
1474 | * @memcg: The memory cgroup that went over limit | |
1475 | */ | |
1476 | void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) | |
1477 | { | |
c8713d0b | 1478 | char *buf; |
e222432b | 1479 | |
3e32cb2e JW |
1480 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1481 | K((u64)page_counter_read(&memcg->memory)), | |
15b42562 | 1482 | K((u64)READ_ONCE(memcg->memory.max)), memcg->memory.failcnt); |
c8713d0b JW |
1483 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
1484 | pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1485 | K((u64)page_counter_read(&memcg->swap)), | |
32d087cd | 1486 | K((u64)READ_ONCE(memcg->swap.max)), memcg->swap.failcnt); |
c8713d0b JW |
1487 | else { |
1488 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1489 | K((u64)page_counter_read(&memcg->memsw)), | |
1490 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); | |
1491 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1492 | K((u64)page_counter_read(&memcg->kmem)), | |
1493 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); | |
58cf188e | 1494 | } |
c8713d0b JW |
1495 | |
1496 | pr_info("Memory cgroup stats for "); | |
1497 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1498 | pr_cont(":"); | |
1499 | buf = memory_stat_format(memcg); | |
1500 | if (!buf) | |
1501 | return; | |
1502 | pr_info("%s", buf); | |
1503 | kfree(buf); | |
e222432b BS |
1504 | } |
1505 | ||
a63d83f4 DR |
1506 | /* |
1507 | * Return the memory (and swap, if configured) limit for a memcg. | |
1508 | */ | |
bbec2e15 | 1509 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1510 | { |
8d387a5f WL |
1511 | unsigned long max = READ_ONCE(memcg->memory.max); |
1512 | ||
1513 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) { | |
1514 | if (mem_cgroup_swappiness(memcg)) | |
1515 | max += min(READ_ONCE(memcg->swap.max), | |
1516 | (unsigned long)total_swap_pages); | |
1517 | } else { /* v1 */ | |
1518 | if (mem_cgroup_swappiness(memcg)) { | |
1519 | /* Calculate swap excess capacity from memsw limit */ | |
1520 | unsigned long swap = READ_ONCE(memcg->memsw.max) - max; | |
1521 | ||
1522 | max += min(swap, (unsigned long)total_swap_pages); | |
1523 | } | |
9a5a8f19 | 1524 | } |
bbec2e15 | 1525 | return max; |
a63d83f4 DR |
1526 | } |
1527 | ||
9783aa99 CD |
1528 | unsigned long mem_cgroup_size(struct mem_cgroup *memcg) |
1529 | { | |
1530 | return page_counter_read(&memcg->memory); | |
1531 | } | |
1532 | ||
b6e6edcf | 1533 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1534 | int order) |
9cbb78bb | 1535 | { |
6e0fc46d DR |
1536 | struct oom_control oc = { |
1537 | .zonelist = NULL, | |
1538 | .nodemask = NULL, | |
2a966b77 | 1539 | .memcg = memcg, |
6e0fc46d DR |
1540 | .gfp_mask = gfp_mask, |
1541 | .order = order, | |
6e0fc46d | 1542 | }; |
1378b37d | 1543 | bool ret = true; |
9cbb78bb | 1544 | |
7775face TH |
1545 | if (mutex_lock_killable(&oom_lock)) |
1546 | return true; | |
1378b37d YS |
1547 | |
1548 | if (mem_cgroup_margin(memcg) >= (1 << order)) | |
1549 | goto unlock; | |
1550 | ||
7775face TH |
1551 | /* |
1552 | * A few threads which were not waiting at mutex_lock_killable() can | |
1553 | * fail to bail out. Therefore, check again after holding oom_lock. | |
1554 | */ | |
1555 | ret = should_force_charge() || out_of_memory(&oc); | |
1378b37d YS |
1556 | |
1557 | unlock: | |
dc56401f | 1558 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1559 | return ret; |
9cbb78bb DR |
1560 | } |
1561 | ||
0608f43d | 1562 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1563 | pg_data_t *pgdat, |
0608f43d AM |
1564 | gfp_t gfp_mask, |
1565 | unsigned long *total_scanned) | |
1566 | { | |
1567 | struct mem_cgroup *victim = NULL; | |
1568 | int total = 0; | |
1569 | int loop = 0; | |
1570 | unsigned long excess; | |
1571 | unsigned long nr_scanned; | |
1572 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1573 | .pgdat = pgdat, |
0608f43d AM |
1574 | }; |
1575 | ||
3e32cb2e | 1576 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1577 | |
1578 | while (1) { | |
1579 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1580 | if (!victim) { | |
1581 | loop++; | |
1582 | if (loop >= 2) { | |
1583 | /* | |
1584 | * If we have not been able to reclaim | |
1585 | * anything, it might because there are | |
1586 | * no reclaimable pages under this hierarchy | |
1587 | */ | |
1588 | if (!total) | |
1589 | break; | |
1590 | /* | |
1591 | * We want to do more targeted reclaim. | |
1592 | * excess >> 2 is not to excessive so as to | |
1593 | * reclaim too much, nor too less that we keep | |
1594 | * coming back to reclaim from this cgroup | |
1595 | */ | |
1596 | if (total >= (excess >> 2) || | |
1597 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1598 | break; | |
1599 | } | |
1600 | continue; | |
1601 | } | |
a9dd0a83 | 1602 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1603 | pgdat, &nr_scanned); |
0608f43d | 1604 | *total_scanned += nr_scanned; |
3e32cb2e | 1605 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1606 | break; |
6d61ef40 | 1607 | } |
0608f43d AM |
1608 | mem_cgroup_iter_break(root_memcg, victim); |
1609 | return total; | |
6d61ef40 BS |
1610 | } |
1611 | ||
0056f4e6 JW |
1612 | #ifdef CONFIG_LOCKDEP |
1613 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1614 | .name = "memcg_oom_lock", | |
1615 | }; | |
1616 | #endif | |
1617 | ||
fb2a6fc5 JW |
1618 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1619 | ||
867578cb KH |
1620 | /* |
1621 | * Check OOM-Killer is already running under our hierarchy. | |
1622 | * If someone is running, return false. | |
1623 | */ | |
fb2a6fc5 | 1624 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1625 | { |
79dfdacc | 1626 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1627 | |
fb2a6fc5 JW |
1628 | spin_lock(&memcg_oom_lock); |
1629 | ||
9f3a0d09 | 1630 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1631 | if (iter->oom_lock) { |
79dfdacc MH |
1632 | /* |
1633 | * this subtree of our hierarchy is already locked | |
1634 | * so we cannot give a lock. | |
1635 | */ | |
79dfdacc | 1636 | failed = iter; |
9f3a0d09 JW |
1637 | mem_cgroup_iter_break(memcg, iter); |
1638 | break; | |
23751be0 JW |
1639 | } else |
1640 | iter->oom_lock = true; | |
7d74b06f | 1641 | } |
867578cb | 1642 | |
fb2a6fc5 JW |
1643 | if (failed) { |
1644 | /* | |
1645 | * OK, we failed to lock the whole subtree so we have | |
1646 | * to clean up what we set up to the failing subtree | |
1647 | */ | |
1648 | for_each_mem_cgroup_tree(iter, memcg) { | |
1649 | if (iter == failed) { | |
1650 | mem_cgroup_iter_break(memcg, iter); | |
1651 | break; | |
1652 | } | |
1653 | iter->oom_lock = false; | |
79dfdacc | 1654 | } |
0056f4e6 JW |
1655 | } else |
1656 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1657 | |
1658 | spin_unlock(&memcg_oom_lock); | |
1659 | ||
1660 | return !failed; | |
a636b327 | 1661 | } |
0b7f569e | 1662 | |
fb2a6fc5 | 1663 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1664 | { |
7d74b06f KH |
1665 | struct mem_cgroup *iter; |
1666 | ||
fb2a6fc5 | 1667 | spin_lock(&memcg_oom_lock); |
5facae4f | 1668 | mutex_release(&memcg_oom_lock_dep_map, _RET_IP_); |
c0ff4b85 | 1669 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1670 | iter->oom_lock = false; |
fb2a6fc5 | 1671 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1672 | } |
1673 | ||
c0ff4b85 | 1674 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1675 | { |
1676 | struct mem_cgroup *iter; | |
1677 | ||
c2b42d3c | 1678 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1679 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1680 | iter->under_oom++; |
1681 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1682 | } |
1683 | ||
c0ff4b85 | 1684 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1685 | { |
1686 | struct mem_cgroup *iter; | |
1687 | ||
867578cb | 1688 | /* |
f0953a1b | 1689 | * Be careful about under_oom underflows because a child memcg |
7a52d4d8 | 1690 | * could have been added after mem_cgroup_mark_under_oom. |
867578cb | 1691 | */ |
c2b42d3c | 1692 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1693 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1694 | if (iter->under_oom > 0) |
1695 | iter->under_oom--; | |
1696 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1697 | } |
1698 | ||
867578cb KH |
1699 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1700 | ||
dc98df5a | 1701 | struct oom_wait_info { |
d79154bb | 1702 | struct mem_cgroup *memcg; |
ac6424b9 | 1703 | wait_queue_entry_t wait; |
dc98df5a KH |
1704 | }; |
1705 | ||
ac6424b9 | 1706 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1707 | unsigned mode, int sync, void *arg) |
1708 | { | |
d79154bb HD |
1709 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1710 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1711 | struct oom_wait_info *oom_wait_info; |
1712 | ||
1713 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1714 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1715 | |
2314b42d JW |
1716 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1717 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1718 | return 0; |
dc98df5a KH |
1719 | return autoremove_wake_function(wait, mode, sync, arg); |
1720 | } | |
1721 | ||
c0ff4b85 | 1722 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1723 | { |
c2b42d3c TH |
1724 | /* |
1725 | * For the following lockless ->under_oom test, the only required | |
1726 | * guarantee is that it must see the state asserted by an OOM when | |
1727 | * this function is called as a result of userland actions | |
1728 | * triggered by the notification of the OOM. This is trivially | |
1729 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1730 | * triggering notification. | |
1731 | */ | |
1732 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1733 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1734 | } |
1735 | ||
29ef680a MH |
1736 | enum oom_status { |
1737 | OOM_SUCCESS, | |
1738 | OOM_FAILED, | |
1739 | OOM_ASYNC, | |
1740 | OOM_SKIPPED | |
1741 | }; | |
1742 | ||
1743 | static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 1744 | { |
7056d3a3 MH |
1745 | enum oom_status ret; |
1746 | bool locked; | |
1747 | ||
29ef680a MH |
1748 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
1749 | return OOM_SKIPPED; | |
1750 | ||
7a1adfdd RG |
1751 | memcg_memory_event(memcg, MEMCG_OOM); |
1752 | ||
867578cb | 1753 | /* |
49426420 JW |
1754 | * We are in the middle of the charge context here, so we |
1755 | * don't want to block when potentially sitting on a callstack | |
1756 | * that holds all kinds of filesystem and mm locks. | |
1757 | * | |
29ef680a MH |
1758 | * cgroup1 allows disabling the OOM killer and waiting for outside |
1759 | * handling until the charge can succeed; remember the context and put | |
1760 | * the task to sleep at the end of the page fault when all locks are | |
1761 | * released. | |
49426420 | 1762 | * |
29ef680a MH |
1763 | * On the other hand, in-kernel OOM killer allows for an async victim |
1764 | * memory reclaim (oom_reaper) and that means that we are not solely | |
1765 | * relying on the oom victim to make a forward progress and we can | |
1766 | * invoke the oom killer here. | |
1767 | * | |
1768 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
1769 | * victim and then we have to bail out from the charge path. | |
867578cb | 1770 | */ |
29ef680a MH |
1771 | if (memcg->oom_kill_disable) { |
1772 | if (!current->in_user_fault) | |
1773 | return OOM_SKIPPED; | |
1774 | css_get(&memcg->css); | |
1775 | current->memcg_in_oom = memcg; | |
1776 | current->memcg_oom_gfp_mask = mask; | |
1777 | current->memcg_oom_order = order; | |
1778 | ||
1779 | return OOM_ASYNC; | |
1780 | } | |
1781 | ||
7056d3a3 MH |
1782 | mem_cgroup_mark_under_oom(memcg); |
1783 | ||
1784 | locked = mem_cgroup_oom_trylock(memcg); | |
1785 | ||
1786 | if (locked) | |
1787 | mem_cgroup_oom_notify(memcg); | |
1788 | ||
1789 | mem_cgroup_unmark_under_oom(memcg); | |
29ef680a | 1790 | if (mem_cgroup_out_of_memory(memcg, mask, order)) |
7056d3a3 MH |
1791 | ret = OOM_SUCCESS; |
1792 | else | |
1793 | ret = OOM_FAILED; | |
1794 | ||
1795 | if (locked) | |
1796 | mem_cgroup_oom_unlock(memcg); | |
29ef680a | 1797 | |
7056d3a3 | 1798 | return ret; |
3812c8c8 JW |
1799 | } |
1800 | ||
1801 | /** | |
1802 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1803 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1804 | * |
49426420 JW |
1805 | * This has to be called at the end of a page fault if the memcg OOM |
1806 | * handler was enabled. | |
3812c8c8 | 1807 | * |
49426420 | 1808 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1809 | * sleep on a waitqueue until the userspace task resolves the |
1810 | * situation. Sleeping directly in the charge context with all kinds | |
1811 | * of locks held is not a good idea, instead we remember an OOM state | |
1812 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1813 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1814 | * |
1815 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1816 | * completed, %false otherwise. |
3812c8c8 | 1817 | */ |
49426420 | 1818 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1819 | { |
626ebc41 | 1820 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1821 | struct oom_wait_info owait; |
49426420 | 1822 | bool locked; |
3812c8c8 JW |
1823 | |
1824 | /* OOM is global, do not handle */ | |
3812c8c8 | 1825 | if (!memcg) |
49426420 | 1826 | return false; |
3812c8c8 | 1827 | |
7c5f64f8 | 1828 | if (!handle) |
49426420 | 1829 | goto cleanup; |
3812c8c8 JW |
1830 | |
1831 | owait.memcg = memcg; | |
1832 | owait.wait.flags = 0; | |
1833 | owait.wait.func = memcg_oom_wake_function; | |
1834 | owait.wait.private = current; | |
2055da97 | 1835 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 1836 | |
3812c8c8 | 1837 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1838 | mem_cgroup_mark_under_oom(memcg); |
1839 | ||
1840 | locked = mem_cgroup_oom_trylock(memcg); | |
1841 | ||
1842 | if (locked) | |
1843 | mem_cgroup_oom_notify(memcg); | |
1844 | ||
1845 | if (locked && !memcg->oom_kill_disable) { | |
1846 | mem_cgroup_unmark_under_oom(memcg); | |
1847 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1848 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1849 | current->memcg_oom_order); | |
49426420 | 1850 | } else { |
3812c8c8 | 1851 | schedule(); |
49426420 JW |
1852 | mem_cgroup_unmark_under_oom(memcg); |
1853 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1854 | } | |
1855 | ||
1856 | if (locked) { | |
fb2a6fc5 JW |
1857 | mem_cgroup_oom_unlock(memcg); |
1858 | /* | |
1859 | * There is no guarantee that an OOM-lock contender | |
1860 | * sees the wakeups triggered by the OOM kill | |
f0953a1b | 1861 | * uncharges. Wake any sleepers explicitly. |
fb2a6fc5 JW |
1862 | */ |
1863 | memcg_oom_recover(memcg); | |
1864 | } | |
49426420 | 1865 | cleanup: |
626ebc41 | 1866 | current->memcg_in_oom = NULL; |
3812c8c8 | 1867 | css_put(&memcg->css); |
867578cb | 1868 | return true; |
0b7f569e KH |
1869 | } |
1870 | ||
3d8b38eb RG |
1871 | /** |
1872 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
1873 | * @victim: task to be killed by the OOM killer | |
1874 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
1875 | * | |
1876 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
1877 | * by killing all belonging OOM-killable tasks. | |
1878 | * | |
1879 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
1880 | */ | |
1881 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
1882 | struct mem_cgroup *oom_domain) | |
1883 | { | |
1884 | struct mem_cgroup *oom_group = NULL; | |
1885 | struct mem_cgroup *memcg; | |
1886 | ||
1887 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
1888 | return NULL; | |
1889 | ||
1890 | if (!oom_domain) | |
1891 | oom_domain = root_mem_cgroup; | |
1892 | ||
1893 | rcu_read_lock(); | |
1894 | ||
1895 | memcg = mem_cgroup_from_task(victim); | |
1896 | if (memcg == root_mem_cgroup) | |
1897 | goto out; | |
1898 | ||
48fe267c RG |
1899 | /* |
1900 | * If the victim task has been asynchronously moved to a different | |
1901 | * memory cgroup, we might end up killing tasks outside oom_domain. | |
1902 | * In this case it's better to ignore memory.group.oom. | |
1903 | */ | |
1904 | if (unlikely(!mem_cgroup_is_descendant(memcg, oom_domain))) | |
1905 | goto out; | |
1906 | ||
3d8b38eb RG |
1907 | /* |
1908 | * Traverse the memory cgroup hierarchy from the victim task's | |
1909 | * cgroup up to the OOMing cgroup (or root) to find the | |
1910 | * highest-level memory cgroup with oom.group set. | |
1911 | */ | |
1912 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
1913 | if (memcg->oom_group) | |
1914 | oom_group = memcg; | |
1915 | ||
1916 | if (memcg == oom_domain) | |
1917 | break; | |
1918 | } | |
1919 | ||
1920 | if (oom_group) | |
1921 | css_get(&oom_group->css); | |
1922 | out: | |
1923 | rcu_read_unlock(); | |
1924 | ||
1925 | return oom_group; | |
1926 | } | |
1927 | ||
1928 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
1929 | { | |
1930 | pr_info("Tasks in "); | |
1931 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1932 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
1933 | } | |
1934 | ||
d7365e78 | 1935 | /** |
bcfe06bf | 1936 | * lock_page_memcg - lock a page and memcg binding |
81f8c3a4 | 1937 | * @page: the page |
32047e2a | 1938 | * |
81f8c3a4 | 1939 | * This function protects unlocked LRU pages from being moved to |
739f79fc JW |
1940 | * another cgroup. |
1941 | * | |
1c824a68 JW |
1942 | * It ensures lifetime of the locked memcg. Caller is responsible |
1943 | * for the lifetime of the page. | |
d69b042f | 1944 | */ |
1c824a68 | 1945 | void lock_page_memcg(struct page *page) |
89c06bd5 | 1946 | { |
9da7b521 | 1947 | struct page *head = compound_head(page); /* rmap on tail pages */ |
89c06bd5 | 1948 | struct mem_cgroup *memcg; |
6de22619 | 1949 | unsigned long flags; |
89c06bd5 | 1950 | |
6de22619 JW |
1951 | /* |
1952 | * The RCU lock is held throughout the transaction. The fast | |
1953 | * path can get away without acquiring the memcg->move_lock | |
1954 | * because page moving starts with an RCU grace period. | |
739f79fc | 1955 | */ |
d7365e78 JW |
1956 | rcu_read_lock(); |
1957 | ||
1958 | if (mem_cgroup_disabled()) | |
1c824a68 | 1959 | return; |
89c06bd5 | 1960 | again: |
bcfe06bf | 1961 | memcg = page_memcg(head); |
29833315 | 1962 | if (unlikely(!memcg)) |
1c824a68 | 1963 | return; |
d7365e78 | 1964 | |
20ad50d6 AS |
1965 | #ifdef CONFIG_PROVE_LOCKING |
1966 | local_irq_save(flags); | |
1967 | might_lock(&memcg->move_lock); | |
1968 | local_irq_restore(flags); | |
1969 | #endif | |
1970 | ||
bdcbb659 | 1971 | if (atomic_read(&memcg->moving_account) <= 0) |
1c824a68 | 1972 | return; |
89c06bd5 | 1973 | |
6de22619 | 1974 | spin_lock_irqsave(&memcg->move_lock, flags); |
bcfe06bf | 1975 | if (memcg != page_memcg(head)) { |
6de22619 | 1976 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
1977 | goto again; |
1978 | } | |
6de22619 JW |
1979 | |
1980 | /* | |
1c824a68 JW |
1981 | * When charge migration first begins, we can have multiple |
1982 | * critical sections holding the fast-path RCU lock and one | |
1983 | * holding the slowpath move_lock. Track the task who has the | |
1984 | * move_lock for unlock_page_memcg(). | |
6de22619 JW |
1985 | */ |
1986 | memcg->move_lock_task = current; | |
1987 | memcg->move_lock_flags = flags; | |
89c06bd5 | 1988 | } |
81f8c3a4 | 1989 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 1990 | |
1c824a68 | 1991 | static void __unlock_page_memcg(struct mem_cgroup *memcg) |
89c06bd5 | 1992 | { |
6de22619 JW |
1993 | if (memcg && memcg->move_lock_task == current) { |
1994 | unsigned long flags = memcg->move_lock_flags; | |
1995 | ||
1996 | memcg->move_lock_task = NULL; | |
1997 | memcg->move_lock_flags = 0; | |
1998 | ||
1999 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2000 | } | |
89c06bd5 | 2001 | |
d7365e78 | 2002 | rcu_read_unlock(); |
89c06bd5 | 2003 | } |
739f79fc JW |
2004 | |
2005 | /** | |
bcfe06bf | 2006 | * unlock_page_memcg - unlock a page and memcg binding |
739f79fc JW |
2007 | * @page: the page |
2008 | */ | |
2009 | void unlock_page_memcg(struct page *page) | |
2010 | { | |
9da7b521 JW |
2011 | struct page *head = compound_head(page); |
2012 | ||
bcfe06bf | 2013 | __unlock_page_memcg(page_memcg(head)); |
739f79fc | 2014 | } |
81f8c3a4 | 2015 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 2016 | |
55927114 | 2017 | struct obj_stock { |
bf4f0599 RG |
2018 | #ifdef CONFIG_MEMCG_KMEM |
2019 | struct obj_cgroup *cached_objcg; | |
68ac5b3c | 2020 | struct pglist_data *cached_pgdat; |
bf4f0599 | 2021 | unsigned int nr_bytes; |
68ac5b3c WL |
2022 | int nr_slab_reclaimable_b; |
2023 | int nr_slab_unreclaimable_b; | |
55927114 WL |
2024 | #else |
2025 | int dummy[0]; | |
bf4f0599 | 2026 | #endif |
55927114 WL |
2027 | }; |
2028 | ||
2029 | struct memcg_stock_pcp { | |
2030 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
2031 | unsigned int nr_pages; | |
2032 | struct obj_stock task_obj; | |
2033 | struct obj_stock irq_obj; | |
bf4f0599 | 2034 | |
cdec2e42 | 2035 | struct work_struct work; |
26fe6168 | 2036 | unsigned long flags; |
a0db00fc | 2037 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2038 | }; |
2039 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2040 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2041 | |
bf4f0599 | 2042 | #ifdef CONFIG_MEMCG_KMEM |
55927114 | 2043 | static void drain_obj_stock(struct obj_stock *stock); |
bf4f0599 RG |
2044 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, |
2045 | struct mem_cgroup *root_memcg); | |
2046 | ||
2047 | #else | |
55927114 | 2048 | static inline void drain_obj_stock(struct obj_stock *stock) |
bf4f0599 RG |
2049 | { |
2050 | } | |
2051 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2052 | struct mem_cgroup *root_memcg) | |
2053 | { | |
2054 | return false; | |
2055 | } | |
2056 | #endif | |
2057 | ||
55927114 WL |
2058 | /* |
2059 | * Most kmem_cache_alloc() calls are from user context. The irq disable/enable | |
2060 | * sequence used in this case to access content from object stock is slow. | |
2061 | * To optimize for user context access, there are now two object stocks for | |
2062 | * task context and interrupt context access respectively. | |
2063 | * | |
2064 | * The task context object stock can be accessed by disabling preemption only | |
2065 | * which is cheap in non-preempt kernel. The interrupt context object stock | |
2066 | * can only be accessed after disabling interrupt. User context code can | |
2067 | * access interrupt object stock, but not vice versa. | |
2068 | */ | |
2069 | static inline struct obj_stock *get_obj_stock(unsigned long *pflags) | |
2070 | { | |
2071 | struct memcg_stock_pcp *stock; | |
2072 | ||
2073 | if (likely(in_task())) { | |
2074 | *pflags = 0UL; | |
2075 | preempt_disable(); | |
2076 | stock = this_cpu_ptr(&memcg_stock); | |
2077 | return &stock->task_obj; | |
2078 | } | |
2079 | ||
2080 | local_irq_save(*pflags); | |
2081 | stock = this_cpu_ptr(&memcg_stock); | |
2082 | return &stock->irq_obj; | |
2083 | } | |
2084 | ||
2085 | static inline void put_obj_stock(unsigned long flags) | |
2086 | { | |
2087 | if (likely(in_task())) | |
2088 | preempt_enable(); | |
2089 | else | |
2090 | local_irq_restore(flags); | |
2091 | } | |
2092 | ||
a0956d54 SS |
2093 | /** |
2094 | * consume_stock: Try to consume stocked charge on this cpu. | |
2095 | * @memcg: memcg to consume from. | |
2096 | * @nr_pages: how many pages to charge. | |
2097 | * | |
2098 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2099 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2100 | * service an allocation will refill the stock. | |
2101 | * | |
2102 | * returns true if successful, false otherwise. | |
cdec2e42 | 2103 | */ |
a0956d54 | 2104 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2105 | { |
2106 | struct memcg_stock_pcp *stock; | |
db2ba40c | 2107 | unsigned long flags; |
3e32cb2e | 2108 | bool ret = false; |
cdec2e42 | 2109 | |
a983b5eb | 2110 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 2111 | return ret; |
a0956d54 | 2112 | |
db2ba40c JW |
2113 | local_irq_save(flags); |
2114 | ||
2115 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 2116 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2117 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2118 | ret = true; |
2119 | } | |
db2ba40c JW |
2120 | |
2121 | local_irq_restore(flags); | |
2122 | ||
cdec2e42 KH |
2123 | return ret; |
2124 | } | |
2125 | ||
2126 | /* | |
3e32cb2e | 2127 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2128 | */ |
2129 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2130 | { | |
2131 | struct mem_cgroup *old = stock->cached; | |
2132 | ||
1a3e1f40 JW |
2133 | if (!old) |
2134 | return; | |
2135 | ||
11c9ea4e | 2136 | if (stock->nr_pages) { |
3e32cb2e | 2137 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 2138 | if (do_memsw_account()) |
3e32cb2e | 2139 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
11c9ea4e | 2140 | stock->nr_pages = 0; |
cdec2e42 | 2141 | } |
1a3e1f40 JW |
2142 | |
2143 | css_put(&old->css); | |
cdec2e42 | 2144 | stock->cached = NULL; |
cdec2e42 KH |
2145 | } |
2146 | ||
cdec2e42 KH |
2147 | static void drain_local_stock(struct work_struct *dummy) |
2148 | { | |
db2ba40c JW |
2149 | struct memcg_stock_pcp *stock; |
2150 | unsigned long flags; | |
2151 | ||
72f0184c | 2152 | /* |
5c49cf9a MH |
2153 | * The only protection from cpu hotplug (memcg_hotplug_cpu_dead) vs. |
2154 | * drain_stock races is that we always operate on local CPU stock | |
2155 | * here with IRQ disabled | |
72f0184c | 2156 | */ |
db2ba40c JW |
2157 | local_irq_save(flags); |
2158 | ||
2159 | stock = this_cpu_ptr(&memcg_stock); | |
55927114 WL |
2160 | drain_obj_stock(&stock->irq_obj); |
2161 | if (in_task()) | |
2162 | drain_obj_stock(&stock->task_obj); | |
cdec2e42 | 2163 | drain_stock(stock); |
26fe6168 | 2164 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c JW |
2165 | |
2166 | local_irq_restore(flags); | |
cdec2e42 KH |
2167 | } |
2168 | ||
2169 | /* | |
3e32cb2e | 2170 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2171 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2172 | */ |
c0ff4b85 | 2173 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2174 | { |
db2ba40c JW |
2175 | struct memcg_stock_pcp *stock; |
2176 | unsigned long flags; | |
2177 | ||
2178 | local_irq_save(flags); | |
cdec2e42 | 2179 | |
db2ba40c | 2180 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 2181 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2182 | drain_stock(stock); |
1a3e1f40 | 2183 | css_get(&memcg->css); |
c0ff4b85 | 2184 | stock->cached = memcg; |
cdec2e42 | 2185 | } |
11c9ea4e | 2186 | stock->nr_pages += nr_pages; |
db2ba40c | 2187 | |
a983b5eb | 2188 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 RG |
2189 | drain_stock(stock); |
2190 | ||
db2ba40c | 2191 | local_irq_restore(flags); |
cdec2e42 KH |
2192 | } |
2193 | ||
2194 | /* | |
c0ff4b85 | 2195 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2196 | * of the hierarchy under it. |
cdec2e42 | 2197 | */ |
6d3d6aa2 | 2198 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2199 | { |
26fe6168 | 2200 | int cpu, curcpu; |
d38144b7 | 2201 | |
6d3d6aa2 JW |
2202 | /* If someone's already draining, avoid adding running more workers. */ |
2203 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2204 | return; | |
72f0184c MH |
2205 | /* |
2206 | * Notify other cpus that system-wide "drain" is running | |
2207 | * We do not care about races with the cpu hotplug because cpu down | |
2208 | * as well as workers from this path always operate on the local | |
2209 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2210 | */ | |
5af12d0e | 2211 | curcpu = get_cpu(); |
cdec2e42 KH |
2212 | for_each_online_cpu(cpu) { |
2213 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2214 | struct mem_cgroup *memcg; |
e1a366be | 2215 | bool flush = false; |
26fe6168 | 2216 | |
e1a366be | 2217 | rcu_read_lock(); |
c0ff4b85 | 2218 | memcg = stock->cached; |
e1a366be RG |
2219 | if (memcg && stock->nr_pages && |
2220 | mem_cgroup_is_descendant(memcg, root_memcg)) | |
2221 | flush = true; | |
27fb0956 | 2222 | else if (obj_stock_flush_required(stock, root_memcg)) |
bf4f0599 | 2223 | flush = true; |
e1a366be RG |
2224 | rcu_read_unlock(); |
2225 | ||
2226 | if (flush && | |
2227 | !test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { | |
d1a05b69 MH |
2228 | if (cpu == curcpu) |
2229 | drain_local_stock(&stock->work); | |
2230 | else | |
2231 | schedule_work_on(cpu, &stock->work); | |
2232 | } | |
cdec2e42 | 2233 | } |
5af12d0e | 2234 | put_cpu(); |
9f50fad6 | 2235 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2236 | } |
2237 | ||
2cd21c89 JW |
2238 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
2239 | { | |
2240 | struct memcg_stock_pcp *stock; | |
a3d4c05a | 2241 | |
2cd21c89 JW |
2242 | stock = &per_cpu(memcg_stock, cpu); |
2243 | drain_stock(stock); | |
a3d4c05a | 2244 | |
308167fc | 2245 | return 0; |
cdec2e42 KH |
2246 | } |
2247 | ||
b3ff9291 CD |
2248 | static unsigned long reclaim_high(struct mem_cgroup *memcg, |
2249 | unsigned int nr_pages, | |
2250 | gfp_t gfp_mask) | |
f7e1cb6e | 2251 | { |
b3ff9291 CD |
2252 | unsigned long nr_reclaimed = 0; |
2253 | ||
f7e1cb6e | 2254 | do { |
e22c6ed9 JW |
2255 | unsigned long pflags; |
2256 | ||
d1663a90 JK |
2257 | if (page_counter_read(&memcg->memory) <= |
2258 | READ_ONCE(memcg->memory.high)) | |
f7e1cb6e | 2259 | continue; |
e22c6ed9 | 2260 | |
e27be240 | 2261 | memcg_memory_event(memcg, MEMCG_HIGH); |
e22c6ed9 JW |
2262 | |
2263 | psi_memstall_enter(&pflags); | |
b3ff9291 CD |
2264 | nr_reclaimed += try_to_free_mem_cgroup_pages(memcg, nr_pages, |
2265 | gfp_mask, true); | |
e22c6ed9 | 2266 | psi_memstall_leave(&pflags); |
4bf17307 CD |
2267 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2268 | !mem_cgroup_is_root(memcg)); | |
b3ff9291 CD |
2269 | |
2270 | return nr_reclaimed; | |
f7e1cb6e JW |
2271 | } |
2272 | ||
2273 | static void high_work_func(struct work_struct *work) | |
2274 | { | |
2275 | struct mem_cgroup *memcg; | |
2276 | ||
2277 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2278 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2279 | } |
2280 | ||
0e4b01df CD |
2281 | /* |
2282 | * Clamp the maximum sleep time per allocation batch to 2 seconds. This is | |
2283 | * enough to still cause a significant slowdown in most cases, while still | |
2284 | * allowing diagnostics and tracing to proceed without becoming stuck. | |
2285 | */ | |
2286 | #define MEMCG_MAX_HIGH_DELAY_JIFFIES (2UL*HZ) | |
2287 | ||
2288 | /* | |
2289 | * When calculating the delay, we use these either side of the exponentiation to | |
2290 | * maintain precision and scale to a reasonable number of jiffies (see the table | |
2291 | * below. | |
2292 | * | |
2293 | * - MEMCG_DELAY_PRECISION_SHIFT: Extra precision bits while translating the | |
2294 | * overage ratio to a delay. | |
ac5ddd0f | 2295 | * - MEMCG_DELAY_SCALING_SHIFT: The number of bits to scale down the |
0e4b01df CD |
2296 | * proposed penalty in order to reduce to a reasonable number of jiffies, and |
2297 | * to produce a reasonable delay curve. | |
2298 | * | |
2299 | * MEMCG_DELAY_SCALING_SHIFT just happens to be a number that produces a | |
2300 | * reasonable delay curve compared to precision-adjusted overage, not | |
2301 | * penalising heavily at first, but still making sure that growth beyond the | |
2302 | * limit penalises misbehaviour cgroups by slowing them down exponentially. For | |
2303 | * example, with a high of 100 megabytes: | |
2304 | * | |
2305 | * +-------+------------------------+ | |
2306 | * | usage | time to allocate in ms | | |
2307 | * +-------+------------------------+ | |
2308 | * | 100M | 0 | | |
2309 | * | 101M | 6 | | |
2310 | * | 102M | 25 | | |
2311 | * | 103M | 57 | | |
2312 | * | 104M | 102 | | |
2313 | * | 105M | 159 | | |
2314 | * | 106M | 230 | | |
2315 | * | 107M | 313 | | |
2316 | * | 108M | 409 | | |
2317 | * | 109M | 518 | | |
2318 | * | 110M | 639 | | |
2319 | * | 111M | 774 | | |
2320 | * | 112M | 921 | | |
2321 | * | 113M | 1081 | | |
2322 | * | 114M | 1254 | | |
2323 | * | 115M | 1439 | | |
2324 | * | 116M | 1638 | | |
2325 | * | 117M | 1849 | | |
2326 | * | 118M | 2000 | | |
2327 | * | 119M | 2000 | | |
2328 | * | 120M | 2000 | | |
2329 | * +-------+------------------------+ | |
2330 | */ | |
2331 | #define MEMCG_DELAY_PRECISION_SHIFT 20 | |
2332 | #define MEMCG_DELAY_SCALING_SHIFT 14 | |
2333 | ||
8a5dbc65 | 2334 | static u64 calculate_overage(unsigned long usage, unsigned long high) |
b23afb93 | 2335 | { |
8a5dbc65 | 2336 | u64 overage; |
b23afb93 | 2337 | |
8a5dbc65 JK |
2338 | if (usage <= high) |
2339 | return 0; | |
e26733e0 | 2340 | |
8a5dbc65 JK |
2341 | /* |
2342 | * Prevent division by 0 in overage calculation by acting as if | |
2343 | * it was a threshold of 1 page | |
2344 | */ | |
2345 | high = max(high, 1UL); | |
9b8b1754 | 2346 | |
8a5dbc65 JK |
2347 | overage = usage - high; |
2348 | overage <<= MEMCG_DELAY_PRECISION_SHIFT; | |
2349 | return div64_u64(overage, high); | |
2350 | } | |
e26733e0 | 2351 | |
8a5dbc65 JK |
2352 | static u64 mem_find_max_overage(struct mem_cgroup *memcg) |
2353 | { | |
2354 | u64 overage, max_overage = 0; | |
e26733e0 | 2355 | |
8a5dbc65 JK |
2356 | do { |
2357 | overage = calculate_overage(page_counter_read(&memcg->memory), | |
d1663a90 | 2358 | READ_ONCE(memcg->memory.high)); |
8a5dbc65 | 2359 | max_overage = max(overage, max_overage); |
e26733e0 CD |
2360 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2361 | !mem_cgroup_is_root(memcg)); | |
2362 | ||
8a5dbc65 JK |
2363 | return max_overage; |
2364 | } | |
2365 | ||
4b82ab4f JK |
2366 | static u64 swap_find_max_overage(struct mem_cgroup *memcg) |
2367 | { | |
2368 | u64 overage, max_overage = 0; | |
2369 | ||
2370 | do { | |
2371 | overage = calculate_overage(page_counter_read(&memcg->swap), | |
2372 | READ_ONCE(memcg->swap.high)); | |
2373 | if (overage) | |
2374 | memcg_memory_event(memcg, MEMCG_SWAP_HIGH); | |
2375 | max_overage = max(overage, max_overage); | |
2376 | } while ((memcg = parent_mem_cgroup(memcg)) && | |
2377 | !mem_cgroup_is_root(memcg)); | |
2378 | ||
2379 | return max_overage; | |
2380 | } | |
2381 | ||
8a5dbc65 JK |
2382 | /* |
2383 | * Get the number of jiffies that we should penalise a mischievous cgroup which | |
2384 | * is exceeding its memory.high by checking both it and its ancestors. | |
2385 | */ | |
2386 | static unsigned long calculate_high_delay(struct mem_cgroup *memcg, | |
2387 | unsigned int nr_pages, | |
2388 | u64 max_overage) | |
2389 | { | |
2390 | unsigned long penalty_jiffies; | |
2391 | ||
e26733e0 CD |
2392 | if (!max_overage) |
2393 | return 0; | |
0e4b01df CD |
2394 | |
2395 | /* | |
0e4b01df CD |
2396 | * We use overage compared to memory.high to calculate the number of |
2397 | * jiffies to sleep (penalty_jiffies). Ideally this value should be | |
2398 | * fairly lenient on small overages, and increasingly harsh when the | |
2399 | * memcg in question makes it clear that it has no intention of stopping | |
2400 | * its crazy behaviour, so we exponentially increase the delay based on | |
2401 | * overage amount. | |
2402 | */ | |
e26733e0 CD |
2403 | penalty_jiffies = max_overage * max_overage * HZ; |
2404 | penalty_jiffies >>= MEMCG_DELAY_PRECISION_SHIFT; | |
2405 | penalty_jiffies >>= MEMCG_DELAY_SCALING_SHIFT; | |
0e4b01df CD |
2406 | |
2407 | /* | |
2408 | * Factor in the task's own contribution to the overage, such that four | |
2409 | * N-sized allocations are throttled approximately the same as one | |
2410 | * 4N-sized allocation. | |
2411 | * | |
2412 | * MEMCG_CHARGE_BATCH pages is nominal, so work out how much smaller or | |
2413 | * larger the current charge patch is than that. | |
2414 | */ | |
ff144e69 | 2415 | return penalty_jiffies * nr_pages / MEMCG_CHARGE_BATCH; |
e26733e0 CD |
2416 | } |
2417 | ||
2418 | /* | |
2419 | * Scheduled by try_charge() to be executed from the userland return path | |
2420 | * and reclaims memory over the high limit. | |
2421 | */ | |
2422 | void mem_cgroup_handle_over_high(void) | |
2423 | { | |
2424 | unsigned long penalty_jiffies; | |
2425 | unsigned long pflags; | |
b3ff9291 | 2426 | unsigned long nr_reclaimed; |
e26733e0 | 2427 | unsigned int nr_pages = current->memcg_nr_pages_over_high; |
d977aa93 | 2428 | int nr_retries = MAX_RECLAIM_RETRIES; |
e26733e0 | 2429 | struct mem_cgroup *memcg; |
b3ff9291 | 2430 | bool in_retry = false; |
e26733e0 CD |
2431 | |
2432 | if (likely(!nr_pages)) | |
2433 | return; | |
2434 | ||
2435 | memcg = get_mem_cgroup_from_mm(current->mm); | |
e26733e0 CD |
2436 | current->memcg_nr_pages_over_high = 0; |
2437 | ||
b3ff9291 CD |
2438 | retry_reclaim: |
2439 | /* | |
2440 | * The allocating task should reclaim at least the batch size, but for | |
2441 | * subsequent retries we only want to do what's necessary to prevent oom | |
2442 | * or breaching resource isolation. | |
2443 | * | |
2444 | * This is distinct from memory.max or page allocator behaviour because | |
2445 | * memory.high is currently batched, whereas memory.max and the page | |
2446 | * allocator run every time an allocation is made. | |
2447 | */ | |
2448 | nr_reclaimed = reclaim_high(memcg, | |
2449 | in_retry ? SWAP_CLUSTER_MAX : nr_pages, | |
2450 | GFP_KERNEL); | |
2451 | ||
e26733e0 CD |
2452 | /* |
2453 | * memory.high is breached and reclaim is unable to keep up. Throttle | |
2454 | * allocators proactively to slow down excessive growth. | |
2455 | */ | |
8a5dbc65 JK |
2456 | penalty_jiffies = calculate_high_delay(memcg, nr_pages, |
2457 | mem_find_max_overage(memcg)); | |
0e4b01df | 2458 | |
4b82ab4f JK |
2459 | penalty_jiffies += calculate_high_delay(memcg, nr_pages, |
2460 | swap_find_max_overage(memcg)); | |
2461 | ||
ff144e69 JK |
2462 | /* |
2463 | * Clamp the max delay per usermode return so as to still keep the | |
2464 | * application moving forwards and also permit diagnostics, albeit | |
2465 | * extremely slowly. | |
2466 | */ | |
2467 | penalty_jiffies = min(penalty_jiffies, MEMCG_MAX_HIGH_DELAY_JIFFIES); | |
2468 | ||
0e4b01df CD |
2469 | /* |
2470 | * Don't sleep if the amount of jiffies this memcg owes us is so low | |
2471 | * that it's not even worth doing, in an attempt to be nice to those who | |
2472 | * go only a small amount over their memory.high value and maybe haven't | |
2473 | * been aggressively reclaimed enough yet. | |
2474 | */ | |
2475 | if (penalty_jiffies <= HZ / 100) | |
2476 | goto out; | |
2477 | ||
b3ff9291 CD |
2478 | /* |
2479 | * If reclaim is making forward progress but we're still over | |
2480 | * memory.high, we want to encourage that rather than doing allocator | |
2481 | * throttling. | |
2482 | */ | |
2483 | if (nr_reclaimed || nr_retries--) { | |
2484 | in_retry = true; | |
2485 | goto retry_reclaim; | |
2486 | } | |
2487 | ||
0e4b01df CD |
2488 | /* |
2489 | * If we exit early, we're guaranteed to die (since | |
2490 | * schedule_timeout_killable sets TASK_KILLABLE). This means we don't | |
2491 | * need to account for any ill-begotten jiffies to pay them off later. | |
2492 | */ | |
2493 | psi_memstall_enter(&pflags); | |
2494 | schedule_timeout_killable(penalty_jiffies); | |
2495 | psi_memstall_leave(&pflags); | |
2496 | ||
2497 | out: | |
2498 | css_put(&memcg->css); | |
b23afb93 TH |
2499 | } |
2500 | ||
c5c8b16b MS |
2501 | static int try_charge_memcg(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2502 | unsigned int nr_pages) | |
8a9f3ccd | 2503 | { |
a983b5eb | 2504 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
d977aa93 | 2505 | int nr_retries = MAX_RECLAIM_RETRIES; |
6539cc05 | 2506 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2507 | struct page_counter *counter; |
e22c6ed9 | 2508 | enum oom_status oom_status; |
6539cc05 | 2509 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2510 | bool may_swap = true; |
2511 | bool drained = false; | |
e22c6ed9 | 2512 | unsigned long pflags; |
a636b327 | 2513 | |
6539cc05 | 2514 | retry: |
b6b6cc72 | 2515 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2516 | return 0; |
8a9f3ccd | 2517 | |
7941d214 | 2518 | if (!do_memsw_account() || |
6071ca52 JW |
2519 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2520 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2521 | goto done_restock; |
7941d214 | 2522 | if (do_memsw_account()) |
3e32cb2e JW |
2523 | page_counter_uncharge(&memcg->memsw, batch); |
2524 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2525 | } else { |
3e32cb2e | 2526 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2527 | may_swap = false; |
3fbe7244 | 2528 | } |
7a81b88c | 2529 | |
6539cc05 JW |
2530 | if (batch > nr_pages) { |
2531 | batch = nr_pages; | |
2532 | goto retry; | |
2533 | } | |
6d61ef40 | 2534 | |
869712fd JW |
2535 | /* |
2536 | * Memcg doesn't have a dedicated reserve for atomic | |
2537 | * allocations. But like the global atomic pool, we need to | |
2538 | * put the burden of reclaim on regular allocation requests | |
2539 | * and let these go through as privileged allocations. | |
2540 | */ | |
2541 | if (gfp_mask & __GFP_ATOMIC) | |
2542 | goto force; | |
2543 | ||
06b078fc JW |
2544 | /* |
2545 | * Unlike in global OOM situations, memcg is not in a physical | |
2546 | * memory shortage. Allow dying and OOM-killed tasks to | |
2547 | * bypass the last charges so that they can exit quickly and | |
2548 | * free their memory. | |
2549 | */ | |
7775face | 2550 | if (unlikely(should_force_charge())) |
10d53c74 | 2551 | goto force; |
06b078fc | 2552 | |
89a28483 JW |
2553 | /* |
2554 | * Prevent unbounded recursion when reclaim operations need to | |
2555 | * allocate memory. This might exceed the limits temporarily, | |
2556 | * but we prefer facilitating memory reclaim and getting back | |
2557 | * under the limit over triggering OOM kills in these cases. | |
2558 | */ | |
2559 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2560 | goto force; | |
2561 | ||
06b078fc JW |
2562 | if (unlikely(task_in_memcg_oom(current))) |
2563 | goto nomem; | |
2564 | ||
d0164adc | 2565 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2566 | goto nomem; |
4b534334 | 2567 | |
e27be240 | 2568 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
241994ed | 2569 | |
e22c6ed9 | 2570 | psi_memstall_enter(&pflags); |
b70a2a21 JW |
2571 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2572 | gfp_mask, may_swap); | |
e22c6ed9 | 2573 | psi_memstall_leave(&pflags); |
6539cc05 | 2574 | |
61e02c74 | 2575 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2576 | goto retry; |
28c34c29 | 2577 | |
b70a2a21 | 2578 | if (!drained) { |
6d3d6aa2 | 2579 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2580 | drained = true; |
2581 | goto retry; | |
2582 | } | |
2583 | ||
28c34c29 JW |
2584 | if (gfp_mask & __GFP_NORETRY) |
2585 | goto nomem; | |
6539cc05 JW |
2586 | /* |
2587 | * Even though the limit is exceeded at this point, reclaim | |
2588 | * may have been able to free some pages. Retry the charge | |
2589 | * before killing the task. | |
2590 | * | |
2591 | * Only for regular pages, though: huge pages are rather | |
2592 | * unlikely to succeed so close to the limit, and we fall back | |
2593 | * to regular pages anyway in case of failure. | |
2594 | */ | |
61e02c74 | 2595 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2596 | goto retry; |
2597 | /* | |
2598 | * At task move, charge accounts can be doubly counted. So, it's | |
2599 | * better to wait until the end of task_move if something is going on. | |
2600 | */ | |
2601 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2602 | goto retry; | |
2603 | ||
9b130619 JW |
2604 | if (nr_retries--) |
2605 | goto retry; | |
2606 | ||
38d38493 | 2607 | if (gfp_mask & __GFP_RETRY_MAYFAIL) |
29ef680a MH |
2608 | goto nomem; |
2609 | ||
6539cc05 | 2610 | if (fatal_signal_pending(current)) |
10d53c74 | 2611 | goto force; |
6539cc05 | 2612 | |
29ef680a MH |
2613 | /* |
2614 | * keep retrying as long as the memcg oom killer is able to make | |
2615 | * a forward progress or bypass the charge if the oom killer | |
2616 | * couldn't make any progress. | |
2617 | */ | |
2618 | oom_status = mem_cgroup_oom(mem_over_limit, gfp_mask, | |
3608de07 | 2619 | get_order(nr_pages * PAGE_SIZE)); |
29ef680a MH |
2620 | switch (oom_status) { |
2621 | case OOM_SUCCESS: | |
d977aa93 | 2622 | nr_retries = MAX_RECLAIM_RETRIES; |
29ef680a MH |
2623 | goto retry; |
2624 | case OOM_FAILED: | |
2625 | goto force; | |
2626 | default: | |
2627 | goto nomem; | |
2628 | } | |
7a81b88c | 2629 | nomem: |
6d1fdc48 | 2630 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2631 | return -ENOMEM; |
10d53c74 TH |
2632 | force: |
2633 | /* | |
2634 | * The allocation either can't fail or will lead to more memory | |
2635 | * being freed very soon. Allow memory usage go over the limit | |
2636 | * temporarily by force charging it. | |
2637 | */ | |
2638 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2639 | if (do_memsw_account()) |
10d53c74 | 2640 | page_counter_charge(&memcg->memsw, nr_pages); |
10d53c74 TH |
2641 | |
2642 | return 0; | |
6539cc05 JW |
2643 | |
2644 | done_restock: | |
2645 | if (batch > nr_pages) | |
2646 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2647 | |
241994ed | 2648 | /* |
b23afb93 TH |
2649 | * If the hierarchy is above the normal consumption range, schedule |
2650 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2651 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2652 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2653 | * not recorded as it most likely matches current's and won't | |
2654 | * change in the meantime. As high limit is checked again before | |
2655 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2656 | */ |
2657 | do { | |
4b82ab4f JK |
2658 | bool mem_high, swap_high; |
2659 | ||
2660 | mem_high = page_counter_read(&memcg->memory) > | |
2661 | READ_ONCE(memcg->memory.high); | |
2662 | swap_high = page_counter_read(&memcg->swap) > | |
2663 | READ_ONCE(memcg->swap.high); | |
2664 | ||
2665 | /* Don't bother a random interrupted task */ | |
2666 | if (in_interrupt()) { | |
2667 | if (mem_high) { | |
f7e1cb6e JW |
2668 | schedule_work(&memcg->high_work); |
2669 | break; | |
2670 | } | |
4b82ab4f JK |
2671 | continue; |
2672 | } | |
2673 | ||
2674 | if (mem_high || swap_high) { | |
2675 | /* | |
2676 | * The allocating tasks in this cgroup will need to do | |
2677 | * reclaim or be throttled to prevent further growth | |
2678 | * of the memory or swap footprints. | |
2679 | * | |
2680 | * Target some best-effort fairness between the tasks, | |
2681 | * and distribute reclaim work and delay penalties | |
2682 | * based on how much each task is actually allocating. | |
2683 | */ | |
9516a18a | 2684 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2685 | set_notify_resume(current); |
2686 | break; | |
2687 | } | |
241994ed | 2688 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2689 | |
2690 | return 0; | |
7a81b88c | 2691 | } |
8a9f3ccd | 2692 | |
c5c8b16b MS |
2693 | static inline int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2694 | unsigned int nr_pages) | |
2695 | { | |
2696 | if (mem_cgroup_is_root(memcg)) | |
2697 | return 0; | |
2698 | ||
2699 | return try_charge_memcg(memcg, gfp_mask, nr_pages); | |
2700 | } | |
2701 | ||
f0e45fb4 | 2702 | #if defined(CONFIG_MEMCG_KMEM) || defined(CONFIG_MMU) |
00501b53 | 2703 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2704 | { |
ce00a967 JW |
2705 | if (mem_cgroup_is_root(memcg)) |
2706 | return; | |
2707 | ||
3e32cb2e | 2708 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2709 | if (do_memsw_account()) |
3e32cb2e | 2710 | page_counter_uncharge(&memcg->memsw, nr_pages); |
d01dd17f | 2711 | } |
f0e45fb4 | 2712 | #endif |
d01dd17f | 2713 | |
d9eb1ea2 | 2714 | static void commit_charge(struct page *page, struct mem_cgroup *memcg) |
0a31bc97 | 2715 | { |
bcfe06bf | 2716 | VM_BUG_ON_PAGE(page_memcg(page), page); |
0a31bc97 | 2717 | /* |
a5eb011a | 2718 | * Any of the following ensures page's memcg stability: |
0a31bc97 | 2719 | * |
a0b5b414 JW |
2720 | * - the page lock |
2721 | * - LRU isolation | |
2722 | * - lock_page_memcg() | |
2723 | * - exclusive reference | |
0a31bc97 | 2724 | */ |
bcfe06bf | 2725 | page->memcg_data = (unsigned long)memcg; |
7a81b88c | 2726 | } |
66e1707b | 2727 | |
e74d2259 MS |
2728 | static struct mem_cgroup *get_mem_cgroup_from_objcg(struct obj_cgroup *objcg) |
2729 | { | |
2730 | struct mem_cgroup *memcg; | |
2731 | ||
2732 | rcu_read_lock(); | |
2733 | retry: | |
2734 | memcg = obj_cgroup_memcg(objcg); | |
2735 | if (unlikely(!css_tryget(&memcg->css))) | |
2736 | goto retry; | |
2737 | rcu_read_unlock(); | |
2738 | ||
2739 | return memcg; | |
2740 | } | |
2741 | ||
84c07d11 | 2742 | #ifdef CONFIG_MEMCG_KMEM |
41eb5df1 WL |
2743 | /* |
2744 | * The allocated objcg pointers array is not accounted directly. | |
2745 | * Moreover, it should not come from DMA buffer and is not readily | |
2746 | * reclaimable. So those GFP bits should be masked off. | |
2747 | */ | |
2748 | #define OBJCGS_CLEAR_MASK (__GFP_DMA | __GFP_RECLAIMABLE | __GFP_ACCOUNT) | |
2749 | ||
10befea9 | 2750 | int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s, |
2e9bd483 | 2751 | gfp_t gfp, bool new_page) |
10befea9 RG |
2752 | { |
2753 | unsigned int objects = objs_per_slab_page(s, page); | |
2e9bd483 | 2754 | unsigned long memcg_data; |
10befea9 RG |
2755 | void *vec; |
2756 | ||
41eb5df1 | 2757 | gfp &= ~OBJCGS_CLEAR_MASK; |
10befea9 RG |
2758 | vec = kcalloc_node(objects, sizeof(struct obj_cgroup *), gfp, |
2759 | page_to_nid(page)); | |
2760 | if (!vec) | |
2761 | return -ENOMEM; | |
2762 | ||
2e9bd483 RG |
2763 | memcg_data = (unsigned long) vec | MEMCG_DATA_OBJCGS; |
2764 | if (new_page) { | |
2765 | /* | |
2766 | * If the slab page is brand new and nobody can yet access | |
2767 | * it's memcg_data, no synchronization is required and | |
2768 | * memcg_data can be simply assigned. | |
2769 | */ | |
2770 | page->memcg_data = memcg_data; | |
2771 | } else if (cmpxchg(&page->memcg_data, 0, memcg_data)) { | |
2772 | /* | |
2773 | * If the slab page is already in use, somebody can allocate | |
2774 | * and assign obj_cgroups in parallel. In this case the existing | |
2775 | * objcg vector should be reused. | |
2776 | */ | |
10befea9 | 2777 | kfree(vec); |
2e9bd483 RG |
2778 | return 0; |
2779 | } | |
10befea9 | 2780 | |
2e9bd483 | 2781 | kmemleak_not_leak(vec); |
10befea9 RG |
2782 | return 0; |
2783 | } | |
2784 | ||
8380ce47 RG |
2785 | /* |
2786 | * Returns a pointer to the memory cgroup to which the kernel object is charged. | |
2787 | * | |
bcfe06bf RG |
2788 | * A passed kernel object can be a slab object or a generic kernel page, so |
2789 | * different mechanisms for getting the memory cgroup pointer should be used. | |
2790 | * In certain cases (e.g. kernel stacks or large kmallocs with SLUB) the caller | |
2791 | * can not know for sure how the kernel object is implemented. | |
2792 | * mem_cgroup_from_obj() can be safely used in such cases. | |
2793 | * | |
8380ce47 RG |
2794 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), |
2795 | * cgroup_mutex, etc. | |
2796 | */ | |
2797 | struct mem_cgroup *mem_cgroup_from_obj(void *p) | |
2798 | { | |
2799 | struct page *page; | |
2800 | ||
2801 | if (mem_cgroup_disabled()) | |
2802 | return NULL; | |
2803 | ||
2804 | page = virt_to_head_page(p); | |
2805 | ||
2806 | /* | |
9855609b RG |
2807 | * Slab objects are accounted individually, not per-page. |
2808 | * Memcg membership data for each individual object is saved in | |
2809 | * the page->obj_cgroups. | |
8380ce47 | 2810 | */ |
270c6a71 | 2811 | if (page_objcgs_check(page)) { |
9855609b RG |
2812 | struct obj_cgroup *objcg; |
2813 | unsigned int off; | |
2814 | ||
2815 | off = obj_to_index(page->slab_cache, page, p); | |
270c6a71 | 2816 | objcg = page_objcgs(page)[off]; |
10befea9 RG |
2817 | if (objcg) |
2818 | return obj_cgroup_memcg(objcg); | |
2819 | ||
2820 | return NULL; | |
9855609b | 2821 | } |
8380ce47 | 2822 | |
bcfe06bf RG |
2823 | /* |
2824 | * page_memcg_check() is used here, because page_has_obj_cgroups() | |
2825 | * check above could fail because the object cgroups vector wasn't set | |
2826 | * at that moment, but it can be set concurrently. | |
2827 | * page_memcg_check(page) will guarantee that a proper memory | |
2828 | * cgroup pointer or NULL will be returned. | |
2829 | */ | |
2830 | return page_memcg_check(page); | |
8380ce47 RG |
2831 | } |
2832 | ||
bf4f0599 RG |
2833 | __always_inline struct obj_cgroup *get_obj_cgroup_from_current(void) |
2834 | { | |
2835 | struct obj_cgroup *objcg = NULL; | |
2836 | struct mem_cgroup *memcg; | |
2837 | ||
279c3393 RG |
2838 | if (memcg_kmem_bypass()) |
2839 | return NULL; | |
2840 | ||
bf4f0599 | 2841 | rcu_read_lock(); |
37d5985c RG |
2842 | if (unlikely(active_memcg())) |
2843 | memcg = active_memcg(); | |
bf4f0599 RG |
2844 | else |
2845 | memcg = mem_cgroup_from_task(current); | |
2846 | ||
2847 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { | |
2848 | objcg = rcu_dereference(memcg->objcg); | |
2849 | if (objcg && obj_cgroup_tryget(objcg)) | |
2850 | break; | |
2f7659a3 | 2851 | objcg = NULL; |
bf4f0599 RG |
2852 | } |
2853 | rcu_read_unlock(); | |
2854 | ||
2855 | return objcg; | |
2856 | } | |
2857 | ||
f3bb3043 | 2858 | static int memcg_alloc_cache_id(void) |
55007d84 | 2859 | { |
f3bb3043 VD |
2860 | int id, size; |
2861 | int err; | |
2862 | ||
dbcf73e2 | 2863 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2864 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2865 | if (id < 0) | |
2866 | return id; | |
55007d84 | 2867 | |
dbcf73e2 | 2868 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2869 | return id; |
2870 | ||
2871 | /* | |
2872 | * There's no space for the new id in memcg_caches arrays, | |
2873 | * so we have to grow them. | |
2874 | */ | |
05257a1a | 2875 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2876 | |
2877 | size = 2 * (id + 1); | |
55007d84 GC |
2878 | if (size < MEMCG_CACHES_MIN_SIZE) |
2879 | size = MEMCG_CACHES_MIN_SIZE; | |
2880 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2881 | size = MEMCG_CACHES_MAX_SIZE; | |
2882 | ||
9855609b | 2883 | err = memcg_update_all_list_lrus(size); |
05257a1a VD |
2884 | if (!err) |
2885 | memcg_nr_cache_ids = size; | |
2886 | ||
2887 | up_write(&memcg_cache_ids_sem); | |
2888 | ||
f3bb3043 | 2889 | if (err) { |
dbcf73e2 | 2890 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2891 | return err; |
2892 | } | |
2893 | return id; | |
2894 | } | |
2895 | ||
2896 | static void memcg_free_cache_id(int id) | |
2897 | { | |
dbcf73e2 | 2898 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2899 | } |
2900 | ||
f1286fae MS |
2901 | /* |
2902 | * obj_cgroup_uncharge_pages: uncharge a number of kernel pages from a objcg | |
2903 | * @objcg: object cgroup to uncharge | |
2904 | * @nr_pages: number of pages to uncharge | |
2905 | */ | |
e74d2259 MS |
2906 | static void obj_cgroup_uncharge_pages(struct obj_cgroup *objcg, |
2907 | unsigned int nr_pages) | |
2908 | { | |
2909 | struct mem_cgroup *memcg; | |
2910 | ||
2911 | memcg = get_mem_cgroup_from_objcg(objcg); | |
e74d2259 | 2912 | |
f1286fae MS |
2913 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
2914 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
2915 | refill_stock(memcg, nr_pages); | |
e74d2259 | 2916 | |
e74d2259 | 2917 | css_put(&memcg->css); |
e74d2259 MS |
2918 | } |
2919 | ||
f1286fae MS |
2920 | /* |
2921 | * obj_cgroup_charge_pages: charge a number of kernel pages to a objcg | |
2922 | * @objcg: object cgroup to charge | |
45264778 | 2923 | * @gfp: reclaim mode |
92d0510c | 2924 | * @nr_pages: number of pages to charge |
45264778 VD |
2925 | * |
2926 | * Returns 0 on success, an error code on failure. | |
2927 | */ | |
f1286fae MS |
2928 | static int obj_cgroup_charge_pages(struct obj_cgroup *objcg, gfp_t gfp, |
2929 | unsigned int nr_pages) | |
7ae1e1d0 | 2930 | { |
f3ccb2c4 | 2931 | struct page_counter *counter; |
f1286fae | 2932 | struct mem_cgroup *memcg; |
7ae1e1d0 GC |
2933 | int ret; |
2934 | ||
f1286fae MS |
2935 | memcg = get_mem_cgroup_from_objcg(objcg); |
2936 | ||
c5c8b16b | 2937 | ret = try_charge_memcg(memcg, gfp, nr_pages); |
52c29b04 | 2938 | if (ret) |
f1286fae | 2939 | goto out; |
52c29b04 JW |
2940 | |
2941 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
2942 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
e55d9d9b MH |
2943 | |
2944 | /* | |
2945 | * Enforce __GFP_NOFAIL allocation because callers are not | |
2946 | * prepared to see failures and likely do not have any failure | |
2947 | * handling code. | |
2948 | */ | |
2949 | if (gfp & __GFP_NOFAIL) { | |
2950 | page_counter_charge(&memcg->kmem, nr_pages); | |
f1286fae | 2951 | goto out; |
e55d9d9b | 2952 | } |
52c29b04 | 2953 | cancel_charge(memcg, nr_pages); |
f1286fae | 2954 | ret = -ENOMEM; |
7ae1e1d0 | 2955 | } |
f1286fae MS |
2956 | out: |
2957 | css_put(&memcg->css); | |
4b13f64d | 2958 | |
f1286fae | 2959 | return ret; |
4b13f64d RG |
2960 | } |
2961 | ||
45264778 | 2962 | /** |
f4b00eab | 2963 | * __memcg_kmem_charge_page: charge a kmem page to the current memory cgroup |
45264778 VD |
2964 | * @page: page to charge |
2965 | * @gfp: reclaim mode | |
2966 | * @order: allocation order | |
2967 | * | |
2968 | * Returns 0 on success, an error code on failure. | |
2969 | */ | |
f4b00eab | 2970 | int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 2971 | { |
b4e0b68f | 2972 | struct obj_cgroup *objcg; |
fcff7d7e | 2973 | int ret = 0; |
7ae1e1d0 | 2974 | |
b4e0b68f MS |
2975 | objcg = get_obj_cgroup_from_current(); |
2976 | if (objcg) { | |
2977 | ret = obj_cgroup_charge_pages(objcg, gfp, 1 << order); | |
4d96ba35 | 2978 | if (!ret) { |
b4e0b68f | 2979 | page->memcg_data = (unsigned long)objcg | |
18b2db3b | 2980 | MEMCG_DATA_KMEM; |
1a3e1f40 | 2981 | return 0; |
4d96ba35 | 2982 | } |
b4e0b68f | 2983 | obj_cgroup_put(objcg); |
c4159a75 | 2984 | } |
d05e83a6 | 2985 | return ret; |
7ae1e1d0 | 2986 | } |
49a18eae | 2987 | |
45264778 | 2988 | /** |
f4b00eab | 2989 | * __memcg_kmem_uncharge_page: uncharge a kmem page |
45264778 VD |
2990 | * @page: page to uncharge |
2991 | * @order: allocation order | |
2992 | */ | |
f4b00eab | 2993 | void __memcg_kmem_uncharge_page(struct page *page, int order) |
7ae1e1d0 | 2994 | { |
1b7e4464 | 2995 | struct folio *folio = page_folio(page); |
b4e0b68f | 2996 | struct obj_cgroup *objcg; |
f3ccb2c4 | 2997 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 2998 | |
1b7e4464 | 2999 | if (!folio_memcg_kmem(folio)) |
7ae1e1d0 GC |
3000 | return; |
3001 | ||
1b7e4464 | 3002 | objcg = __folio_objcg(folio); |
b4e0b68f | 3003 | obj_cgroup_uncharge_pages(objcg, nr_pages); |
1b7e4464 | 3004 | folio->memcg_data = 0; |
b4e0b68f | 3005 | obj_cgroup_put(objcg); |
60d3fd32 | 3006 | } |
bf4f0599 | 3007 | |
68ac5b3c WL |
3008 | void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat, |
3009 | enum node_stat_item idx, int nr) | |
3010 | { | |
68ac5b3c | 3011 | unsigned long flags; |
55927114 | 3012 | struct obj_stock *stock = get_obj_stock(&flags); |
68ac5b3c WL |
3013 | int *bytes; |
3014 | ||
68ac5b3c WL |
3015 | /* |
3016 | * Save vmstat data in stock and skip vmstat array update unless | |
3017 | * accumulating over a page of vmstat data or when pgdat or idx | |
3018 | * changes. | |
3019 | */ | |
3020 | if (stock->cached_objcg != objcg) { | |
3021 | drain_obj_stock(stock); | |
3022 | obj_cgroup_get(objcg); | |
3023 | stock->nr_bytes = atomic_read(&objcg->nr_charged_bytes) | |
3024 | ? atomic_xchg(&objcg->nr_charged_bytes, 0) : 0; | |
3025 | stock->cached_objcg = objcg; | |
3026 | stock->cached_pgdat = pgdat; | |
3027 | } else if (stock->cached_pgdat != pgdat) { | |
3028 | /* Flush the existing cached vmstat data */ | |
7fa0dacb WL |
3029 | struct pglist_data *oldpg = stock->cached_pgdat; |
3030 | ||
68ac5b3c | 3031 | if (stock->nr_slab_reclaimable_b) { |
7fa0dacb | 3032 | mod_objcg_mlstate(objcg, oldpg, NR_SLAB_RECLAIMABLE_B, |
68ac5b3c WL |
3033 | stock->nr_slab_reclaimable_b); |
3034 | stock->nr_slab_reclaimable_b = 0; | |
3035 | } | |
3036 | if (stock->nr_slab_unreclaimable_b) { | |
7fa0dacb | 3037 | mod_objcg_mlstate(objcg, oldpg, NR_SLAB_UNRECLAIMABLE_B, |
68ac5b3c WL |
3038 | stock->nr_slab_unreclaimable_b); |
3039 | stock->nr_slab_unreclaimable_b = 0; | |
3040 | } | |
3041 | stock->cached_pgdat = pgdat; | |
3042 | } | |
3043 | ||
3044 | bytes = (idx == NR_SLAB_RECLAIMABLE_B) ? &stock->nr_slab_reclaimable_b | |
3045 | : &stock->nr_slab_unreclaimable_b; | |
3046 | /* | |
3047 | * Even for large object >= PAGE_SIZE, the vmstat data will still be | |
3048 | * cached locally at least once before pushing it out. | |
3049 | */ | |
3050 | if (!*bytes) { | |
3051 | *bytes = nr; | |
3052 | nr = 0; | |
3053 | } else { | |
3054 | *bytes += nr; | |
3055 | if (abs(*bytes) > PAGE_SIZE) { | |
3056 | nr = *bytes; | |
3057 | *bytes = 0; | |
3058 | } else { | |
3059 | nr = 0; | |
3060 | } | |
3061 | } | |
3062 | if (nr) | |
3063 | mod_objcg_mlstate(objcg, pgdat, idx, nr); | |
3064 | ||
55927114 | 3065 | put_obj_stock(flags); |
68ac5b3c WL |
3066 | } |
3067 | ||
bf4f0599 RG |
3068 | static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) |
3069 | { | |
bf4f0599 | 3070 | unsigned long flags; |
55927114 | 3071 | struct obj_stock *stock = get_obj_stock(&flags); |
bf4f0599 RG |
3072 | bool ret = false; |
3073 | ||
bf4f0599 RG |
3074 | if (objcg == stock->cached_objcg && stock->nr_bytes >= nr_bytes) { |
3075 | stock->nr_bytes -= nr_bytes; | |
3076 | ret = true; | |
3077 | } | |
3078 | ||
55927114 | 3079 | put_obj_stock(flags); |
bf4f0599 RG |
3080 | |
3081 | return ret; | |
3082 | } | |
3083 | ||
55927114 | 3084 | static void drain_obj_stock(struct obj_stock *stock) |
bf4f0599 RG |
3085 | { |
3086 | struct obj_cgroup *old = stock->cached_objcg; | |
3087 | ||
3088 | if (!old) | |
3089 | return; | |
3090 | ||
3091 | if (stock->nr_bytes) { | |
3092 | unsigned int nr_pages = stock->nr_bytes >> PAGE_SHIFT; | |
3093 | unsigned int nr_bytes = stock->nr_bytes & (PAGE_SIZE - 1); | |
3094 | ||
e74d2259 MS |
3095 | if (nr_pages) |
3096 | obj_cgroup_uncharge_pages(old, nr_pages); | |
bf4f0599 RG |
3097 | |
3098 | /* | |
3099 | * The leftover is flushed to the centralized per-memcg value. | |
3100 | * On the next attempt to refill obj stock it will be moved | |
3101 | * to a per-cpu stock (probably, on an other CPU), see | |
3102 | * refill_obj_stock(). | |
3103 | * | |
3104 | * How often it's flushed is a trade-off between the memory | |
3105 | * limit enforcement accuracy and potential CPU contention, | |
3106 | * so it might be changed in the future. | |
3107 | */ | |
3108 | atomic_add(nr_bytes, &old->nr_charged_bytes); | |
3109 | stock->nr_bytes = 0; | |
3110 | } | |
3111 | ||
68ac5b3c WL |
3112 | /* |
3113 | * Flush the vmstat data in current stock | |
3114 | */ | |
3115 | if (stock->nr_slab_reclaimable_b || stock->nr_slab_unreclaimable_b) { | |
3116 | if (stock->nr_slab_reclaimable_b) { | |
3117 | mod_objcg_mlstate(old, stock->cached_pgdat, | |
3118 | NR_SLAB_RECLAIMABLE_B, | |
3119 | stock->nr_slab_reclaimable_b); | |
3120 | stock->nr_slab_reclaimable_b = 0; | |
3121 | } | |
3122 | if (stock->nr_slab_unreclaimable_b) { | |
3123 | mod_objcg_mlstate(old, stock->cached_pgdat, | |
3124 | NR_SLAB_UNRECLAIMABLE_B, | |
3125 | stock->nr_slab_unreclaimable_b); | |
3126 | stock->nr_slab_unreclaimable_b = 0; | |
3127 | } | |
3128 | stock->cached_pgdat = NULL; | |
3129 | } | |
3130 | ||
bf4f0599 RG |
3131 | obj_cgroup_put(old); |
3132 | stock->cached_objcg = NULL; | |
3133 | } | |
3134 | ||
3135 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
3136 | struct mem_cgroup *root_memcg) | |
3137 | { | |
3138 | struct mem_cgroup *memcg; | |
3139 | ||
55927114 WL |
3140 | if (in_task() && stock->task_obj.cached_objcg) { |
3141 | memcg = obj_cgroup_memcg(stock->task_obj.cached_objcg); | |
3142 | if (memcg && mem_cgroup_is_descendant(memcg, root_memcg)) | |
3143 | return true; | |
3144 | } | |
3145 | if (stock->irq_obj.cached_objcg) { | |
3146 | memcg = obj_cgroup_memcg(stock->irq_obj.cached_objcg); | |
bf4f0599 RG |
3147 | if (memcg && mem_cgroup_is_descendant(memcg, root_memcg)) |
3148 | return true; | |
3149 | } | |
3150 | ||
3151 | return false; | |
3152 | } | |
3153 | ||
5387c904 WL |
3154 | static void refill_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes, |
3155 | bool allow_uncharge) | |
bf4f0599 | 3156 | { |
bf4f0599 | 3157 | unsigned long flags; |
55927114 | 3158 | struct obj_stock *stock = get_obj_stock(&flags); |
5387c904 | 3159 | unsigned int nr_pages = 0; |
bf4f0599 | 3160 | |
bf4f0599 RG |
3161 | if (stock->cached_objcg != objcg) { /* reset if necessary */ |
3162 | drain_obj_stock(stock); | |
3163 | obj_cgroup_get(objcg); | |
3164 | stock->cached_objcg = objcg; | |
5387c904 WL |
3165 | stock->nr_bytes = atomic_read(&objcg->nr_charged_bytes) |
3166 | ? atomic_xchg(&objcg->nr_charged_bytes, 0) : 0; | |
3167 | allow_uncharge = true; /* Allow uncharge when objcg changes */ | |
bf4f0599 RG |
3168 | } |
3169 | stock->nr_bytes += nr_bytes; | |
3170 | ||
5387c904 WL |
3171 | if (allow_uncharge && (stock->nr_bytes > PAGE_SIZE)) { |
3172 | nr_pages = stock->nr_bytes >> PAGE_SHIFT; | |
3173 | stock->nr_bytes &= (PAGE_SIZE - 1); | |
3174 | } | |
bf4f0599 | 3175 | |
55927114 | 3176 | put_obj_stock(flags); |
5387c904 WL |
3177 | |
3178 | if (nr_pages) | |
3179 | obj_cgroup_uncharge_pages(objcg, nr_pages); | |
bf4f0599 RG |
3180 | } |
3181 | ||
3182 | int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size) | |
3183 | { | |
bf4f0599 RG |
3184 | unsigned int nr_pages, nr_bytes; |
3185 | int ret; | |
3186 | ||
3187 | if (consume_obj_stock(objcg, size)) | |
3188 | return 0; | |
3189 | ||
3190 | /* | |
5387c904 | 3191 | * In theory, objcg->nr_charged_bytes can have enough |
bf4f0599 | 3192 | * pre-charged bytes to satisfy the allocation. However, |
5387c904 WL |
3193 | * flushing objcg->nr_charged_bytes requires two atomic |
3194 | * operations, and objcg->nr_charged_bytes can't be big. | |
3195 | * The shared objcg->nr_charged_bytes can also become a | |
3196 | * performance bottleneck if all tasks of the same memcg are | |
3197 | * trying to update it. So it's better to ignore it and try | |
3198 | * grab some new pages. The stock's nr_bytes will be flushed to | |
3199 | * objcg->nr_charged_bytes later on when objcg changes. | |
3200 | * | |
3201 | * The stock's nr_bytes may contain enough pre-charged bytes | |
3202 | * to allow one less page from being charged, but we can't rely | |
3203 | * on the pre-charged bytes not being changed outside of | |
3204 | * consume_obj_stock() or refill_obj_stock(). So ignore those | |
3205 | * pre-charged bytes as well when charging pages. To avoid a | |
3206 | * page uncharge right after a page charge, we set the | |
3207 | * allow_uncharge flag to false when calling refill_obj_stock() | |
3208 | * to temporarily allow the pre-charged bytes to exceed the page | |
3209 | * size limit. The maximum reachable value of the pre-charged | |
3210 | * bytes is (sizeof(object) + PAGE_SIZE - 2) if there is no data | |
3211 | * race. | |
bf4f0599 | 3212 | */ |
bf4f0599 RG |
3213 | nr_pages = size >> PAGE_SHIFT; |
3214 | nr_bytes = size & (PAGE_SIZE - 1); | |
3215 | ||
3216 | if (nr_bytes) | |
3217 | nr_pages += 1; | |
3218 | ||
e74d2259 | 3219 | ret = obj_cgroup_charge_pages(objcg, gfp, nr_pages); |
bf4f0599 | 3220 | if (!ret && nr_bytes) |
5387c904 | 3221 | refill_obj_stock(objcg, PAGE_SIZE - nr_bytes, false); |
bf4f0599 | 3222 | |
bf4f0599 RG |
3223 | return ret; |
3224 | } | |
3225 | ||
3226 | void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size) | |
3227 | { | |
5387c904 | 3228 | refill_obj_stock(objcg, size, true); |
bf4f0599 RG |
3229 | } |
3230 | ||
84c07d11 | 3231 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 3232 | |
ca3e0214 | 3233 | /* |
be6c8982 | 3234 | * Because page_memcg(head) is not set on tails, set it now. |
ca3e0214 | 3235 | */ |
be6c8982 | 3236 | void split_page_memcg(struct page *head, unsigned int nr) |
ca3e0214 | 3237 | { |
1b7e4464 MWO |
3238 | struct folio *folio = page_folio(head); |
3239 | struct mem_cgroup *memcg = folio_memcg(folio); | |
e94c8a9c | 3240 | int i; |
ca3e0214 | 3241 | |
be6c8982 | 3242 | if (mem_cgroup_disabled() || !memcg) |
3d37c4a9 | 3243 | return; |
b070e65c | 3244 | |
be6c8982 | 3245 | for (i = 1; i < nr; i++) |
1b7e4464 | 3246 | folio_page(folio, i)->memcg_data = folio->memcg_data; |
b4e0b68f | 3247 | |
1b7e4464 MWO |
3248 | if (folio_memcg_kmem(folio)) |
3249 | obj_cgroup_get_many(__folio_objcg(folio), nr - 1); | |
b4e0b68f MS |
3250 | else |
3251 | css_get_many(&memcg->css, nr - 1); | |
ca3e0214 | 3252 | } |
ca3e0214 | 3253 | |
c255a458 | 3254 | #ifdef CONFIG_MEMCG_SWAP |
02491447 DN |
3255 | /** |
3256 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3257 | * @entry: swap entry to be moved | |
3258 | * @from: mem_cgroup which the entry is moved from | |
3259 | * @to: mem_cgroup which the entry is moved to | |
3260 | * | |
3261 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3262 | * as the mem_cgroup's id of @from. | |
3263 | * | |
3264 | * Returns 0 on success, -EINVAL on failure. | |
3265 | * | |
3e32cb2e | 3266 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
3267 | * both res and memsw, and called css_get(). |
3268 | */ | |
3269 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3270 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3271 | { |
3272 | unsigned short old_id, new_id; | |
3273 | ||
34c00c31 LZ |
3274 | old_id = mem_cgroup_id(from); |
3275 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3276 | |
3277 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
3278 | mod_memcg_state(from, MEMCG_SWAP, -1); |
3279 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
3280 | return 0; |
3281 | } | |
3282 | return -EINVAL; | |
3283 | } | |
3284 | #else | |
3285 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3286 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3287 | { |
3288 | return -EINVAL; | |
3289 | } | |
8c7c6e34 | 3290 | #endif |
d13d1443 | 3291 | |
bbec2e15 | 3292 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 3293 | |
bbec2e15 RG |
3294 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
3295 | unsigned long max, bool memsw) | |
628f4235 | 3296 | { |
3e32cb2e | 3297 | bool enlarge = false; |
bb4a7ea2 | 3298 | bool drained = false; |
3e32cb2e | 3299 | int ret; |
c054a78c YZ |
3300 | bool limits_invariant; |
3301 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 3302 | |
3e32cb2e | 3303 | do { |
628f4235 KH |
3304 | if (signal_pending(current)) { |
3305 | ret = -EINTR; | |
3306 | break; | |
3307 | } | |
3e32cb2e | 3308 | |
bbec2e15 | 3309 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
3310 | /* |
3311 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 3312 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 3313 | */ |
15b42562 | 3314 | limits_invariant = memsw ? max >= READ_ONCE(memcg->memory.max) : |
bbec2e15 | 3315 | max <= memcg->memsw.max; |
c054a78c | 3316 | if (!limits_invariant) { |
bbec2e15 | 3317 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 3318 | ret = -EINVAL; |
8c7c6e34 KH |
3319 | break; |
3320 | } | |
bbec2e15 | 3321 | if (max > counter->max) |
3e32cb2e | 3322 | enlarge = true; |
bbec2e15 RG |
3323 | ret = page_counter_set_max(counter, max); |
3324 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
3325 | |
3326 | if (!ret) | |
3327 | break; | |
3328 | ||
bb4a7ea2 SB |
3329 | if (!drained) { |
3330 | drain_all_stock(memcg); | |
3331 | drained = true; | |
3332 | continue; | |
3333 | } | |
3334 | ||
1ab5c056 AR |
3335 | if (!try_to_free_mem_cgroup_pages(memcg, 1, |
3336 | GFP_KERNEL, !memsw)) { | |
3337 | ret = -EBUSY; | |
3338 | break; | |
3339 | } | |
3340 | } while (true); | |
3e32cb2e | 3341 | |
3c11ecf4 KH |
3342 | if (!ret && enlarge) |
3343 | memcg_oom_recover(memcg); | |
3e32cb2e | 3344 | |
628f4235 KH |
3345 | return ret; |
3346 | } | |
3347 | ||
ef8f2327 | 3348 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
3349 | gfp_t gfp_mask, |
3350 | unsigned long *total_scanned) | |
3351 | { | |
3352 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 3353 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
3354 | unsigned long reclaimed; |
3355 | int loop = 0; | |
ef8f2327 | 3356 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 3357 | unsigned long excess; |
0608f43d AM |
3358 | unsigned long nr_scanned; |
3359 | ||
3360 | if (order > 0) | |
3361 | return 0; | |
3362 | ||
2ab082ba | 3363 | mctz = soft_limit_tree.rb_tree_per_node[pgdat->node_id]; |
d6507ff5 MH |
3364 | |
3365 | /* | |
3366 | * Do not even bother to check the largest node if the root | |
3367 | * is empty. Do it lockless to prevent lock bouncing. Races | |
3368 | * are acceptable as soft limit is best effort anyway. | |
3369 | */ | |
bfc7228b | 3370 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
3371 | return 0; |
3372 | ||
0608f43d AM |
3373 | /* |
3374 | * This loop can run a while, specially if mem_cgroup's continuously | |
3375 | * keep exceeding their soft limit and putting the system under | |
3376 | * pressure | |
3377 | */ | |
3378 | do { | |
3379 | if (next_mz) | |
3380 | mz = next_mz; | |
3381 | else | |
3382 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3383 | if (!mz) | |
3384 | break; | |
3385 | ||
3386 | nr_scanned = 0; | |
ef8f2327 | 3387 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
0608f43d AM |
3388 | gfp_mask, &nr_scanned); |
3389 | nr_reclaimed += reclaimed; | |
3390 | *total_scanned += nr_scanned; | |
0a31bc97 | 3391 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 3392 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
3393 | |
3394 | /* | |
3395 | * If we failed to reclaim anything from this memory cgroup | |
3396 | * it is time to move on to the next cgroup | |
3397 | */ | |
3398 | next_mz = NULL; | |
bc2f2e7f VD |
3399 | if (!reclaimed) |
3400 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3401 | ||
3e32cb2e | 3402 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3403 | /* |
3404 | * One school of thought says that we should not add | |
3405 | * back the node to the tree if reclaim returns 0. | |
3406 | * But our reclaim could return 0, simply because due | |
3407 | * to priority we are exposing a smaller subset of | |
3408 | * memory to reclaim from. Consider this as a longer | |
3409 | * term TODO. | |
3410 | */ | |
3411 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3412 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3413 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3414 | css_put(&mz->memcg->css); |
3415 | loop++; | |
3416 | /* | |
3417 | * Could not reclaim anything and there are no more | |
3418 | * mem cgroups to try or we seem to be looping without | |
3419 | * reclaiming anything. | |
3420 | */ | |
3421 | if (!nr_reclaimed && | |
3422 | (next_mz == NULL || | |
3423 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3424 | break; | |
3425 | } while (!nr_reclaimed); | |
3426 | if (next_mz) | |
3427 | css_put(&next_mz->memcg->css); | |
3428 | return nr_reclaimed; | |
3429 | } | |
3430 | ||
c26251f9 | 3431 | /* |
51038171 | 3432 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
3433 | * |
3434 | * Caller is responsible for holding css reference for memcg. | |
3435 | */ | |
3436 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3437 | { | |
d977aa93 | 3438 | int nr_retries = MAX_RECLAIM_RETRIES; |
c26251f9 | 3439 | |
c1e862c1 KH |
3440 | /* we call try-to-free pages for make this cgroup empty */ |
3441 | lru_add_drain_all(); | |
d12c60f6 JS |
3442 | |
3443 | drain_all_stock(memcg); | |
3444 | ||
f817ed48 | 3445 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3446 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 3447 | int progress; |
c1e862c1 | 3448 | |
c26251f9 MH |
3449 | if (signal_pending(current)) |
3450 | return -EINTR; | |
3451 | ||
b70a2a21 JW |
3452 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3453 | GFP_KERNEL, true); | |
c1e862c1 | 3454 | if (!progress) { |
f817ed48 | 3455 | nr_retries--; |
c1e862c1 | 3456 | /* maybe some writeback is necessary */ |
8aa7e847 | 3457 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3458 | } |
f817ed48 KH |
3459 | |
3460 | } | |
ab5196c2 MH |
3461 | |
3462 | return 0; | |
cc847582 KH |
3463 | } |
3464 | ||
6770c64e TH |
3465 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3466 | char *buf, size_t nbytes, | |
3467 | loff_t off) | |
c1e862c1 | 3468 | { |
6770c64e | 3469 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3470 | |
d8423011 MH |
3471 | if (mem_cgroup_is_root(memcg)) |
3472 | return -EINVAL; | |
6770c64e | 3473 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3474 | } |
3475 | ||
182446d0 TH |
3476 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3477 | struct cftype *cft) | |
18f59ea7 | 3478 | { |
bef8620c | 3479 | return 1; |
18f59ea7 BS |
3480 | } |
3481 | ||
182446d0 TH |
3482 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3483 | struct cftype *cft, u64 val) | |
18f59ea7 | 3484 | { |
bef8620c | 3485 | if (val == 1) |
0b8f73e1 | 3486 | return 0; |
567fb435 | 3487 | |
bef8620c RG |
3488 | pr_warn_once("Non-hierarchical mode is deprecated. " |
3489 | "Please report your usecase to linux-mm@kvack.org if you " | |
3490 | "depend on this functionality.\n"); | |
567fb435 | 3491 | |
bef8620c | 3492 | return -EINVAL; |
18f59ea7 BS |
3493 | } |
3494 | ||
6f646156 | 3495 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 3496 | { |
42a30035 | 3497 | unsigned long val; |
ce00a967 | 3498 | |
3e32cb2e | 3499 | if (mem_cgroup_is_root(memcg)) { |
30def935 JW |
3500 | /* mem_cgroup_threshold() calls here from irqsafe context */ |
3501 | cgroup_rstat_flush_irqsafe(memcg->css.cgroup); | |
0d1c2072 | 3502 | val = memcg_page_state(memcg, NR_FILE_PAGES) + |
be5d0a74 | 3503 | memcg_page_state(memcg, NR_ANON_MAPPED); |
42a30035 JW |
3504 | if (swap) |
3505 | val += memcg_page_state(memcg, MEMCG_SWAP); | |
3e32cb2e | 3506 | } else { |
ce00a967 | 3507 | if (!swap) |
3e32cb2e | 3508 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3509 | else |
3e32cb2e | 3510 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3511 | } |
c12176d3 | 3512 | return val; |
ce00a967 JW |
3513 | } |
3514 | ||
3e32cb2e JW |
3515 | enum { |
3516 | RES_USAGE, | |
3517 | RES_LIMIT, | |
3518 | RES_MAX_USAGE, | |
3519 | RES_FAILCNT, | |
3520 | RES_SOFT_LIMIT, | |
3521 | }; | |
ce00a967 | 3522 | |
791badbd | 3523 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3524 | struct cftype *cft) |
8cdea7c0 | 3525 | { |
182446d0 | 3526 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3527 | struct page_counter *counter; |
af36f906 | 3528 | |
3e32cb2e | 3529 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3530 | case _MEM: |
3e32cb2e JW |
3531 | counter = &memcg->memory; |
3532 | break; | |
8c7c6e34 | 3533 | case _MEMSWAP: |
3e32cb2e JW |
3534 | counter = &memcg->memsw; |
3535 | break; | |
510fc4e1 | 3536 | case _KMEM: |
3e32cb2e | 3537 | counter = &memcg->kmem; |
510fc4e1 | 3538 | break; |
d55f90bf | 3539 | case _TCP: |
0db15298 | 3540 | counter = &memcg->tcpmem; |
d55f90bf | 3541 | break; |
8c7c6e34 KH |
3542 | default: |
3543 | BUG(); | |
8c7c6e34 | 3544 | } |
3e32cb2e JW |
3545 | |
3546 | switch (MEMFILE_ATTR(cft->private)) { | |
3547 | case RES_USAGE: | |
3548 | if (counter == &memcg->memory) | |
c12176d3 | 3549 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 3550 | if (counter == &memcg->memsw) |
c12176d3 | 3551 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
3552 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
3553 | case RES_LIMIT: | |
bbec2e15 | 3554 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
3555 | case RES_MAX_USAGE: |
3556 | return (u64)counter->watermark * PAGE_SIZE; | |
3557 | case RES_FAILCNT: | |
3558 | return counter->failcnt; | |
3559 | case RES_SOFT_LIMIT: | |
3560 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3561 | default: | |
3562 | BUG(); | |
3563 | } | |
8cdea7c0 | 3564 | } |
510fc4e1 | 3565 | |
84c07d11 | 3566 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 3567 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 3568 | { |
bf4f0599 | 3569 | struct obj_cgroup *objcg; |
d6441637 VD |
3570 | int memcg_id; |
3571 | ||
b313aeee VD |
3572 | if (cgroup_memory_nokmem) |
3573 | return 0; | |
3574 | ||
2a4db7eb | 3575 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 3576 | BUG_ON(memcg->kmem_state); |
d6441637 | 3577 | |
f3bb3043 | 3578 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
3579 | if (memcg_id < 0) |
3580 | return memcg_id; | |
d6441637 | 3581 | |
bf4f0599 RG |
3582 | objcg = obj_cgroup_alloc(); |
3583 | if (!objcg) { | |
3584 | memcg_free_cache_id(memcg_id); | |
3585 | return -ENOMEM; | |
3586 | } | |
3587 | objcg->memcg = memcg; | |
3588 | rcu_assign_pointer(memcg->objcg, objcg); | |
3589 | ||
d648bcc7 RG |
3590 | static_branch_enable(&memcg_kmem_enabled_key); |
3591 | ||
900a38f0 | 3592 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 3593 | memcg->kmem_state = KMEM_ONLINE; |
0b8f73e1 JW |
3594 | |
3595 | return 0; | |
d6441637 VD |
3596 | } |
3597 | ||
8e0a8912 JW |
3598 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
3599 | { | |
3600 | struct cgroup_subsys_state *css; | |
3601 | struct mem_cgroup *parent, *child; | |
3602 | int kmemcg_id; | |
3603 | ||
3604 | if (memcg->kmem_state != KMEM_ONLINE) | |
3605 | return; | |
9855609b | 3606 | |
8e0a8912 JW |
3607 | memcg->kmem_state = KMEM_ALLOCATED; |
3608 | ||
8e0a8912 JW |
3609 | parent = parent_mem_cgroup(memcg); |
3610 | if (!parent) | |
3611 | parent = root_mem_cgroup; | |
3612 | ||
bf4f0599 | 3613 | memcg_reparent_objcgs(memcg, parent); |
fb2f2b0a RG |
3614 | |
3615 | kmemcg_id = memcg->kmemcg_id; | |
3616 | BUG_ON(kmemcg_id < 0); | |
3617 | ||
8e0a8912 JW |
3618 | /* |
3619 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3620 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3621 | * to ones of the parent. After we have finished, all list_lrus | |
3622 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3623 | * ordering is imposed by list_lru_node->lock taken by | |
3624 | * memcg_drain_all_list_lrus(). | |
3625 | */ | |
3a06bb78 | 3626 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
8e0a8912 JW |
3627 | css_for_each_descendant_pre(css, &memcg->css) { |
3628 | child = mem_cgroup_from_css(css); | |
3629 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3630 | child->kmemcg_id = parent->kmemcg_id; | |
8e0a8912 | 3631 | } |
3a06bb78 TH |
3632 | rcu_read_unlock(); |
3633 | ||
9bec5c35 | 3634 | memcg_drain_all_list_lrus(kmemcg_id, parent); |
8e0a8912 JW |
3635 | |
3636 | memcg_free_cache_id(kmemcg_id); | |
3637 | } | |
3638 | ||
3639 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3640 | { | |
0b8f73e1 JW |
3641 | /* css_alloc() failed, offlining didn't happen */ |
3642 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
3643 | memcg_offline_kmem(memcg); | |
8e0a8912 | 3644 | } |
d6441637 | 3645 | #else |
0b8f73e1 | 3646 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
3647 | { |
3648 | return 0; | |
3649 | } | |
3650 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
3651 | { | |
3652 | } | |
3653 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3654 | { | |
3655 | } | |
84c07d11 | 3656 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 3657 | |
bbec2e15 RG |
3658 | static int memcg_update_kmem_max(struct mem_cgroup *memcg, |
3659 | unsigned long max) | |
d6441637 | 3660 | { |
b313aeee | 3661 | int ret; |
127424c8 | 3662 | |
bbec2e15 RG |
3663 | mutex_lock(&memcg_max_mutex); |
3664 | ret = page_counter_set_max(&memcg->kmem, max); | |
3665 | mutex_unlock(&memcg_max_mutex); | |
127424c8 | 3666 | return ret; |
d6441637 | 3667 | } |
510fc4e1 | 3668 | |
bbec2e15 | 3669 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
3670 | { |
3671 | int ret; | |
3672 | ||
bbec2e15 | 3673 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 3674 | |
bbec2e15 | 3675 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
3676 | if (ret) |
3677 | goto out; | |
3678 | ||
0db15298 | 3679 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
3680 | /* |
3681 | * The active flag needs to be written after the static_key | |
3682 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
3683 | * function is the last one to run. See mem_cgroup_sk_alloc() |
3684 | * for details, and note that we don't mark any socket as | |
3685 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
3686 | * |
3687 | * We need to do this, because static_keys will span multiple | |
3688 | * sites, but we can't control their order. If we mark a socket | |
3689 | * as accounted, but the accounting functions are not patched in | |
3690 | * yet, we'll lose accounting. | |
3691 | * | |
2d758073 | 3692 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
3693 | * because when this value change, the code to process it is not |
3694 | * patched in yet. | |
3695 | */ | |
3696 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 3697 | memcg->tcpmem_active = true; |
d55f90bf VD |
3698 | } |
3699 | out: | |
bbec2e15 | 3700 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
3701 | return ret; |
3702 | } | |
d55f90bf | 3703 | |
628f4235 KH |
3704 | /* |
3705 | * The user of this function is... | |
3706 | * RES_LIMIT. | |
3707 | */ | |
451af504 TH |
3708 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3709 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3710 | { |
451af504 | 3711 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3712 | unsigned long nr_pages; |
628f4235 KH |
3713 | int ret; |
3714 | ||
451af504 | 3715 | buf = strstrip(buf); |
650c5e56 | 3716 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3717 | if (ret) |
3718 | return ret; | |
af36f906 | 3719 | |
3e32cb2e | 3720 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3721 | case RES_LIMIT: |
4b3bde4c BS |
3722 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3723 | ret = -EINVAL; | |
3724 | break; | |
3725 | } | |
3e32cb2e JW |
3726 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3727 | case _MEM: | |
bbec2e15 | 3728 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 3729 | break; |
3e32cb2e | 3730 | case _MEMSWAP: |
bbec2e15 | 3731 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 3732 | break; |
3e32cb2e | 3733 | case _KMEM: |
0158115f MH |
3734 | pr_warn_once("kmem.limit_in_bytes is deprecated and will be removed. " |
3735 | "Please report your usecase to linux-mm@kvack.org if you " | |
3736 | "depend on this functionality.\n"); | |
bbec2e15 | 3737 | ret = memcg_update_kmem_max(memcg, nr_pages); |
3e32cb2e | 3738 | break; |
d55f90bf | 3739 | case _TCP: |
bbec2e15 | 3740 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 3741 | break; |
3e32cb2e | 3742 | } |
296c81d8 | 3743 | break; |
3e32cb2e JW |
3744 | case RES_SOFT_LIMIT: |
3745 | memcg->soft_limit = nr_pages; | |
3746 | ret = 0; | |
628f4235 KH |
3747 | break; |
3748 | } | |
451af504 | 3749 | return ret ?: nbytes; |
8cdea7c0 BS |
3750 | } |
3751 | ||
6770c64e TH |
3752 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3753 | size_t nbytes, loff_t off) | |
c84872e1 | 3754 | { |
6770c64e | 3755 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3756 | struct page_counter *counter; |
c84872e1 | 3757 | |
3e32cb2e JW |
3758 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3759 | case _MEM: | |
3760 | counter = &memcg->memory; | |
3761 | break; | |
3762 | case _MEMSWAP: | |
3763 | counter = &memcg->memsw; | |
3764 | break; | |
3765 | case _KMEM: | |
3766 | counter = &memcg->kmem; | |
3767 | break; | |
d55f90bf | 3768 | case _TCP: |
0db15298 | 3769 | counter = &memcg->tcpmem; |
d55f90bf | 3770 | break; |
3e32cb2e JW |
3771 | default: |
3772 | BUG(); | |
3773 | } | |
af36f906 | 3774 | |
3e32cb2e | 3775 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3776 | case RES_MAX_USAGE: |
3e32cb2e | 3777 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3778 | break; |
3779 | case RES_FAILCNT: | |
3e32cb2e | 3780 | counter->failcnt = 0; |
29f2a4da | 3781 | break; |
3e32cb2e JW |
3782 | default: |
3783 | BUG(); | |
29f2a4da | 3784 | } |
f64c3f54 | 3785 | |
6770c64e | 3786 | return nbytes; |
c84872e1 PE |
3787 | } |
3788 | ||
182446d0 | 3789 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3790 | struct cftype *cft) |
3791 | { | |
182446d0 | 3792 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3793 | } |
3794 | ||
02491447 | 3795 | #ifdef CONFIG_MMU |
182446d0 | 3796 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3797 | struct cftype *cft, u64 val) |
3798 | { | |
182446d0 | 3799 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3800 | |
1dfab5ab | 3801 | if (val & ~MOVE_MASK) |
7dc74be0 | 3802 | return -EINVAL; |
ee5e8472 | 3803 | |
7dc74be0 | 3804 | /* |
ee5e8472 GC |
3805 | * No kind of locking is needed in here, because ->can_attach() will |
3806 | * check this value once in the beginning of the process, and then carry | |
3807 | * on with stale data. This means that changes to this value will only | |
3808 | * affect task migrations starting after the change. | |
7dc74be0 | 3809 | */ |
c0ff4b85 | 3810 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3811 | return 0; |
3812 | } | |
02491447 | 3813 | #else |
182446d0 | 3814 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3815 | struct cftype *cft, u64 val) |
3816 | { | |
3817 | return -ENOSYS; | |
3818 | } | |
3819 | #endif | |
7dc74be0 | 3820 | |
406eb0c9 | 3821 | #ifdef CONFIG_NUMA |
113b7dfd JW |
3822 | |
3823 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
3824 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
3825 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) | |
3826 | ||
3827 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 | 3828 | int nid, unsigned int lru_mask, bool tree) |
113b7dfd | 3829 | { |
867e5e1d | 3830 | struct lruvec *lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); |
113b7dfd JW |
3831 | unsigned long nr = 0; |
3832 | enum lru_list lru; | |
3833 | ||
3834 | VM_BUG_ON((unsigned)nid >= nr_node_ids); | |
3835 | ||
3836 | for_each_lru(lru) { | |
3837 | if (!(BIT(lru) & lru_mask)) | |
3838 | continue; | |
dd8657b6 SB |
3839 | if (tree) |
3840 | nr += lruvec_page_state(lruvec, NR_LRU_BASE + lru); | |
3841 | else | |
3842 | nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru); | |
113b7dfd JW |
3843 | } |
3844 | return nr; | |
3845 | } | |
3846 | ||
3847 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 SB |
3848 | unsigned int lru_mask, |
3849 | bool tree) | |
113b7dfd JW |
3850 | { |
3851 | unsigned long nr = 0; | |
3852 | enum lru_list lru; | |
3853 | ||
3854 | for_each_lru(lru) { | |
3855 | if (!(BIT(lru) & lru_mask)) | |
3856 | continue; | |
dd8657b6 SB |
3857 | if (tree) |
3858 | nr += memcg_page_state(memcg, NR_LRU_BASE + lru); | |
3859 | else | |
3860 | nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru); | |
113b7dfd JW |
3861 | } |
3862 | return nr; | |
3863 | } | |
3864 | ||
2da8ca82 | 3865 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3866 | { |
25485de6 GT |
3867 | struct numa_stat { |
3868 | const char *name; | |
3869 | unsigned int lru_mask; | |
3870 | }; | |
3871 | ||
3872 | static const struct numa_stat stats[] = { | |
3873 | { "total", LRU_ALL }, | |
3874 | { "file", LRU_ALL_FILE }, | |
3875 | { "anon", LRU_ALL_ANON }, | |
3876 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3877 | }; | |
3878 | const struct numa_stat *stat; | |
406eb0c9 | 3879 | int nid; |
aa9694bb | 3880 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
406eb0c9 | 3881 | |
2d146aa3 JW |
3882 | cgroup_rstat_flush(memcg->css.cgroup); |
3883 | ||
25485de6 | 3884 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
3885 | seq_printf(m, "%s=%lu", stat->name, |
3886 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
3887 | false)); | |
3888 | for_each_node_state(nid, N_MEMORY) | |
3889 | seq_printf(m, " N%d=%lu", nid, | |
3890 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3891 | stat->lru_mask, false)); | |
25485de6 | 3892 | seq_putc(m, '\n'); |
406eb0c9 | 3893 | } |
406eb0c9 | 3894 | |
071aee13 | 3895 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
3896 | |
3897 | seq_printf(m, "hierarchical_%s=%lu", stat->name, | |
3898 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
3899 | true)); | |
3900 | for_each_node_state(nid, N_MEMORY) | |
3901 | seq_printf(m, " N%d=%lu", nid, | |
3902 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3903 | stat->lru_mask, true)); | |
071aee13 | 3904 | seq_putc(m, '\n'); |
406eb0c9 | 3905 | } |
406eb0c9 | 3906 | |
406eb0c9 YH |
3907 | return 0; |
3908 | } | |
3909 | #endif /* CONFIG_NUMA */ | |
3910 | ||
c8713d0b | 3911 | static const unsigned int memcg1_stats[] = { |
0d1c2072 | 3912 | NR_FILE_PAGES, |
be5d0a74 | 3913 | NR_ANON_MAPPED, |
468c3982 JW |
3914 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3915 | NR_ANON_THPS, | |
3916 | #endif | |
c8713d0b JW |
3917 | NR_SHMEM, |
3918 | NR_FILE_MAPPED, | |
3919 | NR_FILE_DIRTY, | |
3920 | NR_WRITEBACK, | |
3921 | MEMCG_SWAP, | |
3922 | }; | |
3923 | ||
3924 | static const char *const memcg1_stat_names[] = { | |
3925 | "cache", | |
3926 | "rss", | |
468c3982 | 3927 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
c8713d0b | 3928 | "rss_huge", |
468c3982 | 3929 | #endif |
c8713d0b JW |
3930 | "shmem", |
3931 | "mapped_file", | |
3932 | "dirty", | |
3933 | "writeback", | |
3934 | "swap", | |
3935 | }; | |
3936 | ||
df0e53d0 | 3937 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 3938 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
3939 | PGPGIN, |
3940 | PGPGOUT, | |
3941 | PGFAULT, | |
3942 | PGMAJFAULT, | |
3943 | }; | |
3944 | ||
2da8ca82 | 3945 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3946 | { |
aa9694bb | 3947 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3e32cb2e | 3948 | unsigned long memory, memsw; |
af7c4b0e JW |
3949 | struct mem_cgroup *mi; |
3950 | unsigned int i; | |
406eb0c9 | 3951 | |
71cd3113 | 3952 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c | 3953 | |
2d146aa3 JW |
3954 | cgroup_rstat_flush(memcg->css.cgroup); |
3955 | ||
71cd3113 | 3956 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
468c3982 JW |
3957 | unsigned long nr; |
3958 | ||
71cd3113 | 3959 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 3960 | continue; |
468c3982 | 3961 | nr = memcg_page_state_local(memcg, memcg1_stats[i]); |
468c3982 | 3962 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], nr * PAGE_SIZE); |
1dd3a273 | 3963 | } |
7b854121 | 3964 | |
df0e53d0 | 3965 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d | 3966 | seq_printf(m, "%s %lu\n", vm_event_name(memcg1_events[i]), |
205b20cc | 3967 | memcg_events_local(memcg, memcg1_events[i])); |
af7c4b0e JW |
3968 | |
3969 | for (i = 0; i < NR_LRU_LISTS; i++) | |
ebc5d83d | 3970 | seq_printf(m, "%s %lu\n", lru_list_name(i), |
205b20cc | 3971 | memcg_page_state_local(memcg, NR_LRU_BASE + i) * |
21d89d15 | 3972 | PAGE_SIZE); |
af7c4b0e | 3973 | |
14067bb3 | 3974 | /* Hierarchical information */ |
3e32cb2e JW |
3975 | memory = memsw = PAGE_COUNTER_MAX; |
3976 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
15b42562 CD |
3977 | memory = min(memory, READ_ONCE(mi->memory.max)); |
3978 | memsw = min(memsw, READ_ONCE(mi->memsw.max)); | |
fee7b548 | 3979 | } |
3e32cb2e JW |
3980 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3981 | (u64)memory * PAGE_SIZE); | |
7941d214 | 3982 | if (do_memsw_account()) |
3e32cb2e JW |
3983 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
3984 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3985 | |
8de7ecc6 | 3986 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
7de2e9f1 | 3987 | unsigned long nr; |
3988 | ||
71cd3113 | 3989 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 3990 | continue; |
7de2e9f1 | 3991 | nr = memcg_page_state(memcg, memcg1_stats[i]); |
8de7ecc6 | 3992 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], |
7de2e9f1 | 3993 | (u64)nr * PAGE_SIZE); |
af7c4b0e JW |
3994 | } |
3995 | ||
8de7ecc6 | 3996 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d KK |
3997 | seq_printf(m, "total_%s %llu\n", |
3998 | vm_event_name(memcg1_events[i]), | |
dd923990 | 3999 | (u64)memcg_events(memcg, memcg1_events[i])); |
af7c4b0e | 4000 | |
8de7ecc6 | 4001 | for (i = 0; i < NR_LRU_LISTS; i++) |
ebc5d83d | 4002 | seq_printf(m, "total_%s %llu\n", lru_list_name(i), |
42a30035 JW |
4003 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
4004 | PAGE_SIZE); | |
14067bb3 | 4005 | |
7f016ee8 | 4006 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 4007 | { |
ef8f2327 MG |
4008 | pg_data_t *pgdat; |
4009 | struct mem_cgroup_per_node *mz; | |
1431d4d1 JW |
4010 | unsigned long anon_cost = 0; |
4011 | unsigned long file_cost = 0; | |
7f016ee8 | 4012 | |
ef8f2327 | 4013 | for_each_online_pgdat(pgdat) { |
a3747b53 | 4014 | mz = memcg->nodeinfo[pgdat->node_id]; |
7f016ee8 | 4015 | |
1431d4d1 JW |
4016 | anon_cost += mz->lruvec.anon_cost; |
4017 | file_cost += mz->lruvec.file_cost; | |
ef8f2327 | 4018 | } |
1431d4d1 JW |
4019 | seq_printf(m, "anon_cost %lu\n", anon_cost); |
4020 | seq_printf(m, "file_cost %lu\n", file_cost); | |
7f016ee8 KM |
4021 | } |
4022 | #endif | |
4023 | ||
d2ceb9b7 KH |
4024 | return 0; |
4025 | } | |
4026 | ||
182446d0 TH |
4027 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
4028 | struct cftype *cft) | |
a7885eb8 | 4029 | { |
182446d0 | 4030 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4031 | |
1f4c025b | 4032 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4033 | } |
4034 | ||
182446d0 TH |
4035 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
4036 | struct cftype *cft, u64 val) | |
a7885eb8 | 4037 | { |
182446d0 | 4038 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4039 | |
37bc3cb9 | 4040 | if (val > 200) |
a7885eb8 KM |
4041 | return -EINVAL; |
4042 | ||
a4792030 | 4043 | if (!mem_cgroup_is_root(memcg)) |
3dae7fec JW |
4044 | memcg->swappiness = val; |
4045 | else | |
4046 | vm_swappiness = val; | |
068b38c1 | 4047 | |
a7885eb8 KM |
4048 | return 0; |
4049 | } | |
4050 | ||
2e72b634 KS |
4051 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4052 | { | |
4053 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 4054 | unsigned long usage; |
2e72b634 KS |
4055 | int i; |
4056 | ||
4057 | rcu_read_lock(); | |
4058 | if (!swap) | |
2c488db2 | 4059 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4060 | else |
2c488db2 | 4061 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4062 | |
4063 | if (!t) | |
4064 | goto unlock; | |
4065 | ||
ce00a967 | 4066 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
4067 | |
4068 | /* | |
748dad36 | 4069 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4070 | * If it's not true, a threshold was crossed after last |
4071 | * call of __mem_cgroup_threshold(). | |
4072 | */ | |
5407a562 | 4073 | i = t->current_threshold; |
2e72b634 KS |
4074 | |
4075 | /* | |
4076 | * Iterate backward over array of thresholds starting from | |
4077 | * current_threshold and check if a threshold is crossed. | |
4078 | * If none of thresholds below usage is crossed, we read | |
4079 | * only one element of the array here. | |
4080 | */ | |
4081 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4082 | eventfd_signal(t->entries[i].eventfd, 1); | |
4083 | ||
4084 | /* i = current_threshold + 1 */ | |
4085 | i++; | |
4086 | ||
4087 | /* | |
4088 | * Iterate forward over array of thresholds starting from | |
4089 | * current_threshold+1 and check if a threshold is crossed. | |
4090 | * If none of thresholds above usage is crossed, we read | |
4091 | * only one element of the array here. | |
4092 | */ | |
4093 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4094 | eventfd_signal(t->entries[i].eventfd, 1); | |
4095 | ||
4096 | /* Update current_threshold */ | |
5407a562 | 4097 | t->current_threshold = i - 1; |
2e72b634 KS |
4098 | unlock: |
4099 | rcu_read_unlock(); | |
4100 | } | |
4101 | ||
4102 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4103 | { | |
ad4ca5f4 KS |
4104 | while (memcg) { |
4105 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 4106 | if (do_memsw_account()) |
ad4ca5f4 KS |
4107 | __mem_cgroup_threshold(memcg, true); |
4108 | ||
4109 | memcg = parent_mem_cgroup(memcg); | |
4110 | } | |
2e72b634 KS |
4111 | } |
4112 | ||
4113 | static int compare_thresholds(const void *a, const void *b) | |
4114 | { | |
4115 | const struct mem_cgroup_threshold *_a = a; | |
4116 | const struct mem_cgroup_threshold *_b = b; | |
4117 | ||
2bff24a3 GT |
4118 | if (_a->threshold > _b->threshold) |
4119 | return 1; | |
4120 | ||
4121 | if (_a->threshold < _b->threshold) | |
4122 | return -1; | |
4123 | ||
4124 | return 0; | |
2e72b634 KS |
4125 | } |
4126 | ||
c0ff4b85 | 4127 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4128 | { |
4129 | struct mem_cgroup_eventfd_list *ev; | |
4130 | ||
2bcf2e92 MH |
4131 | spin_lock(&memcg_oom_lock); |
4132 | ||
c0ff4b85 | 4133 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 4134 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
4135 | |
4136 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
4137 | return 0; |
4138 | } | |
4139 | ||
c0ff4b85 | 4140 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4141 | { |
7d74b06f KH |
4142 | struct mem_cgroup *iter; |
4143 | ||
c0ff4b85 | 4144 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4145 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4146 | } |
4147 | ||
59b6f873 | 4148 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4149 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4150 | { |
2c488db2 KS |
4151 | struct mem_cgroup_thresholds *thresholds; |
4152 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
4153 | unsigned long threshold; |
4154 | unsigned long usage; | |
2c488db2 | 4155 | int i, size, ret; |
2e72b634 | 4156 | |
650c5e56 | 4157 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
4158 | if (ret) |
4159 | return ret; | |
4160 | ||
4161 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4162 | |
05b84301 | 4163 | if (type == _MEM) { |
2c488db2 | 4164 | thresholds = &memcg->thresholds; |
ce00a967 | 4165 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4166 | } else if (type == _MEMSWAP) { |
2c488db2 | 4167 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4168 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4169 | } else |
2e72b634 KS |
4170 | BUG(); |
4171 | ||
2e72b634 | 4172 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4173 | if (thresholds->primary) |
2e72b634 KS |
4174 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4175 | ||
2c488db2 | 4176 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4177 | |
4178 | /* Allocate memory for new array of thresholds */ | |
67b8046f | 4179 | new = kmalloc(struct_size(new, entries, size), GFP_KERNEL); |
2c488db2 | 4180 | if (!new) { |
2e72b634 KS |
4181 | ret = -ENOMEM; |
4182 | goto unlock; | |
4183 | } | |
2c488db2 | 4184 | new->size = size; |
2e72b634 KS |
4185 | |
4186 | /* Copy thresholds (if any) to new array */ | |
e90342e6 GS |
4187 | if (thresholds->primary) |
4188 | memcpy(new->entries, thresholds->primary->entries, | |
4189 | flex_array_size(new, entries, size - 1)); | |
2c488db2 | 4190 | |
2e72b634 | 4191 | /* Add new threshold */ |
2c488db2 KS |
4192 | new->entries[size - 1].eventfd = eventfd; |
4193 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4194 | |
4195 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
61e604e6 | 4196 | sort(new->entries, size, sizeof(*new->entries), |
2e72b634 KS |
4197 | compare_thresholds, NULL); |
4198 | ||
4199 | /* Find current threshold */ | |
2c488db2 | 4200 | new->current_threshold = -1; |
2e72b634 | 4201 | for (i = 0; i < size; i++) { |
748dad36 | 4202 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4203 | /* |
2c488db2 KS |
4204 | * new->current_threshold will not be used until |
4205 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4206 | * it here. |
4207 | */ | |
2c488db2 | 4208 | ++new->current_threshold; |
748dad36 SZ |
4209 | } else |
4210 | break; | |
2e72b634 KS |
4211 | } |
4212 | ||
2c488db2 KS |
4213 | /* Free old spare buffer and save old primary buffer as spare */ |
4214 | kfree(thresholds->spare); | |
4215 | thresholds->spare = thresholds->primary; | |
4216 | ||
4217 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4218 | |
907860ed | 4219 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4220 | synchronize_rcu(); |
4221 | ||
2e72b634 KS |
4222 | unlock: |
4223 | mutex_unlock(&memcg->thresholds_lock); | |
4224 | ||
4225 | return ret; | |
4226 | } | |
4227 | ||
59b6f873 | 4228 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4229 | struct eventfd_ctx *eventfd, const char *args) |
4230 | { | |
59b6f873 | 4231 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4232 | } |
4233 | ||
59b6f873 | 4234 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4235 | struct eventfd_ctx *eventfd, const char *args) |
4236 | { | |
59b6f873 | 4237 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4238 | } |
4239 | ||
59b6f873 | 4240 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4241 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4242 | { |
2c488db2 KS |
4243 | struct mem_cgroup_thresholds *thresholds; |
4244 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 4245 | unsigned long usage; |
7d36665a | 4246 | int i, j, size, entries; |
2e72b634 KS |
4247 | |
4248 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4249 | |
4250 | if (type == _MEM) { | |
2c488db2 | 4251 | thresholds = &memcg->thresholds; |
ce00a967 | 4252 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4253 | } else if (type == _MEMSWAP) { |
2c488db2 | 4254 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4255 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4256 | } else |
2e72b634 KS |
4257 | BUG(); |
4258 | ||
371528ca AV |
4259 | if (!thresholds->primary) |
4260 | goto unlock; | |
4261 | ||
2e72b634 KS |
4262 | /* Check if a threshold crossed before removing */ |
4263 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4264 | ||
4265 | /* Calculate new number of threshold */ | |
7d36665a | 4266 | size = entries = 0; |
2c488db2 KS |
4267 | for (i = 0; i < thresholds->primary->size; i++) { |
4268 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 | 4269 | size++; |
7d36665a CX |
4270 | else |
4271 | entries++; | |
2e72b634 KS |
4272 | } |
4273 | ||
2c488db2 | 4274 | new = thresholds->spare; |
907860ed | 4275 | |
7d36665a CX |
4276 | /* If no items related to eventfd have been cleared, nothing to do */ |
4277 | if (!entries) | |
4278 | goto unlock; | |
4279 | ||
2e72b634 KS |
4280 | /* Set thresholds array to NULL if we don't have thresholds */ |
4281 | if (!size) { | |
2c488db2 KS |
4282 | kfree(new); |
4283 | new = NULL; | |
907860ed | 4284 | goto swap_buffers; |
2e72b634 KS |
4285 | } |
4286 | ||
2c488db2 | 4287 | new->size = size; |
2e72b634 KS |
4288 | |
4289 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4290 | new->current_threshold = -1; |
4291 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4292 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4293 | continue; |
4294 | ||
2c488db2 | 4295 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4296 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4297 | /* |
2c488db2 | 4298 | * new->current_threshold will not be used |
2e72b634 KS |
4299 | * until rcu_assign_pointer(), so it's safe to increment |
4300 | * it here. | |
4301 | */ | |
2c488db2 | 4302 | ++new->current_threshold; |
2e72b634 KS |
4303 | } |
4304 | j++; | |
4305 | } | |
4306 | ||
907860ed | 4307 | swap_buffers: |
2c488db2 KS |
4308 | /* Swap primary and spare array */ |
4309 | thresholds->spare = thresholds->primary; | |
8c757763 | 4310 | |
2c488db2 | 4311 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4312 | |
907860ed | 4313 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4314 | synchronize_rcu(); |
6611d8d7 MC |
4315 | |
4316 | /* If all events are unregistered, free the spare array */ | |
4317 | if (!new) { | |
4318 | kfree(thresholds->spare); | |
4319 | thresholds->spare = NULL; | |
4320 | } | |
371528ca | 4321 | unlock: |
2e72b634 | 4322 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4323 | } |
c1e862c1 | 4324 | |
59b6f873 | 4325 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4326 | struct eventfd_ctx *eventfd) |
4327 | { | |
59b6f873 | 4328 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4329 | } |
4330 | ||
59b6f873 | 4331 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4332 | struct eventfd_ctx *eventfd) |
4333 | { | |
59b6f873 | 4334 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4335 | } |
4336 | ||
59b6f873 | 4337 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4338 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4339 | { |
9490ff27 | 4340 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4341 | |
9490ff27 KH |
4342 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4343 | if (!event) | |
4344 | return -ENOMEM; | |
4345 | ||
1af8efe9 | 4346 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4347 | |
4348 | event->eventfd = eventfd; | |
4349 | list_add(&event->list, &memcg->oom_notify); | |
4350 | ||
4351 | /* already in OOM ? */ | |
c2b42d3c | 4352 | if (memcg->under_oom) |
9490ff27 | 4353 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4354 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4355 | |
4356 | return 0; | |
4357 | } | |
4358 | ||
59b6f873 | 4359 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4360 | struct eventfd_ctx *eventfd) |
9490ff27 | 4361 | { |
9490ff27 | 4362 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4363 | |
1af8efe9 | 4364 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4365 | |
c0ff4b85 | 4366 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4367 | if (ev->eventfd == eventfd) { |
4368 | list_del(&ev->list); | |
4369 | kfree(ev); | |
4370 | } | |
4371 | } | |
4372 | ||
1af8efe9 | 4373 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4374 | } |
4375 | ||
2da8ca82 | 4376 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4377 | { |
aa9694bb | 4378 | struct mem_cgroup *memcg = mem_cgroup_from_seq(sf); |
3c11ecf4 | 4379 | |
791badbd | 4380 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 4381 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
4382 | seq_printf(sf, "oom_kill %lu\n", |
4383 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
4384 | return 0; |
4385 | } | |
4386 | ||
182446d0 | 4387 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4388 | struct cftype *cft, u64 val) |
4389 | { | |
182446d0 | 4390 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4391 | |
4392 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
a4792030 | 4393 | if (mem_cgroup_is_root(memcg) || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4394 | return -EINVAL; |
4395 | ||
c0ff4b85 | 4396 | memcg->oom_kill_disable = val; |
4d845ebf | 4397 | if (!val) |
c0ff4b85 | 4398 | memcg_oom_recover(memcg); |
3dae7fec | 4399 | |
3c11ecf4 KH |
4400 | return 0; |
4401 | } | |
4402 | ||
52ebea74 TH |
4403 | #ifdef CONFIG_CGROUP_WRITEBACK |
4404 | ||
3a8e9ac8 TH |
4405 | #include <trace/events/writeback.h> |
4406 | ||
841710aa TH |
4407 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
4408 | { | |
4409 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
4410 | } | |
4411 | ||
4412 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4413 | { | |
4414 | wb_domain_exit(&memcg->cgwb_domain); | |
4415 | } | |
4416 | ||
2529bb3a TH |
4417 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4418 | { | |
4419 | wb_domain_size_changed(&memcg->cgwb_domain); | |
4420 | } | |
4421 | ||
841710aa TH |
4422 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
4423 | { | |
4424 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4425 | ||
4426 | if (!memcg->css.parent) | |
4427 | return NULL; | |
4428 | ||
4429 | return &memcg->cgwb_domain; | |
4430 | } | |
4431 | ||
c2aa723a TH |
4432 | /** |
4433 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
4434 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
4435 | * @pfilepages: out parameter for number of file pages |
4436 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
4437 | * @pdirty: out parameter for number of dirty pages |
4438 | * @pwriteback: out parameter for number of pages under writeback | |
4439 | * | |
c5edf9cd TH |
4440 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
4441 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
4442 | * is a bit more involved. | |
c2aa723a | 4443 | * |
c5edf9cd TH |
4444 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
4445 | * headroom is calculated as the lowest headroom of itself and the | |
4446 | * ancestors. Note that this doesn't consider the actual amount of | |
4447 | * available memory in the system. The caller should further cap | |
4448 | * *@pheadroom accordingly. | |
c2aa723a | 4449 | */ |
c5edf9cd TH |
4450 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
4451 | unsigned long *pheadroom, unsigned long *pdirty, | |
4452 | unsigned long *pwriteback) | |
c2aa723a TH |
4453 | { |
4454 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4455 | struct mem_cgroup *parent; | |
c2aa723a | 4456 | |
2d146aa3 | 4457 | cgroup_rstat_flush_irqsafe(memcg->css.cgroup); |
c2aa723a | 4458 | |
2d146aa3 JW |
4459 | *pdirty = memcg_page_state(memcg, NR_FILE_DIRTY); |
4460 | *pwriteback = memcg_page_state(memcg, NR_WRITEBACK); | |
4461 | *pfilepages = memcg_page_state(memcg, NR_INACTIVE_FILE) + | |
4462 | memcg_page_state(memcg, NR_ACTIVE_FILE); | |
c2aa723a | 4463 | |
2d146aa3 | 4464 | *pheadroom = PAGE_COUNTER_MAX; |
c2aa723a | 4465 | while ((parent = parent_mem_cgroup(memcg))) { |
15b42562 | 4466 | unsigned long ceiling = min(READ_ONCE(memcg->memory.max), |
d1663a90 | 4467 | READ_ONCE(memcg->memory.high)); |
c2aa723a TH |
4468 | unsigned long used = page_counter_read(&memcg->memory); |
4469 | ||
c5edf9cd | 4470 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
4471 | memcg = parent; |
4472 | } | |
c2aa723a TH |
4473 | } |
4474 | ||
97b27821 TH |
4475 | /* |
4476 | * Foreign dirty flushing | |
4477 | * | |
4478 | * There's an inherent mismatch between memcg and writeback. The former | |
f0953a1b | 4479 | * tracks ownership per-page while the latter per-inode. This was a |
97b27821 TH |
4480 | * deliberate design decision because honoring per-page ownership in the |
4481 | * writeback path is complicated, may lead to higher CPU and IO overheads | |
4482 | * and deemed unnecessary given that write-sharing an inode across | |
4483 | * different cgroups isn't a common use-case. | |
4484 | * | |
4485 | * Combined with inode majority-writer ownership switching, this works well | |
4486 | * enough in most cases but there are some pathological cases. For | |
4487 | * example, let's say there are two cgroups A and B which keep writing to | |
4488 | * different but confined parts of the same inode. B owns the inode and | |
4489 | * A's memory is limited far below B's. A's dirty ratio can rise enough to | |
4490 | * trigger balance_dirty_pages() sleeps but B's can be low enough to avoid | |
4491 | * triggering background writeback. A will be slowed down without a way to | |
4492 | * make writeback of the dirty pages happen. | |
4493 | * | |
f0953a1b | 4494 | * Conditions like the above can lead to a cgroup getting repeatedly and |
97b27821 | 4495 | * severely throttled after making some progress after each |
f0953a1b | 4496 | * dirty_expire_interval while the underlying IO device is almost |
97b27821 TH |
4497 | * completely idle. |
4498 | * | |
4499 | * Solving this problem completely requires matching the ownership tracking | |
4500 | * granularities between memcg and writeback in either direction. However, | |
4501 | * the more egregious behaviors can be avoided by simply remembering the | |
4502 | * most recent foreign dirtying events and initiating remote flushes on | |
4503 | * them when local writeback isn't enough to keep the memory clean enough. | |
4504 | * | |
4505 | * The following two functions implement such mechanism. When a foreign | |
4506 | * page - a page whose memcg and writeback ownerships don't match - is | |
4507 | * dirtied, mem_cgroup_track_foreign_dirty() records the inode owning | |
4508 | * bdi_writeback on the page owning memcg. When balance_dirty_pages() | |
4509 | * decides that the memcg needs to sleep due to high dirty ratio, it calls | |
4510 | * mem_cgroup_flush_foreign() which queues writeback on the recorded | |
4511 | * foreign bdi_writebacks which haven't expired. Both the numbers of | |
4512 | * recorded bdi_writebacks and concurrent in-flight foreign writebacks are | |
4513 | * limited to MEMCG_CGWB_FRN_CNT. | |
4514 | * | |
4515 | * The mechanism only remembers IDs and doesn't hold any object references. | |
4516 | * As being wrong occasionally doesn't matter, updates and accesses to the | |
4517 | * records are lockless and racy. | |
4518 | */ | |
4519 | void mem_cgroup_track_foreign_dirty_slowpath(struct page *page, | |
4520 | struct bdi_writeback *wb) | |
4521 | { | |
bcfe06bf | 4522 | struct mem_cgroup *memcg = page_memcg(page); |
97b27821 TH |
4523 | struct memcg_cgwb_frn *frn; |
4524 | u64 now = get_jiffies_64(); | |
4525 | u64 oldest_at = now; | |
4526 | int oldest = -1; | |
4527 | int i; | |
4528 | ||
3a8e9ac8 TH |
4529 | trace_track_foreign_dirty(page, wb); |
4530 | ||
97b27821 TH |
4531 | /* |
4532 | * Pick the slot to use. If there is already a slot for @wb, keep | |
4533 | * using it. If not replace the oldest one which isn't being | |
4534 | * written out. | |
4535 | */ | |
4536 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4537 | frn = &memcg->cgwb_frn[i]; | |
4538 | if (frn->bdi_id == wb->bdi->id && | |
4539 | frn->memcg_id == wb->memcg_css->id) | |
4540 | break; | |
4541 | if (time_before64(frn->at, oldest_at) && | |
4542 | atomic_read(&frn->done.cnt) == 1) { | |
4543 | oldest = i; | |
4544 | oldest_at = frn->at; | |
4545 | } | |
4546 | } | |
4547 | ||
4548 | if (i < MEMCG_CGWB_FRN_CNT) { | |
4549 | /* | |
4550 | * Re-using an existing one. Update timestamp lazily to | |
4551 | * avoid making the cacheline hot. We want them to be | |
4552 | * reasonably up-to-date and significantly shorter than | |
4553 | * dirty_expire_interval as that's what expires the record. | |
4554 | * Use the shorter of 1s and dirty_expire_interval / 8. | |
4555 | */ | |
4556 | unsigned long update_intv = | |
4557 | min_t(unsigned long, HZ, | |
4558 | msecs_to_jiffies(dirty_expire_interval * 10) / 8); | |
4559 | ||
4560 | if (time_before64(frn->at, now - update_intv)) | |
4561 | frn->at = now; | |
4562 | } else if (oldest >= 0) { | |
4563 | /* replace the oldest free one */ | |
4564 | frn = &memcg->cgwb_frn[oldest]; | |
4565 | frn->bdi_id = wb->bdi->id; | |
4566 | frn->memcg_id = wb->memcg_css->id; | |
4567 | frn->at = now; | |
4568 | } | |
4569 | } | |
4570 | ||
4571 | /* issue foreign writeback flushes for recorded foreign dirtying events */ | |
4572 | void mem_cgroup_flush_foreign(struct bdi_writeback *wb) | |
4573 | { | |
4574 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4575 | unsigned long intv = msecs_to_jiffies(dirty_expire_interval * 10); | |
4576 | u64 now = jiffies_64; | |
4577 | int i; | |
4578 | ||
4579 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4580 | struct memcg_cgwb_frn *frn = &memcg->cgwb_frn[i]; | |
4581 | ||
4582 | /* | |
4583 | * If the record is older than dirty_expire_interval, | |
4584 | * writeback on it has already started. No need to kick it | |
4585 | * off again. Also, don't start a new one if there's | |
4586 | * already one in flight. | |
4587 | */ | |
4588 | if (time_after64(frn->at, now - intv) && | |
4589 | atomic_read(&frn->done.cnt) == 1) { | |
4590 | frn->at = 0; | |
3a8e9ac8 | 4591 | trace_flush_foreign(wb, frn->bdi_id, frn->memcg_id); |
7490a2d2 | 4592 | cgroup_writeback_by_id(frn->bdi_id, frn->memcg_id, |
97b27821 TH |
4593 | WB_REASON_FOREIGN_FLUSH, |
4594 | &frn->done); | |
4595 | } | |
4596 | } | |
4597 | } | |
4598 | ||
841710aa TH |
4599 | #else /* CONFIG_CGROUP_WRITEBACK */ |
4600 | ||
4601 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
4602 | { | |
4603 | return 0; | |
4604 | } | |
4605 | ||
4606 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4607 | { | |
4608 | } | |
4609 | ||
2529bb3a TH |
4610 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4611 | { | |
4612 | } | |
4613 | ||
52ebea74 TH |
4614 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
4615 | ||
3bc942f3 TH |
4616 | /* |
4617 | * DO NOT USE IN NEW FILES. | |
4618 | * | |
4619 | * "cgroup.event_control" implementation. | |
4620 | * | |
4621 | * This is way over-engineered. It tries to support fully configurable | |
4622 | * events for each user. Such level of flexibility is completely | |
4623 | * unnecessary especially in the light of the planned unified hierarchy. | |
4624 | * | |
4625 | * Please deprecate this and replace with something simpler if at all | |
4626 | * possible. | |
4627 | */ | |
4628 | ||
79bd9814 TH |
4629 | /* |
4630 | * Unregister event and free resources. | |
4631 | * | |
4632 | * Gets called from workqueue. | |
4633 | */ | |
3bc942f3 | 4634 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4635 | { |
3bc942f3 TH |
4636 | struct mem_cgroup_event *event = |
4637 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4638 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4639 | |
4640 | remove_wait_queue(event->wqh, &event->wait); | |
4641 | ||
59b6f873 | 4642 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4643 | |
4644 | /* Notify userspace the event is going away. */ | |
4645 | eventfd_signal(event->eventfd, 1); | |
4646 | ||
4647 | eventfd_ctx_put(event->eventfd); | |
4648 | kfree(event); | |
59b6f873 | 4649 | css_put(&memcg->css); |
79bd9814 TH |
4650 | } |
4651 | ||
4652 | /* | |
a9a08845 | 4653 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
4654 | * |
4655 | * Called with wqh->lock held and interrupts disabled. | |
4656 | */ | |
ac6424b9 | 4657 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 4658 | int sync, void *key) |
79bd9814 | 4659 | { |
3bc942f3 TH |
4660 | struct mem_cgroup_event *event = |
4661 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4662 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 4663 | __poll_t flags = key_to_poll(key); |
79bd9814 | 4664 | |
a9a08845 | 4665 | if (flags & EPOLLHUP) { |
79bd9814 TH |
4666 | /* |
4667 | * If the event has been detached at cgroup removal, we | |
4668 | * can simply return knowing the other side will cleanup | |
4669 | * for us. | |
4670 | * | |
4671 | * We can't race against event freeing since the other | |
4672 | * side will require wqh->lock via remove_wait_queue(), | |
4673 | * which we hold. | |
4674 | */ | |
fba94807 | 4675 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4676 | if (!list_empty(&event->list)) { |
4677 | list_del_init(&event->list); | |
4678 | /* | |
4679 | * We are in atomic context, but cgroup_event_remove() | |
4680 | * may sleep, so we have to call it in workqueue. | |
4681 | */ | |
4682 | schedule_work(&event->remove); | |
4683 | } | |
fba94807 | 4684 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4685 | } |
4686 | ||
4687 | return 0; | |
4688 | } | |
4689 | ||
3bc942f3 | 4690 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4691 | wait_queue_head_t *wqh, poll_table *pt) |
4692 | { | |
3bc942f3 TH |
4693 | struct mem_cgroup_event *event = |
4694 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4695 | |
4696 | event->wqh = wqh; | |
4697 | add_wait_queue(wqh, &event->wait); | |
4698 | } | |
4699 | ||
4700 | /* | |
3bc942f3 TH |
4701 | * DO NOT USE IN NEW FILES. |
4702 | * | |
79bd9814 TH |
4703 | * Parse input and register new cgroup event handler. |
4704 | * | |
4705 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4706 | * Interpretation of args is defined by control file implementation. | |
4707 | */ | |
451af504 TH |
4708 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4709 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4710 | { |
451af504 | 4711 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4712 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4713 | struct mem_cgroup_event *event; |
79bd9814 TH |
4714 | struct cgroup_subsys_state *cfile_css; |
4715 | unsigned int efd, cfd; | |
4716 | struct fd efile; | |
4717 | struct fd cfile; | |
fba94807 | 4718 | const char *name; |
79bd9814 TH |
4719 | char *endp; |
4720 | int ret; | |
4721 | ||
451af504 TH |
4722 | buf = strstrip(buf); |
4723 | ||
4724 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4725 | if (*endp != ' ') |
4726 | return -EINVAL; | |
451af504 | 4727 | buf = endp + 1; |
79bd9814 | 4728 | |
451af504 | 4729 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4730 | if ((*endp != ' ') && (*endp != '\0')) |
4731 | return -EINVAL; | |
451af504 | 4732 | buf = endp + 1; |
79bd9814 TH |
4733 | |
4734 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4735 | if (!event) | |
4736 | return -ENOMEM; | |
4737 | ||
59b6f873 | 4738 | event->memcg = memcg; |
79bd9814 | 4739 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4740 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4741 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4742 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4743 | |
4744 | efile = fdget(efd); | |
4745 | if (!efile.file) { | |
4746 | ret = -EBADF; | |
4747 | goto out_kfree; | |
4748 | } | |
4749 | ||
4750 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4751 | if (IS_ERR(event->eventfd)) { | |
4752 | ret = PTR_ERR(event->eventfd); | |
4753 | goto out_put_efile; | |
4754 | } | |
4755 | ||
4756 | cfile = fdget(cfd); | |
4757 | if (!cfile.file) { | |
4758 | ret = -EBADF; | |
4759 | goto out_put_eventfd; | |
4760 | } | |
4761 | ||
4762 | /* the process need read permission on control file */ | |
4763 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
02f92b38 | 4764 | ret = file_permission(cfile.file, MAY_READ); |
79bd9814 TH |
4765 | if (ret < 0) |
4766 | goto out_put_cfile; | |
4767 | ||
fba94807 TH |
4768 | /* |
4769 | * Determine the event callbacks and set them in @event. This used | |
4770 | * to be done via struct cftype but cgroup core no longer knows | |
4771 | * about these events. The following is crude but the whole thing | |
4772 | * is for compatibility anyway. | |
3bc942f3 TH |
4773 | * |
4774 | * DO NOT ADD NEW FILES. | |
fba94807 | 4775 | */ |
b583043e | 4776 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4777 | |
4778 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4779 | event->register_event = mem_cgroup_usage_register_event; | |
4780 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4781 | } else if (!strcmp(name, "memory.oom_control")) { | |
4782 | event->register_event = mem_cgroup_oom_register_event; | |
4783 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4784 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4785 | event->register_event = vmpressure_register_event; | |
4786 | event->unregister_event = vmpressure_unregister_event; | |
4787 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4788 | event->register_event = memsw_cgroup_usage_register_event; |
4789 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4790 | } else { |
4791 | ret = -EINVAL; | |
4792 | goto out_put_cfile; | |
4793 | } | |
4794 | ||
79bd9814 | 4795 | /* |
b5557c4c TH |
4796 | * Verify @cfile should belong to @css. Also, remaining events are |
4797 | * automatically removed on cgroup destruction but the removal is | |
4798 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4799 | */ |
b583043e | 4800 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4801 | &memory_cgrp_subsys); |
79bd9814 | 4802 | ret = -EINVAL; |
5a17f543 | 4803 | if (IS_ERR(cfile_css)) |
79bd9814 | 4804 | goto out_put_cfile; |
5a17f543 TH |
4805 | if (cfile_css != css) { |
4806 | css_put(cfile_css); | |
79bd9814 | 4807 | goto out_put_cfile; |
5a17f543 | 4808 | } |
79bd9814 | 4809 | |
451af504 | 4810 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4811 | if (ret) |
4812 | goto out_put_css; | |
4813 | ||
9965ed17 | 4814 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 4815 | |
4ba9515d | 4816 | spin_lock_irq(&memcg->event_list_lock); |
fba94807 | 4817 | list_add(&event->list, &memcg->event_list); |
4ba9515d | 4818 | spin_unlock_irq(&memcg->event_list_lock); |
79bd9814 TH |
4819 | |
4820 | fdput(cfile); | |
4821 | fdput(efile); | |
4822 | ||
451af504 | 4823 | return nbytes; |
79bd9814 TH |
4824 | |
4825 | out_put_css: | |
b5557c4c | 4826 | css_put(css); |
79bd9814 TH |
4827 | out_put_cfile: |
4828 | fdput(cfile); | |
4829 | out_put_eventfd: | |
4830 | eventfd_ctx_put(event->eventfd); | |
4831 | out_put_efile: | |
4832 | fdput(efile); | |
4833 | out_kfree: | |
4834 | kfree(event); | |
4835 | ||
4836 | return ret; | |
4837 | } | |
4838 | ||
241994ed | 4839 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4840 | { |
0eea1030 | 4841 | .name = "usage_in_bytes", |
8c7c6e34 | 4842 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4843 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4844 | }, |
c84872e1 PE |
4845 | { |
4846 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4847 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4848 | .write = mem_cgroup_reset, |
791badbd | 4849 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4850 | }, |
8cdea7c0 | 4851 | { |
0eea1030 | 4852 | .name = "limit_in_bytes", |
8c7c6e34 | 4853 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4854 | .write = mem_cgroup_write, |
791badbd | 4855 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4856 | }, |
296c81d8 BS |
4857 | { |
4858 | .name = "soft_limit_in_bytes", | |
4859 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4860 | .write = mem_cgroup_write, |
791badbd | 4861 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4862 | }, |
8cdea7c0 BS |
4863 | { |
4864 | .name = "failcnt", | |
8c7c6e34 | 4865 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4866 | .write = mem_cgroup_reset, |
791badbd | 4867 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4868 | }, |
d2ceb9b7 KH |
4869 | { |
4870 | .name = "stat", | |
2da8ca82 | 4871 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4872 | }, |
c1e862c1 KH |
4873 | { |
4874 | .name = "force_empty", | |
6770c64e | 4875 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4876 | }, |
18f59ea7 BS |
4877 | { |
4878 | .name = "use_hierarchy", | |
4879 | .write_u64 = mem_cgroup_hierarchy_write, | |
4880 | .read_u64 = mem_cgroup_hierarchy_read, | |
4881 | }, | |
79bd9814 | 4882 | { |
3bc942f3 | 4883 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4884 | .write = memcg_write_event_control, |
7dbdb199 | 4885 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 4886 | }, |
a7885eb8 KM |
4887 | { |
4888 | .name = "swappiness", | |
4889 | .read_u64 = mem_cgroup_swappiness_read, | |
4890 | .write_u64 = mem_cgroup_swappiness_write, | |
4891 | }, | |
7dc74be0 DN |
4892 | { |
4893 | .name = "move_charge_at_immigrate", | |
4894 | .read_u64 = mem_cgroup_move_charge_read, | |
4895 | .write_u64 = mem_cgroup_move_charge_write, | |
4896 | }, | |
9490ff27 KH |
4897 | { |
4898 | .name = "oom_control", | |
2da8ca82 | 4899 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4900 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4901 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4902 | }, | |
70ddf637 AV |
4903 | { |
4904 | .name = "pressure_level", | |
70ddf637 | 4905 | }, |
406eb0c9 YH |
4906 | #ifdef CONFIG_NUMA |
4907 | { | |
4908 | .name = "numa_stat", | |
2da8ca82 | 4909 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4910 | }, |
4911 | #endif | |
510fc4e1 GC |
4912 | { |
4913 | .name = "kmem.limit_in_bytes", | |
4914 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4915 | .write = mem_cgroup_write, |
791badbd | 4916 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4917 | }, |
4918 | { | |
4919 | .name = "kmem.usage_in_bytes", | |
4920 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4921 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4922 | }, |
4923 | { | |
4924 | .name = "kmem.failcnt", | |
4925 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4926 | .write = mem_cgroup_reset, |
791badbd | 4927 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4928 | }, |
4929 | { | |
4930 | .name = "kmem.max_usage_in_bytes", | |
4931 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4932 | .write = mem_cgroup_reset, |
791badbd | 4933 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4934 | }, |
a87425a3 YS |
4935 | #if defined(CONFIG_MEMCG_KMEM) && \ |
4936 | (defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)) | |
749c5415 GC |
4937 | { |
4938 | .name = "kmem.slabinfo", | |
b047501c | 4939 | .seq_show = memcg_slab_show, |
749c5415 GC |
4940 | }, |
4941 | #endif | |
d55f90bf VD |
4942 | { |
4943 | .name = "kmem.tcp.limit_in_bytes", | |
4944 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
4945 | .write = mem_cgroup_write, | |
4946 | .read_u64 = mem_cgroup_read_u64, | |
4947 | }, | |
4948 | { | |
4949 | .name = "kmem.tcp.usage_in_bytes", | |
4950 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
4951 | .read_u64 = mem_cgroup_read_u64, | |
4952 | }, | |
4953 | { | |
4954 | .name = "kmem.tcp.failcnt", | |
4955 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
4956 | .write = mem_cgroup_reset, | |
4957 | .read_u64 = mem_cgroup_read_u64, | |
4958 | }, | |
4959 | { | |
4960 | .name = "kmem.tcp.max_usage_in_bytes", | |
4961 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
4962 | .write = mem_cgroup_reset, | |
4963 | .read_u64 = mem_cgroup_read_u64, | |
4964 | }, | |
6bc10349 | 4965 | { }, /* terminate */ |
af36f906 | 4966 | }; |
8c7c6e34 | 4967 | |
73f576c0 JW |
4968 | /* |
4969 | * Private memory cgroup IDR | |
4970 | * | |
4971 | * Swap-out records and page cache shadow entries need to store memcg | |
4972 | * references in constrained space, so we maintain an ID space that is | |
4973 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
4974 | * memory-controlled cgroups to 64k. | |
4975 | * | |
b8f2935f | 4976 | * However, there usually are many references to the offline CSS after |
73f576c0 JW |
4977 | * the cgroup has been destroyed, such as page cache or reclaimable |
4978 | * slab objects, that don't need to hang on to the ID. We want to keep | |
4979 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
4980 | * relatively small ID space and prevent the creation of new cgroups | |
4981 | * even when there are much fewer than 64k cgroups - possibly none. | |
4982 | * | |
4983 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
4984 | * be freed and recycled when it's no longer needed, which is usually | |
4985 | * when the CSS is offlined. | |
4986 | * | |
4987 | * The only exception to that are records of swapped out tmpfs/shmem | |
4988 | * pages that need to be attributed to live ancestors on swapin. But | |
4989 | * those references are manageable from userspace. | |
4990 | */ | |
4991 | ||
4992 | static DEFINE_IDR(mem_cgroup_idr); | |
4993 | ||
7e97de0b KT |
4994 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
4995 | { | |
4996 | if (memcg->id.id > 0) { | |
4997 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
4998 | memcg->id.id = 0; | |
4999 | } | |
5000 | } | |
5001 | ||
c1514c0a VF |
5002 | static void __maybe_unused mem_cgroup_id_get_many(struct mem_cgroup *memcg, |
5003 | unsigned int n) | |
73f576c0 | 5004 | { |
1c2d479a | 5005 | refcount_add(n, &memcg->id.ref); |
73f576c0 JW |
5006 | } |
5007 | ||
615d66c3 | 5008 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 5009 | { |
1c2d479a | 5010 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 5011 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
5012 | |
5013 | /* Memcg ID pins CSS */ | |
5014 | css_put(&memcg->css); | |
5015 | } | |
5016 | } | |
5017 | ||
615d66c3 VD |
5018 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) |
5019 | { | |
5020 | mem_cgroup_id_put_many(memcg, 1); | |
5021 | } | |
5022 | ||
73f576c0 JW |
5023 | /** |
5024 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
5025 | * @id: the memcg id to look up | |
5026 | * | |
5027 | * Caller must hold rcu_read_lock(). | |
5028 | */ | |
5029 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
5030 | { | |
5031 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5032 | return idr_find(&mem_cgroup_idr, id); | |
5033 | } | |
5034 | ||
ef8f2327 | 5035 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
5036 | { |
5037 | struct mem_cgroup_per_node *pn; | |
ef8f2327 | 5038 | int tmp = node; |
1ecaab2b KH |
5039 | /* |
5040 | * This routine is called against possible nodes. | |
5041 | * But it's BUG to call kmalloc() against offline node. | |
5042 | * | |
5043 | * TODO: this routine can waste much memory for nodes which will | |
5044 | * never be onlined. It's better to use memory hotplug callback | |
5045 | * function. | |
5046 | */ | |
41e3355d KH |
5047 | if (!node_state(node, N_NORMAL_MEMORY)) |
5048 | tmp = -1; | |
17295c88 | 5049 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
5050 | if (!pn) |
5051 | return 1; | |
1ecaab2b | 5052 | |
7e1c0d6f SB |
5053 | pn->lruvec_stats_percpu = alloc_percpu_gfp(struct lruvec_stats_percpu, |
5054 | GFP_KERNEL_ACCOUNT); | |
5055 | if (!pn->lruvec_stats_percpu) { | |
00f3ca2c JW |
5056 | kfree(pn); |
5057 | return 1; | |
5058 | } | |
5059 | ||
ef8f2327 MG |
5060 | lruvec_init(&pn->lruvec); |
5061 | pn->usage_in_excess = 0; | |
5062 | pn->on_tree = false; | |
5063 | pn->memcg = memcg; | |
5064 | ||
54f72fe0 | 5065 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
5066 | return 0; |
5067 | } | |
5068 | ||
ef8f2327 | 5069 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 5070 | { |
00f3ca2c JW |
5071 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
5072 | ||
4eaf431f MH |
5073 | if (!pn) |
5074 | return; | |
5075 | ||
7e1c0d6f | 5076 | free_percpu(pn->lruvec_stats_percpu); |
00f3ca2c | 5077 | kfree(pn); |
1ecaab2b KH |
5078 | } |
5079 | ||
40e952f9 | 5080 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 5081 | { |
c8b2a36f | 5082 | int node; |
59927fb9 | 5083 | |
c8b2a36f | 5084 | for_each_node(node) |
ef8f2327 | 5085 | free_mem_cgroup_per_node_info(memcg, node); |
871789d4 | 5086 | free_percpu(memcg->vmstats_percpu); |
8ff69e2c | 5087 | kfree(memcg); |
59927fb9 | 5088 | } |
3afe36b1 | 5089 | |
40e952f9 TE |
5090 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
5091 | { | |
5092 | memcg_wb_domain_exit(memcg); | |
5093 | __mem_cgroup_free(memcg); | |
5094 | } | |
5095 | ||
0b8f73e1 | 5096 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 5097 | { |
d142e3e6 | 5098 | struct mem_cgroup *memcg; |
b9726c26 | 5099 | unsigned int size; |
6d12e2d8 | 5100 | int node; |
97b27821 | 5101 | int __maybe_unused i; |
11d67612 | 5102 | long error = -ENOMEM; |
8cdea7c0 | 5103 | |
0b8f73e1 JW |
5104 | size = sizeof(struct mem_cgroup); |
5105 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
5106 | ||
5107 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 5108 | if (!memcg) |
11d67612 | 5109 | return ERR_PTR(error); |
0b8f73e1 | 5110 | |
73f576c0 JW |
5111 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
5112 | 1, MEM_CGROUP_ID_MAX, | |
5113 | GFP_KERNEL); | |
11d67612 YS |
5114 | if (memcg->id.id < 0) { |
5115 | error = memcg->id.id; | |
73f576c0 | 5116 | goto fail; |
11d67612 | 5117 | } |
73f576c0 | 5118 | |
3e38e0aa RG |
5119 | memcg->vmstats_percpu = alloc_percpu_gfp(struct memcg_vmstats_percpu, |
5120 | GFP_KERNEL_ACCOUNT); | |
871789d4 | 5121 | if (!memcg->vmstats_percpu) |
0b8f73e1 | 5122 | goto fail; |
78fb7466 | 5123 | |
3ed28fa1 | 5124 | for_each_node(node) |
ef8f2327 | 5125 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 5126 | goto fail; |
f64c3f54 | 5127 | |
0b8f73e1 JW |
5128 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
5129 | goto fail; | |
28dbc4b6 | 5130 | |
f7e1cb6e | 5131 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 | 5132 | INIT_LIST_HEAD(&memcg->oom_notify); |
d142e3e6 GC |
5133 | mutex_init(&memcg->thresholds_lock); |
5134 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 5135 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
5136 | INIT_LIST_HEAD(&memcg->event_list); |
5137 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 5138 | memcg->socket_pressure = jiffies; |
84c07d11 | 5139 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 5140 | memcg->kmemcg_id = -1; |
bf4f0599 | 5141 | INIT_LIST_HEAD(&memcg->objcg_list); |
900a38f0 | 5142 | #endif |
52ebea74 TH |
5143 | #ifdef CONFIG_CGROUP_WRITEBACK |
5144 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
97b27821 TH |
5145 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) |
5146 | memcg->cgwb_frn[i].done = | |
5147 | __WB_COMPLETION_INIT(&memcg_cgwb_frn_waitq); | |
87eaceb3 YS |
5148 | #endif |
5149 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5150 | spin_lock_init(&memcg->deferred_split_queue.split_queue_lock); | |
5151 | INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue); | |
5152 | memcg->deferred_split_queue.split_queue_len = 0; | |
52ebea74 | 5153 | #endif |
73f576c0 | 5154 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
5155 | return memcg; |
5156 | fail: | |
7e97de0b | 5157 | mem_cgroup_id_remove(memcg); |
40e952f9 | 5158 | __mem_cgroup_free(memcg); |
11d67612 | 5159 | return ERR_PTR(error); |
d142e3e6 GC |
5160 | } |
5161 | ||
0b8f73e1 JW |
5162 | static struct cgroup_subsys_state * __ref |
5163 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 5164 | { |
0b8f73e1 | 5165 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
b87d8cef | 5166 | struct mem_cgroup *memcg, *old_memcg; |
0b8f73e1 | 5167 | long error = -ENOMEM; |
d142e3e6 | 5168 | |
b87d8cef | 5169 | old_memcg = set_active_memcg(parent); |
0b8f73e1 | 5170 | memcg = mem_cgroup_alloc(); |
b87d8cef | 5171 | set_active_memcg(old_memcg); |
11d67612 YS |
5172 | if (IS_ERR(memcg)) |
5173 | return ERR_CAST(memcg); | |
d142e3e6 | 5174 | |
d1663a90 | 5175 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
0b8f73e1 | 5176 | memcg->soft_limit = PAGE_COUNTER_MAX; |
4b82ab4f | 5177 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
0b8f73e1 JW |
5178 | if (parent) { |
5179 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
5180 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
bef8620c | 5181 | |
3e32cb2e | 5182 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 5183 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e | 5184 | page_counter_init(&memcg->kmem, &parent->kmem); |
0db15298 | 5185 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 5186 | } else { |
bef8620c RG |
5187 | page_counter_init(&memcg->memory, NULL); |
5188 | page_counter_init(&memcg->swap, NULL); | |
5189 | page_counter_init(&memcg->kmem, NULL); | |
5190 | page_counter_init(&memcg->tcpmem, NULL); | |
d6441637 | 5191 | |
0b8f73e1 JW |
5192 | root_mem_cgroup = memcg; |
5193 | return &memcg->css; | |
5194 | } | |
5195 | ||
bef8620c | 5196 | /* The following stuff does not apply to the root */ |
b313aeee | 5197 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
5198 | if (error) |
5199 | goto fail; | |
127424c8 | 5200 | |
f7e1cb6e | 5201 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5202 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 5203 | |
0b8f73e1 JW |
5204 | return &memcg->css; |
5205 | fail: | |
7e97de0b | 5206 | mem_cgroup_id_remove(memcg); |
0b8f73e1 | 5207 | mem_cgroup_free(memcg); |
11d67612 | 5208 | return ERR_PTR(error); |
0b8f73e1 JW |
5209 | } |
5210 | ||
73f576c0 | 5211 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 5212 | { |
58fa2a55 VD |
5213 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5214 | ||
0a4465d3 | 5215 | /* |
e4262c4f | 5216 | * A memcg must be visible for expand_shrinker_info() |
0a4465d3 KT |
5217 | * by the time the maps are allocated. So, we allocate maps |
5218 | * here, when for_each_mem_cgroup() can't skip it. | |
5219 | */ | |
e4262c4f | 5220 | if (alloc_shrinker_info(memcg)) { |
0a4465d3 KT |
5221 | mem_cgroup_id_remove(memcg); |
5222 | return -ENOMEM; | |
5223 | } | |
5224 | ||
73f576c0 | 5225 | /* Online state pins memcg ID, memcg ID pins CSS */ |
1c2d479a | 5226 | refcount_set(&memcg->id.ref, 1); |
73f576c0 | 5227 | css_get(css); |
aa48e47e SB |
5228 | |
5229 | if (unlikely(mem_cgroup_is_root(memcg))) | |
5230 | queue_delayed_work(system_unbound_wq, &stats_flush_dwork, | |
5231 | 2UL*HZ); | |
2f7dd7a4 | 5232 | return 0; |
8cdea7c0 BS |
5233 | } |
5234 | ||
eb95419b | 5235 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 5236 | { |
eb95419b | 5237 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5238 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
5239 | |
5240 | /* | |
5241 | * Unregister events and notify userspace. | |
5242 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
5243 | * directory to avoid race between userspace and kernelspace. | |
5244 | */ | |
4ba9515d | 5245 | spin_lock_irq(&memcg->event_list_lock); |
fba94807 | 5246 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { |
79bd9814 TH |
5247 | list_del_init(&event->list); |
5248 | schedule_work(&event->remove); | |
5249 | } | |
4ba9515d | 5250 | spin_unlock_irq(&memcg->event_list_lock); |
ec64f515 | 5251 | |
bf8d5d52 | 5252 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5253 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 5254 | |
567e9ab2 | 5255 | memcg_offline_kmem(memcg); |
a178015c | 5256 | reparent_shrinker_deferred(memcg); |
52ebea74 | 5257 | wb_memcg_offline(memcg); |
73f576c0 | 5258 | |
591edfb1 RG |
5259 | drain_all_stock(memcg); |
5260 | ||
73f576c0 | 5261 | mem_cgroup_id_put(memcg); |
df878fb0 KH |
5262 | } |
5263 | ||
6df38689 VD |
5264 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
5265 | { | |
5266 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5267 | ||
5268 | invalidate_reclaim_iterators(memcg); | |
5269 | } | |
5270 | ||
eb95419b | 5271 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 5272 | { |
eb95419b | 5273 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
97b27821 | 5274 | int __maybe_unused i; |
c268e994 | 5275 | |
97b27821 TH |
5276 | #ifdef CONFIG_CGROUP_WRITEBACK |
5277 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) | |
5278 | wb_wait_for_completion(&memcg->cgwb_frn[i].done); | |
5279 | #endif | |
f7e1cb6e | 5280 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5281 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 5282 | |
0db15298 | 5283 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 5284 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 5285 | |
0b8f73e1 JW |
5286 | vmpressure_cleanup(&memcg->vmpressure); |
5287 | cancel_work_sync(&memcg->high_work); | |
5288 | mem_cgroup_remove_from_trees(memcg); | |
e4262c4f | 5289 | free_shrinker_info(memcg); |
d886f4e4 | 5290 | memcg_free_kmem(memcg); |
0b8f73e1 | 5291 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
5292 | } |
5293 | ||
1ced953b TH |
5294 | /** |
5295 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
5296 | * @css: the target css | |
5297 | * | |
5298 | * Reset the states of the mem_cgroup associated with @css. This is | |
5299 | * invoked when the userland requests disabling on the default hierarchy | |
5300 | * but the memcg is pinned through dependency. The memcg should stop | |
5301 | * applying policies and should revert to the vanilla state as it may be | |
5302 | * made visible again. | |
5303 | * | |
5304 | * The current implementation only resets the essential configurations. | |
5305 | * This needs to be expanded to cover all the visible parts. | |
5306 | */ | |
5307 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
5308 | { | |
5309 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5310 | ||
bbec2e15 RG |
5311 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
5312 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
bbec2e15 RG |
5313 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); |
5314 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 5315 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5316 | page_counter_set_low(&memcg->memory, 0); |
d1663a90 | 5317 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
24d404dc | 5318 | memcg->soft_limit = PAGE_COUNTER_MAX; |
4b82ab4f | 5319 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
2529bb3a | 5320 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
5321 | } |
5322 | ||
aa48e47e SB |
5323 | void mem_cgroup_flush_stats(void) |
5324 | { | |
5325 | if (!spin_trylock(&stats_flush_lock)) | |
5326 | return; | |
5327 | ||
5328 | cgroup_rstat_flush_irqsafe(root_mem_cgroup->css.cgroup); | |
5329 | spin_unlock(&stats_flush_lock); | |
5330 | } | |
5331 | ||
5332 | static void flush_memcg_stats_dwork(struct work_struct *w) | |
5333 | { | |
5334 | mem_cgroup_flush_stats(); | |
5335 | queue_delayed_work(system_unbound_wq, &stats_flush_dwork, 2UL*HZ); | |
5336 | } | |
5337 | ||
2d146aa3 JW |
5338 | static void mem_cgroup_css_rstat_flush(struct cgroup_subsys_state *css, int cpu) |
5339 | { | |
5340 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5341 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); | |
5342 | struct memcg_vmstats_percpu *statc; | |
5343 | long delta, v; | |
7e1c0d6f | 5344 | int i, nid; |
2d146aa3 JW |
5345 | |
5346 | statc = per_cpu_ptr(memcg->vmstats_percpu, cpu); | |
5347 | ||
5348 | for (i = 0; i < MEMCG_NR_STAT; i++) { | |
5349 | /* | |
5350 | * Collect the aggregated propagation counts of groups | |
5351 | * below us. We're in a per-cpu loop here and this is | |
5352 | * a global counter, so the first cycle will get them. | |
5353 | */ | |
5354 | delta = memcg->vmstats.state_pending[i]; | |
5355 | if (delta) | |
5356 | memcg->vmstats.state_pending[i] = 0; | |
5357 | ||
5358 | /* Add CPU changes on this level since the last flush */ | |
5359 | v = READ_ONCE(statc->state[i]); | |
5360 | if (v != statc->state_prev[i]) { | |
5361 | delta += v - statc->state_prev[i]; | |
5362 | statc->state_prev[i] = v; | |
5363 | } | |
5364 | ||
5365 | if (!delta) | |
5366 | continue; | |
5367 | ||
5368 | /* Aggregate counts on this level and propagate upwards */ | |
5369 | memcg->vmstats.state[i] += delta; | |
5370 | if (parent) | |
5371 | parent->vmstats.state_pending[i] += delta; | |
5372 | } | |
5373 | ||
5374 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { | |
5375 | delta = memcg->vmstats.events_pending[i]; | |
5376 | if (delta) | |
5377 | memcg->vmstats.events_pending[i] = 0; | |
5378 | ||
5379 | v = READ_ONCE(statc->events[i]); | |
5380 | if (v != statc->events_prev[i]) { | |
5381 | delta += v - statc->events_prev[i]; | |
5382 | statc->events_prev[i] = v; | |
5383 | } | |
5384 | ||
5385 | if (!delta) | |
5386 | continue; | |
5387 | ||
5388 | memcg->vmstats.events[i] += delta; | |
5389 | if (parent) | |
5390 | parent->vmstats.events_pending[i] += delta; | |
5391 | } | |
7e1c0d6f SB |
5392 | |
5393 | for_each_node_state(nid, N_MEMORY) { | |
5394 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[nid]; | |
5395 | struct mem_cgroup_per_node *ppn = NULL; | |
5396 | struct lruvec_stats_percpu *lstatc; | |
5397 | ||
5398 | if (parent) | |
5399 | ppn = parent->nodeinfo[nid]; | |
5400 | ||
5401 | lstatc = per_cpu_ptr(pn->lruvec_stats_percpu, cpu); | |
5402 | ||
5403 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) { | |
5404 | delta = pn->lruvec_stats.state_pending[i]; | |
5405 | if (delta) | |
5406 | pn->lruvec_stats.state_pending[i] = 0; | |
5407 | ||
5408 | v = READ_ONCE(lstatc->state[i]); | |
5409 | if (v != lstatc->state_prev[i]) { | |
5410 | delta += v - lstatc->state_prev[i]; | |
5411 | lstatc->state_prev[i] = v; | |
5412 | } | |
5413 | ||
5414 | if (!delta) | |
5415 | continue; | |
5416 | ||
5417 | pn->lruvec_stats.state[i] += delta; | |
5418 | if (ppn) | |
5419 | ppn->lruvec_stats.state_pending[i] += delta; | |
5420 | } | |
5421 | } | |
2d146aa3 JW |
5422 | } |
5423 | ||
02491447 | 5424 | #ifdef CONFIG_MMU |
7dc74be0 | 5425 | /* Handlers for move charge at task migration. */ |
854ffa8d | 5426 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 5427 | { |
05b84301 | 5428 | int ret; |
9476db97 | 5429 | |
d0164adc MG |
5430 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
5431 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 5432 | if (!ret) { |
854ffa8d | 5433 | mc.precharge += count; |
854ffa8d DN |
5434 | return ret; |
5435 | } | |
9476db97 | 5436 | |
3674534b | 5437 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 5438 | while (count--) { |
3674534b | 5439 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 5440 | if (ret) |
38c5d72f | 5441 | return ret; |
854ffa8d | 5442 | mc.precharge++; |
9476db97 | 5443 | cond_resched(); |
854ffa8d | 5444 | } |
9476db97 | 5445 | return 0; |
4ffef5fe DN |
5446 | } |
5447 | ||
4ffef5fe DN |
5448 | union mc_target { |
5449 | struct page *page; | |
02491447 | 5450 | swp_entry_t ent; |
4ffef5fe DN |
5451 | }; |
5452 | ||
4ffef5fe | 5453 | enum mc_target_type { |
8d32ff84 | 5454 | MC_TARGET_NONE = 0, |
4ffef5fe | 5455 | MC_TARGET_PAGE, |
02491447 | 5456 | MC_TARGET_SWAP, |
c733a828 | 5457 | MC_TARGET_DEVICE, |
4ffef5fe DN |
5458 | }; |
5459 | ||
90254a65 DN |
5460 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5461 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5462 | { |
25b2995a | 5463 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5464 | |
90254a65 DN |
5465 | if (!page || !page_mapped(page)) |
5466 | return NULL; | |
5467 | if (PageAnon(page)) { | |
1dfab5ab | 5468 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5469 | return NULL; |
1dfab5ab JW |
5470 | } else { |
5471 | if (!(mc.flags & MOVE_FILE)) | |
5472 | return NULL; | |
5473 | } | |
90254a65 DN |
5474 | if (!get_page_unless_zero(page)) |
5475 | return NULL; | |
5476 | ||
5477 | return page; | |
5478 | } | |
5479 | ||
c733a828 | 5480 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 5481 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 5482 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 5483 | { |
90254a65 DN |
5484 | struct page *page = NULL; |
5485 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5486 | ||
9a137153 | 5487 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5488 | return NULL; |
c733a828 JG |
5489 | |
5490 | /* | |
5491 | * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to | |
5492 | * a device and because they are not accessible by CPU they are store | |
5493 | * as special swap entry in the CPU page table. | |
5494 | */ | |
5495 | if (is_device_private_entry(ent)) { | |
af5cdaf8 | 5496 | page = pfn_swap_entry_to_page(ent); |
c733a828 JG |
5497 | /* |
5498 | * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have | |
5499 | * a refcount of 1 when free (unlike normal page) | |
5500 | */ | |
5501 | if (!page_ref_add_unless(page, 1, 1)) | |
5502 | return NULL; | |
5503 | return page; | |
5504 | } | |
5505 | ||
9a137153 RC |
5506 | if (non_swap_entry(ent)) |
5507 | return NULL; | |
5508 | ||
4b91355e KH |
5509 | /* |
5510 | * Because lookup_swap_cache() updates some statistics counter, | |
5511 | * we call find_get_page() with swapper_space directly. | |
5512 | */ | |
f6ab1f7f | 5513 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
2d1c4980 | 5514 | entry->val = ent.val; |
90254a65 DN |
5515 | |
5516 | return page; | |
5517 | } | |
4b91355e KH |
5518 | #else |
5519 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 5520 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
5521 | { |
5522 | return NULL; | |
5523 | } | |
5524 | #endif | |
90254a65 | 5525 | |
87946a72 DN |
5526 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5527 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5528 | { | |
87946a72 DN |
5529 | if (!vma->vm_file) /* anonymous vma */ |
5530 | return NULL; | |
1dfab5ab | 5531 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
5532 | return NULL; |
5533 | ||
87946a72 | 5534 | /* page is moved even if it's not RSS of this task(page-faulted). */ |
aa3b1895 | 5535 | /* shmem/tmpfs may report page out on swap: account for that too. */ |
f5df8635 MWO |
5536 | return find_get_incore_page(vma->vm_file->f_mapping, |
5537 | linear_page_index(vma, addr)); | |
87946a72 DN |
5538 | } |
5539 | ||
b1b0deab CG |
5540 | /** |
5541 | * mem_cgroup_move_account - move account of the page | |
5542 | * @page: the page | |
25843c2b | 5543 | * @compound: charge the page as compound or small page |
b1b0deab CG |
5544 | * @from: mem_cgroup which the page is moved from. |
5545 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
5546 | * | |
3ac808fd | 5547 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
5548 | * |
5549 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
5550 | * from old cgroup. | |
5551 | */ | |
5552 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 5553 | bool compound, |
b1b0deab CG |
5554 | struct mem_cgroup *from, |
5555 | struct mem_cgroup *to) | |
5556 | { | |
ae8af438 KK |
5557 | struct lruvec *from_vec, *to_vec; |
5558 | struct pglist_data *pgdat; | |
6c357848 | 5559 | unsigned int nr_pages = compound ? thp_nr_pages(page) : 1; |
8e88bd2d | 5560 | int nid, ret; |
b1b0deab CG |
5561 | |
5562 | VM_BUG_ON(from == to); | |
5563 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 5564 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
5565 | |
5566 | /* | |
6a93ca8f | 5567 | * Prevent mem_cgroup_migrate() from looking at |
bcfe06bf | 5568 | * page's memory cgroup of its source page while we change it. |
b1b0deab | 5569 | */ |
f627c2f5 | 5570 | ret = -EBUSY; |
b1b0deab CG |
5571 | if (!trylock_page(page)) |
5572 | goto out; | |
5573 | ||
5574 | ret = -EINVAL; | |
bcfe06bf | 5575 | if (page_memcg(page) != from) |
b1b0deab CG |
5576 | goto out_unlock; |
5577 | ||
ae8af438 | 5578 | pgdat = page_pgdat(page); |
867e5e1d JW |
5579 | from_vec = mem_cgroup_lruvec(from, pgdat); |
5580 | to_vec = mem_cgroup_lruvec(to, pgdat); | |
ae8af438 | 5581 | |
abb242f5 | 5582 | lock_page_memcg(page); |
b1b0deab | 5583 | |
be5d0a74 JW |
5584 | if (PageAnon(page)) { |
5585 | if (page_mapped(page)) { | |
5586 | __mod_lruvec_state(from_vec, NR_ANON_MAPPED, -nr_pages); | |
5587 | __mod_lruvec_state(to_vec, NR_ANON_MAPPED, nr_pages); | |
468c3982 | 5588 | if (PageTransHuge(page)) { |
69473e5d MS |
5589 | __mod_lruvec_state(from_vec, NR_ANON_THPS, |
5590 | -nr_pages); | |
5591 | __mod_lruvec_state(to_vec, NR_ANON_THPS, | |
5592 | nr_pages); | |
468c3982 | 5593 | } |
be5d0a74 JW |
5594 | } |
5595 | } else { | |
0d1c2072 JW |
5596 | __mod_lruvec_state(from_vec, NR_FILE_PAGES, -nr_pages); |
5597 | __mod_lruvec_state(to_vec, NR_FILE_PAGES, nr_pages); | |
5598 | ||
5599 | if (PageSwapBacked(page)) { | |
5600 | __mod_lruvec_state(from_vec, NR_SHMEM, -nr_pages); | |
5601 | __mod_lruvec_state(to_vec, NR_SHMEM, nr_pages); | |
5602 | } | |
5603 | ||
49e50d27 JW |
5604 | if (page_mapped(page)) { |
5605 | __mod_lruvec_state(from_vec, NR_FILE_MAPPED, -nr_pages); | |
5606 | __mod_lruvec_state(to_vec, NR_FILE_MAPPED, nr_pages); | |
5607 | } | |
b1b0deab | 5608 | |
49e50d27 JW |
5609 | if (PageDirty(page)) { |
5610 | struct address_space *mapping = page_mapping(page); | |
c4843a75 | 5611 | |
f56753ac | 5612 | if (mapping_can_writeback(mapping)) { |
49e50d27 JW |
5613 | __mod_lruvec_state(from_vec, NR_FILE_DIRTY, |
5614 | -nr_pages); | |
5615 | __mod_lruvec_state(to_vec, NR_FILE_DIRTY, | |
5616 | nr_pages); | |
5617 | } | |
c4843a75 GT |
5618 | } |
5619 | } | |
5620 | ||
b1b0deab | 5621 | if (PageWriteback(page)) { |
ae8af438 KK |
5622 | __mod_lruvec_state(from_vec, NR_WRITEBACK, -nr_pages); |
5623 | __mod_lruvec_state(to_vec, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
5624 | } |
5625 | ||
5626 | /* | |
abb242f5 JW |
5627 | * All state has been migrated, let's switch to the new memcg. |
5628 | * | |
bcfe06bf | 5629 | * It is safe to change page's memcg here because the page |
abb242f5 JW |
5630 | * is referenced, charged, isolated, and locked: we can't race |
5631 | * with (un)charging, migration, LRU putback, or anything else | |
bcfe06bf | 5632 | * that would rely on a stable page's memory cgroup. |
abb242f5 JW |
5633 | * |
5634 | * Note that lock_page_memcg is a memcg lock, not a page lock, | |
bcfe06bf | 5635 | * to save space. As soon as we switch page's memory cgroup to a |
abb242f5 JW |
5636 | * new memcg that isn't locked, the above state can change |
5637 | * concurrently again. Make sure we're truly done with it. | |
b1b0deab | 5638 | */ |
abb242f5 | 5639 | smp_mb(); |
b1b0deab | 5640 | |
1a3e1f40 JW |
5641 | css_get(&to->css); |
5642 | css_put(&from->css); | |
5643 | ||
bcfe06bf | 5644 | page->memcg_data = (unsigned long)to; |
87eaceb3 | 5645 | |
abb242f5 | 5646 | __unlock_page_memcg(from); |
b1b0deab CG |
5647 | |
5648 | ret = 0; | |
8e88bd2d | 5649 | nid = page_to_nid(page); |
b1b0deab CG |
5650 | |
5651 | local_irq_disable(); | |
6e0110c2 | 5652 | mem_cgroup_charge_statistics(to, nr_pages); |
8e88bd2d | 5653 | memcg_check_events(to, nid); |
6e0110c2 | 5654 | mem_cgroup_charge_statistics(from, -nr_pages); |
8e88bd2d | 5655 | memcg_check_events(from, nid); |
b1b0deab CG |
5656 | local_irq_enable(); |
5657 | out_unlock: | |
5658 | unlock_page(page); | |
5659 | out: | |
5660 | return ret; | |
5661 | } | |
5662 | ||
7cf7806c LR |
5663 | /** |
5664 | * get_mctgt_type - get target type of moving charge | |
5665 | * @vma: the vma the pte to be checked belongs | |
5666 | * @addr: the address corresponding to the pte to be checked | |
5667 | * @ptent: the pte to be checked | |
5668 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
5669 | * | |
5670 | * Returns | |
5671 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5672 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5673 | * move charge. if @target is not NULL, the page is stored in target->page | |
5674 | * with extra refcnt got(Callers should handle it). | |
5675 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
5676 | * target for charge migration. if @target is not NULL, the entry is stored | |
5677 | * in target->ent. | |
25b2995a CH |
5678 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PRIVATE |
5679 | * (so ZONE_DEVICE page and thus not on the lru). | |
df6ad698 JG |
5680 | * For now we such page is charge like a regular page would be as for all |
5681 | * intent and purposes it is just special memory taking the place of a | |
5682 | * regular page. | |
c733a828 JG |
5683 | * |
5684 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
5685 | * |
5686 | * Called with pte lock held. | |
5687 | */ | |
5688 | ||
8d32ff84 | 5689 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5690 | unsigned long addr, pte_t ptent, union mc_target *target) |
5691 | { | |
5692 | struct page *page = NULL; | |
8d32ff84 | 5693 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5694 | swp_entry_t ent = { .val = 0 }; |
5695 | ||
5696 | if (pte_present(ptent)) | |
5697 | page = mc_handle_present_pte(vma, addr, ptent); | |
5698 | else if (is_swap_pte(ptent)) | |
48406ef8 | 5699 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 5700 | else if (pte_none(ptent)) |
87946a72 | 5701 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
5702 | |
5703 | if (!page && !ent.val) | |
8d32ff84 | 5704 | return ret; |
02491447 | 5705 | if (page) { |
02491447 | 5706 | /* |
0a31bc97 | 5707 | * Do only loose check w/o serialization. |
1306a85a | 5708 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 5709 | * not under LRU exclusion. |
02491447 | 5710 | */ |
bcfe06bf | 5711 | if (page_memcg(page) == mc.from) { |
02491447 | 5712 | ret = MC_TARGET_PAGE; |
25b2995a | 5713 | if (is_device_private_page(page)) |
c733a828 | 5714 | ret = MC_TARGET_DEVICE; |
02491447 DN |
5715 | if (target) |
5716 | target->page = page; | |
5717 | } | |
5718 | if (!ret || !target) | |
5719 | put_page(page); | |
5720 | } | |
3e14a57b HY |
5721 | /* |
5722 | * There is a swap entry and a page doesn't exist or isn't charged. | |
5723 | * But we cannot move a tail-page in a THP. | |
5724 | */ | |
5725 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 5726 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5727 | ret = MC_TARGET_SWAP; |
5728 | if (target) | |
5729 | target->ent = ent; | |
4ffef5fe | 5730 | } |
4ffef5fe DN |
5731 | return ret; |
5732 | } | |
5733 | ||
12724850 NH |
5734 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5735 | /* | |
d6810d73 HY |
5736 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
5737 | * not support them for now. | |
12724850 NH |
5738 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
5739 | */ | |
5740 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5741 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5742 | { | |
5743 | struct page *page = NULL; | |
12724850 NH |
5744 | enum mc_target_type ret = MC_TARGET_NONE; |
5745 | ||
84c3fc4e ZY |
5746 | if (unlikely(is_swap_pmd(pmd))) { |
5747 | VM_BUG_ON(thp_migration_supported() && | |
5748 | !is_pmd_migration_entry(pmd)); | |
5749 | return ret; | |
5750 | } | |
12724850 | 5751 | page = pmd_page(pmd); |
309381fe | 5752 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 5753 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 5754 | return ret; |
bcfe06bf | 5755 | if (page_memcg(page) == mc.from) { |
12724850 NH |
5756 | ret = MC_TARGET_PAGE; |
5757 | if (target) { | |
5758 | get_page(page); | |
5759 | target->page = page; | |
5760 | } | |
5761 | } | |
5762 | return ret; | |
5763 | } | |
5764 | #else | |
5765 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5766 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5767 | { | |
5768 | return MC_TARGET_NONE; | |
5769 | } | |
5770 | #endif | |
5771 | ||
4ffef5fe DN |
5772 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5773 | unsigned long addr, unsigned long end, | |
5774 | struct mm_walk *walk) | |
5775 | { | |
26bcd64a | 5776 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5777 | pte_t *pte; |
5778 | spinlock_t *ptl; | |
5779 | ||
b6ec57f4 KS |
5780 | ptl = pmd_trans_huge_lock(pmd, vma); |
5781 | if (ptl) { | |
c733a828 JG |
5782 | /* |
5783 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
25b2995a CH |
5784 | * support transparent huge page with MEMORY_DEVICE_PRIVATE but |
5785 | * this might change. | |
c733a828 | 5786 | */ |
12724850 NH |
5787 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5788 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5789 | spin_unlock(ptl); |
1a5a9906 | 5790 | return 0; |
12724850 | 5791 | } |
03319327 | 5792 | |
45f83cef AA |
5793 | if (pmd_trans_unstable(pmd)) |
5794 | return 0; | |
4ffef5fe DN |
5795 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5796 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5797 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5798 | mc.precharge++; /* increment precharge temporarily */ |
5799 | pte_unmap_unlock(pte - 1, ptl); | |
5800 | cond_resched(); | |
5801 | ||
7dc74be0 DN |
5802 | return 0; |
5803 | } | |
5804 | ||
7b86ac33 CH |
5805 | static const struct mm_walk_ops precharge_walk_ops = { |
5806 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5807 | }; | |
5808 | ||
4ffef5fe DN |
5809 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5810 | { | |
5811 | unsigned long precharge; | |
4ffef5fe | 5812 | |
d8ed45c5 | 5813 | mmap_read_lock(mm); |
7b86ac33 | 5814 | walk_page_range(mm, 0, mm->highest_vm_end, &precharge_walk_ops, NULL); |
d8ed45c5 | 5815 | mmap_read_unlock(mm); |
4ffef5fe DN |
5816 | |
5817 | precharge = mc.precharge; | |
5818 | mc.precharge = 0; | |
5819 | ||
5820 | return precharge; | |
5821 | } | |
5822 | ||
4ffef5fe DN |
5823 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5824 | { | |
dfe076b0 DN |
5825 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5826 | ||
5827 | VM_BUG_ON(mc.moving_task); | |
5828 | mc.moving_task = current; | |
5829 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5830 | } |
5831 | ||
dfe076b0 DN |
5832 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5833 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5834 | { |
2bd9bb20 KH |
5835 | struct mem_cgroup *from = mc.from; |
5836 | struct mem_cgroup *to = mc.to; | |
5837 | ||
4ffef5fe | 5838 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5839 | if (mc.precharge) { |
00501b53 | 5840 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5841 | mc.precharge = 0; |
5842 | } | |
5843 | /* | |
5844 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5845 | * we must uncharge here. | |
5846 | */ | |
5847 | if (mc.moved_charge) { | |
00501b53 | 5848 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5849 | mc.moved_charge = 0; |
4ffef5fe | 5850 | } |
483c30b5 DN |
5851 | /* we must fixup refcnts and charges */ |
5852 | if (mc.moved_swap) { | |
483c30b5 | 5853 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5854 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5855 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5856 | |
615d66c3 VD |
5857 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
5858 | ||
05b84301 | 5859 | /* |
3e32cb2e JW |
5860 | * we charged both to->memory and to->memsw, so we |
5861 | * should uncharge to->memory. | |
05b84301 | 5862 | */ |
ce00a967 | 5863 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5864 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5865 | ||
483c30b5 DN |
5866 | mc.moved_swap = 0; |
5867 | } | |
dfe076b0 DN |
5868 | memcg_oom_recover(from); |
5869 | memcg_oom_recover(to); | |
5870 | wake_up_all(&mc.waitq); | |
5871 | } | |
5872 | ||
5873 | static void mem_cgroup_clear_mc(void) | |
5874 | { | |
264a0ae1 TH |
5875 | struct mm_struct *mm = mc.mm; |
5876 | ||
dfe076b0 DN |
5877 | /* |
5878 | * we must clear moving_task before waking up waiters at the end of | |
5879 | * task migration. | |
5880 | */ | |
5881 | mc.moving_task = NULL; | |
5882 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5883 | spin_lock(&mc.lock); |
4ffef5fe DN |
5884 | mc.from = NULL; |
5885 | mc.to = NULL; | |
264a0ae1 | 5886 | mc.mm = NULL; |
2bd9bb20 | 5887 | spin_unlock(&mc.lock); |
264a0ae1 TH |
5888 | |
5889 | mmput(mm); | |
4ffef5fe DN |
5890 | } |
5891 | ||
1f7dd3e5 | 5892 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5893 | { |
1f7dd3e5 | 5894 | struct cgroup_subsys_state *css; |
eed67d75 | 5895 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 5896 | struct mem_cgroup *from; |
4530eddb | 5897 | struct task_struct *leader, *p; |
9f2115f9 | 5898 | struct mm_struct *mm; |
1dfab5ab | 5899 | unsigned long move_flags; |
9f2115f9 | 5900 | int ret = 0; |
7dc74be0 | 5901 | |
1f7dd3e5 TH |
5902 | /* charge immigration isn't supported on the default hierarchy */ |
5903 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
5904 | return 0; |
5905 | ||
4530eddb TH |
5906 | /* |
5907 | * Multi-process migrations only happen on the default hierarchy | |
5908 | * where charge immigration is not used. Perform charge | |
5909 | * immigration if @tset contains a leader and whine if there are | |
5910 | * multiple. | |
5911 | */ | |
5912 | p = NULL; | |
1f7dd3e5 | 5913 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
5914 | WARN_ON_ONCE(p); |
5915 | p = leader; | |
1f7dd3e5 | 5916 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
5917 | } |
5918 | if (!p) | |
5919 | return 0; | |
5920 | ||
1f7dd3e5 | 5921 | /* |
f0953a1b | 5922 | * We are now committed to this value whatever it is. Changes in this |
1f7dd3e5 TH |
5923 | * tunable will only affect upcoming migrations, not the current one. |
5924 | * So we need to save it, and keep it going. | |
5925 | */ | |
5926 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
5927 | if (!move_flags) | |
5928 | return 0; | |
5929 | ||
9f2115f9 TH |
5930 | from = mem_cgroup_from_task(p); |
5931 | ||
5932 | VM_BUG_ON(from == memcg); | |
5933 | ||
5934 | mm = get_task_mm(p); | |
5935 | if (!mm) | |
5936 | return 0; | |
5937 | /* We move charges only when we move a owner of the mm */ | |
5938 | if (mm->owner == p) { | |
5939 | VM_BUG_ON(mc.from); | |
5940 | VM_BUG_ON(mc.to); | |
5941 | VM_BUG_ON(mc.precharge); | |
5942 | VM_BUG_ON(mc.moved_charge); | |
5943 | VM_BUG_ON(mc.moved_swap); | |
5944 | ||
5945 | spin_lock(&mc.lock); | |
264a0ae1 | 5946 | mc.mm = mm; |
9f2115f9 TH |
5947 | mc.from = from; |
5948 | mc.to = memcg; | |
5949 | mc.flags = move_flags; | |
5950 | spin_unlock(&mc.lock); | |
5951 | /* We set mc.moving_task later */ | |
5952 | ||
5953 | ret = mem_cgroup_precharge_mc(mm); | |
5954 | if (ret) | |
5955 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
5956 | } else { |
5957 | mmput(mm); | |
7dc74be0 DN |
5958 | } |
5959 | return ret; | |
5960 | } | |
5961 | ||
1f7dd3e5 | 5962 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5963 | { |
4e2f245d JW |
5964 | if (mc.to) |
5965 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5966 | } |
5967 | ||
4ffef5fe DN |
5968 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5969 | unsigned long addr, unsigned long end, | |
5970 | struct mm_walk *walk) | |
7dc74be0 | 5971 | { |
4ffef5fe | 5972 | int ret = 0; |
26bcd64a | 5973 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5974 | pte_t *pte; |
5975 | spinlock_t *ptl; | |
12724850 NH |
5976 | enum mc_target_type target_type; |
5977 | union mc_target target; | |
5978 | struct page *page; | |
4ffef5fe | 5979 | |
b6ec57f4 KS |
5980 | ptl = pmd_trans_huge_lock(pmd, vma); |
5981 | if (ptl) { | |
62ade86a | 5982 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5983 | spin_unlock(ptl); |
12724850 NH |
5984 | return 0; |
5985 | } | |
5986 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5987 | if (target_type == MC_TARGET_PAGE) { | |
5988 | page = target.page; | |
5989 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 5990 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 5991 | mc.from, mc.to)) { |
12724850 NH |
5992 | mc.precharge -= HPAGE_PMD_NR; |
5993 | mc.moved_charge += HPAGE_PMD_NR; | |
5994 | } | |
5995 | putback_lru_page(page); | |
5996 | } | |
5997 | put_page(page); | |
c733a828 JG |
5998 | } else if (target_type == MC_TARGET_DEVICE) { |
5999 | page = target.page; | |
6000 | if (!mem_cgroup_move_account(page, true, | |
6001 | mc.from, mc.to)) { | |
6002 | mc.precharge -= HPAGE_PMD_NR; | |
6003 | mc.moved_charge += HPAGE_PMD_NR; | |
6004 | } | |
6005 | put_page(page); | |
12724850 | 6006 | } |
bf929152 | 6007 | spin_unlock(ptl); |
1a5a9906 | 6008 | return 0; |
12724850 NH |
6009 | } |
6010 | ||
45f83cef AA |
6011 | if (pmd_trans_unstable(pmd)) |
6012 | return 0; | |
4ffef5fe DN |
6013 | retry: |
6014 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
6015 | for (; addr != end; addr += PAGE_SIZE) { | |
6016 | pte_t ptent = *(pte++); | |
c733a828 | 6017 | bool device = false; |
02491447 | 6018 | swp_entry_t ent; |
4ffef5fe DN |
6019 | |
6020 | if (!mc.precharge) | |
6021 | break; | |
6022 | ||
8d32ff84 | 6023 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
6024 | case MC_TARGET_DEVICE: |
6025 | device = true; | |
e4a9bc58 | 6026 | fallthrough; |
4ffef5fe DN |
6027 | case MC_TARGET_PAGE: |
6028 | page = target.page; | |
53f9263b KS |
6029 | /* |
6030 | * We can have a part of the split pmd here. Moving it | |
6031 | * can be done but it would be too convoluted so simply | |
6032 | * ignore such a partial THP and keep it in original | |
6033 | * memcg. There should be somebody mapping the head. | |
6034 | */ | |
6035 | if (PageTransCompound(page)) | |
6036 | goto put; | |
c733a828 | 6037 | if (!device && isolate_lru_page(page)) |
4ffef5fe | 6038 | goto put; |
f627c2f5 KS |
6039 | if (!mem_cgroup_move_account(page, false, |
6040 | mc.from, mc.to)) { | |
4ffef5fe | 6041 | mc.precharge--; |
854ffa8d DN |
6042 | /* we uncharge from mc.from later. */ |
6043 | mc.moved_charge++; | |
4ffef5fe | 6044 | } |
c733a828 JG |
6045 | if (!device) |
6046 | putback_lru_page(page); | |
8d32ff84 | 6047 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
6048 | put_page(page); |
6049 | break; | |
02491447 DN |
6050 | case MC_TARGET_SWAP: |
6051 | ent = target.ent; | |
e91cbb42 | 6052 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6053 | mc.precharge--; |
8d22a935 HD |
6054 | mem_cgroup_id_get_many(mc.to, 1); |
6055 | /* we fixup other refcnts and charges later. */ | |
483c30b5 DN |
6056 | mc.moved_swap++; |
6057 | } | |
02491447 | 6058 | break; |
4ffef5fe DN |
6059 | default: |
6060 | break; | |
6061 | } | |
6062 | } | |
6063 | pte_unmap_unlock(pte - 1, ptl); | |
6064 | cond_resched(); | |
6065 | ||
6066 | if (addr != end) { | |
6067 | /* | |
6068 | * We have consumed all precharges we got in can_attach(). | |
6069 | * We try charge one by one, but don't do any additional | |
6070 | * charges to mc.to if we have failed in charge once in attach() | |
6071 | * phase. | |
6072 | */ | |
854ffa8d | 6073 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6074 | if (!ret) |
6075 | goto retry; | |
6076 | } | |
6077 | ||
6078 | return ret; | |
6079 | } | |
6080 | ||
7b86ac33 CH |
6081 | static const struct mm_walk_ops charge_walk_ops = { |
6082 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6083 | }; | |
6084 | ||
264a0ae1 | 6085 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 6086 | { |
4ffef5fe | 6087 | lru_add_drain_all(); |
312722cb | 6088 | /* |
81f8c3a4 JW |
6089 | * Signal lock_page_memcg() to take the memcg's move_lock |
6090 | * while we're moving its pages to another memcg. Then wait | |
6091 | * for already started RCU-only updates to finish. | |
312722cb JW |
6092 | */ |
6093 | atomic_inc(&mc.from->moving_account); | |
6094 | synchronize_rcu(); | |
dfe076b0 | 6095 | retry: |
d8ed45c5 | 6096 | if (unlikely(!mmap_read_trylock(mc.mm))) { |
dfe076b0 | 6097 | /* |
c1e8d7c6 | 6098 | * Someone who are holding the mmap_lock might be waiting in |
dfe076b0 DN |
6099 | * waitq. So we cancel all extra charges, wake up all waiters, |
6100 | * and retry. Because we cancel precharges, we might not be able | |
6101 | * to move enough charges, but moving charge is a best-effort | |
6102 | * feature anyway, so it wouldn't be a big problem. | |
6103 | */ | |
6104 | __mem_cgroup_clear_mc(); | |
6105 | cond_resched(); | |
6106 | goto retry; | |
6107 | } | |
26bcd64a NH |
6108 | /* |
6109 | * When we have consumed all precharges and failed in doing | |
6110 | * additional charge, the page walk just aborts. | |
6111 | */ | |
7b86ac33 CH |
6112 | walk_page_range(mc.mm, 0, mc.mm->highest_vm_end, &charge_walk_ops, |
6113 | NULL); | |
0247f3f4 | 6114 | |
d8ed45c5 | 6115 | mmap_read_unlock(mc.mm); |
312722cb | 6116 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
6117 | } |
6118 | ||
264a0ae1 | 6119 | static void mem_cgroup_move_task(void) |
67e465a7 | 6120 | { |
264a0ae1 TH |
6121 | if (mc.to) { |
6122 | mem_cgroup_move_charge(); | |
a433658c | 6123 | mem_cgroup_clear_mc(); |
264a0ae1 | 6124 | } |
67e465a7 | 6125 | } |
5cfb80a7 | 6126 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 6127 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6128 | { |
6129 | return 0; | |
6130 | } | |
1f7dd3e5 | 6131 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6132 | { |
6133 | } | |
264a0ae1 | 6134 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
6135 | { |
6136 | } | |
6137 | #endif | |
67e465a7 | 6138 | |
677dc973 CD |
6139 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
6140 | { | |
6141 | if (value == PAGE_COUNTER_MAX) | |
6142 | seq_puts(m, "max\n"); | |
6143 | else | |
6144 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); | |
6145 | ||
6146 | return 0; | |
6147 | } | |
6148 | ||
241994ed JW |
6149 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
6150 | struct cftype *cft) | |
6151 | { | |
f5fc3c5d JW |
6152 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
6153 | ||
6154 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
6155 | } |
6156 | ||
bf8d5d52 RG |
6157 | static int memory_min_show(struct seq_file *m, void *v) |
6158 | { | |
677dc973 CD |
6159 | return seq_puts_memcg_tunable(m, |
6160 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); | |
bf8d5d52 RG |
6161 | } |
6162 | ||
6163 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
6164 | char *buf, size_t nbytes, loff_t off) | |
6165 | { | |
6166 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6167 | unsigned long min; | |
6168 | int err; | |
6169 | ||
6170 | buf = strstrip(buf); | |
6171 | err = page_counter_memparse(buf, "max", &min); | |
6172 | if (err) | |
6173 | return err; | |
6174 | ||
6175 | page_counter_set_min(&memcg->memory, min); | |
6176 | ||
6177 | return nbytes; | |
6178 | } | |
6179 | ||
241994ed JW |
6180 | static int memory_low_show(struct seq_file *m, void *v) |
6181 | { | |
677dc973 CD |
6182 | return seq_puts_memcg_tunable(m, |
6183 | READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); | |
241994ed JW |
6184 | } |
6185 | ||
6186 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
6187 | char *buf, size_t nbytes, loff_t off) | |
6188 | { | |
6189 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6190 | unsigned long low; | |
6191 | int err; | |
6192 | ||
6193 | buf = strstrip(buf); | |
d2973697 | 6194 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
6195 | if (err) |
6196 | return err; | |
6197 | ||
23067153 | 6198 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
6199 | |
6200 | return nbytes; | |
6201 | } | |
6202 | ||
6203 | static int memory_high_show(struct seq_file *m, void *v) | |
6204 | { | |
d1663a90 JK |
6205 | return seq_puts_memcg_tunable(m, |
6206 | READ_ONCE(mem_cgroup_from_seq(m)->memory.high)); | |
241994ed JW |
6207 | } |
6208 | ||
6209 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
6210 | char *buf, size_t nbytes, loff_t off) | |
6211 | { | |
6212 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6213 | unsigned int nr_retries = MAX_RECLAIM_RETRIES; |
8c8c383c | 6214 | bool drained = false; |
241994ed JW |
6215 | unsigned long high; |
6216 | int err; | |
6217 | ||
6218 | buf = strstrip(buf); | |
d2973697 | 6219 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
6220 | if (err) |
6221 | return err; | |
6222 | ||
e82553c1 JW |
6223 | page_counter_set_high(&memcg->memory, high); |
6224 | ||
8c8c383c JW |
6225 | for (;;) { |
6226 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6227 | unsigned long reclaimed; | |
6228 | ||
6229 | if (nr_pages <= high) | |
6230 | break; | |
6231 | ||
6232 | if (signal_pending(current)) | |
6233 | break; | |
6234 | ||
6235 | if (!drained) { | |
6236 | drain_all_stock(memcg); | |
6237 | drained = true; | |
6238 | continue; | |
6239 | } | |
6240 | ||
6241 | reclaimed = try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
6242 | GFP_KERNEL, true); | |
6243 | ||
6244 | if (!reclaimed && !nr_retries--) | |
6245 | break; | |
6246 | } | |
588083bb | 6247 | |
19ce33ac | 6248 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
6249 | return nbytes; |
6250 | } | |
6251 | ||
6252 | static int memory_max_show(struct seq_file *m, void *v) | |
6253 | { | |
677dc973 CD |
6254 | return seq_puts_memcg_tunable(m, |
6255 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); | |
241994ed JW |
6256 | } |
6257 | ||
6258 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
6259 | char *buf, size_t nbytes, loff_t off) | |
6260 | { | |
6261 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6262 | unsigned int nr_reclaims = MAX_RECLAIM_RETRIES; |
b6e6edcf | 6263 | bool drained = false; |
241994ed JW |
6264 | unsigned long max; |
6265 | int err; | |
6266 | ||
6267 | buf = strstrip(buf); | |
d2973697 | 6268 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
6269 | if (err) |
6270 | return err; | |
6271 | ||
bbec2e15 | 6272 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
6273 | |
6274 | for (;;) { | |
6275 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6276 | ||
6277 | if (nr_pages <= max) | |
6278 | break; | |
6279 | ||
7249c9f0 | 6280 | if (signal_pending(current)) |
b6e6edcf | 6281 | break; |
b6e6edcf JW |
6282 | |
6283 | if (!drained) { | |
6284 | drain_all_stock(memcg); | |
6285 | drained = true; | |
6286 | continue; | |
6287 | } | |
6288 | ||
6289 | if (nr_reclaims) { | |
6290 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
6291 | GFP_KERNEL, true)) | |
6292 | nr_reclaims--; | |
6293 | continue; | |
6294 | } | |
6295 | ||
e27be240 | 6296 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
6297 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
6298 | break; | |
6299 | } | |
241994ed | 6300 | |
2529bb3a | 6301 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
6302 | return nbytes; |
6303 | } | |
6304 | ||
1e577f97 SB |
6305 | static void __memory_events_show(struct seq_file *m, atomic_long_t *events) |
6306 | { | |
6307 | seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW])); | |
6308 | seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH])); | |
6309 | seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX])); | |
6310 | seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM])); | |
6311 | seq_printf(m, "oom_kill %lu\n", | |
6312 | atomic_long_read(&events[MEMCG_OOM_KILL])); | |
6313 | } | |
6314 | ||
241994ed JW |
6315 | static int memory_events_show(struct seq_file *m, void *v) |
6316 | { | |
aa9694bb | 6317 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
241994ed | 6318 | |
1e577f97 SB |
6319 | __memory_events_show(m, memcg->memory_events); |
6320 | return 0; | |
6321 | } | |
6322 | ||
6323 | static int memory_events_local_show(struct seq_file *m, void *v) | |
6324 | { | |
6325 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
241994ed | 6326 | |
1e577f97 | 6327 | __memory_events_show(m, memcg->memory_events_local); |
241994ed JW |
6328 | return 0; |
6329 | } | |
6330 | ||
587d9f72 JW |
6331 | static int memory_stat_show(struct seq_file *m, void *v) |
6332 | { | |
aa9694bb | 6333 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
c8713d0b | 6334 | char *buf; |
1ff9e6e1 | 6335 | |
c8713d0b JW |
6336 | buf = memory_stat_format(memcg); |
6337 | if (!buf) | |
6338 | return -ENOMEM; | |
6339 | seq_puts(m, buf); | |
6340 | kfree(buf); | |
587d9f72 JW |
6341 | return 0; |
6342 | } | |
6343 | ||
5f9a4f4a | 6344 | #ifdef CONFIG_NUMA |
fff66b79 MS |
6345 | static inline unsigned long lruvec_page_state_output(struct lruvec *lruvec, |
6346 | int item) | |
6347 | { | |
6348 | return lruvec_page_state(lruvec, item) * memcg_page_state_unit(item); | |
6349 | } | |
6350 | ||
5f9a4f4a MS |
6351 | static int memory_numa_stat_show(struct seq_file *m, void *v) |
6352 | { | |
6353 | int i; | |
6354 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
6355 | ||
7e1c0d6f SB |
6356 | cgroup_rstat_flush(memcg->css.cgroup); |
6357 | ||
5f9a4f4a MS |
6358 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { |
6359 | int nid; | |
6360 | ||
6361 | if (memory_stats[i].idx >= NR_VM_NODE_STAT_ITEMS) | |
6362 | continue; | |
6363 | ||
6364 | seq_printf(m, "%s", memory_stats[i].name); | |
6365 | for_each_node_state(nid, N_MEMORY) { | |
6366 | u64 size; | |
6367 | struct lruvec *lruvec; | |
6368 | ||
6369 | lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); | |
fff66b79 MS |
6370 | size = lruvec_page_state_output(lruvec, |
6371 | memory_stats[i].idx); | |
5f9a4f4a MS |
6372 | seq_printf(m, " N%d=%llu", nid, size); |
6373 | } | |
6374 | seq_putc(m, '\n'); | |
6375 | } | |
6376 | ||
6377 | return 0; | |
6378 | } | |
6379 | #endif | |
6380 | ||
3d8b38eb RG |
6381 | static int memory_oom_group_show(struct seq_file *m, void *v) |
6382 | { | |
aa9694bb | 6383 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3d8b38eb RG |
6384 | |
6385 | seq_printf(m, "%d\n", memcg->oom_group); | |
6386 | ||
6387 | return 0; | |
6388 | } | |
6389 | ||
6390 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
6391 | char *buf, size_t nbytes, loff_t off) | |
6392 | { | |
6393 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6394 | int ret, oom_group; | |
6395 | ||
6396 | buf = strstrip(buf); | |
6397 | if (!buf) | |
6398 | return -EINVAL; | |
6399 | ||
6400 | ret = kstrtoint(buf, 0, &oom_group); | |
6401 | if (ret) | |
6402 | return ret; | |
6403 | ||
6404 | if (oom_group != 0 && oom_group != 1) | |
6405 | return -EINVAL; | |
6406 | ||
6407 | memcg->oom_group = oom_group; | |
6408 | ||
6409 | return nbytes; | |
6410 | } | |
6411 | ||
241994ed JW |
6412 | static struct cftype memory_files[] = { |
6413 | { | |
6414 | .name = "current", | |
f5fc3c5d | 6415 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
6416 | .read_u64 = memory_current_read, |
6417 | }, | |
bf8d5d52 RG |
6418 | { |
6419 | .name = "min", | |
6420 | .flags = CFTYPE_NOT_ON_ROOT, | |
6421 | .seq_show = memory_min_show, | |
6422 | .write = memory_min_write, | |
6423 | }, | |
241994ed JW |
6424 | { |
6425 | .name = "low", | |
6426 | .flags = CFTYPE_NOT_ON_ROOT, | |
6427 | .seq_show = memory_low_show, | |
6428 | .write = memory_low_write, | |
6429 | }, | |
6430 | { | |
6431 | .name = "high", | |
6432 | .flags = CFTYPE_NOT_ON_ROOT, | |
6433 | .seq_show = memory_high_show, | |
6434 | .write = memory_high_write, | |
6435 | }, | |
6436 | { | |
6437 | .name = "max", | |
6438 | .flags = CFTYPE_NOT_ON_ROOT, | |
6439 | .seq_show = memory_max_show, | |
6440 | .write = memory_max_write, | |
6441 | }, | |
6442 | { | |
6443 | .name = "events", | |
6444 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 6445 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
6446 | .seq_show = memory_events_show, |
6447 | }, | |
1e577f97 SB |
6448 | { |
6449 | .name = "events.local", | |
6450 | .flags = CFTYPE_NOT_ON_ROOT, | |
6451 | .file_offset = offsetof(struct mem_cgroup, events_local_file), | |
6452 | .seq_show = memory_events_local_show, | |
6453 | }, | |
587d9f72 JW |
6454 | { |
6455 | .name = "stat", | |
587d9f72 JW |
6456 | .seq_show = memory_stat_show, |
6457 | }, | |
5f9a4f4a MS |
6458 | #ifdef CONFIG_NUMA |
6459 | { | |
6460 | .name = "numa_stat", | |
6461 | .seq_show = memory_numa_stat_show, | |
6462 | }, | |
6463 | #endif | |
3d8b38eb RG |
6464 | { |
6465 | .name = "oom.group", | |
6466 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
6467 | .seq_show = memory_oom_group_show, | |
6468 | .write = memory_oom_group_write, | |
6469 | }, | |
241994ed JW |
6470 | { } /* terminate */ |
6471 | }; | |
6472 | ||
073219e9 | 6473 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 6474 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6475 | .css_online = mem_cgroup_css_online, |
92fb9748 | 6476 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 6477 | .css_released = mem_cgroup_css_released, |
92fb9748 | 6478 | .css_free = mem_cgroup_css_free, |
1ced953b | 6479 | .css_reset = mem_cgroup_css_reset, |
2d146aa3 | 6480 | .css_rstat_flush = mem_cgroup_css_rstat_flush, |
7dc74be0 DN |
6481 | .can_attach = mem_cgroup_can_attach, |
6482 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 6483 | .post_attach = mem_cgroup_move_task, |
241994ed JW |
6484 | .dfl_cftypes = memory_files, |
6485 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 6486 | .early_init = 0, |
8cdea7c0 | 6487 | }; |
c077719b | 6488 | |
bc50bcc6 JW |
6489 | /* |
6490 | * This function calculates an individual cgroup's effective | |
6491 | * protection which is derived from its own memory.min/low, its | |
6492 | * parent's and siblings' settings, as well as the actual memory | |
6493 | * distribution in the tree. | |
6494 | * | |
6495 | * The following rules apply to the effective protection values: | |
6496 | * | |
6497 | * 1. At the first level of reclaim, effective protection is equal to | |
6498 | * the declared protection in memory.min and memory.low. | |
6499 | * | |
6500 | * 2. To enable safe delegation of the protection configuration, at | |
6501 | * subsequent levels the effective protection is capped to the | |
6502 | * parent's effective protection. | |
6503 | * | |
6504 | * 3. To make complex and dynamic subtrees easier to configure, the | |
6505 | * user is allowed to overcommit the declared protection at a given | |
6506 | * level. If that is the case, the parent's effective protection is | |
6507 | * distributed to the children in proportion to how much protection | |
6508 | * they have declared and how much of it they are utilizing. | |
6509 | * | |
6510 | * This makes distribution proportional, but also work-conserving: | |
6511 | * if one cgroup claims much more protection than it uses memory, | |
6512 | * the unused remainder is available to its siblings. | |
6513 | * | |
6514 | * 4. Conversely, when the declared protection is undercommitted at a | |
6515 | * given level, the distribution of the larger parental protection | |
6516 | * budget is NOT proportional. A cgroup's protection from a sibling | |
6517 | * is capped to its own memory.min/low setting. | |
6518 | * | |
8a931f80 JW |
6519 | * 5. However, to allow protecting recursive subtrees from each other |
6520 | * without having to declare each individual cgroup's fixed share | |
6521 | * of the ancestor's claim to protection, any unutilized - | |
6522 | * "floating" - protection from up the tree is distributed in | |
6523 | * proportion to each cgroup's *usage*. This makes the protection | |
6524 | * neutral wrt sibling cgroups and lets them compete freely over | |
6525 | * the shared parental protection budget, but it protects the | |
6526 | * subtree as a whole from neighboring subtrees. | |
6527 | * | |
6528 | * Note that 4. and 5. are not in conflict: 4. is about protecting | |
6529 | * against immediate siblings whereas 5. is about protecting against | |
6530 | * neighboring subtrees. | |
bc50bcc6 JW |
6531 | */ |
6532 | static unsigned long effective_protection(unsigned long usage, | |
8a931f80 | 6533 | unsigned long parent_usage, |
bc50bcc6 JW |
6534 | unsigned long setting, |
6535 | unsigned long parent_effective, | |
6536 | unsigned long siblings_protected) | |
6537 | { | |
6538 | unsigned long protected; | |
8a931f80 | 6539 | unsigned long ep; |
bc50bcc6 JW |
6540 | |
6541 | protected = min(usage, setting); | |
6542 | /* | |
6543 | * If all cgroups at this level combined claim and use more | |
6544 | * protection then what the parent affords them, distribute | |
6545 | * shares in proportion to utilization. | |
6546 | * | |
6547 | * We are using actual utilization rather than the statically | |
6548 | * claimed protection in order to be work-conserving: claimed | |
6549 | * but unused protection is available to siblings that would | |
6550 | * otherwise get a smaller chunk than what they claimed. | |
6551 | */ | |
6552 | if (siblings_protected > parent_effective) | |
6553 | return protected * parent_effective / siblings_protected; | |
6554 | ||
6555 | /* | |
6556 | * Ok, utilized protection of all children is within what the | |
6557 | * parent affords them, so we know whatever this child claims | |
6558 | * and utilizes is effectively protected. | |
6559 | * | |
6560 | * If there is unprotected usage beyond this value, reclaim | |
6561 | * will apply pressure in proportion to that amount. | |
6562 | * | |
6563 | * If there is unutilized protection, the cgroup will be fully | |
6564 | * shielded from reclaim, but we do return a smaller value for | |
6565 | * protection than what the group could enjoy in theory. This | |
6566 | * is okay. With the overcommit distribution above, effective | |
6567 | * protection is always dependent on how memory is actually | |
6568 | * consumed among the siblings anyway. | |
6569 | */ | |
8a931f80 JW |
6570 | ep = protected; |
6571 | ||
6572 | /* | |
6573 | * If the children aren't claiming (all of) the protection | |
6574 | * afforded to them by the parent, distribute the remainder in | |
6575 | * proportion to the (unprotected) memory of each cgroup. That | |
6576 | * way, cgroups that aren't explicitly prioritized wrt each | |
6577 | * other compete freely over the allowance, but they are | |
6578 | * collectively protected from neighboring trees. | |
6579 | * | |
6580 | * We're using unprotected memory for the weight so that if | |
6581 | * some cgroups DO claim explicit protection, we don't protect | |
6582 | * the same bytes twice. | |
cd324edc JW |
6583 | * |
6584 | * Check both usage and parent_usage against the respective | |
6585 | * protected values. One should imply the other, but they | |
6586 | * aren't read atomically - make sure the division is sane. | |
8a931f80 JW |
6587 | */ |
6588 | if (!(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)) | |
6589 | return ep; | |
cd324edc JW |
6590 | if (parent_effective > siblings_protected && |
6591 | parent_usage > siblings_protected && | |
6592 | usage > protected) { | |
8a931f80 JW |
6593 | unsigned long unclaimed; |
6594 | ||
6595 | unclaimed = parent_effective - siblings_protected; | |
6596 | unclaimed *= usage - protected; | |
6597 | unclaimed /= parent_usage - siblings_protected; | |
6598 | ||
6599 | ep += unclaimed; | |
6600 | } | |
6601 | ||
6602 | return ep; | |
bc50bcc6 JW |
6603 | } |
6604 | ||
241994ed | 6605 | /** |
05395718 | 6606 | * mem_cgroup_calculate_protection - check if memory consumption is in the normal range |
34c81057 | 6607 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
6608 | * @memcg: the memory cgroup to check |
6609 | * | |
23067153 RG |
6610 | * WARNING: This function is not stateless! It can only be used as part |
6611 | * of a top-down tree iteration, not for isolated queries. | |
241994ed | 6612 | */ |
45c7f7e1 CD |
6613 | void mem_cgroup_calculate_protection(struct mem_cgroup *root, |
6614 | struct mem_cgroup *memcg) | |
241994ed | 6615 | { |
8a931f80 | 6616 | unsigned long usage, parent_usage; |
23067153 RG |
6617 | struct mem_cgroup *parent; |
6618 | ||
241994ed | 6619 | if (mem_cgroup_disabled()) |
45c7f7e1 | 6620 | return; |
241994ed | 6621 | |
34c81057 SC |
6622 | if (!root) |
6623 | root = root_mem_cgroup; | |
22f7496f YS |
6624 | |
6625 | /* | |
6626 | * Effective values of the reclaim targets are ignored so they | |
6627 | * can be stale. Have a look at mem_cgroup_protection for more | |
6628 | * details. | |
6629 | * TODO: calculation should be more robust so that we do not need | |
6630 | * that special casing. | |
6631 | */ | |
34c81057 | 6632 | if (memcg == root) |
45c7f7e1 | 6633 | return; |
241994ed | 6634 | |
23067153 | 6635 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 | 6636 | if (!usage) |
45c7f7e1 | 6637 | return; |
bf8d5d52 | 6638 | |
bf8d5d52 | 6639 | parent = parent_mem_cgroup(memcg); |
df2a4196 RG |
6640 | /* No parent means a non-hierarchical mode on v1 memcg */ |
6641 | if (!parent) | |
45c7f7e1 | 6642 | return; |
df2a4196 | 6643 | |
bc50bcc6 | 6644 | if (parent == root) { |
c3d53200 | 6645 | memcg->memory.emin = READ_ONCE(memcg->memory.min); |
03960e33 | 6646 | memcg->memory.elow = READ_ONCE(memcg->memory.low); |
45c7f7e1 | 6647 | return; |
bf8d5d52 RG |
6648 | } |
6649 | ||
8a931f80 JW |
6650 | parent_usage = page_counter_read(&parent->memory); |
6651 | ||
b3a7822e | 6652 | WRITE_ONCE(memcg->memory.emin, effective_protection(usage, parent_usage, |
c3d53200 CD |
6653 | READ_ONCE(memcg->memory.min), |
6654 | READ_ONCE(parent->memory.emin), | |
b3a7822e | 6655 | atomic_long_read(&parent->memory.children_min_usage))); |
23067153 | 6656 | |
b3a7822e | 6657 | WRITE_ONCE(memcg->memory.elow, effective_protection(usage, parent_usage, |
03960e33 CD |
6658 | READ_ONCE(memcg->memory.low), |
6659 | READ_ONCE(parent->memory.elow), | |
b3a7822e | 6660 | atomic_long_read(&parent->memory.children_low_usage))); |
241994ed JW |
6661 | } |
6662 | ||
2c8d8f97 | 6663 | static int charge_memcg(struct page *page, struct mem_cgroup *memcg, gfp_t gfp) |
0add0c77 SB |
6664 | { |
6665 | unsigned int nr_pages = thp_nr_pages(page); | |
6666 | int ret; | |
6667 | ||
6668 | ret = try_charge(memcg, gfp, nr_pages); | |
6669 | if (ret) | |
6670 | goto out; | |
6671 | ||
6672 | css_get(&memcg->css); | |
6673 | commit_charge(page, memcg); | |
6674 | ||
6675 | local_irq_disable(); | |
6e0110c2 | 6676 | mem_cgroup_charge_statistics(memcg, nr_pages); |
8e88bd2d | 6677 | memcg_check_events(memcg, page_to_nid(page)); |
0add0c77 SB |
6678 | local_irq_enable(); |
6679 | out: | |
6680 | return ret; | |
6681 | } | |
6682 | ||
00501b53 | 6683 | /** |
2c8d8f97 | 6684 | * __mem_cgroup_charge - charge a newly allocated page to a cgroup |
00501b53 JW |
6685 | * @page: page to charge |
6686 | * @mm: mm context of the victim | |
6687 | * @gfp_mask: reclaim mode | |
00501b53 JW |
6688 | * |
6689 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
04f94e3f DS |
6690 | * pages according to @gfp_mask if necessary. if @mm is NULL, try to |
6691 | * charge to the active memcg. | |
00501b53 | 6692 | * |
0add0c77 SB |
6693 | * Do not use this for pages allocated for swapin. |
6694 | * | |
f0e45fb4 | 6695 | * Returns 0 on success. Otherwise, an error code is returned. |
00501b53 | 6696 | */ |
2c8d8f97 SB |
6697 | int __mem_cgroup_charge(struct page *page, struct mm_struct *mm, |
6698 | gfp_t gfp_mask) | |
00501b53 | 6699 | { |
0add0c77 SB |
6700 | struct mem_cgroup *memcg; |
6701 | int ret; | |
00501b53 | 6702 | |
0add0c77 | 6703 | memcg = get_mem_cgroup_from_mm(mm); |
2c8d8f97 | 6704 | ret = charge_memcg(page, memcg, gfp_mask); |
0add0c77 | 6705 | css_put(&memcg->css); |
2d1c4980 | 6706 | |
0add0c77 SB |
6707 | return ret; |
6708 | } | |
e993d905 | 6709 | |
0add0c77 SB |
6710 | /** |
6711 | * mem_cgroup_swapin_charge_page - charge a newly allocated page for swapin | |
6712 | * @page: page to charge | |
6713 | * @mm: mm context of the victim | |
6714 | * @gfp: reclaim mode | |
6715 | * @entry: swap entry for which the page is allocated | |
6716 | * | |
6717 | * This function charges a page allocated for swapin. Please call this before | |
6718 | * adding the page to the swapcache. | |
6719 | * | |
6720 | * Returns 0 on success. Otherwise, an error code is returned. | |
6721 | */ | |
6722 | int mem_cgroup_swapin_charge_page(struct page *page, struct mm_struct *mm, | |
6723 | gfp_t gfp, swp_entry_t entry) | |
6724 | { | |
6725 | struct mem_cgroup *memcg; | |
6726 | unsigned short id; | |
6727 | int ret; | |
00501b53 | 6728 | |
0add0c77 SB |
6729 | if (mem_cgroup_disabled()) |
6730 | return 0; | |
00501b53 | 6731 | |
0add0c77 SB |
6732 | id = lookup_swap_cgroup_id(entry); |
6733 | rcu_read_lock(); | |
6734 | memcg = mem_cgroup_from_id(id); | |
6735 | if (!memcg || !css_tryget_online(&memcg->css)) | |
6736 | memcg = get_mem_cgroup_from_mm(mm); | |
6737 | rcu_read_unlock(); | |
00501b53 | 6738 | |
2c8d8f97 | 6739 | ret = charge_memcg(page, memcg, gfp); |
6abb5a86 | 6740 | |
0add0c77 SB |
6741 | css_put(&memcg->css); |
6742 | return ret; | |
6743 | } | |
00501b53 | 6744 | |
0add0c77 SB |
6745 | /* |
6746 | * mem_cgroup_swapin_uncharge_swap - uncharge swap slot | |
6747 | * @entry: swap entry for which the page is charged | |
6748 | * | |
6749 | * Call this function after successfully adding the charged page to swapcache. | |
6750 | * | |
6751 | * Note: This function assumes the page for which swap slot is being uncharged | |
6752 | * is order 0 page. | |
6753 | */ | |
6754 | void mem_cgroup_swapin_uncharge_swap(swp_entry_t entry) | |
6755 | { | |
cae3af62 MS |
6756 | /* |
6757 | * Cgroup1's unified memory+swap counter has been charged with the | |
6758 | * new swapcache page, finish the transfer by uncharging the swap | |
6759 | * slot. The swap slot would also get uncharged when it dies, but | |
6760 | * it can stick around indefinitely and we'd count the page twice | |
6761 | * the entire time. | |
6762 | * | |
6763 | * Cgroup2 has separate resource counters for memory and swap, | |
6764 | * so this is a non-issue here. Memory and swap charge lifetimes | |
6765 | * correspond 1:1 to page and swap slot lifetimes: we charge the | |
6766 | * page to memory here, and uncharge swap when the slot is freed. | |
6767 | */ | |
0add0c77 | 6768 | if (!mem_cgroup_disabled() && do_memsw_account()) { |
00501b53 JW |
6769 | /* |
6770 | * The swap entry might not get freed for a long time, | |
6771 | * let's not wait for it. The page already received a | |
6772 | * memory+swap charge, drop the swap entry duplicate. | |
6773 | */ | |
0add0c77 | 6774 | mem_cgroup_uncharge_swap(entry, 1); |
00501b53 | 6775 | } |
3fea5a49 JW |
6776 | } |
6777 | ||
a9d5adee JG |
6778 | struct uncharge_gather { |
6779 | struct mem_cgroup *memcg; | |
b4e0b68f | 6780 | unsigned long nr_memory; |
a9d5adee | 6781 | unsigned long pgpgout; |
a9d5adee | 6782 | unsigned long nr_kmem; |
8e88bd2d | 6783 | int nid; |
a9d5adee JG |
6784 | }; |
6785 | ||
6786 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 6787 | { |
a9d5adee JG |
6788 | memset(ug, 0, sizeof(*ug)); |
6789 | } | |
6790 | ||
6791 | static void uncharge_batch(const struct uncharge_gather *ug) | |
6792 | { | |
747db954 JW |
6793 | unsigned long flags; |
6794 | ||
b4e0b68f MS |
6795 | if (ug->nr_memory) { |
6796 | page_counter_uncharge(&ug->memcg->memory, ug->nr_memory); | |
7941d214 | 6797 | if (do_memsw_account()) |
b4e0b68f | 6798 | page_counter_uncharge(&ug->memcg->memsw, ug->nr_memory); |
a9d5adee JG |
6799 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem) |
6800 | page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem); | |
6801 | memcg_oom_recover(ug->memcg); | |
ce00a967 | 6802 | } |
747db954 JW |
6803 | |
6804 | local_irq_save(flags); | |
c9019e9b | 6805 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); |
b4e0b68f | 6806 | __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, ug->nr_memory); |
8e88bd2d | 6807 | memcg_check_events(ug->memcg, ug->nid); |
747db954 | 6808 | local_irq_restore(flags); |
f1796544 MH |
6809 | |
6810 | /* drop reference from uncharge_page */ | |
6811 | css_put(&ug->memcg->css); | |
a9d5adee JG |
6812 | } |
6813 | ||
6814 | static void uncharge_page(struct page *page, struct uncharge_gather *ug) | |
6815 | { | |
1b7e4464 | 6816 | struct folio *folio = page_folio(page); |
9f762dbe | 6817 | unsigned long nr_pages; |
b4e0b68f MS |
6818 | struct mem_cgroup *memcg; |
6819 | struct obj_cgroup *objcg; | |
55927114 | 6820 | bool use_objcg = PageMemcgKmem(page); |
9f762dbe | 6821 | |
a9d5adee | 6822 | VM_BUG_ON_PAGE(PageLRU(page), page); |
a9d5adee | 6823 | |
a9d5adee JG |
6824 | /* |
6825 | * Nobody should be changing or seriously looking at | |
b4e0b68f | 6826 | * page memcg or objcg at this point, we have fully |
a9d5adee JG |
6827 | * exclusive access to the page. |
6828 | */ | |
55927114 | 6829 | if (use_objcg) { |
1b7e4464 | 6830 | objcg = __folio_objcg(folio); |
b4e0b68f MS |
6831 | /* |
6832 | * This get matches the put at the end of the function and | |
6833 | * kmem pages do not hold memcg references anymore. | |
6834 | */ | |
6835 | memcg = get_mem_cgroup_from_objcg(objcg); | |
6836 | } else { | |
1b7e4464 | 6837 | memcg = __folio_memcg(folio); |
b4e0b68f | 6838 | } |
a9d5adee | 6839 | |
b4e0b68f MS |
6840 | if (!memcg) |
6841 | return; | |
6842 | ||
6843 | if (ug->memcg != memcg) { | |
a9d5adee JG |
6844 | if (ug->memcg) { |
6845 | uncharge_batch(ug); | |
6846 | uncharge_gather_clear(ug); | |
6847 | } | |
b4e0b68f | 6848 | ug->memcg = memcg; |
8e88bd2d | 6849 | ug->nid = page_to_nid(page); |
f1796544 MH |
6850 | |
6851 | /* pairs with css_put in uncharge_batch */ | |
b4e0b68f | 6852 | css_get(&memcg->css); |
a9d5adee JG |
6853 | } |
6854 | ||
9f762dbe | 6855 | nr_pages = compound_nr(page); |
a9d5adee | 6856 | |
55927114 | 6857 | if (use_objcg) { |
b4e0b68f | 6858 | ug->nr_memory += nr_pages; |
9f762dbe | 6859 | ug->nr_kmem += nr_pages; |
b4e0b68f MS |
6860 | |
6861 | page->memcg_data = 0; | |
6862 | obj_cgroup_put(objcg); | |
6863 | } else { | |
6864 | /* LRU pages aren't accounted at the root level */ | |
6865 | if (!mem_cgroup_is_root(memcg)) | |
6866 | ug->nr_memory += nr_pages; | |
18b2db3b | 6867 | ug->pgpgout++; |
a9d5adee | 6868 | |
b4e0b68f MS |
6869 | page->memcg_data = 0; |
6870 | } | |
6871 | ||
6872 | css_put(&memcg->css); | |
747db954 JW |
6873 | } |
6874 | ||
0a31bc97 | 6875 | /** |
2c8d8f97 | 6876 | * __mem_cgroup_uncharge - uncharge a page |
0a31bc97 JW |
6877 | * @page: page to uncharge |
6878 | * | |
2c8d8f97 | 6879 | * Uncharge a page previously charged with __mem_cgroup_charge(). |
0a31bc97 | 6880 | */ |
2c8d8f97 | 6881 | void __mem_cgroup_uncharge(struct page *page) |
0a31bc97 | 6882 | { |
a9d5adee JG |
6883 | struct uncharge_gather ug; |
6884 | ||
747db954 | 6885 | /* Don't touch page->lru of any random page, pre-check: */ |
bcfe06bf | 6886 | if (!page_memcg(page)) |
0a31bc97 JW |
6887 | return; |
6888 | ||
a9d5adee JG |
6889 | uncharge_gather_clear(&ug); |
6890 | uncharge_page(page, &ug); | |
6891 | uncharge_batch(&ug); | |
747db954 | 6892 | } |
0a31bc97 | 6893 | |
747db954 | 6894 | /** |
2c8d8f97 | 6895 | * __mem_cgroup_uncharge_list - uncharge a list of page |
747db954 JW |
6896 | * @page_list: list of pages to uncharge |
6897 | * | |
6898 | * Uncharge a list of pages previously charged with | |
2c8d8f97 | 6899 | * __mem_cgroup_charge(). |
747db954 | 6900 | */ |
2c8d8f97 | 6901 | void __mem_cgroup_uncharge_list(struct list_head *page_list) |
747db954 | 6902 | { |
c41a40b6 MS |
6903 | struct uncharge_gather ug; |
6904 | struct page *page; | |
6905 | ||
c41a40b6 MS |
6906 | uncharge_gather_clear(&ug); |
6907 | list_for_each_entry(page, page_list, lru) | |
6908 | uncharge_page(page, &ug); | |
6909 | if (ug.memcg) | |
6910 | uncharge_batch(&ug); | |
0a31bc97 JW |
6911 | } |
6912 | ||
6913 | /** | |
6a93ca8f JW |
6914 | * mem_cgroup_migrate - charge a page's replacement |
6915 | * @oldpage: currently circulating page | |
6916 | * @newpage: replacement page | |
0a31bc97 | 6917 | * |
6a93ca8f JW |
6918 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
6919 | * be uncharged upon free. | |
0a31bc97 JW |
6920 | * |
6921 | * Both pages must be locked, @newpage->mapping must be set up. | |
6922 | */ | |
6a93ca8f | 6923 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 6924 | { |
29833315 | 6925 | struct mem_cgroup *memcg; |
44b7a8d3 | 6926 | unsigned int nr_pages; |
d93c4130 | 6927 | unsigned long flags; |
0a31bc97 JW |
6928 | |
6929 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6930 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 6931 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
6932 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6933 | newpage); | |
0a31bc97 JW |
6934 | |
6935 | if (mem_cgroup_disabled()) | |
6936 | return; | |
6937 | ||
6938 | /* Page cache replacement: new page already charged? */ | |
bcfe06bf | 6939 | if (page_memcg(newpage)) |
0a31bc97 JW |
6940 | return; |
6941 | ||
bcfe06bf | 6942 | memcg = page_memcg(oldpage); |
a4055888 | 6943 | VM_WARN_ON_ONCE_PAGE(!memcg, oldpage); |
29833315 | 6944 | if (!memcg) |
0a31bc97 JW |
6945 | return; |
6946 | ||
44b7a8d3 | 6947 | /* Force-charge the new page. The old one will be freed soon */ |
6c357848 | 6948 | nr_pages = thp_nr_pages(newpage); |
44b7a8d3 | 6949 | |
8dc87c7d MS |
6950 | if (!mem_cgroup_is_root(memcg)) { |
6951 | page_counter_charge(&memcg->memory, nr_pages); | |
6952 | if (do_memsw_account()) | |
6953 | page_counter_charge(&memcg->memsw, nr_pages); | |
6954 | } | |
0a31bc97 | 6955 | |
1a3e1f40 | 6956 | css_get(&memcg->css); |
d9eb1ea2 | 6957 | commit_charge(newpage, memcg); |
44b7a8d3 | 6958 | |
d93c4130 | 6959 | local_irq_save(flags); |
6e0110c2 | 6960 | mem_cgroup_charge_statistics(memcg, nr_pages); |
8e88bd2d | 6961 | memcg_check_events(memcg, page_to_nid(newpage)); |
d93c4130 | 6962 | local_irq_restore(flags); |
0a31bc97 JW |
6963 | } |
6964 | ||
ef12947c | 6965 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
6966 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
6967 | ||
2d758073 | 6968 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
6969 | { |
6970 | struct mem_cgroup *memcg; | |
6971 | ||
2d758073 JW |
6972 | if (!mem_cgroup_sockets_enabled) |
6973 | return; | |
6974 | ||
e876ecc6 SB |
6975 | /* Do not associate the sock with unrelated interrupted task's memcg. */ |
6976 | if (in_interrupt()) | |
6977 | return; | |
6978 | ||
11092087 JW |
6979 | rcu_read_lock(); |
6980 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
6981 | if (memcg == root_mem_cgroup) |
6982 | goto out; | |
0db15298 | 6983 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 6984 | goto out; |
8965aa28 | 6985 | if (css_tryget(&memcg->css)) |
11092087 | 6986 | sk->sk_memcg = memcg; |
f7e1cb6e | 6987 | out: |
11092087 JW |
6988 | rcu_read_unlock(); |
6989 | } | |
11092087 | 6990 | |
2d758073 | 6991 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 6992 | { |
2d758073 JW |
6993 | if (sk->sk_memcg) |
6994 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
6995 | } |
6996 | ||
6997 | /** | |
6998 | * mem_cgroup_charge_skmem - charge socket memory | |
6999 | * @memcg: memcg to charge | |
7000 | * @nr_pages: number of pages to charge | |
4b1327be | 7001 | * @gfp_mask: reclaim mode |
11092087 JW |
7002 | * |
7003 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
4b1327be | 7004 | * @memcg's configured limit, %false if it doesn't. |
11092087 | 7005 | */ |
4b1327be WW |
7006 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages, |
7007 | gfp_t gfp_mask) | |
11092087 | 7008 | { |
f7e1cb6e | 7009 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 7010 | struct page_counter *fail; |
f7e1cb6e | 7011 | |
0db15298 JW |
7012 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
7013 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
7014 | return true; |
7015 | } | |
0db15298 | 7016 | memcg->tcpmem_pressure = 1; |
4b1327be WW |
7017 | if (gfp_mask & __GFP_NOFAIL) { |
7018 | page_counter_charge(&memcg->tcpmem, nr_pages); | |
7019 | return true; | |
7020 | } | |
f7e1cb6e | 7021 | return false; |
11092087 | 7022 | } |
d886f4e4 | 7023 | |
4b1327be WW |
7024 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) { |
7025 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); | |
f7e1cb6e | 7026 | return true; |
4b1327be | 7027 | } |
f7e1cb6e | 7028 | |
11092087 JW |
7029 | return false; |
7030 | } | |
7031 | ||
7032 | /** | |
7033 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
7034 | * @memcg: memcg to uncharge |
7035 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
7036 | */ |
7037 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
7038 | { | |
f7e1cb6e | 7039 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 7040 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
7041 | return; |
7042 | } | |
d886f4e4 | 7043 | |
c9019e9b | 7044 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 7045 | |
475d0487 | 7046 | refill_stock(memcg, nr_pages); |
11092087 JW |
7047 | } |
7048 | ||
f7e1cb6e JW |
7049 | static int __init cgroup_memory(char *s) |
7050 | { | |
7051 | char *token; | |
7052 | ||
7053 | while ((token = strsep(&s, ",")) != NULL) { | |
7054 | if (!*token) | |
7055 | continue; | |
7056 | if (!strcmp(token, "nosocket")) | |
7057 | cgroup_memory_nosocket = true; | |
04823c83 VD |
7058 | if (!strcmp(token, "nokmem")) |
7059 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
7060 | } |
7061 | return 0; | |
7062 | } | |
7063 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 7064 | |
2d11085e | 7065 | /* |
1081312f MH |
7066 | * subsys_initcall() for memory controller. |
7067 | * | |
308167fc SAS |
7068 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
7069 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
7070 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
7071 | * should be initialized from here. | |
2d11085e MH |
7072 | */ |
7073 | static int __init mem_cgroup_init(void) | |
7074 | { | |
95a045f6 JW |
7075 | int cpu, node; |
7076 | ||
f3344adf MS |
7077 | /* |
7078 | * Currently s32 type (can refer to struct batched_lruvec_stat) is | |
7079 | * used for per-memcg-per-cpu caching of per-node statistics. In order | |
7080 | * to work fine, we should make sure that the overfill threshold can't | |
7081 | * exceed S32_MAX / PAGE_SIZE. | |
7082 | */ | |
7083 | BUILD_BUG_ON(MEMCG_CHARGE_BATCH > S32_MAX / PAGE_SIZE); | |
7084 | ||
308167fc SAS |
7085 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
7086 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
7087 | |
7088 | for_each_possible_cpu(cpu) | |
7089 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
7090 | drain_local_stock); | |
7091 | ||
7092 | for_each_node(node) { | |
7093 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
7094 | |
7095 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
7096 | node_online(node) ? node : NUMA_NO_NODE); | |
7097 | ||
ef8f2327 | 7098 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 7099 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 7100 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
7101 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
7102 | } | |
7103 | ||
2d11085e MH |
7104 | return 0; |
7105 | } | |
7106 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
7107 | |
7108 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
7109 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
7110 | { | |
1c2d479a | 7111 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
7112 | /* |
7113 | * The root cgroup cannot be destroyed, so it's refcount must | |
7114 | * always be >= 1. | |
7115 | */ | |
7116 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
7117 | VM_BUG_ON(1); | |
7118 | break; | |
7119 | } | |
7120 | memcg = parent_mem_cgroup(memcg); | |
7121 | if (!memcg) | |
7122 | memcg = root_mem_cgroup; | |
7123 | } | |
7124 | return memcg; | |
7125 | } | |
7126 | ||
21afa38e JW |
7127 | /** |
7128 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
7129 | * @page: page whose memsw charge to transfer | |
7130 | * @entry: swap entry to move the charge to | |
7131 | * | |
7132 | * Transfer the memsw charge of @page to @entry. | |
7133 | */ | |
7134 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
7135 | { | |
1f47b61f | 7136 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 7137 | unsigned int nr_entries; |
21afa38e JW |
7138 | unsigned short oldid; |
7139 | ||
7140 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
7141 | VM_BUG_ON_PAGE(page_count(page), page); | |
7142 | ||
76358ab5 AS |
7143 | if (mem_cgroup_disabled()) |
7144 | return; | |
7145 | ||
2d1c4980 | 7146 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
21afa38e JW |
7147 | return; |
7148 | ||
bcfe06bf | 7149 | memcg = page_memcg(page); |
21afa38e | 7150 | |
a4055888 | 7151 | VM_WARN_ON_ONCE_PAGE(!memcg, page); |
21afa38e JW |
7152 | if (!memcg) |
7153 | return; | |
7154 | ||
1f47b61f VD |
7155 | /* |
7156 | * In case the memcg owning these pages has been offlined and doesn't | |
7157 | * have an ID allocated to it anymore, charge the closest online | |
7158 | * ancestor for the swap instead and transfer the memory+swap charge. | |
7159 | */ | |
7160 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
6c357848 | 7161 | nr_entries = thp_nr_pages(page); |
d6810d73 HY |
7162 | /* Get references for the tail pages, too */ |
7163 | if (nr_entries > 1) | |
7164 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
7165 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
7166 | nr_entries); | |
21afa38e | 7167 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7168 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e | 7169 | |
bcfe06bf | 7170 | page->memcg_data = 0; |
21afa38e JW |
7171 | |
7172 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 7173 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 7174 | |
2d1c4980 | 7175 | if (!cgroup_memory_noswap && memcg != swap_memcg) { |
1f47b61f | 7176 | if (!mem_cgroup_is_root(swap_memcg)) |
d6810d73 HY |
7177 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
7178 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
7179 | } |
7180 | ||
ce9ce665 SAS |
7181 | /* |
7182 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 7183 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 7184 | * important here to have the interrupts disabled because it is the |
b93b0163 | 7185 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 SAS |
7186 | */ |
7187 | VM_BUG_ON(!irqs_disabled()); | |
6e0110c2 | 7188 | mem_cgroup_charge_statistics(memcg, -nr_entries); |
8e88bd2d | 7189 | memcg_check_events(memcg, page_to_nid(page)); |
73f576c0 | 7190 | |
1a3e1f40 | 7191 | css_put(&memcg->css); |
21afa38e JW |
7192 | } |
7193 | ||
38d8b4e6 | 7194 | /** |
01c4b28c | 7195 | * __mem_cgroup_try_charge_swap - try charging swap space for a page |
37e84351 VD |
7196 | * @page: page being added to swap |
7197 | * @entry: swap entry to charge | |
7198 | * | |
38d8b4e6 | 7199 | * Try to charge @page's memcg for the swap space at @entry. |
37e84351 VD |
7200 | * |
7201 | * Returns 0 on success, -ENOMEM on failure. | |
7202 | */ | |
01c4b28c | 7203 | int __mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) |
37e84351 | 7204 | { |
6c357848 | 7205 | unsigned int nr_pages = thp_nr_pages(page); |
37e84351 | 7206 | struct page_counter *counter; |
38d8b4e6 | 7207 | struct mem_cgroup *memcg; |
37e84351 VD |
7208 | unsigned short oldid; |
7209 | ||
2d1c4980 | 7210 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
37e84351 VD |
7211 | return 0; |
7212 | ||
bcfe06bf | 7213 | memcg = page_memcg(page); |
37e84351 | 7214 | |
a4055888 | 7215 | VM_WARN_ON_ONCE_PAGE(!memcg, page); |
37e84351 VD |
7216 | if (!memcg) |
7217 | return 0; | |
7218 | ||
f3a53a3a TH |
7219 | if (!entry.val) { |
7220 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 7221 | return 0; |
f3a53a3a | 7222 | } |
bb98f2c5 | 7223 | |
1f47b61f VD |
7224 | memcg = mem_cgroup_id_get_online(memcg); |
7225 | ||
2d1c4980 | 7226 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg) && |
38d8b4e6 | 7227 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
7228 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
7229 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 7230 | mem_cgroup_id_put(memcg); |
37e84351 | 7231 | return -ENOMEM; |
1f47b61f | 7232 | } |
37e84351 | 7233 | |
38d8b4e6 HY |
7234 | /* Get references for the tail pages, too */ |
7235 | if (nr_pages > 1) | |
7236 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
7237 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
37e84351 | 7238 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7239 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 7240 | |
37e84351 VD |
7241 | return 0; |
7242 | } | |
7243 | ||
21afa38e | 7244 | /** |
01c4b28c | 7245 | * __mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 7246 | * @entry: swap entry to uncharge |
38d8b4e6 | 7247 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 7248 | */ |
01c4b28c | 7249 | void __mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
7250 | { |
7251 | struct mem_cgroup *memcg; | |
7252 | unsigned short id; | |
7253 | ||
38d8b4e6 | 7254 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 7255 | rcu_read_lock(); |
adbe427b | 7256 | memcg = mem_cgroup_from_id(id); |
21afa38e | 7257 | if (memcg) { |
2d1c4980 | 7258 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg)) { |
37e84351 | 7259 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
38d8b4e6 | 7260 | page_counter_uncharge(&memcg->swap, nr_pages); |
37e84351 | 7261 | else |
38d8b4e6 | 7262 | page_counter_uncharge(&memcg->memsw, nr_pages); |
37e84351 | 7263 | } |
c9019e9b | 7264 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 7265 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
7266 | } |
7267 | rcu_read_unlock(); | |
7268 | } | |
7269 | ||
d8b38438 VD |
7270 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
7271 | { | |
7272 | long nr_swap_pages = get_nr_swap_pages(); | |
7273 | ||
eccb52e7 | 7274 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
d8b38438 VD |
7275 | return nr_swap_pages; |
7276 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
7277 | nr_swap_pages = min_t(long, nr_swap_pages, | |
bbec2e15 | 7278 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
7279 | page_counter_read(&memcg->swap)); |
7280 | return nr_swap_pages; | |
7281 | } | |
7282 | ||
5ccc5aba VD |
7283 | bool mem_cgroup_swap_full(struct page *page) |
7284 | { | |
7285 | struct mem_cgroup *memcg; | |
7286 | ||
7287 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
7288 | ||
7289 | if (vm_swap_full()) | |
7290 | return true; | |
eccb52e7 | 7291 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
5ccc5aba VD |
7292 | return false; |
7293 | ||
bcfe06bf | 7294 | memcg = page_memcg(page); |
5ccc5aba VD |
7295 | if (!memcg) |
7296 | return false; | |
7297 | ||
4b82ab4f JK |
7298 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { |
7299 | unsigned long usage = page_counter_read(&memcg->swap); | |
7300 | ||
7301 | if (usage * 2 >= READ_ONCE(memcg->swap.high) || | |
7302 | usage * 2 >= READ_ONCE(memcg->swap.max)) | |
5ccc5aba | 7303 | return true; |
4b82ab4f | 7304 | } |
5ccc5aba VD |
7305 | |
7306 | return false; | |
7307 | } | |
7308 | ||
eccb52e7 | 7309 | static int __init setup_swap_account(char *s) |
21afa38e JW |
7310 | { |
7311 | if (!strcmp(s, "1")) | |
5ab92901 | 7312 | cgroup_memory_noswap = false; |
21afa38e | 7313 | else if (!strcmp(s, "0")) |
5ab92901 | 7314 | cgroup_memory_noswap = true; |
21afa38e JW |
7315 | return 1; |
7316 | } | |
eccb52e7 | 7317 | __setup("swapaccount=", setup_swap_account); |
21afa38e | 7318 | |
37e84351 VD |
7319 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
7320 | struct cftype *cft) | |
7321 | { | |
7322 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
7323 | ||
7324 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
7325 | } | |
7326 | ||
4b82ab4f JK |
7327 | static int swap_high_show(struct seq_file *m, void *v) |
7328 | { | |
7329 | return seq_puts_memcg_tunable(m, | |
7330 | READ_ONCE(mem_cgroup_from_seq(m)->swap.high)); | |
7331 | } | |
7332 | ||
7333 | static ssize_t swap_high_write(struct kernfs_open_file *of, | |
7334 | char *buf, size_t nbytes, loff_t off) | |
7335 | { | |
7336 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7337 | unsigned long high; | |
7338 | int err; | |
7339 | ||
7340 | buf = strstrip(buf); | |
7341 | err = page_counter_memparse(buf, "max", &high); | |
7342 | if (err) | |
7343 | return err; | |
7344 | ||
7345 | page_counter_set_high(&memcg->swap, high); | |
7346 | ||
7347 | return nbytes; | |
7348 | } | |
7349 | ||
37e84351 VD |
7350 | static int swap_max_show(struct seq_file *m, void *v) |
7351 | { | |
677dc973 CD |
7352 | return seq_puts_memcg_tunable(m, |
7353 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); | |
37e84351 VD |
7354 | } |
7355 | ||
7356 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
7357 | char *buf, size_t nbytes, loff_t off) | |
7358 | { | |
7359 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7360 | unsigned long max; | |
7361 | int err; | |
7362 | ||
7363 | buf = strstrip(buf); | |
7364 | err = page_counter_memparse(buf, "max", &max); | |
7365 | if (err) | |
7366 | return err; | |
7367 | ||
be09102b | 7368 | xchg(&memcg->swap.max, max); |
37e84351 VD |
7369 | |
7370 | return nbytes; | |
7371 | } | |
7372 | ||
f3a53a3a TH |
7373 | static int swap_events_show(struct seq_file *m, void *v) |
7374 | { | |
aa9694bb | 7375 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
f3a53a3a | 7376 | |
4b82ab4f JK |
7377 | seq_printf(m, "high %lu\n", |
7378 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_HIGH])); | |
f3a53a3a TH |
7379 | seq_printf(m, "max %lu\n", |
7380 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
7381 | seq_printf(m, "fail %lu\n", | |
7382 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
7383 | ||
7384 | return 0; | |
7385 | } | |
7386 | ||
37e84351 VD |
7387 | static struct cftype swap_files[] = { |
7388 | { | |
7389 | .name = "swap.current", | |
7390 | .flags = CFTYPE_NOT_ON_ROOT, | |
7391 | .read_u64 = swap_current_read, | |
7392 | }, | |
4b82ab4f JK |
7393 | { |
7394 | .name = "swap.high", | |
7395 | .flags = CFTYPE_NOT_ON_ROOT, | |
7396 | .seq_show = swap_high_show, | |
7397 | .write = swap_high_write, | |
7398 | }, | |
37e84351 VD |
7399 | { |
7400 | .name = "swap.max", | |
7401 | .flags = CFTYPE_NOT_ON_ROOT, | |
7402 | .seq_show = swap_max_show, | |
7403 | .write = swap_max_write, | |
7404 | }, | |
f3a53a3a TH |
7405 | { |
7406 | .name = "swap.events", | |
7407 | .flags = CFTYPE_NOT_ON_ROOT, | |
7408 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
7409 | .seq_show = swap_events_show, | |
7410 | }, | |
37e84351 VD |
7411 | { } /* terminate */ |
7412 | }; | |
7413 | ||
eccb52e7 | 7414 | static struct cftype memsw_files[] = { |
21afa38e JW |
7415 | { |
7416 | .name = "memsw.usage_in_bytes", | |
7417 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
7418 | .read_u64 = mem_cgroup_read_u64, | |
7419 | }, | |
7420 | { | |
7421 | .name = "memsw.max_usage_in_bytes", | |
7422 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
7423 | .write = mem_cgroup_reset, | |
7424 | .read_u64 = mem_cgroup_read_u64, | |
7425 | }, | |
7426 | { | |
7427 | .name = "memsw.limit_in_bytes", | |
7428 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
7429 | .write = mem_cgroup_write, | |
7430 | .read_u64 = mem_cgroup_read_u64, | |
7431 | }, | |
7432 | { | |
7433 | .name = "memsw.failcnt", | |
7434 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
7435 | .write = mem_cgroup_reset, | |
7436 | .read_u64 = mem_cgroup_read_u64, | |
7437 | }, | |
7438 | { }, /* terminate */ | |
7439 | }; | |
7440 | ||
82ff165c BS |
7441 | /* |
7442 | * If mem_cgroup_swap_init() is implemented as a subsys_initcall() | |
7443 | * instead of a core_initcall(), this could mean cgroup_memory_noswap still | |
7444 | * remains set to false even when memcg is disabled via "cgroup_disable=memory" | |
7445 | * boot parameter. This may result in premature OOPS inside | |
7446 | * mem_cgroup_get_nr_swap_pages() function in corner cases. | |
7447 | */ | |
21afa38e JW |
7448 | static int __init mem_cgroup_swap_init(void) |
7449 | { | |
2d1c4980 JW |
7450 | /* No memory control -> no swap control */ |
7451 | if (mem_cgroup_disabled()) | |
7452 | cgroup_memory_noswap = true; | |
7453 | ||
7454 | if (cgroup_memory_noswap) | |
eccb52e7 JW |
7455 | return 0; |
7456 | ||
7457 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, swap_files)); | |
7458 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, memsw_files)); | |
7459 | ||
21afa38e JW |
7460 | return 0; |
7461 | } | |
82ff165c | 7462 | core_initcall(mem_cgroup_swap_init); |
21afa38e JW |
7463 | |
7464 | #endif /* CONFIG_MEMCG_SWAP */ |