]>
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; | |
2191 | current->memcg_oom_gfp_mask = mask; | |
2192 | current->memcg_oom_order = order; | |
2193 | } | |
2194 | return false; | |
29ef680a MH |
2195 | } |
2196 | ||
7056d3a3 MH |
2197 | mem_cgroup_mark_under_oom(memcg); |
2198 | ||
2199 | locked = mem_cgroup_oom_trylock(memcg); | |
2200 | ||
2201 | if (locked) | |
2202 | mem_cgroup_oom_notify(memcg); | |
2203 | ||
2204 | mem_cgroup_unmark_under_oom(memcg); | |
becdf89d | 2205 | ret = mem_cgroup_out_of_memory(memcg, mask, order); |
7056d3a3 MH |
2206 | |
2207 | if (locked) | |
2208 | mem_cgroup_oom_unlock(memcg); | |
29ef680a | 2209 | |
7056d3a3 | 2210 | return ret; |
3812c8c8 JW |
2211 | } |
2212 | ||
2213 | /** | |
2214 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 2215 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 2216 | * |
49426420 JW |
2217 | * This has to be called at the end of a page fault if the memcg OOM |
2218 | * handler was enabled. | |
3812c8c8 | 2219 | * |
49426420 | 2220 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
2221 | * sleep on a waitqueue until the userspace task resolves the |
2222 | * situation. Sleeping directly in the charge context with all kinds | |
2223 | * of locks held is not a good idea, instead we remember an OOM state | |
2224 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 2225 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
2226 | * |
2227 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 2228 | * completed, %false otherwise. |
3812c8c8 | 2229 | */ |
49426420 | 2230 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 2231 | { |
626ebc41 | 2232 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 2233 | struct oom_wait_info owait; |
49426420 | 2234 | bool locked; |
3812c8c8 JW |
2235 | |
2236 | /* OOM is global, do not handle */ | |
3812c8c8 | 2237 | if (!memcg) |
49426420 | 2238 | return false; |
3812c8c8 | 2239 | |
7c5f64f8 | 2240 | if (!handle) |
49426420 | 2241 | goto cleanup; |
3812c8c8 JW |
2242 | |
2243 | owait.memcg = memcg; | |
2244 | owait.wait.flags = 0; | |
2245 | owait.wait.func = memcg_oom_wake_function; | |
2246 | owait.wait.private = current; | |
2055da97 | 2247 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 2248 | |
3812c8c8 | 2249 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
2250 | mem_cgroup_mark_under_oom(memcg); |
2251 | ||
2252 | locked = mem_cgroup_oom_trylock(memcg); | |
2253 | ||
2254 | if (locked) | |
2255 | mem_cgroup_oom_notify(memcg); | |
2256 | ||
857f2139 HX |
2257 | schedule(); |
2258 | mem_cgroup_unmark_under_oom(memcg); | |
2259 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
49426420 | 2260 | |
18b1d18b | 2261 | if (locked) |
fb2a6fc5 | 2262 | mem_cgroup_oom_unlock(memcg); |
49426420 | 2263 | cleanup: |
626ebc41 | 2264 | current->memcg_in_oom = NULL; |
3812c8c8 | 2265 | css_put(&memcg->css); |
867578cb | 2266 | return true; |
0b7f569e KH |
2267 | } |
2268 | ||
3d8b38eb RG |
2269 | /** |
2270 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
2271 | * @victim: task to be killed by the OOM killer | |
2272 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
2273 | * | |
2274 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
2275 | * by killing all belonging OOM-killable tasks. | |
2276 | * | |
2277 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
2278 | */ | |
2279 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
2280 | struct mem_cgroup *oom_domain) | |
2281 | { | |
2282 | struct mem_cgroup *oom_group = NULL; | |
2283 | struct mem_cgroup *memcg; | |
2284 | ||
2285 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
2286 | return NULL; | |
2287 | ||
2288 | if (!oom_domain) | |
2289 | oom_domain = root_mem_cgroup; | |
2290 | ||
2291 | rcu_read_lock(); | |
2292 | ||
2293 | memcg = mem_cgroup_from_task(victim); | |
7848ed62 | 2294 | if (mem_cgroup_is_root(memcg)) |
3d8b38eb RG |
2295 | goto out; |
2296 | ||
48fe267c RG |
2297 | /* |
2298 | * If the victim task has been asynchronously moved to a different | |
2299 | * memory cgroup, we might end up killing tasks outside oom_domain. | |
2300 | * In this case it's better to ignore memory.group.oom. | |
2301 | */ | |
2302 | if (unlikely(!mem_cgroup_is_descendant(memcg, oom_domain))) | |
2303 | goto out; | |
2304 | ||
3d8b38eb RG |
2305 | /* |
2306 | * Traverse the memory cgroup hierarchy from the victim task's | |
2307 | * cgroup up to the OOMing cgroup (or root) to find the | |
2308 | * highest-level memory cgroup with oom.group set. | |
2309 | */ | |
2310 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
eaf7b66b | 2311 | if (READ_ONCE(memcg->oom_group)) |
3d8b38eb RG |
2312 | oom_group = memcg; |
2313 | ||
2314 | if (memcg == oom_domain) | |
2315 | break; | |
2316 | } | |
2317 | ||
2318 | if (oom_group) | |
2319 | css_get(&oom_group->css); | |
2320 | out: | |
2321 | rcu_read_unlock(); | |
2322 | ||
2323 | return oom_group; | |
2324 | } | |
2325 | ||
2326 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
2327 | { | |
2328 | pr_info("Tasks in "); | |
2329 | pr_cont_cgroup_path(memcg->css.cgroup); | |
2330 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
2331 | } | |
2332 | ||
d7365e78 | 2333 | /** |
f70ad448 MWO |
2334 | * folio_memcg_lock - Bind a folio to its memcg. |
2335 | * @folio: The folio. | |
32047e2a | 2336 | * |
f70ad448 | 2337 | * This function prevents unlocked LRU folios from being moved to |
739f79fc JW |
2338 | * another cgroup. |
2339 | * | |
f70ad448 MWO |
2340 | * It ensures lifetime of the bound memcg. The caller is responsible |
2341 | * for the lifetime of the folio. | |
d69b042f | 2342 | */ |
f70ad448 | 2343 | void folio_memcg_lock(struct folio *folio) |
89c06bd5 KH |
2344 | { |
2345 | struct mem_cgroup *memcg; | |
6de22619 | 2346 | unsigned long flags; |
89c06bd5 | 2347 | |
6de22619 JW |
2348 | /* |
2349 | * The RCU lock is held throughout the transaction. The fast | |
2350 | * path can get away without acquiring the memcg->move_lock | |
2351 | * because page moving starts with an RCU grace period. | |
739f79fc | 2352 | */ |
d7365e78 JW |
2353 | rcu_read_lock(); |
2354 | ||
2355 | if (mem_cgroup_disabled()) | |
1c824a68 | 2356 | return; |
89c06bd5 | 2357 | again: |
f70ad448 | 2358 | memcg = folio_memcg(folio); |
29833315 | 2359 | if (unlikely(!memcg)) |
1c824a68 | 2360 | return; |
d7365e78 | 2361 | |
20ad50d6 AS |
2362 | #ifdef CONFIG_PROVE_LOCKING |
2363 | local_irq_save(flags); | |
2364 | might_lock(&memcg->move_lock); | |
2365 | local_irq_restore(flags); | |
2366 | #endif | |
2367 | ||
bdcbb659 | 2368 | if (atomic_read(&memcg->moving_account) <= 0) |
1c824a68 | 2369 | return; |
89c06bd5 | 2370 | |
6de22619 | 2371 | spin_lock_irqsave(&memcg->move_lock, flags); |
f70ad448 | 2372 | if (memcg != folio_memcg(folio)) { |
6de22619 | 2373 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
2374 | goto again; |
2375 | } | |
6de22619 JW |
2376 | |
2377 | /* | |
1c824a68 JW |
2378 | * When charge migration first begins, we can have multiple |
2379 | * critical sections holding the fast-path RCU lock and one | |
2380 | * holding the slowpath move_lock. Track the task who has the | |
6c77b607 | 2381 | * move_lock for folio_memcg_unlock(). |
6de22619 JW |
2382 | */ |
2383 | memcg->move_lock_task = current; | |
2384 | memcg->move_lock_flags = flags; | |
89c06bd5 | 2385 | } |
f70ad448 | 2386 | |
f70ad448 | 2387 | static void __folio_memcg_unlock(struct mem_cgroup *memcg) |
89c06bd5 | 2388 | { |
6de22619 JW |
2389 | if (memcg && memcg->move_lock_task == current) { |
2390 | unsigned long flags = memcg->move_lock_flags; | |
2391 | ||
2392 | memcg->move_lock_task = NULL; | |
2393 | memcg->move_lock_flags = 0; | |
2394 | ||
2395 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2396 | } | |
89c06bd5 | 2397 | |
d7365e78 | 2398 | rcu_read_unlock(); |
89c06bd5 | 2399 | } |
739f79fc JW |
2400 | |
2401 | /** | |
f70ad448 MWO |
2402 | * folio_memcg_unlock - Release the binding between a folio and its memcg. |
2403 | * @folio: The folio. | |
2404 | * | |
2405 | * This releases the binding created by folio_memcg_lock(). This does | |
2406 | * not change the accounting of this folio to its memcg, but it does | |
2407 | * permit others to change it. | |
739f79fc | 2408 | */ |
f70ad448 | 2409 | void folio_memcg_unlock(struct folio *folio) |
739f79fc | 2410 | { |
f70ad448 MWO |
2411 | __folio_memcg_unlock(folio_memcg(folio)); |
2412 | } | |
9da7b521 | 2413 | |
fead2b86 | 2414 | struct memcg_stock_pcp { |
56751146 | 2415 | local_lock_t stock_lock; |
fead2b86 MH |
2416 | struct mem_cgroup *cached; /* this never be root cgroup */ |
2417 | unsigned int nr_pages; | |
2418 | ||
bf4f0599 RG |
2419 | #ifdef CONFIG_MEMCG_KMEM |
2420 | struct obj_cgroup *cached_objcg; | |
68ac5b3c | 2421 | struct pglist_data *cached_pgdat; |
bf4f0599 | 2422 | unsigned int nr_bytes; |
68ac5b3c WL |
2423 | int nr_slab_reclaimable_b; |
2424 | int nr_slab_unreclaimable_b; | |
bf4f0599 RG |
2425 | #endif |
2426 | ||
cdec2e42 | 2427 | struct work_struct work; |
26fe6168 | 2428 | unsigned long flags; |
a0db00fc | 2429 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 | 2430 | }; |
56751146 SAS |
2431 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock) = { |
2432 | .stock_lock = INIT_LOCAL_LOCK(stock_lock), | |
2433 | }; | |
9f50fad6 | 2434 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2435 | |
bf4f0599 | 2436 | #ifdef CONFIG_MEMCG_KMEM |
56751146 | 2437 | static struct obj_cgroup *drain_obj_stock(struct memcg_stock_pcp *stock); |
bf4f0599 RG |
2438 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, |
2439 | struct mem_cgroup *root_memcg); | |
a8c49af3 | 2440 | static void memcg_account_kmem(struct mem_cgroup *memcg, int nr_pages); |
bf4f0599 RG |
2441 | |
2442 | #else | |
56751146 | 2443 | static inline struct obj_cgroup *drain_obj_stock(struct memcg_stock_pcp *stock) |
bf4f0599 | 2444 | { |
56751146 | 2445 | return NULL; |
bf4f0599 RG |
2446 | } |
2447 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2448 | struct mem_cgroup *root_memcg) | |
2449 | { | |
2450 | return false; | |
2451 | } | |
a8c49af3 YA |
2452 | static void memcg_account_kmem(struct mem_cgroup *memcg, int nr_pages) |
2453 | { | |
2454 | } | |
bf4f0599 RG |
2455 | #endif |
2456 | ||
a0956d54 SS |
2457 | /** |
2458 | * consume_stock: Try to consume stocked charge on this cpu. | |
2459 | * @memcg: memcg to consume from. | |
2460 | * @nr_pages: how many pages to charge. | |
2461 | * | |
2462 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2463 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2464 | * service an allocation will refill the stock. | |
2465 | * | |
2466 | * returns true if successful, false otherwise. | |
cdec2e42 | 2467 | */ |
a0956d54 | 2468 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2469 | { |
2470 | struct memcg_stock_pcp *stock; | |
db2ba40c | 2471 | unsigned long flags; |
3e32cb2e | 2472 | bool ret = false; |
cdec2e42 | 2473 | |
a983b5eb | 2474 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 2475 | return ret; |
a0956d54 | 2476 | |
56751146 | 2477 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
db2ba40c JW |
2478 | |
2479 | stock = this_cpu_ptr(&memcg_stock); | |
f785a8f2 | 2480 | if (memcg == READ_ONCE(stock->cached) && stock->nr_pages >= nr_pages) { |
a0956d54 | 2481 | stock->nr_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 | { | |
f785a8f2 | 2495 | struct mem_cgroup *old = READ_ONCE(stock->cached); |
cdec2e42 | 2496 | |
1a3e1f40 JW |
2497 | if (!old) |
2498 | return; | |
2499 | ||
11c9ea4e | 2500 | if (stock->nr_pages) { |
3e32cb2e | 2501 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 2502 | if (do_memsw_account()) |
3e32cb2e | 2503 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
11c9ea4e | 2504 | stock->nr_pages = 0; |
cdec2e42 | 2505 | } |
1a3e1f40 JW |
2506 | |
2507 | css_put(&old->css); | |
f785a8f2 | 2508 | WRITE_ONCE(stock->cached, NULL); |
cdec2e42 KH |
2509 | } |
2510 | ||
cdec2e42 KH |
2511 | static void drain_local_stock(struct work_struct *dummy) |
2512 | { | |
db2ba40c | 2513 | struct memcg_stock_pcp *stock; |
56751146 | 2514 | struct obj_cgroup *old = NULL; |
db2ba40c JW |
2515 | unsigned long flags; |
2516 | ||
72f0184c | 2517 | /* |
5c49cf9a MH |
2518 | * The only protection from cpu hotplug (memcg_hotplug_cpu_dead) vs. |
2519 | * drain_stock races is that we always operate on local CPU stock | |
2520 | * here with IRQ disabled | |
72f0184c | 2521 | */ |
56751146 | 2522 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
db2ba40c JW |
2523 | |
2524 | stock = this_cpu_ptr(&memcg_stock); | |
56751146 | 2525 | old = drain_obj_stock(stock); |
cdec2e42 | 2526 | drain_stock(stock); |
26fe6168 | 2527 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c | 2528 | |
56751146 | 2529 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
91b71e78 | 2530 | obj_cgroup_put(old); |
cdec2e42 KH |
2531 | } |
2532 | ||
2533 | /* | |
3e32cb2e | 2534 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2535 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2536 | */ |
af9a3b69 | 2537 | static void __refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2538 | { |
db2ba40c | 2539 | struct memcg_stock_pcp *stock; |
cdec2e42 | 2540 | |
db2ba40c | 2541 | stock = this_cpu_ptr(&memcg_stock); |
f785a8f2 | 2542 | if (READ_ONCE(stock->cached) != memcg) { /* reset if necessary */ |
cdec2e42 | 2543 | drain_stock(stock); |
1a3e1f40 | 2544 | css_get(&memcg->css); |
f785a8f2 | 2545 | WRITE_ONCE(stock->cached, memcg); |
cdec2e42 | 2546 | } |
11c9ea4e | 2547 | stock->nr_pages += nr_pages; |
db2ba40c | 2548 | |
a983b5eb | 2549 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 | 2550 | drain_stock(stock); |
af9a3b69 JW |
2551 | } |
2552 | ||
2553 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) | |
2554 | { | |
2555 | unsigned long flags; | |
475d0487 | 2556 | |
56751146 | 2557 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
af9a3b69 | 2558 | __refill_stock(memcg, nr_pages); |
56751146 | 2559 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
cdec2e42 KH |
2560 | } |
2561 | ||
2562 | /* | |
c0ff4b85 | 2563 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2564 | * of the hierarchy under it. |
cdec2e42 | 2565 | */ |
6d3d6aa2 | 2566 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2567 | { |
26fe6168 | 2568 | int cpu, curcpu; |
d38144b7 | 2569 | |
6d3d6aa2 JW |
2570 | /* If someone's already draining, avoid adding running more workers. */ |
2571 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2572 | return; | |
72f0184c MH |
2573 | /* |
2574 | * Notify other cpus that system-wide "drain" is running | |
2575 | * We do not care about races with the cpu hotplug because cpu down | |
2576 | * as well as workers from this path always operate on the local | |
2577 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2578 | */ | |
0790ed62 SAS |
2579 | migrate_disable(); |
2580 | curcpu = smp_processor_id(); | |
cdec2e42 KH |
2581 | for_each_online_cpu(cpu) { |
2582 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2583 | struct mem_cgroup *memcg; |
e1a366be | 2584 | bool flush = false; |
26fe6168 | 2585 | |
e1a366be | 2586 | rcu_read_lock(); |
f785a8f2 | 2587 | memcg = READ_ONCE(stock->cached); |
e1a366be RG |
2588 | if (memcg && stock->nr_pages && |
2589 | mem_cgroup_is_descendant(memcg, root_memcg)) | |
2590 | flush = true; | |
27fb0956 | 2591 | else if (obj_stock_flush_required(stock, root_memcg)) |
bf4f0599 | 2592 | flush = true; |
e1a366be RG |
2593 | rcu_read_unlock(); |
2594 | ||
2595 | if (flush && | |
2596 | !test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { | |
d1a05b69 MH |
2597 | if (cpu == curcpu) |
2598 | drain_local_stock(&stock->work); | |
6a792697 | 2599 | else if (!cpu_is_isolated(cpu)) |
d1a05b69 MH |
2600 | schedule_work_on(cpu, &stock->work); |
2601 | } | |
cdec2e42 | 2602 | } |
0790ed62 | 2603 | migrate_enable(); |
9f50fad6 | 2604 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2605 | } |
2606 | ||
2cd21c89 JW |
2607 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
2608 | { | |
2609 | struct memcg_stock_pcp *stock; | |
a3d4c05a | 2610 | |
2cd21c89 JW |
2611 | stock = &per_cpu(memcg_stock, cpu); |
2612 | drain_stock(stock); | |
a3d4c05a | 2613 | |
308167fc | 2614 | return 0; |
cdec2e42 KH |
2615 | } |
2616 | ||
b3ff9291 CD |
2617 | static unsigned long reclaim_high(struct mem_cgroup *memcg, |
2618 | unsigned int nr_pages, | |
2619 | gfp_t gfp_mask) | |
f7e1cb6e | 2620 | { |
b3ff9291 CD |
2621 | unsigned long nr_reclaimed = 0; |
2622 | ||
f7e1cb6e | 2623 | do { |
e22c6ed9 JW |
2624 | unsigned long pflags; |
2625 | ||
d1663a90 JK |
2626 | if (page_counter_read(&memcg->memory) <= |
2627 | READ_ONCE(memcg->memory.high)) | |
f7e1cb6e | 2628 | continue; |
e22c6ed9 | 2629 | |
e27be240 | 2630 | memcg_memory_event(memcg, MEMCG_HIGH); |
e22c6ed9 JW |
2631 | |
2632 | psi_memstall_enter(&pflags); | |
b3ff9291 | 2633 | nr_reclaimed += try_to_free_mem_cgroup_pages(memcg, nr_pages, |
73b73bac | 2634 | gfp_mask, |
55ab834a | 2635 | MEMCG_RECLAIM_MAY_SWAP); |
e22c6ed9 | 2636 | psi_memstall_leave(&pflags); |
4bf17307 CD |
2637 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2638 | !mem_cgroup_is_root(memcg)); | |
b3ff9291 CD |
2639 | |
2640 | return nr_reclaimed; | |
f7e1cb6e JW |
2641 | } |
2642 | ||
2643 | static void high_work_func(struct work_struct *work) | |
2644 | { | |
2645 | struct mem_cgroup *memcg; | |
2646 | ||
2647 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2648 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2649 | } |
2650 | ||
0e4b01df CD |
2651 | /* |
2652 | * Clamp the maximum sleep time per allocation batch to 2 seconds. This is | |
2653 | * enough to still cause a significant slowdown in most cases, while still | |
2654 | * allowing diagnostics and tracing to proceed without becoming stuck. | |
2655 | */ | |
2656 | #define MEMCG_MAX_HIGH_DELAY_JIFFIES (2UL*HZ) | |
2657 | ||
2658 | /* | |
2659 | * When calculating the delay, we use these either side of the exponentiation to | |
2660 | * maintain precision and scale to a reasonable number of jiffies (see the table | |
2661 | * below. | |
2662 | * | |
2663 | * - MEMCG_DELAY_PRECISION_SHIFT: Extra precision bits while translating the | |
2664 | * overage ratio to a delay. | |
ac5ddd0f | 2665 | * - MEMCG_DELAY_SCALING_SHIFT: The number of bits to scale down the |
0e4b01df CD |
2666 | * proposed penalty in order to reduce to a reasonable number of jiffies, and |
2667 | * to produce a reasonable delay curve. | |
2668 | * | |
2669 | * MEMCG_DELAY_SCALING_SHIFT just happens to be a number that produces a | |
2670 | * reasonable delay curve compared to precision-adjusted overage, not | |
2671 | * penalising heavily at first, but still making sure that growth beyond the | |
2672 | * limit penalises misbehaviour cgroups by slowing them down exponentially. For | |
2673 | * example, with a high of 100 megabytes: | |
2674 | * | |
2675 | * +-------+------------------------+ | |
2676 | * | usage | time to allocate in ms | | |
2677 | * +-------+------------------------+ | |
2678 | * | 100M | 0 | | |
2679 | * | 101M | 6 | | |
2680 | * | 102M | 25 | | |
2681 | * | 103M | 57 | | |
2682 | * | 104M | 102 | | |
2683 | * | 105M | 159 | | |
2684 | * | 106M | 230 | | |
2685 | * | 107M | 313 | | |
2686 | * | 108M | 409 | | |
2687 | * | 109M | 518 | | |
2688 | * | 110M | 639 | | |
2689 | * | 111M | 774 | | |
2690 | * | 112M | 921 | | |
2691 | * | 113M | 1081 | | |
2692 | * | 114M | 1254 | | |
2693 | * | 115M | 1439 | | |
2694 | * | 116M | 1638 | | |
2695 | * | 117M | 1849 | | |
2696 | * | 118M | 2000 | | |
2697 | * | 119M | 2000 | | |
2698 | * | 120M | 2000 | | |
2699 | * +-------+------------------------+ | |
2700 | */ | |
2701 | #define MEMCG_DELAY_PRECISION_SHIFT 20 | |
2702 | #define MEMCG_DELAY_SCALING_SHIFT 14 | |
2703 | ||
8a5dbc65 | 2704 | static u64 calculate_overage(unsigned long usage, unsigned long high) |
b23afb93 | 2705 | { |
8a5dbc65 | 2706 | u64 overage; |
b23afb93 | 2707 | |
8a5dbc65 JK |
2708 | if (usage <= high) |
2709 | return 0; | |
e26733e0 | 2710 | |
8a5dbc65 JK |
2711 | /* |
2712 | * Prevent division by 0 in overage calculation by acting as if | |
2713 | * it was a threshold of 1 page | |
2714 | */ | |
2715 | high = max(high, 1UL); | |
9b8b1754 | 2716 | |
8a5dbc65 JK |
2717 | overage = usage - high; |
2718 | overage <<= MEMCG_DELAY_PRECISION_SHIFT; | |
2719 | return div64_u64(overage, high); | |
2720 | } | |
e26733e0 | 2721 | |
8a5dbc65 JK |
2722 | static u64 mem_find_max_overage(struct mem_cgroup *memcg) |
2723 | { | |
2724 | u64 overage, max_overage = 0; | |
e26733e0 | 2725 | |
8a5dbc65 JK |
2726 | do { |
2727 | overage = calculate_overage(page_counter_read(&memcg->memory), | |
d1663a90 | 2728 | READ_ONCE(memcg->memory.high)); |
8a5dbc65 | 2729 | max_overage = max(overage, max_overage); |
e26733e0 CD |
2730 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2731 | !mem_cgroup_is_root(memcg)); | |
2732 | ||
8a5dbc65 JK |
2733 | return max_overage; |
2734 | } | |
2735 | ||
4b82ab4f JK |
2736 | static u64 swap_find_max_overage(struct mem_cgroup *memcg) |
2737 | { | |
2738 | u64 overage, max_overage = 0; | |
2739 | ||
2740 | do { | |
2741 | overage = calculate_overage(page_counter_read(&memcg->swap), | |
2742 | READ_ONCE(memcg->swap.high)); | |
2743 | if (overage) | |
2744 | memcg_memory_event(memcg, MEMCG_SWAP_HIGH); | |
2745 | max_overage = max(overage, max_overage); | |
2746 | } while ((memcg = parent_mem_cgroup(memcg)) && | |
2747 | !mem_cgroup_is_root(memcg)); | |
2748 | ||
2749 | return max_overage; | |
2750 | } | |
2751 | ||
8a5dbc65 JK |
2752 | /* |
2753 | * Get the number of jiffies that we should penalise a mischievous cgroup which | |
2754 | * is exceeding its memory.high by checking both it and its ancestors. | |
2755 | */ | |
2756 | static unsigned long calculate_high_delay(struct mem_cgroup *memcg, | |
2757 | unsigned int nr_pages, | |
2758 | u64 max_overage) | |
2759 | { | |
2760 | unsigned long penalty_jiffies; | |
2761 | ||
e26733e0 CD |
2762 | if (!max_overage) |
2763 | return 0; | |
0e4b01df CD |
2764 | |
2765 | /* | |
0e4b01df CD |
2766 | * We use overage compared to memory.high to calculate the number of |
2767 | * jiffies to sleep (penalty_jiffies). Ideally this value should be | |
2768 | * fairly lenient on small overages, and increasingly harsh when the | |
2769 | * memcg in question makes it clear that it has no intention of stopping | |
2770 | * its crazy behaviour, so we exponentially increase the delay based on | |
2771 | * overage amount. | |
2772 | */ | |
e26733e0 CD |
2773 | penalty_jiffies = max_overage * max_overage * HZ; |
2774 | penalty_jiffies >>= MEMCG_DELAY_PRECISION_SHIFT; | |
2775 | penalty_jiffies >>= MEMCG_DELAY_SCALING_SHIFT; | |
0e4b01df CD |
2776 | |
2777 | /* | |
2778 | * Factor in the task's own contribution to the overage, such that four | |
2779 | * N-sized allocations are throttled approximately the same as one | |
2780 | * 4N-sized allocation. | |
2781 | * | |
2782 | * MEMCG_CHARGE_BATCH pages is nominal, so work out how much smaller or | |
2783 | * larger the current charge patch is than that. | |
2784 | */ | |
ff144e69 | 2785 | return penalty_jiffies * nr_pages / MEMCG_CHARGE_BATCH; |
e26733e0 CD |
2786 | } |
2787 | ||
2788 | /* | |
63fd3270 JW |
2789 | * Reclaims memory over the high limit. Called directly from |
2790 | * try_charge() (context permitting), as well as from the userland | |
2791 | * return path where reclaim is always able to block. | |
e26733e0 | 2792 | */ |
9ea9cb00 | 2793 | void mem_cgroup_handle_over_high(gfp_t gfp_mask) |
e26733e0 CD |
2794 | { |
2795 | unsigned long penalty_jiffies; | |
2796 | unsigned long pflags; | |
b3ff9291 | 2797 | unsigned long nr_reclaimed; |
e26733e0 | 2798 | unsigned int nr_pages = current->memcg_nr_pages_over_high; |
d977aa93 | 2799 | int nr_retries = MAX_RECLAIM_RETRIES; |
e26733e0 | 2800 | struct mem_cgroup *memcg; |
b3ff9291 | 2801 | bool in_retry = false; |
e26733e0 CD |
2802 | |
2803 | if (likely(!nr_pages)) | |
2804 | return; | |
2805 | ||
2806 | memcg = get_mem_cgroup_from_mm(current->mm); | |
e26733e0 CD |
2807 | current->memcg_nr_pages_over_high = 0; |
2808 | ||
b3ff9291 | 2809 | retry_reclaim: |
63fd3270 JW |
2810 | /* |
2811 | * Bail if the task is already exiting. Unlike memory.max, | |
2812 | * memory.high enforcement isn't as strict, and there is no | |
2813 | * OOM killer involved, which means the excess could already | |
2814 | * be much bigger (and still growing) than it could for | |
2815 | * memory.max; the dying task could get stuck in fruitless | |
2816 | * reclaim for a long time, which isn't desirable. | |
2817 | */ | |
2818 | if (task_is_dying()) | |
2819 | goto out; | |
2820 | ||
b3ff9291 CD |
2821 | /* |
2822 | * The allocating task should reclaim at least the batch size, but for | |
2823 | * subsequent retries we only want to do what's necessary to prevent oom | |
2824 | * or breaching resource isolation. | |
2825 | * | |
2826 | * This is distinct from memory.max or page allocator behaviour because | |
2827 | * memory.high is currently batched, whereas memory.max and the page | |
2828 | * allocator run every time an allocation is made. | |
2829 | */ | |
2830 | nr_reclaimed = reclaim_high(memcg, | |
2831 | in_retry ? SWAP_CLUSTER_MAX : nr_pages, | |
9ea9cb00 | 2832 | gfp_mask); |
b3ff9291 | 2833 | |
e26733e0 CD |
2834 | /* |
2835 | * memory.high is breached and reclaim is unable to keep up. Throttle | |
2836 | * allocators proactively to slow down excessive growth. | |
2837 | */ | |
8a5dbc65 JK |
2838 | penalty_jiffies = calculate_high_delay(memcg, nr_pages, |
2839 | mem_find_max_overage(memcg)); | |
0e4b01df | 2840 | |
4b82ab4f JK |
2841 | penalty_jiffies += calculate_high_delay(memcg, nr_pages, |
2842 | swap_find_max_overage(memcg)); | |
2843 | ||
ff144e69 JK |
2844 | /* |
2845 | * Clamp the max delay per usermode return so as to still keep the | |
2846 | * application moving forwards and also permit diagnostics, albeit | |
2847 | * extremely slowly. | |
2848 | */ | |
2849 | penalty_jiffies = min(penalty_jiffies, MEMCG_MAX_HIGH_DELAY_JIFFIES); | |
2850 | ||
0e4b01df CD |
2851 | /* |
2852 | * Don't sleep if the amount of jiffies this memcg owes us is so low | |
2853 | * that it's not even worth doing, in an attempt to be nice to those who | |
2854 | * go only a small amount over their memory.high value and maybe haven't | |
2855 | * been aggressively reclaimed enough yet. | |
2856 | */ | |
2857 | if (penalty_jiffies <= HZ / 100) | |
2858 | goto out; | |
2859 | ||
b3ff9291 CD |
2860 | /* |
2861 | * If reclaim is making forward progress but we're still over | |
2862 | * memory.high, we want to encourage that rather than doing allocator | |
2863 | * throttling. | |
2864 | */ | |
2865 | if (nr_reclaimed || nr_retries--) { | |
2866 | in_retry = true; | |
2867 | goto retry_reclaim; | |
2868 | } | |
2869 | ||
0e4b01df | 2870 | /* |
63fd3270 JW |
2871 | * Reclaim didn't manage to push usage below the limit, slow |
2872 | * this allocating task down. | |
2873 | * | |
0e4b01df CD |
2874 | * If we exit early, we're guaranteed to die (since |
2875 | * schedule_timeout_killable sets TASK_KILLABLE). This means we don't | |
2876 | * need to account for any ill-begotten jiffies to pay them off later. | |
2877 | */ | |
2878 | psi_memstall_enter(&pflags); | |
2879 | schedule_timeout_killable(penalty_jiffies); | |
2880 | psi_memstall_leave(&pflags); | |
2881 | ||
2882 | out: | |
2883 | css_put(&memcg->css); | |
b23afb93 TH |
2884 | } |
2885 | ||
c5c8b16b MS |
2886 | static int try_charge_memcg(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2887 | unsigned int nr_pages) | |
8a9f3ccd | 2888 | { |
a983b5eb | 2889 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
d977aa93 | 2890 | int nr_retries = MAX_RECLAIM_RETRIES; |
6539cc05 | 2891 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2892 | struct page_counter *counter; |
6539cc05 | 2893 | unsigned long nr_reclaimed; |
a4ebf1b6 | 2894 | bool passed_oom = false; |
73b73bac | 2895 | unsigned int reclaim_options = MEMCG_RECLAIM_MAY_SWAP; |
b70a2a21 | 2896 | bool drained = false; |
d6e103a7 | 2897 | bool raised_max_event = false; |
e22c6ed9 | 2898 | unsigned long pflags; |
a636b327 | 2899 | |
6539cc05 | 2900 | retry: |
b6b6cc72 | 2901 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2902 | return 0; |
8a9f3ccd | 2903 | |
7941d214 | 2904 | if (!do_memsw_account() || |
6071ca52 JW |
2905 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2906 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2907 | goto done_restock; |
7941d214 | 2908 | if (do_memsw_account()) |
3e32cb2e JW |
2909 | page_counter_uncharge(&memcg->memsw, batch); |
2910 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2911 | } else { |
3e32cb2e | 2912 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
73b73bac | 2913 | reclaim_options &= ~MEMCG_RECLAIM_MAY_SWAP; |
3fbe7244 | 2914 | } |
7a81b88c | 2915 | |
6539cc05 JW |
2916 | if (batch > nr_pages) { |
2917 | batch = nr_pages; | |
2918 | goto retry; | |
2919 | } | |
6d61ef40 | 2920 | |
89a28483 JW |
2921 | /* |
2922 | * Prevent unbounded recursion when reclaim operations need to | |
2923 | * allocate memory. This might exceed the limits temporarily, | |
2924 | * but we prefer facilitating memory reclaim and getting back | |
2925 | * under the limit over triggering OOM kills in these cases. | |
2926 | */ | |
2927 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2928 | goto force; | |
2929 | ||
06b078fc JW |
2930 | if (unlikely(task_in_memcg_oom(current))) |
2931 | goto nomem; | |
2932 | ||
d0164adc | 2933 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2934 | goto nomem; |
4b534334 | 2935 | |
e27be240 | 2936 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
d6e103a7 | 2937 | raised_max_event = true; |
241994ed | 2938 | |
e22c6ed9 | 2939 | psi_memstall_enter(&pflags); |
b70a2a21 | 2940 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
55ab834a | 2941 | gfp_mask, reclaim_options); |
e22c6ed9 | 2942 | psi_memstall_leave(&pflags); |
6539cc05 | 2943 | |
61e02c74 | 2944 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2945 | goto retry; |
28c34c29 | 2946 | |
b70a2a21 | 2947 | if (!drained) { |
6d3d6aa2 | 2948 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2949 | drained = true; |
2950 | goto retry; | |
2951 | } | |
2952 | ||
28c34c29 JW |
2953 | if (gfp_mask & __GFP_NORETRY) |
2954 | goto nomem; | |
6539cc05 JW |
2955 | /* |
2956 | * Even though the limit is exceeded at this point, reclaim | |
2957 | * may have been able to free some pages. Retry the charge | |
2958 | * before killing the task. | |
2959 | * | |
2960 | * Only for regular pages, though: huge pages are rather | |
2961 | * unlikely to succeed so close to the limit, and we fall back | |
2962 | * to regular pages anyway in case of failure. | |
2963 | */ | |
61e02c74 | 2964 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2965 | goto retry; |
2966 | /* | |
2967 | * At task move, charge accounts can be doubly counted. So, it's | |
2968 | * better to wait until the end of task_move if something is going on. | |
2969 | */ | |
2970 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2971 | goto retry; | |
2972 | ||
9b130619 JW |
2973 | if (nr_retries--) |
2974 | goto retry; | |
2975 | ||
38d38493 | 2976 | if (gfp_mask & __GFP_RETRY_MAYFAIL) |
29ef680a MH |
2977 | goto nomem; |
2978 | ||
a4ebf1b6 VA |
2979 | /* Avoid endless loop for tasks bypassed by the oom killer */ |
2980 | if (passed_oom && task_is_dying()) | |
2981 | goto nomem; | |
6539cc05 | 2982 | |
29ef680a MH |
2983 | /* |
2984 | * keep retrying as long as the memcg oom killer is able to make | |
2985 | * a forward progress or bypass the charge if the oom killer | |
2986 | * couldn't make any progress. | |
2987 | */ | |
becdf89d SB |
2988 | if (mem_cgroup_oom(mem_over_limit, gfp_mask, |
2989 | get_order(nr_pages * PAGE_SIZE))) { | |
a4ebf1b6 | 2990 | passed_oom = true; |
d977aa93 | 2991 | nr_retries = MAX_RECLAIM_RETRIES; |
29ef680a | 2992 | goto retry; |
29ef680a | 2993 | } |
7a81b88c | 2994 | nomem: |
1461e8c2 SB |
2995 | /* |
2996 | * Memcg doesn't have a dedicated reserve for atomic | |
2997 | * allocations. But like the global atomic pool, we need to | |
2998 | * put the burden of reclaim on regular allocation requests | |
2999 | * and let these go through as privileged allocations. | |
3000 | */ | |
3001 | if (!(gfp_mask & (__GFP_NOFAIL | __GFP_HIGH))) | |
3168ecbe | 3002 | return -ENOMEM; |
10d53c74 | 3003 | force: |
d6e103a7 RG |
3004 | /* |
3005 | * If the allocation has to be enforced, don't forget to raise | |
3006 | * a MEMCG_MAX event. | |
3007 | */ | |
3008 | if (!raised_max_event) | |
3009 | memcg_memory_event(mem_over_limit, MEMCG_MAX); | |
3010 | ||
10d53c74 TH |
3011 | /* |
3012 | * The allocation either can't fail or will lead to more memory | |
3013 | * being freed very soon. Allow memory usage go over the limit | |
3014 | * temporarily by force charging it. | |
3015 | */ | |
3016 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 3017 | if (do_memsw_account()) |
10d53c74 | 3018 | page_counter_charge(&memcg->memsw, nr_pages); |
10d53c74 TH |
3019 | |
3020 | return 0; | |
6539cc05 JW |
3021 | |
3022 | done_restock: | |
3023 | if (batch > nr_pages) | |
3024 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 3025 | |
241994ed | 3026 | /* |
b23afb93 TH |
3027 | * If the hierarchy is above the normal consumption range, schedule |
3028 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 3029 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
3030 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
3031 | * not recorded as it most likely matches current's and won't | |
3032 | * change in the meantime. As high limit is checked again before | |
3033 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
3034 | */ |
3035 | do { | |
4b82ab4f JK |
3036 | bool mem_high, swap_high; |
3037 | ||
3038 | mem_high = page_counter_read(&memcg->memory) > | |
3039 | READ_ONCE(memcg->memory.high); | |
3040 | swap_high = page_counter_read(&memcg->swap) > | |
3041 | READ_ONCE(memcg->swap.high); | |
3042 | ||
3043 | /* Don't bother a random interrupted task */ | |
086f694a | 3044 | if (!in_task()) { |
4b82ab4f | 3045 | if (mem_high) { |
f7e1cb6e JW |
3046 | schedule_work(&memcg->high_work); |
3047 | break; | |
3048 | } | |
4b82ab4f JK |
3049 | continue; |
3050 | } | |
3051 | ||
3052 | if (mem_high || swap_high) { | |
3053 | /* | |
3054 | * The allocating tasks in this cgroup will need to do | |
3055 | * reclaim or be throttled to prevent further growth | |
3056 | * of the memory or swap footprints. | |
3057 | * | |
3058 | * Target some best-effort fairness between the tasks, | |
3059 | * and distribute reclaim work and delay penalties | |
3060 | * based on how much each task is actually allocating. | |
3061 | */ | |
9516a18a | 3062 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
3063 | set_notify_resume(current); |
3064 | break; | |
3065 | } | |
241994ed | 3066 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 | 3067 | |
63fd3270 JW |
3068 | /* |
3069 | * Reclaim is set up above to be called from the userland | |
3070 | * return path. But also attempt synchronous reclaim to avoid | |
3071 | * excessive overrun while the task is still inside the | |
3072 | * kernel. If this is successful, the return path will see it | |
3073 | * when it rechecks the overage and simply bail out. | |
3074 | */ | |
c9afe31e SB |
3075 | if (current->memcg_nr_pages_over_high > MEMCG_CHARGE_BATCH && |
3076 | !(current->flags & PF_MEMALLOC) && | |
63fd3270 | 3077 | gfpflags_allow_blocking(gfp_mask)) |
9ea9cb00 | 3078 | mem_cgroup_handle_over_high(gfp_mask); |
10d53c74 | 3079 | return 0; |
7a81b88c | 3080 | } |
8a9f3ccd | 3081 | |
c5c8b16b MS |
3082 | static inline int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
3083 | unsigned int nr_pages) | |
3084 | { | |
3085 | if (mem_cgroup_is_root(memcg)) | |
3086 | return 0; | |
3087 | ||
3088 | return try_charge_memcg(memcg, gfp_mask, nr_pages); | |
3089 | } | |
3090 | ||
4b569387 NP |
3091 | /** |
3092 | * mem_cgroup_cancel_charge() - cancel an uncommitted try_charge() call. | |
3093 | * @memcg: memcg previously charged. | |
3094 | * @nr_pages: number of pages previously charged. | |
3095 | */ | |
3096 | void mem_cgroup_cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) | |
a3032a2c | 3097 | { |
ce00a967 JW |
3098 | if (mem_cgroup_is_root(memcg)) |
3099 | return; | |
3100 | ||
3e32cb2e | 3101 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 3102 | if (do_memsw_account()) |
3e32cb2e | 3103 | page_counter_uncharge(&memcg->memsw, nr_pages); |
d01dd17f KH |
3104 | } |
3105 | ||
118f2875 | 3106 | static void commit_charge(struct folio *folio, struct mem_cgroup *memcg) |
0a31bc97 | 3107 | { |
118f2875 | 3108 | VM_BUG_ON_FOLIO(folio_memcg(folio), folio); |
0a31bc97 | 3109 | /* |
a5eb011a | 3110 | * Any of the following ensures page's memcg stability: |
0a31bc97 | 3111 | * |
a0b5b414 JW |
3112 | * - the page lock |
3113 | * - LRU isolation | |
6c77b607 | 3114 | * - folio_memcg_lock() |
a0b5b414 | 3115 | * - exclusive reference |
018ee47f | 3116 | * - mem_cgroup_trylock_pages() |
0a31bc97 | 3117 | */ |
118f2875 | 3118 | folio->memcg_data = (unsigned long)memcg; |
7a81b88c | 3119 | } |
66e1707b | 3120 | |
4b569387 NP |
3121 | /** |
3122 | * mem_cgroup_commit_charge - commit a previously successful try_charge(). | |
3123 | * @folio: folio to commit the charge to. | |
3124 | * @memcg: memcg previously charged. | |
3125 | */ | |
3126 | void mem_cgroup_commit_charge(struct folio *folio, struct mem_cgroup *memcg) | |
3127 | { | |
3128 | css_get(&memcg->css); | |
3129 | commit_charge(folio, memcg); | |
3130 | ||
3131 | local_irq_disable(); | |
3132 | mem_cgroup_charge_statistics(memcg, folio_nr_pages(folio)); | |
3133 | memcg_check_events(memcg, folio_nid(folio)); | |
3134 | local_irq_enable(); | |
3135 | } | |
3136 | ||
84c07d11 | 3137 | #ifdef CONFIG_MEMCG_KMEM |
41eb5df1 | 3138 | |
91882c16 SB |
3139 | static inline void __mod_objcg_mlstate(struct obj_cgroup *objcg, |
3140 | struct pglist_data *pgdat, | |
3141 | enum node_stat_item idx, int nr) | |
a7ebf564 WL |
3142 | { |
3143 | struct mem_cgroup *memcg; | |
3144 | struct lruvec *lruvec; | |
3145 | ||
91882c16 SB |
3146 | lockdep_assert_irqs_disabled(); |
3147 | ||
a7ebf564 WL |
3148 | rcu_read_lock(); |
3149 | memcg = obj_cgroup_memcg(objcg); | |
3150 | lruvec = mem_cgroup_lruvec(memcg, pgdat); | |
91882c16 | 3151 | __mod_memcg_lruvec_state(lruvec, idx, nr); |
a7ebf564 WL |
3152 | rcu_read_unlock(); |
3153 | } | |
3154 | ||
fc4db90f RG |
3155 | static __always_inline |
3156 | struct mem_cgroup *mem_cgroup_from_obj_folio(struct folio *folio, void *p) | |
8380ce47 | 3157 | { |
8380ce47 | 3158 | /* |
9855609b RG |
3159 | * Slab objects are accounted individually, not per-page. |
3160 | * Memcg membership data for each individual object is saved in | |
21c690a3 | 3161 | * slab->obj_exts. |
8380ce47 | 3162 | */ |
4b5f8d9a | 3163 | if (folio_test_slab(folio)) { |
21c690a3 | 3164 | struct slabobj_ext *obj_exts; |
4b5f8d9a | 3165 | struct slab *slab; |
9855609b RG |
3166 | unsigned int off; |
3167 | ||
4b5f8d9a | 3168 | slab = folio_slab(folio); |
21c690a3 SB |
3169 | obj_exts = slab_obj_exts(slab); |
3170 | if (!obj_exts) | |
4b5f8d9a VB |
3171 | return NULL; |
3172 | ||
3173 | off = obj_to_index(slab->slab_cache, slab, p); | |
21c690a3 SB |
3174 | if (obj_exts[off].objcg) |
3175 | return obj_cgroup_memcg(obj_exts[off].objcg); | |
10befea9 RG |
3176 | |
3177 | return NULL; | |
9855609b | 3178 | } |
8380ce47 | 3179 | |
bcfe06bf | 3180 | /* |
becacb04 | 3181 | * folio_memcg_check() is used here, because in theory we can encounter |
4b5f8d9a | 3182 | * a folio where the slab flag has been cleared already, but |
21c690a3 | 3183 | * slab->obj_exts has not been freed yet |
becacb04 | 3184 | * folio_memcg_check() will guarantee that a proper memory |
bcfe06bf RG |
3185 | * cgroup pointer or NULL will be returned. |
3186 | */ | |
becacb04 | 3187 | return folio_memcg_check(folio); |
8380ce47 RG |
3188 | } |
3189 | ||
fc4db90f RG |
3190 | /* |
3191 | * Returns a pointer to the memory cgroup to which the kernel object is charged. | |
3192 | * | |
3193 | * A passed kernel object can be a slab object, vmalloc object or a generic | |
3194 | * kernel page, so different mechanisms for getting the memory cgroup pointer | |
3195 | * should be used. | |
3196 | * | |
3197 | * In certain cases (e.g. kernel stacks or large kmallocs with SLUB) the caller | |
3198 | * can not know for sure how the kernel object is implemented. | |
3199 | * mem_cgroup_from_obj() can be safely used in such cases. | |
3200 | * | |
3201 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), | |
3202 | * cgroup_mutex, etc. | |
3203 | */ | |
3204 | struct mem_cgroup *mem_cgroup_from_obj(void *p) | |
3205 | { | |
3206 | struct folio *folio; | |
3207 | ||
3208 | if (mem_cgroup_disabled()) | |
3209 | return NULL; | |
3210 | ||
3211 | if (unlikely(is_vmalloc_addr(p))) | |
3212 | folio = page_folio(vmalloc_to_page(p)); | |
3213 | else | |
3214 | folio = virt_to_folio(p); | |
3215 | ||
3216 | return mem_cgroup_from_obj_folio(folio, p); | |
3217 | } | |
3218 | ||
3219 | /* | |
3220 | * Returns a pointer to the memory cgroup to which the kernel object is charged. | |
3221 | * Similar to mem_cgroup_from_obj(), but faster and not suitable for objects, | |
3222 | * allocated using vmalloc(). | |
3223 | * | |
3224 | * A passed kernel object must be a slab object or a generic kernel page. | |
3225 | * | |
3226 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), | |
3227 | * cgroup_mutex, etc. | |
3228 | */ | |
3229 | struct mem_cgroup *mem_cgroup_from_slab_obj(void *p) | |
3230 | { | |
3231 | if (mem_cgroup_disabled()) | |
3232 | return NULL; | |
3233 | ||
3234 | return mem_cgroup_from_obj_folio(virt_to_folio(p), p); | |
3235 | } | |
3236 | ||
f4840ccf JW |
3237 | static struct obj_cgroup *__get_obj_cgroup_from_memcg(struct mem_cgroup *memcg) |
3238 | { | |
3239 | struct obj_cgroup *objcg = NULL; | |
3240 | ||
7848ed62 | 3241 | for (; !mem_cgroup_is_root(memcg); memcg = parent_mem_cgroup(memcg)) { |
f4840ccf | 3242 | objcg = rcu_dereference(memcg->objcg); |
7d0715d0 | 3243 | if (likely(objcg && obj_cgroup_tryget(objcg))) |
f4840ccf JW |
3244 | break; |
3245 | objcg = NULL; | |
3246 | } | |
3247 | return objcg; | |
3248 | } | |
3249 | ||
1aacbd35 RG |
3250 | static struct obj_cgroup *current_objcg_update(void) |
3251 | { | |
3252 | struct mem_cgroup *memcg; | |
3253 | struct obj_cgroup *old, *objcg = NULL; | |
3254 | ||
3255 | do { | |
3256 | /* Atomically drop the update bit. */ | |
3257 | old = xchg(¤t->objcg, NULL); | |
3258 | if (old) { | |
3259 | old = (struct obj_cgroup *) | |
3260 | ((unsigned long)old & ~CURRENT_OBJCG_UPDATE_FLAG); | |
91b71e78 | 3261 | obj_cgroup_put(old); |
1aacbd35 RG |
3262 | |
3263 | old = NULL; | |
3264 | } | |
3265 | ||
3266 | /* If new objcg is NULL, no reason for the second atomic update. */ | |
3267 | if (!current->mm || (current->flags & PF_KTHREAD)) | |
3268 | return NULL; | |
3269 | ||
3270 | /* | |
3271 | * Release the objcg pointer from the previous iteration, | |
3272 | * if try_cmpxcg() below fails. | |
3273 | */ | |
3274 | if (unlikely(objcg)) { | |
3275 | obj_cgroup_put(objcg); | |
3276 | objcg = NULL; | |
3277 | } | |
3278 | ||
3279 | /* | |
3280 | * Obtain the new objcg pointer. The current task can be | |
3281 | * asynchronously moved to another memcg and the previous | |
3282 | * memcg can be offlined. So let's get the memcg pointer | |
3283 | * and try get a reference to objcg under a rcu read lock. | |
3284 | */ | |
3285 | ||
3286 | rcu_read_lock(); | |
3287 | memcg = mem_cgroup_from_task(current); | |
3288 | objcg = __get_obj_cgroup_from_memcg(memcg); | |
3289 | rcu_read_unlock(); | |
3290 | ||
3291 | /* | |
3292 | * Try set up a new objcg pointer atomically. If it | |
3293 | * fails, it means the update flag was set concurrently, so | |
3294 | * the whole procedure should be repeated. | |
3295 | */ | |
3296 | } while (!try_cmpxchg(¤t->objcg, &old, objcg)); | |
3297 | ||
3298 | return objcg; | |
3299 | } | |
3300 | ||
e86828e5 RG |
3301 | __always_inline struct obj_cgroup *current_obj_cgroup(void) |
3302 | { | |
3303 | struct mem_cgroup *memcg; | |
3304 | struct obj_cgroup *objcg; | |
3305 | ||
3306 | if (in_task()) { | |
3307 | memcg = current->active_memcg; | |
3308 | if (unlikely(memcg)) | |
3309 | goto from_memcg; | |
3310 | ||
3311 | objcg = READ_ONCE(current->objcg); | |
3312 | if (unlikely((unsigned long)objcg & CURRENT_OBJCG_UPDATE_FLAG)) | |
3313 | objcg = current_objcg_update(); | |
3314 | /* | |
3315 | * Objcg reference is kept by the task, so it's safe | |
3316 | * to use the objcg by the current task. | |
3317 | */ | |
3318 | return objcg; | |
3319 | } | |
3320 | ||
3321 | memcg = this_cpu_read(int_active_memcg); | |
3322 | if (unlikely(memcg)) | |
3323 | goto from_memcg; | |
3324 | ||
3325 | return NULL; | |
3326 | ||
3327 | from_memcg: | |
5f79489a | 3328 | objcg = NULL; |
e86828e5 RG |
3329 | for (; !mem_cgroup_is_root(memcg); memcg = parent_mem_cgroup(memcg)) { |
3330 | /* | |
3331 | * Memcg pointer is protected by scope (see set_active_memcg()) | |
3332 | * and is pinning the corresponding objcg, so objcg can't go | |
3333 | * away and can be used within the scope without any additional | |
3334 | * protection. | |
3335 | */ | |
3336 | objcg = rcu_dereference_check(memcg->objcg, 1); | |
3337 | if (likely(objcg)) | |
3338 | break; | |
e86828e5 RG |
3339 | } |
3340 | ||
3341 | return objcg; | |
3342 | } | |
3343 | ||
074e3e26 | 3344 | struct obj_cgroup *get_obj_cgroup_from_folio(struct folio *folio) |
f4840ccf JW |
3345 | { |
3346 | struct obj_cgroup *objcg; | |
3347 | ||
f7a449f7 | 3348 | if (!memcg_kmem_online()) |
f4840ccf JW |
3349 | return NULL; |
3350 | ||
074e3e26 MWO |
3351 | if (folio_memcg_kmem(folio)) { |
3352 | objcg = __folio_objcg(folio); | |
f4840ccf JW |
3353 | obj_cgroup_get(objcg); |
3354 | } else { | |
3355 | struct mem_cgroup *memcg; | |
bf4f0599 | 3356 | |
f4840ccf | 3357 | rcu_read_lock(); |
074e3e26 | 3358 | memcg = __folio_memcg(folio); |
f4840ccf JW |
3359 | if (memcg) |
3360 | objcg = __get_obj_cgroup_from_memcg(memcg); | |
3361 | else | |
3362 | objcg = NULL; | |
3363 | rcu_read_unlock(); | |
3364 | } | |
bf4f0599 RG |
3365 | return objcg; |
3366 | } | |
3367 | ||
a8c49af3 YA |
3368 | static void memcg_account_kmem(struct mem_cgroup *memcg, int nr_pages) |
3369 | { | |
3370 | mod_memcg_state(memcg, MEMCG_KMEM, nr_pages); | |
3371 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { | |
3372 | if (nr_pages > 0) | |
3373 | page_counter_charge(&memcg->kmem, nr_pages); | |
3374 | else | |
3375 | page_counter_uncharge(&memcg->kmem, -nr_pages); | |
3376 | } | |
3377 | } | |
3378 | ||
3379 | ||
f1286fae MS |
3380 | /* |
3381 | * obj_cgroup_uncharge_pages: uncharge a number of kernel pages from a objcg | |
3382 | * @objcg: object cgroup to uncharge | |
3383 | * @nr_pages: number of pages to uncharge | |
3384 | */ | |
e74d2259 MS |
3385 | static void obj_cgroup_uncharge_pages(struct obj_cgroup *objcg, |
3386 | unsigned int nr_pages) | |
3387 | { | |
3388 | struct mem_cgroup *memcg; | |
3389 | ||
3390 | memcg = get_mem_cgroup_from_objcg(objcg); | |
e74d2259 | 3391 | |
a8c49af3 | 3392 | memcg_account_kmem(memcg, -nr_pages); |
f1286fae | 3393 | refill_stock(memcg, nr_pages); |
e74d2259 | 3394 | |
e74d2259 | 3395 | css_put(&memcg->css); |
e74d2259 MS |
3396 | } |
3397 | ||
f1286fae MS |
3398 | /* |
3399 | * obj_cgroup_charge_pages: charge a number of kernel pages to a objcg | |
3400 | * @objcg: object cgroup to charge | |
45264778 | 3401 | * @gfp: reclaim mode |
92d0510c | 3402 | * @nr_pages: number of pages to charge |
45264778 VD |
3403 | * |
3404 | * Returns 0 on success, an error code on failure. | |
3405 | */ | |
f1286fae MS |
3406 | static int obj_cgroup_charge_pages(struct obj_cgroup *objcg, gfp_t gfp, |
3407 | unsigned int nr_pages) | |
7ae1e1d0 | 3408 | { |
f1286fae | 3409 | struct mem_cgroup *memcg; |
7ae1e1d0 GC |
3410 | int ret; |
3411 | ||
f1286fae MS |
3412 | memcg = get_mem_cgroup_from_objcg(objcg); |
3413 | ||
c5c8b16b | 3414 | ret = try_charge_memcg(memcg, gfp, nr_pages); |
52c29b04 | 3415 | if (ret) |
f1286fae | 3416 | goto out; |
52c29b04 | 3417 | |
a8c49af3 | 3418 | memcg_account_kmem(memcg, nr_pages); |
f1286fae MS |
3419 | out: |
3420 | css_put(&memcg->css); | |
4b13f64d | 3421 | |
f1286fae | 3422 | return ret; |
4b13f64d RG |
3423 | } |
3424 | ||
45264778 | 3425 | /** |
f4b00eab | 3426 | * __memcg_kmem_charge_page: charge a kmem page to the current memory cgroup |
45264778 VD |
3427 | * @page: page to charge |
3428 | * @gfp: reclaim mode | |
3429 | * @order: allocation order | |
3430 | * | |
3431 | * Returns 0 on success, an error code on failure. | |
3432 | */ | |
f4b00eab | 3433 | int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 3434 | { |
b4e0b68f | 3435 | struct obj_cgroup *objcg; |
fcff7d7e | 3436 | int ret = 0; |
7ae1e1d0 | 3437 | |
e86828e5 | 3438 | objcg = current_obj_cgroup(); |
b4e0b68f MS |
3439 | if (objcg) { |
3440 | ret = obj_cgroup_charge_pages(objcg, gfp, 1 << order); | |
4d96ba35 | 3441 | if (!ret) { |
e86828e5 | 3442 | obj_cgroup_get(objcg); |
b4e0b68f | 3443 | page->memcg_data = (unsigned long)objcg | |
18b2db3b | 3444 | MEMCG_DATA_KMEM; |
1a3e1f40 | 3445 | return 0; |
4d96ba35 | 3446 | } |
c4159a75 | 3447 | } |
d05e83a6 | 3448 | return ret; |
7ae1e1d0 | 3449 | } |
49a18eae | 3450 | |
45264778 | 3451 | /** |
f4b00eab | 3452 | * __memcg_kmem_uncharge_page: uncharge a kmem page |
45264778 VD |
3453 | * @page: page to uncharge |
3454 | * @order: allocation order | |
3455 | */ | |
f4b00eab | 3456 | void __memcg_kmem_uncharge_page(struct page *page, int order) |
7ae1e1d0 | 3457 | { |
1b7e4464 | 3458 | struct folio *folio = page_folio(page); |
b4e0b68f | 3459 | struct obj_cgroup *objcg; |
f3ccb2c4 | 3460 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 3461 | |
1b7e4464 | 3462 | if (!folio_memcg_kmem(folio)) |
7ae1e1d0 GC |
3463 | return; |
3464 | ||
1b7e4464 | 3465 | objcg = __folio_objcg(folio); |
b4e0b68f | 3466 | obj_cgroup_uncharge_pages(objcg, nr_pages); |
1b7e4464 | 3467 | folio->memcg_data = 0; |
b4e0b68f | 3468 | obj_cgroup_put(objcg); |
60d3fd32 | 3469 | } |
bf4f0599 | 3470 | |
91882c16 | 3471 | static void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat, |
68ac5b3c WL |
3472 | enum node_stat_item idx, int nr) |
3473 | { | |
fead2b86 | 3474 | struct memcg_stock_pcp *stock; |
56751146 | 3475 | struct obj_cgroup *old = NULL; |
68ac5b3c WL |
3476 | unsigned long flags; |
3477 | int *bytes; | |
3478 | ||
56751146 | 3479 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
fead2b86 MH |
3480 | stock = this_cpu_ptr(&memcg_stock); |
3481 | ||
68ac5b3c WL |
3482 | /* |
3483 | * Save vmstat data in stock and skip vmstat array update unless | |
3484 | * accumulating over a page of vmstat data or when pgdat or idx | |
3485 | * changes. | |
3486 | */ | |
3b8abb32 | 3487 | if (READ_ONCE(stock->cached_objcg) != objcg) { |
56751146 | 3488 | old = drain_obj_stock(stock); |
68ac5b3c WL |
3489 | obj_cgroup_get(objcg); |
3490 | stock->nr_bytes = atomic_read(&objcg->nr_charged_bytes) | |
3491 | ? atomic_xchg(&objcg->nr_charged_bytes, 0) : 0; | |
3b8abb32 | 3492 | WRITE_ONCE(stock->cached_objcg, objcg); |
68ac5b3c WL |
3493 | stock->cached_pgdat = pgdat; |
3494 | } else if (stock->cached_pgdat != pgdat) { | |
3495 | /* Flush the existing cached vmstat data */ | |
7fa0dacb WL |
3496 | struct pglist_data *oldpg = stock->cached_pgdat; |
3497 | ||
68ac5b3c | 3498 | if (stock->nr_slab_reclaimable_b) { |
91882c16 | 3499 | __mod_objcg_mlstate(objcg, oldpg, NR_SLAB_RECLAIMABLE_B, |
68ac5b3c WL |
3500 | stock->nr_slab_reclaimable_b); |
3501 | stock->nr_slab_reclaimable_b = 0; | |
3502 | } | |
3503 | if (stock->nr_slab_unreclaimable_b) { | |
91882c16 | 3504 | __mod_objcg_mlstate(objcg, oldpg, NR_SLAB_UNRECLAIMABLE_B, |
68ac5b3c WL |
3505 | stock->nr_slab_unreclaimable_b); |
3506 | stock->nr_slab_unreclaimable_b = 0; | |
3507 | } | |
3508 | stock->cached_pgdat = pgdat; | |
3509 | } | |
3510 | ||
3511 | bytes = (idx == NR_SLAB_RECLAIMABLE_B) ? &stock->nr_slab_reclaimable_b | |
3512 | : &stock->nr_slab_unreclaimable_b; | |
3513 | /* | |
3514 | * Even for large object >= PAGE_SIZE, the vmstat data will still be | |
3515 | * cached locally at least once before pushing it out. | |
3516 | */ | |
3517 | if (!*bytes) { | |
3518 | *bytes = nr; | |
3519 | nr = 0; | |
3520 | } else { | |
3521 | *bytes += nr; | |
3522 | if (abs(*bytes) > PAGE_SIZE) { | |
3523 | nr = *bytes; | |
3524 | *bytes = 0; | |
3525 | } else { | |
3526 | nr = 0; | |
3527 | } | |
3528 | } | |
3529 | if (nr) | |
91882c16 | 3530 | __mod_objcg_mlstate(objcg, pgdat, idx, nr); |
68ac5b3c | 3531 | |
56751146 | 3532 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
91b71e78 | 3533 | obj_cgroup_put(old); |
68ac5b3c WL |
3534 | } |
3535 | ||
bf4f0599 RG |
3536 | static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) |
3537 | { | |
fead2b86 | 3538 | struct memcg_stock_pcp *stock; |
bf4f0599 RG |
3539 | unsigned long flags; |
3540 | bool ret = false; | |
3541 | ||
56751146 | 3542 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
fead2b86 MH |
3543 | |
3544 | stock = this_cpu_ptr(&memcg_stock); | |
3b8abb32 | 3545 | if (objcg == READ_ONCE(stock->cached_objcg) && stock->nr_bytes >= nr_bytes) { |
bf4f0599 RG |
3546 | stock->nr_bytes -= nr_bytes; |
3547 | ret = true; | |
3548 | } | |
3549 | ||
56751146 | 3550 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
bf4f0599 RG |
3551 | |
3552 | return ret; | |
3553 | } | |
3554 | ||
56751146 | 3555 | static struct obj_cgroup *drain_obj_stock(struct memcg_stock_pcp *stock) |
bf4f0599 | 3556 | { |
3b8abb32 | 3557 | struct obj_cgroup *old = READ_ONCE(stock->cached_objcg); |
bf4f0599 RG |
3558 | |
3559 | if (!old) | |
56751146 | 3560 | return NULL; |
bf4f0599 RG |
3561 | |
3562 | if (stock->nr_bytes) { | |
3563 | unsigned int nr_pages = stock->nr_bytes >> PAGE_SHIFT; | |
3564 | unsigned int nr_bytes = stock->nr_bytes & (PAGE_SIZE - 1); | |
3565 | ||
af9a3b69 JW |
3566 | if (nr_pages) { |
3567 | struct mem_cgroup *memcg; | |
3568 | ||
3569 | memcg = get_mem_cgroup_from_objcg(old); | |
3570 | ||
3571 | memcg_account_kmem(memcg, -nr_pages); | |
3572 | __refill_stock(memcg, nr_pages); | |
3573 | ||
3574 | css_put(&memcg->css); | |
3575 | } | |
bf4f0599 RG |
3576 | |
3577 | /* | |
3578 | * The leftover is flushed to the centralized per-memcg value. | |
3579 | * On the next attempt to refill obj stock it will be moved | |
3580 | * to a per-cpu stock (probably, on an other CPU), see | |
3581 | * refill_obj_stock(). | |
3582 | * | |
3583 | * How often it's flushed is a trade-off between the memory | |
3584 | * limit enforcement accuracy and potential CPU contention, | |
3585 | * so it might be changed in the future. | |
3586 | */ | |
3587 | atomic_add(nr_bytes, &old->nr_charged_bytes); | |
3588 | stock->nr_bytes = 0; | |
3589 | } | |
3590 | ||
68ac5b3c WL |
3591 | /* |
3592 | * Flush the vmstat data in current stock | |
3593 | */ | |
3594 | if (stock->nr_slab_reclaimable_b || stock->nr_slab_unreclaimable_b) { | |
3595 | if (stock->nr_slab_reclaimable_b) { | |
91882c16 | 3596 | __mod_objcg_mlstate(old, stock->cached_pgdat, |
68ac5b3c WL |
3597 | NR_SLAB_RECLAIMABLE_B, |
3598 | stock->nr_slab_reclaimable_b); | |
3599 | stock->nr_slab_reclaimable_b = 0; | |
3600 | } | |
3601 | if (stock->nr_slab_unreclaimable_b) { | |
91882c16 | 3602 | __mod_objcg_mlstate(old, stock->cached_pgdat, |
68ac5b3c WL |
3603 | NR_SLAB_UNRECLAIMABLE_B, |
3604 | stock->nr_slab_unreclaimable_b); | |
3605 | stock->nr_slab_unreclaimable_b = 0; | |
3606 | } | |
3607 | stock->cached_pgdat = NULL; | |
3608 | } | |
3609 | ||
3b8abb32 | 3610 | WRITE_ONCE(stock->cached_objcg, NULL); |
56751146 SAS |
3611 | /* |
3612 | * The `old' objects needs to be released by the caller via | |
3613 | * obj_cgroup_put() outside of memcg_stock_pcp::stock_lock. | |
3614 | */ | |
3615 | return old; | |
bf4f0599 RG |
3616 | } |
3617 | ||
3618 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
3619 | struct mem_cgroup *root_memcg) | |
3620 | { | |
3b8abb32 | 3621 | struct obj_cgroup *objcg = READ_ONCE(stock->cached_objcg); |
bf4f0599 RG |
3622 | struct mem_cgroup *memcg; |
3623 | ||
3b8abb32 RG |
3624 | if (objcg) { |
3625 | memcg = obj_cgroup_memcg(objcg); | |
bf4f0599 RG |
3626 | if (memcg && mem_cgroup_is_descendant(memcg, root_memcg)) |
3627 | return true; | |
3628 | } | |
3629 | ||
3630 | return false; | |
3631 | } | |
3632 | ||
5387c904 WL |
3633 | static void refill_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes, |
3634 | bool allow_uncharge) | |
bf4f0599 | 3635 | { |
fead2b86 | 3636 | struct memcg_stock_pcp *stock; |
56751146 | 3637 | struct obj_cgroup *old = NULL; |
bf4f0599 | 3638 | unsigned long flags; |
5387c904 | 3639 | unsigned int nr_pages = 0; |
bf4f0599 | 3640 | |
56751146 | 3641 | local_lock_irqsave(&memcg_stock.stock_lock, flags); |
fead2b86 MH |
3642 | |
3643 | stock = this_cpu_ptr(&memcg_stock); | |
3b8abb32 | 3644 | if (READ_ONCE(stock->cached_objcg) != objcg) { /* reset if necessary */ |
56751146 | 3645 | old = drain_obj_stock(stock); |
bf4f0599 | 3646 | obj_cgroup_get(objcg); |
3b8abb32 | 3647 | WRITE_ONCE(stock->cached_objcg, objcg); |
5387c904 WL |
3648 | stock->nr_bytes = atomic_read(&objcg->nr_charged_bytes) |
3649 | ? atomic_xchg(&objcg->nr_charged_bytes, 0) : 0; | |
3650 | allow_uncharge = true; /* Allow uncharge when objcg changes */ | |
bf4f0599 RG |
3651 | } |
3652 | stock->nr_bytes += nr_bytes; | |
3653 | ||
5387c904 WL |
3654 | if (allow_uncharge && (stock->nr_bytes > PAGE_SIZE)) { |
3655 | nr_pages = stock->nr_bytes >> PAGE_SHIFT; | |
3656 | stock->nr_bytes &= (PAGE_SIZE - 1); | |
3657 | } | |
bf4f0599 | 3658 | |
56751146 | 3659 | local_unlock_irqrestore(&memcg_stock.stock_lock, flags); |
91b71e78 | 3660 | obj_cgroup_put(old); |
5387c904 WL |
3661 | |
3662 | if (nr_pages) | |
3663 | obj_cgroup_uncharge_pages(objcg, nr_pages); | |
bf4f0599 RG |
3664 | } |
3665 | ||
3666 | int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size) | |
3667 | { | |
bf4f0599 RG |
3668 | unsigned int nr_pages, nr_bytes; |
3669 | int ret; | |
3670 | ||
3671 | if (consume_obj_stock(objcg, size)) | |
3672 | return 0; | |
3673 | ||
3674 | /* | |
5387c904 | 3675 | * In theory, objcg->nr_charged_bytes can have enough |
bf4f0599 | 3676 | * pre-charged bytes to satisfy the allocation. However, |
5387c904 WL |
3677 | * flushing objcg->nr_charged_bytes requires two atomic |
3678 | * operations, and objcg->nr_charged_bytes can't be big. | |
3679 | * The shared objcg->nr_charged_bytes can also become a | |
3680 | * performance bottleneck if all tasks of the same memcg are | |
3681 | * trying to update it. So it's better to ignore it and try | |
3682 | * grab some new pages. The stock's nr_bytes will be flushed to | |
3683 | * objcg->nr_charged_bytes later on when objcg changes. | |
3684 | * | |
3685 | * The stock's nr_bytes may contain enough pre-charged bytes | |
3686 | * to allow one less page from being charged, but we can't rely | |
3687 | * on the pre-charged bytes not being changed outside of | |
3688 | * consume_obj_stock() or refill_obj_stock(). So ignore those | |
3689 | * pre-charged bytes as well when charging pages. To avoid a | |
3690 | * page uncharge right after a page charge, we set the | |
3691 | * allow_uncharge flag to false when calling refill_obj_stock() | |
3692 | * to temporarily allow the pre-charged bytes to exceed the page | |
3693 | * size limit. The maximum reachable value of the pre-charged | |
3694 | * bytes is (sizeof(object) + PAGE_SIZE - 2) if there is no data | |
3695 | * race. | |
bf4f0599 | 3696 | */ |
bf4f0599 RG |
3697 | nr_pages = size >> PAGE_SHIFT; |
3698 | nr_bytes = size & (PAGE_SIZE - 1); | |
3699 | ||
3700 | if (nr_bytes) | |
3701 | nr_pages += 1; | |
3702 | ||
e74d2259 | 3703 | ret = obj_cgroup_charge_pages(objcg, gfp, nr_pages); |
bf4f0599 | 3704 | if (!ret && nr_bytes) |
5387c904 | 3705 | refill_obj_stock(objcg, PAGE_SIZE - nr_bytes, false); |
bf4f0599 | 3706 | |
bf4f0599 RG |
3707 | return ret; |
3708 | } | |
3709 | ||
3710 | void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size) | |
3711 | { | |
5387c904 | 3712 | refill_obj_stock(objcg, size, true); |
bf4f0599 RG |
3713 | } |
3714 | ||
e6100a45 VB |
3715 | static inline size_t obj_full_size(struct kmem_cache *s) |
3716 | { | |
3717 | /* | |
3718 | * For each accounted object there is an extra space which is used | |
3719 | * to store obj_cgroup membership. Charge it too. | |
3720 | */ | |
3721 | return s->size + sizeof(struct obj_cgroup *); | |
3722 | } | |
3723 | ||
3724 | bool __memcg_slab_post_alloc_hook(struct kmem_cache *s, struct list_lru *lru, | |
3725 | gfp_t flags, size_t size, void **p) | |
3726 | { | |
3727 | struct obj_cgroup *objcg; | |
3728 | struct slab *slab; | |
3729 | unsigned long off; | |
3730 | size_t i; | |
3731 | ||
3732 | /* | |
3733 | * The obtained objcg pointer is safe to use within the current scope, | |
3734 | * defined by current task or set_active_memcg() pair. | |
3735 | * obj_cgroup_get() is used to get a permanent reference. | |
3736 | */ | |
3737 | objcg = current_obj_cgroup(); | |
3738 | if (!objcg) | |
3739 | return true; | |
3740 | ||
3741 | /* | |
3742 | * slab_alloc_node() avoids the NULL check, so we might be called with a | |
3743 | * single NULL object. kmem_cache_alloc_bulk() aborts if it can't fill | |
3744 | * the whole requested size. | |
3745 | * return success as there's nothing to free back | |
3746 | */ | |
3747 | if (unlikely(*p == NULL)) | |
3748 | return true; | |
3749 | ||
3750 | flags &= gfp_allowed_mask; | |
3751 | ||
3752 | if (lru) { | |
3753 | int ret; | |
3754 | struct mem_cgroup *memcg; | |
3755 | ||
3756 | memcg = get_mem_cgroup_from_objcg(objcg); | |
3757 | ret = memcg_list_lru_alloc(memcg, lru, flags); | |
3758 | css_put(&memcg->css); | |
3759 | ||
3760 | if (ret) | |
3761 | return false; | |
3762 | } | |
3763 | ||
3764 | if (obj_cgroup_charge(objcg, flags, size * obj_full_size(s))) | |
3765 | return false; | |
3766 | ||
3767 | for (i = 0; i < size; i++) { | |
3768 | slab = virt_to_slab(p[i]); | |
3769 | ||
3770 | if (!slab_obj_exts(slab) && | |
3771 | alloc_slab_obj_exts(slab, s, flags, false)) { | |
3772 | obj_cgroup_uncharge(objcg, obj_full_size(s)); | |
3773 | continue; | |
3774 | } | |
3775 | ||
3776 | off = obj_to_index(s, slab, p[i]); | |
3777 | obj_cgroup_get(objcg); | |
3778 | slab_obj_exts(slab)[off].objcg = objcg; | |
3779 | mod_objcg_state(objcg, slab_pgdat(slab), | |
3780 | cache_vmstat_idx(s), obj_full_size(s)); | |
3781 | } | |
3782 | ||
3783 | return true; | |
3784 | } | |
3785 | ||
3786 | void __memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab, | |
3787 | void **p, int objects, struct slabobj_ext *obj_exts) | |
3788 | { | |
3789 | for (int i = 0; i < objects; i++) { | |
3790 | struct obj_cgroup *objcg; | |
3791 | unsigned int off; | |
3792 | ||
3793 | off = obj_to_index(s, slab, p[i]); | |
3794 | objcg = obj_exts[off].objcg; | |
3795 | if (!objcg) | |
3796 | continue; | |
3797 | ||
3798 | obj_exts[off].objcg = NULL; | |
3799 | obj_cgroup_uncharge(objcg, obj_full_size(s)); | |
3800 | mod_objcg_state(objcg, slab_pgdat(slab), cache_vmstat_idx(s), | |
3801 | -obj_full_size(s)); | |
3802 | obj_cgroup_put(objcg); | |
3803 | } | |
3804 | } | |
84c07d11 | 3805 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 3806 | |
ca3e0214 | 3807 | /* |
be6c8982 | 3808 | * Because page_memcg(head) is not set on tails, set it now. |
ca3e0214 | 3809 | */ |
b8791381 | 3810 | void split_page_memcg(struct page *head, int old_order, int new_order) |
ca3e0214 | 3811 | { |
1b7e4464 MWO |
3812 | struct folio *folio = page_folio(head); |
3813 | struct mem_cgroup *memcg = folio_memcg(folio); | |
e94c8a9c | 3814 | int i; |
b8791381 ZY |
3815 | unsigned int old_nr = 1 << old_order; |
3816 | unsigned int new_nr = 1 << new_order; | |
ca3e0214 | 3817 | |
be6c8982 | 3818 | if (mem_cgroup_disabled() || !memcg) |
3d37c4a9 | 3819 | return; |
b070e65c | 3820 | |
b8791381 | 3821 | for (i = new_nr; i < old_nr; i += new_nr) |
1b7e4464 | 3822 | folio_page(folio, i)->memcg_data = folio->memcg_data; |
b4e0b68f | 3823 | |
1b7e4464 | 3824 | if (folio_memcg_kmem(folio)) |
b8791381 | 3825 | obj_cgroup_get_many(__folio_objcg(folio), old_nr / new_nr - 1); |
b4e0b68f | 3826 | else |
b8791381 | 3827 | css_get_many(&memcg->css, old_nr / new_nr - 1); |
ca3e0214 | 3828 | } |
ca3e0214 | 3829 | |
e55b9f96 | 3830 | #ifdef CONFIG_SWAP |
02491447 DN |
3831 | /** |
3832 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3833 | * @entry: swap entry to be moved | |
3834 | * @from: mem_cgroup which the entry is moved from | |
3835 | * @to: mem_cgroup which the entry is moved to | |
3836 | * | |
3837 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3838 | * as the mem_cgroup's id of @from. | |
3839 | * | |
3840 | * Returns 0 on success, -EINVAL on failure. | |
3841 | * | |
3e32cb2e | 3842 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
3843 | * both res and memsw, and called css_get(). |
3844 | */ | |
3845 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3846 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3847 | { |
3848 | unsigned short old_id, new_id; | |
3849 | ||
34c00c31 LZ |
3850 | old_id = mem_cgroup_id(from); |
3851 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3852 | |
3853 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
3854 | mod_memcg_state(from, MEMCG_SWAP, -1); |
3855 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
3856 | return 0; |
3857 | } | |
3858 | return -EINVAL; | |
3859 | } | |
3860 | #else | |
3861 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3862 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3863 | { |
3864 | return -EINVAL; | |
3865 | } | |
8c7c6e34 | 3866 | #endif |
d13d1443 | 3867 | |
bbec2e15 | 3868 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 3869 | |
bbec2e15 RG |
3870 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
3871 | unsigned long max, bool memsw) | |
628f4235 | 3872 | { |
3e32cb2e | 3873 | bool enlarge = false; |
bb4a7ea2 | 3874 | bool drained = false; |
3e32cb2e | 3875 | int ret; |
c054a78c YZ |
3876 | bool limits_invariant; |
3877 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 3878 | |
3e32cb2e | 3879 | do { |
628f4235 KH |
3880 | if (signal_pending(current)) { |
3881 | ret = -EINTR; | |
3882 | break; | |
3883 | } | |
3e32cb2e | 3884 | |
bbec2e15 | 3885 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
3886 | /* |
3887 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 3888 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 3889 | */ |
15b42562 | 3890 | limits_invariant = memsw ? max >= READ_ONCE(memcg->memory.max) : |
bbec2e15 | 3891 | max <= memcg->memsw.max; |
c054a78c | 3892 | if (!limits_invariant) { |
bbec2e15 | 3893 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 3894 | ret = -EINVAL; |
8c7c6e34 KH |
3895 | break; |
3896 | } | |
bbec2e15 | 3897 | if (max > counter->max) |
3e32cb2e | 3898 | enlarge = true; |
bbec2e15 RG |
3899 | ret = page_counter_set_max(counter, max); |
3900 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
3901 | |
3902 | if (!ret) | |
3903 | break; | |
3904 | ||
bb4a7ea2 SB |
3905 | if (!drained) { |
3906 | drain_all_stock(memcg); | |
3907 | drained = true; | |
3908 | continue; | |
3909 | } | |
3910 | ||
73b73bac | 3911 | if (!try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, |
55ab834a | 3912 | memsw ? 0 : MEMCG_RECLAIM_MAY_SWAP)) { |
1ab5c056 AR |
3913 | ret = -EBUSY; |
3914 | break; | |
3915 | } | |
3916 | } while (true); | |
3e32cb2e | 3917 | |
3c11ecf4 KH |
3918 | if (!ret && enlarge) |
3919 | memcg_oom_recover(memcg); | |
3e32cb2e | 3920 | |
628f4235 KH |
3921 | return ret; |
3922 | } | |
3923 | ||
ef8f2327 | 3924 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
3925 | gfp_t gfp_mask, |
3926 | unsigned long *total_scanned) | |
3927 | { | |
3928 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 3929 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
3930 | unsigned long reclaimed; |
3931 | int loop = 0; | |
ef8f2327 | 3932 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 3933 | unsigned long excess; |
0608f43d | 3934 | |
e4dde56c YZ |
3935 | if (lru_gen_enabled()) |
3936 | return 0; | |
3937 | ||
0608f43d AM |
3938 | if (order > 0) |
3939 | return 0; | |
3940 | ||
2ab082ba | 3941 | mctz = soft_limit_tree.rb_tree_per_node[pgdat->node_id]; |
d6507ff5 MH |
3942 | |
3943 | /* | |
3944 | * Do not even bother to check the largest node if the root | |
3945 | * is empty. Do it lockless to prevent lock bouncing. Races | |
3946 | * are acceptable as soft limit is best effort anyway. | |
3947 | */ | |
bfc7228b | 3948 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
3949 | return 0; |
3950 | ||
0608f43d AM |
3951 | /* |
3952 | * This loop can run a while, specially if mem_cgroup's continuously | |
3953 | * keep exceeding their soft limit and putting the system under | |
3954 | * pressure | |
3955 | */ | |
3956 | do { | |
3957 | if (next_mz) | |
3958 | mz = next_mz; | |
3959 | else | |
3960 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3961 | if (!mz) | |
3962 | break; | |
3963 | ||
ef8f2327 | 3964 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
d8f65338 | 3965 | gfp_mask, total_scanned); |
0608f43d | 3966 | nr_reclaimed += reclaimed; |
0a31bc97 | 3967 | spin_lock_irq(&mctz->lock); |
0608f43d AM |
3968 | |
3969 | /* | |
3970 | * If we failed to reclaim anything from this memory cgroup | |
3971 | * it is time to move on to the next cgroup | |
3972 | */ | |
3973 | next_mz = NULL; | |
bc2f2e7f VD |
3974 | if (!reclaimed) |
3975 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3976 | ||
3e32cb2e | 3977 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3978 | /* |
3979 | * One school of thought says that we should not add | |
3980 | * back the node to the tree if reclaim returns 0. | |
3981 | * But our reclaim could return 0, simply because due | |
3982 | * to priority we are exposing a smaller subset of | |
3983 | * memory to reclaim from. Consider this as a longer | |
3984 | * term TODO. | |
3985 | */ | |
3986 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3987 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3988 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3989 | css_put(&mz->memcg->css); |
3990 | loop++; | |
3991 | /* | |
3992 | * Could not reclaim anything and there are no more | |
3993 | * mem cgroups to try or we seem to be looping without | |
3994 | * reclaiming anything. | |
3995 | */ | |
3996 | if (!nr_reclaimed && | |
3997 | (next_mz == NULL || | |
3998 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3999 | break; | |
4000 | } while (!nr_reclaimed); | |
4001 | if (next_mz) | |
4002 | css_put(&next_mz->memcg->css); | |
4003 | return nr_reclaimed; | |
4004 | } | |
4005 | ||
c26251f9 | 4006 | /* |
51038171 | 4007 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
4008 | * |
4009 | * Caller is responsible for holding css reference for memcg. | |
4010 | */ | |
4011 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
4012 | { | |
d977aa93 | 4013 | int nr_retries = MAX_RECLAIM_RETRIES; |
c26251f9 | 4014 | |
c1e862c1 KH |
4015 | /* we call try-to-free pages for make this cgroup empty */ |
4016 | lru_add_drain_all(); | |
d12c60f6 JS |
4017 | |
4018 | drain_all_stock(memcg); | |
4019 | ||
f817ed48 | 4020 | /* try to free all pages in this cgroup */ |
3e32cb2e | 4021 | while (nr_retries && page_counter_read(&memcg->memory)) { |
c26251f9 MH |
4022 | if (signal_pending(current)) |
4023 | return -EINTR; | |
4024 | ||
73b73bac | 4025 | if (!try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, |
55ab834a | 4026 | MEMCG_RECLAIM_MAY_SWAP)) |
f817ed48 | 4027 | nr_retries--; |
f817ed48 | 4028 | } |
ab5196c2 MH |
4029 | |
4030 | return 0; | |
cc847582 KH |
4031 | } |
4032 | ||
6770c64e TH |
4033 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
4034 | char *buf, size_t nbytes, | |
4035 | loff_t off) | |
c1e862c1 | 4036 | { |
6770c64e | 4037 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 4038 | |
d8423011 MH |
4039 | if (mem_cgroup_is_root(memcg)) |
4040 | return -EINVAL; | |
6770c64e | 4041 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
4042 | } |
4043 | ||
182446d0 TH |
4044 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
4045 | struct cftype *cft) | |
18f59ea7 | 4046 | { |
bef8620c | 4047 | return 1; |
18f59ea7 BS |
4048 | } |
4049 | ||
182446d0 TH |
4050 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
4051 | struct cftype *cft, u64 val) | |
18f59ea7 | 4052 | { |
bef8620c | 4053 | if (val == 1) |
0b8f73e1 | 4054 | return 0; |
567fb435 | 4055 | |
bef8620c RG |
4056 | pr_warn_once("Non-hierarchical mode is deprecated. " |
4057 | "Please report your usecase to linux-mm@kvack.org if you " | |
4058 | "depend on this functionality.\n"); | |
567fb435 | 4059 | |
bef8620c | 4060 | return -EINVAL; |
18f59ea7 BS |
4061 | } |
4062 | ||
6f646156 | 4063 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 4064 | { |
42a30035 | 4065 | unsigned long val; |
ce00a967 | 4066 | |
3e32cb2e | 4067 | if (mem_cgroup_is_root(memcg)) { |
a2174e95 | 4068 | /* |
f82a7a86 YA |
4069 | * Approximate root's usage from global state. This isn't |
4070 | * perfect, but the root usage was always an approximation. | |
a2174e95 | 4071 | */ |
f82a7a86 YA |
4072 | val = global_node_page_state(NR_FILE_PAGES) + |
4073 | global_node_page_state(NR_ANON_MAPPED); | |
42a30035 | 4074 | if (swap) |
f82a7a86 | 4075 | val += total_swap_pages - get_nr_swap_pages(); |
3e32cb2e | 4076 | } else { |
ce00a967 | 4077 | if (!swap) |
3e32cb2e | 4078 | val = page_counter_read(&memcg->memory); |
ce00a967 | 4079 | else |
3e32cb2e | 4080 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 4081 | } |
c12176d3 | 4082 | return val; |
ce00a967 JW |
4083 | } |
4084 | ||
3e32cb2e JW |
4085 | enum { |
4086 | RES_USAGE, | |
4087 | RES_LIMIT, | |
4088 | RES_MAX_USAGE, | |
4089 | RES_FAILCNT, | |
4090 | RES_SOFT_LIMIT, | |
4091 | }; | |
ce00a967 | 4092 | |
791badbd | 4093 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 4094 | struct cftype *cft) |
8cdea7c0 | 4095 | { |
182446d0 | 4096 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 4097 | struct page_counter *counter; |
af36f906 | 4098 | |
3e32cb2e | 4099 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 4100 | case _MEM: |
3e32cb2e JW |
4101 | counter = &memcg->memory; |
4102 | break; | |
8c7c6e34 | 4103 | case _MEMSWAP: |
3e32cb2e JW |
4104 | counter = &memcg->memsw; |
4105 | break; | |
510fc4e1 | 4106 | case _KMEM: |
3e32cb2e | 4107 | counter = &memcg->kmem; |
510fc4e1 | 4108 | break; |
d55f90bf | 4109 | case _TCP: |
0db15298 | 4110 | counter = &memcg->tcpmem; |
d55f90bf | 4111 | break; |
8c7c6e34 KH |
4112 | default: |
4113 | BUG(); | |
8c7c6e34 | 4114 | } |
3e32cb2e JW |
4115 | |
4116 | switch (MEMFILE_ATTR(cft->private)) { | |
4117 | case RES_USAGE: | |
4118 | if (counter == &memcg->memory) | |
c12176d3 | 4119 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 4120 | if (counter == &memcg->memsw) |
c12176d3 | 4121 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
4122 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
4123 | case RES_LIMIT: | |
bbec2e15 | 4124 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
4125 | case RES_MAX_USAGE: |
4126 | return (u64)counter->watermark * PAGE_SIZE; | |
4127 | case RES_FAILCNT: | |
4128 | return counter->failcnt; | |
4129 | case RES_SOFT_LIMIT: | |
2178e20c | 4130 | return (u64)READ_ONCE(memcg->soft_limit) * PAGE_SIZE; |
3e32cb2e JW |
4131 | default: |
4132 | BUG(); | |
4133 | } | |
8cdea7c0 | 4134 | } |
510fc4e1 | 4135 | |
6b0ba2ab FS |
4136 | /* |
4137 | * This function doesn't do anything useful. Its only job is to provide a read | |
4138 | * handler for a file so that cgroup_file_mode() will add read permissions. | |
4139 | */ | |
4140 | static int mem_cgroup_dummy_seq_show(__always_unused struct seq_file *m, | |
4141 | __always_unused void *v) | |
4142 | { | |
4143 | return -EINVAL; | |
4144 | } | |
4145 | ||
84c07d11 | 4146 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 4147 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 4148 | { |
bf4f0599 | 4149 | struct obj_cgroup *objcg; |
d6441637 | 4150 | |
9c94bef9 | 4151 | if (mem_cgroup_kmem_disabled()) |
b313aeee VD |
4152 | return 0; |
4153 | ||
da0efe30 MS |
4154 | if (unlikely(mem_cgroup_is_root(memcg))) |
4155 | return 0; | |
d6441637 | 4156 | |
bf4f0599 | 4157 | objcg = obj_cgroup_alloc(); |
f9c69d63 | 4158 | if (!objcg) |
bf4f0599 | 4159 | return -ENOMEM; |
f9c69d63 | 4160 | |
bf4f0599 RG |
4161 | objcg->memcg = memcg; |
4162 | rcu_assign_pointer(memcg->objcg, objcg); | |
675d6c9b RG |
4163 | obj_cgroup_get(objcg); |
4164 | memcg->orig_objcg = objcg; | |
bf4f0599 | 4165 | |
f7a449f7 | 4166 | static_branch_enable(&memcg_kmem_online_key); |
d648bcc7 | 4167 | |
f9c69d63 | 4168 | memcg->kmemcg_id = memcg->id.id; |
0b8f73e1 JW |
4169 | |
4170 | return 0; | |
d6441637 VD |
4171 | } |
4172 | ||
8e0a8912 JW |
4173 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
4174 | { | |
64268868 | 4175 | struct mem_cgroup *parent; |
8e0a8912 | 4176 | |
9c94bef9 | 4177 | if (mem_cgroup_kmem_disabled()) |
da0efe30 MS |
4178 | return; |
4179 | ||
4180 | if (unlikely(mem_cgroup_is_root(memcg))) | |
8e0a8912 | 4181 | return; |
9855609b | 4182 | |
8e0a8912 JW |
4183 | parent = parent_mem_cgroup(memcg); |
4184 | if (!parent) | |
4185 | parent = root_mem_cgroup; | |
4186 | ||
bf4f0599 | 4187 | memcg_reparent_objcgs(memcg, parent); |
fb2f2b0a | 4188 | |
8e0a8912 | 4189 | /* |
64268868 MS |
4190 | * After we have finished memcg_reparent_objcgs(), all list_lrus |
4191 | * corresponding to this cgroup are guaranteed to remain empty. | |
4192 | * The ordering is imposed by list_lru_node->lock taken by | |
1f391eb2 | 4193 | * memcg_reparent_list_lrus(). |
8e0a8912 | 4194 | */ |
1f391eb2 | 4195 | memcg_reparent_list_lrus(memcg, parent); |
8e0a8912 | 4196 | } |
d6441637 | 4197 | #else |
0b8f73e1 | 4198 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
4199 | { |
4200 | return 0; | |
4201 | } | |
4202 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
4203 | { | |
4204 | } | |
84c07d11 | 4205 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 4206 | |
bbec2e15 | 4207 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
4208 | { |
4209 | int ret; | |
4210 | ||
bbec2e15 | 4211 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 4212 | |
bbec2e15 | 4213 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
4214 | if (ret) |
4215 | goto out; | |
4216 | ||
0db15298 | 4217 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
4218 | /* |
4219 | * The active flag needs to be written after the static_key | |
4220 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
4221 | * function is the last one to run. See mem_cgroup_sk_alloc() |
4222 | * for details, and note that we don't mark any socket as | |
4223 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
4224 | * |
4225 | * We need to do this, because static_keys will span multiple | |
4226 | * sites, but we can't control their order. If we mark a socket | |
4227 | * as accounted, but the accounting functions are not patched in | |
4228 | * yet, we'll lose accounting. | |
4229 | * | |
2d758073 | 4230 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
4231 | * because when this value change, the code to process it is not |
4232 | * patched in yet. | |
4233 | */ | |
4234 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 4235 | memcg->tcpmem_active = true; |
d55f90bf VD |
4236 | } |
4237 | out: | |
bbec2e15 | 4238 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
4239 | return ret; |
4240 | } | |
d55f90bf | 4241 | |
628f4235 KH |
4242 | /* |
4243 | * The user of this function is... | |
4244 | * RES_LIMIT. | |
4245 | */ | |
451af504 TH |
4246 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
4247 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 4248 | { |
451af504 | 4249 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 4250 | unsigned long nr_pages; |
628f4235 KH |
4251 | int ret; |
4252 | ||
451af504 | 4253 | buf = strstrip(buf); |
650c5e56 | 4254 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
4255 | if (ret) |
4256 | return ret; | |
af36f906 | 4257 | |
3e32cb2e | 4258 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 4259 | case RES_LIMIT: |
4b3bde4c BS |
4260 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
4261 | ret = -EINVAL; | |
4262 | break; | |
4263 | } | |
3e32cb2e JW |
4264 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
4265 | case _MEM: | |
bbec2e15 | 4266 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 4267 | break; |
3e32cb2e | 4268 | case _MEMSWAP: |
bbec2e15 | 4269 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 4270 | break; |
4597648f MH |
4271 | case _KMEM: |
4272 | pr_warn_once("kmem.limit_in_bytes is deprecated and will be removed. " | |
4273 | "Writing any value to this file has no effect. " | |
4274 | "Please report your usecase to linux-mm@kvack.org if you " | |
4275 | "depend on this functionality.\n"); | |
4276 | ret = 0; | |
4277 | break; | |
d55f90bf | 4278 | case _TCP: |
bbec2e15 | 4279 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 4280 | break; |
3e32cb2e | 4281 | } |
296c81d8 | 4282 | break; |
3e32cb2e | 4283 | case RES_SOFT_LIMIT: |
2343e88d SAS |
4284 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) { |
4285 | ret = -EOPNOTSUPP; | |
4286 | } else { | |
2178e20c | 4287 | WRITE_ONCE(memcg->soft_limit, nr_pages); |
2343e88d SAS |
4288 | ret = 0; |
4289 | } | |
628f4235 KH |
4290 | break; |
4291 | } | |
451af504 | 4292 | return ret ?: nbytes; |
8cdea7c0 BS |
4293 | } |
4294 | ||
6770c64e TH |
4295 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
4296 | size_t nbytes, loff_t off) | |
c84872e1 | 4297 | { |
6770c64e | 4298 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 4299 | struct page_counter *counter; |
c84872e1 | 4300 | |
3e32cb2e JW |
4301 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
4302 | case _MEM: | |
4303 | counter = &memcg->memory; | |
4304 | break; | |
4305 | case _MEMSWAP: | |
4306 | counter = &memcg->memsw; | |
4307 | break; | |
4308 | case _KMEM: | |
4309 | counter = &memcg->kmem; | |
4310 | break; | |
d55f90bf | 4311 | case _TCP: |
0db15298 | 4312 | counter = &memcg->tcpmem; |
d55f90bf | 4313 | break; |
3e32cb2e JW |
4314 | default: |
4315 | BUG(); | |
4316 | } | |
af36f906 | 4317 | |
3e32cb2e | 4318 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 4319 | case RES_MAX_USAGE: |
3e32cb2e | 4320 | page_counter_reset_watermark(counter); |
29f2a4da PE |
4321 | break; |
4322 | case RES_FAILCNT: | |
3e32cb2e | 4323 | counter->failcnt = 0; |
29f2a4da | 4324 | break; |
3e32cb2e JW |
4325 | default: |
4326 | BUG(); | |
29f2a4da | 4327 | } |
f64c3f54 | 4328 | |
6770c64e | 4329 | return nbytes; |
c84872e1 PE |
4330 | } |
4331 | ||
182446d0 | 4332 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
4333 | struct cftype *cft) |
4334 | { | |
182446d0 | 4335 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
4336 | } |
4337 | ||
02491447 | 4338 | #ifdef CONFIG_MMU |
182446d0 | 4339 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
4340 | struct cftype *cft, u64 val) |
4341 | { | |
182446d0 | 4342 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 4343 | |
da34a848 JW |
4344 | pr_warn_once("Cgroup memory moving (move_charge_at_immigrate) is deprecated. " |
4345 | "Please report your usecase to linux-mm@kvack.org if you " | |
4346 | "depend on this functionality.\n"); | |
4347 | ||
1dfab5ab | 4348 | if (val & ~MOVE_MASK) |
7dc74be0 | 4349 | return -EINVAL; |
ee5e8472 | 4350 | |
7dc74be0 | 4351 | /* |
ee5e8472 GC |
4352 | * No kind of locking is needed in here, because ->can_attach() will |
4353 | * check this value once in the beginning of the process, and then carry | |
4354 | * on with stale data. This means that changes to this value will only | |
4355 | * affect task migrations starting after the change. | |
7dc74be0 | 4356 | */ |
c0ff4b85 | 4357 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
4358 | return 0; |
4359 | } | |
02491447 | 4360 | #else |
182446d0 | 4361 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
4362 | struct cftype *cft, u64 val) |
4363 | { | |
4364 | return -ENOSYS; | |
4365 | } | |
4366 | #endif | |
7dc74be0 | 4367 | |
406eb0c9 | 4368 | #ifdef CONFIG_NUMA |
113b7dfd JW |
4369 | |
4370 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
4371 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
4372 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) | |
4373 | ||
4374 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 | 4375 | int nid, unsigned int lru_mask, bool tree) |
113b7dfd | 4376 | { |
867e5e1d | 4377 | struct lruvec *lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); |
113b7dfd JW |
4378 | unsigned long nr = 0; |
4379 | enum lru_list lru; | |
4380 | ||
4381 | VM_BUG_ON((unsigned)nid >= nr_node_ids); | |
4382 | ||
4383 | for_each_lru(lru) { | |
4384 | if (!(BIT(lru) & lru_mask)) | |
4385 | continue; | |
dd8657b6 SB |
4386 | if (tree) |
4387 | nr += lruvec_page_state(lruvec, NR_LRU_BASE + lru); | |
4388 | else | |
4389 | nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru); | |
113b7dfd JW |
4390 | } |
4391 | return nr; | |
4392 | } | |
4393 | ||
4394 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 SB |
4395 | unsigned int lru_mask, |
4396 | bool tree) | |
113b7dfd JW |
4397 | { |
4398 | unsigned long nr = 0; | |
4399 | enum lru_list lru; | |
4400 | ||
4401 | for_each_lru(lru) { | |
4402 | if (!(BIT(lru) & lru_mask)) | |
4403 | continue; | |
dd8657b6 SB |
4404 | if (tree) |
4405 | nr += memcg_page_state(memcg, NR_LRU_BASE + lru); | |
4406 | else | |
4407 | nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru); | |
113b7dfd JW |
4408 | } |
4409 | return nr; | |
4410 | } | |
4411 | ||
2da8ca82 | 4412 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 4413 | { |
25485de6 GT |
4414 | struct numa_stat { |
4415 | const char *name; | |
4416 | unsigned int lru_mask; | |
4417 | }; | |
4418 | ||
4419 | static const struct numa_stat stats[] = { | |
4420 | { "total", LRU_ALL }, | |
4421 | { "file", LRU_ALL_FILE }, | |
4422 | { "anon", LRU_ALL_ANON }, | |
4423 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
4424 | }; | |
4425 | const struct numa_stat *stat; | |
406eb0c9 | 4426 | int nid; |
aa9694bb | 4427 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
406eb0c9 | 4428 | |
7d7ef0a4 | 4429 | mem_cgroup_flush_stats(memcg); |
2d146aa3 | 4430 | |
25485de6 | 4431 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
4432 | seq_printf(m, "%s=%lu", stat->name, |
4433 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
4434 | false)); | |
4435 | for_each_node_state(nid, N_MEMORY) | |
4436 | seq_printf(m, " N%d=%lu", nid, | |
4437 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
4438 | stat->lru_mask, false)); | |
25485de6 | 4439 | seq_putc(m, '\n'); |
406eb0c9 | 4440 | } |
406eb0c9 | 4441 | |
071aee13 | 4442 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
4443 | |
4444 | seq_printf(m, "hierarchical_%s=%lu", stat->name, | |
4445 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
4446 | true)); | |
4447 | for_each_node_state(nid, N_MEMORY) | |
4448 | seq_printf(m, " N%d=%lu", nid, | |
4449 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
4450 | stat->lru_mask, true)); | |
071aee13 | 4451 | seq_putc(m, '\n'); |
406eb0c9 | 4452 | } |
406eb0c9 | 4453 | |
406eb0c9 YH |
4454 | return 0; |
4455 | } | |
4456 | #endif /* CONFIG_NUMA */ | |
4457 | ||
c8713d0b | 4458 | static const unsigned int memcg1_stats[] = { |
0d1c2072 | 4459 | NR_FILE_PAGES, |
be5d0a74 | 4460 | NR_ANON_MAPPED, |
468c3982 JW |
4461 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4462 | NR_ANON_THPS, | |
4463 | #endif | |
c8713d0b JW |
4464 | NR_SHMEM, |
4465 | NR_FILE_MAPPED, | |
4466 | NR_FILE_DIRTY, | |
4467 | NR_WRITEBACK, | |
e09b0b61 YS |
4468 | WORKINGSET_REFAULT_ANON, |
4469 | WORKINGSET_REFAULT_FILE, | |
72a14e82 | 4470 | #ifdef CONFIG_SWAP |
c8713d0b | 4471 | MEMCG_SWAP, |
72a14e82 LS |
4472 | NR_SWAPCACHE, |
4473 | #endif | |
c8713d0b JW |
4474 | }; |
4475 | ||
4476 | static const char *const memcg1_stat_names[] = { | |
4477 | "cache", | |
4478 | "rss", | |
468c3982 | 4479 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
c8713d0b | 4480 | "rss_huge", |
468c3982 | 4481 | #endif |
c8713d0b JW |
4482 | "shmem", |
4483 | "mapped_file", | |
4484 | "dirty", | |
4485 | "writeback", | |
e09b0b61 YS |
4486 | "workingset_refault_anon", |
4487 | "workingset_refault_file", | |
72a14e82 | 4488 | #ifdef CONFIG_SWAP |
c8713d0b | 4489 | "swap", |
72a14e82 LS |
4490 | "swapcached", |
4491 | #endif | |
c8713d0b JW |
4492 | }; |
4493 | ||
df0e53d0 | 4494 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 4495 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
4496 | PGPGIN, |
4497 | PGPGOUT, | |
4498 | PGFAULT, | |
4499 | PGMAJFAULT, | |
4500 | }; | |
4501 | ||
dddb44ff | 4502 | static void memcg1_stat_format(struct mem_cgroup *memcg, struct seq_buf *s) |
d2ceb9b7 | 4503 | { |
3e32cb2e | 4504 | unsigned long memory, memsw; |
af7c4b0e JW |
4505 | struct mem_cgroup *mi; |
4506 | unsigned int i; | |
406eb0c9 | 4507 | |
71cd3113 | 4508 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c | 4509 | |
7d7ef0a4 | 4510 | mem_cgroup_flush_stats(memcg); |
2d146aa3 | 4511 | |
71cd3113 | 4512 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
468c3982 JW |
4513 | unsigned long nr; |
4514 | ||
ff841a06 YA |
4515 | nr = memcg_page_state_local_output(memcg, memcg1_stats[i]); |
4516 | seq_buf_printf(s, "%s %lu\n", memcg1_stat_names[i], nr); | |
1dd3a273 | 4517 | } |
7b854121 | 4518 | |
df0e53d0 | 4519 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
dddb44ff YA |
4520 | seq_buf_printf(s, "%s %lu\n", vm_event_name(memcg1_events[i]), |
4521 | memcg_events_local(memcg, memcg1_events[i])); | |
af7c4b0e JW |
4522 | |
4523 | for (i = 0; i < NR_LRU_LISTS; i++) | |
dddb44ff YA |
4524 | seq_buf_printf(s, "%s %lu\n", lru_list_name(i), |
4525 | memcg_page_state_local(memcg, NR_LRU_BASE + i) * | |
4526 | PAGE_SIZE); | |
af7c4b0e | 4527 | |
14067bb3 | 4528 | /* Hierarchical information */ |
3e32cb2e JW |
4529 | memory = memsw = PAGE_COUNTER_MAX; |
4530 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
15b42562 CD |
4531 | memory = min(memory, READ_ONCE(mi->memory.max)); |
4532 | memsw = min(memsw, READ_ONCE(mi->memsw.max)); | |
fee7b548 | 4533 | } |
dddb44ff YA |
4534 | seq_buf_printf(s, "hierarchical_memory_limit %llu\n", |
4535 | (u64)memory * PAGE_SIZE); | |
840ea53a LS |
4536 | seq_buf_printf(s, "hierarchical_memsw_limit %llu\n", |
4537 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 4538 | |
8de7ecc6 | 4539 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
7de2e9f1 | 4540 | unsigned long nr; |
4541 | ||
ff841a06 | 4542 | nr = memcg_page_state_output(memcg, memcg1_stats[i]); |
dddb44ff | 4543 | seq_buf_printf(s, "total_%s %llu\n", memcg1_stat_names[i], |
ff841a06 | 4544 | (u64)nr); |
af7c4b0e JW |
4545 | } |
4546 | ||
8de7ecc6 | 4547 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
dddb44ff YA |
4548 | seq_buf_printf(s, "total_%s %llu\n", |
4549 | vm_event_name(memcg1_events[i]), | |
4550 | (u64)memcg_events(memcg, memcg1_events[i])); | |
af7c4b0e | 4551 | |
8de7ecc6 | 4552 | for (i = 0; i < NR_LRU_LISTS; i++) |
dddb44ff YA |
4553 | seq_buf_printf(s, "total_%s %llu\n", lru_list_name(i), |
4554 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * | |
4555 | PAGE_SIZE); | |
14067bb3 | 4556 | |
7f016ee8 | 4557 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 4558 | { |
ef8f2327 MG |
4559 | pg_data_t *pgdat; |
4560 | struct mem_cgroup_per_node *mz; | |
1431d4d1 JW |
4561 | unsigned long anon_cost = 0; |
4562 | unsigned long file_cost = 0; | |
7f016ee8 | 4563 | |
ef8f2327 | 4564 | for_each_online_pgdat(pgdat) { |
a3747b53 | 4565 | mz = memcg->nodeinfo[pgdat->node_id]; |
7f016ee8 | 4566 | |
1431d4d1 JW |
4567 | anon_cost += mz->lruvec.anon_cost; |
4568 | file_cost += mz->lruvec.file_cost; | |
ef8f2327 | 4569 | } |
dddb44ff YA |
4570 | seq_buf_printf(s, "anon_cost %lu\n", anon_cost); |
4571 | seq_buf_printf(s, "file_cost %lu\n", file_cost); | |
7f016ee8 KM |
4572 | } |
4573 | #endif | |
d2ceb9b7 KH |
4574 | } |
4575 | ||
182446d0 TH |
4576 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
4577 | struct cftype *cft) | |
a7885eb8 | 4578 | { |
182446d0 | 4579 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4580 | |
1f4c025b | 4581 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4582 | } |
4583 | ||
182446d0 TH |
4584 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
4585 | struct cftype *cft, u64 val) | |
a7885eb8 | 4586 | { |
182446d0 | 4587 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4588 | |
37bc3cb9 | 4589 | if (val > 200) |
a7885eb8 KM |
4590 | return -EINVAL; |
4591 | ||
a4792030 | 4592 | if (!mem_cgroup_is_root(memcg)) |
82b3aa26 | 4593 | WRITE_ONCE(memcg->swappiness, val); |
3dae7fec | 4594 | else |
82b3aa26 | 4595 | WRITE_ONCE(vm_swappiness, val); |
068b38c1 | 4596 | |
a7885eb8 KM |
4597 | return 0; |
4598 | } | |
4599 | ||
2e72b634 KS |
4600 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4601 | { | |
4602 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 4603 | unsigned long usage; |
2e72b634 KS |
4604 | int i; |
4605 | ||
4606 | rcu_read_lock(); | |
4607 | if (!swap) | |
2c488db2 | 4608 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4609 | else |
2c488db2 | 4610 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4611 | |
4612 | if (!t) | |
4613 | goto unlock; | |
4614 | ||
ce00a967 | 4615 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
4616 | |
4617 | /* | |
748dad36 | 4618 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4619 | * If it's not true, a threshold was crossed after last |
4620 | * call of __mem_cgroup_threshold(). | |
4621 | */ | |
5407a562 | 4622 | i = t->current_threshold; |
2e72b634 KS |
4623 | |
4624 | /* | |
4625 | * Iterate backward over array of thresholds starting from | |
4626 | * current_threshold and check if a threshold is crossed. | |
4627 | * If none of thresholds below usage is crossed, we read | |
4628 | * only one element of the array here. | |
4629 | */ | |
4630 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3652117f | 4631 | eventfd_signal(t->entries[i].eventfd); |
2e72b634 KS |
4632 | |
4633 | /* i = current_threshold + 1 */ | |
4634 | i++; | |
4635 | ||
4636 | /* | |
4637 | * Iterate forward over array of thresholds starting from | |
4638 | * current_threshold+1 and check if a threshold is crossed. | |
4639 | * If none of thresholds above usage is crossed, we read | |
4640 | * only one element of the array here. | |
4641 | */ | |
4642 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3652117f | 4643 | eventfd_signal(t->entries[i].eventfd); |
2e72b634 KS |
4644 | |
4645 | /* Update current_threshold */ | |
5407a562 | 4646 | t->current_threshold = i - 1; |
2e72b634 KS |
4647 | unlock: |
4648 | rcu_read_unlock(); | |
4649 | } | |
4650 | ||
4651 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4652 | { | |
ad4ca5f4 KS |
4653 | while (memcg) { |
4654 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 4655 | if (do_memsw_account()) |
ad4ca5f4 KS |
4656 | __mem_cgroup_threshold(memcg, true); |
4657 | ||
4658 | memcg = parent_mem_cgroup(memcg); | |
4659 | } | |
2e72b634 KS |
4660 | } |
4661 | ||
4662 | static int compare_thresholds(const void *a, const void *b) | |
4663 | { | |
4664 | const struct mem_cgroup_threshold *_a = a; | |
4665 | const struct mem_cgroup_threshold *_b = b; | |
4666 | ||
2bff24a3 GT |
4667 | if (_a->threshold > _b->threshold) |
4668 | return 1; | |
4669 | ||
4670 | if (_a->threshold < _b->threshold) | |
4671 | return -1; | |
4672 | ||
4673 | return 0; | |
2e72b634 KS |
4674 | } |
4675 | ||
c0ff4b85 | 4676 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4677 | { |
4678 | struct mem_cgroup_eventfd_list *ev; | |
4679 | ||
2bcf2e92 MH |
4680 | spin_lock(&memcg_oom_lock); |
4681 | ||
c0ff4b85 | 4682 | list_for_each_entry(ev, &memcg->oom_notify, list) |
3652117f | 4683 | eventfd_signal(ev->eventfd); |
2bcf2e92 MH |
4684 | |
4685 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
4686 | return 0; |
4687 | } | |
4688 | ||
c0ff4b85 | 4689 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4690 | { |
7d74b06f KH |
4691 | struct mem_cgroup *iter; |
4692 | ||
c0ff4b85 | 4693 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4694 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4695 | } |
4696 | ||
59b6f873 | 4697 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4698 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4699 | { |
2c488db2 KS |
4700 | struct mem_cgroup_thresholds *thresholds; |
4701 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
4702 | unsigned long threshold; |
4703 | unsigned long usage; | |
2c488db2 | 4704 | int i, size, ret; |
2e72b634 | 4705 | |
650c5e56 | 4706 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
4707 | if (ret) |
4708 | return ret; | |
4709 | ||
4710 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4711 | |
05b84301 | 4712 | if (type == _MEM) { |
2c488db2 | 4713 | thresholds = &memcg->thresholds; |
ce00a967 | 4714 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4715 | } else if (type == _MEMSWAP) { |
2c488db2 | 4716 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4717 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4718 | } else |
2e72b634 KS |
4719 | BUG(); |
4720 | ||
2e72b634 | 4721 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4722 | if (thresholds->primary) |
2e72b634 KS |
4723 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4724 | ||
2c488db2 | 4725 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4726 | |
4727 | /* Allocate memory for new array of thresholds */ | |
67b8046f | 4728 | new = kmalloc(struct_size(new, entries, size), GFP_KERNEL); |
2c488db2 | 4729 | if (!new) { |
2e72b634 KS |
4730 | ret = -ENOMEM; |
4731 | goto unlock; | |
4732 | } | |
2c488db2 | 4733 | new->size = size; |
2e72b634 KS |
4734 | |
4735 | /* Copy thresholds (if any) to new array */ | |
e90342e6 GS |
4736 | if (thresholds->primary) |
4737 | memcpy(new->entries, thresholds->primary->entries, | |
4738 | flex_array_size(new, entries, size - 1)); | |
2c488db2 | 4739 | |
2e72b634 | 4740 | /* Add new threshold */ |
2c488db2 KS |
4741 | new->entries[size - 1].eventfd = eventfd; |
4742 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4743 | |
4744 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
61e604e6 | 4745 | sort(new->entries, size, sizeof(*new->entries), |
2e72b634 KS |
4746 | compare_thresholds, NULL); |
4747 | ||
4748 | /* Find current threshold */ | |
2c488db2 | 4749 | new->current_threshold = -1; |
2e72b634 | 4750 | for (i = 0; i < size; i++) { |
748dad36 | 4751 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4752 | /* |
2c488db2 KS |
4753 | * new->current_threshold will not be used until |
4754 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4755 | * it here. |
4756 | */ | |
2c488db2 | 4757 | ++new->current_threshold; |
748dad36 SZ |
4758 | } else |
4759 | break; | |
2e72b634 KS |
4760 | } |
4761 | ||
2c488db2 KS |
4762 | /* Free old spare buffer and save old primary buffer as spare */ |
4763 | kfree(thresholds->spare); | |
4764 | thresholds->spare = thresholds->primary; | |
4765 | ||
4766 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4767 | |
907860ed | 4768 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4769 | synchronize_rcu(); |
4770 | ||
2e72b634 KS |
4771 | unlock: |
4772 | mutex_unlock(&memcg->thresholds_lock); | |
4773 | ||
4774 | return ret; | |
4775 | } | |
4776 | ||
59b6f873 | 4777 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4778 | struct eventfd_ctx *eventfd, const char *args) |
4779 | { | |
59b6f873 | 4780 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4781 | } |
4782 | ||
59b6f873 | 4783 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4784 | struct eventfd_ctx *eventfd, const char *args) |
4785 | { | |
59b6f873 | 4786 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4787 | } |
4788 | ||
59b6f873 | 4789 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4790 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4791 | { |
2c488db2 KS |
4792 | struct mem_cgroup_thresholds *thresholds; |
4793 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 4794 | unsigned long usage; |
7d36665a | 4795 | int i, j, size, entries; |
2e72b634 KS |
4796 | |
4797 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4798 | |
4799 | if (type == _MEM) { | |
2c488db2 | 4800 | thresholds = &memcg->thresholds; |
ce00a967 | 4801 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4802 | } else if (type == _MEMSWAP) { |
2c488db2 | 4803 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4804 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4805 | } else |
2e72b634 KS |
4806 | BUG(); |
4807 | ||
371528ca AV |
4808 | if (!thresholds->primary) |
4809 | goto unlock; | |
4810 | ||
2e72b634 KS |
4811 | /* Check if a threshold crossed before removing */ |
4812 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4813 | ||
4814 | /* Calculate new number of threshold */ | |
7d36665a | 4815 | size = entries = 0; |
2c488db2 KS |
4816 | for (i = 0; i < thresholds->primary->size; i++) { |
4817 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 | 4818 | size++; |
7d36665a CX |
4819 | else |
4820 | entries++; | |
2e72b634 KS |
4821 | } |
4822 | ||
2c488db2 | 4823 | new = thresholds->spare; |
907860ed | 4824 | |
7d36665a CX |
4825 | /* If no items related to eventfd have been cleared, nothing to do */ |
4826 | if (!entries) | |
4827 | goto unlock; | |
4828 | ||
2e72b634 KS |
4829 | /* Set thresholds array to NULL if we don't have thresholds */ |
4830 | if (!size) { | |
2c488db2 KS |
4831 | kfree(new); |
4832 | new = NULL; | |
907860ed | 4833 | goto swap_buffers; |
2e72b634 KS |
4834 | } |
4835 | ||
2c488db2 | 4836 | new->size = size; |
2e72b634 KS |
4837 | |
4838 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4839 | new->current_threshold = -1; |
4840 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4841 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4842 | continue; |
4843 | ||
2c488db2 | 4844 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4845 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4846 | /* |
2c488db2 | 4847 | * new->current_threshold will not be used |
2e72b634 KS |
4848 | * until rcu_assign_pointer(), so it's safe to increment |
4849 | * it here. | |
4850 | */ | |
2c488db2 | 4851 | ++new->current_threshold; |
2e72b634 KS |
4852 | } |
4853 | j++; | |
4854 | } | |
4855 | ||
907860ed | 4856 | swap_buffers: |
2c488db2 KS |
4857 | /* Swap primary and spare array */ |
4858 | thresholds->spare = thresholds->primary; | |
8c757763 | 4859 | |
2c488db2 | 4860 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4861 | |
907860ed | 4862 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4863 | synchronize_rcu(); |
6611d8d7 MC |
4864 | |
4865 | /* If all events are unregistered, free the spare array */ | |
4866 | if (!new) { | |
4867 | kfree(thresholds->spare); | |
4868 | thresholds->spare = NULL; | |
4869 | } | |
371528ca | 4870 | unlock: |
2e72b634 | 4871 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4872 | } |
c1e862c1 | 4873 | |
59b6f873 | 4874 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4875 | struct eventfd_ctx *eventfd) |
4876 | { | |
59b6f873 | 4877 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4878 | } |
4879 | ||
59b6f873 | 4880 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4881 | struct eventfd_ctx *eventfd) |
4882 | { | |
59b6f873 | 4883 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4884 | } |
4885 | ||
59b6f873 | 4886 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4887 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4888 | { |
9490ff27 | 4889 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4890 | |
9490ff27 KH |
4891 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4892 | if (!event) | |
4893 | return -ENOMEM; | |
4894 | ||
1af8efe9 | 4895 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4896 | |
4897 | event->eventfd = eventfd; | |
4898 | list_add(&event->list, &memcg->oom_notify); | |
4899 | ||
4900 | /* already in OOM ? */ | |
c2b42d3c | 4901 | if (memcg->under_oom) |
3652117f | 4902 | eventfd_signal(eventfd); |
1af8efe9 | 4903 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4904 | |
4905 | return 0; | |
4906 | } | |
4907 | ||
59b6f873 | 4908 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4909 | struct eventfd_ctx *eventfd) |
9490ff27 | 4910 | { |
9490ff27 | 4911 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4912 | |
1af8efe9 | 4913 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4914 | |
c0ff4b85 | 4915 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4916 | if (ev->eventfd == eventfd) { |
4917 | list_del(&ev->list); | |
4918 | kfree(ev); | |
4919 | } | |
4920 | } | |
4921 | ||
1af8efe9 | 4922 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4923 | } |
4924 | ||
2da8ca82 | 4925 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4926 | { |
aa9694bb | 4927 | struct mem_cgroup *memcg = mem_cgroup_from_seq(sf); |
3c11ecf4 | 4928 | |
17c56de6 | 4929 | seq_printf(sf, "oom_kill_disable %d\n", READ_ONCE(memcg->oom_kill_disable)); |
c2b42d3c | 4930 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
4931 | seq_printf(sf, "oom_kill %lu\n", |
4932 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
4933 | return 0; |
4934 | } | |
4935 | ||
182446d0 | 4936 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4937 | struct cftype *cft, u64 val) |
4938 | { | |
182446d0 | 4939 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4940 | |
4941 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
a4792030 | 4942 | if (mem_cgroup_is_root(memcg) || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4943 | return -EINVAL; |
4944 | ||
17c56de6 | 4945 | WRITE_ONCE(memcg->oom_kill_disable, val); |
4d845ebf | 4946 | if (!val) |
c0ff4b85 | 4947 | memcg_oom_recover(memcg); |
3dae7fec | 4948 | |
3c11ecf4 KH |
4949 | return 0; |
4950 | } | |
4951 | ||
52ebea74 TH |
4952 | #ifdef CONFIG_CGROUP_WRITEBACK |
4953 | ||
3a8e9ac8 TH |
4954 | #include <trace/events/writeback.h> |
4955 | ||
841710aa TH |
4956 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
4957 | { | |
4958 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
4959 | } | |
4960 | ||
4961 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4962 | { | |
4963 | wb_domain_exit(&memcg->cgwb_domain); | |
4964 | } | |
4965 | ||
2529bb3a TH |
4966 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4967 | { | |
4968 | wb_domain_size_changed(&memcg->cgwb_domain); | |
4969 | } | |
4970 | ||
841710aa TH |
4971 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
4972 | { | |
4973 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4974 | ||
4975 | if (!memcg->css.parent) | |
4976 | return NULL; | |
4977 | ||
4978 | return &memcg->cgwb_domain; | |
4979 | } | |
4980 | ||
c2aa723a TH |
4981 | /** |
4982 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
4983 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
4984 | * @pfilepages: out parameter for number of file pages |
4985 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
4986 | * @pdirty: out parameter for number of dirty pages |
4987 | * @pwriteback: out parameter for number of pages under writeback | |
4988 | * | |
c5edf9cd TH |
4989 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
4990 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
4991 | * is a bit more involved. | |
c2aa723a | 4992 | * |
c5edf9cd TH |
4993 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
4994 | * headroom is calculated as the lowest headroom of itself and the | |
4995 | * ancestors. Note that this doesn't consider the actual amount of | |
4996 | * available memory in the system. The caller should further cap | |
4997 | * *@pheadroom accordingly. | |
c2aa723a | 4998 | */ |
c5edf9cd TH |
4999 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
5000 | unsigned long *pheadroom, unsigned long *pdirty, | |
5001 | unsigned long *pwriteback) | |
c2aa723a TH |
5002 | { |
5003 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
5004 | struct mem_cgroup *parent; | |
c2aa723a | 5005 | |
d9b3ce87 | 5006 | mem_cgroup_flush_stats_ratelimited(memcg); |
c2aa723a | 5007 | |
2d146aa3 JW |
5008 | *pdirty = memcg_page_state(memcg, NR_FILE_DIRTY); |
5009 | *pwriteback = memcg_page_state(memcg, NR_WRITEBACK); | |
5010 | *pfilepages = memcg_page_state(memcg, NR_INACTIVE_FILE) + | |
5011 | memcg_page_state(memcg, NR_ACTIVE_FILE); | |
c2aa723a | 5012 | |
2d146aa3 | 5013 | *pheadroom = PAGE_COUNTER_MAX; |
c2aa723a | 5014 | while ((parent = parent_mem_cgroup(memcg))) { |
15b42562 | 5015 | unsigned long ceiling = min(READ_ONCE(memcg->memory.max), |
d1663a90 | 5016 | READ_ONCE(memcg->memory.high)); |
c2aa723a TH |
5017 | unsigned long used = page_counter_read(&memcg->memory); |
5018 | ||
c5edf9cd | 5019 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
5020 | memcg = parent; |
5021 | } | |
c2aa723a TH |
5022 | } |
5023 | ||
97b27821 TH |
5024 | /* |
5025 | * Foreign dirty flushing | |
5026 | * | |
5027 | * There's an inherent mismatch between memcg and writeback. The former | |
f0953a1b | 5028 | * tracks ownership per-page while the latter per-inode. This was a |
97b27821 TH |
5029 | * deliberate design decision because honoring per-page ownership in the |
5030 | * writeback path is complicated, may lead to higher CPU and IO overheads | |
5031 | * and deemed unnecessary given that write-sharing an inode across | |
5032 | * different cgroups isn't a common use-case. | |
5033 | * | |
5034 | * Combined with inode majority-writer ownership switching, this works well | |
5035 | * enough in most cases but there are some pathological cases. For | |
5036 | * example, let's say there are two cgroups A and B which keep writing to | |
5037 | * different but confined parts of the same inode. B owns the inode and | |
5038 | * A's memory is limited far below B's. A's dirty ratio can rise enough to | |
5039 | * trigger balance_dirty_pages() sleeps but B's can be low enough to avoid | |
5040 | * triggering background writeback. A will be slowed down without a way to | |
5041 | * make writeback of the dirty pages happen. | |
5042 | * | |
f0953a1b | 5043 | * Conditions like the above can lead to a cgroup getting repeatedly and |
97b27821 | 5044 | * severely throttled after making some progress after each |
f0953a1b | 5045 | * dirty_expire_interval while the underlying IO device is almost |
97b27821 TH |
5046 | * completely idle. |
5047 | * | |
5048 | * Solving this problem completely requires matching the ownership tracking | |
5049 | * granularities between memcg and writeback in either direction. However, | |
5050 | * the more egregious behaviors can be avoided by simply remembering the | |
5051 | * most recent foreign dirtying events and initiating remote flushes on | |
5052 | * them when local writeback isn't enough to keep the memory clean enough. | |
5053 | * | |
5054 | * The following two functions implement such mechanism. When a foreign | |
5055 | * page - a page whose memcg and writeback ownerships don't match - is | |
5056 | * dirtied, mem_cgroup_track_foreign_dirty() records the inode owning | |
5057 | * bdi_writeback on the page owning memcg. When balance_dirty_pages() | |
5058 | * decides that the memcg needs to sleep due to high dirty ratio, it calls | |
5059 | * mem_cgroup_flush_foreign() which queues writeback on the recorded | |
5060 | * foreign bdi_writebacks which haven't expired. Both the numbers of | |
5061 | * recorded bdi_writebacks and concurrent in-flight foreign writebacks are | |
5062 | * limited to MEMCG_CGWB_FRN_CNT. | |
5063 | * | |
5064 | * The mechanism only remembers IDs and doesn't hold any object references. | |
5065 | * As being wrong occasionally doesn't matter, updates and accesses to the | |
5066 | * records are lockless and racy. | |
5067 | */ | |
9d8053fc | 5068 | void mem_cgroup_track_foreign_dirty_slowpath(struct folio *folio, |
97b27821 TH |
5069 | struct bdi_writeback *wb) |
5070 | { | |
9d8053fc | 5071 | struct mem_cgroup *memcg = folio_memcg(folio); |
97b27821 TH |
5072 | struct memcg_cgwb_frn *frn; |
5073 | u64 now = get_jiffies_64(); | |
5074 | u64 oldest_at = now; | |
5075 | int oldest = -1; | |
5076 | int i; | |
5077 | ||
9d8053fc | 5078 | trace_track_foreign_dirty(folio, wb); |
3a8e9ac8 | 5079 | |
97b27821 TH |
5080 | /* |
5081 | * Pick the slot to use. If there is already a slot for @wb, keep | |
5082 | * using it. If not replace the oldest one which isn't being | |
5083 | * written out. | |
5084 | */ | |
5085 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
5086 | frn = &memcg->cgwb_frn[i]; | |
5087 | if (frn->bdi_id == wb->bdi->id && | |
5088 | frn->memcg_id == wb->memcg_css->id) | |
5089 | break; | |
5090 | if (time_before64(frn->at, oldest_at) && | |
5091 | atomic_read(&frn->done.cnt) == 1) { | |
5092 | oldest = i; | |
5093 | oldest_at = frn->at; | |
5094 | } | |
5095 | } | |
5096 | ||
5097 | if (i < MEMCG_CGWB_FRN_CNT) { | |
5098 | /* | |
5099 | * Re-using an existing one. Update timestamp lazily to | |
5100 | * avoid making the cacheline hot. We want them to be | |
5101 | * reasonably up-to-date and significantly shorter than | |
5102 | * dirty_expire_interval as that's what expires the record. | |
5103 | * Use the shorter of 1s and dirty_expire_interval / 8. | |
5104 | */ | |
5105 | unsigned long update_intv = | |
5106 | min_t(unsigned long, HZ, | |
5107 | msecs_to_jiffies(dirty_expire_interval * 10) / 8); | |
5108 | ||
5109 | if (time_before64(frn->at, now - update_intv)) | |
5110 | frn->at = now; | |
5111 | } else if (oldest >= 0) { | |
5112 | /* replace the oldest free one */ | |
5113 | frn = &memcg->cgwb_frn[oldest]; | |
5114 | frn->bdi_id = wb->bdi->id; | |
5115 | frn->memcg_id = wb->memcg_css->id; | |
5116 | frn->at = now; | |
5117 | } | |
5118 | } | |
5119 | ||
5120 | /* issue foreign writeback flushes for recorded foreign dirtying events */ | |
5121 | void mem_cgroup_flush_foreign(struct bdi_writeback *wb) | |
5122 | { | |
5123 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
5124 | unsigned long intv = msecs_to_jiffies(dirty_expire_interval * 10); | |
5125 | u64 now = jiffies_64; | |
5126 | int i; | |
5127 | ||
5128 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
5129 | struct memcg_cgwb_frn *frn = &memcg->cgwb_frn[i]; | |
5130 | ||
5131 | /* | |
5132 | * If the record is older than dirty_expire_interval, | |
5133 | * writeback on it has already started. No need to kick it | |
5134 | * off again. Also, don't start a new one if there's | |
5135 | * already one in flight. | |
5136 | */ | |
5137 | if (time_after64(frn->at, now - intv) && | |
5138 | atomic_read(&frn->done.cnt) == 1) { | |
5139 | frn->at = 0; | |
3a8e9ac8 | 5140 | trace_flush_foreign(wb, frn->bdi_id, frn->memcg_id); |
7490a2d2 | 5141 | cgroup_writeback_by_id(frn->bdi_id, frn->memcg_id, |
97b27821 TH |
5142 | WB_REASON_FOREIGN_FLUSH, |
5143 | &frn->done); | |
5144 | } | |
5145 | } | |
5146 | } | |
5147 | ||
841710aa TH |
5148 | #else /* CONFIG_CGROUP_WRITEBACK */ |
5149 | ||
5150 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
5151 | { | |
5152 | return 0; | |
5153 | } | |
5154 | ||
5155 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
5156 | { | |
5157 | } | |
5158 | ||
2529bb3a TH |
5159 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
5160 | { | |
5161 | } | |
5162 | ||
52ebea74 TH |
5163 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
5164 | ||
3bc942f3 TH |
5165 | /* |
5166 | * DO NOT USE IN NEW FILES. | |
5167 | * | |
5168 | * "cgroup.event_control" implementation. | |
5169 | * | |
5170 | * This is way over-engineered. It tries to support fully configurable | |
5171 | * events for each user. Such level of flexibility is completely | |
5172 | * unnecessary especially in the light of the planned unified hierarchy. | |
5173 | * | |
5174 | * Please deprecate this and replace with something simpler if at all | |
5175 | * possible. | |
5176 | */ | |
5177 | ||
79bd9814 TH |
5178 | /* |
5179 | * Unregister event and free resources. | |
5180 | * | |
5181 | * Gets called from workqueue. | |
5182 | */ | |
3bc942f3 | 5183 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 5184 | { |
3bc942f3 TH |
5185 | struct mem_cgroup_event *event = |
5186 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 5187 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
5188 | |
5189 | remove_wait_queue(event->wqh, &event->wait); | |
5190 | ||
59b6f873 | 5191 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
5192 | |
5193 | /* Notify userspace the event is going away. */ | |
3652117f | 5194 | eventfd_signal(event->eventfd); |
79bd9814 TH |
5195 | |
5196 | eventfd_ctx_put(event->eventfd); | |
5197 | kfree(event); | |
59b6f873 | 5198 | css_put(&memcg->css); |
79bd9814 TH |
5199 | } |
5200 | ||
5201 | /* | |
a9a08845 | 5202 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
5203 | * |
5204 | * Called with wqh->lock held and interrupts disabled. | |
5205 | */ | |
ac6424b9 | 5206 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 5207 | int sync, void *key) |
79bd9814 | 5208 | { |
3bc942f3 TH |
5209 | struct mem_cgroup_event *event = |
5210 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 5211 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 5212 | __poll_t flags = key_to_poll(key); |
79bd9814 | 5213 | |
a9a08845 | 5214 | if (flags & EPOLLHUP) { |
79bd9814 TH |
5215 | /* |
5216 | * If the event has been detached at cgroup removal, we | |
5217 | * can simply return knowing the other side will cleanup | |
5218 | * for us. | |
5219 | * | |
5220 | * We can't race against event freeing since the other | |
5221 | * side will require wqh->lock via remove_wait_queue(), | |
5222 | * which we hold. | |
5223 | */ | |
fba94807 | 5224 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
5225 | if (!list_empty(&event->list)) { |
5226 | list_del_init(&event->list); | |
5227 | /* | |
5228 | * We are in atomic context, but cgroup_event_remove() | |
5229 | * may sleep, so we have to call it in workqueue. | |
5230 | */ | |
5231 | schedule_work(&event->remove); | |
5232 | } | |
fba94807 | 5233 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
5234 | } |
5235 | ||
5236 | return 0; | |
5237 | } | |
5238 | ||
3bc942f3 | 5239 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
5240 | wait_queue_head_t *wqh, poll_table *pt) |
5241 | { | |
3bc942f3 TH |
5242 | struct mem_cgroup_event *event = |
5243 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
5244 | |
5245 | event->wqh = wqh; | |
5246 | add_wait_queue(wqh, &event->wait); | |
5247 | } | |
5248 | ||
5249 | /* | |
3bc942f3 TH |
5250 | * DO NOT USE IN NEW FILES. |
5251 | * | |
79bd9814 TH |
5252 | * Parse input and register new cgroup event handler. |
5253 | * | |
5254 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
5255 | * Interpretation of args is defined by control file implementation. | |
5256 | */ | |
451af504 TH |
5257 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
5258 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 5259 | { |
451af504 | 5260 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 5261 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5262 | struct mem_cgroup_event *event; |
79bd9814 TH |
5263 | struct cgroup_subsys_state *cfile_css; |
5264 | unsigned int efd, cfd; | |
5265 | struct fd efile; | |
5266 | struct fd cfile; | |
4a7ba45b | 5267 | struct dentry *cdentry; |
fba94807 | 5268 | const char *name; |
79bd9814 TH |
5269 | char *endp; |
5270 | int ret; | |
5271 | ||
2343e88d SAS |
5272 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) |
5273 | return -EOPNOTSUPP; | |
5274 | ||
451af504 TH |
5275 | buf = strstrip(buf); |
5276 | ||
5277 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
5278 | if (*endp != ' ') |
5279 | return -EINVAL; | |
451af504 | 5280 | buf = endp + 1; |
79bd9814 | 5281 | |
451af504 | 5282 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
5283 | if ((*endp != ' ') && (*endp != '\0')) |
5284 | return -EINVAL; | |
451af504 | 5285 | buf = endp + 1; |
79bd9814 TH |
5286 | |
5287 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
5288 | if (!event) | |
5289 | return -ENOMEM; | |
5290 | ||
59b6f873 | 5291 | event->memcg = memcg; |
79bd9814 | 5292 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
5293 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
5294 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
5295 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
5296 | |
5297 | efile = fdget(efd); | |
5298 | if (!efile.file) { | |
5299 | ret = -EBADF; | |
5300 | goto out_kfree; | |
5301 | } | |
5302 | ||
5303 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
5304 | if (IS_ERR(event->eventfd)) { | |
5305 | ret = PTR_ERR(event->eventfd); | |
5306 | goto out_put_efile; | |
5307 | } | |
5308 | ||
5309 | cfile = fdget(cfd); | |
5310 | if (!cfile.file) { | |
5311 | ret = -EBADF; | |
5312 | goto out_put_eventfd; | |
5313 | } | |
5314 | ||
5315 | /* the process need read permission on control file */ | |
5316 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
02f92b38 | 5317 | ret = file_permission(cfile.file, MAY_READ); |
79bd9814 TH |
5318 | if (ret < 0) |
5319 | goto out_put_cfile; | |
5320 | ||
4a7ba45b TH |
5321 | /* |
5322 | * The control file must be a regular cgroup1 file. As a regular cgroup | |
5323 | * file can't be renamed, it's safe to access its name afterwards. | |
5324 | */ | |
5325 | cdentry = cfile.file->f_path.dentry; | |
5326 | if (cdentry->d_sb->s_type != &cgroup_fs_type || !d_is_reg(cdentry)) { | |
5327 | ret = -EINVAL; | |
5328 | goto out_put_cfile; | |
5329 | } | |
5330 | ||
fba94807 TH |
5331 | /* |
5332 | * Determine the event callbacks and set them in @event. This used | |
5333 | * to be done via struct cftype but cgroup core no longer knows | |
5334 | * about these events. The following is crude but the whole thing | |
5335 | * is for compatibility anyway. | |
3bc942f3 TH |
5336 | * |
5337 | * DO NOT ADD NEW FILES. | |
fba94807 | 5338 | */ |
4a7ba45b | 5339 | name = cdentry->d_name.name; |
fba94807 TH |
5340 | |
5341 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
5342 | event->register_event = mem_cgroup_usage_register_event; | |
5343 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
5344 | } else if (!strcmp(name, "memory.oom_control")) { | |
5345 | event->register_event = mem_cgroup_oom_register_event; | |
5346 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
5347 | } else if (!strcmp(name, "memory.pressure_level")) { | |
5348 | event->register_event = vmpressure_register_event; | |
5349 | event->unregister_event = vmpressure_unregister_event; | |
5350 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
5351 | event->register_event = memsw_cgroup_usage_register_event; |
5352 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
5353 | } else { |
5354 | ret = -EINVAL; | |
5355 | goto out_put_cfile; | |
5356 | } | |
5357 | ||
79bd9814 | 5358 | /* |
b5557c4c TH |
5359 | * Verify @cfile should belong to @css. Also, remaining events are |
5360 | * automatically removed on cgroup destruction but the removal is | |
5361 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 5362 | */ |
4a7ba45b | 5363 | cfile_css = css_tryget_online_from_dir(cdentry->d_parent, |
ec903c0c | 5364 | &memory_cgrp_subsys); |
79bd9814 | 5365 | ret = -EINVAL; |
5a17f543 | 5366 | if (IS_ERR(cfile_css)) |
79bd9814 | 5367 | goto out_put_cfile; |
5a17f543 TH |
5368 | if (cfile_css != css) { |
5369 | css_put(cfile_css); | |
79bd9814 | 5370 | goto out_put_cfile; |
5a17f543 | 5371 | } |
79bd9814 | 5372 | |
451af504 | 5373 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
5374 | if (ret) |
5375 | goto out_put_css; | |
5376 | ||
9965ed17 | 5377 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 5378 | |
4ba9515d | 5379 | spin_lock_irq(&memcg->event_list_lock); |
fba94807 | 5380 | list_add(&event->list, &memcg->event_list); |
4ba9515d | 5381 | spin_unlock_irq(&memcg->event_list_lock); |
79bd9814 TH |
5382 | |
5383 | fdput(cfile); | |
5384 | fdput(efile); | |
5385 | ||
451af504 | 5386 | return nbytes; |
79bd9814 TH |
5387 | |
5388 | out_put_css: | |
b5557c4c | 5389 | css_put(css); |
79bd9814 TH |
5390 | out_put_cfile: |
5391 | fdput(cfile); | |
5392 | out_put_eventfd: | |
5393 | eventfd_ctx_put(event->eventfd); | |
5394 | out_put_efile: | |
5395 | fdput(efile); | |
5396 | out_kfree: | |
5397 | kfree(event); | |
5398 | ||
5399 | return ret; | |
5400 | } | |
5401 | ||
bc3dcb85 | 5402 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_SLUB_DEBUG) |
c29b5b3d MS |
5403 | static int mem_cgroup_slab_show(struct seq_file *m, void *p) |
5404 | { | |
5405 | /* | |
5406 | * Deprecated. | |
df4ae285 | 5407 | * Please, take a look at tools/cgroup/memcg_slabinfo.py . |
c29b5b3d MS |
5408 | */ |
5409 | return 0; | |
5410 | } | |
5411 | #endif | |
5412 | ||
dddb44ff YA |
5413 | static int memory_stat_show(struct seq_file *m, void *v); |
5414 | ||
241994ed | 5415 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 5416 | { |
0eea1030 | 5417 | .name = "usage_in_bytes", |
8c7c6e34 | 5418 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 5419 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5420 | }, |
c84872e1 PE |
5421 | { |
5422 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 5423 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 5424 | .write = mem_cgroup_reset, |
791badbd | 5425 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 5426 | }, |
8cdea7c0 | 5427 | { |
0eea1030 | 5428 | .name = "limit_in_bytes", |
8c7c6e34 | 5429 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 5430 | .write = mem_cgroup_write, |
791badbd | 5431 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5432 | }, |
296c81d8 BS |
5433 | { |
5434 | .name = "soft_limit_in_bytes", | |
5435 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 5436 | .write = mem_cgroup_write, |
791badbd | 5437 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 5438 | }, |
8cdea7c0 BS |
5439 | { |
5440 | .name = "failcnt", | |
8c7c6e34 | 5441 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 5442 | .write = mem_cgroup_reset, |
791badbd | 5443 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5444 | }, |
d2ceb9b7 KH |
5445 | { |
5446 | .name = "stat", | |
dddb44ff | 5447 | .seq_show = memory_stat_show, |
d2ceb9b7 | 5448 | }, |
c1e862c1 KH |
5449 | { |
5450 | .name = "force_empty", | |
6770c64e | 5451 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 5452 | }, |
18f59ea7 BS |
5453 | { |
5454 | .name = "use_hierarchy", | |
5455 | .write_u64 = mem_cgroup_hierarchy_write, | |
5456 | .read_u64 = mem_cgroup_hierarchy_read, | |
5457 | }, | |
79bd9814 | 5458 | { |
3bc942f3 | 5459 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 5460 | .write = memcg_write_event_control, |
7dbdb199 | 5461 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 5462 | }, |
a7885eb8 KM |
5463 | { |
5464 | .name = "swappiness", | |
5465 | .read_u64 = mem_cgroup_swappiness_read, | |
5466 | .write_u64 = mem_cgroup_swappiness_write, | |
5467 | }, | |
7dc74be0 DN |
5468 | { |
5469 | .name = "move_charge_at_immigrate", | |
5470 | .read_u64 = mem_cgroup_move_charge_read, | |
5471 | .write_u64 = mem_cgroup_move_charge_write, | |
5472 | }, | |
9490ff27 KH |
5473 | { |
5474 | .name = "oom_control", | |
2da8ca82 | 5475 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 5476 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 | 5477 | }, |
70ddf637 AV |
5478 | { |
5479 | .name = "pressure_level", | |
6b0ba2ab | 5480 | .seq_show = mem_cgroup_dummy_seq_show, |
70ddf637 | 5481 | }, |
406eb0c9 YH |
5482 | #ifdef CONFIG_NUMA |
5483 | { | |
5484 | .name = "numa_stat", | |
2da8ca82 | 5485 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
5486 | }, |
5487 | #endif | |
4597648f MH |
5488 | { |
5489 | .name = "kmem.limit_in_bytes", | |
5490 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
5491 | .write = mem_cgroup_write, | |
5492 | .read_u64 = mem_cgroup_read_u64, | |
5493 | }, | |
510fc4e1 GC |
5494 | { |
5495 | .name = "kmem.usage_in_bytes", | |
5496 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 5497 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5498 | }, |
5499 | { | |
5500 | .name = "kmem.failcnt", | |
5501 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 5502 | .write = mem_cgroup_reset, |
791badbd | 5503 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5504 | }, |
5505 | { | |
5506 | .name = "kmem.max_usage_in_bytes", | |
5507 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 5508 | .write = mem_cgroup_reset, |
791badbd | 5509 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 5510 | }, |
bc3dcb85 | 5511 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_SLUB_DEBUG) |
749c5415 GC |
5512 | { |
5513 | .name = "kmem.slabinfo", | |
c29b5b3d | 5514 | .seq_show = mem_cgroup_slab_show, |
749c5415 GC |
5515 | }, |
5516 | #endif | |
d55f90bf VD |
5517 | { |
5518 | .name = "kmem.tcp.limit_in_bytes", | |
5519 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
5520 | .write = mem_cgroup_write, | |
5521 | .read_u64 = mem_cgroup_read_u64, | |
5522 | }, | |
5523 | { | |
5524 | .name = "kmem.tcp.usage_in_bytes", | |
5525 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
5526 | .read_u64 = mem_cgroup_read_u64, | |
5527 | }, | |
5528 | { | |
5529 | .name = "kmem.tcp.failcnt", | |
5530 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
5531 | .write = mem_cgroup_reset, | |
5532 | .read_u64 = mem_cgroup_read_u64, | |
5533 | }, | |
5534 | { | |
5535 | .name = "kmem.tcp.max_usage_in_bytes", | |
5536 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
5537 | .write = mem_cgroup_reset, | |
5538 | .read_u64 = mem_cgroup_read_u64, | |
5539 | }, | |
6bc10349 | 5540 | { }, /* terminate */ |
af36f906 | 5541 | }; |
8c7c6e34 | 5542 | |
73f576c0 JW |
5543 | /* |
5544 | * Private memory cgroup IDR | |
5545 | * | |
5546 | * Swap-out records and page cache shadow entries need to store memcg | |
5547 | * references in constrained space, so we maintain an ID space that is | |
5548 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
5549 | * memory-controlled cgroups to 64k. | |
5550 | * | |
b8f2935f | 5551 | * However, there usually are many references to the offline CSS after |
73f576c0 JW |
5552 | * the cgroup has been destroyed, such as page cache or reclaimable |
5553 | * slab objects, that don't need to hang on to the ID. We want to keep | |
5554 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
5555 | * relatively small ID space and prevent the creation of new cgroups | |
5556 | * even when there are much fewer than 64k cgroups - possibly none. | |
5557 | * | |
5558 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
5559 | * be freed and recycled when it's no longer needed, which is usually | |
5560 | * when the CSS is offlined. | |
5561 | * | |
5562 | * The only exception to that are records of swapped out tmpfs/shmem | |
5563 | * pages that need to be attributed to live ancestors on swapin. But | |
5564 | * those references are manageable from userspace. | |
5565 | */ | |
5566 | ||
60b1e24c | 5567 | #define MEM_CGROUP_ID_MAX ((1UL << MEM_CGROUP_ID_SHIFT) - 1) |
73f576c0 JW |
5568 | static DEFINE_IDR(mem_cgroup_idr); |
5569 | ||
7e97de0b KT |
5570 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
5571 | { | |
5572 | if (memcg->id.id > 0) { | |
5573 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
5574 | memcg->id.id = 0; | |
5575 | } | |
5576 | } | |
5577 | ||
c1514c0a VF |
5578 | static void __maybe_unused mem_cgroup_id_get_many(struct mem_cgroup *memcg, |
5579 | unsigned int n) | |
73f576c0 | 5580 | { |
1c2d479a | 5581 | refcount_add(n, &memcg->id.ref); |
73f576c0 JW |
5582 | } |
5583 | ||
615d66c3 | 5584 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 5585 | { |
1c2d479a | 5586 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 5587 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
5588 | |
5589 | /* Memcg ID pins CSS */ | |
5590 | css_put(&memcg->css); | |
5591 | } | |
5592 | } | |
5593 | ||
615d66c3 VD |
5594 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) |
5595 | { | |
5596 | mem_cgroup_id_put_many(memcg, 1); | |
5597 | } | |
5598 | ||
73f576c0 JW |
5599 | /** |
5600 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
5601 | * @id: the memcg id to look up | |
5602 | * | |
5603 | * Caller must hold rcu_read_lock(). | |
5604 | */ | |
5605 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
5606 | { | |
5607 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5608 | return idr_find(&mem_cgroup_idr, id); | |
5609 | } | |
5610 | ||
c15187a4 RG |
5611 | #ifdef CONFIG_SHRINKER_DEBUG |
5612 | struct mem_cgroup *mem_cgroup_get_from_ino(unsigned long ino) | |
5613 | { | |
5614 | struct cgroup *cgrp; | |
5615 | struct cgroup_subsys_state *css; | |
5616 | struct mem_cgroup *memcg; | |
5617 | ||
5618 | cgrp = cgroup_get_from_id(ino); | |
fa7e439c | 5619 | if (IS_ERR(cgrp)) |
c0f2df49 | 5620 | return ERR_CAST(cgrp); |
c15187a4 RG |
5621 | |
5622 | css = cgroup_get_e_css(cgrp, &memory_cgrp_subsys); | |
5623 | if (css) | |
5624 | memcg = container_of(css, struct mem_cgroup, css); | |
5625 | else | |
5626 | memcg = ERR_PTR(-ENOENT); | |
5627 | ||
5628 | cgroup_put(cgrp); | |
5629 | ||
5630 | return memcg; | |
5631 | } | |
5632 | #endif | |
5633 | ||
ef8f2327 | 5634 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
5635 | { |
5636 | struct mem_cgroup_per_node *pn; | |
8c9bb398 WY |
5637 | |
5638 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, node); | |
6d12e2d8 KH |
5639 | if (!pn) |
5640 | return 1; | |
1ecaab2b | 5641 | |
aab6103b RG |
5642 | pn->lruvec_stats = kzalloc_node(sizeof(struct lruvec_stats), |
5643 | GFP_KERNEL_ACCOUNT, node); | |
70a64b79 SB |
5644 | if (!pn->lruvec_stats) |
5645 | goto fail; | |
5646 | ||
7e1c0d6f SB |
5647 | pn->lruvec_stats_percpu = alloc_percpu_gfp(struct lruvec_stats_percpu, |
5648 | GFP_KERNEL_ACCOUNT); | |
70a64b79 SB |
5649 | if (!pn->lruvec_stats_percpu) |
5650 | goto fail; | |
00f3ca2c | 5651 | |
ef8f2327 | 5652 | lruvec_init(&pn->lruvec); |
ef8f2327 MG |
5653 | pn->memcg = memcg; |
5654 | ||
54f72fe0 | 5655 | memcg->nodeinfo[node] = pn; |
6d12e2d8 | 5656 | return 0; |
70a64b79 SB |
5657 | fail: |
5658 | kfree(pn->lruvec_stats); | |
5659 | kfree(pn); | |
5660 | return 1; | |
6d12e2d8 KH |
5661 | } |
5662 | ||
ef8f2327 | 5663 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 5664 | { |
00f3ca2c JW |
5665 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
5666 | ||
4eaf431f MH |
5667 | if (!pn) |
5668 | return; | |
5669 | ||
7e1c0d6f | 5670 | free_percpu(pn->lruvec_stats_percpu); |
70a64b79 | 5671 | kfree(pn->lruvec_stats); |
00f3ca2c | 5672 | kfree(pn); |
1ecaab2b KH |
5673 | } |
5674 | ||
40e952f9 | 5675 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 5676 | { |
c8b2a36f | 5677 | int node; |
59927fb9 | 5678 | |
91b71e78 | 5679 | obj_cgroup_put(memcg->orig_objcg); |
675d6c9b | 5680 | |
c8b2a36f | 5681 | for_each_node(node) |
ef8f2327 | 5682 | free_mem_cgroup_per_node_info(memcg, node); |
410f8e82 | 5683 | kfree(memcg->vmstats); |
871789d4 | 5684 | free_percpu(memcg->vmstats_percpu); |
8ff69e2c | 5685 | kfree(memcg); |
59927fb9 | 5686 | } |
3afe36b1 | 5687 | |
40e952f9 TE |
5688 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
5689 | { | |
ec1c86b2 | 5690 | lru_gen_exit_memcg(memcg); |
40e952f9 TE |
5691 | memcg_wb_domain_exit(memcg); |
5692 | __mem_cgroup_free(memcg); | |
5693 | } | |
5694 | ||
9cee7e8e | 5695 | static struct mem_cgroup *mem_cgroup_alloc(struct mem_cgroup *parent) |
8cdea7c0 | 5696 | { |
9cee7e8e | 5697 | struct memcg_vmstats_percpu *statc, *pstatc; |
d142e3e6 | 5698 | struct mem_cgroup *memcg; |
9cee7e8e | 5699 | int node, cpu; |
97b27821 | 5700 | int __maybe_unused i; |
11d67612 | 5701 | long error = -ENOMEM; |
8cdea7c0 | 5702 | |
06b2c3b0 | 5703 | memcg = kzalloc(struct_size(memcg, nodeinfo, nr_node_ids), GFP_KERNEL); |
c0ff4b85 | 5704 | if (!memcg) |
11d67612 | 5705 | return ERR_PTR(error); |
0b8f73e1 | 5706 | |
73f576c0 | 5707 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
be740503 | 5708 | 1, MEM_CGROUP_ID_MAX + 1, GFP_KERNEL); |
11d67612 YS |
5709 | if (memcg->id.id < 0) { |
5710 | error = memcg->id.id; | |
73f576c0 | 5711 | goto fail; |
11d67612 | 5712 | } |
73f576c0 | 5713 | |
aab6103b RG |
5714 | memcg->vmstats = kzalloc(sizeof(struct memcg_vmstats), |
5715 | GFP_KERNEL_ACCOUNT); | |
410f8e82 SB |
5716 | if (!memcg->vmstats) |
5717 | goto fail; | |
5718 | ||
3e38e0aa RG |
5719 | memcg->vmstats_percpu = alloc_percpu_gfp(struct memcg_vmstats_percpu, |
5720 | GFP_KERNEL_ACCOUNT); | |
871789d4 | 5721 | if (!memcg->vmstats_percpu) |
0b8f73e1 | 5722 | goto fail; |
78fb7466 | 5723 | |
9cee7e8e YA |
5724 | for_each_possible_cpu(cpu) { |
5725 | if (parent) | |
5726 | pstatc = per_cpu_ptr(parent->vmstats_percpu, cpu); | |
5727 | statc = per_cpu_ptr(memcg->vmstats_percpu, cpu); | |
5728 | statc->parent = parent ? pstatc : NULL; | |
5729 | statc->vmstats = memcg->vmstats; | |
5730 | } | |
5731 | ||
3ed28fa1 | 5732 | for_each_node(node) |
ef8f2327 | 5733 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 5734 | goto fail; |
f64c3f54 | 5735 | |
0b8f73e1 JW |
5736 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
5737 | goto fail; | |
28dbc4b6 | 5738 | |
f7e1cb6e | 5739 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 | 5740 | INIT_LIST_HEAD(&memcg->oom_notify); |
d142e3e6 GC |
5741 | mutex_init(&memcg->thresholds_lock); |
5742 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 5743 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
5744 | INIT_LIST_HEAD(&memcg->event_list); |
5745 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 5746 | memcg->socket_pressure = jiffies; |
84c07d11 | 5747 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 5748 | memcg->kmemcg_id = -1; |
bf4f0599 | 5749 | INIT_LIST_HEAD(&memcg->objcg_list); |
900a38f0 | 5750 | #endif |
52ebea74 TH |
5751 | #ifdef CONFIG_CGROUP_WRITEBACK |
5752 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
97b27821 TH |
5753 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) |
5754 | memcg->cgwb_frn[i].done = | |
5755 | __WB_COMPLETION_INIT(&memcg_cgwb_frn_waitq); | |
87eaceb3 YS |
5756 | #endif |
5757 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5758 | spin_lock_init(&memcg->deferred_split_queue.split_queue_lock); | |
5759 | INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue); | |
5760 | memcg->deferred_split_queue.split_queue_len = 0; | |
52ebea74 | 5761 | #endif |
ec1c86b2 | 5762 | lru_gen_init_memcg(memcg); |
0b8f73e1 JW |
5763 | return memcg; |
5764 | fail: | |
7e97de0b | 5765 | mem_cgroup_id_remove(memcg); |
40e952f9 | 5766 | __mem_cgroup_free(memcg); |
11d67612 | 5767 | return ERR_PTR(error); |
d142e3e6 GC |
5768 | } |
5769 | ||
0b8f73e1 JW |
5770 | static struct cgroup_subsys_state * __ref |
5771 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 5772 | { |
0b8f73e1 | 5773 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
b87d8cef | 5774 | struct mem_cgroup *memcg, *old_memcg; |
d142e3e6 | 5775 | |
b87d8cef | 5776 | old_memcg = set_active_memcg(parent); |
9cee7e8e | 5777 | memcg = mem_cgroup_alloc(parent); |
b87d8cef | 5778 | set_active_memcg(old_memcg); |
11d67612 YS |
5779 | if (IS_ERR(memcg)) |
5780 | return ERR_CAST(memcg); | |
d142e3e6 | 5781 | |
d1663a90 | 5782 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
2178e20c | 5783 | WRITE_ONCE(memcg->soft_limit, PAGE_COUNTER_MAX); |
f4840ccf JW |
5784 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) |
5785 | memcg->zswap_max = PAGE_COUNTER_MAX; | |
501a06fe NP |
5786 | WRITE_ONCE(memcg->zswap_writeback, |
5787 | !parent || READ_ONCE(parent->zswap_writeback)); | |
f4840ccf | 5788 | #endif |
4b82ab4f | 5789 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
0b8f73e1 | 5790 | if (parent) { |
82b3aa26 | 5791 | WRITE_ONCE(memcg->swappiness, mem_cgroup_swappiness(parent)); |
17c56de6 | 5792 | WRITE_ONCE(memcg->oom_kill_disable, READ_ONCE(parent->oom_kill_disable)); |
bef8620c | 5793 | |
3e32cb2e | 5794 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 5795 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e | 5796 | page_counter_init(&memcg->kmem, &parent->kmem); |
0db15298 | 5797 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 5798 | } else { |
ff48c71c | 5799 | init_memcg_stats(); |
8278f1c7 | 5800 | init_memcg_events(); |
bef8620c RG |
5801 | page_counter_init(&memcg->memory, NULL); |
5802 | page_counter_init(&memcg->swap, NULL); | |
5803 | page_counter_init(&memcg->kmem, NULL); | |
5804 | page_counter_init(&memcg->tcpmem, NULL); | |
d6441637 | 5805 | |
0b8f73e1 JW |
5806 | root_mem_cgroup = memcg; |
5807 | return &memcg->css; | |
5808 | } | |
5809 | ||
f7e1cb6e | 5810 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5811 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 5812 | |
b6c1a8af YS |
5813 | #if defined(CONFIG_MEMCG_KMEM) |
5814 | if (!cgroup_memory_nobpf) | |
5815 | static_branch_inc(&memcg_bpf_enabled_key); | |
5816 | #endif | |
5817 | ||
0b8f73e1 | 5818 | return &memcg->css; |
0b8f73e1 JW |
5819 | } |
5820 | ||
73f576c0 | 5821 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 5822 | { |
58fa2a55 VD |
5823 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5824 | ||
da0efe30 MS |
5825 | if (memcg_online_kmem(memcg)) |
5826 | goto remove_id; | |
5827 | ||
0a4465d3 | 5828 | /* |
e4262c4f | 5829 | * A memcg must be visible for expand_shrinker_info() |
0a4465d3 KT |
5830 | * by the time the maps are allocated. So, we allocate maps |
5831 | * here, when for_each_mem_cgroup() can't skip it. | |
5832 | */ | |
da0efe30 MS |
5833 | if (alloc_shrinker_info(memcg)) |
5834 | goto offline_kmem; | |
0a4465d3 | 5835 | |
13ef7424 | 5836 | if (unlikely(mem_cgroup_is_root(memcg)) && !mem_cgroup_disabled()) |
aa48e47e | 5837 | queue_delayed_work(system_unbound_wq, &stats_flush_dwork, |
396faf88 | 5838 | FLUSH_TIME); |
e4dde56c | 5839 | lru_gen_online_memcg(memcg); |
6f0df8e1 JW |
5840 | |
5841 | /* Online state pins memcg ID, memcg ID pins CSS */ | |
5842 | refcount_set(&memcg->id.ref, 1); | |
5843 | css_get(css); | |
5844 | ||
5845 | /* | |
5846 | * Ensure mem_cgroup_from_id() works once we're fully online. | |
5847 | * | |
5848 | * We could do this earlier and require callers to filter with | |
5849 | * css_tryget_online(). But right now there are no users that | |
5850 | * need earlier access, and the workingset code relies on the | |
5851 | * cgroup tree linkage (mem_cgroup_get_nr_swap_pages()). So | |
5852 | * publish it here at the end of onlining. This matches the | |
5853 | * regular ID destruction during offlining. | |
5854 | */ | |
5855 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); | |
5856 | ||
2f7dd7a4 | 5857 | return 0; |
da0efe30 MS |
5858 | offline_kmem: |
5859 | memcg_offline_kmem(memcg); | |
5860 | remove_id: | |
5861 | mem_cgroup_id_remove(memcg); | |
5862 | return -ENOMEM; | |
8cdea7c0 BS |
5863 | } |
5864 | ||
eb95419b | 5865 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 5866 | { |
eb95419b | 5867 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5868 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
5869 | |
5870 | /* | |
5871 | * Unregister events and notify userspace. | |
5872 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
5873 | * directory to avoid race between userspace and kernelspace. | |
5874 | */ | |
4ba9515d | 5875 | spin_lock_irq(&memcg->event_list_lock); |
fba94807 | 5876 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { |
79bd9814 TH |
5877 | list_del_init(&event->list); |
5878 | schedule_work(&event->remove); | |
5879 | } | |
4ba9515d | 5880 | spin_unlock_irq(&memcg->event_list_lock); |
ec64f515 | 5881 | |
bf8d5d52 | 5882 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5883 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 5884 | |
a65b0e76 DC |
5885 | zswap_memcg_offline_cleanup(memcg); |
5886 | ||
567e9ab2 | 5887 | memcg_offline_kmem(memcg); |
a178015c | 5888 | reparent_shrinker_deferred(memcg); |
52ebea74 | 5889 | wb_memcg_offline(memcg); |
e4dde56c | 5890 | lru_gen_offline_memcg(memcg); |
73f576c0 | 5891 | |
591edfb1 RG |
5892 | drain_all_stock(memcg); |
5893 | ||
73f576c0 | 5894 | mem_cgroup_id_put(memcg); |
df878fb0 KH |
5895 | } |
5896 | ||
6df38689 VD |
5897 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
5898 | { | |
5899 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5900 | ||
5901 | invalidate_reclaim_iterators(memcg); | |
e4dde56c | 5902 | lru_gen_release_memcg(memcg); |
6df38689 VD |
5903 | } |
5904 | ||
eb95419b | 5905 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 5906 | { |
eb95419b | 5907 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
97b27821 | 5908 | int __maybe_unused i; |
c268e994 | 5909 | |
97b27821 TH |
5910 | #ifdef CONFIG_CGROUP_WRITEBACK |
5911 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) | |
5912 | wb_wait_for_completion(&memcg->cgwb_frn[i].done); | |
5913 | #endif | |
f7e1cb6e | 5914 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5915 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 5916 | |
0db15298 | 5917 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 5918 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 5919 | |
b6c1a8af YS |
5920 | #if defined(CONFIG_MEMCG_KMEM) |
5921 | if (!cgroup_memory_nobpf) | |
5922 | static_branch_dec(&memcg_bpf_enabled_key); | |
5923 | #endif | |
5924 | ||
0b8f73e1 JW |
5925 | vmpressure_cleanup(&memcg->vmpressure); |
5926 | cancel_work_sync(&memcg->high_work); | |
5927 | mem_cgroup_remove_from_trees(memcg); | |
e4262c4f | 5928 | free_shrinker_info(memcg); |
0b8f73e1 | 5929 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
5930 | } |
5931 | ||
1ced953b TH |
5932 | /** |
5933 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
5934 | * @css: the target css | |
5935 | * | |
5936 | * Reset the states of the mem_cgroup associated with @css. This is | |
5937 | * invoked when the userland requests disabling on the default hierarchy | |
5938 | * but the memcg is pinned through dependency. The memcg should stop | |
5939 | * applying policies and should revert to the vanilla state as it may be | |
5940 | * made visible again. | |
5941 | * | |
5942 | * The current implementation only resets the essential configurations. | |
5943 | * This needs to be expanded to cover all the visible parts. | |
5944 | */ | |
5945 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
5946 | { | |
5947 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5948 | ||
bbec2e15 RG |
5949 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
5950 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
bbec2e15 RG |
5951 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); |
5952 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 5953 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5954 | page_counter_set_low(&memcg->memory, 0); |
d1663a90 | 5955 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
2178e20c | 5956 | WRITE_ONCE(memcg->soft_limit, PAGE_COUNTER_MAX); |
4b82ab4f | 5957 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
2529bb3a | 5958 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
5959 | } |
5960 | ||
2d146aa3 JW |
5961 | static void mem_cgroup_css_rstat_flush(struct cgroup_subsys_state *css, int cpu) |
5962 | { | |
5963 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5964 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); | |
5965 | struct memcg_vmstats_percpu *statc; | |
f82e6bf9 | 5966 | long delta, delta_cpu, v; |
7e1c0d6f | 5967 | int i, nid; |
2d146aa3 JW |
5968 | |
5969 | statc = per_cpu_ptr(memcg->vmstats_percpu, cpu); | |
5970 | ||
ff48c71c | 5971 | for (i = 0; i < MEMCG_VMSTAT_SIZE; i++) { |
2d146aa3 JW |
5972 | /* |
5973 | * Collect the aggregated propagation counts of groups | |
5974 | * below us. We're in a per-cpu loop here and this is | |
5975 | * a global counter, so the first cycle will get them. | |
5976 | */ | |
410f8e82 | 5977 | delta = memcg->vmstats->state_pending[i]; |
2d146aa3 | 5978 | if (delta) |
410f8e82 | 5979 | memcg->vmstats->state_pending[i] = 0; |
2d146aa3 JW |
5980 | |
5981 | /* Add CPU changes on this level since the last flush */ | |
f82e6bf9 | 5982 | delta_cpu = 0; |
2d146aa3 JW |
5983 | v = READ_ONCE(statc->state[i]); |
5984 | if (v != statc->state_prev[i]) { | |
f82e6bf9 YA |
5985 | delta_cpu = v - statc->state_prev[i]; |
5986 | delta += delta_cpu; | |
2d146aa3 JW |
5987 | statc->state_prev[i] = v; |
5988 | } | |
5989 | ||
2d146aa3 | 5990 | /* Aggregate counts on this level and propagate upwards */ |
f82e6bf9 YA |
5991 | if (delta_cpu) |
5992 | memcg->vmstats->state_local[i] += delta_cpu; | |
5993 | ||
5994 | if (delta) { | |
5995 | memcg->vmstats->state[i] += delta; | |
5996 | if (parent) | |
5997 | parent->vmstats->state_pending[i] += delta; | |
5998 | } | |
2d146aa3 JW |
5999 | } |
6000 | ||
8278f1c7 | 6001 | for (i = 0; i < NR_MEMCG_EVENTS; i++) { |
410f8e82 | 6002 | delta = memcg->vmstats->events_pending[i]; |
2d146aa3 | 6003 | if (delta) |
410f8e82 | 6004 | memcg->vmstats->events_pending[i] = 0; |
2d146aa3 | 6005 | |
f82e6bf9 | 6006 | delta_cpu = 0; |
2d146aa3 JW |
6007 | v = READ_ONCE(statc->events[i]); |
6008 | if (v != statc->events_prev[i]) { | |
f82e6bf9 YA |
6009 | delta_cpu = v - statc->events_prev[i]; |
6010 | delta += delta_cpu; | |
2d146aa3 JW |
6011 | statc->events_prev[i] = v; |
6012 | } | |
6013 | ||
f82e6bf9 YA |
6014 | if (delta_cpu) |
6015 | memcg->vmstats->events_local[i] += delta_cpu; | |
2d146aa3 | 6016 | |
f82e6bf9 YA |
6017 | if (delta) { |
6018 | memcg->vmstats->events[i] += delta; | |
6019 | if (parent) | |
6020 | parent->vmstats->events_pending[i] += delta; | |
6021 | } | |
2d146aa3 | 6022 | } |
7e1c0d6f SB |
6023 | |
6024 | for_each_node_state(nid, N_MEMORY) { | |
6025 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[nid]; | |
70a64b79 SB |
6026 | struct lruvec_stats *lstats = pn->lruvec_stats; |
6027 | struct lruvec_stats *plstats = NULL; | |
7e1c0d6f SB |
6028 | struct lruvec_stats_percpu *lstatc; |
6029 | ||
6030 | if (parent) | |
70a64b79 | 6031 | plstats = parent->nodeinfo[nid]->lruvec_stats; |
7e1c0d6f SB |
6032 | |
6033 | lstatc = per_cpu_ptr(pn->lruvec_stats_percpu, cpu); | |
6034 | ||
ff48c71c | 6035 | for (i = 0; i < NR_MEMCG_NODE_STAT_ITEMS; i++) { |
70a64b79 | 6036 | delta = lstats->state_pending[i]; |
7e1c0d6f | 6037 | if (delta) |
70a64b79 | 6038 | lstats->state_pending[i] = 0; |
7e1c0d6f | 6039 | |
f82e6bf9 | 6040 | delta_cpu = 0; |
7e1c0d6f SB |
6041 | v = READ_ONCE(lstatc->state[i]); |
6042 | if (v != lstatc->state_prev[i]) { | |
f82e6bf9 YA |
6043 | delta_cpu = v - lstatc->state_prev[i]; |
6044 | delta += delta_cpu; | |
7e1c0d6f SB |
6045 | lstatc->state_prev[i] = v; |
6046 | } | |
6047 | ||
f82e6bf9 | 6048 | if (delta_cpu) |
70a64b79 | 6049 | lstats->state_local[i] += delta_cpu; |
7e1c0d6f | 6050 | |
f82e6bf9 | 6051 | if (delta) { |
70a64b79 SB |
6052 | lstats->state[i] += delta; |
6053 | if (plstats) | |
6054 | plstats->state_pending[i] += delta; | |
f82e6bf9 | 6055 | } |
7e1c0d6f SB |
6056 | } |
6057 | } | |
78ec6f9d | 6058 | WRITE_ONCE(statc->stats_updates, 0); |
8d59d221 YA |
6059 | /* We are in a per-cpu loop here, only do the atomic write once */ |
6060 | if (atomic64_read(&memcg->vmstats->stats_updates)) | |
6061 | atomic64_set(&memcg->vmstats->stats_updates, 0); | |
2d146aa3 JW |
6062 | } |
6063 | ||
02491447 | 6064 | #ifdef CONFIG_MMU |
7dc74be0 | 6065 | /* Handlers for move charge at task migration. */ |
854ffa8d | 6066 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 6067 | { |
05b84301 | 6068 | int ret; |
9476db97 | 6069 | |
d0164adc MG |
6070 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
6071 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 6072 | if (!ret) { |
854ffa8d | 6073 | mc.precharge += count; |
854ffa8d DN |
6074 | return ret; |
6075 | } | |
9476db97 | 6076 | |
3674534b | 6077 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 6078 | while (count--) { |
3674534b | 6079 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 6080 | if (ret) |
38c5d72f | 6081 | return ret; |
854ffa8d | 6082 | mc.precharge++; |
9476db97 | 6083 | cond_resched(); |
854ffa8d | 6084 | } |
9476db97 | 6085 | return 0; |
4ffef5fe DN |
6086 | } |
6087 | ||
4ffef5fe | 6088 | union mc_target { |
b46777da | 6089 | struct folio *folio; |
02491447 | 6090 | swp_entry_t ent; |
4ffef5fe DN |
6091 | }; |
6092 | ||
4ffef5fe | 6093 | enum mc_target_type { |
8d32ff84 | 6094 | MC_TARGET_NONE = 0, |
4ffef5fe | 6095 | MC_TARGET_PAGE, |
02491447 | 6096 | MC_TARGET_SWAP, |
c733a828 | 6097 | MC_TARGET_DEVICE, |
4ffef5fe DN |
6098 | }; |
6099 | ||
90254a65 DN |
6100 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
6101 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 6102 | { |
25b2995a | 6103 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 6104 | |
58f341f7 | 6105 | if (!page) |
90254a65 DN |
6106 | return NULL; |
6107 | if (PageAnon(page)) { | |
1dfab5ab | 6108 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 6109 | return NULL; |
1dfab5ab JW |
6110 | } else { |
6111 | if (!(mc.flags & MOVE_FILE)) | |
6112 | return NULL; | |
6113 | } | |
58f341f7 | 6114 | get_page(page); |
90254a65 DN |
6115 | |
6116 | return page; | |
6117 | } | |
6118 | ||
c733a828 | 6119 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 6120 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 6121 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 6122 | { |
90254a65 DN |
6123 | struct page *page = NULL; |
6124 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
6125 | ||
9a137153 | 6126 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 6127 | return NULL; |
c733a828 JG |
6128 | |
6129 | /* | |
27674ef6 CH |
6130 | * Handle device private pages that are not accessible by the CPU, but |
6131 | * stored as special swap entries in the page table. | |
c733a828 JG |
6132 | */ |
6133 | if (is_device_private_entry(ent)) { | |
af5cdaf8 | 6134 | page = pfn_swap_entry_to_page(ent); |
27674ef6 | 6135 | if (!get_page_unless_zero(page)) |
c733a828 JG |
6136 | return NULL; |
6137 | return page; | |
6138 | } | |
6139 | ||
9a137153 RC |
6140 | if (non_swap_entry(ent)) |
6141 | return NULL; | |
6142 | ||
4b91355e | 6143 | /* |
cb691e2f | 6144 | * Because swap_cache_get_folio() updates some statistics counter, |
4b91355e KH |
6145 | * we call find_get_page() with swapper_space directly. |
6146 | */ | |
f6ab1f7f | 6147 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
2d1c4980 | 6148 | entry->val = ent.val; |
90254a65 DN |
6149 | |
6150 | return page; | |
6151 | } | |
4b91355e KH |
6152 | #else |
6153 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 6154 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
6155 | { |
6156 | return NULL; | |
6157 | } | |
6158 | #endif | |
90254a65 | 6159 | |
87946a72 | 6160 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
48384b0b | 6161 | unsigned long addr, pte_t ptent) |
87946a72 | 6162 | { |
524984ff MWO |
6163 | unsigned long index; |
6164 | struct folio *folio; | |
6165 | ||
87946a72 DN |
6166 | if (!vma->vm_file) /* anonymous vma */ |
6167 | return NULL; | |
1dfab5ab | 6168 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
6169 | return NULL; |
6170 | ||
524984ff | 6171 | /* folio is moved even if it's not RSS of this task(page-faulted). */ |
aa3b1895 | 6172 | /* shmem/tmpfs may report page out on swap: account for that too. */ |
524984ff MWO |
6173 | index = linear_page_index(vma, addr); |
6174 | folio = filemap_get_incore_folio(vma->vm_file->f_mapping, index); | |
66dabbb6 | 6175 | if (IS_ERR(folio)) |
524984ff MWO |
6176 | return NULL; |
6177 | return folio_file_page(folio, index); | |
87946a72 DN |
6178 | } |
6179 | ||
b1b0deab | 6180 | /** |
b267e1a3 MWO |
6181 | * mem_cgroup_move_account - move account of the folio |
6182 | * @folio: The folio. | |
25843c2b | 6183 | * @compound: charge the page as compound or small page |
b267e1a3 MWO |
6184 | * @from: mem_cgroup which the folio is moved from. |
6185 | * @to: mem_cgroup which the folio is moved to. @from != @to. | |
b1b0deab | 6186 | * |
b267e1a3 | 6187 | * The folio must be locked and not on the LRU. |
b1b0deab CG |
6188 | * |
6189 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
6190 | * from old cgroup. | |
6191 | */ | |
b267e1a3 | 6192 | static int mem_cgroup_move_account(struct folio *folio, |
f627c2f5 | 6193 | bool compound, |
b1b0deab CG |
6194 | struct mem_cgroup *from, |
6195 | struct mem_cgroup *to) | |
6196 | { | |
ae8af438 KK |
6197 | struct lruvec *from_vec, *to_vec; |
6198 | struct pglist_data *pgdat; | |
fcce4672 | 6199 | unsigned int nr_pages = compound ? folio_nr_pages(folio) : 1; |
8e88bd2d | 6200 | int nid, ret; |
b1b0deab CG |
6201 | |
6202 | VM_BUG_ON(from == to); | |
4e0cf05f | 6203 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
fcce4672 | 6204 | VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); |
9c325215 | 6205 | VM_BUG_ON(compound && !folio_test_large(folio)); |
b1b0deab | 6206 | |
b1b0deab | 6207 | ret = -EINVAL; |
fcce4672 | 6208 | if (folio_memcg(folio) != from) |
4e0cf05f | 6209 | goto out; |
b1b0deab | 6210 | |
fcce4672 | 6211 | pgdat = folio_pgdat(folio); |
867e5e1d JW |
6212 | from_vec = mem_cgroup_lruvec(from, pgdat); |
6213 | to_vec = mem_cgroup_lruvec(to, pgdat); | |
ae8af438 | 6214 | |
fcce4672 | 6215 | folio_memcg_lock(folio); |
b1b0deab | 6216 | |
fcce4672 MWO |
6217 | if (folio_test_anon(folio)) { |
6218 | if (folio_mapped(folio)) { | |
be5d0a74 JW |
6219 | __mod_lruvec_state(from_vec, NR_ANON_MAPPED, -nr_pages); |
6220 | __mod_lruvec_state(to_vec, NR_ANON_MAPPED, nr_pages); | |
6199277b | 6221 | if (folio_test_pmd_mappable(folio)) { |
69473e5d MS |
6222 | __mod_lruvec_state(from_vec, NR_ANON_THPS, |
6223 | -nr_pages); | |
6224 | __mod_lruvec_state(to_vec, NR_ANON_THPS, | |
6225 | nr_pages); | |
468c3982 | 6226 | } |
be5d0a74 JW |
6227 | } |
6228 | } else { | |
0d1c2072 JW |
6229 | __mod_lruvec_state(from_vec, NR_FILE_PAGES, -nr_pages); |
6230 | __mod_lruvec_state(to_vec, NR_FILE_PAGES, nr_pages); | |
6231 | ||
fcce4672 | 6232 | if (folio_test_swapbacked(folio)) { |
0d1c2072 JW |
6233 | __mod_lruvec_state(from_vec, NR_SHMEM, -nr_pages); |
6234 | __mod_lruvec_state(to_vec, NR_SHMEM, nr_pages); | |
6235 | } | |
6236 | ||
fcce4672 | 6237 | if (folio_mapped(folio)) { |
49e50d27 JW |
6238 | __mod_lruvec_state(from_vec, NR_FILE_MAPPED, -nr_pages); |
6239 | __mod_lruvec_state(to_vec, NR_FILE_MAPPED, nr_pages); | |
6240 | } | |
b1b0deab | 6241 | |
fcce4672 MWO |
6242 | if (folio_test_dirty(folio)) { |
6243 | struct address_space *mapping = folio_mapping(folio); | |
c4843a75 | 6244 | |
f56753ac | 6245 | if (mapping_can_writeback(mapping)) { |
49e50d27 JW |
6246 | __mod_lruvec_state(from_vec, NR_FILE_DIRTY, |
6247 | -nr_pages); | |
6248 | __mod_lruvec_state(to_vec, NR_FILE_DIRTY, | |
6249 | nr_pages); | |
6250 | } | |
c4843a75 GT |
6251 | } |
6252 | } | |
6253 | ||
c449deb2 HD |
6254 | #ifdef CONFIG_SWAP |
6255 | if (folio_test_swapcache(folio)) { | |
6256 | __mod_lruvec_state(from_vec, NR_SWAPCACHE, -nr_pages); | |
6257 | __mod_lruvec_state(to_vec, NR_SWAPCACHE, nr_pages); | |
6258 | } | |
6259 | #endif | |
fcce4672 | 6260 | if (folio_test_writeback(folio)) { |
ae8af438 KK |
6261 | __mod_lruvec_state(from_vec, NR_WRITEBACK, -nr_pages); |
6262 | __mod_lruvec_state(to_vec, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
6263 | } |
6264 | ||
6265 | /* | |
abb242f5 JW |
6266 | * All state has been migrated, let's switch to the new memcg. |
6267 | * | |
bcfe06bf | 6268 | * It is safe to change page's memcg here because the page |
abb242f5 JW |
6269 | * is referenced, charged, isolated, and locked: we can't race |
6270 | * with (un)charging, migration, LRU putback, or anything else | |
bcfe06bf | 6271 | * that would rely on a stable page's memory cgroup. |
abb242f5 | 6272 | * |
6c77b607 | 6273 | * Note that folio_memcg_lock is a memcg lock, not a page lock, |
bcfe06bf | 6274 | * to save space. As soon as we switch page's memory cgroup to a |
abb242f5 JW |
6275 | * new memcg that isn't locked, the above state can change |
6276 | * concurrently again. Make sure we're truly done with it. | |
b1b0deab | 6277 | */ |
abb242f5 | 6278 | smp_mb(); |
b1b0deab | 6279 | |
1a3e1f40 JW |
6280 | css_get(&to->css); |
6281 | css_put(&from->css); | |
6282 | ||
fcce4672 | 6283 | folio->memcg_data = (unsigned long)to; |
87eaceb3 | 6284 | |
f70ad448 | 6285 | __folio_memcg_unlock(from); |
b1b0deab CG |
6286 | |
6287 | ret = 0; | |
fcce4672 | 6288 | nid = folio_nid(folio); |
b1b0deab CG |
6289 | |
6290 | local_irq_disable(); | |
6e0110c2 | 6291 | mem_cgroup_charge_statistics(to, nr_pages); |
8e88bd2d | 6292 | memcg_check_events(to, nid); |
6e0110c2 | 6293 | mem_cgroup_charge_statistics(from, -nr_pages); |
8e88bd2d | 6294 | memcg_check_events(from, nid); |
b1b0deab | 6295 | local_irq_enable(); |
b1b0deab CG |
6296 | out: |
6297 | return ret; | |
6298 | } | |
6299 | ||
7cf7806c LR |
6300 | /** |
6301 | * get_mctgt_type - get target type of moving charge | |
6302 | * @vma: the vma the pte to be checked belongs | |
6303 | * @addr: the address corresponding to the pte to be checked | |
6304 | * @ptent: the pte to be checked | |
6305 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
6306 | * | |
853f62a3 MWO |
6307 | * Context: Called with pte lock held. |
6308 | * Return: | |
6309 | * * MC_TARGET_NONE - If the pte is not a target for move charge. | |
6310 | * * MC_TARGET_PAGE - If the page corresponding to this pte is a target for | |
b46777da | 6311 | * move charge. If @target is not NULL, the folio is stored in target->folio |
853f62a3 MWO |
6312 | * with extra refcnt taken (Caller should release it). |
6313 | * * MC_TARGET_SWAP - If the swap entry corresponding to this pte is a | |
6314 | * target for charge migration. If @target is not NULL, the entry is | |
6315 | * stored in target->ent. | |
6316 | * * MC_TARGET_DEVICE - Like MC_TARGET_PAGE but page is device memory and | |
6317 | * thus not on the lru. For now such page is charged like a regular page | |
6318 | * would be as it is just special memory taking the place of a regular page. | |
6319 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c | 6320 | */ |
8d32ff84 | 6321 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
6322 | unsigned long addr, pte_t ptent, union mc_target *target) |
6323 | { | |
6324 | struct page *page = NULL; | |
b67fa6e4 | 6325 | struct folio *folio; |
8d32ff84 | 6326 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
6327 | swp_entry_t ent = { .val = 0 }; |
6328 | ||
6329 | if (pte_present(ptent)) | |
6330 | page = mc_handle_present_pte(vma, addr, ptent); | |
5c041f5d PX |
6331 | else if (pte_none_mostly(ptent)) |
6332 | /* | |
6333 | * PTE markers should be treated as a none pte here, separated | |
6334 | * from other swap handling below. | |
6335 | */ | |
6336 | page = mc_handle_file_pte(vma, addr, ptent); | |
90254a65 | 6337 | else if (is_swap_pte(ptent)) |
48406ef8 | 6338 | page = mc_handle_swap_pte(vma, ptent, &ent); |
90254a65 | 6339 | |
b67fa6e4 MWO |
6340 | if (page) |
6341 | folio = page_folio(page); | |
4e0cf05f | 6342 | if (target && page) { |
b67fa6e4 MWO |
6343 | if (!folio_trylock(folio)) { |
6344 | folio_put(folio); | |
4e0cf05f JW |
6345 | return ret; |
6346 | } | |
6347 | /* | |
6348 | * page_mapped() must be stable during the move. This | |
6349 | * pte is locked, so if it's present, the page cannot | |
6350 | * become unmapped. If it isn't, we have only partial | |
6351 | * control over the mapped state: the page lock will | |
6352 | * prevent new faults against pagecache and swapcache, | |
6353 | * so an unmapped page cannot become mapped. However, | |
6354 | * if the page is already mapped elsewhere, it can | |
6355 | * unmap, and there is nothing we can do about it. | |
6356 | * Alas, skip moving the page in this case. | |
6357 | */ | |
6358 | if (!pte_present(ptent) && page_mapped(page)) { | |
b67fa6e4 MWO |
6359 | folio_unlock(folio); |
6360 | folio_put(folio); | |
4e0cf05f JW |
6361 | return ret; |
6362 | } | |
6363 | } | |
6364 | ||
90254a65 | 6365 | if (!page && !ent.val) |
8d32ff84 | 6366 | return ret; |
02491447 | 6367 | if (page) { |
02491447 | 6368 | /* |
0a31bc97 | 6369 | * Do only loose check w/o serialization. |
1306a85a | 6370 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 6371 | * not under LRU exclusion. |
02491447 | 6372 | */ |
b67fa6e4 | 6373 | if (folio_memcg(folio) == mc.from) { |
02491447 | 6374 | ret = MC_TARGET_PAGE; |
b67fa6e4 MWO |
6375 | if (folio_is_device_private(folio) || |
6376 | folio_is_device_coherent(folio)) | |
c733a828 | 6377 | ret = MC_TARGET_DEVICE; |
02491447 | 6378 | if (target) |
b67fa6e4 | 6379 | target->folio = folio; |
02491447 | 6380 | } |
4e0cf05f JW |
6381 | if (!ret || !target) { |
6382 | if (target) | |
b67fa6e4 MWO |
6383 | folio_unlock(folio); |
6384 | folio_put(folio); | |
4e0cf05f | 6385 | } |
02491447 | 6386 | } |
3e14a57b HY |
6387 | /* |
6388 | * There is a swap entry and a page doesn't exist or isn't charged. | |
6389 | * But we cannot move a tail-page in a THP. | |
6390 | */ | |
6391 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 6392 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
6393 | ret = MC_TARGET_SWAP; |
6394 | if (target) | |
6395 | target->ent = ent; | |
4ffef5fe | 6396 | } |
4ffef5fe DN |
6397 | return ret; |
6398 | } | |
6399 | ||
12724850 NH |
6400 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
6401 | /* | |
d6810d73 HY |
6402 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
6403 | * not support them for now. | |
12724850 NH |
6404 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
6405 | */ | |
6406 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6407 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6408 | { | |
6409 | struct page *page = NULL; | |
f6c7590b | 6410 | struct folio *folio; |
12724850 NH |
6411 | enum mc_target_type ret = MC_TARGET_NONE; |
6412 | ||
84c3fc4e ZY |
6413 | if (unlikely(is_swap_pmd(pmd))) { |
6414 | VM_BUG_ON(thp_migration_supported() && | |
6415 | !is_pmd_migration_entry(pmd)); | |
6416 | return ret; | |
6417 | } | |
12724850 | 6418 | page = pmd_page(pmd); |
309381fe | 6419 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
f6c7590b | 6420 | folio = page_folio(page); |
1dfab5ab | 6421 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 6422 | return ret; |
f6c7590b | 6423 | if (folio_memcg(folio) == mc.from) { |
12724850 NH |
6424 | ret = MC_TARGET_PAGE; |
6425 | if (target) { | |
f6c7590b MWO |
6426 | folio_get(folio); |
6427 | if (!folio_trylock(folio)) { | |
6428 | folio_put(folio); | |
4e0cf05f JW |
6429 | return MC_TARGET_NONE; |
6430 | } | |
f6c7590b | 6431 | target->folio = folio; |
12724850 NH |
6432 | } |
6433 | } | |
6434 | return ret; | |
6435 | } | |
6436 | #else | |
6437 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6438 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6439 | { | |
6440 | return MC_TARGET_NONE; | |
6441 | } | |
6442 | #endif | |
6443 | ||
4ffef5fe DN |
6444 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
6445 | unsigned long addr, unsigned long end, | |
6446 | struct mm_walk *walk) | |
6447 | { | |
26bcd64a | 6448 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
6449 | pte_t *pte; |
6450 | spinlock_t *ptl; | |
6451 | ||
b6ec57f4 KS |
6452 | ptl = pmd_trans_huge_lock(pmd, vma); |
6453 | if (ptl) { | |
c733a828 JG |
6454 | /* |
6455 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
25b2995a CH |
6456 | * support transparent huge page with MEMORY_DEVICE_PRIVATE but |
6457 | * this might change. | |
c733a828 | 6458 | */ |
12724850 NH |
6459 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
6460 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 6461 | spin_unlock(ptl); |
1a5a9906 | 6462 | return 0; |
12724850 | 6463 | } |
03319327 | 6464 | |
4ffef5fe | 6465 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
04dee9e8 HD |
6466 | if (!pte) |
6467 | return 0; | |
4ffef5fe | 6468 | for (; addr != end; pte++, addr += PAGE_SIZE) |
c33c7948 | 6469 | if (get_mctgt_type(vma, addr, ptep_get(pte), NULL)) |
4ffef5fe DN |
6470 | mc.precharge++; /* increment precharge temporarily */ |
6471 | pte_unmap_unlock(pte - 1, ptl); | |
6472 | cond_resched(); | |
6473 | ||
7dc74be0 DN |
6474 | return 0; |
6475 | } | |
6476 | ||
7b86ac33 CH |
6477 | static const struct mm_walk_ops precharge_walk_ops = { |
6478 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
49b06385 | 6479 | .walk_lock = PGWALK_RDLOCK, |
7b86ac33 CH |
6480 | }; |
6481 | ||
4ffef5fe DN |
6482 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
6483 | { | |
6484 | unsigned long precharge; | |
4ffef5fe | 6485 | |
d8ed45c5 | 6486 | mmap_read_lock(mm); |
ba0aff8e | 6487 | walk_page_range(mm, 0, ULONG_MAX, &precharge_walk_ops, NULL); |
d8ed45c5 | 6488 | mmap_read_unlock(mm); |
4ffef5fe DN |
6489 | |
6490 | precharge = mc.precharge; | |
6491 | mc.precharge = 0; | |
6492 | ||
6493 | return precharge; | |
6494 | } | |
6495 | ||
4ffef5fe DN |
6496 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
6497 | { | |
dfe076b0 DN |
6498 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
6499 | ||
6500 | VM_BUG_ON(mc.moving_task); | |
6501 | mc.moving_task = current; | |
6502 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
6503 | } |
6504 | ||
dfe076b0 DN |
6505 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
6506 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 6507 | { |
2bd9bb20 KH |
6508 | struct mem_cgroup *from = mc.from; |
6509 | struct mem_cgroup *to = mc.to; | |
6510 | ||
4ffef5fe | 6511 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 6512 | if (mc.precharge) { |
4b569387 | 6513 | mem_cgroup_cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
6514 | mc.precharge = 0; |
6515 | } | |
6516 | /* | |
6517 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
6518 | * we must uncharge here. | |
6519 | */ | |
6520 | if (mc.moved_charge) { | |
4b569387 | 6521 | mem_cgroup_cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 6522 | mc.moved_charge = 0; |
4ffef5fe | 6523 | } |
483c30b5 DN |
6524 | /* we must fixup refcnts and charges */ |
6525 | if (mc.moved_swap) { | |
483c30b5 | 6526 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 6527 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 6528 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 6529 | |
615d66c3 VD |
6530 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
6531 | ||
05b84301 | 6532 | /* |
3e32cb2e JW |
6533 | * we charged both to->memory and to->memsw, so we |
6534 | * should uncharge to->memory. | |
05b84301 | 6535 | */ |
ce00a967 | 6536 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
6537 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
6538 | ||
483c30b5 DN |
6539 | mc.moved_swap = 0; |
6540 | } | |
dfe076b0 DN |
6541 | memcg_oom_recover(from); |
6542 | memcg_oom_recover(to); | |
6543 | wake_up_all(&mc.waitq); | |
6544 | } | |
6545 | ||
6546 | static void mem_cgroup_clear_mc(void) | |
6547 | { | |
264a0ae1 TH |
6548 | struct mm_struct *mm = mc.mm; |
6549 | ||
dfe076b0 DN |
6550 | /* |
6551 | * we must clear moving_task before waking up waiters at the end of | |
6552 | * task migration. | |
6553 | */ | |
6554 | mc.moving_task = NULL; | |
6555 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 6556 | spin_lock(&mc.lock); |
4ffef5fe DN |
6557 | mc.from = NULL; |
6558 | mc.to = NULL; | |
264a0ae1 | 6559 | mc.mm = NULL; |
2bd9bb20 | 6560 | spin_unlock(&mc.lock); |
264a0ae1 TH |
6561 | |
6562 | mmput(mm); | |
4ffef5fe DN |
6563 | } |
6564 | ||
1f7dd3e5 | 6565 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 6566 | { |
1f7dd3e5 | 6567 | struct cgroup_subsys_state *css; |
eed67d75 | 6568 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 6569 | struct mem_cgroup *from; |
4530eddb | 6570 | struct task_struct *leader, *p; |
9f2115f9 | 6571 | struct mm_struct *mm; |
1dfab5ab | 6572 | unsigned long move_flags; |
9f2115f9 | 6573 | int ret = 0; |
7dc74be0 | 6574 | |
1f7dd3e5 TH |
6575 | /* charge immigration isn't supported on the default hierarchy */ |
6576 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
6577 | return 0; |
6578 | ||
4530eddb TH |
6579 | /* |
6580 | * Multi-process migrations only happen on the default hierarchy | |
6581 | * where charge immigration is not used. Perform charge | |
6582 | * immigration if @tset contains a leader and whine if there are | |
6583 | * multiple. | |
6584 | */ | |
6585 | p = NULL; | |
1f7dd3e5 | 6586 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
6587 | WARN_ON_ONCE(p); |
6588 | p = leader; | |
1f7dd3e5 | 6589 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
6590 | } |
6591 | if (!p) | |
6592 | return 0; | |
6593 | ||
1f7dd3e5 | 6594 | /* |
f0953a1b | 6595 | * We are now committed to this value whatever it is. Changes in this |
1f7dd3e5 TH |
6596 | * tunable will only affect upcoming migrations, not the current one. |
6597 | * So we need to save it, and keep it going. | |
6598 | */ | |
6599 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
6600 | if (!move_flags) | |
6601 | return 0; | |
6602 | ||
9f2115f9 TH |
6603 | from = mem_cgroup_from_task(p); |
6604 | ||
6605 | VM_BUG_ON(from == memcg); | |
6606 | ||
6607 | mm = get_task_mm(p); | |
6608 | if (!mm) | |
6609 | return 0; | |
6610 | /* We move charges only when we move a owner of the mm */ | |
6611 | if (mm->owner == p) { | |
6612 | VM_BUG_ON(mc.from); | |
6613 | VM_BUG_ON(mc.to); | |
6614 | VM_BUG_ON(mc.precharge); | |
6615 | VM_BUG_ON(mc.moved_charge); | |
6616 | VM_BUG_ON(mc.moved_swap); | |
6617 | ||
6618 | spin_lock(&mc.lock); | |
264a0ae1 | 6619 | mc.mm = mm; |
9f2115f9 TH |
6620 | mc.from = from; |
6621 | mc.to = memcg; | |
6622 | mc.flags = move_flags; | |
6623 | spin_unlock(&mc.lock); | |
6624 | /* We set mc.moving_task later */ | |
6625 | ||
6626 | ret = mem_cgroup_precharge_mc(mm); | |
6627 | if (ret) | |
6628 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
6629 | } else { |
6630 | mmput(mm); | |
7dc74be0 DN |
6631 | } |
6632 | return ret; | |
6633 | } | |
6634 | ||
1f7dd3e5 | 6635 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 6636 | { |
4e2f245d JW |
6637 | if (mc.to) |
6638 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
6639 | } |
6640 | ||
4ffef5fe DN |
6641 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
6642 | unsigned long addr, unsigned long end, | |
6643 | struct mm_walk *walk) | |
7dc74be0 | 6644 | { |
4ffef5fe | 6645 | int ret = 0; |
26bcd64a | 6646 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
6647 | pte_t *pte; |
6648 | spinlock_t *ptl; | |
12724850 NH |
6649 | enum mc_target_type target_type; |
6650 | union mc_target target; | |
b267e1a3 | 6651 | struct folio *folio; |
4ffef5fe | 6652 | |
b6ec57f4 KS |
6653 | ptl = pmd_trans_huge_lock(pmd, vma); |
6654 | if (ptl) { | |
62ade86a | 6655 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 6656 | spin_unlock(ptl); |
12724850 NH |
6657 | return 0; |
6658 | } | |
6659 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
6660 | if (target_type == MC_TARGET_PAGE) { | |
b46777da | 6661 | folio = target.folio; |
b267e1a3 MWO |
6662 | if (folio_isolate_lru(folio)) { |
6663 | if (!mem_cgroup_move_account(folio, true, | |
1306a85a | 6664 | mc.from, mc.to)) { |
12724850 NH |
6665 | mc.precharge -= HPAGE_PMD_NR; |
6666 | mc.moved_charge += HPAGE_PMD_NR; | |
6667 | } | |
b267e1a3 | 6668 | folio_putback_lru(folio); |
12724850 | 6669 | } |
b267e1a3 MWO |
6670 | folio_unlock(folio); |
6671 | folio_put(folio); | |
c733a828 | 6672 | } else if (target_type == MC_TARGET_DEVICE) { |
b46777da | 6673 | folio = target.folio; |
b267e1a3 | 6674 | if (!mem_cgroup_move_account(folio, true, |
c733a828 JG |
6675 | mc.from, mc.to)) { |
6676 | mc.precharge -= HPAGE_PMD_NR; | |
6677 | mc.moved_charge += HPAGE_PMD_NR; | |
6678 | } | |
b267e1a3 MWO |
6679 | folio_unlock(folio); |
6680 | folio_put(folio); | |
12724850 | 6681 | } |
bf929152 | 6682 | spin_unlock(ptl); |
1a5a9906 | 6683 | return 0; |
12724850 NH |
6684 | } |
6685 | ||
4ffef5fe DN |
6686 | retry: |
6687 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
04dee9e8 HD |
6688 | if (!pte) |
6689 | return 0; | |
4ffef5fe | 6690 | for (; addr != end; addr += PAGE_SIZE) { |
c33c7948 | 6691 | pte_t ptent = ptep_get(pte++); |
c733a828 | 6692 | bool device = false; |
02491447 | 6693 | swp_entry_t ent; |
4ffef5fe DN |
6694 | |
6695 | if (!mc.precharge) | |
6696 | break; | |
6697 | ||
8d32ff84 | 6698 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
6699 | case MC_TARGET_DEVICE: |
6700 | device = true; | |
e4a9bc58 | 6701 | fallthrough; |
4ffef5fe | 6702 | case MC_TARGET_PAGE: |
b46777da | 6703 | folio = target.folio; |
53f9263b KS |
6704 | /* |
6705 | * We can have a part of the split pmd here. Moving it | |
6706 | * can be done but it would be too convoluted so simply | |
6707 | * ignore such a partial THP and keep it in original | |
6708 | * memcg. There should be somebody mapping the head. | |
6709 | */ | |
b267e1a3 | 6710 | if (folio_test_large(folio)) |
53f9263b | 6711 | goto put; |
b267e1a3 | 6712 | if (!device && !folio_isolate_lru(folio)) |
4ffef5fe | 6713 | goto put; |
b267e1a3 | 6714 | if (!mem_cgroup_move_account(folio, false, |
f627c2f5 | 6715 | mc.from, mc.to)) { |
4ffef5fe | 6716 | mc.precharge--; |
854ffa8d DN |
6717 | /* we uncharge from mc.from later. */ |
6718 | mc.moved_charge++; | |
4ffef5fe | 6719 | } |
c733a828 | 6720 | if (!device) |
b267e1a3 | 6721 | folio_putback_lru(folio); |
4e0cf05f | 6722 | put: /* get_mctgt_type() gets & locks the page */ |
b267e1a3 MWO |
6723 | folio_unlock(folio); |
6724 | folio_put(folio); | |
4ffef5fe | 6725 | break; |
02491447 DN |
6726 | case MC_TARGET_SWAP: |
6727 | ent = target.ent; | |
e91cbb42 | 6728 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6729 | mc.precharge--; |
8d22a935 HD |
6730 | mem_cgroup_id_get_many(mc.to, 1); |
6731 | /* we fixup other refcnts and charges later. */ | |
483c30b5 DN |
6732 | mc.moved_swap++; |
6733 | } | |
02491447 | 6734 | break; |
4ffef5fe DN |
6735 | default: |
6736 | break; | |
6737 | } | |
6738 | } | |
6739 | pte_unmap_unlock(pte - 1, ptl); | |
6740 | cond_resched(); | |
6741 | ||
6742 | if (addr != end) { | |
6743 | /* | |
6744 | * We have consumed all precharges we got in can_attach(). | |
6745 | * We try charge one by one, but don't do any additional | |
6746 | * charges to mc.to if we have failed in charge once in attach() | |
6747 | * phase. | |
6748 | */ | |
854ffa8d | 6749 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6750 | if (!ret) |
6751 | goto retry; | |
6752 | } | |
6753 | ||
6754 | return ret; | |
6755 | } | |
6756 | ||
7b86ac33 CH |
6757 | static const struct mm_walk_ops charge_walk_ops = { |
6758 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
49b06385 | 6759 | .walk_lock = PGWALK_RDLOCK, |
7b86ac33 CH |
6760 | }; |
6761 | ||
264a0ae1 | 6762 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 6763 | { |
4ffef5fe | 6764 | lru_add_drain_all(); |
312722cb | 6765 | /* |
6c77b607 | 6766 | * Signal folio_memcg_lock() to take the memcg's move_lock |
81f8c3a4 JW |
6767 | * while we're moving its pages to another memcg. Then wait |
6768 | * for already started RCU-only updates to finish. | |
312722cb JW |
6769 | */ |
6770 | atomic_inc(&mc.from->moving_account); | |
6771 | synchronize_rcu(); | |
dfe076b0 | 6772 | retry: |
d8ed45c5 | 6773 | if (unlikely(!mmap_read_trylock(mc.mm))) { |
dfe076b0 | 6774 | /* |
c1e8d7c6 | 6775 | * Someone who are holding the mmap_lock might be waiting in |
dfe076b0 DN |
6776 | * waitq. So we cancel all extra charges, wake up all waiters, |
6777 | * and retry. Because we cancel precharges, we might not be able | |
6778 | * to move enough charges, but moving charge is a best-effort | |
6779 | * feature anyway, so it wouldn't be a big problem. | |
6780 | */ | |
6781 | __mem_cgroup_clear_mc(); | |
6782 | cond_resched(); | |
6783 | goto retry; | |
6784 | } | |
26bcd64a NH |
6785 | /* |
6786 | * When we have consumed all precharges and failed in doing | |
6787 | * additional charge, the page walk just aborts. | |
6788 | */ | |
ba0aff8e | 6789 | walk_page_range(mc.mm, 0, ULONG_MAX, &charge_walk_ops, NULL); |
d8ed45c5 | 6790 | mmap_read_unlock(mc.mm); |
312722cb | 6791 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
6792 | } |
6793 | ||
264a0ae1 | 6794 | static void mem_cgroup_move_task(void) |
67e465a7 | 6795 | { |
264a0ae1 TH |
6796 | if (mc.to) { |
6797 | mem_cgroup_move_charge(); | |
a433658c | 6798 | mem_cgroup_clear_mc(); |
264a0ae1 | 6799 | } |
67e465a7 | 6800 | } |
1aacbd35 | 6801 | |
5cfb80a7 | 6802 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 6803 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6804 | { |
6805 | return 0; | |
6806 | } | |
1f7dd3e5 | 6807 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6808 | { |
6809 | } | |
264a0ae1 | 6810 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
6811 | { |
6812 | } | |
6813 | #endif | |
67e465a7 | 6814 | |
1aacbd35 RG |
6815 | #ifdef CONFIG_MEMCG_KMEM |
6816 | static void mem_cgroup_fork(struct task_struct *task) | |
6817 | { | |
6818 | /* | |
6819 | * Set the update flag to cause task->objcg to be initialized lazily | |
6820 | * on the first allocation. It can be done without any synchronization | |
6821 | * because it's always performed on the current task, so does | |
6822 | * current_objcg_update(). | |
6823 | */ | |
6824 | task->objcg = (struct obj_cgroup *)CURRENT_OBJCG_UPDATE_FLAG; | |
6825 | } | |
6826 | ||
6827 | static void mem_cgroup_exit(struct task_struct *task) | |
6828 | { | |
6829 | struct obj_cgroup *objcg = task->objcg; | |
6830 | ||
6831 | objcg = (struct obj_cgroup *) | |
6832 | ((unsigned long)objcg & ~CURRENT_OBJCG_UPDATE_FLAG); | |
91b71e78 | 6833 | obj_cgroup_put(objcg); |
1aacbd35 RG |
6834 | |
6835 | /* | |
6836 | * Some kernel allocations can happen after this point, | |
6837 | * but let's ignore them. It can be done without any synchronization | |
6838 | * because it's always performed on the current task, so does | |
6839 | * current_objcg_update(). | |
6840 | */ | |
6841 | task->objcg = NULL; | |
6842 | } | |
6843 | #endif | |
6844 | ||
bd74fdae | 6845 | #ifdef CONFIG_LRU_GEN |
1aacbd35 | 6846 | static void mem_cgroup_lru_gen_attach(struct cgroup_taskset *tset) |
bd74fdae YZ |
6847 | { |
6848 | struct task_struct *task; | |
6849 | struct cgroup_subsys_state *css; | |
6850 | ||
6851 | /* find the first leader if there is any */ | |
6852 | cgroup_taskset_for_each_leader(task, css, tset) | |
6853 | break; | |
6854 | ||
6855 | if (!task) | |
6856 | return; | |
6857 | ||
6858 | task_lock(task); | |
6859 | if (task->mm && READ_ONCE(task->mm->owner) == task) | |
6860 | lru_gen_migrate_mm(task->mm); | |
6861 | task_unlock(task); | |
6862 | } | |
6863 | #else | |
1aacbd35 RG |
6864 | static void mem_cgroup_lru_gen_attach(struct cgroup_taskset *tset) {} |
6865 | #endif /* CONFIG_LRU_GEN */ | |
6866 | ||
6867 | #ifdef CONFIG_MEMCG_KMEM | |
6868 | static void mem_cgroup_kmem_attach(struct cgroup_taskset *tset) | |
6869 | { | |
6870 | struct task_struct *task; | |
6871 | struct cgroup_subsys_state *css; | |
6872 | ||
6873 | cgroup_taskset_for_each(task, css, tset) { | |
6874 | /* atomically set the update bit */ | |
6875 | set_bit(CURRENT_OBJCG_UPDATE_BIT, (unsigned long *)&task->objcg); | |
6876 | } | |
6877 | } | |
6878 | #else | |
6879 | static void mem_cgroup_kmem_attach(struct cgroup_taskset *tset) {} | |
6880 | #endif /* CONFIG_MEMCG_KMEM */ | |
6881 | ||
6882 | #if defined(CONFIG_LRU_GEN) || defined(CONFIG_MEMCG_KMEM) | |
bd74fdae YZ |
6883 | static void mem_cgroup_attach(struct cgroup_taskset *tset) |
6884 | { | |
1aacbd35 RG |
6885 | mem_cgroup_lru_gen_attach(tset); |
6886 | mem_cgroup_kmem_attach(tset); | |
bd74fdae | 6887 | } |
1aacbd35 | 6888 | #endif |
bd74fdae | 6889 | |
677dc973 CD |
6890 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
6891 | { | |
6892 | if (value == PAGE_COUNTER_MAX) | |
6893 | seq_puts(m, "max\n"); | |
6894 | else | |
6895 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); | |
6896 | ||
6897 | return 0; | |
6898 | } | |
6899 | ||
241994ed JW |
6900 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
6901 | struct cftype *cft) | |
6902 | { | |
f5fc3c5d JW |
6903 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
6904 | ||
6905 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
6906 | } |
6907 | ||
8e20d4b3 GR |
6908 | static u64 memory_peak_read(struct cgroup_subsys_state *css, |
6909 | struct cftype *cft) | |
6910 | { | |
6911 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
6912 | ||
6913 | return (u64)memcg->memory.watermark * PAGE_SIZE; | |
6914 | } | |
6915 | ||
bf8d5d52 RG |
6916 | static int memory_min_show(struct seq_file *m, void *v) |
6917 | { | |
677dc973 CD |
6918 | return seq_puts_memcg_tunable(m, |
6919 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); | |
bf8d5d52 RG |
6920 | } |
6921 | ||
6922 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
6923 | char *buf, size_t nbytes, loff_t off) | |
6924 | { | |
6925 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6926 | unsigned long min; | |
6927 | int err; | |
6928 | ||
6929 | buf = strstrip(buf); | |
6930 | err = page_counter_memparse(buf, "max", &min); | |
6931 | if (err) | |
6932 | return err; | |
6933 | ||
6934 | page_counter_set_min(&memcg->memory, min); | |
6935 | ||
6936 | return nbytes; | |
6937 | } | |
6938 | ||
241994ed JW |
6939 | static int memory_low_show(struct seq_file *m, void *v) |
6940 | { | |
677dc973 CD |
6941 | return seq_puts_memcg_tunable(m, |
6942 | READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); | |
241994ed JW |
6943 | } |
6944 | ||
6945 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
6946 | char *buf, size_t nbytes, loff_t off) | |
6947 | { | |
6948 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6949 | unsigned long low; | |
6950 | int err; | |
6951 | ||
6952 | buf = strstrip(buf); | |
d2973697 | 6953 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
6954 | if (err) |
6955 | return err; | |
6956 | ||
23067153 | 6957 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
6958 | |
6959 | return nbytes; | |
6960 | } | |
6961 | ||
6962 | static int memory_high_show(struct seq_file *m, void *v) | |
6963 | { | |
d1663a90 JK |
6964 | return seq_puts_memcg_tunable(m, |
6965 | READ_ONCE(mem_cgroup_from_seq(m)->memory.high)); | |
241994ed JW |
6966 | } |
6967 | ||
6968 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
6969 | char *buf, size_t nbytes, loff_t off) | |
6970 | { | |
6971 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6972 | unsigned int nr_retries = MAX_RECLAIM_RETRIES; |
8c8c383c | 6973 | bool drained = false; |
241994ed JW |
6974 | unsigned long high; |
6975 | int err; | |
6976 | ||
6977 | buf = strstrip(buf); | |
d2973697 | 6978 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
6979 | if (err) |
6980 | return err; | |
6981 | ||
e82553c1 JW |
6982 | page_counter_set_high(&memcg->memory, high); |
6983 | ||
8c8c383c JW |
6984 | for (;;) { |
6985 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6986 | unsigned long reclaimed; | |
6987 | ||
6988 | if (nr_pages <= high) | |
6989 | break; | |
6990 | ||
6991 | if (signal_pending(current)) | |
6992 | break; | |
6993 | ||
6994 | if (!drained) { | |
6995 | drain_all_stock(memcg); | |
6996 | drained = true; | |
6997 | continue; | |
6998 | } | |
6999 | ||
7000 | reclaimed = try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
55ab834a | 7001 | GFP_KERNEL, MEMCG_RECLAIM_MAY_SWAP); |
8c8c383c JW |
7002 | |
7003 | if (!reclaimed && !nr_retries--) | |
7004 | break; | |
7005 | } | |
588083bb | 7006 | |
19ce33ac | 7007 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
7008 | return nbytes; |
7009 | } | |
7010 | ||
7011 | static int memory_max_show(struct seq_file *m, void *v) | |
7012 | { | |
677dc973 CD |
7013 | return seq_puts_memcg_tunable(m, |
7014 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); | |
241994ed JW |
7015 | } |
7016 | ||
7017 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
7018 | char *buf, size_t nbytes, loff_t off) | |
7019 | { | |
7020 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 7021 | unsigned int nr_reclaims = MAX_RECLAIM_RETRIES; |
b6e6edcf | 7022 | bool drained = false; |
241994ed JW |
7023 | unsigned long max; |
7024 | int err; | |
7025 | ||
7026 | buf = strstrip(buf); | |
d2973697 | 7027 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
7028 | if (err) |
7029 | return err; | |
7030 | ||
bbec2e15 | 7031 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
7032 | |
7033 | for (;;) { | |
7034 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
7035 | ||
7036 | if (nr_pages <= max) | |
7037 | break; | |
7038 | ||
7249c9f0 | 7039 | if (signal_pending(current)) |
b6e6edcf | 7040 | break; |
b6e6edcf JW |
7041 | |
7042 | if (!drained) { | |
7043 | drain_all_stock(memcg); | |
7044 | drained = true; | |
7045 | continue; | |
7046 | } | |
7047 | ||
7048 | if (nr_reclaims) { | |
7049 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
55ab834a | 7050 | GFP_KERNEL, MEMCG_RECLAIM_MAY_SWAP)) |
b6e6edcf JW |
7051 | nr_reclaims--; |
7052 | continue; | |
7053 | } | |
7054 | ||
e27be240 | 7055 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
7056 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
7057 | break; | |
7058 | } | |
241994ed | 7059 | |
2529bb3a | 7060 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
7061 | return nbytes; |
7062 | } | |
7063 | ||
664dc218 DR |
7064 | /* |
7065 | * Note: don't forget to update the 'samples/cgroup/memcg_event_listener' | |
7066 | * if any new events become available. | |
7067 | */ | |
1e577f97 SB |
7068 | static void __memory_events_show(struct seq_file *m, atomic_long_t *events) |
7069 | { | |
7070 | seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW])); | |
7071 | seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH])); | |
7072 | seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX])); | |
7073 | seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM])); | |
7074 | seq_printf(m, "oom_kill %lu\n", | |
7075 | atomic_long_read(&events[MEMCG_OOM_KILL])); | |
b6bf9abb DS |
7076 | seq_printf(m, "oom_group_kill %lu\n", |
7077 | atomic_long_read(&events[MEMCG_OOM_GROUP_KILL])); | |
1e577f97 SB |
7078 | } |
7079 | ||
241994ed JW |
7080 | static int memory_events_show(struct seq_file *m, void *v) |
7081 | { | |
aa9694bb | 7082 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
241994ed | 7083 | |
1e577f97 SB |
7084 | __memory_events_show(m, memcg->memory_events); |
7085 | return 0; | |
7086 | } | |
7087 | ||
7088 | static int memory_events_local_show(struct seq_file *m, void *v) | |
7089 | { | |
7090 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
241994ed | 7091 | |
1e577f97 | 7092 | __memory_events_show(m, memcg->memory_events_local); |
241994ed JW |
7093 | return 0; |
7094 | } | |
7095 | ||
587d9f72 JW |
7096 | static int memory_stat_show(struct seq_file *m, void *v) |
7097 | { | |
aa9694bb | 7098 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
68aaee14 | 7099 | char *buf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
5b42360c | 7100 | struct seq_buf s; |
1ff9e6e1 | 7101 | |
c8713d0b JW |
7102 | if (!buf) |
7103 | return -ENOMEM; | |
5b42360c YA |
7104 | seq_buf_init(&s, buf, PAGE_SIZE); |
7105 | memory_stat_format(memcg, &s); | |
c8713d0b JW |
7106 | seq_puts(m, buf); |
7107 | kfree(buf); | |
587d9f72 JW |
7108 | return 0; |
7109 | } | |
7110 | ||
5f9a4f4a | 7111 | #ifdef CONFIG_NUMA |
fff66b79 MS |
7112 | static inline unsigned long lruvec_page_state_output(struct lruvec *lruvec, |
7113 | int item) | |
7114 | { | |
ff841a06 YA |
7115 | return lruvec_page_state(lruvec, item) * |
7116 | memcg_page_state_output_unit(item); | |
fff66b79 MS |
7117 | } |
7118 | ||
5f9a4f4a MS |
7119 | static int memory_numa_stat_show(struct seq_file *m, void *v) |
7120 | { | |
7121 | int i; | |
7122 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
7123 | ||
7d7ef0a4 | 7124 | mem_cgroup_flush_stats(memcg); |
7e1c0d6f | 7125 | |
5f9a4f4a MS |
7126 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { |
7127 | int nid; | |
7128 | ||
7129 | if (memory_stats[i].idx >= NR_VM_NODE_STAT_ITEMS) | |
7130 | continue; | |
7131 | ||
7132 | seq_printf(m, "%s", memory_stats[i].name); | |
7133 | for_each_node_state(nid, N_MEMORY) { | |
7134 | u64 size; | |
7135 | struct lruvec *lruvec; | |
7136 | ||
7137 | lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); | |
fff66b79 MS |
7138 | size = lruvec_page_state_output(lruvec, |
7139 | memory_stats[i].idx); | |
5f9a4f4a MS |
7140 | seq_printf(m, " N%d=%llu", nid, size); |
7141 | } | |
7142 | seq_putc(m, '\n'); | |
7143 | } | |
7144 | ||
7145 | return 0; | |
7146 | } | |
7147 | #endif | |
7148 | ||
3d8b38eb RG |
7149 | static int memory_oom_group_show(struct seq_file *m, void *v) |
7150 | { | |
aa9694bb | 7151 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3d8b38eb | 7152 | |
eaf7b66b | 7153 | seq_printf(m, "%d\n", READ_ONCE(memcg->oom_group)); |
3d8b38eb RG |
7154 | |
7155 | return 0; | |
7156 | } | |
7157 | ||
7158 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
7159 | char *buf, size_t nbytes, loff_t off) | |
7160 | { | |
7161 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7162 | int ret, oom_group; | |
7163 | ||
7164 | buf = strstrip(buf); | |
7165 | if (!buf) | |
7166 | return -EINVAL; | |
7167 | ||
7168 | ret = kstrtoint(buf, 0, &oom_group); | |
7169 | if (ret) | |
7170 | return ret; | |
7171 | ||
7172 | if (oom_group != 0 && oom_group != 1) | |
7173 | return -EINVAL; | |
7174 | ||
eaf7b66b | 7175 | WRITE_ONCE(memcg->oom_group, oom_group); |
3d8b38eb RG |
7176 | |
7177 | return nbytes; | |
7178 | } | |
7179 | ||
94968384 SB |
7180 | static ssize_t memory_reclaim(struct kernfs_open_file *of, char *buf, |
7181 | size_t nbytes, loff_t off) | |
7182 | { | |
7183 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7184 | unsigned int nr_retries = MAX_RECLAIM_RETRIES; | |
7185 | unsigned long nr_to_reclaim, nr_reclaimed = 0; | |
55ab834a MH |
7186 | unsigned int reclaim_options; |
7187 | int err; | |
12a5d395 MA |
7188 | |
7189 | buf = strstrip(buf); | |
55ab834a MH |
7190 | err = page_counter_memparse(buf, "", &nr_to_reclaim); |
7191 | if (err) | |
7192 | return err; | |
12a5d395 | 7193 | |
55ab834a | 7194 | reclaim_options = MEMCG_RECLAIM_MAY_SWAP | MEMCG_RECLAIM_PROACTIVE; |
94968384 | 7195 | while (nr_reclaimed < nr_to_reclaim) { |
287d5fed M |
7196 | /* Will converge on zero, but reclaim enforces a minimum */ |
7197 | unsigned long batch_size = (nr_to_reclaim - nr_reclaimed) / 4; | |
94968384 SB |
7198 | unsigned long reclaimed; |
7199 | ||
7200 | if (signal_pending(current)) | |
7201 | return -EINTR; | |
7202 | ||
7203 | /* | |
7204 | * This is the final attempt, drain percpu lru caches in the | |
7205 | * hope of introducing more evictable pages for | |
7206 | * try_to_free_mem_cgroup_pages(). | |
7207 | */ | |
7208 | if (!nr_retries) | |
7209 | lru_add_drain_all(); | |
7210 | ||
7211 | reclaimed = try_to_free_mem_cgroup_pages(memcg, | |
287d5fed | 7212 | batch_size, GFP_KERNEL, reclaim_options); |
94968384 SB |
7213 | |
7214 | if (!reclaimed && !nr_retries--) | |
7215 | return -EAGAIN; | |
7216 | ||
7217 | nr_reclaimed += reclaimed; | |
7218 | } | |
7219 | ||
7220 | return nbytes; | |
7221 | } | |
7222 | ||
241994ed JW |
7223 | static struct cftype memory_files[] = { |
7224 | { | |
7225 | .name = "current", | |
f5fc3c5d | 7226 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
7227 | .read_u64 = memory_current_read, |
7228 | }, | |
8e20d4b3 GR |
7229 | { |
7230 | .name = "peak", | |
7231 | .flags = CFTYPE_NOT_ON_ROOT, | |
7232 | .read_u64 = memory_peak_read, | |
7233 | }, | |
bf8d5d52 RG |
7234 | { |
7235 | .name = "min", | |
7236 | .flags = CFTYPE_NOT_ON_ROOT, | |
7237 | .seq_show = memory_min_show, | |
7238 | .write = memory_min_write, | |
7239 | }, | |
241994ed JW |
7240 | { |
7241 | .name = "low", | |
7242 | .flags = CFTYPE_NOT_ON_ROOT, | |
7243 | .seq_show = memory_low_show, | |
7244 | .write = memory_low_write, | |
7245 | }, | |
7246 | { | |
7247 | .name = "high", | |
7248 | .flags = CFTYPE_NOT_ON_ROOT, | |
7249 | .seq_show = memory_high_show, | |
7250 | .write = memory_high_write, | |
7251 | }, | |
7252 | { | |
7253 | .name = "max", | |
7254 | .flags = CFTYPE_NOT_ON_ROOT, | |
7255 | .seq_show = memory_max_show, | |
7256 | .write = memory_max_write, | |
7257 | }, | |
7258 | { | |
7259 | .name = "events", | |
7260 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 7261 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
7262 | .seq_show = memory_events_show, |
7263 | }, | |
1e577f97 SB |
7264 | { |
7265 | .name = "events.local", | |
7266 | .flags = CFTYPE_NOT_ON_ROOT, | |
7267 | .file_offset = offsetof(struct mem_cgroup, events_local_file), | |
7268 | .seq_show = memory_events_local_show, | |
7269 | }, | |
587d9f72 JW |
7270 | { |
7271 | .name = "stat", | |
587d9f72 JW |
7272 | .seq_show = memory_stat_show, |
7273 | }, | |
5f9a4f4a MS |
7274 | #ifdef CONFIG_NUMA |
7275 | { | |
7276 | .name = "numa_stat", | |
7277 | .seq_show = memory_numa_stat_show, | |
7278 | }, | |
7279 | #endif | |
3d8b38eb RG |
7280 | { |
7281 | .name = "oom.group", | |
7282 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
7283 | .seq_show = memory_oom_group_show, | |
7284 | .write = memory_oom_group_write, | |
7285 | }, | |
94968384 SB |
7286 | { |
7287 | .name = "reclaim", | |
7288 | .flags = CFTYPE_NS_DELEGATABLE, | |
7289 | .write = memory_reclaim, | |
7290 | }, | |
241994ed JW |
7291 | { } /* terminate */ |
7292 | }; | |
7293 | ||
073219e9 | 7294 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 7295 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 7296 | .css_online = mem_cgroup_css_online, |
92fb9748 | 7297 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 7298 | .css_released = mem_cgroup_css_released, |
92fb9748 | 7299 | .css_free = mem_cgroup_css_free, |
1ced953b | 7300 | .css_reset = mem_cgroup_css_reset, |
2d146aa3 | 7301 | .css_rstat_flush = mem_cgroup_css_rstat_flush, |
7dc74be0 | 7302 | .can_attach = mem_cgroup_can_attach, |
1aacbd35 | 7303 | #if defined(CONFIG_LRU_GEN) || defined(CONFIG_MEMCG_KMEM) |
bd74fdae | 7304 | .attach = mem_cgroup_attach, |
1aacbd35 | 7305 | #endif |
7dc74be0 | 7306 | .cancel_attach = mem_cgroup_cancel_attach, |
264a0ae1 | 7307 | .post_attach = mem_cgroup_move_task, |
1aacbd35 RG |
7308 | #ifdef CONFIG_MEMCG_KMEM |
7309 | .fork = mem_cgroup_fork, | |
7310 | .exit = mem_cgroup_exit, | |
7311 | #endif | |
241994ed JW |
7312 | .dfl_cftypes = memory_files, |
7313 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 7314 | .early_init = 0, |
8cdea7c0 | 7315 | }; |
c077719b | 7316 | |
bc50bcc6 JW |
7317 | /* |
7318 | * This function calculates an individual cgroup's effective | |
7319 | * protection which is derived from its own memory.min/low, its | |
7320 | * parent's and siblings' settings, as well as the actual memory | |
7321 | * distribution in the tree. | |
7322 | * | |
7323 | * The following rules apply to the effective protection values: | |
7324 | * | |
7325 | * 1. At the first level of reclaim, effective protection is equal to | |
7326 | * the declared protection in memory.min and memory.low. | |
7327 | * | |
7328 | * 2. To enable safe delegation of the protection configuration, at | |
7329 | * subsequent levels the effective protection is capped to the | |
7330 | * parent's effective protection. | |
7331 | * | |
7332 | * 3. To make complex and dynamic subtrees easier to configure, the | |
7333 | * user is allowed to overcommit the declared protection at a given | |
7334 | * level. If that is the case, the parent's effective protection is | |
7335 | * distributed to the children in proportion to how much protection | |
7336 | * they have declared and how much of it they are utilizing. | |
7337 | * | |
7338 | * This makes distribution proportional, but also work-conserving: | |
7339 | * if one cgroup claims much more protection than it uses memory, | |
7340 | * the unused remainder is available to its siblings. | |
7341 | * | |
7342 | * 4. Conversely, when the declared protection is undercommitted at a | |
7343 | * given level, the distribution of the larger parental protection | |
7344 | * budget is NOT proportional. A cgroup's protection from a sibling | |
7345 | * is capped to its own memory.min/low setting. | |
7346 | * | |
8a931f80 JW |
7347 | * 5. However, to allow protecting recursive subtrees from each other |
7348 | * without having to declare each individual cgroup's fixed share | |
7349 | * of the ancestor's claim to protection, any unutilized - | |
7350 | * "floating" - protection from up the tree is distributed in | |
7351 | * proportion to each cgroup's *usage*. This makes the protection | |
7352 | * neutral wrt sibling cgroups and lets them compete freely over | |
7353 | * the shared parental protection budget, but it protects the | |
7354 | * subtree as a whole from neighboring subtrees. | |
7355 | * | |
7356 | * Note that 4. and 5. are not in conflict: 4. is about protecting | |
7357 | * against immediate siblings whereas 5. is about protecting against | |
7358 | * neighboring subtrees. | |
bc50bcc6 JW |
7359 | */ |
7360 | static unsigned long effective_protection(unsigned long usage, | |
8a931f80 | 7361 | unsigned long parent_usage, |
bc50bcc6 JW |
7362 | unsigned long setting, |
7363 | unsigned long parent_effective, | |
7364 | unsigned long siblings_protected) | |
7365 | { | |
7366 | unsigned long protected; | |
8a931f80 | 7367 | unsigned long ep; |
bc50bcc6 JW |
7368 | |
7369 | protected = min(usage, setting); | |
7370 | /* | |
7371 | * If all cgroups at this level combined claim and use more | |
08e0f49e | 7372 | * protection than what the parent affords them, distribute |
bc50bcc6 JW |
7373 | * shares in proportion to utilization. |
7374 | * | |
7375 | * We are using actual utilization rather than the statically | |
7376 | * claimed protection in order to be work-conserving: claimed | |
7377 | * but unused protection is available to siblings that would | |
7378 | * otherwise get a smaller chunk than what they claimed. | |
7379 | */ | |
7380 | if (siblings_protected > parent_effective) | |
7381 | return protected * parent_effective / siblings_protected; | |
7382 | ||
7383 | /* | |
7384 | * Ok, utilized protection of all children is within what the | |
7385 | * parent affords them, so we know whatever this child claims | |
7386 | * and utilizes is effectively protected. | |
7387 | * | |
7388 | * If there is unprotected usage beyond this value, reclaim | |
7389 | * will apply pressure in proportion to that amount. | |
7390 | * | |
7391 | * If there is unutilized protection, the cgroup will be fully | |
7392 | * shielded from reclaim, but we do return a smaller value for | |
7393 | * protection than what the group could enjoy in theory. This | |
7394 | * is okay. With the overcommit distribution above, effective | |
7395 | * protection is always dependent on how memory is actually | |
7396 | * consumed among the siblings anyway. | |
7397 | */ | |
8a931f80 JW |
7398 | ep = protected; |
7399 | ||
7400 | /* | |
7401 | * If the children aren't claiming (all of) the protection | |
7402 | * afforded to them by the parent, distribute the remainder in | |
7403 | * proportion to the (unprotected) memory of each cgroup. That | |
7404 | * way, cgroups that aren't explicitly prioritized wrt each | |
7405 | * other compete freely over the allowance, but they are | |
7406 | * collectively protected from neighboring trees. | |
7407 | * | |
7408 | * We're using unprotected memory for the weight so that if | |
7409 | * some cgroups DO claim explicit protection, we don't protect | |
7410 | * the same bytes twice. | |
cd324edc JW |
7411 | * |
7412 | * Check both usage and parent_usage against the respective | |
7413 | * protected values. One should imply the other, but they | |
7414 | * aren't read atomically - make sure the division is sane. | |
8a931f80 JW |
7415 | */ |
7416 | if (!(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)) | |
7417 | return ep; | |
cd324edc JW |
7418 | if (parent_effective > siblings_protected && |
7419 | parent_usage > siblings_protected && | |
7420 | usage > protected) { | |
8a931f80 JW |
7421 | unsigned long unclaimed; |
7422 | ||
7423 | unclaimed = parent_effective - siblings_protected; | |
7424 | unclaimed *= usage - protected; | |
7425 | unclaimed /= parent_usage - siblings_protected; | |
7426 | ||
7427 | ep += unclaimed; | |
7428 | } | |
7429 | ||
7430 | return ep; | |
bc50bcc6 JW |
7431 | } |
7432 | ||
241994ed | 7433 | /** |
05395718 | 7434 | * mem_cgroup_calculate_protection - check if memory consumption is in the normal range |
34c81057 | 7435 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
7436 | * @memcg: the memory cgroup to check |
7437 | * | |
23067153 RG |
7438 | * WARNING: This function is not stateless! It can only be used as part |
7439 | * of a top-down tree iteration, not for isolated queries. | |
241994ed | 7440 | */ |
45c7f7e1 CD |
7441 | void mem_cgroup_calculate_protection(struct mem_cgroup *root, |
7442 | struct mem_cgroup *memcg) | |
241994ed | 7443 | { |
8a931f80 | 7444 | unsigned long usage, parent_usage; |
23067153 RG |
7445 | struct mem_cgroup *parent; |
7446 | ||
241994ed | 7447 | if (mem_cgroup_disabled()) |
45c7f7e1 | 7448 | return; |
241994ed | 7449 | |
34c81057 SC |
7450 | if (!root) |
7451 | root = root_mem_cgroup; | |
22f7496f YS |
7452 | |
7453 | /* | |
7454 | * Effective values of the reclaim targets are ignored so they | |
7455 | * can be stale. Have a look at mem_cgroup_protection for more | |
7456 | * details. | |
7457 | * TODO: calculation should be more robust so that we do not need | |
7458 | * that special casing. | |
7459 | */ | |
34c81057 | 7460 | if (memcg == root) |
45c7f7e1 | 7461 | return; |
241994ed | 7462 | |
23067153 | 7463 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 | 7464 | if (!usage) |
45c7f7e1 | 7465 | return; |
bf8d5d52 | 7466 | |
bf8d5d52 | 7467 | parent = parent_mem_cgroup(memcg); |
df2a4196 | 7468 | |
bc50bcc6 | 7469 | if (parent == root) { |
c3d53200 | 7470 | memcg->memory.emin = READ_ONCE(memcg->memory.min); |
03960e33 | 7471 | memcg->memory.elow = READ_ONCE(memcg->memory.low); |
45c7f7e1 | 7472 | return; |
bf8d5d52 RG |
7473 | } |
7474 | ||
8a931f80 JW |
7475 | parent_usage = page_counter_read(&parent->memory); |
7476 | ||
b3a7822e | 7477 | WRITE_ONCE(memcg->memory.emin, effective_protection(usage, parent_usage, |
c3d53200 CD |
7478 | READ_ONCE(memcg->memory.min), |
7479 | READ_ONCE(parent->memory.emin), | |
b3a7822e | 7480 | atomic_long_read(&parent->memory.children_min_usage))); |
23067153 | 7481 | |
b3a7822e | 7482 | WRITE_ONCE(memcg->memory.elow, effective_protection(usage, parent_usage, |
03960e33 CD |
7483 | READ_ONCE(memcg->memory.low), |
7484 | READ_ONCE(parent->memory.elow), | |
b3a7822e | 7485 | atomic_long_read(&parent->memory.children_low_usage))); |
241994ed JW |
7486 | } |
7487 | ||
8f425e4e MWO |
7488 | static int charge_memcg(struct folio *folio, struct mem_cgroup *memcg, |
7489 | gfp_t gfp) | |
0add0c77 | 7490 | { |
0add0c77 SB |
7491 | int ret; |
7492 | ||
4b569387 | 7493 | ret = try_charge(memcg, gfp, folio_nr_pages(folio)); |
0add0c77 SB |
7494 | if (ret) |
7495 | goto out; | |
7496 | ||
4b569387 | 7497 | mem_cgroup_commit_charge(folio, memcg); |
0add0c77 SB |
7498 | out: |
7499 | return ret; | |
7500 | } | |
7501 | ||
8f425e4e | 7502 | int __mem_cgroup_charge(struct folio *folio, struct mm_struct *mm, gfp_t gfp) |
00501b53 | 7503 | { |
0add0c77 SB |
7504 | struct mem_cgroup *memcg; |
7505 | int ret; | |
00501b53 | 7506 | |
0add0c77 | 7507 | memcg = get_mem_cgroup_from_mm(mm); |
8f425e4e | 7508 | ret = charge_memcg(folio, memcg, gfp); |
0add0c77 | 7509 | css_put(&memcg->css); |
2d1c4980 | 7510 | |
0add0c77 SB |
7511 | return ret; |
7512 | } | |
e993d905 | 7513 | |
8cba9576 NP |
7514 | /** |
7515 | * mem_cgroup_hugetlb_try_charge - try to charge the memcg for a hugetlb folio | |
7516 | * @memcg: memcg to charge. | |
7517 | * @gfp: reclaim mode. | |
7518 | * @nr_pages: number of pages to charge. | |
7519 | * | |
7520 | * This function is called when allocating a huge page folio to determine if | |
7521 | * the memcg has the capacity for it. It does not commit the charge yet, | |
7522 | * as the hugetlb folio itself has not been obtained from the hugetlb pool. | |
7523 | * | |
7524 | * Once we have obtained the hugetlb folio, we can call | |
7525 | * mem_cgroup_commit_charge() to commit the charge. If we fail to obtain the | |
7526 | * folio, we should instead call mem_cgroup_cancel_charge() to undo the effect | |
7527 | * of try_charge(). | |
7528 | * | |
7529 | * Returns 0 on success. Otherwise, an error code is returned. | |
7530 | */ | |
7531 | int mem_cgroup_hugetlb_try_charge(struct mem_cgroup *memcg, gfp_t gfp, | |
7532 | long nr_pages) | |
7533 | { | |
7534 | /* | |
7535 | * If hugetlb memcg charging is not enabled, do not fail hugetlb allocation, | |
7536 | * but do not attempt to commit charge later (or cancel on error) either. | |
7537 | */ | |
7538 | if (mem_cgroup_disabled() || !memcg || | |
7539 | !cgroup_subsys_on_dfl(memory_cgrp_subsys) || | |
7540 | !(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_HUGETLB_ACCOUNTING)) | |
7541 | return -EOPNOTSUPP; | |
7542 | ||
7543 | if (try_charge(memcg, gfp, nr_pages)) | |
7544 | return -ENOMEM; | |
7545 | ||
7546 | return 0; | |
7547 | } | |
7548 | ||
0add0c77 | 7549 | /** |
65995918 MWO |
7550 | * mem_cgroup_swapin_charge_folio - Charge a newly allocated folio for swapin. |
7551 | * @folio: folio to charge. | |
0add0c77 SB |
7552 | * @mm: mm context of the victim |
7553 | * @gfp: reclaim mode | |
65995918 | 7554 | * @entry: swap entry for which the folio is allocated |
0add0c77 | 7555 | * |
65995918 MWO |
7556 | * This function charges a folio allocated for swapin. Please call this before |
7557 | * adding the folio to the swapcache. | |
0add0c77 SB |
7558 | * |
7559 | * Returns 0 on success. Otherwise, an error code is returned. | |
7560 | */ | |
65995918 | 7561 | int mem_cgroup_swapin_charge_folio(struct folio *folio, struct mm_struct *mm, |
0add0c77 SB |
7562 | gfp_t gfp, swp_entry_t entry) |
7563 | { | |
7564 | struct mem_cgroup *memcg; | |
7565 | unsigned short id; | |
7566 | int ret; | |
00501b53 | 7567 | |
0add0c77 SB |
7568 | if (mem_cgroup_disabled()) |
7569 | return 0; | |
00501b53 | 7570 | |
0add0c77 SB |
7571 | id = lookup_swap_cgroup_id(entry); |
7572 | rcu_read_lock(); | |
7573 | memcg = mem_cgroup_from_id(id); | |
7574 | if (!memcg || !css_tryget_online(&memcg->css)) | |
7575 | memcg = get_mem_cgroup_from_mm(mm); | |
7576 | rcu_read_unlock(); | |
00501b53 | 7577 | |
8f425e4e | 7578 | ret = charge_memcg(folio, memcg, gfp); |
6abb5a86 | 7579 | |
0add0c77 SB |
7580 | css_put(&memcg->css); |
7581 | return ret; | |
7582 | } | |
00501b53 | 7583 | |
0add0c77 SB |
7584 | /* |
7585 | * mem_cgroup_swapin_uncharge_swap - uncharge swap slot | |
7586 | * @entry: swap entry for which the page is charged | |
7587 | * | |
7588 | * Call this function after successfully adding the charged page to swapcache. | |
7589 | * | |
7590 | * Note: This function assumes the page for which swap slot is being uncharged | |
7591 | * is order 0 page. | |
7592 | */ | |
7593 | void mem_cgroup_swapin_uncharge_swap(swp_entry_t entry) | |
7594 | { | |
cae3af62 MS |
7595 | /* |
7596 | * Cgroup1's unified memory+swap counter has been charged with the | |
7597 | * new swapcache page, finish the transfer by uncharging the swap | |
7598 | * slot. The swap slot would also get uncharged when it dies, but | |
7599 | * it can stick around indefinitely and we'd count the page twice | |
7600 | * the entire time. | |
7601 | * | |
7602 | * Cgroup2 has separate resource counters for memory and swap, | |
7603 | * so this is a non-issue here. Memory and swap charge lifetimes | |
7604 | * correspond 1:1 to page and swap slot lifetimes: we charge the | |
7605 | * page to memory here, and uncharge swap when the slot is freed. | |
7606 | */ | |
0add0c77 | 7607 | if (!mem_cgroup_disabled() && do_memsw_account()) { |
00501b53 JW |
7608 | /* |
7609 | * The swap entry might not get freed for a long time, | |
7610 | * let's not wait for it. The page already received a | |
7611 | * memory+swap charge, drop the swap entry duplicate. | |
7612 | */ | |
0add0c77 | 7613 | mem_cgroup_uncharge_swap(entry, 1); |
00501b53 | 7614 | } |
3fea5a49 JW |
7615 | } |
7616 | ||
a9d5adee JG |
7617 | struct uncharge_gather { |
7618 | struct mem_cgroup *memcg; | |
b4e0b68f | 7619 | unsigned long nr_memory; |
a9d5adee | 7620 | unsigned long pgpgout; |
a9d5adee | 7621 | unsigned long nr_kmem; |
8e88bd2d | 7622 | int nid; |
a9d5adee JG |
7623 | }; |
7624 | ||
7625 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 7626 | { |
a9d5adee JG |
7627 | memset(ug, 0, sizeof(*ug)); |
7628 | } | |
7629 | ||
7630 | static void uncharge_batch(const struct uncharge_gather *ug) | |
7631 | { | |
747db954 JW |
7632 | unsigned long flags; |
7633 | ||
b4e0b68f MS |
7634 | if (ug->nr_memory) { |
7635 | page_counter_uncharge(&ug->memcg->memory, ug->nr_memory); | |
7941d214 | 7636 | if (do_memsw_account()) |
b4e0b68f | 7637 | page_counter_uncharge(&ug->memcg->memsw, ug->nr_memory); |
a8c49af3 YA |
7638 | if (ug->nr_kmem) |
7639 | memcg_account_kmem(ug->memcg, -ug->nr_kmem); | |
a9d5adee | 7640 | memcg_oom_recover(ug->memcg); |
ce00a967 | 7641 | } |
747db954 JW |
7642 | |
7643 | local_irq_save(flags); | |
c9019e9b | 7644 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); |
b4e0b68f | 7645 | __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, ug->nr_memory); |
8e88bd2d | 7646 | memcg_check_events(ug->memcg, ug->nid); |
747db954 | 7647 | local_irq_restore(flags); |
f1796544 | 7648 | |
c4ed6ebf | 7649 | /* drop reference from uncharge_folio */ |
f1796544 | 7650 | css_put(&ug->memcg->css); |
a9d5adee JG |
7651 | } |
7652 | ||
c4ed6ebf | 7653 | static void uncharge_folio(struct folio *folio, struct uncharge_gather *ug) |
a9d5adee | 7654 | { |
c4ed6ebf | 7655 | long nr_pages; |
b4e0b68f MS |
7656 | struct mem_cgroup *memcg; |
7657 | struct obj_cgroup *objcg; | |
9f762dbe | 7658 | |
c4ed6ebf | 7659 | VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); |
b7b098cf MWO |
7660 | VM_BUG_ON_FOLIO(folio_order(folio) > 1 && |
7661 | !folio_test_hugetlb(folio) && | |
7662 | !list_empty(&folio->_deferred_list), folio); | |
a9d5adee | 7663 | |
a9d5adee JG |
7664 | /* |
7665 | * Nobody should be changing or seriously looking at | |
c4ed6ebf MWO |
7666 | * folio memcg or objcg at this point, we have fully |
7667 | * exclusive access to the folio. | |
a9d5adee | 7668 | */ |
fead2b86 | 7669 | if (folio_memcg_kmem(folio)) { |
1b7e4464 | 7670 | objcg = __folio_objcg(folio); |
b4e0b68f MS |
7671 | /* |
7672 | * This get matches the put at the end of the function and | |
7673 | * kmem pages do not hold memcg references anymore. | |
7674 | */ | |
7675 | memcg = get_mem_cgroup_from_objcg(objcg); | |
7676 | } else { | |
1b7e4464 | 7677 | memcg = __folio_memcg(folio); |
b4e0b68f | 7678 | } |
a9d5adee | 7679 | |
b4e0b68f MS |
7680 | if (!memcg) |
7681 | return; | |
7682 | ||
7683 | if (ug->memcg != memcg) { | |
a9d5adee JG |
7684 | if (ug->memcg) { |
7685 | uncharge_batch(ug); | |
7686 | uncharge_gather_clear(ug); | |
7687 | } | |
b4e0b68f | 7688 | ug->memcg = memcg; |
c4ed6ebf | 7689 | ug->nid = folio_nid(folio); |
f1796544 MH |
7690 | |
7691 | /* pairs with css_put in uncharge_batch */ | |
b4e0b68f | 7692 | css_get(&memcg->css); |
a9d5adee JG |
7693 | } |
7694 | ||
c4ed6ebf | 7695 | nr_pages = folio_nr_pages(folio); |
a9d5adee | 7696 | |
fead2b86 | 7697 | if (folio_memcg_kmem(folio)) { |
b4e0b68f | 7698 | ug->nr_memory += nr_pages; |
9f762dbe | 7699 | ug->nr_kmem += nr_pages; |
b4e0b68f | 7700 | |
c4ed6ebf | 7701 | folio->memcg_data = 0; |
b4e0b68f MS |
7702 | obj_cgroup_put(objcg); |
7703 | } else { | |
7704 | /* LRU pages aren't accounted at the root level */ | |
7705 | if (!mem_cgroup_is_root(memcg)) | |
7706 | ug->nr_memory += nr_pages; | |
18b2db3b | 7707 | ug->pgpgout++; |
a9d5adee | 7708 | |
c4ed6ebf | 7709 | folio->memcg_data = 0; |
b4e0b68f MS |
7710 | } |
7711 | ||
7712 | css_put(&memcg->css); | |
747db954 JW |
7713 | } |
7714 | ||
bbc6b703 | 7715 | void __mem_cgroup_uncharge(struct folio *folio) |
0a31bc97 | 7716 | { |
a9d5adee JG |
7717 | struct uncharge_gather ug; |
7718 | ||
bbc6b703 MWO |
7719 | /* Don't touch folio->lru of any random page, pre-check: */ |
7720 | if (!folio_memcg(folio)) | |
0a31bc97 JW |
7721 | return; |
7722 | ||
a9d5adee | 7723 | uncharge_gather_clear(&ug); |
bbc6b703 | 7724 | uncharge_folio(folio, &ug); |
a9d5adee | 7725 | uncharge_batch(&ug); |
747db954 | 7726 | } |
0a31bc97 | 7727 | |
4882c809 MWO |
7728 | void __mem_cgroup_uncharge_folios(struct folio_batch *folios) |
7729 | { | |
7730 | struct uncharge_gather ug; | |
7731 | unsigned int i; | |
7732 | ||
7733 | uncharge_gather_clear(&ug); | |
7734 | for (i = 0; i < folios->nr; i++) | |
7735 | uncharge_folio(folios->folios[i], &ug); | |
7736 | if (ug.memcg) | |
7737 | uncharge_batch(&ug); | |
0a31bc97 JW |
7738 | } |
7739 | ||
7740 | /** | |
85ce2c51 | 7741 | * mem_cgroup_replace_folio - Charge a folio's replacement. |
d21bba2b MWO |
7742 | * @old: Currently circulating folio. |
7743 | * @new: Replacement folio. | |
0a31bc97 | 7744 | * |
d21bba2b | 7745 | * Charge @new as a replacement folio for @old. @old will |
85ce2c51 NP |
7746 | * be uncharged upon free. This is only used by the page cache |
7747 | * (in replace_page_cache_folio()). | |
0a31bc97 | 7748 | * |
d21bba2b | 7749 | * Both folios must be locked, @new->mapping must be set up. |
0a31bc97 | 7750 | */ |
85ce2c51 | 7751 | void mem_cgroup_replace_folio(struct folio *old, struct folio *new) |
0a31bc97 | 7752 | { |
29833315 | 7753 | struct mem_cgroup *memcg; |
d21bba2b | 7754 | long nr_pages = folio_nr_pages(new); |
d93c4130 | 7755 | unsigned long flags; |
0a31bc97 | 7756 | |
d21bba2b MWO |
7757 | VM_BUG_ON_FOLIO(!folio_test_locked(old), old); |
7758 | VM_BUG_ON_FOLIO(!folio_test_locked(new), new); | |
7759 | VM_BUG_ON_FOLIO(folio_test_anon(old) != folio_test_anon(new), new); | |
7760 | VM_BUG_ON_FOLIO(folio_nr_pages(old) != nr_pages, new); | |
0a31bc97 JW |
7761 | |
7762 | if (mem_cgroup_disabled()) | |
7763 | return; | |
7764 | ||
d21bba2b MWO |
7765 | /* Page cache replacement: new folio already charged? */ |
7766 | if (folio_memcg(new)) | |
0a31bc97 JW |
7767 | return; |
7768 | ||
d21bba2b MWO |
7769 | memcg = folio_memcg(old); |
7770 | VM_WARN_ON_ONCE_FOLIO(!memcg, old); | |
29833315 | 7771 | if (!memcg) |
0a31bc97 JW |
7772 | return; |
7773 | ||
44b7a8d3 | 7774 | /* Force-charge the new page. The old one will be freed soon */ |
8dc87c7d MS |
7775 | if (!mem_cgroup_is_root(memcg)) { |
7776 | page_counter_charge(&memcg->memory, nr_pages); | |
7777 | if (do_memsw_account()) | |
7778 | page_counter_charge(&memcg->memsw, nr_pages); | |
7779 | } | |
0a31bc97 | 7780 | |
1a3e1f40 | 7781 | css_get(&memcg->css); |
d21bba2b | 7782 | commit_charge(new, memcg); |
44b7a8d3 | 7783 | |
d93c4130 | 7784 | local_irq_save(flags); |
6e0110c2 | 7785 | mem_cgroup_charge_statistics(memcg, nr_pages); |
d21bba2b | 7786 | memcg_check_events(memcg, folio_nid(new)); |
d93c4130 | 7787 | local_irq_restore(flags); |
0a31bc97 JW |
7788 | } |
7789 | ||
85ce2c51 NP |
7790 | /** |
7791 | * mem_cgroup_migrate - Transfer the memcg data from the old to the new folio. | |
7792 | * @old: Currently circulating folio. | |
7793 | * @new: Replacement folio. | |
7794 | * | |
7795 | * Transfer the memcg data from the old folio to the new folio for migration. | |
7796 | * The old folio's data info will be cleared. Note that the memory counters | |
7797 | * will remain unchanged throughout the process. | |
7798 | * | |
7799 | * Both folios must be locked, @new->mapping must be set up. | |
7800 | */ | |
7801 | void mem_cgroup_migrate(struct folio *old, struct folio *new) | |
7802 | { | |
7803 | struct mem_cgroup *memcg; | |
7804 | ||
7805 | VM_BUG_ON_FOLIO(!folio_test_locked(old), old); | |
7806 | VM_BUG_ON_FOLIO(!folio_test_locked(new), new); | |
7807 | VM_BUG_ON_FOLIO(folio_test_anon(old) != folio_test_anon(new), new); | |
7808 | VM_BUG_ON_FOLIO(folio_nr_pages(old) != folio_nr_pages(new), new); | |
7809 | ||
7810 | if (mem_cgroup_disabled()) | |
7811 | return; | |
7812 | ||
7813 | memcg = folio_memcg(old); | |
8cba9576 NP |
7814 | /* |
7815 | * Note that it is normal to see !memcg for a hugetlb folio. | |
7816 | * For e.g, itt could have been allocated when memory_hugetlb_accounting | |
7817 | * was not selected. | |
7818 | */ | |
7819 | VM_WARN_ON_ONCE_FOLIO(!folio_test_hugetlb(old) && !memcg, old); | |
85ce2c51 NP |
7820 | if (!memcg) |
7821 | return; | |
7822 | ||
7823 | /* Transfer the charge and the css ref */ | |
7824 | commit_charge(new, memcg); | |
9bcef597 BW |
7825 | /* |
7826 | * If the old folio is a large folio and is in the split queue, it needs | |
7827 | * to be removed from the split queue now, in case getting an incorrect | |
7828 | * split queue in destroy_large_folio() after the memcg of the old folio | |
7829 | * is cleared. | |
7830 | * | |
7831 | * In addition, the old folio is about to be freed after migration, so | |
7832 | * removing from the split queue a bit earlier seems reasonable. | |
7833 | */ | |
7834 | if (folio_test_large(old) && folio_test_large_rmappable(old)) | |
7835 | folio_undo_large_rmappable(old); | |
85ce2c51 NP |
7836 | old->memcg_data = 0; |
7837 | } | |
7838 | ||
ef12947c | 7839 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
7840 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
7841 | ||
2d758073 | 7842 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
7843 | { |
7844 | struct mem_cgroup *memcg; | |
7845 | ||
2d758073 JW |
7846 | if (!mem_cgroup_sockets_enabled) |
7847 | return; | |
7848 | ||
e876ecc6 | 7849 | /* Do not associate the sock with unrelated interrupted task's memcg. */ |
086f694a | 7850 | if (!in_task()) |
e876ecc6 SB |
7851 | return; |
7852 | ||
11092087 JW |
7853 | rcu_read_lock(); |
7854 | memcg = mem_cgroup_from_task(current); | |
7848ed62 | 7855 | if (mem_cgroup_is_root(memcg)) |
f7e1cb6e | 7856 | goto out; |
0db15298 | 7857 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 7858 | goto out; |
8965aa28 | 7859 | if (css_tryget(&memcg->css)) |
11092087 | 7860 | sk->sk_memcg = memcg; |
f7e1cb6e | 7861 | out: |
11092087 JW |
7862 | rcu_read_unlock(); |
7863 | } | |
11092087 | 7864 | |
2d758073 | 7865 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 7866 | { |
2d758073 JW |
7867 | if (sk->sk_memcg) |
7868 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
7869 | } |
7870 | ||
7871 | /** | |
7872 | * mem_cgroup_charge_skmem - charge socket memory | |
7873 | * @memcg: memcg to charge | |
7874 | * @nr_pages: number of pages to charge | |
4b1327be | 7875 | * @gfp_mask: reclaim mode |
11092087 JW |
7876 | * |
7877 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
4b1327be | 7878 | * @memcg's configured limit, %false if it doesn't. |
11092087 | 7879 | */ |
4b1327be WW |
7880 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages, |
7881 | gfp_t gfp_mask) | |
11092087 | 7882 | { |
f7e1cb6e | 7883 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 7884 | struct page_counter *fail; |
f7e1cb6e | 7885 | |
0db15298 JW |
7886 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
7887 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
7888 | return true; |
7889 | } | |
0db15298 | 7890 | memcg->tcpmem_pressure = 1; |
4b1327be WW |
7891 | if (gfp_mask & __GFP_NOFAIL) { |
7892 | page_counter_charge(&memcg->tcpmem, nr_pages); | |
7893 | return true; | |
7894 | } | |
f7e1cb6e | 7895 | return false; |
11092087 | 7896 | } |
d886f4e4 | 7897 | |
4b1327be WW |
7898 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) { |
7899 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); | |
f7e1cb6e | 7900 | return true; |
4b1327be | 7901 | } |
f7e1cb6e | 7902 | |
11092087 JW |
7903 | return false; |
7904 | } | |
7905 | ||
7906 | /** | |
7907 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
7908 | * @memcg: memcg to uncharge |
7909 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
7910 | */ |
7911 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
7912 | { | |
f7e1cb6e | 7913 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 7914 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
7915 | return; |
7916 | } | |
d886f4e4 | 7917 | |
c9019e9b | 7918 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 7919 | |
475d0487 | 7920 | refill_stock(memcg, nr_pages); |
11092087 JW |
7921 | } |
7922 | ||
f7e1cb6e JW |
7923 | static int __init cgroup_memory(char *s) |
7924 | { | |
7925 | char *token; | |
7926 | ||
7927 | while ((token = strsep(&s, ",")) != NULL) { | |
7928 | if (!*token) | |
7929 | continue; | |
7930 | if (!strcmp(token, "nosocket")) | |
7931 | cgroup_memory_nosocket = true; | |
04823c83 VD |
7932 | if (!strcmp(token, "nokmem")) |
7933 | cgroup_memory_nokmem = true; | |
b6c1a8af YS |
7934 | if (!strcmp(token, "nobpf")) |
7935 | cgroup_memory_nobpf = true; | |
f7e1cb6e | 7936 | } |
460a79e1 | 7937 | return 1; |
f7e1cb6e JW |
7938 | } |
7939 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 7940 | |
2d11085e | 7941 | /* |
1081312f MH |
7942 | * subsys_initcall() for memory controller. |
7943 | * | |
308167fc SAS |
7944 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
7945 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
7946 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
7947 | * should be initialized from here. | |
2d11085e MH |
7948 | */ |
7949 | static int __init mem_cgroup_init(void) | |
7950 | { | |
95a045f6 JW |
7951 | int cpu, node; |
7952 | ||
f3344adf MS |
7953 | /* |
7954 | * Currently s32 type (can refer to struct batched_lruvec_stat) is | |
7955 | * used for per-memcg-per-cpu caching of per-node statistics. In order | |
7956 | * to work fine, we should make sure that the overfill threshold can't | |
7957 | * exceed S32_MAX / PAGE_SIZE. | |
7958 | */ | |
7959 | BUILD_BUG_ON(MEMCG_CHARGE_BATCH > S32_MAX / PAGE_SIZE); | |
7960 | ||
308167fc SAS |
7961 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
7962 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
7963 | |
7964 | for_each_possible_cpu(cpu) | |
7965 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
7966 | drain_local_stock); | |
7967 | ||
7968 | for_each_node(node) { | |
7969 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 | 7970 | |
91f0dcce | 7971 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, node); |
95a045f6 | 7972 | |
ef8f2327 | 7973 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 7974 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 7975 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
7976 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
7977 | } | |
7978 | ||
2d11085e MH |
7979 | return 0; |
7980 | } | |
7981 | subsys_initcall(mem_cgroup_init); | |
21afa38e | 7982 | |
e55b9f96 | 7983 | #ifdef CONFIG_SWAP |
358c07fc AB |
7984 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
7985 | { | |
1c2d479a | 7986 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
7987 | /* |
7988 | * The root cgroup cannot be destroyed, so it's refcount must | |
7989 | * always be >= 1. | |
7990 | */ | |
7848ed62 | 7991 | if (WARN_ON_ONCE(mem_cgroup_is_root(memcg))) { |
358c07fc AB |
7992 | VM_BUG_ON(1); |
7993 | break; | |
7994 | } | |
7995 | memcg = parent_mem_cgroup(memcg); | |
7996 | if (!memcg) | |
7997 | memcg = root_mem_cgroup; | |
7998 | } | |
7999 | return memcg; | |
8000 | } | |
8001 | ||
21afa38e JW |
8002 | /** |
8003 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
3ecb0087 | 8004 | * @folio: folio whose memsw charge to transfer |
21afa38e JW |
8005 | * @entry: swap entry to move the charge to |
8006 | * | |
3ecb0087 | 8007 | * Transfer the memsw charge of @folio to @entry. |
21afa38e | 8008 | */ |
3ecb0087 | 8009 | void mem_cgroup_swapout(struct folio *folio, swp_entry_t entry) |
21afa38e | 8010 | { |
1f47b61f | 8011 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 8012 | unsigned int nr_entries; |
21afa38e JW |
8013 | unsigned short oldid; |
8014 | ||
3ecb0087 MWO |
8015 | VM_BUG_ON_FOLIO(folio_test_lru(folio), folio); |
8016 | VM_BUG_ON_FOLIO(folio_ref_count(folio), folio); | |
21afa38e | 8017 | |
76358ab5 AS |
8018 | if (mem_cgroup_disabled()) |
8019 | return; | |
8020 | ||
b94c4e94 | 8021 | if (!do_memsw_account()) |
21afa38e JW |
8022 | return; |
8023 | ||
3ecb0087 | 8024 | memcg = folio_memcg(folio); |
21afa38e | 8025 | |
3ecb0087 | 8026 | VM_WARN_ON_ONCE_FOLIO(!memcg, folio); |
21afa38e JW |
8027 | if (!memcg) |
8028 | return; | |
8029 | ||
1f47b61f VD |
8030 | /* |
8031 | * In case the memcg owning these pages has been offlined and doesn't | |
8032 | * have an ID allocated to it anymore, charge the closest online | |
8033 | * ancestor for the swap instead and transfer the memory+swap charge. | |
8034 | */ | |
8035 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
3ecb0087 | 8036 | nr_entries = folio_nr_pages(folio); |
d6810d73 HY |
8037 | /* Get references for the tail pages, too */ |
8038 | if (nr_entries > 1) | |
8039 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
8040 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
8041 | nr_entries); | |
3ecb0087 | 8042 | VM_BUG_ON_FOLIO(oldid, folio); |
c9019e9b | 8043 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e | 8044 | |
3ecb0087 | 8045 | folio->memcg_data = 0; |
21afa38e JW |
8046 | |
8047 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 8048 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 8049 | |
b25806dc | 8050 | if (memcg != swap_memcg) { |
1f47b61f | 8051 | if (!mem_cgroup_is_root(swap_memcg)) |
d6810d73 HY |
8052 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
8053 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
8054 | } |
8055 | ||
ce9ce665 SAS |
8056 | /* |
8057 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 8058 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 8059 | * important here to have the interrupts disabled because it is the |
b93b0163 | 8060 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 | 8061 | */ |
be3e67b5 | 8062 | memcg_stats_lock(); |
6e0110c2 | 8063 | mem_cgroup_charge_statistics(memcg, -nr_entries); |
be3e67b5 | 8064 | memcg_stats_unlock(); |
3ecb0087 | 8065 | memcg_check_events(memcg, folio_nid(folio)); |
73f576c0 | 8066 | |
1a3e1f40 | 8067 | css_put(&memcg->css); |
21afa38e JW |
8068 | } |
8069 | ||
38d8b4e6 | 8070 | /** |
e2e3fdc7 MWO |
8071 | * __mem_cgroup_try_charge_swap - try charging swap space for a folio |
8072 | * @folio: folio being added to swap | |
37e84351 VD |
8073 | * @entry: swap entry to charge |
8074 | * | |
e2e3fdc7 | 8075 | * Try to charge @folio's memcg for the swap space at @entry. |
37e84351 VD |
8076 | * |
8077 | * Returns 0 on success, -ENOMEM on failure. | |
8078 | */ | |
e2e3fdc7 | 8079 | int __mem_cgroup_try_charge_swap(struct folio *folio, swp_entry_t entry) |
37e84351 | 8080 | { |
e2e3fdc7 | 8081 | unsigned int nr_pages = folio_nr_pages(folio); |
37e84351 | 8082 | struct page_counter *counter; |
38d8b4e6 | 8083 | struct mem_cgroup *memcg; |
37e84351 VD |
8084 | unsigned short oldid; |
8085 | ||
b94c4e94 | 8086 | if (do_memsw_account()) |
37e84351 VD |
8087 | return 0; |
8088 | ||
e2e3fdc7 | 8089 | memcg = folio_memcg(folio); |
37e84351 | 8090 | |
e2e3fdc7 | 8091 | VM_WARN_ON_ONCE_FOLIO(!memcg, folio); |
37e84351 VD |
8092 | if (!memcg) |
8093 | return 0; | |
8094 | ||
f3a53a3a TH |
8095 | if (!entry.val) { |
8096 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 8097 | return 0; |
f3a53a3a | 8098 | } |
bb98f2c5 | 8099 | |
1f47b61f VD |
8100 | memcg = mem_cgroup_id_get_online(memcg); |
8101 | ||
b25806dc | 8102 | if (!mem_cgroup_is_root(memcg) && |
38d8b4e6 | 8103 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
8104 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
8105 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 8106 | mem_cgroup_id_put(memcg); |
37e84351 | 8107 | return -ENOMEM; |
1f47b61f | 8108 | } |
37e84351 | 8109 | |
38d8b4e6 HY |
8110 | /* Get references for the tail pages, too */ |
8111 | if (nr_pages > 1) | |
8112 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
8113 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
e2e3fdc7 | 8114 | VM_BUG_ON_FOLIO(oldid, folio); |
c9019e9b | 8115 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 8116 | |
37e84351 VD |
8117 | return 0; |
8118 | } | |
8119 | ||
21afa38e | 8120 | /** |
01c4b28c | 8121 | * __mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 8122 | * @entry: swap entry to uncharge |
38d8b4e6 | 8123 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 8124 | */ |
01c4b28c | 8125 | void __mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
8126 | { |
8127 | struct mem_cgroup *memcg; | |
8128 | unsigned short id; | |
8129 | ||
38d8b4e6 | 8130 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 8131 | rcu_read_lock(); |
adbe427b | 8132 | memcg = mem_cgroup_from_id(id); |
21afa38e | 8133 | if (memcg) { |
b25806dc | 8134 | if (!mem_cgroup_is_root(memcg)) { |
b94c4e94 | 8135 | if (do_memsw_account()) |
38d8b4e6 | 8136 | page_counter_uncharge(&memcg->memsw, nr_pages); |
b94c4e94 JW |
8137 | else |
8138 | page_counter_uncharge(&memcg->swap, nr_pages); | |
37e84351 | 8139 | } |
c9019e9b | 8140 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 8141 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
8142 | } |
8143 | rcu_read_unlock(); | |
8144 | } | |
8145 | ||
d8b38438 VD |
8146 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
8147 | { | |
8148 | long nr_swap_pages = get_nr_swap_pages(); | |
8149 | ||
b25806dc | 8150 | if (mem_cgroup_disabled() || do_memsw_account()) |
d8b38438 | 8151 | return nr_swap_pages; |
7848ed62 | 8152 | for (; !mem_cgroup_is_root(memcg); memcg = parent_mem_cgroup(memcg)) |
d8b38438 | 8153 | nr_swap_pages = min_t(long, nr_swap_pages, |
bbec2e15 | 8154 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
8155 | page_counter_read(&memcg->swap)); |
8156 | return nr_swap_pages; | |
8157 | } | |
8158 | ||
9202d527 | 8159 | bool mem_cgroup_swap_full(struct folio *folio) |
5ccc5aba VD |
8160 | { |
8161 | struct mem_cgroup *memcg; | |
8162 | ||
9202d527 | 8163 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
5ccc5aba VD |
8164 | |
8165 | if (vm_swap_full()) | |
8166 | return true; | |
b25806dc | 8167 | if (do_memsw_account()) |
5ccc5aba VD |
8168 | return false; |
8169 | ||
9202d527 | 8170 | memcg = folio_memcg(folio); |
5ccc5aba VD |
8171 | if (!memcg) |
8172 | return false; | |
8173 | ||
7848ed62 | 8174 | for (; !mem_cgroup_is_root(memcg); memcg = parent_mem_cgroup(memcg)) { |
4b82ab4f JK |
8175 | unsigned long usage = page_counter_read(&memcg->swap); |
8176 | ||
8177 | if (usage * 2 >= READ_ONCE(memcg->swap.high) || | |
8178 | usage * 2 >= READ_ONCE(memcg->swap.max)) | |
5ccc5aba | 8179 | return true; |
4b82ab4f | 8180 | } |
5ccc5aba VD |
8181 | |
8182 | return false; | |
8183 | } | |
8184 | ||
eccb52e7 | 8185 | static int __init setup_swap_account(char *s) |
21afa38e | 8186 | { |
118642d7 JW |
8187 | bool res; |
8188 | ||
8189 | if (!kstrtobool(s, &res) && !res) | |
8190 | pr_warn_once("The swapaccount=0 commandline option is deprecated " | |
8191 | "in favor of configuring swap control via cgroupfs. " | |
8192 | "Please report your usecase to linux-mm@kvack.org if you " | |
8193 | "depend on this functionality.\n"); | |
21afa38e JW |
8194 | return 1; |
8195 | } | |
eccb52e7 | 8196 | __setup("swapaccount=", setup_swap_account); |
21afa38e | 8197 | |
37e84351 VD |
8198 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
8199 | struct cftype *cft) | |
8200 | { | |
8201 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
8202 | ||
8203 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
8204 | } | |
8205 | ||
e0e0b412 LD |
8206 | static u64 swap_peak_read(struct cgroup_subsys_state *css, |
8207 | struct cftype *cft) | |
8208 | { | |
8209 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
8210 | ||
8211 | return (u64)memcg->swap.watermark * PAGE_SIZE; | |
8212 | } | |
8213 | ||
4b82ab4f JK |
8214 | static int swap_high_show(struct seq_file *m, void *v) |
8215 | { | |
8216 | return seq_puts_memcg_tunable(m, | |
8217 | READ_ONCE(mem_cgroup_from_seq(m)->swap.high)); | |
8218 | } | |
8219 | ||
8220 | static ssize_t swap_high_write(struct kernfs_open_file *of, | |
8221 | char *buf, size_t nbytes, loff_t off) | |
8222 | { | |
8223 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
8224 | unsigned long high; | |
8225 | int err; | |
8226 | ||
8227 | buf = strstrip(buf); | |
8228 | err = page_counter_memparse(buf, "max", &high); | |
8229 | if (err) | |
8230 | return err; | |
8231 | ||
8232 | page_counter_set_high(&memcg->swap, high); | |
8233 | ||
8234 | return nbytes; | |
8235 | } | |
8236 | ||
37e84351 VD |
8237 | static int swap_max_show(struct seq_file *m, void *v) |
8238 | { | |
677dc973 CD |
8239 | return seq_puts_memcg_tunable(m, |
8240 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); | |
37e84351 VD |
8241 | } |
8242 | ||
8243 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
8244 | char *buf, size_t nbytes, loff_t off) | |
8245 | { | |
8246 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
8247 | unsigned long max; | |
8248 | int err; | |
8249 | ||
8250 | buf = strstrip(buf); | |
8251 | err = page_counter_memparse(buf, "max", &max); | |
8252 | if (err) | |
8253 | return err; | |
8254 | ||
be09102b | 8255 | xchg(&memcg->swap.max, max); |
37e84351 VD |
8256 | |
8257 | return nbytes; | |
8258 | } | |
8259 | ||
f3a53a3a TH |
8260 | static int swap_events_show(struct seq_file *m, void *v) |
8261 | { | |
aa9694bb | 8262 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
f3a53a3a | 8263 | |
4b82ab4f JK |
8264 | seq_printf(m, "high %lu\n", |
8265 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_HIGH])); | |
f3a53a3a TH |
8266 | seq_printf(m, "max %lu\n", |
8267 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
8268 | seq_printf(m, "fail %lu\n", | |
8269 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
8270 | ||
8271 | return 0; | |
8272 | } | |
8273 | ||
37e84351 VD |
8274 | static struct cftype swap_files[] = { |
8275 | { | |
8276 | .name = "swap.current", | |
8277 | .flags = CFTYPE_NOT_ON_ROOT, | |
8278 | .read_u64 = swap_current_read, | |
8279 | }, | |
4b82ab4f JK |
8280 | { |
8281 | .name = "swap.high", | |
8282 | .flags = CFTYPE_NOT_ON_ROOT, | |
8283 | .seq_show = swap_high_show, | |
8284 | .write = swap_high_write, | |
8285 | }, | |
37e84351 VD |
8286 | { |
8287 | .name = "swap.max", | |
8288 | .flags = CFTYPE_NOT_ON_ROOT, | |
8289 | .seq_show = swap_max_show, | |
8290 | .write = swap_max_write, | |
8291 | }, | |
e0e0b412 LD |
8292 | { |
8293 | .name = "swap.peak", | |
8294 | .flags = CFTYPE_NOT_ON_ROOT, | |
8295 | .read_u64 = swap_peak_read, | |
8296 | }, | |
f3a53a3a TH |
8297 | { |
8298 | .name = "swap.events", | |
8299 | .flags = CFTYPE_NOT_ON_ROOT, | |
8300 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
8301 | .seq_show = swap_events_show, | |
8302 | }, | |
37e84351 VD |
8303 | { } /* terminate */ |
8304 | }; | |
8305 | ||
eccb52e7 | 8306 | static struct cftype memsw_files[] = { |
21afa38e JW |
8307 | { |
8308 | .name = "memsw.usage_in_bytes", | |
8309 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
8310 | .read_u64 = mem_cgroup_read_u64, | |
8311 | }, | |
8312 | { | |
8313 | .name = "memsw.max_usage_in_bytes", | |
8314 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
8315 | .write = mem_cgroup_reset, | |
8316 | .read_u64 = mem_cgroup_read_u64, | |
8317 | }, | |
8318 | { | |
8319 | .name = "memsw.limit_in_bytes", | |
8320 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
8321 | .write = mem_cgroup_write, | |
8322 | .read_u64 = mem_cgroup_read_u64, | |
8323 | }, | |
8324 | { | |
8325 | .name = "memsw.failcnt", | |
8326 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
8327 | .write = mem_cgroup_reset, | |
8328 | .read_u64 = mem_cgroup_read_u64, | |
8329 | }, | |
8330 | { }, /* terminate */ | |
8331 | }; | |
8332 | ||
f4840ccf JW |
8333 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) |
8334 | /** | |
8335 | * obj_cgroup_may_zswap - check if this cgroup can zswap | |
8336 | * @objcg: the object cgroup | |
8337 | * | |
8338 | * Check if the hierarchical zswap limit has been reached. | |
8339 | * | |
8340 | * This doesn't check for specific headroom, and it is not atomic | |
8341 | * either. But with zswap, the size of the allocation is only known | |
be16dd76 | 8342 | * once compression has occurred, and this optimistic pre-check avoids |
f4840ccf JW |
8343 | * spending cycles on compression when there is already no room left |
8344 | * or zswap is disabled altogether somewhere in the hierarchy. | |
8345 | */ | |
8346 | bool obj_cgroup_may_zswap(struct obj_cgroup *objcg) | |
8347 | { | |
8348 | struct mem_cgroup *memcg, *original_memcg; | |
8349 | bool ret = true; | |
8350 | ||
8351 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
8352 | return true; | |
8353 | ||
8354 | original_memcg = get_mem_cgroup_from_objcg(objcg); | |
7848ed62 | 8355 | for (memcg = original_memcg; !mem_cgroup_is_root(memcg); |
f4840ccf JW |
8356 | memcg = parent_mem_cgroup(memcg)) { |
8357 | unsigned long max = READ_ONCE(memcg->zswap_max); | |
8358 | unsigned long pages; | |
8359 | ||
8360 | if (max == PAGE_COUNTER_MAX) | |
8361 | continue; | |
8362 | if (max == 0) { | |
8363 | ret = false; | |
8364 | break; | |
8365 | } | |
8366 | ||
7d7ef0a4 YA |
8367 | /* |
8368 | * mem_cgroup_flush_stats() ignores small changes. Use | |
8369 | * do_flush_stats() directly to get accurate stats for charging. | |
8370 | */ | |
8371 | do_flush_stats(memcg); | |
f4840ccf JW |
8372 | pages = memcg_page_state(memcg, MEMCG_ZSWAP_B) / PAGE_SIZE; |
8373 | if (pages < max) | |
8374 | continue; | |
8375 | ret = false; | |
8376 | break; | |
8377 | } | |
8378 | mem_cgroup_put(original_memcg); | |
8379 | return ret; | |
8380 | } | |
8381 | ||
8382 | /** | |
8383 | * obj_cgroup_charge_zswap - charge compression backend memory | |
8384 | * @objcg: the object cgroup | |
8385 | * @size: size of compressed object | |
8386 | * | |
3a1060c2 | 8387 | * This forces the charge after obj_cgroup_may_zswap() allowed |
f4840ccf JW |
8388 | * compression and storage in zwap for this cgroup to go ahead. |
8389 | */ | |
8390 | void obj_cgroup_charge_zswap(struct obj_cgroup *objcg, size_t size) | |
8391 | { | |
8392 | struct mem_cgroup *memcg; | |
8393 | ||
8394 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
8395 | return; | |
8396 | ||
8397 | VM_WARN_ON_ONCE(!(current->flags & PF_MEMALLOC)); | |
8398 | ||
8399 | /* PF_MEMALLOC context, charging must succeed */ | |
8400 | if (obj_cgroup_charge(objcg, GFP_KERNEL, size)) | |
8401 | VM_WARN_ON_ONCE(1); | |
8402 | ||
8403 | rcu_read_lock(); | |
8404 | memcg = obj_cgroup_memcg(objcg); | |
8405 | mod_memcg_state(memcg, MEMCG_ZSWAP_B, size); | |
8406 | mod_memcg_state(memcg, MEMCG_ZSWAPPED, 1); | |
8407 | rcu_read_unlock(); | |
8408 | } | |
8409 | ||
8410 | /** | |
8411 | * obj_cgroup_uncharge_zswap - uncharge compression backend memory | |
8412 | * @objcg: the object cgroup | |
8413 | * @size: size of compressed object | |
8414 | * | |
8415 | * Uncharges zswap memory on page in. | |
8416 | */ | |
8417 | void obj_cgroup_uncharge_zswap(struct obj_cgroup *objcg, size_t size) | |
8418 | { | |
8419 | struct mem_cgroup *memcg; | |
8420 | ||
8421 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
8422 | return; | |
8423 | ||
8424 | obj_cgroup_uncharge(objcg, size); | |
8425 | ||
8426 | rcu_read_lock(); | |
8427 | memcg = obj_cgroup_memcg(objcg); | |
8428 | mod_memcg_state(memcg, MEMCG_ZSWAP_B, -size); | |
8429 | mod_memcg_state(memcg, MEMCG_ZSWAPPED, -1); | |
8430 | rcu_read_unlock(); | |
8431 | } | |
8432 | ||
501a06fe NP |
8433 | bool mem_cgroup_zswap_writeback_enabled(struct mem_cgroup *memcg) |
8434 | { | |
8435 | /* if zswap is disabled, do not block pages going to the swapping device */ | |
8436 | return !is_zswap_enabled() || !memcg || READ_ONCE(memcg->zswap_writeback); | |
8437 | } | |
8438 | ||
f4840ccf JW |
8439 | static u64 zswap_current_read(struct cgroup_subsys_state *css, |
8440 | struct cftype *cft) | |
8441 | { | |
7d7ef0a4 YA |
8442 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
8443 | ||
8444 | mem_cgroup_flush_stats(memcg); | |
8445 | return memcg_page_state(memcg, MEMCG_ZSWAP_B); | |
f4840ccf JW |
8446 | } |
8447 | ||
8448 | static int zswap_max_show(struct seq_file *m, void *v) | |
8449 | { | |
8450 | return seq_puts_memcg_tunable(m, | |
8451 | READ_ONCE(mem_cgroup_from_seq(m)->zswap_max)); | |
8452 | } | |
8453 | ||
8454 | static ssize_t zswap_max_write(struct kernfs_open_file *of, | |
8455 | char *buf, size_t nbytes, loff_t off) | |
8456 | { | |
8457 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
8458 | unsigned long max; | |
8459 | int err; | |
8460 | ||
8461 | buf = strstrip(buf); | |
8462 | err = page_counter_memparse(buf, "max", &max); | |
8463 | if (err) | |
8464 | return err; | |
8465 | ||
8466 | xchg(&memcg->zswap_max, max); | |
8467 | ||
8468 | return nbytes; | |
8469 | } | |
8470 | ||
501a06fe NP |
8471 | static int zswap_writeback_show(struct seq_file *m, void *v) |
8472 | { | |
8473 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
8474 | ||
8475 | seq_printf(m, "%d\n", READ_ONCE(memcg->zswap_writeback)); | |
8476 | return 0; | |
8477 | } | |
8478 | ||
8479 | static ssize_t zswap_writeback_write(struct kernfs_open_file *of, | |
8480 | char *buf, size_t nbytes, loff_t off) | |
8481 | { | |
8482 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
8483 | int zswap_writeback; | |
8484 | ssize_t parse_ret = kstrtoint(strstrip(buf), 0, &zswap_writeback); | |
8485 | ||
8486 | if (parse_ret) | |
8487 | return parse_ret; | |
8488 | ||
8489 | if (zswap_writeback != 0 && zswap_writeback != 1) | |
8490 | return -EINVAL; | |
8491 | ||
8492 | WRITE_ONCE(memcg->zswap_writeback, zswap_writeback); | |
8493 | return nbytes; | |
8494 | } | |
8495 | ||
f4840ccf JW |
8496 | static struct cftype zswap_files[] = { |
8497 | { | |
8498 | .name = "zswap.current", | |
8499 | .flags = CFTYPE_NOT_ON_ROOT, | |
8500 | .read_u64 = zswap_current_read, | |
8501 | }, | |
8502 | { | |
8503 | .name = "zswap.max", | |
8504 | .flags = CFTYPE_NOT_ON_ROOT, | |
8505 | .seq_show = zswap_max_show, | |
8506 | .write = zswap_max_write, | |
8507 | }, | |
501a06fe NP |
8508 | { |
8509 | .name = "zswap.writeback", | |
8510 | .seq_show = zswap_writeback_show, | |
8511 | .write = zswap_writeback_write, | |
8512 | }, | |
f4840ccf JW |
8513 | { } /* terminate */ |
8514 | }; | |
8515 | #endif /* CONFIG_MEMCG_KMEM && CONFIG_ZSWAP */ | |
8516 | ||
21afa38e JW |
8517 | static int __init mem_cgroup_swap_init(void) |
8518 | { | |
2d1c4980 | 8519 | if (mem_cgroup_disabled()) |
eccb52e7 JW |
8520 | return 0; |
8521 | ||
8522 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, swap_files)); | |
8523 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, memsw_files)); | |
f4840ccf JW |
8524 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_ZSWAP) |
8525 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, zswap_files)); | |
8526 | #endif | |
21afa38e JW |
8527 | return 0; |
8528 | } | |
b25806dc | 8529 | subsys_initcall(mem_cgroup_swap_init); |
21afa38e | 8530 | |
e55b9f96 | 8531 | #endif /* CONFIG_SWAP */ |