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> |
4882c809 | 36 | #include <linux/pagevec.h> |
1ff9e6e1 | 37 | #include <linux/vm_event_item.h> |
d52aa412 | 38 | #include <linux/smp.h> |
8a9f3ccd | 39 | #include <linux/page-flags.h> |
66e1707b | 40 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
41 | #include <linux/bit_spinlock.h> |
42 | #include <linux/rcupdate.h> | |
e222432b | 43 | #include <linux/limits.h> |
b9e15baf | 44 | #include <linux/export.h> |
8c7c6e34 | 45 | #include <linux/mutex.h> |
bb4cc1a8 | 46 | #include <linux/rbtree.h> |
b6ac57d5 | 47 | #include <linux/slab.h> |
66e1707b | 48 | #include <linux/swap.h> |
02491447 | 49 | #include <linux/swapops.h> |
66e1707b | 50 | #include <linux/spinlock.h> |
2e72b634 | 51 | #include <linux/eventfd.h> |
79bd9814 | 52 | #include <linux/poll.h> |
2e72b634 | 53 | #include <linux/sort.h> |
66e1707b | 54 | #include <linux/fs.h> |
d2ceb9b7 | 55 | #include <linux/seq_file.h> |
70ddf637 | 56 | #include <linux/vmpressure.h> |
dc90f084 | 57 | #include <linux/memremap.h> |
b69408e8 | 58 | #include <linux/mm_inline.h> |
5d1ea48b | 59 | #include <linux/swap_cgroup.h> |
cdec2e42 | 60 | #include <linux/cpu.h> |
158e0a2d | 61 | #include <linux/oom.h> |
0056f4e6 | 62 | #include <linux/lockdep.h> |
79bd9814 | 63 | #include <linux/file.h> |
03248add | 64 | #include <linux/resume_user_mode.h> |
0e4b01df | 65 | #include <linux/psi.h> |
c8713d0b | 66 | #include <linux/seq_buf.h> |
6a792697 | 67 | #include <linux/sched/isolation.h> |
6011be59 | 68 | #include <linux/kmemleak.h> |
08e552c6 | 69 | #include "internal.h" |
d1a4c0b3 | 70 | #include <net/sock.h> |
4bd2c1ee | 71 | #include <net/ip.h> |
f35c3a8e | 72 | #include "slab.h" |
014bb1de | 73 | #include "swap.h" |
8cdea7c0 | 74 | |
7c0f6ba6 | 75 | #include <linux/uaccess.h> |
8697d331 | 76 | |
cc8e970c KM |
77 | #include <trace/events/vmscan.h> |
78 | ||
073219e9 TH |
79 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
80 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 81 | |
7d828602 JW |
82 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
83 | ||
37d5985c RG |
84 | /* Active memory cgroup to use from an interrupt context */ |
85 | DEFINE_PER_CPU(struct mem_cgroup *, int_active_memcg); | |
c74d40e8 | 86 | EXPORT_PER_CPU_SYMBOL_GPL(int_active_memcg); |
37d5985c | 87 | |
f7e1cb6e | 88 | /* Socket memory accounting disabled? */ |
0f0cace3 | 89 | static bool cgroup_memory_nosocket __ro_after_init; |
f7e1cb6e | 90 | |
04823c83 | 91 | /* Kernel memory accounting disabled? */ |
17c17367 | 92 | static bool cgroup_memory_nokmem __ro_after_init; |
04823c83 | 93 | |
b6c1a8af YS |
94 | /* BPF memory accounting disabled? */ |
95 | static bool cgroup_memory_nobpf __ro_after_init; | |
96 | ||
97b27821 TH |
97 | #ifdef CONFIG_CGROUP_WRITEBACK |
98 | static DECLARE_WAIT_QUEUE_HEAD(memcg_cgwb_frn_waitq); | |
99 | #endif | |
100 | ||
7941d214 JW |
101 | /* Whether legacy memory+swap accounting is active */ |
102 | static bool do_memsw_account(void) | |
103 | { | |
b25806dc | 104 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys); |
7941d214 JW |
105 | } |
106 | ||
a0db00fc KS |
107 | #define THRESHOLDS_EVENTS_TARGET 128 |
108 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
e9f8974f | 109 | |
bb4cc1a8 AM |
110 | /* |
111 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
112 | * their hierarchy representation | |
113 | */ | |
114 | ||
ef8f2327 | 115 | struct mem_cgroup_tree_per_node { |
bb4cc1a8 | 116 | struct rb_root rb_root; |
fa90b2fd | 117 | struct rb_node *rb_rightmost; |
bb4cc1a8 AM |
118 | spinlock_t lock; |
119 | }; | |
120 | ||
bb4cc1a8 AM |
121 | struct mem_cgroup_tree { |
122 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
123 | }; | |
124 | ||
125 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
126 | ||
9490ff27 KH |
127 | /* for OOM */ |
128 | struct mem_cgroup_eventfd_list { | |
129 | struct list_head list; | |
130 | struct eventfd_ctx *eventfd; | |
131 | }; | |
2e72b634 | 132 | |
79bd9814 TH |
133 | /* |
134 | * cgroup_event represents events which userspace want to receive. | |
135 | */ | |
3bc942f3 | 136 | struct mem_cgroup_event { |
79bd9814 | 137 | /* |
59b6f873 | 138 | * memcg which the event belongs to. |
79bd9814 | 139 | */ |
59b6f873 | 140 | struct mem_cgroup *memcg; |
79bd9814 TH |
141 | /* |
142 | * eventfd to signal userspace about the event. | |
143 | */ | |
144 | struct eventfd_ctx *eventfd; | |
145 | /* | |
146 | * Each of these stored in a list by the cgroup. | |
147 | */ | |
148 | struct list_head list; | |
fba94807 TH |
149 | /* |
150 | * register_event() callback will be used to add new userspace | |
151 | * waiter for changes related to this event. Use eventfd_signal() | |
152 | * on eventfd to send notification to userspace. | |
153 | */ | |
59b6f873 | 154 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 155 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
156 | /* |
157 | * unregister_event() callback will be called when userspace closes | |
158 | * the eventfd or on cgroup removing. This callback must be set, | |
159 | * if you want provide notification functionality. | |
160 | */ | |
59b6f873 | 161 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 162 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
163 | /* |
164 | * All fields below needed to unregister event when | |
165 | * userspace closes eventfd. | |
166 | */ | |
167 | poll_table pt; | |
168 | wait_queue_head_t *wqh; | |
ac6424b9 | 169 | wait_queue_entry_t wait; |
79bd9814 TH |
170 | struct work_struct remove; |
171 | }; | |
172 | ||
c0ff4b85 R |
173 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
174 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 175 | |
7dc74be0 DN |
176 | /* Stuffs for move charges at task migration. */ |
177 | /* | |
1dfab5ab | 178 | * Types of charges to be moved. |
7dc74be0 | 179 | */ |
1dfab5ab JW |
180 | #define MOVE_ANON 0x1U |
181 | #define MOVE_FILE 0x2U | |
182 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 183 | |
4ffef5fe DN |
184 | /* "mc" and its members are protected by cgroup_mutex */ |
185 | static struct move_charge_struct { | |
b1dd693e | 186 | spinlock_t lock; /* for from, to */ |
264a0ae1 | 187 | struct mm_struct *mm; |
4ffef5fe DN |
188 | struct mem_cgroup *from; |
189 | struct mem_cgroup *to; | |
1dfab5ab | 190 | unsigned long flags; |
4ffef5fe | 191 | unsigned long precharge; |
854ffa8d | 192 | unsigned long moved_charge; |
483c30b5 | 193 | unsigned long moved_swap; |
8033b97c DN |
194 | struct task_struct *moving_task; /* a task moving charges */ |
195 | wait_queue_head_t waitq; /* a waitq for other context */ | |
196 | } mc = { | |
2bd9bb20 | 197 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
198 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
199 | }; | |
4ffef5fe | 200 | |
4e416953 | 201 | /* |
f4d005af | 202 | * Maximum loops in mem_cgroup_soft_reclaim(), used for soft |
4e416953 BS |
203 | * limit reclaim to prevent infinite loops, if they ever occur. |
204 | */ | |
a0db00fc | 205 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 206 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 207 | |
8c7c6e34 | 208 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
209 | enum res_type { |
210 | _MEM, | |
211 | _MEMSWAP, | |
510fc4e1 | 212 | _KMEM, |
d55f90bf | 213 | _TCP, |
86ae53e1 GC |
214 | }; |
215 | ||
a0db00fc KS |
216 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
217 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 KH |
218 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
219 | ||
b05706f1 KT |
220 | /* |
221 | * Iteration constructs for visiting all cgroups (under a tree). If | |
222 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
223 | * be used for reference counting. | |
224 | */ | |
225 | #define for_each_mem_cgroup_tree(iter, root) \ | |
226 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ | |
227 | iter != NULL; \ | |
228 | iter = mem_cgroup_iter(root, iter, NULL)) | |
229 | ||
230 | #define for_each_mem_cgroup(iter) \ | |
231 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ | |
232 | iter != NULL; \ | |
233 | iter = mem_cgroup_iter(NULL, iter, NULL)) | |
234 | ||
a4ebf1b6 | 235 | static inline bool task_is_dying(void) |
7775face TH |
236 | { |
237 | return tsk_is_oom_victim(current) || fatal_signal_pending(current) || | |
238 | (current->flags & PF_EXITING); | |
239 | } | |
240 | ||
70ddf637 AV |
241 | /* Some nice accessors for the vmpressure. */ |
242 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
243 | { | |
244 | if (!memcg) | |
245 | memcg = root_mem_cgroup; | |
246 | return &memcg->vmpressure; | |
247 | } | |
248 | ||
9647875b | 249 | struct mem_cgroup *vmpressure_to_memcg(struct vmpressure *vmpr) |
70ddf637 | 250 | { |
9647875b | 251 | return container_of(vmpr, struct mem_cgroup, vmpressure); |
70ddf637 AV |
252 | } |
253 | ||
1aacbd35 RG |
254 | #define CURRENT_OBJCG_UPDATE_BIT 0 |
255 | #define CURRENT_OBJCG_UPDATE_FLAG (1UL << CURRENT_OBJCG_UPDATE_BIT) | |
256 | ||
84c07d11 | 257 | #ifdef CONFIG_MEMCG_KMEM |
0764db9b | 258 | static DEFINE_SPINLOCK(objcg_lock); |
bf4f0599 | 259 | |
4d5c8aed RG |
260 | bool mem_cgroup_kmem_disabled(void) |
261 | { | |
262 | return cgroup_memory_nokmem; | |
263 | } | |
264 | ||
f1286fae MS |
265 | static void obj_cgroup_uncharge_pages(struct obj_cgroup *objcg, |
266 | unsigned int nr_pages); | |
c1a660de | 267 | |
bf4f0599 RG |
268 | static void obj_cgroup_release(struct percpu_ref *ref) |
269 | { | |
270 | struct obj_cgroup *objcg = container_of(ref, struct obj_cgroup, refcnt); | |
bf4f0599 RG |
271 | unsigned int nr_bytes; |
272 | unsigned int nr_pages; | |
273 | unsigned long flags; | |
274 | ||
275 | /* | |
276 | * At this point all allocated objects are freed, and | |
277 | * objcg->nr_charged_bytes can't have an arbitrary byte value. | |
278 | * However, it can be PAGE_SIZE or (x * PAGE_SIZE). | |
279 | * | |
280 | * The following sequence can lead to it: | |
281 | * 1) CPU0: objcg == stock->cached_objcg | |
282 | * 2) CPU1: we do a small allocation (e.g. 92 bytes), | |
283 | * PAGE_SIZE bytes are charged | |
284 | * 3) CPU1: a process from another memcg is allocating something, | |
285 | * the stock if flushed, | |
286 | * objcg->nr_charged_bytes = PAGE_SIZE - 92 | |
287 | * 5) CPU0: we do release this object, | |
288 | * 92 bytes are added to stock->nr_bytes | |
289 | * 6) CPU0: stock is flushed, | |
290 | * 92 bytes are added to objcg->nr_charged_bytes | |
291 | * | |
292 | * In the result, nr_charged_bytes == PAGE_SIZE. | |
293 | * This page will be uncharged in obj_cgroup_release(). | |
294 | */ | |
295 | nr_bytes = atomic_read(&objcg->nr_charged_bytes); | |
296 | WARN_ON_ONCE(nr_bytes & (PAGE_SIZE - 1)); | |
297 | nr_pages = nr_bytes >> PAGE_SHIFT; | |
298 | ||
bf4f0599 | 299 | if (nr_pages) |
f1286fae | 300 | obj_cgroup_uncharge_pages(objcg, nr_pages); |
271dd6b1 | 301 | |
0764db9b | 302 | spin_lock_irqsave(&objcg_lock, flags); |
bf4f0599 | 303 | list_del(&objcg->list); |
0764db9b | 304 | spin_unlock_irqrestore(&objcg_lock, flags); |
bf4f0599 RG |
305 | |
306 | percpu_ref_exit(ref); | |
307 | kfree_rcu(objcg, rcu); | |
308 | } | |
309 | ||
310 | static struct obj_cgroup *obj_cgroup_alloc(void) | |
311 | { | |
312 | struct obj_cgroup *objcg; | |
313 | int ret; | |
314 | ||
315 | objcg = kzalloc(sizeof(struct obj_cgroup), GFP_KERNEL); | |
316 | if (!objcg) | |
317 | return NULL; | |
318 | ||
319 | ret = percpu_ref_init(&objcg->refcnt, obj_cgroup_release, 0, | |
320 | GFP_KERNEL); | |
321 | if (ret) { | |
322 | kfree(objcg); | |
323 | return NULL; | |
324 | } | |
325 | INIT_LIST_HEAD(&objcg->list); | |
326 | return objcg; | |
327 | } | |
328 | ||
329 | static void memcg_reparent_objcgs(struct mem_cgroup *memcg, | |
330 | struct mem_cgroup *parent) | |
331 | { | |
332 | struct obj_cgroup *objcg, *iter; | |
333 | ||
334 | objcg = rcu_replace_pointer(memcg->objcg, NULL, true); | |
335 | ||
0764db9b | 336 | spin_lock_irq(&objcg_lock); |
bf4f0599 | 337 | |
9838354e MS |
338 | /* 1) Ready to reparent active objcg. */ |
339 | list_add(&objcg->list, &memcg->objcg_list); | |
340 | /* 2) Reparent active objcg and already reparented objcgs to parent. */ | |
341 | list_for_each_entry(iter, &memcg->objcg_list, list) | |
342 | WRITE_ONCE(iter->memcg, parent); | |
343 | /* 3) Move already reparented objcgs to the parent's list */ | |
bf4f0599 RG |
344 | list_splice(&memcg->objcg_list, &parent->objcg_list); |
345 | ||
0764db9b | 346 | spin_unlock_irq(&objcg_lock); |
bf4f0599 RG |
347 | |
348 | percpu_ref_kill(&objcg->refcnt); | |
349 | } | |
350 | ||
d7f25f8a GC |
351 | /* |
352 | * A lot of the calls to the cache allocation functions are expected to be | |
9f9796b4 | 353 | * inlined by the compiler. Since the calls to memcg_slab_post_alloc_hook() are |
d7f25f8a GC |
354 | * conditional to this static branch, we'll have to allow modules that does |
355 | * kmem_cache_alloc and the such to see this symbol as well | |
356 | */ | |
f7a449f7 RG |
357 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_online_key); |
358 | EXPORT_SYMBOL(memcg_kmem_online_key); | |
b6c1a8af YS |
359 | |
360 | DEFINE_STATIC_KEY_FALSE(memcg_bpf_enabled_key); | |
361 | EXPORT_SYMBOL(memcg_bpf_enabled_key); | |
0a432dcb | 362 | #endif |
17cc4dfe | 363 | |
ad7fa852 | 364 | /** |
75376c6f MWO |
365 | * mem_cgroup_css_from_folio - css of the memcg associated with a folio |
366 | * @folio: folio of interest | |
ad7fa852 TH |
367 | * |
368 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
75376c6f | 369 | * with @folio is returned. The returned css remains associated with @folio |
ad7fa852 TH |
370 | * until it is released. |
371 | * | |
372 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
373 | * is returned. | |
ad7fa852 | 374 | */ |
75376c6f | 375 | struct cgroup_subsys_state *mem_cgroup_css_from_folio(struct folio *folio) |
ad7fa852 | 376 | { |
75376c6f | 377 | struct mem_cgroup *memcg = folio_memcg(folio); |
ad7fa852 | 378 | |
9e10a130 | 379 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
380 | memcg = root_mem_cgroup; |
381 | ||
ad7fa852 TH |
382 | return &memcg->css; |
383 | } | |
384 | ||
2fc04524 VD |
385 | /** |
386 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
387 | * @page: the page | |
388 | * | |
389 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
390 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
391 | * is safe to call this function without holding a reference to @page. | |
392 | * | |
393 | * Note, this function is inherently racy, because there is nothing to prevent | |
394 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
395 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
396 | * do not care (such as procfs interfaces). | |
397 | */ | |
398 | ino_t page_cgroup_ino(struct page *page) | |
399 | { | |
400 | struct mem_cgroup *memcg; | |
401 | unsigned long ino = 0; | |
402 | ||
403 | rcu_read_lock(); | |
ec342603 YA |
404 | /* page_folio() is racy here, but the entire function is racy anyway */ |
405 | memcg = folio_memcg_check(page_folio(page)); | |
286e04b8 | 406 | |
2fc04524 VD |
407 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) |
408 | memcg = parent_mem_cgroup(memcg); | |
409 | if (memcg) | |
410 | ino = cgroup_ino(memcg->css.cgroup); | |
411 | rcu_read_unlock(); | |
412 | return ino; | |
413 | } | |
414 | ||
ef8f2327 MG |
415 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
416 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 417 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
418 | { |
419 | struct rb_node **p = &mctz->rb_root.rb_node; | |
420 | struct rb_node *parent = NULL; | |
ef8f2327 | 421 | struct mem_cgroup_per_node *mz_node; |
fa90b2fd | 422 | bool rightmost = true; |
bb4cc1a8 AM |
423 | |
424 | if (mz->on_tree) | |
425 | return; | |
426 | ||
427 | mz->usage_in_excess = new_usage_in_excess; | |
428 | if (!mz->usage_in_excess) | |
429 | return; | |
430 | while (*p) { | |
431 | parent = *p; | |
ef8f2327 | 432 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 | 433 | tree_node); |
fa90b2fd | 434 | if (mz->usage_in_excess < mz_node->usage_in_excess) { |
bb4cc1a8 | 435 | p = &(*p)->rb_left; |
fa90b2fd | 436 | rightmost = false; |
378876b0 | 437 | } else { |
bb4cc1a8 | 438 | p = &(*p)->rb_right; |
378876b0 | 439 | } |
bb4cc1a8 | 440 | } |
fa90b2fd DB |
441 | |
442 | if (rightmost) | |
443 | mctz->rb_rightmost = &mz->tree_node; | |
444 | ||
bb4cc1a8 AM |
445 | rb_link_node(&mz->tree_node, parent, p); |
446 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
447 | mz->on_tree = true; | |
448 | } | |
449 | ||
ef8f2327 MG |
450 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
451 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
452 | { |
453 | if (!mz->on_tree) | |
454 | return; | |
fa90b2fd DB |
455 | |
456 | if (&mz->tree_node == mctz->rb_rightmost) | |
457 | mctz->rb_rightmost = rb_prev(&mz->tree_node); | |
458 | ||
bb4cc1a8 AM |
459 | rb_erase(&mz->tree_node, &mctz->rb_root); |
460 | mz->on_tree = false; | |
461 | } | |
462 | ||
ef8f2327 MG |
463 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
464 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 465 | { |
0a31bc97 JW |
466 | unsigned long flags; |
467 | ||
468 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 469 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 470 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
471 | } |
472 | ||
3e32cb2e JW |
473 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
474 | { | |
475 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 476 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
477 | unsigned long excess = 0; |
478 | ||
479 | if (nr_pages > soft_limit) | |
480 | excess = nr_pages - soft_limit; | |
481 | ||
482 | return excess; | |
483 | } | |
bb4cc1a8 | 484 | |
658b69c9 | 485 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, int nid) |
bb4cc1a8 | 486 | { |
3e32cb2e | 487 | unsigned long excess; |
ef8f2327 MG |
488 | struct mem_cgroup_per_node *mz; |
489 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 490 | |
e4dde56c | 491 | if (lru_gen_enabled()) { |
36c7b4db | 492 | if (soft_limit_excess(memcg)) |
5c7e7a0d | 493 | lru_gen_soft_reclaim(memcg, nid); |
e4dde56c YZ |
494 | return; |
495 | } | |
496 | ||
2ab082ba | 497 | mctz = soft_limit_tree.rb_tree_per_node[nid]; |
bfc7228b LD |
498 | if (!mctz) |
499 | return; | |
bb4cc1a8 AM |
500 | /* |
501 | * Necessary to update all ancestors when hierarchy is used. | |
502 | * because their event counter is not touched. | |
503 | */ | |
504 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
658b69c9 | 505 | mz = memcg->nodeinfo[nid]; |
3e32cb2e | 506 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
507 | /* |
508 | * We have to update the tree if mz is on RB-tree or | |
509 | * mem is over its softlimit. | |
510 | */ | |
511 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
512 | unsigned long flags; |
513 | ||
514 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
515 | /* if on-tree, remove it */ |
516 | if (mz->on_tree) | |
cf2c8127 | 517 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
518 | /* |
519 | * Insert again. mz->usage_in_excess will be updated. | |
520 | * If excess is 0, no tree ops. | |
521 | */ | |
cf2c8127 | 522 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 523 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
524 | } |
525 | } | |
526 | } | |
527 | ||
528 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
529 | { | |
ef8f2327 MG |
530 | struct mem_cgroup_tree_per_node *mctz; |
531 | struct mem_cgroup_per_node *mz; | |
532 | int nid; | |
bb4cc1a8 | 533 | |
e231875b | 534 | for_each_node(nid) { |
a3747b53 | 535 | mz = memcg->nodeinfo[nid]; |
2ab082ba | 536 | mctz = soft_limit_tree.rb_tree_per_node[nid]; |
bfc7228b LD |
537 | if (mctz) |
538 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
539 | } |
540 | } | |
541 | ||
ef8f2327 MG |
542 | static struct mem_cgroup_per_node * |
543 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 544 | { |
ef8f2327 | 545 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
546 | |
547 | retry: | |
548 | mz = NULL; | |
fa90b2fd | 549 | if (!mctz->rb_rightmost) |
bb4cc1a8 AM |
550 | goto done; /* Nothing to reclaim from */ |
551 | ||
fa90b2fd DB |
552 | mz = rb_entry(mctz->rb_rightmost, |
553 | struct mem_cgroup_per_node, tree_node); | |
bb4cc1a8 AM |
554 | /* |
555 | * Remove the node now but someone else can add it back, | |
556 | * we will to add it back at the end of reclaim to its correct | |
557 | * position in the tree. | |
558 | */ | |
cf2c8127 | 559 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 560 | if (!soft_limit_excess(mz->memcg) || |
8965aa28 | 561 | !css_tryget(&mz->memcg->css)) |
bb4cc1a8 AM |
562 | goto retry; |
563 | done: | |
564 | return mz; | |
565 | } | |
566 | ||
ef8f2327 MG |
567 | static struct mem_cgroup_per_node * |
568 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 569 | { |
ef8f2327 | 570 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 571 | |
0a31bc97 | 572 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 573 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 574 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
575 | return mz; |
576 | } | |
577 | ||
ff48c71c SB |
578 | /* Subset of node_stat_item for memcg stats */ |
579 | static const unsigned int memcg_node_stat_items[] = { | |
580 | NR_INACTIVE_ANON, | |
581 | NR_ACTIVE_ANON, | |
582 | NR_INACTIVE_FILE, | |
583 | NR_ACTIVE_FILE, | |
584 | NR_UNEVICTABLE, | |
585 | NR_SLAB_RECLAIMABLE_B, | |
586 | NR_SLAB_UNRECLAIMABLE_B, | |
587 | WORKINGSET_REFAULT_ANON, | |
588 | WORKINGSET_REFAULT_FILE, | |
589 | WORKINGSET_ACTIVATE_ANON, | |
590 | WORKINGSET_ACTIVATE_FILE, | |
591 | WORKINGSET_RESTORE_ANON, | |
592 | WORKINGSET_RESTORE_FILE, | |
593 | WORKINGSET_NODERECLAIM, | |
594 | NR_ANON_MAPPED, | |
595 | NR_FILE_MAPPED, | |
596 | NR_FILE_PAGES, | |
597 | NR_FILE_DIRTY, | |
598 | NR_WRITEBACK, | |
599 | NR_SHMEM, | |
600 | NR_SHMEM_THPS, | |
601 | NR_FILE_THPS, | |
602 | NR_ANON_THPS, | |
603 | NR_KERNEL_STACK_KB, | |
604 | NR_PAGETABLE, | |
605 | NR_SECONDARY_PAGETABLE, | |
606 | #ifdef CONFIG_SWAP | |
607 | NR_SWAPCACHE, | |
608 | #endif | |
609 | }; | |
610 | ||
611 | static const unsigned int memcg_stat_items[] = { | |
612 | MEMCG_SWAP, | |
613 | MEMCG_SOCK, | |
614 | MEMCG_PERCPU_B, | |
615 | MEMCG_VMALLOC, | |
616 | MEMCG_KMEM, | |
617 | MEMCG_ZSWAP_B, | |
618 | MEMCG_ZSWAPPED, | |
619 | }; | |
620 | ||
621 | #define NR_MEMCG_NODE_STAT_ITEMS ARRAY_SIZE(memcg_node_stat_items) | |
622 | #define MEMCG_VMSTAT_SIZE (NR_MEMCG_NODE_STAT_ITEMS + \ | |
623 | ARRAY_SIZE(memcg_stat_items)) | |
624 | static int8_t mem_cgroup_stats_index[MEMCG_NR_STAT] __read_mostly; | |
625 | ||
626 | static void init_memcg_stats(void) | |
627 | { | |
628 | int8_t i, j = 0; | |
629 | ||
630 | BUILD_BUG_ON(MEMCG_NR_STAT >= S8_MAX); | |
631 | ||
632 | for (i = 0; i < NR_MEMCG_NODE_STAT_ITEMS; ++i) | |
633 | mem_cgroup_stats_index[memcg_node_stat_items[i]] = ++j; | |
634 | ||
635 | for (i = 0; i < ARRAY_SIZE(memcg_stat_items); ++i) | |
636 | mem_cgroup_stats_index[memcg_stat_items[i]] = ++j; | |
637 | } | |
638 | ||
639 | static inline int memcg_stats_index(int idx) | |
640 | { | |
641 | return mem_cgroup_stats_index[idx] - 1; | |
642 | } | |
643 | ||
70a64b79 SB |
644 | struct lruvec_stats_percpu { |
645 | /* Local (CPU and cgroup) state */ | |
ff48c71c | 646 | long state[NR_MEMCG_NODE_STAT_ITEMS]; |
70a64b79 SB |
647 | |
648 | /* Delta calculation for lockless upward propagation */ | |
ff48c71c | 649 | long state_prev[NR_MEMCG_NODE_STAT_ITEMS]; |
70a64b79 SB |
650 | }; |
651 | ||
652 | struct lruvec_stats { | |
653 | /* Aggregated (CPU and subtree) state */ | |
ff48c71c | 654 | long state[NR_MEMCG_NODE_STAT_ITEMS]; |
70a64b79 SB |
655 | |
656 | /* Non-hierarchical (CPU aggregated) state */ | |
ff48c71c | 657 | long state_local[NR_MEMCG_NODE_STAT_ITEMS]; |
70a64b79 SB |
658 | |
659 | /* Pending child counts during tree propagation */ | |
ff48c71c | 660 | long state_pending[NR_MEMCG_NODE_STAT_ITEMS]; |
70a64b79 SB |
661 | }; |
662 | ||
663 | unsigned long lruvec_page_state(struct lruvec *lruvec, enum node_stat_item idx) | |
664 | { | |
665 | struct mem_cgroup_per_node *pn; | |
666 | long x; | |
ff48c71c | 667 | int i; |
70a64b79 SB |
668 | |
669 | if (mem_cgroup_disabled()) | |
670 | return node_page_state(lruvec_pgdat(lruvec), idx); | |
671 | ||
ff48c71c | 672 | i = memcg_stats_index(idx); |
acb5fe2f | 673 | if (WARN_ONCE(i < 0, "%s: missing stat item %d\n", __func__, idx)) |
ff48c71c SB |
674 | return 0; |
675 | ||
70a64b79 | 676 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
ff48c71c | 677 | x = READ_ONCE(pn->lruvec_stats->state[i]); |
70a64b79 SB |
678 | #ifdef CONFIG_SMP |
679 | if (x < 0) | |
680 | x = 0; | |
681 | #endif | |
682 | return x; | |
683 | } | |
684 | ||
685 | unsigned long lruvec_page_state_local(struct lruvec *lruvec, | |
686 | enum node_stat_item idx) | |
687 | { | |
688 | struct mem_cgroup_per_node *pn; | |
acb5fe2f | 689 | long x; |
ff48c71c | 690 | int i; |
70a64b79 SB |
691 | |
692 | if (mem_cgroup_disabled()) | |
693 | return node_page_state(lruvec_pgdat(lruvec), idx); | |
694 | ||
ff48c71c | 695 | i = memcg_stats_index(idx); |
acb5fe2f | 696 | if (WARN_ONCE(i < 0, "%s: missing stat item %d\n", __func__, idx)) |
ff48c71c SB |
697 | return 0; |
698 | ||
70a64b79 | 699 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
ff48c71c | 700 | x = READ_ONCE(pn->lruvec_stats->state_local[i]); |
70a64b79 SB |
701 | #ifdef CONFIG_SMP |
702 | if (x < 0) | |
703 | x = 0; | |
704 | #endif | |
705 | return x; | |
706 | } | |
707 | ||
d396def5 SB |
708 | /* Subset of vm_event_item to report for memcg event stats */ |
709 | static const unsigned int memcg_vm_event_stat[] = { | |
8278f1c7 SB |
710 | PGPGIN, |
711 | PGPGOUT, | |
d396def5 SB |
712 | PGSCAN_KSWAPD, |
713 | PGSCAN_DIRECT, | |
57e9cc50 | 714 | PGSCAN_KHUGEPAGED, |
d396def5 SB |
715 | PGSTEAL_KSWAPD, |
716 | PGSTEAL_DIRECT, | |
57e9cc50 | 717 | PGSTEAL_KHUGEPAGED, |
d396def5 SB |
718 | PGFAULT, |
719 | PGMAJFAULT, | |
720 | PGREFILL, | |
721 | PGACTIVATE, | |
722 | PGDEACTIVATE, | |
723 | PGLAZYFREE, | |
724 | PGLAZYFREED, | |
725 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) | |
726 | ZSWPIN, | |
727 | ZSWPOUT, | |
e0bf1dc8 | 728 | ZSWPWB, |
d396def5 SB |
729 | #endif |
730 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
731 | THP_FAULT_ALLOC, | |
732 | THP_COLLAPSE_ALLOC, | |
811244a5 XH |
733 | THP_SWPOUT, |
734 | THP_SWPOUT_FALLBACK, | |
d396def5 SB |
735 | #endif |
736 | }; | |
737 | ||
8278f1c7 | 738 | #define NR_MEMCG_EVENTS ARRAY_SIZE(memcg_vm_event_stat) |
59142d87 | 739 | static int8_t mem_cgroup_events_index[NR_VM_EVENT_ITEMS] __read_mostly; |
8278f1c7 SB |
740 | |
741 | static void init_memcg_events(void) | |
742 | { | |
59142d87 SB |
743 | int8_t i; |
744 | ||
745 | BUILD_BUG_ON(NR_VM_EVENT_ITEMS >= S8_MAX); | |
8278f1c7 SB |
746 | |
747 | for (i = 0; i < NR_MEMCG_EVENTS; ++i) | |
748 | mem_cgroup_events_index[memcg_vm_event_stat[i]] = i + 1; | |
749 | } | |
750 | ||
751 | static inline int memcg_events_index(enum vm_event_item idx) | |
752 | { | |
753 | return mem_cgroup_events_index[idx] - 1; | |
754 | } | |
755 | ||
410f8e82 | 756 | struct memcg_vmstats_percpu { |
9cee7e8e YA |
757 | /* Stats updates since the last flush */ |
758 | unsigned int stats_updates; | |
759 | ||
760 | /* Cached pointers for fast iteration in memcg_rstat_updated() */ | |
761 | struct memcg_vmstats_percpu *parent; | |
762 | struct memcg_vmstats *vmstats; | |
763 | ||
764 | /* The above should fit a single cacheline for memcg_rstat_updated() */ | |
765 | ||
410f8e82 | 766 | /* Local (CPU and cgroup) page state & events */ |
ff48c71c | 767 | long state[MEMCG_VMSTAT_SIZE]; |
8278f1c7 | 768 | unsigned long events[NR_MEMCG_EVENTS]; |
410f8e82 SB |
769 | |
770 | /* Delta calculation for lockless upward propagation */ | |
ff48c71c | 771 | long state_prev[MEMCG_VMSTAT_SIZE]; |
8278f1c7 | 772 | unsigned long events_prev[NR_MEMCG_EVENTS]; |
410f8e82 SB |
773 | |
774 | /* Cgroup1: threshold notifications & softlimit tree updates */ | |
775 | unsigned long nr_page_events; | |
776 | unsigned long targets[MEM_CGROUP_NTARGETS]; | |
9cee7e8e | 777 | } ____cacheline_aligned; |
410f8e82 SB |
778 | |
779 | struct memcg_vmstats { | |
780 | /* Aggregated (CPU and subtree) page state & events */ | |
ff48c71c | 781 | long state[MEMCG_VMSTAT_SIZE]; |
8278f1c7 | 782 | unsigned long events[NR_MEMCG_EVENTS]; |
410f8e82 | 783 | |
f82e6bf9 | 784 | /* Non-hierarchical (CPU aggregated) page state & events */ |
ff48c71c | 785 | long state_local[MEMCG_VMSTAT_SIZE]; |
f82e6bf9 YA |
786 | unsigned long events_local[NR_MEMCG_EVENTS]; |
787 | ||
410f8e82 | 788 | /* Pending child counts during tree propagation */ |
ff48c71c | 789 | long state_pending[MEMCG_VMSTAT_SIZE]; |
8278f1c7 | 790 | unsigned long events_pending[NR_MEMCG_EVENTS]; |
8d59d221 YA |
791 | |
792 | /* Stats updates since the last flush */ | |
793 | atomic64_t stats_updates; | |
410f8e82 SB |
794 | }; |
795 | ||
11192d9c SB |
796 | /* |
797 | * memcg and lruvec stats flushing | |
798 | * | |
799 | * Many codepaths leading to stats update or read are performance sensitive and | |
800 | * adding stats flushing in such codepaths is not desirable. So, to optimize the | |
801 | * flushing the kernel does: | |
802 | * | |
803 | * 1) Periodically and asynchronously flush the stats every 2 seconds to not let | |
804 | * rstat update tree grow unbounded. | |
805 | * | |
806 | * 2) Flush the stats synchronously on reader side only when there are more than | |
807 | * (MEMCG_CHARGE_BATCH * nr_cpus) update events. Though this optimization | |
808 | * will let stats be out of sync by atmost (MEMCG_CHARGE_BATCH * nr_cpus) but | |
809 | * only for 2 seconds due to (1). | |
810 | */ | |
811 | static void flush_memcg_stats_dwork(struct work_struct *w); | |
812 | static DECLARE_DEFERRABLE_WORK(stats_flush_dwork, flush_memcg_stats_dwork); | |
508bed88 | 813 | static u64 flush_last_time; |
9b301615 SB |
814 | |
815 | #define FLUSH_TIME (2UL*HZ) | |
11192d9c | 816 | |
be3e67b5 SAS |
817 | /* |
818 | * Accessors to ensure that preemption is disabled on PREEMPT_RT because it can | |
819 | * not rely on this as part of an acquired spinlock_t lock. These functions are | |
820 | * never used in hardirq context on PREEMPT_RT and therefore disabling preemtion | |
821 | * is sufficient. | |
822 | */ | |
823 | static void memcg_stats_lock(void) | |
824 | { | |
e575d401 TG |
825 | preempt_disable_nested(); |
826 | VM_WARN_ON_IRQS_ENABLED(); | |
be3e67b5 SAS |
827 | } |
828 | ||
829 | static void __memcg_stats_lock(void) | |
830 | { | |
e575d401 | 831 | preempt_disable_nested(); |
be3e67b5 SAS |
832 | } |
833 | ||
834 | static void memcg_stats_unlock(void) | |
835 | { | |
e575d401 | 836 | preempt_enable_nested(); |
be3e67b5 SAS |
837 | } |
838 | ||
8d59d221 | 839 | |
9cee7e8e | 840 | static bool memcg_vmstats_needs_flush(struct memcg_vmstats *vmstats) |
8d59d221 | 841 | { |
9cee7e8e | 842 | return atomic64_read(&vmstats->stats_updates) > |
8d59d221 YA |
843 | MEMCG_CHARGE_BATCH * num_online_cpus(); |
844 | } | |
845 | ||
5b3be698 | 846 | static inline void memcg_rstat_updated(struct mem_cgroup *memcg, int val) |
11192d9c | 847 | { |
9cee7e8e | 848 | struct memcg_vmstats_percpu *statc; |
8d59d221 | 849 | int cpu = smp_processor_id(); |
78ec6f9d | 850 | unsigned int stats_updates; |
5b3be698 | 851 | |
f9d911ca YA |
852 | if (!val) |
853 | return; | |
854 | ||
8d59d221 | 855 | cgroup_rstat_updated(memcg->css.cgroup, cpu); |
9cee7e8e YA |
856 | statc = this_cpu_ptr(memcg->vmstats_percpu); |
857 | for (; statc; statc = statc->parent) { | |
78ec6f9d BL |
858 | stats_updates = READ_ONCE(statc->stats_updates) + abs(val); |
859 | WRITE_ONCE(statc->stats_updates, stats_updates); | |
860 | if (stats_updates < MEMCG_CHARGE_BATCH) | |
8d59d221 | 861 | continue; |
5b3be698 | 862 | |
873f64b7 | 863 | /* |
8d59d221 YA |
864 | * If @memcg is already flush-able, increasing stats_updates is |
865 | * redundant. Avoid the overhead of the atomic update. | |
873f64b7 | 866 | */ |
9cee7e8e | 867 | if (!memcg_vmstats_needs_flush(statc->vmstats)) |
78ec6f9d | 868 | atomic64_add(stats_updates, |
9cee7e8e | 869 | &statc->vmstats->stats_updates); |
78ec6f9d | 870 | WRITE_ONCE(statc->stats_updates, 0); |
5b3be698 | 871 | } |
11192d9c SB |
872 | } |
873 | ||
7d7ef0a4 | 874 | static void do_flush_stats(struct mem_cgroup *memcg) |
11192d9c | 875 | { |
7d7ef0a4 YA |
876 | if (mem_cgroup_is_root(memcg)) |
877 | WRITE_ONCE(flush_last_time, jiffies_64); | |
9fad9aee | 878 | |
7d7ef0a4 | 879 | cgroup_rstat_flush(memcg->css.cgroup); |
11192d9c SB |
880 | } |
881 | ||
7d7ef0a4 YA |
882 | /* |
883 | * mem_cgroup_flush_stats - flush the stats of a memory cgroup subtree | |
884 | * @memcg: root of the subtree to flush | |
885 | * | |
886 | * Flushing is serialized by the underlying global rstat lock. There is also a | |
887 | * minimum amount of work to be done even if there are no stat updates to flush. | |
888 | * Hence, we only flush the stats if the updates delta exceeds a threshold. This | |
889 | * avoids unnecessary work and contention on the underlying lock. | |
890 | */ | |
891 | void mem_cgroup_flush_stats(struct mem_cgroup *memcg) | |
11192d9c | 892 | { |
7d7ef0a4 YA |
893 | if (mem_cgroup_disabled()) |
894 | return; | |
895 | ||
896 | if (!memcg) | |
897 | memcg = root_mem_cgroup; | |
898 | ||
9cee7e8e | 899 | if (memcg_vmstats_needs_flush(memcg->vmstats)) |
7d7ef0a4 | 900 | do_flush_stats(memcg); |
9fad9aee YA |
901 | } |
902 | ||
7d7ef0a4 | 903 | void mem_cgroup_flush_stats_ratelimited(struct mem_cgroup *memcg) |
9b301615 | 904 | { |
508bed88 YA |
905 | /* Only flush if the periodic flusher is one full cycle late */ |
906 | if (time_after64(jiffies_64, READ_ONCE(flush_last_time) + 2*FLUSH_TIME)) | |
7d7ef0a4 | 907 | mem_cgroup_flush_stats(memcg); |
9b301615 SB |
908 | } |
909 | ||
11192d9c SB |
910 | static void flush_memcg_stats_dwork(struct work_struct *w) |
911 | { | |
9fad9aee | 912 | /* |
9cee7e8e | 913 | * Deliberately ignore memcg_vmstats_needs_flush() here so that flushing |
8d59d221 | 914 | * in latency-sensitive paths is as cheap as possible. |
9fad9aee | 915 | */ |
7d7ef0a4 | 916 | do_flush_stats(root_mem_cgroup); |
9b301615 | 917 | queue_delayed_work(system_unbound_wq, &stats_flush_dwork, FLUSH_TIME); |
11192d9c SB |
918 | } |
919 | ||
410f8e82 SB |
920 | unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx) |
921 | { | |
ff48c71c SB |
922 | long x; |
923 | int i = memcg_stats_index(idx); | |
924 | ||
acb5fe2f | 925 | if (WARN_ONCE(i < 0, "%s: missing stat item %d\n", __func__, idx)) |
ff48c71c SB |
926 | return 0; |
927 | ||
928 | x = READ_ONCE(memcg->vmstats->state[i]); | |
410f8e82 SB |
929 | #ifdef CONFIG_SMP |
930 | if (x < 0) | |
931 | x = 0; | |
932 | #endif | |
933 | return x; | |
934 | } | |
935 | ||
7bd5bc3c YA |
936 | static int memcg_page_state_unit(int item); |
937 | ||
938 | /* | |
939 | * Normalize the value passed into memcg_rstat_updated() to be in pages. Round | |
940 | * up non-zero sub-page updates to 1 page as zero page updates are ignored. | |
941 | */ | |
942 | static int memcg_state_val_in_pages(int idx, int val) | |
943 | { | |
944 | int unit = memcg_page_state_unit(idx); | |
945 | ||
946 | if (!val || unit == PAGE_SIZE) | |
947 | return val; | |
948 | else | |
949 | return max(val * unit / PAGE_SIZE, 1UL); | |
950 | } | |
951 | ||
db9adbcb JW |
952 | /** |
953 | * __mod_memcg_state - update cgroup memory statistics | |
954 | * @memcg: the memory cgroup | |
955 | * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item | |
956 | * @val: delta to add to the counter, can be negative | |
957 | */ | |
a94032b3 SB |
958 | void __mod_memcg_state(struct mem_cgroup *memcg, enum memcg_stat_item idx, |
959 | int val) | |
db9adbcb | 960 | { |
ff48c71c SB |
961 | int i = memcg_stats_index(idx); |
962 | ||
acb5fe2f SB |
963 | if (mem_cgroup_disabled()) |
964 | return; | |
965 | ||
966 | if (WARN_ONCE(i < 0, "%s: missing stat item %d\n", __func__, idx)) | |
db9adbcb JW |
967 | return; |
968 | ||
ff48c71c | 969 | __this_cpu_add(memcg->vmstats_percpu->state[i], val); |
7bd5bc3c | 970 | memcg_rstat_updated(memcg, memcg_state_val_in_pages(idx, val)); |
db9adbcb JW |
971 | } |
972 | ||
2d146aa3 | 973 | /* idx can be of type enum memcg_stat_item or node_stat_item. */ |
a18e6e6e JW |
974 | static unsigned long memcg_page_state_local(struct mem_cgroup *memcg, int idx) |
975 | { | |
ff48c71c SB |
976 | long x; |
977 | int i = memcg_stats_index(idx); | |
978 | ||
acb5fe2f | 979 | if (WARN_ONCE(i < 0, "%s: missing stat item %d\n", __func__, idx)) |
ff48c71c | 980 | return 0; |
a18e6e6e | 981 | |
ff48c71c | 982 | x = READ_ONCE(memcg->vmstats->state_local[i]); |
a18e6e6e JW |
983 | #ifdef CONFIG_SMP |
984 | if (x < 0) | |
985 | x = 0; | |
986 | #endif | |
987 | return x; | |
988 | } | |
989 | ||
91882c16 SB |
990 | static void __mod_memcg_lruvec_state(struct lruvec *lruvec, |
991 | enum node_stat_item idx, | |
992 | int val) | |
db9adbcb JW |
993 | { |
994 | struct mem_cgroup_per_node *pn; | |
42a30035 | 995 | struct mem_cgroup *memcg; |
ff48c71c SB |
996 | int i = memcg_stats_index(idx); |
997 | ||
acb5fe2f | 998 | if (WARN_ONCE(i < 0, "%s: missing stat item %d\n", __func__, idx)) |
ff48c71c | 999 | return; |
db9adbcb | 1000 | |
db9adbcb | 1001 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
42a30035 | 1002 | memcg = pn->memcg; |
db9adbcb | 1003 | |
be3e67b5 | 1004 | /* |
be16dd76 | 1005 | * The caller from rmap relies on disabled preemption because they never |
be3e67b5 SAS |
1006 | * update their counter from in-interrupt context. For these two |
1007 | * counters we check that the update is never performed from an | |
1008 | * interrupt context while other caller need to have disabled interrupt. | |
1009 | */ | |
1010 | __memcg_stats_lock(); | |
e575d401 | 1011 | if (IS_ENABLED(CONFIG_DEBUG_VM)) { |
be3e67b5 SAS |
1012 | switch (idx) { |
1013 | case NR_ANON_MAPPED: | |
1014 | case NR_FILE_MAPPED: | |
1015 | case NR_ANON_THPS: | |
be3e67b5 SAS |
1016 | WARN_ON_ONCE(!in_task()); |
1017 | break; | |
1018 | default: | |
e575d401 | 1019 | VM_WARN_ON_IRQS_ENABLED(); |
be3e67b5 SAS |
1020 | } |
1021 | } | |
1022 | ||
db9adbcb | 1023 | /* Update memcg */ |
ff48c71c | 1024 | __this_cpu_add(memcg->vmstats_percpu->state[i], val); |
db9adbcb | 1025 | |
b4c46484 | 1026 | /* Update lruvec */ |
ff48c71c | 1027 | __this_cpu_add(pn->lruvec_stats_percpu->state[i], val); |
11192d9c | 1028 | |
7bd5bc3c | 1029 | memcg_rstat_updated(memcg, memcg_state_val_in_pages(idx, val)); |
be3e67b5 | 1030 | memcg_stats_unlock(); |
db9adbcb JW |
1031 | } |
1032 | ||
eedc4e5a RG |
1033 | /** |
1034 | * __mod_lruvec_state - update lruvec memory statistics | |
1035 | * @lruvec: the lruvec | |
1036 | * @idx: the stat item | |
1037 | * @val: delta to add to the counter, can be negative | |
1038 | * | |
1039 | * The lruvec is the intersection of the NUMA node and a cgroup. This | |
1040 | * function updates the all three counters that are affected by a | |
1041 | * change of state at this level: per-node, per-cgroup, per-lruvec. | |
1042 | */ | |
1043 | void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, | |
1044 | int val) | |
1045 | { | |
1046 | /* Update node */ | |
1047 | __mod_node_page_state(lruvec_pgdat(lruvec), idx, val); | |
1048 | ||
1049 | /* Update memcg and lruvec */ | |
1050 | if (!mem_cgroup_disabled()) | |
1051 | __mod_memcg_lruvec_state(lruvec, idx, val); | |
1052 | } | |
1053 | ||
c701123b | 1054 | void __lruvec_stat_mod_folio(struct folio *folio, enum node_stat_item idx, |
c47d5032 SB |
1055 | int val) |
1056 | { | |
b4e0b68f | 1057 | struct mem_cgroup *memcg; |
c701123b | 1058 | pg_data_t *pgdat = folio_pgdat(folio); |
c47d5032 SB |
1059 | struct lruvec *lruvec; |
1060 | ||
b4e0b68f | 1061 | rcu_read_lock(); |
c701123b | 1062 | memcg = folio_memcg(folio); |
c47d5032 | 1063 | /* Untracked pages have no memcg, no lruvec. Update only the node */ |
d635a69d | 1064 | if (!memcg) { |
b4e0b68f | 1065 | rcu_read_unlock(); |
c47d5032 SB |
1066 | __mod_node_page_state(pgdat, idx, val); |
1067 | return; | |
1068 | } | |
1069 | ||
d635a69d | 1070 | lruvec = mem_cgroup_lruvec(memcg, pgdat); |
c47d5032 | 1071 | __mod_lruvec_state(lruvec, idx, val); |
b4e0b68f | 1072 | rcu_read_unlock(); |
c47d5032 | 1073 | } |
c701123b | 1074 | EXPORT_SYMBOL(__lruvec_stat_mod_folio); |
c47d5032 | 1075 | |
da3ceeff | 1076 | void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, int val) |
ec9f0238 | 1077 | { |
4f103c63 | 1078 | pg_data_t *pgdat = page_pgdat(virt_to_page(p)); |
ec9f0238 RG |
1079 | struct mem_cgroup *memcg; |
1080 | struct lruvec *lruvec; | |
1081 | ||
1082 | rcu_read_lock(); | |
fc4db90f | 1083 | memcg = mem_cgroup_from_slab_obj(p); |
ec9f0238 | 1084 | |
8faeb1ff MS |
1085 | /* |
1086 | * Untracked pages have no memcg, no lruvec. Update only the | |
1087 | * node. If we reparent the slab objects to the root memcg, | |
1088 | * when we free the slab object, we need to update the per-memcg | |
1089 | * vmstats to keep it correct for the root memcg. | |
1090 | */ | |
1091 | if (!memcg) { | |
ec9f0238 RG |
1092 | __mod_node_page_state(pgdat, idx, val); |
1093 | } else { | |
867e5e1d | 1094 | lruvec = mem_cgroup_lruvec(memcg, pgdat); |
ec9f0238 RG |
1095 | __mod_lruvec_state(lruvec, idx, val); |
1096 | } | |
1097 | rcu_read_unlock(); | |
1098 | } | |
1099 | ||
db9adbcb JW |
1100 | /** |
1101 | * __count_memcg_events - account VM events in a cgroup | |
1102 | * @memcg: the memory cgroup | |
1103 | * @idx: the event item | |
f0953a1b | 1104 | * @count: the number of events that occurred |
db9adbcb JW |
1105 | */ |
1106 | void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, | |
1107 | unsigned long count) | |
1108 | { | |
acb5fe2f | 1109 | int i = memcg_events_index(idx); |
8278f1c7 | 1110 | |
acb5fe2f SB |
1111 | if (mem_cgroup_disabled()) |
1112 | return; | |
1113 | ||
1114 | if (WARN_ONCE(i < 0, "%s: missing stat item %d\n", __func__, idx)) | |
db9adbcb JW |
1115 | return; |
1116 | ||
be3e67b5 | 1117 | memcg_stats_lock(); |
acb5fe2f | 1118 | __this_cpu_add(memcg->vmstats_percpu->events[i], count); |
5b3be698 | 1119 | memcg_rstat_updated(memcg, count); |
be3e67b5 | 1120 | memcg_stats_unlock(); |
db9adbcb JW |
1121 | } |
1122 | ||
42a30035 | 1123 | static unsigned long memcg_events(struct mem_cgroup *memcg, int event) |
e9f8974f | 1124 | { |
acb5fe2f | 1125 | int i = memcg_events_index(event); |
8278f1c7 | 1126 | |
acb5fe2f | 1127 | if (WARN_ONCE(i < 0, "%s: missing stat item %d\n", __func__, event)) |
8278f1c7 | 1128 | return 0; |
acb5fe2f SB |
1129 | |
1130 | return READ_ONCE(memcg->vmstats->events[i]); | |
e9f8974f JW |
1131 | } |
1132 | ||
42a30035 JW |
1133 | static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event) |
1134 | { | |
acb5fe2f | 1135 | int i = memcg_events_index(event); |
8278f1c7 | 1136 | |
acb5fe2f | 1137 | if (WARN_ONCE(i < 0, "%s: missing stat item %d\n", __func__, event)) |
8278f1c7 | 1138 | return 0; |
815744d7 | 1139 | |
acb5fe2f | 1140 | return READ_ONCE(memcg->vmstats->events_local[i]); |
42a30035 JW |
1141 | } |
1142 | ||
c0ff4b85 | 1143 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
3fba69a5 | 1144 | int nr_pages) |
d52aa412 | 1145 | { |
e401f176 KH |
1146 | /* pagein of a big page is an event. So, ignore page size */ |
1147 | if (nr_pages > 0) | |
c9019e9b | 1148 | __count_memcg_events(memcg, PGPGIN, 1); |
3751d604 | 1149 | else { |
c9019e9b | 1150 | __count_memcg_events(memcg, PGPGOUT, 1); |
3751d604 KH |
1151 | nr_pages = -nr_pages; /* for event */ |
1152 | } | |
e401f176 | 1153 | |
871789d4 | 1154 | __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages); |
6d12e2d8 KH |
1155 | } |
1156 | ||
f53d7ce3 JW |
1157 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
1158 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
1159 | { |
1160 | unsigned long val, next; | |
1161 | ||
871789d4 CD |
1162 | val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events); |
1163 | next = __this_cpu_read(memcg->vmstats_percpu->targets[target]); | |
7a159cc9 | 1164 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 1165 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
1166 | switch (target) { |
1167 | case MEM_CGROUP_TARGET_THRESH: | |
1168 | next = val + THRESHOLDS_EVENTS_TARGET; | |
1169 | break; | |
bb4cc1a8 AM |
1170 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
1171 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
1172 | break; | |
f53d7ce3 JW |
1173 | default: |
1174 | break; | |
1175 | } | |
871789d4 | 1176 | __this_cpu_write(memcg->vmstats_percpu->targets[target], next); |
f53d7ce3 | 1177 | return true; |
7a159cc9 | 1178 | } |
f53d7ce3 | 1179 | return false; |
d2265e6f KH |
1180 | } |
1181 | ||
1182 | /* | |
1183 | * Check events in order. | |
1184 | * | |
1185 | */ | |
8e88bd2d | 1186 | static void memcg_check_events(struct mem_cgroup *memcg, int nid) |
d2265e6f | 1187 | { |
2343e88d SAS |
1188 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) |
1189 | return; | |
1190 | ||
d2265e6f | 1191 | /* threshold event is triggered in finer grain than soft limit */ |
f53d7ce3 JW |
1192 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
1193 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 1194 | bool do_softlimit; |
f53d7ce3 | 1195 | |
bb4cc1a8 AM |
1196 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
1197 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
c0ff4b85 | 1198 | mem_cgroup_threshold(memcg); |
bb4cc1a8 | 1199 | if (unlikely(do_softlimit)) |
8e88bd2d | 1200 | mem_cgroup_update_tree(memcg, nid); |
0a31bc97 | 1201 | } |
d2265e6f KH |
1202 | } |
1203 | ||
cf475ad2 | 1204 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 1205 | { |
31a78f23 BS |
1206 | /* |
1207 | * mm_update_next_owner() may clear mm->owner to NULL | |
1208 | * if it races with swapoff, page migration, etc. | |
1209 | * So this can be called with p == NULL. | |
1210 | */ | |
1211 | if (unlikely(!p)) | |
1212 | return NULL; | |
1213 | ||
073219e9 | 1214 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 1215 | } |
33398cf2 | 1216 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 1217 | |
04f94e3f DS |
1218 | static __always_inline struct mem_cgroup *active_memcg(void) |
1219 | { | |
55a68c82 | 1220 | if (!in_task()) |
04f94e3f DS |
1221 | return this_cpu_read(int_active_memcg); |
1222 | else | |
1223 | return current->active_memcg; | |
1224 | } | |
1225 | ||
d46eb14b SB |
1226 | /** |
1227 | * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg. | |
1228 | * @mm: mm from which memcg should be extracted. It can be NULL. | |
1229 | * | |
04f94e3f DS |
1230 | * Obtain a reference on mm->memcg and returns it if successful. If mm |
1231 | * is NULL, then the memcg is chosen as follows: | |
1232 | * 1) The active memcg, if set. | |
1233 | * 2) current->mm->memcg, if available | |
1234 | * 3) root memcg | |
1235 | * If mem_cgroup is disabled, NULL is returned. | |
d46eb14b SB |
1236 | */ |
1237 | struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) | |
54595fe2 | 1238 | { |
d46eb14b SB |
1239 | struct mem_cgroup *memcg; |
1240 | ||
1241 | if (mem_cgroup_disabled()) | |
1242 | return NULL; | |
0b7f569e | 1243 | |
2884b6b7 MS |
1244 | /* |
1245 | * Page cache insertions can happen without an | |
1246 | * actual mm context, e.g. during disk probing | |
1247 | * on boot, loopback IO, acct() writes etc. | |
1248 | * | |
1249 | * No need to css_get on root memcg as the reference | |
1250 | * counting is disabled on the root level in the | |
1251 | * cgroup core. See CSS_NO_REF. | |
1252 | */ | |
04f94e3f DS |
1253 | if (unlikely(!mm)) { |
1254 | memcg = active_memcg(); | |
1255 | if (unlikely(memcg)) { | |
1256 | /* remote memcg must hold a ref */ | |
1257 | css_get(&memcg->css); | |
1258 | return memcg; | |
1259 | } | |
1260 | mm = current->mm; | |
1261 | if (unlikely(!mm)) | |
1262 | return root_mem_cgroup; | |
1263 | } | |
2884b6b7 | 1264 | |
54595fe2 KH |
1265 | rcu_read_lock(); |
1266 | do { | |
2884b6b7 MS |
1267 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1268 | if (unlikely(!memcg)) | |
df381975 | 1269 | memcg = root_mem_cgroup; |
00d484f3 | 1270 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 1271 | rcu_read_unlock(); |
c0ff4b85 | 1272 | return memcg; |
54595fe2 | 1273 | } |
d46eb14b SB |
1274 | EXPORT_SYMBOL(get_mem_cgroup_from_mm); |
1275 | ||
4b569387 NP |
1276 | /** |
1277 | * get_mem_cgroup_from_current - Obtain a reference on current task's memcg. | |
1278 | */ | |
1279 | struct mem_cgroup *get_mem_cgroup_from_current(void) | |
1280 | { | |
1281 | struct mem_cgroup *memcg; | |
1282 | ||
1283 | if (mem_cgroup_disabled()) | |
1284 | return NULL; | |
1285 | ||
1286 | again: | |
1287 | rcu_read_lock(); | |
1288 | memcg = mem_cgroup_from_task(current); | |
1289 | if (!css_tryget(&memcg->css)) { | |
1290 | rcu_read_unlock(); | |
1291 | goto again; | |
1292 | } | |
1293 | rcu_read_unlock(); | |
1294 | return memcg; | |
1295 | } | |
1296 | ||
5660048c JW |
1297 | /** |
1298 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1299 | * @root: hierarchy root | |
1300 | * @prev: previously returned memcg, NULL on first invocation | |
1301 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1302 | * | |
1303 | * Returns references to children of the hierarchy below @root, or | |
1304 | * @root itself, or %NULL after a full round-trip. | |
1305 | * | |
1306 | * Caller must pass the return value in @prev on subsequent | |
1307 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1308 | * to cancel a hierarchy walk before the round-trip is complete. | |
1309 | * | |
05bdc520 ML |
1310 | * Reclaimers can specify a node in @reclaim to divide up the memcgs |
1311 | * in the hierarchy among all concurrent reclaimers operating on the | |
1312 | * same node. | |
5660048c | 1313 | */ |
694fbc0f | 1314 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1315 | struct mem_cgroup *prev, |
694fbc0f | 1316 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1317 | { |
3f649ab7 | 1318 | struct mem_cgroup_reclaim_iter *iter; |
5ac8fb31 | 1319 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 1320 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1321 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1322 | |
694fbc0f AM |
1323 | if (mem_cgroup_disabled()) |
1324 | return NULL; | |
5660048c | 1325 | |
9f3a0d09 JW |
1326 | if (!root) |
1327 | root = root_mem_cgroup; | |
7d74b06f | 1328 | |
542f85f9 | 1329 | rcu_read_lock(); |
5f578161 | 1330 | |
5ac8fb31 | 1331 | if (reclaim) { |
ef8f2327 | 1332 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 1333 | |
a3747b53 | 1334 | mz = root->nodeinfo[reclaim->pgdat->node_id]; |
9da83f3f | 1335 | iter = &mz->iter; |
5ac8fb31 | 1336 | |
a9320aae WY |
1337 | /* |
1338 | * On start, join the current reclaim iteration cycle. | |
1339 | * Exit when a concurrent walker completes it. | |
1340 | */ | |
1341 | if (!prev) | |
1342 | reclaim->generation = iter->generation; | |
1343 | else if (reclaim->generation != iter->generation) | |
5ac8fb31 JW |
1344 | goto out_unlock; |
1345 | ||
6df38689 | 1346 | while (1) { |
4db0c3c2 | 1347 | pos = READ_ONCE(iter->position); |
6df38689 VD |
1348 | if (!pos || css_tryget(&pos->css)) |
1349 | break; | |
5ac8fb31 | 1350 | /* |
6df38689 VD |
1351 | * css reference reached zero, so iter->position will |
1352 | * be cleared by ->css_released. However, we should not | |
1353 | * rely on this happening soon, because ->css_released | |
1354 | * is called from a work queue, and by busy-waiting we | |
1355 | * might block it. So we clear iter->position right | |
1356 | * away. | |
5ac8fb31 | 1357 | */ |
6df38689 VD |
1358 | (void)cmpxchg(&iter->position, pos, NULL); |
1359 | } | |
89d8330c WY |
1360 | } else if (prev) { |
1361 | pos = prev; | |
5ac8fb31 JW |
1362 | } |
1363 | ||
1364 | if (pos) | |
1365 | css = &pos->css; | |
1366 | ||
1367 | for (;;) { | |
1368 | css = css_next_descendant_pre(css, &root->css); | |
1369 | if (!css) { | |
1370 | /* | |
1371 | * Reclaimers share the hierarchy walk, and a | |
1372 | * new one might jump in right at the end of | |
1373 | * the hierarchy - make sure they see at least | |
1374 | * one group and restart from the beginning. | |
1375 | */ | |
1376 | if (!prev) | |
1377 | continue; | |
1378 | break; | |
527a5ec9 | 1379 | } |
7d74b06f | 1380 | |
5ac8fb31 JW |
1381 | /* |
1382 | * Verify the css and acquire a reference. The root | |
1383 | * is provided by the caller, so we know it's alive | |
1384 | * and kicking, and don't take an extra reference. | |
1385 | */ | |
41555dad WY |
1386 | if (css == &root->css || css_tryget(css)) { |
1387 | memcg = mem_cgroup_from_css(css); | |
0b8f73e1 | 1388 | break; |
41555dad | 1389 | } |
9f3a0d09 | 1390 | } |
5ac8fb31 JW |
1391 | |
1392 | if (reclaim) { | |
5ac8fb31 | 1393 | /* |
6df38689 VD |
1394 | * The position could have already been updated by a competing |
1395 | * thread, so check that the value hasn't changed since we read | |
1396 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 1397 | */ |
6df38689 VD |
1398 | (void)cmpxchg(&iter->position, pos, memcg); |
1399 | ||
5ac8fb31 JW |
1400 | if (pos) |
1401 | css_put(&pos->css); | |
1402 | ||
1403 | if (!memcg) | |
1404 | iter->generation++; | |
9f3a0d09 | 1405 | } |
5ac8fb31 | 1406 | |
542f85f9 MH |
1407 | out_unlock: |
1408 | rcu_read_unlock(); | |
c40046f3 MH |
1409 | if (prev && prev != root) |
1410 | css_put(&prev->css); | |
1411 | ||
9f3a0d09 | 1412 | return memcg; |
14067bb3 | 1413 | } |
7d74b06f | 1414 | |
5660048c JW |
1415 | /** |
1416 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1417 | * @root: hierarchy root | |
1418 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1419 | */ | |
1420 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1421 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1422 | { |
1423 | if (!root) | |
1424 | root = root_mem_cgroup; | |
1425 | if (prev && prev != root) | |
1426 | css_put(&prev->css); | |
1427 | } | |
7d74b06f | 1428 | |
54a83d6b MC |
1429 | static void __invalidate_reclaim_iterators(struct mem_cgroup *from, |
1430 | struct mem_cgroup *dead_memcg) | |
6df38689 | 1431 | { |
6df38689 | 1432 | struct mem_cgroup_reclaim_iter *iter; |
ef8f2327 MG |
1433 | struct mem_cgroup_per_node *mz; |
1434 | int nid; | |
6df38689 | 1435 | |
54a83d6b | 1436 | for_each_node(nid) { |
a3747b53 | 1437 | mz = from->nodeinfo[nid]; |
9da83f3f YS |
1438 | iter = &mz->iter; |
1439 | cmpxchg(&iter->position, dead_memcg, NULL); | |
6df38689 VD |
1440 | } |
1441 | } | |
1442 | ||
54a83d6b MC |
1443 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1444 | { | |
1445 | struct mem_cgroup *memcg = dead_memcg; | |
1446 | struct mem_cgroup *last; | |
1447 | ||
1448 | do { | |
1449 | __invalidate_reclaim_iterators(memcg, dead_memcg); | |
1450 | last = memcg; | |
1451 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1452 | ||
1453 | /* | |
b8dd3ee9 | 1454 | * When cgroup1 non-hierarchy mode is used, |
54a83d6b MC |
1455 | * parent_mem_cgroup() does not walk all the way up to the |
1456 | * cgroup root (root_mem_cgroup). So we have to handle | |
1457 | * dead_memcg from cgroup root separately. | |
1458 | */ | |
7848ed62 | 1459 | if (!mem_cgroup_is_root(last)) |
54a83d6b MC |
1460 | __invalidate_reclaim_iterators(root_mem_cgroup, |
1461 | dead_memcg); | |
1462 | } | |
1463 | ||
7c5f64f8 VD |
1464 | /** |
1465 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1466 | * @memcg: hierarchy root | |
1467 | * @fn: function to call for each task | |
1468 | * @arg: argument passed to @fn | |
1469 | * | |
1470 | * This function iterates over tasks attached to @memcg or to any of its | |
1471 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
025b7799 Z |
1472 | * value, the function breaks the iteration loop. Otherwise, it will iterate |
1473 | * over all tasks and return 0. | |
7c5f64f8 VD |
1474 | * |
1475 | * This function must not be called for the root memory cgroup. | |
1476 | */ | |
025b7799 Z |
1477 | void mem_cgroup_scan_tasks(struct mem_cgroup *memcg, |
1478 | int (*fn)(struct task_struct *, void *), void *arg) | |
7c5f64f8 VD |
1479 | { |
1480 | struct mem_cgroup *iter; | |
1481 | int ret = 0; | |
1482 | ||
7848ed62 | 1483 | BUG_ON(mem_cgroup_is_root(memcg)); |
7c5f64f8 VD |
1484 | |
1485 | for_each_mem_cgroup_tree(iter, memcg) { | |
1486 | struct css_task_iter it; | |
1487 | struct task_struct *task; | |
1488 | ||
f168a9a5 | 1489 | css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it); |
7c5f64f8 VD |
1490 | while (!ret && (task = css_task_iter_next(&it))) |
1491 | ret = fn(task, arg); | |
1492 | css_task_iter_end(&it); | |
1493 | if (ret) { | |
1494 | mem_cgroup_iter_break(memcg, iter); | |
1495 | break; | |
1496 | } | |
1497 | } | |
7c5f64f8 VD |
1498 | } |
1499 | ||
6168d0da | 1500 | #ifdef CONFIG_DEBUG_VM |
e809c3fe | 1501 | void lruvec_memcg_debug(struct lruvec *lruvec, struct folio *folio) |
6168d0da AS |
1502 | { |
1503 | struct mem_cgroup *memcg; | |
1504 | ||
1505 | if (mem_cgroup_disabled()) | |
1506 | return; | |
1507 | ||
e809c3fe | 1508 | memcg = folio_memcg(folio); |
6168d0da AS |
1509 | |
1510 | if (!memcg) | |
7848ed62 | 1511 | VM_BUG_ON_FOLIO(!mem_cgroup_is_root(lruvec_memcg(lruvec)), folio); |
6168d0da | 1512 | else |
e809c3fe | 1513 | VM_BUG_ON_FOLIO(lruvec_memcg(lruvec) != memcg, folio); |
6168d0da AS |
1514 | } |
1515 | #endif | |
1516 | ||
6168d0da | 1517 | /** |
e809c3fe MWO |
1518 | * folio_lruvec_lock - Lock the lruvec for a folio. |
1519 | * @folio: Pointer to the folio. | |
6168d0da | 1520 | * |
d7e3aba5 | 1521 | * These functions are safe to use under any of the following conditions: |
e809c3fe MWO |
1522 | * - folio locked |
1523 | * - folio_test_lru false | |
1524 | * - folio_memcg_lock() | |
1525 | * - folio frozen (refcount of 0) | |
1526 | * | |
1527 | * Return: The lruvec this folio is on with its lock held. | |
6168d0da | 1528 | */ |
e809c3fe | 1529 | struct lruvec *folio_lruvec_lock(struct folio *folio) |
6168d0da | 1530 | { |
e809c3fe | 1531 | struct lruvec *lruvec = folio_lruvec(folio); |
6168d0da | 1532 | |
6168d0da | 1533 | spin_lock(&lruvec->lru_lock); |
e809c3fe | 1534 | lruvec_memcg_debug(lruvec, folio); |
6168d0da AS |
1535 | |
1536 | return lruvec; | |
1537 | } | |
1538 | ||
e809c3fe MWO |
1539 | /** |
1540 | * folio_lruvec_lock_irq - Lock the lruvec for a folio. | |
1541 | * @folio: Pointer to the folio. | |
1542 | * | |
1543 | * These functions are safe to use under any of the following conditions: | |
1544 | * - folio locked | |
1545 | * - folio_test_lru false | |
1546 | * - folio_memcg_lock() | |
1547 | * - folio frozen (refcount of 0) | |
1548 | * | |
1549 | * Return: The lruvec this folio is on with its lock held and interrupts | |
1550 | * disabled. | |
1551 | */ | |
1552 | struct lruvec *folio_lruvec_lock_irq(struct folio *folio) | |
6168d0da | 1553 | { |
e809c3fe | 1554 | struct lruvec *lruvec = folio_lruvec(folio); |
6168d0da | 1555 | |
6168d0da | 1556 | spin_lock_irq(&lruvec->lru_lock); |
e809c3fe | 1557 | lruvec_memcg_debug(lruvec, folio); |
6168d0da AS |
1558 | |
1559 | return lruvec; | |
1560 | } | |
1561 | ||
e809c3fe MWO |
1562 | /** |
1563 | * folio_lruvec_lock_irqsave - Lock the lruvec for a folio. | |
1564 | * @folio: Pointer to the folio. | |
1565 | * @flags: Pointer to irqsave flags. | |
1566 | * | |
1567 | * These functions are safe to use under any of the following conditions: | |
1568 | * - folio locked | |
1569 | * - folio_test_lru false | |
1570 | * - folio_memcg_lock() | |
1571 | * - folio frozen (refcount of 0) | |
1572 | * | |
1573 | * Return: The lruvec this folio is on with its lock held and interrupts | |
1574 | * disabled. | |
1575 | */ | |
1576 | struct lruvec *folio_lruvec_lock_irqsave(struct folio *folio, | |
1577 | unsigned long *flags) | |
6168d0da | 1578 | { |
e809c3fe | 1579 | struct lruvec *lruvec = folio_lruvec(folio); |
6168d0da | 1580 | |
6168d0da | 1581 | spin_lock_irqsave(&lruvec->lru_lock, *flags); |
e809c3fe | 1582 | lruvec_memcg_debug(lruvec, folio); |
6168d0da AS |
1583 | |
1584 | return lruvec; | |
1585 | } | |
1586 | ||
925b7673 | 1587 | /** |
fa9add64 HD |
1588 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1589 | * @lruvec: mem_cgroup per zone lru vector | |
1590 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1591 | * @zid: zone id of the accounted pages |
fa9add64 | 1592 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1593 | * |
ca707239 | 1594 | * This function must be called under lru_lock, just before a page is added |
07ca7606 | 1595 | * to or just after a page is removed from an lru list. |
3f58a829 | 1596 | */ |
fa9add64 | 1597 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1598 | int zid, int nr_pages) |
3f58a829 | 1599 | { |
ef8f2327 | 1600 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1601 | unsigned long *lru_size; |
ca707239 | 1602 | long size; |
3f58a829 MK |
1603 | |
1604 | if (mem_cgroup_disabled()) | |
1605 | return; | |
1606 | ||
ef8f2327 | 1607 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1608 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1609 | |
1610 | if (nr_pages < 0) | |
1611 | *lru_size += nr_pages; | |
1612 | ||
1613 | size = *lru_size; | |
b4536f0c MH |
1614 | if (WARN_ONCE(size < 0, |
1615 | "%s(%p, %d, %d): lru_size %ld\n", | |
1616 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1617 | VM_BUG_ON(1); |
1618 | *lru_size = 0; | |
1619 | } | |
1620 | ||
1621 | if (nr_pages > 0) | |
1622 | *lru_size += nr_pages; | |
08e552c6 | 1623 | } |
544122e5 | 1624 | |
19942822 | 1625 | /** |
9d11ea9f | 1626 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1627 | * @memcg: the memory cgroup |
19942822 | 1628 | * |
9d11ea9f | 1629 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1630 | * pages. |
19942822 | 1631 | */ |
c0ff4b85 | 1632 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1633 | { |
3e32cb2e JW |
1634 | unsigned long margin = 0; |
1635 | unsigned long count; | |
1636 | unsigned long limit; | |
9d11ea9f | 1637 | |
3e32cb2e | 1638 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1639 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1640 | if (count < limit) |
1641 | margin = limit - count; | |
1642 | ||
7941d214 | 1643 | if (do_memsw_account()) { |
3e32cb2e | 1644 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1645 | limit = READ_ONCE(memcg->memsw.max); |
1c4448ed | 1646 | if (count < limit) |
3e32cb2e | 1647 | margin = min(margin, limit - count); |
cbedbac3 LR |
1648 | else |
1649 | margin = 0; | |
3e32cb2e JW |
1650 | } |
1651 | ||
1652 | return margin; | |
19942822 JW |
1653 | } |
1654 | ||
32047e2a | 1655 | /* |
bdcbb659 | 1656 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1657 | * |
bdcbb659 QH |
1658 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1659 | * moving cgroups. This is for waiting at high-memory pressure | |
1660 | * caused by "move". | |
32047e2a | 1661 | */ |
c0ff4b85 | 1662 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1663 | { |
2bd9bb20 KH |
1664 | struct mem_cgroup *from; |
1665 | struct mem_cgroup *to; | |
4b534334 | 1666 | bool ret = false; |
2bd9bb20 KH |
1667 | /* |
1668 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1669 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1670 | */ | |
1671 | spin_lock(&mc.lock); | |
1672 | from = mc.from; | |
1673 | to = mc.to; | |
1674 | if (!from) | |
1675 | goto unlock; | |
3e92041d | 1676 | |
2314b42d JW |
1677 | ret = mem_cgroup_is_descendant(from, memcg) || |
1678 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1679 | unlock: |
1680 | spin_unlock(&mc.lock); | |
4b534334 KH |
1681 | return ret; |
1682 | } | |
1683 | ||
c0ff4b85 | 1684 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1685 | { |
1686 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1687 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1688 | DEFINE_WAIT(wait); |
1689 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1690 | /* moving charge context might have finished. */ | |
1691 | if (mc.moving_task) | |
1692 | schedule(); | |
1693 | finish_wait(&mc.waitq, &wait); | |
1694 | return true; | |
1695 | } | |
1696 | } | |
1697 | return false; | |
1698 | } | |
1699 | ||
5f9a4f4a MS |
1700 | struct memory_stat { |
1701 | const char *name; | |
5f9a4f4a MS |
1702 | unsigned int idx; |
1703 | }; | |
1704 | ||
57b2847d | 1705 | static const struct memory_stat memory_stats[] = { |
fff66b79 MS |
1706 | { "anon", NR_ANON_MAPPED }, |
1707 | { "file", NR_FILE_PAGES }, | |
a8c49af3 | 1708 | { "kernel", MEMCG_KMEM }, |
fff66b79 MS |
1709 | { "kernel_stack", NR_KERNEL_STACK_KB }, |
1710 | { "pagetables", NR_PAGETABLE }, | |
ebc97a52 | 1711 | { "sec_pagetables", NR_SECONDARY_PAGETABLE }, |
fff66b79 MS |
1712 | { "percpu", MEMCG_PERCPU_B }, |
1713 | { "sock", MEMCG_SOCK }, | |
4e5aa1f4 | 1714 | { "vmalloc", MEMCG_VMALLOC }, |
fff66b79 | 1715 | { "shmem", NR_SHMEM }, |
f4840ccf JW |
1716 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) |
1717 | { "zswap", MEMCG_ZSWAP_B }, | |
1718 | { "zswapped", MEMCG_ZSWAPPED }, | |
1719 | #endif | |
fff66b79 MS |
1720 | { "file_mapped", NR_FILE_MAPPED }, |
1721 | { "file_dirty", NR_FILE_DIRTY }, | |
1722 | { "file_writeback", NR_WRITEBACK }, | |
b6038942 SB |
1723 | #ifdef CONFIG_SWAP |
1724 | { "swapcached", NR_SWAPCACHE }, | |
1725 | #endif | |
5f9a4f4a | 1726 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
fff66b79 MS |
1727 | { "anon_thp", NR_ANON_THPS }, |
1728 | { "file_thp", NR_FILE_THPS }, | |
1729 | { "shmem_thp", NR_SHMEM_THPS }, | |
5f9a4f4a | 1730 | #endif |
fff66b79 MS |
1731 | { "inactive_anon", NR_INACTIVE_ANON }, |
1732 | { "active_anon", NR_ACTIVE_ANON }, | |
1733 | { "inactive_file", NR_INACTIVE_FILE }, | |
1734 | { "active_file", NR_ACTIVE_FILE }, | |
1735 | { "unevictable", NR_UNEVICTABLE }, | |
1736 | { "slab_reclaimable", NR_SLAB_RECLAIMABLE_B }, | |
1737 | { "slab_unreclaimable", NR_SLAB_UNRECLAIMABLE_B }, | |
5f9a4f4a MS |
1738 | |
1739 | /* The memory events */ | |
fff66b79 MS |
1740 | { "workingset_refault_anon", WORKINGSET_REFAULT_ANON }, |
1741 | { "workingset_refault_file", WORKINGSET_REFAULT_FILE }, | |
1742 | { "workingset_activate_anon", WORKINGSET_ACTIVATE_ANON }, | |
1743 | { "workingset_activate_file", WORKINGSET_ACTIVATE_FILE }, | |
1744 | { "workingset_restore_anon", WORKINGSET_RESTORE_ANON }, | |
1745 | { "workingset_restore_file", WORKINGSET_RESTORE_FILE }, | |
1746 | { "workingset_nodereclaim", WORKINGSET_NODERECLAIM }, | |
5f9a4f4a MS |
1747 | }; |
1748 | ||
ff841a06 | 1749 | /* The actual unit of the state item, not the same as the output unit */ |
fff66b79 MS |
1750 | static int memcg_page_state_unit(int item) |
1751 | { | |
1752 | switch (item) { | |
1753 | case MEMCG_PERCPU_B: | |
f4840ccf | 1754 | case MEMCG_ZSWAP_B: |
fff66b79 MS |
1755 | case NR_SLAB_RECLAIMABLE_B: |
1756 | case NR_SLAB_UNRECLAIMABLE_B: | |
ff841a06 YA |
1757 | return 1; |
1758 | case NR_KERNEL_STACK_KB: | |
1759 | return SZ_1K; | |
1760 | default: | |
1761 | return PAGE_SIZE; | |
1762 | } | |
1763 | } | |
1764 | ||
1765 | /* Translate stat items to the correct unit for memory.stat output */ | |
1766 | static int memcg_page_state_output_unit(int item) | |
1767 | { | |
1768 | /* | |
1769 | * Workingset state is actually in pages, but we export it to userspace | |
1770 | * as a scalar count of events, so special case it here. | |
1771 | */ | |
1772 | switch (item) { | |
fff66b79 MS |
1773 | case WORKINGSET_REFAULT_ANON: |
1774 | case WORKINGSET_REFAULT_FILE: | |
1775 | case WORKINGSET_ACTIVATE_ANON: | |
1776 | case WORKINGSET_ACTIVATE_FILE: | |
1777 | case WORKINGSET_RESTORE_ANON: | |
1778 | case WORKINGSET_RESTORE_FILE: | |
1779 | case WORKINGSET_NODERECLAIM: | |
1780 | return 1; | |
fff66b79 | 1781 | default: |
ff841a06 | 1782 | return memcg_page_state_unit(item); |
fff66b79 MS |
1783 | } |
1784 | } | |
1785 | ||
1786 | static inline unsigned long memcg_page_state_output(struct mem_cgroup *memcg, | |
1787 | int item) | |
1788 | { | |
ff841a06 YA |
1789 | return memcg_page_state(memcg, item) * |
1790 | memcg_page_state_output_unit(item); | |
1791 | } | |
1792 | ||
1793 | static inline unsigned long memcg_page_state_local_output( | |
1794 | struct mem_cgroup *memcg, int item) | |
1795 | { | |
1796 | return memcg_page_state_local(memcg, item) * | |
1797 | memcg_page_state_output_unit(item); | |
fff66b79 MS |
1798 | } |
1799 | ||
dddb44ff | 1800 | static void memcg_stat_format(struct mem_cgroup *memcg, struct seq_buf *s) |
c8713d0b | 1801 | { |
c8713d0b | 1802 | int i; |
71cd3113 | 1803 | |
c8713d0b JW |
1804 | /* |
1805 | * Provide statistics on the state of the memory subsystem as | |
1806 | * well as cumulative event counters that show past behavior. | |
1807 | * | |
1808 | * This list is ordered following a combination of these gradients: | |
1809 | * 1) generic big picture -> specifics and details | |
1810 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
1811 | * | |
1812 | * Current memory state: | |
1813 | */ | |
7d7ef0a4 | 1814 | mem_cgroup_flush_stats(memcg); |
c8713d0b | 1815 | |
5f9a4f4a MS |
1816 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { |
1817 | u64 size; | |
c8713d0b | 1818 | |
fff66b79 | 1819 | size = memcg_page_state_output(memcg, memory_stats[i].idx); |
5b42360c | 1820 | seq_buf_printf(s, "%s %llu\n", memory_stats[i].name, size); |
c8713d0b | 1821 | |
5f9a4f4a | 1822 | if (unlikely(memory_stats[i].idx == NR_SLAB_UNRECLAIMABLE_B)) { |
fff66b79 MS |
1823 | size += memcg_page_state_output(memcg, |
1824 | NR_SLAB_RECLAIMABLE_B); | |
5b42360c | 1825 | seq_buf_printf(s, "slab %llu\n", size); |
5f9a4f4a MS |
1826 | } |
1827 | } | |
c8713d0b JW |
1828 | |
1829 | /* Accumulated memory events */ | |
5b42360c | 1830 | seq_buf_printf(s, "pgscan %lu\n", |
c8713d0b | 1831 | memcg_events(memcg, PGSCAN_KSWAPD) + |
57e9cc50 JW |
1832 | memcg_events(memcg, PGSCAN_DIRECT) + |
1833 | memcg_events(memcg, PGSCAN_KHUGEPAGED)); | |
5b42360c | 1834 | seq_buf_printf(s, "pgsteal %lu\n", |
c8713d0b | 1835 | memcg_events(memcg, PGSTEAL_KSWAPD) + |
57e9cc50 JW |
1836 | memcg_events(memcg, PGSTEAL_DIRECT) + |
1837 | memcg_events(memcg, PGSTEAL_KHUGEPAGED)); | |
c8713d0b | 1838 | |
8278f1c7 SB |
1839 | for (i = 0; i < ARRAY_SIZE(memcg_vm_event_stat); i++) { |
1840 | if (memcg_vm_event_stat[i] == PGPGIN || | |
1841 | memcg_vm_event_stat[i] == PGPGOUT) | |
1842 | continue; | |
1843 | ||
5b42360c | 1844 | seq_buf_printf(s, "%s %lu\n", |
673520f8 QZ |
1845 | vm_event_name(memcg_vm_event_stat[i]), |
1846 | memcg_events(memcg, memcg_vm_event_stat[i])); | |
8278f1c7 | 1847 | } |
c8713d0b JW |
1848 | |
1849 | /* The above should easily fit into one page */ | |
5b42360c | 1850 | WARN_ON_ONCE(seq_buf_has_overflowed(s)); |
c8713d0b | 1851 | } |
71cd3113 | 1852 | |
dddb44ff YA |
1853 | static void memcg1_stat_format(struct mem_cgroup *memcg, struct seq_buf *s); |
1854 | ||
1855 | static void memory_stat_format(struct mem_cgroup *memcg, struct seq_buf *s) | |
1856 | { | |
1857 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
1858 | memcg_stat_format(memcg, s); | |
1859 | else | |
1860 | memcg1_stat_format(memcg, s); | |
1861 | WARN_ON_ONCE(seq_buf_has_overflowed(s)); | |
1862 | } | |
1863 | ||
e222432b | 1864 | /** |
f0c867d9 | 1865 | * mem_cgroup_print_oom_context: Print OOM information relevant to |
1866 | * memory controller. | |
e222432b BS |
1867 | * @memcg: The memory cgroup that went over limit |
1868 | * @p: Task that is going to be killed | |
1869 | * | |
1870 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1871 | * enabled | |
1872 | */ | |
f0c867d9 | 1873 | void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) |
e222432b | 1874 | { |
e222432b BS |
1875 | rcu_read_lock(); |
1876 | ||
f0c867d9 | 1877 | if (memcg) { |
1878 | pr_cont(",oom_memcg="); | |
1879 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1880 | } else | |
1881 | pr_cont(",global_oom"); | |
2415b9f5 | 1882 | if (p) { |
f0c867d9 | 1883 | pr_cont(",task_memcg="); |
2415b9f5 | 1884 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
2415b9f5 | 1885 | } |
e222432b | 1886 | rcu_read_unlock(); |
f0c867d9 | 1887 | } |
1888 | ||
1889 | /** | |
1890 | * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to | |
1891 | * memory controller. | |
1892 | * @memcg: The memory cgroup that went over limit | |
1893 | */ | |
1894 | void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) | |
1895 | { | |
68aaee14 TH |
1896 | /* Use static buffer, for the caller is holding oom_lock. */ |
1897 | static char buf[PAGE_SIZE]; | |
5b42360c | 1898 | struct seq_buf s; |
68aaee14 TH |
1899 | |
1900 | lockdep_assert_held(&oom_lock); | |
e222432b | 1901 | |
3e32cb2e JW |
1902 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1903 | K((u64)page_counter_read(&memcg->memory)), | |
15b42562 | 1904 | K((u64)READ_ONCE(memcg->memory.max)), memcg->memory.failcnt); |
c8713d0b JW |
1905 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
1906 | pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1907 | K((u64)page_counter_read(&memcg->swap)), | |
32d087cd | 1908 | K((u64)READ_ONCE(memcg->swap.max)), memcg->swap.failcnt); |
c8713d0b JW |
1909 | else { |
1910 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1911 | K((u64)page_counter_read(&memcg->memsw)), | |
1912 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); | |
1913 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1914 | K((u64)page_counter_read(&memcg->kmem)), | |
1915 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); | |
58cf188e | 1916 | } |
c8713d0b JW |
1917 | |
1918 | pr_info("Memory cgroup stats for "); | |
1919 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1920 | pr_cont(":"); | |
5b42360c YA |
1921 | seq_buf_init(&s, buf, sizeof(buf)); |
1922 | memory_stat_format(memcg, &s); | |
1923 | seq_buf_do_printk(&s, KERN_INFO); | |
e222432b BS |
1924 | } |
1925 | ||
a63d83f4 DR |
1926 | /* |
1927 | * Return the memory (and swap, if configured) limit for a memcg. | |
1928 | */ | |
bbec2e15 | 1929 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1930 | { |
8d387a5f WL |
1931 | unsigned long max = READ_ONCE(memcg->memory.max); |
1932 | ||
b94c4e94 | 1933 | if (do_memsw_account()) { |
8d387a5f WL |
1934 | if (mem_cgroup_swappiness(memcg)) { |
1935 | /* Calculate swap excess capacity from memsw limit */ | |
1936 | unsigned long swap = READ_ONCE(memcg->memsw.max) - max; | |
1937 | ||
1938 | max += min(swap, (unsigned long)total_swap_pages); | |
1939 | } | |
b94c4e94 JW |
1940 | } else { |
1941 | if (mem_cgroup_swappiness(memcg)) | |
1942 | max += min(READ_ONCE(memcg->swap.max), | |
1943 | (unsigned long)total_swap_pages); | |
9a5a8f19 | 1944 | } |
bbec2e15 | 1945 | return max; |
a63d83f4 DR |
1946 | } |
1947 | ||
9783aa99 CD |
1948 | unsigned long mem_cgroup_size(struct mem_cgroup *memcg) |
1949 | { | |
1950 | return page_counter_read(&memcg->memory); | |
1951 | } | |
1952 | ||
b6e6edcf | 1953 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1954 | int order) |
9cbb78bb | 1955 | { |
6e0fc46d DR |
1956 | struct oom_control oc = { |
1957 | .zonelist = NULL, | |
1958 | .nodemask = NULL, | |
2a966b77 | 1959 | .memcg = memcg, |
6e0fc46d DR |
1960 | .gfp_mask = gfp_mask, |
1961 | .order = order, | |
6e0fc46d | 1962 | }; |
1378b37d | 1963 | bool ret = true; |
9cbb78bb | 1964 | |
7775face TH |
1965 | if (mutex_lock_killable(&oom_lock)) |
1966 | return true; | |
1378b37d YS |
1967 | |
1968 | if (mem_cgroup_margin(memcg) >= (1 << order)) | |
1969 | goto unlock; | |
1970 | ||
7775face TH |
1971 | /* |
1972 | * A few threads which were not waiting at mutex_lock_killable() can | |
1973 | * fail to bail out. Therefore, check again after holding oom_lock. | |
1974 | */ | |
a4ebf1b6 | 1975 | ret = task_is_dying() || out_of_memory(&oc); |
1378b37d YS |
1976 | |
1977 | unlock: | |
dc56401f | 1978 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1979 | return ret; |
9cbb78bb DR |
1980 | } |
1981 | ||
0608f43d | 1982 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1983 | pg_data_t *pgdat, |
0608f43d AM |
1984 | gfp_t gfp_mask, |
1985 | unsigned long *total_scanned) | |
1986 | { | |
1987 | struct mem_cgroup *victim = NULL; | |
1988 | int total = 0; | |
1989 | int loop = 0; | |
1990 | unsigned long excess; | |
1991 | unsigned long nr_scanned; | |
1992 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1993 | .pgdat = pgdat, |
0608f43d AM |
1994 | }; |
1995 | ||
3e32cb2e | 1996 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1997 | |
1998 | while (1) { | |
1999 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
2000 | if (!victim) { | |
2001 | loop++; | |
2002 | if (loop >= 2) { | |
2003 | /* | |
2004 | * If we have not been able to reclaim | |
2005 | * anything, it might because there are | |
2006 | * no reclaimable pages under this hierarchy | |
2007 | */ | |
2008 | if (!total) | |
2009 | break; | |
2010 | /* | |
2011 | * We want to do more targeted reclaim. | |
2012 | * excess >> 2 is not to excessive so as to | |
2013 | * reclaim too much, nor too less that we keep | |
2014 | * coming back to reclaim from this cgroup | |
2015 | */ | |
2016 | if (total >= (excess >> 2) || | |
2017 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
2018 | break; | |
2019 | } | |
2020 | continue; | |
2021 | } | |
a9dd0a83 | 2022 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 2023 | pgdat, &nr_scanned); |
0608f43d | 2024 | *total_scanned += nr_scanned; |
3e32cb2e | 2025 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 2026 | break; |
6d61ef40 | 2027 | } |
0608f43d AM |
2028 | mem_cgroup_iter_break(root_memcg, victim); |
2029 | return total; | |
6d61ef40 BS |
2030 | } |
2031 | ||
0056f4e6 JW |
2032 | #ifdef CONFIG_LOCKDEP |
2033 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
2034 | .name = "memcg_oom_lock", | |
2035 | }; | |
2036 | #endif | |
2037 | ||
fb2a6fc5 JW |
2038 | static DEFINE_SPINLOCK(memcg_oom_lock); |
2039 | ||
867578cb KH |
2040 | /* |
2041 | * Check OOM-Killer is already running under our hierarchy. | |
2042 | * If someone is running, return false. | |
2043 | */ | |
fb2a6fc5 | 2044 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 2045 | { |
79dfdacc | 2046 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 2047 | |
fb2a6fc5 JW |
2048 | spin_lock(&memcg_oom_lock); |
2049 | ||
9f3a0d09 | 2050 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 2051 | if (iter->oom_lock) { |
79dfdacc MH |
2052 | /* |
2053 | * this subtree of our hierarchy is already locked | |
2054 | * so we cannot give a lock. | |
2055 | */ | |
79dfdacc | 2056 | failed = iter; |
9f3a0d09 JW |
2057 | mem_cgroup_iter_break(memcg, iter); |
2058 | break; | |
23751be0 JW |
2059 | } else |
2060 | iter->oom_lock = true; | |
7d74b06f | 2061 | } |
867578cb | 2062 | |
fb2a6fc5 JW |
2063 | if (failed) { |
2064 | /* | |
2065 | * OK, we failed to lock the whole subtree so we have | |
2066 | * to clean up what we set up to the failing subtree | |
2067 | */ | |
2068 | for_each_mem_cgroup_tree(iter, memcg) { | |
2069 | if (iter == failed) { | |
2070 | mem_cgroup_iter_break(memcg, iter); | |
2071 | break; | |
2072 | } | |
2073 | iter->oom_lock = false; | |
79dfdacc | 2074 | } |
0056f4e6 JW |
2075 | } else |
2076 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
2077 | |
2078 | spin_unlock(&memcg_oom_lock); | |
2079 | ||
2080 | return !failed; | |
a636b327 | 2081 | } |
0b7f569e | 2082 | |
fb2a6fc5 | 2083 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 2084 | { |
7d74b06f KH |
2085 | struct mem_cgroup *iter; |
2086 | ||
fb2a6fc5 | 2087 | spin_lock(&memcg_oom_lock); |
5facae4f | 2088 | mutex_release(&memcg_oom_lock_dep_map, _RET_IP_); |
c0ff4b85 | 2089 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 2090 | iter->oom_lock = false; |
fb2a6fc5 | 2091 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
2092 | } |
2093 | ||
c0ff4b85 | 2094 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2095 | { |
2096 | struct mem_cgroup *iter; | |
2097 | ||
c2b42d3c | 2098 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 2099 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
2100 | iter->under_oom++; |
2101 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
2102 | } |
2103 | ||
c0ff4b85 | 2104 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2105 | { |
2106 | struct mem_cgroup *iter; | |
2107 | ||
867578cb | 2108 | /* |
f0953a1b | 2109 | * Be careful about under_oom underflows because a child memcg |
7a52d4d8 | 2110 | * could have been added after mem_cgroup_mark_under_oom. |
867578cb | 2111 | */ |
c2b42d3c | 2112 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 2113 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
2114 | if (iter->under_oom > 0) |
2115 | iter->under_oom--; | |
2116 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
2117 | } |
2118 | ||
867578cb KH |
2119 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
2120 | ||
dc98df5a | 2121 | struct oom_wait_info { |
d79154bb | 2122 | struct mem_cgroup *memcg; |
ac6424b9 | 2123 | wait_queue_entry_t wait; |
dc98df5a KH |
2124 | }; |
2125 | ||
ac6424b9 | 2126 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
2127 | unsigned mode, int sync, void *arg) |
2128 | { | |
d79154bb HD |
2129 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
2130 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
2131 | struct oom_wait_info *oom_wait_info; |
2132 | ||
2133 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 2134 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 2135 | |
2314b42d JW |
2136 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
2137 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 2138 | return 0; |
dc98df5a KH |
2139 | return autoremove_wake_function(wait, mode, sync, arg); |
2140 | } | |
2141 | ||
c0ff4b85 | 2142 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 2143 | { |
c2b42d3c TH |
2144 | /* |
2145 | * For the following lockless ->under_oom test, the only required | |
2146 | * guarantee is that it must see the state asserted by an OOM when | |
2147 | * this function is called as a result of userland actions | |
2148 | * triggered by the notification of the OOM. This is trivially | |
2149 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
2150 | * triggering notification. | |
2151 | */ | |
2152 | if (memcg && memcg->under_oom) | |
f4b90b70 | 2153 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
2154 | } |
2155 | ||
becdf89d SB |
2156 | /* |
2157 | * Returns true if successfully killed one or more processes. Though in some | |
2158 | * corner cases it can return true even without killing any process. | |
2159 | */ | |
2160 | static bool mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 2161 | { |
becdf89d | 2162 | bool locked, ret; |
7056d3a3 | 2163 | |
29ef680a | 2164 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
becdf89d | 2165 | return false; |
29ef680a | 2166 | |
7a1adfdd RG |
2167 | memcg_memory_event(memcg, MEMCG_OOM); |
2168 | ||
867578cb | 2169 | /* |
49426420 JW |
2170 | * We are in the middle of the charge context here, so we |
2171 | * don't want to block when potentially sitting on a callstack | |
2172 | * that holds all kinds of filesystem and mm locks. | |
2173 | * | |
29ef680a MH |
2174 | * cgroup1 allows disabling the OOM killer and waiting for outside |
2175 | * handling until the charge can succeed; remember the context and put | |
2176 | * the task to sleep at the end of the page fault when all locks are | |
2177 | * released. | |
49426420 | 2178 | * |
29ef680a MH |
2179 | * On the other hand, in-kernel OOM killer allows for an async victim |
2180 | * memory reclaim (oom_reaper) and that means that we are not solely | |
2181 | * relying on the oom victim to make a forward progress and we can | |
2182 | * invoke the oom killer here. | |
2183 | * | |
2184 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
2185 | * victim and then we have to bail out from the charge path. | |
867578cb | 2186 | */ |
17c56de6 | 2187 | if (READ_ONCE(memcg->oom_kill_disable)) { |
becdf89d SB |
2188 | if (current->in_user_fault) { |
2189 | css_get(&memcg->css); | |
2190 | current->memcg_in_oom = memcg; | |
becdf89d SB |
2191 | } |
2192 | return false; | |
29ef680a MH |
2193 | } |
2194 | ||
7056d3a3 MH |
2195 | mem_cgroup_mark_under_oom(memcg); |
2196 | ||
2197 | locked = mem_cgroup_oom_trylock(memcg); | |
2198 | ||
2199 | if (locked) | |
2200 | mem_cgroup_oom_notify(memcg); | |
2201 | ||
2202 | mem_cgroup_unmark_under_oom(memcg); | |
becdf89d | 2203 | ret = mem_cgroup_out_of_memory(memcg, mask, order); |
7056d3a3 MH |
2204 | |
2205 | if (locked) | |
2206 | mem_cgroup_oom_unlock(memcg); | |
29ef680a | 2207 | |
7056d3a3 | 2208 | return ret; |
3812c8c8 JW |
2209 | } |
2210 | ||
2211 | /** | |
2212 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 2213 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 2214 | * |
49426420 JW |
2215 | * This has to be called at the end of a page fault if the memcg OOM |
2216 | * handler was enabled. | |
3812c8c8 | 2217 | * |
49426420 | 2218 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
2219 | * sleep on a waitqueue until the userspace task resolves the |
2220 | * situation. Sleeping directly in the charge context with all kinds | |
2221 | * of locks held is not a good idea, instead we remember an OOM state | |
2222 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 2223 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
2224 | * |
2225 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 2226 | * completed, %false otherwise. |
3812c8c8 | 2227 | */ |
49426420 | 2228 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 2229 | { |
626ebc41 | 2230 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 2231 | struct oom_wait_info owait; |
49426420 | 2232 | bool locked; |
3812c8c8 JW |
2233 | |
2234 | /* OOM is global, do not handle */ | |
3812c8c8 | 2235 | if (!memcg) |
49426420 | 2236 | return false; |
3812c8c8 | 2237 | |
7c5f64f8 | 2238 | if (!handle) |
49426420 | 2239 | goto cleanup; |
3812c8c8 JW |
2240 | |
2241 | owait.memcg = memcg; | |
2242 | owait.wait.flags = 0; | |
2243 | owait.wait.func = memcg_oom_wake_function; | |
2244 | owait.wait.private = current; | |
2055da97 | 2245 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 2246 | |
3812c8c8 | 2247 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
2248 | mem_cgroup_mark_under_oom(memcg); |
2249 | ||
2250 | locked = mem_cgroup_oom_trylock(memcg); | |
2251 | ||
2252 | if (locked) | |
2253 | mem_cgroup_oom_notify(memcg); | |
2254 | ||
857f2139 HX |
2255 | schedule(); |
2256 | mem_cgroup_unmark_under_oom(memcg); | |
2257 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
49426420 | 2258 | |
18b1d18b | 2259 | if (locked) |
fb2a6fc5 | 2260 | mem_cgroup_oom_unlock(memcg); |
49426420 | 2261 | cleanup: |
626ebc41 | 2262 | current->memcg_in_oom = NULL; |
3812c8c8 | 2263 | css_put(&memcg->css); |
867578cb | 2264 | return true; |
0b7f569e KH |
2265 | } |
2266 | ||
3d8b38eb RG |
2267 | /** |
2268 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
2269 | * @victim: task to be killed by the OOM killer | |
2270 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
2271 | * | |
2272 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
2273 | * by killing all belonging OOM-killable tasks. | |
2274 | * | |
2275 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
2276 | */ | |
2277 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
2278 | struct mem_cgroup *oom_domain) | |
2279 | { | |
2280 | struct mem_cgroup *oom_group = NULL; | |
2281 | struct mem_cgroup *memcg; | |
2282 | ||
2283 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
2284 | return NULL; | |
2285 | ||
2286 | if (!oom_domain) | |
2287 | oom_domain = root_mem_cgroup; | |
2288 | ||
2289 | rcu_read_lock(); | |
2290 | ||
2291 | memcg = mem_cgroup_from_task(victim); | |
7848ed62 | 2292 | if (mem_cgroup_is_root(memcg)) |
3d8b38eb RG |
2293 | goto out; |
2294 | ||
48fe267c RG |
2295 | /* |
2296 | * If the victim task has been asynchronously moved to a different | |
2297 | * memory cgroup, we might end up killing tasks outside oom_domain. | |
2298 | * In this case it's better to ignore memory.group.oom. | |
2299 | */ | |
2300 | if (unlikely(!mem_cgroup_is_descendant(memcg, oom_domain))) | |
2301 | goto out; | |
2302 | ||
3d8b38eb RG |
2303 | /* |
2304 | * Traverse the memory cgroup hierarchy from the victim task's | |
2305 | * cgroup up to the OOMing cgroup (or root) to find the | |
2306 | * highest-level memory cgroup with oom.group set. | |
2307 | */ | |
2308 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
eaf7b66b | 2309 | if (READ_ONCE(memcg->oom_group)) |
3d8b38eb RG |
2310 | oom_group = memcg; |
2311 | ||
2312 | if (memcg == oom_domain) | |
2313 | break; | |
2314 | } | |
2315 | ||
2316 | if (oom_group) | |
2317 | css_get(&oom_group->css); | |
2318 | out: | |
2319 | rcu_read_unlock(); | |
2320 | ||
2321 | return oom_group; | |
2322 | } | |
2323 | ||
2324 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
2325 | { | |
2326 | pr_info("Tasks in "); | |
2327 | pr_cont_cgroup_path(memcg->css.cgroup); | |
2328 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
2329 | } | |
2330 | ||
d7365e78 | 2331 | /** |
f70ad448 MWO |
2332 | * folio_memcg_lock - Bind a folio to its memcg. |
2333 | * @folio: The folio. | |
32047e2a | 2334 | * |
f70ad448 | 2335 | * This function prevents unlocked LRU folios from being moved to |
739f79fc JW |
2336 | * another cgroup. |
2337 | * | |
f70ad448 MWO |
2338 | * It ensures lifetime of the bound memcg. The caller is responsible |
2339 | * for the lifetime of the folio. | |
d69b042f | 2340 | */ |
f70ad448 | 2341 | void folio_memcg_lock(struct folio *folio) |
89c06bd5 KH |
2342 | { |
2343 | struct mem_cgroup *memcg; | |
6de22619 | 2344 | unsigned long flags; |
89c06bd5 | 2345 | |
6de22619 JW |
2346 | /* |
2347 | * The RCU lock is held throughout the transaction. The fast | |
2348 | * path can get away without acquiring the memcg->move_lock | |
2349 | * because page moving starts with an RCU grace period. | |
739f79fc | 2350 | */ |
d7365e78 JW |
2351 | rcu_read_lock(); |
2352 | ||
2353 | if (mem_cgroup_disabled()) | |
1c824a68 | 2354 | return; |
89c06bd5 | 2355 | again: |
f70ad448 | 2356 | memcg = folio_memcg(folio); |
29833315 | 2357 | if (unlikely(!memcg)) |
1c824a68 | 2358 | return; |
d7365e78 | 2359 | |
20ad50d6 AS |
2360 | #ifdef CONFIG_PROVE_LOCKING |
2361 | local_irq_save(flags); | |
2362 | might_lock(&memcg->move_lock); | |
2363 | local_irq_restore(flags); | |
2364 | #endif | |
2365 | ||
bdcbb659 | 2366 | if (atomic_read(&memcg->moving_account) <= 0) |
1c824a68 | 2367 | return; |
89c06bd5 | 2368 | |
6de22619 | 2369 | spin_lock_irqsave(&memcg->move_lock, flags); |
f70ad448 | 2370 | if (memcg != folio_memcg(folio)) { |
6de22619 | 2371 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
2372 | goto again; |
2373 | } | |
6de22619 JW |
2374 | |
2375 | /* | |
1c824a68 JW |
2376 | * When charge migration first begins, we can have multiple |
2377 | * critical sections holding the fast-path RCU lock and one | |
2378 | * holding the slowpath move_lock. Track the task who has the | |
6c77b607 | 2379 | * move_lock for folio_memcg_unlock(). |
6de22619 JW |
2380 | */ |
2381 | memcg->move_lock_task = current; | |
2382 | memcg->move_lock_flags = flags; | |
89c06bd5 | 2383 | } |
f70ad448 | 2384 | |
f70ad448 | 2385 | static void __folio_memcg_unlock(struct mem_cgroup *memcg) |
89c06bd5 | 2386 | { |
6de22619 JW |
2387 | if (memcg && memcg->move_lock_task == current) { |
2388 | unsigned long flags = memcg->move_lock_flags; | |
2389 | ||
2390 | memcg->move_lock_task = NULL; | |
2391 | memcg->move_lock_flags = 0; | |
2392 | ||
2393 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2394 | } | |
89c06bd5 | 2395 | |
d7365e78 | 2396 | rcu_read_unlock(); |
89c06bd5 | 2397 | } |
739f79fc JW |
2398 | |
2399 | /** | |
f70ad448 MWO |
2400 | * folio_memcg_unlock - Release the binding between a folio and its memcg. |
2401 | * @folio: The folio. | |
2402 | * | |
2403 | * This releases the binding created by folio_memcg_lock(). This does | |
2404 | * not change the accounting of this folio to its memcg, but it does | |
2405 | * permit others to change it. | |
739f79fc | 2406 | */ |
f70ad448 | 2407 | void folio_memcg_unlock(struct folio *folio) |
739f79fc | 2408 | { |
f70ad448 MWO |
2409 | __folio_memcg_unlock(folio_memcg(folio)); |
2410 | } | |
9da7b521 | 2411 | |
fead2b86 | 2412 | struct memcg_stock_pcp { |
56751146 | 2413 | local_lock_t stock_lock; |
fead2b86 MH |
2414 | struct mem_cgroup *cached; /* this never be root cgroup */ |
2415 | unsigned int nr_pages; | |
2416 | ||
bf4f0599 RG |
2417 | #ifdef CONFIG_MEMCG_KMEM |
2418 | struct obj_cgroup *cached_objcg; | |
68ac5b3c | 2419 | struct pglist_data *cached_pgdat; |
bf4f0599 | 2420 | unsigned int nr_bytes; |
68ac5b3c WL |
2421 | int nr_slab_reclaimable_b; |
2422 | int nr_slab_unreclaimable_b; | |
bf4f0599 RG |
2423 | #endif |
2424 | ||
cdec2e42 | 2425 | struct work_struct work; |
26fe6168 | 2426 | unsigned long flags; |
a0db00fc | 2427 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 | 2428 | }; |
56751146 SAS |
2429 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock) = { |
2430 | .stock_lock = INIT_LOCAL_LOCK(stock_lock), | |
2431 | }; | |
9f50fad6 | 2432 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2433 | |
bf4f0599 | 2434 | #ifdef CONFIG_MEMCG_KMEM |
56751146 | 2435 | static struct obj_cgroup *drain_obj_stock(struct memcg_stock_pcp *stock); |
bf4f0599 RG |
2436 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, |
2437 | struct mem_cgroup *root_memcg); | |
a8c49af3 | 2438 | static void memcg_account_kmem(struct mem_cgroup *memcg, int nr_pages); |
bf4f0599 RG |
2439 | |
2440 | #else | |
56751146 | 2441 | static inline struct obj_cgroup *drain_obj_stock(struct memcg_stock_pcp *stock) |
bf4f0599 | 2442 | { |
56751146 | 2443 | return NULL; |
bf4f0599 RG |
2444 | } |
2445 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2446 | struct mem_cgroup *root_memcg) | |
2447 | { | |
2448 | return false; | |
2449 | } | |
a8c49af3 YA |
2450 | static void memcg_account_kmem(struct mem_cgroup *memcg, int nr_pages) |
2451 | { | |
2452 | } | |
bf4f0599 RG |
2453 | #endif |
2454 | ||
a0956d54 SS |
2455 | /** |
2456 | * consume_stock: Try to consume stocked charge on this cpu. | |
2457 | * @memcg: memcg to consume from. | |
2458 | * @nr_pages: how many pages to charge. | |
2459 | * | |
2460 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2461 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2462 | * service an allocation will refill the stock. | |
2463 | * | |
2464 | * returns true if successful, false otherwise. | |
cdec2e42 | 2465 | */ |
a0956d54 | 2466 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2467 | { |
2468 | struct memcg_stock_pcp *stock; | |
1872b3bc | 2469 | unsigned int stock_pages; |
db2ba40c | 2470 | unsigned long flags; |
3e32cb2e | 2471 | bool ret = false; |
cdec2e42 | 2472 | |
a983b5eb | 2473 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 2474 | return ret; |
a0956d54 | 2475 | |
56751146 | 2476 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
db2ba40c JW |
2477 | |
2478 | stock = this_cpu_ptr(&memcg_stock); | |
1872b3bc BL |
2479 | stock_pages = READ_ONCE(stock->nr_pages); |
2480 | if (memcg == READ_ONCE(stock->cached) && stock_pages >= nr_pages) { | |
2481 | WRITE_ONCE(stock->nr_pages, stock_pages - nr_pages); | |
3e32cb2e JW |
2482 | ret = true; |
2483 | } | |
db2ba40c | 2484 | |
56751146 | 2485 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
db2ba40c | 2486 | |
cdec2e42 KH |
2487 | return ret; |
2488 | } | |
2489 | ||
2490 | /* | |
3e32cb2e | 2491 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2492 | */ |
2493 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2494 | { | |
1872b3bc | 2495 | unsigned int stock_pages = READ_ONCE(stock->nr_pages); |
f785a8f2 | 2496 | struct mem_cgroup *old = READ_ONCE(stock->cached); |
cdec2e42 | 2497 | |
1a3e1f40 JW |
2498 | if (!old) |
2499 | return; | |
2500 | ||
1872b3bc BL |
2501 | if (stock_pages) { |
2502 | page_counter_uncharge(&old->memory, stock_pages); | |
7941d214 | 2503 | if (do_memsw_account()) |
1872b3bc BL |
2504 | page_counter_uncharge(&old->memsw, stock_pages); |
2505 | ||
2506 | WRITE_ONCE(stock->nr_pages, 0); | |
cdec2e42 | 2507 | } |
1a3e1f40 JW |
2508 | |
2509 | css_put(&old->css); | |
f785a8f2 | 2510 | WRITE_ONCE(stock->cached, NULL); |
cdec2e42 KH |
2511 | } |
2512 | ||
cdec2e42 KH |
2513 | static void drain_local_stock(struct work_struct *dummy) |
2514 | { | |
db2ba40c | 2515 | struct memcg_stock_pcp *stock; |
56751146 | 2516 | struct obj_cgroup *old = NULL; |
db2ba40c JW |
2517 | unsigned long flags; |
2518 | ||
72f0184c | 2519 | /* |
5c49cf9a MH |
2520 | * The only protection from cpu hotplug (memcg_hotplug_cpu_dead) vs. |
2521 | * drain_stock races is that we always operate on local CPU stock | |
2522 | * here with IRQ disabled | |
72f0184c | 2523 | */ |
56751146 | 2524 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
db2ba40c JW |
2525 | |
2526 | stock = this_cpu_ptr(&memcg_stock); | |
56751146 | 2527 | old = drain_obj_stock(stock); |
cdec2e42 | 2528 | drain_stock(stock); |
26fe6168 | 2529 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c | 2530 | |
56751146 | 2531 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
91b71e78 | 2532 | obj_cgroup_put(old); |
cdec2e42 KH |
2533 | } |
2534 | ||
2535 | /* | |
3e32cb2e | 2536 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2537 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2538 | */ |
af9a3b69 | 2539 | static void __refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2540 | { |
db2ba40c | 2541 | struct memcg_stock_pcp *stock; |
1872b3bc | 2542 | unsigned int stock_pages; |
cdec2e42 | 2543 | |
db2ba40c | 2544 | stock = this_cpu_ptr(&memcg_stock); |
f785a8f2 | 2545 | if (READ_ONCE(stock->cached) != memcg) { /* reset if necessary */ |
cdec2e42 | 2546 | drain_stock(stock); |
1a3e1f40 | 2547 | css_get(&memcg->css); |
f785a8f2 | 2548 | WRITE_ONCE(stock->cached, memcg); |
cdec2e42 | 2549 | } |
1872b3bc BL |
2550 | stock_pages = READ_ONCE(stock->nr_pages) + nr_pages; |
2551 | WRITE_ONCE(stock->nr_pages, stock_pages); | |
db2ba40c | 2552 | |
1872b3bc | 2553 | if (stock_pages > MEMCG_CHARGE_BATCH) |
475d0487 | 2554 | drain_stock(stock); |
af9a3b69 JW |
2555 | } |
2556 | ||
2557 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) | |
2558 | { | |
2559 | unsigned long flags; | |
475d0487 | 2560 | |
56751146 | 2561 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
af9a3b69 | 2562 | __refill_stock(memcg, nr_pages); |
56751146 | 2563 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
cdec2e42 KH |
2564 | } |
2565 | ||
2566 | /* | |
c0ff4b85 | 2567 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2568 | * of the hierarchy under it. |
cdec2e42 | 2569 | */ |
6d3d6aa2 | 2570 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2571 | { |
26fe6168 | 2572 | int cpu, curcpu; |
d38144b7 | 2573 | |
6d3d6aa2 JW |
2574 | /* If someone's already draining, avoid adding running more workers. */ |
2575 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2576 | return; | |
72f0184c MH |
2577 | /* |
2578 | * Notify other cpus that system-wide "drain" is running | |
2579 | * We do not care about races with the cpu hotplug because cpu down | |
2580 | * as well as workers from this path always operate on the local | |
2581 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2582 | */ | |
0790ed62 SAS |
2583 | migrate_disable(); |
2584 | curcpu = smp_processor_id(); | |
cdec2e42 KH |
2585 | for_each_online_cpu(cpu) { |
2586 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2587 | struct mem_cgroup *memcg; |
e1a366be | 2588 | bool flush = false; |
26fe6168 | 2589 | |
e1a366be | 2590 | rcu_read_lock(); |
f785a8f2 | 2591 | memcg = READ_ONCE(stock->cached); |
1872b3bc | 2592 | if (memcg && READ_ONCE(stock->nr_pages) && |
e1a366be RG |
2593 | mem_cgroup_is_descendant(memcg, root_memcg)) |
2594 | flush = true; | |
27fb0956 | 2595 | else if (obj_stock_flush_required(stock, root_memcg)) |
bf4f0599 | 2596 | flush = true; |
e1a366be RG |
2597 | rcu_read_unlock(); |
2598 | ||
2599 | if (flush && | |
2600 | !test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { | |
d1a05b69 MH |
2601 | if (cpu == curcpu) |
2602 | drain_local_stock(&stock->work); | |
6a792697 | 2603 | else if (!cpu_is_isolated(cpu)) |
d1a05b69 MH |
2604 | schedule_work_on(cpu, &stock->work); |
2605 | } | |
cdec2e42 | 2606 | } |
0790ed62 | 2607 | migrate_enable(); |
9f50fad6 | 2608 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2609 | } |
2610 | ||
2cd21c89 JW |
2611 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
2612 | { | |
2613 | struct memcg_stock_pcp *stock; | |
a3d4c05a | 2614 | |
2cd21c89 JW |
2615 | stock = &per_cpu(memcg_stock, cpu); |
2616 | drain_stock(stock); | |
a3d4c05a | 2617 | |
308167fc | 2618 | return 0; |
cdec2e42 KH |
2619 | } |
2620 | ||
b3ff9291 CD |
2621 | static unsigned long reclaim_high(struct mem_cgroup *memcg, |
2622 | unsigned int nr_pages, | |
2623 | gfp_t gfp_mask) | |
f7e1cb6e | 2624 | { |
b3ff9291 CD |
2625 | unsigned long nr_reclaimed = 0; |
2626 | ||
f7e1cb6e | 2627 | do { |
e22c6ed9 JW |
2628 | unsigned long pflags; |
2629 | ||
d1663a90 JK |
2630 | if (page_counter_read(&memcg->memory) <= |
2631 | READ_ONCE(memcg->memory.high)) | |
f7e1cb6e | 2632 | continue; |
e22c6ed9 | 2633 | |
e27be240 | 2634 | memcg_memory_event(memcg, MEMCG_HIGH); |
e22c6ed9 JW |
2635 | |
2636 | psi_memstall_enter(&pflags); | |
b3ff9291 | 2637 | nr_reclaimed += try_to_free_mem_cgroup_pages(memcg, nr_pages, |
73b73bac | 2638 | gfp_mask, |
55ab834a | 2639 | MEMCG_RECLAIM_MAY_SWAP); |
e22c6ed9 | 2640 | psi_memstall_leave(&pflags); |
4bf17307 CD |
2641 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2642 | !mem_cgroup_is_root(memcg)); | |
b3ff9291 CD |
2643 | |
2644 | return nr_reclaimed; | |
f7e1cb6e JW |
2645 | } |
2646 | ||
2647 | static void high_work_func(struct work_struct *work) | |
2648 | { | |
2649 | struct mem_cgroup *memcg; | |
2650 | ||
2651 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2652 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2653 | } |
2654 | ||
0e4b01df CD |
2655 | /* |
2656 | * Clamp the maximum sleep time per allocation batch to 2 seconds. This is | |
2657 | * enough to still cause a significant slowdown in most cases, while still | |
2658 | * allowing diagnostics and tracing to proceed without becoming stuck. | |
2659 | */ | |
2660 | #define MEMCG_MAX_HIGH_DELAY_JIFFIES (2UL*HZ) | |
2661 | ||
2662 | /* | |
2663 | * When calculating the delay, we use these either side of the exponentiation to | |
2664 | * maintain precision and scale to a reasonable number of jiffies (see the table | |
2665 | * below. | |
2666 | * | |
2667 | * - MEMCG_DELAY_PRECISION_SHIFT: Extra precision bits while translating the | |
2668 | * overage ratio to a delay. | |
ac5ddd0f | 2669 | * - MEMCG_DELAY_SCALING_SHIFT: The number of bits to scale down the |
0e4b01df CD |
2670 | * proposed penalty in order to reduce to a reasonable number of jiffies, and |
2671 | * to produce a reasonable delay curve. | |
2672 | * | |
2673 | * MEMCG_DELAY_SCALING_SHIFT just happens to be a number that produces a | |
2674 | * reasonable delay curve compared to precision-adjusted overage, not | |
2675 | * penalising heavily at first, but still making sure that growth beyond the | |
2676 | * limit penalises misbehaviour cgroups by slowing them down exponentially. For | |
2677 | * example, with a high of 100 megabytes: | |
2678 | * | |
2679 | * +-------+------------------------+ | |
2680 | * | usage | time to allocate in ms | | |
2681 | * +-------+------------------------+ | |
2682 | * | 100M | 0 | | |
2683 | * | 101M | 6 | | |
2684 | * | 102M | 25 | | |
2685 | * | 103M | 57 | | |
2686 | * | 104M | 102 | | |
2687 | * | 105M | 159 | | |
2688 | * | 106M | 230 | | |
2689 | * | 107M | 313 | | |
2690 | * | 108M | 409 | | |
2691 | * | 109M | 518 | | |
2692 | * | 110M | 639 | | |
2693 | * | 111M | 774 | | |
2694 | * | 112M | 921 | | |
2695 | * | 113M | 1081 | | |
2696 | * | 114M | 1254 | | |
2697 | * | 115M | 1439 | | |
2698 | * | 116M | 1638 | | |
2699 | * | 117M | 1849 | | |
2700 | * | 118M | 2000 | | |
2701 | * | 119M | 2000 | | |
2702 | * | 120M | 2000 | | |
2703 | * +-------+------------------------+ | |
2704 | */ | |
2705 | #define MEMCG_DELAY_PRECISION_SHIFT 20 | |
2706 | #define MEMCG_DELAY_SCALING_SHIFT 14 | |
2707 | ||
8a5dbc65 | 2708 | static u64 calculate_overage(unsigned long usage, unsigned long high) |
b23afb93 | 2709 | { |
8a5dbc65 | 2710 | u64 overage; |
b23afb93 | 2711 | |
8a5dbc65 JK |
2712 | if (usage <= high) |
2713 | return 0; | |
e26733e0 | 2714 | |
8a5dbc65 JK |
2715 | /* |
2716 | * Prevent division by 0 in overage calculation by acting as if | |
2717 | * it was a threshold of 1 page | |
2718 | */ | |
2719 | high = max(high, 1UL); | |
9b8b1754 | 2720 | |
8a5dbc65 JK |
2721 | overage = usage - high; |
2722 | overage <<= MEMCG_DELAY_PRECISION_SHIFT; | |
2723 | return div64_u64(overage, high); | |
2724 | } | |
e26733e0 | 2725 | |
8a5dbc65 JK |
2726 | static u64 mem_find_max_overage(struct mem_cgroup *memcg) |
2727 | { | |
2728 | u64 overage, max_overage = 0; | |
e26733e0 | 2729 | |
8a5dbc65 JK |
2730 | do { |
2731 | overage = calculate_overage(page_counter_read(&memcg->memory), | |
d1663a90 | 2732 | READ_ONCE(memcg->memory.high)); |
8a5dbc65 | 2733 | max_overage = max(overage, max_overage); |
e26733e0 CD |
2734 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2735 | !mem_cgroup_is_root(memcg)); | |
2736 | ||
8a5dbc65 JK |
2737 | return max_overage; |
2738 | } | |
2739 | ||
4b82ab4f JK |
2740 | static u64 swap_find_max_overage(struct mem_cgroup *memcg) |
2741 | { | |
2742 | u64 overage, max_overage = 0; | |
2743 | ||
2744 | do { | |
2745 | overage = calculate_overage(page_counter_read(&memcg->swap), | |
2746 | READ_ONCE(memcg->swap.high)); | |
2747 | if (overage) | |
2748 | memcg_memory_event(memcg, MEMCG_SWAP_HIGH); | |
2749 | max_overage = max(overage, max_overage); | |
2750 | } while ((memcg = parent_mem_cgroup(memcg)) && | |
2751 | !mem_cgroup_is_root(memcg)); | |
2752 | ||
2753 | return max_overage; | |
2754 | } | |
2755 | ||
8a5dbc65 JK |
2756 | /* |
2757 | * Get the number of jiffies that we should penalise a mischievous cgroup which | |
2758 | * is exceeding its memory.high by checking both it and its ancestors. | |
2759 | */ | |
2760 | static unsigned long calculate_high_delay(struct mem_cgroup *memcg, | |
2761 | unsigned int nr_pages, | |
2762 | u64 max_overage) | |
2763 | { | |
2764 | unsigned long penalty_jiffies; | |
2765 | ||
e26733e0 CD |
2766 | if (!max_overage) |
2767 | return 0; | |
0e4b01df CD |
2768 | |
2769 | /* | |
0e4b01df CD |
2770 | * We use overage compared to memory.high to calculate the number of |
2771 | * jiffies to sleep (penalty_jiffies). Ideally this value should be | |
2772 | * fairly lenient on small overages, and increasingly harsh when the | |
2773 | * memcg in question makes it clear that it has no intention of stopping | |
2774 | * its crazy behaviour, so we exponentially increase the delay based on | |
2775 | * overage amount. | |
2776 | */ | |
e26733e0 CD |
2777 | penalty_jiffies = max_overage * max_overage * HZ; |
2778 | penalty_jiffies >>= MEMCG_DELAY_PRECISION_SHIFT; | |
2779 | penalty_jiffies >>= MEMCG_DELAY_SCALING_SHIFT; | |
0e4b01df CD |
2780 | |
2781 | /* | |
2782 | * Factor in the task's own contribution to the overage, such that four | |
2783 | * N-sized allocations are throttled approximately the same as one | |
2784 | * 4N-sized allocation. | |
2785 | * | |
2786 | * MEMCG_CHARGE_BATCH pages is nominal, so work out how much smaller or | |
2787 | * larger the current charge patch is than that. | |
2788 | */ | |
ff144e69 | 2789 | return penalty_jiffies * nr_pages / MEMCG_CHARGE_BATCH; |
e26733e0 CD |
2790 | } |
2791 | ||
2792 | /* | |
63fd3270 JW |
2793 | * Reclaims memory over the high limit. Called directly from |
2794 | * try_charge() (context permitting), as well as from the userland | |
2795 | * return path where reclaim is always able to block. | |
e26733e0 | 2796 | */ |
9ea9cb00 | 2797 | void mem_cgroup_handle_over_high(gfp_t gfp_mask) |
e26733e0 CD |
2798 | { |
2799 | unsigned long penalty_jiffies; | |
2800 | unsigned long pflags; | |
b3ff9291 | 2801 | unsigned long nr_reclaimed; |
e26733e0 | 2802 | unsigned int nr_pages = current->memcg_nr_pages_over_high; |
d977aa93 | 2803 | int nr_retries = MAX_RECLAIM_RETRIES; |
e26733e0 | 2804 | struct mem_cgroup *memcg; |
b3ff9291 | 2805 | bool in_retry = false; |
e26733e0 CD |
2806 | |
2807 | if (likely(!nr_pages)) | |
2808 | return; | |
2809 | ||
2810 | memcg = get_mem_cgroup_from_mm(current->mm); | |
e26733e0 CD |
2811 | current->memcg_nr_pages_over_high = 0; |
2812 | ||
b3ff9291 | 2813 | retry_reclaim: |
63fd3270 JW |
2814 | /* |
2815 | * Bail if the task is already exiting. Unlike memory.max, | |
2816 | * memory.high enforcement isn't as strict, and there is no | |
2817 | * OOM killer involved, which means the excess could already | |
2818 | * be much bigger (and still growing) than it could for | |
2819 | * memory.max; the dying task could get stuck in fruitless | |
2820 | * reclaim for a long time, which isn't desirable. | |
2821 | */ | |
2822 | if (task_is_dying()) | |
2823 | goto out; | |
2824 | ||
b3ff9291 CD |
2825 | /* |
2826 | * The allocating task should reclaim at least the batch size, but for | |
2827 | * subsequent retries we only want to do what's necessary to prevent oom | |
2828 | * or breaching resource isolation. | |
2829 | * | |
2830 | * This is distinct from memory.max or page allocator behaviour because | |
2831 | * memory.high is currently batched, whereas memory.max and the page | |
2832 | * allocator run every time an allocation is made. | |
2833 | */ | |
2834 | nr_reclaimed = reclaim_high(memcg, | |
2835 | in_retry ? SWAP_CLUSTER_MAX : nr_pages, | |
9ea9cb00 | 2836 | gfp_mask); |
b3ff9291 | 2837 | |
e26733e0 CD |
2838 | /* |
2839 | * memory.high is breached and reclaim is unable to keep up. Throttle | |
2840 | * allocators proactively to slow down excessive growth. | |
2841 | */ | |
8a5dbc65 JK |
2842 | penalty_jiffies = calculate_high_delay(memcg, nr_pages, |
2843 | mem_find_max_overage(memcg)); | |
0e4b01df | 2844 | |
4b82ab4f JK |
2845 | penalty_jiffies += calculate_high_delay(memcg, nr_pages, |
2846 | swap_find_max_overage(memcg)); | |
2847 | ||
ff144e69 JK |
2848 | /* |
2849 | * Clamp the max delay per usermode return so as to still keep the | |
2850 | * application moving forwards and also permit diagnostics, albeit | |
2851 | * extremely slowly. | |
2852 | */ | |
2853 | penalty_jiffies = min(penalty_jiffies, MEMCG_MAX_HIGH_DELAY_JIFFIES); | |
2854 | ||
0e4b01df CD |
2855 | /* |
2856 | * Don't sleep if the amount of jiffies this memcg owes us is so low | |
2857 | * that it's not even worth doing, in an attempt to be nice to those who | |
2858 | * go only a small amount over their memory.high value and maybe haven't | |
2859 | * been aggressively reclaimed enough yet. | |
2860 | */ | |
2861 | if (penalty_jiffies <= HZ / 100) | |
2862 | goto out; | |
2863 | ||
b3ff9291 CD |
2864 | /* |
2865 | * If reclaim is making forward progress but we're still over | |
2866 | * memory.high, we want to encourage that rather than doing allocator | |
2867 | * throttling. | |
2868 | */ | |
2869 | if (nr_reclaimed || nr_retries--) { | |
2870 | in_retry = true; | |
2871 | goto retry_reclaim; | |
2872 | } | |
2873 | ||
0e4b01df | 2874 | /* |
63fd3270 JW |
2875 | * Reclaim didn't manage to push usage below the limit, slow |
2876 | * this allocating task down. | |
2877 | * | |
0e4b01df CD |
2878 | * If we exit early, we're guaranteed to die (since |
2879 | * schedule_timeout_killable sets TASK_KILLABLE). This means we don't | |
2880 | * need to account for any ill-begotten jiffies to pay them off later. | |
2881 | */ | |
2882 | psi_memstall_enter(&pflags); | |
2883 | schedule_timeout_killable(penalty_jiffies); | |
2884 | psi_memstall_leave(&pflags); | |
2885 | ||
2886 | out: | |
2887 | css_put(&memcg->css); | |
b23afb93 TH |
2888 | } |
2889 | ||
c5c8b16b MS |
2890 | static int try_charge_memcg(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2891 | unsigned int nr_pages) | |
8a9f3ccd | 2892 | { |
a983b5eb | 2893 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
d977aa93 | 2894 | int nr_retries = MAX_RECLAIM_RETRIES; |
6539cc05 | 2895 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2896 | struct page_counter *counter; |
6539cc05 | 2897 | unsigned long nr_reclaimed; |
a4ebf1b6 | 2898 | bool passed_oom = false; |
73b73bac | 2899 | unsigned int reclaim_options = MEMCG_RECLAIM_MAY_SWAP; |
b70a2a21 | 2900 | bool drained = false; |
d6e103a7 | 2901 | bool raised_max_event = false; |
e22c6ed9 | 2902 | unsigned long pflags; |
a636b327 | 2903 | |
6539cc05 | 2904 | retry: |
b6b6cc72 | 2905 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2906 | return 0; |
8a9f3ccd | 2907 | |
7941d214 | 2908 | if (!do_memsw_account() || |
6071ca52 JW |
2909 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2910 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2911 | goto done_restock; |
7941d214 | 2912 | if (do_memsw_account()) |
3e32cb2e JW |
2913 | page_counter_uncharge(&memcg->memsw, batch); |
2914 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2915 | } else { |
3e32cb2e | 2916 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
73b73bac | 2917 | reclaim_options &= ~MEMCG_RECLAIM_MAY_SWAP; |
3fbe7244 | 2918 | } |
7a81b88c | 2919 | |
6539cc05 JW |
2920 | if (batch > nr_pages) { |
2921 | batch = nr_pages; | |
2922 | goto retry; | |
2923 | } | |
6d61ef40 | 2924 | |
89a28483 JW |
2925 | /* |
2926 | * Prevent unbounded recursion when reclaim operations need to | |
2927 | * allocate memory. This might exceed the limits temporarily, | |
2928 | * but we prefer facilitating memory reclaim and getting back | |
2929 | * under the limit over triggering OOM kills in these cases. | |
2930 | */ | |
2931 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2932 | goto force; | |
2933 | ||
06b078fc JW |
2934 | if (unlikely(task_in_memcg_oom(current))) |
2935 | goto nomem; | |
2936 | ||
d0164adc | 2937 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2938 | goto nomem; |
4b534334 | 2939 | |
e27be240 | 2940 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
d6e103a7 | 2941 | raised_max_event = true; |
241994ed | 2942 | |
e22c6ed9 | 2943 | psi_memstall_enter(&pflags); |
b70a2a21 | 2944 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
55ab834a | 2945 | gfp_mask, reclaim_options); |
e22c6ed9 | 2946 | psi_memstall_leave(&pflags); |
6539cc05 | 2947 | |
61e02c74 | 2948 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2949 | goto retry; |
28c34c29 | 2950 | |
b70a2a21 | 2951 | if (!drained) { |
6d3d6aa2 | 2952 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2953 | drained = true; |
2954 | goto retry; | |
2955 | } | |
2956 | ||
28c34c29 JW |
2957 | if (gfp_mask & __GFP_NORETRY) |
2958 | goto nomem; | |
6539cc05 JW |
2959 | /* |
2960 | * Even though the limit is exceeded at this point, reclaim | |
2961 | * may have been able to free some pages. Retry the charge | |
2962 | * before killing the task. | |
2963 | * | |
2964 | * Only for regular pages, though: huge pages are rather | |
2965 | * unlikely to succeed so close to the limit, and we fall back | |
2966 | * to regular pages anyway in case of failure. | |
2967 | */ | |
61e02c74 | 2968 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2969 | goto retry; |
2970 | /* | |
2971 | * At task move, charge accounts can be doubly counted. So, it's | |
2972 | * better to wait until the end of task_move if something is going on. | |
2973 | */ | |
2974 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2975 | goto retry; | |
2976 | ||
9b130619 JW |
2977 | if (nr_retries--) |
2978 | goto retry; | |
2979 | ||
38d38493 | 2980 | if (gfp_mask & __GFP_RETRY_MAYFAIL) |
29ef680a MH |
2981 | goto nomem; |
2982 | ||
a4ebf1b6 VA |
2983 | /* Avoid endless loop for tasks bypassed by the oom killer */ |
2984 | if (passed_oom && task_is_dying()) | |
2985 | goto nomem; | |
6539cc05 | 2986 | |
29ef680a MH |
2987 | /* |
2988 | * keep retrying as long as the memcg oom killer is able to make | |
2989 | * a forward progress or bypass the charge if the oom killer | |
2990 | * couldn't make any progress. | |
2991 | */ | |
becdf89d SB |
2992 | if (mem_cgroup_oom(mem_over_limit, gfp_mask, |
2993 | get_order(nr_pages * PAGE_SIZE))) { | |
a4ebf1b6 | 2994 | passed_oom = true; |
d977aa93 | 2995 | nr_retries = MAX_RECLAIM_RETRIES; |
29ef680a | 2996 | goto retry; |
29ef680a | 2997 | } |
7a81b88c | 2998 | nomem: |
1461e8c2 SB |
2999 | /* |
3000 | * Memcg doesn't have a dedicated reserve for atomic | |
3001 | * allocations. But like the global atomic pool, we need to | |
3002 | * put the burden of reclaim on regular allocation requests | |
3003 | * and let these go through as privileged allocations. | |
3004 | */ | |
3005 | if (!(gfp_mask & (__GFP_NOFAIL | __GFP_HIGH))) | |
3168ecbe | 3006 | return -ENOMEM; |
10d53c74 | 3007 | force: |
d6e103a7 RG |
3008 | /* |
3009 | * If the allocation has to be enforced, don't forget to raise | |
3010 | * a MEMCG_MAX event. | |
3011 | */ | |
3012 | if (!raised_max_event) | |
3013 | memcg_memory_event(mem_over_limit, MEMCG_MAX); | |
3014 | ||
10d53c74 TH |
3015 | /* |
3016 | * The allocation either can't fail or will lead to more memory | |
3017 | * being freed very soon. Allow memory usage go over the limit | |
3018 | * temporarily by force charging it. | |
3019 | */ | |
3020 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 3021 | if (do_memsw_account()) |
10d53c74 | 3022 | page_counter_charge(&memcg->memsw, nr_pages); |
10d53c74 TH |
3023 | |
3024 | return 0; | |
6539cc05 JW |
3025 | |
3026 | done_restock: | |
3027 | if (batch > nr_pages) | |
3028 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 3029 | |
241994ed | 3030 | /* |
b23afb93 TH |
3031 | * If the hierarchy is above the normal consumption range, schedule |
3032 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 3033 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
3034 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
3035 | * not recorded as it most likely matches current's and won't | |
3036 | * change in the meantime. As high limit is checked again before | |
3037 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
3038 | */ |
3039 | do { | |
4b82ab4f JK |
3040 | bool mem_high, swap_high; |
3041 | ||
3042 | mem_high = page_counter_read(&memcg->memory) > | |
3043 | READ_ONCE(memcg->memory.high); | |
3044 | swap_high = page_counter_read(&memcg->swap) > | |
3045 | READ_ONCE(memcg->swap.high); | |
3046 | ||
3047 | /* Don't bother a random interrupted task */ | |
086f694a | 3048 | if (!in_task()) { |
4b82ab4f | 3049 | if (mem_high) { |
f7e1cb6e JW |
3050 | schedule_work(&memcg->high_work); |
3051 | break; | |
3052 | } | |
4b82ab4f JK |
3053 | continue; |
3054 | } | |
3055 | ||
3056 | if (mem_high || swap_high) { | |
3057 | /* | |
3058 | * The allocating tasks in this cgroup will need to do | |
3059 | * reclaim or be throttled to prevent further growth | |
3060 | * of the memory or swap footprints. | |
3061 | * | |
3062 | * Target some best-effort fairness between the tasks, | |
3063 | * and distribute reclaim work and delay penalties | |
3064 | * based on how much each task is actually allocating. | |
3065 | */ | |
9516a18a | 3066 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
3067 | set_notify_resume(current); |
3068 | break; | |
3069 | } | |
241994ed | 3070 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 | 3071 | |
63fd3270 JW |
3072 | /* |
3073 | * Reclaim is set up above to be called from the userland | |
3074 | * return path. But also attempt synchronous reclaim to avoid | |
3075 | * excessive overrun while the task is still inside the | |
3076 | * kernel. If this is successful, the return path will see it | |
3077 | * when it rechecks the overage and simply bail out. | |
3078 | */ | |
c9afe31e SB |
3079 | if (current->memcg_nr_pages_over_high > MEMCG_CHARGE_BATCH && |
3080 | !(current->flags & PF_MEMALLOC) && | |
63fd3270 | 3081 | gfpflags_allow_blocking(gfp_mask)) |
9ea9cb00 | 3082 | mem_cgroup_handle_over_high(gfp_mask); |
10d53c74 | 3083 | return 0; |
7a81b88c | 3084 | } |
8a9f3ccd | 3085 | |
c5c8b16b MS |
3086 | static inline int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
3087 | unsigned int nr_pages) | |
3088 | { | |
3089 | if (mem_cgroup_is_root(memcg)) | |
3090 | return 0; | |
3091 | ||
3092 | return try_charge_memcg(memcg, gfp_mask, nr_pages); | |
3093 | } | |
3094 | ||
4b569387 NP |
3095 | /** |
3096 | * mem_cgroup_cancel_charge() - cancel an uncommitted try_charge() call. | |
3097 | * @memcg: memcg previously charged. | |
3098 | * @nr_pages: number of pages previously charged. | |
3099 | */ | |
3100 | void mem_cgroup_cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) | |
a3032a2c | 3101 | { |
ce00a967 JW |
3102 | if (mem_cgroup_is_root(memcg)) |
3103 | return; | |
3104 | ||
3e32cb2e | 3105 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 3106 | if (do_memsw_account()) |
3e32cb2e | 3107 | page_counter_uncharge(&memcg->memsw, nr_pages); |
d01dd17f KH |
3108 | } |
3109 | ||
118f2875 | 3110 | static void commit_charge(struct folio *folio, struct mem_cgroup *memcg) |
0a31bc97 | 3111 | { |
118f2875 | 3112 | VM_BUG_ON_FOLIO(folio_memcg(folio), folio); |
0a31bc97 | 3113 | /* |
a5eb011a | 3114 | * Any of the following ensures page's memcg stability: |
0a31bc97 | 3115 | * |
a0b5b414 JW |
3116 | * - the page lock |
3117 | * - LRU isolation | |
6c77b607 | 3118 | * - folio_memcg_lock() |
a0b5b414 | 3119 | * - exclusive reference |
018ee47f | 3120 | * - mem_cgroup_trylock_pages() |
0a31bc97 | 3121 | */ |
118f2875 | 3122 | folio->memcg_data = (unsigned long)memcg; |
7a81b88c | 3123 | } |
66e1707b | 3124 | |
4b569387 NP |
3125 | /** |
3126 | * mem_cgroup_commit_charge - commit a previously successful try_charge(). | |
3127 | * @folio: folio to commit the charge to. | |
3128 | * @memcg: memcg previously charged. | |
3129 | */ | |
3130 | void mem_cgroup_commit_charge(struct folio *folio, struct mem_cgroup *memcg) | |
3131 | { | |
3132 | css_get(&memcg->css); | |
3133 | commit_charge(folio, memcg); | |
3134 | ||
3135 | local_irq_disable(); | |
3136 | mem_cgroup_charge_statistics(memcg, folio_nr_pages(folio)); | |
3137 | memcg_check_events(memcg, folio_nid(folio)); | |
3138 | local_irq_enable(); | |
3139 | } | |
3140 | ||
84c07d11 | 3141 | #ifdef CONFIG_MEMCG_KMEM |
41eb5df1 | 3142 | |
91882c16 SB |
3143 | static inline void __mod_objcg_mlstate(struct obj_cgroup *objcg, |
3144 | struct pglist_data *pgdat, | |
3145 | enum node_stat_item idx, int nr) | |
a7ebf564 WL |
3146 | { |
3147 | struct mem_cgroup *memcg; | |
3148 | struct lruvec *lruvec; | |
3149 | ||
91882c16 SB |
3150 | lockdep_assert_irqs_disabled(); |
3151 | ||
a7ebf564 WL |
3152 | rcu_read_lock(); |
3153 | memcg = obj_cgroup_memcg(objcg); | |
3154 | lruvec = mem_cgroup_lruvec(memcg, pgdat); | |
91882c16 | 3155 | __mod_memcg_lruvec_state(lruvec, idx, nr); |
a7ebf564 WL |
3156 | rcu_read_unlock(); |
3157 | } | |
3158 | ||
fc4db90f RG |
3159 | static __always_inline |
3160 | struct mem_cgroup *mem_cgroup_from_obj_folio(struct folio *folio, void *p) | |
8380ce47 | 3161 | { |
8380ce47 | 3162 | /* |
9855609b RG |
3163 | * Slab objects are accounted individually, not per-page. |
3164 | * Memcg membership data for each individual object is saved in | |
21c690a3 | 3165 | * slab->obj_exts. |
8380ce47 | 3166 | */ |
4b5f8d9a | 3167 | if (folio_test_slab(folio)) { |
21c690a3 | 3168 | struct slabobj_ext *obj_exts; |
4b5f8d9a | 3169 | struct slab *slab; |
9855609b RG |
3170 | unsigned int off; |
3171 | ||
4b5f8d9a | 3172 | slab = folio_slab(folio); |
21c690a3 SB |
3173 | obj_exts = slab_obj_exts(slab); |
3174 | if (!obj_exts) | |
4b5f8d9a VB |
3175 | return NULL; |
3176 | ||
3177 | off = obj_to_index(slab->slab_cache, slab, p); | |
21c690a3 SB |
3178 | if (obj_exts[off].objcg) |
3179 | return obj_cgroup_memcg(obj_exts[off].objcg); | |
10befea9 RG |
3180 | |
3181 | return NULL; | |
9855609b | 3182 | } |
8380ce47 | 3183 | |
bcfe06bf | 3184 | /* |
becacb04 | 3185 | * folio_memcg_check() is used here, because in theory we can encounter |
4b5f8d9a | 3186 | * a folio where the slab flag has been cleared already, but |
21c690a3 | 3187 | * slab->obj_exts has not been freed yet |
becacb04 | 3188 | * folio_memcg_check() will guarantee that a proper memory |
bcfe06bf RG |
3189 | * cgroup pointer or NULL will be returned. |
3190 | */ | |
becacb04 | 3191 | return folio_memcg_check(folio); |
8380ce47 RG |
3192 | } |
3193 | ||
fc4db90f RG |
3194 | /* |
3195 | * Returns a pointer to the memory cgroup to which the kernel object is charged. | |
3196 | * | |
3197 | * A passed kernel object can be a slab object, vmalloc object or a generic | |
3198 | * kernel page, so different mechanisms for getting the memory cgroup pointer | |
3199 | * should be used. | |
3200 | * | |
3201 | * In certain cases (e.g. kernel stacks or large kmallocs with SLUB) the caller | |
3202 | * can not know for sure how the kernel object is implemented. | |
3203 | * mem_cgroup_from_obj() can be safely used in such cases. | |
3204 | * | |
3205 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), | |
3206 | * cgroup_mutex, etc. | |
3207 | */ | |
3208 | struct mem_cgroup *mem_cgroup_from_obj(void *p) | |
3209 | { | |
3210 | struct folio *folio; | |
3211 | ||
3212 | if (mem_cgroup_disabled()) | |
3213 | return NULL; | |
3214 | ||
3215 | if (unlikely(is_vmalloc_addr(p))) | |
3216 | folio = page_folio(vmalloc_to_page(p)); | |
3217 | else | |
3218 | folio = virt_to_folio(p); | |
3219 | ||
3220 | return mem_cgroup_from_obj_folio(folio, p); | |
3221 | } | |
3222 | ||
3223 | /* | |
3224 | * Returns a pointer to the memory cgroup to which the kernel object is charged. | |
3225 | * Similar to mem_cgroup_from_obj(), but faster and not suitable for objects, | |
3226 | * allocated using vmalloc(). | |
3227 | * | |
3228 | * A passed kernel object must be a slab object or a generic kernel page. | |
3229 | * | |
3230 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), | |
3231 | * cgroup_mutex, etc. | |
3232 | */ | |
3233 | struct mem_cgroup *mem_cgroup_from_slab_obj(void *p) | |
3234 | { | |
3235 | if (mem_cgroup_disabled()) | |
3236 | return NULL; | |
3237 | ||
3238 | return mem_cgroup_from_obj_folio(virt_to_folio(p), p); | |
3239 | } | |
3240 | ||
f4840ccf JW |
3241 | static struct obj_cgroup *__get_obj_cgroup_from_memcg(struct mem_cgroup *memcg) |
3242 | { | |
3243 | struct obj_cgroup *objcg = NULL; | |
3244 | ||
7848ed62 | 3245 | for (; !mem_cgroup_is_root(memcg); memcg = parent_mem_cgroup(memcg)) { |
f4840ccf | 3246 | objcg = rcu_dereference(memcg->objcg); |
7d0715d0 | 3247 | if (likely(objcg && obj_cgroup_tryget(objcg))) |
f4840ccf JW |
3248 | break; |
3249 | objcg = NULL; | |
3250 | } | |
3251 | return objcg; | |
3252 | } | |
3253 | ||
1aacbd35 RG |
3254 | static struct obj_cgroup *current_objcg_update(void) |
3255 | { | |
3256 | struct mem_cgroup *memcg; | |
3257 | struct obj_cgroup *old, *objcg = NULL; | |
3258 | ||
3259 | do { | |
3260 | /* Atomically drop the update bit. */ | |
3261 | old = xchg(¤t->objcg, NULL); | |
3262 | if (old) { | |
3263 | old = (struct obj_cgroup *) | |
3264 | ((unsigned long)old & ~CURRENT_OBJCG_UPDATE_FLAG); | |
91b71e78 | 3265 | obj_cgroup_put(old); |
1aacbd35 RG |
3266 | |
3267 | old = NULL; | |
3268 | } | |
3269 | ||
3270 | /* If new objcg is NULL, no reason for the second atomic update. */ | |
3271 | if (!current->mm || (current->flags & PF_KTHREAD)) | |
3272 | return NULL; | |
3273 | ||
3274 | /* | |
3275 | * Release the objcg pointer from the previous iteration, | |
3276 | * if try_cmpxcg() below fails. | |
3277 | */ | |
3278 | if (unlikely(objcg)) { | |
3279 | obj_cgroup_put(objcg); | |
3280 | objcg = NULL; | |
3281 | } | |
3282 | ||
3283 | /* | |
3284 | * Obtain the new objcg pointer. The current task can be | |
3285 | * asynchronously moved to another memcg and the previous | |
3286 | * memcg can be offlined. So let's get the memcg pointer | |
3287 | * and try get a reference to objcg under a rcu read lock. | |
3288 | */ | |
3289 | ||
3290 | rcu_read_lock(); | |
3291 | memcg = mem_cgroup_from_task(current); | |
3292 | objcg = __get_obj_cgroup_from_memcg(memcg); | |
3293 | rcu_read_unlock(); | |
3294 | ||
3295 | /* | |
3296 | * Try set up a new objcg pointer atomically. If it | |
3297 | * fails, it means the update flag was set concurrently, so | |
3298 | * the whole procedure should be repeated. | |
3299 | */ | |
3300 | } while (!try_cmpxchg(¤t->objcg, &old, objcg)); | |
3301 | ||
3302 | return objcg; | |
3303 | } | |
3304 | ||
e86828e5 RG |
3305 | __always_inline struct obj_cgroup *current_obj_cgroup(void) |
3306 | { | |
3307 | struct mem_cgroup *memcg; | |
3308 | struct obj_cgroup *objcg; | |
3309 | ||
3310 | if (in_task()) { | |
3311 | memcg = current->active_memcg; | |
3312 | if (unlikely(memcg)) | |
3313 | goto from_memcg; | |
3314 | ||
3315 | objcg = READ_ONCE(current->objcg); | |
3316 | if (unlikely((unsigned long)objcg & CURRENT_OBJCG_UPDATE_FLAG)) | |
3317 | objcg = current_objcg_update(); | |
3318 | /* | |
3319 | * Objcg reference is kept by the task, so it's safe | |
3320 | * to use the objcg by the current task. | |
3321 | */ | |
3322 | return objcg; | |
3323 | } | |
3324 | ||
3325 | memcg = this_cpu_read(int_active_memcg); | |
3326 | if (unlikely(memcg)) | |
3327 | goto from_memcg; | |
3328 | ||
3329 | return NULL; | |
3330 | ||
3331 | from_memcg: | |
5f79489a | 3332 | objcg = NULL; |
e86828e5 RG |
3333 | for (; !mem_cgroup_is_root(memcg); memcg = parent_mem_cgroup(memcg)) { |
3334 | /* | |
3335 | * Memcg pointer is protected by scope (see set_active_memcg()) | |
3336 | * and is pinning the corresponding objcg, so objcg can't go | |
3337 | * away and can be used within the scope without any additional | |
3338 | * protection. | |
3339 | */ | |
3340 | objcg = rcu_dereference_check(memcg->objcg, 1); | |
3341 | if (likely(objcg)) | |
3342 | break; | |
e86828e5 RG |
3343 | } |
3344 | ||
3345 | return objcg; | |
3346 | } | |
3347 | ||
074e3e26 | 3348 | struct obj_cgroup *get_obj_cgroup_from_folio(struct folio *folio) |
f4840ccf JW |
3349 | { |
3350 | struct obj_cgroup *objcg; | |
3351 | ||
f7a449f7 | 3352 | if (!memcg_kmem_online()) |
f4840ccf JW |
3353 | return NULL; |
3354 | ||
074e3e26 MWO |
3355 | if (folio_memcg_kmem(folio)) { |
3356 | objcg = __folio_objcg(folio); | |
f4840ccf JW |
3357 | obj_cgroup_get(objcg); |
3358 | } else { | |
3359 | struct mem_cgroup *memcg; | |
bf4f0599 | 3360 | |
f4840ccf | 3361 | rcu_read_lock(); |
074e3e26 | 3362 | memcg = __folio_memcg(folio); |
f4840ccf JW |
3363 | if (memcg) |
3364 | objcg = __get_obj_cgroup_from_memcg(memcg); | |
3365 | else | |
3366 | objcg = NULL; | |
3367 | rcu_read_unlock(); | |
3368 | } | |
bf4f0599 RG |
3369 | return objcg; |
3370 | } | |
3371 | ||
a8c49af3 YA |
3372 | static void memcg_account_kmem(struct mem_cgroup *memcg, int nr_pages) |
3373 | { | |
3374 | mod_memcg_state(memcg, MEMCG_KMEM, nr_pages); | |
3375 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { | |
3376 | if (nr_pages > 0) | |
3377 | page_counter_charge(&memcg->kmem, nr_pages); | |
3378 | else | |
3379 | page_counter_uncharge(&memcg->kmem, -nr_pages); | |
3380 | } | |
3381 | } | |
3382 | ||
3383 | ||
f1286fae MS |
3384 | /* |
3385 | * obj_cgroup_uncharge_pages: uncharge a number of kernel pages from a objcg | |
3386 | * @objcg: object cgroup to uncharge | |
3387 | * @nr_pages: number of pages to uncharge | |
3388 | */ | |
e74d2259 MS |
3389 | static void obj_cgroup_uncharge_pages(struct obj_cgroup *objcg, |
3390 | unsigned int nr_pages) | |
3391 | { | |
3392 | struct mem_cgroup *memcg; | |
3393 | ||
3394 | memcg = get_mem_cgroup_from_objcg(objcg); | |
e74d2259 | 3395 | |
a8c49af3 | 3396 | memcg_account_kmem(memcg, -nr_pages); |
f1286fae | 3397 | refill_stock(memcg, nr_pages); |
e74d2259 | 3398 | |
e74d2259 | 3399 | css_put(&memcg->css); |
e74d2259 MS |
3400 | } |
3401 | ||
f1286fae MS |
3402 | /* |
3403 | * obj_cgroup_charge_pages: charge a number of kernel pages to a objcg | |
3404 | * @objcg: object cgroup to charge | |
45264778 | 3405 | * @gfp: reclaim mode |
92d0510c | 3406 | * @nr_pages: number of pages to charge |
45264778 VD |
3407 | * |
3408 | * Returns 0 on success, an error code on failure. | |
3409 | */ | |
f1286fae MS |
3410 | static int obj_cgroup_charge_pages(struct obj_cgroup *objcg, gfp_t gfp, |
3411 | unsigned int nr_pages) | |
7ae1e1d0 | 3412 | { |
f1286fae | 3413 | struct mem_cgroup *memcg; |
7ae1e1d0 GC |
3414 | int ret; |
3415 | ||
f1286fae MS |
3416 | memcg = get_mem_cgroup_from_objcg(objcg); |
3417 | ||
c5c8b16b | 3418 | ret = try_charge_memcg(memcg, gfp, nr_pages); |
52c29b04 | 3419 | if (ret) |
f1286fae | 3420 | goto out; |
52c29b04 | 3421 | |
a8c49af3 | 3422 | memcg_account_kmem(memcg, nr_pages); |
f1286fae MS |
3423 | out: |
3424 | css_put(&memcg->css); | |
4b13f64d | 3425 | |
f1286fae | 3426 | return ret; |
4b13f64d RG |
3427 | } |
3428 | ||
45264778 | 3429 | /** |
f4b00eab | 3430 | * __memcg_kmem_charge_page: charge a kmem page to the current memory cgroup |
45264778 VD |
3431 | * @page: page to charge |
3432 | * @gfp: reclaim mode | |
3433 | * @order: allocation order | |
3434 | * | |
3435 | * Returns 0 on success, an error code on failure. | |
3436 | */ | |
f4b00eab | 3437 | int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 3438 | { |
b4e0b68f | 3439 | struct obj_cgroup *objcg; |
fcff7d7e | 3440 | int ret = 0; |
7ae1e1d0 | 3441 | |
e86828e5 | 3442 | objcg = current_obj_cgroup(); |
b4e0b68f MS |
3443 | if (objcg) { |
3444 | ret = obj_cgroup_charge_pages(objcg, gfp, 1 << order); | |
4d96ba35 | 3445 | if (!ret) { |
e86828e5 | 3446 | obj_cgroup_get(objcg); |
b4e0b68f | 3447 | page->memcg_data = (unsigned long)objcg | |
18b2db3b | 3448 | MEMCG_DATA_KMEM; |
1a3e1f40 | 3449 | return 0; |
4d96ba35 | 3450 | } |
c4159a75 | 3451 | } |
d05e83a6 | 3452 | return ret; |
7ae1e1d0 | 3453 | } |
49a18eae | 3454 | |
45264778 | 3455 | /** |
f4b00eab | 3456 | * __memcg_kmem_uncharge_page: uncharge a kmem page |
45264778 VD |
3457 | * @page: page to uncharge |
3458 | * @order: allocation order | |
3459 | */ | |
f4b00eab | 3460 | void __memcg_kmem_uncharge_page(struct page *page, int order) |
7ae1e1d0 | 3461 | { |
1b7e4464 | 3462 | struct folio *folio = page_folio(page); |
b4e0b68f | 3463 | struct obj_cgroup *objcg; |
f3ccb2c4 | 3464 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 3465 | |
1b7e4464 | 3466 | if (!folio_memcg_kmem(folio)) |
7ae1e1d0 GC |
3467 | return; |
3468 | ||
1b7e4464 | 3469 | objcg = __folio_objcg(folio); |
b4e0b68f | 3470 | obj_cgroup_uncharge_pages(objcg, nr_pages); |
1b7e4464 | 3471 | folio->memcg_data = 0; |
b4e0b68f | 3472 | obj_cgroup_put(objcg); |
60d3fd32 | 3473 | } |
bf4f0599 | 3474 | |
91882c16 | 3475 | static void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat, |
68ac5b3c WL |
3476 | enum node_stat_item idx, int nr) |
3477 | { | |
fead2b86 | 3478 | struct memcg_stock_pcp *stock; |
56751146 | 3479 | struct obj_cgroup *old = NULL; |
68ac5b3c WL |
3480 | unsigned long flags; |
3481 | int *bytes; | |
3482 | ||
56751146 | 3483 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
fead2b86 MH |
3484 | stock = this_cpu_ptr(&memcg_stock); |
3485 | ||
68ac5b3c WL |
3486 | /* |
3487 | * Save vmstat data in stock and skip vmstat array update unless | |
3488 | * accumulating over a page of vmstat data or when pgdat or idx | |
3489 | * changes. | |
3490 | */ | |
3b8abb32 | 3491 | if (READ_ONCE(stock->cached_objcg) != objcg) { |
56751146 | 3492 | old = drain_obj_stock(stock); |
68ac5b3c WL |
3493 | obj_cgroup_get(objcg); |
3494 | stock->nr_bytes = atomic_read(&objcg->nr_charged_bytes) | |
3495 | ? atomic_xchg(&objcg->nr_charged_bytes, 0) : 0; | |
3b8abb32 | 3496 | WRITE_ONCE(stock->cached_objcg, objcg); |
68ac5b3c WL |
3497 | stock->cached_pgdat = pgdat; |
3498 | } else if (stock->cached_pgdat != pgdat) { | |
3499 | /* Flush the existing cached vmstat data */ | |
7fa0dacb WL |
3500 | struct pglist_data *oldpg = stock->cached_pgdat; |
3501 | ||
68ac5b3c | 3502 | if (stock->nr_slab_reclaimable_b) { |
91882c16 | 3503 | __mod_objcg_mlstate(objcg, oldpg, NR_SLAB_RECLAIMABLE_B, |
68ac5b3c WL |
3504 | stock->nr_slab_reclaimable_b); |
3505 | stock->nr_slab_reclaimable_b = 0; | |
3506 | } | |
3507 | if (stock->nr_slab_unreclaimable_b) { | |
91882c16 | 3508 | __mod_objcg_mlstate(objcg, oldpg, NR_SLAB_UNRECLAIMABLE_B, |
68ac5b3c WL |
3509 | stock->nr_slab_unreclaimable_b); |
3510 | stock->nr_slab_unreclaimable_b = 0; | |
3511 | } | |
3512 | stock->cached_pgdat = pgdat; | |
3513 | } | |
3514 | ||
3515 | bytes = (idx == NR_SLAB_RECLAIMABLE_B) ? &stock->nr_slab_reclaimable_b | |
3516 | : &stock->nr_slab_unreclaimable_b; | |
3517 | /* | |
3518 | * Even for large object >= PAGE_SIZE, the vmstat data will still be | |
3519 | * cached locally at least once before pushing it out. | |
3520 | */ | |
3521 | if (!*bytes) { | |
3522 | *bytes = nr; | |
3523 | nr = 0; | |
3524 | } else { | |
3525 | *bytes += nr; | |
3526 | if (abs(*bytes) > PAGE_SIZE) { | |
3527 | nr = *bytes; | |
3528 | *bytes = 0; | |
3529 | } else { | |
3530 | nr = 0; | |
3531 | } | |
3532 | } | |
3533 | if (nr) | |
91882c16 | 3534 | __mod_objcg_mlstate(objcg, pgdat, idx, nr); |
68ac5b3c | 3535 | |
56751146 | 3536 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
91b71e78 | 3537 | obj_cgroup_put(old); |
68ac5b3c WL |
3538 | } |
3539 | ||
bf4f0599 RG |
3540 | static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) |
3541 | { | |
fead2b86 | 3542 | struct memcg_stock_pcp *stock; |
bf4f0599 RG |
3543 | unsigned long flags; |
3544 | bool ret = false; | |
3545 | ||
56751146 | 3546 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
fead2b86 MH |
3547 | |
3548 | stock = this_cpu_ptr(&memcg_stock); | |
3b8abb32 | 3549 | if (objcg == READ_ONCE(stock->cached_objcg) && stock->nr_bytes >= nr_bytes) { |
bf4f0599 RG |
3550 | stock->nr_bytes -= nr_bytes; |
3551 | ret = true; | |
3552 | } | |
3553 | ||
56751146 | 3554 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
bf4f0599 RG |
3555 | |
3556 | return ret; | |
3557 | } | |
3558 | ||
56751146 | 3559 | static struct obj_cgroup *drain_obj_stock(struct memcg_stock_pcp *stock) |
bf4f0599 | 3560 | { |
3b8abb32 | 3561 | struct obj_cgroup *old = READ_ONCE(stock->cached_objcg); |
bf4f0599 RG |
3562 | |
3563 | if (!old) | |
56751146 | 3564 | return NULL; |
bf4f0599 RG |
3565 | |
3566 | if (stock->nr_bytes) { | |
3567 | unsigned int nr_pages = stock->nr_bytes >> PAGE_SHIFT; | |
3568 | unsigned int nr_bytes = stock->nr_bytes & (PAGE_SIZE - 1); | |
3569 | ||
af9a3b69 JW |
3570 | if (nr_pages) { |
3571 | struct mem_cgroup *memcg; | |
3572 | ||
3573 | memcg = get_mem_cgroup_from_objcg(old); | |
3574 | ||
3575 | memcg_account_kmem(memcg, -nr_pages); | |
3576 | __refill_stock(memcg, nr_pages); | |
3577 | ||
3578 | css_put(&memcg->css); | |
3579 | } | |
bf4f0599 RG |
3580 | |
3581 | /* | |
3582 | * The leftover is flushed to the centralized per-memcg value. | |
3583 | * On the next attempt to refill obj stock it will be moved | |
3584 | * to a per-cpu stock (probably, on an other CPU), see | |
3585 | * refill_obj_stock(). | |
3586 | * | |
3587 | * How often it's flushed is a trade-off between the memory | |
3588 | * limit enforcement accuracy and potential CPU contention, | |
3589 | * so it might be changed in the future. | |
3590 | */ | |
3591 | atomic_add(nr_bytes, &old->nr_charged_bytes); | |
3592 | stock->nr_bytes = 0; | |
3593 | } | |
3594 | ||
68ac5b3c WL |
3595 | /* |
3596 | * Flush the vmstat data in current stock | |
3597 | */ | |
3598 | if (stock->nr_slab_reclaimable_b || stock->nr_slab_unreclaimable_b) { | |
3599 | if (stock->nr_slab_reclaimable_b) { | |
91882c16 | 3600 | __mod_objcg_mlstate(old, stock->cached_pgdat, |
68ac5b3c WL |
3601 | NR_SLAB_RECLAIMABLE_B, |
3602 | stock->nr_slab_reclaimable_b); | |
3603 | stock->nr_slab_reclaimable_b = 0; | |
3604 | } | |
3605 | if (stock->nr_slab_unreclaimable_b) { | |
91882c16 | 3606 | __mod_objcg_mlstate(old, stock->cached_pgdat, |
68ac5b3c WL |
3607 | NR_SLAB_UNRECLAIMABLE_B, |
3608 | stock->nr_slab_unreclaimable_b); | |
3609 | stock->nr_slab_unreclaimable_b = 0; | |
3610 | } | |
3611 | stock->cached_pgdat = NULL; | |
3612 | } | |
3613 | ||
3b8abb32 | 3614 | WRITE_ONCE(stock->cached_objcg, NULL); |
56751146 SAS |
3615 | /* |
3616 | * The `old' objects needs to be released by the caller via | |
3617 | * obj_cgroup_put() outside of memcg_stock_pcp::stock_lock. | |
3618 | */ | |
3619 | return old; | |
bf4f0599 RG |
3620 | } |
3621 | ||
3622 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
3623 | struct mem_cgroup *root_memcg) | |
3624 | { | |
3b8abb32 | 3625 | struct obj_cgroup *objcg = READ_ONCE(stock->cached_objcg); |
bf4f0599 RG |
3626 | struct mem_cgroup *memcg; |
3627 | ||
3b8abb32 RG |
3628 | if (objcg) { |
3629 | memcg = obj_cgroup_memcg(objcg); | |
bf4f0599 RG |
3630 | if (memcg && mem_cgroup_is_descendant(memcg, root_memcg)) |
3631 | return true; | |
3632 | } | |
3633 | ||
3634 | return false; | |
3635 | } | |
3636 | ||
5387c904 WL |
3637 | static void refill_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes, |
3638 | bool allow_uncharge) | |
bf4f0599 | 3639 | { |
fead2b86 | 3640 | struct memcg_stock_pcp *stock; |
56751146 | 3641 | struct obj_cgroup *old = NULL; |
bf4f0599 | 3642 | unsigned long flags; |
5387c904 | 3643 | unsigned int nr_pages = 0; |
bf4f0599 | 3644 | |
56751146 | 3645 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
fead2b86 MH |
3646 | |
3647 | stock = this_cpu_ptr(&memcg_stock); | |
3b8abb32 | 3648 | if (READ_ONCE(stock->cached_objcg) != objcg) { /* reset if necessary */ |
56751146 | 3649 | old = drain_obj_stock(stock); |
bf4f0599 | 3650 | obj_cgroup_get(objcg); |
3b8abb32 | 3651 | WRITE_ONCE(stock->cached_objcg, objcg); |
5387c904 WL |
3652 | stock->nr_bytes = atomic_read(&objcg->nr_charged_bytes) |
3653 | ? atomic_xchg(&objcg->nr_charged_bytes, 0) : 0; | |
3654 | allow_uncharge = true; /* Allow uncharge when objcg changes */ | |
bf4f0599 RG |
3655 | } |
3656 | stock->nr_bytes += nr_bytes; | |
3657 | ||
5387c904 WL |
3658 | if (allow_uncharge && (stock->nr_bytes > PAGE_SIZE)) { |
3659 | nr_pages = stock->nr_bytes >> PAGE_SHIFT; | |
3660 | stock->nr_bytes &= (PAGE_SIZE - 1); | |
3661 | } | |
bf4f0599 | 3662 | |
56751146 | 3663 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
91b71e78 | 3664 | obj_cgroup_put(old); |
5387c904 WL |
3665 | |
3666 | if (nr_pages) | |
3667 | obj_cgroup_uncharge_pages(objcg, nr_pages); | |
bf4f0599 RG |
3668 | } |
3669 | ||
3670 | int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size) | |
3671 | { | |
bf4f0599 RG |
3672 | unsigned int nr_pages, nr_bytes; |
3673 | int ret; | |
3674 | ||
3675 | if (consume_obj_stock(objcg, size)) | |
3676 | return 0; | |
3677 | ||
3678 | /* | |
5387c904 | 3679 | * In theory, objcg->nr_charged_bytes can have enough |
bf4f0599 | 3680 | * pre-charged bytes to satisfy the allocation. However, |
5387c904 WL |
3681 | * flushing objcg->nr_charged_bytes requires two atomic |
3682 | * operations, and objcg->nr_charged_bytes can't be big. | |
3683 | * The shared objcg->nr_charged_bytes can also become a | |
3684 | * performance bottleneck if all tasks of the same memcg are | |
3685 | * trying to update it. So it's better to ignore it and try | |
3686 | * grab some new pages. The stock's nr_bytes will be flushed to | |
3687 | * objcg->nr_charged_bytes later on when objcg changes. | |
3688 | * | |
3689 | * The stock's nr_bytes may contain enough pre-charged bytes | |
3690 | * to allow one less page from being charged, but we can't rely | |
3691 | * on the pre-charged bytes not being changed outside of | |
3692 | * consume_obj_stock() or refill_obj_stock(). So ignore those | |
3693 | * pre-charged bytes as well when charging pages. To avoid a | |
3694 | * page uncharge right after a page charge, we set the | |
3695 | * allow_uncharge flag to false when calling refill_obj_stock() | |
3696 | * to temporarily allow the pre-charged bytes to exceed the page | |
3697 | * size limit. The maximum reachable value of the pre-charged | |
3698 | * bytes is (sizeof(object) + PAGE_SIZE - 2) if there is no data | |
3699 | * race. | |
bf4f0599 | 3700 | */ |
bf4f0599 RG |
3701 | nr_pages = size >> PAGE_SHIFT; |
3702 | nr_bytes = size & (PAGE_SIZE - 1); | |
3703 | ||
3704 | if (nr_bytes) | |
3705 | nr_pages += 1; | |
3706 | ||
e74d2259 | 3707 | ret = obj_cgroup_charge_pages(objcg, gfp, nr_pages); |
bf4f0599 | 3708 | if (!ret && nr_bytes) |
5387c904 | 3709 | refill_obj_stock(objcg, PAGE_SIZE - nr_bytes, false); |
bf4f0599 | 3710 | |
bf4f0599 RG |
3711 | return ret; |
3712 | } | |
3713 | ||
3714 | void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size) | |
3715 | { | |
5387c904 | 3716 | refill_obj_stock(objcg, size, true); |
bf4f0599 RG |
3717 | } |
3718 | ||
e6100a45 VB |
3719 | static inline size_t obj_full_size(struct kmem_cache *s) |
3720 | { | |
3721 | /* | |
3722 | * For each accounted object there is an extra space which is used | |
3723 | * to store obj_cgroup membership. Charge it too. | |
3724 | */ | |
3725 | return s->size + sizeof(struct obj_cgroup *); | |
3726 | } | |
3727 | ||
3728 | bool __memcg_slab_post_alloc_hook(struct kmem_cache *s, struct list_lru *lru, | |
3729 | gfp_t flags, size_t size, void **p) | |
3730 | { | |
3731 | struct obj_cgroup *objcg; | |
3732 | struct slab *slab; | |
3733 | unsigned long off; | |
3734 | size_t i; | |
3735 | ||
3736 | /* | |
3737 | * The obtained objcg pointer is safe to use within the current scope, | |
3738 | * defined by current task or set_active_memcg() pair. | |
3739 | * obj_cgroup_get() is used to get a permanent reference. | |
3740 | */ | |
3741 | objcg = current_obj_cgroup(); | |
3742 | if (!objcg) | |
3743 | return true; | |
3744 | ||
3745 | /* | |
3746 | * slab_alloc_node() avoids the NULL check, so we might be called with a | |
3747 | * single NULL object. kmem_cache_alloc_bulk() aborts if it can't fill | |
3748 | * the whole requested size. | |
3749 | * return success as there's nothing to free back | |
3750 | */ | |
3751 | if (unlikely(*p == NULL)) | |
3752 | return true; | |
3753 | ||
3754 | flags &= gfp_allowed_mask; | |
3755 | ||
3756 | if (lru) { | |
3757 | int ret; | |
3758 | struct mem_cgroup *memcg; | |
3759 | ||
3760 | memcg = get_mem_cgroup_from_objcg(objcg); | |
3761 | ret = memcg_list_lru_alloc(memcg, lru, flags); | |
3762 | css_put(&memcg->css); | |
3763 | ||
3764 | if (ret) | |
3765 | return false; | |
3766 | } | |
3767 | ||
3768 | if (obj_cgroup_charge(objcg, flags, size * obj_full_size(s))) | |
3769 | return false; | |
3770 | ||
3771 | for (i = 0; i < size; i++) { | |
3772 | slab = virt_to_slab(p[i]); | |
3773 | ||
3774 | if (!slab_obj_exts(slab) && | |
3775 | alloc_slab_obj_exts(slab, s, flags, false)) { | |
3776 | obj_cgroup_uncharge(objcg, obj_full_size(s)); | |
3777 | continue; | |
3778 | } | |
3779 | ||
3780 | off = obj_to_index(s, slab, p[i]); | |
3781 | obj_cgroup_get(objcg); | |
3782 | slab_obj_exts(slab)[off].objcg = objcg; | |
3783 | mod_objcg_state(objcg, slab_pgdat(slab), | |
3784 | cache_vmstat_idx(s), obj_full_size(s)); | |
3785 | } | |
3786 | ||
3787 | return true; | |
3788 | } | |
3789 | ||
3790 | void __memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab, | |
3791 | void **p, int objects, struct slabobj_ext *obj_exts) | |
3792 | { | |
3793 | for (int i = 0; i < objects; i++) { | |
3794 | struct obj_cgroup *objcg; | |
3795 | unsigned int off; | |
3796 | ||
3797 | off = obj_to_index(s, slab, p[i]); | |
3798 | objcg = obj_exts[off].objcg; | |
3799 | if (!objcg) | |
3800 | continue; | |
3801 | ||
3802 | obj_exts[off].objcg = NULL; | |
3803 | obj_cgroup_uncharge(objcg, obj_full_size(s)); | |
3804 | mod_objcg_state(objcg, slab_pgdat(slab), cache_vmstat_idx(s), | |
3805 | -obj_full_size(s)); | |
3806 | obj_cgroup_put(objcg); | |
3807 | } | |
3808 | } | |
84c07d11 | 3809 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 3810 | |
ca3e0214 | 3811 | /* |
be6c8982 | 3812 | * Because page_memcg(head) is not set on tails, set it now. |
ca3e0214 | 3813 | */ |
b8791381 | 3814 | void split_page_memcg(struct page *head, int old_order, int new_order) |
ca3e0214 | 3815 | { |
1b7e4464 MWO |
3816 | struct folio *folio = page_folio(head); |
3817 | struct mem_cgroup *memcg = folio_memcg(folio); | |
e94c8a9c | 3818 | int i; |
b8791381 ZY |
3819 | unsigned int old_nr = 1 << old_order; |
3820 | unsigned int new_nr = 1 << new_order; | |
ca3e0214 | 3821 | |
be6c8982 | 3822 | if (mem_cgroup_disabled() || !memcg) |
3d37c4a9 | 3823 | return; |
b070e65c | 3824 | |
b8791381 | 3825 | for (i = new_nr; i < old_nr; i += new_nr) |
1b7e4464 | 3826 | folio_page(folio, i)->memcg_data = folio->memcg_data; |
b4e0b68f | 3827 | |
1b7e4464 | 3828 | if (folio_memcg_kmem(folio)) |
b8791381 | 3829 | obj_cgroup_get_many(__folio_objcg(folio), old_nr / new_nr - 1); |
b4e0b68f | 3830 | else |
b8791381 | 3831 | css_get_many(&memcg->css, old_nr / new_nr - 1); |
ca3e0214 | 3832 | } |
ca3e0214 | 3833 | |
e55b9f96 | 3834 | #ifdef CONFIG_SWAP |
02491447 DN |
3835 | /** |
3836 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3837 | * @entry: swap entry to be moved | |
3838 | * @from: mem_cgroup which the entry is moved from | |
3839 | * @to: mem_cgroup which the entry is moved to | |
3840 | * | |
3841 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3842 | * as the mem_cgroup's id of @from. | |
3843 | * | |
3844 | * Returns 0 on success, -EINVAL on failure. | |
3845 | * | |
3e32cb2e | 3846 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
3847 | * both res and memsw, and called css_get(). |
3848 | */ | |
3849 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3850 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3851 | { |
3852 | unsigned short old_id, new_id; | |
3853 | ||
34c00c31 LZ |
3854 | old_id = mem_cgroup_id(from); |
3855 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3856 | |
3857 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
3858 | mod_memcg_state(from, MEMCG_SWAP, -1); |
3859 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
3860 | return 0; |
3861 | } | |
3862 | return -EINVAL; | |
3863 | } | |
3864 | #else | |
3865 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3866 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3867 | { |
3868 | return -EINVAL; | |
3869 | } | |
8c7c6e34 | 3870 | #endif |
d13d1443 | 3871 | |
bbec2e15 | 3872 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 3873 | |
bbec2e15 RG |
3874 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
3875 | unsigned long max, bool memsw) | |
628f4235 | 3876 | { |
3e32cb2e | 3877 | bool enlarge = false; |
bb4a7ea2 | 3878 | bool drained = false; |
3e32cb2e | 3879 | int ret; |
c054a78c YZ |
3880 | bool limits_invariant; |
3881 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 3882 | |
3e32cb2e | 3883 | do { |
628f4235 KH |
3884 | if (signal_pending(current)) { |
3885 | ret = -EINTR; | |
3886 | break; | |
3887 | } | |
3e32cb2e | 3888 | |
bbec2e15 | 3889 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
3890 | /* |
3891 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 3892 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 3893 | */ |
15b42562 | 3894 | limits_invariant = memsw ? max >= READ_ONCE(memcg->memory.max) : |
bbec2e15 | 3895 | max <= memcg->memsw.max; |
c054a78c | 3896 | if (!limits_invariant) { |
bbec2e15 | 3897 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 3898 | ret = -EINVAL; |
8c7c6e34 KH |
3899 | break; |
3900 | } | |
bbec2e15 | 3901 | if (max > counter->max) |
3e32cb2e | 3902 | enlarge = true; |
bbec2e15 RG |
3903 | ret = page_counter_set_max(counter, max); |
3904 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
3905 | |
3906 | if (!ret) | |
3907 | break; | |
3908 | ||
bb4a7ea2 SB |
3909 | if (!drained) { |
3910 | drain_all_stock(memcg); | |
3911 | drained = true; | |
3912 | continue; | |
3913 | } | |
3914 | ||
73b73bac | 3915 | if (!try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, |
55ab834a | 3916 | memsw ? 0 : MEMCG_RECLAIM_MAY_SWAP)) { |
1ab5c056 AR |
3917 | ret = -EBUSY; |
3918 | break; | |
3919 | } | |
3920 | } while (true); | |
3e32cb2e | 3921 | |
3c11ecf4 KH |
3922 | if (!ret && enlarge) |
3923 | memcg_oom_recover(memcg); | |
3e32cb2e | 3924 | |
628f4235 KH |
3925 | return ret; |
3926 | } | |
3927 | ||
ef8f2327 | 3928 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
3929 | gfp_t gfp_mask, |
3930 | unsigned long *total_scanned) | |
3931 | { | |
3932 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 3933 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
3934 | unsigned long reclaimed; |
3935 | int loop = 0; | |
ef8f2327 | 3936 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 3937 | unsigned long excess; |
0608f43d | 3938 | |
e4dde56c YZ |
3939 | if (lru_gen_enabled()) |
3940 | return 0; | |
3941 | ||
0608f43d AM |
3942 | if (order > 0) |
3943 | return 0; | |
3944 | ||
2ab082ba | 3945 | mctz = soft_limit_tree.rb_tree_per_node[pgdat->node_id]; |
d6507ff5 MH |
3946 | |
3947 | /* | |
3948 | * Do not even bother to check the largest node if the root | |
3949 | * is empty. Do it lockless to prevent lock bouncing. Races | |
3950 | * are acceptable as soft limit is best effort anyway. | |
3951 | */ | |
bfc7228b | 3952 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
3953 | return 0; |
3954 | ||
0608f43d AM |
3955 | /* |
3956 | * This loop can run a while, specially if mem_cgroup's continuously | |
3957 | * keep exceeding their soft limit and putting the system under | |
3958 | * pressure | |
3959 | */ | |
3960 | do { | |
3961 | if (next_mz) | |
3962 | mz = next_mz; | |
3963 | else | |
3964 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3965 | if (!mz) | |
3966 | break; | |
3967 | ||
ef8f2327 | 3968 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
d8f65338 | 3969 | gfp_mask, total_scanned); |
0608f43d | 3970 | nr_reclaimed += reclaimed; |
0a31bc97 | 3971 | spin_lock_irq(&mctz->lock); |
0608f43d AM |
3972 | |
3973 | /* | |
3974 | * If we failed to reclaim anything from this memory cgroup | |
3975 | * it is time to move on to the next cgroup | |
3976 | */ | |
3977 | next_mz = NULL; | |
bc2f2e7f VD |
3978 | if (!reclaimed) |
3979 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3980 | ||
3e32cb2e | 3981 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3982 | /* |
3983 | * One school of thought says that we should not add | |
3984 | * back the node to the tree if reclaim returns 0. | |
3985 | * But our reclaim could return 0, simply because due | |
3986 | * to priority we are exposing a smaller subset of | |
3987 | * memory to reclaim from. Consider this as a longer | |
3988 | * term TODO. | |
3989 | */ | |
3990 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3991 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3992 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3993 | css_put(&mz->memcg->css); |
3994 | loop++; | |
3995 | /* | |
3996 | * Could not reclaim anything and there are no more | |
3997 | * mem cgroups to try or we seem to be looping without | |
3998 | * reclaiming anything. | |
3999 | */ | |
4000 | if (!nr_reclaimed && | |
4001 | (next_mz == NULL || | |
4002 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
4003 | break; | |
4004 | } while (!nr_reclaimed); | |
4005 | if (next_mz) | |
4006 | css_put(&next_mz->memcg->css); | |
4007 | return nr_reclaimed; | |
4008 | } | |
4009 | ||
c26251f9 | 4010 | /* |
51038171 | 4011 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
4012 | * |
4013 | * Caller is responsible for holding css reference for memcg. | |
4014 | */ | |
4015 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
4016 | { | |
d977aa93 | 4017 | int nr_retries = MAX_RECLAIM_RETRIES; |
c26251f9 | 4018 | |
c1e862c1 KH |
4019 | /* we call try-to-free pages for make this cgroup empty */ |
4020 | lru_add_drain_all(); | |
d12c60f6 JS |
4021 | |
4022 | drain_all_stock(memcg); | |
4023 | ||
f817ed48 | 4024 | /* try to free all pages in this cgroup */ |
3e32cb2e | 4025 | while (nr_retries && page_counter_read(&memcg->memory)) { |
c26251f9 MH |
4026 | if (signal_pending(current)) |
4027 | return -EINTR; | |
4028 | ||
73b73bac | 4029 | if (!try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, |
55ab834a | 4030 | MEMCG_RECLAIM_MAY_SWAP)) |
f817ed48 | 4031 | nr_retries--; |
f817ed48 | 4032 | } |
ab5196c2 MH |
4033 | |
4034 | return 0; | |
cc847582 KH |
4035 | } |
4036 | ||
6770c64e TH |
4037 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
4038 | char *buf, size_t nbytes, | |
4039 | loff_t off) | |
c1e862c1 | 4040 | { |
6770c64e | 4041 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 4042 | |
d8423011 MH |
4043 | if (mem_cgroup_is_root(memcg)) |
4044 | return -EINVAL; | |
6770c64e | 4045 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
4046 | } |
4047 | ||
182446d0 TH |
4048 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
4049 | struct cftype *cft) | |
18f59ea7 | 4050 | { |
bef8620c | 4051 | return 1; |
18f59ea7 BS |
4052 | } |
4053 | ||
182446d0 TH |
4054 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
4055 | struct cftype *cft, u64 val) | |
18f59ea7 | 4056 | { |
bef8620c | 4057 | if (val == 1) |
0b8f73e1 | 4058 | return 0; |
567fb435 | 4059 | |
bef8620c RG |
4060 | pr_warn_once("Non-hierarchical mode is deprecated. " |
4061 | "Please report your usecase to linux-mm@kvack.org if you " | |
4062 | "depend on this functionality.\n"); | |
567fb435 | 4063 | |
bef8620c | 4064 | return -EINVAL; |
18f59ea7 BS |
4065 | } |
4066 | ||
6f646156 | 4067 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 4068 | { |
42a30035 | 4069 | unsigned long val; |
ce00a967 | 4070 | |
3e32cb2e | 4071 | if (mem_cgroup_is_root(memcg)) { |
a2174e95 | 4072 | /* |
f82a7a86 YA |
4073 | * Approximate root's usage from global state. This isn't |
4074 | * perfect, but the root usage was always an approximation. | |
a2174e95 | 4075 | */ |
f82a7a86 YA |
4076 | val = global_node_page_state(NR_FILE_PAGES) + |
4077 | global_node_page_state(NR_ANON_MAPPED); | |
42a30035 | 4078 | if (swap) |
f82a7a86 | 4079 | val += total_swap_pages - get_nr_swap_pages(); |
3e32cb2e | 4080 | } else { |
ce00a967 | 4081 | if (!swap) |
3e32cb2e | 4082 | val = page_counter_read(&memcg->memory); |
ce00a967 | 4083 | else |
3e32cb2e | 4084 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 4085 | } |
c12176d3 | 4086 | return val; |
ce00a967 JW |
4087 | } |
4088 | ||
3e32cb2e JW |
4089 | enum { |
4090 | RES_USAGE, | |
4091 | RES_LIMIT, | |
4092 | RES_MAX_USAGE, | |
4093 | RES_FAILCNT, | |
4094 | RES_SOFT_LIMIT, | |
4095 | }; | |
ce00a967 | 4096 | |
791badbd | 4097 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 4098 | struct cftype *cft) |
8cdea7c0 | 4099 | { |
182446d0 | 4100 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 4101 | struct page_counter *counter; |
af36f906 | 4102 | |
3e32cb2e | 4103 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 4104 | case _MEM: |
3e32cb2e JW |
4105 | counter = &memcg->memory; |
4106 | break; | |
8c7c6e34 | 4107 | case _MEMSWAP: |
3e32cb2e JW |
4108 | counter = &memcg->memsw; |
4109 | break; | |
510fc4e1 | 4110 | case _KMEM: |
3e32cb2e | 4111 | counter = &memcg->kmem; |
510fc4e1 | 4112 | break; |
d55f90bf | 4113 | case _TCP: |
0db15298 | 4114 | counter = &memcg->tcpmem; |
d55f90bf | 4115 | break; |
8c7c6e34 KH |
4116 | default: |
4117 | BUG(); | |
8c7c6e34 | 4118 | } |
3e32cb2e JW |
4119 | |
4120 | switch (MEMFILE_ATTR(cft->private)) { | |
4121 | case RES_USAGE: | |
4122 | if (counter == &memcg->memory) | |
c12176d3 | 4123 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 4124 | if (counter == &memcg->memsw) |
c12176d3 | 4125 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
4126 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
4127 | case RES_LIMIT: | |
bbec2e15 | 4128 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
4129 | case RES_MAX_USAGE: |
4130 | return (u64)counter->watermark * PAGE_SIZE; | |
4131 | case RES_FAILCNT: | |
4132 | return counter->failcnt; | |
4133 | case RES_SOFT_LIMIT: | |
2178e20c | 4134 | return (u64)READ_ONCE(memcg->soft_limit) * PAGE_SIZE; |
3e32cb2e JW |
4135 | default: |
4136 | BUG(); | |
4137 | } | |
8cdea7c0 | 4138 | } |
510fc4e1 | 4139 | |
6b0ba2ab FS |
4140 | /* |
4141 | * This function doesn't do anything useful. Its only job is to provide a read | |
4142 | * handler for a file so that cgroup_file_mode() will add read permissions. | |
4143 | */ | |
4144 | static int mem_cgroup_dummy_seq_show(__always_unused struct seq_file *m, | |
4145 | __always_unused void *v) | |
4146 | { | |
4147 | return -EINVAL; | |
4148 | } | |
4149 | ||
84c07d11 | 4150 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 4151 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 4152 | { |
bf4f0599 | 4153 | struct obj_cgroup *objcg; |
d6441637 | 4154 | |
9c94bef9 | 4155 | if (mem_cgroup_kmem_disabled()) |
b313aeee VD |
4156 | return 0; |
4157 | ||
da0efe30 MS |
4158 | if (unlikely(mem_cgroup_is_root(memcg))) |
4159 | return 0; | |
d6441637 | 4160 | |
bf4f0599 | 4161 | objcg = obj_cgroup_alloc(); |
f9c69d63 | 4162 | if (!objcg) |
bf4f0599 | 4163 | return -ENOMEM; |
f9c69d63 | 4164 | |
bf4f0599 RG |
4165 | objcg->memcg = memcg; |
4166 | rcu_assign_pointer(memcg->objcg, objcg); | |
675d6c9b RG |
4167 | obj_cgroup_get(objcg); |
4168 | memcg->orig_objcg = objcg; | |
bf4f0599 | 4169 | |
f7a449f7 | 4170 | static_branch_enable(&memcg_kmem_online_key); |
d648bcc7 | 4171 | |
f9c69d63 | 4172 | memcg->kmemcg_id = memcg->id.id; |
0b8f73e1 JW |
4173 | |
4174 | return 0; | |
d6441637 VD |
4175 | } |
4176 | ||
8e0a8912 JW |
4177 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
4178 | { | |
64268868 | 4179 | struct mem_cgroup *parent; |
8e0a8912 | 4180 | |
9c94bef9 | 4181 | if (mem_cgroup_kmem_disabled()) |
da0efe30 MS |
4182 | return; |
4183 | ||
4184 | if (unlikely(mem_cgroup_is_root(memcg))) | |
8e0a8912 | 4185 | return; |
9855609b | 4186 | |
8e0a8912 JW |
4187 | parent = parent_mem_cgroup(memcg); |
4188 | if (!parent) | |
4189 | parent = root_mem_cgroup; | |
4190 | ||
bf4f0599 | 4191 | memcg_reparent_objcgs(memcg, parent); |
fb2f2b0a | 4192 | |
8e0a8912 | 4193 | /* |
64268868 MS |
4194 | * After we have finished memcg_reparent_objcgs(), all list_lrus |
4195 | * corresponding to this cgroup are guaranteed to remain empty. | |
4196 | * The ordering is imposed by list_lru_node->lock taken by | |
1f391eb2 | 4197 | * memcg_reparent_list_lrus(). |
8e0a8912 | 4198 | */ |
1f391eb2 | 4199 | memcg_reparent_list_lrus(memcg, parent); |
8e0a8912 | 4200 | } |
d6441637 | 4201 | #else |
0b8f73e1 | 4202 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
4203 | { |
4204 | return 0; | |
4205 | } | |
4206 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
4207 | { | |
4208 | } | |
84c07d11 | 4209 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 4210 | |
bbec2e15 | 4211 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
4212 | { |
4213 | int ret; | |
4214 | ||
bbec2e15 | 4215 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 4216 | |
bbec2e15 | 4217 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
4218 | if (ret) |
4219 | goto out; | |
4220 | ||
0db15298 | 4221 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
4222 | /* |
4223 | * The active flag needs to be written after the static_key | |
4224 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
4225 | * function is the last one to run. See mem_cgroup_sk_alloc() |
4226 | * for details, and note that we don't mark any socket as | |
4227 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
4228 | * |
4229 | * We need to do this, because static_keys will span multiple | |
4230 | * sites, but we can't control their order. If we mark a socket | |
4231 | * as accounted, but the accounting functions are not patched in | |
4232 | * yet, we'll lose accounting. | |
4233 | * | |
2d758073 | 4234 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
4235 | * because when this value change, the code to process it is not |
4236 | * patched in yet. | |
4237 | */ | |
4238 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 4239 | memcg->tcpmem_active = true; |
d55f90bf VD |
4240 | } |
4241 | out: | |
bbec2e15 | 4242 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
4243 | return ret; |
4244 | } | |
d55f90bf | 4245 | |
628f4235 KH |
4246 | /* |
4247 | * The user of this function is... | |
4248 | * RES_LIMIT. | |
4249 | */ | |
451af504 TH |
4250 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
4251 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 4252 | { |
451af504 | 4253 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 4254 | unsigned long nr_pages; |
628f4235 KH |
4255 | int ret; |
4256 | ||
451af504 | 4257 | buf = strstrip(buf); |
650c5e56 | 4258 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
4259 | if (ret) |
4260 | return ret; | |
af36f906 | 4261 | |
3e32cb2e | 4262 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 4263 | case RES_LIMIT: |
4b3bde4c BS |
4264 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
4265 | ret = -EINVAL; | |
4266 | break; | |
4267 | } | |
3e32cb2e JW |
4268 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
4269 | case _MEM: | |
bbec2e15 | 4270 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 4271 | break; |
3e32cb2e | 4272 | case _MEMSWAP: |
bbec2e15 | 4273 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 4274 | break; |
4597648f MH |
4275 | case _KMEM: |
4276 | pr_warn_once("kmem.limit_in_bytes is deprecated and will be removed. " | |
4277 | "Writing any value to this file has no effect. " | |
4278 | "Please report your usecase to linux-mm@kvack.org if you " | |
4279 | "depend on this functionality.\n"); | |
4280 | ret = 0; | |
4281 | break; | |
d55f90bf | 4282 | case _TCP: |
bbec2e15 | 4283 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 4284 | break; |
3e32cb2e | 4285 | } |
296c81d8 | 4286 | break; |
3e32cb2e | 4287 | case RES_SOFT_LIMIT: |
2343e88d SAS |
4288 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) { |
4289 | ret = -EOPNOTSUPP; | |
4290 | } else { | |
2178e20c | 4291 | WRITE_ONCE(memcg->soft_limit, nr_pages); |
2343e88d SAS |
4292 | ret = 0; |
4293 | } | |
628f4235 KH |
4294 | break; |
4295 | } | |
451af504 | 4296 | return ret ?: nbytes; |
8cdea7c0 BS |
4297 | } |
4298 | ||
6770c64e TH |
4299 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
4300 | size_t nbytes, loff_t off) | |
c84872e1 | 4301 | { |
6770c64e | 4302 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 4303 | struct page_counter *counter; |
c84872e1 | 4304 | |
3e32cb2e JW |
4305 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
4306 | case _MEM: | |
4307 | counter = &memcg->memory; | |
4308 | break; | |
4309 | case _MEMSWAP: | |
4310 | counter = &memcg->memsw; | |
4311 | break; | |
4312 | case _KMEM: | |
4313 | counter = &memcg->kmem; | |
4314 | break; | |
d55f90bf | 4315 | case _TCP: |
0db15298 | 4316 | counter = &memcg->tcpmem; |
d55f90bf | 4317 | break; |
3e32cb2e JW |
4318 | default: |
4319 | BUG(); | |
4320 | } | |
af36f906 | 4321 | |
3e32cb2e | 4322 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 4323 | case RES_MAX_USAGE: |
3e32cb2e | 4324 | page_counter_reset_watermark(counter); |
29f2a4da PE |
4325 | break; |
4326 | case RES_FAILCNT: | |
3e32cb2e | 4327 | counter->failcnt = 0; |
29f2a4da | 4328 | break; |
3e32cb2e JW |
4329 | default: |
4330 | BUG(); | |
29f2a4da | 4331 | } |
f64c3f54 | 4332 | |
6770c64e | 4333 | return nbytes; |
c84872e1 PE |
4334 | } |
4335 | ||
182446d0 | 4336 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
4337 | struct cftype *cft) |
4338 | { | |
182446d0 | 4339 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
4340 | } |
4341 | ||
02491447 | 4342 | #ifdef CONFIG_MMU |
182446d0 | 4343 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
4344 | struct cftype *cft, u64 val) |
4345 | { | |
182446d0 | 4346 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 4347 | |
da34a848 JW |
4348 | pr_warn_once("Cgroup memory moving (move_charge_at_immigrate) is deprecated. " |
4349 | "Please report your usecase to linux-mm@kvack.org if you " | |
4350 | "depend on this functionality.\n"); | |
4351 | ||
1dfab5ab | 4352 | if (val & ~MOVE_MASK) |
7dc74be0 | 4353 | return -EINVAL; |
ee5e8472 | 4354 | |
7dc74be0 | 4355 | /* |
ee5e8472 GC |
4356 | * No kind of locking is needed in here, because ->can_attach() will |
4357 | * check this value once in the beginning of the process, and then carry | |
4358 | * on with stale data. This means that changes to this value will only | |
4359 | * affect task migrations starting after the change. | |
7dc74be0 | 4360 | */ |
c0ff4b85 | 4361 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
4362 | return 0; |
4363 | } | |
02491447 | 4364 | #else |
182446d0 | 4365 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
4366 | struct cftype *cft, u64 val) |
4367 | { | |
4368 | return -ENOSYS; | |
4369 | } | |
4370 | #endif | |
7dc74be0 | 4371 | |
406eb0c9 | 4372 | #ifdef CONFIG_NUMA |
113b7dfd JW |
4373 | |
4374 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
4375 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
4376 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) | |
4377 | ||
4378 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 | 4379 | int nid, unsigned int lru_mask, bool tree) |
113b7dfd | 4380 | { |
867e5e1d | 4381 | struct lruvec *lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); |
113b7dfd JW |
4382 | unsigned long nr = 0; |
4383 | enum lru_list lru; | |
4384 | ||
4385 | VM_BUG_ON((unsigned)nid >= nr_node_ids); | |
4386 | ||
4387 | for_each_lru(lru) { | |
4388 | if (!(BIT(lru) & lru_mask)) | |
4389 | continue; | |
dd8657b6 SB |
4390 | if (tree) |
4391 | nr += lruvec_page_state(lruvec, NR_LRU_BASE + lru); | |
4392 | else | |
4393 | nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru); | |
113b7dfd JW |
4394 | } |
4395 | return nr; | |
4396 | } | |
4397 | ||
4398 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 SB |
4399 | unsigned int lru_mask, |
4400 | bool tree) | |
113b7dfd JW |
4401 | { |
4402 | unsigned long nr = 0; | |
4403 | enum lru_list lru; | |
4404 | ||
4405 | for_each_lru(lru) { | |
4406 | if (!(BIT(lru) & lru_mask)) | |
4407 | continue; | |
dd8657b6 SB |
4408 | if (tree) |
4409 | nr += memcg_page_state(memcg, NR_LRU_BASE + lru); | |
4410 | else | |
4411 | nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru); | |
113b7dfd JW |
4412 | } |
4413 | return nr; | |
4414 | } | |
4415 | ||
2da8ca82 | 4416 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 4417 | { |
25485de6 GT |
4418 | struct numa_stat { |
4419 | const char *name; | |
4420 | unsigned int lru_mask; | |
4421 | }; | |
4422 | ||
4423 | static const struct numa_stat stats[] = { | |
4424 | { "total", LRU_ALL }, | |
4425 | { "file", LRU_ALL_FILE }, | |
4426 | { "anon", LRU_ALL_ANON }, | |
4427 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
4428 | }; | |
4429 | const struct numa_stat *stat; | |
406eb0c9 | 4430 | int nid; |
aa9694bb | 4431 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
406eb0c9 | 4432 | |
7d7ef0a4 | 4433 | mem_cgroup_flush_stats(memcg); |
2d146aa3 | 4434 | |
25485de6 | 4435 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
4436 | seq_printf(m, "%s=%lu", stat->name, |
4437 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
4438 | false)); | |
4439 | for_each_node_state(nid, N_MEMORY) | |
4440 | seq_printf(m, " N%d=%lu", nid, | |
4441 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
4442 | stat->lru_mask, false)); | |
25485de6 | 4443 | seq_putc(m, '\n'); |
406eb0c9 | 4444 | } |
406eb0c9 | 4445 | |
071aee13 | 4446 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
4447 | |
4448 | seq_printf(m, "hierarchical_%s=%lu", stat->name, | |
4449 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
4450 | true)); | |
4451 | for_each_node_state(nid, N_MEMORY) | |
4452 | seq_printf(m, " N%d=%lu", nid, | |
4453 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
4454 | stat->lru_mask, true)); | |
071aee13 | 4455 | seq_putc(m, '\n'); |
406eb0c9 | 4456 | } |
406eb0c9 | 4457 | |
406eb0c9 YH |
4458 | return 0; |
4459 | } | |
4460 | #endif /* CONFIG_NUMA */ | |
4461 | ||
c8713d0b | 4462 | static const unsigned int memcg1_stats[] = { |
0d1c2072 | 4463 | NR_FILE_PAGES, |
be5d0a74 | 4464 | NR_ANON_MAPPED, |
468c3982 JW |
4465 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4466 | NR_ANON_THPS, | |
4467 | #endif | |
c8713d0b JW |
4468 | NR_SHMEM, |
4469 | NR_FILE_MAPPED, | |
4470 | NR_FILE_DIRTY, | |
4471 | NR_WRITEBACK, | |
e09b0b61 YS |
4472 | WORKINGSET_REFAULT_ANON, |
4473 | WORKINGSET_REFAULT_FILE, | |
72a14e82 | 4474 | #ifdef CONFIG_SWAP |
c8713d0b | 4475 | MEMCG_SWAP, |
72a14e82 LS |
4476 | NR_SWAPCACHE, |
4477 | #endif | |
c8713d0b JW |
4478 | }; |
4479 | ||
4480 | static const char *const memcg1_stat_names[] = { | |
4481 | "cache", | |
4482 | "rss", | |
468c3982 | 4483 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
c8713d0b | 4484 | "rss_huge", |
468c3982 | 4485 | #endif |
c8713d0b JW |
4486 | "shmem", |
4487 | "mapped_file", | |
4488 | "dirty", | |
4489 | "writeback", | |
e09b0b61 YS |
4490 | "workingset_refault_anon", |
4491 | "workingset_refault_file", | |
72a14e82 | 4492 | #ifdef CONFIG_SWAP |
c8713d0b | 4493 | "swap", |
72a14e82 LS |
4494 | "swapcached", |
4495 | #endif | |
c8713d0b JW |
4496 | }; |
4497 | ||
df0e53d0 | 4498 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 4499 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
4500 | PGPGIN, |
4501 | PGPGOUT, | |
4502 | PGFAULT, | |
4503 | PGMAJFAULT, | |
4504 | }; | |
4505 | ||
dddb44ff | 4506 | static void memcg1_stat_format(struct mem_cgroup *memcg, struct seq_buf *s) |
d2ceb9b7 | 4507 | { |
3e32cb2e | 4508 | unsigned long memory, memsw; |
af7c4b0e JW |
4509 | struct mem_cgroup *mi; |
4510 | unsigned int i; | |
406eb0c9 | 4511 | |
71cd3113 | 4512 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c | 4513 | |
7d7ef0a4 | 4514 | mem_cgroup_flush_stats(memcg); |
2d146aa3 | 4515 | |
71cd3113 | 4516 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
468c3982 JW |
4517 | unsigned long nr; |
4518 | ||
ff841a06 YA |
4519 | nr = memcg_page_state_local_output(memcg, memcg1_stats[i]); |
4520 | seq_buf_printf(s, "%s %lu\n", memcg1_stat_names[i], nr); | |
1dd3a273 | 4521 | } |
7b854121 | 4522 | |
df0e53d0 | 4523 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
dddb44ff YA |
4524 | seq_buf_printf(s, "%s %lu\n", vm_event_name(memcg1_events[i]), |
4525 | memcg_events_local(memcg, memcg1_events[i])); | |
af7c4b0e JW |
4526 | |
4527 | for (i = 0; i < NR_LRU_LISTS; i++) | |
dddb44ff YA |
4528 | seq_buf_printf(s, "%s %lu\n", lru_list_name(i), |
4529 | memcg_page_state_local(memcg, NR_LRU_BASE + i) * | |
4530 | PAGE_SIZE); | |
af7c4b0e | 4531 | |
14067bb3 | 4532 | /* Hierarchical information */ |
3e32cb2e JW |
4533 | memory = memsw = PAGE_COUNTER_MAX; |
4534 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
15b42562 CD |
4535 | memory = min(memory, READ_ONCE(mi->memory.max)); |
4536 | memsw = min(memsw, READ_ONCE(mi->memsw.max)); | |
fee7b548 | 4537 | } |
dddb44ff YA |
4538 | seq_buf_printf(s, "hierarchical_memory_limit %llu\n", |
4539 | (u64)memory * PAGE_SIZE); | |
840ea53a LS |
4540 | seq_buf_printf(s, "hierarchical_memsw_limit %llu\n", |
4541 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 4542 | |
8de7ecc6 | 4543 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
7de2e9f1 | 4544 | unsigned long nr; |
4545 | ||
ff841a06 | 4546 | nr = memcg_page_state_output(memcg, memcg1_stats[i]); |
dddb44ff | 4547 | seq_buf_printf(s, "total_%s %llu\n", memcg1_stat_names[i], |
ff841a06 | 4548 | (u64)nr); |
af7c4b0e JW |
4549 | } |
4550 | ||
8de7ecc6 | 4551 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
dddb44ff YA |
4552 | seq_buf_printf(s, "total_%s %llu\n", |
4553 | vm_event_name(memcg1_events[i]), | |
4554 | (u64)memcg_events(memcg, memcg1_events[i])); | |
af7c4b0e | 4555 | |
8de7ecc6 | 4556 | for (i = 0; i < NR_LRU_LISTS; i++) |
dddb44ff YA |
4557 | seq_buf_printf(s, "total_%s %llu\n", lru_list_name(i), |
4558 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * | |
4559 | PAGE_SIZE); | |
14067bb3 | 4560 | |
7f016ee8 | 4561 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 4562 | { |
ef8f2327 MG |
4563 | pg_data_t *pgdat; |
4564 | struct mem_cgroup_per_node *mz; | |
1431d4d1 JW |
4565 | unsigned long anon_cost = 0; |
4566 | unsigned long file_cost = 0; | |
7f016ee8 | 4567 | |
ef8f2327 | 4568 | for_each_online_pgdat(pgdat) { |
a3747b53 | 4569 | mz = memcg->nodeinfo[pgdat->node_id]; |
7f016ee8 | 4570 | |
1431d4d1 JW |
4571 | anon_cost += mz->lruvec.anon_cost; |
4572 | file_cost += mz->lruvec.file_cost; | |
ef8f2327 | 4573 | } |
dddb44ff YA |
4574 | seq_buf_printf(s, "anon_cost %lu\n", anon_cost); |
4575 | seq_buf_printf(s, "file_cost %lu\n", file_cost); | |
7f016ee8 KM |
4576 | } |
4577 | #endif | |
d2ceb9b7 KH |
4578 | } |
4579 | ||
182446d0 TH |
4580 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
4581 | struct cftype *cft) | |
a7885eb8 | 4582 | { |
182446d0 | 4583 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4584 | |
1f4c025b | 4585 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4586 | } |
4587 | ||
182446d0 TH |
4588 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
4589 | struct cftype *cft, u64 val) | |
a7885eb8 | 4590 | { |
182446d0 | 4591 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4592 | |
37bc3cb9 | 4593 | if (val > 200) |
a7885eb8 KM |
4594 | return -EINVAL; |
4595 | ||
a4792030 | 4596 | if (!mem_cgroup_is_root(memcg)) |
82b3aa26 | 4597 | WRITE_ONCE(memcg->swappiness, val); |
3dae7fec | 4598 | else |
82b3aa26 | 4599 | WRITE_ONCE(vm_swappiness, val); |
068b38c1 | 4600 | |
a7885eb8 KM |
4601 | return 0; |
4602 | } | |
4603 | ||
2e72b634 KS |
4604 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4605 | { | |
4606 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 4607 | unsigned long usage; |
2e72b634 KS |
4608 | int i; |
4609 | ||
4610 | rcu_read_lock(); | |
4611 | if (!swap) | |
2c488db2 | 4612 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4613 | else |
2c488db2 | 4614 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4615 | |
4616 | if (!t) | |
4617 | goto unlock; | |
4618 | ||
ce00a967 | 4619 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
4620 | |
4621 | /* | |
748dad36 | 4622 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4623 | * If it's not true, a threshold was crossed after last |
4624 | * call of __mem_cgroup_threshold(). | |
4625 | */ | |
5407a562 | 4626 | i = t->current_threshold; |
2e72b634 KS |
4627 | |
4628 | /* | |
4629 | * Iterate backward over array of thresholds starting from | |
4630 | * current_threshold and check if a threshold is crossed. | |
4631 | * If none of thresholds below usage is crossed, we read | |
4632 | * only one element of the array here. | |
4633 | */ | |
4634 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3652117f | 4635 | eventfd_signal(t->entries[i].eventfd); |
2e72b634 KS |
4636 | |
4637 | /* i = current_threshold + 1 */ | |
4638 | i++; | |
4639 | ||
4640 | /* | |
4641 | * Iterate forward over array of thresholds starting from | |
4642 | * current_threshold+1 and check if a threshold is crossed. | |
4643 | * If none of thresholds above usage is crossed, we read | |
4644 | * only one element of the array here. | |
4645 | */ | |
4646 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3652117f | 4647 | eventfd_signal(t->entries[i].eventfd); |
2e72b634 KS |
4648 | |
4649 | /* Update current_threshold */ | |
5407a562 | 4650 | t->current_threshold = i - 1; |
2e72b634 KS |
4651 | unlock: |
4652 | rcu_read_unlock(); | |
4653 | } | |
4654 | ||
4655 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4656 | { | |
ad4ca5f4 KS |
4657 | while (memcg) { |
4658 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 4659 | if (do_memsw_account()) |
ad4ca5f4 KS |
4660 | __mem_cgroup_threshold(memcg, true); |
4661 | ||
4662 | memcg = parent_mem_cgroup(memcg); | |
4663 | } | |
2e72b634 KS |
4664 | } |
4665 | ||
4666 | static int compare_thresholds(const void *a, const void *b) | |
4667 | { | |
4668 | const struct mem_cgroup_threshold *_a = a; | |
4669 | const struct mem_cgroup_threshold *_b = b; | |
4670 | ||
2bff24a3 GT |
4671 | if (_a->threshold > _b->threshold) |
4672 | return 1; | |
4673 | ||
4674 | if (_a->threshold < _b->threshold) | |
4675 | return -1; | |
4676 | ||
4677 | return 0; | |
2e72b634 KS |
4678 | } |
4679 | ||
c0ff4b85 | 4680 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4681 | { |
4682 | struct mem_cgroup_eventfd_list *ev; | |
4683 | ||
2bcf2e92 MH |
4684 | spin_lock(&memcg_oom_lock); |
4685 | ||
c0ff4b85 | 4686 | list_for_each_entry(ev, &memcg->oom_notify, list) |
3652117f | 4687 | eventfd_signal(ev->eventfd); |
2bcf2e92 MH |
4688 | |
4689 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
4690 | return 0; |
4691 | } | |
4692 | ||
c0ff4b85 | 4693 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4694 | { |
7d74b06f KH |
4695 | struct mem_cgroup *iter; |
4696 | ||
c0ff4b85 | 4697 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4698 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4699 | } |
4700 | ||
59b6f873 | 4701 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4702 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4703 | { |
2c488db2 KS |
4704 | struct mem_cgroup_thresholds *thresholds; |
4705 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
4706 | unsigned long threshold; |
4707 | unsigned long usage; | |
2c488db2 | 4708 | int i, size, ret; |
2e72b634 | 4709 | |
650c5e56 | 4710 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
4711 | if (ret) |
4712 | return ret; | |
4713 | ||
4714 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4715 | |
05b84301 | 4716 | if (type == _MEM) { |
2c488db2 | 4717 | thresholds = &memcg->thresholds; |
ce00a967 | 4718 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4719 | } else if (type == _MEMSWAP) { |
2c488db2 | 4720 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4721 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4722 | } else |
2e72b634 KS |
4723 | BUG(); |
4724 | ||
2e72b634 | 4725 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4726 | if (thresholds->primary) |
2e72b634 KS |
4727 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4728 | ||
2c488db2 | 4729 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4730 | |
4731 | /* Allocate memory for new array of thresholds */ | |
67b8046f | 4732 | new = kmalloc(struct_size(new, entries, size), GFP_KERNEL); |
2c488db2 | 4733 | if (!new) { |
2e72b634 KS |
4734 | ret = -ENOMEM; |
4735 | goto unlock; | |
4736 | } | |
2c488db2 | 4737 | new->size = size; |
2e72b634 KS |
4738 | |
4739 | /* Copy thresholds (if any) to new array */ | |
e90342e6 GS |
4740 | if (thresholds->primary) |
4741 | memcpy(new->entries, thresholds->primary->entries, | |
4742 | flex_array_size(new, entries, size - 1)); | |
2c488db2 | 4743 | |
2e72b634 | 4744 | /* Add new threshold */ |
2c488db2 KS |
4745 | new->entries[size - 1].eventfd = eventfd; |
4746 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4747 | |
4748 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
61e604e6 | 4749 | sort(new->entries, size, sizeof(*new->entries), |
2e72b634 KS |
4750 | compare_thresholds, NULL); |
4751 | ||
4752 | /* Find current threshold */ | |
2c488db2 | 4753 | new->current_threshold = -1; |
2e72b634 | 4754 | for (i = 0; i < size; i++) { |
748dad36 | 4755 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4756 | /* |
2c488db2 KS |
4757 | * new->current_threshold will not be used until |
4758 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4759 | * it here. |
4760 | */ | |
2c488db2 | 4761 | ++new->current_threshold; |
748dad36 SZ |
4762 | } else |
4763 | break; | |
2e72b634 KS |
4764 | } |
4765 | ||
2c488db2 KS |
4766 | /* Free old spare buffer and save old primary buffer as spare */ |
4767 | kfree(thresholds->spare); | |
4768 | thresholds->spare = thresholds->primary; | |
4769 | ||
4770 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4771 | |
907860ed | 4772 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4773 | synchronize_rcu(); |
4774 | ||
2e72b634 KS |
4775 | unlock: |
4776 | mutex_unlock(&memcg->thresholds_lock); | |
4777 | ||
4778 | return ret; | |
4779 | } | |
4780 | ||
59b6f873 | 4781 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4782 | struct eventfd_ctx *eventfd, const char *args) |
4783 | { | |
59b6f873 | 4784 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4785 | } |
4786 | ||
59b6f873 | 4787 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4788 | struct eventfd_ctx *eventfd, const char *args) |
4789 | { | |
59b6f873 | 4790 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4791 | } |
4792 | ||
59b6f873 | 4793 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4794 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4795 | { |
2c488db2 KS |
4796 | struct mem_cgroup_thresholds *thresholds; |
4797 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 4798 | unsigned long usage; |
7d36665a | 4799 | int i, j, size, entries; |
2e72b634 KS |
4800 | |
4801 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4802 | |
4803 | if (type == _MEM) { | |
2c488db2 | 4804 | thresholds = &memcg->thresholds; |
ce00a967 | 4805 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4806 | } else if (type == _MEMSWAP) { |
2c488db2 | 4807 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4808 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4809 | } else |
2e72b634 KS |
4810 | BUG(); |
4811 | ||
371528ca AV |
4812 | if (!thresholds->primary) |
4813 | goto unlock; | |
4814 | ||
2e72b634 KS |
4815 | /* Check if a threshold crossed before removing */ |
4816 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4817 | ||
4818 | /* Calculate new number of threshold */ | |
7d36665a | 4819 | size = entries = 0; |
2c488db2 KS |
4820 | for (i = 0; i < thresholds->primary->size; i++) { |
4821 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 | 4822 | size++; |
7d36665a CX |
4823 | else |
4824 | entries++; | |
2e72b634 KS |
4825 | } |
4826 | ||
2c488db2 | 4827 | new = thresholds->spare; |
907860ed | 4828 | |
7d36665a CX |
4829 | /* If no items related to eventfd have been cleared, nothing to do */ |
4830 | if (!entries) | |
4831 | goto unlock; | |
4832 | ||
2e72b634 KS |
4833 | /* Set thresholds array to NULL if we don't have thresholds */ |
4834 | if (!size) { | |
2c488db2 KS |
4835 | kfree(new); |
4836 | new = NULL; | |
907860ed | 4837 | goto swap_buffers; |
2e72b634 KS |
4838 | } |
4839 | ||
2c488db2 | 4840 | new->size = size; |
2e72b634 KS |
4841 | |
4842 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4843 | new->current_threshold = -1; |
4844 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4845 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4846 | continue; |
4847 | ||
2c488db2 | 4848 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4849 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4850 | /* |
2c488db2 | 4851 | * new->current_threshold will not be used |
2e72b634 KS |
4852 | * until rcu_assign_pointer(), so it's safe to increment |
4853 | * it here. | |
4854 | */ | |
2c488db2 | 4855 | ++new->current_threshold; |
2e72b634 KS |
4856 | } |
4857 | j++; | |
4858 | } | |
4859 | ||
907860ed | 4860 | swap_buffers: |
2c488db2 KS |
4861 | /* Swap primary and spare array */ |
4862 | thresholds->spare = thresholds->primary; | |
8c757763 | 4863 | |
2c488db2 | 4864 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4865 | |
907860ed | 4866 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4867 | synchronize_rcu(); |
6611d8d7 MC |
4868 | |
4869 | /* If all events are unregistered, free the spare array */ | |
4870 | if (!new) { | |
4871 | kfree(thresholds->spare); | |
4872 | thresholds->spare = NULL; | |
4873 | } | |
371528ca | 4874 | unlock: |
2e72b634 | 4875 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4876 | } |
c1e862c1 | 4877 | |
59b6f873 | 4878 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4879 | struct eventfd_ctx *eventfd) |
4880 | { | |
59b6f873 | 4881 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4882 | } |
4883 | ||
59b6f873 | 4884 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4885 | struct eventfd_ctx *eventfd) |
4886 | { | |
59b6f873 | 4887 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4888 | } |
4889 | ||
59b6f873 | 4890 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4891 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4892 | { |
9490ff27 | 4893 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4894 | |
9490ff27 KH |
4895 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4896 | if (!event) | |
4897 | return -ENOMEM; | |
4898 | ||
1af8efe9 | 4899 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4900 | |
4901 | event->eventfd = eventfd; | |
4902 | list_add(&event->list, &memcg->oom_notify); | |
4903 | ||
4904 | /* already in OOM ? */ | |
c2b42d3c | 4905 | if (memcg->under_oom) |
3652117f | 4906 | eventfd_signal(eventfd); |
1af8efe9 | 4907 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4908 | |
4909 | return 0; | |
4910 | } | |
4911 | ||
59b6f873 | 4912 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4913 | struct eventfd_ctx *eventfd) |
9490ff27 | 4914 | { |
9490ff27 | 4915 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4916 | |
1af8efe9 | 4917 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4918 | |
c0ff4b85 | 4919 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4920 | if (ev->eventfd == eventfd) { |
4921 | list_del(&ev->list); | |
4922 | kfree(ev); | |
4923 | } | |
4924 | } | |
4925 | ||
1af8efe9 | 4926 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4927 | } |
4928 | ||
2da8ca82 | 4929 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4930 | { |
aa9694bb | 4931 | struct mem_cgroup *memcg = mem_cgroup_from_seq(sf); |
3c11ecf4 | 4932 | |
17c56de6 | 4933 | seq_printf(sf, "oom_kill_disable %d\n", READ_ONCE(memcg->oom_kill_disable)); |
c2b42d3c | 4934 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
4935 | seq_printf(sf, "oom_kill %lu\n", |
4936 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
4937 | return 0; |
4938 | } | |
4939 | ||
182446d0 | 4940 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4941 | struct cftype *cft, u64 val) |
4942 | { | |
182446d0 | 4943 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4944 | |
4945 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
a4792030 | 4946 | if (mem_cgroup_is_root(memcg) || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4947 | return -EINVAL; |
4948 | ||
17c56de6 | 4949 | WRITE_ONCE(memcg->oom_kill_disable, val); |
4d845ebf | 4950 | if (!val) |
c0ff4b85 | 4951 | memcg_oom_recover(memcg); |
3dae7fec | 4952 | |
3c11ecf4 KH |
4953 | return 0; |
4954 | } | |
4955 | ||
52ebea74 TH |
4956 | #ifdef CONFIG_CGROUP_WRITEBACK |
4957 | ||
3a8e9ac8 TH |
4958 | #include <trace/events/writeback.h> |
4959 | ||
841710aa TH |
4960 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
4961 | { | |
4962 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
4963 | } | |
4964 | ||
4965 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4966 | { | |
4967 | wb_domain_exit(&memcg->cgwb_domain); | |
4968 | } | |
4969 | ||
2529bb3a TH |
4970 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4971 | { | |
4972 | wb_domain_size_changed(&memcg->cgwb_domain); | |
4973 | } | |
4974 | ||
841710aa TH |
4975 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
4976 | { | |
4977 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4978 | ||
4979 | if (!memcg->css.parent) | |
4980 | return NULL; | |
4981 | ||
4982 | return &memcg->cgwb_domain; | |
4983 | } | |
4984 | ||
c2aa723a TH |
4985 | /** |
4986 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
4987 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
4988 | * @pfilepages: out parameter for number of file pages |
4989 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
4990 | * @pdirty: out parameter for number of dirty pages |
4991 | * @pwriteback: out parameter for number of pages under writeback | |
4992 | * | |
c5edf9cd TH |
4993 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
4994 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
4995 | * is a bit more involved. | |
c2aa723a | 4996 | * |
c5edf9cd TH |
4997 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
4998 | * headroom is calculated as the lowest headroom of itself and the | |
4999 | * ancestors. Note that this doesn't consider the actual amount of | |
5000 | * available memory in the system. The caller should further cap | |
5001 | * *@pheadroom accordingly. | |
c2aa723a | 5002 | */ |
c5edf9cd TH |
5003 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
5004 | unsigned long *pheadroom, unsigned long *pdirty, | |
5005 | unsigned long *pwriteback) | |
c2aa723a TH |
5006 | { |
5007 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
5008 | struct mem_cgroup *parent; | |
c2aa723a | 5009 | |
d9b3ce87 | 5010 | mem_cgroup_flush_stats_ratelimited(memcg); |
c2aa723a | 5011 | |
2d146aa3 JW |
5012 | *pdirty = memcg_page_state(memcg, NR_FILE_DIRTY); |
5013 | *pwriteback = memcg_page_state(memcg, NR_WRITEBACK); | |
5014 | *pfilepages = memcg_page_state(memcg, NR_INACTIVE_FILE) + | |
5015 | memcg_page_state(memcg, NR_ACTIVE_FILE); | |
c2aa723a | 5016 | |
2d146aa3 | 5017 | *pheadroom = PAGE_COUNTER_MAX; |
c2aa723a | 5018 | while ((parent = parent_mem_cgroup(memcg))) { |
15b42562 | 5019 | unsigned long ceiling = min(READ_ONCE(memcg->memory.max), |
d1663a90 | 5020 | READ_ONCE(memcg->memory.high)); |
c2aa723a TH |
5021 | unsigned long used = page_counter_read(&memcg->memory); |
5022 | ||
c5edf9cd | 5023 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
5024 | memcg = parent; |
5025 | } | |
c2aa723a TH |
5026 | } |
5027 | ||
97b27821 TH |
5028 | /* |
5029 | * Foreign dirty flushing | |
5030 | * | |
5031 | * There's an inherent mismatch between memcg and writeback. The former | |
f0953a1b | 5032 | * tracks ownership per-page while the latter per-inode. This was a |
97b27821 TH |
5033 | * deliberate design decision because honoring per-page ownership in the |
5034 | * writeback path is complicated, may lead to higher CPU and IO overheads | |
5035 | * and deemed unnecessary given that write-sharing an inode across | |
5036 | * different cgroups isn't a common use-case. | |
5037 | * | |
5038 | * Combined with inode majority-writer ownership switching, this works well | |
5039 | * enough in most cases but there are some pathological cases. For | |
5040 | * example, let's say there are two cgroups A and B which keep writing to | |
5041 | * different but confined parts of the same inode. B owns the inode and | |
5042 | * A's memory is limited far below B's. A's dirty ratio can rise enough to | |
5043 | * trigger balance_dirty_pages() sleeps but B's can be low enough to avoid | |
5044 | * triggering background writeback. A will be slowed down without a way to | |
5045 | * make writeback of the dirty pages happen. | |
5046 | * | |
f0953a1b | 5047 | * Conditions like the above can lead to a cgroup getting repeatedly and |
97b27821 | 5048 | * severely throttled after making some progress after each |
f0953a1b | 5049 | * dirty_expire_interval while the underlying IO device is almost |
97b27821 TH |
5050 | * completely idle. |
5051 | * | |
5052 | * Solving this problem completely requires matching the ownership tracking | |
5053 | * granularities between memcg and writeback in either direction. However, | |
5054 | * the more egregious behaviors can be avoided by simply remembering the | |
5055 | * most recent foreign dirtying events and initiating remote flushes on | |
5056 | * them when local writeback isn't enough to keep the memory clean enough. | |
5057 | * | |
5058 | * The following two functions implement such mechanism. When a foreign | |
5059 | * page - a page whose memcg and writeback ownerships don't match - is | |
5060 | * dirtied, mem_cgroup_track_foreign_dirty() records the inode owning | |
5061 | * bdi_writeback on the page owning memcg. When balance_dirty_pages() | |
5062 | * decides that the memcg needs to sleep due to high dirty ratio, it calls | |
5063 | * mem_cgroup_flush_foreign() which queues writeback on the recorded | |
5064 | * foreign bdi_writebacks which haven't expired. Both the numbers of | |
5065 | * recorded bdi_writebacks and concurrent in-flight foreign writebacks are | |
5066 | * limited to MEMCG_CGWB_FRN_CNT. | |
5067 | * | |
5068 | * The mechanism only remembers IDs and doesn't hold any object references. | |
5069 | * As being wrong occasionally doesn't matter, updates and accesses to the | |
5070 | * records are lockless and racy. | |
5071 | */ | |
9d8053fc | 5072 | void mem_cgroup_track_foreign_dirty_slowpath(struct folio *folio, |
97b27821 TH |
5073 | struct bdi_writeback *wb) |
5074 | { | |
9d8053fc | 5075 | struct mem_cgroup *memcg = folio_memcg(folio); |
97b27821 TH |
5076 | struct memcg_cgwb_frn *frn; |
5077 | u64 now = get_jiffies_64(); | |
5078 | u64 oldest_at = now; | |
5079 | int oldest = -1; | |
5080 | int i; | |
5081 | ||
9d8053fc | 5082 | trace_track_foreign_dirty(folio, wb); |
3a8e9ac8 | 5083 | |
97b27821 TH |
5084 | /* |
5085 | * Pick the slot to use. If there is already a slot for @wb, keep | |
5086 | * using it. If not replace the oldest one which isn't being | |
5087 | * written out. | |
5088 | */ | |
5089 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
5090 | frn = &memcg->cgwb_frn[i]; | |
5091 | if (frn->bdi_id == wb->bdi->id && | |
5092 | frn->memcg_id == wb->memcg_css->id) | |
5093 | break; | |
5094 | if (time_before64(frn->at, oldest_at) && | |
5095 | atomic_read(&frn->done.cnt) == 1) { | |
5096 | oldest = i; | |
5097 | oldest_at = frn->at; | |
5098 | } | |
5099 | } | |
5100 | ||
5101 | if (i < MEMCG_CGWB_FRN_CNT) { | |
5102 | /* | |
5103 | * Re-using an existing one. Update timestamp lazily to | |
5104 | * avoid making the cacheline hot. We want them to be | |
5105 | * reasonably up-to-date and significantly shorter than | |
5106 | * dirty_expire_interval as that's what expires the record. | |
5107 | * Use the shorter of 1s and dirty_expire_interval / 8. | |
5108 | */ | |
5109 | unsigned long update_intv = | |
5110 | min_t(unsigned long, HZ, | |
5111 | msecs_to_jiffies(dirty_expire_interval * 10) / 8); | |
5112 | ||
5113 | if (time_before64(frn->at, now - update_intv)) | |
5114 | frn->at = now; | |
5115 | } else if (oldest >= 0) { | |
5116 | /* replace the oldest free one */ | |
5117 | frn = &memcg->cgwb_frn[oldest]; | |
5118 | frn->bdi_id = wb->bdi->id; | |
5119 | frn->memcg_id = wb->memcg_css->id; | |
5120 | frn->at = now; | |
5121 | } | |
5122 | } | |
5123 | ||
5124 | /* issue foreign writeback flushes for recorded foreign dirtying events */ | |
5125 | void mem_cgroup_flush_foreign(struct bdi_writeback *wb) | |
5126 | { | |
5127 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
5128 | unsigned long intv = msecs_to_jiffies(dirty_expire_interval * 10); | |
5129 | u64 now = jiffies_64; | |
5130 | int i; | |
5131 | ||
5132 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
5133 | struct memcg_cgwb_frn *frn = &memcg->cgwb_frn[i]; | |
5134 | ||
5135 | /* | |
5136 | * If the record is older than dirty_expire_interval, | |
5137 | * writeback on it has already started. No need to kick it | |
5138 | * off again. Also, don't start a new one if there's | |
5139 | * already one in flight. | |
5140 | */ | |
5141 | if (time_after64(frn->at, now - intv) && | |
5142 | atomic_read(&frn->done.cnt) == 1) { | |
5143 | frn->at = 0; | |
3a8e9ac8 | 5144 | trace_flush_foreign(wb, frn->bdi_id, frn->memcg_id); |
7490a2d2 | 5145 | cgroup_writeback_by_id(frn->bdi_id, frn->memcg_id, |
97b27821 TH |
5146 | WB_REASON_FOREIGN_FLUSH, |
5147 | &frn->done); | |
5148 | } | |
5149 | } | |
5150 | } | |
5151 | ||
841710aa TH |
5152 | #else /* CONFIG_CGROUP_WRITEBACK */ |
5153 | ||
5154 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
5155 | { | |
5156 | return 0; | |
5157 | } | |
5158 | ||
5159 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
5160 | { | |
5161 | } | |
5162 | ||
2529bb3a TH |
5163 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
5164 | { | |
5165 | } | |
5166 | ||
52ebea74 TH |
5167 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
5168 | ||
3bc942f3 TH |
5169 | /* |
5170 | * DO NOT USE IN NEW FILES. | |
5171 | * | |
5172 | * "cgroup.event_control" implementation. | |
5173 | * | |
5174 | * This is way over-engineered. It tries to support fully configurable | |
5175 | * events for each user. Such level of flexibility is completely | |
5176 | * unnecessary especially in the light of the planned unified hierarchy. | |
5177 | * | |
5178 | * Please deprecate this and replace with something simpler if at all | |
5179 | * possible. | |
5180 | */ | |
5181 | ||
79bd9814 TH |
5182 | /* |
5183 | * Unregister event and free resources. | |
5184 | * | |
5185 | * Gets called from workqueue. | |
5186 | */ | |
3bc942f3 | 5187 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 5188 | { |
3bc942f3 TH |
5189 | struct mem_cgroup_event *event = |
5190 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 5191 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
5192 | |
5193 | remove_wait_queue(event->wqh, &event->wait); | |
5194 | ||
59b6f873 | 5195 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
5196 | |
5197 | /* Notify userspace the event is going away. */ | |
3652117f | 5198 | eventfd_signal(event->eventfd); |
79bd9814 TH |
5199 | |
5200 | eventfd_ctx_put(event->eventfd); | |
5201 | kfree(event); | |
59b6f873 | 5202 | css_put(&memcg->css); |
79bd9814 TH |
5203 | } |
5204 | ||
5205 | /* | |
a9a08845 | 5206 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
5207 | * |
5208 | * Called with wqh->lock held and interrupts disabled. | |
5209 | */ | |
ac6424b9 | 5210 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 5211 | int sync, void *key) |
79bd9814 | 5212 | { |
3bc942f3 TH |
5213 | struct mem_cgroup_event *event = |
5214 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 5215 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 5216 | __poll_t flags = key_to_poll(key); |
79bd9814 | 5217 | |
a9a08845 | 5218 | if (flags & EPOLLHUP) { |
79bd9814 TH |
5219 | /* |
5220 | * If the event has been detached at cgroup removal, we | |
5221 | * can simply return knowing the other side will cleanup | |
5222 | * for us. | |
5223 | * | |
5224 | * We can't race against event freeing since the other | |
5225 | * side will require wqh->lock via remove_wait_queue(), | |
5226 | * which we hold. | |
5227 | */ | |
fba94807 | 5228 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
5229 | if (!list_empty(&event->list)) { |
5230 | list_del_init(&event->list); | |
5231 | /* | |
5232 | * We are in atomic context, but cgroup_event_remove() | |
5233 | * may sleep, so we have to call it in workqueue. | |
5234 | */ | |
5235 | schedule_work(&event->remove); | |
5236 | } | |
fba94807 | 5237 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
5238 | } |
5239 | ||
5240 | return 0; | |
5241 | } | |
5242 | ||
3bc942f3 | 5243 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
5244 | wait_queue_head_t *wqh, poll_table *pt) |
5245 | { | |
3bc942f3 TH |
5246 | struct mem_cgroup_event *event = |
5247 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
5248 | |
5249 | event->wqh = wqh; | |
5250 | add_wait_queue(wqh, &event->wait); | |
5251 | } | |
5252 | ||
5253 | /* | |
3bc942f3 TH |
5254 | * DO NOT USE IN NEW FILES. |
5255 | * | |
79bd9814 TH |
5256 | * Parse input and register new cgroup event handler. |
5257 | * | |
5258 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
5259 | * Interpretation of args is defined by control file implementation. | |
5260 | */ | |
451af504 TH |
5261 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
5262 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 5263 | { |
451af504 | 5264 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 5265 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5266 | struct mem_cgroup_event *event; |
79bd9814 TH |
5267 | struct cgroup_subsys_state *cfile_css; |
5268 | unsigned int efd, cfd; | |
5269 | struct fd efile; | |
5270 | struct fd cfile; | |
4a7ba45b | 5271 | struct dentry *cdentry; |
fba94807 | 5272 | const char *name; |
79bd9814 TH |
5273 | char *endp; |
5274 | int ret; | |
5275 | ||
2343e88d SAS |
5276 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) |
5277 | return -EOPNOTSUPP; | |
5278 | ||
451af504 TH |
5279 | buf = strstrip(buf); |
5280 | ||
5281 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
5282 | if (*endp != ' ') |
5283 | return -EINVAL; | |
451af504 | 5284 | buf = endp + 1; |
79bd9814 | 5285 | |
451af504 | 5286 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
5287 | if ((*endp != ' ') && (*endp != '\0')) |
5288 | return -EINVAL; | |
451af504 | 5289 | buf = endp + 1; |
79bd9814 TH |
5290 | |
5291 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
5292 | if (!event) | |
5293 | return -ENOMEM; | |
5294 | ||
59b6f873 | 5295 | event->memcg = memcg; |
79bd9814 | 5296 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
5297 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
5298 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
5299 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
5300 | |
5301 | efile = fdget(efd); | |
5302 | if (!efile.file) { | |
5303 | ret = -EBADF; | |
5304 | goto out_kfree; | |
5305 | } | |
5306 | ||
5307 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
5308 | if (IS_ERR(event->eventfd)) { | |
5309 | ret = PTR_ERR(event->eventfd); | |
5310 | goto out_put_efile; | |
5311 | } | |
5312 | ||
5313 | cfile = fdget(cfd); | |
5314 | if (!cfile.file) { | |
5315 | ret = -EBADF; | |
5316 | goto out_put_eventfd; | |
5317 | } | |
5318 | ||
5319 | /* the process need read permission on control file */ | |
5320 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
02f92b38 | 5321 | ret = file_permission(cfile.file, MAY_READ); |
79bd9814 TH |
5322 | if (ret < 0) |
5323 | goto out_put_cfile; | |
5324 | ||
4a7ba45b TH |
5325 | /* |
5326 | * The control file must be a regular cgroup1 file. As a regular cgroup | |
5327 | * file can't be renamed, it's safe to access its name afterwards. | |
5328 | */ | |
5329 | cdentry = cfile.file->f_path.dentry; | |
5330 | if (cdentry->d_sb->s_type != &cgroup_fs_type || !d_is_reg(cdentry)) { | |
5331 | ret = -EINVAL; | |
5332 | goto out_put_cfile; | |
5333 | } | |
5334 | ||
fba94807 TH |
5335 | /* |
5336 | * Determine the event callbacks and set them in @event. This used | |
5337 | * to be done via struct cftype but cgroup core no longer knows | |
5338 | * about these events. The following is crude but the whole thing | |
5339 | * is for compatibility anyway. | |
3bc942f3 TH |
5340 | * |
5341 | * DO NOT ADD NEW FILES. | |
fba94807 | 5342 | */ |
4a7ba45b | 5343 | name = cdentry->d_name.name; |
fba94807 TH |
5344 | |
5345 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
5346 | event->register_event = mem_cgroup_usage_register_event; | |
5347 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
5348 | } else if (!strcmp(name, "memory.oom_control")) { | |
5349 | event->register_event = mem_cgroup_oom_register_event; | |
5350 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
5351 | } else if (!strcmp(name, "memory.pressure_level")) { | |
5352 | event->register_event = vmpressure_register_event; | |
5353 | event->unregister_event = vmpressure_unregister_event; | |
5354 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
5355 | event->register_event = memsw_cgroup_usage_register_event; |
5356 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
5357 | } else { |
5358 | ret = -EINVAL; | |
5359 | goto out_put_cfile; | |
5360 | } | |
5361 | ||
79bd9814 | 5362 | /* |
b5557c4c TH |
5363 | * Verify @cfile should belong to @css. Also, remaining events are |
5364 | * automatically removed on cgroup destruction but the removal is | |
5365 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 5366 | */ |
4a7ba45b | 5367 | cfile_css = css_tryget_online_from_dir(cdentry->d_parent, |
ec903c0c | 5368 | &memory_cgrp_subsys); |
79bd9814 | 5369 | ret = -EINVAL; |
5a17f543 | 5370 | if (IS_ERR(cfile_css)) |
79bd9814 | 5371 | goto out_put_cfile; |
5a17f543 TH |
5372 | if (cfile_css != css) { |
5373 | css_put(cfile_css); | |
79bd9814 | 5374 | goto out_put_cfile; |
5a17f543 | 5375 | } |
79bd9814 | 5376 | |
451af504 | 5377 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
5378 | if (ret) |
5379 | goto out_put_css; | |
5380 | ||
9965ed17 | 5381 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 5382 | |
4ba9515d | 5383 | spin_lock_irq(&memcg->event_list_lock); |
fba94807 | 5384 | list_add(&event->list, &memcg->event_list); |
4ba9515d | 5385 | spin_unlock_irq(&memcg->event_list_lock); |
79bd9814 TH |
5386 | |
5387 | fdput(cfile); | |
5388 | fdput(efile); | |
5389 | ||
451af504 | 5390 | return nbytes; |
79bd9814 TH |
5391 | |
5392 | out_put_css: | |
b5557c4c | 5393 | css_put(css); |
79bd9814 TH |
5394 | out_put_cfile: |
5395 | fdput(cfile); | |
5396 | out_put_eventfd: | |
5397 | eventfd_ctx_put(event->eventfd); | |
5398 | out_put_efile: | |
5399 | fdput(efile); | |
5400 | out_kfree: | |
5401 | kfree(event); | |
5402 | ||
5403 | return ret; | |
5404 | } | |
5405 | ||
bc3dcb85 | 5406 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_SLUB_DEBUG) |
c29b5b3d MS |
5407 | static int mem_cgroup_slab_show(struct seq_file *m, void *p) |
5408 | { | |
5409 | /* | |
5410 | * Deprecated. | |
df4ae285 | 5411 | * Please, take a look at tools/cgroup/memcg_slabinfo.py . |
c29b5b3d MS |
5412 | */ |
5413 | return 0; | |
5414 | } | |
5415 | #endif | |
5416 | ||
dddb44ff YA |
5417 | static int memory_stat_show(struct seq_file *m, void *v); |
5418 | ||
241994ed | 5419 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 5420 | { |
0eea1030 | 5421 | .name = "usage_in_bytes", |
8c7c6e34 | 5422 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 5423 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5424 | }, |
c84872e1 PE |
5425 | { |
5426 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 5427 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 5428 | .write = mem_cgroup_reset, |
791badbd | 5429 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 5430 | }, |
8cdea7c0 | 5431 | { |
0eea1030 | 5432 | .name = "limit_in_bytes", |
8c7c6e34 | 5433 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 5434 | .write = mem_cgroup_write, |
791badbd | 5435 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5436 | }, |
296c81d8 BS |
5437 | { |
5438 | .name = "soft_limit_in_bytes", | |
5439 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 5440 | .write = mem_cgroup_write, |
791badbd | 5441 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 5442 | }, |
8cdea7c0 BS |
5443 | { |
5444 | .name = "failcnt", | |
8c7c6e34 | 5445 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 5446 | .write = mem_cgroup_reset, |
791badbd | 5447 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5448 | }, |
d2ceb9b7 KH |
5449 | { |
5450 | .name = "stat", | |
dddb44ff | 5451 | .seq_show = memory_stat_show, |
d2ceb9b7 | 5452 | }, |
c1e862c1 KH |
5453 | { |
5454 | .name = "force_empty", | |
6770c64e | 5455 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 5456 | }, |
18f59ea7 BS |
5457 | { |
5458 | .name = "use_hierarchy", | |
5459 | .write_u64 = mem_cgroup_hierarchy_write, | |
5460 | .read_u64 = mem_cgroup_hierarchy_read, | |
5461 | }, | |
79bd9814 | 5462 | { |
3bc942f3 | 5463 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 5464 | .write = memcg_write_event_control, |
7dbdb199 | 5465 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 5466 | }, |
a7885eb8 KM |
5467 | { |
5468 | .name = "swappiness", | |
5469 | .read_u64 = mem_cgroup_swappiness_read, | |
5470 | .write_u64 = mem_cgroup_swappiness_write, | |
5471 | }, | |
7dc74be0 DN |
5472 | { |
5473 | .name = "move_charge_at_immigrate", | |
5474 | .read_u64 = mem_cgroup_move_charge_read, | |
5475 | .write_u64 = mem_cgroup_move_charge_write, | |
5476 | }, | |
9490ff27 KH |
5477 | { |
5478 | .name = "oom_control", | |
2da8ca82 | 5479 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 5480 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 | 5481 | }, |
70ddf637 AV |
5482 | { |
5483 | .name = "pressure_level", | |
6b0ba2ab | 5484 | .seq_show = mem_cgroup_dummy_seq_show, |
70ddf637 | 5485 | }, |
406eb0c9 YH |
5486 | #ifdef CONFIG_NUMA |
5487 | { | |
5488 | .name = "numa_stat", | |
2da8ca82 | 5489 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
5490 | }, |
5491 | #endif | |
4597648f MH |
5492 | { |
5493 | .name = "kmem.limit_in_bytes", | |
5494 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
5495 | .write = mem_cgroup_write, | |
5496 | .read_u64 = mem_cgroup_read_u64, | |
5497 | }, | |
510fc4e1 GC |
5498 | { |
5499 | .name = "kmem.usage_in_bytes", | |
5500 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 5501 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5502 | }, |
5503 | { | |
5504 | .name = "kmem.failcnt", | |
5505 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 5506 | .write = mem_cgroup_reset, |
791badbd | 5507 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5508 | }, |
5509 | { | |
5510 | .name = "kmem.max_usage_in_bytes", | |
5511 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 5512 | .write = mem_cgroup_reset, |
791badbd | 5513 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 5514 | }, |
bc3dcb85 | 5515 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_SLUB_DEBUG) |
749c5415 GC |
5516 | { |
5517 | .name = "kmem.slabinfo", | |
c29b5b3d | 5518 | .seq_show = mem_cgroup_slab_show, |
749c5415 GC |
5519 | }, |
5520 | #endif | |
d55f90bf VD |
5521 | { |
5522 | .name = "kmem.tcp.limit_in_bytes", | |
5523 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
5524 | .write = mem_cgroup_write, | |
5525 | .read_u64 = mem_cgroup_read_u64, | |
5526 | }, | |
5527 | { | |
5528 | .name = "kmem.tcp.usage_in_bytes", | |
5529 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
5530 | .read_u64 = mem_cgroup_read_u64, | |
5531 | }, | |
5532 | { | |
5533 | .name = "kmem.tcp.failcnt", | |
5534 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
5535 | .write = mem_cgroup_reset, | |
5536 | .read_u64 = mem_cgroup_read_u64, | |
5537 | }, | |
5538 | { | |
5539 | .name = "kmem.tcp.max_usage_in_bytes", | |
5540 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
5541 | .write = mem_cgroup_reset, | |
5542 | .read_u64 = mem_cgroup_read_u64, | |
5543 | }, | |
6bc10349 | 5544 | { }, /* terminate */ |
af36f906 | 5545 | }; |
8c7c6e34 | 5546 | |
73f576c0 JW |
5547 | /* |
5548 | * Private memory cgroup IDR | |
5549 | * | |
5550 | * Swap-out records and page cache shadow entries need to store memcg | |
5551 | * references in constrained space, so we maintain an ID space that is | |
5552 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
5553 | * memory-controlled cgroups to 64k. | |
5554 | * | |
b8f2935f | 5555 | * However, there usually are many references to the offline CSS after |
73f576c0 JW |
5556 | * the cgroup has been destroyed, such as page cache or reclaimable |
5557 | * slab objects, that don't need to hang on to the ID. We want to keep | |
5558 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
5559 | * relatively small ID space and prevent the creation of new cgroups | |
5560 | * even when there are much fewer than 64k cgroups - possibly none. | |
5561 | * | |
5562 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
5563 | * be freed and recycled when it's no longer needed, which is usually | |
5564 | * when the CSS is offlined. | |
5565 | * | |
5566 | * The only exception to that are records of swapped out tmpfs/shmem | |
5567 | * pages that need to be attributed to live ancestors on swapin. But | |
5568 | * those references are manageable from userspace. | |
5569 | */ | |
5570 | ||
60b1e24c | 5571 | #define MEM_CGROUP_ID_MAX ((1UL << MEM_CGROUP_ID_SHIFT) - 1) |
73f576c0 JW |
5572 | static DEFINE_IDR(mem_cgroup_idr); |
5573 | ||
7e97de0b KT |
5574 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
5575 | { | |
5576 | if (memcg->id.id > 0) { | |
5577 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
5578 | memcg->id.id = 0; | |
5579 | } | |
5580 | } | |
5581 | ||
c1514c0a VF |
5582 | static void __maybe_unused mem_cgroup_id_get_many(struct mem_cgroup *memcg, |
5583 | unsigned int n) | |
73f576c0 | 5584 | { |
1c2d479a | 5585 | refcount_add(n, &memcg->id.ref); |
73f576c0 JW |
5586 | } |
5587 | ||
615d66c3 | 5588 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 5589 | { |
1c2d479a | 5590 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 5591 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
5592 | |
5593 | /* Memcg ID pins CSS */ | |
5594 | css_put(&memcg->css); | |
5595 | } | |
5596 | } | |
5597 | ||
615d66c3 VD |
5598 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) |
5599 | { | |
5600 | mem_cgroup_id_put_many(memcg, 1); | |
5601 | } | |
5602 | ||
73f576c0 JW |
5603 | /** |
5604 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
5605 | * @id: the memcg id to look up | |
5606 | * | |
5607 | * Caller must hold rcu_read_lock(). | |
5608 | */ | |
5609 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
5610 | { | |
5611 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5612 | return idr_find(&mem_cgroup_idr, id); | |
5613 | } | |
5614 | ||
c15187a4 RG |
5615 | #ifdef CONFIG_SHRINKER_DEBUG |
5616 | struct mem_cgroup *mem_cgroup_get_from_ino(unsigned long ino) | |
5617 | { | |
5618 | struct cgroup *cgrp; | |
5619 | struct cgroup_subsys_state *css; | |
5620 | struct mem_cgroup *memcg; | |
5621 | ||
5622 | cgrp = cgroup_get_from_id(ino); | |
fa7e439c | 5623 | if (IS_ERR(cgrp)) |
c0f2df49 | 5624 | return ERR_CAST(cgrp); |
c15187a4 RG |
5625 | |
5626 | css = cgroup_get_e_css(cgrp, &memory_cgrp_subsys); | |
5627 | if (css) | |
5628 | memcg = container_of(css, struct mem_cgroup, css); | |
5629 | else | |
5630 | memcg = ERR_PTR(-ENOENT); | |
5631 | ||
5632 | cgroup_put(cgrp); | |
5633 | ||
5634 | return memcg; | |
5635 | } | |
5636 | #endif | |
5637 | ||
a8248bb7 | 5638 | static bool alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
5639 | { |
5640 | struct mem_cgroup_per_node *pn; | |
8c9bb398 WY |
5641 | |
5642 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, node); | |
6d12e2d8 | 5643 | if (!pn) |
a8248bb7 | 5644 | return false; |
1ecaab2b | 5645 | |
aab6103b RG |
5646 | pn->lruvec_stats = kzalloc_node(sizeof(struct lruvec_stats), |
5647 | GFP_KERNEL_ACCOUNT, node); | |
70a64b79 SB |
5648 | if (!pn->lruvec_stats) |
5649 | goto fail; | |
5650 | ||
7e1c0d6f SB |
5651 | pn->lruvec_stats_percpu = alloc_percpu_gfp(struct lruvec_stats_percpu, |
5652 | GFP_KERNEL_ACCOUNT); | |
70a64b79 SB |
5653 | if (!pn->lruvec_stats_percpu) |
5654 | goto fail; | |
00f3ca2c | 5655 | |
ef8f2327 | 5656 | lruvec_init(&pn->lruvec); |
ef8f2327 MG |
5657 | pn->memcg = memcg; |
5658 | ||
54f72fe0 | 5659 | memcg->nodeinfo[node] = pn; |
a8248bb7 | 5660 | return true; |
70a64b79 SB |
5661 | fail: |
5662 | kfree(pn->lruvec_stats); | |
5663 | kfree(pn); | |
a8248bb7 | 5664 | return false; |
6d12e2d8 KH |
5665 | } |
5666 | ||
ef8f2327 | 5667 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 5668 | { |
00f3ca2c JW |
5669 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
5670 | ||
4eaf431f MH |
5671 | if (!pn) |
5672 | return; | |
5673 | ||
7e1c0d6f | 5674 | free_percpu(pn->lruvec_stats_percpu); |
70a64b79 | 5675 | kfree(pn->lruvec_stats); |
00f3ca2c | 5676 | kfree(pn); |
1ecaab2b KH |
5677 | } |
5678 | ||
40e952f9 | 5679 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 5680 | { |
c8b2a36f | 5681 | int node; |
59927fb9 | 5682 | |
91b71e78 | 5683 | obj_cgroup_put(memcg->orig_objcg); |
675d6c9b | 5684 | |
c8b2a36f | 5685 | for_each_node(node) |
ef8f2327 | 5686 | free_mem_cgroup_per_node_info(memcg, node); |
410f8e82 | 5687 | kfree(memcg->vmstats); |
871789d4 | 5688 | free_percpu(memcg->vmstats_percpu); |
8ff69e2c | 5689 | kfree(memcg); |
59927fb9 | 5690 | } |
3afe36b1 | 5691 | |
40e952f9 TE |
5692 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
5693 | { | |
ec1c86b2 | 5694 | lru_gen_exit_memcg(memcg); |
40e952f9 TE |
5695 | memcg_wb_domain_exit(memcg); |
5696 | __mem_cgroup_free(memcg); | |
5697 | } | |
5698 | ||
9cee7e8e | 5699 | static struct mem_cgroup *mem_cgroup_alloc(struct mem_cgroup *parent) |
8cdea7c0 | 5700 | { |
9cee7e8e | 5701 | struct memcg_vmstats_percpu *statc, *pstatc; |
d142e3e6 | 5702 | struct mem_cgroup *memcg; |
9cee7e8e | 5703 | int node, cpu; |
97b27821 | 5704 | int __maybe_unused i; |
11d67612 | 5705 | long error = -ENOMEM; |
8cdea7c0 | 5706 | |
06b2c3b0 | 5707 | memcg = kzalloc(struct_size(memcg, nodeinfo, nr_node_ids), GFP_KERNEL); |
c0ff4b85 | 5708 | if (!memcg) |
11d67612 | 5709 | return ERR_PTR(error); |
0b8f73e1 | 5710 | |
73f576c0 | 5711 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
be740503 | 5712 | 1, MEM_CGROUP_ID_MAX + 1, GFP_KERNEL); |
11d67612 YS |
5713 | if (memcg->id.id < 0) { |
5714 | error = memcg->id.id; | |
73f576c0 | 5715 | goto fail; |
11d67612 | 5716 | } |
73f576c0 | 5717 | |
aab6103b RG |
5718 | memcg->vmstats = kzalloc(sizeof(struct memcg_vmstats), |
5719 | GFP_KERNEL_ACCOUNT); | |
410f8e82 SB |
5720 | if (!memcg->vmstats) |
5721 | goto fail; | |
5722 | ||
3e38e0aa RG |
5723 | memcg->vmstats_percpu = alloc_percpu_gfp(struct memcg_vmstats_percpu, |
5724 | GFP_KERNEL_ACCOUNT); | |
871789d4 | 5725 | if (!memcg->vmstats_percpu) |
0b8f73e1 | 5726 | goto fail; |
78fb7466 | 5727 | |
9cee7e8e YA |
5728 | for_each_possible_cpu(cpu) { |
5729 | if (parent) | |
5730 | pstatc = per_cpu_ptr(parent->vmstats_percpu, cpu); | |
5731 | statc = per_cpu_ptr(memcg->vmstats_percpu, cpu); | |
5732 | statc->parent = parent ? pstatc : NULL; | |
5733 | statc->vmstats = memcg->vmstats; | |
5734 | } | |
5735 | ||
3ed28fa1 | 5736 | for_each_node(node) |
a8248bb7 | 5737 | if (!alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 5738 | goto fail; |
f64c3f54 | 5739 | |
0b8f73e1 JW |
5740 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
5741 | goto fail; | |
28dbc4b6 | 5742 | |
f7e1cb6e | 5743 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 | 5744 | INIT_LIST_HEAD(&memcg->oom_notify); |
d142e3e6 GC |
5745 | mutex_init(&memcg->thresholds_lock); |
5746 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 5747 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
5748 | INIT_LIST_HEAD(&memcg->event_list); |
5749 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 5750 | memcg->socket_pressure = jiffies; |
84c07d11 | 5751 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 5752 | memcg->kmemcg_id = -1; |
bf4f0599 | 5753 | INIT_LIST_HEAD(&memcg->objcg_list); |
900a38f0 | 5754 | #endif |
52ebea74 TH |
5755 | #ifdef CONFIG_CGROUP_WRITEBACK |
5756 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
97b27821 TH |
5757 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) |
5758 | memcg->cgwb_frn[i].done = | |
5759 | __WB_COMPLETION_INIT(&memcg_cgwb_frn_waitq); | |
87eaceb3 YS |
5760 | #endif |
5761 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5762 | spin_lock_init(&memcg->deferred_split_queue.split_queue_lock); | |
5763 | INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue); | |
5764 | memcg->deferred_split_queue.split_queue_len = 0; | |
52ebea74 | 5765 | #endif |
ec1c86b2 | 5766 | lru_gen_init_memcg(memcg); |
0b8f73e1 JW |
5767 | return memcg; |
5768 | fail: | |
7e97de0b | 5769 | mem_cgroup_id_remove(memcg); |
40e952f9 | 5770 | __mem_cgroup_free(memcg); |
11d67612 | 5771 | return ERR_PTR(error); |
d142e3e6 GC |
5772 | } |
5773 | ||
0b8f73e1 JW |
5774 | static struct cgroup_subsys_state * __ref |
5775 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 5776 | { |
0b8f73e1 | 5777 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
b87d8cef | 5778 | struct mem_cgroup *memcg, *old_memcg; |
d142e3e6 | 5779 | |
b87d8cef | 5780 | old_memcg = set_active_memcg(parent); |
9cee7e8e | 5781 | memcg = mem_cgroup_alloc(parent); |
b87d8cef | 5782 | set_active_memcg(old_memcg); |
11d67612 YS |
5783 | if (IS_ERR(memcg)) |
5784 | return ERR_CAST(memcg); | |
d142e3e6 | 5785 | |
d1663a90 | 5786 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
2178e20c | 5787 | WRITE_ONCE(memcg->soft_limit, PAGE_COUNTER_MAX); |
f4840ccf JW |
5788 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) |
5789 | memcg->zswap_max = PAGE_COUNTER_MAX; | |
501a06fe NP |
5790 | WRITE_ONCE(memcg->zswap_writeback, |
5791 | !parent || READ_ONCE(parent->zswap_writeback)); | |
f4840ccf | 5792 | #endif |
4b82ab4f | 5793 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
0b8f73e1 | 5794 | if (parent) { |
82b3aa26 | 5795 | WRITE_ONCE(memcg->swappiness, mem_cgroup_swappiness(parent)); |
17c56de6 | 5796 | WRITE_ONCE(memcg->oom_kill_disable, READ_ONCE(parent->oom_kill_disable)); |
bef8620c | 5797 | |
3e32cb2e | 5798 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 5799 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e | 5800 | page_counter_init(&memcg->kmem, &parent->kmem); |
0db15298 | 5801 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 5802 | } else { |
ff48c71c | 5803 | init_memcg_stats(); |
8278f1c7 | 5804 | init_memcg_events(); |
bef8620c RG |
5805 | page_counter_init(&memcg->memory, NULL); |
5806 | page_counter_init(&memcg->swap, NULL); | |
5807 | page_counter_init(&memcg->kmem, NULL); | |
5808 | page_counter_init(&memcg->tcpmem, NULL); | |
d6441637 | 5809 | |
0b8f73e1 JW |
5810 | root_mem_cgroup = memcg; |
5811 | return &memcg->css; | |
5812 | } | |
5813 | ||
f7e1cb6e | 5814 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5815 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 5816 | |
b6c1a8af YS |
5817 | #if defined(CONFIG_MEMCG_KMEM) |
5818 | if (!cgroup_memory_nobpf) | |
5819 | static_branch_inc(&memcg_bpf_enabled_key); | |
5820 | #endif | |
5821 | ||
0b8f73e1 | 5822 | return &memcg->css; |
0b8f73e1 JW |
5823 | } |
5824 | ||
73f576c0 | 5825 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 5826 | { |
58fa2a55 VD |
5827 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5828 | ||
da0efe30 MS |
5829 | if (memcg_online_kmem(memcg)) |
5830 | goto remove_id; | |
5831 | ||
0a4465d3 | 5832 | /* |
e4262c4f | 5833 | * A memcg must be visible for expand_shrinker_info() |
0a4465d3 KT |
5834 | * by the time the maps are allocated. So, we allocate maps |
5835 | * here, when for_each_mem_cgroup() can't skip it. | |
5836 | */ | |
da0efe30 MS |
5837 | if (alloc_shrinker_info(memcg)) |
5838 | goto offline_kmem; | |
0a4465d3 | 5839 | |
13ef7424 | 5840 | if (unlikely(mem_cgroup_is_root(memcg)) && !mem_cgroup_disabled()) |
aa48e47e | 5841 | queue_delayed_work(system_unbound_wq, &stats_flush_dwork, |
396faf88 | 5842 | FLUSH_TIME); |
e4dde56c | 5843 | lru_gen_online_memcg(memcg); |
6f0df8e1 JW |
5844 | |
5845 | /* Online state pins memcg ID, memcg ID pins CSS */ | |
5846 | refcount_set(&memcg->id.ref, 1); | |
5847 | css_get(css); | |
5848 | ||
5849 | /* | |
5850 | * Ensure mem_cgroup_from_id() works once we're fully online. | |
5851 | * | |
5852 | * We could do this earlier and require callers to filter with | |
5853 | * css_tryget_online(). But right now there are no users that | |
5854 | * need earlier access, and the workingset code relies on the | |
5855 | * cgroup tree linkage (mem_cgroup_get_nr_swap_pages()). So | |
5856 | * publish it here at the end of onlining. This matches the | |
5857 | * regular ID destruction during offlining. | |
5858 | */ | |
5859 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); | |
5860 | ||
2f7dd7a4 | 5861 | return 0; |
da0efe30 MS |
5862 | offline_kmem: |
5863 | memcg_offline_kmem(memcg); | |
5864 | remove_id: | |
5865 | mem_cgroup_id_remove(memcg); | |
5866 | return -ENOMEM; | |
8cdea7c0 BS |
5867 | } |
5868 | ||
eb95419b | 5869 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 5870 | { |
eb95419b | 5871 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5872 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
5873 | |
5874 | /* | |
5875 | * Unregister events and notify userspace. | |
5876 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
5877 | * directory to avoid race between userspace and kernelspace. | |
5878 | */ | |
4ba9515d | 5879 | spin_lock_irq(&memcg->event_list_lock); |
fba94807 | 5880 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { |
79bd9814 TH |
5881 | list_del_init(&event->list); |
5882 | schedule_work(&event->remove); | |
5883 | } | |
4ba9515d | 5884 | spin_unlock_irq(&memcg->event_list_lock); |
ec64f515 | 5885 | |
bf8d5d52 | 5886 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5887 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 5888 | |
a65b0e76 DC |
5889 | zswap_memcg_offline_cleanup(memcg); |
5890 | ||
567e9ab2 | 5891 | memcg_offline_kmem(memcg); |
a178015c | 5892 | reparent_shrinker_deferred(memcg); |
52ebea74 | 5893 | wb_memcg_offline(memcg); |
e4dde56c | 5894 | lru_gen_offline_memcg(memcg); |
73f576c0 | 5895 | |
591edfb1 RG |
5896 | drain_all_stock(memcg); |
5897 | ||
73f576c0 | 5898 | mem_cgroup_id_put(memcg); |
df878fb0 KH |
5899 | } |
5900 | ||
6df38689 VD |
5901 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
5902 | { | |
5903 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5904 | ||
5905 | invalidate_reclaim_iterators(memcg); | |
e4dde56c | 5906 | lru_gen_release_memcg(memcg); |
6df38689 VD |
5907 | } |
5908 | ||
eb95419b | 5909 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 5910 | { |
eb95419b | 5911 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
97b27821 | 5912 | int __maybe_unused i; |
c268e994 | 5913 | |
97b27821 TH |
5914 | #ifdef CONFIG_CGROUP_WRITEBACK |
5915 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) | |
5916 | wb_wait_for_completion(&memcg->cgwb_frn[i].done); | |
5917 | #endif | |
f7e1cb6e | 5918 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5919 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 5920 | |
0db15298 | 5921 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 5922 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 5923 | |
b6c1a8af YS |
5924 | #if defined(CONFIG_MEMCG_KMEM) |
5925 | if (!cgroup_memory_nobpf) | |
5926 | static_branch_dec(&memcg_bpf_enabled_key); | |
5927 | #endif | |
5928 | ||
0b8f73e1 JW |
5929 | vmpressure_cleanup(&memcg->vmpressure); |
5930 | cancel_work_sync(&memcg->high_work); | |
5931 | mem_cgroup_remove_from_trees(memcg); | |
e4262c4f | 5932 | free_shrinker_info(memcg); |
0b8f73e1 | 5933 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
5934 | } |
5935 | ||
1ced953b TH |
5936 | /** |
5937 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
5938 | * @css: the target css | |
5939 | * | |
5940 | * Reset the states of the mem_cgroup associated with @css. This is | |
5941 | * invoked when the userland requests disabling on the default hierarchy | |
5942 | * but the memcg is pinned through dependency. The memcg should stop | |
5943 | * applying policies and should revert to the vanilla state as it may be | |
5944 | * made visible again. | |
5945 | * | |
5946 | * The current implementation only resets the essential configurations. | |
5947 | * This needs to be expanded to cover all the visible parts. | |
5948 | */ | |
5949 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
5950 | { | |
5951 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5952 | ||
bbec2e15 RG |
5953 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
5954 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
bbec2e15 RG |
5955 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); |
5956 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 5957 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5958 | page_counter_set_low(&memcg->memory, 0); |
d1663a90 | 5959 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
2178e20c | 5960 | WRITE_ONCE(memcg->soft_limit, PAGE_COUNTER_MAX); |
4b82ab4f | 5961 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
2529bb3a | 5962 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
5963 | } |
5964 | ||
2d146aa3 JW |
5965 | static void mem_cgroup_css_rstat_flush(struct cgroup_subsys_state *css, int cpu) |
5966 | { | |
5967 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5968 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); | |
5969 | struct memcg_vmstats_percpu *statc; | |
f82e6bf9 | 5970 | long delta, delta_cpu, v; |
7e1c0d6f | 5971 | int i, nid; |
2d146aa3 JW |
5972 | |
5973 | statc = per_cpu_ptr(memcg->vmstats_percpu, cpu); | |
5974 | ||
ff48c71c | 5975 | for (i = 0; i < MEMCG_VMSTAT_SIZE; i++) { |
2d146aa3 JW |
5976 | /* |
5977 | * Collect the aggregated propagation counts of groups | |
5978 | * below us. We're in a per-cpu loop here and this is | |
5979 | * a global counter, so the first cycle will get them. | |
5980 | */ | |
410f8e82 | 5981 | delta = memcg->vmstats->state_pending[i]; |
2d146aa3 | 5982 | if (delta) |
410f8e82 | 5983 | memcg->vmstats->state_pending[i] = 0; |
2d146aa3 JW |
5984 | |
5985 | /* Add CPU changes on this level since the last flush */ | |
f82e6bf9 | 5986 | delta_cpu = 0; |
2d146aa3 JW |
5987 | v = READ_ONCE(statc->state[i]); |
5988 | if (v != statc->state_prev[i]) { | |
f82e6bf9 YA |
5989 | delta_cpu = v - statc->state_prev[i]; |
5990 | delta += delta_cpu; | |
2d146aa3 JW |
5991 | statc->state_prev[i] = v; |
5992 | } | |
5993 | ||
2d146aa3 | 5994 | /* Aggregate counts on this level and propagate upwards */ |
f82e6bf9 YA |
5995 | if (delta_cpu) |
5996 | memcg->vmstats->state_local[i] += delta_cpu; | |
5997 | ||
5998 | if (delta) { | |
5999 | memcg->vmstats->state[i] += delta; | |
6000 | if (parent) | |
6001 | parent->vmstats->state_pending[i] += delta; | |
6002 | } | |
2d146aa3 JW |
6003 | } |
6004 | ||
8278f1c7 | 6005 | for (i = 0; i < NR_MEMCG_EVENTS; i++) { |
410f8e82 | 6006 | delta = memcg->vmstats->events_pending[i]; |
2d146aa3 | 6007 | if (delta) |
410f8e82 | 6008 | memcg->vmstats->events_pending[i] = 0; |
2d146aa3 | 6009 | |
f82e6bf9 | 6010 | delta_cpu = 0; |
2d146aa3 JW |
6011 | v = READ_ONCE(statc->events[i]); |
6012 | if (v != statc->events_prev[i]) { | |
f82e6bf9 YA |
6013 | delta_cpu = v - statc->events_prev[i]; |
6014 | delta += delta_cpu; | |
2d146aa3 JW |
6015 | statc->events_prev[i] = v; |
6016 | } | |
6017 | ||
f82e6bf9 YA |
6018 | if (delta_cpu) |
6019 | memcg->vmstats->events_local[i] += delta_cpu; | |
2d146aa3 | 6020 | |
f82e6bf9 YA |
6021 | if (delta) { |
6022 | memcg->vmstats->events[i] += delta; | |
6023 | if (parent) | |
6024 | parent->vmstats->events_pending[i] += delta; | |
6025 | } | |
2d146aa3 | 6026 | } |
7e1c0d6f SB |
6027 | |
6028 | for_each_node_state(nid, N_MEMORY) { | |
6029 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[nid]; | |
70a64b79 SB |
6030 | struct lruvec_stats *lstats = pn->lruvec_stats; |
6031 | struct lruvec_stats *plstats = NULL; | |
7e1c0d6f SB |
6032 | struct lruvec_stats_percpu *lstatc; |
6033 | ||
6034 | if (parent) | |
70a64b79 | 6035 | plstats = parent->nodeinfo[nid]->lruvec_stats; |
7e1c0d6f SB |
6036 | |
6037 | lstatc = per_cpu_ptr(pn->lruvec_stats_percpu, cpu); | |
6038 | ||
ff48c71c | 6039 | for (i = 0; i < NR_MEMCG_NODE_STAT_ITEMS; i++) { |
70a64b79 | 6040 | delta = lstats->state_pending[i]; |
7e1c0d6f | 6041 | if (delta) |
70a64b79 | 6042 | lstats->state_pending[i] = 0; |
7e1c0d6f | 6043 | |
f82e6bf9 | 6044 | delta_cpu = 0; |
7e1c0d6f SB |
6045 | v = READ_ONCE(lstatc->state[i]); |
6046 | if (v != lstatc->state_prev[i]) { | |
f82e6bf9 YA |
6047 | delta_cpu = v - lstatc->state_prev[i]; |
6048 | delta += delta_cpu; | |
7e1c0d6f SB |
6049 | lstatc->state_prev[i] = v; |
6050 | } | |
6051 | ||
f82e6bf9 | 6052 | if (delta_cpu) |
70a64b79 | 6053 | lstats->state_local[i] += delta_cpu; |
7e1c0d6f | 6054 | |
f82e6bf9 | 6055 | if (delta) { |
70a64b79 SB |
6056 | lstats->state[i] += delta; |
6057 | if (plstats) | |
6058 | plstats->state_pending[i] += delta; | |
f82e6bf9 | 6059 | } |
7e1c0d6f SB |
6060 | } |
6061 | } | |
78ec6f9d | 6062 | WRITE_ONCE(statc->stats_updates, 0); |
8d59d221 YA |
6063 | /* We are in a per-cpu loop here, only do the atomic write once */ |
6064 | if (atomic64_read(&memcg->vmstats->stats_updates)) | |
6065 | atomic64_set(&memcg->vmstats->stats_updates, 0); | |
2d146aa3 JW |
6066 | } |
6067 | ||
02491447 | 6068 | #ifdef CONFIG_MMU |
7dc74be0 | 6069 | /* Handlers for move charge at task migration. */ |
854ffa8d | 6070 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 6071 | { |
05b84301 | 6072 | int ret; |
9476db97 | 6073 | |
d0164adc MG |
6074 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
6075 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 6076 | if (!ret) { |
854ffa8d | 6077 | mc.precharge += count; |
854ffa8d DN |
6078 | return ret; |
6079 | } | |
9476db97 | 6080 | |
3674534b | 6081 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 6082 | while (count--) { |
3674534b | 6083 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 6084 | if (ret) |
38c5d72f | 6085 | return ret; |
854ffa8d | 6086 | mc.precharge++; |
9476db97 | 6087 | cond_resched(); |
854ffa8d | 6088 | } |
9476db97 | 6089 | return 0; |
4ffef5fe DN |
6090 | } |
6091 | ||
4ffef5fe | 6092 | union mc_target { |
b46777da | 6093 | struct folio *folio; |
02491447 | 6094 | swp_entry_t ent; |
4ffef5fe DN |
6095 | }; |
6096 | ||
4ffef5fe | 6097 | enum mc_target_type { |
8d32ff84 | 6098 | MC_TARGET_NONE = 0, |
4ffef5fe | 6099 | MC_TARGET_PAGE, |
02491447 | 6100 | MC_TARGET_SWAP, |
c733a828 | 6101 | MC_TARGET_DEVICE, |
4ffef5fe DN |
6102 | }; |
6103 | ||
90254a65 DN |
6104 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
6105 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 6106 | { |
25b2995a | 6107 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 6108 | |
58f341f7 | 6109 | if (!page) |
90254a65 DN |
6110 | return NULL; |
6111 | if (PageAnon(page)) { | |
1dfab5ab | 6112 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 6113 | return NULL; |
1dfab5ab JW |
6114 | } else { |
6115 | if (!(mc.flags & MOVE_FILE)) | |
6116 | return NULL; | |
6117 | } | |
58f341f7 | 6118 | get_page(page); |
90254a65 DN |
6119 | |
6120 | return page; | |
6121 | } | |
6122 | ||
c733a828 | 6123 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 6124 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 6125 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 6126 | { |
90254a65 DN |
6127 | struct page *page = NULL; |
6128 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
6129 | ||
9a137153 | 6130 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 6131 | return NULL; |
c733a828 JG |
6132 | |
6133 | /* | |
27674ef6 CH |
6134 | * Handle device private pages that are not accessible by the CPU, but |
6135 | * stored as special swap entries in the page table. | |
c733a828 JG |
6136 | */ |
6137 | if (is_device_private_entry(ent)) { | |
af5cdaf8 | 6138 | page = pfn_swap_entry_to_page(ent); |
27674ef6 | 6139 | if (!get_page_unless_zero(page)) |
c733a828 JG |
6140 | return NULL; |
6141 | return page; | |
6142 | } | |
6143 | ||
9a137153 RC |
6144 | if (non_swap_entry(ent)) |
6145 | return NULL; | |
6146 | ||
4b91355e | 6147 | /* |
cb691e2f | 6148 | * Because swap_cache_get_folio() updates some statistics counter, |
4b91355e KH |
6149 | * we call find_get_page() with swapper_space directly. |
6150 | */ | |
f6ab1f7f | 6151 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
2d1c4980 | 6152 | entry->val = ent.val; |
90254a65 DN |
6153 | |
6154 | return page; | |
6155 | } | |
4b91355e KH |
6156 | #else |
6157 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 6158 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
6159 | { |
6160 | return NULL; | |
6161 | } | |
6162 | #endif | |
90254a65 | 6163 | |
87946a72 | 6164 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
48384b0b | 6165 | unsigned long addr, pte_t ptent) |
87946a72 | 6166 | { |
524984ff MWO |
6167 | unsigned long index; |
6168 | struct folio *folio; | |
6169 | ||
87946a72 DN |
6170 | if (!vma->vm_file) /* anonymous vma */ |
6171 | return NULL; | |
1dfab5ab | 6172 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
6173 | return NULL; |
6174 | ||
524984ff | 6175 | /* folio is moved even if it's not RSS of this task(page-faulted). */ |
aa3b1895 | 6176 | /* shmem/tmpfs may report page out on swap: account for that too. */ |
524984ff MWO |
6177 | index = linear_page_index(vma, addr); |
6178 | folio = filemap_get_incore_folio(vma->vm_file->f_mapping, index); | |
66dabbb6 | 6179 | if (IS_ERR(folio)) |
524984ff MWO |
6180 | return NULL; |
6181 | return folio_file_page(folio, index); | |
87946a72 DN |
6182 | } |
6183 | ||
b1b0deab | 6184 | /** |
b267e1a3 MWO |
6185 | * mem_cgroup_move_account - move account of the folio |
6186 | * @folio: The folio. | |
25843c2b | 6187 | * @compound: charge the page as compound or small page |
b267e1a3 MWO |
6188 | * @from: mem_cgroup which the folio is moved from. |
6189 | * @to: mem_cgroup which the folio is moved to. @from != @to. | |
b1b0deab | 6190 | * |
b267e1a3 | 6191 | * The folio must be locked and not on the LRU. |
b1b0deab CG |
6192 | * |
6193 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
6194 | * from old cgroup. | |
6195 | */ | |
b267e1a3 | 6196 | static int mem_cgroup_move_account(struct folio *folio, |
f627c2f5 | 6197 | bool compound, |
b1b0deab CG |
6198 | struct mem_cgroup *from, |
6199 | struct mem_cgroup *to) | |
6200 | { | |
ae8af438 KK |
6201 | struct lruvec *from_vec, *to_vec; |
6202 | struct pglist_data *pgdat; | |
fcce4672 | 6203 | unsigned int nr_pages = compound ? folio_nr_pages(folio) : 1; |
8e88bd2d | 6204 | int nid, ret; |
b1b0deab CG |
6205 | |
6206 | VM_BUG_ON(from == to); | |
4e0cf05f | 6207 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
fcce4672 | 6208 | VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); |
9c325215 | 6209 | VM_BUG_ON(compound && !folio_test_large(folio)); |
b1b0deab | 6210 | |
b1b0deab | 6211 | ret = -EINVAL; |
fcce4672 | 6212 | if (folio_memcg(folio) != from) |
4e0cf05f | 6213 | goto out; |
b1b0deab | 6214 | |
fcce4672 | 6215 | pgdat = folio_pgdat(folio); |
867e5e1d JW |
6216 | from_vec = mem_cgroup_lruvec(from, pgdat); |
6217 | to_vec = mem_cgroup_lruvec(to, pgdat); | |
ae8af438 | 6218 | |
fcce4672 | 6219 | folio_memcg_lock(folio); |
b1b0deab | 6220 | |
fcce4672 MWO |
6221 | if (folio_test_anon(folio)) { |
6222 | if (folio_mapped(folio)) { | |
be5d0a74 JW |
6223 | __mod_lruvec_state(from_vec, NR_ANON_MAPPED, -nr_pages); |
6224 | __mod_lruvec_state(to_vec, NR_ANON_MAPPED, nr_pages); | |
6199277b | 6225 | if (folio_test_pmd_mappable(folio)) { |
69473e5d MS |
6226 | __mod_lruvec_state(from_vec, NR_ANON_THPS, |
6227 | -nr_pages); | |
6228 | __mod_lruvec_state(to_vec, NR_ANON_THPS, | |
6229 | nr_pages); | |
468c3982 | 6230 | } |
be5d0a74 JW |
6231 | } |
6232 | } else { | |
0d1c2072 JW |
6233 | __mod_lruvec_state(from_vec, NR_FILE_PAGES, -nr_pages); |
6234 | __mod_lruvec_state(to_vec, NR_FILE_PAGES, nr_pages); | |
6235 | ||
fcce4672 | 6236 | if (folio_test_swapbacked(folio)) { |
0d1c2072 JW |
6237 | __mod_lruvec_state(from_vec, NR_SHMEM, -nr_pages); |
6238 | __mod_lruvec_state(to_vec, NR_SHMEM, nr_pages); | |
6239 | } | |
6240 | ||
fcce4672 | 6241 | if (folio_mapped(folio)) { |
49e50d27 JW |
6242 | __mod_lruvec_state(from_vec, NR_FILE_MAPPED, -nr_pages); |
6243 | __mod_lruvec_state(to_vec, NR_FILE_MAPPED, nr_pages); | |
6244 | } | |
b1b0deab | 6245 | |
fcce4672 MWO |
6246 | if (folio_test_dirty(folio)) { |
6247 | struct address_space *mapping = folio_mapping(folio); | |
c4843a75 | 6248 | |
f56753ac | 6249 | if (mapping_can_writeback(mapping)) { |
49e50d27 JW |
6250 | __mod_lruvec_state(from_vec, NR_FILE_DIRTY, |
6251 | -nr_pages); | |
6252 | __mod_lruvec_state(to_vec, NR_FILE_DIRTY, | |
6253 | nr_pages); | |
6254 | } | |
c4843a75 GT |
6255 | } |
6256 | } | |
6257 | ||
c449deb2 HD |
6258 | #ifdef CONFIG_SWAP |
6259 | if (folio_test_swapcache(folio)) { | |
6260 | __mod_lruvec_state(from_vec, NR_SWAPCACHE, -nr_pages); | |
6261 | __mod_lruvec_state(to_vec, NR_SWAPCACHE, nr_pages); | |
6262 | } | |
6263 | #endif | |
fcce4672 | 6264 | if (folio_test_writeback(folio)) { |
ae8af438 KK |
6265 | __mod_lruvec_state(from_vec, NR_WRITEBACK, -nr_pages); |
6266 | __mod_lruvec_state(to_vec, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
6267 | } |
6268 | ||
6269 | /* | |
abb242f5 JW |
6270 | * All state has been migrated, let's switch to the new memcg. |
6271 | * | |
bcfe06bf | 6272 | * It is safe to change page's memcg here because the page |
abb242f5 JW |
6273 | * is referenced, charged, isolated, and locked: we can't race |
6274 | * with (un)charging, migration, LRU putback, or anything else | |
bcfe06bf | 6275 | * that would rely on a stable page's memory cgroup. |
abb242f5 | 6276 | * |
6c77b607 | 6277 | * Note that folio_memcg_lock is a memcg lock, not a page lock, |
bcfe06bf | 6278 | * to save space. As soon as we switch page's memory cgroup to a |
abb242f5 JW |
6279 | * new memcg that isn't locked, the above state can change |
6280 | * concurrently again. Make sure we're truly done with it. | |
b1b0deab | 6281 | */ |
abb242f5 | 6282 | smp_mb(); |
b1b0deab | 6283 | |
1a3e1f40 JW |
6284 | css_get(&to->css); |
6285 | css_put(&from->css); | |
6286 | ||
fcce4672 | 6287 | folio->memcg_data = (unsigned long)to; |
87eaceb3 | 6288 | |
f70ad448 | 6289 | __folio_memcg_unlock(from); |
b1b0deab CG |
6290 | |
6291 | ret = 0; | |
fcce4672 | 6292 | nid = folio_nid(folio); |
b1b0deab CG |
6293 | |
6294 | local_irq_disable(); | |
6e0110c2 | 6295 | mem_cgroup_charge_statistics(to, nr_pages); |
8e88bd2d | 6296 | memcg_check_events(to, nid); |
6e0110c2 | 6297 | mem_cgroup_charge_statistics(from, -nr_pages); |
8e88bd2d | 6298 | memcg_check_events(from, nid); |
b1b0deab | 6299 | local_irq_enable(); |
b1b0deab CG |
6300 | out: |
6301 | return ret; | |
6302 | } | |
6303 | ||
7cf7806c LR |
6304 | /** |
6305 | * get_mctgt_type - get target type of moving charge | |
6306 | * @vma: the vma the pte to be checked belongs | |
6307 | * @addr: the address corresponding to the pte to be checked | |
6308 | * @ptent: the pte to be checked | |
6309 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
6310 | * | |
853f62a3 MWO |
6311 | * Context: Called with pte lock held. |
6312 | * Return: | |
6313 | * * MC_TARGET_NONE - If the pte is not a target for move charge. | |
6314 | * * MC_TARGET_PAGE - If the page corresponding to this pte is a target for | |
b46777da | 6315 | * move charge. If @target is not NULL, the folio is stored in target->folio |
853f62a3 MWO |
6316 | * with extra refcnt taken (Caller should release it). |
6317 | * * MC_TARGET_SWAP - If the swap entry corresponding to this pte is a | |
6318 | * target for charge migration. If @target is not NULL, the entry is | |
6319 | * stored in target->ent. | |
6320 | * * MC_TARGET_DEVICE - Like MC_TARGET_PAGE but page is device memory and | |
6321 | * thus not on the lru. For now such page is charged like a regular page | |
6322 | * would be as it is just special memory taking the place of a regular page. | |
6323 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c | 6324 | */ |
8d32ff84 | 6325 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
6326 | unsigned long addr, pte_t ptent, union mc_target *target) |
6327 | { | |
6328 | struct page *page = NULL; | |
b67fa6e4 | 6329 | struct folio *folio; |
8d32ff84 | 6330 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
6331 | swp_entry_t ent = { .val = 0 }; |
6332 | ||
6333 | if (pte_present(ptent)) | |
6334 | page = mc_handle_present_pte(vma, addr, ptent); | |
5c041f5d PX |
6335 | else if (pte_none_mostly(ptent)) |
6336 | /* | |
6337 | * PTE markers should be treated as a none pte here, separated | |
6338 | * from other swap handling below. | |
6339 | */ | |
6340 | page = mc_handle_file_pte(vma, addr, ptent); | |
90254a65 | 6341 | else if (is_swap_pte(ptent)) |
48406ef8 | 6342 | page = mc_handle_swap_pte(vma, ptent, &ent); |
90254a65 | 6343 | |
b67fa6e4 MWO |
6344 | if (page) |
6345 | folio = page_folio(page); | |
4e0cf05f | 6346 | if (target && page) { |
b67fa6e4 MWO |
6347 | if (!folio_trylock(folio)) { |
6348 | folio_put(folio); | |
4e0cf05f JW |
6349 | return ret; |
6350 | } | |
6351 | /* | |
6352 | * page_mapped() must be stable during the move. This | |
6353 | * pte is locked, so if it's present, the page cannot | |
6354 | * become unmapped. If it isn't, we have only partial | |
6355 | * control over the mapped state: the page lock will | |
6356 | * prevent new faults against pagecache and swapcache, | |
6357 | * so an unmapped page cannot become mapped. However, | |
6358 | * if the page is already mapped elsewhere, it can | |
6359 | * unmap, and there is nothing we can do about it. | |
6360 | * Alas, skip moving the page in this case. | |
6361 | */ | |
6362 | if (!pte_present(ptent) && page_mapped(page)) { | |
b67fa6e4 MWO |
6363 | folio_unlock(folio); |
6364 | folio_put(folio); | |
4e0cf05f JW |
6365 | return ret; |
6366 | } | |
6367 | } | |
6368 | ||
90254a65 | 6369 | if (!page && !ent.val) |
8d32ff84 | 6370 | return ret; |
02491447 | 6371 | if (page) { |
02491447 | 6372 | /* |
0a31bc97 | 6373 | * Do only loose check w/o serialization. |
1306a85a | 6374 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 6375 | * not under LRU exclusion. |
02491447 | 6376 | */ |
b67fa6e4 | 6377 | if (folio_memcg(folio) == mc.from) { |
02491447 | 6378 | ret = MC_TARGET_PAGE; |
b67fa6e4 MWO |
6379 | if (folio_is_device_private(folio) || |
6380 | folio_is_device_coherent(folio)) | |
c733a828 | 6381 | ret = MC_TARGET_DEVICE; |
02491447 | 6382 | if (target) |
b67fa6e4 | 6383 | target->folio = folio; |
02491447 | 6384 | } |
4e0cf05f JW |
6385 | if (!ret || !target) { |
6386 | if (target) | |
b67fa6e4 MWO |
6387 | folio_unlock(folio); |
6388 | folio_put(folio); | |
4e0cf05f | 6389 | } |
02491447 | 6390 | } |
3e14a57b HY |
6391 | /* |
6392 | * There is a swap entry and a page doesn't exist or isn't charged. | |
6393 | * But we cannot move a tail-page in a THP. | |
6394 | */ | |
6395 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 6396 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
6397 | ret = MC_TARGET_SWAP; |
6398 | if (target) | |
6399 | target->ent = ent; | |
4ffef5fe | 6400 | } |
4ffef5fe DN |
6401 | return ret; |
6402 | } | |
6403 | ||
12724850 NH |
6404 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
6405 | /* | |
d6810d73 HY |
6406 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
6407 | * not support them for now. | |
12724850 NH |
6408 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
6409 | */ | |
6410 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6411 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6412 | { | |
6413 | struct page *page = NULL; | |
f6c7590b | 6414 | struct folio *folio; |
12724850 NH |
6415 | enum mc_target_type ret = MC_TARGET_NONE; |
6416 | ||
84c3fc4e ZY |
6417 | if (unlikely(is_swap_pmd(pmd))) { |
6418 | VM_BUG_ON(thp_migration_supported() && | |
6419 | !is_pmd_migration_entry(pmd)); | |
6420 | return ret; | |
6421 | } | |
12724850 | 6422 | page = pmd_page(pmd); |
309381fe | 6423 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
f6c7590b | 6424 | folio = page_folio(page); |
1dfab5ab | 6425 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 6426 | return ret; |
f6c7590b | 6427 | if (folio_memcg(folio) == mc.from) { |
12724850 NH |
6428 | ret = MC_TARGET_PAGE; |
6429 | if (target) { | |
f6c7590b MWO |
6430 | folio_get(folio); |
6431 | if (!folio_trylock(folio)) { | |
6432 | folio_put(folio); | |
4e0cf05f JW |
6433 | return MC_TARGET_NONE; |
6434 | } | |
f6c7590b | 6435 | target->folio = folio; |
12724850 NH |
6436 | } |
6437 | } | |
6438 | return ret; | |
6439 | } | |
6440 | #else | |
6441 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6442 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6443 | { | |
6444 | return MC_TARGET_NONE; | |
6445 | } | |
6446 | #endif | |
6447 | ||
4ffef5fe DN |
6448 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
6449 | unsigned long addr, unsigned long end, | |
6450 | struct mm_walk *walk) | |
6451 | { | |
26bcd64a | 6452 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
6453 | pte_t *pte; |
6454 | spinlock_t *ptl; | |
6455 | ||
b6ec57f4 KS |
6456 | ptl = pmd_trans_huge_lock(pmd, vma); |
6457 | if (ptl) { | |
c733a828 JG |
6458 | /* |
6459 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
25b2995a CH |
6460 | * support transparent huge page with MEMORY_DEVICE_PRIVATE but |
6461 | * this might change. | |
c733a828 | 6462 | */ |
12724850 NH |
6463 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
6464 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 6465 | spin_unlock(ptl); |
1a5a9906 | 6466 | return 0; |
12724850 | 6467 | } |
03319327 | 6468 | |
4ffef5fe | 6469 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
04dee9e8 HD |
6470 | if (!pte) |
6471 | return 0; | |
4ffef5fe | 6472 | for (; addr != end; pte++, addr += PAGE_SIZE) |
c33c7948 | 6473 | if (get_mctgt_type(vma, addr, ptep_get(pte), NULL)) |
4ffef5fe DN |
6474 | mc.precharge++; /* increment precharge temporarily */ |
6475 | pte_unmap_unlock(pte - 1, ptl); | |
6476 | cond_resched(); | |
6477 | ||
7dc74be0 DN |
6478 | return 0; |
6479 | } | |
6480 | ||
7b86ac33 CH |
6481 | static const struct mm_walk_ops precharge_walk_ops = { |
6482 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
49b06385 | 6483 | .walk_lock = PGWALK_RDLOCK, |
7b86ac33 CH |
6484 | }; |
6485 | ||
4ffef5fe DN |
6486 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
6487 | { | |
6488 | unsigned long precharge; | |
4ffef5fe | 6489 | |
d8ed45c5 | 6490 | mmap_read_lock(mm); |
ba0aff8e | 6491 | walk_page_range(mm, 0, ULONG_MAX, &precharge_walk_ops, NULL); |
d8ed45c5 | 6492 | mmap_read_unlock(mm); |
4ffef5fe DN |
6493 | |
6494 | precharge = mc.precharge; | |
6495 | mc.precharge = 0; | |
6496 | ||
6497 | return precharge; | |
6498 | } | |
6499 | ||
4ffef5fe DN |
6500 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
6501 | { | |
dfe076b0 DN |
6502 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
6503 | ||
6504 | VM_BUG_ON(mc.moving_task); | |
6505 | mc.moving_task = current; | |
6506 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
6507 | } |
6508 | ||
dfe076b0 DN |
6509 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
6510 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 6511 | { |
2bd9bb20 KH |
6512 | struct mem_cgroup *from = mc.from; |
6513 | struct mem_cgroup *to = mc.to; | |
6514 | ||
4ffef5fe | 6515 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 6516 | if (mc.precharge) { |
4b569387 | 6517 | mem_cgroup_cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
6518 | mc.precharge = 0; |
6519 | } | |
6520 | /* | |
6521 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
6522 | * we must uncharge here. | |
6523 | */ | |
6524 | if (mc.moved_charge) { | |
4b569387 | 6525 | mem_cgroup_cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 6526 | mc.moved_charge = 0; |
4ffef5fe | 6527 | } |
483c30b5 DN |
6528 | /* we must fixup refcnts and charges */ |
6529 | if (mc.moved_swap) { | |
483c30b5 | 6530 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 6531 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 6532 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 6533 | |
615d66c3 VD |
6534 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
6535 | ||
05b84301 | 6536 | /* |
3e32cb2e JW |
6537 | * we charged both to->memory and to->memsw, so we |
6538 | * should uncharge to->memory. | |
05b84301 | 6539 | */ |
ce00a967 | 6540 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
6541 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
6542 | ||
483c30b5 DN |
6543 | mc.moved_swap = 0; |
6544 | } | |
dfe076b0 DN |
6545 | memcg_oom_recover(from); |
6546 | memcg_oom_recover(to); | |
6547 | wake_up_all(&mc.waitq); | |
6548 | } | |
6549 | ||
6550 | static void mem_cgroup_clear_mc(void) | |
6551 | { | |
264a0ae1 TH |
6552 | struct mm_struct *mm = mc.mm; |
6553 | ||
dfe076b0 DN |
6554 | /* |
6555 | * we must clear moving_task before waking up waiters at the end of | |
6556 | * task migration. | |
6557 | */ | |
6558 | mc.moving_task = NULL; | |
6559 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 6560 | spin_lock(&mc.lock); |
4ffef5fe DN |
6561 | mc.from = NULL; |
6562 | mc.to = NULL; | |
264a0ae1 | 6563 | mc.mm = NULL; |
2bd9bb20 | 6564 | spin_unlock(&mc.lock); |
264a0ae1 TH |
6565 | |
6566 | mmput(mm); | |
4ffef5fe DN |
6567 | } |
6568 | ||
1f7dd3e5 | 6569 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 6570 | { |
1f7dd3e5 | 6571 | struct cgroup_subsys_state *css; |
eed67d75 | 6572 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 6573 | struct mem_cgroup *from; |
4530eddb | 6574 | struct task_struct *leader, *p; |
9f2115f9 | 6575 | struct mm_struct *mm; |
1dfab5ab | 6576 | unsigned long move_flags; |
9f2115f9 | 6577 | int ret = 0; |
7dc74be0 | 6578 | |
1f7dd3e5 TH |
6579 | /* charge immigration isn't supported on the default hierarchy */ |
6580 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
6581 | return 0; |
6582 | ||
4530eddb TH |
6583 | /* |
6584 | * Multi-process migrations only happen on the default hierarchy | |
6585 | * where charge immigration is not used. Perform charge | |
6586 | * immigration if @tset contains a leader and whine if there are | |
6587 | * multiple. | |
6588 | */ | |
6589 | p = NULL; | |
1f7dd3e5 | 6590 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
6591 | WARN_ON_ONCE(p); |
6592 | p = leader; | |
1f7dd3e5 | 6593 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
6594 | } |
6595 | if (!p) | |
6596 | return 0; | |
6597 | ||
1f7dd3e5 | 6598 | /* |
f0953a1b | 6599 | * We are now committed to this value whatever it is. Changes in this |
1f7dd3e5 TH |
6600 | * tunable will only affect upcoming migrations, not the current one. |
6601 | * So we need to save it, and keep it going. | |
6602 | */ | |
6603 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
6604 | if (!move_flags) | |
6605 | return 0; | |
6606 | ||
9f2115f9 TH |
6607 | from = mem_cgroup_from_task(p); |
6608 | ||
6609 | VM_BUG_ON(from == memcg); | |
6610 | ||
6611 | mm = get_task_mm(p); | |
6612 | if (!mm) | |
6613 | return 0; | |
6614 | /* We move charges only when we move a owner of the mm */ | |
6615 | if (mm->owner == p) { | |
6616 | VM_BUG_ON(mc.from); | |
6617 | VM_BUG_ON(mc.to); | |
6618 | VM_BUG_ON(mc.precharge); | |
6619 | VM_BUG_ON(mc.moved_charge); | |
6620 | VM_BUG_ON(mc.moved_swap); | |
6621 | ||
6622 | spin_lock(&mc.lock); | |
264a0ae1 | 6623 | mc.mm = mm; |
9f2115f9 TH |
6624 | mc.from = from; |
6625 | mc.to = memcg; | |
6626 | mc.flags = move_flags; | |
6627 | spin_unlock(&mc.lock); | |
6628 | /* We set mc.moving_task later */ | |
6629 | ||
6630 | ret = mem_cgroup_precharge_mc(mm); | |
6631 | if (ret) | |
6632 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
6633 | } else { |
6634 | mmput(mm); | |
7dc74be0 DN |
6635 | } |
6636 | return ret; | |
6637 | } | |
6638 | ||
1f7dd3e5 | 6639 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 6640 | { |
4e2f245d JW |
6641 | if (mc.to) |
6642 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
6643 | } |
6644 | ||
4ffef5fe DN |
6645 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
6646 | unsigned long addr, unsigned long end, | |
6647 | struct mm_walk *walk) | |
7dc74be0 | 6648 | { |
4ffef5fe | 6649 | int ret = 0; |
26bcd64a | 6650 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
6651 | pte_t *pte; |
6652 | spinlock_t *ptl; | |
12724850 NH |
6653 | enum mc_target_type target_type; |
6654 | union mc_target target; | |
b267e1a3 | 6655 | struct folio *folio; |
4ffef5fe | 6656 | |
b6ec57f4 KS |
6657 | ptl = pmd_trans_huge_lock(pmd, vma); |
6658 | if (ptl) { | |
62ade86a | 6659 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 6660 | spin_unlock(ptl); |
12724850 NH |
6661 | return 0; |
6662 | } | |
6663 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
6664 | if (target_type == MC_TARGET_PAGE) { | |
b46777da | 6665 | folio = target.folio; |
b267e1a3 MWO |
6666 | if (folio_isolate_lru(folio)) { |
6667 | if (!mem_cgroup_move_account(folio, true, | |
1306a85a | 6668 | mc.from, mc.to)) { |
12724850 NH |
6669 | mc.precharge -= HPAGE_PMD_NR; |
6670 | mc.moved_charge += HPAGE_PMD_NR; | |
6671 | } | |
b267e1a3 | 6672 | folio_putback_lru(folio); |
12724850 | 6673 | } |
b267e1a3 MWO |
6674 | folio_unlock(folio); |
6675 | folio_put(folio); | |
c733a828 | 6676 | } else if (target_type == MC_TARGET_DEVICE) { |
b46777da | 6677 | folio = target.folio; |
b267e1a3 | 6678 | if (!mem_cgroup_move_account(folio, true, |
c733a828 JG |
6679 | mc.from, mc.to)) { |
6680 | mc.precharge -= HPAGE_PMD_NR; | |
6681 | mc.moved_charge += HPAGE_PMD_NR; | |
6682 | } | |
b267e1a3 MWO |
6683 | folio_unlock(folio); |
6684 | folio_put(folio); | |
12724850 | 6685 | } |
bf929152 | 6686 | spin_unlock(ptl); |
1a5a9906 | 6687 | return 0; |
12724850 NH |
6688 | } |
6689 | ||
4ffef5fe DN |
6690 | retry: |
6691 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
04dee9e8 HD |
6692 | if (!pte) |
6693 | return 0; | |
4ffef5fe | 6694 | for (; addr != end; addr += PAGE_SIZE) { |
c33c7948 | 6695 | pte_t ptent = ptep_get(pte++); |
c733a828 | 6696 | bool device = false; |
02491447 | 6697 | swp_entry_t ent; |
4ffef5fe DN |
6698 | |
6699 | if (!mc.precharge) | |
6700 | break; | |
6701 | ||
8d32ff84 | 6702 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
6703 | case MC_TARGET_DEVICE: |
6704 | device = true; | |
e4a9bc58 | 6705 | fallthrough; |
4ffef5fe | 6706 | case MC_TARGET_PAGE: |
b46777da | 6707 | folio = target.folio; |
53f9263b KS |
6708 | /* |
6709 | * We can have a part of the split pmd here. Moving it | |
6710 | * can be done but it would be too convoluted so simply | |
6711 | * ignore such a partial THP and keep it in original | |
6712 | * memcg. There should be somebody mapping the head. | |
6713 | */ | |
b267e1a3 | 6714 | if (folio_test_large(folio)) |
53f9263b | 6715 | goto put; |
b267e1a3 | 6716 | if (!device && !folio_isolate_lru(folio)) |
4ffef5fe | 6717 | goto put; |
b267e1a3 | 6718 | if (!mem_cgroup_move_account(folio, false, |
f627c2f5 | 6719 | mc.from, mc.to)) { |
4ffef5fe | 6720 | mc.precharge--; |
854ffa8d DN |
6721 | /* we uncharge from mc.from later. */ |
6722 | mc.moved_charge++; | |
4ffef5fe | 6723 | } |
c733a828 | 6724 | if (!device) |
b267e1a3 | 6725 | folio_putback_lru(folio); |
4e0cf05f | 6726 | put: /* get_mctgt_type() gets & locks the page */ |
b267e1a3 MWO |
6727 | folio_unlock(folio); |
6728 | folio_put(folio); | |
4ffef5fe | 6729 | break; |
02491447 DN |
6730 | case MC_TARGET_SWAP: |
6731 | ent = target.ent; | |
e91cbb42 | 6732 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6733 | mc.precharge--; |
8d22a935 HD |
6734 | mem_cgroup_id_get_many(mc.to, 1); |
6735 | /* we fixup other refcnts and charges later. */ | |
483c30b5 DN |
6736 | mc.moved_swap++; |
6737 | } | |
02491447 | 6738 | break; |
4ffef5fe DN |
6739 | default: |
6740 | break; | |
6741 | } | |
6742 | } | |
6743 | pte_unmap_unlock(pte - 1, ptl); | |
6744 | cond_resched(); | |
6745 | ||
6746 | if (addr != end) { | |
6747 | /* | |
6748 | * We have consumed all precharges we got in can_attach(). | |
6749 | * We try charge one by one, but don't do any additional | |
6750 | * charges to mc.to if we have failed in charge once in attach() | |
6751 | * phase. | |
6752 | */ | |
854ffa8d | 6753 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6754 | if (!ret) |
6755 | goto retry; | |
6756 | } | |
6757 | ||
6758 | return ret; | |
6759 | } | |
6760 | ||
7b86ac33 CH |
6761 | static const struct mm_walk_ops charge_walk_ops = { |
6762 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
49b06385 | 6763 | .walk_lock = PGWALK_RDLOCK, |
7b86ac33 CH |
6764 | }; |
6765 | ||
264a0ae1 | 6766 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 6767 | { |
4ffef5fe | 6768 | lru_add_drain_all(); |
312722cb | 6769 | /* |
6c77b607 | 6770 | * Signal folio_memcg_lock() to take the memcg's move_lock |
81f8c3a4 JW |
6771 | * while we're moving its pages to another memcg. Then wait |
6772 | * for already started RCU-only updates to finish. | |
312722cb JW |
6773 | */ |
6774 | atomic_inc(&mc.from->moving_account); | |
6775 | synchronize_rcu(); | |
dfe076b0 | 6776 | retry: |
d8ed45c5 | 6777 | if (unlikely(!mmap_read_trylock(mc.mm))) { |
dfe076b0 | 6778 | /* |
c1e8d7c6 | 6779 | * Someone who are holding the mmap_lock might be waiting in |
dfe076b0 DN |
6780 | * waitq. So we cancel all extra charges, wake up all waiters, |
6781 | * and retry. Because we cancel precharges, we might not be able | |
6782 | * to move enough charges, but moving charge is a best-effort | |
6783 | * feature anyway, so it wouldn't be a big problem. | |
6784 | */ | |
6785 | __mem_cgroup_clear_mc(); | |
6786 | cond_resched(); | |
6787 | goto retry; | |
6788 | } | |
26bcd64a NH |
6789 | /* |
6790 | * When we have consumed all precharges and failed in doing | |
6791 | * additional charge, the page walk just aborts. | |
6792 | */ | |
ba0aff8e | 6793 | walk_page_range(mc.mm, 0, ULONG_MAX, &charge_walk_ops, NULL); |
d8ed45c5 | 6794 | mmap_read_unlock(mc.mm); |
312722cb | 6795 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
6796 | } |
6797 | ||
264a0ae1 | 6798 | static void mem_cgroup_move_task(void) |
67e465a7 | 6799 | { |
264a0ae1 TH |
6800 | if (mc.to) { |
6801 | mem_cgroup_move_charge(); | |
a433658c | 6802 | mem_cgroup_clear_mc(); |
264a0ae1 | 6803 | } |
67e465a7 | 6804 | } |
1aacbd35 | 6805 | |
5cfb80a7 | 6806 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 6807 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6808 | { |
6809 | return 0; | |
6810 | } | |
1f7dd3e5 | 6811 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6812 | { |
6813 | } | |
264a0ae1 | 6814 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
6815 | { |
6816 | } | |
6817 | #endif | |
67e465a7 | 6818 | |
1aacbd35 RG |
6819 | #ifdef CONFIG_MEMCG_KMEM |
6820 | static void mem_cgroup_fork(struct task_struct *task) | |
6821 | { | |
6822 | /* | |
6823 | * Set the update flag to cause task->objcg to be initialized lazily | |
6824 | * on the first allocation. It can be done without any synchronization | |
6825 | * because it's always performed on the current task, so does | |
6826 | * current_objcg_update(). | |
6827 | */ | |
6828 | task->objcg = (struct obj_cgroup *)CURRENT_OBJCG_UPDATE_FLAG; | |
6829 | } | |
6830 | ||
6831 | static void mem_cgroup_exit(struct task_struct *task) | |
6832 | { | |
6833 | struct obj_cgroup *objcg = task->objcg; | |
6834 | ||
6835 | objcg = (struct obj_cgroup *) | |
6836 | ((unsigned long)objcg & ~CURRENT_OBJCG_UPDATE_FLAG); | |
91b71e78 | 6837 | obj_cgroup_put(objcg); |
1aacbd35 RG |
6838 | |
6839 | /* | |
6840 | * Some kernel allocations can happen after this point, | |
6841 | * but let's ignore them. It can be done without any synchronization | |
6842 | * because it's always performed on the current task, so does | |
6843 | * current_objcg_update(). | |
6844 | */ | |
6845 | task->objcg = NULL; | |
6846 | } | |
6847 | #endif | |
6848 | ||
bd74fdae | 6849 | #ifdef CONFIG_LRU_GEN |
1aacbd35 | 6850 | static void mem_cgroup_lru_gen_attach(struct cgroup_taskset *tset) |
bd74fdae YZ |
6851 | { |
6852 | struct task_struct *task; | |
6853 | struct cgroup_subsys_state *css; | |
6854 | ||
6855 | /* find the first leader if there is any */ | |
6856 | cgroup_taskset_for_each_leader(task, css, tset) | |
6857 | break; | |
6858 | ||
6859 | if (!task) | |
6860 | return; | |
6861 | ||
6862 | task_lock(task); | |
6863 | if (task->mm && READ_ONCE(task->mm->owner) == task) | |
6864 | lru_gen_migrate_mm(task->mm); | |
6865 | task_unlock(task); | |
6866 | } | |
6867 | #else | |
1aacbd35 RG |
6868 | static void mem_cgroup_lru_gen_attach(struct cgroup_taskset *tset) {} |
6869 | #endif /* CONFIG_LRU_GEN */ | |
6870 | ||
6871 | #ifdef CONFIG_MEMCG_KMEM | |
6872 | static void mem_cgroup_kmem_attach(struct cgroup_taskset *tset) | |
6873 | { | |
6874 | struct task_struct *task; | |
6875 | struct cgroup_subsys_state *css; | |
6876 | ||
6877 | cgroup_taskset_for_each(task, css, tset) { | |
6878 | /* atomically set the update bit */ | |
6879 | set_bit(CURRENT_OBJCG_UPDATE_BIT, (unsigned long *)&task->objcg); | |
6880 | } | |
6881 | } | |
6882 | #else | |
6883 | static void mem_cgroup_kmem_attach(struct cgroup_taskset *tset) {} | |
6884 | #endif /* CONFIG_MEMCG_KMEM */ | |
6885 | ||
6886 | #if defined(CONFIG_LRU_GEN) || defined(CONFIG_MEMCG_KMEM) | |
bd74fdae YZ |
6887 | static void mem_cgroup_attach(struct cgroup_taskset *tset) |
6888 | { | |
1aacbd35 RG |
6889 | mem_cgroup_lru_gen_attach(tset); |
6890 | mem_cgroup_kmem_attach(tset); | |
bd74fdae | 6891 | } |
1aacbd35 | 6892 | #endif |
bd74fdae | 6893 | |
677dc973 CD |
6894 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
6895 | { | |
6896 | if (value == PAGE_COUNTER_MAX) | |
6897 | seq_puts(m, "max\n"); | |
6898 | else | |
6899 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); | |
6900 | ||
6901 | return 0; | |
6902 | } | |
6903 | ||
241994ed JW |
6904 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
6905 | struct cftype *cft) | |
6906 | { | |
f5fc3c5d JW |
6907 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
6908 | ||
6909 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
6910 | } |
6911 | ||
8e20d4b3 GR |
6912 | static u64 memory_peak_read(struct cgroup_subsys_state *css, |
6913 | struct cftype *cft) | |
6914 | { | |
6915 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
6916 | ||
6917 | return (u64)memcg->memory.watermark * PAGE_SIZE; | |
6918 | } | |
6919 | ||
bf8d5d52 RG |
6920 | static int memory_min_show(struct seq_file *m, void *v) |
6921 | { | |
677dc973 CD |
6922 | return seq_puts_memcg_tunable(m, |
6923 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); | |
bf8d5d52 RG |
6924 | } |
6925 | ||
6926 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
6927 | char *buf, size_t nbytes, loff_t off) | |
6928 | { | |
6929 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6930 | unsigned long min; | |
6931 | int err; | |
6932 | ||
6933 | buf = strstrip(buf); | |
6934 | err = page_counter_memparse(buf, "max", &min); | |
6935 | if (err) | |
6936 | return err; | |
6937 | ||
6938 | page_counter_set_min(&memcg->memory, min); | |
6939 | ||
6940 | return nbytes; | |
6941 | } | |
6942 | ||
241994ed JW |
6943 | static int memory_low_show(struct seq_file *m, void *v) |
6944 | { | |
677dc973 CD |
6945 | return seq_puts_memcg_tunable(m, |
6946 | READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); | |
241994ed JW |
6947 | } |
6948 | ||
6949 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
6950 | char *buf, size_t nbytes, loff_t off) | |
6951 | { | |
6952 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6953 | unsigned long low; | |
6954 | int err; | |
6955 | ||
6956 | buf = strstrip(buf); | |
d2973697 | 6957 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
6958 | if (err) |
6959 | return err; | |
6960 | ||
23067153 | 6961 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
6962 | |
6963 | return nbytes; | |
6964 | } | |
6965 | ||
6966 | static int memory_high_show(struct seq_file *m, void *v) | |
6967 | { | |
d1663a90 JK |
6968 | return seq_puts_memcg_tunable(m, |
6969 | READ_ONCE(mem_cgroup_from_seq(m)->memory.high)); | |
241994ed JW |
6970 | } |
6971 | ||
6972 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
6973 | char *buf, size_t nbytes, loff_t off) | |
6974 | { | |
6975 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6976 | unsigned int nr_retries = MAX_RECLAIM_RETRIES; |
8c8c383c | 6977 | bool drained = false; |
241994ed JW |
6978 | unsigned long high; |
6979 | int err; | |
6980 | ||
6981 | buf = strstrip(buf); | |
d2973697 | 6982 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
6983 | if (err) |
6984 | return err; | |
6985 | ||
e82553c1 JW |
6986 | page_counter_set_high(&memcg->memory, high); |
6987 | ||
8c8c383c JW |
6988 | for (;;) { |
6989 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6990 | unsigned long reclaimed; | |
6991 | ||
6992 | if (nr_pages <= high) | |
6993 | break; | |
6994 | ||
6995 | if (signal_pending(current)) | |
6996 | break; | |
6997 | ||
6998 | if (!drained) { | |
6999 | drain_all_stock(memcg); | |
7000 | drained = true; | |
7001 | continue; | |
7002 | } | |
7003 | ||
7004 | reclaimed = try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
55ab834a | 7005 | GFP_KERNEL, MEMCG_RECLAIM_MAY_SWAP); |
8c8c383c JW |
7006 | |
7007 | if (!reclaimed && !nr_retries--) | |
7008 | break; | |
7009 | } | |
588083bb | 7010 | |
19ce33ac | 7011 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
7012 | return nbytes; |
7013 | } | |
7014 | ||
7015 | static int memory_max_show(struct seq_file *m, void *v) | |
7016 | { | |
677dc973 CD |
7017 | return seq_puts_memcg_tunable(m, |
7018 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); | |
241994ed JW |
7019 | } |
7020 | ||
7021 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
7022 | char *buf, size_t nbytes, loff_t off) | |
7023 | { | |
7024 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 7025 | unsigned int nr_reclaims = MAX_RECLAIM_RETRIES; |
b6e6edcf | 7026 | bool drained = false; |
241994ed JW |
7027 | unsigned long max; |
7028 | int err; | |
7029 | ||
7030 | buf = strstrip(buf); | |
d2973697 | 7031 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
7032 | if (err) |
7033 | return err; | |
7034 | ||
bbec2e15 | 7035 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
7036 | |
7037 | for (;;) { | |
7038 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
7039 | ||
7040 | if (nr_pages <= max) | |
7041 | break; | |
7042 | ||
7249c9f0 | 7043 | if (signal_pending(current)) |
b6e6edcf | 7044 | break; |
b6e6edcf JW |
7045 | |
7046 | if (!drained) { | |
7047 | drain_all_stock(memcg); | |
7048 | drained = true; | |
7049 | continue; | |
7050 | } | |
7051 | ||
7052 | if (nr_reclaims) { | |
7053 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
55ab834a | 7054 | GFP_KERNEL, MEMCG_RECLAIM_MAY_SWAP)) |
b6e6edcf JW |
7055 | nr_reclaims--; |
7056 | continue; | |
7057 | } | |
7058 | ||
e27be240 | 7059 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
7060 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
7061 | break; | |
7062 | } | |
241994ed | 7063 | |
2529bb3a | 7064 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
7065 | return nbytes; |
7066 | } | |
7067 | ||
664dc218 DR |
7068 | /* |
7069 | * Note: don't forget to update the 'samples/cgroup/memcg_event_listener' | |
7070 | * if any new events become available. | |
7071 | */ | |
1e577f97 SB |
7072 | static void __memory_events_show(struct seq_file *m, atomic_long_t *events) |
7073 | { | |
7074 | seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW])); | |
7075 | seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH])); | |
7076 | seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX])); | |
7077 | seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM])); | |
7078 | seq_printf(m, "oom_kill %lu\n", | |
7079 | atomic_long_read(&events[MEMCG_OOM_KILL])); | |
b6bf9abb DS |
7080 | seq_printf(m, "oom_group_kill %lu\n", |
7081 | atomic_long_read(&events[MEMCG_OOM_GROUP_KILL])); | |
1e577f97 SB |
7082 | } |
7083 | ||
241994ed JW |
7084 | static int memory_events_show(struct seq_file *m, void *v) |
7085 | { | |
aa9694bb | 7086 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
241994ed | 7087 | |
1e577f97 SB |
7088 | __memory_events_show(m, memcg->memory_events); |
7089 | return 0; | |
7090 | } | |
7091 | ||
7092 | static int memory_events_local_show(struct seq_file *m, void *v) | |
7093 | { | |
7094 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
241994ed | 7095 | |
1e577f97 | 7096 | __memory_events_show(m, memcg->memory_events_local); |
241994ed JW |
7097 | return 0; |
7098 | } | |
7099 | ||
587d9f72 JW |
7100 | static int memory_stat_show(struct seq_file *m, void *v) |
7101 | { | |
aa9694bb | 7102 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
68aaee14 | 7103 | char *buf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
5b42360c | 7104 | struct seq_buf s; |
1ff9e6e1 | 7105 | |
c8713d0b JW |
7106 | if (!buf) |
7107 | return -ENOMEM; | |
5b42360c YA |
7108 | seq_buf_init(&s, buf, PAGE_SIZE); |
7109 | memory_stat_format(memcg, &s); | |
c8713d0b JW |
7110 | seq_puts(m, buf); |
7111 | kfree(buf); | |
587d9f72 JW |
7112 | return 0; |
7113 | } | |
7114 | ||
5f9a4f4a | 7115 | #ifdef CONFIG_NUMA |
fff66b79 MS |
7116 | static inline unsigned long lruvec_page_state_output(struct lruvec *lruvec, |
7117 | int item) | |
7118 | { | |
ff841a06 YA |
7119 | return lruvec_page_state(lruvec, item) * |
7120 | memcg_page_state_output_unit(item); | |
fff66b79 MS |
7121 | } |
7122 | ||
5f9a4f4a MS |
7123 | static int memory_numa_stat_show(struct seq_file *m, void *v) |
7124 | { | |
7125 | int i; | |
7126 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
7127 | ||
7d7ef0a4 | 7128 | mem_cgroup_flush_stats(memcg); |
7e1c0d6f | 7129 | |
5f9a4f4a MS |
7130 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { |
7131 | int nid; | |
7132 | ||
7133 | if (memory_stats[i].idx >= NR_VM_NODE_STAT_ITEMS) | |
7134 | continue; | |
7135 | ||
7136 | seq_printf(m, "%s", memory_stats[i].name); | |
7137 | for_each_node_state(nid, N_MEMORY) { | |
7138 | u64 size; | |
7139 | struct lruvec *lruvec; | |
7140 | ||
7141 | lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); | |
fff66b79 MS |
7142 | size = lruvec_page_state_output(lruvec, |
7143 | memory_stats[i].idx); | |
5f9a4f4a MS |
7144 | seq_printf(m, " N%d=%llu", nid, size); |
7145 | } | |
7146 | seq_putc(m, '\n'); | |
7147 | } | |
7148 | ||
7149 | return 0; | |
7150 | } | |
7151 | #endif | |
7152 | ||
3d8b38eb RG |
7153 | static int memory_oom_group_show(struct seq_file *m, void *v) |
7154 | { | |
aa9694bb | 7155 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3d8b38eb | 7156 | |
eaf7b66b | 7157 | seq_printf(m, "%d\n", READ_ONCE(memcg->oom_group)); |
3d8b38eb RG |
7158 | |
7159 | return 0; | |
7160 | } | |
7161 | ||
7162 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
7163 | char *buf, size_t nbytes, loff_t off) | |
7164 | { | |
7165 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7166 | int ret, oom_group; | |
7167 | ||
7168 | buf = strstrip(buf); | |
7169 | if (!buf) | |
7170 | return -EINVAL; | |
7171 | ||
7172 | ret = kstrtoint(buf, 0, &oom_group); | |
7173 | if (ret) | |
7174 | return ret; | |
7175 | ||
7176 | if (oom_group != 0 && oom_group != 1) | |
7177 | return -EINVAL; | |
7178 | ||
eaf7b66b | 7179 | WRITE_ONCE(memcg->oom_group, oom_group); |
3d8b38eb RG |
7180 | |
7181 | return nbytes; | |
7182 | } | |
7183 | ||
94968384 SB |
7184 | static ssize_t memory_reclaim(struct kernfs_open_file *of, char *buf, |
7185 | size_t nbytes, loff_t off) | |
7186 | { | |
7187 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7188 | unsigned int nr_retries = MAX_RECLAIM_RETRIES; | |
7189 | unsigned long nr_to_reclaim, nr_reclaimed = 0; | |
55ab834a MH |
7190 | unsigned int reclaim_options; |
7191 | int err; | |
12a5d395 MA |
7192 | |
7193 | buf = strstrip(buf); | |
55ab834a MH |
7194 | err = page_counter_memparse(buf, "", &nr_to_reclaim); |
7195 | if (err) | |
7196 | return err; | |
12a5d395 | 7197 | |
55ab834a | 7198 | reclaim_options = MEMCG_RECLAIM_MAY_SWAP | MEMCG_RECLAIM_PROACTIVE; |
94968384 | 7199 | while (nr_reclaimed < nr_to_reclaim) { |
287d5fed M |
7200 | /* Will converge on zero, but reclaim enforces a minimum */ |
7201 | unsigned long batch_size = (nr_to_reclaim - nr_reclaimed) / 4; | |
94968384 SB |
7202 | unsigned long reclaimed; |
7203 | ||
7204 | if (signal_pending(current)) | |
7205 | return -EINTR; | |
7206 | ||
7207 | /* | |
7208 | * This is the final attempt, drain percpu lru caches in the | |
7209 | * hope of introducing more evictable pages for | |
7210 | * try_to_free_mem_cgroup_pages(). | |
7211 | */ | |
7212 | if (!nr_retries) | |
7213 | lru_add_drain_all(); | |
7214 | ||
7215 | reclaimed = try_to_free_mem_cgroup_pages(memcg, | |
287d5fed | 7216 | batch_size, GFP_KERNEL, reclaim_options); |
94968384 SB |
7217 | |
7218 | if (!reclaimed && !nr_retries--) | |
7219 | return -EAGAIN; | |
7220 | ||
7221 | nr_reclaimed += reclaimed; | |
7222 | } | |
7223 | ||
7224 | return nbytes; | |
7225 | } | |
7226 | ||
241994ed JW |
7227 | static struct cftype memory_files[] = { |
7228 | { | |
7229 | .name = "current", | |
f5fc3c5d | 7230 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
7231 | .read_u64 = memory_current_read, |
7232 | }, | |
8e20d4b3 GR |
7233 | { |
7234 | .name = "peak", | |
7235 | .flags = CFTYPE_NOT_ON_ROOT, | |
7236 | .read_u64 = memory_peak_read, | |
7237 | }, | |
bf8d5d52 RG |
7238 | { |
7239 | .name = "min", | |
7240 | .flags = CFTYPE_NOT_ON_ROOT, | |
7241 | .seq_show = memory_min_show, | |
7242 | .write = memory_min_write, | |
7243 | }, | |
241994ed JW |
7244 | { |
7245 | .name = "low", | |
7246 | .flags = CFTYPE_NOT_ON_ROOT, | |
7247 | .seq_show = memory_low_show, | |
7248 | .write = memory_low_write, | |
7249 | }, | |
7250 | { | |
7251 | .name = "high", | |
7252 | .flags = CFTYPE_NOT_ON_ROOT, | |
7253 | .seq_show = memory_high_show, | |
7254 | .write = memory_high_write, | |
7255 | }, | |
7256 | { | |
7257 | .name = "max", | |
7258 | .flags = CFTYPE_NOT_ON_ROOT, | |
7259 | .seq_show = memory_max_show, | |
7260 | .write = memory_max_write, | |
7261 | }, | |
7262 | { | |
7263 | .name = "events", | |
7264 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 7265 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
7266 | .seq_show = memory_events_show, |
7267 | }, | |
1e577f97 SB |
7268 | { |
7269 | .name = "events.local", | |
7270 | .flags = CFTYPE_NOT_ON_ROOT, | |
7271 | .file_offset = offsetof(struct mem_cgroup, events_local_file), | |
7272 | .seq_show = memory_events_local_show, | |
7273 | }, | |
587d9f72 JW |
7274 | { |
7275 | .name = "stat", | |
587d9f72 JW |
7276 | .seq_show = memory_stat_show, |
7277 | }, | |
5f9a4f4a MS |
7278 | #ifdef CONFIG_NUMA |
7279 | { | |
7280 | .name = "numa_stat", | |
7281 | .seq_show = memory_numa_stat_show, | |
7282 | }, | |
7283 | #endif | |
3d8b38eb RG |
7284 | { |
7285 | .name = "oom.group", | |
7286 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
7287 | .seq_show = memory_oom_group_show, | |
7288 | .write = memory_oom_group_write, | |
7289 | }, | |
94968384 SB |
7290 | { |
7291 | .name = "reclaim", | |
7292 | .flags = CFTYPE_NS_DELEGATABLE, | |
7293 | .write = memory_reclaim, | |
7294 | }, | |
241994ed JW |
7295 | { } /* terminate */ |
7296 | }; | |
7297 | ||
073219e9 | 7298 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 7299 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 7300 | .css_online = mem_cgroup_css_online, |
92fb9748 | 7301 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 7302 | .css_released = mem_cgroup_css_released, |
92fb9748 | 7303 | .css_free = mem_cgroup_css_free, |
1ced953b | 7304 | .css_reset = mem_cgroup_css_reset, |
2d146aa3 | 7305 | .css_rstat_flush = mem_cgroup_css_rstat_flush, |
7dc74be0 | 7306 | .can_attach = mem_cgroup_can_attach, |
1aacbd35 | 7307 | #if defined(CONFIG_LRU_GEN) || defined(CONFIG_MEMCG_KMEM) |
bd74fdae | 7308 | .attach = mem_cgroup_attach, |
1aacbd35 | 7309 | #endif |
7dc74be0 | 7310 | .cancel_attach = mem_cgroup_cancel_attach, |
264a0ae1 | 7311 | .post_attach = mem_cgroup_move_task, |
1aacbd35 RG |
7312 | #ifdef CONFIG_MEMCG_KMEM |
7313 | .fork = mem_cgroup_fork, | |
7314 | .exit = mem_cgroup_exit, | |
7315 | #endif | |
241994ed JW |
7316 | .dfl_cftypes = memory_files, |
7317 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 7318 | .early_init = 0, |
8cdea7c0 | 7319 | }; |
c077719b | 7320 | |
bc50bcc6 JW |
7321 | /* |
7322 | * This function calculates an individual cgroup's effective | |
7323 | * protection which is derived from its own memory.min/low, its | |
7324 | * parent's and siblings' settings, as well as the actual memory | |
7325 | * distribution in the tree. | |
7326 | * | |
7327 | * The following rules apply to the effective protection values: | |
7328 | * | |
7329 | * 1. At the first level of reclaim, effective protection is equal to | |
7330 | * the declared protection in memory.min and memory.low. | |
7331 | * | |
7332 | * 2. To enable safe delegation of the protection configuration, at | |
7333 | * subsequent levels the effective protection is capped to the | |
7334 | * parent's effective protection. | |
7335 | * | |
7336 | * 3. To make complex and dynamic subtrees easier to configure, the | |
7337 | * user is allowed to overcommit the declared protection at a given | |
7338 | * level. If that is the case, the parent's effective protection is | |
7339 | * distributed to the children in proportion to how much protection | |
7340 | * they have declared and how much of it they are utilizing. | |
7341 | * | |
7342 | * This makes distribution proportional, but also work-conserving: | |
7343 | * if one cgroup claims much more protection than it uses memory, | |
7344 | * the unused remainder is available to its siblings. | |
7345 | * | |
7346 | * 4. Conversely, when the declared protection is undercommitted at a | |
7347 | * given level, the distribution of the larger parental protection | |
7348 | * budget is NOT proportional. A cgroup's protection from a sibling | |
7349 | * is capped to its own memory.min/low setting. | |
7350 | * | |
8a931f80 JW |
7351 | * 5. However, to allow protecting recursive subtrees from each other |
7352 | * without having to declare each individual cgroup's fixed share | |
7353 | * of the ancestor's claim to protection, any unutilized - | |
7354 | * "floating" - protection from up the tree is distributed in | |
7355 | * proportion to each cgroup's *usage*. This makes the protection | |
7356 | * neutral wrt sibling cgroups and lets them compete freely over | |
7357 | * the shared parental protection budget, but it protects the | |
7358 | * subtree as a whole from neighboring subtrees. | |
7359 | * | |
7360 | * Note that 4. and 5. are not in conflict: 4. is about protecting | |
7361 | * against immediate siblings whereas 5. is about protecting against | |
7362 | * neighboring subtrees. | |
bc50bcc6 JW |
7363 | */ |
7364 | static unsigned long effective_protection(unsigned long usage, | |
8a931f80 | 7365 | unsigned long parent_usage, |
bc50bcc6 JW |
7366 | unsigned long setting, |
7367 | unsigned long parent_effective, | |
7368 | unsigned long siblings_protected) | |
7369 | { | |
7370 | unsigned long protected; | |
8a931f80 | 7371 | unsigned long ep; |
bc50bcc6 JW |
7372 | |
7373 | protected = min(usage, setting); | |
7374 | /* | |
7375 | * If all cgroups at this level combined claim and use more | |
08e0f49e | 7376 | * protection than what the parent affords them, distribute |
bc50bcc6 JW |
7377 | * shares in proportion to utilization. |
7378 | * | |
7379 | * We are using actual utilization rather than the statically | |
7380 | * claimed protection in order to be work-conserving: claimed | |
7381 | * but unused protection is available to siblings that would | |
7382 | * otherwise get a smaller chunk than what they claimed. | |
7383 | */ | |
7384 | if (siblings_protected > parent_effective) | |
7385 | return protected * parent_effective / siblings_protected; | |
7386 | ||
7387 | /* | |
7388 | * Ok, utilized protection of all children is within what the | |
7389 | * parent affords them, so we know whatever this child claims | |
7390 | * and utilizes is effectively protected. | |
7391 | * | |
7392 | * If there is unprotected usage beyond this value, reclaim | |
7393 | * will apply pressure in proportion to that amount. | |
7394 | * | |
7395 | * If there is unutilized protection, the cgroup will be fully | |
7396 | * shielded from reclaim, but we do return a smaller value for | |
7397 | * protection than what the group could enjoy in theory. This | |
7398 | * is okay. With the overcommit distribution above, effective | |
7399 | * protection is always dependent on how memory is actually | |
7400 | * consumed among the siblings anyway. | |
7401 | */ | |
8a931f80 JW |
7402 | ep = protected; |
7403 | ||
7404 | /* | |
7405 | * If the children aren't claiming (all of) the protection | |
7406 | * afforded to them by the parent, distribute the remainder in | |
7407 | * proportion to the (unprotected) memory of each cgroup. That | |
7408 | * way, cgroups that aren't explicitly prioritized wrt each | |
7409 | * other compete freely over the allowance, but they are | |
7410 | * collectively protected from neighboring trees. | |
7411 | * | |
7412 | * We're using unprotected memory for the weight so that if | |
7413 | * some cgroups DO claim explicit protection, we don't protect | |
7414 | * the same bytes twice. | |
cd324edc JW |
7415 | * |
7416 | * Check both usage and parent_usage against the respective | |
7417 | * protected values. One should imply the other, but they | |
7418 | * aren't read atomically - make sure the division is sane. | |
8a931f80 JW |
7419 | */ |
7420 | if (!(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)) | |
7421 | return ep; | |
cd324edc JW |
7422 | if (parent_effective > siblings_protected && |
7423 | parent_usage > siblings_protected && | |
7424 | usage > protected) { | |
8a931f80 JW |
7425 | unsigned long unclaimed; |
7426 | ||
7427 | unclaimed = parent_effective - siblings_protected; | |
7428 | unclaimed *= usage - protected; | |
7429 | unclaimed /= parent_usage - siblings_protected; | |
7430 | ||
7431 | ep += unclaimed; | |
7432 | } | |
7433 | ||
7434 | return ep; | |
bc50bcc6 JW |
7435 | } |
7436 | ||
241994ed | 7437 | /** |
05395718 | 7438 | * mem_cgroup_calculate_protection - check if memory consumption is in the normal range |
34c81057 | 7439 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
7440 | * @memcg: the memory cgroup to check |
7441 | * | |
23067153 RG |
7442 | * WARNING: This function is not stateless! It can only be used as part |
7443 | * of a top-down tree iteration, not for isolated queries. | |
241994ed | 7444 | */ |
45c7f7e1 CD |
7445 | void mem_cgroup_calculate_protection(struct mem_cgroup *root, |
7446 | struct mem_cgroup *memcg) | |
241994ed | 7447 | { |
8a931f80 | 7448 | unsigned long usage, parent_usage; |
23067153 RG |
7449 | struct mem_cgroup *parent; |
7450 | ||
241994ed | 7451 | if (mem_cgroup_disabled()) |
45c7f7e1 | 7452 | return; |
241994ed | 7453 | |
34c81057 SC |
7454 | if (!root) |
7455 | root = root_mem_cgroup; | |
22f7496f YS |
7456 | |
7457 | /* | |
7458 | * Effective values of the reclaim targets are ignored so they | |
7459 | * can be stale. Have a look at mem_cgroup_protection for more | |
7460 | * details. | |
7461 | * TODO: calculation should be more robust so that we do not need | |
7462 | * that special casing. | |
7463 | */ | |
34c81057 | 7464 | if (memcg == root) |
45c7f7e1 | 7465 | return; |
241994ed | 7466 | |
23067153 | 7467 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 | 7468 | if (!usage) |
45c7f7e1 | 7469 | return; |
bf8d5d52 | 7470 | |
bf8d5d52 | 7471 | parent = parent_mem_cgroup(memcg); |
df2a4196 | 7472 | |
bc50bcc6 | 7473 | if (parent == root) { |
c3d53200 | 7474 | memcg->memory.emin = READ_ONCE(memcg->memory.min); |
03960e33 | 7475 | memcg->memory.elow = READ_ONCE(memcg->memory.low); |
45c7f7e1 | 7476 | return; |
bf8d5d52 RG |
7477 | } |
7478 | ||
8a931f80 JW |
7479 | parent_usage = page_counter_read(&parent->memory); |
7480 | ||
b3a7822e | 7481 | WRITE_ONCE(memcg->memory.emin, effective_protection(usage, parent_usage, |
c3d53200 CD |
7482 | READ_ONCE(memcg->memory.min), |
7483 | READ_ONCE(parent->memory.emin), | |
b3a7822e | 7484 | atomic_long_read(&parent->memory.children_min_usage))); |
23067153 | 7485 | |
b3a7822e | 7486 | WRITE_ONCE(memcg->memory.elow, effective_protection(usage, parent_usage, |
03960e33 CD |
7487 | READ_ONCE(memcg->memory.low), |
7488 | READ_ONCE(parent->memory.elow), | |
b3a7822e | 7489 | atomic_long_read(&parent->memory.children_low_usage))); |
241994ed JW |
7490 | } |
7491 | ||
8f425e4e MWO |
7492 | static int charge_memcg(struct folio *folio, struct mem_cgroup *memcg, |
7493 | gfp_t gfp) | |
0add0c77 | 7494 | { |
0add0c77 SB |
7495 | int ret; |
7496 | ||
4b569387 | 7497 | ret = try_charge(memcg, gfp, folio_nr_pages(folio)); |
0add0c77 SB |
7498 | if (ret) |
7499 | goto out; | |
7500 | ||
4b569387 | 7501 | mem_cgroup_commit_charge(folio, memcg); |
0add0c77 SB |
7502 | out: |
7503 | return ret; | |
7504 | } | |
7505 | ||
8f425e4e | 7506 | int __mem_cgroup_charge(struct folio *folio, struct mm_struct *mm, gfp_t gfp) |
00501b53 | 7507 | { |
0add0c77 SB |
7508 | struct mem_cgroup *memcg; |
7509 | int ret; | |
00501b53 | 7510 | |
0add0c77 | 7511 | memcg = get_mem_cgroup_from_mm(mm); |
8f425e4e | 7512 | ret = charge_memcg(folio, memcg, gfp); |
0add0c77 | 7513 | css_put(&memcg->css); |
2d1c4980 | 7514 | |
0add0c77 SB |
7515 | return ret; |
7516 | } | |
e993d905 | 7517 | |
8cba9576 NP |
7518 | /** |
7519 | * mem_cgroup_hugetlb_try_charge - try to charge the memcg for a hugetlb folio | |
7520 | * @memcg: memcg to charge. | |
7521 | * @gfp: reclaim mode. | |
7522 | * @nr_pages: number of pages to charge. | |
7523 | * | |
7524 | * This function is called when allocating a huge page folio to determine if | |
7525 | * the memcg has the capacity for it. It does not commit the charge yet, | |
7526 | * as the hugetlb folio itself has not been obtained from the hugetlb pool. | |
7527 | * | |
7528 | * Once we have obtained the hugetlb folio, we can call | |
7529 | * mem_cgroup_commit_charge() to commit the charge. If we fail to obtain the | |
7530 | * folio, we should instead call mem_cgroup_cancel_charge() to undo the effect | |
7531 | * of try_charge(). | |
7532 | * | |
7533 | * Returns 0 on success. Otherwise, an error code is returned. | |
7534 | */ | |
7535 | int mem_cgroup_hugetlb_try_charge(struct mem_cgroup *memcg, gfp_t gfp, | |
7536 | long nr_pages) | |
7537 | { | |
7538 | /* | |
7539 | * If hugetlb memcg charging is not enabled, do not fail hugetlb allocation, | |
7540 | * but do not attempt to commit charge later (or cancel on error) either. | |
7541 | */ | |
7542 | if (mem_cgroup_disabled() || !memcg || | |
7543 | !cgroup_subsys_on_dfl(memory_cgrp_subsys) || | |
7544 | !(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_HUGETLB_ACCOUNTING)) | |
7545 | return -EOPNOTSUPP; | |
7546 | ||
7547 | if (try_charge(memcg, gfp, nr_pages)) | |
7548 | return -ENOMEM; | |
7549 | ||
7550 | return 0; | |
7551 | } | |
7552 | ||
0add0c77 | 7553 | /** |
65995918 MWO |
7554 | * mem_cgroup_swapin_charge_folio - Charge a newly allocated folio for swapin. |
7555 | * @folio: folio to charge. | |
0add0c77 SB |
7556 | * @mm: mm context of the victim |
7557 | * @gfp: reclaim mode | |
65995918 | 7558 | * @entry: swap entry for which the folio is allocated |
0add0c77 | 7559 | * |
65995918 MWO |
7560 | * This function charges a folio allocated for swapin. Please call this before |
7561 | * adding the folio to the swapcache. | |
0add0c77 SB |
7562 | * |
7563 | * Returns 0 on success. Otherwise, an error code is returned. | |
7564 | */ | |
65995918 | 7565 | int mem_cgroup_swapin_charge_folio(struct folio *folio, struct mm_struct *mm, |
0add0c77 SB |
7566 | gfp_t gfp, swp_entry_t entry) |
7567 | { | |
7568 | struct mem_cgroup *memcg; | |
7569 | unsigned short id; | |
7570 | int ret; | |
00501b53 | 7571 | |
0add0c77 SB |
7572 | if (mem_cgroup_disabled()) |
7573 | return 0; | |
00501b53 | 7574 | |
0add0c77 SB |
7575 | id = lookup_swap_cgroup_id(entry); |
7576 | rcu_read_lock(); | |
7577 | memcg = mem_cgroup_from_id(id); | |
7578 | if (!memcg || !css_tryget_online(&memcg->css)) | |
7579 | memcg = get_mem_cgroup_from_mm(mm); | |
7580 | rcu_read_unlock(); | |
00501b53 | 7581 | |
8f425e4e | 7582 | ret = charge_memcg(folio, memcg, gfp); |
6abb5a86 | 7583 | |
0add0c77 SB |
7584 | css_put(&memcg->css); |
7585 | return ret; | |
7586 | } | |
00501b53 | 7587 | |
0add0c77 SB |
7588 | /* |
7589 | * mem_cgroup_swapin_uncharge_swap - uncharge swap slot | |
7590 | * @entry: swap entry for which the page is charged | |
7591 | * | |
7592 | * Call this function after successfully adding the charged page to swapcache. | |
7593 | * | |
7594 | * Note: This function assumes the page for which swap slot is being uncharged | |
7595 | * is order 0 page. | |
7596 | */ | |
7597 | void mem_cgroup_swapin_uncharge_swap(swp_entry_t entry) | |
7598 | { | |
cae3af62 MS |
7599 | /* |
7600 | * Cgroup1's unified memory+swap counter has been charged with the | |
7601 | * new swapcache page, finish the transfer by uncharging the swap | |
7602 | * slot. The swap slot would also get uncharged when it dies, but | |
7603 | * it can stick around indefinitely and we'd count the page twice | |
7604 | * the entire time. | |
7605 | * | |
7606 | * Cgroup2 has separate resource counters for memory and swap, | |
7607 | * so this is a non-issue here. Memory and swap charge lifetimes | |
7608 | * correspond 1:1 to page and swap slot lifetimes: we charge the | |
7609 | * page to memory here, and uncharge swap when the slot is freed. | |
7610 | */ | |
0add0c77 | 7611 | if (!mem_cgroup_disabled() && do_memsw_account()) { |
00501b53 JW |
7612 | /* |
7613 | * The swap entry might not get freed for a long time, | |
7614 | * let's not wait for it. The page already received a | |
7615 | * memory+swap charge, drop the swap entry duplicate. | |
7616 | */ | |
0add0c77 | 7617 | mem_cgroup_uncharge_swap(entry, 1); |
00501b53 | 7618 | } |
3fea5a49 JW |
7619 | } |
7620 | ||
a9d5adee JG |
7621 | struct uncharge_gather { |
7622 | struct mem_cgroup *memcg; | |
b4e0b68f | 7623 | unsigned long nr_memory; |
a9d5adee | 7624 | unsigned long pgpgout; |
a9d5adee | 7625 | unsigned long nr_kmem; |
8e88bd2d | 7626 | int nid; |
a9d5adee JG |
7627 | }; |
7628 | ||
7629 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 7630 | { |
a9d5adee JG |
7631 | memset(ug, 0, sizeof(*ug)); |
7632 | } | |
7633 | ||
7634 | static void uncharge_batch(const struct uncharge_gather *ug) | |
7635 | { | |
747db954 JW |
7636 | unsigned long flags; |
7637 | ||
b4e0b68f MS |
7638 | if (ug->nr_memory) { |
7639 | page_counter_uncharge(&ug->memcg->memory, ug->nr_memory); | |
7941d214 | 7640 | if (do_memsw_account()) |
b4e0b68f | 7641 | page_counter_uncharge(&ug->memcg->memsw, ug->nr_memory); |
a8c49af3 YA |
7642 | if (ug->nr_kmem) |
7643 | memcg_account_kmem(ug->memcg, -ug->nr_kmem); | |
a9d5adee | 7644 | memcg_oom_recover(ug->memcg); |
ce00a967 | 7645 | } |
747db954 JW |
7646 | |
7647 | local_irq_save(flags); | |
c9019e9b | 7648 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); |
b4e0b68f | 7649 | __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, ug->nr_memory); |
8e88bd2d | 7650 | memcg_check_events(ug->memcg, ug->nid); |
747db954 | 7651 | local_irq_restore(flags); |
f1796544 | 7652 | |
c4ed6ebf | 7653 | /* drop reference from uncharge_folio */ |
f1796544 | 7654 | css_put(&ug->memcg->css); |
a9d5adee JG |
7655 | } |
7656 | ||
c4ed6ebf | 7657 | static void uncharge_folio(struct folio *folio, struct uncharge_gather *ug) |
a9d5adee | 7658 | { |
c4ed6ebf | 7659 | long nr_pages; |
b4e0b68f MS |
7660 | struct mem_cgroup *memcg; |
7661 | struct obj_cgroup *objcg; | |
9f762dbe | 7662 | |
c4ed6ebf | 7663 | VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); |
b7b098cf MWO |
7664 | VM_BUG_ON_FOLIO(folio_order(folio) > 1 && |
7665 | !folio_test_hugetlb(folio) && | |
7666 | !list_empty(&folio->_deferred_list), folio); | |
a9d5adee | 7667 | |
a9d5adee JG |
7668 | /* |
7669 | * Nobody should be changing or seriously looking at | |
c4ed6ebf MWO |
7670 | * folio memcg or objcg at this point, we have fully |
7671 | * exclusive access to the folio. | |
a9d5adee | 7672 | */ |
fead2b86 | 7673 | if (folio_memcg_kmem(folio)) { |
1b7e4464 | 7674 | objcg = __folio_objcg(folio); |
b4e0b68f MS |
7675 | /* |
7676 | * This get matches the put at the end of the function and | |
7677 | * kmem pages do not hold memcg references anymore. | |
7678 | */ | |
7679 | memcg = get_mem_cgroup_from_objcg(objcg); | |
7680 | } else { | |
1b7e4464 | 7681 | memcg = __folio_memcg(folio); |
b4e0b68f | 7682 | } |
a9d5adee | 7683 | |
b4e0b68f MS |
7684 | if (!memcg) |
7685 | return; | |
7686 | ||
7687 | if (ug->memcg != memcg) { | |
a9d5adee JG |
7688 | if (ug->memcg) { |
7689 | uncharge_batch(ug); | |
7690 | uncharge_gather_clear(ug); | |
7691 | } | |
b4e0b68f | 7692 | ug->memcg = memcg; |
c4ed6ebf | 7693 | ug->nid = folio_nid(folio); |
f1796544 MH |
7694 | |
7695 | /* pairs with css_put in uncharge_batch */ | |
b4e0b68f | 7696 | css_get(&memcg->css); |
a9d5adee JG |
7697 | } |
7698 | ||
c4ed6ebf | 7699 | nr_pages = folio_nr_pages(folio); |
a9d5adee | 7700 | |
fead2b86 | 7701 | if (folio_memcg_kmem(folio)) { |
b4e0b68f | 7702 | ug->nr_memory += nr_pages; |
9f762dbe | 7703 | ug->nr_kmem += nr_pages; |
b4e0b68f | 7704 | |
c4ed6ebf | 7705 | folio->memcg_data = 0; |
b4e0b68f MS |
7706 | obj_cgroup_put(objcg); |
7707 | } else { | |
7708 | /* LRU pages aren't accounted at the root level */ | |
7709 | if (!mem_cgroup_is_root(memcg)) | |
7710 | ug->nr_memory += nr_pages; | |
18b2db3b | 7711 | ug->pgpgout++; |
a9d5adee | 7712 | |
c4ed6ebf | 7713 | folio->memcg_data = 0; |
b4e0b68f MS |
7714 | } |
7715 | ||
7716 | css_put(&memcg->css); | |
747db954 JW |
7717 | } |
7718 | ||
bbc6b703 | 7719 | void __mem_cgroup_uncharge(struct folio *folio) |
0a31bc97 | 7720 | { |
a9d5adee JG |
7721 | struct uncharge_gather ug; |
7722 | ||
bbc6b703 MWO |
7723 | /* Don't touch folio->lru of any random page, pre-check: */ |
7724 | if (!folio_memcg(folio)) | |
0a31bc97 JW |
7725 | return; |
7726 | ||
a9d5adee | 7727 | uncharge_gather_clear(&ug); |
bbc6b703 | 7728 | uncharge_folio(folio, &ug); |
a9d5adee | 7729 | uncharge_batch(&ug); |
747db954 | 7730 | } |
0a31bc97 | 7731 | |
4882c809 MWO |
7732 | void __mem_cgroup_uncharge_folios(struct folio_batch *folios) |
7733 | { | |
7734 | struct uncharge_gather ug; | |
7735 | unsigned int i; | |
7736 | ||
7737 | uncharge_gather_clear(&ug); | |
7738 | for (i = 0; i < folios->nr; i++) | |
7739 | uncharge_folio(folios->folios[i], &ug); | |
7740 | if (ug.memcg) | |
7741 | uncharge_batch(&ug); | |
0a31bc97 JW |
7742 | } |
7743 | ||
7744 | /** | |
85ce2c51 | 7745 | * mem_cgroup_replace_folio - Charge a folio's replacement. |
d21bba2b MWO |
7746 | * @old: Currently circulating folio. |
7747 | * @new: Replacement folio. | |
0a31bc97 | 7748 | * |
d21bba2b | 7749 | * Charge @new as a replacement folio for @old. @old will |
85ce2c51 NP |
7750 | * be uncharged upon free. This is only used by the page cache |
7751 | * (in replace_page_cache_folio()). | |
0a31bc97 | 7752 | * |
d21bba2b | 7753 | * Both folios must be locked, @new->mapping must be set up. |
0a31bc97 | 7754 | */ |
85ce2c51 | 7755 | void mem_cgroup_replace_folio(struct folio *old, struct folio *new) |
0a31bc97 | 7756 | { |
29833315 | 7757 | struct mem_cgroup *memcg; |
d21bba2b | 7758 | long nr_pages = folio_nr_pages(new); |
d93c4130 | 7759 | unsigned long flags; |
0a31bc97 | 7760 | |
d21bba2b MWO |
7761 | VM_BUG_ON_FOLIO(!folio_test_locked(old), old); |
7762 | VM_BUG_ON_FOLIO(!folio_test_locked(new), new); | |
7763 | VM_BUG_ON_FOLIO(folio_test_anon(old) != folio_test_anon(new), new); | |
7764 | VM_BUG_ON_FOLIO(folio_nr_pages(old) != nr_pages, new); | |
0a31bc97 JW |
7765 | |
7766 | if (mem_cgroup_disabled()) | |
7767 | return; | |
7768 | ||
d21bba2b MWO |
7769 | /* Page cache replacement: new folio already charged? */ |
7770 | if (folio_memcg(new)) | |
0a31bc97 JW |
7771 | return; |
7772 | ||
d21bba2b MWO |
7773 | memcg = folio_memcg(old); |
7774 | VM_WARN_ON_ONCE_FOLIO(!memcg, old); | |
29833315 | 7775 | if (!memcg) |
0a31bc97 JW |
7776 | return; |
7777 | ||
44b7a8d3 | 7778 | /* Force-charge the new page. The old one will be freed soon */ |
8dc87c7d MS |
7779 | if (!mem_cgroup_is_root(memcg)) { |
7780 | page_counter_charge(&memcg->memory, nr_pages); | |
7781 | if (do_memsw_account()) | |
7782 | page_counter_charge(&memcg->memsw, nr_pages); | |
7783 | } | |
0a31bc97 | 7784 | |
1a3e1f40 | 7785 | css_get(&memcg->css); |
d21bba2b | 7786 | commit_charge(new, memcg); |
44b7a8d3 | 7787 | |
d93c4130 | 7788 | local_irq_save(flags); |
6e0110c2 | 7789 | mem_cgroup_charge_statistics(memcg, nr_pages); |
d21bba2b | 7790 | memcg_check_events(memcg, folio_nid(new)); |
d93c4130 | 7791 | local_irq_restore(flags); |
0a31bc97 JW |
7792 | } |
7793 | ||
85ce2c51 NP |
7794 | /** |
7795 | * mem_cgroup_migrate - Transfer the memcg data from the old to the new folio. | |
7796 | * @old: Currently circulating folio. | |
7797 | * @new: Replacement folio. | |
7798 | * | |
7799 | * Transfer the memcg data from the old folio to the new folio for migration. | |
7800 | * The old folio's data info will be cleared. Note that the memory counters | |
7801 | * will remain unchanged throughout the process. | |
7802 | * | |
7803 | * Both folios must be locked, @new->mapping must be set up. | |
7804 | */ | |
7805 | void mem_cgroup_migrate(struct folio *old, struct folio *new) | |
7806 | { | |
7807 | struct mem_cgroup *memcg; | |
7808 | ||
7809 | VM_BUG_ON_FOLIO(!folio_test_locked(old), old); | |
7810 | VM_BUG_ON_FOLIO(!folio_test_locked(new), new); | |
7811 | VM_BUG_ON_FOLIO(folio_test_anon(old) != folio_test_anon(new), new); | |
7812 | VM_BUG_ON_FOLIO(folio_nr_pages(old) != folio_nr_pages(new), new); | |
7813 | ||
7814 | if (mem_cgroup_disabled()) | |
7815 | return; | |
7816 | ||
7817 | memcg = folio_memcg(old); | |
8cba9576 NP |
7818 | /* |
7819 | * Note that it is normal to see !memcg for a hugetlb folio. | |
7820 | * For e.g, itt could have been allocated when memory_hugetlb_accounting | |
7821 | * was not selected. | |
7822 | */ | |
7823 | VM_WARN_ON_ONCE_FOLIO(!folio_test_hugetlb(old) && !memcg, old); | |
85ce2c51 NP |
7824 | if (!memcg) |
7825 | return; | |
7826 | ||
7827 | /* Transfer the charge and the css ref */ | |
7828 | commit_charge(new, memcg); | |
9bcef597 BW |
7829 | /* |
7830 | * If the old folio is a large folio and is in the split queue, it needs | |
7831 | * to be removed from the split queue now, in case getting an incorrect | |
7832 | * split queue in destroy_large_folio() after the memcg of the old folio | |
7833 | * is cleared. | |
7834 | * | |
7835 | * In addition, the old folio is about to be freed after migration, so | |
7836 | * removing from the split queue a bit earlier seems reasonable. | |
7837 | */ | |
7838 | if (folio_test_large(old) && folio_test_large_rmappable(old)) | |
7839 | folio_undo_large_rmappable(old); | |
85ce2c51 NP |
7840 | old->memcg_data = 0; |
7841 | } | |
7842 | ||
ef12947c | 7843 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
7844 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
7845 | ||
2d758073 | 7846 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
7847 | { |
7848 | struct mem_cgroup *memcg; | |
7849 | ||
2d758073 JW |
7850 | if (!mem_cgroup_sockets_enabled) |
7851 | return; | |
7852 | ||
e876ecc6 | 7853 | /* Do not associate the sock with unrelated interrupted task's memcg. */ |
086f694a | 7854 | if (!in_task()) |
e876ecc6 SB |
7855 | return; |
7856 | ||
11092087 JW |
7857 | rcu_read_lock(); |
7858 | memcg = mem_cgroup_from_task(current); | |
7848ed62 | 7859 | if (mem_cgroup_is_root(memcg)) |
f7e1cb6e | 7860 | goto out; |
0db15298 | 7861 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 7862 | goto out; |
8965aa28 | 7863 | if (css_tryget(&memcg->css)) |
11092087 | 7864 | sk->sk_memcg = memcg; |
f7e1cb6e | 7865 | out: |
11092087 JW |
7866 | rcu_read_unlock(); |
7867 | } | |
11092087 | 7868 | |
2d758073 | 7869 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 7870 | { |
2d758073 JW |
7871 | if (sk->sk_memcg) |
7872 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
7873 | } |
7874 | ||
7875 | /** | |
7876 | * mem_cgroup_charge_skmem - charge socket memory | |
7877 | * @memcg: memcg to charge | |
7878 | * @nr_pages: number of pages to charge | |
4b1327be | 7879 | * @gfp_mask: reclaim mode |
11092087 JW |
7880 | * |
7881 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
4b1327be | 7882 | * @memcg's configured limit, %false if it doesn't. |
11092087 | 7883 | */ |
4b1327be WW |
7884 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages, |
7885 | gfp_t gfp_mask) | |
11092087 | 7886 | { |
f7e1cb6e | 7887 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 7888 | struct page_counter *fail; |
f7e1cb6e | 7889 | |
0db15298 JW |
7890 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
7891 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
7892 | return true; |
7893 | } | |
0db15298 | 7894 | memcg->tcpmem_pressure = 1; |
4b1327be WW |
7895 | if (gfp_mask & __GFP_NOFAIL) { |
7896 | page_counter_charge(&memcg->tcpmem, nr_pages); | |
7897 | return true; | |
7898 | } | |
f7e1cb6e | 7899 | return false; |
11092087 | 7900 | } |
d886f4e4 | 7901 | |
4b1327be WW |
7902 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) { |
7903 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); | |
f7e1cb6e | 7904 | return true; |
4b1327be | 7905 | } |
f7e1cb6e | 7906 | |
11092087 JW |
7907 | return false; |
7908 | } | |
7909 | ||
7910 | /** | |
7911 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
7912 | * @memcg: memcg to uncharge |
7913 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
7914 | */ |
7915 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
7916 | { | |
f7e1cb6e | 7917 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 7918 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
7919 | return; |
7920 | } | |
d886f4e4 | 7921 | |
c9019e9b | 7922 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 7923 | |
475d0487 | 7924 | refill_stock(memcg, nr_pages); |
11092087 JW |
7925 | } |
7926 | ||
f7e1cb6e JW |
7927 | static int __init cgroup_memory(char *s) |
7928 | { | |
7929 | char *token; | |
7930 | ||
7931 | while ((token = strsep(&s, ",")) != NULL) { | |
7932 | if (!*token) | |
7933 | continue; | |
7934 | if (!strcmp(token, "nosocket")) | |
7935 | cgroup_memory_nosocket = true; | |
04823c83 VD |
7936 | if (!strcmp(token, "nokmem")) |
7937 | cgroup_memory_nokmem = true; | |
b6c1a8af YS |
7938 | if (!strcmp(token, "nobpf")) |
7939 | cgroup_memory_nobpf = true; | |
f7e1cb6e | 7940 | } |
460a79e1 | 7941 | return 1; |
f7e1cb6e JW |
7942 | } |
7943 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 7944 | |
2d11085e | 7945 | /* |
1081312f MH |
7946 | * subsys_initcall() for memory controller. |
7947 | * | |
308167fc SAS |
7948 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
7949 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
7950 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
7951 | * should be initialized from here. | |
2d11085e MH |
7952 | */ |
7953 | static int __init mem_cgroup_init(void) | |
7954 | { | |
95a045f6 JW |
7955 | int cpu, node; |
7956 | ||
f3344adf MS |
7957 | /* |
7958 | * Currently s32 type (can refer to struct batched_lruvec_stat) is | |
7959 | * used for per-memcg-per-cpu caching of per-node statistics. In order | |
7960 | * to work fine, we should make sure that the overfill threshold can't | |
7961 | * exceed S32_MAX / PAGE_SIZE. | |
7962 | */ | |
7963 | BUILD_BUG_ON(MEMCG_CHARGE_BATCH > S32_MAX / PAGE_SIZE); | |
7964 | ||
308167fc SAS |
7965 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
7966 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
7967 | |
7968 | for_each_possible_cpu(cpu) | |
7969 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
7970 | drain_local_stock); | |
7971 | ||
7972 | for_each_node(node) { | |
7973 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 | 7974 | |
91f0dcce | 7975 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, node); |
95a045f6 | 7976 | |
ef8f2327 | 7977 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 7978 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 7979 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
7980 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
7981 | } | |
7982 | ||
2d11085e MH |
7983 | return 0; |
7984 | } | |
7985 | subsys_initcall(mem_cgroup_init); | |
21afa38e | 7986 | |
e55b9f96 | 7987 | #ifdef CONFIG_SWAP |
358c07fc AB |
7988 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
7989 | { | |
1c2d479a | 7990 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
7991 | /* |
7992 | * The root cgroup cannot be destroyed, so it's refcount must | |
7993 | * always be >= 1. | |
7994 | */ | |
7848ed62 | 7995 | if (WARN_ON_ONCE(mem_cgroup_is_root(memcg))) { |
358c07fc AB |
7996 | VM_BUG_ON(1); |
7997 | break; | |
7998 | } | |
7999 | memcg = parent_mem_cgroup(memcg); | |
8000 | if (!memcg) | |
8001 | memcg = root_mem_cgroup; | |
8002 | } | |
8003 | return memcg; | |
8004 | } | |
8005 | ||
21afa38e JW |
8006 | /** |
8007 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
3ecb0087 | 8008 | * @folio: folio whose memsw charge to transfer |
21afa38e JW |
8009 | * @entry: swap entry to move the charge to |
8010 | * | |
3ecb0087 | 8011 | * Transfer the memsw charge of @folio to @entry. |
21afa38e | 8012 | */ |
3ecb0087 | 8013 | void mem_cgroup_swapout(struct folio *folio, swp_entry_t entry) |
21afa38e | 8014 | { |
1f47b61f | 8015 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 8016 | unsigned int nr_entries; |
21afa38e JW |
8017 | unsigned short oldid; |
8018 | ||
3ecb0087 MWO |
8019 | VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); |
8020 | VM_BUG_ON_FOLIO(folio_ref_count(folio), folio); | |
21afa38e | 8021 | |
76358ab5 AS |
8022 | if (mem_cgroup_disabled()) |
8023 | return; | |
8024 | ||
b94c4e94 | 8025 | if (!do_memsw_account()) |
21afa38e JW |
8026 | return; |
8027 | ||
3ecb0087 | 8028 | memcg = folio_memcg(folio); |
21afa38e | 8029 | |
3ecb0087 | 8030 | VM_WARN_ON_ONCE_FOLIO(!memcg, folio); |
21afa38e JW |
8031 | if (!memcg) |
8032 | return; | |
8033 | ||
1f47b61f VD |
8034 | /* |
8035 | * In case the memcg owning these pages has been offlined and doesn't | |
8036 | * have an ID allocated to it anymore, charge the closest online | |
8037 | * ancestor for the swap instead and transfer the memory+swap charge. | |
8038 | */ | |
8039 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
3ecb0087 | 8040 | nr_entries = folio_nr_pages(folio); |
d6810d73 HY |
8041 | /* Get references for the tail pages, too */ |
8042 | if (nr_entries > 1) | |
8043 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
8044 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
8045 | nr_entries); | |
3ecb0087 | 8046 | VM_BUG_ON_FOLIO(oldid, folio); |
c9019e9b | 8047 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e | 8048 | |
3ecb0087 | 8049 | folio->memcg_data = 0; |
21afa38e JW |
8050 | |
8051 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 8052 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 8053 | |
b25806dc | 8054 | if (memcg != swap_memcg) { |
1f47b61f | 8055 | if (!mem_cgroup_is_root(swap_memcg)) |
d6810d73 HY |
8056 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
8057 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
8058 | } |
8059 | ||
ce9ce665 SAS |
8060 | /* |
8061 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 8062 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 8063 | * important here to have the interrupts disabled because it is the |
b93b0163 | 8064 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 | 8065 | */ |
be3e67b5 | 8066 | memcg_stats_lock(); |
6e0110c2 | 8067 | mem_cgroup_charge_statistics(memcg, -nr_entries); |
be3e67b5 | 8068 | memcg_stats_unlock(); |
3ecb0087 | 8069 | memcg_check_events(memcg, folio_nid(folio)); |
73f576c0 | 8070 | |
1a3e1f40 | 8071 | css_put(&memcg->css); |
21afa38e JW |
8072 | } |
8073 | ||
38d8b4e6 | 8074 | /** |
e2e3fdc7 MWO |
8075 | * __mem_cgroup_try_charge_swap - try charging swap space for a folio |
8076 | * @folio: folio being added to swap | |
37e84351 VD |
8077 | * @entry: swap entry to charge |
8078 | * | |
e2e3fdc7 | 8079 | * Try to charge @folio's memcg for the swap space at @entry. |
37e84351 VD |
8080 | * |
8081 | * Returns 0 on success, -ENOMEM on failure. | |
8082 | */ | |
e2e3fdc7 | 8083 | int __mem_cgroup_try_charge_swap(struct folio *folio, swp_entry_t entry) |
37e84351 | 8084 | { |
e2e3fdc7 | 8085 | unsigned int nr_pages = folio_nr_pages(folio); |
37e84351 | 8086 | struct page_counter *counter; |
38d8b4e6 | 8087 | struct mem_cgroup *memcg; |
37e84351 VD |
8088 | unsigned short oldid; |
8089 | ||
b94c4e94 | 8090 | if (do_memsw_account()) |
37e84351 VD |
8091 | return 0; |
8092 | ||
e2e3fdc7 | 8093 | memcg = folio_memcg(folio); |
37e84351 | 8094 | |
e2e3fdc7 | 8095 | VM_WARN_ON_ONCE_FOLIO(!memcg, folio); |
37e84351 VD |
8096 | if (!memcg) |
8097 | return 0; | |
8098 | ||
f3a53a3a TH |
8099 | if (!entry.val) { |
8100 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 8101 | return 0; |
f3a53a3a | 8102 | } |
bb98f2c5 | 8103 | |
1f47b61f VD |
8104 | memcg = mem_cgroup_id_get_online(memcg); |
8105 | ||
b25806dc | 8106 | if (!mem_cgroup_is_root(memcg) && |
38d8b4e6 | 8107 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
8108 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
8109 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 8110 | mem_cgroup_id_put(memcg); |
37e84351 | 8111 | return -ENOMEM; |
1f47b61f | 8112 | } |
37e84351 | 8113 | |
38d8b4e6 HY |
8114 | /* Get references for the tail pages, too */ |
8115 | if (nr_pages > 1) | |
8116 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
8117 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
e2e3fdc7 | 8118 | VM_BUG_ON_FOLIO(oldid, folio); |
c9019e9b | 8119 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 8120 | |
37e84351 VD |
8121 | return 0; |
8122 | } | |
8123 | ||
21afa38e | 8124 | /** |
01c4b28c | 8125 | * __mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 8126 | * @entry: swap entry to uncharge |
38d8b4e6 | 8127 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 8128 | */ |
01c4b28c | 8129 | void __mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
8130 | { |
8131 | struct mem_cgroup *memcg; | |
8132 | unsigned short id; | |
8133 | ||
38d8b4e6 | 8134 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 8135 | rcu_read_lock(); |
adbe427b | 8136 | memcg = mem_cgroup_from_id(id); |
21afa38e | 8137 | if (memcg) { |
b25806dc | 8138 | if (!mem_cgroup_is_root(memcg)) { |
b94c4e94 | 8139 | if (do_memsw_account()) |
38d8b4e6 | 8140 | page_counter_uncharge(&memcg->memsw, nr_pages); |
b94c4e94 JW |
8141 | else |
8142 | page_counter_uncharge(&memcg->swap, nr_pages); | |
37e84351 | 8143 | } |
c9019e9b | 8144 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 8145 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
8146 | } |
8147 | rcu_read_unlock(); | |
8148 | } | |
8149 | ||
d8b38438 VD |
8150 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
8151 | { | |
8152 | long nr_swap_pages = get_nr_swap_pages(); | |
8153 | ||
b25806dc | 8154 | if (mem_cgroup_disabled() || do_memsw_account()) |
d8b38438 | 8155 | return nr_swap_pages; |
7848ed62 | 8156 | for (; !mem_cgroup_is_root(memcg); memcg = parent_mem_cgroup(memcg)) |
d8b38438 | 8157 | nr_swap_pages = min_t(long, nr_swap_pages, |
bbec2e15 | 8158 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
8159 | page_counter_read(&memcg->swap)); |
8160 | return nr_swap_pages; | |
8161 | } | |
8162 | ||
9202d527 | 8163 | bool mem_cgroup_swap_full(struct folio *folio) |
5ccc5aba VD |
8164 | { |
8165 | struct mem_cgroup *memcg; | |
8166 | ||
9202d527 | 8167 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
5ccc5aba VD |
8168 | |
8169 | if (vm_swap_full()) | |
8170 | return true; | |
b25806dc | 8171 | if (do_memsw_account()) |
5ccc5aba VD |
8172 | return false; |
8173 | ||
9202d527 | 8174 | memcg = folio_memcg(folio); |
5ccc5aba VD |
8175 | if (!memcg) |
8176 | return false; | |
8177 | ||
7848ed62 | 8178 | for (; !mem_cgroup_is_root(memcg); memcg = parent_mem_cgroup(memcg)) { |
4b82ab4f JK |
8179 | unsigned long usage = page_counter_read(&memcg->swap); |
8180 | ||
8181 | if (usage * 2 >= READ_ONCE(memcg->swap.high) || | |
8182 | usage * 2 >= READ_ONCE(memcg->swap.max)) | |
5ccc5aba | 8183 | return true; |
4b82ab4f | 8184 | } |
5ccc5aba VD |
8185 | |
8186 | return false; | |
8187 | } | |
8188 | ||
eccb52e7 | 8189 | static int __init setup_swap_account(char *s) |
21afa38e | 8190 | { |
118642d7 JW |
8191 | bool res; |
8192 | ||
8193 | if (!kstrtobool(s, &res) && !res) | |
8194 | pr_warn_once("The swapaccount=0 commandline option is deprecated " | |
8195 | "in favor of configuring swap control via cgroupfs. " | |
8196 | "Please report your usecase to linux-mm@kvack.org if you " | |
8197 | "depend on this functionality.\n"); | |
21afa38e JW |
8198 | return 1; |
8199 | } | |
eccb52e7 | 8200 | __setup("swapaccount=", setup_swap_account); |
21afa38e | 8201 | |
37e84351 VD |
8202 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
8203 | struct cftype *cft) | |
8204 | { | |
8205 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
8206 | ||
8207 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
8208 | } | |
8209 | ||
e0e0b412 LD |
8210 | static u64 swap_peak_read(struct cgroup_subsys_state *css, |
8211 | struct cftype *cft) | |
8212 | { | |
8213 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
8214 | ||
8215 | return (u64)memcg->swap.watermark * PAGE_SIZE; | |
8216 | } | |
8217 | ||
4b82ab4f JK |
8218 | static int swap_high_show(struct seq_file *m, void *v) |
8219 | { | |
8220 | return seq_puts_memcg_tunable(m, | |
8221 | READ_ONCE(mem_cgroup_from_seq(m)->swap.high)); | |
8222 | } | |
8223 | ||
8224 | static ssize_t swap_high_write(struct kernfs_open_file *of, | |
8225 | char *buf, size_t nbytes, loff_t off) | |
8226 | { | |
8227 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
8228 | unsigned long high; | |
8229 | int err; | |
8230 | ||
8231 | buf = strstrip(buf); | |
8232 | err = page_counter_memparse(buf, "max", &high); | |
8233 | if (err) | |
8234 | return err; | |
8235 | ||
8236 | page_counter_set_high(&memcg->swap, high); | |
8237 | ||
8238 | return nbytes; | |
8239 | } | |
8240 | ||
37e84351 VD |
8241 | static int swap_max_show(struct seq_file *m, void *v) |
8242 | { | |
677dc973 CD |
8243 | return seq_puts_memcg_tunable(m, |
8244 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); | |
37e84351 VD |
8245 | } |
8246 | ||
8247 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
8248 | char *buf, size_t nbytes, loff_t off) | |
8249 | { | |
8250 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
8251 | unsigned long max; | |
8252 | int err; | |
8253 | ||
8254 | buf = strstrip(buf); | |
8255 | err = page_counter_memparse(buf, "max", &max); | |
8256 | if (err) | |
8257 | return err; | |
8258 | ||
be09102b | 8259 | xchg(&memcg->swap.max, max); |
37e84351 VD |
8260 | |
8261 | return nbytes; | |
8262 | } | |
8263 | ||
f3a53a3a TH |
8264 | static int swap_events_show(struct seq_file *m, void *v) |
8265 | { | |
aa9694bb | 8266 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
f3a53a3a | 8267 | |
4b82ab4f JK |
8268 | seq_printf(m, "high %lu\n", |
8269 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_HIGH])); | |
f3a53a3a TH |
8270 | seq_printf(m, "max %lu\n", |
8271 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
8272 | seq_printf(m, "fail %lu\n", | |
8273 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
8274 | ||
8275 | return 0; | |
8276 | } | |
8277 | ||
37e84351 VD |
8278 | static struct cftype swap_files[] = { |
8279 | { | |
8280 | .name = "swap.current", | |
8281 | .flags = CFTYPE_NOT_ON_ROOT, | |
8282 | .read_u64 = swap_current_read, | |
8283 | }, | |
4b82ab4f JK |
8284 | { |
8285 | .name = "swap.high", | |
8286 | .flags = CFTYPE_NOT_ON_ROOT, | |
8287 | .seq_show = swap_high_show, | |
8288 | .write = swap_high_write, | |
8289 | }, | |
37e84351 VD |
8290 | { |
8291 | .name = "swap.max", | |
8292 | .flags = CFTYPE_NOT_ON_ROOT, | |
8293 | .seq_show = swap_max_show, | |
8294 | .write = swap_max_write, | |
8295 | }, | |
e0e0b412 LD |
8296 | { |
8297 | .name = "swap.peak", | |
8298 | .flags = CFTYPE_NOT_ON_ROOT, | |
8299 | .read_u64 = swap_peak_read, | |
8300 | }, | |
f3a53a3a TH |
8301 | { |
8302 | .name = "swap.events", | |
8303 | .flags = CFTYPE_NOT_ON_ROOT, | |
8304 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
8305 | .seq_show = swap_events_show, | |
8306 | }, | |
37e84351 VD |
8307 | { } /* terminate */ |
8308 | }; | |
8309 | ||
eccb52e7 | 8310 | static struct cftype memsw_files[] = { |
21afa38e JW |
8311 | { |
8312 | .name = "memsw.usage_in_bytes", | |
8313 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
8314 | .read_u64 = mem_cgroup_read_u64, | |
8315 | }, | |
8316 | { | |
8317 | .name = "memsw.max_usage_in_bytes", | |
8318 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
8319 | .write = mem_cgroup_reset, | |
8320 | .read_u64 = mem_cgroup_read_u64, | |
8321 | }, | |
8322 | { | |
8323 | .name = "memsw.limit_in_bytes", | |
8324 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
8325 | .write = mem_cgroup_write, | |
8326 | .read_u64 = mem_cgroup_read_u64, | |
8327 | }, | |
8328 | { | |
8329 | .name = "memsw.failcnt", | |
8330 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
8331 | .write = mem_cgroup_reset, | |
8332 | .read_u64 = mem_cgroup_read_u64, | |
8333 | }, | |
8334 | { }, /* terminate */ | |
8335 | }; | |
8336 | ||
f4840ccf JW |
8337 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) |
8338 | /** | |
8339 | * obj_cgroup_may_zswap - check if this cgroup can zswap | |
8340 | * @objcg: the object cgroup | |
8341 | * | |
8342 | * Check if the hierarchical zswap limit has been reached. | |
8343 | * | |
8344 | * This doesn't check for specific headroom, and it is not atomic | |
8345 | * either. But with zswap, the size of the allocation is only known | |
be16dd76 | 8346 | * once compression has occurred, and this optimistic pre-check avoids |
f4840ccf JW |
8347 | * spending cycles on compression when there is already no room left |
8348 | * or zswap is disabled altogether somewhere in the hierarchy. | |
8349 | */ | |
8350 | bool obj_cgroup_may_zswap(struct obj_cgroup *objcg) | |
8351 | { | |
8352 | struct mem_cgroup *memcg, *original_memcg; | |
8353 | bool ret = true; | |
8354 | ||
8355 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
8356 | return true; | |
8357 | ||
8358 | original_memcg = get_mem_cgroup_from_objcg(objcg); | |
7848ed62 | 8359 | for (memcg = original_memcg; !mem_cgroup_is_root(memcg); |
f4840ccf JW |
8360 | memcg = parent_mem_cgroup(memcg)) { |
8361 | unsigned long max = READ_ONCE(memcg->zswap_max); | |
8362 | unsigned long pages; | |
8363 | ||
8364 | if (max == PAGE_COUNTER_MAX) | |
8365 | continue; | |
8366 | if (max == 0) { | |
8367 | ret = false; | |
8368 | break; | |
8369 | } | |
8370 | ||
7d7ef0a4 YA |
8371 | /* |
8372 | * mem_cgroup_flush_stats() ignores small changes. Use | |
8373 | * do_flush_stats() directly to get accurate stats for charging. | |
8374 | */ | |
8375 | do_flush_stats(memcg); | |
f4840ccf JW |
8376 | pages = memcg_page_state(memcg, MEMCG_ZSWAP_B) / PAGE_SIZE; |
8377 | if (pages < max) | |
8378 | continue; | |
8379 | ret = false; | |
8380 | break; | |
8381 | } | |
8382 | mem_cgroup_put(original_memcg); | |
8383 | return ret; | |
8384 | } | |
8385 | ||
8386 | /** | |
8387 | * obj_cgroup_charge_zswap - charge compression backend memory | |
8388 | * @objcg: the object cgroup | |
8389 | * @size: size of compressed object | |
8390 | * | |
3a1060c2 | 8391 | * This forces the charge after obj_cgroup_may_zswap() allowed |
f4840ccf JW |
8392 | * compression and storage in zwap for this cgroup to go ahead. |
8393 | */ | |
8394 | void obj_cgroup_charge_zswap(struct obj_cgroup *objcg, size_t size) | |
8395 | { | |
8396 | struct mem_cgroup *memcg; | |
8397 | ||
8398 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
8399 | return; | |
8400 | ||
8401 | VM_WARN_ON_ONCE(!(current->flags & PF_MEMALLOC)); | |
8402 | ||
8403 | /* PF_MEMALLOC context, charging must succeed */ | |
8404 | if (obj_cgroup_charge(objcg, GFP_KERNEL, size)) | |
8405 | VM_WARN_ON_ONCE(1); | |
8406 | ||
8407 | rcu_read_lock(); | |
8408 | memcg = obj_cgroup_memcg(objcg); | |
8409 | mod_memcg_state(memcg, MEMCG_ZSWAP_B, size); | |
8410 | mod_memcg_state(memcg, MEMCG_ZSWAPPED, 1); | |
8411 | rcu_read_unlock(); | |
8412 | } | |
8413 | ||
8414 | /** | |
8415 | * obj_cgroup_uncharge_zswap - uncharge compression backend memory | |
8416 | * @objcg: the object cgroup | |
8417 | * @size: size of compressed object | |
8418 | * | |
8419 | * Uncharges zswap memory on page in. | |
8420 | */ | |
8421 | void obj_cgroup_uncharge_zswap(struct obj_cgroup *objcg, size_t size) | |
8422 | { | |
8423 | struct mem_cgroup *memcg; | |
8424 | ||
8425 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
8426 | return; | |
8427 | ||
8428 | obj_cgroup_uncharge(objcg, size); | |
8429 | ||
8430 | rcu_read_lock(); | |
8431 | memcg = obj_cgroup_memcg(objcg); | |
8432 | mod_memcg_state(memcg, MEMCG_ZSWAP_B, -size); | |
8433 | mod_memcg_state(memcg, MEMCG_ZSWAPPED, -1); | |
8434 | rcu_read_unlock(); | |
8435 | } | |
8436 | ||
501a06fe NP |
8437 | bool mem_cgroup_zswap_writeback_enabled(struct mem_cgroup *memcg) |
8438 | { | |
8439 | /* if zswap is disabled, do not block pages going to the swapping device */ | |
8440 | return !is_zswap_enabled() || !memcg || READ_ONCE(memcg->zswap_writeback); | |
8441 | } | |
8442 | ||
f4840ccf JW |
8443 | static u64 zswap_current_read(struct cgroup_subsys_state *css, |
8444 | struct cftype *cft) | |
8445 | { | |
7d7ef0a4 YA |
8446 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
8447 | ||
8448 | mem_cgroup_flush_stats(memcg); | |
8449 | return memcg_page_state(memcg, MEMCG_ZSWAP_B); | |
f4840ccf JW |
8450 | } |
8451 | ||
8452 | static int zswap_max_show(struct seq_file *m, void *v) | |
8453 | { | |
8454 | return seq_puts_memcg_tunable(m, | |
8455 | READ_ONCE(mem_cgroup_from_seq(m)->zswap_max)); | |
8456 | } | |
8457 | ||
8458 | static ssize_t zswap_max_write(struct kernfs_open_file *of, | |
8459 | char *buf, size_t nbytes, loff_t off) | |
8460 | { | |
8461 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
8462 | unsigned long max; | |
8463 | int err; | |
8464 | ||
8465 | buf = strstrip(buf); | |
8466 | err = page_counter_memparse(buf, "max", &max); | |
8467 | if (err) | |
8468 | return err; | |
8469 | ||
8470 | xchg(&memcg->zswap_max, max); | |
8471 | ||
8472 | return nbytes; | |
8473 | } | |
8474 | ||
501a06fe NP |
8475 | static int zswap_writeback_show(struct seq_file *m, void *v) |
8476 | { | |
8477 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
8478 | ||
8479 | seq_printf(m, "%d\n", READ_ONCE(memcg->zswap_writeback)); | |
8480 | return 0; | |
8481 | } | |
8482 | ||
8483 | static ssize_t zswap_writeback_write(struct kernfs_open_file *of, | |
8484 | char *buf, size_t nbytes, loff_t off) | |
8485 | { | |
8486 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
8487 | int zswap_writeback; | |
8488 | ssize_t parse_ret = kstrtoint(strstrip(buf), 0, &zswap_writeback); | |
8489 | ||
8490 | if (parse_ret) | |
8491 | return parse_ret; | |
8492 | ||
8493 | if (zswap_writeback != 0 && zswap_writeback != 1) | |
8494 | return -EINVAL; | |
8495 | ||
8496 | WRITE_ONCE(memcg->zswap_writeback, zswap_writeback); | |
8497 | return nbytes; | |
8498 | } | |
8499 | ||
f4840ccf JW |
8500 | static struct cftype zswap_files[] = { |
8501 | { | |
8502 | .name = "zswap.current", | |
8503 | .flags = CFTYPE_NOT_ON_ROOT, | |
8504 | .read_u64 = zswap_current_read, | |
8505 | }, | |
8506 | { | |
8507 | .name = "zswap.max", | |
8508 | .flags = CFTYPE_NOT_ON_ROOT, | |
8509 | .seq_show = zswap_max_show, | |
8510 | .write = zswap_max_write, | |
8511 | }, | |
501a06fe NP |
8512 | { |
8513 | .name = "zswap.writeback", | |
8514 | .seq_show = zswap_writeback_show, | |
8515 | .write = zswap_writeback_write, | |
8516 | }, | |
f4840ccf JW |
8517 | { } /* terminate */ |
8518 | }; | |
8519 | #endif /* CONFIG_MEMCG_KMEM && CONFIG_ZSWAP */ | |
8520 | ||
21afa38e JW |
8521 | static int __init mem_cgroup_swap_init(void) |
8522 | { | |
2d1c4980 | 8523 | if (mem_cgroup_disabled()) |
eccb52e7 JW |
8524 | return 0; |
8525 | ||
8526 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, swap_files)); | |
8527 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, memsw_files)); | |
f4840ccf JW |
8528 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) |
8529 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, zswap_files)); | |
8530 | #endif | |
21afa38e JW |
8531 | return 0; |
8532 | } | |
b25806dc | 8533 | subsys_initcall(mem_cgroup_swap_init); |
21afa38e | 8534 | |
e55b9f96 | 8535 | #endif /* CONFIG_SWAP */ |